HOW D E A D L A N G UAG E S WO R K

How Dead
Languages Work

C OU LT E R  H .  G E O R G E

1
 1
Great Clarendon Street, Oxford, OX2 6DP,
United Kingdom
Oxford University Press is a department of the University of Oxford.
It furthers the University’s objective of excellence in research, scholarship,
and education by publishing worldwide. Oxford is a registered trade mark of
Oxford University Press in the UK and in certain other countries
© Coulter H. George 2020
The moral rights of the author have been asserted
First Edition published in 2020
Impression: 1
All rights reserved. No part of this publication may be reproduced, stored in
a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by licence or under terms agreed with the appropriate reprographics
rights organization. Enquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above
You must not circulate this work in any other form
and you must impose this same condition on any acquirer
Published in the United States of America by Oxford University Press
198 Madison Avenue, New York, NY 10016, United States of America
British Library Cataloguing in Publication Data
Data available
Library of Congress Control Number: 2019950243
ISBN 978–0–19–885282–7
Printed and bound in Great Britain by
Clays Ltd, Elcograf S.p.A.
Links to third party websites are provided by Oxford in good faith and
for information only. Oxford disclaims any responsibility for the materials
contained in any third party website referenced in this work.
 Acknowledgments

Languages, especially dead ones, are an endless source of intellectual delight,

and my first debt of gratitude is to the many outstanding teachers I’ve had the
immense good fortune to study with over the years. Collectively, they have
advocated for Greek, Latin, and all the rest of the languages I cover in this book
in a kaleidoscopic variety of ways. While Harvey Yunis and James Clackson
were particularly instrumental in giving me the intellectual tools to work with
Greek and Latin, I would here especially like to record my gratitude to my
Sanskrit teacher, the late Douglas Mitchell, who never ceased in his own inim-
itable manner to underline the importance of viewing every linguistic expres-
sion as possessed of its own unique value. I have learned much from all of them
and, indeed, from the rich secondary literature in the field—although I have
preferred in what follows to give pride of place to the languages themselves
rather than scholarly debates about controversial points, and so have been
sparing with footnotes.
The opportunities provided by my home institution, the University of
Virginia, were also crucial in seeing this project through. For several years I
taught an introductory seminar for first-year undergraduates here, in which we
looked at texts in seven different ancient and medieval languages over the
course of the semester: I am grateful to my students for their willingness to
serve as guinea pigs for much of the material in this book. The university’s
generous Sesqui Leave (2016–17) freed up the time for actually writing much of
the manuscript. And I especially continue to treasure the wonderful commu-
nity of classicists here. For the many evenings of good cheer they’ve provided,
particular thanks go to the Wednesday Night Bachelors—including those who
have moved away from Charlottesville and are much missed. Within their
number, Tony Woodman deserves special credit for reading through the entire
manuscript and saving me from one appallingly egregious Tacitean blunder, as
does Elizabeth Meyer, whose unfailing thoughtfulness in her giving of gifts
helped inspire me to write a book for non-specialists in the first place.
I am grateful to Oxford University Press for their work in seeing the book
through production: Charlotte Loveridge, Georgie Leighton, Jenny King, Lucía
Pérez, and Michael Janes have all been very helpful along the way. Both Joshua
Katz and an anonymous reader for the press read through the whole book and
offered invaluable corrections and suggestions. Hilary Bouxsein also improved
the book immeasurably with her consistently perspicacious advice on multiple
drafts as well as her general good spirits: I only hope that the words that follow
are not only ἐτύμοισιν ὁμοῖα but downright ἀληθέα. Finally, I would like to
thank two other people, who, if indirectly and unknowingly, also played a role
vi Acknowledgments

in the book’s genesis. My grandmothers, Miriam Young and Lila Gene George,
to whom I owe my appreciation of good food, fine music, and countless else
besides, always showed unflagging interest in my linguistic pursuits, no matter
how arcane they became. While they are no longer with us, it is in no small part
thanks to them that it’s been so important to me to explain to a wider audience
what it is that I find so deeply fascinating about these languages.
Coulter H. George
Charlottesville
 Contents

1. Introduction 1
2. Greek 13
The sounds of Greek 13
Word forms 18
The Iliad 27
Formulas 35
Thucydides and abstract language 41
Pauline prepositions 49
3. Latin 57
How Latin works 59
Lucretius 67
Horace and Housman 78
Tacitus 88
4. Old English and the Germanic languages 99
Grimm’s Law and umlaut 100
Verbs, strong and weak 107
Old English 114
Beowulf 119
5. Sanskrit 133
The sounds of Sanskrit 136
Sandhi 141
Nouns in Sanskrit 142
The Rig Veda 146
How to kill a dragon 148
The hidden names of the dawn-cows 152
6. Old Irish and the Celtic languages 157
The eccentricities of the Irish language 157
Old Irish in action 173
Welsh 178
viii Contents

7. Hebrew 189
The sounds of Semitic 189
How Semitic words change shape 195
Let there be light 199
Noun chains 205
Epilogue and further reading 213

Index 219
 1

Introduction

Why study Ancient Greek? Classicists are asked this all the time, and most will
answer by pointing to the cultural achievements of the Greeks: this is the
­language of the first great epic poetry (Homer), historians (Herodotus and
Thucydides), playwrights (Aeschylus, Sophocles, Euripides, Aristophanes),
and philosophers (Plato, Aristotle). Sure enough, wanting to gain a proper
appreciation and understanding of any one of these texts would be more than
enough reason on its own to learn Ancient Greek. But it’s not actually the main
reason that I’ve stayed fascinated by the language for so many years. When
asked once what I liked best about Greek literature, I had to respond that it
wasn’t so much the content of all these great works, as impressive a cultural
achievement as they represent, but rather everything about them that can’t be
translated into English. Because Greek, like all languages, has its own dis­tinct­
ive features that don’t survive the process of translation. It has its own flavor.
And this book is, in part, an attempt to explain to the Greekless reader what
that means. Of course, the best way to understand what is lost in translation is
to learn Greek. But it’s not as if most people have time for that, and it is with
them in mind that I’ve tried to explain as best as I can what makes Greek Greek.
But Greek is only the starting point for this book. Since it’s easier to under­
stand what I mean by the flavor of a language if different languages are explicitly
contrasted with one another, the book as a whole is designed as a tour of a
half-dozen ancient languages of Europe and Asia, which offer among them
enough variety to illustrate why the polyglot might think of languages as being
like different cuisines, and why one would no more want to limit oneself to
reading only Greek than to eating only at Italian restaurants. There is pleasure
to be had not only in the connoisseurship of food and drink but also in that of
language. In his recent book Babel No More: The Search for the World’s Most
Extraordinary Language Learners, Michael Erard does a good job of describing
many of the characteristics of polyglots, but, by his own admission, he is not
one himself, and because he’s looking at his subject from the outside, he occa­
sionally offers a view of the solitary language enthusiast as a somewhat joyless
creature that is, to my mind, more dismal than is called for. I hope that this
book will serve as a corrective to this portrayal by enabling readers to get a

How Dead Languages Work. Coulter H. George, Oxford University Press (2020).
© Coulter H. George. DOI: 10.1093/oso/9780198852827.001.0001
2 How dead languages work

better sense of what polyglots see in all these different languages they devote so
much time to. It may seem far-fetched to the layperson to view the mastering
of pages of vocabulary lists and verb endings as fun—indeed, the sheer
­mem­or­iza­tion necessary in order to read Greek or Latin is presumably the chief
draw for very few people indeed—but this is to grossly discount the rewards
that lie in wait once you’ve gone to that trouble and can finally read Homer or
Virgil in the original language.
Languages are amazingly rich systems, and each chapter of the book takes a
language or couple of related languages and progresses more or less method­ic­
al­ly through the aspects of a language’s grammar that account for its dis­tinct­
ive­ness. Most chapters begin with a look at the sounds of the language. We all
associate particular sounds with particular languages, working from our per­
ceptions of how native speakers of that language handle English or latching
onto sounds that are completely unfamiliar to us: the nasal vowels of French or
the guttural fricatives of German. But where do these different sounds come
from? And what do they look like when represented on the page? If you have
ever wondered why written Welsh looks as though it has no vowels, or what
gives rise to the sorts of spellings that characterize words borrowed from Greek
(ichthyology and rhythm don’t look very Anglo-Saxon), then these sections will
help explain in historical terms how these traits arose. Once the sounds of the
language have been introduced, one is in a position to see how words are built
up out of those sounds and, as is important for the languages covered in this
book, how the forms of those words change in order to indicate what their
grammatical role in a sentence is. English may not possess many endings beyond
-s, ‑ing, and -ed, but most of its relatives in the Indo-European language family
have at their disposal numerous such changes in form (so-called inflections)
that can be used to indicate whether a noun is the subject or object of a sen­
tence, or whether the subject of a verb is you, me, or a third party. All of these
features can help impart a distinctive quality to a given language, especially
when a profusion of endings indicating grammatical functions frees up word
order for purposes other than simply indicating which nouns are subjects and
objects, or which adjectives modify which nouns.
But one of the more obvious ways in which a language can acquire a particu­
lar personality is through the words it happens to maintain in its lexicon.
Indeed, one of the things that’s always entertaining about picking up a new
language is seeing what vocabulary items textbook writers deem appropriate
for beginners. Often, modern languages come across as being relatively alike:
one learns how to greet new people, order food, tell time, and understand a
weather forecast. Still, most textbooks are also garnished with sidebars offering
various cultural nuggets: learners of Dutch will be told of gezelligheid as an
ideal important to speakers of the language (generally said to be untranslatable,
but “cozy conviviality” is close, if clunky), and the student of French will be
given additional vocabulary to contend with the rich variety of wine and cheese
 Introduction3

in the land of Brie and Bordeaux. But those who write the primers of ancient
languages have even more scope and reason to tailor their vocabulary to the
cultures that spoke them. After all, you don’t learn Ancient Greek so you can
order a Coke in Sparta, and the typical student of Sanskrit doesn’t care about
understanding announcements at train stations. People learn these languages
so that they can read them, and those teaching them, recognizing this, design
their vocabulary lists to match the texts that inspire students to pick up these
languages in the first place. To read Caesar’s military narrative of the conquest
of Gaul, beginning Latin students typically learn a rich variety of words
­referring to weapons and death. Greek textbooks often take a philosophical
turn and opt for more vocabulary in the spheres of being and understanding—
reasonably enough, since both ontology and epistemology as fields take their
names from the Greek words in question. That some words are language-
specific is sometimes seen in translators’ occasional reluctance to translate
them; for instance, Greek polis hovers between “city” and “state” and is often
simply transliterated into English. At other times it’s the juxtaposition of
words in two different languages that highlights a point of cultural difference:
someone learning Greek is taught that the word barbaros is the adjective that
signifies anyone who doesn’t speak Greek—but someone learning Sanskrit
from Perry’s Sanskrit Primer finds the word yavana- glossed as meaning both
“Greek” and “barbarian”, since Sanskrit speakers will naturally have had a
different idea about what constitutes barbarianness than a Greek.
That said, a disconnected sequence of lexical factoids isn’t in the end as good
a way of getting across a feel for a language as a sentence or two of actual litera­
ture, since, really, a language can only be understood as the sum of all of these
elements. So each of the chapters contains a few sample passages drawn from
some of the most famous texts in each language. I have chosen them not just as
examples of good literature (although they are that, too) but as illustrations of
what one misses when one reads Homer or the Rig Veda in translation rather
than in the original. In doing so, I have a couple of aims. One, of course, is to
encourage people to consider learning these or other languages: they open up
new worlds that can only be made out fuzzily through English-tinted glasses.
But in showing how inadequate translations are in capturing the nuances of
Horace’s poetry or God’s first words in the book of Genesis, I also aim for this
book to take part in the debate over the Sapir–Whorf hypothesis, which con­
tinues to loom large in discussions of linguistics in the popular press. Stated
robustly, Sapir–Whorf (named after two linguists important in its develop­
ment) suggests that our thoughts are necessarily shaped by the particular lan­
guages we happen to speak. It’s a little surprising that the hypothesis is still as
prominent as it is since pretty much universally the linguistics community has
agreed that, in this strong form, it’s patently false. In particular, the linguists
who contribute to the wonderful Language Log blog do an excellent job of
­collecting all the times that fallacious ideas related to Sapir–Whorf surface in
4 How dead languages work

the wider media, with particular scorn for the common trope that the Eskimos’
50 words for snow supposedly show a cognitive sophistication in this area
that can’t be matched by speakers of other languages. The standard argument
against it is, first, that those different words for snow have parallels in other
languages too: to reach the magic number of 50, one also has to include words
for various types of wintry mix, slush, and sleet, and any speaker of any lan­
guage who goes skiing regularly will have acquired plenty of fine gradations in
their vocabulary for snow. More important, though, is the fact that even those
types of snow that don’t have good single-word English equivalents can still be
expressed by phrases in English, and so we’re not really at any ter­mino­logic­al
disadvantage when it comes to discussions of wintry weather. All this is true,
so it’s quite right that linguists like John McWhorter have gone out of their way
to try to put the final nail in Sapir–Whorf ’s coffin.1 Yet frustratingly, even
though the strong version of the hypothesis is indubitably wrong, it ­keeps
­coming back—most recently in the movie Arrival (2016), when (spoiler alert!)
it’s suggested that learning a new language can give you an ability to see into
the future.
So why do people continue to be so attracted to it? In a review of McWhorter’s
book in Pacific Standard, Graeme Wood noted that Sapir–Whorf can, to some
extent, absolve us of our cognitive shortcomings. If speakers of one language
don’t attain the same excellence in some field as those of a different language,
they can say they were bound to fail because their language didn’t predispose
them to think in the right way for the particular intellectual task at hand: maybe
there’s something about German as a language that just makes it easier for a
Kant or a Hegel to formulate philosophical ideas. But this argument doesn’t hold
water. Translations may not be perfect, but, when combined with explanatory
notes and commentaries, philosophers outside the German-speaking world seem
perfectly capable of understanding what their Teutonic colleagues are getting
at. And when an idea—say, Zeitgeist—doesn’t already have an exact equivalent
in another language, it’s easy enough to borrow the word in order to get the idea
across. Besides, Kant’s prose hardly represents ordinary German: it’s so difficult
that even those whose native language is German often reach for English
translations, I’m told, in an attempt to understand what he’s trying to say.
Another downside to suggesting that German might be an especially appro­
priate language for philosophy—or, more generally, that some languages are
better suited for particular ends—is that other languages would then by def­in­
ition be inferior tools for those communicative tasks. And, these days, it’s not
very popular among linguists to make the value judgment of saying that a
particular language is somehow “primitive”. This is not just political correct­
ness at work: if anything, the languages that the layperson might expect to be

1  J.  H.  McWhorter, The Language Hoax: Why the World Looks the Same in Any Language
(Oxford, 2012).
 Introduction5

primitive tend to have more, not fewer, grammatical categories than the
­mainstream languages of the West, which suggests a complexity at odds with a
simplistic notion of primitiveness. The same is also often true if we track a
particular language over time. As we’ll see in Chapter 4, Modern English, com­
pared to Old English, has lost a considerable number of noun and verb endings,
showing nicely the complete uselessness of “primitive” as a parameter in
describing languages: Would we assign that label to Old English, even though
it’s far richer in inflections than Modern English? Or to Modern English, even
though the vocabulary of the present-day language is vastly greater? As sys­
tems, languages are simply too complex for such a reductionist label. What’s
more, the prevailing direction of research in linguistics over the past 60 years,
thanks to Noam Chomsky, has been the attempt to work out what is universal
to the grammar of all languages. Given that ideological backdrop, it’s no
­wonder that the idea that different languages might somehow have different
strengths has not met with a warm reception among general linguists.
In light of all this, why has it proven so hard to kill Sapir–Whorf? Perhaps
because, at least in a much weaker formulation, it’s not entirely wrong. There
may be some basic underlying grammar that’s constant from language to lan­
guage, and new words can always be coined as necessary to express new con­
cepts, but, at another level, non-linguists’ intuition that translation is imperfect,
that what one language does is not quite the same as what another does, is a
correct one. No matter how much linguists—rightly—shrink from denying
one language a capacity for expression that’s enjoyed by another, anyone who’s
tried to translate anything more complicated than “I’d like a pint of beer, please”
knows that there’s not a one-to-one mapping of one language onto another, and
that that entails that what you say in one language won’t be quite the same as
what you say in another. This is true at a host of different levels. For now, we can
consider two examples, drawn from either end of the spectrum. First, there’s
the more narrowly linguistic matter of grammatical gender. In English, one
generally uses he to refer to animate males, she to animate females, and it to all
inanimate objects (although the increasing desire not to make a binary division
between two animate categories has furthered the rise of singular they). But in
other European languages, the correspondence between grammatical gender,
as marked in pronouns, and the real-world sex of the noun is much looser: a
two-gender language like Spanish assigns all inanimate nouns either masculine
or feminine gender, and a three-gender language like German apportions them
out among masculine, feminine, and neuter—and can even use the neuter for
animate nouns. Neo-Whorfians have noted that speakers of such languages are
quicker to assign gender-stereotypical adjectives to inanimate objects based on
the grammatical gender of the noun in question. The word for “bridge” in
German is feminine (die Brücke), but masculine in Spanish (el puente), and,
sure enough, when asked to pick adjectives to describe bridges, German
­speakers go for words like “elegant” and “slender”, while Spanish speakers opt
6 How dead languages work

for “strong” and “sturdy”.2 Vigorous anti-Whorfians like McWhorter generally

acknowledge that effects like this are real, but downplay their importance.3
That is fair enough: since it’s not as if speakers of these languages are unable to
use the dispreferred adjectives if they so desire, both languages can, at a basic
level, express the same thing. But, once one moves past the rudimentary com­
munication of propositions to higher-order linguistic spheres like rhetoric and
poetry, then such resonances become harder to ignore.
Second, and more broadly, there is the cultural baggage associated with
­different languages. German and Italian are both immensely expressive lan­
guages, both with celebrated literary traditions that have also enriched the
musical world. No right-thinking person would want to claim that one is inher­
ently better than the other: Does Dante surpass Goethe? Does Wagner outrank
Verdi? The Divine Comedy and Faust, Tristan and Traviata—these all belong to
the great and lasting cultural heritage of the West, and there’s room for all at the
table. But can one have a preference for one or the other? Of course, and while
it’s often the case that those preferences are based on non-linguistic criteria—
perhaps you just don’t fancy reading a Christian vision of the afterlife, or maybe
you prefer your operas to be two hours long, not four—those who respond
aesthetically to these works are often basing their judgments in part on the
language itself. Many non-Germans think of German as a harsh language, with
all of its consonant clusters and guttural ch’s. Especially in the case of opera, it
took time for a full-fledged German-language tradition to develop: the first
major German-speaking composers to turn to the genre, like Handel and
Mozart, worked primarily with Italian libretti. This makes a certain amount of
sense. Not only did the operatic tradition originate in Italy but Italian also
works well as a singing language, with few diphthongs or word-final consonant
clusters. And after a time it becomes hard to separate cultural inertia from
the  inherent phonological properties of Italian in explaining its success in
this field.
In short, both these examples—subconscious biases in the viewing of inani­
mate objects according to their grammatical gender, and the association of
opera with Italian—illustrate well the extent to which languages aren’t in fact

2  L.  Boroditsky, L.  Schmidt, and W.  Phillips, “Sex, Syntax, and Semantics”, in Language in
Mind: Advances in the Study of Language and Cognition, edited by D. Gentner and S. Goldin-
Meadow (Cambridge, MA, 2003), p. 70.
3  A particularly clear indication of this is McWhorter’s reaction (The Language Hoax, pp. 7–12)
to an experiment by Jonathan Winawer and others (including Lera Boroditsky) testing the rate
at which Russian and English speakers could match different shades of blue. Because Russian
has different basic color terms for dark blue (siniy) versus light blue (goluboy), whereas English
has only the one principal word, Russian speakers were 124 milliseconds slower at identifying a
mismatched blue shade within a single half of the blue spectrum than one from the other half of
the spectrum; no such lag was found with English speakers. McWhorter is scornful of the signifi­
cance of the effect—or, perhaps better put, scornful of the exaggerated importance placed on the
finding, especially in the popular media. It’s true that a tenth of a second is not a lot of time, but to
me at least, the fact that there should have been any difference at all is fascinating.
 Introduction7

entirely fungible. No, there’s no constraint on Spanish speakers using stereo­

typically masculine or feminine adjectives with nouns of the opposite gender,
and one is free to write opera in whatever language one chooses. But, all the
same, the meaning of a work, or even just a sentence, will be subtly affected
by the different resonances at play in one language compared to another—
re­son­ances often determined by the broader cultural context, but sometimes,
as will be more prominent in the current book, even by the underlying struc­
tural features of the language. Indeed, both of these types of difference—the
cultural and the more narrowly linguistic—are particularly marked when
one compares English with ancient languages, making the latter an especially
fruitful source of examples where some nuance or other is lost in translation.
That Sanskrit or Ancient Greek would be culturally more different from English
than is Spanish or French probably requires little explanation. But the linguistic
differences are also important, and rather less obvious to those who don’t know
the relevant languages.
I’ve already mentioned that Old English differs from Modern English in that
the former has many more endings for nouns and verbs than the latter does.
Something similar is the case with most of the other languages of Europe as
well: Latin has more forms than Spanish, French, or Italian, Ancient Greek
than Modern Greek, Old Irish than Modern Irish, and so on.4 One of the con­
sequences of each word’s having more forms is that a language can do more
with a single word. What English expresses, for instance, in the three words she
will see, Latin can combine into just one, vidēbit, with vidē- equivalent to “see”,
-bi- to “will”, and -t to “she”. As a result of all these endings, we’ll see that there’s
often a conciseness to the ancient languages covered in this book that their
modern descendants find it hard (though not impossible) to match. A further
consequence of the wealth of endings is that word order is more flexible in most
of these languages than in their descendants: since endings can indicate which
adjectives go with which nouns, or which nouns are the subjects or objects of
verbs, individual words have more freedom to move to different parts of the
sentence without obscuring the syntax. Especially in poetry, this allows for
effects that cannot be replicated directly in a language with more rigid word
order, like English.
But while one can get a rough sense of how these sorts of linguistic differ­
ences work in an abstract summary such as I’ve just given, it’s far more effective
to look at the details of specific examples. And that is exactly what this book
aims to do. It’s all very well to say that Latin poetry can do different things from
English poetry because of the more flexible word order, but if you don’t know

4  It might be tempting to see this as a simple linear progression: languages begin with lots of
endings, but speakers are lazy and fail to articulate them properly, and they’re lost, end of story.
In fact, though, new endings arise out of the phonological reduction of helping verbs, postpositions,
and the like, so over a long enough timescale the process can be cyclical.
8 How dead languages work

Latin, then it’s hard to get much of a feeling for what that means—unless someone
takes you through some actual lines of Horace. Accordingly, the book works in
turn through six ancient languages or language groups that are especially import­
ant to the West, and which illustrate the wide range of per­son­al­ities that different
languages can have. Throughout, the fundamental aim is to show readers just
how much they miss when they read the great works of ancient literature in
translation—no matter how good the translation—rather than in the original.
We start with Ancient Greek (Chapter 2), because that’s where European litera­
ture begins—more specifically with the two epic poems attributed to Homer, the
Iliad and the Odyssey, thought by most scholars to have been ­composed in the
eighth century bc. And “composed”, rather than “written”, is the operative word
here: features of the Greek strongly suggest that they are not the product of a liter­
ate poet but rather of a bard working in an oral tradition. In particular, the sort of
repetition of lines and half-lines that pervades these works matches the compos­
itional style of oral poems in modern times, and this is best seen by looking at
some of the original Greek. This chapter also serves as an introduction to many
of the basic grammatical terms that will be used over the course of the book.
While this is not a book for specialists, much of the discussion can be conducted
more elegantly if certain technical terms are allowed to come into play: it’s easier
to understand the sounds of languages, and how they change, if one has a basic
understanding of how and where they’re produced in the mouth, and an account
of the various endings that nouns take in most of these languages is easier if one
can call those endings by their proper names. In addition to Homer, the Greek
chapter also considers small excerpts of two later texts: Thucydides’ history of
the Peloponnesian War (written in the Classical period of Greek, in the late
fifth century bc) and Paul’s Letter to the Romans (representative of the Koine
Greek of the first century ad). An im­port­ant lesson to be drawn from both of
these is the extent to which translations5 often flatten out difficulties in the
Greek: abstract nouns used in uncommon ways are often paraphrased with
more natural equivalents in English, and prepositions that are doing more than
an “in” or a “to” typically would can be expanded into entire verbal phrases.
Compared to English, all of the languages covered in this book are highly
inflected: the shapes of words change to indicate their grammatical role in a
sentence. The chapter on Latin (Chapter 3) brings out two important ramifica­
tions of this, both already noted. First, such languages are often more concise
than English, at least when conciseness is judged by the crude metric of word
count, since prepositions and helping verbs can be replaced with various inflec­
tions instead. Second, the flexible word order enabled by such endings is par­
ticularly neatly illustrated with Latin, since so many phrases that can be used to
explain these characteristics have been borrowed into English. That is, rather

5  Scripture quotations have been taken from The Holy Bible: King James Version and from New
Revised Standard Version Bible (copyright © 1989 National Council of the Churches of Christ in
the United States of America. Used by permission. All rights reserved worldwide), and noted
accordingly.
 Introduction9

than just being a study of linguistic compression in the abstract, one can see
directly how it is that a phrase like mūtātīs mūtandīs, when translated, has to
expand to “what needs to be changed having been changed”. Again, short
excerpts from three Roman authors show how the differences between Latin
and English play out at a more literary level: Lucretius’ Dē rērum nātūrā (“On
the Nature of Things”), a poetic account of the Epicurean world-view, written
in the mid first century bc, exemplifies not only the general advantages to a
Roman poet of writing in a language with freer word order but also the more
particular difficulties of conveying wordplay in translation, along with the
more pervasive problem that the resonance in Latin of a word like immortālis
is different from that in English of its most obvious equivalent, immortal.
A comparison of Horace’s Ode 4.7, a famous meditation on mortality published
about a generation after Lucretius’ poem, with the skillful translation by the
classicist poet A. E. Housman shows the extent to which Latin and English lend
themselves to very different poetic techniques. And that prose, too, can be
much denser in Latin comes across clearly in some passages from the historian
Tacitus, writing in the early second century ad.
Chapter 4 turns to the Germanic languages, with a concentration on Old
English, the form of English between the time when it was brought to Britain
by the Anglo-Saxons (fifth century ad) to the Norman Conquest in 1066.
Modern English is a very different language from Greek and Latin, but it arose
from the same source (all three belong to the Indo-European family of
­languages), so the fundamental questions considered here have to do with
­language change. What sorts of linguistic developments took place when the
Germanic languages split off as their own branch of the Indo-European family
tree? To what extent did Old English remain similar to the classical languages?
And what further changes caused Modern English to evolve so far from the
language of the Anglo-Saxon period that, to the untrained eye, the two scarcely
look like the same language? How do languages evolve from being highly
inflected like Greek and Latin to being more like Modern English, with its
pared-down store of forms and more rigid word order? Since this isn’t the place
for exhaustive consideration of these questions, several key developments
stand in for the general pathway of language change: the consonant changes
that are so characteristic of Germanic, the origins of plural formations like
mouse ~ mice, and the reasons that English has two main classes of verbs, those
which form their past tense and past participle with a suffix (walk ~ walked),
and those that form it by changing the vowel of the verb (sing ~ sang ~ sung).
The second half of the chapter then shows how these developments are reflected
in actual texts, with commentaries on short passages from the Old English
Bible and Beowulf offering concrete examples of how Old English is both an
Indo-European language in its wider affiliations and more narrowly a Germanic
one, as well as of how these features play out in the modern language.
Because of the close cultural connection between the ancient Greco-Roman
world and the modern European societies that developed out of the Renaissance,
10 How dead languages work

it is perhaps not so strange to think that nearly all the languages of Europe
would belong to a single family. What’s more counter-intuitive is that Sanskrit,
the great sacred language of India, should be part of this linguistic group as
well—hence its designation as the Indo-European family. Chapter 5 sets out
how this is in fact the case, showing how some of the superficially obvious
­differences between Sanskrit and English in fact obscure some underlying
­likenesses. Not only are the basic vocabulary items of Sanskrit, such as nu­merals
and kinship terms, clearly related to their European counterparts but even
grammatical patterns are often the same, showing that the similarities are due
not to the simple borrowing of individual words but to all of these languages’
having arisen from an earlier common language, dubbed Proto-Indo-European
(PIE), which no longer survives but which has been reconstructed on the basis
of correspondences between the numerous daughter languages. Excerpts from
the hymns of the Rig Veda, the earliest Sanskrit text to survive (probably dating
from some time in the late second millennium bc), are analyzed so that readers
can see that, after one has accounted for some of the sound changes that make
the Indic languages distinctive, there are actually quite a few words and gram­
matical forms found in Sanskrit that have relatives in Europe. Moreover, while
it is the correspondences between individual words that are the most ubiqui­
tous, there are also even broader patterns of poetic expression shared by the
Rig Veda and by the early poetic traditions of Europe. Accordingly, drawing
especially on the work of the Indo-Europeanist Calvert Watkins, the chapter
concludes by showing some of these common turns of phrase in action.
In Chapter 6, the scene moves back to the far west of the Indo-European
world, to the Celtic languages. Irish and Welsh, like Sanskrit, look very foreign
to the Anglophone, even though they’re neighbors, so it’s especially instructive
to see how the linguistic developments that led to the separation of the Celtic
branch have given them such a distinctive appearance: all those Irish words
with bh and fh in them, and those Welsh ll’s, w’s, and y’s. But while more
ex­amples are taken from the modern representatives of the branch than in the
other chapters,6 pride of place belongs to Old Irish, arguably the most difficult
member of the entire Indo-European family. Ultimately that assessment must
rest on subjective grounds, but it’s worth considering some of the more ob­ject­
ive features of the language that make it such a good candidate for that title, in
part to show just how baroque a language can become.
After a couple of chapters in which even some very strange-looking languages
are seen to be related to English, Chapter 7 moves outside the Indo-European
family altogether to consider Biblical Hebrew. It occupies a somewhat para­dox­
ical position. On the one hand, as a Semitic language, it isn’t related to English
at all, so it mostly serves as a linguistic foil: by looking at a non-Indo-European

6  Even today, the Celtic languages behave unusually enough by the standards of most of their
European kin that they pose the same difficulties more often associated with ancient languages.
 Introduction11

language, one can better see, thanks to the contrast, what makes Indo-European
Indo-European. On the other, the language of the Hebrew Bible is at times
oddly familiar, because it has exercised a considerable influence on English
through its cultural importance. Particular turns of phrase that are idiomatic in
Hebrew, but not in Greek or Latin, entered into the language of Jewish writers
who, starting in the third century bc, first translated the Hebrew Bible into
Greek (the Septuagint); then, Christian writers began to write texts of their
own, notably the New Testament (first century ad), in Greek that had been
stylistically influenced by the Septuagint; after that, they then translated all of
the above into Latin, most prominently Jerome’s Vulgate (late fourth to early
fifth century ad). And so it came to pass that, eventually, “And it came to pass”
became a phrase associated with Biblical language even in English. Languages
don’t exist in a cultural vacuum: they may each have their own flavor, but that
doesn’t keep them from taking on words and expressions from other languages
that cross paths with them.
 2

Greek

From the epic poetry of Homer to the three great tragedians Aeschylus,
Sophocles, and Euripides, to the historians Herodotus and Thucydides, to
­philosophers like Plato and Aristotle, and, a few centuries later, to the sacred
Christian texts of the New Testament, Greek was the language in which many
of the most influential works in the Western tradition were written. The redis-
covery of Greek literature in the West after the fall of Constantinople in 1453
gave significant impetus to the Renaissance, and, along with Latin, Greek
formed the backbone of classical education in western Europe for centuries.
There’s no better place for this book to start. In this chapter, as in later ones,
we’ll first examine the language as language—starting with the smallest element,
the sounds that are the elemental building-blocks of the language, ­followed by
consideration of the ways in which words change their shape to express differ-
ent grammatical categories—then turn to sample texts (in this case, the epic
poetry of the Iliad, Thucydides’ history, and the letters of Paul) that illustrate
some of the key characteristics of the language in action.

THE SOUNDS OF GREEK

Much of the core vocabulary of English fits a standard pattern: consonant,

vowel, consonant, and then perhaps a silent e at the end. That gives us man,
mane, name, rat, rate, and so forth—the bread and butter of early phonics les-
sons. But some words just look different. It would clearly be cruel to put rhythm
or ichthyology on a kindergartner’s spelling test. Now the reasons why English
has accumulated so many words that are tricky to spell are too numerous and
complicated to explore here, but the two words I’ve just mentioned both owe
their appearance to their status as words that have been borrowed from Greek.
As in any language, Greek words are built from an inventory of sounds that are
particular to it as a language. If one sees how that set of sounds works as a sys-
tem, then it becomes easier to understand why such borrowed words look the
way they do.

How Dead Languages Work. Coulter H. George, Oxford University Press (2020).
© Coulter H. George. DOI: 10.1093/oso/9780198852827.001.0001
14 How dead languages work

So what is so Greek about rhythm and ichthyology? Let’s start with the v­ owels,
in particular the y that appears in so many Greek words. It seems a strange
spelling. Anglophones would hardly pronounce the words differently if they
were spelled with an i instead, and many of the Romance languages do just that:
in Spanish, songs have a ritmo and a fish-doctor has studied ictiología. Even
stranger, the i and y in words borrowed from Greek are also pronounced the
same in Modern Greek. Compare the words Phil-adelphia and chloro-phyll. The
first half of the former comes from a root that means “dear, loved”; the second
half of the latter from one that means “leaf ”; we pronounce both the same. Both
roots exist in Modern Greek, where a friend is a philos (ϕίλος) and a leaf is a
phyllo1 (ϕύλλο)—and the first syllables of both are pronounced fee, even though
they’re spelled with different vowels, an iota (ι) and an upsilon (υ) respectively.2
What’s going on?
As is often the case with spelling, it’s a holdover from a time when these were
two different vowels: the i pronounced as in Modern Greek, but the y like a
French u or a German ü. If you’re not familiar with these sounds, do the follow-
ing: slide from ee (as in fee) to oo (as in boo). You should notice two changes:
your tongue retracts in your mouth (at least in most dialects), and your lips
round. The French u, German ü, and Ancient Greek y are in between: to pro-
nounce it, keep the tongue forward, as for ee, but round the lips, as for oo. This
sound, though common enough in the world’s languages—Old English had it
too—has often proven unstable. The lips become unrounded, and it merges
with ee. But since the differentiation in spelling persists, we’re left, in the case of
borrowings from Greek, with a whole host of words with y: from hydrogen to
xylophone, from dynamic to crypt.3
Another feature that distinguishes many Greek borrowings in English is
the frequency of the letter h, in particular in the combinations th, ch (when
pronounced like a k), ph, and rh. Now these are of course not surefire proof of
Greek origin: the th of thick and thin is perfectly Germanic,4 and the usual
pronunciation of ch seen in chin and cheap developed within English from an
inherited Germanic k (still seen in German Kinn “chin” and kaufen “to buy”).
But nearly all the words that begin with rh are Greek in origin: rhythm, rhap-

1  To show the connection between Greek words and their English derivatives—and as a
reminder that the Classical Greek upsilon is not just a u—I have transliterated it as y rather than
the usual u (except when it is the second element of a diphthong). Dyo “two” and lȳō “loosen” may
look odd by the usual transcriptional conventions, but they at least have the virtue of more closely
resembling dy-ad and ana-ly-sis.
2  The metaphor of phyllo dough is the same as that of mille-feuille pastry, French for
“thousand-leaf ”.
3  Three of these words show another pronunciation of y, as an English long i: this difference is
due to the effects of the so-called Great Vowel Shift, which took place at the end of the Middle
English period; we will meet it again in Chapter 4.
4  The origin of the Germanic th sound will be explained in Chapter  4 in the section on
Grimm’s Law.
 Greek15

sode, rhetoric, and rhinoplasty; so too those with rrh in the middle, like diarrhea
and cirrhosis. Probably the most prominent exception is rhyme, which was
once spelled rime like the good Germanic word it is, as, for instance, in
Coleridge’s Rime of the Ancient Mariner. In time, though, its spelling was artifi-
cially made to look more Greek with the -hy- sequence introduced through
assimilation to its frequent companion, rhythm. The “hard” ch is also a good
hallmark of Greek origin, as seen in chaos, chorus, chronology, and chrysanthe-
mum; so too the ph of phobia, physics, and photography. As with y, though, the
spellings seem unmotivated by the pronunciations. Why spell with ph, ch, and
rh sounds that could be adequately represented with f, k (or c), and r? The
Spanish, after all, manage just fine with cronología, retórica, and física. Also,
why use an h, anyway? It makes sense in a word like uphill, where there really is
a p followed by an h, but not in photo, where there’s only a single f sound at the
beginning.
Again, the explanation lies in the change of pronunciation over time. Leaving
aside rh, which is a different matter,5 take the digraphs (two-letter combin­
ations) ph, ch, and th. When used for words of Greek origin, these represent the
single letters phi (ϕ), chi (χ), and theta (θ). In Modern Greek, these are pro-
nounced respectively f, ch-as-in-Bach, and th-as-in-theta—what linguists
call fricatives, in which the airflow through the mouth isn’t completely blocked
off but merely restricted. Not so, however, in Ancient Greek, when they were
aspirated stops—that is, a p, k, or t (all examples of stops, that is, sounds where
the airflow is completely blocked) each followed by an h sound (that is, aspiration).
In other words, the th in a word like Ancient Greek anthos (“flower”) would not
have been pronounced as it is in our word anthology but as in ant-hill. In late
antiquity, the earlier pronunciation of these sounds as p+h, k+h, t+h would
eventually weaken to the fricative pronunciation they have today—but not
before Latin, which itself borrowed a lot of words from Greek, had standard-
ized the practice of spelling them ph, ch, and th like the aspirated stops they
were. But even when the pronunciation changed, the spelling with the h
remained.
Before leaving h behind, there’s one more place it occurs in Greek that
deserves mention: at the start of words. In this context, it serves as a convenient
illustration of how regularly the sounds of language change over time. To see
how this works, we need some basic vocabulary. The numbers one through ten,
as it happens, will do nicely, since they are usually among the most conservative
elements in a language’s lexicon, or store of words: they’ll undergo the same

5  The phonetics of Greek rh are somewhat more complicated: suffice it to say that Greek words
beginning with an r sound were regularly notated with the same sign used to indicate an initial h,
although this is usually taken to mean not that it was an aspirated r but rather that it was voiceless.
For further information on how Ancient Greek was pronounced (and how we know how it was
pronounced), the standard reference is W.  Sidney Allen, Vox Graeca: The Pronunciation of
Classical Greek, 3rd ed. (Cambridge, 1987).
16 How dead languages work

phonetic changes that any word does, but they are seldom replaced by com-
pletely different words altogether. Lining up the Latin and Greek words for the
numerals, many similarities are obvious, especially in the words for “two”,
“three”, “eight”, and “ten” (we’ll get to the column on the right shortly):

  Latin Greek Proto-Indo-European6

one ūnus heis *oinos, *sems
two duo dyo *duō
three trēs treis *treyes
four quattuor tettares *kwetwores
five quı̄nque pente *penkwe
six sex hex *seks
seven septem hepta *septm
eight octō oktō *oktō
nine novem ennea *hnewn
ten decem deka *dekm

With some of the other words, the match is not quite as close. Only in the case
of “one”, however, is this because the two words lack a common origin.
Elsewhere, the numbers began life from a single ancestral form, which, among
the ancestors of the Greeks, underwent one set of sound changes, but, among
the ancestors of the Romans, underwent another—and, among the ancestors of
English speakers, yet another. Over the past couple of centuries, historical lin-
guists have used the comparative evidence of dozens of related languages
throughout Europe and South Asia to work out what the ancestral language
(dubbed Proto-Indo-European (PIE)) would have sounded like, and the
results are given in the right column; the asterisk before the words warns that
these forms are only reconstructions, not attested in any written texts.
Now pretend for a moment that we don’t have the reconstructions, and just
consider the words for “six” and “seven”. There are certainly enough similarities
to suggest a common origin: (i) both numerals begin with an s in both English
and Latin; (ii) in the case of “six”, all three numbers end with an x; (iii) in the
“seven” word, both Latin and Greek have a -pt- cluster in the middle. Why,
then, does Greek have h as the initial consonant? The most economical ex­plan­
ation is to assume that, in the parent language, the words began with s, but that
this s somehow changed to an h in Greek. There are two main ways to support
this hypothesis: one is to find other examples of s changing to h in Greek, so
that what’s going on in the numerals doesn’t just seem to be a one-off arbitrary
event; the other is to find examples of s changing to h in other ­languages,

6  Those with particular interest in Indo-European linguistics should be aware that, for ease of
presentation, I offer simplified reconstructions for now, most egregiously in omitting *-u̯- from
the onset of *su̯eks.́ More refined notation will be introduced for several of these sounds starting
in Chapter 4.
 Greek17

where we do have both the earlier and the later form attested, to show that this
is a plausible direction to posit for the sound change (as opposed to, say, an
original h turning into an s). We can in fact do both.
First, here are some other words which start with an s in Latin and English,
but with an h in Greek; I’ve used hyphens to isolate the roots of the words:

  Latin Greek  
sal-t sāl hal-s (> halogen, i.e. element like chlorine that forms salts)
sun sōl hēl-ios (> heliotropic)
seat sēd-ēs hed-rā (> cat-hedral, i.e. the bishop’s seat)
sweet suā-vis hēd-ys (> hedonism)

Since it would be quite a coincidence if all of these words simply happened at

random to start with an s in English and Latin, and an h in Greek, it’s far better
to assume that some sort of principled process is causing this particular set of
correspondences. The case becomes even stronger if we note other such cor-
respondences, such as the initial d of the Greek and Latin words for “two” and
“ten” mapping onto the t of their English counterparts—an equation also found
in the final consonant of the seat and, Latin aside, sweet roots.
As for why the change from s to h is more natural than the other way around,
one could in the first instance appeal to what seems to fit the natural articula-
tion of sounds in the mouth better. When we’re mumbling, we notice that cer-
tain sounds lose their crispness, and this generally takes the form of the tongue
not moving quite as far to one end or another of the mouth as it ought to: an s
requires actually getting the tongue to the alveolar ridge behind the teeth,
whereas an h is just a puff of aspiration. But a far better way to justify assuming
that s became h, rather than vice versa, is to find other examples of this change
in the histories of the world’s languages. And, sure enough, there are parallels.
In Sanskrit, for instance, what was once a final s (as suggested by comparative
evidence) turns into an h if the word occurs at the end of a phrase or before a
certain set of consonants. Nor is the evidence for such a change limited to his-
torical times: in many dialects of Spanish, a similar weakening of s is taking
place, with estar coming to be pronounced more like ehtar. Something similar
must also have taken place in French, where words that in Latin began with the
sequences sc-, ­st-, and sp- first acquired an e- before the s to ease the pro­nun­ci­
ation of the cluster, then lost the s altogether; Spanish underwent the first
change, but is only now getting started on the second, giving us respectively:

  Late Latin Spanish French

school schola escuela école
write scrı̄bere escribir écrire
establish stabilı̄re establecer établir
spouse spōnsa esposa épouse
18 How dead languages work

So, since there are good examples of the sound s turning into an h—and since
it’s a lot harder to find examples of the spontaneous creation of an s out of an
h—historical linguists are content with the hypothesis that the parent lan-
guage had an s, which, in Greek, systematically became an h at the start of a
word. After repeating this comparative process dozens of times over with
different sets of sounds (there will be further examples in subsequent chap-
ters), one can eventually build up a remarkably good picture of what the
­parent language must have sounded like, as represented in the asterisked
forms above.

WORD FORMS

Now that we’ve seen some of the most basic building-blocks of Greek—the
sounds that make up its words—we can move up a linguistic level and start
making words out of those sounds. At this point, additional technical vocabu-
lary will become useful, which it’s fitting to introduce in a section on how you
build words in Greek since the technical terms needed are themselves Greek in
origin. In moving from sounds to word forms, we’re progressing from phon­
ology to morphology. In both of these words, the suffix -ology, widespread in
English to denote fields of study, is added to a Greek root: phōn- “sound”, and
morph- “form, shape”, respectively. Phonology, accordingly, is the study of the
sounds of a language, whereas morphology deals with (i) how you can change
the form of a given word (like adding -s to nouns to make them plural) and
(ii) how you can create a new word by building it up from component parts
(like adding -ful to help to form an adjective, then adding un- to helpful to
negate it). The former is called inflectional morphology (since an ending like
-s is said simply to inflect an existing word), and the latter is derivational
­morphology (since it’s used to derive a whole new word from an earlier one).
Units like un- and -s that are the smallest building-blocks of words that can be
analyzed as having meaning are called morphemes, and the fact that I write
un- with a hyphen after it and -s with a hyphen before it indicates that these are
bound morphemes: they can’t occur on their own but only joined to another
morpheme in the spot indicated by the hyphen.
Now monolingual speakers of English tend not to think very much about
morphology because English is not a very morphology-rich language. (That said,
even if English were morphology-rich, monolingual speakers still might
not  think about it very much since most native speakers don’t worry about
what characterizes their own language until forced to confront something dif-
ferent in another language.) What sort of morphology does English have?
Well, to take just inflectional morphology, the average noun in spoken English
just has two forms, one basic form, and one with an s at the end: book and
 Greek19

books. Now, as it happens, that second form plays three different roles, as is
indicated in the writing system: if it’s simply showing that the noun is plural,
then one leaves it alone; if it’s equivalent to of a/the book, then it gets an apos-
trophe before the s (the end of the book ≈ the book’s end); if it’s equivalent to of
(the) books, then it gets an apostrophe after the s (the covers of the books ≈ the
books’ covers). To the readership of this book, this is all probably obvious, but
there’s no reason, other than universal public education, that it should be. The
œuvre of Lynne Truss demonstrates amply that, however easy it is for fluent
English speakers to know whether to pronounce the s, it’s simply a matter of
learned convention to know what to do with the apostrophe. So why make the
distinction at all? Is complaining about apostrophe abuse just a big fuss about
an artificial precept that has no firm basis in what many linguists would
­consider linguistic reality—namely, the spoken language? Well, not exactly.
Because there are words that do have four different forms, even in the spoken
language. All it takes is for a word to have an irregular plural—say, man and
men—and then it becomes clear that a four-way morphological distinction
really is operative: man, man’s, men, and men’s are not just all spelled differ-
ently but also pronounced differently, and native speakers are a lot less likely
to mix them up in speech than they are to misplace an apostrophe in writing.
In short, then, the typical English noun has four different forms—although in
most instances, three of the four are pronounced the same, and so they are
only kept distinct in writing. And those four forms can be arranged in a 2×2
array of singular vs plural, and (so as not to prejudice the discussion that
­follows by adopting ­additional technical terminology too quickly) the default
form and the of-form.
The situation is similar with the verb, which in most Indo-European lan-
guages offers the greatest morphological complexity. In English, if we pick a
regular verb like walk, there are really only four forms to contend with: walk
(which forms the infinitive and most of the simple present tense), walks (the
present form used if the subject is third-person singular), walked (past tense
and past participle), and walking (present participle and gerund, at least as it’s
described in traditional grammars). There’s also a large class of verbs, called
strong verbs, which have a fifth form, because they distinguish between the
past tense (sang) and past participle (sung), although the other forms remain
the same (sing, sings, singing). But to get more distinctions than this, one has
to turn to to be, where the present tense has not a two-way, but a three-way
division between am, are, and is, and the past tense distinguishes between was
and were.
If we look now at Greek, it will become clear why English can be considered
morphologically poor. The typical Ancient Greek noun, for instance, rather
than having only two or four forms as in English, has eleven—although, to be
fair, only nine of these are common. Here, for example, are the different forms
of the word hippos (“horse”):
20 How dead languages work

  singular plural dual

nominative hippos hippoi hippō
genitive hippou hippōn hippoin
dative hippōi hippois = gen.
accusative hippon hippous = nom.
vocative hippe = nom. = nom.

First, a point of transcription. The ō with the line over it, called a macron, is
pronounced longer than the unmarked short o. So the distinction between hip-
pon and hippōn, unlike that between horses and horse’s, is there in the spoken
language as well. These sounds are indicated in Greek by two different letters,
and it’s only the poverty of the Roman alphabet that means that, to distinguish
between them here, one has to resort to a diacritical mark (the general term for
the accents and other squiggles that alter how a letter is read). The names of the
two letters, familiar to many Americans from fraternities and sororities, reveal
openly enough which is which: the short o is o-micron (“little o”), and the long
ō is ō-mega (“big o”).
More pressingly, why are there so many forms? Singular and plural forms
seem reasonable enough—a distinction of grammatical number found in lan-
guages across Europe—but why have a dual? And what are the labels in the
left-hand column? In effect, they represent further variations on the distinction
that English makes between a default form of the word, and one where the
ending ’s or s’ is roughly interchangeable with the preposition of. Called cases,
such endings indicate the grammatical function of a noun in a clause. Nouns in
English retain only the slightest vestiges of this feature, with the ’s ending cor-
responding to what in other languages is called the genitive case.7 At a basic
level, the genitive signifies possession: in the phrase Plato’s book, Plato would be
in the genitive. But it also extends to a whole host of figurative uses, most of
which relate one noun to another, just as of does in English: for instance, the
second noun in Game of Thrones would also be in the genitive case in Greek,
even though the thrones do not possess the game, but rather specify which
game is meant. Thus, what makes Greek nouns so much more complicated (or,
to be more positive, richer) than their English counterparts is that they have
five such cases, all of which signify different roles played by nouns in the struc-
ture of the sentence. All the cases (except the vocative) have many different
uses, so here I’ll just single out the most important role for each: (1) The nom­
inative is, in some sense, the most basic case. It’s the form the noun takes if it’s
the subject of the sentence, and it’s how words are listed in the dictionary. (2)
The accusative is arguably the next more important case, since it marks the

7  Most personal pronouns, however, retain an additional distinction between nominative

forms (I, he, she, we, they) and accusative (me, him, her, us, them), used as subjects and objects
respectively.
 Greek21

objects of most verbs. (3) The genitive, as already noted, is more or less equiva-
lent to putting an of in front of the noun. (4) The dative signals the indirect
object of the verb (prototypically, the recipient of a gift), and can usually be
translated in English by putting to or for in front of the noun. (5) The vocative
is most commonly found with personal names, since it’s the form used to
­indicate the addressee of an utterance.
What advantages does a language gain by having cases in its morphological
toolbox? Paradoxically, having to fix an ending onto nouns to indicate their
function gives speakers greater flexibility in how they arrange words to form a
sentence—that is, in the syntax of the language. This is clearest to see in the
distinction of nominative and accusative, of subject and object. In an ordinary
English sentence, because the nouns themselves look the same, these two roles
have to be distinguished by word order: The man sees the horse vs The horse
sees  the man. But in a language that has case at its disposal, who’s seeing
whom can be differentiated through endings, leaving word order free to express
other nuances.

The man sees the horse → either Ho anthrōpos horāi ton hippon
    the man.nom sees the horse.acc
  or Ton hippon horāi ho anthrōpos
    the horse.acc sees the man.nom
The horse sees the man → either Ho hippos horāi ton anthrōpon
    the horse.nom sees the man.acc
  or Ton anthrōpon horāi ho hippos
    the man.acc sees the horse.nom

Indeed, while I’ve given only two possible versions of the Greek, all six permu-
tations of subject, verb, and object could be grammatically correct, depending
on the context of the sentence. The main thing to note for now is that, not just
for Greek but also for several other Indo-European languages—and this is cer-
tainly a trait inherited from the parent language—many singular nouns will
end in -s if they’re the subject of the verb, in -n or -m (both nasal sounds) if
they’re the object. As for what nuances might be conveyed by playing around
with the word order, that will gradually become clear as we look at specific
examples in languages that inflect for case.
A second area in which word order is freed up by case lies in the relationship
between nouns and adjectives. Like nouns, adjectives also have different end-
ings to mark case and number, as well as an additional parameter, gender.
(Grammatical gender only bears a loose relationship to natural gender. In
the Romance languages, for instance, every noun is assigned to either the mas-
culine or feminine gender, no matter how inanimate and sexless it is.) If an
22 How dead languages work

adjective has to agree with the noun it modifies in case, number, and gender,
then it doesn’t necessarily have to occur right next to it for that connection to
be understood. In English writing, one has to be wary of dangling participles—
verbal adjectives that have wandered far enough away from the noun they
modify that one mistakenly reads them as modifying a different noun, with
nonsensical results. This is much less of a problem in an inflected language like
Greek, where the ending on the participle serves as a link back to its noun: we’ll
see this at work when we look at the opening of the Iliad.
Finally, that dual column: as the name suggests, these are special endings
deployed when you want to talk about a pair of items. This might seem too
narrow a meaning to merit a separate set of endings, and, sure enough, the
Greeks themselves gradually got rid of them. One could even guess that they
were on their way out judging simply from the comparative paucity of forms:
whereas five cases are distinguished in the singular, and four in the plural, the
dual had to make do with two endings, one for the nominative, accusative, and
vocative, the other for the genitive and dative.
So much for the noun. How many forms does the Greek verb have in com-
parison to the five of the English verb? It depends a little on how one counts
because participles have all the possibilities of inflection that nouns and adjec-
tives have, but if we discount those inflections as not being proper to the verb,
then a reasonable answer would be: about 283. It’s difficult to give an exact
number because few verbs are actually found in all of these forms, some of
which are more theoretical than real. But that is the number of distinct forms
of lȳō (“I free, loosen, dissolve”), inflected in full in Smyth’s Greek Grammar as
the paradigmatic example of a regular verb.8 How does the total number rise so
high? Because Greek verbs change their form for an even greater number of
linguistic variables than do Greek nouns:

Person: Leaving aside forms of to be, the only distinction of person in the
English verb is that the third-person singular of the present is
marked by an ­-s: I drink and you drink, but he or she drinks. In
Greek, all three of these would have separate endings, and so the
subject pronoun can be dispensed with: pı̄nō, pı̄neis, pı̄nei
respectively.
Number: As with person, the English verb only distinguishes number in
the third person singular. But all three of the Greek forms cited
in the previous paragraph also have plural counterparts: pı̄nomen

8  For Hellenists keeping score: I only counted one set of aorist optative endings, and a form like
ἔλυον counts only once, as both first singular and third plural imperfect active indicative; but I did
include duals.
 Greek23

“we drink”, pı̄nete “y’all9 drink”, pı̄nousi “they drink”. What’s

more, there are dual forms as well. As with the dual forms of
the noun, they are not as widespread (the first-person dual is
vanishingly rare), and in the present only one form, pı̄neton, is
available to express either “the two of you drink” or “the two of
them drink”. In some tenses, however, like the imperfect, the
second- and third-person dual are distinguished: epı̄neton
“the two of you were drinking”, but epı̄netēn “the two of them
were drinking”.
Tense: While English certainly has plenty of tenses of its own, only the
simple present and past tense are distinguished solely by the
morphology of the main verb: I drink vs I drank. Other tenses,
like the future (I will drink) or present perfect (I have drunk),
require a helping verb. In Greek, all four of these could be
expressed by different forms of the main verb on its own: pı̄nō,
epion, pı̄omai, pepōka. In total, Greek has seven different tenses:
present (I drink, I am drinking), imperfect (I was drinking), future
(I will drink), aorist (which, in simple declarative sentences,
most often serves as a past tense: I drank), perfect (I have drunk),
pluperfect (I had drunk), future perfect (I will have drunk,
although in Greek this tense is rare in the active voice, for
which see below).
Mood: Nowadays mood hardly ever impinges on the inflection of the
English verb. All the examples given in the past few paragraphs
are in the indicative mood, which expresses matter-of-fact
statements. Commands are issued in the imperative mood,
which in English looks like the present indicative (You drink
this.) only without the subject pronoun (Drink this!). English
also retains traces of the subjunctive mood, used to indicate
that the speaker does not commit to the reality of the
statement because it is merely potentially true (I asked that
he close the door) or clearly counterfactual (if I were you).
Greek, however, has separate endings for all three of these
moods—as well as a fourth mood, the optative, which evades
simple description, but can roughly be understood as
expressing even less commitment to the reality of the
statement than the subjunctive.

9  I grew up in Texas, so this, for me, is the correct form of the second-person plural pronoun.
24 How dead languages work

Voice: English has a two-way voice distinction between the active (The
student read the book) and the much-maligned passive (The book
was read by the student), in which the object of the active verb (the
book) is promoted to become the subject of its passive counterpart.
As with tense, voice can only be marked with helping verbs in
English, but Greek can manage this simply with verbal inflections:
the verbs in the two previous sentences, both third-person
singular aorist indicative, would become anegnō and anegnōsthē
respectively. Additionally, Greek has a third voice, the middle,
which, at a basic level, indicates that the subject of the verb is
somehow affected by the action of the sentence (it thus resembles
the reflexive verbs of Spanish, French, and German), but it often
takes on idiomatic senses depending on the verb in question:
egrapse (aorist active) means “he wrote”, but egrapsato (aorist
middle), which will have originally meant “he wrote for his own
benefit”, acquired the specialized sense “he issued an indictment”.
Both of these can be contrasted with the aorist passive egraphē “it
was written”.

As if the various combinations of person, number, tense, mood, and voice

weren’t enough, Greek also has at its disposal a rich panoply of infinitives and
participles which act as verbal nouns and adjectives respectively.
Take the infinitives, which in English are generally preceded by the word
to, and whose status as nouns can be seen by simple substitution tests, in which
the infinitive is replaced by a noun of equivalent force and syntax: to err is
human, to forgive divine corresponds roughly to fallibility is human, forgiveness
divine. English has the basic present infinitive to write, but other forms require
somewhat clumsy periphrases like to have written; a Greek verb like lȳō, by
contrast, has no fewer than eleven infinitives:

present active lȳein “to set free (as an ongoing event)”

future active lȳsein “to be going to set free”
aorist active lȳsai “to set free (as a one-time event)”
perfect active lelykenai “to have set free”
present middle-passive10 lȳesthai “to get someone set free (ongoing
event)”
    or “to be set free (ongoing event)”
future middle lȳsesthai “to be going to get someone set free”

10  In the present and perfect, there are no distinctions of form between the middle and passive
voice.
 Greek25

aorist middle lȳsasthai “to get someone set free (one-time

event)”
perfect middle-passive lelysthai “to have gotten someone set free”
    or “to have been set free”
future passive lythēsesthai “to be going to be set free”
aorist passive lythēnai “to be set free (one-time event)”
future perfect passive lelȳsesthai “to be going to have been set free”

Now a couple of these forms are extremely rare: the Ancient Greeks didn’t feel
the need to say “to be going to have been set free” all that much more often than
English speakers do. But most of them are a vibrant part of the language—espe-
cially since the definite article, with its ability to mark case, can be placed in
front of any of them, thereby showing more clearly how they act as nouns and
making them an incredibly supple tool for expressing fine nuances of thought.
The distinction between the present and aorist infinitive, for instance, is not
the same as that between the present and aorist indicative, which is often, in
effect, a difference of tense, with the aorist playing the role of the simple past
tense. In the infinitive, by contrast, another important opposition between the
present and aorist comes to the fore, since, rather than just marking tense, the
stems of the verb associated with these two forms fundamentally mark a differ-
ence in aspect. In some languages, like Greek (and also Russian), there is a
basic division between verb forms that present the verbal event as incomplete
and open-ended on the one hand (the so-called imperfective, not to be con-
fused with the imperfect), and those that set a boundary on the event, viewing
it as a complete whole (the perfective, again, not to be confused with the per-
fect). The closest analogue in English lies in the difference between the past
progressive (I was walking) and simple past (I walked): the former typically
indicates that the event was still under way when something else happened
(I was walking along minding my own business, when all of a sudden an anteater
appeared out of nowhere), whereas the latter presents it in its entirety (I walked
to the store without seeing a single anteater).
In Greek, this distinction is widespread in the language, leading to the differ-
ent translations of the present and aorist infinitives given above: one depicts the
verb as ongoing, the other as a complete, one-time event. To see how this works
in practice, consider the following pair of sentences from Plato’s dialogue Crito
(50c and 50a), which show the contrasting use of the present and aorist infini-
tives of apokrı̄nomai “answer”, given in bold:11

11  In the four rows, I first give the Greek, then gloss it (translate it word for word, identifying
relevant linguistic features), then finesse that gloss into a word-for-word translation, then give an
idiomatic translation.
26 How dead languages work

eiōthas      chrēsthai   tōi    erōtān     te12  

you’re.accustomed  use.pres.inf  article  ask.pres.inf  both 
You’re accustomed  to use    the   asking        
kai apokrı̄nesthai
and answer.pres.inf
and answering
You’re accustomed to make use of both asking and answering.

ouk echō  apokrı̄nasthai  pros ho  erōtāis    

not I.can  answer.aor.inf to what you’re.asking   
I cannot  answer     to what you’re asking   
I cannot answer what you’re asking.

In the first sentence, Socrates’ interlocutors (in this case, the personified laws of
the city of Athens) have asked him a question, and they expect him to engage
in the same back-and-forth of question and answer that has marked the whole
of his philosophical life. Because it’s a matter of repeated, habitual, open-ended
asking and answering, the present infinitive apokrı̄nesthai is used. In the
­second, however, Crito is expressing his inability to answer a single, particular
question that Socrates has just posed, and so Plato uses the aorist infinitive
apokrı̄nasthai instead.
That leaves the participles, yet another verb form that is especially well rep-
resented in Greek. Leaving aside constructions formed with helping verbs,
English only has two of them, one present active (-ing), the other past passive
(most often -ed), both of which turn verbs into adjectives that can modify
nouns: the sleeping giant is a reduction of the giant who sleeps, while the
ripped jeans is equivalent to the jeans that were or had been ripped. Greek, once
again, has a much fuller complement: present, future, aorist, and perfect active;
present and perfect middle-passive; future and aorist middle; future, aorist,
and future ­perfect passive. The combination of the various tenses and moods—
not to mention the fact that a Greek participle, because it’s marked for
case, gender, and number, is still linked to its noun even when it drifts away to
different parts of the sentence—results in a linguistic feature of considerable
versatility.
Now readers might be forgiven for wondering whether it’s worth it to have to
learn such a rich inventory of forms in order to read Greek. Indeed, beginning
students of the language wonder this all the time. One can partially answer the
question by noting that the various endings do by and large fall into regular
patterns, so it’s not as if all 283 forms have to be learned in isolation from one
other. Consider the present indicative, future indicative, and present subjunct­
ive of lȳō in the active voice:

12  The placement of this particle after its position in English will be discussed below.
 Greek27

  present indicative future indicative present subjunctive

  “. . . set(s) free” “. . . will set free” “. . . ever set(s) free”13
1sg “I . . .” lȳ-ō lȳ-s-ō lȳ-ō
2sg “you . . .” lȳ-eis lȳ-s-eis lȳ-ēis
3sg “(s)he . . .” lȳ-ei lȳ-s-ei lȳ-ēi
1pl “we . . .” lȳ-omen lȳ-s-omen lȳ-ōmen
2pl “y’all . . .” lȳ-ete lȳ-s-ete lȳ-ēte
3pl “they . . .” lȳ-ousi lȳ-s-ousi lȳ-ōsi

The future indicative is the same as the present indicative, apart from the fact
that an s is added to the stem as a sign of the future. Similarly, except in the
third-person plural, the present subjunctive is the same as the indicative, only
with the first vowel of the ending lengthened (if it is not already long). So, rather
than having to learn 18 independent forms for these inflectional matrices, one
really only has to learn the six forms of the present, plus two rules for deriving
the future and subjunctive, and a third for coping with the irregularity in the
third-person plural of the subjunctive. Attentive readers will also notice that
I’ve not given any dual forms, which make up about a fifth of the total number
of 283: they’re uncommon enough that it’s fairly standard practice not to teach
them to beginning students because one can pick them up later on as one actu-
ally stumbles across them in texts.
Still, no amount of rationalization along these lines will magically sweep
away all the difficulties of learning the Greek verb; it just takes time. But in the
end, it repays the effort because a knowledge of Greek gives you access to one
of the richest literary traditions in the world. And it is to that tradition that we
now turn.

THE ILIAD

Even though we have seen only a small fraction of the subtleties that the copi-
ous morphology of Greek enables the language to express, at this point it will
give a much more well-rounded picture of the language if we move away from
the abstract and turn to some actual examples. We start with what may rightly
be considered the beginning of Western literature, the opening of the Iliad.
I give first the text in Greek, then a transliteration (including vowel quantities,

13  It’s very hard to come up with an English equivalent for the subjunctive, since its meaning
varies considerably depending on context. Because it’s so common in subordinate conditional
and temporal clauses that express an indefiniteness regarding the occurrence of the event in
­question, I have opted for “ever”.
28 How dead languages work

but omitting the accents), with a word-by-word interlinear translation as well

as a smoother translation adapted to the word order of English:

μῆνιν ἄειδε, θεά, Πηληϊάδεω Ἀχιλῆος

οὐλομένην, ἣ μυρί’ Ἀχαιοῖς ἄλγε’ ἔθηκε,
πολλὰς δ’ ἰϕθίμους ψυχὰς Ἄϊδι προΐαψεν
ἡρώων, αὐτοὺς δὲ ἑλώρια τεῦχε κύνεσσιν
οἰωνοῖσί τε πᾶσι, Διὸς δ’ ἐτελείετο βουλή,
ἐξ οὗ δὴ τὰ πρῶτα διαστήτην ἐρίσαντε
Ἀτρεΐδης τε ἄναξ ἀνδρῶν καὶ δῖος Αχιλλεύς. (Il. 1.1–7)
mēnin  aeide,  theā,   Pēlēïadeō   Achilēos   
wrath  sing,  goddess, of.Peleus’.son Achilles   
Sing the wrath, goddess, of Peleus’ son Achilles,
oulomenēn,  hē    mȳri’   Achaiois     alge’  ethēke  
accursed,  which countless for.the.Achaeans pains made  
accursed wrath, which made countless pains for the Achaeans
pollās  d’   iphthimous  psȳchās  Aïdi    proïapsen  
many and noble    souls  to.Hades sent.forth  
and sent forth many noble souls to Hades,
hērōōn,  autous  de  helōria  teuche  kynessin  
of.heroes, them  and prey  it.made for.the.dogs  
(souls) of heroes, and it made them prey for the dogs
oiōnoisi   te   pāsi,  Dios   d’  eteleieto        boulē,
for.the.birds and all,  of.Zeus and was.being.accomplished the.will
and for all the birds, and the will of Zeus was being accomplished,
ex   hou   dē    ta     prōta  diastētēn    erisante 
from when indeed at.the first  they.stood.apart having.quarreled
from the point when first they were at odds, having quarreled,
Atreïdēs te anax andrōn kai dı̄os Achilleus
Atreus’.son both lord of.men and godlike Achilles
both Atreus’ son (Agamemnon), lord of men, and godlike Achilles.

There’s a lot here to unpack. Look first at some of the individual words, several
of which should immediately be familiar to English speakers.
• In the first line, “goddess” is theā, as in theology, but with the feminine
ending -ā, related to the feminine -a seen in Latin and Spanish as well.
• The word for “countless” here is my r̄ ia, with the final vowel elided before a
following word beginning with a vowel. It’s also the word for “ten thousand”,
 Greek29

but like the derivative myriad in English, so too the Greek adjective, with
a slight change in accentuation, came to denote any large number.
• Next we see that the anger of Achilles sends down to Hades the souls of
heroes: psȳchās hērōōn. Our word hero is a direct borrowing from the
Greek, as is only fitting given the extent to which Achilles is a paradigmatic
hero in the Western tradition. And the word for soul, psȳchē in the singular,
has given us such derivatives as psychology and psychiatry.
• Sometimes, however, the Greek uses a word that isn’t the normal word.
Consider oiōnoisi, the word here used for “birds”. This has no obvious
English derivatives—in keeping with its status as a marked word for
bird—whereas the standard Greek word, ornis (plural ornithes), does:
ornithology. There’s no easy way for a translator to capture the fact that this
is a special word, used in part to distinguish large birds of prey from
smaller birds, and in part to refer to birds as creatures whose paths of flight
were read as omens.
• Boulē, too, has resonances that elude easy translation. On the one hand, it
is the basic noun derived from the verb boulomai “want, be willing”, so
“will” here is an easy translation. But it goes beyond “will” to “plan,
design”,14 and it makes a difference to the theology of the Iliad if one here
translates “will”, suggesting only a general desire on the part of Zeus, or
“plan”, assuming he had a more specific design in mind.
But while individual words can reveal points of contact between Greek and
English, there are also some fundamental differences in how the languages
work. Perhaps the most striking of these is the word order. One particularly
pervasive example is the (by English standards) strange placement of conjunc-
tions like the words for “and”, de (often elided to d’) and te. This follows from a
rule that was inherited by a wide range of older Indo-European languages:
Wackernagel’s Law. In its simplest form, it can be understood as the principle
that unstressed words don’t go first in the sentence: instead they are placed after
a stressed word, which, generally located first in its clause, acts as their host.
There’s not much of this left in Modern English—except for contexts such as
the start of this very sentence, where a form of to be, in this case is, is reduced
to being phonologically dependent on the preceding word: it no longer has an
accented vowel of its own, and becomes just ’s. In a language like Ancient Greek,
many more words regularly have this status, in which they lack full stress of
their own and instead gravitate to the second position in the sentence: not just
forms of to be but also unstressed object pronouns (e.g. me, us) and certain
conjunctions, as seen here. This has the strange effect that, if one uses one of

14  It can also mean “council”, and in Classical Athens refers to the administrative body of the
Council of 500, often simply transliterated in modern editions as “the Boule”.
30 How dead languages work

these words for and, then rather than saying A and B, one says A B te.15 As it
happens, this particular word for and (which, as seen in the last line of the
excerpt, can also mean both), was passed down from PIE not only into Greek
but also into Sanskrit (as ca) and Latin, where the standard orthography
attaches it directly to its host word, as -que. This is why the Senate and People
of Rome are the Senātus Populusque Rōmānus (SPQR) and not the Senātus que
Populus Rōmānus. Because they are thus placed after, rather than before, the
words they connect, such conjunctions as te are called postpositives.
At other times, Greek word order diverges from English not because the
default location for a word is different but because it simply has more flexibility.
We already have seen that, thanks to case endings, Greek doesn’t need to use
word order to distinguish between, say, Achilles killed Hector and Hector killed
Achilles; instead, it can rearrange the words of a sentence like this to throw dif-
ferent types of emphasis on the various constituents. The opening line of the
Iliad shows nicely how this plays out, especially if we compare two competing
translations, both published by Penguin. Robert Fagles offers “Rage—Goddess,
sing the rage of Peleus’ son Achilles”, while Martin Hammond has “Sing, god-
dess, of the anger of Achilleus,16 son of Peleus”. On the whole, Hammond stays
closer to the Greek in an unadorned prose that, as far as possible, keeps a one-
to-one correspondence between the original and the translation. So since the
word for anger, mēnin, only occurs once, and since, in English, objects have to
come after their verbs, he can’t begin his translation with anger, since the
imperative sing has to come first. Fagles allows himself more freedom, and is
happy to double up on Achilles’ rage, including it first at the start of the line,
syntactically isolated from the rest of the clause with an em dash, then again
after the imperative, in the position required by English syntax. Does it repre-
sent a distortion of the Greek to repeat rage in the translation? Not really: the
beginning of a poem is prime literary real estate, and comparison with the two
other great epics of the classical world shows that, in all three, the first word or
words are chosen to indicate the thematic center of gravity of the poem. The
Odyssey opens Andra moi ennepe, once again an imperative addressed to the
Muse: “Recount (ennepe) to me (moi) the man (andra)”, which highlights
the central role of Odysseus; and Virgil’s Aeneid begins Arma virumque canō
“I sing (canō) of arms (arma) and of a man (virumque)” (note postpositive -que
once more), announcing the poet’s intention of combining both the Iliad’s
atmosphere of war and the Odyssey’s focus on a particular individual into a
single epic poem. In this light, Fagles’s decision to begin his poem with Achilles’
rage seems the better choice.

15  Some readers may object: but wouldn’t te be in second position in A te B? If one started
numbering positions from A, then yes. But because te adds a new unit to the preceding one, it
needs to go second in the new unit, which starts with B.
16  The spelling Achilleus is closer to the Greek; in Latin, however, it was spelled Achilles, which
became the traditional form of his name in English as well.
 Greek31

A second point of word order: earlier I noted that Greek has less trouble with
dangling participles because case endings link participles back to the nouns
they modify. We see this in play at the start of the second line: oulomenēn sim-
ply means “accursed”, and the reason it’s clear that it’s the wrath that’s accursed,
rather than Achilles or Peleus, is the -ēn ending: this marks the participle as
feminine, accusative (note the -n), and singular, and it therefore needs to
modify something else that’s feminine, accusative, and singular—for which the
only candidate in this sentence is mēnin, the wrath.
One might expect that proper nouns could simply be transliterated from
Greek into English, but in fact there are differences between the two languages
even here: the word Dios, for instance, is not immediately recognizable as the
name of Zeus. Now, had he been the subject of the sentence, then Zeus is indeed
the form his name would have taken. But, in keeping with the conservative
nature of religious terminology, his name in Greek has preserved an extremely
old alternation going back to PIE times. How Zeus and Dios could come to be
different forms of the same word is best understood by ­starting from two
premises:
• First, words in PIE were built to a root, a sort of stable consonantal skel­
eton, with the vowel at the center of the syllable changing or even dis­
appearing depending on the form. This feature survives in such English
alternations as sing – sang – sung.
• Second, sounds like i and u had a status in between that of vowels and
consonants. If they occurred between consonants, then, in contrast to
those consonants, they behaved more like vowels, and were simply pro-
nounced i and u. But if they occurred next to vowels like e and o, they
seemed more like consonants by comparison, and so were pronounced
like y and w respectively.
To return then to our sky god. In the almost algebraic notation used for PIE, his
name can be represented as *Di(e)u-, with *Diu- as the root, and the paren­
theses around the e in the middle of the syllable indicating that the vowel is
present in some cases but absent in others. As a subject, in the nominative case,
his name would have been *Diēus, with a full, lengthened ē between the di and
the u that formed the beginning and end of the root, and a final -s to mark the
nominative case. Because the i was sandwiched between a consonant and a
vowel, it was treated like a y, which was unstable in Greek, and, through a
lengthy sequence of sound changes, the dy cluster ended up, eventually, as a z.17

17  It’s not only in Greek that the dy cluster has proved unstable. One may compare the French
word for goodbye, adieu (< à Dieu “to God”—with Dieu ultimately from the same Indo-European
root as Greek Zeus). When it was borrowed into German, where it became the most common
informal word for goodbye, it lost the first vowel and the dy in the middle was reduced to the
sound of English ch, with the word as a whole ending up as tschüss.
32 How dead languages work

In the genitive case of PIE (once again, this is roughly equivalent to the noun in
English preceded by of), this noun showed a weaker form of the stem without
the e, followed by an ending *-os: *Diu-os. In this environment, it’s the u that is
lost in Greek, while the i remains intact, leading to Dios. As it happens, this root
shows up again in the final line of the selection. Achilles is, as often, described
as dı̄os, which, according as one prefers to see the root as reflecting the divinity
of the sky-god or the brightness of the sky, can be translated as either “godlike” or
“brilliant”: either one suits our hero.
One great advantage of studying the Iliad in Greek is that themes obscured
by translation can come to the fore in the original language. The second word
of the poem, aeide, is an imperative addressed to the Muse, inviting her to sing
the wrath of Achilles. The next time this word form recurs in the Iliad, it comes
in the ninth book, often regarded as one of the poem’s high points because of
the way it foregrounds the system of values that underlies Achilles’ decision to
stay out of the fighting. Achilles’ honor had been slighted by Agamemnon
in the opening book, when the latter, the supreme commander of the Greek
forces, had taken away from Achilles his war-prize, the captive slave Briseis.
By  the ninth book, Agamemnon has come to regret his decision and sends
Ajax, Odysseus, and Phoenix on an embassy to Achilles to win him over. When
the men reach Achilles in his camp, the scene is one of the most famous in
the poem:

τὸν δ’ εὗρον ϕρένα τερπόμενον ϕόρμιγγι λιγείῃ

καλῇ δαιδαλέῃ, ἐπὶ δ’ ἀργύρεον ζυγὸν ἦεν,
τὴν ἄρετ’ ἐξ ἐνάρων πόλιν Ἠετίωνος ὀλέσσας·
τῇ ὅ γε θυμὸν ἔτερπεν, ἄειδε δ’ ἄρα κλέα ἀνδρῶν. (Il. 9.186–9)
ton  d’   heuron   phrena   terpomenon  phormingi  ligeiēi   
him  and  they.found  his.heart  delighting   with.a.lyre  clear-voiced  
And they found him delighting his heart with a clear-voiced lyre,
kālēi   daidaleēi,   epi   d’  argyreon  zygon    ēen,   
beautiful  well-wrought  on.it  and  silver   cross-beam  there.was  
beautiful and well-wrought—there was a silver cross-beam on it—
tēn   aret’      ex enarōn    polin Ēetiōnos   olessas;     
which  he.had.taken  from the.spoils  the.city of.Eetion  having.sacked    
which he had taken from the spoils when he sacked the city of Eetion;
tēi   ho ge  thȳmon   eterpen,  aeide  d’ ara  klea      andrōn.
with.it  he  his.spirit  delighted  sang  and   the.famous.deeds  of.men
with it, he delighted his spirit and sang of the famous deeds of men.

First, notice in passing that, once again, every line has words that have made
it into English somehow or other: heuron “they found” is the same verb that
 Greek33

gave us Eureka! (from Greek heurēka “I have found it”) and heuristics. While
I have here translated phrena as “heart”, it is often closer to “mind”, and is found
in this sense in phrenology. The noun derived from kālēi, kallos “beauty”, is
the source of calligraphy; and the cross-beam of the lyre, the zygon, is literally
a “yoke”, just as a zygote is so called in English because it represents the yoking
together of two cells. Eetion’s city is none other than a polis (whence English
metropolis and political), having become polin, in the accusative case, because it
serves as the object of a verb. The correspondence between a final -n marking
the object and a final -s marking the subject also holds true for the word for
“spirit”, which appears as thȳmon in the final line, but as thȳmos in the nomina-
tive case, the etymological source of our ­thymus gland.
What makes this passage stand out, however, is not the fact that it uses words
which have relatives in English—that would be true of virtually any passage of
Greek of comparable length. Rather, it is the picture of how Achilles spends his
free time: we might expect him to be just a rough, uncultured soldier—in the
first book of the poem, Agamemnon had accused him of being a bloodthirsty
brute—but instead the men on the embassy find him singing. This is not to
say that his warrior side is somehow sidelined. After all, his lyre is explicitly
described as the spoils of war. In particular, he won it through sacking the city
of Eetion. This is no throwaway detail: Eetion was the father of Andromache,
Hector’s wife and one of the women who will suffer most when Troy is finally
taken by the Greeks. There is a poignant irony to Achilles’ artistic pursuits, which
we might think to represent his softer side, taking place on an instrument so
closely associated with the grief he is to cause his Trojan adversaries.
After introducing this aside, the poet returns to the main strand of the nar-
rative with his usual technique of ring composition: the digression on the lyre
began with phrena terpomenon “delighting his heart”, and it ends with a
resumptive thȳmon eterpen “he delighted his spirit”. This then leads directly
into the verb aeide, seen here in this form for the first time since the very first
line of the poem. But its function has changed: in Homeric Greek, the same
form can serve either as a second-person singular present imperative (the
directive Sing! that opens the poem) or as a third-person singular imperfect
indicative (he was singing), and it is in that second use that we have it here. The
poet’s initial appeal to the Muses is thus picked up here as a scene that is taking
place within the poem, with Achilles as the agent. Not only does the poet
­himself sing (the bards who sang epic poetry were called aoidoi, from the same
root as aeide), not only do the Muses sing—but the poet, ennobling himself,
even goes so far as to depict the hero of his poem taking part in precisely
the same activity that he himself is engaged in as he performs the Iliad before
his ­audience. (While this trick of turning an imperative aeide into an imper-
fect  of  the same form is possible only in Greek, the root in question has in
fact found its way into English. The word for “song” in this period was aoidē—
essentially the feminine form of the word for “bard”, aoidos. By Classical Greek,
34 How dead languages work

however, this had contracted to ōidē, and, by the time of the Roman Empire,
the -i- had dropped as well, leaving us with what would become our English
word ode.)
What is more, the object of song is also significant: it is the famous deeds
(klea) of men. If any word sums up the Greek epic tradition, it is this one.
Generally, one sees it in the singular—kleos—as an abstract noun usually trans-
lated “fame”. But one characteristic of Greek is that abstract nouns can be plur­
alized to refer to multiple concrete examples of that abstract noun: as if we
could say fames in English to mean instances or examples of fame. It is this sort
of plural that klea represents, and, even without paying close attention to the
particular words Homer uses, we can see that Achilles, in singing of famous
deeds, is doing precisely that which the epic poet himself does. But, if we turn
to the word kleos itself, we find our way into an even richer web of meaning. It
goes back to a PIE root *kleu-, which formed words referring to hearing and
fame in the various daughter languages, especially in their poetic traditions.
As a basic verbal root, it meant “to hear”. But with the ending -os added to it
(*kleuos, which—remember the equivalence of u and w—can also be repre-
sented *klewos), it was “that which is heard”: in other words, “that which is
famous”, or simply “fame”. This form survives not only in Greek kleos but also
in Sanskrit śravaḥ (the ś is pronounced like an English sh; the ḥ used to be an s,
but it weakened at the end of a word to become a murmured h), and in both
traditions it is also associated with an adjective meaning “imperishable” (aph-
thiton in Greek, akṣita- or akṣiti- in Sanskrit). While the attempt to work out the
poetic language of the Indo-Europeans must inevitably be a tentative enterprise,18
something like n̥dhgwhitom klewos is very likely to have been a phrase that
already existed in the parent language in reference to undying fame; it was then
inherited as aphthiton kleos in Greek and akṣitam śravaḥ in Sanskrit.
Nor is the *kleu- root restricted to these two languages. It’s also prominent in
the personal names found in several other branches as well. While it is most
transparent in the numerous Greek names in -cles, such as Sophocles “wise-
famous” or Pericles “very-famous”, it also occurs in Slavic languages as the -slav
element in Bohuslav or Mstislav. (The change of k to s in Slavic parallels the
change of k to ś in Sanskrit.) Further disguised by phonological change, it
occurs in Germanic in the name Ludwig, itself a linguistic cousin of French
Louis.19 This requires some further explanation. First, we have to start from a
different form, in effect a past participle *klu‑tos or *klū-tos “that which is
heard”, which survived in Greek klytos and Latin in-clutus, both “famous”.

18  For more on the Indo-European poetic tradition, see Chapter 5, as well as the convenient
introductions in B.  W.  Fortson, Indo-European Language and Culture (Malden, MA, 2010),
pp. 32–7, and J. Clackson, Indo-European Linguistics (Cambridge, 2007), pp. 180–4.
19  For still more examples of Indo-European personal names built from words not only for
“fame” but also for “god”, “battle”, and the like, see B. W. Fortson, Indo-European Language and
Culture (Malden, MA, 2010), pp. 38–9.
 Greek35

Second, in Germanic, the initial k sound weakened to an h, as also happened in

English heart (whereas the original sound was preserved in Latin cor and Greek
kardia), and in this particular phonetic context, the t would soften to a d.20 This
means that the equivalent Germanic form ought to have ended up with the
stem *hlūd-. In fact, Old English still has words that begin with hl-.21 But
because the h- was comparatively difficult to articulate in this environment, it
was soon lost, leaving just Lud- as the first element in Lud-wig. Within English,
hlūd survived as an adjective in Old English, but with the loss of the h and the
change of ū to ou in the Great Vowel Shift that occurred at the end of the Middle
English period, it became our modern word loud.
So Achilles sings of the famous deeds of men, using a verb also found at the
start of the poem. Now to someone versed in the criticism of most English
poetry, it might seem obvious that one should look to the careful deployment
of key words, presumed to be deliberate, as a clue towards understanding the
work as a whole. But this is in fact not always the case in the poems attributed
to Homer, because the circumstances of composition were considerably differ-
ent. This brings us to formulas.

FORMULAS

Since Achilles is the leading Greek hero in the Iliad, our poet often wants to
compose sentences in which he is the subject of a verb, or, in Greek terms, in
the nominative case. Here are all the examples from the first book of the Iliad
(611 lines of poetry) where Achilles occurs in the nominative:

Atreïdēs te anax andrōn kai dı̄os Achilleus (7)

tēi dekatēi d’ agorēn de kalessato lāon Achilleus (54)
toisi d’ anistamenos metephē podas ōkys Achilleus (58)
ton d’ apameibomenos prosephē podas ōkys Achilleus (84)
ton d’ ēmeibet’ epeita podarkēs dı̄os Achilleus (121)
ton d’ ar’ hypodra idōn prosephē podas ōkys Achilleus (148)
thambēsen d’ Achileus, meta d’ etrapet’, autika d’ egnō (199)
tēn d’ apameibomenos prosephē podas ōkys Achilleus (215)
ton d’ ar’ hypoblēdēn ēmeibeto dı̄os Achilleus (292)
hēmenon; oud’ ara tō ge idōn gēthēsen Achilleus (330)
hē d’ aekous’ hama toisi gynē kien; autar Achilleus (348)
tēn de bary stenachōn prosephē podas ōkys Achilleus (364)
dı̄ogenēs Pēlēos hyios podas ōkys Achilleus (489)

20  The change of k to h in Germanic will be discussed more in Chapter 4 in the section on
Grimm’s Law.
21  Readers of Beowulf may recall the similar cluster hr- in the name of king Hrothgar.
36 How dead languages work

Even if you don’t know a single word of Greek, a couple of patterns stand out.
First, there is the placement of the name in the line: every time but once, it
comes at the end of the line. The exception, in line 199, is also different in that
there is only one l in his name, which changes its metrical shape; this license
also applies to the name of Achil(l)es’ heroic—and prosodic—counterpart,
Odys(s)eus. More interesting, however, for understanding the composition of
Homeric poetry are the line-end examples. Look what happens if these lines
are sorted according to the word that precedes Achilles’ name:

toisi d’ anistamenos metephē podas ōkys Achilleus (58)

ton d’ apameibomenos prosephē podas ōkys Achilleus (84)
ton d’ ar’ hypodra idōn prosephē podas ōkys Achilleus (148)
tēn d’ apameibomenos prosephē podas ōkys Achilleus (215)
tēn de bary stenachōn prosephē podas ōkys Achilleus (364)
dı̄ogenēs Pēlēos hyios podas ōkys Achilleus (489)
Atreïdēs te anax andrōn kai dı̄os Achilleus (7)
ton d’ ar’ hypoblēdēn ēmeibeto dı̄os Achilleus (292)
ton d’ ēmeibet’ epeita podarkēs dı̄os Achilleus (121)
tēi dekatēi d’ agorēn de kalessato lāon Achilleus (54)
hēmenon; oud’ ara tō ge idōn gēthēsen Achilleus (330)
hē d’ aekous’ hama toisi gynē kien; autar Achilleus (348)

Six out of the twelve have the phrase podas ōkys, another three have dı̄os, and
only three do not fall into such a pattern. That is, in the first nine examples,
Achilles’ name is preceded by one of two epithets: podas ōkys, which means
“swift-footed”, or dı̄os, either “godlike” or “brilliant”. But what exactly is an
­epithet? The English word is derived from Greek epitheton, which just means
“adjective”—and adjective itself is nothing other than the Romans’ take on
Greek epitheton: both words signify simply a word that has been placed
(-ject- and -thet- in Latin and Greek respectively) next to (ad‑, epi-) another.
But in English an epithet is more than just any old adjective: it’s one that’s par-
ticularly closely associated with the noun it’s modifying, such as here where the
poet clearly had favorite adjectives to use in conjunction with Achilles.
But why so much repetition anyway? To answer this question, we need to
consider the metrical structure of the line. Greek epic poetry was composed in
dactylic hexameter: that is, there are six feet (hence, hexa-meter), each of which
has the dactyl as its basic unit: a long syllable followed by two short syllables.22
More or less, long syllables are those with a long vowel (those marked with
the macron, e.g. ē) or diphthong, or a short vowel followed by two or more

22  Daktylos is Greek for “finger”, and the metrical foot takes its name from the fact that a finger
is composed of a long bone followed by two short ones.
 Greek37

consonants.23 In dactylic hexameter, the dactyl can also be replaced by the so-
called spondee, a sequence of two long syllables. This substitution is common in
the first four feet of the line, rare in the penultimate, and obligatory in the last foot.
A verse can thus be scanned as follows, with vertical lines marking foot boundar­
ies; in this case, only the final foot is a spondee, the rest being all dactyls:

tōn d’ ăpă|meı̄bŏmĕ|nōs prŏsĕ|phē pŏdăs | ōky̆s Ă|chīlleūs (84)

And swift-footed Achilles, in reply, spoke to him

As it happens, the line usually breaks in two in the middle of either the third or
the fourth foot at a dividing point, signaled by a word boundary, called the
caesura; another common place for a word boundary lies between the fourth
and the fifth foot (the so-called bucolic diaeresis). Such points of articulation
within the line are extremely useful for the poet because—and this is where
Homeric poetry is significantly different from Virgil or Shakespeare—it was
composed orally. Rather than working with pen and paper, the Greek epic poet
sang his poetry extemporaneously, an impressive feat of memory that required
certain compositional tricks. Among these was the existence of set phrases, called
formulas by modern scholars, that slotted into certain parts of the line, gener-
ally before or after a caesura or diaeresis. Take line 84, for instance. It begins
with ton, a masculine singular pronoun indicating the object (“him”), followed
by de, the common postpositive word for “and” that marks a new unit of speech;
then comes the participle apameibomenos “replying”.24 Next comes the main verb,
prosephē, followed by the hephthemimeral (“of the seventh half-part”) caes­
ura—so-called because, if one divides each dactylic foot into two beats, a caes­
ura in the middle of the fourth foot comes after the seventh beat of the line. We
can thus fill out the metrical scheme given above with the numbers of the beats
as well as the hephthemimeral caesura, marked with a double vertical line:25

1tōn d’ ăpă|3meı̄bŏmĕ|5nōs prŏsĕ|7phē || pŏdăs |9 ōky̆s Ă|11chı̄lleūs

The line divides here not only metrically but also syntactically. What comes
before the caesura is the predicate of the sentence,26 what comes after it is the

23  For these purposes, the sequences ph, th, and ch count as single consonants, as is more
apparent in the Greek alphabet, where they are written with single letters.
24  The root of the word is ‑ameib- “change”, the same as underlies the word amoeba, the chan­
ging organism par excellence. With the prefix ap-, it comes to mean “reply”, just as adding the
prefix ex- to change can lead to the specific sense of a verbal exchange.
25  One could also note a penthemimeral (“of the fifth half-part”) caesura in the middle of the
third foot, between apameibomenos and prosephē.
26  Not true from a strict syntactic viewpoint, since the participial phrase in the first half of the
line is in agreement with the subject, but Greek often uses participles modifying the subject where
English would use adverbial phrases modifying the predicate, and, in terms of the sense, that’s
what’s going on here.
38 How dead languages work

subject: Achilles, preceded by the epithet “swift-footed”. Apart from line 489,
the other lines that end with swift-footed Achilles have a remarkably similar
structure. All five (counting line 84 as well) begin with some form of the third-
person pronoun (once toisi “among them”, twice ton “him”, twice tēn “her”),
followed by the particle combination de or d’ ara (both meaning roughly “and”),
followed by a participial phrase describing the circumstances of the action
(anistamenos “standing up”, apameibomenos “replying”, hypodra idōn “scowling”,
bary stenachōn “groaning deeply”), followed by either prosephē “he addressed”
or metephē “he spoke among”, both compounds of ephē “he spoke”. That these
five lines should follow such a regular pattern is already enough to suggest that
lines in which Homeric poetry narrates who said something are built following
a ready-made template. This impression is further confirmed by looking at the
rest of the poem: in total podas ōkys Achilleus occurs 30 times at line-end in the
Iliad; of these, it’s preceded by a compound of ephē 26 times.
One can also observe the template-like pattern from the other direction: ini­
tial half-lines of this same structure are followed regularly by the names of other
heroes and gods, accompanied by a particular epithet of the appropriate met­
rical length. To cite just three of the most important characters in the poems:

ton d’ apameibomenos prosephē kreiōn Agamemnōn (5× in the Iliad)

Then mighty Agamemnon spoke to him in reply
ton/tēn d’ apameibomenos prosephē nephelēgereta Zeus (8× Iliad, 6× Odyssey)
Then cloud-gathering Zeus spoke to him/her in reply
ton/tēn d’ apameibomenos prosephē polymētis Odysseus (5× Iliad, 45× Odyssey)
Then many-counseled Odysseus spoke to him/her in reply

Agamemnon has a longer name than Achilles, so his epithet kreiōn is corres­
pondingly shorter; since Zeus is just a monosyllable, his epithet nephelēgereta
is a full six syllables long. Now not only do these particular epithets come up
extremely frequently (the figures for Odysseus are one sign that his words dir­
ect the action of the Odyssey to a remarkable extent) but, even more important­
ly, every time these initial half-lines are followed by the names of these particular
heroes, one never finds a different epithet instead: every major character of the
poem has one and only one epithet that occurs in this metrical context. The
significance of this distribution was first noticed in the ground-breaking work
of the early twentieth-century classicist Milman Parry. In particular, what this
so-called economy in the use of epithets resembled was not later literary epic
(such as Virgil) but rather contemporary oral traditions of epic poetry as prac-
ticed in what was then Yugoslavia. Rather than being solely the creation of a
single poet Homer—a position already questioned by a whole host of nineteenth-
century scholars—the Iliad and Odyssey appeared instead to be the product of
 Greek39

a long line of illiterate poets who, over the years, built up a highly systematic
set of epithets and other formulaic phrases that formed the basis of the poetry
that they then composed extemporaneously. This still of course leaves open
the possibility that a single “monumental” poet, coming at the end of the trad­
ition (probably in the eighth century bc), took the material he had inherited
and worked it into something closely resembling the form in which we have
it today.
Some modern readers have been troubled by the implications of the formula:
if it’s applied so mechanically when a particular character is named at a particu-
lar point in the line, what freedom does that leave the poet to be a poet? There
are a couple of responses to this concern. First is that there are a great many
lines of Homer that do not fall into as neat a pattern as those found in the lines
that introduce direct speech. To get a more balanced picture of the amount of
repetition found in Homeric diction, take the very first word form of the Iliad,
mēnin, the accusative of mēnis, “wrath”. It occurs twelve times in Homer, six
times at the start of the line, five times at the end, and once in the middle. There
are, to be sure, some recurring collocations: of the five line-final occurrences,
it’s twice the object of a verb of fearing in the second-person plural (edeisate at
Il. 13.624 and hypodeisate at Od. 2.66); it’s twice the object of a verb of caution
(epopizeo, the imperative “show respect for” at Od. 5.146, the imperfect indica-
tive ōpizeto at Od. 14.283); but one example has no exact verbal parallels, at Il.
5.34, where it is the object of the verb aleōmetha “let us avoid”. That said, it does
occur at the start of the line later in the same book (5.444), where it’s immedi-
ately followed by the related participle aleuamenos “having avoided”, and in all
these passages anger is, reasonably enough, regarded as something to be cau-
tious of. On the whole, the impression given is that the poet has considerable
flexibility in how he deploys the word—but that it tends to recur in certain
thematic contexts. Indeed, it’s often the case that a word can occur more than
once in close succession in such a way as to draw attention to a parallelism: after
kicking off the poem, the next occurrence of mēnin comes already at line 75 of
the first book—quite soon for a word that, on average, only occurs once every
four books. There, however, rather than referring to the wrath of Achilles, it
refers to that of the god Apollo. Indeed, most times this word for anger is found,
it’s a god who’s angry. Now, apart from being swift-footed, Achilles is also char-
acterized by the epithet dı̄os. As already noted, it’s ambiguous in meaning, but
ultimately is derived from the word that refers both to the bright sky—and to
the god of that sky, Zeus—and so is translated either “godlike” or “brilliant”. By
starting the Iliad off with mēnin, the poet not only puts Achilles’ anger at the
center of the work from the start but also hints at his superhuman status—a
hint that’s further confirmed when mēnin makes its next appearance a mere 75
lines later. To highlight this nexus between Achilles, anger, and the gods, espe-
cially Zeus, it’s worth citing lines 74 and 75 in full:
40 How dead languages work

ὦ Ἀχιλεῦ, κέλεαί με, Διῒ ϕίλε, μυθήσασθαι

μῆνιν Ἀπόλλωνος ἑκατηβελέταο ἄνακτος
ō Achileu keleai me Diı̄ phile mȳthēsasthai
O Achilles, you.bid me, to.Zeus dear, to.recount
mēnin Apollōnos hekatēbeletāo anaktos  
the.wrath of.Apollo far.shooting27 lord  
Achilles, dear to Zeus, you tell me to recount to you the wrath of Apollo, the
lord who shoots from afar

These lines are spoken by the prophet Calchas, who has been summoned by
Achilles to explain to the Greeks why it is that Apollo has sent a plague against
them. Achilles is thus the addressee of lines that, with his name substituted in
for Apollo’s, could equally well be used of the poet’s own narration of the wrath
of Achilles. Here a prophet is doing the telling rather than a poet, but the two
occupations were not considered so different from one another: when the
priestess of Apollo at Delphi, the Pythia, uttered the god’s oracles, they too were
cast in dactylic hexameter, and, when Plato, in his Phaedrus, enumerated the
types of madness that were beneficial, love may have been the most important,
but poetry and prophecy were included in the list as well. One final word
in  these lines might also draw some attention: the verb for “recount” is
mȳthēsasthai, an infinitive that contains in it the same root as that found in
the word myth. Can we go on to say that, just as the epic poet tells a myth
about  the wrath of Achilles, so too the prophet Calchas is asked to tell a
myth about the wrath of Apollo? Yes and no. If we replace English “myth” with
Greek mȳthos in that question, then we’re on safe ground: at this period, the
Greek word refers to a wide range of types of speech, and both these types of
narration can be subsumed under it. Its English derivative, however, has under-
gone considerable semantic narrowing through its general restriction to the
traditional tales of early peoples on the one hand and widely-held false beliefs,
like urban legends, on the other. Here we must bear in mind the dangers of the
so-called etymological fallacy: one mustn’t be too clever in seeing linguistic
relationships between words and assuming thereby that those relationships are
always perceptible to the speakers. Homer didn’t compose his poems with an
Indo-European dictionary by his side.
In sum, there’s repetition in Greek epic poetry: whole lines can be repeated,
as can the epithets of characters within those lines; and at a less systematic level,
thematic motifs can be highlighted by the clustering of a particular word at a
particular place in the metrical line. What, then, is the translator to do with

27  Or so it was interpreted by Homeric scholars in antiquity. In fact, the word originally meant
“who shoots at his will”, and the first element was wrongly understood to be the same as that in
hekas “at a distance, afar”.
 Greek41

this—mechanically translate the same words the same way each time? That’s
faithful to the Greek in an obvious sense. But it also does a certain amount of
injustice to the poetry: what would have worked well in the original oral per-
formance of the poems—repetition of certain lines as a sort of irregular refrain
(as happens, for instance, in certain styles of preaching)—comes across as
monotonous on the page for an audience used to the conventions of poetry that
was intended to be read. The other option is to vary the translation each time,
perhaps fleshing out ideas that are latent in the text. Some (Hammond) opt for
the former, others (Fagles) for the latter. Either way, since we can’t recover the
original audience’s experience of listening to Homer in a world not yet steeped
in writing, our best chance of appreciating the poet’s artistry comes from read-
ing the poem in Greek.

THUCYDIDES AND ABSTRACT LANGUAGE

Most readers of this book will probably be sympathetic to the idea that poetry,
in particular, loses a great deal in translation. But what about prose? It’s a more
pedestrian form of language—literally, in fact, in Greek, where the adjective
pezos can mean either “on foot” or “in prose”.28 Its ostensibly more straightfor-
ward nature might suggest that prose could be translated into English with less
damage done to the sense of the original, and in some cases this may well be
true. But much of the greatest Greek prose is also extremely difficult, and dif-
ficulty in Greek correlates closely to difficulty in translation. No author illus-
trates this better than Thucydides. In his history of the Peloponnesian War, the
late fifth-century bc conflict between Athens and Sparta that marked the end
of the age of Pericles, he wanted, as he puts it, to do more than just tell crowd-
pleasing anecdotes (something of a dig at his predecessor Herodotus). He
aimed to describe the events of the war in such a way that future generations
could profit from his account by drawing universal lessons from the actions of
the warring states—and their consequences.29 For his work to have lasting

28  At this point, I cannot resist sharing a favorite misreading of the Odyssey. When strangers
arrive at the island of Ithaca, they’re greeted, several times in the poem, with the words: “Who are
you, and from what race of men? Where’s your city and your parents? On what ship did you
arrive? How did sailors bring you to Ithaca? Who did they claim to be? After all, I don’t think you
arrived here pezos” (1.170–3). The correct translation is, of course, “on foot”, but I’ve often been
tempted to read this pezos in a grotesquely anachronistic fashion as “in prose”, to yield the almost
Cavafy-like idea that one can only get to Ithaca through poetry. (Readers unfamiliar with Cavafy
are warmly encouraged to go to https://www.onassis.org/initiatives/cavafy-archive/the-canon
and browse the poems collected there, starting with ‘Ithaka’.)
29  Considering how often Thucydides continues to appear in the news—recently, for instance,
in the context of political scientist Graham Allison’s ‘Thucydides trap’, whereby a dominant power
responds to a threatening new power with war—he was apparently successful in this aim.
42 How dead languages work

value in this way, Thucydides found it important to move beyond the concrete
details of who defeated whom in which battle: he had to describe the motiv­
ations of the opposing parties, often in paired sets of speeches that argued for
and against a given course of action. These debates, at least as he represents
them, were couched in terms of grand abstractions, in the language of justice
and expediency, of the contrast of word and deed, that was certainly character-
istic of Greek political discourse generally but is unusually concentrated in
Thucydides. Indeed, it is precisely because his work is so dense with abstract
language that it becomes in places so difficult to understand.
One excellent starting point for getting to know this particularly challenging
type of Thucydidean language comes from an episode in which he is describing
the civil strife on the island of Corcyra (modern Corfu), with the aristocratic
and democratic factions in the city at odds over how they should align them-
selves in the war. The situation spirals out of control, and Thucydides presents
a grim picture of cruel bloodshed in the city, as even the sacred bonds of family
are torn apart in the conflict: father kills child, and temples provide no sanctu-
ary from murder (3.81.5). But for the linguist, what is even more noteworthy is
the violence done to language, violence that Thucydides takes pains to describe
at length. Words no longer mean what they used to: what had been considered
reckless daring is now labeled loyal courage, prudent delay is rebranded spe-
cious cowardice. What is perhaps most striking is that this violence is even
reflected in Thucydides’ own expressions to a degree that is difficult to convey
in translation. Consider the sentence with which he opens this discussion
(3.82.4), which I quote here in several translations to show just how much
­variation there is in how the translators cope with the Greek:

Robert B. Strassler’s 1996 revision of Richard Crawley’s 1874 translation

(the Landmark Thucydides): “Words had to change their ordinary mean-
ing and to take that which was now given them.”
Benjamin Jowett’s revised 1900 Oxford translation: “The meaning of words
had no longer the same relation to things, but was changed by them as they
thought proper.”
Charles Forster Smith’s revised 1920 Loeb: “The ordinary acceptation of
words in their relation to things was changed as men thought fit.”
Rex Warner’s 1954 Penguin: “To fit in with the change of events, words,
too, had to change their usual meaning.”
A. W. Gomme’s 1956 Oxford commentary: “The customary meanings of
words were changed as men claimed the right to use them as they would to
suit their actions.”
Simon Hornblower’s 1991 Oxford commentary: “And they exchanged their
usual verbal evaluations of actions for new ones, in the light of what they
thought justified.”
 Greek43

While the general sense of the words is clear from all six, the dissimilarities are
striking: only Hornblower depicts the men who change the meanings of the words
as the subject of an active verb; four of the translators speak of the meanings of
words, but Forster Smith prefers “acceptation” and Hornblower “verbal evalu-
ations”; as for the nature of the change, Jowett, Forster Smith, and Hornblower
state that the new meanings are what men thought “proper”, “fit”, or “justified”;
Gomme hints at this with “claimed the right to use them . . . to suit their actions”,
but Warner merely says that the changes were made “to fit in with the change of
events”, and Crawley passes over in silence the idea that there was a conscious
calculation of what the new meanings ought to be. Without consulting the ori­
ginal, even a reader who compared these translations scrupulously would be
hard-pressed to work out that it is Hornblower’s translation that is closest to the
Greek: an unspecified “they” is the subject of an active verb; the word that most
translate as “meaning” is not a customary word for “meaning”; and there is a
word that justifies his use of “justified”, although it is, in fact, an abstract noun
rather than a participle.
In short, to understand what Thucydides is really saying here, one has no
choice but to look at the Greek, to which I append first word-by-word glossing,
then as literal a translation as possible within the constraints imposed by
English idiom:

καὶ τὴν εἰωθυῖαν ἀξίωσιν τῶν ὀνομάτων ἐς τὰ ἔργα ἀντήλλαξαν τῇ δικαιώσει

kai tēn eiōthyian axiōsin tōn onomatōn
and the accustomed evaluation of.the words
es   ta  
erga    antēllaxan   tēi   dikaiōsei
to the actions they.exchanged by.the justification
And they exchanged the accustomed evaluation of words to their actions by
justification.

Before considering the details, one may notice here the compression of
Thucydides’ language. Whereas the translations vary between 16 and 22 words,
and my version comes in at 13, Thucydides fits this all into just 12 words.
(Readers who have some Latin may be accustomed to such succinctness and
find it natural, but Greek is actually quite different from Latin in this respect,
not least because, unlike Latin, it has a definite article.30) And the unintelligibility

30  To be precise, the article only comes into use in between the time of Homer and Thucydides:
what later came to be the article of Classical Greek was in Homer still the demonstrative pronoun
seen at the beginning of the formulaic lines of speech discussed in the previous section. The shift
in usage whereby a demonstrative becomes an article is a common one: Romance articles like
Spanish el and la go back to the Latin demonstrative ille, illa “that”, and the English article the owes
its initial consonant to the th- of that.
44 How dead languages work

of my translation goes a long way towards explaining why the translators

expand: one needs to, if the English is to make any sense. What readers of the
translations won’t realize, however, is the fact that the Greek, as Greek, is prob-
ably as difficult to understand as my English version is. For the language here is
so dense that virtually every word in this passage could be (and has been) the
subject of scholarly study.
Consider first the verb, antēllaxan, which I have translated “they exchanged”.
This is a form of the compound verb ant-allassō, where the main verb allassō
means “change”, and the prefix, an elided form of anti-, here introduces the idea
of a reciprocal act. (This force is also seen in the English word antiphonal, used
of music in which different voices, or phōnai, answer one another in turn.) But
of the translations offered above, only Hornblower’s commentary renders the
verb as “exchange”, with the other versions all reducing it to simply “change”.
Why do they do this? The best answer is probably: because it makes for more
natural English. When we speak of exchange, there’s generally a clear quid pro
quo: one gives one thing, and receives another in return. But this sort of reci-
procity doesn’t seem to align well with the meanings of words, which usually
shift in non-discrete ways, so the translators select a plainer, more generic
verb.31 They are not alone in this: Liddell–Scott–Jones (LSJ), the standard
scholarly dictionary of Ancient Greek, does so too, and, tellingly, of the eight
examples of this verb that they translate, only in this passage do they render it
simply “change” rather than “exchange” or “interchange”. But had Thucydides
wanted to write just that words changed their meanings, there were other verbs
he could have used to do so. To understand why he chose so marked a verb, it
helps to pay attention to the wider context, such as the examples he goes on to
give in the very next sentence:
τόλμα μὲν γὰρ ἀλόγιστος ἀνδρεία ϕιλέταιρος ἐνομίσθη, μέλλησις δὲ προμηθὴς
δειλία εὐπρεπής
tolma men gar alogistos andreiā philetairos enomisthē
recklessness -- for unreasoning courage loyal was.considered
mellēsis  de  promēthēs  deilia     euprepēs   
delay  --  prudent  cowardice specious   
For unreasoning recklessness was considered loyal courage, and prudent
delay, a cowardice that only looked good.

This pair of examples—and it goes on in this vein—shows that Thucydides isn’t

talking about gradual semantic shifts here but about a wholesale inversion of

31  That said, words themselves, as opposed to their meanings, are of course frequently prone to
exchange in ordinary conversation, as is reflected in the Greek lexicon: as pointed out in n. 24
above, the root ameib- “change” underlies the common Homeric verb apameibomai “exchange,
reply”.
 Greek45

the language of virtue and vice. This is very much in line with the tenor of the
passage as a whole, which describes the horrendous violence that accompanied
the civil strife on Corcyra as a complete reversal in moral polarity. That there’s
a complete switch of positions is also suggested at a local level by the two words
I’ve left untranslated in the sentence above, men and de. These particles are
among the words most characteristic of Greek, and they are generally taught as
meaning “on the one hand” and “on the other hand”, only less ungainly. (Replace
a one-syllable word in one language with a four- or five-syllable phrase in
another, and chances are you’ve not captured the nimbleness of the original.)
Their basic function is to set up a gentle contrast between two opposed ideas,
usually throwing some emphasis on the words that immediately precede them
(in this case, tolma “recklessness” and mellēsis “delay”). Here they highlight the
balance between the two halves of the semantic situation Thucydides describes:
acting quickly has been reevaluated as a strength, and acting slowly as a weak-
ness. And that this is a question of reevaluation is foregrounded in turn by
Thucydides’ choice of axiōsin, as we’ll see below, rather than a more straightfor-
ward word for “meaning”, as most of the translators have it.
Now verbs like exchanged don’t operate in a syntactic vacuum, and antēllaxan
here works closely together with the phrase es ta erga, introduced by the prep­
osition es, which in its most basic spatial sense means “into, to”. Thucydides,
however, frequently requires his prepositions to carry more weight than most
other Greek authors do (more on this in the next section), in which case one
has to flesh out the translation to yield passable English. Here, “to suit” or “to
match” works well (that is, not “they exchanged the evaluation to the actions”
but “to suit or match the actions”), but in any case the translator has to make an
interpretative choice.
Such questions become even trickier when one turns to the nouns, especially
when, as here, the nouns are abstract. Ostensibly, onomatō n (nominative sin-
gular onoma) ought to be straightforward, as it is the basic Greek word for
“name”—a word with which it is in fact cognate, and a meaning seen in the
derivative onomastics, the study of personal names. But another common
English derivative hints at its broader use to refer to nouns or, indeed, words
generally, rather than proper names in particular, for the compound noun
onomatopoeia can be understood as either “noun-” or “word-making”. We here
see one of the more striking instances of the lack of a one-to-one correspond-
ence between Greek and English vocabulary. Most students of Ancient Greek
learn that the word for “word” is logos, often in connection with the opening of
the Gospel of John: “In the beginning was the Word (Logos).” But while this is
probably the best translation,32 it is nevertheless a misleading one, because
logos does not mean “word” in the sense “that which is distinguished in a text

32  Goethe’s Faust, a somewhat more advanced student of Greek, wavers over the best equiva-
lent: Wort “word”, Sinn “mind”, Kraft “force”, or Tat “act” (Part I, 1225–38).
46 How dead languages work

by having spaces placed on either side of it” but rather something closer to
“rational thought as expressed through language”, traditionally opposed to
ergon “deed, action” (whose plural erga is found in the sentence above). The
former sense of “word”, by contrast, has no single Ancient Greek equivalent—
in part, one suspects, because written Greek of this period did not generally
indicate word division, with all the words instead written continuously without
any breaks. It instead made do with onoma, used of nouns, and rhēma, of verbs.
If, then, Thucydides wanted to discuss the fact that words had changed their
meaning, onoma was the natural word to use in this particular context, since
the words he’s particularly concerned with are abstract nouns. That said, trans-
lating “noun” would not be right here, since it would imply too technical a
grammatical concept. But even the translation “word”, which is no doubt best,
is still potentially confusing, since it suggests the standard opposition of logos
(in the broad sense) and ergon, whereas Thucydides is here focusing on the
­lexical details.
Before turning to the other two nouns in the passage, we must also consider
even so humble a word as the definite article, which occurs four times in this
sentence, each time in a different form because of the case, number, and gender
of the noun it agrees with. The two that I have glossed “of the” and “by the” are
in the genitive and dative case respectively, and the preposition in the English
is determined by the case. For the genitive, “of ” is a fairly straightforward
equivalent here, but “by” is by no means the only option for the dative, a multi-
purpose case, which indicates not only (i) the indirect object (“to, for”) but also
(ii) location (“in, at”—in Classical Greek, usually a preposition is required in
addition) and (iii) means or instrument (“by, with, through”). The other two
articles, tēn and ta, are both in the accusative case, in the first instance to show
that its noun, axiōsin, is the direct object of the verb, in the second because erga
is the object of the preposition es. Here no special action is required in the
translation to bring out the force of the case, but it should be reiterated that
the fact that tēn eiōthyian axiōsin (“the accustomed evaluation”) is marked as
accusative by the endings of all three words enables Greek to put it anywhere
in the sentence, either before or after the verb, and it will still be clear that it is
the object.
But the case-marking on the article is far from the only complication for the
translator: generally labeled a definite article (with good reason), it plays a dif-
ferent role in Greek from in English. Sometimes it maps closely onto its English
equivalent, and so tēn axiōsin becomes “the evaluation”. But other times Greek
uses it when English sees no need to do so: if one translated “the accustomed
evaluation of the words” and “by the justification”, the English reader would feel
obliged to ask “Which words?” and “Which justification?”, since, in these
instances, the English article requires that the noun be more specifically defined
than the Greek one does. (In tōn onomatōn, for instance, the article simply
 Greek47

denotes that words are to be considered as a generic class.) In the remaining

example, where ta erga becomes “their actions”, we see another important use
of the Greek article: as a weak possessive pronoun. Greek has equivalents of the
English possessive pronouns, to be sure, and they become increasingly common,
but at this date they are emphatic enough that to use one here would throw the
sentence off-balance (as if one had italicized “their actions” in the translation).
As a result, it’s up to the translator to gauge the best equivalent for the article
each time: keep the article, omit it, or replace it with a possessive. In this
­sentence, for example, one has the option of keeping it more specific to the
­situation at hand by rendering tōn onomatōn as “their words”, or, as I have
­chosen, to generalize the semantic upheaval on Corcyra by leaving the article
here untranslated, suggesting that the linguistic ramifications of the events
there extended throughout the Greek world.
Everything I’ve just said about the article is characteristic of Greek generally.
But the last point to make—and it’s the most important—turns on a particular
stylistic feature of Thucydides’: his love of abstract nouns. Much of the vocabu-
lary of the older Indo-European languages is built on the basis of verbal roots,
as various suffixes can be added to turn these roots into different types of nouns,
adjectives, or indeed different verbs. The process of creating a noun from a
verb by adding a suffix is not so different from the gerund of English. One can
start from a verb like assemble and add an -ing, thereby yielding a form that
can be used in noun-like ways, e.g. as a sentence subject: Assembling IKEA
furniture is reasonably straightforward. Here the verb retains an important
verbal quality—it can take a direct object—but in other constructions, the
gerund can be still more noun-like and mark its notional object with the
preposition of, bringing it more into line with other nouns: syntactically,
the  assembling of IKEA furniture and the assembly of IKEA furniture work
the same way. Where English is less like Greek or Latin (and indeed most
modern European languages) is that it is extremely flexible about using nouns
as verbs and vice versa. Because it has gotten rid of nearly all the endings that
specifically characterize nouns and verbs as such in the other Indo-European
languages, there’s nothing about the form of a word that makes it look par-
ticularly like the part of speech that it is. Love, hate, play, and work all serve
perfectly well as either noun or verb—and, for that matter, can even be com-
pounded and turned into an adjective, in a love–hate relationship. (The free-
dom with which English can turn other parts of speech into verbs was once
brilliantly exploited by Bill Watterson in the punchline to a Calvin & Hobbes
cartoon: Verbing weirds language.) Words in Greek and Latin, by contrast,
have to have very specific endings to function as nouns or verbs, and so a rich
set of suffixes exists to turn verbal roots into nouns of different sorts. One
particularly important pair, Greek -si- and Latin -tiōn-, both go back to the
same PIE suffix, *-ti-, which underwent slightly different developments in the
48 How dead languages work

two languages. In Latin, the suffix survived as such, but to it was added an
additional suffix, ‑ōn-. In Greek (at least in the Attic dialect spoken in Athens),
the consonant t became an s when it preceded the vowel i, yielding ‑si‑,33 or,
with the nominative singular ending, ­-sis. Both suffixes have found their way
into English by way of numerous borrowings from Greek (thesis, stasis, meta-
morphosis) and Latin (faction, ration, and—from the same verbal root as
stasis—station).
Before proceeding further, it will probably be prudent to repeat the sentence
under consideration for ease of reference:

kai tēn eiōthyian axiōsin tōn onomatōn

and the accustomed evaluation of.the words
es   ta   erga    antēllaxan   tēi    dikaiōsei
to the actions they.exchanged by.the justification
And they exchanged the accustomed evaluation of words to their actions by
justification.

In our sentence, the -sis suffix occurs twice, in axiōsin and in dikaiōsei, which,
in the nominative case, would be axiōsis and dikaiōsis respectively. Now such
abstract nouns are formed from verbs, in this case axioō (“think worthy or fit,
expect”) and dikaioō (“think just, justify”) respectively. That both of these verbs
end in the sequence -oō is in turn an indication that they are not primary verbs
but rather verbs formed in turn from adjectives: axios “worthy, fit” and dikaios
“just”. Thus, whereas English translations have to rely on a vocabulary of
abstraction that’s borrowed largely from Latin (evaluation, justification), the
words Thucydides uses are derived from some of the most basic vocabulary
items of Greek. But his usage of axiōsis and dikaiōsis in this passage is so con-
torted as to require a separate Roman numeral heading in the entries of both
words in LSJ, just to deal with this one sentence. Following Hornblower, I’ve
allowed axiōsis to slide from its more usual sense “that which is thought worthy”
(with a range extending most often from “dignity, reputation” to “demand,
claim”), to “the assessing of the worth of something”, i.e. “evaluation”; and
dikaiōsis is not so much “the assertion that something is just” (the usual English
sense of “justification”) as “the assessment of what is just”. Both of these shifts
are reasonable enough given the force of the roots of the words, but neither has

33  This is a common sound change, and it can also be seen occurring independently in the
Romance languages, which have all softened the pronunciation of the -tiōn- words they inherited
from Latin: French (nation), Portuguese (nação), and Latin American Spanish (nación) have
turned it into an s, Castilian Spanish into the voiceless th of thin. Italian (nazione) did not get this
far: the z, which is essentially a t followed by an s, shows that it’s still kept the original t sound, but
has started to move in the direction of its Romance peers by relaxing the end of the articulation
into an s.
 Greek49

any obvious parallels in Greek, and they certainly can’t be ignored in interpret-
ing the sentence.
The change in the meaning of dikaiōsis, for instance, brings us back to one of
the misrepresentations found in nearly all the translations. Because dikaiōsis,
like English “justification”, usually refers to a statement justifying a position,
the actual weighing of the evidence is already complete in the mind of the
speaker offering the justification: the position, in his view, is just, and it simply
needs to be demonstrated to the audience. But here it seems to be more closely
connected with the initial step of working out what is just in the first place. The
assessment itself comes to the fore, and this in turn suggests the active engage-
ment of those who are changing the use of words. We thus return to the main
verb, antēllaxan. Nearly all the translators turn the words themselves into the
subject of the verb, either making the verb passive, or even keeping the verb
active, such that the words change their own meaning. This, however, downplays
the role of the human subjects of the verb in the Greek: it’s an active verb, and
those who change the meanings of words know what they’re doing. How do we
know that it’s a deliberate process? Because it happens tēi dikaiōsei, according
to what they think just. There are plenty of legitimate circumstances in which
to use the passive voice, but to deploy it here is to render Thucydides’ thought
too impersonal.
All in all, this single, 12-word clause is a remarkable tour de force on
Thucydides’ part: not only has he observed that the Greeks changed their lan-
guage to match the political circumstances but he’s even made that observation
in language that itself represents, mimetically, a deviation from what Greek
words had previously been able to express. There’s simply no adequate way
of  getting this across in translation, and that’s why it’s only when reading
Thucydides in Greek that one can properly appreciate what he’s doing with
his words.

PAULINE PREPOSITIONS

In the previous section, we saw Thucydides use the word dikaiōsis, which, as
the abstract verbal action noun from the verb dikaioō “justify”, means “justifi-
cation”. While the English word justification is most often used now to refer to
the reason or excuse one gives for a particular course of action, in one area of
discourse it has what might be considered a more basic meaning: not the
ex­plan­ation of why something is just but the actual process of making it just.
That is how Christian theology uses the word, in reference to God’s freeing
people from sin. As is often the case, theologians are drawing here on the lan-
guage of Paul’s epistles, notably the Letter to the Romans, which twice uses the
Greek word dikaiōsis:
50 How dead languages work

ὃς παρεδόθη διὰ τὰ παραπτώματα ἡμῶν καὶ ἠγέρθη διὰ τὴν δικαίωσιν ἡμῶν
(Rom. 4:25)
hos   paredothē    dia    ta paraptōmata hēmōn
who was.handed.over because.of the wrongdoings of.us
kai  
ēgerthē    dia    tēn dikaiōsin hēmōn
and was.raised because.of the justification of.us
Who was handed over to death for our trespasses and was raised for our
justification (NRSV)

ἄρα οὖν ὡς δι’ ἑνὸς παραπτώματος εἰς πάντας ἀνθρώπους εἰς κατάκριμα,
οὕτως καὶ δι’ ἑνὸς δικαιώματος εἰς πάντας ἀνθρώπους εἰς δικαίωσιν ζωῆς
(Rom. 5:18)
ara oun   hōs di’ henos paraptōmatos    eis pantas anthrōpous 
therefore  as through (of.)one wrongdoing  to all people     
eis katakrima,
to condemnation
houtōs kai di’ henos dikaiōmatos eis pantas anthrōpous eis dikaiōsin zōēs
so too through (of.)one just.act to all people to justification of.life
Therefore just as one man’s trespass led to condemnation for all, so one
man’s act of righteousness leads to justification and life for all (NRSV)

Paul is not generally regarded by the classicists who study Classical Greek as a
model of good prose style—after all, he’s writing in the first century ad, half a
millennium after the flourishing of Periclean Athens, in the later variety of
Greek known as Koine.34 Still, when set beside Thucydides, some curious par-
allels emerge.35 First, there is the same fondness for abstract nouns formed
from verbs—not just dikaiōsis (“a making just”) but also dikaiōma (“that which
is done justly”) from the same verb, but with a different suffix, as well as
paraptōma (“that which has fallen astray, wrongdoing”, from parapiptō “fall
astray”) and katakrima (“that which has been condemned, condemnation”,
from katakrı̄nō “condemn”). And while the strict parallelism of Paul’s con­
structions here is un-Thucydidean (in 4:25, verb, preposition, and prepositio­n­
al object are all balanced; so too the three prepositional phrases in 5:18), there
is some of the same compression of language, notably in the second verse,

34  Koine is simply the Greek word for “common”. It’s called this because, rather than being the
dialect of a specific place (Greece being a mountainous country, there was considerable dialectal
differentiation from city to city), it was a more neutral variety of the language that spread in the
Hellenistic Age (323–31 bc) into the territory conquered by Alexander the Great.
35  To be fair, most teachers of Greek prose composition would also frown on students’ produ­
cing sentences that resembled Thuc. 3.82.4 above.
 Greek51

where, if you look closely at the strict glossing of the Greek, there is not a verb
in sight. Two important questions arise:
• First, where did the translators of the New Revised Standard Version
(NRSV) get “led” and “leads” from?
• Second, if they’ve introduced verbs that aren’t in the Greek into the trans-
lation, where do these verbs’ respective subjects (“one man’s trespass”, “one
man’s act of righteousness”) come from?
Before answering these questions, it’s worth comparing the King James Version
(KJV) of the same verse:
Therefore as by the offence of one judgment came upon all men to condemnation;
even so by the righteousness of one the free gift came upon all men unto justifica-
tion of life.
The italics here are not my own addition; rather, this is the way that modern
editions of the King James indicate words that, in the 1611 original, were printed
in Roman type rather than Gothic—a typographical device to indicate those
places where the translators felt it necessary to insert words into the English to
produce a grammatical translation.36 Normally, the words that are omitted are
semantically light pronouns; we saw already when looking at Thucydides that
Greek needs fewer of these than English. But in this case, what’s missing in the
Greek is nothing less than both the subject and the verb of the two clauses. Now
a missing subject is perhaps not such a big deal. Greek, like Spanish, encodes
the person and number of the subject with endings on the verb (hablo “I speak”,
but hablamos “we speak”—no pronouns necessary), so as long as the identity of
the subject can be understood from context, there’s no need for a subject pro-
noun. To be fair, it is, by the usual standards of subject-omission, comparatively
difficult to extrapolate “judgment” and “free gift” as subjects of these clauses
from what precedes, but it’s also not impossible: just two verses earlier, in 5:16,
krima and charisma (“judgment” and “free gift”) had been subjects of two bal-
anced clauses (KJV: “For the judgment was by one to condemnation, but the
free gift is of many offences unto justification”), so it’s not too great a stretch to
understand them in the later verse as well. Still, the NRSV translators felt
uncomfortable with taking this step, so to give a preliminary answer to our
second question: they took nouns that were the objects of the preposition di(a)
in the Greek (rendered “by” in the KJV) and turned them into the subjects of
verbs in English. But we can only fully explain why they make this choice once
we’ve considered how they deal with the dearth of verbs in Paul’s Greek.

36  Paradoxically, this choice often leads to quite the wrong cadence when modern readers
approach the King James: if they’re unaware of the practice, they accent the italicized words
according to the usual convention, when the fact that the words are not in the original language is
usually a good sign that they are actually among the weaker words of the sentence.
52 How dead languages work

Typically, when a verb is missing in Greek, one supplies a form of “to be”. This
is common in other ancient languages too, such as Hebrew and Latin, and it’s
also true of Russian, which only uses the present tense of “to be” in a very
restricted set of circumstances. Thus, reading through the KJV of Luke (gener-
ally agreed to have the best Greek prose style of the evangelists), one finds early
on, in 1:42, “Blessed art thou among women, and blessed is the fruit of thy
womb”, with the translators supplementing “to be” in both clauses. Such omis-
sion is easier in Greek than in English because the participle translated
“blessed” has an ending that indicates that it either needs to agree with the
subject (“the blessed fruit”), or be the nominal predicate of the subject (“the
fruit is blessed”), and the word order excludes the former option. But if one
tries to turn the Greek of Romans 5:18 into good English simply by adding
in forms of “to be”, fleshed out with a dummy subject pronoun, the result,
while certainly more grammatical, is hardly more comprehensible than the
verb-free version:
Therefore just as through a wrongdoing of one, it was to all people to condemna-
tion, so too through a just act of one, it was to all people to justification of life.
To make sense of the sentence at this point, we then have to choose the approach
of either the King James or the NRSV. The former is simpler, since in effect it
requires simply replacing “it” with an implied subject in both instances, then
adjusting the verb from was to came. The first of these changes we’ve already
explored, but what’s the basis for the second?
The answer lies in the prepositions in the sentence. Most of the time, when
one needs to supply a verb in translating a sentence, there’s a noun (or pronoun)
that’s a subject, and there’s another noun or adjective that’s describing the sub-
ject: she is blessed, or he is a carpenter. But in this case, the predicate, to which
the verb must be accommodated, consists in each clause of a pair of prepos-
itional phrases: to all people to condemnation and to all people to justification of
life. Now eis, the preposition here translated “to”, is primarily used to indicate
the place to which the action of the verb is leading, as seen in a rather easier
verse to translate: “And he took him to Jerusalem (eis Ierousalēm)” (Luke 4:9).37
So too in Romans 5:18, what follows eis is the destination of the action, but
since expressions of place are typically adverbial rather than adjectival, the eis
phrases need a proper verb of motion to follow rather than just the gram­mat­
ical placeholder that is “to be”. Thus, not “it was to all people” but “it came to
all people”.
There remains the little matter that eis occurs twice in Greek, both with
the people as its object and with the abstract destination of condemnation or

37  It is in fact the same preposition as es, seen above in the sample sentence from Thucydides,
and the difference in form is only one of dialect.
 Greek53

justification. Here the best starting point for distinguishing between the two
usages is to note that the animate and the inanimate often elicit different lin-
guistic treatment. Languages, for instance, tend to structure narratives from
the perspective of people (especially the first person), rather than things, so
people are more likely to be sentence subjects, things to be objects. Indeed, this
gives rise to one common environment where the English passive makes good
stylistic sense. Rather than saying John was crossing the street, and a car hit him,
it’s often more natural to say John was crossing the street, and (he) was hit by a
car: the use of the passive enables John to be retained as the subject in the sec-
ond clause, which allows for a smoother presentation of the information con-
tent of the sentence, especially given that it enables the possibility of altogether
omitting a pronoun referring to the subject in the second clause.
If animacy is a useful parameter to introduce in discussing what gets chosen
as a sentence subject, what about our sentence here, with a preposition that
means to used twice? Broadly speaking, the English preposition to has two
major uses: one is to denote the indirect object, the recipient of a gift (I gave the
book to Anna); the other is to denote the destination (I went to the store).
Typically, animate nouns serve as indirect objects, and inanimate nouns as des-
tinations. There’s no iron law saying that this is always the case, but, given the
way most people view the world, an indirect object typically benefits from
the action, and we tend to do things to benefit people, not things; and if we’re
describing where we’re going, we usually pick fixed landmarks, and so a stable
location like a building or a city is more common than a moving target, like a
person. If we apply this tendency to Romans 5:18, then we’re led to maintain to
as the preposition used before justification and condemnation, as inanimate
destinations. But as the KJV and NRSV indicate, the choice is less clear in deal-
ing with the animate object: they offer “upon all men” and “for all” respectively.
The difference seems to reflect a varying degree of optimism about the message
of the verse. If the movement towards condemnation or justification is happen-
ing without regard to the people affected by it, then it can be seen, with the KJV
translators, as almost monolithically coming upon them. But if one takes a
kinder view of God’s grace, one might prefer the NRSV’s phrasing that it occurs
for them.
In either case, readers who know Greek may have been uncomfortable with
how I’ve been handling eis pantas anthrōpous till now. I’ve blithely used the fact
that to serves in English to mark both indirect object and destination as a way
to get at the Greek—when Greek in fact does not have this equivalence. It’s the
dative case that should signal the personal beneficiary of an action, not eis. And
that’s true, but only up to a point, since the dative case gradually went into a
decline in the post-Classical period. It has dropped out of Modern Greek
entirely, and the preposition that has replaced it is none other than se, the
ultimate descendant of eis. Already in the time of Paul, eis was frequently used
of “actions or feelings directed in someone’s direction in [a] hostile or friendly
54 How dead languages work

sense”, as the leading dictionary of New Testament Greek puts it.38 According
as one views the actions of condemnation and justification as hostile or friendly,
either upon or for might do the trick. This then gets us to a translation along
these lines:
Therefore just as through a wrongdoing of one, it came for all people to condem-
nation, so too through a just act of one, it came for all people to justification of life.
At this point, we’re quite close to a serviceable translation, and the main point
requiring clarification would be the specification of it, which the KJV takes
from the verses above. Still, while this solution might seem elegant, it doesn’t
quite do justice to the Greek, since it does require fleshing out an implied
subject, it, into e.g. judgment and free gift, when that very implied subject was
introduced into the rough draft English simply to fill out the syntax of a verb
that also wasn’t there in the Greek. So here the NRSV takes a different route.
It takes what had been the object of the preposition dia, henos paraptōmatos
“the wrongdoing of one” and henos dikaiōmatos “the just act of one”, and
treats these as the subject of a different verb of motion, lead. This might seem
to be more of a stretch than the path taken by the KJV, but it’s not unreason-
able. It copes better with the lack of a verb in the original by picking a verb
that doesn’t require importing a subject from outside the sentence, instead
promoting it from within. What’s required is simply the awareness that the
preposition dia (at least when it takes the genitive case, as here) marks the
instrument with which an action is completed. In Greek, such instruments
don’t work well as sentence subjects, since it prefers animate nouns in this
syntactic slot. But English is happy to allow a sentence like The arrow killed
the warrior (Greek would prefer The warrior was killed by the arrow). This
then gives the NRSV translators another way out of the syntactic impasse of
Romans 5:18. The wrongdoing and the justification, rather than playing a
subordinate role as objects of a preposition, can become the subjects of their
respective clauses, since English can treat instruments as subjects, whereas
Greek avoids this.
Whether one prefers the KJV or the NRSV, I’d like to stress two points. First,
there’s no ideal solution. Neither option allows for a perfect, one-to-one corres-
pondence between the Greek of the New Testament and modern-day English.
To understand the discrepancies between the countless translations of the
Bible, one has no recourse but to consider the phrasing of the original language.
People who’ve made it this far in the book probably don’t need to be told
that, but it’s still important to see a few concrete examples of the ways that
Greek operates differently enough from English that an ideal translation is a

38  F.  W.  Danker, A Greek-English Lexicon of the New Testament and Other Early Christian
Literature, 3rd ed. (Chicago, 2000), s.v. εἰς 4.c.
 Greek55

theoretical impossibility. Second, more specifically, note just how much rides
on the prepositions: it’s because of the prepositions that translators know to
supply a verb of motion; deciding between upon all people or for all people
requires a judgment call regarding a preposition; and the NRSV translators
rewrite a prepositional object as the sentence subject based on an awareness of
what the preposition dia can do. Prepositions may not be as glamorous as
nouns and verbs, but, like loyal servants, they stand ready to take up whatever
semantic burden the sentence requires them to carry.
 3

Latin

Nec mē animī fallit Grāiōrum obscūra reperta

difficile inlūstrāre Latīnīs versibus esse,
multa novīs verbīs praesertim cum sit agendum
propter egestātem linguae et rērum novitātem.
Nor does it escape my notice that it is difficult to shed light on the dark
discoveries of the Greeks with Latin verse, especially since one must treat
many subjects with new words on account of the poverty of the language
and the novelty of the concepts.
(Lucretius, On the Nature of Things, 1.136–9)

For the average American, Latin is the dead language par excellence. True,
many have come across Biblical Hebrew or Sanskrit for religious reasons, and
a handful of high schools still teach Ancient Greek; but only Latin is widely
taught enough to have its own Advanced Placement or SAT subject test. Why
does it have this pride of place?
In part, the answer is historical. For centuries after the fall of the Roman
Empire, Latin remained the language of the literate class. To be viewed as edu-
cated, one had to have the ability to read and write in Latin, and this only ceased
to be the case over the course of the seventeenth and eighteenth centuries as the
vernacular languages of Europe gained the upper hand. If a language spends
sufficient time as the chief means for the learned to communicate, that’s bound
to give it some staying power. This is only partially due to educational inertia,
for it also reflects the substantive ascendancy Latin acquired over time, as seen
in the legacy of hundreds of years of artistic, scientific, and religious discourse
conducted in the language. The part of that legacy that we’ll be exploring in this
chapter is found not in the ideas conveyed by the language, important though
they are, but in the language itself. This is most obvious in the case of lexical
borrowings. While it’s certainly possible to write a paragraph of standard
English prose without using any words derived from Latin, it takes a conscious
effort to do so; indeed, in the first half of this sentence alone, such basic words
as certainly, possible, using, and conscious all come from Latin. Which brings us
to one reason commonly given for the continued study of Latin in schools: it

How Dead Languages Work. Coulter H. George, Oxford University Press (2020).
© Coulter H. George. DOI: 10.1093/oso/9780198852827.001.0001
58 How dead languages work

helps your English vocabulary. But this, though not a bad reason to undertake
the language, is also not the best one. Latin is a complicated language, after all,
and if your only interest in studying it is to learn vocabulary for the SAT, there
are more efficient ways of doing so. Still, of the utilitarian reasons frequently
given for learning Latin, the idea that it improves one’s English more generally
holds a little more water. Any time someone learns a foreign language, it leads
them to consider the workings of their own language in greater detail, and,
since the syntactic structures of the canonical works of Latin prose and poetry
are relatively complicated, familiarity with the language can give students the
understanding needed to work with a greater repertory of grammatical possi-
bilities in English as well. Since the particular character of Latin can be seen
most fully in the rich literature the Romans have given us, in this chapter we’ll
look at three literary high points—the poetry of Lucretius and Horace, and the
prose of Tacitus—to get a sense of how Latin does things in ways that English
typically doesn’t.
Before turning to these examples, however, some historical context is
important. An observer of the fifth century bc would hardly have predicted
that Latin would be so wildly successful as a world language. At that point,
when Greek civilization was at its peak, the age of the great tragedians, of
Pericles and Socrates, Latin was spoken by only an insignificant number of
people in the immediate vicinity of Rome. Other languages loomed larger in
Italy: Greek was spoken in colonies all around Sicily and southern Italy; to the
north of Rome, the Etruscans spoke a completely non-Indo-European lan-
guage; and there were also languages like Oscan and Umbrian, which were
related to Latin—belonging to the same Italic branch of Indo-European—but
still different enough to count as separate languages rather than just dialects.
As  commendably detailed by Denis Feeney in a recent book, it was by no
means obvious that Latin would flourish as it did, and its history as a literary
language is an unusual one.1 For most of the languages covered in this book,
the earliest writings to survive are generally poems, often religious in nature,
arising chiefly out of a native oral tradition: the first Greek, apart from
Mycenaean administrative documents preserved by chance, is Homeric epic;
the earliest Sanskrit texts are the equally poetic hymns of the Rig Veda; and
many of what are thought to be the oldest passages of the Hebrew Bible, like the
Song of the Sea in Exodus 15, or the Song of Deborah in Judges 5, are poetry as
well. Eventually, in each of these cases, a literary prose language would then
develop to complement the poetic tradition.
With Latin, the situation is different. On the basis of the languages just
­mentioned, we might expect the earliest major literary work to be some sort
of  religious or mythological poem, building on native Italic oral traditions.
But instead, the first literary texts to survive in more than fragments are the

1  D. Feeney, Beyond Greek: The Beginnings of Latin Literature (Cambridge, MA, 2016).
 Latin59

comedies of Plautus, written in the late third to early second century bc—
slapstick comedies with stock characters like the braggart soldier and the clever
slave, which stand near the beginning of a tradition that ultimately led to A
Funny Thing Happened on the Way to the Forum. Not actually at the beginning,
however, since these plays are in turn based on Greek originals by playwrights
like Menander (active in the late fourth century bc). And if we turn to authors
of Latin slightly earlier than Plautus, whose works only survive in fragments,
Greek influence is again ubiquitous. The first name in Roman literature is
Livius Andronicus, who came from Tarentum, in Greek-speaking southern
Italy: according to Cicero’s Brutus, he put on the first play in Rome in 240 bc,
and is best known for his reworking of the Odyssey into Saturnians, a native
Italic verse form. Roughly contemporary, the tragedies and comedies of Naevius
also are also modeled on Greek originals.
But while the influence of the Greeks is a constant theme in Roman literary
history, it will only be an accompaniment to the main idea followed in this
chapter: the way in which Latin idiom—both what it inherited from Proto-
Indo-European and what it borrowed from Greek—encourages certain types
of expression that give it a different flavor from most modern European lan-
guages. That is to say, for English speakers, Latin has the reputation, compared
to Spanish or French, of being difficult. Some of that difficulty is due to the
nature of the texts assigned beginning Latin students: we give them Virgil and
Cicero much earlier in their studies than a Spanish student will have to read
Cervantes, or a French student Proust. And some of it is due to the foreignness
of literature written two thousand years ago: allusions are harder to under-
stand, the material culture isn’t self-explanatory, and even differences in social
norms can cause confusion. But there’s another force at work that we’ll explore
here: sentences are simply structured in Latin differently enough from in
English that there’s even less of a one-to-one correspondence between a Latin
sentence and possible English equivalents than would be the case with Spanish
or French.

HOW LATIN WORKS

Before turning to more complicated literature, we can see some of these dis-
tinctly Latin features in a couple of short common phrases, many of which have
found their way into English. One common theme that will run through this
section is the fact that Latin is generally a very concise language compared to
English, as it can compress into one word ideas that might require two or three
in English. It achieves this in part through the use of numerous endings to
express different grammatical relationships that require helping words of vari-
ous sorts in English. Much of this is comparable to what was discussed in the
60 How dead languages work

chapter on Greek, but Latin sometimes goes farther with this than Greek does.
Take cases, for instance. These are the endings that both Greek and Latin place
on nouns and adjectives to indicate what a noun, or the adjective modifying it,
is doing in a sentence: whether it’s the subject, object, or the like. Greek has
five of these, but Latin has a sixth, the ablative, which is a heterogeneous case
that, for historical reasons, serves three main functions. It can show that the
noun marks:
1. a location in space or time, as if the noun were preceded by a preposition
like “in”
2. the instrument with which an action is carried out, as if the noun f­ ollowed
a preposition like “with”; as is also true of English “with”, instrumental
usages of the ablative extend from denoting the instrument in a narrow
sense (“I wrote this with a pencil”) to accompaniment (“I arrived with
a friend”)
3. an origin or starting point, in space or time, as if the preposition “from”
came before the noun.
As an example of the first, locative usage, consider the English abbreviation ad,
which stands for annō Dominī “in the year of the Lord”. Here we see two Latin
words expand to six in English. This comes about for two main reasons. First,
Latin doesn’t have a definite article equivalent to English the, or, for that matter,
an indefinite article such as a(n). While it might not seem like a big difference,
the presence or absence of an article in English often tells the reader whether
the noun in question is something new in the discourse (in which case in­def­
in­ite a is more common), or whether it’s already active and present in the
addressee’s mind (in which case it gets the definite the), as in the contrast
between The girl saw a bird and The girl saw the bird. Second, Latin can be more
concise because of case endings. As the subject of a sentence in the nominative,
the word for Lord is Dominus, and the substitution of the ending ‑ī for -us here,
indicating that the noun is in the genitive case, does the same work that the
preposition of does in English. The same principle is at work with annō: the
word for year is annus in its usual dictionary form, and the change of -us to -ō
here marks the ablative case. Given the regular use of nouns of time like annus
in expressions that specify the temporal location of an event, that ending -ō
serves as the Latin counterpart to the preposition in in English.
The second, instrumental usage has also made it into English in the phrase
ipsō factō “by the very fact”. But the fact that the preposition here is by rather
than with is one clue that it’s not quite a prototypical example—a fact isn’t as
instrument-y an instrument as a knife or a hammer—but if it is the means by
which one clinches an argument, it’s close enough to fall under this category.
One can see how such classifications of case usages would come to have fuzzy
boundaries, and a couple more examples of Latin ablatives in English show
 Latin61

how the locative bleeds into the instrumental, and why it makes sense that a
single form could do service for both. Take the phrase prīmā faciē “at first
appearance”, where again both words are marked with the ablative. (Not all
words take the same endings to express the same cases.) Translated with at, this
ablative would seem to belong to the locative variety, but insofar as it is often
used of evidence or proof of an argument, it’s not so very different from the
ipsō factō example. Or take bonā fidē “in good faith”: the preposition in again
suggests location, but if something is done in good faith, is that stipulation
better described as the location where something takes place (as if it could be
expanded to “in an environment of good faith”) or as a sort of instrument or
means by which it happens (as if equivalent to “through the catalyst of good
faith”)? It’s not clear that one would always want to make a sharp distinction
between the two, and Latin captures that by allowing both ideas to be expressed
through its ablative case.
At times, though, Latin speakers did feel the need to distinguish more care-
fully what kind of ablative was in play, and they did so by means of the same
linguistic tool that English uses to express these relationships: prepositions.
In phrases like requiēscat in pāce “may s/he rest in peace” or in vīnō vēritās “in
wine, truth”, rather than using the bare ablatives pāce or vīnō, the preposition
in is added, in the same sense as in English, to specify that peace is viewed
as the surroundings in which the resting will take place, and the wine as the
place where the truth resides, rather than simply a tool for eliciting the truth.
Similarly, the preposition cum “with” may be used to strengthen the instru-
mental uses of the ablative, as in magnā cum laude “with great praise”. And the
third use of the ablative, to indicate the origin or starting point of an event, is
very often reinforced with a preposition, as in ē plūribus ūnum “from many,
one”, where ē, a reduced form of the preposition ex “out of ”, clarifies that the
many are the source of the one, rather than, say, the location of it.
Now annō Dominī is not the only chronological designation to display a
characteristically Latin linguistic feature. One means by which the Romans
reckoned time was to count the year in which, according to tradition, the city
was founded, 753 bc, as Year One, then number sequentially from there, such
that something that happened in, say, 44 bc, like the assassination of Julius
Caesar, would be said to have occurred in the 710th year since the founding of
the city. But how does Latin actually express “since the founding of the city”?
Where English requires six words, Latin does it in three: ab urbe conditā, liter-
ally “from the-city having-been-founded”. Latin has achieved its concision here
through two main differences. First, as already noted, it doesn’t have a definite
article, so neither “the” carries over into Latin. Second, English takes the verbal
idea of the city’s founding and treats it as a noun, the so-called gerund founding.
Its noun-like behavior can be seen both in the definite article that is used before
it and in the fact that its object, the city, is not simply tacked on after it, as in
They founded the city, but is governed by the preposition of, thus making the
62 How dead languages work

expression parallel to since the establishment of the city, where the word
­replacing founding is more obviously a noun.2 Latin, on the other hand, is a
little less strictly logical: rather than having the preposition equivalent to since
governing the word that refers to the founding of the city, which, in a narrow
sense, is what provides the starting point for the time-reckoning, it reframes
the idea such that the preposition ab governs the noun for “city”, urbe, and
recasts the word for the founding as a perfect passive participle—that is, a ver-
bal adjective meaning “having been founded”. This construction was imitated
by Early Modern writers steeped in the Latin tradition of epic: the very title
of Milton’s Paradise Lost, it has been argued, means not so much “Paradise,
which has been lost” as “The losing of Paradise”.3 In any event, this pattern,
in  which a phrase of the shape “the doing of something” is expressed as
­“something (having been) done”, is common enough in Latin—and ab urbe
conditā is a prominent enough example of it—that it is regularly referred to as
the ab urbe conditā construction.
In this type of expression, we see Latin’s preference for using a participle
(that is, a verbal adjective) rather than a gerund (a verbal noun). This favoring
of participles extends to another characteristically Latin expression, the abla-
tive absolute. This is a phrase that stands apart from the syntax of the rest of the
sentence (hence “absolute”), and in which the two main elements both stand in
the ablative case. In most examples, those two elements are a noun and a par­ti­
ciple, as in Deō volente “God willing”. As with English God willing, the con-
struction is equivalent to an subordinate clause, with the noun as the subject,
and the participle as the verb, as if this were an abbreviated form of if God is
willing. In this instance, the equivalent clause is generally understood to be
conditional (an if-clause), but one reason the absolute construction is so useful
in Latin is that the type of subordinate clause is left unspecified, to be deter-
mined from context.
With God willing, English is just as concise as Latin (more so, in fact, if one’s
counting syllables rather than words), but in other examples, Latin capitalizes
on the efficiency of its participles so as to produce very succinct turns of phrase.
An especially concentrated example of this is the ablative absolute mūtātīs
mūtandīs “having-been-changed what-needs-to-be-changed”. Here Latin
exploits to the fullest the fact that it has both perfect passive participles (“hav-
ing been changed”) and future passive participles (“going to be changed”) at its
disposal. The latter participle, marked by the -nd- suffix, is unusually common
in Latin, in part because it develops a more particularized usage as the so-called
gerundive: rather than signifying simply that which is going to be done, its
meaning extends to cover that which needs to be done or must be done. This

2  See also the discussion in Chapter 2 of Thucydides’ use of abstract nouns.

3  This, and further examples, may be found in J. K. Hale, Milton’s Languages: The Impact of
Multilingualism on Style (Cambridge, 1997), p. 111.
 Latin63

broadening of its usage may not seem especially predictable or natural, but we
have an expression of similar ambiguity in English. Consider first the sentence
If this is to be done, it should be done well. Here, is to be done could stand in for
a future passive: If this is going to be done, it should be done well. Perhaps it could
suggest that the action in question needs to be done, but it doesn’t have to do so.
(That is, there’s nothing ungrammatical about further specifying: We all agree
that this is an optional activity, but if it is to be done, it should be done well.) Now
contrast that first sentence with I’m not sure what is to be done. In this context,
what is to be done is not equivalent to what is going to be done, but is instead
closer to what ought to be done or what needs to be done. So too with mūtātīs
mūtandīs, the gerundive mūtandīs could be rendered simply “what is to be
changed”. (This particular gerundive also takes on a life of its own in Italian:
mutande, the feminine plural form, has become the word for “underwear”,
since, if nothing else, those clothes at least should be changed regularly.)
Another Latin gerundive that has found its way into English can also be
­translated neatly with “to be verb-ed”: QED, which stands for quod erat
dēmōnstrandum “which was to-be-proved”. At other times, the idea of obliga-
tion or necessity becomes more pronounced: an addendum is that which need­
ed to be added for completeness’ sake; since addenda are often already
incorporated into a document, the nuance of futurity is downplayed. Or, to
quote the words with which Cato the Elder is supposed to have closed all his
speeches, hammering home his implacable hatred of Rome’s enemy Carthage:
Carthāgō dēlenda est would be rather weakly translated as “Carthage is to be
destroyed”, and “Carthage must be destroyed” comes closer to the mark.
Of course, mūtātīs mūtandīs achieves its brevity not only because of the par­
ti­ciples but also because of the ease which with Latin can treat an adjective or
participle as a noun. That is to say, to put mūtandīs into English, one not only
needs to expand the verbal idea to to be changed but one also has to understand
along with this an implicit noun that it’s modifying: the things that are to be
changed. This is what’s called substantivization: taking an adjective or par­ti­
ciple, and treating it as a noun by giving it a generic default noun to modify, like
person or thing. English can do this as well, to be sure. It’s especially natural to
add a definite article and supply an understood noun “people”, such that one
can speak of “the poor” or “the young” as rough equivalents to “poor people”
and “young people”. Inanimate examples are also possible: philosophers can
search for “the good”, and an eccentric can have a taste for “the outlandish”. So
it’s just about conceivable that one could say something along the lines of “the
inadequate having been changed”, but, if one tries to combine substantivization
of this sort with the underlying participial nature of Latin mūtandīs, one ends
up with something like “the needing-to-be-changed having been changed”,
which really doesn’t count as English any more.
Participles aren’t the only possible second element in a Latin ablative abso-
lute: they can be replaced by an adjective, leading to phrases like cēterīs paribus
64 How dead languages work

“other things being equal”. Once again, Latin needs only half as many words
as  English. Here too, substantivization is partly responsible: English needs
the word “things”, but Latin can simply use cēterīs “other”, with the ending -īs
indicating that a plurality of things is in play. The other word Latin can dispense
with is “being”, as paribus means simply “equal”.4 Once again, it is the ending
that enables Latin to do without some sort of participle connecting the two
words. If, in English, one said simply “other equal” or, to capture the plural
“others equal”, there would be nothing to clarify either what role these words
played in the sentence as a whole, or that they belong together in a single
­syntactic unit. In Latin, by contrast, the fact that the -īs of cēterīs and the -ibus
of paribus are both ablative plural endings serves as a sort of index that says
both that they constitute a phrase in their own right and that the particular sort
of phrase that they constitute is an ablative absolute.
Or, at any rate, that it is likely to be an ablative absolute. Because there are, as
we’ve seen, other types of ablative as well, and, as we’ve also seen, it’s not always
clear exactly what sort one is dealing with. How does one tell? Partly through
contextual clues: if the word for some sort of tool is in the ablative, then there’s
a good chance that it’s an instrumental ablative. Similarly, if there’s a noun and
a participle together in the ablative, it’s likely to be an absolute construction.
But even though the ablative absolute is often presented as a discrete usage, the
line between it and other types of ablative is also blurry at times. (Indeed, this
is one reason why the ablative case is used for the ablative absolute in the first
place, rather than, say, the genitive or the dative.) So, for one final example,
consider the phrase vice versā. The first word, vice, is the ablative of a noun that
means, among other things, “place” or “position”, and versā is the perfect pas-
sive participle of vertere “to turn, change”. So one possibility would be to regard
this as a straightforward ablative absolute, equivalent to “the position having
been changed”, or, if expanded into a subordinate clause, “if the position has
been changed”, or the like. At the same time, it’s also possible that such a trans-
lation exaggerates the verbal component of the phrase, and that one would be
better off treating it as one of the other ablatives discussed, as “in or with a
changed position”. That might seem a little awkward in English, since there’s
some tendency to place a participle like changed after, rather than before its
noun, in a way that highlights its status as a predicate—that is, with the position
changed—getting us closer again to translation as an absolute construction. But
adjust the particular English words used to other potential equivalents—e.g. in
the reverse order—and the line between the ablative absolute and some sort of
locative-instrumental ablative becomes rather fuzzy again.
Indeed, it’s an especially fuzzy line in Latin if we recall the prominence of the
ab urbe conditā construction: if phrases like this can slide so easily from “from

4  Indeed, so little does Latin need a participle equivalent to English “being” that the classical
language does without it altogether.
 Latin65

the city having-been-founded” to “from the founding of the city”, it follows that
it would also be problematic to draw too neat a distinction between “with
the  position having-been-changed” (treating vice as a sort of instrumental
ablative that happens to be modified by a participle) to “with a change of
­position” (prioritizing the role played by the verb). But this is exactly what we
expect in language: meaning exists along a continuum, not in discrete chunks,
and the tools languages use to express that meaning can also shift seamlessly
from one usage to another.
So far, this chapter has been looking at ways in which Latin shows versatility
and conciseness that are difficult in English. But it began with a quotation from
Lucretius, in which the poet regrets that his task of bringing the Greek phil­oso­
phy of Epicurus to the Romans is rendered difficult by the poverty of the Latin
language. This is not the place to offer an exhaustive list of things Greek can do
that Latin can’t, but a couple of differences should be mentioned. First, Greek
has a richer panoply of participles at its disposal: in particular, most Latin verbs
have only an active participle in the present tense and a passive participle in the
past. This is similar to English, where a regular transitive verb—one that takes
an object and therefore has a passive voice—has only two simple participles: in
The audience loved the person playing the piano, the participle playing is a pre­
sent active participle (the noun it’s modifying, person, is performing the action,
so it’s active, and the action is taking place at the same time as the main verb, so
it’s present), but in The bar sold the piano played by the musician, played is a past
passive participle (the noun it’s modifying, piano, is what’s having the action of
the verb done to it, so it’s passive, and the playing of the piano took place before
the sale referenced by the main verb, so it’s past). But if one wants a present
passive or a past active participle, then English and Latin both have to reword
somehow, with English having constructions like being played and having played
to take care of those two options. Greek, however, has one-word p ­ articiples for
all four of these possibilities.
A second linguistic resource that gives Greek an expressive edge over Latin
is the definite article: virtually any word in Greek can be turned into a noun by
putting an article in front of it, as if English, rather than needing a specific noun
to refer to the present, could simply speak of “the now”. In particular, Greek gets
a lot of mileage out of putting definite articles in front of all sorts of infinitives
and participles, allowing philosophers, for instance, to distinguish between
“being” as an abstract verbal concept (by putting the article in front of the
infinitive, as if English could say “the to-be”), and “being” as that which is (if the
article is put in front of the present participle). A difference like this is as hard
to express in Latin as it is in English. It doesn’t help that Classical Latin, unlike
Greek, didn’t have a present active participle from the verb “to be”, even though
regular verbs do have such a form.
In any event, this is enough to give a sense of some of the grammatical
­shortcomings of Latin that Lucretius had to contend with in setting out Greek
66 How dead languages work

philosophy in Latin verse. But there were also deficiencies of vocabulary. By

Lucretius’ day, in the mid-first century bc, it had been three centuries since
Plato had been teaching in Athens; since then, a formidable technical language
for philosophy had developed, and Latin didn’t have ready-made equivalents
for all these words. Accordingly, a corresponding Latin vocabulary had to be
developed, in part simply through borrowing new words wholesale from
Greek—philosophia “philosophy” and poēta “poet”, for instance, which have
in turn been borrowed into English—but also through the use of calques.
A calque is a word that has been built out of elements that already exist in a
language, but whose meaning has shifted, or which are put together in a new
way, on the model of the analogous elements in another language. In this case,
Latin words develop new usages on the basis of their Greek counterparts, or
altogether new words are coined, through combining roots and affixes in the
same way as in Greek. The first of these two patterns is particularly easy to
detect in Christian vocabulary, where the range of the Greek source words had
often already been influenced by contact with Hebrew and Aramaic. In pagan
Greek, for instance, an angelos is simply a messenger. But because the Hebrew
word mal’āk referred to both ordinary human messengers and also the divine
sort—that is, the beings that we would call an angel in English—the range of
the Greek word was extended to refer to the latter category as well. From Greek,
the word was then simply borrowed as such into Latin as angelus and English
as angel. In this particular instance, Greek has a calque (the extended use of
angelos, patterned on Hebrew) whereas Latin simply borrows the Greek word
rather than, say, broadening the use of the native word for messenger, nūntius,
in this function. At other times, however, Latin also calques, and, in philo­soph­
ic­al language, one example is the word mundus. In its original sense, it referred
to ornaments and decorations such as jewelry. But the Greeks had taken the
corresponding word in their own language, kosmos, which had meant both
“ornament, decoration” and “(proper) order”, and extended it to refer to the
world order of the universe as well. (This is how both cosmetics and cosmos can
come from the same Greek root.) Similarly, mundus takes on this latter mean-
ing in Latin, as Cicero makes explicit in his translation of Plato’s Timaeus (§35).
From that point, it develops in its own right in Latin, passing from “universe”
to “world” to “earth”, and because the earthly can be opposed to the heavenly,
eventually gives us the more prosaic adjective mundane.
In the second type of calquing, a new word is created by matching up con-
stituent elements of existing words on a one-to-one basis with those in a word
in the source language. Thus, Greek had already taken the interrogative adjec-
tive poios “of what sort”, and combined it with the suffix -tēs “-ness”, to yield the
noun poiotēs “of-what-sort-ness”. This, of course, isn’t a well-formed English
word, but we’ll make do with it for now. Latin, for its part, also didn’t have a
word for this concept until the Romans decided, thanks to contact with the
Greeks, that they needed one too. In their case, however, since the Latin process
 Latin67

of word formation is a bit closer to Greek than the English one is, it was natural
enough for the Romans to take quālis “of what sort” and -tās “-ness”, to yield
quālitās—which was later borrowed into English as quality, which is how we
actually express this idea. Similarly, a Greek word like megathȳmos, built out
of the elements mega- “great” and thȳmos “spirit”, could be calqued as Latin
magnanimus, with magn- “great” and animus “spirit” taking the place of their
Greek counterparts.5 Once again, the word is borrowed directly into English, as
magnanimous.

LUCRETIUS

After this brief glimpse at some of the things that Latin can do—and that it
can’t—we’re ready to look at our first text, Lucretius’ Dē Rērum Nātūrā, usually
translated “On the Nature of Things”.6 It’s a lengthy poem (roughly half the
length of epics like the Iliad or Aeneid), written in the mid-first century bc
towards the end of the Roman Republic, in which the poet aims to convince his
Roman readership that, really, they’d all be much better off if they’d adopt the
philosophy of Epicurus as their guiding principle. Once they’d seen that the
world is nothing but atoms and void, and that we simply disintegrate after
death, and so have no need to fear the torments of Hell, they would be in a
better position to cultivate ataraxia—the Greek word for the calm, unruffled
composure that was the Epicurean ideal. Lucretius carefully structured the
poem into six books, which deal with phenomena of increasing size, starting
with atoms in the first two books, the human body and soul in the next two, and
ending with a discussion of the wider natural world in the final two. It’s a com-
prehensive undertaking, as suggested by the breadth of the title, which doesn’t
exactly exclude a whole lot.
But while that very title, Dē Rērum Nātūrā, is most straightforwardly trans-
lated as “On the Nature of Things”, the English obscures some of the resonances
of the two main words in the Latin title. (For once, I will ignore a preposition:
dē here does little more than indicate what the poem is “on” or “about”; its
object must stand in the ablative case, hence the long final a in nātūrā.) Rērum

5  I have taken the quālitās example from L. R. Palmer, The Latin Language (London, 1954),
p. 129, and the magnanimus example from M. Fruyt, “Latin vocabulary”, in A Companion to the
Latin Language, edited by J. P. T. Clackson (Malden, MA, 2011), p. 152. As it happens, Cicero
claims credit for coining quālitās in particular (Academica 1.25).
6  Lucretius has become a little more prominent again in recent years thanks in part to
S. Greenblatt, The Swerve: How the World Became Modern (New York, 2011), which tells the story
of the rediscovery of Lucretius’ poem in the Renaissance. Readers wanting a more thorough intro-
duction to his work may be directed to S. Gillespie and P. Hardie, eds., The Cambridge Companion
to Lucretius (Cambridge, 2007).
68 How dead languages work

is the genitive plural of rēs, the Latin word for “thing”, and, since the genitive,
as  a case, is most often equivalent to of in English, “of things” is a natural
way  to  render it. But Latin teachers rightly tell their students to be wary of
mechanically translating rēs as “thing”: it may be the basic English counterpart,
but whenever a word is as broad in scope as rēs or thing, it’s unlikely that any
one translation will be best in every situation. Our English word republic,
for instance, comes from the Latin phrase rēs pūblica, and while a republic is
indeed the archetypal “public thing”, it doesn’t make for as elegant a translation
of the Latin as “the public interest”, “the public affair”, or, as it has traditionally
been rendered, “the commonwealth”. Another prominent occurrence of the
word in Latin is in the title of the famous inscription in which the emperor
Augustus listed his accomplishments, the Rēs Gestae Dīvī Augustī “The Things
Done of the Divine Augustus”. Here it serves as little more than a syntactic hook
on which to hang the participle gestae “done”, and rēs gestae together hardly
means more than just “deeds”. Indeed, in the contemporary Greek translation
of the inscription, the equivalent word is simply the noun praxeis “deeds”.
As for the title of Lucretius’ poem, it makes use of yet another meaning of rēs,
in which the plural of the noun is understood to refer not just to “things”, but to
“all things”, that is, “the universe”.
Even more complex is the network of meanings associated with the Latin
word nātūra, borrowed into English as nature with a similarly wide range of
usages, from one’s inborn character to the natural world seen as a whole. It’s no
wonder that it’s the semantic field around this word that C. S. Lewis chose for the
first main chapter in his masterful Studies in Words. First, take the etymology of
the Latin word. It’s built to a root nā- (simplified from an earlier gnā-) “to be
born”, seen also in English derivatives na-tal “related to birth”, na-tive “by birth”,
and co-gna-te “of shared birth”. The suffix -tūra turns the root into a noun that
will originally have meant simply “birth”, but this meaning is very rare in attested
Latin, since the extended sense “quality or character since birth” quickly took
off. That this particular meaning became so common is due in large part to the
process of calquing. For the range of meanings that nātūra took on in Latin, the
Greeks had used the word physis, which, like nātūra, is formed from the com­
bin­ation of a root meaning “to be born”, phy-, with a suffix, -sis, used to form
nouns. When first-century bc philosophical writers in Latin gave their works
titles with the word nātūra—not just Lucretius’ Dē rērum nātūrā but also Cicero’s
prose dialogue Dē nātūrā deōrum (“On the Nature of the Gods”)—it was in
conscious imitation of their Greek predecessors: the major work of Lucretius’
philosophical hero Epicurus was a 37-book treatise Peri physeōs (“On Nature”),
and Empedocles, a fifth-century bc philosopher-poet from Greek Sicily whom
Lucretius took as a poetic model, composed a poem with that same title.
The double nature of Latin nātūra—that it refers not only to the static condi-
tion of things but, more specifically, to their birth or coming into being—is
seen especially clearly in the first occurrence of the phrase rērum nātūra in
 Latin69

Lucretius’ poem. In the first 20 lines, the poet has been praising Venus, the
goddess of love, as a generative force that gives rise to life, and justifies his
choice of her as the divine addressee of the poem as follows:
quae quoniam rērum nātūram sōla gubernās
Because you alone govern the nature of things
nec sine tē quicquam dīās in lūminis ōrās
Nor, without you, does anything, into the bright bounds of light,
exoritur neque fit laetum neque amābile quicquam,
Arise, nor does anything become fertile or lovely,
tē sociam studeō scrībendīs versibus esse,
I am eager for you to be a companion for the writing of the verses,
quōs ego dē rērum nātūrā pangere cōnor
Which I am attempting to compose on the nature of things
In these lines (1.21–5), Lucretius slides easily from one sense of nātūra to another:
since the previous lines have just detailed Venus’ role in the blooming of fl ­ owers
and the breeding of animals, the derivation of nātūra from the “to be born” root
comes to the fore in the first line. But, in the second instance of the phrase,
when he gives rērum nātūra as the subject of his poem, a broader meaning
seems to be in play. In any case, the work is certainly not just about the begin-
nings of life, and the sixth and final book famously concludes with a dismal
description of a great plague ending all sorts of lives, human and animal alike.
Many other words in these lines lose much of their resonance in translation
as well. The word that I’ve translated “bright”, dīās, is the Latin counterpart of a
word we saw in the chapter on Greek: dīos, the epithet of Achilles that means
either “godlike” or “brilliant”. Here too there’s ambiguity as to whether Lucretius
is characterizing the region of light more as “divine” or as “dazzling”—or, most
likely, a combination of both. One question that has long troubled commenta-
tors on the Dē Rērum Nātūrā is the prominent part given to Venus in the
­prologue: the Epicureans didn’t believe that the gods played an active role in
human affairs, so Lucretius comes across as having abandoned right at the very
start of the poem the philosophical principles he supposedly professes so
ardently. Later on, he suggests that it’s perfectly fine to speak of the gods in this
way, as long as one is aware that it’s just a poetic conceit. This might not seem a
good enough reason on its own to make Venus so active a participant in the
prologue, but it does help matters that even the word dīās acts to blur the line
between the natural and the supernatural: does it mean “bright” or “divine”? Even
the shape of the word, too, is significant: in most authors (and, for that matter,
elsewhere in Lucretius), one would find not dīās, but its near-equivalent dīvās,
the adjective that modifies Augustus’ name in the Rēs Gestae Dīvī Augustī. The
alternative without the v makes it look a little more like its Greek counterpart,
establishing yet another subtle link between Lucretius’ poem and the Greek
70 How dead languages work

tradition onto which it has been grafted: not only are the philosophical ideas
those of Epicurus but also the metrical form of the poem, dactylic hexameter,
is the same as that used by Homer. Another adjective worth singling out for
its double meaning is laetum, which I have translated “fertile”. More often, it
means “joyful, happy”, but the idea of abundance is foregrounded when it is
used in agricultural contexts, prosperity being the shared common element
between fertility and joy. But, once again, the fact that both senses are poten-
tially in play enables Lucretius to obscure the boundary between natural phe-
nomena pertaining to the agricultural world and the emotional experiences
associated with the goddess Venus—note that laetum is paired with amābile
“lovely”—at the divine level of the prologue.
With scrībendīs versibus, we can see in action one of the syntactic features of
Latin mentioned above, the gerundive. While I’ve translated it “for the writing
of the verses”, a stricter, if misleading, rendering would be “for the verses to be
written”, as scrībendīs is the gerundive of scrībere “to write”. But as seen in the
ab urbe conditā construction, it is the Latin way to take the verbal idea that
predominates in idiomatic English (in which “the writing” is syntactically more
prominent, with “of the verses” dependent on it), and subordinate it to the
noun that is its object. (That is, relative to English, versibus “for the verses” rises
higher in the syntactic hierarchy, with scrībendīs “to be written” relegated to
the  status of a dependent.) But why should this interest anyone other than
­linguists? How does it affect Lucretius as poetry? Because translating the phrase
as “I am eager for you to be a companion for the writing of the verses” flattens
out with an abstraction what in Latin is more concrete and immediate. In the
English rewording, the act of writing becomes introduced as an intermediary
between Venus, whom the poet seeks as a companion, and the verses them-
selves. In Latin, by contrast, Lucretius calls on the goddess to be a companion
directly to the verses. In this way, a sort of physical immediacy is suggested,
which is relevant to broader issues of materialism raised in the poem. Lucretius,
following Epicurus, reduces everything in the natural world to atoms and void,
and, while I wouldn’t want to insist that he deliberately uses the gerundive to
achieve this poetic end—it is, after all, an unmarked expression in Latin—there
are many places in the work where the concrete imagery employed by the poet
not only embellishes the abstract philosophical argument of the poem but
actually itself embodies that argument.7 This is especially evident in Lucretius’
love of wordplay: just as the world consists of atoms that, when combined
in  different arrangements, create different objects, so too language consists
of  elements that, when rearranged, create different words. This analogy is
­especially easy in Latin, since the word elementum signifies both the building
blocks of the physical universe and the letters of the alphabet.

7  This point is brought out especially well by D. West, The Imagery and Poetry of Lucretius
(Edinburgh, 1969).
 Latin71

In the final line of the passage above, the verb pangere “compose” exempli-
fies nicely Lucretius’ habit of using language at two different levels at once.
On the one hand, it’s a comparatively common word for the writing or com-
posing of poetry. But its original, physical meaning was “to fasten, fix in
place”. (It is from a different development of this sense, “to settle, agree upon”,
that the related word pact “that which is agreed upon” derives.) Now one of
the most notorious features of Epicurus’ philosophy, as conveyed to us by
Lucretius, is the idea of the atomic swerve: unless one allows that atoms occa-
sionally undergo unpredictable motion, then it’s hard to find room for free
will.8 Since Lucretius doesn’t want people simply to be machines whose every
move is predetermined by the physical configuration of their atoms at the
start of their life, the swerve is introduced as a way of allowing for the unfore-
seeable. In a world, then, which consists of atoms mostly behaving according
to the laws of physics, but occasionally deviating from their expected path, it
works well for Lucretius to characterize the act of poetic composition with
the verb pangere: the elements of words, too, can swerve, and he is trying to
fasten them in place.
Perhaps the most famous instance of wordplay in Lucretius comes a little
later in Book One, when he is praising Epicurus for encouraging men to resist
what he regards as the greatest evil in life: a religious superstition that keeps
humankind downcast in anxious fear (1.62–7):
hūmāna ante oculōs foedē cum vīta iacēret
When human life lay foully before our eyes
in terrīs oppressa gravī sub religiōne9
Pressed down on the earth beneath heavy Religion
quae caput a caelī regiōnibus ostendēbat
Who showed forth her head from the regions of the sky
horribilī super aspectū mortālibus īnstāns
Standing over mortals with a terrible appearance
prīmum Grāius homō mortālīs tollere contrā
A Greek human first, in opposition, to raise mortal
est oculōs ausus prīmusque obsistere contrā
Eyes did dare, and first to take a stand, in opposition
The first word we must attend to is religiōne, the ablative (indicating location,
with the preposition sub “beneath”) of religiō, a word that Lucretius elsewhere in
the poem10 connects with ligāre “to bind, tie (fast)”. One could write an entire

8  Stephen Greenblatt’s book on Lucretius takes its title from this phenomenon as well.
9  While religiō normally has a short e, its first syllable regularly scans long in Lucretius.
10  artīs | religiōnum animum nōdīs exsolvere “to loosen the mind from the tight knots of
­superstitious feelings” (1.931–2).
72 How dead languages work

book on the wealth of meaning packed into just this one word, but for our pur-
poses it’s enough to say that its semantic center of gravity, in contrast to English
religion, lies not so much with the institution of organized religion as with the
internal inhibitions and scruples that arise from religious belief; as such, trans­
lators of Lucretius often render it as “superstition”—although this is perhaps
misleading, since in most authors, religiō is a more positive word than that, and
Lucretius comes across as less provocative if he’s condemning superstition than
if he’s attacking religion. Still, “superstition” is an understandable word to reach
for in this passage, given the way in which religiō is personified: she has a head
(caput), and stands over (super . . . īnstāns) mortals. The latter phrase, with super
as a preposition-like adverb meaning “above, over”, and īnstāns, a participle
formed from the root stā- “to stand”, combined with the prefix in-, which further
emphasizes the position of religiō over mortals, contains precisely the same
elements that are found in the word superstitiō. Not unlike its English deriva-
tive superstition, the Latin word has negative connotations, suggesting unrea-
sonable or excessive fear of the gods. In short, Lucretius here takes a generally
positive word for religious belief, religiō, but personifies it as a monster that, by
standing over mortals, enacts superstitiō, the negative counterpart of religiō,
effectively blurring the lines between the two, and calling on readers to question
whether even ostensibly good religiō is not in fact superstitiō after all.
But the wordplay in Lucretius is not limited to the double meaning of a phrase
like super īnstāns; it extends down to the level of individual letters. As a poet who
compares letters to atoms, he is fond of showing how the rearranging of those
letters can cause changes of meaning analogous to transformations in the phys­ic­
al world. Thus, in the third line of this passage, the elements of religiō, which has
just been named, recur, scrambled, in caelī regiōnibus “regions of the sky”, thereby
highlighting through linguistic legerdemain the connection between religious
superstition and the celestial phenomena, like lightning, that were so closely
associated with the gods in antiquity. (Much of the final book of the poem is
devoted to showing how thunder, earthquakes, volcanoes, and the like, are more
plausibly explained as produced by the motions of atoms than by Jupiter, Neptune,
and Vulcan, respectively. Such arguments go a long way towards giving the Dē
Rērum Nātūrā its modern feel.) Another fine example of such atom-level
wordplay occurs in Book Four, when Lucretius attacks the madness of love in a
unrelentingly bitter screed—so bitter a screed, in fact, that four hundred odd
years later, Jerome, translator of the Latin Vulgate Bible, thought he’d been driven
mad by a love potion. In this attack, Lucretius adheres to his usual materialist
view and reduces love to a simple question of fluid dynamics: in order to extin-
guish a fire, fluid from one body is drawn into another body. This is, poetically, an
easier equation to make when, as David West points out, the word for love, amor,
only differs from that for fluid, ūmor, by a single letter.11

11  D. West, The Imagery and Poetry of Lucretius (Edinburgh, 1969), pp. 94–5.
 Latin73

One final aspect of the personification worth mentioning is, in fact, the word
aspectū “appearance”. It is a sad fact that many of the Latin words English has
borrowed have been rather abused, thanks to historical happenstance. There
are many lexical doublets in English, where a native Germanic word competes
with one of Latin origin—borrowed either directly from Latin or by way of
French, thanks especially to the Norman Conquest. Indeed, sometimes there
are triplets:
1. kingly, from Germanic king (compare German König, seen in the city
name Königsberg “King’s Mountain”)
2. regal, from Latin rēx (whose stem is rēg-)
3. royal, from French roi (the natural development of Latin rēx in French;
the final consonant wasn’t stable in the other Romance languages either:
compare Spanish rey or Italian re)
The most famous discussion of the parallel native and French vocabularies
of English is no doubt that found in the first chapter of Walter Scott’s Ivanhoe,
in which the jester Wamba points out that animals keep their native Germanic
names when tended by the Anglo-Saxon herdsmen, but are called by their French
names when served as food to their Norman overlords: the swine (German
Schwein) becomes pork (French porc), the ox (German Ochse) becomes beef
(French bœuf), the calf (German Kalb) becomes veal (French veau). The rele-
vant sociolinguistic point illustrated here is that, when a language acquires
doublets like this, they often specialize in particular spheres of language use.
As Latin was associated with the prestige of higher learning, words borrowed
from it often came across as more impressive, and clustered in comparatively
grandiose contexts. Unfortunately, this has led to a tendency on the part of
some to inflate their language by using Latin-derived words when simple
English ones would do just as well. Now such is the extent to which Latin
vocabulary has infiltrated English that it’s virtually impossible to do without it.
In this paragraph alone, doubt, use, serve, and language are all such natural
English words now that it would be hard to come up with native Germanic
equivalents that would somehow be simpler. But it’s still a common rookie
­mistake for students to dress up their writing by replacing shorter, native words,
with longer, Latinate ones: take becomes acquire, go becomes proceed, talk
becomes converse. Even worse, this sort of language bloating has mushroomed
in bureaucratese as well, where words like excellence and engagement are ban-
died about without any real substance behind them. Now this is obviously not
to say that such borrowed words have no place in English. As an exercise for
anyone who’s gotten tired of reading words of this register misused in student
essays or administrative blather, I heartily recommend searching the corpus of
a first-class writer like Shakespeare or Jefferson for the same, for, sure enough,
they find ways of making them work. But, to get back to Lucretius, a word like
74 How dead languages work

aspect belongs very much to this sort of diction, and, when an English speaker
sees it in Latin, it’s unfairly tarnished by the lexical company it keeps in our own
language.
That is to say, it is very easy for a Latin student, on coming across a word like
aspectū, to observe (rightly) that is the source of our word aspect and, if it’s
a weaker student, simply translate it as aspect (“standing over mortals with a
horrible aspect”), thus leading to translationese that, because it incorporates a
word that can be used vaguely as a synonym of “feature” or “trait”, sim­ul­tan­
eous­ly sounds like the bland prose of officialdom. Now stronger students, who
will be aware that there have been many shifts in meaning between Latin source
words and their English derivatives, will look in their dictionary, and translate
it with “appearance” or “look”, which are closer to the sense here. But even so,
it’s hard, in the aesthetic appreciation of an English speaker, for the Latin word
not to lose a little bit of the poetry it ought to have had, because of this frequent
association of the Latin words in English with a very non-poetic register of
language. And in a passage like this, it would be especially regrettable for the
abstract connotations of aspect to get in the way of Lucretius’ imagery. This is
a passage, after all, in which vision is very prominent: not only does the per-
sonification of Religion cause the reader to picture her looming over men but,
in the first line, human life is said to lie ante oculōs—literally just “before eyes”,
but one can easily supply a possessive like “before our eyes” or “before every-
one’s eyes”—and, in the last line, our hero Epicurus has dared to raise up his
oculōs against this adversary. So when Lucretius speaks of Religion’s horribilī
aspectū, the etymology of the latter word, derived as it is from the root spec- “to
see”, contributes to the concreteness of Lucretius’ description.
This is just one example, however, of the constant struggle it takes for anglo-
phones to defamiliarize individual words when reading Latin poetry. In the
second line, human life is said to be oppressa: it’s very easy simply to read this as
“oppressed”, and indeed it’s not wrong to see that connotation in the Latin word
here. But if we translate it that way, then once again we lose the physical image
of Religion literally weighing people down, crushing them, which is the basic
sense of this verb. (Note that Lucretius has taken care to describe Religion as
gravī “heavy”.) Oppressa is also the first of three verbs in this passage that are
linked through the use of the prefix ob- (which changes into op- before the p- of
pressa, and os- before the t- of tendere, but remains ob­- before the s- of sistere).
Like many prefixes that are the counterparts of prepositions, it has a wide range
of uses, but for the most part the common denominator is that ob- indicates
that an action occurs “face to face”—with the frequent additional implication,
as in the expression face-off in English, that such actions involve a degree of
hostility, with the agents working against each other. It is no coincidence that
this is the first element in the words op‑ponent and op‑posite. By repeating this
prefix three times in this section—with Religion pressing against (oppressa)
mortals, showing forth her head face to face (ostendēbat), and Epicurus taking
 Latin75

a stand against her (obsistere)—a common thread of physical confrontation is

brought out that is lost in translations where the verbs are not linked together
by this shared element.
One final word that is not as close in feel to its English derivative as it appears
is mortālis, here used both as a noun mortālibus, in reference to the mortals
over whom Religion stands, and as an adjective mortālis or mortālīs (the manu-
script tradition is ambiguous, since vowel length is not marked by macrons in
the original text), which (if the former) would be modifying Epicurus, “a Greek
human, a mortal one”, or (if the latter) the “mortal eyes” he raises against
Religion. Either way, both times, it is perfectly sensible to translate it as “mor-
tal” in English, which, like Latin mortālis, means “subject to death”. And yet
there is a difference: since English mortal is a borrowing from Latin, it doesn’t
actually have the word death in it; Latin mortālis, by contrast, is formed from
mors (stem: mort-), the blunt, direct Latin word for death. When we hear the
word mortal, the image of death doesn’t necessarily spring to mind as happens
when something is described as deadly or deathly. Rather, we hear the op­pos­
ition to immortal—an opposition that Latin frequently takes advantage of as
well—and perhaps medical discussion of mortality. In either case, instead of
the cold, stark reality of death conjured up by the Germanic word, we hear it
mediated through the detachment of a learned Latin word. But Lucretius, here,
is not detached. He goes on to describe death in grim detail at the end of the
poem, as, it is often thought, a sort of test for his readers: if they’ve properly
imbibed the Epicurean message of the poem, they won’t be perturbed by the
listing of the symptoms of the plague—the phlegm, the bleeding, the ulcers, the
cold sweat, the limbs lost to gangrene. English simply doesn’t have a word that
refers to humans, as a class, opposed in particular to the gods, with the word
“death” right at the heart of it. Latin does, so when Lucretius wants to decry
human superstition about the gods as something that ultimately arises from the
fear of death, in the word mortālis he has a lexical resource at his disposal that
no English translator can match.
But the difficulties of translating Latin poetry go beyond the impossibility of
capturing the resonance of individual words. The structure of the language also
gets in the way of turning it into English. In this last example, as noted, it isn’t
clear in a standard, macron-free Latin text whether mortalis modifies homo or
oculos. This ambiguity arises from Latin’s rich inflectional morphology. Latin
can, for the most part, indicate the grammatical role of a word in a sentence by
its ending, which means, as already seen, that it doesn’t have to rely on word
order, as English does, to clarify, for instance, whether a noun is the subject or
the object of a verb. Since the endings of the words serve under most circum-
stances to link adjectives with the nouns they modify, the former are generally
free to drift away from their noun without any loss of clarity. An English teacher
can reprimand a student for a dangling participle in a sentence like Throwing
the stick, the dog watched the boy, as the participle throwing is closer to the dog
76 How dead languages work

than it is to the boy who, we assume, is actually throwing the stick. In Latin,
however, that participle can change form to indicate what it’s modifying: no
matter where it was in the sentence, iaciēns would indicate that the subject of
the verb was doing the throwing, whereas iacientem would signal that it was
that object that did so. Very often, this ability to index a participle or adjective
to its noun through the use of the ending allows for considerable flexibility in
word order, without subsequent loss of clarity. In the first two lines of the
­quotation above, for instance, it is unambiguous that hūmāna “human” modi-
fies vīta “life”, even though the four words ante oculōs foedē cum “before (our)
eyes foully when” intervene. By moving hūmāna so far to the front of the sen-
tence, Lucretius can emphasize its importance to what follows in a way that
cannot be imitated directly in English.12 It is equally unambiguous in the Latin
that oppressa also modifies vīta, even across a line break. This is precisely the
sort of situation in which the danger of a dangling participle lurks in English:
cum vīta iacēret in terrīs oppressa sub religiōne is, word for word, “when life lay on
earth crushed beneath religion”. Leave it like this in English, and it’s unclear
whether it’s life or the earth that’s crushed, but oppressa in Latin can only ­modify
vīta; if it were the earth that was crushed, the form would need to be oppressīs.13
Now, in that last phrase, I left out the adjective gravī “heavy” from the original
text, gravī sub religiōne, for the sake of simplicity in the English, but, in the
Latin, its position in the sentence is unremarkable: we can only say “beneath
heavy religion”, but Latin prefers the equivalent of “heavy beneath religion”.
That the adjective gravī can be separated from religiōne by the preposition sub
“beneath”, or, in close succession, horribilī “terrible” from aspectū “appearance”
by super “over”, suggests how ordinary such discontinuity is in Latin. Indeed,
it’s evident even in a couple of phrases familiar to English speakers: magnā cum
laude and summā cum laude both reflect the regularity with which Latin slips a
preposition in between an adjective and its noun. The expected Latin order is
thus not “with great praise” and “with highest praise”, but “great with praise”
and “highest with praise”. Use the latter word order in English, and it will be
assumed that the adjectives modified the person who is the subject of the
phrase. But in Latin, the ending links them with the noun laude.
Or at least that is the case in spoken Latin. While the system of rules for
spelling standard Latin—its orthography—is fairly sensible on the whole, with
something close to a one-to-one correspondence between the letters of its
alphabet and the sounds they represent, one important shortcoming concerns
the vowels. Vowel length in Latin is contrastive: that is, in certain words,

12  To anticipate a technical term introduced below, placing hūmāna this early establishes it as
a topic for what follows.
13  If this were a prose text without macrons to indicate vowel length, then oppressa could in
theory be modifying religiōne. But the meter of the poem makes clear that this is oppressa with a
short a, not with a long a, and so must be nominative, not ablative, and therefore in agreement
with vīta, not religiōne. For more on the ambiguity of vowel length in Latin, see below.
 Latin77

whether or not a vowel is long can change the meaning of the word. Thus,
mălum with a short a meant “bad”, but mālum with a long a meant “apple”.14
And, while I’ve been supplying macrons all along to indicate vowel length
(as  linguists often do), they are not in fact found in the standard written
­language. Now you may well ask how we know that they were pronounced
differently if this wasn’t written down in ancient texts. Partly because in certain
places in a line of poetry, only a long or a short vowel would fit the meter, partly
because most pairs of long and short vowels develop differently in the Romance
languages, partly because a few ancient inscriptions do indicate vowel length
with a diacritical mark, and partly because ancient grammarians tell us that
vowel quantity mattered. Be that as it may, in purely written documents, there
was potentially scope for confusion, and the identical spelling of malum “bad”
and malum “apple” is what led the fruit that Eve gave to Adam to be identified
in the Middle Ages as the apple. (It had been left unspecified in Hebrew.)
Unfortunately for those trying to read Latin, some of the endings that indi-
cate what adjectives belong with what nouns rely precisely on vowel length as a
distinguishing feature. So the most characteristic feminine nominative singu-
lar ending, used of the subject of a sentence, is -ă, whereas the corresponding
ablative ending is -ā. Thus, without any context, a phrase like summa cum laude
could in theory be understood (correctly) as summā cum laude, with the long
vowel at the end of summā signaling that it is to be understood with laude, or
summă cum laude, with the short vowel marking it as a subject, and therefore
to be taken as modifying an unspecified woman, rather than the praise.15 This
is a somewhat contrived example, but an actual ambiguity can be found in the
last two lines of the snippet of Lucretius: prīmum Grāius homō mortālis/mortālīs
tollere contrā | est oculōs ausus prīmusque obsistere contrā (“A [mortal?] Greek
human first dared in opposition to raise [mortal?] eyes, and to take a stand in
opposition”). Now, at least on the page, mortalis is ambiguous, as it could either
modify homō “human” or oculōs “eyes”. But if we could hear Lucretius reciting
this passage, we could instantly tell which it was: if he pronounced it mortālĭs,
then it would have to be a nominative singular adjective, and therefore in
agreement with homō; but if he drew out the last syllable to mortālīs, then it
would need to be accusative plural, and therefore modify oculōs. In the absence,
however, of any recordings from the first century bc, we have to rely on other
sorts of reasoning to determine which makes better sense. In favor of taking it
with homō is the fact that it occurs right after it. But while that may seem a
strong argument to an English speaker, it’s actually rather weak when dealing
with Latin poetry, since the examples of separation of adjective from noun are

14  As an anonymous reviewer has pointed out, this particular near-homonym makes it into the
libretto of Benjamin Britten’s Turn of the Screw, in which “Mālō1 mālō2 mălō3 mălō4” is glossed as
“I-would-rather-be1 in-an-apple-tree2 than-a-naughty-boy3 in-adversity4”.
15  Yet another possibility would be to read summa as a different word altogether, the derivative
noun meaning “chief point” or “sum”.
78 How dead languages work

so common. Two other considerations suggest that it is better understood with

oculōs. First, there’s the question of what works better from a poetic standpoint.
It’s somewhat otiose for homō, which already means “human”, to be modified
by an adjective that is often used with the same denotation. It’s less tautological
if it’s the eyes that are described as mortal. The second point is a stylistic one.
While Latin poetry is perfectly happy for an adjective not to be next to its noun,
what it doesn’t like is for there to be an uneven distribution of adjectives relative
to nouns: if there are two adjectives in a clause, and two nouns, then it’s much
preferred for there to be one adjective for each noun than for one noun to be
decked out with two adjectives, and the other to be left bare. Since homō here is
already modified by Grāius “Greek”, taking mortālīs with oculōs means that
each noun gets one and only one adjective.
Lucretius stands comparatively early in the Latin poetic tradition, at least as
it has been transmitted to us. While there had been written Latin poetry for
nearly two hundred years by his time, Lucretius comes before nearly all the
other names that loom largest in the canon of Latin verse: Catullus was his
contemporary, but the great Augustan poets Virgil, Horace, Propertius, and
Ovid all follow. He himself was aware of the difficulties of turning Greek phil­
oso­phy into Latin poetry, and occasionally the writing is clunky, especially in
the presentation of some of the drier arguments. Coleridge was unfair when he
remarked to Wordsworth, “Whatever in Lucretius is poetry is not philosophical,
whatever is philosophical is not poetry,” but one can at times see where he was
coming from, what with the profusion of transitional phrases like praetereā
quoniam “furthermore, since . . .” or tum porrō “then, moreover . . .” and the
somewhat unrelenting Epicurus-worship. But even if critical judgments that
took later Augustan models as the pinnacle of poetic practice could carp about
Lucretius’ lack of polish, the skillful wordplay and arrangement of words in
the lines just considered—linguistic craftsmanship that is virtually impossible16 to
capture in translation—is more than enough to justify a prominent position for
him in the Roman canon.

HORACE AND HOUSMAN

For our next sample of Latin, we move forward in time from the last years of the
Roman Republic (we don’t know exactly when Lucretius was writing, but
sometime in the 60s or early 50s bc) to the early years of the Empire—13 bc, to

16  This doesn’t mean that people haven’t tried. E. J. Kenney has recently noted the continued
appearance of new translations of Lucretius as a sign of lively interest in the poem (Lucretius: De
Rerum Natura Book III, 2nd ed. (Cambridge, 2014), p. xi): Ronald Melville’s translation, later
reissued with Oxford World’s Classics, first appeared in 1997, A. E. Stallings’s Penguin translation
in 2007, and David R. Slavitt’s for the University of California Press in 2008.
 Latin79

be precise, when Horace published his fourth book of Odes. Only half a century
had passed, but the Roman world looked very different: the ructions of the
great civil war were over, and Augustus was now in charge. In the (compara-
tively) calm period of his rule, poetry flourished: this is the time not just of
Horace but also of Virgil’s Aeneid and Ovid’s Metamorphoses. Rather than
­tackling one of these great epic poems, however, we’ll focus on a poem that the
great classical scholar, A. E. Housman, better known to most laypeople as the
poet of A Shropshire Lad, regarded as “the most beautiful in ancient literature”,
Odes 4.7. Here are the first four stanzas of the Latin, side by side with Housman’s
translation:
Diffūgēre nivēs, redeunt iam grāmina campīs The snows are fled away, leaves on the shaws
arboribusque comae; And grasses in the mead renew their birth,
mūtat terra vicēs, et dēcrēscentia rīpās The river to the river-bed withdraws,
flūmina praetereunt; And altered is the fashion of the earth.

Grātia cum Nymphīs geminīsque sorōribus audet The Nymphs and Graces three put off their fear
dūcere nūda chorōs. And unapparelled in the woodland play.
immortālia nē spērēs, monet annus et almum The swift hour and the brief prime of the year
quae rapit hōra diem. Say to the soul, Thou wast not born for aye.

frīgora mītēscunt Zephyrīs, vēr prōterit aestās Thaw follows frost; hard on the heel of spring
interitūra, simul Treads summer sure to die, for hard on hers
pōmifer autumnus frūgēs effūderit, et mox Comes autumn, with his apples scattering;
brūma recurrit iners. Then back to wintertide, when nothing stirs.

damna tamen celerēs reparant caelestia lūnae: But oh, whate’er the sky-led seasons mar,
nōs ubi dēcidimus Moon upon moon rebuilds it with her beams:
quō pater Aenēās, quō dīves Tullus et Ancus, Come we where Tullus and where Ancus are,
pulvis et umbra sumus. And good Aeneas, we are dust and dreams.

Even without looking at individual words, some differences between the ori­
ginal and the translation are clear. Housman needs a lot more words than
Horace does to get the same ideas across, and the line lengths are different:
while Housman’s lines are all iambic pentameters, Horace’s alternate between
dactylic hexameters—the same meter used by Homer, Lucretius, and Virgil—
and shorter lines that are also dactylic, but only extend for two and a half feet
rather than six.17 In effect, what Horace can express in a line and a half fills up
two full lines of Housman. Partly this is the inevitable consequence of translat-
ing poetry, a form of language in which every word bears special weight, such
that meaning is often unusually compressed. Moreover, the content of the
original somehow needs to be adapted to the rhyme scheme and meter of the
target language, and, when there’s not an easy fit, the natural temptation is to
pad the translation with additional words rather than to leave anything out

17  This particular pattern is rare in antiquity, although, if the even-numbered lines consisted of
two of these half-lines each, rather than just one, the poem would then consist of elegiac couplets,
which are very common.
80 How dead languages work

that’s in the original. But density of expression is characteristic of English

poetry, too, and much of the expansion in the appearance of Housman’s text on
the page is thus best attributed to the general temperament of the two lan-
guages. How all of this plays out in practice can be seen neatly in the first two
lines of the poem. First, Diffūgēre nivēs becomes The snows are fled away. While
a straight word count—two words have expanded to five—might suggest a
degree of padding, this apparent bloating results entirely from the more ana­­
lytic­al structure of English. Nivēs turns into the snows because English requires
a definite article, and diffūgēre becomes are fled away because Latin fuses into
one word elements that English expresses individually: dif- (an assimilated
form of the prefix dis‑) corresponds to away, -fūg- to fled, and the ending -ēre,
which indicates person, number, and tense, to are.18 Indeed, it’s hardly clear
that counting words is the best way to assess conciseness of expression: English
may take more words, but they’re also shorter than the Latin, and the total
number of syllables, six, is the same in both.
Comparing the two versions becomes more complicated in the next line and
a half, since the translator alters the order of the ideas expressed in the Latin.
For orientation, here is (a) a word-for-word glossing of the Latin, (b) a more
polished version that preserves the word order, and (c) Housman’s translation:

  redeunt iam grāmina campīs arboribus-que comae

(a) they return already grasses to/in fields to/in trees-and leaves
(b) They’re returning already, the grasses to the fields, and to the trees,
their leaves
(c) Leaves on the shaws and grasses in the mead renew their birth

Once again, Housman’s English has more words: six Latin words have become
12 in English. Once again, judging conciseness by the word count is misleading:
the translator has actually reduced the number of syllables, from 16 to 14. Some
of the shifts parallel those seen with the first two words of the poem: grāmina
campīs, for instance, becomes grasses in the mead, with additional words—but
not syllables—needed in English because the definite article comes into play
and the idea expressed by the case ending -īs in campīs requires a preposition
like in in English. But which preposition is one to choose? With both campīs
“fields” and arboribus “trees”, the case ending is ambiguous, since, with plural
nouns in Latin, the dative and ablative cases are always identical. If these are
understood as datives, then the fields and trees are regarded as the beneficiaries

18  Note also that Housman chooses are rather than have to form the present perfect. This used
to be a possibility in English when the verb in question was intransitive, much as in German,
where intransitive verbs of motion or change in state take sein (“to be”) rather than haben (“to
have”) as the corresponding auxiliary. It works very well for Housman to use the older construc-
tion here insofar as the Latin ending -ēre is also an archaizing poeticism, in this case an earlier
form of the third-person plural perfect active ending, generally replaced by -ērunt in prose.
 Latin81

of the renewal, and the grasses and leaves viewed as their possessions, an
­interpretation best conveyed by “to” in English. But they could equally well be
ablatives, and, of the three main uses of the ablative, the locative fits best here,
and is Housman’s choice; thus, “on the shaws” and “in the mead”.
But why pick “shaws” to translate arboribus, which is the plain Latin word for
tree? Or “mead” for campīs, which simply means “field”? Because Housman is
working in an English poetic tradition that is set apart from the ordinary lan-
guage of prose in part through the use of distinctive vocabulary. Moreover,
English is no different from Latin in this respect. While arbor and campus
(as these words would occur in a dictionary, in the nominative singular) may be
prose words, grāmina (singular grāmen) is not as common a word for grass as
herba, and comae (singular coma) is not the regular word for “leaves”, which is
folia, the source of English foliage. Comae, by contrast, is a word favored by
poets, used primarily of hair. Astronomers will have come across the singular
coma in Coma Berenices “The Hair of Berenice”, a constellation named after a
Ptolemaic queen whose hair had been transformed into the stars in question.
(At least that’s what the court astronomer claimed.) Because poems praising this
metamorphosis had been written by the poet Callimachus in Greek, then again
by Catullus in Latin, the constellation ended up with a name featuring the poetic
word for hair rather than either of the main prose words, capillus or pilus. Here
in Horace’s poem, however, it is used in a figurative sense, very common in Latin
poetry, in reference to leaves, which are to the tree as hair is to the head. Indeed,
it’s so commonly used in this metaphorical sense that it’s easy to overlook that it
actually means hair, a meaning that should, however, be kept in the back of one’s
mind since the fundamental point of the poem is to contrast the renewal of
nature, collectively, with the inevitable mortal decline of humans, individually.
The particular image employed here can only be expressed coarsely in English—
tree hair grows back, human hair, eventually, doesn’t—but Latin poetic diction
can do this very elegantly with the single word comae.
Finally, there’s the first word of the sentence, the verb redeunt, which, as the
compound of red- “back, re‑” and eunt “go”, simply means “return”. But here
Housman moves furthest away from a literal rendering of the Latin, offering
“renew their birth”. This might seem like excessive liberty on the part of the
translator, drawing out too explicitly the contrast between Nature’s springtime
renewal and humankind’s inexorable decline. But considering that he has
had to let go of the double meaning of comae, this is a good way of bringing
the  subtle anthropomorphization of the natural world back into the poem.
Conveniently, the use of a stationary verb (“renew”) rather than one suggesting
more motion (“return”) also works well with his choice of the similarly station-
ary prepositions “on” and “in” to translate the case endings of campīs and arbo-
ribus. This decision, though, is not without its cost: in Latin, there’s a symmetrical
balance of movement, lost in Housman’s version, between the snows fleeing
and the grass and leaves returning.
82 How dead languages work

Thus even as skilled a translator as Housman can’t capture everything that’s

going on in the Latin. Horace, to give a further example, can throw shades of
emphasis on different words by manipulating the word order in a way that’s not
possible, or at least not natural, in English. To understand how he plays with the
word order, though, one first needs to have an idea of the baseline he’s starting
from. In the typical English sentence with a transitive verb, the default order of
the main elements will be subject–verb–object (SVO): The poet wrote the poem.
In Latin, while there is considerably greater freedom of word order, it would be
most common for the verb to come last (SOV), as if English said The poet the
poem wrote. For the first three verbs in this poem, however, Horace places the
verb first in its clause: diffūgēre nivēs, redeunt iam grāmina, mūtat terra.
Understanding what he gains by this requires a brief consideration of how sen-
tences are shaped by the wider context in which they occur, the sort of question
studied by the branch of linguistics known as pragmatics.19
Now subjects come first across a wide range of the world’s languages,20 and
a major reason for this has to do with the flow of information in a sentence.
On  the whole, it’s easier for the listener to process a sentence if the given,
­accessible material (often called by the technical term topic) comes first as an
orientation, and the new material that’s the reason for making the utterance
in the first place (called the focus) is only presented afterwards. What’s more,
just as the discourse, at a pragmatic level, is organized around a topic (that is,
the  person or  thing that the narrative or conversation is about), so too the
­sentence, at a syntactic level, is centered on a noun or pronoun subject of which
some ­verbal predicate is said to be true. Sentences flow best when these two
levels are aligned, so if you’re writing a poem with Achilles as the topic, Achilles
is likely to be a frequent sentence subject as well. And because it works best to
put topics first, so too subjects are also often first in their sentence.
But there’s also no requirement that the topic of a sentence be the same as its
subject. Sometimes a grammatical element other than the subject is a better
starting point for the thought being expressed. Consider the simple exchange:

A: Do you like To Kill a Mockingbird?

B: The movie’s very good, but the book, embarrassingly, I’ve never read.

Since speaker B has been asked about To Kill a Mockingbird by speaker A, it

makes sense for it to be treated as the topic of the reply. Now English, in most
circumstances, doesn’t rearrange the elements of a sentence to highlight the

19  For a general introduction to pragmatics, see G. Yule, Pragmatics (Oxford, 1996).
20  For answering such questions as “How many languages have such and such a feature?”, the
World Atlas of Language Structures Online (https://wals.info/) is an outstanding resource.
Feature 81A catalogues the order of subject, object, and verb in no fewer than 1377 languages,
complete with a color-coded map; the subject comes first in 1053 of these, or 76%.
 Latin83

topic as much as certain other languages do, so a straightforward answer might

be simply, “Yes, I like it a lot”, with the usual SVO word order. But here, speaker
B wishes to distinguish between two different versions of Mockingbird, the
movie and the book, either of which could potentially be treated as the topic of
speaker B’s answer, once speaker A asked a question about it. In such cases,
where a topic (Mockingbird) encompasses two or more subtopics (the movie
and the book), English has the option of bringing the subtopics forward in the
sentence, even if (as is the case with “the book”) it’s the object of the verb and
would normally follow it.
Topics, thus, have a tendency to move further forward in the sentence than
they might otherwise appear, and this can even happen with verbs in some
languages. Here Horace’s diffūgēre nivēs, redeunt iam grāmina, and mūtat terra
all have this sort of gentle dislocation, in which the verb moves earlier in its
clause, thereby emphasizing that it is the ideas expressed by the verbs—the
cycle of seasonal change—that serve as the topic of the poem rather than
what happens specifically to snow, grasses, and the earth. While this sort of
topicalization of verbs doesn’t happen as such in English, a similar effect can be
achieved through the use of the gerund or other verbal nouns: “There’s a fleeing
of the snows, a return of the grass, and a changing of the earth.” But it comes at
a cost, since the change from finite verbs to verbal nouns not only obscures the
shift in tense in the Latin (the snows have fled in the perfect tense, whereas the
grasses and the earth are returning and changing in the present) but also leads
to a much more static picture (not to mention stilted language) than is offered
by the Latin.
Some of the other features of Horace’s word order also fail to translate well.
In redeunt iam grāmina campīs arboribusque comae, the final four words fall
into a pattern common in Greco-Roman poetry: rather than grouping two
conceptually parallel pairs as ABAB—grassesA to the fieldsB, leavesA to the
treesB—Latin often arranges them symmetrically, in the order ABBA. If one
writes the pairs over one another, and draws lines between the corresponding
members (e.g. subject to subject, object to object), a cross is formed, giving this
pattern its technical name, chiasmus, after the Greek letter chi, which is shaped
like a St Andrew’s cross (X):

grāmina (grasses) campīs (to the fields)

arboribus (to the trees) comae (leaves)

While chiasmus is often treated as if it were no more than an ornamental styl­
is­tic device, something a poet includes simply to dress up an otherwise or­din­
ary sequence, in fact it is based on a rationale similar to that which causes topics
to come early in the sentence.21 The tendency for clauses to start from the given
84 How dead languages work

and proceed to the new is part of a general strategy whereby speakers make
what they’re saying cohesive for the listener by moving from what is either
already in the addressee’s mind, or could reasonably be inferred on the basis of
what has already been said, towards what is less expected. In a chiasmus like
grāmina campīs, arboribus comae, at the end of the first pair, the B item of the
pair, campīs “to the fields”, is what’s in the speaker’s mind, so it works well for
the second pair to begin with the corresponding element, arboribus “to the
trees”, and only then move on to comae “leaves”. This sequence of thought is
rendered even easier in Latin thanks to a quirk of the word that here means
“and”, which I’ve left out of the last couple of quotations of the phrase to isolate
the chiasmus more clearly. As mentioned in Chapter  2, Greek, Latin, and
Sanskrit all have a word for “and”—que in Latin, te in Greek, and ca in Sanskrit—
that, as a postpositive conjunction, comes after the second word of the pair it
joins together rather than between them: not A que B, but A B-que, as if we said
in English “salt pepper-and”. Here Horace uses -que22 to join not just two nouns
but two clauses. And using postpositive -que, rather than the alternative et,
which goes between the two elements it coordinates, leads to two felicitous
effects difficult to replicate in English. First, its position after, rather than before,
arboribus puts campīs and arboribus directly next to each other, thereby work-
ing together with the chiasmus to achieve an especially smooth flow from one
idea to the next, with the two locations which turn green in spring contiguous
with one another. Second, recalling that arboribusque comae constitutes an
entire line in the poem, we see that the post­pos­ition of the conjunction allows
the two most important words, arboribus and comae, to take pride of place at
the start and end of the line, yielding a symmetry in which the two full nouns
are located on either side of the weak conjunction.
Untranslatable intricacies of word order continue throughout the 16 lines
I’ve excerpted. To touch on just a few of these effects, from the second stanza,
consider again these lines:
immortālia nē spērēs, monet annus et almum The swift hour and the brief prime of the year
quae rapit hōra diem. Say to the soul, Thou wast not born for aye.

Housman has here altered the sequence of thought considerably. A word-

for-word translation would run something like this, and be virtually incompre-
hensible without recourse to the original poem:
immortālia nē spērēs, monet annus et almum
immortality that not you hope for, warns the year and nourishing

21  Readers who already have a grounding in Latin and/or theoretical linguistics can find a
considerably more detailed treatment of the pragmatic factors underlying chiastic constructions
in A.  M.  Devine and L.  D.  Stephens, Latin Word Order: Structured Meaning and Information
(Oxford, 2006), pp. 242–9.
22  When written in its own right, -que is usually given a leading hyphen, since it’s attached to
the previous word in standard Latin orthography.
 Latin85

quae rapit hōra diem.

which snatches away the hour the day.

Rearranged into something a little more idiomatic, this becomes: The year
warns that you shouldn’t hope for immortality, as does the hour which snatches
away the nourishing day. Whereas Housman’s translation begins, as is natural
for English, with the subjects of the main verb, the seasons of the year that
remind us of our transience, Horace places the idea of immortality front and
center. In English syntax, this would be difficult, since not only is immortālia
the object of the verb, namely spērēs “hope for”, but the whole clause in which
it’s situated, immortālia nē spērēs (“. . . that you shouldn’t hope for immortal­
ity”), is also acting as the object of the verb monet (“it warns . . .”), the reverse of
the natural English order. Nor, for that matter, is “immortality” all that great a
translation of immortālia. First, it suffers from the same problem that mortālis
does in the passage of Lucretius discussed above: it belongs to the Latinate
register of English diction, and is therefore one degree more distant from the
blunt expression of death than is immortālia in Latin. Second, immortality is an
abstract noun in English, whereas immortālia is the neuter plural of the adjec-
tive immortālis, which has undergone the same substantivization seen earlier
in the examples mūtātīs mūtandīs and cēterīs paribus. That is, what the address-
ee of Horace’s poem shouldn’t hope for isn’t immortality in the abstract but
deathless things in the concrete. Of course, the exact identity of those theoret­
ically concrete deathless things is left unspecified, so the concept is effectively
reabstractified. But the fact that the Latin word is actually the plural form of an
adjective, and not a noun, leaves space for the reader to wonder what those
multiple deathless things might be—put syntactically, to wonder what the
adjective could potentially be modifying.
A much more complicated word order marks the main clause—complicated
even by the standards of Latin, as this sort of interlacing goes beyond what one
would expect to find in the prose of a Cicero. It starts off simply enough: monet
annus “the year warns”, with the verb fronted, both to bring it closer to the
content of the warning, just given and, as with the other examples of clause-
initial verbs seen so far, to topicalize it. But then the simple conjunction et
introduces a second subject of monet. It comes across as a sort of afterthought,
since monet itself is marked as having only a singular subject. (With an or­din­
ary plural subject, the verb form would be monent.) But Horace leaves the reader
guessing as to what exactly that second subject is. First, we hear almum, an
adjective meaning “nourishing” (also found in the phrase alma māter “nour-
ishing mother”), whose ending -um tells us that it’s likely to be modifying an
object, but that it could also be in agreement with a neuter subject. Next comes
quae, a relative pronoun “that, which”, which again is ambiguous, but could
either refer to a plurality of neuter things or, as will turn out to be the case, a
feminine subject. With rapit, we get a second verb, which the reader is expecting
86 How dead languages work

once the relative pronoun was introduced, but we don’t know yet what its sub-
ject is, or how any of this is coordinated with monet annus. Only with the last
two words of the line do things fall into place: hōra is feminine, and it can only
be a subject, so it easily slots into place both as the antecedent of the relative
pronoun quae and as the noun that et coordinates with annus; and diem is the
object that almum is modifying. If this seems unusually jumbled, that’s because
it is. In ordinary Latin prose, while there would still be a lot of flexibility, we
would probably expect the order:

et hōra quae almum diem rapit

and the hour which the nourishing day snatches away

Now an English speaker is likely to process even this order incorrectly: arranged
thus, we have to understand the relative pronoun which as the object of the
verb, and the nourishing day as its subject. But the only transposition needed to
turn this into English is to move the verb in between which and the nourishing
day. Importantly, everything that belongs together syntactically is grouped
together in continuous units: almum modifies diem, so they appear next to
each other. And quae almum diem rapit is also a single constituent—a relative
clause—and it too appears as a single, uninterrupted sequence. Not so, how-
ever, with Horace’s poem, where almum is separated from diem by three words,
and hōra, rather than preceding the relative clause as its antecedent, appears in
the middle of it. Such dislocation, especially the positioning of almum, is strik-
ing even by the standards of Latin, and the technical term for it is hyperbaton.
Why does it happen? One explanation might be that Horace had to rearrange
the words for them to fit the meter, and this is what he came up with. But that’s
intolerably weak: there are plenty of lines of Latin verse where this doesn’t hap-
pen, and we’d be accusing Horace of being an awfully lousy poet if we assumed
that it was only faute de mieux that he fell back on this word order. A better
explanation would be to propose that some sort of movement has taken
place, like the topicalization of verbs mentioned earlier in this section. While
hyperbaton involving the dislocation of this many words would be unusual in
prose, there are plenty of parallels for individual topics and foci being brought
forward in the sentence because of the particular emphasis placed on them.
One should prefer, where possible, the economical explanation, so if we can
argue that something similar is happening here, only to a greater degree,
then we haven’t introduced any unnecessary complications into the grammar
of Latin.
But what’s going on in the information flow of this sentence that this sort of
movement would be called for? Most of the relevant syntactic theories hold that
a default order A shifts to a marked order B through the leftward movement of
topics, foci, and the like. We’ve already seen an example of this with the verb
redeunt fronted because it’s a topic:
 Latin87

A iam grāmina campīs redeunt

B redeunt iam grāmina campīs

That much is fairly clear and not impossible in Latin prose either. But in the
example at hand, we see that no fewer than three of the five words have to
move forwards in order for hōra, which should come first, to be relegated to the
­penultimate position:
A hōra, quae almum diem rapit

B almum quae rapit hōra diem

Just as the end result is not the sort of word order one would expect in prose, so
too this sort of movement is rather too extreme to be accounted for by the usual
pragmatic rules whereby topics and foci are fronted. We could certainly try to do
so: almum could be in a topic slot; quae, as a relative pronoun could be fronted in
its role of introducing a subordinate clause (something similar happens regularly
in English: in the person whom I saw, the relative pronoun whom, though an
object, comes before the subject); and rapit could be regarded as a focal element,
since one could argue that the snatching away of the day is what’s most salient
here. But that all seems a little ad hoc. The topic should be something that the
sentence is about, which connects it to what precedes, and almum, as an adjective
meaning “nourishing”, probably fails on both counts. A better case could be made
for taking rapit as the focus of the subordinate clause, but its movement in front
of the noun hōra, which is the antecedent of that clause, would still be strange.
No, in a situation like this, it’s better to remember that Horace is a poet, and
think about other effects that are achieved by these dislocations. In terms of the
overall impression, there’s the very confusion caused by all this syntactic move-
ment: even the most fluent Latin speakers would have to wait until hora before
they could start to make sense of the relative clause. Such uncertainty is very
much at home in a line of poetry that reminds us how quickly the passage of
time snatches the days away from us. This general effect is further enhanced by
the particular residue that’s left after the fronting has taken place: the line ends
with the two words in the clause most associated with time, hōra diem. By for­
cing the reader to wait until the very end of the line to hear the two words that
enable the preceding syntax to fall into place, the poet can emphasize them all
the more strongly.
88 How dead languages work

All of this, of course, poses immense difficulties for the translator, who
c­ annot jumble the words in this way in English, because we don’t have the
case  endings on words that would enable the reader to make sense out of
Horace’s syntactic spaghetti bowl. But this doesn’t mean that an English poet
can’t draw attention to some of the same words and ideas, just that it must be
done in way more suitable for the particular character of our language. One
means of underlining words that does work well in English is alliteration:
Housman’s “We are dust and dreams” makes for a very effective close to the
fourth stanza and, while alliteration occurs from time to time in Latin poetry,
it’s nowhere near as pervasive a poetic device as in Germanic languages like
English, as we’ll see at greater length in Chapter 4. Furthermore, while Housman
can’t put immortality at the start of the first line without sounding stilted
(“Immortality, that it should not be hoped for, the year warns . . .”), he can place
“aye” at the end of the verse, another strong position in the line. Indeed, verse-
end is stronger in English than it is even in Latin because of a common struc-
tural feature of English verse that is absent altogether from Classical Latin
poetry: rhyme. In a stanza like Housman’s, the sonic reinforcement created
between the line-end rhymes ( fear ~ year, play ~ aye) inevitably draws the ear’s
attention to these words in particular and goes a long way towards restoring the
prominence of immortālia in Horace’s original. But while that much can be
conveyed in translation, there is very little that can be done in English to
­capture the complex interweaving that permeates the rest of that couplet.
As a final note, though, before moving on from word order, it is also im­port­
ant to forestall any impression that this sort of complexity is unique to Latin. In
fact, most of the other languages discussed in this book, like Greek and Sanskrit,
take advantage of the flexibility enabled by case endings and other inflections
to a similar extent. In Greek, for instance, equally intricate word order can
be  seen in the lyric poetry of Pindar, a poet just as difficult to translate as
Horace and, not coincidentally, one of Horace’s models. (The second poem in
Horace’s Odes 4 warns that imitating Pindar is no easy task, with the explicitly
agonistic first line: Pindarum quisquis studet aemulārī “Whoever desires to
rival Pindar.”) But the centrality of such syntactic play to Horace’s poetry makes
him especially appropriate as an illustration of what inflected languages can do
that English can’t.

TACITUS

Some may wonder whether I have exaggerated Latin’s flexibility by choosing

poets as the source of the first two examples of “Things Latin can do that English
can’t”: Surely poetic license exists in Latin just as much as in English? Yes, it
does, but most of the features of Latin we’ve seen so far can be paralleled in
 Latin89

prose language—it’s just that their prominence in poetry makes them easier to
single out for observation there than in more pedestrian texts where linguistic
craftsmanship isn’t at such a premium. But for our third text, we turn to the
prose of the historian Tacitus, who was writing a hundred-odd years after
Horace, in the early second century ad. Most of his works—which are anything
but pedestrian—are marked not only by a delightfully bitter cynicism but also
by a style that exhibits an extreme compression of thought, eschewing the
­balance and fullness of Cicero, the great master of earlier, Republican prose.23
Tacitus’ economy of language can be seen at a glance if one opens pretty much
any page of the bilingual Loeb edition of his works: the left page, with the
Latin, is virtually double-spaced, or, if closer to single-spaced, is filled out with
­copious notes at the bottom of the page; the facing page, with the English
­translation, is crammed with text from top to bottom. That said, the same is
often true of other Latin authors as well, so, rather than relying on such anec-
dotal evidence, it’s better to turn to Tacitus’ own words to get a sense of how he
takes  advantage of the linguistic resources of Latin in order to achieve his
extreme brevity.
In this, we’ll focus on a couple of passages from the Annals, Tacitus’ final
work, in which he covered the history of the Julio-Claudian emperors, start-
ing from the death of Augustus in ad 14, and ending with the reign of Nero
in 68. Much of the work is lost—we don’t even know how many books it
consisted of, with 16 or 18 being the best guess—but eight books, including
the first four, survive intact, as do substantial fragments from four more.
Right from the very  first sentence, the work is a study in untranslatable
succinctness:

urbem Rōmam ā prīncipiō rēgēs habuēre;

the city Rome from the beginning kings held
lībertātem et cōnsulātum Lūcius Brūtus īnstituit
freedom and the consulship Lucius Brutus established

Until recently, the most widely available translation of the Annals was probably
Michael Grant’s 1956 Penguin edition, which renders these lines as such:
When Rome was first a city, its rulers were kings. Then Lucius Junius Brutus
­created the consulate and free Republican institutions in general.
Twelve words have become 23, partly because of the exigencies of English
­syntax—all those definite articles—but also because of the extent to which
Grant expands the phrasing in order to give more context for the contemporary

23  In both these respects, Tacitus is often compared to Thucydides; we will return to this at the
end of the chapter.
90 How dead languages work

reader: most noticeably, lībertātem becomes “free Republican institutions”, as

a way of slipping into the text the reminder that, for a Roman audience, the
mention of Brutus signals the start of the Roman Republic.
A couple of new translations, however, have appeared in the past few
years, both of which stay rather closer to the Latin. In 2012, Penguin published
a ­completely new version, by Cynthia Damon, who renders the opening
as follows:
The city of Rome was originally in the hands of kings; liberty and the consulship
were instituted by Lucius Brutus.
At 20 words, Damon has only shaved off three from Grant’s version, but much
more of Tacitus’ style comes through. Particularly noticeable is the abruptness
of the transition from the first to the second half of the sentence, with only
a semicolon in Damon’s rendering where Grant slips in “Then” as a sort of
lubricant to smooth out the sequence of thought. Such absence of a connective
particle is not uncommon in Latin (it’s given the Greek name asyndeton,
which simply means “not (a-) tied (-deton) together (-syn-)”), and it often
implies a contrast between the two clauses joined thereby, such as here,
between kings and liberty. Perhaps the most important difference, though,
between these two translations lies in the treatment of lībertātem et cōnsulātum.
Tacitus here makes grammatically parallel two words that are not (or at least
not straightforwardly) on the same semantic plane: lībertātem is the abstract
concept of freedom or liberty, cōnsulātum the more concrete institution of the
consulship, the highest public office in Republican Rome. Once again, Grant
completely erases this somewhat jarring combination by rewriting lībertātem
as “free Republican institutions in general”, thus artificially restoring a seman-
tic parallelism with “the consulate”. Damon’s version is much closer to this
typically Tacitean turn of phrase both in its brevity and in its deliberately
incongruous juxtaposition of the two terms. Less obviously, Damon also
rewrites the syntax of the Latin so that the subjects of the two clauses are not
people (“kings” and “Brutus”) but the inanimate entities that are the gram-
matical objects in the Latin (“the city of Rome”, “liberty and the consulship”).
While this superficially makes Grant’s a more faithful version, Damon has
probably captured the spirit of the original better. As seen above in the case
of Horace, Latin word order is often determined not by syntactic functions
like subject and object but by the pragmatic roles played by the participants in
the clause: items that come early on—here urbem Rōmam and lībertātem et
cōnsulātum—are often placed where they are to signal that they are the topic
of the sentence. But in order for them to register as topics in English, they have
to come earlier in the sentence than is generally possible for grammatical
objects. Thus, if the information flow is to be preserved, then the sentences
need to be rewritten with passive verbs (“were instituted by”) or equivalent
periphrases (“was in the hands of ”), so that these topical elements can be
 Latin91

brought forward as sentence-initial subjects. Recasting the verbs in this way

allows the translator to make clearer that these sentences are not so much
about kings and Brutus as about Rome and freedom.
For a final translational comparandum (yet another English word that comes
from a Latin gerundive), we may turn to A. J. Woodman’s 2004 translation for
Hackett, which is especially useful for our purposes because of its deliberate
attempt to remain as faithful to Tacitus as allowed by the constraints of
English:24
The City of Rome from its inception was held by kings; freedom and the consul-
ship were established by L. Brutus.
This version, which also comes to 20 words, is clearly much closer to Damon’s
than to Grant’s: Woodman has anticipated Damon in adopting the same sen-
tence structure, asyndeton, and a single-word translation of lībertātem. That
said, his choice of “freedom” better captures the fact that lībertātem in Latin is
transparently derived from līber “free”, and thus has a visceral straightforward-
ness not as apparent with the more abstract “liberty” in English. Still, both of
these words acquired such baggage in English in the wake of the American and
French Revolutions that either choice is ­inevitably loaded with anachronistic
resonances.25 The issue is reminiscent of the earlier discussion of the best way
to render Lucretius’ mortālis: “liberty” here, like “mortal” there, preserves the
Latin root, but at the cost of directness.
The other noticeable difference lies in the rendering of ā prīncipiō, with
Woodman’s “from its inception”, unlike Damon’s “originally”, retaining the syn-
tax of a prepositional phrase. This may seem like a minor difference, and, con-
sidering that “inception” is a fairly rarefied word in English while ā prīncipiō is
perfectly ordinary Latin, one might be tempted to see Damon’s version as here
adhering more closely to the underlying feel of the original. Indeed, the choice
of “inception” makes sense primarily in the context of Woodman’s general aim
of allowing as much as possible of Tacitus’ wordplay to show through in the
translation (just as prīncipium “beginning” echoes prīncipātus “principate,
empire”, a key concept in Tacitus’ œuvre, so too “inception” has the same
central root as “principate”).26 Yet it is the decision to retain the prepositional

24  For an outstanding introduction to the impossibility of understanding what Tacitus has
really said if one can’t read Latin, with further examples along these lines, see especially
A.  J.  Woodman “Readers and reception: A text case”, in A Companion to Greek and Roman
Historiography, vol. 1, edited by J. Marincola (Malden, MA, 2007).
25  Geoffrey Nunberg’s editorial “The Nation: Freedom vs. Liberty; More Than Just Another
Word for Nothing Left to Lose” (New York Times, March 23, 2003) nicely draws attention to the
historical shift in the relative popularity of the two words in American political discourse, with
liberty more prominent at the time of the Revolution, and freedom coming to the fore with
Roosevelt’s Four Freedoms and the civil rights movement.
26  It’s not absurd to see prīncipium in the neighborhood of lībertās and surmise that Tacitus
might at some level have had the word prīncipātus in mind. After all, a memorable statement
92 How dead languages work

phrase that actually has more significant repercussions for the construal of
the Latin.
There are several different ways of saying “at or in the beginning” in Latin:
one can either use the ablative case on its own (prīncipiō), which, as noted
­earlier, can be used to mark the location, including the temporal location, of an
event; or one can add a preposition, either in “in” or ab “from”, as Tacitus does
here (with the final -b dropped before the initial p- of prīncipiō). Of these three,
it is probably in prīncipiō that will be the alternative most familiar to readers
because of its prominence in ecclesiastical Latin and classical music: In prīncipiō
erat Verbum (“In the beginning was the Word”) at the start of the Gospel of
John, and Sīcut erat in prīncipiō (“As it was in the beginning”) in the Gloria.
Still, all three are perfectly good Latin (Cicero, for instance, uses all three), and
it is not always easy to determine how they differ from one another. But while
there may be some contexts in which they would be interchangeable, their
­distribution is also not entirely random. In the collected works of Tacitus, for
instance, prīncipiō occurs 13 times in the ablative, 12 times without a prep­os­
ition (mostly expressions with a dependent genitive, e.g. annī prīncipiō “at the
start of the year”), only here with ab, and never with in. If a writer as meticulous
with his words—and as sparing with them—as Tacitus only uses the prep­os­
ition­al option here, then readers should take note, especially since this is the
very first sentence of his magnum opus, and thus one that he will presumably
have crafted as painstakingly as possible.27 And what the historian achieves by
adding ab is the introduction of a useful ambiguity.
On the one hand, ā prīncipiō can be as neutral as Damon’s translation
­suggests: the preposition’s original sense “from” has frequently been bleached
away, and the phrase means simply “in the beginning” or “originally”.28 On the
other hand, there are also examples where ab still retains its full force, as is clear
in more extended expressions like ā prīncipiō ūsque ad hoc tempus “from the
beginning right up to the present time”. By the time, then, that the reader has

on the (in)compatibility of prīncipātus and lībertās is found in the Agricola, Tacitus’ encomium
of his father-in-law: quamquam . . . Nerva Caesar rēs ōlim dissociābilēs miscuerit, prīncipātum ac
lībertātem “Although . . . the emperor Nerva blended things that were once irreconcilable, the
principate and freedom” (3). For further pairings of prīncipātus and lībertās, see Woodman’s
­commentary ad loc.
27  The first lines of works were at least as important in ancient literature as they are now.
Dionysius of Halicarnassus, a teacher of rhetoric in the Augustan age, records that, when Plato
died, there was found among his possessions a tablet containing the opening eight words of the
Republic arranged in various orders, so determined had he been to get it just right.
28  This shift from “from” to “in” is prominent in the spatial question-words of Latin and
the Romance languages, with interrogatives that mean “from where?” coming to mean simply
“where?”, as we’ll see in Chapter 5.
 Latin93

read Urbem Rōmam ā prīncipiō rēgēs habuēre, some uncertainty hovers in

the air: “Kings held Rome in the beginning [but no longer do]” or “Kings held
Rome from the beginning [and perhaps they still do]”. Indeed, the potential for
the latter reading is even greater than that translation suggests since the verb
form habuēre, as a Latin perfect, can be understood either as equivalent to an
English simple past (“held”) or present perfect (“have held”). Now, since the
basic facts of Roman history are what they are, and all of Tacitus’ audience
knew of Brutus’ expulsion of the kings, no reader would hear this second option
as literally true. In any case, Tacitus goes on to mention the end of the kingship
in the very next clause, thereby canceling the possibility of the second in­ter­
pret­ation. But given that he portrays the emperors of his own era as engaged in
the same sort of tyrannical behavior that led to the end of the kings, one can
certainly see ā prīncipiō rēgēs habuēre as inviting the reader to ask: Just how
long did the kings’ rule last for? Is it in fact over? And this nuance is easier to
detect in Woodman’s “from its inception was held by kings” than in translations
that flatten ā prīncipiō to “originally” or the like.
Some may find it a bit excessive to extract so much meaning out of the ā
prīncipiō of the first clause, especially if that reading only works if that clause is
considered in isolation from what follows immediately afterwards. But there is
in fact an excellent reason to take these six words as a statement in their own
right, and it is a final important feature of this opening that is inevitably lost in
translation. They form a perfect dactylic hexameter, the same metrical pattern
used by Lucretius in the Dē Rērum Nātūrā, by Virgil in the Aeneid, by Horace
in the odd lines of Odes 4.7, and, for that matter, in Greek, by Homer. To recap,
lines in this meter consist of six feet, of which the first four may be either a
dactyl (a long followed by two shorts) or a spondee (two longs), the fifth nearly
always a dactyl, and the sixth a spondee. Compare the first lines of Tacitus and
Horace’s ode, with the rhythms marked out; the musically-inclined may wish to
hear this as a 2/4 rhythm, with the long syllables as quarter notes, the short
syllables as eighth notes:

– – | –  – | – ˘ ˘  | – – | – ˘  ˘ |  – –
ur- bem   Rō-  mam ā29   prīn- cipi- ō rēg-   ēs habu-   ēre

– – | – ˘  ˘ | – ˘  ˘ | – – |   –   ˘  ˘ |  –  –
dif- fūg-   ē- re niv-   ēs, rede-  unt iam  grā- mina  campīs

29  Final -m in Latin was pronounced weakly—in all probability it represents simply nasaliza-
tion of the preceding vowel, with -am pronounced as if it were French an—so the final syllable of
Rōmam is elided into the preposition ā to become a single long beat in the meter of the verse.
Incidentally, the prosodic shape of the line would be the same if the preposition were omitted or
changed to in, so the use of ā cannot be attributed to metrical factors.
94 How dead languages work

That Tacitus should work such a clearly poetic rhythm into his prose history
apparently goes against Aristotle’s injunction, in the Rhetoric, that prose should
be neither metrical (emmetron) nor without rhythm (arrhythmon).30 If it’s
metrical, says Aristotle, then it comes across as artificial and draws too much
attention to itself; but if a sense of rhythm is altogether lacking, then sentences
never come to a proper end, and that’s unpleasant too. But if Tacitus here seems
to be disobeying Aristotle, he does so with good reason. First, the genre of
history was described by the important rhetorician Quintilian ( fl. first century
ad, about a generation older than Tacitus) as proxima poētīs (“closest to the
poets”), in part because it aimed to preserve the renown of great events and
people for posterity—also a key goal of the sort of epic poetry composed in
dactylic hexameters. Second, Tacitus is not the first Roman historian to have
directed readers’ attention by casting an introductory sentence in a dactylic
rhythm. While it is not the opening line of the work, Sallust, a historian in the
late Republic, signals that the heart of his account of the Jugurthine War is
getting started with this line:

– – | – – | –   – |   –   ˘  ˘ | – –  |  – –

bel- lum   scrīp- tū-   rus sum,   quod popu-   lus Rō-  mānus
I am going to write of the war which the Roman people…

With the sentence-initial object (bellum), first-person verb (sum), and relative
clause giving further specifics about the object (quod  . . .), this is not so different
an opening gambit from that followed a generation later by Virgil in the Aeneid,
an actual epic poem:

– ˘  ˘ | – ˘  ˘ | –    – | –  – |   – ˘  ˘ |   – –
ar- ma vir   um- que ca-   nō, Trō-  iae quī   prī- mus ab  ōrīs
I sing of arms and of the man who first from the shores of Troy…

And Tacitus’ other great predecessor, Livy, writing in the time of Augustus,
begins his history with what’s trying to be a dactylic hexameter but doesn’t
quite get there in the final foot:31

–       –   |   – ˘ ˘  |  – ˘ ˘  |  –     –  | –     – |  –  ˘ –

fac- tū- rus- ne ope- rae preti- um sim,  sī ā prīm- ōrdiō
Whether I’m going to do something worth the effort, if from the
beginning . . .

30  τὸ δὲ σχῆμα τῆς λέξεως δεῖ μήτε ἔμμετρον εἶναι μήτε ἄρρυθμον (1408b21–2).
31  In both this line and the Sallust, the penultimate foot is a spondee, which is much rarer than
a dactyl at this position in the verse, but not unheard of. It is also just possible that the final word
could in fact scan as a spondee if the short i in prīmōrdiō is treated as a glide (i.e. as if it were an
English y) rather than as a full vowel.
 Latin95

It is also worth noting how Livy’s sentence proceeds in full: factūrusne operae
pretium sim, sī ā prīmōrdiō urbis rēs populī Rōmānī perscrīpserim, nec satis
sciō nec, sī sciam, dīcere ausim, quippe quī cum veterem tum vulgātam esse rem
videam, dum novī semper scrīptōrēs aut in rēbus certius aliquid allātūrōs sē aut
scrībendī arte rudem vetustātem superātūrōs crēdunt “Whether I’m going to
do something worth the effort, if I write the affairs of the Roman people all
the way through from the beginning of the city, I neither know very well nor,
if did know, would I dare to say, inasmuch as I see that it is both an old and
common thing, while new writers each in turn believe either that in the con-
tents of their work they’re going to contribute some sort of greater certainty
or that they’ll surpass old-fashioned roughness in their craft of writing.” Two
points deserve attention. First, Livy uses the phrase ā prīmōrdiō “from the
beginning”, with prīmōrdiō a near-synonym of Tacitus’ prīncipiō, in a way
that makes clear that the preposition here must bear its full sense “from”,
rendering more likely a similar reading of Tacitus’ ā prīncipiō.32 Second,
whereas Tacitus’ initial sentence is a mere 12 words, with no sub­or­din­ate
clauses, Livy’s extends to 47, and has considerable syntactic complexity, with
an indirect question (“Whether . . .”) and a conditional clause (“if ”) at the
start, and causal “inasmuch as” and temporal “while” clauses at the end.
Tacitus’ opening shows that Latin can be very concise; Livy’s, that it doesn’t
have to be.
Before leaving Tacitus and Latin behind, an additional example of its
brevity is worth noting quickly, simply to emphasize what the language is
capable of. As a contrast to the first passage, the very opening of the work,
consider next a comparatively ordinary and unremarkable couple of sen-
tences from the middle of the Annals. It’s the year ad 28, the Romans have
just been defeated by the Frisians, a Germanic tribe, and Tacitus reports the
effects as follows (Annals 4.74.1); to give a sense of where Latin takes short-
cuts, I’ve put hyphens in the translation between words that are expressed
through a single word in Latin; words in parentheses have to be supplied
from context:
clārum inde inter Germānōs Frīsium nōmen,
Illustrious then among Germans (was) the-Frisian name,
dissimulante Tiberiō damna
Tiberius concealing the-losses,
nē cui bellum permitteret.
so-as-not to-entrust the-war to-anyone.
neque senātūs in eō cūra,
nor (was) the-senate’s concern in this,

32  The similarity of ā prīmōrdiō and ā prīncipiō—as well as their usefulness in programmatic
statements by historians—can also be seen in their both occurring as readings in one fragment of
Sallust’s History (1.8).
96 How dead languages work

an imperiī extrēma dehonestārentur:

whether the-empire’s boundaries were-dishonored:
pavor internus occupāverat animōs,
domestic fright had-seized their-minds,
cui remedium adūlātiōne quaerēbātur.
whose remedy in-flattery was-sought.

At this point, most of the reasons why Latin has a lower word count have been
seen already: it doesn’t have a distinct definite article, prepositions are often
expressed through case endings, and the auxiliary verbs of English often cor-
respond to suffixes or endings in Latin. But we also see here the fact that Latin
doesn’t need to use a form of the verb “to be” in sentences which simply connect
a subject with a predicate noun (“Their name was illustrious”) or state where
something is located (“Nor was the senate’s concern in this”). Such omission is
relatively unambiguous in a language that already signals subjects and objects
through case endings, and is found in other languages, too. But not having a
verb specifically marked for tense in such sentences does allow them to come
across as timeless truths more than is the case in English, where the syntax
requires a verb that, in this case, gives them away as past tense. (That an impres-
sion of lasting grandeur is aimed at in the first words is suggested not only by
the presence of clārum, exactly the sort of adjective that an epic hero would
want applied to himself, but also by the fact that the first four and a half words,
clārum ‿ īnde ‿ īntēr Gērmānōs Frī-, could be the first four spondees of a dactylic
hexameter. The ties between the first two words indicate that the final syllable
of one word is elided, that is, runs together and counts as one with the first
syllable of the next.)
Furthermore, that Tacitus’ material here does not have to be as anchored
to  the specific temporal context recalls his Greek predecessor Thucydides’
description of his own history (1.22.4): “It has been composed as a possession
for all time rather than as a showpiece to be heard in the moment.” Because of
this tendency to abstract away from the present to arrive at general truths—not
to mention a shared propensity for thorny language and pessimism—the two
historians have often been associated with one another. The pairing occurs in
passing in a letter of Thomas Jefferson’s, who, on January 21, 1812, three years
after the end of his presidency, wrote: “I have given up newspapers in exchange
for Tacitus and Thucydides, for Newton and Euclid; and I find myself much
the happier.” Considering the reputation that both historians have for a rather
gloomy view of human nature, there’s a somewhat mordant irony in Jefferson’s
finding even their writings to be a source of greater happiness than current
politics. Certainly, that gloominess, made all the more piquant by the concise-
ness of Tacitus’ expression, is in evidence in this passage just considered, with
its characteristically cynical content: the depiction of the paranoid Tiberius,
 Latin97

together with the bitter understatement that the senators chose to save their
own skin through sycophancy instead of taking steps to defend the empire’s
borders.
Still, Rome was able to hold off those barbarian Germans to the north for
about four hundred more years: after sporadic incursions in the following cen-
turies, in 406 Germanic tribes crossed over the Rhine, previously the border
with the Empire, and Rome itself was sacked in 410 by Alaric and the Visigoths.
The situation in Britain was also grim: already in the previous century, the
Romano-British (that is, the inhabitants at the time of what is now England and
Wales, whose linguistic culture was a fusion of Latin and Celtic elements) had
been coping with attacks from the west by the Irish, from the north by the Picts,
and the east by the Saxons. And when, in 410, they appealed to the emperor
Honorius for help with their defenses, they were told they were on their own—
not surprising, considering the turmoil in Italy itself. Whatever the exact cir-
cumstances, they are said to have invited Germanic speakers to come over from
the Continent as mercenaries, notably Angles and Saxons. Over the course of
the fifth and sixth centuries, it is these peoples whose language became rooted
in what at this point became England, as the newcomers gradually pushed the
Celtic speakers back to Wales and Cornwall in the west. And it is to this lan-
guage branch, Germanic—to which not only English and German but also
Dutch and the Scandinavian languages belong—that we now turn.
 4

Old English and the Germanic languages

What makes a Germanic language Germanic? To answer this question, this

chapter will begin by examining some of the specific linguistic features that
separate the Germanic languages from the rest of their Indo-European kin.
There are a couple of reasons to do so. First, by singling out some of these dis-
tinctive traits, we can see how a whole set of languages can be classified together
on purely linguistic grounds. Second, we can introduce a historical perspective,
since such traits arise through language change, and, in this case, the changes
became significant enough that a whole community of speakers ended up
forming a separate new group. In any case, we start from the basic fact that all
the Germanic languages—with English, German, Dutch, Swedish, Danish,
Norwegian, and Icelandic as the leading modern-day representatives—are
marked by particular features that indicate that they broke away from the ori­
gin­al parent language as a group rather than singly. For instance, they have the
sound f, rather than p, at the start of the father word, and they use endings like
-ed to form the past tense of most verbs. When such features represent a new
way of doing things linguistically, they are called shared innovations, and
they are generally the best means of assessing which languages belong together
in a common branch of a linguistic family tree. A shared archaism can be
retained independently in two different branches, but if, say, the oldest attested
Germanic languages all show evidence of a particular change in the articula-
tion of consonants and a new way of making the past tense, this is most eco-
nomically explained as arising from their ancestors’ having formed a common
speech community.
After considering these features, we will then have a look at Old English, the
language of the Anglo-Saxons, to see what makes the oldest recorded form of
our own language characteristically Germanic. Furthermore, we’ve now seen
how Greek and Latin differ from Modern English with their rich arsenal of
noun and verb endings, and Old English will give us a glimpse of what English
was like when it was still closer to Greek and Latin in this respect, and had only
just started down the path that would make it the morphologically pared-down
language we speak today. As sample texts, we will look both at the relatively

How Dead Languages Work. Coulter H. George, Oxford University Press (2020).
© Coulter H. George. DOI: 10.1093/oso/9780198852827.001.0001
100 How dead languages work

plain prose of the Old English translation of the New Testament, as well as the
more elaborate poetry of the most famous Old English poem, Beowulf.

GRIMM’S LAW AND UMLAUT

Our story begins with a couple of the German linguists who came to be most
associated with grouping the Germanic languages together in the first place,
the brothers Grimm. For they didn’t just collect fairy tales. They also col-
lected words, compiling an immense historical dictionary of the German
language, a counterpart to the Oxford English Dictionary (OED). And, after
devoting considerable time to comparing words in the Germanic languages
with those in Latin and Greek, Jacob Grimm noticed that there were certain
regular correspondences:

Greek Latin English German

patēr pater father Vater1
pod- ped- foot Fuß
—2 pisc- fish Fisch

Whenever Greek and Latin words started with a p, the corresponding Germanic
words started with an f sound (spelled in German with either an f or a v). Nor
was this the only such correspondence. Something similar could be seen with
initial k sounds in Greek and Latin (spelled in the latter with a c) matching
initial h in Germanic:

Greek Latin English German

kardiā cord- heart Herz
keras cornū horn Horn
he-katon3 centum hundred hundert

Finally, initial t in Greek and Latin corresponds to initial th in English, and

initial d in German. As will be explained below, an additional change has
affected German here, so I have also cited the corresponding words in Old
Norse, the language of the Vikings, as evidence that German, not English, is the
odd one out in this respect. For in Old Norse, these words also have the same th

1  German nouns are spelled with an initial capital letter in the standard orthography.
2  Greek ichthys is a different kettle of fish, so to speak.
3  The he- of Greek hekaton is of disputed origin; some have argued that it is a form of the word
for “one” that has become fused with the expected Greek outcome of the basic word for “hundred”,
katon, but the question remains contentious.
 Old English and Germanic languages 101

sound as English, even if it looks different, since it is spelled with þ, a letter

called thorn, which we will also see used in Old English.

Greek Latin English German Old Norse

tany-4 tenui- thin dünn þunn-
ty5 tū thou du þú
treis trēs three drei þrí-

What makes these sets of matching words (called word equations) important is
the regularity with which the correspondences between sounds occur; even

is, any time the Indo-European parent language had a *p, *t, or *k sound (with
apparent exceptions can generally be shown to follow subsidiary patterns. That

the asterisks indicating that these sounds are reconstructed rather than attested),
it remained such in Greek and Latin, but shifted to a f, th, or h respectively in
the Germanic languages. This led linguists in the late nineteenth century to posit
what is known as the Neogrammarian Hypothesis, that sound change is regular
and exceptionless (or, in its catchy German formulation, die Ausnahmslosigkeit
der Lautgesetze “the exceptionlessness of sound laws”). Furthermore, in addition
to the individual sets of correspondences here, a higher-order pattern encom-
passes all three: p, t, and k all belong to the class of sounds known as voiceless
stops. They’re called voiceless because the vocal cords don’t vibrate (in contrast
to their voiced counterparts b, d, g), and stops because the airflow is completely
blocked—in contrast to the sounds they turn into in Germanic, f, th, and h,
which are called fricatives, as air continues to pass through the mouth as
they are pronounced. The three sounds p, t, and k thus have the same manner
of articulation—a complete blockage of airflow—but differ in their point of
articulation: with p, airflow is blocked at the lips; with t, with the tongue behind
the teeth; with k, with the back of the tongue raised up to touch the soft palate.
This analysis paves the way for a unitary description of all three changes: in the

stops (*p *t *k) turn into the corresponding fricatives (f th h).6

Germanic languages, the Proto-Indo-European (PIE) word-initial voiceless

What’s more, this form of description, in which one generalized phonological

rule covers three separate individual sound changes, can be extended to account

4  This is not an independent adjective in Greek but is found as the first element in compounds
in the sense “long, stretched out”.
5  This is the form found in Doric and Aeolic Greek; it has changed to sy in the standard Attic
dialect.
6  One must allow, to be sure, a certain amount of phonetic fuzziness for this account to work:
f is technically a labiodental (with the lower lips touching the upper teeth) rather than a true
bilabial like p (with upper and lower lips touching); and since the fricative that properly cor­re­
sponds to k should be the ch of German, the actual outcome of h is usually thought to be a later
development. But in broad terms, the formulation that voiceless stops become voiceless fricatives
in Germanic works well.
102 How dead languages work

guages. First, the inherited Indo-European voiced stops *b, *d, *g turn into the
for two other classes of consonant change that characterize the Germanic lan-

voiceless stops p, t, k (thereby filling the gap left by the voiceless stops p, t, k
once they had turned into the fricatives f, th, h): thus Russian yabloko cor­re­
sponds to English apple, Latin duo, decem, and dent- to two, ten, and tooth,
and Latin gen-us and Greek gen‑esis have the same root as English kin. Second,

to have been voiced in the parent language (*bh, *dh, *gh—more on this in
where Greek has the aspirated stops ph, th, ch, which are most often thought

Chapter  5), Germanic has the voiced stops b, d, g (thereby filling the gap
caused by the change of b, d, g to p, t, k): Greek pherō ~ English bear (the verb,
not the animal), Greek thyrā ~ English door, Greek cholē ~ English gall.7
Again, these are sound changes that characterize all the Germanic languages:
apple matches Icelandic epli; in Gothic, two, ten, and tooth are twai, taihun,

gall. Taken together, these three sets of sound changes (*p *t *k > f th h; *b *d
and tunþus; and bear, door, and gall correspond to Old Norse bera, dyrr, and

*g > p t k; *bh *dh *gh > b d g),8 in which particular consonants, inherited
from PIE and largely preserved in Greek, Latin, and Sanskrit, shift as noted
in  Germanic, are called Grimm’s Law, in honor of their discoverer, Jacob
Grimm.
In citing these examples, I’ve mostly avoided using German itself as an
example. That’s because this sort of sound change is not uncommon, and a sub-
sequent change along the same lines affected the consonant system of Modern
High German (the standard variety of the language), which therefore differs
from the other Germanic languages in distinctive ways. Indeed, to the ears of
most anglophones, what most characterizes German is the apparent harshness
of its consonants. In particular, the rolling ch of Bach (called a velar fricative
because it’s produced by friction at the velum, or soft palate) can come across as
rebarbative to those whose languages lack the sound. Additionally, German is
also a language of affricates. These are composite sounds, like ts and pf, in
which the first element (a stop) is created by the complete closure of airflow
through the mouth, and the second (a fricative) by the turbulence caused when
that closure is opened up slightly. Of course, we have affricates in English as
well—the first and last sounds in both church and judge—but they are espe-
cially common in German owing to the so-called Second Germanic Consonant
Shift. This was a series of sound changes that affected the High German dialects

7  The Greek word for bile has found its way into more English words than you might expect:
thanks to the theory of the four humors, a depressed person was thought to suffer from an excess
of black bile (melan-choly), and an angry one from too much of the usual yellow bile (chole-ric);
and chole-sterol (“bile solid”) was so named because it was first discovered in gallstones.
8  Inevitably, I am simplifying the picture by e.g. leaving out discussion of non-initial stops and
labiovelars, which would have taken things too far afield. For further details about Indo-European
matters, readers can consult B. W. Fortson, Indo-European Language and Culture (Malden, MA ,
2010) and J. Clackson, Indo-European Linguistics (Cambridge, 2007).
 Old English and Germanic languages 103

in the mid-first millennium ad, roughly spanning the time between the fall
of the Western Roman Empire (traditionally dated to 476) and Charlemagne
(who died in 814). It affected different Germanic dialects to different extents,
and its effects were generally strongest in the south, in the direction of the
Alps—hence High German, as opposed to the Low German spoken closer to
the North Sea coast. In this consonant shift, what had earlier been simple stops
became affricates at the start of a word: words that had begun with p now had a
pf; those that had begun with t now had a ts sound, spelled z.9 Furthermore, in
the middle of a word, the stops p, t, and k mostly developed into the simple
fricatives, spelled ff, ss, and ch respectively. This one change accounts for many
of the superficial differences between the core vocabulary of English and
German. It is why we have pepper, but they have Pfeffer; why our numbers
include two and ten, theirs zwei and zehn; why they drink Wasser rather than
water—at least as an occasional alternative to Bier. And with this last word, we
see that words that don’t have consonants subject to this change stand a better
chance of looking alike in both languages: German Finger, Mann, and Ring all
mean exactly what you would expect. In any event, these initial affricates give
German much of its distinctive crispness.
Before we leave phonology and move on to a second major diagnostic fea-
ture of Germanic, one further phenomenon found in many—but not all—of
the Germanic languages needs to be mentioned: umlaut. This is a word that
does double duty. Those English speakers who have come across it are probably
mostly familiar with it as a term for the two dots that sometimes occur over the
German vowels ä, ö, and ü. Graphically, these letters are so characteristically
German that they have been deployed by heavy metal bands like Mötley Crüe
to brand themselves in an appropriately edgy way.10 But why does German use
this notation in the first place?
The key lies in the meaning of the word umlaut: um- is a German prefix that
often signifies change (like meta- in Greek, or trans- in Latin), and Laut is sim-
ply the German word for “sound”. Thus the diacritical device of the double dot
gets its name from the fact that it marks a change of sound. Now this could be
viewed simply as a synchronic feature: that is, without taking historical devel-
opments into account, it could simply be noted that the umlaut indicates that
the following umlauted vowels have a different quality from their counterpart
without the umlaut:

9  The change of word-initial k to the corresponding affricate (k followed by the ch of Bach)

affected only a comparatively small number of dialects, like Bavarian, and did not make it into the

might expect since its precursor, p, was not very common in Germanic, because, in turn, PIE *b
standard language. Also, pf, while certainly characteristic of German, is found less often than one

(the source of Germanic p by Grimm’s Law) was also unusually rare.

10  Readers are especially encouraged to consult the Wikipedia article on these so-called “metal
umlauts”.
104 How dead languages work

German word Meaning Vowel pronounced roughly like English . . .

Mann “man” father
Männer “men” bet
Tochter “daughter” (UK) dog
Töchter “daughters” no English equivalent: like bet, but with lips
rounded as if pronouncing dog (or like dog, but
with tongue moved forward as if pronouncing
bet); French eu
Bruder “brother” brood
Brüder “brothers” no English equivalent: like breed, but with lips
rounded as if pronouncing brood (or like brood,
but with tongue moved forward as if
pronouncing breed); French u; this is also the
same sound as the Greek y seen in Chapter 2

But the fact that words with the umlaut are often related to words without the
umlaut suggests that there’s a historical, or diachronic, relationship between
the two forms. Take the common pattern, seen here, in which a vowel with-
out umlaut in the singular (Mann) acquires the umlaut in the plural (Männer).
Since this pattern extends to numerous words (e.g. Wald “forest” also forms the
plural Wälder, and the plural of Blatt “leaf ” is Blätter), it is more efficient to
explain such plurals by assuming that originally the noun had the same vowel
in singular and plural, but that some sound change limited to one or the other
form caused a divergence between the two, than it is to posit that the system of
two different vowels arose independently and arbitrarily in numerous words.
That a historical process is in play is further suggested by the correspondence
between such plurals as Männer and Brüder in German and men and brethren
in English: it would be very unlikely that two different languages had in­de­
pend­ent­ly innovated their plural markers in the same way. So what happened
in English and German? How did they both come to have this vowel change in
the plurals of certain nouns?
Originally the nouns that would come to be subject to umlaut formed their

for mouse, for instance, would have been something like *mūs, and the plural
plural with an ending that had an i in it, whereas the singular didn’t.11 The word

*mūsiz. But speakers of a language sometimes anticipate a sound they’re about

to pronounce by altering an earlier sound to make it closer to the sound that’s
coming along after it. In this case, when pronouncing the ū, speakers anticipated
the i in the next syllable by bringing the tongue further forward in the mouth

11  For a more detailed, but still relatively accessible, historical account of the origins of umlaut in
English, see R. Lass, Old English: A Historical Linguistic Companion (Cambridge, 1994), pp. 59–71.
 Old English and Germanic languages 105

than it would have been for a normal ū. This change essentially matches the
description of the pronunciation of Brüder given above: the lips stay rounded,
as if to pronounce brood, but the tongue moves forward to the position where

also seen in the i of *mūsiz. This new vowel in the first syllable of *mūsiz is most
it needs to be to pronounce the vowel of breed—that is to say, the sort of vowel

form *mȳsiz, while the singular remained *mūs, since it didn’t have the ending
often transcribed as a y (or ȳ if a long vowel, as here), thus leading to a plural

with an i that caused the change of u to y.

At this point, the most salient marker of the plural would still have been
the ending, but this was about to change. Germanic, as it happens, is a language
branch that had a strong initial stress accent on most words, which caused the

had been a pair *mūs ~ *mȳsiz became simply mūs ~ mȳs, and it was now only
loss of many final syllables, especially those with short vowels. As a result, what

the vowel in the root of the word that served to distinguish the singular and
plural. (Once this change happens, the words have taken on the form they have
in Old English, so the asterisks can be dropped, as the forms are actually attested
rather than only reconstructed.) It wouldn’t be until the Great Vowel Shift12 at
the end of the Middle English period that these vowels changed to the diph-
thongs that would give rise to Modern English mouse and mice, but the struc-
tural change had already taken place: what used to be the plural ending was
lost, but not before it had had a chance to alter the main vowel of the word, so
there was still something there to indicate plurality even after the disappear-
ance of the vowel in the ending that had caused the change in the first place.
Plurals, moreover, are not the only place that umlaut of this sort has taken
place. Several other suffixes and endings that had once had an i or j in them
caused it as well. (Here j has its German value, i.e. English y, since the symbol y
is generally reserved to represent the umlauted ü vowel in this context.) As with
the plural marker, though, once the vowel had been affected by that i or j, the
latter sound was lost, leaving only the umlaut behind as a sign that it had once
been there. This leads to pairs in English where the historical relationship
between the vowels is no longer immediately clear, although in German it’s
often still visible because of the use of the umlaut as a diacritical mark. The
adjective old, for instance, has the regular comparative and superlative forms
older and oldest, but has also retained elder and eldest, which show e as the
outcome of umlauted o; this is more obvious in German where the cognate
adjective alt has älter and ältest- as the corresponding forms. Indeed, as with
plurals, German has retained the umlaut in more such adjectives than English
has: whereas long has only longer as a comparative, German lang forms länger,

12  The Great Vowel Shift is a complicated set of changes in (chiefly) the long vowels of English
that, in extremely rough terms, accounts for why our long vowels are pronounced as the letters
A E I O U rather than ah eh ee oh oo, as in most European languages.
106 How dead languages work

preserving a vowel change that Old English (lengra) and Middle English (lenger,
seen in the Prologue to the Canterbury Tales (330): Of his array telle I no lenger
tale) still had, but which Modern English has leveled away in favor of forms that
are more transparently related to the starting point long.
But in words where the umlauted form was no longer as closely connected

stabler even in English. In PIE, there was a suffix *-ti- that was used to form
with the base form as is the case with comparatives and plurals, umlaut proved

abstract nouns. As already seen,13 it became -si- in Greek, leading to English

borrowings like the-si-s and sta-si-s, and it acquired the additional suffix -ōn- in
Latin, thus giving rise to fac-ti-on and sta-ti-on (built, as it happens, to the same
roots as thesis and stasis respectively). Because of Grimm’s Law, the inherited t
turned into a th in Germanic, and the i was lost, but not before it had caused
umlaut: as a result, the abstract noun formed from long is length, that formed
from hale is health, and that formed from foul is filth. Incidentally, the last
example offers another good example of Grimm’s Law in play: the initial f in
English foul (Old English fūl) leads us to expect Latin cognates that start with a

*pū- “to stink”. One final set of forms that shows umlaut is a group of verbs
p, which is exactly what we find in putrid and pus, all formed from a PIE root

formed with a suffix whose original shape in Germanic was *-jan. It had a
couple of main functions, chiefly (i) to turn intransitive verbs (verbs that don’t
take a direct object, like to fall, to sit, to lie) into causative verbs (verbs that
mean, relative to the base form, to cause something to X or to make something
X), and (ii) to form verbs from nouns and adjectives. Here too, the suffix that
caused the umlaut in the first place was lost, but the umlaut remained, leaving
us with pairs like:

intransitive to fall → causative to fell (= to make something fall)

intransitive to sit → causative to set (= to make something sit)
intransitive to lie → causative to lay (= to make something lie)
noun blood → verb to bleed
adjective full → verb to fill

As far as the operation of the umlaut is concerned, that fall should form fell is
straightforward enough (it’s a vowel change along the same lines as that which
caused man to have men as its plural), and the same is true of blood ~ bleed,
which, at least graphically, shows the same pattern as foot ~ feet or goose ~ geese.
And the vowel shift that led to fill is roughly the same as that which led to filth.
But set and lay require further explanation: if anything, sit looks as if it should
be the form with umlaut, not set, since umlaut should be moving vowels towards
i, not away from it. As for lie and lay, a notoriously difficult pair for native
speakers to keep straight, if we strip away the effects of the Great Vowel Shift

13  See the discussion of abstract nouns in Thucydides in Chapter 2.

 Old English and Germanic languages 107

and other later changes and compare the corresponding Old English forms
licgan (pronounced “lidge-on”) and lecgan (“ledge-on”), we see that they once
followed the same pattern as sit and set. In seeking an explanation for this
apparent reversal of the workings of umlaut, it’s important to bear in mind a
further fact about the base verbs sit and lie: they were already subject to a dif-
ferent, earlier vowel alternation (ultimately going back to PIE times) that meant
that their past tense was formed with a different vowel from the present stem.
Thus the past tenses of intransitive sit and lie are sat and lay in Modern English,
and were, in the singular, sæt and læġ in Old English (sæt was pronounced
pretty much exactly like the modern word, and læġ had that same vowel, but the
final ġ was pronounced less like a g and more like the modern y).14 In short,
what’s happening here is that the vowel changes caused by umlaut are not being
applied to the present-tense stem but to something resembling the past-tense
stem: set may not look like an umlauted form compared to sit, but it does
when compared to sat. And the same is ultimately true of lay as well. Transitive
lay (of to lay the book on the table) goes back to the umlauted form of the intransi-
tive past lay (that is, the lay of the book just lay there on the table): there’s no sign
of umlaut left in Modern English, but the two forms were lecgan (with the vowel
of ledge) and læġ (with the vowel of lack) respectively in Old English, making it
clearer that umlaut has taken place in the former but not the latter. (Considering
how complicated all this is, English speakers can be forgiven for being confused
by their own language on this particular point.) But why in the world were
vowel alternations used to form the past tenses of some verbs in the first place?
To answer this question, we need to turn to a second characteristic feature of
the Germanic languages.

VERBS, STRONG AND WEAK

While umlaut is widespread in the Germanic languages, it differs from Grimm’s

Law in that it’s not a diagnostic feature: it’s a phonologically natural change, and
so, rather than being something that took place only once at the particular time
when the Germanic speakers separated from the other branches of Indo-
European (hence, diagnostic), it happened later on, independently, in slightly
different ways in each of the Germanic languages that it affects. Now in deter-
mining whether or not a group of languages really counts as a distinct branch
of a linguistic family tree, one really wants more than just a single distinguish-
ing feature, like Grimm’s Law. In the case of Germanic, there are several that we
could mention, but we’ll focus on just one, as it represents a different part of the
linguistic system—morphology rather than phonology—and is particularly

14  For more on the use of the dot over the g, see note 28.
108 How dead languages work

prominent in all the major representatives of the group still spoken today: the
reorganization of the inherited verbal system of Indo-European into two main
patterns, strong and weak verbs.
In this context, “strong” and “weak” do not have anything to do with strength
in a literal sense but rather relate to the regularity, frequency, and age of the two
patterns in question. All three of these factors are in fact connected with one
another because of a common and natural sequence of language change:
(1) a language creates or inflects words according to a regular pattern;
(2) something, often a sound change, happens to disrupt the pattern;
(3) a new, regular pattern is created, often by a process called leveling,
in which a single ending or pattern is extended beyond its original
domain;
(4a) old words that are used frequently continue to inflect according to the
old, now irregular pattern, because they’re common enough that learners
can cope with the irregularity;
(4b) but old words that weren’t used frequently, and all new words, adopt the
new pattern.
This is, in a nutshell, why the verb “to be” is so often irregular in the world’s
languages: because it’s usually the most common verb, children have plenty of
opportunity to hear the irregular forms, and in turn they soon adopt them as
well. In the end, regularity, age, and frequency come to relate to one another as
follows: an irregular pattern in a language is likely to be older than a regular
pattern; and it is likely to occur with words that are used more frequently than
is the regular pattern. In what follows, I adopt the traditional nomenclature of
calling the older, more irregular pattern “strong”, and the newer, more regular
one “weak”. One important corollary of all this is that historical linguists are
very much drawn to whatever is irregular in a language because it often pro-
vides a glimpse at an earlier stage in the language when such forms were still
regular. Generally, language users don’t create irregularity spontaneously; but
they are certainly happy to retain it.
With that background, consider the principal parts—present stem, past tense,
and past participle—of two English verbs:

strong verb: I sing I sang I have sung

weak verb: I lip-synch I lip-synched I have lip-synched

In the first verb, the different stems are formed by changing the vowel of the
verbal root; in the second, they are formed by adding the suffix -ed. (Since this
suffix is pronounced as either a t or a d depending on the preceding sound, it is
usually called the dental suffix to capture the fact that it can be either con­son­
ant.) In keeping with the general principle described above, the first verb is
 Old English and Germanic languages 109

older, more common, and, at least by the standards of Modern English, more
irregular; the latter is younger, less common, and forms its past tense and past
participle in a predictable way. This is not to say that sing ~ sang ~ sung is com-
pletely unparalleled, of course: it matches ring ~ rang ~ rung and drink ~ drank
~ drunk.15 But newer verbs of a similar structure, like to ping, will almost never16
be coopted into this pattern, but instead form regular ping ~ pinged ~ pinged.
Now this does not mean that all weak verbs are new and infrequent: plenty
of them are already common in Old English: ask ~ asked, gather ~ gathered,
hear ~ heard (in their Old English forms ascian ~ ascode, gaderian ~ gaderode,
hı̄eran ~ hı̄erde). What’s more, they couldn’t very well be a shared feature of the
Germanic languages if they didn’t date back even further, to the time when a
single Proto-Germanic linguistic community was still intact. With hear ~ heard,
for instance, one can compare not only Modern High German hören ~ hörte
and Dutch horen ~ hoorde but also Old Norse heyra ~ heyrði (where ð repre-
sents the voiced th of English the) and Gothic hausjan ~ hausida. Because it is
found so widely, the innovation must have taken place before all these languages
went their separate ways.
By contrast, widespread as this way of forming the past tense is in all the old
Germanic languages, one never sees it outside Germanic—unlike the alternations
of sing ~ sang ~ sung, which do have parallels in other Indo-European branches.
These three forms, for instance, derive from the same inherited pattern as
Ancient Greek leip-ō “I leave”, le-loip-a “I have left”, e-lip-on “I left”. Comparing
what happens in Greek with Germanic, it’s easy to understand why speakers of
the Germanic languages would have decided they needed a new way of form-
ing the past tense: Greek largely kept the diphthongs of PIE intact, so it was
relatively clear that the pattern that was inherited involved the alternation of
an e, an o, and ‘zero’, in e.g. ei ~ oi ~ i (as in leipō, above) or eu ~ ou ~ u.17 In
Germanic, however, these diphthongs underwent considerable phonological

15  For that matter, it also matches sink ~ sank ~ sunk, whose first principal part is hom­oph­
onous with the second syllable of lip-synch, and which could thus have provided an analogical
starting point for lip-synch ~ †lip-sanch ~ †lip-sunch. That the neologism has not been coopted
into this pattern is an indication of the general resistance of English towards forming new strong
verbs, even when there’s a clear model.
16  This being linguistics rather than mathematics, there are of course some exceptions: the
American Heritage Dictionary’s usage note under dive mentions dove, wore, and spat as examples
of strong past tenses built to verbs that had previously formed a weak past tense—indeed, dived
is still considered acceptable by most of their Usage Panel, and is the regular form in UK
English. To these we may add shat as a past tense and participle of shit, which only came into
existence in the nineteenth century on the model of sit ~ sat. (While, as is so often the case, the
exact development in the Middle English period is complicated, if it had retained its earlier
Germanic form to the present day, it would probably be shite ~ shote ~ shitten, following the
pattern of ride or drive.)
17  This type of vowel alternation that was present already in PIE is called ablaut and is not to be
confused with umlaut.
110 How dead languages work

change, especially as one moves forward to the attested languages, like Old
English, where these patterns developed (for the most part) as follows, to give
just three variations on what was once a single pattern:

Proto-Indo-European Old English Modern English

ei ~ oi ~ i ı̄ ~ ā ~ i ride ~ rode ~ ridden
eu ~ ou ~ u ēo ~ ēa ~ o/u freeze ~ froze ~ frozen
en ~ on ~ n̥ 18 in ~ an ~ un sing ~ sang ~ sung

In the left column, everything is tidy, but, already by the time of the middle
column, what had previously been a unified pattern has splintered into vari-
ous subclasses, and the fact that so many of these verbs still exist in Modern
English is powerful evidence for just how conservative languages can be at
retaining inflections that haven’t really made good sense for perhaps two
thousand years. Nor is this persistence of the strong verbs limited to English.
One has to learn principal parts comparable to sing ~ sang ~ sung in the other
Germanic languages too: German singen ~ sang ~ ge-sungen, Dutch zingen ~
zong ~ ge-zongen, Norwegian synge ~ sang ~ sunget. But these irregularities
are opaque enough even for fluent speakers of the language that it is no sur-
prise that, when a new verb is created—say, to paralyze—it is not especially
tempting to inflect it as paralyze ~ paraloze ~ paralizzen, and one defaults to
the weak forms instead.
One might further wonder how such patterns of tense formation arise in the
first place. While the vowel alternations of the strong verbs go back to Indo-
European times, and their origin is therefore still a subject of dispute, the dental
suffix of the Germanic weak verbs has arisen through a process that is rather
better understood, partly because it came about more recently and so is some-
what more transparent, partly because it has parallels in languages whose his-
tory is actually attested. While exploring the origins of the past tense in ‑ed will
require a certain amount of digression, it’s very much worthwhile because the
history of this ending will serve so well as a model example of how languages
acquire endings in the first place.
First, note that a particularly common way of expressing a tense for which a
language doesn’t have special endings is to employ a helping verb. We don’t
have separate future or present perfect forms in English, so we have to use
so-called periphrastic constructions like She will sing tomorrow or He has
walked to the store, in which the verbs will and has are pressed into service to
mark these tenses. What’s more, these helping verbs are subject to phonological

18  In PIE, an n could be treated as a vowel, pronounced much as it is in English words like
button and written, where the o and e before the n have ceased to represent distinct vowels. In such
cases, Indo-Europeanists write it with a small ring underneath to indicate that it is a vocalic n.
 Old English and Germanic languages 111

weakening—that is, loss of stress—that makes them a little less like a separate
word and a little more like an inflection, since they lose their main vowel and
require a host word to attach to, thus yielding She’ll sing tomorrow and He’s
walked to the store. Now, in English, the full forms can still be used for em­phasis,
so one would not want to say that a complete fusion of pronoun and helping
verb had taken place. But in everyday speech, the contracted forms are much
more common, and we’re not that far from a linguistic situation in which, if
we were drawing up the rules of English spelling from scratch, it might be
tempting to mark these tenses differently than we do now, to show how what
used to be an independent verb is starting to work more like an affix. For
instance, one could indicate the future tense in English by inserting an l in
between the subject pronoun and the present stem of the verb (and dropping
the final -s in the third-person singular): aising “I sing” → ailsing “I’ll sing”,
yusing → yulsing, ­wising → wilsing, but, in the third singular, shisingz → shils-
ing. Similarly one could rewrite the present perfect of weak verbs as being
formed through the insertion of a v between the subject and the past stem,
which again would change in the third-person singular, this time to a z. That
is, instead of pairs like I walked and I’ve walked, we’d have aiwakt ~ aivwakt,
yuwakt ~ yuvwakt, wiwakt ~ wivwakt, but third singular hiwakt ~ hizwakt.
At the end of the day, treating such subject (“I”) + auxiliary verb (“’ve”) +
main verb (“walked”) sequences as single words is unlikely to be the best
analysis—one can insert adverbs, for instance, between the subject–auxiliary
complex and the main verb (He’s always walked to the store), showing that
the auxiliary is not bound closely enough to the main verb to count as a
prefix the same way that the -ed counts as a suffix: one can separate ’s from
walked, but not walk from -ed. But it certainly looks as if one day the auxil-
iary could become bound to the main verb, if English happened to progress
in that direction.
In fact, just such a development has given rise to the future tense of the Romance
languages. Here are the infinitives of the Spanish, French, and Portuguese verbs
“to love”, as well as the singular forms of the future tense:

  Spanish French Portuguese

to love amar aimer amar
I will love amar-é aimer-ai amar-ei
you will love amar-ás aimer-as amar-ás
s/he will love amar-á aimer-a amar-á

In all three languages, the future can thus easily be analyzed as the addition
of endings to the infinitive. Those endings, as it happens, are the same as the
present-tense forms of the verb “to have”; the initial h- in the Spanish and
Portuguese forms is silent:
112 How dead languages work

  Spanish French Portuguese

to have haber avoir haver
I have he ai hei
you have has as hás
s/he has ha a há

It is thus as clear as anything in historical linguistics ever is that the future tense
of the Romance languages is not the direct outgrowth of the Latin future tense
(which for these forms was amābō, amābis, amābit), but of what in Latin would
have been a periphrastic construction amāre habeō, amāre habēs, amāre habet
(“I have to love, you have to love, s/he has to love”). In English, of course, “to
have to verb” indicates obligation or necessity rather than futurity. But it’s not
hard to see how, in a different language, the idea of “having” a verb could
instead suggest the capacity to do it or possibility thereof, thus leading to the
development of a future sense. What’s more, while the forms of “to have” have
completely fused with the infinitive in Spanish and French, formal European
Portuguese remains at an earlier stage in this process: unstressed personal object
pronouns can still be inserted in between the infinitive stem and the forms of
haver. Thus “he will love you”, in elevated speech, is amar-te-á, with the inser-
tion of a so-called mesoclitic pronoun, here te “you”. Finally it should also be
noted that haber/haver is only one of two verbs in Spanish and Portuguese that
mean “to have”. Much more common, as a full verb, is Spanish tener, Portuguese
ter. Here we see another common linguistic phenomenon in play: when a word
like habēre loses ground to an upstart like tenēre (originally “to hold”), one
context in which the older verb can survive is in grammatical constructions
like these. And the whole process whereby a word gradually goes from being a
full word with its own semantic force (“I have a car”), to an auxiliary word used
to mark a grammatical feature, like tense (“I have walked”), to an unstressed
unit that no longer counts as a full word, but, as a so-called enclitic, leans on
another (“I’ve walked”), to a ending that’s become fused with its host word
(“aimerai”) is what linguists call grammaticalization.
Armed with this example of grammaticalization taking place over a period
of time that’s left written records, we can now turn to endings of the older Indo-
European languages where we can only see the end result. First, take the Latin
future endings that were replaced by the construction with “to have” in the
Romance languages: amā-bō, amā-bis, amā-bit “I will love, you will love, s/he
will love” all look as if they could have arisen from the same sort of fusing of an
auxiliary verb (bō, bis, bit) with an unchanging verbal stem. That some sort of
b-initial verb was brought into play in the prehistory of Latin in order to form
the future is further suggested by the imperfect tense, which has a similar for-
mation: amā-bam, amā-bās, amā-bat “I was loving, you were loving, s/he was
loving”. (Unlike the Latin future, this tense survived rather better in Romance,
becoming Spanish amaba and Portuguese amava “I or s/he loved”.) Exactly
 Old English and Germanic languages 113

what verb this is likely to have been is comparatively easy to work out once one
observes that only a fairly limited number of verbs typically become used as
auxiliaries, many of which we can see at work in English: the basic verbs of
possession (I have done that), of motion (I’m going to do that), of volition (I will
do that), of action (I don’t know), and of being (I am doing that). In the case of
Latin, it’s the last class of auxiliary that is at work here: the -b- of the Latin
imperfect and future formations is, in origin, the same as the b- at the start of
English be.19 And to return now to the Germanic weak verbs, which led to this
whole excursus, here too there’s a good source for the -ed that marks their past
tense: do. Essentially, a past tense like I walked is the end result of the weaken-
ing of a structure comparable to I walk-did. That a fuller verb form ultimately
underlies the -ed suffix can be seen most clearly in the oldest Germanic lan-
guage to be attested in any sizable corpus, Gothic.20 The weak verb hausjan “to
hear” has the following past-tense conjugation:

  singular plural
1 hausi-da hausi-dēdum
2 hausi-dēs hausi-dēduþ
3 hausi-da hausi-dēdun

While the exact origin of this paradigm is beyond the scope of this book, the
plural endings, with their d-vowel-d structure, are excellent evidence that the
past tense of these Germanic verbs was originally an early form of did.21
With these two features, Grimm’s Law and the weak verbs, we now have a
good sense not only of what makes Germanic Germanic (shared innovations

19  The verbal root in question more or less had the form *bhū- in PIE (to simplify the picture,
I omit consideration of laryngeals here, in note 21, and in the discussion of the root *gen- below);
by Grimm’s Law, the *bh- regularly became a b in English (hence be), but it had two primary out-
comes in Latin depending on the environment: f- at the start of the word (leading to the Latin
perfect tense fuı̄ “I was, I have been” as well as the future participle futūrus, borrowed into English
as future), but -b- in the middle of the word (hence amābam and amābō).
20  Sadly, the corpus consists almost exclusively of a translation of the Bible made by Wulfila,
missionary to the Goths in the fourth century ad. Well, most of the Bible: he left out Kings
because it had so much military material, and he figured his audience needed little encourage-
ment on that front.
21  For the details, see D. Ringe, From Proto-Indo-European to Proto-Germanic (Oxford, 2006),
pp. 166–8. The sharp-eyed may well ask: if did is what became the -ed suffix, then what was the
source of the final -d in did itself? As it happens, do is not historically a weak verb, even though it
appears to form its past tense with a dental suffix synchronically within present-day English. In
fact, the final -d of did is actually the initial d of the root, and the di- at the start of the form goes
back to a reduplicating syllable used to form the present and imperfect tenses of certain PIE verbs
(see Ringe again, pp. 157–60); thus, from the root dō- “give”, Greek forms di-dō-mi “I give”, with

Since both English do and Greek tithēmi “I place” go back to the PIE root *dhē-, the past tense did
the initial consonant of the root (the d of dō-) copied at the start of a prefixed syllable (the d of di-).

corresponds directly to the tith- of the (unaugmented) Greek imperfect tithēn “I was placing”.
Considering its frequency, we should not be surprised that did offers a relic of so archaic a form.
114 How dead languages work

like consonant changes and a new way to form the past tense) but also of some
of the most important concepts in historical linguistics: how sound changes
affect languages, and how new endings are formed. At this point, we can pro-
ceed to examples of our model Germanic language, Old English. After an initial
look at some of the morphology specific to Old English, we’ll first go through a
verse of the New Testament, which will show how Old English serves as a lin-
guistic bridge between its sister languages, Greek and Latin on the one hand,
and its daughter language, Modern English, on the other. Then, a reading from
Beowulf will illustrate some of the literary features characteristic of the older
Germanic languages, such as the use of compound nouns and alliteration,
which give this poetry its striking intensity.

OLD ENGLISH

For the speaker of Modern English, the languages covered in Chapters 2 and 3,
Greek and Latin, stand out as difficult in large part because of their compara-
tively rich morphology—all those endings to denote cases of nouns and tenses
of verbs. Even the other modern languages of Europe have for the most part
retained more complexity in this regard than English: in the Romance lan-
guages like Spanish and French, verbs still sport dozens of endings to denote
person, number, tense, and mood, and both German and Russian still inflect
their nouns and adjectives to mark case, number, and gender. Much of the
morphological wealth inherited from Indo-European has been lost in these
languages too, but English has gone the furthest in reducing the number of forms
speakers have to master, relying instead on word order, prepositions, obligatory
subject pronouns, and auxiliary verbs to signal what used to be expressed
through endings. In this general arc of linguistic change, Old English—the
form of the language spoken between the fifth century ad, when Germanic
speakers first came to Britain in large numbers, and (more or less) 1066, when
the Normans drastically altered the linguistic landscape—occupies a usefully
intermediate position. In the Old English period, certain changes had taken
place that distinguished English from its close Germanic kin, but it had not yet
jettisoned the principle that most nouns would have a half-dozen different
forms, and most verbs a couple dozen, to show what they were doing syn­tac­tic­
al­ly in a sentence. In this respect, it was only moderately more complex than
modern German.
Take the word for “stone”, stān. (The Great Vowel Shift had not yet taken
place, so the long ā of stān had roughly the same quality as the vowel of don.)
Whereas the modern word stone only has two forms in speech (stone, which
serves as the basic singular form, and stones, which is not only the basic plural
form but also the possessive of both singular and plural, differentiated only in
 Old English and Germanic languages 115

writing as stone’s and stones’), the Old English word had no fewer than six dis-
tinct forms:22

  singular plural
nominative-accusative (subject or direct object)23 stān stānas
genitive (possessive) stānes stāna
dative (indirect object) stāne stānum24

This represents, on the one hand, a considerable decline in the number of forms
relative to its PIE ancestor; for the corresponding declension (pattern of case
endings) in Sanskrit, even ignoring the preservation of a dual number, there are
eight cases, each with a distinct form in the singular and with six different
forms in the plural. With 14 different forms at its disposal, Sanskrit could thus
enjoy a more flexible word order and dispense with prepositions to a much
greater extent than Old English. For example, rather than needing special
words to express the instrument with which something happened or the place
from which something went, it could simply put what would in English be the
object of the preposition into the appropriate case (the instrumental and the
ablative, respectively).
But while the Germanic languages arose from a parent language that was as
rich in endings as Sanskrit, another of their characteristic features is that they
had a strong initial stress accent, which in many instances led to the weakening

al nominative singular ending in this declension, *-os, which became -us in

of syllables later in the word, especially in the endings. For instance, the ori­gin­

Latin and survives to this day as -os in Greek, weakened simply to -s in Gothic
and -r in Old Norse, and, as evidenced by stān, was lost entirely in Old English.
Phonological reduction of this sort not only leads to the loss of individual
inflections but can also undermine the case system altogether, especially if
prepositions and a more rigid word order are already being employed on an
optional basis to clarify the functions of ailing case endings.25 Certainly, the
trend continued in Middle English. By around 1150, further reduction of final
syllables had weakened the system as seen in Old English stān still further:

22  See also the discussion of noun morphology in Chapter 2.

23  In this particular declension, the Old English counterpart of the Latin and Greek second
declension, the nominative and accusative have the same form in both singular and plural, but
this is not the case for all declensions, so they still need to be set up as distinct cases.
24  The dative plural ending, -um, which is found in all declensions, is historically the same
as the corresponding Modern High German ending -en, the most consistently preserved noun
inflection in the language.

instance, in the Latin second declension, when the original accusative singular ending *-om and
25  That said, phonological reduction does not necessarily lead to the end of a case system. For

genitive plural ending *-ōm both merged as -um, Latin speakers didn’t give up the inflectional
distinction. Rather, they created a new genitive plural ending ‑ōrum (on the basis of the feminine
ending ‑ārum), thereby maintaining a lively case system.
116 How dead languages work

as nearly all the vowels in these unstressed endings turned into schwa (the
unstressed neutral vowel of upon), -es and -as merged as -es, -e and -a as ‑e, and
-um became -en, leaving only four distinct forms, which would require only a
little more simplification before yielding the minimalist paradigm of Modern
English.26 (Of course, it is not as minimalist as it could be: some languages, like
Chinese, make do without even having a plural ending for nouns.)
The situation is similar for other parts of speech too, with adjectives, pro-
nouns, and verbs all having a greater variety of endings in Old English than
in the modern language. Take just the present active indicative of strong
verbs. In Latin or Ancient Greek (again ignoring the dual), verbs have six
different forms—a 2×3 array for singular and plural, and the first, second,
and third persons. Old English, however, has reduced this to four forms, with
the three persons distinguished in the singular, but all three persons using
the same ending in the plural. Thus, with Latin canō “I sing”, compare Old
English singan:

  singular plural singular plural

1 canō canimus iċ singe wē singaþ
2 canis canitis þū sing(e)st ġē singaþ
3 canit canunt hē sing(e)þ hı̄e singaþ

In Latin the endings are distinct for each slot in the paradigm, so there’s no
need for subject pronouns: you don’t have to say ego canō or tū canis if the -ō
and -is already indicate that the subjects are “I” and “you” respectively. One can
add the pronouns as a supplement but, because they’re not needed to distin-
guish who’s doing the singing, they generally only occur to mark emphasis
(“I’m singing, but you’re only lip-synching”). If one confined oneself to the
singular, the same would be true of Old English, but the plural has entirely lost
the differentiation of the three persons: if one simply said singaþ,27 it would be
unclear whether we, y’all, or they were singing, so the pronouns wē, ġē,28 and
hı̄e need to be deployed in addition. By the time we get to Modern English, the
endings have been leveled out even more, with only the third singular marked
by a special ending, -s, and the rest left unmarked. While the exact sequence
whereby the endings are reduced and the subject pronouns become correspond-
ingly more obligatory is too complicated to be detailed here, it is worth compar-

26  For an accessible account of the loss of case endings in Old and Middle English, see R. Lass,
Old English: A Historical Linguistic Companion (Cambridge, 1994), pp. 228–40, 250–2.
27  As noted earlier, the final letter in this word is called thorn and is pronounced like the th
of thin.
28  The dot over the g in ġē indicates that this is a palatalized g, well on its way to becoming the
y of its Early Modern English descendant, ye. We will return to this in greater detail in the discus-
sion of alliteration; for now, it’s sufficient to note that the word would have been pronounced
roughly like Modern English yay. Palatalization is also signaled by the dot over the c in iċ: the c
should be pronounced like Modern English ch, and the word as a whole somewhat like itch.
 Old English and Germanic languages 117

ing a similar development in the Romance languages: in Spanish and Italian,

where verb endings have remained more differentiated, the subject pronouns
may be omitted, but in French, which has under most circumstances reduced
the number of distinct forms to three (in the spoken language at least), the sub-
ject pronouns have become obligatory.
But, while Old English thus differs structurally from Modern English in hav-
ing a lot more endings, what really makes it seem as foreign as it does is a
combination of different spelling conventions and the shift in vocabulary over
the past ten or twelve centuries. Take, for instance, the version of John 3:16
found in the tenth-century Wessex Gospels:29
God lufode middaneard, swā þæt hē sealde his āncennedan Sunu, þæt nān ne
forweorðe þe on hine ġelyfð, ac hæbbe þæt ēċe lı̄f.
For God so loved the world, that he gave his only begotten Son, that whosoever
believeth in him should not perish, but have everlasting life. (KJV, 1611)
Consider for now just the first ten words in Old English. All but one of them,
āncennedan, has a fairly clear Modern English descendant, and we could rewrite
it as follows:
God loved mid-yard, so that he sold his ‘one-kind’ Son
In rewriting it to update the spelling (and, with it, the phonology) to modern
standards, we’ve come a long way towards making it look more familiar; seven
of the ten words map fairly closely onto their current descendants:
(i) God and his have stayed the same, and hē has simply lost a mark of
vowel length.30
(ii) When one recalls that final unstressed vowels are lost, and if one also
knows that an f between two vowels was pronounced as a v, then it’s not
hard to see how one gets from lufode and Sunu to loved and Son.
(iii) The form þæt, to be sure, looks alien to the uninitiated, but as long as
one recalls that that initial letter, thorn, yielded the Modern English th,
and that the æ digraph, called ash, is pronounced like the a of ash, then
it becomes self-evident that this is modern that.
(iv) Updating swā to so is a less obvious move, but it fits in with a broader
change whereby Old English ā generally becomes a long o in Modern
English, as seen above in the shift from stān to stone.

29  Thanks to the educational program of Alfred the Great (king of Wessex from 871 to 899), the
West Saxon dialect of inter alia the Wessex Gospels became the closest thing to a standard, “clas-
sical” dialect of Old English.
30  And even this change is an artificial one: long vowels were mostly left unmarked in manu-
scripts of the time, and it is only the convention of modern linguists to indicate them regularly in
citing Old English forms.
118 How dead languages work

With the remaining three words, however, sealde, middaneard, and āncennedan,
we see larger shifts in the language over the past millennium:
(i) Drop the final -e, and sealde is only a vowel away from its modern coun-
terpart sold, but there’s also been an important change in sense, too:
Old English sellan meant simply to give something away, whereas
Modern English sell entails the receipt of money in exchange.
(ii) Middaneard is a compound of middan and eard. The first half is com­
parable to the modern prefix mid-, but it is less clear what exactly eard
represents, since this second element was subject to variation within
Old English between eard and ġeard. The latter variant, with the word
that gave rise to Modern English yard (showing the same change of ġ to
y seen above, with the pronoun ġē becoming ye), shows especially well
the close cultural relationship between Old English and its sister lan-
guage Old Norse. In Norse mythology, the world as inhabited by ordin­
ary people was known as Miðgarðr, generally anglicized to Midgard.31
It was called this because it was enclosed (a yard is, ultimately, an
­enclosure) in the middle of various other worlds that were not access­
ible to ordinary mortals. But, as a term referring to the world of men,
the second element of Old English middanġeard was later on reinter-
preted as earth, giving rise to the term Middle Earth, popularized by
J. R. R. Tolkien, himself a scholar of the early Germanic languages.
In  either case, the Old English version of John 3:16 comes across as
decidedly more Germanic once one recognizes that the word for “world”
is virtually the same as a key place name in one or the other of these two
mythologies, the Vikings’ and Tolkien’s.
(iii) Finally, the word with the least obvious connections to Modern English
is āncennedan “only begotten”. But just because the connections aren’t
as obvious doesn’t mean they’re not there. It’s helpful here to compare
the corresponding word in the Latin Vulgate translation, ūnigenitum,
familiar to singers from its appearance in the Credo of the Mass. The
Old English and Latin are almost perfect counterparts. Both begin with

back to an original Indo-European word *oinos. While the phonetic

a form of the word for “one”, ān in Old English, ūnus in Latin, both going

development from Old English ān to Modern one is complicated, it at

least graphically parallels that of stān into stone. And the second elem­
ent in both is a past participle of a word meaning “to beget”, whose root
takes the shape gen- in Latin and cen- in Old English (with a hard c).

31  The Old Norse letter ð, known as eth or edh, had the voiced th sound of the, but is generally
turned into a d in English versions of Norse names, as also in Odin, who in Old Norse is Óðinn.
As for the final -r of Miðgarðr—and, for that matter, the second -n of Óðinn—this is a common
nominative singular ending in Old Norse, cognate with Latin -us, as noted above.
 Old English and Germanic languages 119

Once again, these both go back to the same Indo-European root, *gen-,32
with the Latin preserving the inherited form, but Old English, as a
Germanic language, changing the g to a c, in keeping with Grimm’s Law.
While no verb ken meaning “to beget” survives in Modern English—
according to the OED (s.v. †ken, v.2), it didn’t make it past the Middle
English period—other relatives, like kin and kind, are still around.
At this point, we can take a step back and summarize why even a fairly simple
sentence of Old English is so foreign to a speaker of the modern language. As a
reminder, here are the relevant Old English and (Early) Modern English versions:
God lufode middaneard, swā þæt hē sealde his āncennedan Sunu
For God so loved the world, that he gave his only begotten Son
Yes, Old English has some endings that have been lost (final -e in lufode and
sealde, -an in āncennedan, and -u in Sunu), but what really makes it hard to
read is a combination of some comparatively superficial changes in how we
spell (þæt looks very different from its descendant that, but was pronounced
nearly the same), and some more substantial changes in the meanings of words
(sell no longer means “give”) and the loss of some words altogether (āncennedan).
Finally, before we move on to a more complicated passage of Old English, it’s
also worth noting just how uneven the rate of linguistic change was. While it’s
true that we don’t use the word begotten so much anymore (and indeed the
reason it’s still familiar at all is probably precisely that it occurs in this most
famous of Bible verses), all the relevant changes just detailed had taken place
before the translation of the King James Version (KJV) in 1611. English has
certainly not remained stationary since then—moving further on in the verse,
we see that early seventeenth-century English could still use -eth as a third-person
singular ending (with the final consonant of believeth matching that of Old
English ġelyfð)—but it remains the case that, in the last four hundred years,
which haven’t seen a historical disruption comparable to the Norman Conquest,
English has been a lot more stable than it was in the six hundred years before that.

BEOWULF

As a counterpart to the passages of the Iliad seen in the chapter on Greek, it is

fitting, in our second example of Old English, to draw on Beowulf, partly
because of its importance as a literary monument, and more specifically
32  The shared vowel e is deceptive: the Latin e is an inherited e, whereas the Old English e is the
outcome of umlauted a: the verb cennan was a causative verb (“to cause to be born”) and its e is
thus parallel with that of fell and set, as seen above in the discussion of causative verbs and umlaut.
As it happens, this root, in an alternate form that developed in Latin as (g)nā-, also led to the word
nātūra, as discussed in Chapter 3.
120 How dead languages work

because it too is centered on the deeds of a hero in a warlike society. At 3182

lines of verse, the poem is about one-fifth the length of the Iliad, and its dating
is similarly contentious: the manuscript dates to around ad 1000, but, as some
linguistic features belong to an earlier period of the language, and Christianity
is alluded to in ways that suggest it has not yet fully taken hold in England, the
poem is perhaps to be dated to the eighth century. While the poem’s main char-
acter, Beowulf himself, belonged to a Germanic tribe called the Geats, who
lived in what is now southern Sweden, the poem opens with a brief history of
the Danes—understandably enough, since the action begins with the attacks
of the monster Grendel on Heorot, the hall of the Danish king Hrothgar. From
that opening consider the following lines (20–5), first in Old English, then in
two translations, both E. Talbot Donaldson’s comparatively literal prose ver-
sion (placed to the side, and spaced out to make the lines of text correspond
approximately to those of the original) and, below, the verse translation of
Seamus Heaney:33

Swā sceal geong guma  gōde gewyrcean In this way a young man ought by his good deeds,
fromum feohgiftum  on fæder bearme by giving splendid gifts while still in his father’s house,
þæt hine on ylde  eft gewunigen to make sure that later in life beloved companions
wilgesı̄þas  þonne wı̄g cume, will stand by him, that people will serve him
lē ode gelǣ sten: lofdǣ dum sceal when war comes. Through deeds that bring praise,
in mǣ gþa gehwǣ re  man geþē on. a man shall prosper in every country. 

And a young prince must be prudent like that,

giving freely while his father lives
so that afterwards in age when fighting starts
steadfast companions will stand by him
and hold the line. Behaviour that’s admired
is the path to power among people everywhere.

In these lines, a commentary on the actions of Hrothgar’s grandfather Beow,

the poet describes the sort of behavior that the hero ought to follow if he is to
be praised. Now the importance of gift-giving to establish alliances that will
prove useful in war would certainly not be a strange concept to the audience of
the Iliad. But how does Old English express this? Compared to the excerpt
from the Gospel of John, this passage is even further removed from ordinary
Modern English. The only words to remain unchanged are man and the two
prepositions in and on (although the latter is used twice in the passage above
where Modern English would also use in). Two more words, swā and þæt, we’ve

33  The translations are taken from Beowulf: The Donaldson Translation, Backgrounds and
Sources, Criticism, edited by J. F. Tuso (Norton, New York, 1975); and from Beowulf, translated by
Seamus Heaney (copyright © 2000 by Seamus Heaney. Used by permission of W. W. Norton &
Company, Inc., and Faber & Faber Ltd). In citing the Old English of Beowulf, I omit the dots that
indicate palatalized ċ and ġ so as not to prejudice the subsequent discussion of alliteration.
 Old English and Germanic languages 121

already seen as the ancestors of so and that. And several more words have fairly
straightforward descendants and only look foreign because of changes in spell-
ing and pronunciation: sceal > shall, geong > young, gōde > good, fæder > father,
cume > come. But that still leaves a considerable number of words that have
either been lost altogether or else have undergone major shifts in form or
meaning. Consider the passage again line by line, with an interlinear transla-
tion and commentary to explain the relationship between the Old English and
the modern language.

Swā sceal geong guma gōde gewyrcean

So shall (a) young man (with) good bring about

While sceal is the direct ancestor of modern shall, there has been a shift in
meaning: in Old English, it expressed obligation rather than futurity, hence the
translators’ ought and must.
While it might not be obvious that geong would eventually develop into
modern “young”, the letter g before the vowels e and i would eventually soften
to the Modern English y in most instances (as in ġē > ye, seen above). Note also
that Old English has no need for the indefinite article. As the glossing of gōde
indicates, it can also dispense with the preposition with here because the case
ending -e is sufficient to indicate that the “good” is regarded as the instrument
with which the young man brings something about.
guma is a word for “man” that has largely been lost, but survives in an altered
form as the second element in bridegroom, which, in the Middle English period,
was still bridegome.
gewyrcean: While most Germanic words are stressed on the initial syllable,
there are some unstressed prefixes, of which ge- is the most common. It sur-
vives as such in German and Dutch, where it is chiefly used to mark the past
participle, but was almost entirely lost in English during the Middle English
period, by which point it had been reduced merely to i- or y-, as in the medieval
song Sumer is icumen in “Summer has come in” and in the word yclept “called”,
a favorite of archaizing poets.34 But it was also used to mark more generally that
an event was brought to completion (in technical terms, it was a perfectivizing
suffix), a function derived from its earlier concrete sense “together”. This is a
natural enough shift in meaning, at least to judge from the fact that the Latin
cognate of this prefix underwent the same development: as a preposition (cum),
it meant “(together) with”, but as a verbal prefix (com-, con-), it could either

34  That German is closer to Middle English than to the modern language is seen especially
nicely in its equivalent for Sumer is icumen: Sommer ist gekommen. Not only does it share with
Middle English the retention of the prefix ge-/i-, it also uses the equivalent of is, rather than has,
to form the present perfect tense of an intransitive verb of motion.
122 How dead languages work

mean “together” or be used simply to indicate that an activity was carried out
in its entirety.35 For the former meaning, one may contrast ferō “bring” with
conferō “bring together, collect”; for the latter, one may compare edō “eat” and
comedō “eat up, consume”, which in Spanish and Portuguese, as comer, eventu-
ally replaced the shorter (and irregular) edō as the basic verb for eating. Indeed,
the prefix could also serve both ends simultaneously, which is how it slid from
the spatial to the more abstract meaning in the first place: nectō already means
“tie, fasten” on its own, and adding the prefix to form connectō arguably empha-
sizes either that two items are being tied together or, simply, that they have been
completely connected.
To return, then, to Old English gewyrcean: it is composed simply of the ge-
prefix (here indicating completion), the root wyrc- “work”, and an infinitive
ending -ean. And, as it happens, this form is closely related to one of the few
scattered words where the prefix survives in Modern English: one rarely gives
much thought to why there’s an -i- in handiwork—it sounds natural enough
given the existence of the adjective handy—but it is in fact the outcome of this
same prefix, still there in all its glory in its Old English ancestor, handgeweorc.
Nor is this the only word to preserve the prefix: it’s also present in e-nough (a
word well suited for retaining a prefix signifying completion), which in Old
English was still genōg, clearly cognate with the modern German equivalent,
genug.

fromum feoh-giftum on fæder bearme

(with) rich, abundant money-gifts in (his) father’s lap, bosom

Once again the bracketing of “with” and “his” in the glossing of the line shows
just how compact Old English can be compared to the modern language. As in
the previous line, it can do without the preposition because this is indicated
sufficiently with the dative plural ending ‑um (the same as German -en) which
links together both the adjective from and the noun feohgift. And de­ter­miners—
the various articles and possessive pronouns that precede most singular nouns
in Modern English—are also more optional, as seen in the absence of any word
for “his” with fæder. Even the ’s of father’s doesn’t exactly have an equivalent:
normally, in Old English, possession of this sort would be signaled with the
genitive ending, and, in a sense, that’s true here too, except that the genitive of
fæder happens to be the same as the nominative, so one simply has to work out
the syntax from context.

35  According to the rules given above in the discussion of Grimm’s Law, a Latin c should cor-
respond to a Germanic h, but depending on the location of the accent in the ancestral word in PIE,
the same inherited sound could also develop into a g in Germanic; this refinement of Grimm’s
Law is called Verner’s Law.
 Old English and Germanic languages 123

The line also offers two good examples of how the simple loss of words can
make Old English look alien: fromum is a form of the adjective from (both
“firm, strong” and “rich, abundant”) which has disappeared from the modern
language. The cognate adjective fromm “pious” is still found in German, whose
Yiddish counterpart frum has been borrowed into English in the same sense.
That might seem a different enough meaning as to call the relationship between

ring to forward motion, that had forms beginning *pr- and *pro- in PIE—but
the two into question, but both belong to a large group of words, chiefly refer-

fur-, far‑, and fro- in Germanic, once Grimm’s Law had changed the original *p
into an f. Thus words like further, farther, and forwards share their first two
consonants, etymologically speaking, with the Latin prefix pro-, so common in
English derivatives like promote and proceed. Old English from and German
fromm belong here as well: ultimately, the Germanic adjective simply meant
something like “that which moves forward, that which furthers you”. The
other word which has simply been lost from the language is bearme, the dative
singular (with the case ending indicating that it is the object of the spatial
preposition on) of a now obsolete word barm “lap, bosom”, last attested as an
independent noun in 1522, according to the OED.
But the most typically Old English word in the line is feohgiftum “money-
gifts”. The noun feoh (one source, along with a cognate word used by the

*peku-, that could be used either of livestock or of the wealth that such animals
Normans, of Modern English fee) is derived from an Indo-European root,

represented. Once again, an initial f corresponds to a p in Latin, and (as with

the horn : corn- and heart : cord- equations), Germanic h to Latin c, yielding a
clear relationship between feoh and pecū‑nia “money”. By way of Italian, the
concrete sense of the root is also still in evidence in pecorino cheese, made of
sheep’s milk (< Latin pecus “livestock, esp. sheep”). In any case, the idea of gift-
exchange as a means of establishing mutual networks of support was central to
Germanic society—as, indeed, it was to other early Indo-Europeans too.
More importantly from a linguistic standpoint, feohgiftum is a compound
noun, a type of formation that, while inherited from Indo-European, was to be
especially productive in the Germanic languages. This is most obvious in the
long compound words of German, but even English is more Germanic in this
respect than it would appear at first glance. Consider how one says “string
quartet” in the following languages:

German: Streichquartett
English: string quartet
French: quatuor à cordes
Italian: quartetto d’archi

True to its reputation, German can express the concept in one long word, con-
sisting simply of the word for the stringed instruments (literally “stroke”, referring
124 How dead languages work

to the action of the bow) and that for quartet written together as one. This is, in
effect, the same way that English handles the expression, the only difference
being the superficial convention of orthography that English puts a space
between the two constituent components. Contrast the Romance languages,
which are less amenable to simply juxtaposing two words in order to form a
compound: French speaks of a “quartet for strings”, Italian of a “quartet of
bows”. While the simple act of spacing out the two words may make English
appear a little less forbidding, it in fact stays very close to its Germanic roots
here. Such compound nouns are especially common in the language of poetry,
with the next few lines of Beowulf also offering wilgesı̄þas “pleasant-companions”
and lofdǣdum “praise-deeds”. While the compounds just cited all mean exactly
what you’d expect from their component parts, the proclivity of Germanic
speakers to link words together in this way also helped lead to the widespread
use of kennings in their poetry: that is, compound expressions which contain
a compressed metaphor, such as calling the body a bānhūs “bone-house” or
blood heaþoswāt “battle-sweat”. As these last two examples suggest, the mean-
ing of the word is not necessarily clear at first—one could also imagine a bānhūs
referring to an ossuary (such, indeed, is the meaning of the German cognate
Beinhaus), or heaþoswāt to literal sweat—but, as coinages, they have become
fixed in these particular senses (bānhūs occurs twice in Beowulf, heaþoswāt
three times).

þæt hine on ylde eft gewunigen

that him in old age afterwards will stand by

In this line, nearly all the words have a modern relative that is clear, at least once
they have been glossed, leaving only gewunigen to be explained. The root of the
verb here is simply wun-, the same as found in German wohnen “to live (in a
place), dwell”, and both Old English wunian and prefixed gewunian can have
this same sense “to dwell, stay, remain”.36 Here, though, with hine as a direct
object, it takes on the stronger meaning “to stay firmly behind someone”. Where
this line differs most from the modern language, then, is not so much in the
words themselves but in the syntax with which they’re combined into a clause.
The natural order nowadays, including wilgesı̄þas from the next line, would be
something like “That afterwards in old age companions will support him”; but
Old English has “That him in old age afterwards support companions”. Two
differences in particular are significant. First, and most immediately obvious,
the order of the basic elements of the sentence is different. Under most circum-

36  In another point of tangency between Old English and Modern German, this root also takes
on the sense “to be accustomed to” in both languages, leading to the somewhat antiquated word
wont in English, and transitive gewöhnen “to accustom” in German, with the umlaut signaling that
this is a causative verb.
 Old English and Germanic languages 125

stances (questions are an important exception), the modern language strongly

prefers the order subject–verb–object, i.e. companions – would support – him.
But Old English inverts this here to hine – gewunigen – wilgesı̄þas. Now one
might generally argue that it makes sense that Old English can do this: it has all
these additional endings, so it should have a little more flexibility as to where
subjects and objects can go relative to verbs. In this particular instance, how-
ever, that explanation isn’t very effective. In Modern English him is clearly an
object, not a subject, and companions could be either subject or object, and, if
we consider a pared-down version of the sentence without the auxiliary verb
will, the lack of an -s on support would also indicate that a third-person plural,
not singular, subject was needed. The amount of syntactic information pro-
vided by the Old English forms is no different: hine is clearly an object, gewuni-
gen a plural verb form, and wilgesı̄þas could be either a subject or an object,
since the nominative and accusative plural have the same form. So why does
the object-first order work in Old English, whereas “That him in old age after-
wards support companions” is so hard for today’s speakers to process? Largely
because languages display considerable inertia: change something in one area
of a language’s grammar, such as morphology, and it may take a while for the
effects of that to play out in another area, like syntax. Old English had already
gone a long way towards reducing the number of distinct case endings on
nouns (recall the discussion of stān above), but there were still enough nouns
around that did have different nominative and accusative endings—in other
words, nouns which looked different depending on whether they were the sub­
ject or the object—that speakers would have retained a keener dis­crim­in­ation
for processing nouns in real time as potentially either subjects or objects
according to their endings. Nor should one forget that this is poetry. Just
because this object–verb–subject order was possible doesn’t mean it was the
default choice; indeed, Old English was already starting to favor subject–verb–
object order. But in this instance, delaying wilgesı̄þas until the following line
allows it to alliterate with wı̄g “war”, thereby binding together the themes of
companionship and battle at a phonetic level.
The second significant difference between the Old and Modern English syn-
tax in this clause is the need, felt by both translators, to introduce the helping
verb will. While such auxiliary verbs were already in use in Old English—wit-
ness sceal gewyrcean “must bring about” in the first line of the excerpt—their
role has increased over the last dozen centuries. To a large extent, this is simply
the verbal counterpart of what we’ve already seen with nouns: as nouns lost
case endings, it gradually became necessary to have more prepositions and a
more fixed word order to make syntactic relationships clear; so too with verbs,
the decline of the subjunctive mood increased the need for auxiliaries to express
what had once been indicated simply through endings. To understand how this
works, consider a pair of sentences that illustrate where the subjunctive is still
just barely holding on in the modern language:
126 How dead languages work

The manager insists that his intern arrives on time.

The manager insists that his intern arrive on time.
In those dialects of English that still maintain a subjunctive, the presence or
absence of the final s on arrive makes a difference: in the first sentence, arrives
is in the default indicative mood, which presents statements as realities, so the
manager is simply stating forcefully that it is indeed the case that the intern
arrives punctually; in the second sentence, arrive, because it no longer has
the -s of the third-person singular indicative, is marked as a subjunctive, which
presents a statement as somehow falling short of reality because, in this case, it’s
an ideal that needs to be worked towards as a general rule rather than a state-
ment of fact: the manager is setting the rules that the intern needs to follow
in the future, not commenting on a state of affairs that already exists. Now, even
for those speakers of English for whom this distinction remains in play, it can
only be made explicit if the subject of the subordinate clause is, as here, third-
person singular, since with all other persons and numbers, the indicative and
subjunctive look alike: The manager insists that his interns arrive on time is
formally ambiguous. Accordingly, with the diminished utility of the sub­junct­
ive mood as a means of distinguishing between these two usages, auxiliary
verbs have filled the gap, with expressions like The manager insists that his
interns should arrive on time clarifying the semantics.
To return to the Old English example, þæt hine on ylde eft gewunigen wilgesгþas
“so that dear companions will support him afterwards in old age”, gewunigen is
marked by its ending as a subjunctive, with the ending -en rather than indica-
tive plural -aþ. While the modern language needs to pick an auxiliary verb to
indicate that the clause is only a potential event in the future—and here will is
the obvious choice—Old English can do this more subtly by simply employing
the subjunctive.

wil-gesı̄þas þonne wı̄g cume,

dear-companions when war comes

This line provides further examples both of Old English’s love of compound
nouns and of the subjunctive. Wilgesı þ ̄ as is formed simply by juxtaposing
wil “will, pleasure” and gesı̄þ “companion”, with the addition of the plural
ending -as. The first element, wil, survives in English, but the second does
not, although German still preserves the cognate Gesinde “domestic servants,
farmhands”, if as a rather old-fashioned word. The conjunction þonne is
the ancestor of modern then and than (which were still used interchangeably
in the sixteenth century), but in Old English it could be used either like its
modern descendants or, as here, where the modern language would use the
subordinating conjunction when. While wıḡ “war” has also been lost as an
independent noun, it can still be found in the unimpeachably Germanic
 Old English and Germanic languages 127

compound name Ludwig “famous-war”.37 With cume, we have another sub-

junctive: the coming of war is presented as a future contingency, not a present
reality, a force that one can render in the modern language by replacing when
with whenever.

lēode gelǣ sten lof-dǣ dum sceal

people38 will stand by (with) praise-deeds shall

This line too presents us both with words that survive in the modern language,
but with a different appearance, and with those that have been lost—in English
but not in Old English’s niece language, German. In the first half-line (“So that
the people will stand by him”), the subject lē ode is a common word for
“­people”,39 the same as German Leute, and the verb gelǣ sten, which looks so
foreign, is none other than Modern English last, but with the prefix ge- and the
subjunctive ending -en added on either side of a verbal root that’s pronounced
essentially the same as its descendant. With the second half-line, the next sen-
tence begins, starting with lofdǣ dum, another dative plural in -um like feohgif-
tum above, and here too the presence of the dative ending obviates the need for
a preposition like “with”. Once again, the noun is a compound. The first elem­
ent, lof “praise”, like lēode, is no longer current in English, but is still found in
German, as Lob (seen in the sobriquet of Mendelssohn’s Second Symphony, the
Lobgesang, or “Praise-song”, which yet again exemplifies the ubiquity of the
ge- prefix in West Germanic languages); the second element, dǣ d (pronounced
like modern Dad), survives as deed. In theory, one could simply translate this
compound as “with praise-deeds”. But while Modern English certainly allows
such compound nouns as grammatical possibilities (recall the discussion of
string quartets above), it has become more hesitant to deploy them than was
Old English poetry: thus Heaney expands it to “behaviour that’s admired” and
even the comparatively literal Donaldson offers “through deeds that bring

37  For the Lud- element, cognate with English loud, see the discussion in Chapter 2. As for wı̄g,
it is also related to the Latin verb vincere “to conquer”, best known to English speakers as the third
in Caesar’s famous triad Vēnı̄, vı̄dı̄, vı̄cı̄ “I came, I saw, I conquered”. As with the equation between
the Germanic prefix ge- and Latin com- (see n. 35), Verner’s Law must be invoked for the Latin c
to correspond to Germanic g rather than h.
38  I follow R. D. Fulk, R. E. Bjork, and J. D. Niles, eds., Klaeber’s Beowulf, 4th ed. (Toronto, 2008)
in taking lēode here as a nominative plural subject of gelǣ sten; B. Mitchell and F. C. Robinson,
Beowulf: An Edition (Oxford, 1998) take it to be a dative singular object, with wilgesı̄þas continu-
ing on as subject (“support the prince”), but this adds little to the force of hine . . . gewunigen,
whereas taking lēode as the subject enables this clause to broaden the scope of the previous state-
ment: not just his companions but the people generally will support him.
39  For the complications regarding the relationship between the three words lēod (fem.)
“­people”, lēode (masc.pl.) “people”, lēod (masc.) “man, prince”, as well as the gradual decline of the
word in Middle English, see OED s.v. †lede, n.1.
128 How dead languages work

praise”. These are understandable choices, but in making the syntax more
familiar to the modern reader, both translations lose some of the intensity of
the original.
Finally, the verb sceal, pronounced nearly the same as modern shall, seems
somewhat closer here to its modern sense, expressing a sort of inevitable fu­tur­
ity, than it does in the first line. That is, following Donaldson’s translation, and
seeing a difference between the two uses, one could read it the first time as
signaling the beginning of an exhortation (“In this way one ought to bring it
about that . . .”), but in this second instance as merely marking the conclusion of
what the poet has just said in a summary statement (“Through deeds of praise
[such as have just been mentioned], one shall [or even just will] prosper”). Now
the reasoning behind suggesting the two different senses rests on more than
just an intuition about the poet’s likely line of argument. Let’s say one wants to
see the stronger, prescriptive sense of the verb here. After all, couldn’t the poet
simply be repeating the advice of the first line? This would theoretically lead to
a translation such as “Through deeds of praise, one ought to prosper”—but this
doesn’t capture the right meaning, since “ought to” here wouldn’t express so
much what one is obliged to do as what ideally will take place if the advice is
followed. This difference is due to the nature of the verb that follows sceal: in the
first line, sceal takes an infinitive (gewyrcean “bring it about”) over which the
agent has control: the young man has a choice over whether he’ll be the sort of
prince who will be properly generous or not, and therefore a verb of this sort
represents an appropriate course of action to serve as the substance of a pre-
scriptive instruction. By contrast, the second example of sceal occurs with
geþēon (“prosper”), which does not describe an action over which the agent has
control. One can certainly issue an imperative of the form “Live long and pros-
per”, but because the addressee has no direct control over either of these acts,
we automatically interpret this as a wish rather than a command. So too here,
sceal geþēon doesn’t make sense as an exhortation, and consequently reads
more naturally as the expression of an inevitable conclusion. All this goes to
show the importance of paying close attention to the immediate context when
determining the meaning of a helping verb like sceal. If we want to track its shift
from meaning “ought to” to “will” and to suggest that, in this passage, it’s used
once in one way, once in the other, then it’s important to have these sorts of
syntactic tests to assess where it stands on this semantic continuum.

in mǣ gþa gehwǣ re man geþēon

in of nations each a person prosper

The final line of the excerpt also contains the now familiar mix of words that
survive intact (in and man), one that has a modern descendant, but not an
obvious one (gehwǣ re), and a couple that have simply fallen out of use (mǣ gþa,
geþēon). But even man, superficially identical to its descendant, is here used in
 Old English and Germanic languages 129

an obsolete fashion, as an indefinite pronoun (“one, a person”)—a use that

survives in German. Gehwǣ re can be analyzed as the prefix ­ge-, a root syllable
hwǣ -, and the case ending -re. The root here is actually the same as that found
in the modern word who, and the prefix essentially turns what would otherwise
be an interrogative pronoun (“what person?”) into a distributive form (“each
person”). In a sign that the case system of Old English enables certain syntactic
constructions that are no longer possible in the language, the phrase here does
not translate literally as “in each or every nation” but as “in each one of nations”.
While this so-called partitive construction would be comparatively clunky in
Modern English, the genitive plural ending -a on mǣ gþa (signaling “of . . .s”)
and the dative singular ending -re on gehwǣ re (indicating that it is the object of
the preposition in) convey the same syntactic information in a succinct way—
the sort of syntactic efficiency that helps give Old English poetry its compressed
intensity. While neither mǣ gþa nor geþē on is still an active part of the modern
language, both may be found in the OED: the former under maegth, a term
referring to the concept of a person’s kin as understood specifically in the
Anglo-Saxon world, the latter under the entry †thee, v.1 (not to be confused
with thee, the object form of the pronoun thou), which records that it had
already become archaic by the sixteenth century.
Now all this linguistic commentary has shown in detail the extent to which
the language of Beowulf differs from that spoken today, but it has almost entirely
left aside one of the most important differences between the excerpts from
Beowulf and the New Testament: this one is poetry. We see this playing out in a
couple of different ways on the page. First, there’s the unusual layout of the
lines: each line is divided into two halves, with a strong break between the two,
called the caesura, marked in modern texts through the additional spacing at
the midpoint. This reflects the metrical structure of Old English poetry in par-
ticular, and most early Germanic poetry in general. Unlike the main Greek and
Latin meters, which are based on repeating patterns of long and short syllables,
Old English meter is governed by the principle of stress: each half-line has two
syllables that are stressed syllables, and anywhere from two to four that are
unstressed or half-stressed. That would not be an especially strict meter in its
own right, but it’s combined with a second principle: two or three of the stressed
syllables in a line should alliterate, either by sharing an initial consonant or
by starting with any vowel. Thus, with stressed syllables underlined, and alliterat-
ing sounds in bold, our excerpt scans as follows:

Swā sceal geong guma  gō de gewyrcean

fromum feohgiftum on fæder bearme
þæt hine on ylde  eft gewunigen
wilgesı̄þas þonne wı̄g cume,
lēode gelǣ sten:  lofdǣ dum sceal
in mǣ gþa gehwǣ re  man geþēon.
130 How dead languages work

The end result is a poetic form that is much terser than Greek and Latin dactylic
hexameter: not only are the lines noticeably shorter (around ten syllables each,
compared to around fifteen on average in a hexameter) but the regular break at
the halfway point makes them still choppier. (While the classical hexameter
also has a break somewhere in the middle of the line,40 it can come at several
different points in the line, and virtually never at the actual halfway point,
which keeps the ear from splitting the line into two segments as easily as is the
case with Germanic alliterative verse.) What’s more, the simple fact that the
rhythm is organized around stressed syllables—rather than the alternation of
long and short syllables—emphasizes the regularity of the two strong beats per
half-line, especially since those same syllables are further emphasized by shar-
ing an initial alliterative sound in most instances. As already noted in the case
of wilgesı̄þas “will-companions” and wı̄g “war”, this is a very effective poetic
device for highlighting thematic connections between words, much as also
happens with line-end rhyme in later English poetry. The extent to which one
sees such connections is of course up to the individual interpreter: the initial
sounds of fromum feohgiftum (“abundant money-gifts”) of the second line
seem closely bound to one another, but one may not want to see their relation-
ship to fæder in the second half-line as anything more than alliteratively con-
venient. Of greater linguistic significance is the alliteration in the first line
between geong “young”, guma “man”, and gōde “good”. Now, earlier in the chap-
ter (see n. 28), the letter g was occasionally topped with a dot (ġ) to indicate that
it was a palatalized g, that is, a g that was pronounced with the tongue touching
not the soft palate at the back of the mouth (as in goose) but the hard palate
further forwards (as in geese). Furthermore, in addition to the different point of
articulation, the latter palatalized g also had a tendency to weaken such that the
tongue no longer came fully into contact with the roof of the mouth, eventually
turning the sound into that of the modern consonant y. This sound change did
not take place in German, and thus accounts for many instances of words
where a g in German corresponds to a y in English: at word-end, we see this in
pairs like Tag ~ day, lag ~ lay (as in Bach’s Easter cantata Christ lag in Todes
Banden “Christ lay in death’s bonds”); at the start of the word, it occurs in cog-
nates like gern “gladly” ~ yearn, and even Geld “money” ~ yield (where there is
a cognitive equivalence between yielding something and paying it over as
money to another).41 But regularly in Beowulf, words that would come in later

40  See the discussion of caesura in the section on the Iliad in Chapter 2.
41  Geong, inconveniently, does not follow this pattern (young pairs with German jung rather
than a hypothetical gung), as its initial g had a different origin. But, even though this g is the out-
come of an original Indo-European y sound, as the German and Latin (iuvenis) cognates suggest,
it had still come to alliterate with the regular velar g, as seen in e.g. line 2626 of Beowulf, where
geongan alliterates with gūðe, and that is the only alliterative pair in the line. (The line Swā sceal
geong guma gōde gewyrcean doesn’t prove that geong alliterates with hard g since the alliteration of
guma and gōde would already be sufficient to meet the metrical requirements of the verse.)
 Old English and Germanic languages 131

Old English to be pronounced with an initial palatal g still alliterate with those
that start with a regular “hard” (properly, velar) g, suggesting that this sound
change had not yet taken place at the time it was composed as poetry, and thus
forming one of the linguistic arguments for dating the poem to perhaps the
eighth century ad.42
The clipped rhythm of Beowulf, with its emphasis on repeated sounds at the
onsets of syllables that are grouped into comparatively short segments, is diffi-
cult to replicate convincingly in the modern language. But one reason I’ve
brought in Seamus Heaney’s translation is to show that, thanks to his outstand-
ing skill as a poet, he has succeeded in doing just that. Consider his translation
again, but set with the typographical conventions followed in the Old English
example above, i.e. spacing to indicate caesurae, alliteration in bold, and stressed
syllables underlined:

And a young prince  must be prudent like that,

giving freely  while his father lives
so that afterwards in age when fighting starts
steadfast companions will stand by him
and hold the line. Behaviour that’s admired
is the path to power among people everywhere.

As he details in his introduction (pp. xxviii–ix), Heaney keeps remarkably close

to the metrical rules of Old English throughout his translation. What’s also
clear is the power of the sheer visual appearance of the words to affect how we
read poetry: open up the lines with double-spacing at the caesura, and it’s hard
not to read the Modern English with more of the choppiness that is so charac-
teristic of the original poem. Indeed, one then comes to question how much of
that choppiness is itself an artificial by-product of the conventional spacing.
(Answer: Some of it can be ascribed to typographical illusion, but, no matter
how you print it, the syntax remains much more compact than is typical of the
modern language.) It’s true that Heaney has to venture further from a strict
literal translation to achieve this metrical effect: swā sceal geong guma gōde
gewyrcean “so shall a young man bring about by good”, which Donaldson ren-
ders “In this way a young man ought by his good deeds . . . to make sure”,
becomes Heaney’s “And a young prince must be prudent like that”, with the
concrete accomplishing of good deeds turned into a more abstract exhortation
to prudence—in part in order to achieve the alliteration of prince and prudence.
But Heaney is too good a poet for this not to be part of a broader goal: a similar
shift towards the abstract is also evident in the translation of lofdǣ dum, with

42  See B. Mitchell and F. C. Robinson, Beowulf: An Edition (Oxford, 1998), p. 11.

132 How dead languages work

Donaldson’s “deeds that bring praise” contrasting with Heaney’s “behaviour that’s
admired”. At some level, these translations amount to the same thing. After all,
what is behavior other than a set of deeds that one has done? And, given the
slightly stilted flavor of deed in contemporary language, one understands
Heaney’s reluctance to use it. Yet there is a difference: speak of behavior in the
singular, and one has turned a set of individual acts into a single generalized
pattern; retain deeds in the plural, and the focus remains more on the discrete
actions that could potentially bring praise, less on the overall character of the
person performing them. If, in observing the whole sweep of Western litera-
ture, a critic like Harold Bloom is right to descry an increasingly sophisticated
portrayal of characters’ psychology,43 then Heaney’s Beowulf could here be said
to jump the gun: what the poet saw in terms of specific actions performed, the
translator reimagines as a more holistic concept of character. Here, as ever, in
conducting even the most basic literary analysis of a text, one ignores the
phrasing of the original language at one’s peril.

43  In his Shakespeare: The Invention of the Human (New York, 1998), he gives Shakespeare
pride of place in this development.
 5

Sanskrit

With Sanskrit, the ancient sacred language of India, we move far to the east.
But that does not mean that we’ve left the Indo-European family of languages
behind. Indeed, European contact with India in the eighteenth century was
crucial in piquing scholars’ interest in the prehistory of the languages of
Europe. The more that was known about Sanskrit in the West, the more it
became clear that any account of the origins of Greek, Latin, and the Germanic
languages had to take into consideration the similarities they shared with
Sanskrit, the language of such important Hindu texts as the Vedas, the
Upanishads, and the Bhagavad Gita. This awareness was stated most famously
by Sir William Jones, in a speech delivered before the Asiatick Society of
Calcutta on February 2, 1786, a date now generally given as the birthday of
Indo-European linguistics. Introductory textbooks in the field invariably
quote his words at length:
The Sanscrit language, whatever be its antiquity, is of a wonderful structure; more
perfect than the Greek, more copious than the Latin, and more exquisitely refined
than either; yet bearing to both of them a stronger affinity, both in the roots of
verbs, and in the forms of grammar, than could possibly have been produced by
accident; so strong, indeed, that no philologer could examine them all three, with-
out believing them to have sprung from some common source, which, perhaps, no
longer exists.
Hellenists might dispute whether Sanskrit is really more perfect than Greek,
and Latin is copious enough to have served as the lingua franca of Europe for
centuries. But in addition to his glowing admiration of Sanskrit’s refinement—
which is certainly exquisite, whatever one thinks of his hierarchical ranking of
so subjective a quality—Jones’s words also offer a superb distillation of the lin-
guistic place of Sanskrit relative to the classical languages of Europe.
The first point is the statement that the affinity between these languages lies
both in the roots of individual words and in the forms those words take,
depending on the syntax of the sentence. Limiting ourselves, as Jones does, to
the roots of verbs, we see correspondences like the following:

How Dead Languages Work. Coulter H. George, Oxford University Press (2020).
© Coulter H. George. DOI: 10.1093/oso/9780198852827.001.0001
134 How dead languages work

  Latin Greek Sanskrit

“bear” fer- pher- bhar-
“give” dō- dō- dā-
“give birth” gen- gen- jan-

But shared words, in themselves, are not the most secure way of establishing
that a group of languages descend from a common source. Even relatively com-
mon words can be borrowed from one language to another: such English words
as people, place, point, and use are all ultimately taken from Latin, by way of
French. Still, the sheer number of core vocabulary items shared by Greek, Latin,
and Sanskrit—not just verbs but nouns, pronouns, adjectives, and numerals as
well—is suggestive even on its own. But what makes watertight the argument
that all three languages derive from a single, earlier language is the shared pat-
terns of inflection, patterns that, based on the historical record, are far less
likely to have spread simply through diffusion. Consider the first verbal root
given above. In the three languages, the forms it takes in the singular of the
imperfect active indicative are as follows:

  Latin Greek Sanskrit

“I was bearing” ferēba-m e-phero-n a-bhara-m
“you (s.) were bearing” ferēbā-s e-phere-s a-bhara-s
“he/she/it was bearing” ferēba-t e-phere a-bhara-t

In all three forms, Latin and Sanskrit have exactly the same ending, marked
respectively by ‑m, -s, and -t. Greek, at first glance, doesn’t fit quite as well, since
the corresponding forms show -n, -s, and a null ending respectively. But the
shared s of the second person already points to a common origin, and the -n of
the first and the null ending of the third can also be shown to go back to an ‑m
and a -t respectively. A full demonstration of this would lead to too much of a
digression, but, as partial proof, note that no Greek words end in -m or -t, and
that there are many other examples of languages where final -m and -n have
merged as -n, such as Spanish, where Bethlehem has become Belén, and
Columbus’s name is shortened to Colón. There remains a difference in the part
of the word that’s not the ending. Whereas Greek and Sanskrit both add a prefix
to mark these forms as past indicative (the so-called augment), Latin—along
with the other Italic languages—has innovated by inserting a -ba- element to
form the imperfect.1 Since this is exactly the sort of restructuring of how verbs
work that we can see taking place in languages whose history is attested over a
long period of time, the absence of correspondence in this feature is not as sig-
nificant as is the presence of the correspondence in the endings.

1  Originally this was a helping verb, as seen in the discussion of the Germanic weak verbs in
Chapter 4.
 Sanskrit135

Nor are such grammatical similarities limited to verbs. The case endings of
nouns also match up from language to language. Here is the word for “wolf ” in
the nominative and accusative singular; in the Sanskrit, the r with the subscript
ring indicates an r pronounced as a vowel (as in English butter, but with a slight
trill):

  Latin Greek Sanskrit

nominative lupus lykos vr̥kas
accusative lupum lykon vr̥kam

From these endings, which are paralleled in other Indo-European languages as

well, it seems safe to conclude that these languages all derive from a parent that
marked sentence subjects (in the singular) with a final -s, and sentence objects
with a final -m; in Greek, this final -m has become an -n, just as happened with
the first-person singular verbal ending.
But while Jones’s statement thus neatly encapsulates two of the main types of
similarity among the three languages that suggest that they arose from a single
source, the truly innovative aspect of his proclamation was the awareness that
that source might no longer exist. Up to that point, linguists had tried to estab-
lish one language or another as the original, pre-Babel tongue, and all other
languages as corruptions of this Ur-language. One might reasonably have put
forward Sanskrit as a candidate: as Jones indicated, it was ancient, and it pos-
sessed a fuller set of grammatical inflections than are found in Latin and Greek.
(It has, for instance, eight cases rather than only five or six, and a more com-
plete  set of distinct dual endings.) But—and it took a while for linguists to
realize this—Sanskrit had also collapsed together sounds that remained dis-
tinct in Greek, Latin, and the other Indo-European languages. This is clearest
with the vowels. If you look back over the Sanskrit forms already given, you’ll
notice that the vowel a is particularly common, with the short a pronounced
like a schwa, and the long ā like the a in father. What is more, it corresponds to
several different vowels in the cognate words in Greek and Latin. First, take
those three verbal roots, bhar-, jan-, and dā-. In the first two, where Sanskrit has
a short a, both Latin and Greek have a short e; in the last, where Sanskrit has
long ā, the other two languages have a long ō. Next, look at the forms of the
imperfect and the “wolf ” word. Here, Greek has a distinction between e as the
past tense augment at the start of the word (e-), the vowel in the root (-pher-),
and the final vowel in the second- and third-person forms (‑es, -e), as opposed
to an o as the final vowel in the first-person form (-on) and the two case endings
(‑os, ‑on). Sanskrit, by contrast, has an a in both sets of environments. Latin
can’t be used as a comparandum for the vowels of the verbal forms, since they’ve
clearly been restructured, but older Latin inscriptions show that the case end-
ings -us and -um had been -os and -om, with the same vowel as preserved in
Greek, until the third century bc.
136 How dead languages work

There are two explanations, then, for the vowel patterns of the languages in
question: either (i) Sanskrit represents the original situation, and it, or a parent
language like it, used the vowel a in many different environments, but at some
point a language ancestral to Greek and Latin split off and arbitrarily changed
some of those a vowels to e, and others to o; or (ii) the vowels of the parent
language were closer to those of Greek and Latin, but Sanskrit changed e and o
to a across the board. The second explanation is clearly better, since, crucially,
it doesn’t require that a major change like the wholesale creation of new vowels
take place on an arbitrary basis, and in the same way in both Greek and Latin
(and many of the other Indo-European languages too). Thus Sanskrit cannot
represent, at least in this feature, a more archaic form of the language than
Greek and Latin. But since Sanskrit is in other respects more conservative than
Greek and Latin, it becomes clear that no attested languages are themselves the
parent language, which must instead be reconstructed on the basis of the most
ancient features of all the attested daughter languages. And as a reconstruction,
it is referred to as a proto-language: Proto-Indo-European (PIE).

THE SOUNDS OF SANSKRIT

What makes Sanskrit look like Sanskrit? We’ve already seen one change that
gives it its distinctive appearance, namely the merger of inherited e, a, and o as
the single vowel a, and that of long ē, ā, and ō as long ā. We may add the follow-
ing to the examples above to show just how characteristic of the core vocabu-
lary of Sanskrit this vowel is:2

  PIE Sanskrit  
“be” *es- as- Latin es-se “to be”, es-t “s/he is”,
Greek es-ti “s/he is”
“lead” *aǵ- aj- Lat. ag-ō, Gk. ag-ō, both “I lead”
“make; place” *dhē- dhā- Lat. fē-cī “I made”, Gk. e-thē-ka “I placed”
“three” *treyes trayas Lat. trēs, Gk. treis (in both, the e vowels
have contracted)
“seven” *septm̥ sapta Lat. septem, Gk. hepta
“ten” *deḱm̥ daśa Lat. decem, Gk. deka

Now this does not mean that Sanskrit only has the vowel a. First, the vowels i
and u, both long and short, survived as such, so where the Greek for “s/he is” is

2  Some additional diacritical marks creep into play here: the circle under the Indo-European m̥
simply indicates that this m is to be pronounced as a vowel, as in bottom; the acute accents over
PIE ḱ and ǵ and Sanskrit ś we’ll return to shortly. So as not to complicate the presentation unduly,
I have omitted laryngeals from the reconstructions.
 Sanskrit137

esti, the Sanskrit is asti, and where the Latin for “smoke” is fūmus, the Sanskrit
is dhūmas.3 But Sanskrit also filled the gaps left by the merger of *e and *o with
a by creating new e and o vowels out of what used to be diphthongs:
• Inherited *ai, *ei, and *oi all became e.
• Inherited *au, *eu, and *ou all became o.
There’s a pleasing symmetry to this, which becomes clear if one charts out
where the tongue is positioned for the pronunciation of all these vowels. (In
what follows, because we’re dealing in broad brushstrokes with prehistoric
vowel changes, strict phonetic accuracy will not be the goal, and it will be
assumed that the five inherited vowels *a *e *i *o *u were pronounced roughly
as in Spanish or Italian, i.e. ah eh ee oh oo.) First, both i and u are capable of
serving either as vowels in their own right or as the second element of a diph-
thong, with i pronounced with the tongue towards the front of the mouth, and
u with the tongue towards the back; in either case, the tongue is comparatively
high, close to the roof of the mouth. Second, like i and u, e and o form a pair in
which the first vowel is pronounced with the tongue farther forward, and the
second farther back—but this time with the tongue not so close to the top of the
mouth. Third, even further down is the vowel a. It’s because the tongue is par-
ticularly low for this vowel that doctors have you say ah when they want to peer
into your throat. Linguists typically represent all this with a vowel triangle:

  Front   Central   Back

High i       u
Mid   e   o  
Low       a    

In Sanskrit, then, certain vowel sounds have a tendency to drift downwards, at

least in terms of the scheme just outlined:4 just as the original vowels e and o
shift closer to a, merging with it in the end, so too diphthongs with i and u as
second elements shift closer to a, but only get as far as the mid vowels e and o
respectively.

3  The sharp-eyed may object that the conservation of ū holds good in the first syllable but not
in the second. But here, as it happens, both languages have innovated, with the original vowel *o
changing to a in Sanskrit, and to u in Latin, as regularly happened in unstressed final syllables of
this shape. (Note the correspondence in the endings of lupus and lykos seen above.) The original
final vowel is still apparent in the Greek reflex of this word, thȳmos, whose meaning has shifted
from “smoke” to “spirit”, as seen in the excerpt from Iliad 9 in Chapter 2.
4  A disclaimer: As noted above, the Sanskrit short a is actually a schwa—that is, a mid central
rather than low central vowel, so, in stricter phonetic terms, the movement of *ĕ *ă *ŏ to a is a
centering rather than a lowering, but the merger of *ē *ā and *ō as ā does still represent a down-
ward drift on the part of the former mid vowels.
138 How dead languages work

All this is important because the relationship between diphthongs and sim-
ple vowels played a major structural role in the workings of PIE. We’ve already
seen one example of this in the discussion of the Germanic strong verbs: pat-
terns like drink ~ drank ~ drunk go back to alternations between forms with an
e, an o, and no vowel. So too in Sanskrit—and in words borrowed from it into
English—we see that the alternation between ei or oi and i becomes one between
e and i, and that between eu or ou and u becomes one between o and u. This
leads to doublets like the following:
• Buddha “the enlightened one” continues an original u, but bodhisattva
“one with enlightenment (bodhi-) in their being (sattva-)” originally had
the diphthong eu in its first syllable.
• In the Hare Krishna mantra, Hare, in the vocative case of direct address,
had an ei diphthong in its final syllable, but the name is generally cited as
Hari on its own, since its basic stem ended in the plain vowel i.
Finally, to anticipate the chapter’s sample Sanskrit text from the Rig Veda, the e
in Veda is a sign that the word, which means “knowledge”, goes back to a root
that originally had an i in it: *weid-. The moment the i is restored, it looks much
closer to its cognates in other Indo-European branches, such as wit in English
and vidēre “see” in Latin. In earlier English, wit referred more widely to the
intellectual faculties, and to know something, in the metaphor of the parent
language, was to have seen it.
One final vowel worth discussing is the i of Krishna and Sanskrit. This repre-
sents a vowel different from the actual Sanskrit i: together with the preceding r,
it is instead a common way of transcribing the vocalic r seen above in the “wolf ”
word, vr̥kas. Because this sound has come to be pronounced by most modern
Indians as the sequence ri, that is a good representation for most purposes. But
since it is thought historically to have been a pure vocalic r sound, scholarly
transliterations of Sanskrit would give the words at the start of this paragraph
as Kr̥ṣṇa- and Saṃ skr̥ta-, with the ring beneath the letter indicating the vocalic
pronunciation of the r. Of course, the r is not the only letter there with a diacrit­
ical mark below it. While ṃ is beyond the scope of this book, the ṣ and ṇ both
deserve attention, as they, along with Sanskrit ṭ and ḍ, belong to a class of
sounds especially characteristic of both Sanskrit and its modern-day Indian
descendants: retroflex consonants. Now the sounds t d n s are all classified,
broadly, as dentals, since they are made with the tongue in the vicinity of the
teeth. (In English, they are in fact alveolars, since the tongue is not right at the
teeth, as it is for, say, a Spanish t, but a little bit farther back, at the alveolar
ridge.) Sanskrit, as it happens, not only retained the dental consonants t d n s
but also developed another series, notated with underdots, in which the tongue
curls so far upwards that the bottom of the tongue comes into contact with the
roof of the mouth. In the case of the retroflex s, this results in a sound closer to
 Sanskrit139

English sh than s, hence the rendering of Kr̥ṣṇa- as Krishna. They also tend to
occur in clusters: the r̥ triggers the change of a regular s to retroflex ṣ, which in
turn shifts the n to its retroflex counterpart ṇ.
The other set of sounds that gives Sanskrit a distinctive appearance is the
series of voiced aspirates, chiefly bh, dh, and gh, which may be compared with
the voiceless aspirates of Greek: ph, th, and kh. Both sets of consonants consist
of a stop followed by aspiration (h), and they occur in cognate words, such as
the pairs bhar- : pher- “bear” and dhā- : thē- “place”. Other examples of voiced
aspirates, seen in borrowings into English, include the dh in Buddha, as well as
the initial sounds of Bhagavad Gita, dharma, and ghee. As suggested by some of
the reconstructed forms given above, these are traditionally described as voiced
aspirates in PIE as well (e.g. *bh), with devoicing occurring in Greek (thus, ph),
loss of aspiration in Germanic, Celtic, and Slavic (giving the b of bear), and
more complicated changes in Latin. The exact phonetic nature of the sound in
the parent language is still the subject of debate, since voiced aspirated stops are
very rare in the world’s languages. But however they were pronounced in PIE,
it’s at least convenient to speak of them as voiced aspirates since in most of the
daughter languages they give rise to sounds that have either voicing or as­pir­
ation, if not both.
Finally, the accent over the s of Sanskrit daśa “ten” requires some ex­plan­
ation. Pronounced like the regular English sh (and therefore distinct from the
retroflex ṣ of Krishna), the sound is illustrative of what was once seen as an
important dialectal split in the Indo-European family as a whole. Named after
the words for “hundred” in Latin and Avestan, an ancient Iranian language
closely related to Sanskrit, the division sets the centum languages of the west
(Germanic, Celtic, Italic, Greek) against the satəm languages of the east (Balto-
Slavic, Indo-Iranian).5 By and large, where the centum languages have hard k
and g sounds—centum is to be pronounced with a hard c, as in Classical Latin—
the satəm languages have softer s, z, and j sounds; and where the centum lan-
guages have compound kw and gw sounds, the satəm languages have simple k
and g sounds. Here are some words that show the correspondences, with the
relevant sounds in bold:

  centum languages satəm languages

“hundred” Latin centum Russian sto
  Welsh cant Sanskrit śatam
  Greek hekaton Avestan satəm
“ten” Latin decem Russian desyat’
  Welsh deg Sanskrit daśa
  Greek deka  

5  The ə in the Avestan word satəm is simply the vowel schwa.

140 How dead languages work

“know” Latin cognōscō “I know” Russian znat’ “to know”

  Old Irish ad·gnin “recognize” Avestan zan- “to know”
  Greek gnōsis “knowledge” Sanskrit jñā- “to know”
  English know  
“who?” Latin quis? Russian kto?
  Old English hwā? Sanskrit kas?
“come” Latin venīre “to come” Sanskrit gam- “to go, step”
  Gothic qiman “to come”  
  English come  

The correlation is easiest to see in the first three sets of words, the first two of
which illustrate outcomes of the PIE *ḱ, and the third that of PIE *ǵ. The acute
accent here indicates that these are so-called palatal velars: a velar is a sound
produced with the tongue touching the soft palate (the velum), and the further
specification that it is palatal signals that it’s a kind of velar that’s articulated
farther forward, towards the hard palate, like the k in key (as opposed to the c
of car). That these PIE sounds were articulated at a point intermediate between
the soft and hard palate is suggested by the fact that they become velar stops
in some languages but a variety of palatal and alveolar sounds, like s, sh, and z,
in others.
The final two words, who and come, represent the PIE sounds *kw and *gw
respectively, the labiovelars. They get their name from their compound articu-
lation: at the same time that you pronounce the stop at the velum (k, g), you also
round the lips, as if to say w. (It’s because this second feature is a simultaneous
co-articulation that the w in the sound is represented as a small superscript
letter rather than as a sequence of k or g followed by w.) The compound nature
of these two sounds is also responsible for the seemingly small number of good
cognate words in the daughter languages. From an articulatory standpoint, it’s
comparatively difficult to pronounce labiovelars, so they have a tendency to
decompose into only having the stop or the w element but not both. The satəm
languages lost the w across the board, as did English come, although the q (pro-
nounced kw) at the start of the Gothic cognate qiman shows that the w at least
made it into Germanic before disappearing. Latin venīre shows the other pos-
sible simplification: the original g was lost, but the w sound, as the letter v was
pronounced in classical times, survived at the start of the verb.
While the centum/satəm split still makes it into introductory historical lin-
guistic textbooks—it is, after all, a useful shorthand for discussing a set of
sound changes that make the Balto-Slavic and Indo-Iranian languages look
rather different from Latin or Greek—it lost much of its importance in the early
twentieth century after the decipherment of Hittite, spoken in the second mil-
lennium bc in what is now Turkey, and Tocharian A and B, a pair of closely-
related languages spoken in the Xinjiang region of far northwest China in the
 Sanskrit141

sixth to eight centuries ad. Previously, there had been a straightforward geo-
graphical division between centum languages in the west and satəm languages
in the east. But the word for “hundred” in Tocharian A turned out to be känt,
and the Hittite word for “who?” is kuiš?, showing that sound changes charac-
teristic of the satəm languages did not in fact affect all the languages spoken to
the east. Instead, it’s more accurate to view them as having affected a smaller
grouping of languages in the center. This fits much better in any case with our
general understanding of the geography of sound changes—they spread out
like waves6 from particular points—than does the obsolete nineteenth-century
notion of a simple opposition between East and West.

SANDHI

A confession is now in order. Throughout this chapter, when I’ve cited Sanskrit
words, I’ve tilted the evidence ever so slightly in favor of bringing out the con-
nection with Greek and Latin. How so? The words that I’ve given as ending in
-as, like trayas “three” or vr̥kas “wolf ”, would only occasionally appear in texts
in this form, because Sanskrit practice is to show in writing how the sound at
the end of one word adapts to match that at the beginning of the next. This
sometimes happens in English, too. In the two-word phrase Don’t you, the final
t of Don’t is usually only pronounced as a t if one is saying the words slowly and
separately. In normal speech, it combines with the initial y of you to yield a ch
sound, with the resulting phrase pronounced doncha, a spelling sometimes
encountered in texts that aim to suggest a colloquial register. But whereas this
is only a very occasional feature of written English (gonna and wanna also
come to mind), it’s systematic and universal in Sanskrit, which has an elaborate
set of rules covering every possible combination of sounds that can occur at a
word boundary. The Sanskrit term for this is sandhi, which is a compound
word consisting of the prefix saṃ - “with” (comparable in force to Latin com-
and Greek ­syn-) and dhi-, a form of the root dhā- seen above, meaning “put,
place”, and cognate with Greek thē-. Calqued into Greek, then, the word is
equivalent to syn-thesis; into Latin, com-position.
How does it work in practice? Consider the word for god, devas, which is
cognate with Latin deus. In the following combinations, the underlying form to
the left of the arrow would in fact be written as the form to the right:

6  For a very basic discussion of the wave model, see e.g. A. Fox, Linguistic Reconstruction: An
Introduction to Theory and Method (Oxford, 1995), pp. 128–33, or L.  Campbell, Historical
Linguistics: An Introduction (Edinburgh, 1998), pp. 188–91; for a more up-to-date, if also some-
what more advanced account of how sound changes spread, see D. Ringe and J. F. Eska, Historical
Linguistics: Towards a Twenty-First Century Reintegration (Cambridge, 2013), pp. 45–58.
142 How dead languages work

(1) devas tiṣtḥ ati → devas tiṣtḥ ati “the god stands”
(2) devas karoti → devaḥ karoti “the god does”
(3) devas → devaḥ “god”
(4) devas dadāti → devo dadāti “the god gives”
(5) devas asti → devo ’sti “the god is”
(6) devas āsīt → deva āsīt “the god was”

To be fair, the treatment of final -as brings into play the most complicated of the
sandhi rules,7 but the behavior of other sounds is not all that much simpler, even
if there are linguistically reasonable explanations for each of the individual rules.
One practical consequence of sandhi is the difficulty in determining exactly
how to cite Sanskrit words in English-language works. Not only does the typ­
ical Sanskrit noun occur in well over a dozen forms, thanks to its wealth of case
endings and its fully-functioning dual number, but even a form that one would
normally expect to serve well as a citation form, the nominative singular, can
occur in several different guises. As a result, Sanskrit words are, more often
than their Greek or Latin counterparts, referred to simply in terms of the bare
stem. Below, then, rather than referring to the “god” word as devas, in deference
to its ancestral form in PIE, or devaḥ, as it would occur in isolation in Sanskrit
as a nominative singular, it will instead be cited as deva-, with the hyphen indi-
cating that this is only the stem, to which endings are then added.

NOUNS IN SANSKRIT

To return to the statement by Sir William Jones that opened this chapter: What
did he mean when he said that Sanskrit was more perfect than Greek, more
copious than Latin? For the most part, he’s referring to an incredible richness in
the morphology of the language, the various endings, prefixes, and changes of
stem that can affect a Sanskrit word depending on its function in a sentence.
Consider a regular noun like deva-, the word for god just introduced. How
does one arrive at a total of 17 forms, even ignoring the variations caused by
sandhi? The full paradigm looks as follows:

  singular dual plural

nominative devas devau devās
accusative devam devau devān

7  For those who want the gory details, here are the relevant rules. (1) When -as occurs before two
sounds, t and th, it remains stable. But in all other circumstances, it changes: (2) before most other
voiceless consonants, like k and p, it becomes a murmured h sound, notated ḥ ; (3) that’s also what
happens to it when the word occurs in isolation or as the final word of a sentence; (4) before voiced
consonants, like d, it turns into an o; (5) that also happens when it precedes a word with a short a,
but that a is then generally dropped; (6) before all other vowels, the s is lost, but the a remains.
 Sanskrit143

instrumental devena devābhyām devais

dative devāya devābhyām devebhyas
ablative devāt devābhyām devebhyas
genitive devasya devayos devānām
locative deve devayos deveṣu
vocative deva devau devās

Eight cases multiplied by three numbers should yield 24 different forms, but in
practice the total is never that high: there are never more than three different
forms of the dual, the vocative is always the same as the nominative in the dual
and plural, and the dative and ablative plural always look alike. (It’s no accident
that those last two rules hold true for Latin as well, as seen in Chapter 3 in the
section on Horace.) But even given those reductions in complexity, it remains
a fuller paradigm than that found in Greek or Latin: Greek has a dual, but three
fewer cases, and while Latin is only one or two cases short of Sanskrit (depend-
ing on whether one counts certain locative forms as a separate case), it has no
dual at all. But what does Sanskrit gain, relative to Latin and Greek, from hav-
ing so many cases?
Chiefly, it can dispense with prepositions in a much greater number of situ­
ations. To understand how this works, it’s best to consider the question from a
historical perspective. Sanskrit is generally agreed, in respect of the inventory of
its cases, to be closer to the parent language than are the classical languages of
Europe. Indeed, the standard reconstruction of PIE has precisely the same eight
cases as those attested in Sanskrit. Thus Latin and, to a slightly greater extent,
Greek have carried out a merger of what had been distinct cases into fewer, but
more functionally complex, forms. Take three cases in particular, the instru-
mental, ablative, and locative. In PIE, much as in Sanskrit itself, these cases will
have had three distinct functions: the instrumental, predictably enough, denoted
the instrument with which one did something; the ablative indicated the place
from which one was moving; while the locative specified the place at or in which
one was situated. I’ve italicized those prepositions for a reason: in Sanskrit, you
don’t need to include them in the sentence, because the case ending already does
the work of the preposition for you. Thus, in the following sentences, the under-
lined phrases could be translated with a single Sanskrit word:

He obtained wealth by means of the god → devena (instrumental)

She rejoiced in the god → deve (locative)
They received help from the god → devāt (ablative)

Now one might be troubled by the slight discrepancies between how I’ve
described the cases and what exactly they’re doing in these sample sentences.
Most obviously, just as to rejoice in the god is a figurative expression in English,
so too Sanskrit deve, in such a context, should not be taken to imply that one was
144 How dead languages work

somehow located within the god. But this brings us to a couple of important
points: first, even with as many as eight cases, each case still needs to take on a
range of uses, since there are far more than eight different roles that a noun can
play in a sentence;8 second, depending on the noun, some uses are a lot more
common than others. In particular, animate nouns often act differently from
inanimate nouns. The usage of English with—a preposition roughly equivalent
to Sanskrit’s instrumental case—is instructive here. With an inanimate object,
like pencil, with is likelier to indicate an instrument than a companion: She wrote
with a pencil is fine, but She traveled to town with a pencil rather stranger. With
an animate noun, on the other hand, the reverse is true: She traveled to town with
a friend is perfectly fine, and, while She wrote with a friend is grammatical, the
sense is once again that the friend is a companion, not the tool or intermediary
used for the writing. A friend could in fact play that role, but because it’s not the
most natural role for the friend, a more explicit stand-in for with would be need­
ed: She wrote to her senator through a more powerful friend. Two points, then, to
take away from this: First, what a preposition does can depend on what its object
is, and it’s no different with cases—the instrumental case of a noun designating
a person is likely to do something different from that of one referring to a thing.
Second, if a simpler preposition or case isn’t clear enough on its own, the lan-
guage may need to switch to using a more specific and, probably, more complex
expression in order to clarify what’s going on.
But before we can see what led to the situation in Greek and Latin, we first
need to observe that there’s plenty of scope for these three cases, the instrumen-
tal, locative, and ablative, to overlap in their usage.9 Take the instrumental and
locative, i.e. the with-case and the in-case, and consider how much difference
there is between the following sentences: He held the pencil in his hand and He
held the pencil with his hand. They are clearly not identical: the former, which
to my ear sounds slightly more natural, stresses the hand as the location of the
pencil, the latter throws emphasis on the hand’s role as the instrument that
holds the pencil. But both still refer to more or less the same state of affairs. As
a result, it could potentially be superfluous for a language to have both expres-
sions. Now, when it comes to prepositions, there’s only a single word form in
play, so it’s easy enough for a language to cope with some lexical excess. But to
have an entirely separate case for one of these semantic roles requires a whole
slew of endings that may differ considerably, depending on how many different
patterns (or, to use the technical term, declensions) of case endings a language
has. That is to say, maintaining in and with as distinct expressions in English
only requires those two different forms, since no matter the object, whether

8  Indeed, Hungarian is often said to have 18 cases, with nine cases just to deal with spatial
relations. But as there is little formal variation in the endings from noun to noun, many of them
resemble postpositions more than they do the case endings of the Indo-European languages.
9  For additional examples along these lines, see the discussion of the ablative case in Chapter 3.
 Sanskrit145

singular or plural, noun or pronoun, in will remain in, and with will remain
with. But preserving that same distinction between the locative and instrumental
cases in Sanskrit is a greater cognitive work-out: even just keeping to the end-
ings of deva- given above, one has to remember that the two contrastive forms
would be deve and devena in the singular, devayos and devābhyām in the dual,
and deveṣu and devais in the plural.
Nor are the instrumental and locative the only two cases ripe for such a merger.
The ablative (the from-case) can also be folded into these other two. One might
think, of course, that the ablative and the locative would not be particularly sus-
ceptible to collapse: there should be a pretty clear distinction between saying that
something is coming from a place as opposed to being stationary at a place. But
most of the time this difference is indicated clearly enough with the verb of a
sentence: She is coming from the house vs She is at the house. And many languages
have words that are ambiguous as to whether they designate place where or place
from which: the Latin word undique, for instance, can mean either “on all sides”
or “from all sides”. Moreover, this ambiguity has repercussions for the develop-
ment of the Latin interrogatives of place in the Romance languages. The Classical
Latin word unde? meant “from what place?”, but then, during the development of
Vulgar Latin in the Iberian peninsula, it acquired the meaning “at what place?”;
after a slight sound change, this became the form onde?, which remains the basic
Portuguese word for “where?” Now, because onde? had come to mean “where?”,
the preposition de “from” had to be prefixed in order to specify “from where?”,
yielding donde? (after elision of the -e of de), and this, likewise, remains the basic
word for “from where?” in European Portuguese. But Spanish didn’t stop there:
once again, the word that meant “from where?”, now spelled ¿dónde?, came to
mean simply “where?”, so, to ask “from where?”, the preposition de had to be
prefixed once more to yield ¿de dónde?, a phrase which, if we were to analyze it
into its component parts, de-de-unde?, and give them their original Latin mean-
ing, would mean “from from from where?” In short, however strange it may seem
that languages would merge ablative from-expressions with locative in- or at-
expressions, there’s plenty of evidence that it does happen.
Now we’re in a better position to see what’s happened in Latin and Greek:
case endings that were kept distinct in Sanskrit were potentially subject to
functional overlap. As prepositions—like de, in Romance—were brought into
play to clarify semantic roles, the case endings themselves bore less of the load
in conveying the meaning of the sentence, so there was less reason to maintain
so full a paradigm. In Latin this gives rise to a so-called ablative case that, as
seen in Chapter 3, actually fills the role not only of the PIE ablative (ab equō
“from the horse”) but also of the instrumental (cum equō “with the horse”) and
the locative (in equō “in the horse”). There is some movement in this direction
in Sanskrit—there are adpositions (prepositions that precede and postposi-
tions that follow their nouns) that can clarify the semantic role played by a
noun—but on the whole they are far less important to the language than in
146 How dead languages work

Greek or Latin. Now that we’ve seen the basics of the sounds of Sanskrit, as well
as a good example of what Jones meant when he spoke of Sanskrit as copious,
we can turn to some passages of actual Sanskrit.

THE RIG VEDA

Because Sanskrit literature is far too vast to survey properly, we’ll restrict ourselves
to the Rig Veda, the very earliest text. A collection of 1028 hymns (sūkta-) organ-
ized into ten books (maṇdạ la-), it amounts to about as much poetry as the Iliad
and Odyssey combined. The individual hymns praise various of the Hindu gods,
such as Indra and Vishnu, with particular attention paid to the twin principles of
the divine fire, the god Agni, and the divine liquid, Soma, to whom the entirety of
Book 9 is devoted. Dating the Rig Veda, as with the rest of early Indic literature, is
notoriously difficult, but, on purely linguistic grounds, it clearly belongs to an
earlier stratum of language than other important texts, such as the two great epic
poems, the Mahābhārata and Rāmāyaṇa. In its retention of some older endings
and syntactic patterns, as well as in its belonging to a pre-literate, oral tradition,
Vedic Sanskrit bears about the same relationship to Classical Sanskrit as Homeric
Greek does to Classical Greek.10 In any event, most scholars are content with pla­
cing the Rig Veda in the late second millennium bc, making it earlier than Homer
but later, probably, than most of the evidence we have for Hittite.
The name Rig Veda (R̥gveda-) derives from the internal structure of the
hymns, each of which is divided into stanzas (r̥c-). (According to the rules of
sandhi, r̥c- changes to r̥g- when it occurs before a v.) As a noun, r̥c- can refer not
only to the stanzas as formal entities but also the praise they contain, so the
whole title could loosely be rendered “praise-poem knowledge”. The other
three Vedas, the Sāmaveda (essentially, the Rig Veda rearranged and set to
music), Yajurveda (a compilation of liturgical formulas), and Atharvaveda (not
only hymns but also various folk incantations, attributed to the legendary seer
Atharvan), are less prominent in studies of Indo-European linguistics because
they were compiled at a later date.
Most of the ten books of the Rig Veda open with a hymn to Agni, and Book
1 is no different. Here is the first stanza of RV 1.1:11

10  Vedic Sanskrit also preserves the earlier Indo-European mobile accent, but I have silently
omitted this from the passages below in the interests of typographical simplicity.
11  A discussion of Devanāgarī, the script in which Sanskrit is standardly written, would take us
too far afield, but one feature in particular is apparent even to the untrained eye: in it, words are
far more likely to be run together than is the case in present-day European alphabets. While not
too much should be read into this—so-called scriptio continua used to be common practice in
European orthography as well—it does provide a fitting graphic counterpart to the phonological
principle of sandhi, that no word should be considered in isolation.
 Sanskrit147

अग्निमीळे पुरोहितं
यज्ञस्य देवमृत्विजम् ।
होतारं रत्नधातमम् ॥
Agnim  īḷe   purohitaṃ
Agni   I praise  domestic priest
yajñasya  devam  r̥tvijam,
of sacrifice divine  priest
hotāraṃ ratna-dhā-tamam  
chief priest treasure-granting-most  
I praise Agni, the domestic priest, the divine priest of the sacrifice, the
chief priest, most giving of treasure.

At first one might well question to what extent such a strange language could
belong to the same family as Greek and Latin. Of the eight words in question,
only devam has an obvious cognate: it is indeed related to Latin dīvīnus, whence
English divine. But other lexical relatives are lurking there. Start with the first
word, Agnim. This is not only the name of the god of fire but also the basic word
for fire itself. The two consonants of the root, agni-, are the same as those in the
corresponding Latin word, ignis, and the Russian word, огонь (ogón’), and all
three go back to a single common source. What is more, the fact that the ending
is an -m also matches Latin perfectly: Agni’s name here, as the object of the verb
of praise, is in the accusative case, and both Latin and Sanskrit have preserved
the same ending. Indeed, five more words in this sentence are also in the accusa-
tive (purohitaṃ , devam, r̥tvijam, hotāraṃ , and ratnadhātamam), and they also
all end in ‑m. (The dot under some of the m’s is simply a product of sandhi and
can be ignored for our purposes.) This shared ending highlights the extent to
which this first stanza is built up through simply piling up extra nouns as apposi-
tives to that first object. Now the main verb, īḷe, does not have any clear relatives
in European languages, but the next word purohitaṃ does. It consists of puro-
“before” and hita- “placed” (the domestic priest is one “placed in front” of the
family to protect it): puro- is ultimately related to the Greek and Latin prefix
pro-, as well as to the first element in the Germanic words forward and further
(as seen in Chapter 4), and hita- is the past participle of the dhā- root we’ve
already seen in saṃ dhi, and so cognate with the Greek-derived forms ‑thesis and
-thetic. Moving to the next line, the word for “sacrifice” yajña- is related to the
Greek word for “holy”, hagios, which has made it into the English word hagiog-
raphy as well as the name of the great basilica Hagia Sophia in Istanbul.
In the final line, the chief priest is, etymologically, a “pourer”: the root ho-
means “to pour”, the suffix -tar-, like English -er, forms an agent noun (one who
does something), and the ‑am ending, as we’ve seen, marks it as accusative. The
verbal root, with its o—originally a u-diphthong—goes back to an Indo-European
148 How dead languages work

form *ǵheu-; in the form *ǵhu-, this root became Greek chy-, which was the basis
for various words for “juice”, notably chȳlos and chȳmos, both of which have
entered English as medical terms for specific bodily fluids, chyle and chyme. As
for the suffix, this too has its relatives in English: it’s the same as the -tor in Latin-
derived words like actor and cantor. Finally, there’s the superlative ratnadhātamam.
It’s a compound adjective—Sanskrit is very rich in such compounds, especially
later on, by the time of the epics—formed from the noun ratna- “treasure” and
the verbal root dhā- “place, bestow”, thus “treasure-bestowing”, to which the
superlative suffix -tama- is then added, as if it were “treasure-bestowingest”. The
etymology of ratna- is unclear, and dhā- is familiar by now, so it’s the suffix ‑tama-
that will repay further comment as it too is related to forms that have been bor-
rowed into English from Latin.
While most Latin superlatives are formed with the suffix -issimus, a handful
of prepositional adverbs indicating spatial orientation use -timus instead:

in “inside” → in-timus “innermost” → English intimate

ul-s “on the far side” → ul-timus “farthest” → English ultimate
ob “at the front” → op-timus “best” (orig. “foremost”) → English optimal

Even taken in isolation, the sequence of t and m is distinctive enough that it

would be reasonable to derive Latin -timus and Sanskrit -tama- from the same
Indo-European source. But the connection becomes even closer when we con-
sider the corresponding comparative formations. In both languages, the com-
parative has an -r- where the superlative has an -m-, and English has borrowed
the Latin comparative forms directly:

  comparative (-er) superlative (-est)

Sanskrit -tara- -tama-
Latin in-ter-ior in-tim-us
  ul-ter-ior ul-tim-us

This is precisely the sort of linguistic correspondence that can only be explained
as a shared inheritance from the parent language. So, however alien the open-
ing of the Rig Veda may seem at first, it is still full of words and grammatical
patterns that are related to those found in the familiar languages of Europe, as
long as you know where to look.

HOW TO KILL A DRAGON

But the connections between Vedic Sanskrit and the classical languages of Europe
extend beyond grammatical endings and vocabulary to a shared Indo-European
poetic tradition. For a Vedic hymn that is much cited as evidence for this, ­consider
 Sanskrit149

next the opening of RV 1.32, in praise of Indra, king of the gods. The hymn as a
whole tells of Indra’s slaying of the serpent Vr̥tra (“Resistance”), who had
imprisoned the rain waters in the clouds—or perhaps in the mountains; the
­relevant word, parvata-, can mean either one:

इन्द्रस्य नु वीर्याणि प्र वोचं यानि चकार प्रथमानि वज्री ।

अहन्नहिमन्वपस्ततर्द प्र वक्षणा अभिनत्पर्वतानाम् ॥
indrasya nu   vīryāṇ i    pra   vocaṃ
of Indra now  manly deeds  forth  let me speak
yāni cakāra prathamāni vajrī
which he did first thunderbolt wielder
ahann ahim anv-apas-tatarda
he slew serpent up-waters-opened
pra vakṣaṇ ā abhinat parvatānām
forth bellies he split of the clouds/mountains

Let me speak forth now the manly deeds of Indra, which he did first, the wielder
of the thunderbolt. He slew the serpent; he opened up the waters; he split forth
the bellies of the cloud-mountains.
Once again a hymn begins with the name of the god, though this time in the
genitive case, with the ending -asya. (This disyllabic ending is cognate with the
archaic -oio genitive familiar to classicists from Greek epic poetry.) The line con-
tinues with a series of words that have clear relatives in more familiar languages.
It is no coincidence that nu and now both start with an n, that the word for “manly
deeds”, vīryāṇi, starts with the same three letters as virile, that pra means the same
as the Greek and Latin prefix pro, or that vocam, “let me speak”, looks like vocal.12
The word for “first”, prathama-, is probably a distant cousin of Greek prōtos, and
the fact that Indra’s weapon is a thunderbolt connects him with various import­
ant gods in European traditions as well, such as Zeus, Jupiter, and Thor.13
But the two words in this passage that have drawn the most attention are the
next two: ahann ahim “he slew the serpent”. Their fame is largely due to a 1995
study by the Indo-Europeanist Calvert Watkins, How to Kill a Dragon: Aspects

12  The last word equation requires a disclaimer: while Sanskrit vocam is from the same Indo-
European root as Latin vōx, vōcis “voice”, it’s partly through chance that the two resemble each
other as much as they do. Recall that PIE *o turned into a in Sanskrit, and that Sanskrit o actually
goes back to diphthongs like eu. As it happens, vocam is the outcome of a protoform *wewkwom,
in which the basic root *wekw- has been reduplicated (that is, an extra syllable *we- has been added
at the start) and then lost its own e vowel (yielding *-wkw‑); to this, the ending *-om is added. What
matches this formation far more closely than the Latin is Greek (ϝ)εἶπον (w)eipon “I said”, which
is also in origin a reduplicated aorist. The Sanskrit here corresponds to the unaugmented form of
the Greek and is therefore a so-called ‘injunctive’, acting as a sort of hortatory expression. A final
difference is that the Sanskrit c is pronounced like an English ch.
13  The former two, to be fair, may not originally have been associated with the thunderbolt: see
M. L. West, Indo-European Poetry and Myth (Oxford, 2007), p. 238.
150 How dead languages work

of Indo-European Poetics. In this influential book, which draws on data from an

almost inconceivable wealth of ancient sources, Watkins showed the extent
to  which the oldest poetic languages of the various Indo-European peoples
show a common phraseology that is most economically explained as an inher-
itance from a PIE poetic tradition. We have already come across this idea in
Chapter 2, as both Homer and the Rig Veda show expressions for “im­per­ish­
able fame” that can be reconstructed back to a single phrase in the parent lan-
guage. That equation, to be sure, had been noticed already in the nineteenth
century, but many other patterns are there to be found, provided one is will-
ing to sift patiently through sources ranging as widely afield as rituals per-
formed by Hittite palace officials, odes sung on behalf of the winners of Greek
athletic contests, and Old English medical charms. There’s certainly room for
disputing whether or not every similarity noticed by Watkins necessarily points
to a shared poetic inheritance, but the sheer amount of material assembled
is telling.
So how does ahann ahim fit in? First, it’s significant that they both start with
the same two sounds, a phonetic doubling that gains significance through repeti-
tion: the two-word phrase occurs 11 times in the Rig Veda, thereby taking on a
formulaic feel.14 Next, the source of the h in both words is important: both go
back to one of the rarest, and therefore most expressively marked, sounds in
PIE: *gwh. (Technically described as a voiced aspirated labiovelar stop in the
standard reconstruction, it would have been pronounced something like a gw
with an additional puff of air following.) Turning to the individual words, the
noun for “serpent”, ahim, is in the accusative case, as the object of the verb, and
indeed shows the same ending seen in Agnim in the first passage. But what’s
most striking is that the verb, ahann “he slew”, far from being an ordinary word
for “to kill”, is derived from a root that goes on to appear in especially dramatic
contexts in the daughter languages. Discussion of this point, however, will be
easier once the component parts of the verb have been established. The initial
a- is an augment,15 marking this as a past-time event, and the final -n is, in
effect, the ending of the verb, so the verbal root in isolation is merely han-, or,
as it would have been in PIE, *gwhen-. Because the articulation of this initial
consonant, like that of all the labiovelars, was so complex (as seen earlier in the
chapter), the descendants of this root diverged wildly in the various daughter
languages. Consider just two examples:
• In Greek, the relevant forms of the inherited root *gwhen- are those in
which the e was replaced by an o (thus, *gwhon-) or lost altogether (*gwhn-);
in these phonetic conditions, the first consonant turned into a ph, with

14  C. Watkins, How to Kill a Dragon: Aspects of Indo-European Poetics (Oxford, 1995), p. 305.
15  This is cognate with the past-tense augment e- found in Greek, seen in the comparison of
verbal forms at the start of this chapter.
 Sanskrit151

the root becoming phon- and phn- respectively.16 The first of these is seen
most transparently in the basic Greek word for “murder”, phonos, and the
second is found in the stem used for the main past17 tense of the verb “to
slay”, pephn-. (In particular, this verbal stem is formed by taking the
vowel-less form of the root, phn‑, but with an extra syllable, copying the
first consonant, added at the start (i.e. pe-), to yield pephn-.18) Be that
historical morphology as it may, what about the actual usage of the verb?
Well, there’s a lot of killing in the Iliad, and it’s usually described with the
verb kteinō. But the pephn- root can be used as well, and what Watkins
noticed is that it clusters particularly in one section of the sixth book of
the poem, when the deeds of Bellerophon are being described.19 Not only
does Bellerophon himself have a name whose second element consists
precisely of this root but the verb itself occurs three times in quick succes-
sion (6.180, 183, 186) when the hero kills not only the Amazons but also
the Chimera, a beast described as a lion in front, a goat in the middle, and
a snake in the rear. For Watkins, it is no coincidence that this cluster of
examples of the root *gwhen- occurs around the slaying of an animal that
was part snake.
• While *gwhen- did not survive as a verb in the Germanic languages, it is
found in what would become the Modern English noun bane. While the
modern word, in phrases like the bane of one’s existence, refers somewhat
weakly to anything seen as an affliction, its Old English ancestor bana or
bona referred more specifically to a murderer or slayer:20 in Beowulf, for
instance, it is used both of Grendel (line 2082) and of the dragon (Old
English wyrm) that finally defeats Beowulf (line 2824).21 In Old English’s
sister language Old Norse, the cognate word bani and its derivatives come
up repeatedly in Snorri Sturluson’s Prose Edda in the description of
Ragnarök, the famous Twilight of the Gods—most suggestively when
Thor kills the World Serpent (Miðgarðsormr), the second element of

16  This root has a short o and is thus different from the root phōn- “sound, voice”, which has a
different etymology altogether.
17  Properly speaking, this is an aorist stem, not a past-tense stem, but in the indicative mood,
it most often served as the equivalent of a simple past, so it seems a justifiable simplification to
label it here as such. See the discussion of aspect in Chapter 2.
18  As it happens, because of the divergent outcomes in Greek of PIE *gwh, this verb has an
unusually irregular paradigm. Since the present stem of the verb had an e as its vowel (i.e. *gwhen‑),
the *gwh turned into a th rather than a ph, so the present stem ended up as thein-, not exactly the
most obvious counterpart to that past-tense stem pephn-.
19  C. Watkins, How to Kill a Dragon: Aspects of Indo-European Poetics (Oxford, 1995), pp. 357–8.
In addition to killing the Chimera, Bellerophon is especially famous for taming the winged horse
Pegasus.
20  A hint of this stronger use has survived in the name of the villain in the Batman movie The
Dark Night Rises, Bane.
21  C. Watkins, How to Kill a Dragon: Aspects of Indo-European Poetics (Oxford, 1995), p. 418.
152 How dead languages work

whose name, ormr, is the Old Norse equivalent of the Old English word
wyrm used of Beowulf ’s foe.22
There is plenty of death in ancient Sanskrit, Greek, and Germanic literature, and
several words vied to be the basic word for “kill”. But what makes the descendants
of *gwhen- stand out is the extent to which they all cluster around particularly
violent acts of killing: when you have to kill a dragon, only *gwhen- will do.

THE HIDDEN NAMES OF THE DAWN-COWS

Yet another shared motif in early Indo-European poetry is the idea that speech
can be deliberately obscure. In the Theogony, an account of the origins of the
cosmos and the gods, the poet Hesiod, frequently mentioned by the ancient
Greeks in the same breath as Homer, describes his initiation by the Muses into
the arts of song. As he’s tending his sheep on Mount Helicon, the goddesses
address him as follows:
poimenes agrauloi, kak’ elenchea, gasteres oion,
idmen pseudea polla legein etymoisin homoia,
idmen d’, eut’ ethelōmen, alēthea gērȳsasthai (Hesiod Theogony 26–8)
Field-dwelling shepherds, base reproaches, mere bellies,
We know (idmen) how to speak (legein) many false things (pseudea) that are like
 (homoia) true things (etymoisin),
And we know, when we’re willing, how to utter (gērȳsasthai) true things (alēthea).

It is one of the most vigorously debated passages in early Greek poetry, but one
thing seems clear: for the goddesses of poetry, there is no simple correspond-
ence between the words they utter and the truth of the content. And a general
contrast between the low physical needs of the shepherds, and the higher-order
sophistication of the Muses’ art is also apparent. But virtually every word in the
passage invites further investigation:
• The main verb is of impeccable Indo-European heritage, for it is with the
verb idmen that the Muses present their gifts as knowledge: divided into a
root id- and a first-person plural ending ‑men, this verb originally had a
w- at the start. Once this is restored, the kinship between (w)id-men and
the “knowledge” root seen in the name of the Vedas becomes clearer.
• What follows seems to be a simple opposition: the Muses can either tell the
truth or they can tell lies that are like the truth. But there are three reasons

22  Ibid., p. 422. As seen in Chapter 4, the -r on the end of ormr is the Old Norse nominative
singular ending equivalent to Latin -us and Greek -os.
 Sanskrit153

that this opposition is not so clear-cut, none of which come across readily
in translation:
(i) two different words for truth are used: etymos (the source of English
“etymology”, though, ironically, its own etymology is unclear) and
alēthēs (which originally will have meant something like “unhidden,
unconcealed”);
(ii) two different words for speech are used: the Muses speak the false-
hoods with the ordinary verb legein “to speak” (this is the verb that
corresponds to the noun logos); but the truths they are said to
gērȳsasthai, a comparatively rare verb found exclusively in poetry,
used often of song (including that of birds), which seems to throw
weight on the aural quality of the truths—more “utter” or “voice”
than “speak”;
(iii) the word translated “like”, homoios, may range in meaning from “the
same as, identical to” to “resembling”. This semantic ambiguity is
actually quite remarkable from the perspective of later controversies
in Christian theology. In the fourth century ad, when the relation-
ship of God the Father and Jesus was being worked out in writings
such as the Nicene Creed, an important distinction was drawn
between the homoousians, who held that substance (ousia) of God
the Father was the same (homos) as that of Jesus, and the homoious­
ians, who believed that Jesus’ substance was only similar (homoios)
to that of the Father. By that point, the iota had come to make a much
bigger difference than it did a thousand years earlier, in Hesiod’s day.
Thankfully, for our purposes here, we do not have to solve any of these prob-
lems, which would take a great deal of ink to disentangle. What matters here is
only this: poetry is difficult to understand—in part because it deceives you into
thinking it’s easy to understand—and early Greek poets made note of this.
This brings us to our final passage of Vedic Sanskrit, from a hymn to the sky
god Varuṇa, RV 8.41.5:

. . . य उस्राणामपीच्या वेद नामानिगुह्या ।

स कविः काव्या पुरु रूपं द्यौरिव पुष्यति . . .
ya usrāṇ ām apīcyā
who of the dawn cows secret
veda nāmāni guhyā
knows names hidden
sa kaviḥ kāvyā puru
that poet poetic arts many
rūpaṃ dyaur iva puṣyati
beauty sky like causes to flourish
154 How dead languages work

The one who knows the secret, hidden names of the dawn-cows, that poet
causes his many poetic arts to flourish, as the sky does its beauty.

The language in this passage once again draws heavily on inherited Indo-
European vocabulary.
• At a basic grammatical level, the two words ya and sa are directly related
to the equivalent words in Greek, the relative pronoun hos (ὅς), and the
demonstrative pronoun ho (ὁ), which later turned into the definite
article.
• The verb in the subordinate clause, veda, is the third-person singular of
the verb “to know”, which, curiously, has the form of a perfect (as if it were
“has known”), but regularly has the sense of a present tense. If it were in the
first-person plural, “we know”, it would take the form vidma. This matches
exactly the corresponding forms of the Greek verb—also perfect in form
but present in meaning—which was in the passage above from Hesiod’s
Theogony:

  Sanskrit Greek PIE

“s/he knows” ved-a (w)oid-e *woid-e
“we know” vid-ma (w)id-men *wid-me-

In other words, both Greek and Sanskrit show (i) a change in the vowel of the
root between the singular and plural, (ii) a third-singular ending that is just a
vowel, (iii) a first-plural ending that starts with an m (although the original
termination of that ending is unclear); (iv) endings that look like the perfect
tense but have the meaning of the present tense. And this verb is used to
describe poetic knowledge in both passages.23
• The word for “names”, nāmāni, is related to English name, just as one
would expect, and also to Greek onoma and Latin nōmen. In the singular,
it would take the form nāma, and the n that crops up in forming the plural
is the same n that occurs at the end of the Latin word.24 Something similar
occurs with the German word for “name”: as the subject of a sentence, it’s
Name, but it belongs to a small class of nouns that adds a final -n when
used as an object.
• The adjective puru “many” is related to both the Latin word plūs “more”
and the Greek word poly “much, a lot”. A little more is going on with the
Latin word, but the ancestral word from which both puru and poly derive

23  That these similarities go back to Indo-European and are not just specific to a “Greco-Indic”
branch can be seen by turning to Modern German, where the verb for “to know”, wissen, not only
goes back to the same root but also shows the same alternation between a fuller vowel in the sin-
gular (er weiß “he knows”) and a weaker vowel in the plural (wir wissen “we know”).
24  That n is also there in the -a of Greek onom-a, but that’s too long a story to tell here.
 Sanskrit155

was probably pronounced something like “pull-hoo” (or, as Indo-

Europeanists would spell it, *pl h̥ 1u), with the schwa-like vowel of the first
syllable developing into a u in Sanskrit but an o in Greek.25 That leaves the
Sanskrit r to be explained. As it happens, l and r are similar sounds to
begin with, and many of the world’s languages have only one or the other.
Japanese, for instance, has only r, so when foreign words with l are bor-
rowed into Japanese, the l’s all turn into r’s: thus, the English words salary
man become Japanese sarariman. So too in Sanskrit, inherited l’s generally
became r’s, leading not only to puru but also to the following equation:
English light (i.e. the opposite of darkness), Latin lūc- “light”, Greek leukos
“white”, but Sanskrit ruc- “to shine”.
• The word for sky, dyaur, comes from another root that’s already made an
appearance. The root itself is dyau-, and the -r is the nominative ending
that marks this as a subject or, as here, an appositive to the subject (“as the
sky”). Previously we’ve seen this ending as -s, but because of the sandhi
rules discussed earlier, it turns into an -r here. The root, meanwhile, is the
same as that in the names of the Greek and Roman sky-gods Zeu-s and
Ju-piter, which would have had the form *dieu- in PIE.26
• At this point, it’s time to go back to the first line and look at usrāṇām,
which, following Watkins, I have glossed as “the dawn-cows”. It’s not, per-
haps, the most obvious of compound nouns in English, but, as the main
etymological dictionary of Sanskrit tells us, there are in fact two words
with the stem usra-, one meaning “reddish, of the morning”, the other
meaning “cow, ox”, and the poetic language of the Rig Veda is obscure
enough not to allow a sharp line to be drawn between the two.27 The latter
sense is probably secondary, arising from the reddish color of the cattle in
question, as etymological connections outside the Indic languages suggest
that the former meaning was the earlier of the two. As an adjective, usra- is
a derivative of the noun uṣas-, which belongs to a well-attested set of Indo-
European words for “dawn”:

1. Variation in the form of the word for “dawn” in different Greek dia-
lects—Attic heōs, Ionic ēōs (seen in the geologic epoch called the

25  It is also possible that the Greek and Sanskrit forms derive from slightly different ancestral
forms, with the Greek going back to *polh1u (R. Beekes, Etymological Dictionary of Greek (Leiden,
2009), p. 1221) rather than *pl̥h1u (H. Rix, Historische Grammatik des Griechischen: Laut- und
Formenlehre (Darmstadt, 1992), p. 163, A. L. Sihler, New Comparative Grammar of Greek and Latin
(Oxford, 1995), pp. 43, 104). Either way, the symbol *h1 represents one of the so-called laryngeal
consonants reconstructed for PIE; its exact pronunciation is very unclear, but one possibility is
that it was like an English h.
26  For more on this root, see the discussion of Zeus’ name in Chapter 2 in the section on the
Iliad.
27  M. Mayrhofer, Etymologisches Wörterbuch des Altindoarischen, vol. 1 (Heidelberg, 1992), p. 239.
156 How dead languages work

Eocene), and Aeolic auōs—points to an original Proto-Greek *auhōs—

which in turn would have arisen from an earlier *ausōs, since inter­
vocal­ic h in Greek typically developed from s.
2. In Latin, the same putative starting point *ausōs, which will have had a
fairly complicated inflection, was regularized by the addition of -a to
the stem, making it a straightforward feminine noun of the first declen-
sion. Moreover, whereas Greek turned intervocalic s’s into h’s that were
then lost, Latin turned them into r’s—the same sort of change seen
above in Sanskrit, with the final -s of dyaus turning into an r before the
initial vowel of the following word iva. This left Latin with aurōra as its
word for “dawn”.
3. The Germanic languages developed new words for “dawn”—dawn itself
is related to day—but the old word survived in a somewhat shifted
semantic domain. Dawn breaks pretty regularly in one particular part of
the sky, and the Germanic languages preserved the derivative *austa- as
the designation of the east. (Germanic au most often becomes ea in
Modern English: this is not the most transparent of sound changes, but
it does account for equations like German Baum “tree” = English beam,
and Laub “foliage” = leaf.) Furthermore, when Germanic speakers were
searching for the best term to designate the central holiday of Christianity,
the day of the Resurrection, rather than simply borrowing the Hebrew
word for Passover, pesaḥ, which passed into Greek and Latin as Pascha
(whence Spanish Pascua and French Pâques), they turned instead to the
old root that referred to the dawn—because, if we are to trust Bede (De
temporum ratione 15), the festival of a pagan goddess of the dawn was
held at about the same time. And so we now celebrate Easter.
Having spent this much time looking at trees, we should step back and
examine the forest. Once again, here’s a translation of the excerpt: “The one
who knows the secret, hidden names of the dawn-cows, that poet causes his
many poetic arts to flourish, as the sky does its beauty.” This sentence encapsu-
lates wonderfully the poetry of the Rig Veda: it draws on core vocabulary com-
mon to all the Indo-European languages—from grammatical items like relative
pronouns and demonstratives to basic vocabulary like the words for “sky”,
“dawn”, “name”, and “know”—and then stretches that vocabulary to its limits.
The successful poet—and it’s an oversimplification to translate kavi simply as
“poet” here, considering that it has religious connotations that also suggest
“priest”—does not just traffic in superficial, obvious words but in secret, hidden
names as well. The poet can obscure the difference between the red of dawn and
the red of a cow or, to go back to parvatānām in the second passage, can blend
together two strands of imagery when speaking of waters enclosed in a mountain—­
or perhaps in a cloud.
 6

Old Irish and the Celtic languages

When I was considering different titles for this book, one of my crueler friends
suggested Welsh, Irish, and Other Dead Languages. I’m especially fond of the
Celtic languages, so I couldn’t quite inflict such negative publicity on them, but
it’s not hard to see where he was coming from: look at a map of the places in
Europe where the Celtic languages are still spoken, and their future appears
precarious indeed. Occupying only a small fringe of land along the Atlantic—
the western parts of Ireland, Wales, Scotland, and Brittany—these languages are
minority languages at best. Still, in recent years, lovers of linguistic diversity have
been pleased to see renewed interest in keeping them alive and, indeed, in bring-
ing back to life those, like Cornish and Manx, that had died out by the early
twentieth century. Curiously, the resurgence of the Celtic languages is not seen as
a good thing by all. I have heard otherwise liberal, tolerant Englishmen complain
about bilingual road signs in Wales, with the reasoning: “If everyone here speaks
English, why clutter up the signs with Welsh?” But apart from the fact that
speakers of Celtic languages ought to be afforded the same protections as any
other minority, the question raised does deserve a response. What do the Celtic
languages offer us? Quite a lot, in fact. The earlier stages of the languages devel-
oped some very unusual features, like initial consonant mutations, conjugated
prepositions, and infixed pronouns, many of which are still preserved today.
Most of the chapter will focus on Irish—both the medieval and modern versions
of the language—but we’ll finish off with Welsh, which will illustrate well how
two languages can be closely related yet still end up looking very ­different indeed.

THE ECCENTRICITIES OF THE IRISH LANGUAGE

Old Irish, the form of Irish spoken from roughly the seventh through ninth
centuries ad, is arguably the most difficult of the Indo-European languages,
almost willfully so. To understand why, consider the following sequence,

How Dead Languages Work. Coulter H. George, Oxford University Press (2020).
© Coulter H. George. DOI: 10.1093/oso/9780198852827.001.0001
158 How dead languages work

which, as languages change over time, repeats itself many times over. While
these general principles were first introduced in Chapter 4, in the discussion of
weak and strong verbs in Germanic, it is worth expanding and reframing it
here to highlight the particular way it helps explain the strangeness of Irish:
1. A language starts off with a regular pattern of endings to indicate, say,
case and number. Latin, for instance, had nouns whose stem originally
ended in an -s, like honōs “honor”, with a nominative plural formed by
adding -es on to this: honōsēs.
2. Then a sound change affects part of the paradigm but not all of it: in early
Latin, any -s- between two vowels changed to an -r-, meaning that honōs
now had a plural honōrēs. This was now a more complicated plural forma-
tion, since it required not only adding an -ēs but also changing an s to an r.
3. Because the untidy state of affairs caused by the sound change presents
something of a cognitive challenge, the r then spreads to the singular,
where it wasn’t the result of regular phonological change, in order to
restore the pattern whereby the nominative singular looks like the nom­
ina­tive plural but without the final -ēs (and with shortening of the vowel
in the final syllable). This phenomenon is dubbed analogical leveling,
since it proceeds according to the operation of a so-called four-part analogy,
based on other nouns, like āctor “driver, actor”, that didn’t have a change
in consonant in the stem:

If the nominative plural is . . . Then the nominative singular should be . . .
āctōrēs           āctor
honōrēs          [not honōs, which no longer follows the
“drop the -ēs” rule, but rather . . .] honor

Given that there were plenty of nouns, like āctor “driver, actor”, with this
pattern whereby the nominative singular was (apart from vowel quantity)
the same as the nominative plural minus the -ēs, honōs was brought into
line with that same pattern, and the singular duly became honor.1
4. At this point a new regularity has been established, and we’re back to stage
1, where, once again, a new phonological change could potentially disrupt
the situation, triggering a further round of leveling.
How is this relevant to the difficulties of Old Irish? Probably the most import-
ant reason why it’s such a challenge is the sheer irregularity of some of the
paradigms for various nouns and verbs, which results from the general failure

1  There was in fact much more than just the nominative plural to act as a force encouraging the
leveling: all the other case forms, like the genitive singular honōris, the dative singular honōrī,
the accusative singular honōrem, and so forth, would have also increased the pressure on honōs
to conform.
 Old Irish and Celtic languages 159

of analogical leveling to take place in Old Irish. Take the word for “woman”
(which, admittedly, is irregular even by Old Irish standards): in the singular, it’s
ben if it’s the subject of a verb, but mnaí if it’s the object; in the plural, it’s mná as
either the subject or object, but ban if it’s possessive, or mnáib if it’s the object of
certain prepositions. Most languages would have eliminated the variation
between b- and m- by leveling one or the other throughout the entire set of
forms. Or, to conjugate the verb “to go” in the present tense (“I go, you go, he/
she/it goes, we go, y’all go, they go”), one must remember the sequence tíagu,
téigi, téit, tíagmai, tíagthae,2 tíagait. (Because of the variation in the endings, no
subject pronouns are necessary.) And even that only gets you half of the way to
knowing the present, because, in a second, even more egregious complication,
the forms just mentioned (the so-called absolute) are only used if the verb
comes first in its clause; if it comes after certain particles, like ní “not”, then a
different set of endings (the conjunct) is used. Thus, “I do not go etc.” becomes:
ní tíag, ní téig, ní tét, ní tíagam, ní téit, ní tíagat. Side by side, the two sets of
forms are as follows:
1sg tíagu    ní tíag
2sg téigi      ní téig
3sg téit     ní tét
1pl tíagmai ní tíagam
2pl tíagthae ní téit
3pl tíagait ní tíagat

It gets worse: the verb “to come” is a compound built to téit by adding the prefix
do· to it. When the compound verb comes first in its clause, then the prefix do·
has the same effect that ní does, and “I come etc.” is do·tíag, do·téig, do·tét, etc.
But if one has to negate the compound verb with ní, then another round of
phonological reductions takes place, such that “I don’t come etc.” becomes ní
táeg, ní taíg, ní táet. But wait: it gets even worse. Throughout this paragraph, I’ve
only given one variant form per slot in the paradigm. In fact, there is consider-
able variation in the spelling of our manuscripts of Old Irish: while interlinear
glosses in eighth- and ninth-century manuscripts give us a good idea of what a
classical Old Irish orthography would have looked like, the bulk of the interest-
ing literary material (such as the great prose epic, the Táin Bó Cúailnge “The
Cattle-Raid of Cooley”) survives only in manuscripts from the Middle Irish
period, such as the twelfth-century Lebor na hUidre (“The Book of the Dun
Cow”), in which the spellings are haphazardly updated to reflect contemporary
developments in the language.
In short, Old Irish is a bit of a mess: not only does the lack of analogical
­leveling mean that individual paradigms often look very irregular and that verb

2  This form is not actually attested but is a reasonable guess for what it might have been.
160 How dead languages work

forms have proliferated beyond what would be expected in other languages but
we can’t even get a clear picture of much of this linguistic exuberance because
the actual texts of most medieval Irish literature jumble up forms from differ-
ent periods of the language.
But back to the Lebor na hUidre. There’s something unusual about the spell-
ing of the title of that manuscript: what’s a lower-case h doing in front of the
capital U? True, readers will have come across such sequences in early twenty-
first-century brand names like eBay or iPod. But the spelling hUidre is not the
product of an orthographic fad in tech companies. Rather, it shows one of
the most characteristic features of the Celtic languages: an initial mutation, the
change of the initial sound of one word that is triggered by the word that pre-
cedes it. In this instance, the word na is the feminine genitive singular definite
article (“of the”), and, in the mutation called aspiration, it causes a following
word beginning with a vowel to be prefixed by an h sound. The other two muta-
tions in Old Irish are lenition, which generally changes stops like p, t, and c to
fricatives like ph, th, and ch (pronounced as in phone, thin, and loch respective-
ly), and nasalization, which originally added a nasal sound (n or m) to the
following word, although by the late Old Irish period, the prefixed nasal had in
turn caused further changes to following consonants, e.g. turning voiceless
stops p t c into their voiced counterparts b d g (though this is not actually
­indicated in Old Irish orthography).
But why would a language be so perverse as to have words cause changes at
the beginning of the words that come after them?3 The explanation, as so often,
is historical. We have already seen that nouns in the older Indo-European lan-
guages change their ending in order to indicate its grammatical function in the
sentence. Thus, in Latin, a horse that is a sentence subject, in the nominative
case, is equus. If it’s the object, it’s in the accusative: equum. And if it is posses-
sive (i.e., if you say that something is “of the horse” or “the horse’s”), then it’s in
the genitive: equī. In well-behaved Indo-European languages like Latin or
Ancient Greek, these endings, inherited from the parent language Proto-Indo-
European (PIE), survive as markers of the noun’s syntactic role in the sentence.
Celtic originally had such endings too, and indeed they survive intact in the
earliest Celtic inscriptions, notably the Ogam inscriptions of Irish (largely
fifth–sixth century ad), in which -i is often still found as a genitive ending (as
in Latin equ-ī). But by the time of classical Old Irish (eighth–ninth century ad),
all but the longest of these endings had disappeared.

3  Some readers may ask: Is this really any different from the sandhi seen in Sanskrit, where e.g.
devas asti “the god is” becomes devo ’sti? While it’s certainly related—like sandhi, the initial muta-
tions involve sound changes across word boundaries—it’s also different in that sandhi is (virtually
always) simply a phonological process: it’s predictable purely on the basis of the sounds involved.
The initial mutations in Celtic, however, can’t be predicted in the same way because (as will be
seen below) they are determined also by the grammatical form of the word causing the change.
 Old Irish and Celtic languages 161

Disappeared—but not without leaving a trace behind. As it happens, the

system of initial mutations maps very closely onto the lost case endings.
Consider the following table, which compares the Latin and Old Irish for
“white horse” in various cases (with the adjective following the noun in both
languages):

case Latin Old Irish mutation

nominative equus albus ech bán (b pronounced like English b) none
genitive equī albī eich báin (b pronounced like English v) lenition
accusative equum album ech mbán (mb pronounced like English m) nasalization

In the Old Irish nominative singular (at least of this particular class of
nouns), the ending, which as in Latin had originally ended in an -s, was lost
without affecting the following word at all. But in the genitive singular, while
the ending in -ī still existed, the initial b of the adjective báin was sandwiched
between two vowels, which caused it to be lenited (that is, softened) to the v
sound. (Intervocalic consonants are often prone to this sort of change: as a
parallel, consider the pronunciation of d in Spanish nada, which has weakened
to something like the th of other.) Then, in the accusative singular, where the
ending originally ended in an ‑m, that -m in effect became detached from the
noun: if there were no adjective following, ech would be the same in both nom­
ina­tive and accusative; but with an adjective like bán following, the m survived,
first as a sound pronounced before the b, then later replacing the b altogether.
If we look at the genitive, we see another characteristic feature of the sound
system of Old Irish: palatalization. The vowel that had marked the genitive, -ī,
is what’s called a high front vowel because the tongue is far forward in the
mouth (like ee in English, when contrasted with oo) and also close to the roof
of the mouth (again like English ee, but this time contrasted with ah).4 High
front vowels often exert an influence on the consonants that come before them,
pulling a stop like k, pronounced at the soft palate (the velum), forward to the
hard palate: feel the difference in where the tongue hits the roof of your mouth
in car and key. This movement of consonant sounds to the hard palate is called
palatalization, and it is a systematic part of Old Irish. It has also occurred in
English, turning earlier k sounds into ch sounds: that’s why the German Kinn
equates to English chin, and why the noun formed from the adjective cool is
chill and not kill.5 (That said, it didn’t occur regularly everywhere in all English
dialects, which has led to doublets like kirk and church.) In what sense, then, is
such palatalization systematic in Old Irish? That -ī genitive singular ending
regularly caused palatalization of the preceding consonant before it was lost.
And this is what’s indicated with the i’s in eich báin: the ei and ái aren’t actually

4  One may also compare the location of i in the vowel triangle in Chapter 5.
5  See also the discussion of alliterating g in the final section of Chapter 4.
162 How dead languages work

diphthongs; rather, the vowel stays the same as in the nominative ech bán, and
what’s pronounced differently is the final consonant: the ch of ech is the ch of
Bach and loch, but the ch of eich is that of German ich, halfway between the
non-palatalized sound and the sh of ship.6 Likewise, the n of bán is close to the
regular English n, whereas the n of báin is pronounced with the tongue flat-
tened against the roof of the mouth, in a sound similar to Spanish ñ or Italian
gn. One final instance of palatalization familiar to English speakers: Irish s is
always palatalized before e and i, in which case it is pronounced like English sh:
hence Sinn Fein, Sinead, and Seamus.
If we turn now to Modern Irish, where the spelling rules are somewhat more
regular than in Old Irish, we can see that the necessity of representing both
initial mutations and palatalization is what makes it look so distinctive on the
page: the former leads to unusual consonant clusters, often involving h, and the
latter to outré sequences of vowels. Taken together, one can end up with a word
like bhfaoileán (“sea-gull”, with initial nasalization; pronounced roughly vee-
lon (IPA: /viːljɑːn/)).7 How do modern spellings such as this come to pass?
First, lenition, the softening of articulation of previously intervocalic con­
son­ants is now regularly indicated with an h (and not just at the start of words,
where it’s caused by what used to be at the end of the preceding word, but also
when the same changes took place word-internally). In the Old Irish example I
gave above, with bán in the nominative, and báin in the genitive, there’s nothing
in the spelling to indicate that the b in the second is pronounced like a v. Now,
however, that first sound of the latter word would be written bh to show the
mutation. This is true of a whole host of sounds, leading to the following cor­
res­pond­ences in Modern Irish:

The following lenited sounds . . . are pronounced like English . . .

ph f
th, sh h
ch ch as in loch
bh, mh v or w, depending on dialect and
phonetic context
dh, gh like ch in loch, but voiced (a bit like gargling)
fh silent
Some of these pronunciations are extremely unintuitive for English speakers: fh
is strange enough on the page that skipping over it altogether seems a reasonable

6  Because the palatalization of the final consonant affects the transition from the vowel to that
consonant in such a way that one can actually hear an i-glide after the vowel, notating palataliza-
tion with the i before the relevant consonant makes a certain amount of sense. We’ll return to
this below.
7  So as not to complicate the broad phonological picture with phonetic particulars, especially
the differences between Modern Irish dialects, I adopt minimalist IPA transcriptions in line with
Foclóir Póca: Irish Dictionary (Dublin, 2009).
 Old Irish and Celtic languages 163

course of action. But the fact that th is weakened simply to an h can cause the
novice real trouble when tackling the Irish name for Dublin, Baile Átha Cliath.
Not only are both vowels of the first word and the final vowel of the second so
weakly pronounced as to be essentially lost in ordinary speech but this leaves
the two th’s in effect stranded at the ends of words, where the h, too, goes miss-
ing, leaving something like blaw-clee-uh as the phonological residue. (In
theory, it would be by-law-huh-clee-uh if it were actually articulated as it’s
spelled.) As for the pronunciation of sh as h, most anglophones will have come
across this without knowing it in the doublet of personal names, Seamus and
Hamish. The latter originated as the lenited variant of the former, as the voca-
tive particle a, used before personal names in direct address, caused lenition:
Séamas in the nominative, but a Shéamais (“O Seamus!”) in the vocative (we’ll
return to that additional i shortly). The th has one other curious effect: when it
occurs after a voiced stop, like g, it devoices it, in the case of g to a k, and the h
itself is lost. Thus the word fágtha “left (behind)” is pronounced roughly fock-
uh (IPA: /fɑːkə/), making it a remarkable example of an Irish spelling that looks
like one rude word in English but sounds like another.
The other major initial mutation is nasalization, seen in the accusative above,
ech mbán, where an original nasal consonant (either n or, as here, m) has been
lost, but not before affecting the first sound of the next word, here a b that has
turned into an m. Other consonants mutate in similar ways:

Original . . . becomes nasalized . . . which is pronounced . . .

p bp b
t dt d
c gc g (as in got)
b mb m
d nd n
g ng ng (as in sing)
f bhf v or w, depending on dialect and
phonetic context
For the most part, the sound changes here are straightforward: the voiceless
stops, p t c, become the voiced stops b d g, and the voiced stops b d g become the
corresponding nasals m n ng. The voiceless fricative f turns into the voiced frica-
tive v, but this often weakens to w—just as also happens to the lenited b. In all
seven, the new sound is written before the old sound, which it then replaces. This
overshadowing of the one sound by the other gives rise to the name nasaliza-
tion usually goes by in Modern Irish (rather than Old Irish) grammars: eclipsis.
For the purposes of understanding odd-looking Irish spellings, the most
­relevant of the seven is the convention of spelling the v or w sound that results
from nasalized f as bhf: the f is there to indicate what the original sound was,
and the bh is there to tell you that it’s now pronounced as if it were a lenited b.
164 How dead languages work

The last feature of Irish phonology which has major effects on the orthography
of the modern language is palatalization. As already noted, this refers to the shift
in the place where consonants are produced in the mouth, usually conditioned by
a front vowel, e or i, causing consonants articulated farther back, like velars, to
move forward (once again, that’s the c of car becoming the k of key) and those
produced farther forward, like dentals, to move back (say, a t becoming more like
a ts, with the tip of the tongue flatter against the hard palate). Such a distinction in
consonant articulation based on the following vowel sound is very common in the
world’s languages, but Irish goes further with it than most, in that every consonant
except h has two variants, one palatalized (or slender, i.e. key-sounds), one non-
palatalized (or broad, i.e. car-sounds). Now Irish orthography never acquired a
dedicated symbol to indicate palatalization—unlike Russian, which has a similar
feature, and uses the so-called мягкий знак (myágkiy znak “soft sign”), ь, to mark
palatalized consonants.8 Instead, Irish employs an intricate system of vowel-letter
sequences to indicate which consonants are slender and which are broad. In the
left-hand column of the following examples, which are by no means exhaustive,
(i) italics are used for slender (palatalized) consonants, (ii) bold for broad
(non-palatalized) consonants, (iii) underlining for vowel letters that are intro-
duced specifically to show palatalization (or not) of neighboring consonants:
1. Because e and i naturally trigger palatalization, and a, o, and u don’t, the
basic rule is that consonants next to e and i are slender, and those next to
a, o, and u are broad.

dom “to me” One expects broad consonants next to o.

agus “and” One expects broad consonants next to a and u.
sin “that” One expects slender consonants next to i.
tine “fire” One expects slender consonants next to e and i.

2. Sometimes, however, consonants next to a, o, and u are slender, so these

need to be marked with additional e and i signs, which are not vowels in
their own right but simply show palatalization:
baile “town, place” Normally an a would be followed by a broad
consonant, so the i is needed to show that
the l is slender.
Seán “Sean” The e is inserted to show that the s before
the á is slender.
3. And sometimes consonants next to e and i are broad, so these need to be
marked with additional a, o, and u signs (with the choice of sign dependent
on the particulars of the phonetic environment):

8  To be more precise, it marks those palatalized consonants that are not already signaled as
such by the following vowel.
 Old Irish and Celtic languages 165

Séamas “Seamus” Normally an e would be followed by a slender consonant,

so the a is needed to show that the m is broad.
Fionn “Finn” The o shows that the nn is broad, even though it is
preceded by i.
muid “we” Here u indicates that the preceding m is broad,
despite the i.

4. The system is further complicated by the fact that all vowels have long and
short counterparts, and these also need to be distinguished. Normally, an
acute accent marks long vowels, and short vowels are left unmarked, but
sometimes particular combinations of vowels that look like they indicate
diphthongs in fact show long vowels.

Gaeltacht “Gaeltacht”9 The spelling ae is how what might otherwise be

é is spelled when it has a broad consonant on
either side. The a says, “Don’t palatalize the
preceding consonant even though there’s an é
vowel”, while the e, which ought to indicate
that the following consonant is slender,
misbehaves, and doesn’t do so here.
Gaeilge “Irish” Once again the actual vowel is long é; once
again it’s preceded by a broad consonant, and
so requires the a. Here, however, it’s followed
by a slender consonant, so an additional i is
necessary to override the convention that ae
has broad consonants on either side.
laoch “warrior” The vowel in this word is a long í, but because
the consonants on either side are both broad, it
is spelled ao.
taoiseach “prime Once again the actual vowel in the first syllable
minister” is long í; once again it’s preceded by a broad
consonant, so it’s spelled ao. Here, however, it’s
followed by a slender consonant, so an additional
i is necessary. The vowel of the final syllable is
actually a schwa, so one could theoretically regard
the e as the representation of the schwa, with
the a showing that the final ch is broad—or, as
I’ve notated it, with the a as the schwa, and the
e confirming that the preceding s is slender.

9  This is the word that designates the part of Ireland where Irish is still spoken.
166 How dead languages work

This is all very complicated, of course, but what it has in its favor is that, to a
certain extent, the vowels that are inserted in order to indicate whether a con­
son­ant is broad or not often mimic the sounds that are produced as the tongue
shifts from the “proper” vowel to the consonant. To understand the idea of a
transitional sound like this, consider what happens in English when we hear a
sequence of a stop consonant (like p or t) plus vowel. What we typically per-
ceive as a difference in consonant (such as pack vs tack) is actually, from an
acoustic perspective, a difference in the very initial stage of the production of
the neighboring vowel. This makes a certain amount of sense: when one pro-
nounces a p or a t, it’s the result of a momentary stoppage of air flow through
the mouth that only resumes once the lips open up or the tongue is released
back down from the top of the mouth. Only once that air is flowing again—that
is, only once the stoppage in the mouth that defined the stop as a stop in the first
place has ceased—can an actual sound be heard. In other words, a p sounds
different from a t not because of what’s happening in the mouth at the time of
its articulation as a stop (because no sound is being produced during the stoppage
proper) but rather because of what happens during the very quick transition
from the stop to the neighboring vowel. So too here, in the case of Irish palatal-
ized and non-palatalized consonants, if one contrasts bád “boat” and báid
“boats”, the standard description is that the final d of the singular is broad
(roughly pronounced bod; IPA: /bɑːd/), whereas the final d of the plural is slen-
der (roughly bodge; /bɑːdj/), and that palatalization is marked by the insertion
of the i. But as the tongue moves from its position for the low central vowel á to
the palatalized d, it’s very hard for it not to go by way of a slight i sound—often
called, because of this transitional role, a glide. The same is also true of at least
some of the vowels that indicate broad consonants: the latter are not just “nor-
mal” variants of their slender counterparts. Rather, they’re pronounced with
what’s called velarization: the back of the tongue is raised up towards the soft
palate at the same time as the consonant is pronounced. Thus the sequence mui
in muid (“we”) is not just a regular m followed by an i, in which case the u would
not be an especially natural orthographic choice; instead, the broad m itself
has a slight w glide after it, and a u is a reasonable way to indicate this.
We’re now in a position to see why it is that Irish and English spellings of the
same proper noun can differ so much, and, indeed, why some proper nouns
that have found their way into English have alternate forms. Consider once
more the pair of names Seamus and Hamish: we saw that they went back to two
different case forms, nominative Séamas and vocative a Shéamais. The voca-
tive, then, is marked not only by the leniting particle a but also by palatalization
of the final consonant. Readers who know Latin may recall that, in masculine
nouns of the second declension (those that end in -us, like equus “horse” at
the start of our discussion of initial mutations), the ending of the vocative is -e.
That is, if Marcus is the subject of the sentence, he’s Marcus, but if you’re
addressing him by name, you say Marce. That final -e was also the mark of the
 Old Irish and Celtic languages 167

vocative in the earliest stages of Irish as well, and, as a front vowel, it caused
palatalization of the final consonant of the stem. This didn’t happen with the
nominative, however, so Séamas ends with a broad s (as seen by the fact that it’s
preceded by an a), whereas Shéamais ends with a slender s (as seen by the inser-
tion of the i). The resulting contrast then plays out in the more common English
forms of the name, where Seamus is pronounced with an s at the end, but
Hamish with an sh.
The following Irish spellings should also now be at least a little more
comprehensible:
• The surname O’Flaherty is spelled Ó Flaithearta in Irish. The th is just an
h, and the i and e on either side just serve to mark it as slender.10
• The given name Saoirse, pronounced Seer-shuh, means “freedom”, and is
derived from the adjective saor “free”. As noted above, the ao sequence indi-
cates a long í vowel, but with broad consonants on either side: thus, the s is
an s, and not an sh sound. But the noun has the suffix ‑se, which itself has a
slender s (thus the sh of the second syllable), and that in turn causes the final
r of the adjectival root to turn slender, which requires the insertion of the i.
• The surname Ward corresponds to Irish Mac an Bhaird. The basic noun here
is bard “poet”, a word borrowed into English in its own right. But the sur-
name as a whole means “the son of the poet”; bard, then, needs to go into the
genitive case. And, just as with the vocative, this originally ended in a vowel
that caused palatalization (recall the -ī of the Latin genitive singular in the
eich báin example above), so the final rd cluster needs to be slender, as indi-
cated by the i. As for the start of the word, it’s preceded by the definite article
an, which, because it’s masculine genitive singular (a form which had earlier
ended in a vowel), causes the initial b to lenite to bh, pronounced w.
But while the orthography of Irish is curious enough, what really makes the
language stand out is its syntax. Whereas most of the Indo-European languages
place the sentence subject first, then either the verb followed by the object (as
in English or French), or the object followed by the verb (as often in Latin), the
Celtic languages place the verb first, then the subject, then the object. Consider
the following sentence, from the opening paragraph of Scéla Muicce Meic Dathó
“The Tale of the Pig of Mac Dathó”, often the first Old Irish beginners read
(much as Caesar is a traditional entry text for Latin):

im·díched in cú Laigniu uili

protected the dog Leinstermen all
The dog protected all the Leinstermen.

10  As it happens, now that the th is pronounced h, there’s no difference between the broad and
slender pronunciation, but earlier it was pronounced like the th of thin, and did have a distinctively
slender pronunciation.
168 How dead languages work

In a Germanic language, if we were to keep the order of the main constitu-

ents, “Protected the dog all the Leinstermen”, it would be most natural to read
this as a question. In English, of course, such questions now require the use of
the helping verb do at the start—Did the dog protect them?—but in Shakespeare’s
day (and up through the present in German and Dutch), simple inversion of
subject and verb was enough to turn a statement into a question. Not so in Old
Irish, where the simple statement has the verb first and, to turn it into a ques-
tion, one would have to add the particle in at the start.
But even languages that have verb-initial word order often like to move
either the subject or the object in front of the verb. What comes first is more
prominent in the ears of the addressee, and so if a speaker wishes to emphasize
the object of the verb, even English makes use of various strategies to move that
object to the front of the sentence, a process often designated by the remarkably
straightforward technical term fronting.11 Emphasis on its own is often enough
(I want THIS one), but there’s also the option of reordering the sentence, either
(i) by simply moving the object before the subject (THIS one I want, THAT one
I don’t), or (ii) by placing the object in a clause of its own at the start of the
sentence, then relegating what would otherwise be the main clause to a subor-
dinate status: THIS is the one that I want, or It’s CAKE that she wants. This sec-
ond strategy becomes very common in the Celtic languages as a work-around
to get not only objects but also subjects into the slot before the verb. Compare
the following two Modern Irish sentences; first, the default way to say “I spent
a year in Ireland”:

chaith mé bliain in Éirinn

spent I a year in Ireland

Second, the way to emphasize that it’s a year (and not some other length of
time):

(is) bliain a chaith mé in Éirinn

(it’s) a year that spent I in Ireland

In the second formulation, the verb “it is” comes first, acting as a peg from which
to hang the object, bliain “year”; then comes the relative pronoun, followed by
what would otherwise have been the main verb: “it’s a year that I spent in
Ireland”. But, as indicated by the parentheses, the verb is may be omitted, lead-
ing to a structure which looks a little less like it’s based on a relative clause, and
a little more like a simple object–verb–subject sentence, but with the addition
of the particle a to facilitate the fronting of the object. In any event, because
Irish moves elements of the sentence forward in this way so frequently, it’s even

11  In addition to emphasis, fronting can also be used to indicate topicalization, as seen in the
discussion of Latin word order in Chapter 3.
 Old Irish and Celtic languages 169

caused the same structure to be particularly common in the English spoken in

Ireland as well (“’Tis joking you are, I suppose”).12
Another hallmark of Irish is the use of so-called conjugated prepositions.
Those who have studied some Spanish or Latin are accustomed to the conjuga-
tion of verbs, but what does it mean for a language to conjugate a preposition?
Whereas verbs in the Indo-European languages are marked with endings to
indicate the subject, prepositions in Irish (and Welsh) have endings that mark
their object. Consider the Modern Irish preposition ag “at”. If it has a noun as its
object, then it works like a preposition in any other language: ag an teach is
“at the house”. But if it has a pronoun as its object, then the preposition and
­pronoun are expressed as a single unit, and this can be laid out in a paradigm
like those used for verb conjugations:

at me agam at us againn
at you (sg.) agat at you (pl.) agaibh
at him aige at them acu
at her aici    
The paradigm is a bit of a mess, and it’s representative of the amount of irregu-
larity one has to contend with in learning the Irish prepositions. Still, some
patterns are fairly consistent, especially in the first and second persons, where
the singular ends in -m and -t, and the plural in ‑inn and -ibh. The bh here is
pronounced like a v, and the Indo-European origin of these endings is clear to
anyone who knows Latin or a Romance language, since they correspond dir-
ectly to the first consonant of object pronouns like French me, te, nous, vous.
But the third-person forms are beset with seemingly haphazard forms. One
could describe those for ag by saying that the (i) masculine and (ii) feminine
singular, and (iii) plural forms are characterized respectively by (i) palatalization
(-i-) of the final consonant and the ending -e, (ii) palatalization (-i-) and devoi-
cing (-c- for -g-) of the final consonant and the ending -i, (iii) devoicing of the
final consonant and the ending -u. Some of those patterns can be seen elsewhere,
but, on the whole, one is left in the uncomfortable position of having to learn
multiple forms to cope with what in other languages would just be a single prep­
os­ition; here, for instance, are the corresponding forms of ar “on” and le “with”:

orm orainn liom linn

ort oraibh leat libh
air orthu leis leo
uirthi   léi  

12  R. Ó hÚrdail, “Hiberno-English: historical background and synchronic features and vari-
ation”, in The Celtic Englishes, edited by H.  L.  C.  Tristam (Heidelberg, 1997), p. 190, cited by
M. Filppula, “The making of Hiberno-English and other ‘Celtic Englishes’ ”, in The Handbook of
the History of English, edited by A. van Kemenade and B. Los (Malden, MA, 2006), pp. 526–7;
other examples include “’tis at home we should be, ’twas money they wanted, ’tis well you looked”.
170 How dead languages work

Again, in the first and second persons, the basic -m -t -inn -ibh pattern is evi-
dent, and one can count on a fairly stable stem (or- and li-, apart from the lea- of
leat).13 But the third-person forms are only vaguely similar to those for ag.
With ar, there’s a different vowel in the stem for all three, no ending on the
masculine singular, and rather than devoicing of the final consonant, there’s the
introduction of th—which, since it’s pronounced h, is actually not that far off
from the devoicing seen in aici and acu. (Recall that th devoices g to c in the
word fágtha.) As for le, the masculine singular is equipped with an s, and the u
of the plural has combined with the e to become a long ó—spelled eo because
the preceding consonant is slender. Naturally.
The student of Irish might well be tempted to muddle through all this as
quickly as possible, but learning these prepositions thoroughly is important, in
part because they do a lot more work than their English equivalents. Many
common phrases that, in English, have the pronoun as the subject of a verb are
impersonal in Irish (that is, they have no personal subject), and what would
otherwise be the subject becomes instead the object of a preposition:

I have a book → There’s a book at me

    tá leabhar agam  
    there’s book at me  
I like music → Music is good with me
    is   maith  liom    ceol
    it’s good  with me music
I’m sorry → There’s sorrow on me
    tá brón orm  
    there’s sorrow on me  
I want a pint → There’s a pint from me
    tá pionta uaim  
    there’s a pint from me  

It’s not uncommon for languages to express possession not with a dedicated
verb “to have” but with a construction like that used by Irish. Among the more
widely-known European languages, Russian is a good parallel, with I have a
book becoming “at me book”: у меня книга (u menyá kníga). And German too
is happy to use impersonal constructions more widely than English: I’m sorry is
Es tut mir Leid (lit. It does me sorrow). But Irish goes further than most languages
in just how much it gives its prepositions to do. In the last example, uaim, which
is essentially just ó combined with the first person suffix -m, the preposition,
which would normally mean “from”, acquires the weightier sense “wanted by”.

13  Note that even these stems, which one might expect to be a bit more regular, don’t match the
form the prepositions take if they’re occurring on their own with a normal noun as an object,
namely ar and le.
 Old Irish and Celtic languages 171

(The semantic pathway behind that shift in meaning: that which is from me is
away from me, which is to say distant from me, therefore lacked by me and
potentially needed or wanted by me.) If the speaker wants, the force of the
preposition on its own can be strengthened by changing the verb to the more
specific teastaíonn “it is wanted”, but the sense is generally clear just from the
preposition. In effect, Irish has gone further in giving its prepositions a seman-
tic workout than even Thucydides and Paul: in the sentences quoted from these
authors in Chapter 2, prepositions bore a similar functional load, but in Greek
it comes across as a marked expression, while in Irish it’s perfectly normal.
Now if a language conjugates prepositions, then that eliminates a lot of the
need for object pronouns. That is, if the me of of me, to me, or from me is just an
ending on a preposition, then there’s less need for a separate word for me. True,
one might want a separate word for me if one wants to emphasize it—“this pint
is for me, not for you”—but Modern Irish takes cares of this by using special
suffixes on the conjugated prepositions that, except in a couple of instances, are
not connected with particular persons or numbers: “with me” is liomsa, “with
you” is leatsa, with -sa added to the base form in both cases. Still, one also needs
pronominal forms to act as objects of verbs. And while the modern language
has indeed developed independent forms of the pronouns that serve this pur-
pose, the final curiosity of Old Irish to deserve mention represents a different
way of dealing with this: the infixed pronoun.
At the start of this chapter, we saw that the Old Irish verb is rather messy to
begin with, in that (a) it has different forms depending on whether it comes at
the very beginning of its clause or not, and (b) there are lots of compound verbs
with prepositional prefixes. Moreover, these prefixes count as one of the initial
particles that shift the ending to the one that’s used if the verb doesn’t come first
in the sentence. It’s best to review this here with an additional example. Take the
verb gaibid “s/he takes”, which has the following third-person singular forms:

gaibid in n-ech vs ní gaib in n-ech

he takes the horse vs he does not take the horse

In the first sentence, gaibid occurs at the very beginning of the sentence, and so
takes the fuller, absolute ending; in the second, the negative particle ní pre-
cedes, and so the verb takes the shorter, conjunct ending, which in this slot in
the paradigm is no ending at all, since gaib is, on its own, the verbal stem. (Ech,
the word for “horse”, is nasalized to n-ech, since the definite article in front of it,
in, is in the accusative case, and it will thus have originally ended in a nasal, just
as is true of the accusative singular endings of Latin -um and Greek -on.)
Now, if one puts the prefix fo· in front of the verb (in origin, the same prefix
as Greek hypo-), the meaning changes from “take” to “find”, and the verb can
now only draw on the second set of endings. Anyone who likes imposing gram-
matical terminology will be pleased to learn that compound verbs in this
­context are called deuterotonic, since they have the accent (Greek tonos) on
172 How dead languages work

their second (deuteros) element; this is indicated by the raised dot, which pre-
cedes the accented syllable:
fo·gaib in n-ech (accent on gaib)
he finds the horse  

But what if one wants to put yet another particle, like the negative, in front?
Then the accent shifts to the prefix (leading to a prototonic verb), and other
changes take place as well, especially in the vowels of the word:

ní fagaib in n-ech (accent on fa-)

he doesn’t find the horse  

Now, in all these clauses, the object of the verb has been a noun, so there’s been
no need for an object pronoun. But if Old Irish wants to say He finds me or, for
that matter, He doesn’t find me, then a slot has to be found for the object pronoun
(since it’s not a stressed word in its own right). In particular, it goes after what-
ever the first particle is, so, given that the infix for me is -m, and that for you
(singular) is -t, one finds forms like:

fo·gaib “he finds” fom·ġaib “he finds me”

̇
ní fagaib “he doesn’t find” ním·fagaib “he doesn’t find me”
fot·ġaib “he finds you”
̇
nít·fagaib “he doesn’t find you”

In the first form of each pair, fo is the first particle, so the object pronoun leans on
that; in the second, ní is the first particle, so the object pronoun comes after that
and before the preverb fo—which, when stressed, becomes fa. As if that were not
enough, these particular pronouns cause lenition, indicated here with the dot
over the following letter, meaning that the g is pronounced like a German ch but
voiced (a gargling sound), and the f is silent.14 This is all rather complicated, but
at least there’s the silver lining that, since Old Irish is an Indo-European language,
the fact that it’s m and t that serve as these pronouns once again maps directly on
to Latin, where me and you, as accusative objects, are mē and tē, and the Romance
languages have preserved this pair as well. Even remembering the lenition
caused by the infixes -m and -t is made easier by bearing in mind that the
ancestral forms from which both they and mē and tē derive ended in a vowel.
Now, with a compound verb like fo·gaib, there’s always a preverbal element
after which the infix can be placed. With simple verbs, however, the situation is
a little trickier. If there’s a particle, like the negative, in front, then the infix can
go between that and the main verb:

14  Indication of lenition with a superscript dot is standard Old Irish orthography only for f and
s, but I have extended it here to g for convenience’ sake. As lenition of g is not indicated in normal-
ized texts, which would instead print fom·gaib, the reader would simply have to know that this -m
would cause lenition of the following g.
 Old Irish and Celtic languages 173

ní gaib “he doesn’t take” ním·ġaib “he doesn’t take me”

But if the verb comes in the absolute first position, then a dummy particle, no-,
is added at the start of the sentence to act as a host for the particle:

gaibid “he takes” nom·ġaib “he takes me”

So far, I’ve used first- and second-person infixes for all the examples because
these are the easiest to spot: -m and -t in the singular, and -nn and -b in the
plural, all of which recall the endings seen in the conjugated prepositions.
With the third-person forms, however, distinctively Celtic phonology and
syntax come together, since careful attention to the consonant mutations is critical
to understanding which infixed pronoun one is dealing with. The masculine
third-person singular (“him”) is an -a that causes nasalization of the following
consonant, while the neuter (“it”) is an -a that causes lenition of the same. To
make matters still more confusing, this -a swallows up the preceding o of the
prepositional prefixes, but is itself swallowed up by the í of the negative particle,
such that the only indication that there’s an infixed pronoun is the mutation of
the following consonant. Some examples will help clarify all of this.
Before adding the infixed pronoun:

fo·gaib “he finds” ní gaib “he does not take”

After adding the infixed pronoun; in the left column, it can be seen in the
replacement of fo· by fa·, as well as the mutational effects on the g; in the right
column, where the vowel of ní overrides the pronoun itself, only the consonant
mutations reveal its presence:

fa·ngaib “he finds him” ní·ngaib “he does not take him”
fa·ġaib “he finds it” ní·ġaib15 “he does not take it”

Finally, to give one last idea of how rich (or frustrating) Old Irish is: all the
examples so far have only drawn on the so-called Class A infixed pronouns.
There are two more classes, and which class is used depends not only on the
phonology of the prefix (Class B is used if the prefix originally ended in a con­son­
ant) but also on the syntax (Class C is restricted to particular types of clauses).

OLD IRISH IN ACTION

After this quick tour of some of the most peculiar features of Old Irish, we’re in
a better position to see what Old Irish literature has to offer. But first, a brief

15  As an actual Old Irish text would neither punctuate prefixes consistently nor have a dot over
the g to indicate lenition (see previous note), there would be no way of distinguishing between ní
gaib “he does not take” and ní·ġaib “he does not take it” apart from context.
174 How dead languages work

disclaimer. We have a good idea of how the classical Old Irish of the eighth and
ninth centuries worked thanks to the Irish monks of the time, roughly contem-
poraneous with Charlemagne: their texts of, say, the Psalms or Pauline epistles
were in Latin, but in between the lines of Latin, they would write in the Irish
equivalents, much as modern students of Latin are prone to do with English in
their textbooks. (It never ceases to amaze me that there were once people who
found it easier to understand Old Irish than Latin.) These interlinear glosses
preserve a remarkably uniform language, and it is on them that the main gram-
mars of Old Irish are based. But reading Old Irish glosses on biblical material is
not particularly fulfilling from a literary standpoint, and the works of broader
cultural interest that are preserved directly from the Old Irish period include
only a handful of poems in early manuscripts. Fortunately, more extensive prose
tales survive in manuscripts from the Middle Irish period (tenth through twelfth
centuries). Yet, while these stories have their origins in the Old Irish period, or
even earlier, the language in which they have come down to us contains a mix-
ture of Old and Middle Irish forms, leading to considerable headaches in work-
ing out how best to normalize the language in modern scholarly editions.
Still, it’s worth trying to get at the language of these tales because they contain
wonderful stories in a distinctively Celtic tradition. We’ll limit ourselves here to
a couple of sentences from the most important of them, the Táin Bó Cúailnge
(“Cattle Raid of Cooley”). The Táin belongs to what is known as the Ulster Cycle,
an important set of tales featuring the king of Ulster, Conchobor, and his chief
warrior Cú Chulainn. In the Táin, the rival king of Connacht, Ailill, and his
rather more impressive queen Medb,16 mount a raid to seize a particularly fine
bull that’s kept in Ulster, but their aims are repeatedly thwarted through the
actions of Cú Chulainn, who acts as a sort of Irish Achilles. What makes Cú
Chulainn different from Achilles—and, with it, the Táin from the Iliad—is a
much more outlandish degree of fantasy in the description of his prowess in
battle. He is particularly associated with the ríastrad, variously translated as
“warp-spasm” or “torque”. Here are just a few of its more picturesque effects:
The first warp-spasm seized [Cú Chulainn], and made him into a monstrous
thing, hideous and shapeless, unheard of. His shanks and his joints, every knuckle
and angle and organ from head to foot, shook like a tree in the flood or a reed in
the stream. His body made a furious twist inside his skin, so that his feet and shins
and knees switched to the rear and his heels and calves switched to the front . . . His
face and features became a red bowl: he sucked one eye so deep into his head that
a wild crane couldn’t probe it onto his cheek out of the depths of his skull; the other
eye fell out along his cheek . . . [I]f a royal apple tree with all its kingly fruit were shaken
above him, scarce an apple would reach the ground but each would be spiked on a
bristle of his hair as it stood up on his scalp with rage. (trans. T. Kinsella, pp. 150–3)

16  Her name is often anglicized as Maeve: both the d and b are lenited, so would be pronounced
as a sequence of the voiced th of then, followed by a v.
 Old Irish and Celtic languages 175

As a flashback within the Táin, some Ulstermen who happen to be with the
Connacht army let their hosts know the sort of fighter they’re facing by recount-
ing some of the deeds he accomplished while still a young boy being fostered in
the household of Conchobor. One of the more striking episodes occurs when
Cú Chulainn is a seven-year-old, and Cathbad the druid arrives.17 It so happened
that Cathbad was asked what that day would be good for, and the wise druid
said that the warrior who took up arms that day would achieve such fame that
the stories about him would last forever. The young Cú Chulainn goes and tells
Conchobor that he would like to take up arms. Precocious child that he is, all
the arms he tries out shatter under his strength until he’s given Conchobor’s
own arms. Cathbad then finds out and is upset: the warrior who took up arms
that day would indeed be famous, but his life would also be short. The seven-
year-old replies that that’s fine by him:
Acht roba airdirc-se, maith lem
provided that be famous-I, good with me
ceni    beinn  acht  óen-láa  for  domun
though not I be  but  one-day on  earth
As long as I’m famous, I’m happy, though I should not spend but one day on earth.

That Cú Chulainn should prefer fame to a long life has an exact parallel in
Achilles’ words in the ninth book of the Iliad, when he tells of the dilemma his
divine mother Thetis has warned him he faces: “If I stay here and fight for the
city of the Trojans, my homecoming (nostos) is destroyed, but my fame (kleos)
will be imperishable (aphthiton). But if I go home to my dear fatherland, my
noble fame (kleos) is destroyed, but my life will be long, nor would the final
limit of death reach me quickly” (9.412–16). Achilles, of course, stays to fight,
and, within the Iliad, at least, this choice of kleos over long life remains in place.
That said, in the Odyssey, Achilles’ shade in the Underworld famously appears
to question the wisdom of this decision, saying that it would be better to be the
lowliest serf and still be alive than to be the king of all the dead (11.488–91). It is
probably no coincidence that the hero of that poem manages to achieve both
kleos and nostos.
As for the words with which Cú Chulainn expresses his eagerness for fame,
they show a couple of the features of Irish we’ve already seen:

• Rather than a personal expression of happiness (I’m happy), Cú Chulainn

says, just as one would in Modern Irish, “(it is) good with me” (maith lem),

17  The word druid, borrowed from a Celtic language by way of Latin, has a first-rate Indo-
European etymology as well. The first element, dru-, means (and, thanks to Grimm’s Law, is
­cognate with) “tree” or, more specifically, “oak”, and the second, -id, is from the *weid- root “see,
know”, making druids “tree/oak-knowers/seers”.
176 How dead languages work

with ellipsis of the dummy subject it and verb is; lem is a conjugated
­ reposition, equivalent to Modern Irish liom in the examples given earlier.
p
English of course has similar structures (It’s fine by me), but they are
­particularly typical of Irish.
• To express the first person in the first clause, an emphasizing pronoun -se
is used. The verb form roba is present subjunctive, but could be either
first- or second-person singular, and the ‑se specifies that this is first-person,
since the second-person would be marked with -siu instead. (This particular
distinction in the emphasizing pronouns is no longer maintained in
Modern Irish.)
Individual words in the Old Irish offer additional points of interest:
• The word for “famous”, airdirc, ultimately means “very visible”, with -dirc,
the second element, connected with the PIE root *derk-, which as a verb
means “to see”. This same root is found in Greek drakōn “serpent”, whence
English dragon; originally, the snake was “the seeing one”.
• Both forms of “to be”, roba and beinn, have a b that is, etymologically, the
same as the b- in English be. The first form, roba, has a preverb ro-, which has
a wide range of meanings, mostly perfectivizing, i.e. indicating that the ver-
bal event is viewed as complete.18 It has the same origin as the Latin prefix
pro-, but, in one of the most distinctive Celtic sound changes, the p was lost.
This loss led to the following correspondences, which show how different a
word can come to look once it loses an initial sound as prominent as a p:

  Latin Old Irish

“father” pater athair
“fish” piscis íasc
  penna “feather” én “bird”
• Finally, the preposition for “on” also reflects the loss of an original p.
The parent language had a preposition meaning “above” that had the shape
*(s)uper, with a poorly understood optional s, which gave rise to Greek
hyper, Latin super, and English over. In Celtic, the p was lost, and so *uper
became *uer, with the u pronounced as a w in this environment.19 In Irish,
however, this initial w sound regularly turned into an f (more on this
below, in the section on Welsh), bringing *uer- one step closer to for. As for

18  One may thus compare the Germanic prefix ge- and the Latin prefix com- discussed in
Chapter 4.
19  This prefix still had the form uer- in Gaulish, the Celtic language spoken by the inhabitants
of what is now France when Julius Caesar conquered it in the 50s bc. It’s found in the name of the
Gauls’ most important leader, Vercingetorix (the V would have been pronounced as a w in Caesar’s
day), which can be analyzed as follows: Ver- “over” + cingeto- “step” + -rix “king” (cf. Latin rēx),
i.e. “the king who steps over”.
 Old Irish and Celtic languages 177

how the e became an o, that was caused by contamination from the vowel
of the word with which “over” is often paired, “under”, *(s)upo (whence
Greek hypo and Latin sub), which became first *uo, then, in Old Irish,
fo-, the prefix we already saw in the compound verb fo·gaib “s/he finds”.
We’ll close this section with the declaration Cathbad had made about the
fame of the warrior born that day, the words that elicited the response of
Cú Chulainn’s discussed above:

óclach no gébad gaisced and, for·biad a ainm

warrior (that) would take up weapons on it, over·would be his name
ar gnímaib  gaiscid   firu   Érenn    ocus  no mértais
for deeds    of weapons  men  of Ireland  and   would remain
a airscélae co bráth
his great tales for ever
A warrior that would take up weapons on it, his name would be over the men
of Ireland for deeds of weapons, and great tales of him would remain for ever.

By now, much of what is characteristically Irish about this has already

been seen:
• In the second two clauses, the verb (for·biad and no mértais) comes before
the subject (a ainm, a airscélae).
• In and, we have another conjugated preposition.
• With for·biad we see the “over” prefix again.
• With firu “men”, once more a word that starts with a w sound elsewhere
(Latin vir “man”, whence English virile) begins with an f in Irish.
• We’ve already seen both elements in airscélae: air- as an intensive prefix in
airdirc, and scélae “tales” in the title of Scéla Muicce Meic Dathó “The Tale
of the Pig of Mac Dathó”.
• With no gébad “would take up”, we see both the verb gaibid “take” again, as
well as the dummy particle no, here used for a different purpose, namely
to form the conditional tense (more or less the “would” of the English).
• The word for name, ainm, belongs to the same Indo-European root as the
words in the other major languages seen so far: Greek onoma, Latin
nōmen, Sanskrit nāma, and English name. It’s no coincidence that all five
have both n and m in that sequence.
• The most unexpected connection, however, is afforded by the last phrase,
co bráth, as this is the earlier form of the final two words of Erin go bragh
“Ireland forever!”, the cry heard on St. Patrick’s Day in Irish immigrant
communities around the world. “Forever”, however, is too bland a transla-
tion to do justice to the literal meaning of the phrase: “until Doomsday”.
178 How dead languages work

WELSH

We next turn to Welsh, today the most widely spoken of the Celtic languages.
This sometimes comes as a surprise: given the size (and boisterous pride) of the
Irish diaspora, not to mention the fact that the Republic of Ireland is an inde-
pendent country, Irish would at first glance seem to have had a better chance of
holding its own against English. But only about 100,000 people, less than 2% of
the population, live in the Gaeltacht, the part of Ireland where Irish is still spoken
regularly outside the classroom, whereas about 562,000 people in Wales can
speak Welsh—nearly 20% of the population.20 Of course, this should not be
taken to mean that there’s a particularly broad understanding of the language
throughout the country: in 2008, an official road sign appeared in Swansea with
the English “No entry for heavy goods vehicles. Residential site only” translated
into Welsh as Nid wyf yn y swyddfa ar hyn o bryd. Anfonwch unrhyw waith i’w
gyfieithu, which actually means “I am not in the office at the moment. Send any
work for translation.” Now a more rigorous proofreading system should no doubt
be in place to prevent such slips, but, for our purposes, it’s worth pointing out that
there’s not a single word in the Welsh sentences above that resembles anything in
either the actual or the faulty English translations, with the possible exception of
nid, which looks like it could be a negative word (and negatives are indeed there
in both English versions). Contrast the first sentence of the mistranslation with
the same phrase in French or German: Je ne suis pas au bureau pour le moment,
or Ich bin im Moment nicht im Büro: in both of these other languages, there are a
couple of words that have entered the common vocabulary of many European
languages, bureau and moment, that would act as a clue that this did not actually
mean No entry for heavy goods vehicles. Welsh, though, with swyddfa for office
and ar hyn o bryd (“at this point of time”), offers no such easily recognizable
words as a starting point for decipherment. Instead, it offers a confusing assort-
ment of w’s and y’s where one might expect to see proper vowels.
First, then, we should clear up what is probably the most common miscon-
ception about the language, namely, that it’s short on vowels. It actually has
plenty of them, and the only reason it doesn’t look like it is that all those w’s and
y’s are doing full duty as symbols for vowels. The Welsh vowel w is pronounced
like the oo in English pool, and thus is indeed a sort of “double u”, in keeping
with the name of the letter in English. Once one knows this, then the names of
various countries, like Hwngari, Rwsia, and Twrci look decidedly less alien. It
also helps with borrowed words like pwdin “pudding” and, with the additional
information that a circumflex accent indicates a long vowel, siampŵ. Why
couldn’t Welsh just use u to indicate this sound? Because it has come to be used
for a different vowel, the high central vowel (IPA /ɨ/), the second vowel in

20  For those keeping score, Breton has about 200,000 speakers, and Scottish Gaelic about
60,000.
 Old Irish and Celtic languages 179

roses.21 One might at first think that that vowel is just a schwa, but most US
English speakers in fact have a contrast between Rosa’s and roses, with the e of
roses pronounced with the tongue closer to the roof of the mouth: this is the
value of Welsh u. As for y, it’s more complicated because it has two different
sounds depending on where it falls in the word. It has a stronger sound—identical
to the Welsh u—if it occurs in a stressed monosyllable or in the final syllable of
a polysyllabic word; but it is pronounced more weakly, as a schwa, if it’s in an
unstressed monosyllable or in any syllable other than the final syllable of a
polysyllabic word. To untangle that last sentence, some examples will no doubt
be helpful. First, consider the monosyllables, since with them the rule makes
sense: in a stressed word like rhyd “ford”, y has the stronger /ɨ/ sound; in the
unstressed definite article y, it’s just a schwa. But the polysyllabic rule is counter-
intuitive. Normally, it’s the penultimate syllable that’s stressed, so, in mynydd
“mountain”, the first y is stressed, but pronounced like a schwa, while the sec-
ond is unstressed, but pronounced /ɨ/. The whole word, then, sounds like the
mon of money followed by something resembling the ith of with (assuming a
voiced pronunciation of th as in breathe). The same two vowels are also heard
in that sequence in Cymru, the Welsh name for their own country.
That the dd of mynydd should be pronounced like a voiced th is also odd to
most English speakers, and that brings us to the consonants. But before tack-
ling dd, we should first cover another sound, the humble p, which lies at the
heart of one common classification of the Celtic languages, a division based on
what happens to the inherited sound *kw. In the so-called P-Celtic languages,
like Welsh, it turned into a p; but in the Q-Celtic languages, like Irish, the kw was
retained at first, although it would later simplify to just a k sound, spelled c in
Irish. This is easiest to see at the start of words, as in the following examples,
with Latin retaining the original *kw sound as qu:

  Latin Welsh Old Irish

“four” quattuor pedwar cethair
“five” quīnque pump cóic
“head” -- penn cenn

In the middle or at the end of words, the sounds in question were subject to other
changes. In the word for “son”, for instance, Irish retains the c in mac, but Welsh,
after changing the *kw to p, then voiced the p to b, leading to mab—which is prob-
ably the first element in the name of the most famous collection of Middle Welsh
tales, the Mabinogion. (The exact meaning of the rest of it is subject to considerable
debate.) Now complications related to the classification of ancient Celtic lan-
guages, such as Gaulish, have rendered the specific terms P-Celtic and Q-Celtic

21  Strictly speaking, this is true only for North Welsh dialects; in South Welsh, u has drifted still
further forward to become simply the i of English tip. For simplicity’s sake, I restrict discussion
here to the short-vowel values of u and y.
180 How dead languages work

somewhat contentious. But the differing treatment of inherited *kw remains one of
the main diagnostic features underlying the primary dichotomy of the modern
Celtic languages: the P-Celtic languages Welsh, Cornish, and Breton are all classi-
fied as Brythonic, in keeping with their British origins, whereas Q-Celtic Irish,
Scottish Gaelic, and Manx are all termed Goidelic languages. (The word Gaelic
itself is essentially just an alternative form of Goidelic; while it can be applied to
any of these three languages, it is mostly used of the Scottish variety.)
Another sound characteristic of Welsh is gw. It’s particularly frequent because
it’s the outcome of one of the most common sounds in PIE, *w. In contrast to
Welsh, the *w was preserved in Latin (though spelled v in the  traditional
orthography), and it became an f in Irish. One thus gets cor­re­spond­ence sets
like the following:

  Latin Old Irish Welsh

“men” virī fir gwŷr
“true” vērus fír gwir
“wine” vīnum fín gwin
“white, fair” -- find gwyn

(Combining the words for “head” and “white” from these two lists, we see that
the Welsh for “white head” is penn gwyn. While the etymological dictionaries
aren’t entirely convinced that this is the origin of the English word penguin, it’s
too charming an idea to leave unmentioned.22)
Three further consonants that are especially characteristic of the written appear-
ance of Welsh are the double letters ff, dd, and ll. First, a single f in Welsh is pro-
nounced like an English v, leaving the sound of English f to be spelt with the double
ff: a farm is thus a fferm, a telephone is a ffôn (as before, the circumflex indicating a
long vowel), and to say goodbye is ffarwelio. And the voiced v sound represented by
a single Welsh f is quite common because consonants in Welsh, as in Irish, were
subject to lenition. The results are, to be sure, somewhat different, but just as Irish
lenites b and m to bh and mh, both pronounced v or w, so too Welsh has a lot of
loanwords from Latin in which original b and m were lenited to Welsh f:
• The Welsh for “capital city”, prifddinas, has as its first element prif  “chief ”,
a borrowing from Latin prīmus “first”, with the m lenited to the v sound,
spelled f.
• Latin fōrma “form, shape” became Welsh ffurf: the initial f remained, but
is spelled ff; the m, meanwhile, lenited to f.

22 The OED, which has a lengthy and excellent etymological note for the word, says it’s probably
correct, but the American Heritage Dictionary only that it’s possibly so. Readers may also wish to
turn to P. Durkin, The Oxford Guide to Etymology (Oxford, 2009), pp. 264–5, which explores the
possibility that since (i) the word penguin was probably first used of the now-extinct great auk,
which did not in fact have a white head, and (ii) there is an island in Newfoundland formerly called
Penguin Island (i.e. White Headland Island) that was frequented by the great auk, it may well be
that the birds take their name from the place rather than from a description of their own features.
 Old Irish and Celtic languages 181

• In Latin barba “beard”, the first b remained intact, but the second was
lenited; with the loss of the final vowel seen in forma as well, we’re left with
Welsh barf (rhymes with “starve”).
Readers of J. R. R. Tolkien will have come across this value of f before since,
along with Finnish, Welsh was one of the two languages that most inspired him
in his creation of the linguistic world of Middle Earth. Those familiar with
Appendix E of The Return of the King, in which the basic phonological and
orthographic rules of his languages are set out, will know that the f of Gandalf
is to be pronounced as a v, a quirk that is lifted straight from Welsh (though
Tolkien restricts it to word-final position). But one can also find a parallel for
these two spellings within English itself: of and off, where the f is voiced v and
the ff voiceless f. The final f of of actually arose in much the same way as the
voiced f of Welsh: originally the same word as off, and thus ending with a voice-
less f, the preposition of was pronounced with weakened stress, causing what
amounts to the lenition of the final consonant to the v sound it has today.
Next, there’s dd. Welsh, like English, has a contrast between voiceless and
voiced th sounds, the th of thin and them respectively. The former is spelled th,
the latter dd, and it’s often the result of leniting a regular d, as can be seen in the
Welsh word for “sword”, cleddyf. Before lenition took place, the Celtic word
would have had a d in the middle, a sound still there when Latin borrowed the
word as gladius (whence English gladiator). The lenition of what was once
intervocalic d has gone still further in Irish and Scottish Gaelic: while the cor-
responding Old Irish word, claideb, still had a voiced th in the middle and a v at
the end, this has become simply claíomh in Modern Irish (pronounced—but
not related to—cleave /kliːv/), claidheamh in Gaelic (pronounced cleye-of
/ˈklaiə̃v/). The Scottish Gaelic form is especially worth citing here because it’s
the source of the English loan word claymore, from claidheamh mòr “big sword”.
The word for “big” in the various Celtic languages is also familiar to anglo-
phones from various place names in both Welsh (Bryn Mawr “big hill”) and
Gaelic, in a host of names for single-malt whiskies: Ardmore (“big height”),
Bowmore (“big reef ”), Dalmore (“big field”).
But the most distinctive sound in Welsh is indubitably the one spelled ll. In
proper phonetic terms, it’s a voiceless lateral alveolar fricative (IPA /ɬ/), which
means that it belongs to the same general family of sounds as the regular l, only
with more turbulent airflow around the tongue, and without vibration of the
vocal cords. In layman’s terms, it’s a bit like a cross between the th of thin and
the l of lip. Historically, this sound is actually the regular outcome of inherited
l, so a Welsh dictionary will list a lot more words starting with ll than with simple
l (most often, simple l in Welsh is merely the lenited variant of original ll): a
letter is a llythyr; Monday, the day of the moon (Latin lūna), is Dydd Llun, and
Latin itself is Lladin.
One final consonant digraph, analogous to ll, is rh. A trilled, voiceless r, this
sound is to a regular trilled r as an s is to a z, and, like the ll, is the r-sound that
182 How dead languages work

occurs by default at the start of a word, with the regular voiced r being its lenited
variant. It also represents another feature of Welsh phonology and orthography
that Tolkien borrowed in creating Sindarin, where it occurs in place names like
Rhûn and Rhovanion.
The phonological details described above, together with the orthographic
conventions for representing them, give Welsh much of its distinctive look. It
would be hard to confuse written Irish and Welsh (though Welsh-language
signs have mistakenly turned up in Scotland, and Scottish Gaelic signs in
Wales, so evidently this sort of confusion isn’t impossible). But these compara-
tively superficial differences obscure considerable similarities between Welsh
and Irish that justify the placement of both these languages in a single Celtic
branch of the Indo-European family tree. In addition to a wealth of shared
lex­ic­al items, some of which have been seen in the tables above, there are also
numerous grammatical correspondences:
• Like Irish, Welsh has consonant mutations. In the sample sentence of
Welsh at the start of this section, neither the word for time, bryd, nor the
word for work, waith, would be listed under b or w in the dictionary. Both
are lenited because of the preceding words, and would otherwise have
been pryd and gwaith respectively.
• Like Irish, Welsh has verb–subject–object word order:

darllenith y ferch y llyfr

will read the girl the book
The girl will read the book
• As in Irish, the verb is often a form of “to be”, which is very common as a
helping verb. In such constructions, the main verb of the sentence takes
the form of a verbal noun—which, as the name suggests, is a form of the
verb that acts as a noun, in the example below as the object of the particle
yn, which acts like a preposition:

mae ’r ferch yn darllen y llyfr

is the girl particle reading the book
The girl is reading23 the book

23  The -ing form has different values in the gloss and in the translation. In the gloss, which
mimics the syntax of the Welsh, it is in effect the object of a preposition and is therefore a gerund.
In the translation, it is acting as a predicate adjective with is and is thus a participle. Furthermore,
while it is tempting to gloss the particle yn as “in” because it can be used with regular nouns in this
sense, I have opted for a more agnostic position: since the particle used before the verbal noun
does not have the same mutational effects as the regular preposition “in” (indeed, they have sep-
arate entries, as yn2 and yn1 respectively, in the chief Welsh dictionary, the Geiriadur Prifysgol
Cymru), they may not be the same word in origin.
 Old Irish and Celtic languages 183

• Once again, prepositions are important. Whereas the present progressive,

for example, is formed with yn (as above), the perfect tense uses the
­preposition that means “after”:

mae hi wedi darllen y llyfr

is she after reading the book
She has read the book (lit. “She is after reading the book”)

• Because of the verb-initial word order, the English speaker is tempted to

understand all the word-by-word glosses above as questions, even though
they are declarative statements. To turn them into questions, other strat­
egies must be used, as in Irish. With the verb “to be”, special forms of the
verb are used:

ydy ’r ferch yn darllen y llyfr?

is? the girl particle reading the book
Is the girl reading the book?

In the case of the future tense (seen a few examples above in the form
­darllenith), lenition of the verb marks it as an interrogative:

ddarllenith y ferch y llyfr

will read? the girl the book
Will the girl read the book?

• Welsh has conjugated prepositions. For the most part, though, the forms
follow more regular patterns than they do in Irish. Contrast what both
languages do with ar “on”:

Irish   Welsh  
orm orainn arna i arnon ni
ort oraibh arnat ti arnoch chi
air orthu arno fe arnyn nhw
uirthi   arni hi  

While the Irish paradigm sports a variable stem and, when compared with
other prepositions’ conjugations, the endings are far from consistent, the Welsh
forms are built to a stable stem arn- with the endings much the same from
preposition to preposition. But an even more prominent difference is that in
Welsh the person and number are marked doubly: not only does the prep­os­
ition inflect but a prepositional object is included as well, rendering the inflec-
tion itself somewhat superfluous.
184 How dead languages work

• Welsh has numerous impersonal constructions, often matching those of Irish:

I have a book → There’s a book at me

    mae llyfr ’da fi (South Welsh)
    there’s  book  with  me   
I’m sorry → It’s in a bad way with me
    mae  ’n     ddrwg  ’da  fi 
    it’s  particle24 bad  with me 

But Welsh does not go quite as far as Irish in this respect. To say that one likes
or wants something, one regularly expresses the person whose preferences are
related as the subject of the verb, and the item that is being assessed as the
object, as with English like and want; that said, Welsh has a Celtic propensity
for using the helping verb and verbal noun in what looks like it ought to be
equivalent to an English progressive formation:

I like music → I am liking music

    dw i ’n hoffi cerddoriaeth
    am I particle liking music

Indeed, Welsh has even borrowed the English verb like as leicio—not to men-
tion, the noun music as miwsig (with the diphthong iw representing the com-
position of English long u as y + oo), yielding the anglicized alternative dw i’n
leicio miwsig. Recalling the faulty translation at the start of this section, we can
see even better now how easily such a mishap could happen with Welsh: with
this last example, we have a sentence where both the verb and the object are
borrowed directly from English, and the subject pronoun i, through historical
accident, is spelled with exactly the same letter, but the combination of Welsh
orthography (leicio and miwsig don’t look very much like like and music) with
Welsh syntax (the helping verb dw at the start, the seemingly random ’n that
introduces the verbal noun) would make it virtually impossible for an anglo-
phone to guess its meaning without some training in the language.
Such features can give Welsh an even greater sense of strangeness when
they’re embedded in texts from an earlier period of the language. To see this in
action, we’ll look at a sentence from one of the stories generally known as the
Mabinogion.25 These eleven tales, found in two major manuscripts dated to the
fourteenth to early fifteenth centuries, but containing language perhaps three
hundred years old by that point, draw not only on some of the literary t­ raditions

24  The particle’n (elided from yn) introduces predicate adjectives. Its mutational effects are
different from either of the two words yn mentioned in the previous note, and it too gets its own
entry in the Geiriadur Prifysgol Cymru, as yn3, as it is certainly a separate word in origin.
25  As noted earlier, the mab element in the name is probably the Welsh word for “son”, cognate
with Irish mac, but the rest of the name is of uncertain meaning.
 Old Irish and Celtic languages 185

seen in Irish myth but also on elements of Arthurian legend. Four stories at the
start of the collection, identified in the texts as the Four Branches of the
Mabinogi, are particularly tightly knit and present some of the clearest examples
of Celtic motifs shared with Ireland. Indeed, while the action of the Second
Branch, often styled Branwen, Daughter of Llŷr after a key character, begins at
Harlech, in Wales, events central to the story take place in Ireland. Branwen,
sister of Bran the Blessed, king of the Island of the Mighty (i.e. Britain), is to be
married to Matholwch, king of Ireland, but their trouble-making half-brother
Efnysien mutilates Matholwch’s horses. By way of compensation, Bran offers
Matholwch a magic cauldron with the following words (which I have given in
Modern Welsh spelling to make them more directly comparable to the rest of
this section):

Mi a roddaf it pair
(it is) I who will give to you a cauldron
a chynneddf y pair yw
and a peculiarity of the cauldron is
y gŵr a ladder heddiw it
the man who is killed today to you (i.e. “of yours”)
ei fwrw yn y pair
his throwing in the cauldron
ac erbyn yfory ei fod
and by tomorrow his being
yn gystal ag y bu orau
particle26 as good as he was best
eithyr na bydd lleferydd  ganddo
except that not will be speech with him
I will give you a cauldron, and it is a peculiarity of the cauldron that the man
of yours who is killed today—throw him in the cauldron, and by tomorrow he
will be as good as when he was best, except that he will not have the power
of speech.

While some of the most basic words in this passage are of common Indo-
European origin, what stands out more is the extent to which, even glossed, it
is hard to spot immediate connections between Welsh and more familiar lan-
guages. Instead, what emerges more clearly is the linguistic heritage it shares
with Irish.

26  The particle yn indicates that gystal is a predicate adjective; it is the same particle as the’n
mentioned in note 24 above.
186 How dead languages work

First, what here is Indo-European? Most obvious are a couple of forms of  “to
be”, bu “was” and bydd “he will be”: these start with b for the same reason that
English be does, as cognate descendants of the PIE root *bhū-. The same is true
of some of the other smaller function words: it’s no surprise that the negative
word na starts with an n-, and the conjugated preposition it “to you” can be
broken down into i “to” (which used to have a d- at the start of it and is related
to English to) and -t “you”, which has the same t as seen in second-person forms
in Latin and the Romance languages. Similarly, at the start of the passage, the
pronoun mi “I” shows the m characteristic of the first-person pronoun across
the Indo-European family. Also, the word for “today”, heddiw, is built up from
elements analogous to those in the corresponding Latin word hodiē (whence
Italian oggi, Spanish hoy, and the hui of French aujourd’hui, literally “on the day
of today”). The Latin word is a compound of the elements ho- “this” (whose
form as an independent word is the hōc of Constantine’s In hōc signō vincēs “In
this sign you will conquer”) and diē “day”, with the ablative case giving the sense
“on”. So too in Welsh, the first element, he-, comes from an old demonstrative
“this”, while the second, -ddiw, is a lenited form of the same PIE root for “day”
that became Latin diē. Finally, gŵr “man”, is the singular of the form gwŷr noted
above, and so cognate with Latin vir.
Much of the vocabulary, however, only has obvious connections to other Celtic
languages. The most conspicuous word in the passage, pair “cauldron”, for
instance, is cognate with the Old Irish word of the same meaning, coire, follow-
ing the same p ~ c correspondence seen above. And the word for “speech”,
lleferydd, is related to the common Old Irish verb labraithir “speaks”, but clear
connections outside Celtic are lacking. But what makes the passage come
across as Celtic at a deeper level is its syntax. First, there are the conjugated
prepositions, it and ganddo, which, here in Middle Welsh, have not yet acquired
the additional object pronoun to reinforce the ending, and so still look more
like their Old Irish counterparts. Also Celtic is the structure seen at the very
beginning of the passage: mi a roddaf, literally “I who will-give”, as the Middle
Welsh for “I will give”. This is another example of the fronting already seen in
Irish as a strategy for allowing the subject to come first in a language where the
default word order is verb–subject–object. To move the subject, mi, before the
verb, one has to treat it as a sort of reduced clause in its own right, as if it were
not just “I”, but “it is I”, then insert the relative pronoun a “who” before the verb.
So often does this structure occur in the Mabinogion that one could be tempted
to describe Middle Welsh as a subject–verb–object language, but with an
obligatory particle a preceding the verb. Also noteworthy is the phrase yn gys-
tal ag y bu orau “as good as he was when he was best”. In addition to a couple of
consonant mutations (gystal and orau are mutated from cystal and gorau), the
word cystal “as good as” itself represents a particularly Celtic feature. Speakers
of most Indo-European languages are accustomed to a three-way gradation of
adjectives: good (the positive), better (the comparative), and best (the superlative).
 Old Irish and Celtic languages 187

But Welsh and Irish both have a fourth degree, the equative, which gets across
in one word, here cystal, the sense “as good (as)”.
But what really gives this passage a distinctively Celtic (and indeed a particu-
larly Middle Welsh) feel is the prominent role played by the verbal noun. Take
what I have translated “Throw him in the cauldron, and by tomorrow he will be
as good.” Here, the verbs throw and will be are translating combinations of the
possessive pronoun and the verbal noun (hence my use of the -ing gerund in
the gloss): ei fwrw “his throwing” and ei fod “his being”. While it is easy enough
to take a more literal translation—“his throwing in the cauldron and by tomor-
row his being as good”—and make sense out of it by supplying a conditional
structure (“If one throws him in the cauldron, then by tomorrow he will be as
good”), that explicit logic isn’t actually there in the Welsh.27 As a result, the
language of the original has a fast-paced breathlessness to it, combined with a
strangely impersonal detachment, that can’t be captured properly in translation.
Clearly, there are a lot of reasons that a passage like this can look forbidding
at first sight. But, despite all these complications that Middle Welsh by and large
shares with Old Irish, in other respects it is much the simpler language:
• It doesn’t have the contrast between palatalized and non-palatalized
­consonants, and so the orthography is considerably simpler.
• It has given up case endings.
• It does not have the system of double verbal inflection, with different
­endings depending on whether a particle precedes the verb.
• Along with that, there are no infixed object pronouns.
Most anglophones who set about learning any Welsh don’t find it especially
easy. But if you come to it after Irish, especially Old Irish, then it is actually quite
straightforward by comparison. As Celtic languages, both Welsh and Irish
share numerous features that set them apart from their Indo-European kin,
but, curiously, two of most prominent of these, verb-initial word order and
conjugated prepositions, are also found in the final, non-Indo-European
­language to which we’ll turn, Hebrew.

27  As Derick Thomson notes in his commentary on the Second Branch, “[T]here is no proper
logical sequence between yw and [ei fwrw]” (Branwen Uerch Lyr (Dublin, 1961), p. 27).
 7

Hebrew

So far, this book has remained safely within the confines of the Indo-European
family—a language family that spread so widely that, even before the age of
colonization ushered in the modern world, the range of its speakers extended
from Iceland and Ireland in the west to Iran and India in the east. But while the
comparison of five representative branches of the family has been a reasonable
way of establishing the basic outline of its character, it’s also important to
consider it from the outside rather than only from within. And so we proceed
to Hebrew, which, as a Semitic language, provides the necessary linguistic
contrast—but which also, as one of the three sacred languages of the West,
is culturally familiar enough to offer some points of tangency as well.

THE SOUNDS OF SEMITIC

In previous chapters, we have seen numerous features of grammar and

vocabulary that are shared by the Indo-European languages. To reiterate some
obvious examples, similar words are found for basic items in the lexicon, such
as numerals and kinship terms:

English Latin Greek Sanskrit

two duo dyo dvā
three trēs treis trayas
four quattuor tettares catvāras
father pater patēr pitar-
mother māter mētēr mātar-
brother frāter phrātēr1 bhrātar-

1  The normal Ancient Greek word for brother is adelphos, but phrātēr survives as a word refer-
ring to a member of the same clan as oneself, much as, in English, one might refer to a fraternity
brother.

How Dead Languages Work. Coulter H. George, Oxford University Press (2020).
© Coulter H. George. DOI: 10.1093/oso/9780198852827.001.0001
190 How dead languages work

Sometimes, as in the words for “four”, sound changes have obscured the
underlying similarity between the words, but they can still be traced back to a
single original form, here something like *kwetwores. English four may look
like it doesn’t belong—and the initial f is indeed a problem—but the r at the
end is reassuring, and the corresponding numeral in Gothic, the earliest-
attested Germanic ­language, is fidwor, with a ­-dw- sequence in the middle that
shows that the Proto-Germanic form must have looked a lot more like Latin
and Sanskrit than does the more phonologically streamlined word attested
in English.
The same words in Hebrew, however, look completely different: 

šənayim “two”
šəlōšāh “three”
’arbāāh “four”
’āb “father”
’ēm “mother”
’āḥ “brother”

True, one can find a shared consonant or two—perhaps the final m of ’ēm
matches the initial m of mother, and, if so, the final b of ’āb could conceivably be
related to the initial p of pater and friends—but the overall impression is one of
dissimilarity. Indeed, this is true even at the level of the letters and diacritical
marks used to represent Hebrew in transcription: there are so-called hačeks
over the three s’s, indicating that they are pronounced like the English sh, as
well as both forwards and backwards apostrophes which, in Semitic linguistics,
indicate not omissions but consonants (the details of which will be discussed
below).2 Now, to some extent, such differences could be dismissed as mere
typographical convention: the haček, for instance, owes its name to its use in
the Slavic languages, which, though not discussed in this book, are perfectly
Indo-European. But the presence of š, as well as that of ‘ and ’, in these basic
Hebrew words is in fact indicative of the very different phonological inventory
of the Semitic languages.
For the most part, the vowels are self-explanatory: the basic a e i o u vowels
have approximately the values they have in Spanish or Italian, and they come
in short and long varieties, with the latter marked by either a macron (ā) or a
circumflex (ô) depending on how it’s represented in Hebrew spelling. The only
vowel that may look out of place is the schwa, ə, which one also finds in phon­
et­ic transcriptions of English, where it stands for the same sound as it does

2  The concave-left and concave-right ’ and ‘ match their IPA equivalents ʔ and ʕ respectively.
As a mnemonic, Hellenists may treat the former as roughly equivalent to the smooth breathing,
since it simply indicates a glottal stop, while the latter resembles the rough breathing in that it
represents a more h-like sound.
 Hebrew191

here: the weak, unstressed, neutral vowel found, for instance, at the start of the
word about. Ultimately, schwa is itself a Hebrew word (although it comes into
English by way of German, as the sch- spelling indicates), and it refers to the
pair of dots placed below a consonant in the Hebrew alphabet to indicate a
following schwa vowel, so this is certainly its home turf here.
No, it’s really in the consonants that the Semitic languages stand out as
­different from the Indo-European family. In particular, they are much richer
both in fricative consonants—those, like f and s, in which air continues to pass
through the mouth since there isn’t a complete obstruction of the airflow—and
in consonants that are produced farther back in the throat than in most
European languages. To begin with the fricatives: while Proto-Indo-European
(PIE) only had one distinct s-sound, Hebrew has no fewer than four—five if
one counts z, too. One of these, ṣ, we’ll pass over for now, and we’ve already seen
both s (= English s) and the š (= English sh), but there’s also an ś, whose original
value was probably similar to that of the Welsh ll discussed in Chapter 6, a sort
of cross between a voiceless th and an l.3 It’s not the most common of sounds,
and, already by the time of the Rabbinic Hebrew of the first few centuries ad, it
had become standard to pronounce it the same as a regular s. The potential for
confusion caused by this wealth of fricatives has also led directly to the English
word shibboleth. It’s borrowed from the Hebrew word šibbōlet, which, on the
one hand, is simply a word for “flood”, as found in Psalm 69:2. More famously,
though, it occurs in Judges 12:6, when Jephthah and his men, who are fighting
the Ephraimites, want to test fugitives to see whether they are the enemy:
because the Ephraimites did not have the sh sound in their dialect, they pro-
nounced š as a plain s, were revealed as the enemy, and then slain for their
phonological failure. Since then, shibboleth has become the general term for
any word used in this way to distinguish speakers who don’t belong to one’s
own community.
But the sibilant s-sounds are not the only reason Hebrew is rich in fricatives.
Here one needs to pay attention to the final consonant of šibbōlet, which I’ve
transcribed simply with a t, but which has become a th in the English loanword.
As it happens, six of the core stops of Hebrew, labial p and b, dental t and d, and
velar k and g, are all subject to a sort of weakening: in certain environments,
especially when preceded by a vowel, they become the corresponding fricatives,
e.g. f and v in the case of the two labial stops, or, for the t, the voiceless th of thin.
For the most part, it’s fairly predictable (at least if you know Hebrew well)
whether what’s written as a stop is pronounced as a stop or a fricative, so I don’t

3  P. Joüon, A Grammar of Biblical Hebrew, vol. 1, trans. and rev. T. Muraoka (Rome, 1993),
p. 29 n. 4, following especially R. C. Steiner, The Case for Fricative-Laterals in Proto-Semitic (New
Haven, 1977). One piece of evidence adduced for a lateral (i.e. l-like) value for ś is that Hebrew
bōśem ‘balsam’ was borrowed into Greek as balsamon, whence English balsam (Steiner, ibid.,
pp. 123–9, L. Kogan, “Proto-Semitic phonetics and phonology”, in The Semitic Languages: An
International Handbook, edited by S. Weininger (Berlin, 2011), p. 78).
192 How dead languages work

always distinguish between the two values in transliteration. But there will be
times later in the chapter when this phenomenon, called begadkefat after the six
consonants in question, will explain what might otherwise seem like in­ac­cur­
acies in the transliteration, and, where necessary, the fricative variants will be
indicated with macrons above or below the letter (ḇ ḡ ḏ ḵ p̄ ṯ).4
In addition to its especially rich inventory of fricatives, the other respect in
which Hebrew consonants differ from those of most Indo-European languages
is that many more of them are pronounced farther back in the throat. Thus,
with a p, airflow is blocked at the lips; with a t, just behind the teeth, at the
alveolar ridge; with a k, with the back of the tongue against the soft palate.
But the Semitic q is a uvular stop, articulated, as the name would suggest, by
bringing the base of the tongue against the uvula. Produced still farther back
are the pharyngeals ḥ and ‘ (often called ayin, after the letter that represents it
in Hebrew), which involve the constriction of airflow through the pharynx, the
part of the throat just above its separation into the windpipe and esophagus.
Both of these sounds are fricatives, with ḥ unvoiced and ayin voiced, so they
bear the same relationship to each other as s and z. Describing on the page what
they sound like or how to produce them in practice is rather more difficult: ḥ is a
bit like a regular English h, but the tenseness in the throat when it’s pronounced
makes it a little Darth-Vaderesque; for ayin, one adds voicing on top of that,
yielding a sound I’ve heard compared to that of retching. Still further back in
the throat are two glottal sounds: h, the same voiceless fricative that we have in
English, and ’, the second of the two sounds designated by apostrophes. The latter
sound, also called aleph after the letter that represents it in Semitic alphabets, is a
glottal stop: the catch in the throat in the Cockney pronunciation of bottle as bo’ul
or the pause before the final syllable in the native pronunciation of the state
Hawai‘i. (Confusingly, in Hawaiian orthography it is rendered with the apostro-
phe facing the opposite direction.) While those are the textbook examples, in fact
one can remain within standard US English and still find the sound: it’s how
most speakers pronounce the t in written or before an m in a word like litmus.
So we now have the following sounds in play, from front to back: 

q ‫ק‬ voiceless uvular stop

ḥ ‫ח‬ voiceless pharyngeal fricative
‘ ‫ע‬ voiced pharyngeal fricative (ayin)
h ‫ה‬ voiceless glottal fricative
’ ‫א‬ voiceless glottal stop (aleph)

4  In the pronunciation of Modern Hebrew, which is often applied to the Biblical language
anachronistically (if understandably), only b, p, and k are subject to this weakening, with t, d, and
g all maintaining their value as stops even in phonetic environments in which the Biblical lan-
guage would have changed them to fricatives.
 Hebrew193

At this point, one might well wonder what the connection is between aleph of
Semitic, which is a consonant, and the alpha of Greek, which is the vowel a.
First, note that it would be more accurate to transcribe the Hebrew letter names
as ’ālep̄ and ‘ayin:5 the usual simplification of dropping the apostrophes doesn’t
reflect the reality that, in biblical times, the former started with a glottal stop,
the latter with the voiced pharyngeal fricative. Now in a language that doesn’t
have such sounds, the former is likely not to be recognized as a distinct con­son­
ant­al sound at the start of the word. Thus, when the alphabet was first borrowed
into Greek from its Semitic source, and Greeks, learning it for the first time,
were matching up the values of the letters with the first sounds of their names,
a Semitic speaker who pronounced the letter-name ’ālep̄ would have heard the
first sound of the word as the glottal stop, but a Greek, not distinguishing that
as a separate sound, would instead have heard it as starting with the vowel a (in
particular, as a long variant of the vowel, as indicated by the macron in the
transcription).
With ‘ayin, the situation is a little more complicated. Its position in the
Semitic alphabet is right before the letter that gave rise to p, and its shape in
Phoenician, the variety of Semitic alphabet that was the most direct source of
the Greek alphabet, is a circle—appropriately enough, considering that ‘ayin
means “eye” (this word will return later in the chapter). So it is clearly the source
of the Greek omicron, which in turn became the Roman O. Once again, what
represented a consonant in Semitic was redeployed as a vowel sign in Greek.
That this should have happened makes sense since Greek doesn’t have a voiced
pharyngeal fricative. But why was it taken over to represent o rather than a?6
Partly because there was already a sign for a and it would have been inefficient
to have two different letters to represent one vowel. (Writing systems do
­occasionally tolerate such inefficiencies: Latin borrowed both C and K from
Etruscan to represent a hard k sound, and ended up using the latter in a small
number of words even though C would have been just as serviceable.) More
importantly, the a of ‘ayin was probably pronounced differently from the ā of
’āleph—in particular, it may have been more like an o. One might then reason-
ably ask: Why transcribe it as ‘ayin rather than ‘oyin? The answer lies in the
concept of the phoneme.
The organs of speech can produce a nearly infinite variety of different sounds,
and speakers of a language need some way to organize them into a meaningful
system. Now, to a monolingual speaker of English, it might seem obvious that,

5  As noted above, the macron over the p in ’al̄ ep̄ indicates that it has its fricative value, namely f.
6  I follow here the suggestion of I. J. Gelb (A Study of Writing, 2nd ed. (Chicago, 1963), p. 292
n. 5) more optimistically than B. B. Powell does, who ascribes the redeployment of ‘ayin as o
simply to “free invention” (Homer and the Origin of the Greek Alphabet (Cambridge, 1991), p. 43).
R. D. Woodard is more or less silent on the matter, but sees “arbitrariness” in play (Greek Writing
from Knossos to Homer (New York, 1997), pp. 135–6, 148).
194 How dead languages work

say, a t is a t, and not going to be randomly confused with an s or a y. But in fact

there’s more than one kind of t. In a word like stop, it’s, well, just that: a voiceless
dental stop, with the airflow through the mouth temporarily blocked with the
tongue parked just behind the teeth. In written and litmus, however, as we’ve
already seen, it’s instead pronounced as a glottal stop—but we still perceive it as
a t, because, in the way US English speakers divide up the continuum of sounds,
the glottal stop and the dental stop both get placed in the same pigeonhole.
What’s more, these variations in how a t is pronounced are predictable on the
basis of the surrounding sounds: that is to say, in other words of the same phonetic
shape, where t comes before an m or the vocalic n of written, it will also be
realized as a glottal stop: boatman and atmosphere, and mutton and bitten.7
And this is where some technical terminology comes in handy: specific types
of t, like the dental stop or the glottal stop, are said to be allophones of the
overarching phoneme t. One other point is important: different languages div-
ide up this continuum of sounds in different ways. What is a mere allophone in
one language may be a full phoneme in another. Thus, in English, a glottal stop
is just an allophone of t; it will never occur in an environment where it might be
contrasted with one of the other possible pronunciations of the t phoneme,
such as the dental stop. We might usually say wri’en with a glottal stop, but if we
heard written enunciated carefully with a dental stop instead, we’d still process
it the same way. Not so in Hebrew, where the glottal stop is a phoneme in
its own right, and it can contrast with a t: šātal means “he planted”, but šā’al
“he asked”.
To return now to the first vowel of ‘ayin, and why it may have been pro-
nounced more like an o: that initial consonant, the voiced pharyngeal frica-
tive, as is often the case with pharyngeal sounds, had an effect on the following
vowel sound, here potentially making it sound a little more like an o. This
makes anatomical sense. After all, if your throat is still constricted from
articulating the pharyngeal, then the back of the tongue will still be fairly close
to the soft palate, and the airflow won’t be as open as is necessary for a true a.
As a result, the a of ‘ayin still counts as the phoneme a—structurally, it occurs
in the sort of place one would expect an a—but it is realized as an allophone
that’s a little closer to an o. That the pronunciation of a vowel can be affected
by that of a neighboring consonant is also something seen in the final two
Hebrew consonants that lack counterparts in the languages considered so far:
the so-called emphatic s and t, generally transcribed as ṣ and ṭ. Exactly how
these would have been articulated, as distinct from regular s and t, in various
ancient Semitic dialects is not entirely certain, but in modern Arabic, which
preserves them well, they are most often characterized by pharyngealization:
that is, at the same time the s or t is pronounced, the throat is constricted in the

7  For more on the phonology of the glottal stop in English, see D. Odden, Introducing Phonology
(Cambridge, 2005), pp. 48–9.
 Hebrew195

same way as for ‘ayin, which in turn causes the quality of a neighboring vowel
to “darken”.8
In discussing the phonetics of all these sounds, Hebrew’s close cousin Arabic
is often a more useful point of reference than Modern Hebrew itself, as nearly
all dialects of Modern Hebrew have greatly reduced the number of pharyngeals
and related sounds—in large part because the language is more a consciously
revived version of the Biblical language than a simple, organic survival from
antiquity. In any case, nowadays the q is simply pronounced like a k, the ḥ like
a velar fricative (i.e. like the ch in Bach), and the ‘ is pronounced as a glottal
stop—that is, confusingly, the same as ’. The emphatic ṭ is pronounced no dif-
ferently from a regular t, and what used to be the emphatic ṣ is now just articu-
lated as the ts sequence, often spelled tz, as in mitzvah, “commandment”, which,
as a Biblical Hebrew word, would be transliterated miṣwāh or miṣwâ. (It makes
some sense that an emphatic s should end up as a ts: if “emphasis” means some
sort of additional constriction of airflow through the mouth, then that could
easily lead to the dental stop t developing in front of the corresponding dental
fricative s.) Be that as it may, that all these additional characters and diacritical
marks are necessary to distinguish between these sounds is one reason—along
with, in Arabic, a vast amount of dialect variation—that Semitic names can be
spelled in such different ways when they are transliterated: witness the various
alternative spelling of, say, Hanukkah or Muammar al-Qaddafi. For the most
part, however, many of these subtleties in Semitic phonology simply vanish in
our alphabet: the q at the end of Iraq serves as a clue that something other than
the usual k sound is in play, but nothing in the Romanization would indicate
that there’s also an ‘ayin at the start of it.

HOW SEMITIC WORDS CHANGE SHAPE

But it’s not just all these different sounds that make Semitic languages unlike
their Indo-European counterparts: the way they change shape to indicate
grammatical categories like plural or the past tense also differs considerably.
Perhaps the most distinctive feature of Semitic morphology is that most of
the words in these languages are derived from basic roots of three consonants
each. These consonants then act as a sort of skeleton that remains stable, while
changes in the vowels between them, and the addition of prefixes and suffixes,
either mark the grammatical function of a given word, or change it into a

8  For the phonetic realization of the emphatic consonants in modern Semitic languages
(including both glottalized and pharyngealized varieties), as well as the likelihood that they were
in fact glottalized in Proto-Semitic, see L. Kogan, “Proto-Semitic phonetics and phonology”, in
The Semitic Languages: An International Handbook, edited by S.  Weininger (Berlin, 2011),
pp. 59–61.
196 How dead languages work

­ ifferent but related word. For example, there is a root š-l-m, whose basic verbal
d
sense is “to be complete, whole”. If one fills the first slot between consonants
with ā and the second with ō, then one has šālōm, the Hebrew word for “peace”
(by way of the meaning “wholeness, soundness, well-being”). In the same way,
one can add the same vowels to the root k-b-d “to be weighty, honored” to yield
the noun kābōd “honor, glory”. If, however, one wants to build verbs to these
roots, one uses different vowels: thus, šālēm means “he is complete, whole”,
and kābēd “he is honored”. Further possibilities of the system can be seen in
the  formation of personal names: in Absalom (transcribed more precisely,
’ab-šālōm “father(-is-)peace”), the basic noun šālōm is the second element in a
compound name, and in Solomon (šəlōm-ōh “his peace”), a possessive suffix
“his” has been added to the noun, causing the first vowel to be reduced to a
schwa. (For those paying close attention, something similar has happened in
’ab-šālōm, where the “father” word has only a short a, not the long ā it has when
used as an independent noun, as when it was cited above.) Similar patterns are
at work in Arabic: the š sound became a plain s, but apart from that minor
change the š-l-m root is evident in salām, the word for “peace” that corresponds
to Hebrew šālōm, which has been borrowed into English as salaam, the bow of
greeting in Arabic countries—and is also seen in Dar es Salaam “House of
Peace”, the name of Tanzania’s largest city. The same root, extended with a prefix
and suffix, but with the vowel dropped between the first two consonants, gives
rise to the Arabic verb “to surrender, submit”, ’aslama, from which are derived
both ’islām, the noun “surrender, submission”, and the active participle muslim
“one who surrenders, submits”, retaining the clear triconsonantal pattern of the
root throughout.
It’s worth comparing this facet of Semitic with Indo-European to highlight
both what’s distinctively Semitic, and what isn’t—because the idea that a lan-
guage family builds up its vocabulary out of a set of roots, and that the con­son­
ants of those roots are particularly stable, is certainly not something on which
Semitic has a monopoly. Even just within English, the relationship between
give, gave, given, and forgiven could be analyzed in similar terms: a basic root
structure of g-v, with various changes to the internal vowel, the addition of a
suffix -en, or a prefix for- changing the grammatical form or meaning of the
word. And it makes phonological sense that the consonants should be the
backbone of such roots. Because pronouncing a consonant requires the tongue
or lips to commit to an articulatory position with clearer landmarks than a
vowel—the soft palate with a k, or the lips with a p, as compared to the fuzzier
difference between an a and an o—they are generally more resistant to change
than vowels. As a reality check on this principle, consider the different dialects
of English: while British and American speakers don’t pronounce all con­son­
ants in exactly the same way (most noticeably, UK Received Pronunciation
loses syllable-final r’s), it’s in the vowels that one hears the greatest differences:
the a of grant that rhymes with (US) want rather than rant, or the different
 Hebrew197

starting point for the diphthong in know. So if it’s so linguistically natural for
a language to have words built to roots where the consonants are mostly stable,
why do introductions to Semitic languages make such a big deal of the tricon­
son­ant­al root structure? Partly it’s because such a great majority of the roots
have three consonants in particular. To be sure, it’s not true of all roots—some
common nouns, like the family relationship terms cited above, ’āb “father”
and ’āḥ “brother”, have only two—but for the most part the canonical Semitic
root really does have a shape like š-l-m, whereas Indo-European roots, while
they may well have three consonants, only have one slot in the middle for a
vowel. That is, an Indo-European root of the simplest form would have a
shape like p-d “foot”, with the vowel between the two consonants subject to
variation between e and o (thus ped-al and pod-iatrist) or lost altogether. And
while a different root could have additional consonants, such as the root sp-nd,
here too there is only one slot for introducing a vowel, between the p and the n
(yielding e.g. re-spond, via Latin), but no forms in which the vowel was
­inserted between the s and p or between the n and d. This contrasts with the
considerable flexibility the Semitic system acquires through being able to
place vowels between either the first and second or the second and third
consonants.
But an even more important reason to single out the root structure of the
Semitic languages is that it remains such a fundamental organizational prin-
ciple for the modern languages: one can quite reasonably draw up a Hebrew or
Arabic dictionary with all the derivatives of each root grouped together rather
than on strict alphabetic principles. That is, there’s a convenience to having both
Islam and Muslim under s-l-m rather than one under i and the other under m.
And it would also make good sense to organize a PIE dictionary along the same
lines. But English or French would be much less amen­able to such treatment.
While one can detect traces of the old Indo-European root structure in the
modern languages of Europe, they remain just that: traces. The variations of g-v
mentioned above don’t form any sort of productive pattern in English, such
that you could take another pair of consonants, fill in the same vowels, and add
the same suffixes, and get words of analogous meaning. That is, one can’t start
from the pattern give ~ gave ~ given ~ forgive, and assume that one also finds
live ~ ×lave ~ ×liven ~ ×forlive. In Semitic languages, by contrast, such patterns
are regular, and there are dozens of Hebrew verbs, for instance, that can be
formed by putting ā and a respectively into the two main slots in the root.
Finally, the writing systems of most Semitic languages, including Arabic and
Hebrew, inevitably end up highlighting the triconsonantal root because, in the
basic form of the script, only the consonants are written down, and vowels have
to be inferred from context. If necessary, they can be specified (in Hebrew by
diacritical marks called vowel points), but in most texts these are omitted—
apart from the Bible, since in the case of a sacred text, there’s a premium placed
on accuracy of reading.
198 How dead languages work

Much of the rest of Semitic morphology can be discussed along the same
lines: the basic categories and principles are more or less the same as what
one finds in Indo-European, but the specific sounds and affixes are different.
Take noun plurals. Like Latin or Greek, Hebrew marks plural nouns by adding
­endings, and, just as Latin and Greek have some endings that are more charac-
teristic of masculine nouns, others of feminine nouns, so too Hebrew has one
ending, -îm, used chiefly of the former, and another, -ôt, used of the latter. The
masculine plural is perhaps most obviously present in the English loanwords
cherubim and seraphim, which, in a closer transcription of the Hebrew, would
be kərûb-îm and śərāp-îm respectively. As for feminine -ôt, it makes it into
words like Sukkot, the Jewish festival of Booths (sukk-āh “booth”, sukk-ôt
“booths”), the phrase in the Latin Mass Dominus Deus Sabaoth “Lord God
of  Hosts” (ṣābā’ “host”, plural ṣəbā’ôt), and it’s even in the word behemoth.
While bəhēmāh was an ordinary word for animal (in Genesis 1:24, it’s opposed
to  wild beasts and translated “cattle” in the New Revised Standard Version
(NRSV)), its plural bəhēmôt was used as a sort of intensive formation, perhaps
originally in reference to the hippopotamus but then extended to any monster.
In any event, the categories of masculine and feminine, and singular and plural,
are essentially the same as in Indo-European languages—it’s just the particular
suffixes, like ‑îm and -ôt, that reveal Semitic as a different language family
altogether.
The same holds for verbs, too: like Indo-European verbs, they inflect for
­person and number, and tense, voice, and mood, so the categories are more or
less the same. But the details of the forms are different, as are some of the
boundaries between the categories. As an example of these differences, one
could consider the conjugation of the Hebrew verb tenses called the perfect
(most often a simple past tense) and the imperfect (despite its name, more often
used as a future). The perfect inflects with endings; thus, just as Latin has
scrīps-ī, scrīps-istī, scrīps-it “I wrote, you (sg.) wrote, he/she/it wrote”, so too
Hebrew has kātab-tî, kātab-tā, kātab in very nearly the same sense. The chief
difference: whereas the inherited verbal endings of Indo-European languages
don’t show gender, the second- and third-person forms just cited for Hebrew
are specifically masculine. Thus, “you (fem.) wrote” and “she wrote” would be
kātab-t and kātəb-āh respectively.9 As for the imperfect tense of Hebrew,
­whereas the Latin forms of equivalent meaning, namely the future, would again
inflect with endings, but added to a different stem (scrīb-am, scrīb-ēs, scrīb-et
“I will write, you will write, he/she/it will write”), Hebrew chiefly uses prefixes,
also added to a different stem: ’e-ktōb, ti-ktōb, yi-ktōb. Again, the last two forms
are only used for the masculine second- and third-person singular. For “you

9  Indo-European verbs do sometimes inflect for gender, but only in forms that either had been
or still are participles, i.e. verbal adjectives. The last form, kātəb-āh, shows a predictable vowel
weakening in the second syllable of the stem.
 Hebrew199

(fem.) will write”, one adds a suffix in addition to the prefix, with vowel
­weakening in the root: ti-ktəb-î. And, confusingly, “she will write” is hom­oph­
onous with “you (masc.) will write”: ti-ktōb. In passing, we can also see the
­triconsonantal stem in operation: in the two slots between the three con­son­
ants, the perfect is characterized by the vowels ā and a, the imperfect by the
absence of a vowel and ō, with both of these patterns subject to reduction
depending on the following ending—but in all these forms, the basic three
­consonants remain intact as the various prefixes and suffixes are added and the
internal vowels are altered.
Now all these formal differences in verb inflection are somewhat superficial
in the sense that they rarely cause the translator any problems: one simply uses
the equivalent verb form in the language into which one is translating, and all
is well. But to turn towards some of the features of Hebrew that pose more dif-
ficulties, one could point to the functional difference between the two tenses
whose morphology I’ve just described. While the Hebrew perfect most often
serves as a simple past tense, with certain verbs that express mental or emotion-
al states (e.g. to know, to love), it can be the equivalent of the present tense; the
imperfect, for its part, not only fills the role of a future tense but also has several
modal functions—that is, yi-ktōb isn’t just “he will write”, but also “he may/
would/could write”—and can also act as a general habitual tense (“he writes
(regularly)” or even “he used to write”). This might seem confusing enough,
but there’s an additional complication that has a more pervasive influence on
the language of the Old Testament: when the word for “and”, wə, is added to the
start of the sentence, it essentially reverses the polarity of the two tenses. What
had been a perfect is translated as if it were an imperfect, what had been an
imperfect is translated as if it were a perfect. Because wə is represented in
Hebrew with just the letter ‫ ו‬wāw, this construction has traditionally been
called the waw-conversive.

LET THERE BE LIGHT

Before seeing examples of how this works, we must first note a feature of Semitic
syntax that makes the waw-conversive a slightly more comprehensible phe-
nomenon than it would otherwise be: just as in the Celtic languages, verbs
typically come before, rather than after, the subject. As a result, a conjunction
like “and” comes immediately before the verb, interacting with it directly, so it
makes more sense that it could have such a strong effect on the verb. The appar-
ent conversion caused by adding wə can be seen in the following pair, in which
I have glossed kātab and yiktōb simply as “he write” so as not to prejudice the
interpretation of the verb in favor of one English tense or another. First, the
regular perfect and the converted perfect:
200 How dead languages work

kātab han-nābî’ sēper  

he write the-prophet book → “The prophet wrote a book”
wə-kātab han-nābî’ sēper  
and-he write the-prophet book → “And the prophet will write a book”
Second, the regular imperfect, and the converted imperfect:

yiktōb han-nābî’ sēper  

he write the-prophet book → “The prophet will write a book”
way-yiktōb han-nābî’ sēper  
and-he write the-prophet book → “And the prophet wrote a book”

Now there is a difference between these two pairs: in the former, the conjunc-
tion really is just a simply wə-; but in the second, I’ve transliterated it as way-.
Thus two things are in play that should make us suspect that something other
than just the ordinary conjunction is at work, at least in the second pair:

• First, the first consonant of the verb is doubled after the conjunction. That is
to say, the y at the end of way- is not a sort of independent y but rather
­something that emerges from the join of the conjunction and the root.
Change the first consonant of the verb, and one changes the consonant at the
end of the conjunction: “she will write” is tiktōb, but “and she wrote” is
wattiktōb; “we will write” is niktōb, but “and we wrote” is wanniktōb. This
doubling is not as readily apparent in a Hebrew-alphabet text because
the doubling in question is indicated simply by a dot in the letter: thus, bear-
ing in mind that Hebrew is written right-to-left, yiktōb is ‫יִ כְ ּתֹב‬, but wayyiktōb
is ‫וַ ּיִ כְ ּתֹב‬, with the far right letter, together with the line underneath it, repre-
senting the sequence wa, and an extra dot in the letter that’s second from the
right (‫י‬, i.e. y). Consequently, in a text that lacked the diacritical marks, the
only difference between the two forms would be the additional ‫ ו‬at the start.
• Second, the vowel following the w is not the weak schwa but a full a.

Taken together, these differences suggest that what’s happening is something

more complicated than just the addition of the conjunction to the ordinary
imperfect; in any case, the ultimate origins of the construction lie in a
­complicated reorganization of the Semitic tense system.10
At this point, you might well ask: How would this cause a translator much
difficulty? Doesn’t one simply translate with the converted tense, and leave it at

10  For a short account, see P. K. McCarter, Jr., “Hebrew”, in The Cambridge Encyclopedia of the
World’s Ancient Languages, edited by R. D. Woodard (Cambridge, 2004), pp. 347–8; there are full-
er details in P. Joüon, A Grammar of Biblical Hebrew, vol. 2, trans. and rev. T. Muraoka (Rome,
1993), pp. 379–81, 386–9.
 Hebrew201

that? While this is largely true, some distinctive effects are then missed in the
resulting translation. One excellent example of this occurs in the very first
chapter of Genesis, in verse 3, when God creates light: 

way-yō’mer ’ĕlōhîm yəhî ’ôr wa-yhî ’ôr

and-said God let there be light and-there was light

The first word here is a fairly standard waw conversive, with the imperfect
of  the root ’āmar used as a simple past tense (the prefix yō- is functionally
equivalent to the yi- of yiktōb). Next comes God, the word for whom happens
to contain the plural marker -îm, probably (as in the Hebrew source of behe-
moth) a sort of intensive formation, here used as an honorific. Next comes the
verb form yəhî, which is the so-called jussive of the verb hāyāh “to be”. The
jussive is essentially the third-person equivalent of an imperative: whereas an
imperative has the addressee as its understood subject (“Walk!”), the jussive is
an order issued to a third party (“Let him walk!”, or in this case “Let there be
light!”). Most of the time, Hebrew verbs look the same in the imperfect as they
do in the  jussive, but this particular root—partly because the  verb “to be”
is usually irregular, but more directly because its final root consonant is the
­weakly articulated -h—distinguishes between the two:
• The third-singular imperfect yihyeh “he will be” has the usual prefix yi-
and retains all three consonants of the root (h-y-h). (We don’t need this
form for the sentence above, but we’ll return to it shortly.)
• The jussive yəhî “let him/it be” has a somewhat reduced prefix (rather
than yi-, it is simply yə-), and a greatly reduced root, -hî. To be fair, in the
Hebrew spelling, the y of the root is still visible as what becomes the
­circumflex over the i in the transliteration, but the final h is lost altogether.
Now the distinction between the imperfect and jussive is important here
because it is the jussive, not the imperfect, which, strictly speaking, is the start-
ing point for the waw conversive. Putting this all together, to say “And there
was” in Hebrew, one says not literally “And there was”, but rather “And (wa-) let
there be (yəhî)”, with the wa- having two effects: first, it mechanically causes the
schwa of yəhî to disappear, yielding the form wayhî; second, it changes the
sense of the verb from “let there be” to “there was”.
This creates a striking literary effect in the original that’s unavoidably absent
from the translation. In English, there’s a slight distance between God’s uttering
the command that light be created and the actual creation of the light. True,
there’s not a lot of distance, and it seems pretty clear that the creation follows
more or less immediately after the command. But in Hebrew, where the verb
form with which God orders that light be created is transformed by the waw
conversive into the exactly the same verb form with which the light comes into
being, the acts of speech and creation come to overlap entirely. This sort of
202 How dead languages work

wordplay, which is founded on the fact that a single verb form can have two
different grammatical meanings, has its counterpart in the Iliad, as noted
in Chapter 2: the form aeide, the imperative “Sing!” addressed to the Muse in
the poem’s first line, doesn’t recur until the ninth book of the poem—where it
functions as a third-person singular imperfect indicative, with Achilles as the
subject (“He was singing”), as he broods in his tent and sings of the famous
deeds of men.
One final word in this passage worth looking at is ’ôr, the word for “light”.
While there aren’t a lot of direct English derivatives of the words we come
across in this chapter—we’ve left Indo-European behind, after all—there’s
probably a connection between this word and ancient objects referred to in
English by their Hebrew name: the Urim and Thummim. To judge from pas-
sages like Exodus 28:30 and 1 Samuel 14:41, these were, it seems, some sort of
item placed in the breastpiece of the high priest that were used in the determin-
ation of judgments by the casting of lots—essentially, the Magic 8 Ball of the
ancient Israelites. Whatever the exact physical make-up of these objects (and,
for that matter, whatever the actual etymology of the phrase, which remains
disputed), the first word, ’ûrîm, was taken to be yet another intensive plural, in
this case to a word ’ûr, understood to be from the same triconsonantal root,
’‑w-r, as ’ôr, the word for “light”.11 And the second word, tummîm, was regarded
as a derivative of the root t-m-m “to be complete, have integrity”. Accordingly,
in the Latin Vulgate of Exodus 28:30, they’re translated as doctrīna “teaching”
(viewed as a type of illumination) and vēritās “truth”, following the Greek
Septuagint fairly closely, which offers dēlōsis “clarification” and alētheia “truth”.
Somewhat less expected, the phrase also makes an appearance as the motto
of Yale University, which both depicts the Hebrew letters ‫’( אורים ותמים‬ûrîm
wə-tummîm) across the pages of an open book and the Latin translation Lūx
et vēritās (“Light and truth”) below—a gloss on the Hebrew that, while an
excellent guiding principle for a university, doesn’t do justice to the semantic
complexities of the original phrase.
Next, there’s more to say about the waw-conversive form wa-yhî “and there
was”. Now the effect found in Genesis 1:3 of using the converted yhî (“there
was”) in such close proximity to unconverted yəhî (“let there be”) is inevitably
lost in English. But another, far more widespread usage of wa-yhî is retained
in some translations, leading to a phrase that’s become a hallmark of Biblical
language in English thanks to its ubiquity in the long-authoritative King James
Version (KJV): “and it came to pass”. To understand how this works, consider
two alternative translations of Genesis 4:8, the KJV and the NRSV: 

11  The central consonant of the root, w, is absorbed by the vowel in both forms, and, written in
the Hebrew alphabet, the only difference between ’ôr and ’ûr is the placement of the vowel point
on the central waw: ‫ אֹור‬and ‫ אּור‬respectively.
 Hebrew203

KJV And it came to pass, when they were in the field, that Cain rose up
against Abel his brother, and slew him.
NRSV And when they were in the field, Cain rose up against his brother
Abel, and killed him.

Apart from replacing “slew” with “killed”, the most significant change between
the two versions is the simplification of the syntax: whereas the KJV has as
a  main clause “And it came to pass”, followed by a temporal when-clause,
­followed by a subordinate that-clause, which presents the primary content of
the verse, the NRSV omits this framing clause, which means it can also omit the
“that”, thus elevating the “Cain rose up . . .” material to become the main clause
of the sentence. It’s exactly the sort of change one would recommend in
­marking a student paper as a way of reducing linguistic clutter. So why does the
KJV include the additional layer of syntax? Because something similar is there
in Hebrew, which, word for word, runs as follows: 

wa-yhî bi-hyôt-ām baś-śādeh way-yāqom

and-it was in-being-their in the-field and-he rose up
qayin  ’el-hebel  ’āḥ-îw    way-yaharg-ēhû
Cain    to-Abel  brother-his  and-he killed-him

With English word order and prepositional usage, this would be something
like “And it was during their being in the field, and Cain rose up against Abel
his brother, and he killed him,” which strains English idiom too severely to be
left like that in a formal translation. Accordingly, the KJV captures the rhythm
of the Hebrew syntax inasmuch as it retains the initial main verb, but it devi-
ates from it by making “rose up . . . and slew” subordinate to “it came to pass”,
when the original text simply has three main verbs all connected by the waw-
conversive (“and it was . . . and he rose up . . . and he killed”, with the third-
person masculine singular y(a/ā)- prefix on all three verbs). Now, in English,
this syntactic structure comes across as artificial, but, in the Hebrew prose
style of the Old Testament, it’s completely unremarkable for a phrase indicat-
ing the time of the main event to be introduced by wayhî in this way. Two more
examples: 

Genesis 7:10
KJV And it came to pass after seven days, that the waters of the flood were
upon the earth.
NRSV And after seven days the waters of the flood came on the earth.
literal And it was after seven days, and the waters of the flood were upon the
earth.
204 How dead languages work

2 Samuel 11:14
KJV And it came to pass in the morning, that David wrote a letter to Joab,
and sent it by the hand of Uriah.
NRSV In the morning David wrote a letter to Joab, and sent it by the hand of
Uriah.
literal And it was in the morning, and David wrote (wayyiktōb) a letter to
Joab, and he sent it by the hand of Uriah.

The translator, then, is left with an unenviable choice: Attempt to stay close to
the Hebrew, and translate wayhî fully every time with a phrase like “And it came
to pass”, at the cost of throwing more emphasis than is warranted on an expres-
sion that’s simply the ordinary way of marking when an event took place? Or
else omit it in favor of a more natural English idiom, but then lose an expression
that, for all that it’s the unmarked way of doing this in Biblical Hebrew, still
involves a full verb form that gives these sentences a distinctive cadence? And
it is distinctive. Mark Twain—no linguistic fool—also noted the usefulness of
“And it came to pass” as a marker of Biblical language when he speaks, in
Roughing It, of Joseph Smith’s translationese in the Book of Mormon:
The author labored to give his words and phrases the quaint, old-fashioned sound
and structure of our King James’s translation of the Scriptures; and the result is a
mongrel—half modern glibness, and half ancient simplicity and gravity. The latter
is awkward and constrained; the former natural, but grotesque by the contrast.
Whenever he found his speech growing too modern—which was about every
­sentence or two—he ladled in a few such Scriptural phrases as ‘exceeding sore,’
‘and it came to pass,’ etc., and made things satisfactory again. ‘And it came to pass’
was his pet. If he had left that out, his Bible would have been only a pamphlet.
Unfortunately, the golden plates from which Smith is said to have translated
the Book of Mormon were dutifully returned to the angel Moroni, so we are
unable to consider here whether Twain’s stylistic criticisms should more aptly
have been directed at the original “Reformed Egyptian” text.
We are not quite done, however, with h-y-h, the “to be” root. It is also ­prominent
in Exodus 3:14, when God speaks to Moses: 

KJV And God said unto Moses, I Am That I Am: and he said, Thus shalt
thou say unto the children of Israel, I Am hath sent me unto you.
NRSV God said to Moses, “I am who I am.” He said further, “Thus you shall
say to the Israelites, ‘I am has sent me to you.’ ”

Now the NRSV includes a cautionary note that the phrase “I am who I am”
might also be rendered “I am what I am” or “I will be what I will be”. Small caps
aside—these signal, as with the frequent “Lord”, that the words in question are
 Hebrew205

to be understood as a name of God—why the uncertainty? Here’s the Hebrew

of the first half of the verse: 

way-yō’mer ’ĕlōhîm ’el-mōšeh ’ehyeh ’ăšer ’ehyeh

and-he said God to-Moses I am/will be who/what I am/will be

The first three words are straightforward enough—though it’s worth noting
that whereas the KJV preserves the “and” of the waw-conversive, the NRSV
gets rid of it in its efforts to achieve a more streamlined English style. The real
difficulties arise in the second half. First, ’ehyeh is the first-person singular
imperfect of hāyāh “to be”: ’e- is the prefix for the first-person singular, func-
tionally equivalent to English I, and -hyeh is the regular imperfect stem as
opposed to the shorter -hî seen in the jussive.12 To this extent, the form is
­perfectly easy to understand—but, as mentioned earlier, the Hebrew imperfect
has a wide range of functions: most often a future tense (I will be) but also used
to express modality (I would be, could be), and as a habitual (I am regularly).
In  most circumstances, contextual clues are sufficient to disambiguate, but
when the verb form is used in an isolated way like this, as a name expressing the
nature of God, then the question of how best to translate it moves from the
realm of linguistics to that of theology. The same could be said of the pronoun
’ăšer, which, unlike English who or what, is used indifferently of animate and
inanimate objects: Should the translator choose the former, and make God a
little more human? Or the latter, and characterize Him more as an abstraction?
In a situation like this, one respects the NRSV’s decision to include a footnote:
no single translation can capture the alternative possibilities.

NOUN CHAINS

To move on to a second notable stylistic feature of Hebrew, consider the diver-

gence in the translations of Exodus 3:14 between the KJV’s “the children of Israel”
and the NRSV’s “the Israelites”. In what is now emerging as a regular pattern, the
KJV stays a little closer to the exact wording of the Hebrew, whereas the NRSV
turns it into more natural English. For the Hebrew does in fact say: bənê yiśrā’ēl
“(the) children (of) Israel”. If that’s the case, then why does the NRSV flatten out

12  One other verse where this verb is prominent, to striking acoustic effect, is Ecclesiastes 1:9:
“What has been is what will be, and what has been done is what will be done; there is nothing new
under the sun” (NRSV). The first third, in Hebrew, is mah-šše-hāyāh hû’ šey-yihyeh “what-which-
was [is] that which-will.be”, with a sequence of wispy, whispering sh, h, and y sounds that phonet­
ic­al­ly imitates the transitory vanity of life that is the overarching message of Ecclesiastes. The
sounds at the start of the verse are then neatly echoed in the final word, haš-šāmeš “the sun”. It is
difficult to see how one could effectively capture this sound-play in English.
206 How dead languages work

the reference to children, and simply resort to “Israelites”? As one might expect, it’s
because the Hebrew expression lies somewhere in between the two alternatives,
but this isn’t clear until one learns a little about the so-called construct chain.
There are three clues in the expression bənê yiśrā’ēl and my glossing of it
that suggest that even “(the) children (of) Israel” involves a slight translational
fudge. First, there’s the fact that I’ve had to put parentheses around the word
“the”. Hebrew has a definite article, so its omission here requires some explan­
ation. Second, there’s the next set of parentheses, around “of ”. There’s no prepos­
ition here in the Hebrew, and the word yiśrā’ēl isn’t inflected for, say, a genitive
case, so what justifies the inclusion of it in a translation? Does Hebrew simply
juxtapose two nouns and assume that the reader will supply an “of ” as needed?
While that’s certainly not impossible, the answer is in fact no, but to understand
why, we need the third observation: the word for “children”, bənê, doesn’t end
in either of the plural endings we’ve seen already. In the singular, “son” is bēn
(the first element in the name Benjamin, either “son of the right hand” or “son
of the south”13); in the plural, “sons” is bānîm, with the root vowel undergoing
an irregular change to ā, but with the usual masculine plural ending -îm. Taking
all three of these clues together, we can set out what’s going on.
When Hebrew wants to say anything of the form “the Noun of the Noun” (e.g.
“the son of the king” or “the voice of the prophet”), the following things happen:
1. The first noun loses its phonological independence: its vowels and end-
ings are reduced, thereby signaling that it’s entering into a syntactic rela-
tionship with the noun that follows. Here, the independent form bānîm is
weakened to bənê. This reduced form is called the construct state, and the
resulting unit formed by the nouns is a construct chain—a chain because
more nouns can be added to it, as needed.
2. Because the construct state as such is enough to indicate that a following
noun stands in an “of ”-relationship to the first noun, no preposition or
case-marking for “of ” is necessary.
3. Once nouns are combined in such a chain, they act as a single unit. Thus
it’s treated as either entirely definite (“the Noun of the Noun”) or entirely
indefinite (“a Noun of a Noun”). If a definite article is needed, then it’s
only used before the final noun of the chain. In this particular example,
no definite article is needed because Israel, as a proper noun, automatic-
ally counts as definite, and, since that extends to the whole phrase, “chil-
dren” doesn’t need an article. (If definiteness doesn’t apply over the whole
phrase, as in “a Noun of the Noun”, then other constructions have to be
used.) The fact that the phrase counts as a single unit can also be seen in

13  Directions were given from the perspective of one facing east, an orientation in common
with that which led to the word orientation itself, which presupposes the orient as one’s chief
directional reference point.
 Hebrew207

the treatment of adjectives: they can only follow after the phrase as a
whole, and can modify either noun, with gender and number marking on
the adjective generally clarifying which of the nouns they modify.
To understand the translator’s dilemma, what’s most important is the idea that a
construct chain, like bənê yiśrā’ēl, forms a single unit. One can, to some extent,
compare the difference in English between “David’s son” and the surname
“Davidson”. The element son is clearly present in the latter expression, and yet, in
part because it’s not stressed, it doesn’t register fully. Now it’s not a perfect compari-
son, since “Davidson” has become fossilized as a proper noun, and is not a regular
way of saying “the son of David”. By contrast, bənê yiśrā’ēl has in fact become the
default expression for “Israelites”, occurring hundreds of times in the Hebrew Bible.
Thus, while one can’t deny that some resonance of “the children of Israel” remains,
translate it that way, and it exaggerates the weight placed on “children” in Hebrew.
What’s more, this is far from the only phrase in which the use of construct
chains gives a particular stylistic nuance to the language of the Old Testament.
Especially common are phrases in which the first noun in the chain is a body
part, used metaphorically. Twice already in the second verse of Genesis one
finds such an expression: 

KJV And the earth was without form, and void; and darkness was upon the
face of the deep (‘al-pənê təhôm). And the Spirit of God moved upon
the face of the waters (‘al-pənê ham-māyim).
NRSV The earth was a formless void and darkness covered the face of the
deep, while a wind from God swept over the face of the waters.

Here, in the poetic description of the creation of the world, both the KJV and
NRSV translators were happy to retain “the face of the deep” and “the face of
the waters”; the translators of the Greek Septuagint, however, rendered ‘al-pənê
“upon the face of ” simply with epanō “over”. In any case, the shape of the
Hebrew expression is reminiscent of bənê, with good reason: pənê is the con-
struct form of the noun for “face”, pānîm (which only occurs as a plural), and
“the deep” and “the waters” form the second element in the construct chain.
With these expressions, is the writer of Genesis here personifying the deep
and the waters by giving them a face? No, this is simply the regular way that
Hebrew says “upon” or “over”, and, while we might expect the NRSV translators
to have simplified it accordingly, as they did with “the children of Israel”, it is
retained even in fairly ordinary contexts, such as Genesis 6:1: 

KJV And it came to pass, when men began to multiply on the face of the
earth (‘al-pənê hā-’ădāmāh), and daughters were born unto them . . .
NRSV When people began to multiply on the face of the ground, and
daughters were born to them . . .
208 How dead languages work

In fact, the NRSV here seems to have awkwardly split the difference between
two competing visions for this line: had it kept the “earth” of the KJV, then the
line would have the same sort of cosmic grandeur as seen in Genesis 1:2; but
with the physical expanse at our feet emphasized by “ground”, the retention of
“on the face of ” is somewhat jarring.
At other times, though, both translations omit it in favor of a wording that
more naturally expresses a local meaning in English. For instance, moving fur-
ther on, to the destruction of Sodom and Gomorrah (Genesis 19:28), we find
the following: 

KJV And he looked toward Sodom and Gomorrah, and toward all the land
of the plain . . .
NRSV And he looked down toward Sodom and Gomorrah and toward all
the land of the Plain . . .

Had the translators approached ‘al-pənê in the same way here, we might have
expected, “And he looked down on the face of Sodom and Gomorrah [more
accurately, ‘ămōrāh, with an ayin at the start], and on the whole face of the land
of the plain”. Similarly, in the clause of 2 Samuel 11:11 translated in the NRSV as
“The servants of my lord are camping in the open field”, the Hebrew phrase is
closer to “. . . are camping on the face of the field” (‘al-pənê haś-śādeh ḥōnîm).
In pointing out these inconsistencies, the last thing I want to do is fault the
translators but rather to draw attention to the dilemma they face. Translate
‘al-pənê in full as “on the face of ”, or reduce it to a simple preposition like “on”
or “at”? Not unreasonably, they tend to the former solution in passages when
the expression seems to have a broader significance, the latter when a more
pedestrian spatial designation is in play—but this still requires a judgment call
about where to draw that line. And, whenever they make the latter choice, they
sacrifice some of the flavor of the original.
This predilection of Biblical Hebrew for the fleshing out of prepositions with
body parts isn’t limited to expressions with “face”. Consider another marker
of Biblical style, the phrase “in the eyes of ”: in Hebrew, bə‘ênê, where bə- is
the preposition “in”, and -‘ên- is what ‘ayin reduces to in the construct state.
(We saw ‘ayin earlier as the word for “eye” as well as the letter and sound named
after it.) In dealing with this expression, the KJV and NRSV once again diverge,
with the older translation preferring to retain the Hebrew phrasing, and the
NRSV opting for more idiomatic English. In Genesis 6:8, the change is fairly
slight, with the syntax retained, and only a slight shift from “eyes” to the more
abstract “sight”: 

KJV But Noah found grace in the eyes of the Lord (bə‘ênê yhwh).
NRSV But Noah found favor in the sight of the Lord.
 Hebrew209

But in Genesis 41:37, the translations differ considerably, with the NRSV
­rewriting so as to remove the expression entirely: 

KJV And the thing was good in the eyes of Pharaoh, and in the eyes of all
his servants.
NRSV The proposal pleased Pharaoh and all his servants.

As a glance at the original will show, it is the KJV that has stayed closer to the
syntax of the Hebrew: 

way-yîṭab    had-dābār  bə-‘ênê      par‘ōh
and-it was good  the-thing     in-(the) eyes of  Pharaoh
û-bə-‘ênê kol-‘ăbādâw    
and-in-(the) eyes of all-his servants    

This is not to say, though, that the NRSV should be faulted for removing the
expression: since “it was good in the eyes of X” is a regular way of saying “it
pleased X” in Hebrew but not in English, there’s a lot to be said for using the
latter, simpler translation, so that the text doesn’t come across as more unnat-
ural than it was intended to be. That said, the substitution does make the
­language a little blander, less connected with a widespread type of Hebrew
expression, than is the case with the KJV’s alternative.
In passing, we may observe two other features of the Hebrew of this verse.
First, tucked away in the word glossed “it was good”, wayyîṭab, is a form related
to one of the Hebrew words most familiar to Americans: the tov of Mazel tov!
(literally “Luck good!”). Strip away the waw conversive and the yî- prefix of the
imperfect, then ṭab is left—and, as b belongs to the begadkefat constellation of
letters, it’s pronounced as a v when, as here, it follows a vowel. Second, note the
word for Pharaoh, par‘ōh. The word pharaoh is easy to misspell in English, with
the unusual sequence ‑ao- frequently reversed to the more familiar -oa-, most
regrettably in the name of the Triple Crown–winning horse American Pharoah.
But why does it have -ao- in the first place? Because those translating the Hebrew
Bible into the Greek Septuagint needed some way to represent the voiced
pharyngeal fricative ‘ayin in the Greek alphabet, and an alpha was the best
solution they could come up with (ϕαραώ). From there it passed into Latin and
then into English.
One final peculiarity of Biblical Hebrew worth noting in this context is that
it is comparatively poor in adjectives. This is not to say that it doesn’t have
them, just that they’re not used as often as in English. And the reason for this
isn’t that Hebrew leaves all its nouns unadorned by any descriptive modifiers at
all: rather, where English would use an adjective to modify a noun (e.g. “the
golden altar”) Hebrew will often take a second noun and have it modify the first
210 How dead languages work

noun in a construct chain (“the altar of gold”).14 For instance, there is an

­adjective qādôš “holy”, which is not uncommon in the Old Testament; it occurs
117 times.15 But there is also a noun built to the same root, qōdeš “holiness”,
which is found 470 times. Obviously not all of these examples of the noun
count as “nouns used where we might expect an adjective in English”, and
there’s no clear way to distinguish those instances from the rest, but a sample of
some of its usages will give a sense of the different flavor of the Hebrew here.
Take Psalm 98:1, which the NRSV translates:
O sing to the Lord a new song,
  for he has done marvelous things.
His right hand and his holy arm
  have gotten him victory.
In the third line, there’s a parallelism in the English, a pair of “his Adjective
Noun” phrases, that isn’t there in the original—although it’s certainly in keep-
ing with the sort of parallel structures of which Hebrew poetry is particularly
fond. In Hebrew, one finds instead: 

yəmîn-ô û-zərôa‘ qodš-ô

right hand-his and-arm of holiness-his

Both halves of the line, in fact, differ structurally from each other and from
their English counterparts. First, rather than having a separate adjective and
noun to express “right hand”, Hebrew simply has a noun yāmîn which combines
both concepts into a single word, whose ā vowel weakens to a schwa when the
suffix -ô is added to mark the possessive “his”. (This is the second element in the
name Ben-jamin seen above.) Second, “his holy arm” is, in Hebrew, a construct
chain: “his arm-of-holiness”, with the same possessive suffix -ô attached to the
end of the chain (here causing the reduction of the noun stem qōdeš to qodš-).

14  It is probably because such “Noun of Noun” phrases are so typical of Biblical language that
they have infiltrated the language of fantasy novels. As the “Honest Trailers” YouTube video for
Seasons 1–3 of George R. R. Martin’s Game of Thrones put it beautifully, “Travel to Westeros,
a place where everything is the Thing of Nouns”, citing such phrases as the Hand of the King, Lord
of Light, Master of Coin, and Mother of Dragons, to name just a few (https://www.youtube.com/
watch?v=SVaD8rouJn0). All of these would make for well-formed construct chains in Hebrew.
Indeed, at times, Hebrew goes even further in this direction, especially with its fondness for body
metaphors: ‘eṣem, the word for “bone”, can also be used in the sense “substance”, leading to the
expression bə-‘eṣem hay-yôm haz-zeh “on the very same day” (NRSV), literally “on the bone of this
day” (Gen. 7:13). Further examples of Hebrew’s propensity for the construct chain are prominent
in Psalm 23: the “green pastures” are literally “pastures of grass”, and the “still waters” are “waters
of rest”. And while “the valley of the shadow of death” (gê’ ṣalmāwet) is technically only a two-part
chain, since ṣalmāwet “shadow-of-death” is a compound noun rather than itself a construct chain
(which would be ṣēl māwet), the phrase still ultimately consists of the concatenation of three nouns.
15  Frequency counts are taken from M. V. Van Pelt and G. D. Pratico, The Vocabulary Guide to
Biblical Hebrew (Grand Rapids, 2003).
 Hebrew211

While in that first example neither the KJV nor the NRSV hints at the pres-
ence of a construct chain in the Hebrew, in another famous phrase it does come
across in translation: the Holy of Holies. It’s a familiar enough expression in
English that one doesn’t necessarily pause to recognize just how strange it is:
we don’t often pluralize an adjective like holy as holies. Even when adjectives
are treated as nouns referring to a plural entity, they tend to resist the ending -s,
as in The Young and the Restless. But here the translation Holies is a literal rendi­
tion of the second noun in the construct chain qōdeš haq-qădāšîm, found,
for instance, at Exodus 26:33, in which haq- is the definite article and qădāšîm
is simply the regular plural of qōdeš. (Once more, the pervasiveness of the
triconsonantal root is apparent: in this perfectly ordinary plural, the vowels
change completely but the q-d-š framework remains intact.) Even here, how-
ever, where a literal translation has become a possibility in English thanks to
the nativization of the phrase Holy of Holies, both the KJV and NRSV shy away
from it in order to suggest what the more fundamental English equivalent of
the Hebrew phrase would be: “The curtain shall separate for you the holy place
(qōdeš) from the most holy (qōdeš haq-qădāšîm)” (NRSV). For, in contrast to
English, where adjectives are inflected for three different degrees—positive
(old), comparative (older), superlative (oldest)—Hebrew adjectives do not have
this option, so the comparative or, as here, superlative has to be expressed dif-
ferently, with the phrase “X of Xs” being a favored way of getting across the idea
“the most X”. But since this is a structure that requires a plural and a construct
chain, and therefore works better with nouns, we can see part of what it means
for Hebrew to be a language that lets nouns do a little more work relative to
adjectives than would be the case in English.
Other examples of this typically Hebrew formation are not hard to find:
­simply scanning through the table of contents of a Bible, one finds the Song
of Songs (šîr haš-šîrîm). And anyone who’s ever heard Handel’s Hallelujah
chorus is likely to remember the phrase “King of Kings, and Lord of Lords”,
taken in this case from the Greek New Testament: basileus basileōn kai kyrios
kyriōn, with “of . . . -s” marked both times by the genitive plural ending -ōn
(Revelation 19:16). In Greek, while this is perfectly grammatical, just as
“King of Kings” is in English, it is not nearly as idiomatic a construction as
in Hebrew, and this in turn apparently affected the KJV translation of
Hebrews 9:3: 

KJV And after the second veil, the tabernacle which is called the Holiest
of all.
NRSV Behind the second curtain was a tent called the Holy of Holies
(Hagia Hagiōn).

Here, for once, it is the NRSV that stays closer to the wording of the original
text—presumably to capture the strangeness of the phrase in Greek, as opposed
212 How dead languages work

to Hebrew, idiom—while the KJV adapts the phrasing to the usual English
superlative construction.
This last verse is a fitting example with which to end the chapter and, indeed,
the book: a Hebrew turn of phrase that has been taken up into the Greek of
the  New Testament and, from there, has made its way into English, too.
Languages don’t exist in isolation. Words spoken in one language are some-
times of the utmost importance to speakers of another language, and there’s no
one-size-fits-all solution to the tension seen repeatedly in this chapter: Should
one be as literal as possible at the cost of producing a translated text that is
strange or even misleading in not using the natural idiom of the target ­language?
Or should one adapt the language to produce a more natural translation, but
sacrifice nuances of phrasing present in the original? That this tension has been
at work in two millennia’s worth of translations of the Bible has enriched the
linguistic possibilities available to speakers of languages like Greek and English:
any time the translator chooses the former option, it introduces a new element,
like “and it came to pass”, into the second language. Even in English, the
­language of the Old Testament still retains echoes of the Hebrew original—in
some translations more than others—and by far the best way of developing a
real sense for this is to sit down and actually learn a bit of Hebrew.
 Epilogue and further reading

This is a book that resists an easy conclusion. From the formulas of Homeric
poetry to the riddling language of the Rig Veda, from the intricate word order
of Horace to the almost willful complications of the Old Irish verb, from the
alliteration of Beowulf to the construct chains of the Hebrew Bible, what stands
out is the languages’ distinctive personalities. Given this diversity of expres-
sion, if asked how dead languages work, one is tempted to answer: each in its
own special way. On the other hand, because the book largely stays within the
confines of the Indo-European languages, certain similarities also come to the
fore, whether it’s the shared history of the case endings that signal what nouns
are doing in a sentence or the roots that underlie individual words. And if this
is what looms larger in the reader’s mind, one may prefer instead to draw out
these common features as at least a partial answer to the question. But it would
be irresponsible, in that case, not to offer the crucial disclaimer that the lan-
guages considered here cannot be regarded as a representative sample of the
world’s dead languages at large but only of those that have particular cultural
resonance for the average anglophone. I wish, of course, that I could have
included more: chapters on Classic Mayan, the Egyptian of the hieroglyphs,
and Ancient Chinese could all have helped to create a richer portrayal of the
linguistic environment of the ancient world. The three languages just mentioned,
for instance, did not have a case system of the sort that the Indo-European
languages had; and while Mayan and Egyptian do certainly have plenty of mor-
phological complexity, the same is not true of Classical Chinese, in which
words do not give the explicit indications of grammatical function that have
been so characteristic of the languages featured in this book.1 But while it
would have rounded out the strictly linguistic picture to have expanded it in
this way, it would have led to a diffuseness in the cultural landscape presented.
Languages are embedded in the societies that speak them, and the narrative

1  As I have no proficiency myself in any of these three languages, I here rely on the descriptions
offered in the relevant chapters of R. D. Woodard’s Cambridge Encyclopedia of the World’s Ancient
Languages (Cambridge, 2004), namely those by V. R. Bricker (“Mayan”), A. Loprieno (“Ancient
Egyptian and Coptic”), and A. Peyraube (“Ancient Chinese”).

How Dead Languages Work. Coulter H. George, Oxford University Press (2020).
© Coulter H. George. DOI: 10.1093/oso/9780198852827.001.0001
214 Epilogue and further reading

thread can be woven more tightly if the connections drawn between the sample
texts include not only specific points of grammar but also shared motifs, like the
dragon-slaying myth, and the broader influence one language can exert on
another, with Lucretius extending what Latin can do as he proselytizes for
Greek ideas, or the translators of the King James Version (KJV) repeating the
phrase It came to pass in order to reflect Hebrew syntax.
That said, if we do limit ourselves to the languages included here, one could
still reasonably ask: Are they so different from the modern languages that des-
cend from them that we can lump them together as their own separate cat-
egory? Do dead languages, as such, simply work differently? If we rely only on
this narrow sample, the answer may well be yes: if one compares, say, Ancient
Greek against Modern Greek, Latin against Spanish or French, and Old English
against Modern English, it’s hard not to see morphological simplification in
play. Since antiquity, Greek has lost one of its five cases (the dative), the optative
mood, and even its infinitives; Spanish and French no longer have cases at all;
and, in Chapter 4, we saw in some detail the drastic reduction in the number of
forms English has at its disposal. To be sure, one also needs to note certain signs
of renewal as well. Yes, Spanish and French got rid of the old Latin future, but
they still managed to create a new one through the fusing of an old helping verb
with the main verb. Whatever forms Greek’s verbs may have shed, they have
also acquired a new distinction of aspect in the future tense along the way. But,
in the end, looking at the descendants of Greek, Latin, Old English, Sanskrit,
and Old Irish, it is difficult to find examples of languages that are more mor-
phologically complex than their forebears, even if some modern Indo-European
languages, like Icelandic, remain extremely conservative in this respect. One
needs to hedge a little more than usual, but by and large it’s a fair simplification
to observe that the major Indo-European languages of antiquity have tended to
shed more forms than they’ve gained as they’ve entered the modern era.2 But
rather than tying everything up in a neat package—languages tend to bridle at
the imposition of too much regularity in any case—I much prefer to celebrate
the exuberance of expression that they offer: to understand the workings of the
Latin gerundive or Hebrew’s waw-conversive and the implications for how
those languages structure ideas. This, then, is the point at which I’d like to send
readers off to explore these languages and cultures in greater detail. But the
obvious problem is that the best way to do so is to learn the languages properly,
and teaching oneself Old Irish or Biblical Hebrew, while not impossible, isn’t
especially easy either. So I’ll offer a mix of different avenues one could poten-
tially try out, depending on one’s particular interests.
First, for those interested in the connections between the Indo-European
languages, the shortest and most accessible introduction to Indo-European

2  For more on this general pattern, see G. Deutscher, The Unfolding of Language: An Evolutionary
Tour of Mankind’s Greatest Invention (New York, 2005), especially pp. 266–74.
 Epilogue and further reading 215

linguistics is tucked away as an appendix to the American Heritage Dictionary,

5th ed. (Boston, 2016). Written by Calvert Watkins (whose work played a sig-
nificant role in the discussion of the Indo-European background of Sanskrit
poetry in Chapter 5), it begins with a general essay covering not only how the
comparative method is used to reconstruct the proto-language but also how the
resulting panoply of reconstructed words can give us an idea of the cultural world
of the Indo-Europeans (setting out, for example, some of the terms that the
speakers of Proto-Indo-European (PIE) will have had for the natural world,
kinship relationships, and religious concepts). This is followed by a short, but
thorough lexicon of the roots whose descendants have made their way into
English, either directly or through borrowings from Latin, Greek, and the other
languages of the family. As a bonus, John Huehnergard has complemented the
Indo-European appendix with a matching discussion of the Semitic languages,
together with a lexicon of Semitic roots that have ended up in the dictionary—
often, in this case, as proper nouns, given the relative paucity of direct borrowings
from Semitic languages into English, but if you want to know how shekel and
scallion could ultimately go back to the same root, then this is the place to go. Both
lexica of roots are conveniently available online (https://www.ahdictionary.
com/), and this site is also more generally an excellent place to get reliable
etymological information on words online.
Speaking of online resources: If one googles etymologies of words, the
Online Etymological Dictionary (https://www.etymonline.com) is often close
to the top of the search results, but it should only be used with considerable
caution, as it is often out of date and leaves out key diacritical marks like ­macrons.
Remarkably, Wiktionary (https://en.wiktionary.org) generally seems to be a
better resource, even if it cannot be regarded as definitive for obvious reasons.
Moving beyond simple etymologies, those who wish to dabble in the early
Indo-European languages without committing to a full course of study can also
profitably turn to the Early Indo-European Online site maintained by the
Linguistics Research Center of the University of Texas at Austin (https://lrc.
la.utexas.edu/eieol), which has ten-lesson introductions for over a dozen lan-
guages (including all five main Indo-European languages covered in this book),
structured around glossed samples of actual ancient texts.
For more substantial introductions to the field, two books stand out:
Benjamin Fortson’s Indo-European Language and Culture, 2nd ed. (Malden,
MA, 2010) and James Clackson’s Indo-European Linguistics (Cambridge, 2007).
The former, divided roughly into two halves, offers first a well-balanced
account of what is generally agreed to be known about PIE, followed by chap-
ters that detail the major sound changes and morphological in­nov­ations that
characterize the various daughter branches. The latter focuses more on meth-
odology and the limits of our understanding, with particular attention paid
to points of disagreement, such as the curious scarcity of *b in reconstruc-
tions, or the origins of the system of grammatical gender. Those who are more
interested in the vocabulary of the Indo-Europeans can also turn to J. P. Mallory
216 Epilogue and further reading

and D. Q. Adams’s Oxford Introduction to Proto-Indo-European and the Proto-

Indo-European World (Oxford, 2006), most of which is devoted to surveying
the roots that can be reconstructed, one semantic field at a time.
For individual languages, it is harder to single out useful books to consult,
since the most thoughtful accounts of the linguistic worlds of Homer, Beowulf,
and the like typically require knowing the languages in question. And while
self-study materials are available for all these languages, they often require a lot
of patience (and often previous experience with a morphologically rich lan-
guage like Greek or Latin) in order to work through them fully. Start with
Greek: there are any number of books available to introduce readers to the litera-
ture of the Greeks, but very few of them spend any time on how the language as
such contributes to their achievement. Nor is there an obvious beginning text-
book to suggest for self-study. Since it judiciously avoids overwhelming the
student with material more suited to the intermediate level, but offers compara-
tively full explanations of what it does present, Anne Groton’s From Alpha to
Omega: A Beginning Course in Classical Greek, 4th ed. (Newburyport, MA,
2013) would probably be my recommendation—but one still imagines that all
but the most motivated students will be better off if they can work with a teach-
er. Readers may also be interested in A Companion to the Ancient Greek
Language (Malden, MA, 2010), edited by Egbert Bakker, which assembles
essays on a wide variety of subjects related to the history of Greek. Most of
them, to be sure, assume at least a passing familiarity with the language, but
some of those that deal with broader cultural issues are fairly accessible.
Unsurprisingly, Latin is somewhat better served, considering that it is often
treated as a sort of entry-level dead language. Joseph Solodow’s Latin Alive: The
Survival of Latin in English and the Romance Languages (Cambridge, 2009) has
accessible discussions both of the language as it was spoken in its heyday and of
the changes it underwent as it developed into the Romance languages, along
with the multiple borrowings into English at several points along the way. Those
more interested in what Latin can do as a vehicle for literary expression can
turn to William Fitzgerald’s How to Read a Latin Poem If You Can’t Read Latin
Yet (Oxford, 2013) for many more examples of the effects not just Lucretius and
Horace but also Catullus, Virgil, Ovid, and others achieve with their language.
There are also numerous introductions to Old English. Those who want to
take it up as a language in its own right can work, for instance, with Peter Baker’s
Introduction to Old English, 3rd ed. (Malden, MA, 2012) or Bruce Mitchell and
Fred Robinson’s A Guide to Old English, 8th ed. (Malden, MA, 2011), although
those chiefly interested in the cultural world may prefer Mitchell’s An Invitation
to Old English and Anglo-Saxon England (Oxford, 1995). Meanwhile, Orrin
Robinson’s Old English and Its Closest Relatives: A Survey of the Earliest
Germanic Languages (Stanford, CA, 1992) situates Old English among its
Germanic kin, with brief grammatical and historical introductions to all the
old Germanic languages, together with extensive samples of text in each of them.
 Epilogue and further reading 217

Sanskrit, again, is not the easiest language to teach oneself, but it’s certainly a
much more manageable proposition than it used to be, thanks to the publica-
tion of Antonia Ruppel’s The Cambridge Introduction to Sanskrit (Cambridge,
2017), together with the numerous resources available through the associated
website (http://cambridge-sanskrit.org/). Curiously, Old Irish, arguably the
most obscure of the languages discussed here, also boasts an exceptionally
good textbook, David Stifter’s Sengoídelc: Old Irish for Beginners (Syracuse, NY,
2006), which offers paradigms that often give the PIE and Proto-Celtic ante-
cedents to the Old Irish forms so that learners can get a sense of how the
­language ended up where it is. (At the risk of appearing frivolous, I should also
note that it’s illustrated with utterly charming drawings of Old Irish–speaking
sheep, which go a long way towards giving one the stamina needed to get
through one irregular verb table after another.) While nothing comparable
exists for Middle Welsh, one can get a basic introduction to the language of the
Mabinogion from Gareth Morgan’s online site Reading Middle Welsh (http://
www.mit.edu/~dfm/canol/).
For Hebrew, Thomas Lambdin’s Introduction to Biblical Hebrew (New York,
1971) is solid but perhaps a bit daunting for self-study, with Jo Ann Hackett’s A
Basic Introduction to Biblical Hebrew, with CD (Peabody, MA, 2010), a good
alternative. Those who would like to know more in particular about the KJV of the
Bible can profit from the surge of interest in the translation that accompanied
its four-hundredth anniversary in 2011: Gordon Campbell’s Bible: The Story of the
King James Version 1611–2011 (Oxford, 2010) and David Norton’s The King James
Bible: A Short History from Tyndale to Today (Cambridge, 2011) both shed use-
ful light on the kinds of decisions that the translators had to face in producing
their version. A broader discussion of the difficulties of translation can be
found in David Bellos’s Is That a Fish in Your Ear? Translation and the Meaning
of Everything (New York, 2011).
Finally, for those who want to step back and look at the big picture of the
histories of languages and how they change, Nicholas Ostler’s Empires of the
Word: A Language History of the World (New York, 2005) examines the his­tor­ic­al
circumstances surrounding the flourishing of an astonishingly wide range of
the world’s languages, and Guy Deutscher’s The Unfolding of Language: An
Evolutionary Tour of Mankind’s Greatest Invention (New York, 2005) offers an
excellent account of the general mechanisms of language change (with an espe-
cially accessible description of the origins of the triconsonantal roots of the
Semitic languages).
What emerges from all of these works is the sheer wealth of material dead
languages give us to explore—more than one can possibly get to in a lifetime.
And whether you end up chasing down any of these leads or not, I hope you’ll
excuse me now: there’s some Greek I’d like to go read. Or maybe it’s time for
some Latin. Or Hebrew. You get the picture.
 Index

Words discussed as linguistic examples are listed under their respective languages.

ablative, see case, ablative Biblical Hebrew, see Hebrew, Biblical

ablative absolute  62–5 broad consonant  164–7
ablaut  109 n. 17 bucolic diaeresis  37
absolute (verbal inflection)  159, 171–3
abstract language  41–9, 90, 131–2 Caesar, Julius  3
see also noun, abstract caesura  37, 129–31
Achilles  28–41, 174–5 Callimachus 81
accusative, see case, accusative calque  66–7, 68, 141
active, see voice, active Calvin & Hobbes  47
adjective  21–2, 63–4, 78, 85, 105–6, case  20–22, 59–65, 114–16, 122, 125, 129, 135,
209–12 142–6, 160–1, 187, 214
comparative  105–6, 148 ablative  59–65, 80–1, 92, 143–6, 186;
compound 148 see also ablative absolute
equative 187 accusative  20–21, 31, 33, 46, 147, 150,
superlative  105–6, 148, 211–2 160–1, 171
Aeneid 30 dative  21, 46, 53, 80–1
affricates 102–3 genitive  20–21, 32, 46, 60, 68, 129, 149,
Agni 146–7 160–1, 206
aleph 192–4 in Greek  20–22
alliteration  88, 125, 129–31 in Latin  59–65
allophone 194 in Old English  114–16, 122, 125, 129
alveolar ridge  17, 138 in Old Irish  160–1
analogical leveling, see leveling in Sanskrit  142–6
Ancient Greek, see Greek, Ancient instrumental 143–6
animacy  53–4, 144 locative 143–6
aorist, see tense, aorist nominative  20–21, 31, 33, 35, 160–1
apostrophe 19 vocative  21, 166–7
appositive  147, 155 Catullus 81
Arabic  195, 196 causative, see verb, causative
Aristotle 94 Celtic languages  157–87, 199
ash (æ) 117 centum/satəm split  139–41
aspect  25–6, 214 chiasmus 83–4
imperfective 25–6 Celtic languages  10
perfective 25–6 Cicero  66, 67 n. 5, 68, 89
aspirated stop, see stop, aspirated classification of languages  99–114
aspiration  15, 139 comparative, see adjective, comparative
mutation 160 compound adjective, see adjective,
asyndeton 90 compound
augment  134, 150 compound noun, see noun, compound
auxiliary verb, see verb, auxiliary compound verb, see verb, compound
ayin, see Hebrew words, ‘ayin conciseness of expression  59–65, 80,
88–97
Baile Átha Cliath  163 conjugated preposition, see preposition,
barbarian, words for  3 conjugated
begadkefat  192, 209 conjunct (verbal inflection)  159, 171–3
Bellerophon 151 construct chain  205–12
Beowulf  9, 119–32, 151–2 Cú Chulainn  174–7
220 Index
dactylic hexameter  36–7, 40, 70, 79, 93–5, Epicurus  67–71, 74–5, 78
96, 130 epithet  36, 38–9
Damon, Cynthia  90–2 equative, see adjective, equative
dative, see case, dative Erard, Michael  1
dawn-cows 152–6 etymological fallacy  40
declension  115–16, 144
definite article  25, 43, 46–7, 60, 65, 80, 88–9, Feeney, Denis  58
96, 206 focus  82, 86–7
dental suffix  108, 110, 113 formula 35–41
deuterotonic 171–3 French  2, 14, 17, 73, 111–12, 123–4, 214
diachronic 104 fricative  15, 101–3, 191–2
diacritical mark  20, 77, 103, 200 fronting  168–9, 186
digraph 15 future, see tense, future
Donaldson, E. Talbot  120, 127, 131–2
dragon, how to kill a  148–52 Game of Thrones  210 n. 14
dual  22, 23 gender, grammatical  5–6, 21–2
Dutch 2 genitive, see case, genitive
German  2, 4–6, 14, 80 n. 18, 100–6, 121–2,
eclipsis 163 123–4, 170
Empedocles 68 High German vs. Low German  103
emphasizing pronoun, see pronoun, Germanic languages  9, 73, 97, 99–132, 138,
emphasizing 151–2
emphatic consonant  194–5 gerund  61–2, 83
endings, see inflections gerundive  62–3, 70, 91
English 18–19 glide 166
morphology of  18–19 glossing  25 n. 11
English words glottal consonant  192, 194
balsam  191 n. 3 Gothic  113, 190
bane 151 grammatical gender, see gender, grammatical
be  52, 96, 108, 113 grammaticalization 112–13
behemoth 198 Grant, Michael  89–91
bridegroom 121 Great Vowel Shift  35, 105
cherubim 198 Greek, Ancient  1–2, 8, 13–55, 58–9, 88, 109,
claymore 181 150–1, 152–3
do 113 morphology of  18–27
dragon 176 phonology of  13–18
druid  175 n. 17 Greek, Modern  14, 214
east 156 Greek words
Easter 156 aeide  32–4, 202
enough 122 angelos 66
handiwork 122 antallassō 44–5
Holy of Holies 211–12 axiōsis 48–9
Islam 196–7 dia 54–5
lay 106–7 dikaiōsis 48–50
lie 106–7 dīos  32, 36, 69–70
mazel tov 209 eis 52–4
mitzvah 195 ēōs 155–6
Muslim 196–7 ergon 45–6
name  154, 177 etymos 153
orient  206 n. 13 homoios 153
penguin 180 homos 153
pharaoh 209 kleos  34, 175
salaam 196 kosmos 66
seraphim 198 leipō 109
shall  121, 128 logos  45–6, 153
shibboleth 191 lykos 135
sing 108–10 men . . . de 45
 Index221
mēnis 39–40 infinitive  24–5, 65
nostos 175 initial mutation, see mutation
onoma  45–6, 154, 177 Irish 10
pherō 134–5 Middle 174
physis 68 Modern 162–71
polis 3 see also Old Irish
Zeus  31–2, 155 Italian  6, 63, 123–4
Grimm’s Law  35, 100–2, 107, 123
Jefferson, Thomas  97
haček 190 Jerome 72
Heaney, Seamus  120, 127, 131–2 Jones, William  133, 142
Hebrew Bible  11, 58, 199–212 jussive  201, 205
Hebrew, Biblical  10–11, 189–212
morphology of  195–9 kenning 124
phonology of  189–95 King James Version  51–55, 202–12
Hebrew, Modern  195 kinship terms  189–90
Hebrew words KJV, see King James Version
‘al-pənê 207–8 Koine 50
‘ayin  192–5, 208–9
bə‘ênê 208–9 labiovelar  140–1, 150
bəhēmôt 198 Language Log blog  3
bənê yiśrā’ēl 205–7 Latin  8–9, 16–17, 30, 43, 57–97, 112–13,
hāyāh 201–5 158, 198
kātab 198–200 Latin words and phrases
miṣwāh 195 ab urbe conditā  61–2, 64–5, 70
’ôr 201–2 addendum 63
pānîm 207–8 alma māter 85
qōdeš 210–12 annō Dominī 60
šālōm 196 ā prīncipiō  91–3, 95
wə 199–205 aspectus 73–4
Hesiod 152–3 aurōra 156
Hittite 140–1 bonā fidē 61
Homer 8 canō 116
see also Iliad, Odyssey Carthāgō dēlenda est 63
Horace  9, 78–88 cēterīs paribus 63–4
Housman, A. E.  9, 78–88 com-, con-  121–2, 176 n. 18
hyperbaton 86–7 coma 81
Deō volente 62
iambic pentameter  79 dī(v)us 69–70
Iliad  27–41, 120, 146, 174–5 elementum 70
imperative, see mood, imperative ē plūribus ūnum 61
imperfective, see aspect, imperfective ferō  134, 135
impersonal verb, see verb, impersonal gladius 181
indicative, see mood, indicative hodiē 186
Indo-European languages  10–11 honor 158
Indo-European linguistics  133–6 immortālis 85
Indo-European poetics  148–56 in vīnō vēritās 61
Indra 148–52 ipsō factō 60–1
infixed pronoun, see pronoun, infixed laetus 70
inflections  2, 7, 8–9, 18–27, 47, 51, 59–65, lībertās 90–1
75–6, 114–17, 125–6, 142–6, 195–9 lupus 135
of Greek  18–27 magnā cum laude  61, 76
of Hebrew  195–9 magnanimus 67
of Latin  59–65 malum 77
of Old English  114–17, 125–6 mālum 77
of Sanskrit  142–6 mortālis  75, 77–8, 85
see also morphology, inflectional mundus 66
222 Index
Latin words and phrases (cont.) Muses 152–3
mūtātīs mūtandīs 62–3 mutation  160–3, 172–3, 182, 186
nātūra 68–9
nōmen  154, 177 names, personal  34
ob- 74–5 Absalom 196
oppressus 74 Benjamin  206, 210
pangere 71 Hamish  163, 166–7
pecūnia 123 Ludwig  34–5, 127
prīmā faciē 61 Seamus  163, 166–7
prīncipium  91–3, 95 Solomon 196
pro- 123 nasalization  160–3, 171, 173
proxima poētīs 94 Neogrammarian Hypothesis  101
quālitās 67 New Revised Standard Version  50–5,
-que  30, 84 202–12
quod erat dēmōnstrandum 63 New Testament  11, 49–55
religiō 71–2 nominative, see case, nominative
requiēscat in pāce 61 noun
rēs 67–8 abstract  8, 45–9, 50, 106
SPQR 30 agent 147–8
summā cum laude 76–77 compound  123–4, 127, 141, 155
superstitiō 72 Greek 19–22
undique 145 Sanskrit 142–6
vice versā 64–5 see also verbal noun
Lebor na hUidre 159–60 NRSV, see New Revised Standard Version
lenition  160–3, 172–3, 180–2 number, grammatical  20, 22–3
leveling  108, 158–9 numerals  16, 189–90
Lewis, C. S.  68
lexicon  15, 66 Odyssey  30, 38, 59, 146
Livy 94–5 Ogam 160
Lucretius  9, 65–66, 67–78 Old English  9, 99–132, 151–2
Old English words
Mabinogion  179, 184–7 feoh 123
macron  20, 36, 75–8 from 123
McWhorter, John  4, 6 ge-  121–2, 127, 176 n. 18
middle, see voice, middle geong  121, 130
Middle Earth  118 guma  121, 130
Midgard 118 man 128–9
Modern Greek, see Greek, Modern middaneard 118
Modern Hebrew, see Hebrew, Modern sceal  121, 128
Modern Irish, see Irish, Modern singan 116
mood 23 wīg 126–7
imperative  23, 32, 33, 201–2 wyrm 151–2
indicative  23, 126 Old Irish  10, 157–87
optative 23 phonology of  160–2
subjunctive  23, 26–7, 125–6, 127 syntax of  167–73
morpheme 18 Old Irish words
bound 18 ainm 177
morphology 18–27 ben 159
derivational  18, 47–8 claideb 181
inflectional  18–27, 75–6 gaibid 171–3
of English  18–19 ríastrad 174
of Greek  18–27 Old Norse  100–1, 118, 151–2
of Hebrew  195–9 Old Testament  11, 58, 199–212
of Latin  59–65 opera 6
of Sanskrit  142–6 optative, see mood, optative
 Index223
oral poetry  8, 38–41, 58 Rēs Gestae Dīvī Augustī  68, 69
orthography 76 retroflex consonant  138–9
rhyme  88, 130
palatalization  130, 161–2, 164–7, 187 Rig Veda  10, 58, 146–56
palatal velar  140–1 root  31, 47–8
Paradise Lost 62 Russian 170
Parry, Milman  38
participle  22, 26, 61–5 Sallust 94
dangling  22, 31, 75–6 sandhi  141–2, 146, 155, 160 n. 3
passive, see voice, passive Sanskrit  9–10, 17, 115, 133–56
Paul  8, 49–55, 171 morphology of  142–6
perfective, see aspect, perfective phonology of  136–42
periphrastic construction  110–13 vowels 135–9
person 22 Sanskrit words
pharyngeal consonant  192–5 agni- 147
phoneme 193–4 ahann ahim 149–52
phonology  13–18, 136–42 bhar-  134, 135
of Greek  13–18 bodhisattva 138
of Hebrew  189–95 Buddha 138
of Old Irish  160–2 deva-  141–5, 147
of Sanskrit  136–42 Hare Krishna 138–9
Pindar 88 nāma-  154, 177
Plato  25–6, 40 puro- 147
Plautus 59 Sanskrit 138
plurals uṣas- 155–6
English  19, 104–5 Veda  138, 152, 154
German 104–5 vr̥ka- 135
Hebrew  198, 201 Sapir–Whorf hypothesis  3–6
poetic license  88–9 Scéla Muicce Meic Dathó  167, 177
polyglots 1–2 schwa  116, 179, 190–1, 196
Portuguese  111–12, 145 Scott, Walter  73
possession 170 Second Germanic Consonant Shift  102–3
postpositive  30, 84 Semitic languages  10–11, 189–99
pragmatics  82, 90 see also Arabic, Hebrew
preposition  8, 45, 49–55, 61, 91–3, 96, 143–6, Septuagint  11, 202, 207, 209
169–71, 183, 206 shared innovations  99–114
conjugated  169–71, 176, 177, 183, 186 slender consonant  164–7
present, see tense, present sound change  15–18
pronoun Spanish  5, 14, 15, 17, 51, 111–12, 134, 145,
emphasizing  171, 176 161, 214
infixed  171–3, 187 stop 15
prose  41, 58, 88–9, 94 aspirated  15, 102
Proto-Indo-European (PIE)  10, 16, 31, 101–2, voiced 101–2
106–7, 109–10, 123, 136, 139–40, 143, 150, voiced aspirated  139, 150
160, 186, 191 voiceless 101–2
Proto-Indo-European roots stress accent  105, 115
*gwhen- 148–52 subjunctive, see mood, subjunctive
*weid-  138, 152, 154 substantivization  63–4, 85
prototonic 172 subtopic 83
superlative, see adjective, superlative
question 183 synchronic 103
Quintilian 94 syntax 21

reconstruction 16 Tacitus  9, 88–97

reflexive verb  24 Táin Bó Cúailnge  159, 174–7
224 Index
tense 23 strong  107–14, 138, 158
aorist  23, 25–6 weak  107–14, 158
future  26–7, 111–12, 214 see also reflexive verb
imperfect  33, 112, 134, 198–205 verbal noun  182–3, 187
past 107–14 vocative, see case, vocative
perfect  93, 198–200 voice 24
present 25–6 active  24, 65
thorn (þ) 101 and participles  65
thunderbolt 149 middle 24
Thucydides  8, 41–9, 50, 96–7, 171 passive  24, 49, 53, 65, 90–1
Tocharian 140–1 vowel
Tolkien, J. R. R.  118, 181–2 in Germanic  103–7
topic  76 n. 12, 82–3, 85–7, 90–1 in Greek  13–14
topicalization  83, 85–7, 90–1 in Hebrew  190–1
Truss, Lynne  19 in Sanskrit  135–9
Twain, Mark  204 vowel triangle  137
Vulgate 202
ü  14, 104
umlaut 103–7 Wackernagel’s Law  29–30, 84
metal umlaut  103 Wamba 73
upsilon 14 Watkins, Calvert  10, 149–56, 215
Urim and Thummim  202 waw-conversive 199–205
uvular consonant  192 Welsh  2, 10, 178–87
Welsh words
velarization 166 cleddyf 181
Venus 69–70 Cymru 179
Vercingetorix  176 n. 19 West, David  72
verb  22–7, 107–14 Wood, Graeme  4
auxiliary  96, 110–13, 125–6, 128, 182 Woodman, A. J.  91–3
causative 106–7 word order  7, 8–9, 21–2, 29–31, 46, 75–6,
compound 159 82–8, 90–1, 124–5, 167–9
impersonal  170, 184 see also Wackernagel’s Law
Greek 22–7 wordplay  70–72, 91