

Towards standardization of morphosyntactic

terminology for general linguistics

Martin Haspelmath
MPI-SHH Jena and Leipzig University

is paper proposes that just like phonologists, linguists working on

morphosyntax should have a core set of standard terms that are understood in
exactly the same way across the discipline. Most of these terms are traditional
terms that are given a standard retro-de nition, because linguists already
behave as if these terms had the same meaning for everyone. e de nitions
are de nitions of general concepts (i.e., comparative concepts, applicable to all
languages in exactly the same way), but they are expected to be highly similar
to language-particular categories with the same labels. If linguists were close
to nding out the true natural-kind categories of Human Language that all
grammars consist of, there would be no need for de nitions, but since this seems
to be a remote goal, research on general linguistics must rely on uniformly
de ned general terms.

. Terminological consistency and standardization

Standardization has proved highly bene cial in many domains of technology and
trade, and it is impossible to imagine today’s world without worldwide standards
for basic units of time and space (the International System of Units, the Coordi-
nated Universal Time, and others).
In addition to such practical standards that ensure technological interoper-
ability, there are also standard nomenclatures in many elds of science that a ect
how scientists talk about their subject domain. Biologists have had generally
accepted conventions for naming species since the 18th century, and chemists
started to organize terminological standards in the 1860s. Linguists have had a
standard for representing general phonetic categories of segments since the 1890s:
what is now known as the International Phonetic Alphabet ( ).

https://doi.org/ . /tsl. . has

©  John Benjamins Publishing Company
 Martin Haspelmath

In this paper, I would like to make the case for a morphosyntactic counterpart
of the : a standard set of morphosyntactic terms for general linguistics. If such a
standard were possible, its advantages should be evident to every linguist who has
had more than a few years of experience in the eld. Many terms in linguistics are
used in a variety of ways that are o en confusing, and unless one is a specialist in
a particular area, one may be unaware of these ambiguities. As a result, automatic
literature search is o en problematic or impossible, and unfortunately, there is also
quite a bit of incomprehension and talking past each other.
A widespread attitude in the eld seems to be that the di culties of our subject
matter – the enormous complexities of the many di erent language systems that
linguists are grappling with – make it impossible to have a standard terminology,
at least at the present stage of our knowledge. It is this attitude that I would like to
challenge in the present paper. One of the reasons for being more optimistic is that
I have observed not only big di culties, but also quite a bit of sloppy terminological
use by linguists. Grammatical terms o en change their meanings through a novel
use of an existing term that is primarily motivated by the desire to avoid coining a
new term. For example, the term oblique used to refer to all non-nominative cases
(there was the nominative case and the oblique cases). Since the 1970s, however, it
has come to be used for all cases apart from the nominative and the accusative, as
well as (more generally) for all cases and adpositions that are used for arguments
other than the transitive subject and direct object (e.g., Nichols 1984). ere was no
particular reason for this change, and it would have been easy to coin a new term
instead. e confusion generated by the semantic shi of the term oblique was thus
unrelated to any particular di culties, and entirely due to the attitude that it does
not matter much if an older term is used with a new meaning. And indeed, within
a given narrow context, it is o en fairly clear what a term means. But from the
broader perspective of the entire eld of the language sciences, the current level of
terminological unclarity and ambiguity is undesirable.
us, while there are no doubt many di culties in the eld of general mor-
phosyntax, this need not make it impossible to have a standard terminology. Other
elds have di culties as well (including the phoneticians, who have the ), but
many elds of science are at least making an e ort to have terminological stan-
dards. I thus want to argue that the eld of general linguistics should have at least a
limited set of standard terms (say, a few dozen, comparable to the 107 letters).
It is important to note that my proposal in this paper concerns exclusively ter-
minology for general linguistics, and that I say nothing about language-particular
terms here. ere are many phenomena that can be talked about only at the level
of a particular language, e.g., the French Passé Surcomposé (as in elle l’a eu vue ‘she
had seen her’), or the Genitive Absolute in Ancient Greek, or even the “f-word” in
English. ere are no counterparts of these phenomena in most other languages,
 Chapter 2. Towards standardization of morphosyntactic terminology for general linguistics 

so they are not directly relevant to general linguistics. Terminology of the lan-
guage-particular kind can perhaps be standardized as well,1 but in this paper, I
will con ne myself to terms of general linguistics – in other words, to terms for
comparative concepts.
e notion of in grammar has been fairly widely
adopted since I rst coined the term in 2010 (see Haspelmath 2010, 2018a; Brown
& Chumakina 2013; Cro 2016; Dryer 2016), but it seems that many linguists are
still not fully clear about it. I will therefore brie y explain the distinction between
comparative concepts and language-particular categories in the next section, and
contrast both with the notion of (innate) natural-kind categories.

. Comparative concepts, language-particular categories,

and natural kinds

Confusingly, grammatical terms such as complementizer or consonant are used in

three di erent senses in the literature:

– as (category-like) comparative concepts

– as language-particular categories
– as natural-kind categories

A category-like comparative concept is a term that can be applied to any language

and that is identi ed in all languages by the same criteria.2 For example, a con-
sonant can be de ned as a sound segment that is articulated with at least partial
closure of the vocal tract, and a complementizer can be said to be a marker that
indicates that the clause in which it occurs is a complement clause. Comparative
concepts of this kind are needed for universal claims, e.g., about the position of
the complementizer within a complement clause (Dryer 2009). e comparative

. In view of the proliferation of grammatical terminology in different (West) German

school textbooks in the 1970s, politicians entrusted some linguistics professors with the task
of setting up a standard set of terms for German, English, French and Latin grammar for use in
secondary schools. is work has been ongoing for quite a while (see Hennig 2012), showing
that grammatical terminology can also be relevant to applied concerns such as teaching of
grammar and languages in schools.
. A comparative concept need not be category-like. A nonverbal stimulus or a text passage
in a parallel text can also be a comparative concept (such concepts are called etic comparative
concepts; Haspelmath 2018a: 87–88). Another type of comparative concept is the standard
lexical meaning, as in the Concepticon (List et al. 2019). Such comparative concepts are never
confused with descriptive categories, probably because of their specific names.
 Martin Haspelmath

de nition of a complementizer merely says that it is a marker (not that it is a

word), because many languages have elements that seem to be part of the verbal
morphology but function just like the English word that or the Italian word che.
And in fact, English has a marker like this as well: the su x -ing. is is used as
a complementizer in cases like She considered leav-ing. us, from a comparative
perspective, English has not only clause-initial complementizer words, but also an
a xed complementizer that is postposed to the verb.
Moving now on to language-particular terms, we can talk about English Com-
plementizers and Latin Consonants (with capitalization of unique entities), but
these categories are de ned di erently from the comparative concepts. In English
grammar, a Complementizer is generally said to be a word, so that -ing is not con-
sidered as a Complementizer. And what counts as a word (or rather, an English
Word) is determined by English-speci c criteria. Likewise, what counts as a Latin
Consonant is determined by Latin-speci c criteria. us, the semivowels [j] and
[w] are usually treated as Consonants for the purposes of (Latin-speci c) pho-
notactics, even though by the phonetic criterion, they would not be consonants
because there is no closure of the vocal tract.
Language-particular terms are sometimes written with capitalization, in order
to distinguish them clearly from comparative terms (Comrie 1976; Haspelmath
2010: Section 5), but most of the time, linguists rely on context to make it clear
whether they are talking about concepts of general linguistics or about language-
particular categories.
Both comparative concepts and language-particular categories must be
de ned in a precise way. It is sometimes thought that a prototype de nition is suf-
cient for comparative purposes. For example, Dingemanse (2019: 20) says that
“typological de nitions generally aim to capture the centre of gravity of a phenom-
enon rather than providing a list of necessary and su cient properties”, but this is
not true. If there were no list of necessary and su cient properties, it could not be
clear whether a phenomenon should be grouped under a comparative concept or
not, and it would not be possible to make a cross-linguistic database that records
the properties of a sample of languages in terms of the comparative concept.
However, comparative concepts need not be comprehensive, because lan-
guage comparison is always partial (only language description must ultimately be
complete). Languages have many structures that can be compared with similar
structures in other languages, but they o en have completely unique phenomena
that are not amenable to comparison (e.g., German Weak vs. Strong adjectives, or
the Latin Attributive Gerundive construction). us, it is possible to have de ni-
tions of category-like comparative concepts that focus on a shared core, where
the corresponding descriptive categories must be extensionally broader (see the
 Chapter 2. Towards standardization of morphosyntactic terminology for general linguistics 

discussion in Section 5 below). is may give the wrong impression that the com-
parative concept is vague or covers only a “prototype” (like the de nition of noun
in (5) below).
In addition to comparative concepts and language-particular categories,
many linguists also work with natural-kind categories. ese are di erent from
language-particular categories in that they occur not only in a single language
but are assumed to be potentially applicable to any language. is is so because
they are thought to be innately given (as part of a genetically determined gram-
mar blueprint called “universal grammar”), in advance of language learning (as
“pre-established categories”, cf. Haspelmath 2007). us, it is o en claimed that
there is a natural-kind category complementizer (o en written or simply
C), and that di erent languages instantiate it in di erent ways (o en by zero, and
sometimes even by movement of a verb). Similarly, consonants (o en written
“[+cons]”) can be thought of as an innate natural-kind category of phonology,
instantiated in di erent ways (o en by zero, as in “ phonology”, Clements &
Keyser 1983, or by non-movement in sign languages, Brentari 2002).
Natural-kind categories are very di erent from comparative concepts and lan-
guage-particular categories, because they are not instruments for research, but are
hypothetical results of research. ey have the status of chemical elements in 18th
century chemistry, before scientists had gured out what underlies the diversity of
chemical compounds (cf. Baker (2001) on the similarities between natural-kind
based generative typology and chemistry). Many linguists are skeptical of the natu-
ral-kind approach, and even those who assume the basic correctness of the approach
admit that the goal of discovering the true natural kinds of the grammar blueprint
is still distant. us, it would be premature to talk about standardization of natural-
kind categories, because we know too little about them, if they exist at all.3
In some other elds, the standard terminology concerns natural-kind catego-
ries, e.g., in chemistry (whose elements and compounds are natural kinds) and in
biology (whose species are also o en considered as natural kinds). And still other
elds, such as economics or political science, have no natural-kind categories at
all, as far as I am aware. Categories such as income, tax, state, and government are

. It seems to me that those linguists who think that standardization of grammatical termi-
nology is too difficult o en have natural-kind categories in mind, and if so, I fully agree with
them. Many linguists do not distinguish clearly between comparative concepts and natural-
kind categories, and this may account for the skeptical attitude toward standardization of
grammatical terms.
 Martin Haspelmath

social categories, and nobody would suggest that they are innate properties of the
human mind.4 us, there are three types of disciplines:

i. social sciences, which have culture-speci c social categories, but no natural-

kind categories;
ii. natural sciences, which have natural-kind categories but no social categories;
iii. sciences at the intersection of social and natural sciences such as linguistics,
which have both social categories (such as English Complementizer and Latin
Consonant) and natural-kind categories (though the latter are controversial in
linguistics).

But importantly, sciences have (observer-made) comparative concepts in

addition to the independently existing categories (social or natural) that they
encounter in the world.
In the natural and social sciences, comparative concepts are not so easily con-
fused with the independently existing categories, so the distinction is not high-
lighted frequently. But some well-known nomenclature systems in the natural
sciences are systems of conventional comparative concepts, e.g., the Yerkes spectral
classi cation of stars in astronomy,5 or the International Cloud Atlas Classi cation
of clouds in meteorology.6 Likewise, the social sciences have some well-known
nomenclature systems of conventional comparative concepts, e.g., the Hornbostel-
Sachs system of musical instrument classi cation in the comparative anthropol-
ogy of music,7 or the Human Development Index in comparative development
economics. ese systems of comparative concepts exist alongside culture-speci c
categories, and there is no danger of confusing them.

. Examples of possible standard de nitions of well-known terms

Before continuing the discussion of issues arising in standardization of terms, let

us now look at a few concrete examples of terms and their de nitions that I think

. is might be different with kinship categories like ‘mother’ or ‘brother’, which might con-
ceivably be innate categories, as they seem to be shared with other mammals that have kinship
organization, but nothing like humans’ complex culture.
. <https://en.wikipedia.org/wiki/Stellar_classification#Yerkes_spectral_classification>
. <https://en.wikipedia.org/wiki/International_Cloud_Atlas>
. <https://en.wikipedia.org/wiki/Hornbostel%E2%80%93Sachs>
 Chapter 2. Towards standardization of morphosyntactic terminology for general linguistics 

might be suitable for standardization. Naturally, many of these are taken from my
earlier or current work.
(1) sentence8
A sentence is a maximal clause, i.e., a clause that is not part of another
clause.
(2) clause
A clause is a combination of a predicate (full verb or nonverbal predicate)
and its arguments plus modi ers.
(3) morph9
A morph is a minimal form, i.e., a form that does not consist of other forms.
(4) root10
A root is a morph that denotes a thing, an action or a property.
(5) noun11
A noun is a morph (or a root) that denotes a thing, i.e., a physical object or
a person.
(6) a x12
An a x is a non-promiscuous bound morph that is not a root.
(7) bound form13
A bound form is a form that cannot occur on its own.
(8) marker14
A marker is a bound form that is not a root.

. <https://dlc.hypotheses.org/1725>
. Haspelmath (2020).
. Haspelmath (2012).
. A colleague has suggested that noun may be better defined as ‘a root that can be case-
marked’, but what does it mean to be case-marked? As seen in (16) and (15) below, ‘case-
marker’ can be defined with reference to ‘flag’, which is defined with reference to ‘nominal’.
Maybe the latter term can be defined without reference to ‘noun’, but I leave this open here.
. Haspelmath (2021a).
. Haspelmath (2021a); <https://dlc.hypotheses.org/1779>
. A reviewer notes that Pollard and Sag (1994: 44–45) provide an exclusively semantic
definition of marker (“a word whose semantic content is purely logical in nature”), which is
perhaps more in line with linguists’ intuitions. But I do not think that a free form would be
regarded as a marker, and the “non-root” part of my definition amounts to much the same (see
(4)), except that it is easier to apply than a vague concept like “purely logical”.
 Martin Haspelmath

(9) A-argument15
e A-argument is the argument of a two-participant clause that is coded
like the ‘breaker’ or ‘killer’ argument of ‘break/kill’, if the other argument
(the P-argument) is coded like the ‘broken thing’/’killed animal’.
(10) subject16
e subject of a clause is its A-argument or its S-argument.
(11) transitive clause (cf. footnote 12)
A transitive clause is a clause that has an A-argument and a P-argument.
(12) ergative construction17
An ergative construction is a construction with a transitive verb in
which the P-argument is coded like the intransitive S-argument, and the
A-argument is coded di erently.
(13) argument coding
Argument coding is the marking of an argument’s semantic or syntactic role
by means of a ag or a person index.
(14) passive construction
A passive construction is a construction (i) which shares the verb root
with the transitive construction, (ii) whose S-argument corresponds to
the transitive P-argument, and (iii) which requires oblique agging of the
argument corresponding to the transitive A-argument, if it can be expressed
at all.
(15) ag18
A ag is a bound morph that occurs with a nominal and that expresses its
semantic role.
(16) case-marker
A case-marker is a ag that is an a x.
(17) serial verb construction19
A serial verb construction is a monoclausal construction consisting of
multiple independent verbs with no element linking them and with no
predicate–argument relation between the verbs.

. Lazard (2002).

. Haspelmath (2011).
. Comrie (1978).
. Haspelmath (2005, 2019).
. Haspelmath (2016).
 Chapter 2. Towards standardization of morphosyntactic terminology for general linguistics 

(18) re exive construction20

A re exive construction is a grammatical construction
i. that can only be used when two participant positions of a clause are
lled by the same participant
ii. and that contains a special form (a re exivizer) that signals this
coreference.
(19) gender system21
A gender system is a nomi er system with up to 20 nomi er classes (= gender
classes) whose nomi ers are not restricted to occurring on numerals and
possibly other adnominal modi ers, or restricted to occurring on possessors.
(20) nomi er system
A nomi er system is a paradigm of grammatical markers which occur
on noun-associated forms and each of which expresses (partly re ects, or
partly contributes) a broad property of the selecting noun other than person
and number.
Several of these de nitions will look strange to experienced readers, and many
linguists will nd it easy to raise objections. But my proposal here is not that
these de nitions (which happen to be the ones that I use in my own work) should
become standard. And it seems that those colleagues who criticize the de nitions
generally nd it less easy to come up with more appropriate de nitions, so I have
decided to list them anyway here. e main purpose of this article is to make a
general case for standard de nitions of terms, and by giving some de nitions of
some basic terms, I provide a proof of concept how this can be done concretely,
even in di cult cases. (More speci c terms such as “free relative clause” or “sub-
sective adjective” are presumably easier to de ne than the basic terms.)

. Principles for standard morphosyntactic terms

A er having seen some concrete examples, let us consider a number of general

principles for choosing terms and de nitions.

. Haspelmath (2021b).

. Haspelmath (2018c). I have been asked whether 20 classes is not an arbitrary limit, and
indeed it is, but this number is sometimes mentioned in the literature (e.g., Corbett 2007: 242:
“the number of genders is not limited to two, nor to three: four is common and twenty is
possible”). is arbitrary definition is given here in order to show that some traditional terms
cannot be retro-defined in a way that appears natural. But whether a clearly defined concept is
“natural” (and what this might mean) is not a question that I address in this paper.
 Martin Haspelmath

First of all, we would like to have standard de nitions of well-known gram-

matical terms, so many of the de nitions in Section 3 concern well-known terms.
is is completely analogous to the , which provides standard de nitions of the
well-known letters of the Latin alphabet. I call such de nitions retro-de nitions,
because they assign a precise meaning to an existing term that does not have a
widely recognized precise meaning yet. (Some widely known established terms
do have a precise meaning, e.g., interrogative pronoun, or concessive clause; these
terms do not need to be retro-de ned, because their de nition is not in question.)
But this does not imply that widely needed comparative concepts cannot be
described by completely novel terms. Just as the includes many novel letters,
we need many additional comparative concepts for morphosyntax, and some of
the terms in Section 3 are quite novel (morph, bound form, marker, and ergative
construction were added in the 20th century, and ag and nomi er are even more
recent additions). Of course, the list is completely open-ended, and any general
linguist who feels the need to use a new concept should feel free to coin a new
term. Once a new term has been picked up by a certain number of other linguists,
it could be added to the list of standard terms. ( is is di erent from the : pho-
neticians do not seem to think that the list of possible symbols is completely
22
open-ended.) I do not expect the list of category-like comparative concepts ever
to be complete, because the range of morphosyntactic constructions that might be
compared across languages is very large and open-ended.
Each of the de nitions in Section 3 presupposes a number of other terms,
which must either be de ned in turn (as comparative concepts), or must be
assumed to be generally understood in the same way by everyone. Very basic lin-
guistic concepts such as ‘form’, ‘action’, and ‘semantic role’ can (or must) be le
unde ned (as primitives), and in addition the de nitions may of course contain
general nontechnical concepts such as ‘not’, ‘part of ’, or ‘require’. But many techni-
cal terms will need to be de ned in turn, as in the case of oblique in the de nition
of passive construction, or associated form in the de nition of nomi er system.23
Not all of the sample de nitions in Section 3 are thus complete in the sense of
being fully comprehensible. Some of them contain terms that are not widely
known yet (e.g., A-argument, ag, nomi er system), which I decided to include

. However, the actual descriptions of segment inventories in the world’s languages contain
an extremely wide variety of segment types. e Phoible database (Moran & McCloy 2019)
includes over 3,000 different segmental comparative concepts, and for some of them, having a
non-compound designation by means of a novel letter might well be useful.
. Similarly, Corbett (2007: 242) notes that since gender is defined in terms of ‘agreement’,
“the definition of agreement itself becomes important”.
 Chapter 2. Towards standardization of morphosyntactic terminology for general linguistics 

here in order to show how a number of more familiar terms (subject, case-marker,
gender system) are de ned.
Ultimately, all standard terms must be de ned in such a way that their de ni-
tions only include primitive concepts or other well-de ned concepts. is is not an
easy task, of course, so I do not foresee it to be nished within a few years, regard-
less of whether such proposals will be widely accepted or not.
e examples in Section 3 show that fairly straightforward de nitions are pos-
sible for frequently used terms, e.g., for sentence, morph, root and marker. Such
terms are not usually de ned by linguists, and are typically learned by ostension,
like everyday words. For example, Booij’s (2005) morphology textbook does not
provide a usable de nition of root: On p 29, we read that “Stems can be either
simplex or complex. If they are simplex, they are called roots”. is would seem to
exclude roots which have no in ection and are therefore not stems (like English
gold or solid). For the stereotypical Indo-European language, this may not be a big
problem (because most verbs, nouns and adjectives show in ection), but it does
not work for languages in general, because many languages have nouns that can-
not be in ected, but we would still call them roots. Moreover, Booij provides no
de nition of stem. ere are similar problems with the terms clause and sentence,
which are rarely de ned in a way that corresponds to the actual usage of the terms.
For example, a well-known online glossary de nes a sentence as “a grammatical
unit that is composed of one or more clauses”,24 but this is not the way the term is
used, because a clause may of course contain another clause (e.g., a relative clause,
or a complement clause), and not all such clauses would be called sentences. It is
clear that linguists have simply not invested a lot of energy into providing de ni-
tions, and that in many cases it is not di cult to improve on the current situation.
But other frequently used terms are harder to de ne in such a way that their
de nition broadly corresponds to their current use. For example, subject can be
de ned only through the terms A-argument and S-argument, which are not easy
to de ne (see (9), and Haspelmath 2011a). And the term gender system is par-
ticularly di cult to de ne. Numeral classi ers are very similar to gender markers,
and it seems that the main reason they are never included in discussions of gen-
der systems is that they are stereotypically characteristic of East Asian languages,
while gender systems are stereotypically characteristic of European (and African)
languages.25 us, numeral classi ers have to be speci cally excluded, as is done
in (19), and a new term (nomi er, short for ‘nominal classi cation marker’) needs

. <https://glossary.sil.org/term/sentence>
. Numeral classifiers are very similar to gender markers, cf. Japanese niwatori san-ba
[chicken three- . ] ‘three chickens’, Italian casa nuov-a [house new- . ].
 Martin Haspelmath

to be introduced as a general term that has numeral classi ers and gender markers
as subtypes.
us, standardization of grammatical terms has at least two aspects: Retro-
de nitions of existing widely-used terms (such as a x, sentence, subject) and
creation of new terms (such as A-argument, nomi er) when needed in order to
provide retro-de nitions.26
What are general principles for retro-de nitions? A rst principle is that an
established term should not be de ned in such a way that its de nition is at vari-
ance with traditional use. It should cover the core of the phenomenon designated
by the term (as generally understood), it should cover at least 80% of the cases
where the term has been applied, and it should not include too many cases which
would not be included traditionally. ere are some well-known cases where the
meaning of existing terms has been changed by prominent linguists (thus lead-
ing to much confusion),27 and this experience should not be repeated. It will
o en be impossible to nd a de nition that covers 100% of the traditional usage,
because this usage is frequently somewhat inconsistent, but 80% accuracy should
be enough to justify continuing the term.
If a traditional term is used so inconsistently that it is not possible to de ne it
in such a way that the de nition covers most of its uses, the term should be aban-
doned. Examples of such terms whose traditional use is not su ciently coherent
are in ection, (non) nite, and clitic; I do not know how to de ne them in such a
way that their de nitions would correspond very largely to traditional usage (see
Cristofaro (2007) on nite, and Haspelmath (2015) on clitic). And even though
many people still use word in a technical sense, it is not clear either how to de ne
it objectively (Haspelmath 2011b).
e de nition of a comparative term should be as simple as possible, even if
this means that not all cases that are traditionally subsumed under the term are
included. For example, if marker is de ned as ‘a bound form that is not a root’ (see
(8) above), this is a simple de nition, but it does not fully capture the intuition
that a marker expresses a grammatical meaning. Bound forms such as however
or basically are not roots (according to the de nition in (4)), but they would not
be considered as typical markers. us, the criterion of having a simple de nition

. But of course, one may also create completely novel terms that may become standard
once they prove to be useful and are picked up by many linguists.
. E.g., ergative for ‘unaccusative’ (e.g., Grewendorf 1989), government in Chomsky’s (1981)
sense (contrasting with the earlier established sense, cf. Lehmann 1983; Kibort 2010), or
anaphor in Chomsky’s (1981) sense (contrasting with the sense in computational linguistics).
 Chapter 2. Towards standardization of morphosyntactic terminology for general linguistics 

may con ict with the criterion of having a good match with traditional usage, and
one needs to strike a balance between the two criteria.
ese principles will not cover all cases, and there will always be a certain
amount of arbitrariness in de nitions of comparative concepts. is is as it should
be, because comparative concepts are not discoveries, but instruments for research
(like units of measurement in physics). If there were no arbitrariness, no standard-
ization would be required. e social implications of this will brie y be discussed
in the nal section.
Another important aspect of retro-de nitions is that they may have a restricted
extension in comparison to descriptive categories with the same name, as will be
discussed next.

. Shared-core de nitions of comparative concepts

e de nition of category-like terms need not correspond very closely to language-

particular categories, but may be semantically simpler and merely correspond
to the shared core of features of di erent language-particular categories. is is
because it is o en clearly meaningful to compare languages with respect to salient
“core” concepts, regardless of how exactly the language-particular categories are
delimited. For example, it is clearly very useful to compare languages with respect
to how they express things (physical objects and persons), e.g., how they form plu-
rals and how they express actions involving persons and things (‘the girl took the
pens’). We can thus de ne the term noun as a comparative concept as in (5) (‘a root
that denotes a thing’). Of course, in most or all languages, the class called “noun”
goes beyond this core set of possible denotations (e.g., Russian vojna ‘war’, svoboda
‘freedom’). us, the class of Russian Nouns cannot be de ned semantically – this
is known to all linguists from their syntax textbooks. But language-particular cat-
egories are de ned by language-particular criteria (e.g., by being combinable with
articles, or showing number distinctions, or in ecting for case), and these cannot
be applied to all languages. e reason why we call di erent classes in di erent
languages “nouns” is that they all include roots denoting objects and roots denot-
ing persons, so this is the de nition of the comparative concept. e fact that the
categories called “noun” in di erent languages usually include more elements is not
relevant to the de nition of noun is a comparative concept, because the meanings
of these other elements play no role in mapping the language-particular classes to
the comparative concept.
In a very similar way, Nikolaeva and Spencer (2013: 219–220) mention the
example of the term adjective, which they propose to de ne in terms of the con-
cept of ‘gradable property’. is decision leaves aside adjectives like ‘dead’ or ‘blue’
 Martin Haspelmath

(Nikolaeva and Spencer also mention more exotic cases like ‘alleged’), but again,
everyone seems to agree that gradable-property adjectives do indeed represent the
shared core of the various categories in di erent languages that we call adjectives.
If a language had a class of words that does not include any gradable property con-
cepts, it would not be called “adjective”.
Another example comes from the domain of gender. Languages with gender
classes o en have a feminine class, which can be de ned semantically as a com-
parative concept:
For some values, cross-linguistic comparison is straightforward: feminine gender
is the value which includes nouns denoting females, and the interesting typologi-
cal considerations are what other nouns may be included in this gender value [… ]
We need to de ne the core meanings and functions: we call a gender value the
feminine if it includes nouns denoting females, whether or not it also includes
diminutives. (Corbett 2009: 137)

e German Feminine class and the Arabic Feminine class are quite di erent in
their extension (both contain many inanimates), but they share the common core
of female animates, which makes it meaningful to compare them.
Similarly, the English preposition to and the Russian Dative su xes -u/-e/
-am share the recipient meaning (cf. Haspelmath 2010: 666) and can thus be said
to match the comparative concept ‘dative’, even though they are otherwise quite
di erent (the English preposition also marks spatial goals, and the Russian Dative
case su xes are also governed by some prepositions).28
Finally, the term subject as a comparative concept is de ned in (10) in terms of
the A-argument of two-argument clauses expressing a physical e ect and patien-
tive single-argument clauses, as discussed in detail in Haspelmath (2011a). By
contrast, language-particular descriptions must include all the argument of all the
verbs, including atypical two-argument verbs (such as ‘to look’ or ‘to like’), and
there is much less cross-linguistic uniformity with these other verbs. e literature
is full of discussions of how to use the term “subject” with these more hetero-
geneous verbs, but there is no doubt about physical-e ect verbs and patientive
single-argument verbs.29

. Corbett (2009: 137) mentions this example, too: “Similarly we call a case value the dative
if used for recipients, whether or not it can also be governed by prepositions.” (However,
Corbett is only concerned with “case features”, in which he does not include prepositions.)
. e need for shared-core comparative concepts is not o en mentioned in the literature,
but Cro (2016: 378–379) notes something very similar:

In general, extensionally “large” semantic categories that are given monosemous

definitions … do in fact o en fail as comparative concepts. For example, property
 Chapter 2. Towards standardization of morphosyntactic terminology for general linguistics 

It may well be that Dingemanse’s formulation in terms of a “centre of gravity”

(cited in Section 2 above) refers to what I call “shared core” here. However, what I
mean here is not a “core” of a phenomenon that obviates the need for a de nition
in terms of necessary and su cient conditions. e de nitions must be precise
(rather than prototype-based, see Section 6), but the mapping of a language-par-
ticular category onto the precisely de ned comparative concept is cross-linguisti-
cally somewhat variable.
e fact that comparative concepts o en refer to a shared core and are exten-
sionally smaller compared to language-particular categories means that not all
parts of every language enter the relevant comparison. Inanimate Feminines in
German are not part of comparisons of feminine gender classes across languages,
and arguments of experiential verbs (which are not typical transitive verbs) are
not part of comparisons of subjects across languages. is is not a problem, but
it must be kept in mind if one evaluates claims based on such comparative con-
cepts. In order to compare arguments of experiential verbs (such as ‘to like’), one
needs more ne-grained comparative concepts (cf. the microroles of Hartmann
et al. 2014).

. Stereotypes and prototypes

e literature on grammatical patterns worldwide is full of stereotypes that are

widely known, e.g.,

– Latin has free word order, but English has rigid word order
– Italian is a pro-drop language, but English cannot drop its personal pronouns
– Turkish is an agglutinating language, and Chinese is an isolating language
– English is poor in in ection, but richer in derivation
– some North American languages have noun incorporation
– Latin makes much use of non nite clauses
– German has many compounds
– the Romance languages have clitic pronouns
– English has gender only in personal pronouns

concepts taken as a broad category (stative, unary valency, gradable, inherent) do

not serve well as a comparative concept for understanding “adjectives”, because
semantic subclasses of property concepts – age, dimension, color, value, etc. –
exhibit different grammatical behavior in one and the same language […] e usual
solution to this problem is to use finer-grained categories, such as the property sub-
classes, or for core participant roles, the division into A, S, and P.
 Martin Haspelmath

Linguists almost never ask whether these stereotypes are true, but if they were true,
this would be very interesting. We see outside of linguistics that many stereotypes
are not true (e.g., most dogs are not called Fido), but some stereotypes are true
(Smith is indeed the most frequent surname in the United Kingdom).30 us, I nd
it important to know whether the above stereotypes are true (they may or may not
be), and in order to assess them, we need objective de nitions of these terms.
is should go without saying, but many linguists seem to treat the terms
involved in such stereotypes as somehow having an independent existence, regard-
less of their de nition. For example, at the end of their book on clitics, Spencer and
Luís (2012: 321) admit that they have not been able to come up with a de nition
of the term clitic that encompasses all and only those phenomena that they discuss
in their book. Still, they do not conclude from this that the phenomena they dis-
cussed may not be coherent (but united merely by the fact that some people use the
same label clitic for these phenomena). Similarly, Reuland’s (2018) overview paper
of “re exives and re exivity” contains no de nition of “re exive”, and the author is
apparently not interested in providing one (because he takes his goal as discovering
the relevant aspects of the innate grammar blueprint, not as comparing languages
systematically). And even though my (2011b) paper on the de nition of “word” has
been widely cited, many authors continue to use the term word, even in technical
contexts, as if the term had a clear meaning (perhaps hoping that such a meaning
will be provided by someone eventually). Another example of this nonchalant atti-
tude is Massam’s (2017) overview paper on noun incorporation. Massam writes:
“ ere is a lot of disagreement about exactly what constitutes noun incorporation
… the eld is rich with proposals and counter-proposals as to its true nature” (2017:
Section 7), but she provides no de nition. She seems to presuppose that noun incor-
poration has an independent existence, even if we have not found its de nition yet.
And Lieber and Štekauer (2009: 14) conclude their introduction to a handbook on
compounding by saying that “there are (almost) no reliable criteria for distinguish-
ing compounds from phrases or from other sorts of derived words” – but for some
reason, they still say that “it’s worth looking further”, as if nding a de nition were
a research result (rather than an indispensable methodological prerequisite).
us, many linguists have been unable to provide clear de nitions of terms,
but have nevertheless been unwilling to abandon the traditional terms. While
many have been content to simply ignore these problems (hoping that they are
not too serious, and/or that someone else will eventually solve them), others have
noticed the problem and have reacted by invoking “prototypes”. e idea that lin-
guistic categories are (sometimes) based on cognitive prototypes was made famous

. <https://en.wikipedia.org/wiki/List_of_most_common_surnames_in_Europe>
 Chapter 2. Towards standardization of morphosyntactic terminology for general linguistics 

by Lako (1987) and was explored in more detail by Taylor (1989). is work
was based on mental representations of particular knowledge systems, including
grammatical knowledge. But the cognitive notion of a prototype category cannot
simply be transferred to cross-linguistic categories, which are not represented in
any individual speaker’s mind. ere was some discussion of cross-linguistic pro-
totypes in the 1980s and 1990s (e.g., Newmeyer 1998: Chapter 4), but this idea has
not been pursued systematically in recent decades, as far as I am aware. So, I do
not see any basis for a systematic “prototype view” of cross-linguistic categories.
Cross-linguistic phenomena o en seem to cluster in certain ways, but the extent
to which these clusters are real or based on our stereotypes can be assessed only if
we have precise ways of measuring di erences between languages.
us, I see the standardization of well-known terms as one way to help the
discipline move beyond the traditional vague stereotypes.

. Standard comparative terms and language-particular description

Linguists encounter grammatical terminology most o en in the context of par-

ticular languages, rather than in a comparative context, as most linguists study
particular languages most of the time. us, the kind of standardization discussed
here might seem not to a ect most activities of linguists.
But appearances are deceptive. Even when a linguist talks about a particular
language, they typically want their ndings to be relevant to other linguists study-
ing similar phenomena in other languages. For example, García García (2020)
studies causative verbs in Old English (e.g., þwīnan ‘dwindle’, þwænan ‘cause
to dwindle’), and puts these verbs explicitly in a typological context. Likewise,
Nordlinger (2014) studies serial verbs in Wambaya (an Australian language), and
she puts these constructions in a typological context. e idea that the study of
particular languages has relevance for general linguistics has become almost uni-
versal, perhaps because there are nowadays many linguists who study languages
outside of an applied context (where languages are studied to facilitate language
learning, or understanding of important literary text). And when the research
question is theoretical rather than applied, then it is attractive to link one’s lan-
guage-particular insights to larger generalizations.31

. e comparative perspective is so deeply engrained in current linguistics that we hardly

notice its presence in everyday expressions such as “Turkish has a passive construction”, or
“Vietnamese has serial verb constructions”. Every time when we say that “a language has a
category X”, we make reference to a category-like comparative concept.
 Martin Haspelmath

Moreover, even though each language has its own categories (Haspelmath
2007), there are many similarities between the categories of di erent languages,
and we do not want completely di erent terminologies for di erent languages.
us, we call German verb forms like sind (‘are’) ird Person forms, even though
they are also used with the polite address form Sie (‘you’) (e.g., Sind Sie fertig?
‘Are you ready?’). From a language-particular point of view, these verb forms are
di erent from third person forms in, say, French, but it would not serve transpar-
ency to give them any other name. Similarly, the Russian Imperative has a use in
conditional clauses (E.g., bud’ ja na vašem meste [be. I on your place] ‘if I
were in your place’), so it is di erent from, say, the Latin Imperative, but it would
be strange to give it any other name.
So, from the point of view of terminological transparency, it is best if language-
particular categories are given names that correspond closely to comparative-
concept names. It is therefore also from a purely descriptive, language-particular
perspective that standard terminology is relevant; but because of the fundamental
di erence between comparative concepts and descriptive categories (Section 2), the
standard terms as discussed here are not crucial for language-particular description.
Finally, it should be noted that in practice, characters are used somewhat
di erently from morphosyntactic comparative concepts as envisaged here, because
they are o en used for language-particular notation. In fact, linguistics students
typically learn characters as a method of transcribing the pronunciation of
words in particular languages, and it is only later that they may be confronted with
for the purposes of cross-linguistic comparison. But it is a misunderstanding
to think of characters as a list of all possible sounds – rather, characters
are a list of well-known comparative concepts for comparing sound inventories
(Ladd 2011), which happen to work quite well for the practical purposes of tran-
scribing pronunciations. But it should be kept in mind that the , too, arose in
a comparative context: English-language teachers in France and French-language
teachers in Britain got together in order to create a tool facilitating language learn-
ing – which usefully starts with comparing one’s original language to the language
to be learned. For the purposes of scienti cally describing a language (in terms
of its own categories), one needs language-particular categories in phonology as
well, just as in morphosyntax. e o ers a convenient (widely understood) set
of symbols, but these are not actually crucial to language-particular description
(which must be based on contrasts and phonetic characterizations).
In this connection, a question that sometimes arises concerns the status of
gloss abbreviations in interlinear text. ere is a standard set of abbreviations in the
appendix of the Leipzig Glossing Rules,32 and this has become very popular (e.g.,

. <https://www.eva.mpg.de/lingua/resources/glossing-rules.php>
 Chapter 2. Towards standardization of morphosyntactic terminology for general linguistics 

for accustive, for genitive, for plural, and so on). But some linguists
seem to take it as the main aspect of the rules (while in fact, the gloss abbreviations
are merely the appendix), and I fear that a few even think that these categories are
intended to be universal categories. However, what is standardized here is merely
the relation between the abbreviations and the terms (e.g., the use of rather
than for the term instrumental). e gloss abbreviations say nothing about
the meanings of the terms themselves. And since interlinear glossing is typically
done from a language-particular point of view, they are normally interpreted as
representing language-particular categories. us, in the above gloss [be. ]
(for Russian bud’), the abbreviation stands for “(Russian) Imperative”, not for
a comparative concept. However, in context where we compare languages (e.g., in
a typological study), it makes good sense to provide “comparative glosses”, rather
than language-particular glosses. So in a context where word order in conditional
clauses in di erent languages is discussed, it is probably best to gloss the Russian
form bud’ as [be. ], because the Imperative form serves to indicate a condi-
tional clause in this context. e fact that the actual form is called Imperative and
is also used for imperative clauses is irrelevant here.
us, the technical terms of phonetics/phonology and morphosyntax play a
dual role, which helps us understand that some linguists feel that the distinction
between comparative concepts and language-particular descriptive categories is
somehow problematic or di cult to draw. But it is not a problematic conceptual
distinction, and all linguists agree that we should use the same technical terms in
both roles (rather than, say, having two completely distinct sets of terms).

. Concluding remarks

In this paper, I have argued that there should be some standard terminology for
morphosyntax, just as there are standards of terminology or notation in other
disciplines, including subdisciplines of linguistics such as phonetics. I have given
some concrete examples of possible standard de nitions of well-known (and novel)
terms, and I have addressed some issues and brie y compared the standardization
approach with alternative ideas making use of prototypes (or actually stereotypes).
Perhaps the most pressing question that readers have at this point is the ques-
tion of implementation: What needs to happen so that linguists actually adopt a
standard? But this question is outside the scope of the present paper. Adopting
a standard is a collective action problem, and there are many di erent proposals
for how to solve such problems. e task that I have set for myself in this paper is
merely to address the issue of standardization from a conceptual point of view. e
principles discussed in Section 4 should help to make the proposals acceptable to a
maximal set of linguists, but there will always be some arbitrariness in any standard.
 Martin Haspelmath

And as noted in Section 2, many linguists seem to conceive of their gram-

matical terms as universal mental entities (natural-kind categories of the innate
grammar blueprint), though this is not o en stated clearly. In order for the stan-
dardization programme to make progress, it is crucial to make a clear distinc-
tion between hypothetical universal innate categories (as true research results)
and universally applicable comparative concepts (as historically arbitrary research
instruments which are subject to standardization).

Acknowledgements

I thank the editors for inviting me to submit this paper, and two reviewers (including Eitan
Grossman) for helpful comments. I also thank the numerous colleagues who discussed the issues
raised in the paper in an Academia.edu session. And I am particularly grateful to Erich Round
and Greville Corbett for extensive discussions of some potential problems of my approach, as
well as Christian Lehmann, Edith Moravcsik and Andrew Spencer for further discussion.

References

Baker, Mark C. 2001. e Atoms of Language. New York NY: Basic Books.
Booij, Geert E. 2005. e Grammar of Words: An Introduction to Linguistic Morphology. Oxford:
OUP.
Brentari, Diane. 2002. Modality di erences in sign language phonology and morphophone-
mics. In Modality and Structure in Signed and Spoken Languages, Richard P. Meier, Kearsy
Cormier & David Quinto-Pozos (eds), 35–64. Cambridge: CUP.
https://doi.org/10.1017/CBO978051148 777.00
Brown, Dunstan & Chumakina, Marina. 2013. What there might be and what there is: An intro-
duction to Canonical Typology. In Canonical Morphology and Syntax, Dunstan Brown,
Marina Chumakina & Greville G. Corbett (eds), 1–19. Oxford: OUP.
Chomsky, Noam A. 1981. Lectures on Government and Binding. Dordrecht: Foris.
Clements, George N. & Keyser, Samuel Jay. 1985. CV Phonology: A Generative eory of the Syl-
lable. Cambridge MA: e MIT Press.
Comrie, Bernard. 1976. Aspect: An Introduction to the Study of Verbal Aspect and Related Prob-
lems. Cambridge: CUP.
Comrie, Bernard. 1978. Ergativity. In Syntactic Typology: Studies in the Phenomenology of Lan-
guage, Winfred P. Lehmann (ed.), 329–394. Austin TX: University of Texas Press.
Corbett, Greville G. 2007. Gender and noun classes. In Language Typology and Syntactic Descrip-
tion, Vol. III: Grammatical Categories and the Lexicon, Timothy Shopen (ed.), 241–279.
Cambridge: CUP. https://doi.org/10.1017/CBO9780511 184 7.004
Corbett, Greville G. 2009. Universals and features. In Universals of Language Today, Sergio
Scalise, Elisabetta Magni & Antonietta Bisetto (eds), 129–143. Dordrecht: Springer.
https://doi.org/10.1007/978-1-4020-8825-4_7