Tracing Translation Universals and

Translator Development by Word

Aligning a Harry Potter Corpus

Sofia Helgegren
2005-05-24

LIU-KOGVET-D--05/09--SE

Magisteruppsats i kognitionsvetenskap
Instutitionen för datavetenskap
Linköpings universitet

Handledare och examinator: Magnus Merkel

Abstract
For the purpose of this descriptive translation study, a translation corpus was
built from roughly the first 20,000 words of each of the first four Harry Potter
books by J.K. Rowling, and their respective translations into Swedish. I*Link, a
new type of word alignment tool, was used to align the samples on a word level
and to investigate and analyse the aligned corpus. The purpose of the study
was threefold: to investigate manifestations of translation universals, to search
for evidence of translator development and to study the efficiency of different
strategies for using the alignment tools.
The results show that all three translation universals were manifested in
the corpus, both on a general pattern level and on a more specific lexical level.
Additionally, a clear pattern of translator development was discovered, showing
that there are differences between the four different samples. The tendency is
that the translations become further removed from the original texts, and this
difference occurs homogeneously and sequentially. In the word alignment, four
different ways of using the tools were tested, and one strategy was found to
be more efficient than the others. This strategy uses dynamic resources from
previous alignment sessions as input to I*Trix, an automatic alignment tool,
and the output file is manually post-edited in I*Link.
In conclusion, the study shows how new tools and methods can be used in
descriptive translation studies to extract information that is not readily obtain-
able with traditional tools and methods.
Acknowledgements
First and foremost, I would like to thank Magnus Merkel, my tutor, for all
his patience, not to mention the invaluable support and feedback. Michael
Petterstedt was a great help, especially in the early stages, by assisting in the
setup and guiding in the use of the alignment tools. I am also grateful to
Helge Dyvik for creating semantic mirrors from the resources built up during
the alignment.
A big thanks to my family, who have supported and encouraged me through-
out this sometimes slightly overwhelming process. Lastly, I am forever indebted
to Oscar, who always listens and cares when I need it the most.
Contents

Contents

1 Introduction 1

2 Background 5
2.1 A Brief Introduction to the Harry Potter Series . . . . . . . . . . 5
2.1.1 The Harry Potter Series and Culture . . . . . . . . . . . . 6
2.2 The HP Series from a Translation Studies Perspective . . . . . . 7
2.2.1 A Note on the Translator . . . . . . . . . . . . . . . . . . 7
2.2.2 The Harry Potter Books as Novels . . . . . . . . . . . . . 7
2.3 Previous Studies of the Harry Potter Books . . . . . . . . . . . . 8

3 Translation Theory 9
3.1 Translation and Culture . . . . . . . . . . . . . . . . . . . . . . . 9
3.2 Descriptive Translation Studies . . . . . . . . . . . . . . . . . . . 10
3.3 The Effect of the Translator . . . . . . . . . . . . . . . . . . . . . 11
3.4 Translation Universals . . . . . . . . . . . . . . . . . . . . . . . . 11
3.4.1 Explicitation . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.4.2 Simplification . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.4.3 Normalisation . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.5 Translation of Fiction . . . . . . . . . . . . . . . . . . . . . . . . 14
3.6 Children’s Literature in Translation . . . . . . . . . . . . . . . . 15
3.7 Constraints on Translation of Children’s Literature . . . . . . . . 16

4 Studying Translations 19
4.1 Parallel Corpora . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.2 Sentence Alignment . . . . . . . . . . . . . . . . . . . . . . . . . 20
4.3 Part-of-speech Tagging . . . . . . . . . . . . . . . . . . . . . . . . 20
4.4 Word Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4.4.1 Guidelines for Manual Word Alignment . . . . . . . . . . 21
4.5 Non-1-to-1-operations . . . . . . . . . . . . . . . . . . . . . . . . 22
4.6 Lexical Shifts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.6.1 Strategy for Lexical Shifts . . . . . . . . . . . . . . . . . . 25
4.7 Paraphrasing and Lexical Choice . . . . . . . . . . . . . . . . . . 25
 CONTENTS

4.7.1 Examples of Rejected Lexical Choices in the HP-corpus . 26

5 Methodology 27
5.1 The Sequence of Work . . . . . . . . . . . . . . . . . . . . . . . . 27
5.2 A Presentation of the Tools . . . . . . . . . . . . . . . . . . . . . 28
5.2.1 I*Link . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
5.2.2 I*Trix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.2.3 New Tools, New Possibilities . . . . . . . . . . . . . . . . 30

6 The Making of the HP-corpus 33

6.1 The Corpus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
6.2 Word Aligning the Corpus . . . . . . . . . . . . . . . . . . . . . . 34
6.2.1 Aligning HP1 . . . . . . . . . . . . . . . . . . . . . . . . . 35
6.2.2 Aligning HP2 . . . . . . . . . . . . . . . . . . . . . . . . . 36
6.2.3 Aligning HP3 . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.2.4 Aligning HP4 . . . . . . . . . . . . . . . . . . . . . . . . . 37
6.3 Comments on the Alignment Process . . . . . . . . . . . . . . . . 37
6.3.1 Problems Common to the Samples . . . . . . . . . . . . . 38
6.4 Post-editing HP1 . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

7 Results 41
7.1 The HP-Corpus . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.2 Translational Results . . . . . . . . . . . . . . . . . . . . . . . . . 41
7.2.1 Additions and Deletions . . . . . . . . . . . . . . . . . . . 42
7.2.2 Translation Universals . . . . . . . . . . . . . . . . . . . . 49
7.2.3 Investigating Translational Choices . . . . . . . . . . . . . 53
7.3 Methodological Results . . . . . . . . . . . . . . . . . . . . . . . . 58
7.3.1 Evaluation of the Different Strategies . . . . . . . . . . . 59

8 Discussion 61
8.1 Discussion on the Translational Results . . . . . . . . . . . . . . 61
8.1.1 FDG Imperfections . . . . . . . . . . . . . . . . . . . . . . 61
8.1.2 The Relationship between Additions, Deletions and Lex-
ical Shifts . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
8.1.3 Translation Universals . . . . . . . . . . . . . . . . . . . . 63
8.1.4 The Development of the Translator . . . . . . . . . . . . . 64
8.1.5 Sources of Error for the Translational Results . . . . . . . 64
8.2 Discussion on Tools and Methodological Results . . . . . . . . . 65
8.2.1 Using the Alignment Tools . . . . . . . . . . . . . . . . . 65
8.2.2 Advantages and Disadvantages of Using I*Link . . . . . . 65
8.2.3 Specifics of I*Link as Sources of Error . . . . . . . . . . . 66
8.2.4 Suggestions for Improvements of the Tools . . . . . . . . . 67
8.3 Suggestions for Further Research . . . . . . . . . . . . . . . . . . 68

Bibliography 71
Chapter 1

Introduction

Translations and the original text they are supposed to be the equivalent of are
not exactly the same. They differ in many ways, some changes occur naturally
as the basic structure of different languages is not the same, and some are due
to choices made by the translator. However, the differences between the two
texts might not be caused just in the process of translation, but also by the
process of translation.
Studies have shown that there are structural differences not only between
a specific original and its translation, but also between translations and other
texts written in the same language in general (Baker 1996). Translated text
often has certain characteristics that sets it apart from other texts written in
the same language. These characteristics are claimed to be the result of a
subconscious process in translators to ensure that the text is understandable to
the new readers in the new context.
Translation is all about context. It is about taking one text out of its cul-
tural context and making it available to a whole new readership that is not
a part of that cultural context, and therefore cannot have the same vantage
point as a reader from the source culture reading the source text. Because of
this, in translating the words of the text, the translator must also take the for-
eignness of the text into consideration, and decide whether that is something
worth preserving for the foreign feel, or if it should be adapted to the target
culture readers. Over the years, the translation studies community has shifted
from favouring the source-oriented approach and its very close rendering of the
original text, to the target-oriented approach that focuses on readability and
achieving an equivalent effect in the target culture (Tabbert 2002). It is a shift
from smaller segments to larger and from closeness to ease of understanding.
The objects chosen for this study are the first four books in the astoundingly
successful Harry Potter series written by J.K. Rowling, and their translations
into Swedish. There are many reasons behind this choice, but a fact that makes
them so interesting to study is that they belong to the genre of children’s litera-
ture but have not been treated exclusively as such. They have attracted readers
both older and younger than the intended one, and through their success, they

1
2 CHAPTER 1. INTRODUCTION

have gained a unique status in children’s literature. Moreover, the Harry Potter
books belong to a sub-genre, namely fantasy.
There are two additional reasons to why the Harry Potter books were cho-
sen for this study. Firstly, they are all translated into Swedish by the same
translator, Lena Fries-Gedin. This fact makes it possible to study the books
contrastively, and see if there are any structural differences between the trans-
lations. The translations may have changed over time and therefore, it is inter-
esting to analyse the samples sequentially. With a formal description of actual
changes, it would be possible to ascertain if the translator’s work is consistent
over time or if changes can be detected.
Secondly, the four books were written, translated and published within a
relatively short period of time, which makes it more likely that any contrastive
differences between the samples are actually due to changes in the approach
of the translator, and not any of the other possible sources of change, such as
a change in the cultural climate due to long periods of time passing between
the publication of the original text and the translation. This makes it possible
to pose questions concerning whether the translator in some way develops over
time and whether that is traceable in the produced texts.
In order to study the translations, the samples of source and target texts
were aligned. Alignment is a method in which each sentence in the source text
is paired with the corresponding sentence in the target text. This method was
also used on a word-level, i.e. each word or cluster of words, depending on the
nature of the text segments, was paired with the corresponding units. This
allows for all translated segments to be linked together in units of source and
target words, and makes visible the words in the source texts that were omitted
in the translation process, as well as any words that were added to it, i.e. exists
in the target text but not in the original. Consequently, changes in the text
that occur in translation can be studied, which is why alignment is the chosen
method for the study. This study is data-driven, and the hypotheses took a
preliminary shape during the manual sentence alignment.
The purpose of this study is threefold. In the field of translation studies,
one purpose is to investigate whether the so called translation universals are
manifested in these texts, and if they are, what form do they take? The second
purpose is to contrastively study the samples to discover whether there are
detectable differences between them that could indicate that the translator’s
approach has in some way changed from the first book to the fourth.
The methodological purpose is to investigate alignment and to evaluate the
different alignment strategies used. Aligning as a method for studying transla-
tions is also evaluated, especially in relation to the new kind of information the
new type of alignment tools used in this study can provide in comparison with
traditional tools.
The hypotheses that will be investigated are:

• The translation universals are in some way manifested in the samples.

• The translator’s style in translating the texts will have changed over time,
 3

primarily measured in the number of additions and deletions made to the

texts.
• Different strategies in using the alignment tool should affect the efficiency
of the process of using the tools in significantly detectable ways.
The set of alignment tools that is used in this study is new and unexplored,
which means that this is a new type of study. Therefore, there are no existing
frameworks for analysis available, and a great deal of effort has gone into the
analysis of the material. The lack of a framework also means that a critical
approach must be taken concerning the usefulness of the tools. Consequently,
the advantages and disadvantages of the tools and the associated methods will
be discussed.
This is a first attempt at a new way of studying translations, and it must be
seen as such. No similar studies have been done, to my knowledge, neither using
the same type of tools, nor attempting to investigate the change over time in
the translations of one particular translator, of books from one particular genre
written by the same author. The point of this study is not to uncover universal
truths about translations, but to study one particular type of translation made
by one translator, and to present a way to systematically investigate translations
using new tools and methods.
4 CHAPTER 1. INTRODUCTION
Chapter 2

Background

It is difficult to find words to describe the success of the Harry Potter books, and
considering the number of copies sold in both English and various translations
world wide, perhaps an introduction seems superfluous. Nevertheless, the books
present a rather specific mix of two different worlds which presents difficulties
to both translators of the series, and to readers of this thesis that are unfamiliar
with the books. Therefore, a brief summary of some important aspects of the
series is provided below. In addition, an explanation is given to why the Harry
Potter books were chosen for this study.

2.1 A Brief Introduction to the Harry Potter

Series
To date, five books have been published in the Harry Potter series. Each book
is set in two different worlds, one being the suburban boredom of number four,
Privet Drive, in some fictitious town in middle England. The other is the
exciting and action-packed magic world, predominantly set at Hogwarts School
of Witchcraft and Wizardry.
The protagonist is the young Harry Potter, a lonely, unloved, bespectacled,
friendless 11-year-old orphan who lives in Privet Drive with the Dursley family,
consisting of his aunt Petunia, her husband Vernon, and their son Dudley. As
the reader soon learns, Harry’s parents were a witch and wizard, and they
were killed by an evil wizard named Voldemort when Harry was a baby. For
some reason, Harry survived the attack with only a lightning-shaped scar on his
forehead. Friends of his parents brought him to the Dursleys, who unwillingly
accepted to bring him up.
The Dursleys want nothing to do with the magical world, and by ignoring
it they hope to eradicate Harry’s potential magical powers. This proves to be
fruitless, and as Harry is repeatedly invited to come to Hogwarts, they are in
the end forced to give up. The game keeper and keeper of the keys at Hogwarts,
Rubeus Hagrid, simply comes and takes Harry with him to go shopping for his

5
6 CHAPTER 2. BACKGROUND

school things. This is the introduction to the magical world, for Harry as well
as for the reader.
Apart from this introduction to the magical world in the first book, the
books all follow more or less the same format. They start in Privet Drive, in
the summer holidays, with a bored and lonely Harry harassed by Dudley. As
the school year starts, Harry by some means, usually the chartered Hogwarts
Express leaving from platform 9 3 /4 at King’s Cross station, goes to the magical
world of Hogwarts where great adventures of different sorts happen. The books
end with a crisis and a sometimes bitter-sweet triumph for Harry in a fight in
which he defeats the Dark Side, i.e. Lord Voldemort or some of his followers.

2.1.1 The Harry Potter Series and Culture

There are essentially three layers of culture in the Harry Potter books. The first
is of course the image of normality, or life as the reader knows it, portrayed in
the life in Privet Drive.
The second is the British public school system that the stories are so depen-
dent on, particularly the fact that Hogwarts is a boarding school (Davies 2003).
The Hogwarts culture is vividly described by Rowling through the use of board-
ing school elements, which contribute greatly to the very explicit Britishness of
the books. A few examples of this is the Hogwarts Express, the chartered train
that takes the students directly to Hogwarts, the school houses, dormitories
and the Head Boys and Girls. Through the boarding school setting, the books
portray a very British world, and one question the translator needs to ask him-
or herself is if this should or should not be retained in translation.
The third layer is the magical world, which places the books in the fantasy
genre. This layer is very much woven into the boarding school setting, and
separating the two is perhaps not necessary for the purpose of this study. Suffice
it to say that the basis in the books is always the normality and boredom of
Privet Drive, skillfully contrasted with the other layers that serve to trigger the
imagination and capture the interest of the reader. The complex interaction of
the different layers of culture present an interesting challenge in the translation
process.

Specifics of the Magical World

Apart from the Britishness of the books, they are very much characterised by
the magical elements. Shopping for magical things (such as cloaks, spell books,
pewters, potion ingredients and wands) is done in Diagon Alley, a street in
a magical, parallel part of London unreachable for the non-magic people, or
Muggles, i.e. those without magic power. The students at Hogwarts School of
Witchcraft and Wizardry take classes such as Transfiguration, Defence Against
the Dark Arts, Potions, and Care of Magical Creatures.
In the names of the professors and the rest of the characters, Rowling has
used a lot of imagery and cultural references. This would normally pose a
problem to translators and could be an interesting area of research, but the
2.2. THE HP SERIES FROM A TRANSLATION STUDIES PERSPECTIVE7

Swedish translator has chosen to keep all names of the leading characters in
their original English versions, and has only translated minor characters and
some animal names.

2.2 The HP Series from a Translation Studies

Perspective
The plots in the samples are of little interest as this is not a literary study, but
some aspects of the magical world need explaining. This is because both the
world of magic, witches and wizards, and the boarding school setting of Hog-
warts closely woven into the magic world, pose a problem to translators. There
is a vast terminology related to magic, which is, in effect, a subculture, and the
usage of some terms differs between English and Swedish. In addition, Rowl-
ing frequently coins new terms and invents new concepts that are not normally
associated with magic, for example the game Quidditch. These new concepts
increase the complexity of the magical world, and are perhaps even more dif-
ficult to transfer in translation because they are completely new. Sometimes,
these will carry certain connotations and cultural references that the translator
must both recognise and succeed in translating.
Translating such a complex mix of worlds is neither easy nor straightforward.
That is why this study focuses on the general patterns of choices made by the
translator, rather than isolated mistakes or successful translational choices. This
is also why the study focuses on a contrastive investigation of the samples; it
would not be unreasonable to expect some sort of development in the transla-
tions, because of the large amount of text written by the same author in the
same genre translated by the same translator.

2.2.1 A Note on the Translator

As mentioned above, the HP books are all translated by the same translator,
Lena Fries-Gedin. She has been translating for nearly fifty years, starting when
she was a student, and continuing parallel to her teaching career, but increasing
heavily after her retirement. Fries-Gedin has mainly translated literature for
adults, but because she had translated some books about a princess and dragons,
she was offered to take on the first Harry Potter book, Harry Potter and the
Philosopher’s Stone (Bergius 2003).

2.2.2 The Harry Potter Books as Novels

Placing the Harry Potter series in a genre is not as straightforward as could be
expected. The obvious solution would be to state that they are children’s books,
but I argue that this is not the whole truth. As O’Connell points out (1999),
all children’s books are, to some extent, written at least in part for adults (see
section 3.7), and for a number of reasons, this is even more so with the Harry
Potter books.
8 CHAPTER 2. BACKGROUND

As is obvious to any reader of the series, the length of the books has increased
with every new published piece. Particularly the later books that span 636
pages for the fourth book (Rowling 2001a) and 766 pages for the fifth book,
Harry Potter and the Order of the Phoenix (which is not included in the HP-
corpus), demand very much more of a young reader than ordinary children’s
fiction does. The length alone suggests that the books are meant to be read
by fairly accomplished readers with a certain amount of patience and stamina,
and for children perhaps even more so since no pictures or illustrations are used.
Moreover, as Harry Potter grows older (as he does with every book, because each
book describes the event of one school year), the plot becomes more complicated
and the demands on the reader therefore increase. Consequently, at least the
later books in the series merit discussion as novels, in my opinion, at least from
a purely literary perspective.
The conclusion of the discussion above is that first and foremost, the books
are fiction, as they portray fictitious events. Secondly, they contain many ele-
ments from the fantasy-genre. Thirdly, and naturally, they are children’s books.
In general, however, I state that they can be seen as novels targeted on both
adults and children. From a translational perspective, however, it is important
to consider the fact that at least one part of the targeted audience is children,
which will be explained in section 3.7.

2.3 Previous Studies of the Harry Potter Books

There are a few published studies on the Harry Potter books from a translation
studies perspective. Eirlys E. Davies, for example, has studied the treatment
of culture-specific items, or CSIs as she calls them, in Harry Potter and the
Philosopher’s Stone and several of its translations (2003). This article is a very
interesting read for anyone with a scholarly interest in the Harry Potter books,
although most of what it covers is beyond the scope of this study.
The process of translating Harry Potter into Brazilian Portuguese is re-
counted in an article by professional translator Lia Wyler (2003). Though it
is a reflection of her personal experience, it discusses the books from an insid-
ers point of view, as well as gives an interesting peek into the Harry Potter
phenomenon and its reception in Brazil.
Chapter 3

Translation Theory

In this chapter, relevant theory from the translation studies field is presented.
The particular research questions investigated in this study are explained in
connection with the corresponding background theories.

3.1 Translation and Culture

Translation is, in the words of Peter Newmark, “rendering the meaning of a
text into another language in the way that the author intended the text” (1988,
p. 5). The text to be rendered, the original, is commonly referred to as the
source text, or ST. The text that the translator produces is the target text,
or TT. Some words, phrases and concepts in the source language, or SL, have
one-to-one correspondences in the target language, or TL, and are fairly simple
to render in the new language.
However, “since no two languages are identical...it stands to reason that there
can be no absolute correspondence between languages. Hence there can be no
fully exact translations” (Nida 2000, p. 126). Languages are not identical,
because a language and the culture in which it is used are very intimately
connected, and any text that is produced in a certain language is an artifact of
the accompanying culture. Naturally, this has implications when a text is to be
translated, because “translation is a kind of activity which inevitably involves
at least two languages and two cultural traditions, i.e., at least two sets of norm-
systems” (Toury 1995, p. 56). Translating is taking a text out of its cultural
context and bringing it into another, foreign context.
Because there can be no absolute correspondence between languages, trans-
lations must be closer to either the source or the target language. The source-
oriented approach is literal translation, in which closeness to the original text
is pivotal, whereas free translation favours the target language and culture
(Newmark 1988). The distinction between the two is by no means absolute,
and most translations are not fully, but to some degree, oriented towards either
the SL or the TL.

9
10 CHAPTER 3. TRANSLATION THEORY

In the history of translation studies, much discussion has pivoted around the
concept of free and literal translation, and which one is to be preferred. Until the
beginning of the nineteenth century, a free style that emphasised the spirit and
sense of the text was favoured. After this, the study of cultural anthropology
dictated that language “was entirely the product of culture”, which brought
with it the idea that translation was nearly impossible, and that it at any rate
needed to be as literal as possible (ibid., p. 45). This rather extreme point
of view was gradually abandoned, however, and today, translations tend to
be more target oriented (Baker 1996). Moreover, in translation studies, the
prescriptive approach saying what a translation should be like has been replaced
by a descriptive approach, aiming instead to explain what a translation is really
like (Tabbert 2002).

3.2 Descriptive Translation Studies

According to Toury (1995), translation studies can be divided into sub-genres on
different levels. On the first level it is a question of “pure” or applied translation
studies. The latter concerns translator training, translation aids and translation
criticism, which is beyond the scope of this study. The interest here is in pure
translation studies, which can be either theoretical or descriptive. In turn, the
descriptive branch is focused on either the product, i.e. the text itself, the
process of translation, or the function of the text. Toury claims that the three
are not as separate as the division implies, but that they are in fact to some
degree interdependent on each other.
This study focuses on the product of the process, that is the text in itself,
and the possible differences between the source and target versions. It is not
a study of the process of translation, as the only artifact that is available for
study is the text, and the text says very little about the process. The process
is cognitive, and as with all cognitive processes there is a black box problem,
in that processes that take place in the human brain cannot be studied in
a simple way (Holmes 2000). However, with the help of the alignment tools
used in the study, certain aspects of the process can be investigated through
the linguistic patterns the translator produces, as the tools allow consistent
differences between the source and target texts to be discovered. Patterns are,
naturally, not inconclusive evidence of the translation process, but if strong and
general patterns can be detected, this is in the very least an indication that
the linguistic choices that are the basis of the patterns are indeed part of the
process, and not just coincidence. What lies behind the specific choices made
by the translator is, however, impossible to determine simply through studying
the text and is beyond the scope of this study.
The received opinion, nowadays, is that the source text is just one factor of
many that come into play in the translation process (Newmark 1988). Trans-
lations are instead seen as the product of a situational process, where elements
like the translator in question, the target culture and the particular constraints
on the situation (such as deadlines, payment etc.) interact and influence the
3.3. THE EFFECT OF THE TRANSLATOR 11

produced text.

3.3 The Effect of the Translator

Traditionally, translation has not been seen as a creative activity, and translators
are not supposed to have a style of writing of their own that is visible in the
target text (Baker 2000). However, it is a truth universally acknowledged in
the field of translation studies that if a number of translators were all given
the same source text to translate into the same language, not many sentences
would be translated in exactly the same way. If there is so much variation in the
way different people translate, there must be an effect of the translator. The
question is how, and indeed if, such an effect can be studied.
A small-scale study made by Mona Baker suggests that it is possible to
“identify patterns of choice which together form a particular thumb-print or
style of an individual literary translator” (ibid., p. 260). The focus in such
studies, Baker emphasises, must be on the patterns the translator produces,
rather than on the specific cases that could be brought up in order to prove a
certain point. These patterns can be investigated using a corpus made up of
large parts of the translator’s production.
In investigating the style of a translator, his or her background and what
is known about it must be taken into consideration, and “whatever we manage
to establish as attributable to the translator’s own linguistic choices must be
placed in the context of what we know about the translator in question” (ibid., p.
258). In addition, the relationship between the cultures involved is significant,
specifically whether they are closely related or disparate.
The HP-corpus only contains texts from one genre, written by the same
author, and is in no way representative of the scope of Lena Fries-Gedin’s work.
Therefore, it is natural that anything that can be said here about her translating
style is limited to the material used in this study. It is specific to this genre
of text, written by this author. However, the four samples can be compared
and contrasted sequentially, in order to reveal whether the specific style in this
context has changed from the first to the fourth book.

3.4 Translation Universals

Translations have certain universal features that separate them from original
texts, and these features are caused in and by the process of translation. Mona
Baker has given this issue a lot of attention, and states that the universal features
come natural, since “the nature and pressures of the translation process must
leave traces in the language that translators produce” (Baker 1996, p. 177).
One challenge that faces scholars interested in the universals of translation is
that they are rather vague notions and studying them is by no means straight-
forward or easy. The first question to ask is in what way each feature might be
manifested in a particular text, and how these manifestations can be located.
12 CHAPTER 3. TRANSLATION THEORY

When this is done properly, a computerised corpus should be able to provide

a lot of information and is the proposed basis for a study of the translation
universals.
Baker focuses on three universals of translation, namely explicitation, sim-
plification and normalisation, and the combined effect of the three is that trans-
lations are usually less complex than their original texts. This is particularly
interesting in a study of the translation of children’s literature, since the strate-
gies that are less faithful to the original but serve to adapt the text to the target
language are used more freely for this genre in order to achieve texts that are
easy to read (O’Connell 2003).

3.4.1 Explicitation
The theory of explicitation concerns the tendency in translations to “spell things
out rather than leave them implicit” (Baker 1996, p. 180). Explicitation can
be expressed syntactically or lexically. For example, translated texts tend to
have a higher degree of conjunctions than original texts. Lexical explicitation
can be made through various means, but oftentimes it is made by adding nouns
in order to explain some piece of information that needs to be explained to a
target culture reader.
Another possible manifestation of explicitation is the fact that translations
tend to be longer than their original texts. When translations become longer,
the additions to the ST are often made to explain features in the ST that
might not be known to readers in a TT-culture. Thus the translation becomes
more understandable than a more faithful rendering. This manifestation has
the advantage of being relatively easy to examine.
In this study, explicitation is thought to manifest itself in two ways. Firstly,
that the TTs are longer than the STs was evident on a very early stage. Sec-
ondly, if more information has been added to the target texts than removed
from the source texts, this also indicates that they have been explicitated.

3.4.2 Simplification
Simplification is the tendency of translated texts to contain simplified language
compared to the original text (ibid.). For example, long sentences are often
divided into several shorter ones.
One indicator of simplification is a relatively low lexical density, meaning
that the number of function words or grammatical words is high, in proportion
to the number of lexical words. Lexical words contain more information than
grammatical words, and using fewer lexical words means that the reader will
have to keep track of less information. Using less variated vocabulary is also
one manifestation of simplification.
Another possible sign of simplification is that punctuation tends to change in
translations. According to Malmkjaer (1997), punctuation is rateable on a scale
from weak to strong in the order comma, semicolon and full stop. In translations,
punctuation usually becomes stronger, in that commas are often translated into
3.4. TRANSLATION UNIVERSALS 13

semicolons or full stops, and semicolons are translated into full stops. If the
punctuation is stronger, it is highly likely that there are more sentences in the
TT than in the ST, which indicates that long and complex sentences have been
divided into several shorter ones, and thereby the complexity of the text has
been decreased.
In the HP-corpus, simplification is assumed to be manifested in long sen-
tences being divided into several shorter ones, stronger punctuation and the
removal of the regional dialects that some characters speak in (see discussion
below).

3.4.3 Normalisation
Normalisation or conservatism is what Baker calls the “tendency to exaggerate
features of the target language and to conform to its typical patterns” (1996,
p. 183). This can take the shape of the translator over-using clichés or typical
grammatical structures of the TL, often grammaticising elements of texts that
are ungrammatical in the source.
Normalising also involves adapting the punctuation to the typical usage of
the TL. For example, commas are used much more in English than in Swedish.
Ingo states that a Swedish reader is much disturbed by an overuse of commas,
and strongly recommends that the amount of commas is adapted to the usage
of the target language (1991). One of the ways in which normalisation will
be investigated in the HP-corpus is through the treatment of punctuation, and
whether or not any evidence can be found of it being adapted to fit Swedish
usage.
Another element of the Harry Potter books in which normalisation might be
manifested is in the treatment of the different dialects used for certain characters
in the source texts dialogues. Dialect “differs from person to person primarily
in the phonic medium” and “has to do with the user in a particular language
event: who (or what) the speaker/writer is” (Hatim & Mason 1990, p. 39). The
effect of changing a character’s dialect can be considerable, as in the French
version of the first Harry Potter book, where the dialect of Rubeus Hagrid has
been normalised and grammaticised (Davies 2003). In the English versions of
the books, Hagrid’s speech casts him as a “down-to-earth, simple, uneducated
and in some ways childlike character” but in the French version, his utterances
are “characterized by impeccable grammar and standard, even somewhat formal
vocabulary” (ibid., p. 82).
Dialect is a language variation that is dependent on the user, and Hatim
and Mason distinguish between idiolectical, geographical, temporal, social and
standard/non-standard variation (Hatim & Mason 1990). For the purpose of
this study, the main interest in dialect is the use of different geographical di-
alects, or accents. Accent is the variation in language that roughly corresponds
to the geographical origin of the speaker. Accents can carry ideological and
political implications that translators must be aware of, and because of this
translation of accent is problematic (ibid.).
14 CHAPTER 3. TRANSLATION THEORY

In the Harry Potter series, accent is used actively in the depiction of dif-
ferent characters, not only for Rubeus Hagrid, but also for Stan Shunpike, the
conductor on the Knight Bus in Harry Potter and the Prisoner of Azkaban
(Rowling 2000a). Through alternative spelling in the utterances of Hagrid and
Stan Shunpike, that clearly deviates from standard English spelling, Rowling
represents the phonic qualities specific to two very different geographical di-
alects.

Example of Hagrid’s dialect (Rowling 1998, p. 48):

’It’s gettin’ late and we’ve got lots ter do tomorrow,’ said Hagrid loudly.
’Gotta get up ter town, get all yer books an’ that.’

Example of Stan Shunpike’s dialect (Rowling 2000a, p. 31):

’Can’t do nuffink underwater. ’Ere,’ he said, looking suspicious again, ’you
did flag us down, dincha? Stuck out your wand ’and, dincha?’

Both dialects are to certain extents ungrammatical, and it could prove in-
teresting to see if the translator has chosen to grammaticise the utterances, or
adapted them to Swedish in some other way. Significantly, the dialects are very
different, and should this difference not have been retained in the target texts,
this is not only an instance of normalisation, but also of simplification, since it
decreases the complexity of the texts.

3.5 Translation of Fiction

The books in the HP series belong to the fantasy genre, which also entails that
they are fiction. In translation theory, it is very difficult to find theorists that
speak about fiction with any interest at all. The focus tends to be on literary
texts, which are considered to be serious and artistic, and neither fiction nor
children’s literature is usually included in this category. However, due to the
reasons stated in section 2.2.2, I argue that the Harry Potter books have many
of the elements that characterise serious literature, and are therefore subject to
some of the same constraints.
Bearing this in mind, there are a number of issues particular to the transla-
tion of literary texts that put constraints of the translator, demanding a lot of
effort. Newmark distinguishes between three functions a translation must meet,
namely the expressive, the informative and the vocative functions (1988). There
is no strict division between these, and elements of all three can usually be found
in most texts, although to different degrees. Fiction, in the form of novels, is
placed among the serious imaginative literature as having mainly expressive and
vocative functions.
Prominent for the expressive function is the mind of the writer, who “uses
[the] utterance to express his feelings irrespective of any response” (ibid., p.
39). This is reflected in the writer’s personal use of language, and Newmark
3.6. CHILDREN’S LITERATURE IN TRANSLATION 15

emphasises that those personal, expressive, elements must not be normalised in

translation. Examples of expressive elements can be “’untranslatable’ words”,
unconventional syntax, neologisms and uses of dialect (ibid., p. 40).
The vocative function concerns the readership, and the intended effect of the
text to make the reader “act, think or feel, or indeed ’react’ in the way intended
by the text” (ibid., p. 41). One factor in these texts is the relationship between
the writer and the readership. Another is the fact that “these texts must be
written in a language that is immediately comprehensible to the readership”
(ibid., p. 41-42). It can be argued that in the case of children’s literature, this
is especially important due to an assumed difference in linguistic skills and world
knowledge between the translator and the readership (see discussion below).

3.6 Children’s Literature in Translation

Toury claims that translations usually occupy peripheral positions in the target
literary system (1995). The more peripheral a text or its genre seems to be to
the target culture, the more adjustments of the text will the translator tend to
make in order to adapt it to the norms of the receiving culture.
Children’s books and translations of children’s literature tend to be seen as
peripheral in most systems, something that can affect the process of translation.
Shavit (1981) argues that translators of children’s literature have a much greater
degree of freedom in relation to the source text, and “can permit himself great
liberties regarding the text because of the peripheral position [of] children’s
literature” (p. 171). However, this generalisation does not hold for the Harry
Potter series, as it cannot be rightfully described as peripheral. Nevertheless,
Shavit’s argument is still valid for the first book, which was still peripheral at
the time of translation into Swedish.
Stolze (2003) has indicated that it can be questioned whether or not the
translation of children’s literature is indeed different from the translation of
adult literature, since the original of a translation for children was also written
for children, as adult novels in translation are originally written for adults.
Stolze’s opinion is that this is dependent on the way children are seen in different
cultures, and that they should not be looked down upon as not being able to
understand many things. However, translating takes place in the publishing
industry, in which children are indeed marginalised (O’Connell 1999).
Consequently, the translation of children’s literature is subject to certain
constraints that sets it apart from translating for adults. O’Connell (2003)
points out that children have their own culture into which adults, among them
the translator, have limited insight. Moreover, there is a significant difference
“between the knowledge and linguistic skills of the translating adult and the
children who make up the target language audience” and in translating for
adults the translator can “expect the target readership to have approximately
corresponding levels of linguistic skills, general knowledge and world experience”
(ibid., p. 229). The knowledge level of the receiving audience is indeed a
constraint in the translation of the Harry Potter series, because of the fact that
16 CHAPTER 3. TRANSLATION THEORY

it is set in such a British environment and contains so many concepts that are
completely foreign to Swedish children.

3.7 Constraints on Translation of Children’s Lit-

erature
Children’s literature was for a long time a neglected area in translation studies
(O’Connell 1999). Today, it enjoys much more attention and both descriptive
and theoretical studies on the subject abound (Tabbert 2002).
Eithne O’Connell points out four features specific to this genre, indicating
some issues that separate translating children’s literature from translating adult
literature (1999). Children’s literature:

1. has two specific audiences, namely children and adults.

2. has ambivalent texts, with both literal meaning and a deeper, interpretable
meaning.

3. is written and purchased by others than the primary readership, i.e. adults.

4. has many functions and cultural constraints, in that they are intended to
both entertain and educate.

The fact that the genre has two audiences has some interesting implications.
In the relationship between adults and children, the power is with the former
group, which is very much reflected in the area of children’s literature. Adults
write, edit, publish, market and buy the books that are intended for children,
which means that the primary audience is more or less without say when it
comes to what they read. Parents decide what is suitable for their child, but
children and adults are not likely to have the same taste in literature (ibid.).
Number two above, although worth investigating, is not something that will
be pursued further in this study, as it is more interesting to do so from a literary
angle.
Because works of this genre are produced in a more or less exclusively adult
environment, it is important for the adults in that environment to be very
much in touch with current children’s culture. In all literary production, the
writers, publishers, editors and indeed, translators, have to be aware of the
current trends in the culture for which they produce, which is not a trivial
matter, and in children’s literature, it is complicated by the fact that adults
cannot be equal members of the child community. Still, they must know and
understand the culture, in terms of what children find interesting, how they
speak and think, current vocabulary, and so on. Otherwise, the style of the
language used in the translation risks being dated, and the readers will notice
this. As Eirlys E. Davies points out, “translating for children may present more
of a challenge than translating for adults; young readers are perhaps less likely
to be tolerant of the occasional obscurity, awkwardness or unnatural-sounding
3.7. CONSTRAINTS ON TRANSLATION OF CHILDREN’S LITERATURE17

phrasing which adults, conscious that they are dealing with a translation, may
be more accepting of” (2003, p. 66).
Due to the educational goal of children’s literature, studying explicitation,
simplification and normalisation might be of particular relevance, as there is an
even greater need to make texts understandable for the readership in order to
meet with the goal to educate. One important part of the purpose to educate
is, as Puurtinen (1998) points out, that adults expect children’s literature to
help in the development of the child’s linguistic skills. Therefore, there might
be a stronger tendency for translators of children’s literature to normalise the
texts by grammaticising them, in order to avoid the readership learning faulty
grammar from the books.
18 CHAPTER 3. TRANSLATION THEORY
Chapter 4

Studying Translations

This chapter gives background information on corpora and how alignment of

corpora can be used in studies of translations. It also explains how some complex
changes that translators make in translating texts are treated in the alignment.

4.1 Parallel Corpora

Originally, the word corpus was used for a collection of writings, usually writ-
ten by the same author. In modern corpus linguistics, it has come to mean “a
collection of texts held in machine-readable form and capable of being analysed
automatically or semi-automatically in a variety of ways” (Baker 1995, p. 225).
Corpora are created for specific purposes, and can be of different types depend-
ing on the intended use.
Parallel corpora consist of texts that in some way are parallel. The typical
parallel corpus contains original texts written in one language or language vari-
ety, and one or more translations of this text into one or more target languages,
or language varieties (Borin 2002). The relationship between the text(s) and its
translation(s) is one of translation equivalence (ibid.).
With parallel corpora, translated text can be studied in a number of ways,
but in this study, the point is to discover translation effects. The basic idea
behind this concept is that translated text can be linguistically and structurally
different from original text, and in what way they differ can be discovered com-
paring STs with their TTs through the use of parallel corpora.
When starting a parallel corpus project, the first step is to select the texts
to be included and create electronic versions of them. This can often be quite
time-consuming, as it usually involves a great deal of manual work, such as
typing, scanning and proofreading the material (ibid.). Borin also points out
that the use that can be made of parallel corpora depends heavily on which type
of tools are available to the researcher. However, the next step in the process
can be done manually without the use of specialised tools.

19
20 CHAPTER 4. STUDYING TRANSLATIONS

4.2 Sentence Alignment

Alignment of the corpus texts is a process performed on parallel corpora. Align-
ing a corpus is “the process of identifying and pairing up corresponding units
in the two (or more) languages making up the parallel corpus” (ibid., p. 20).
This can be done on different levels, for example sentence alignment and word
alignment.
In sentence alignment pairs of more or less equivalent source and target
sentences are by some means put next to each other, which can be done by using
simple tables. This is done to discover the most obvious changes to the text, such
as elements of meaning being transferred to another sentence in the TT, long
sentences being translated as several short ones, and extensive omissions and
additions. An excerpt of the sentence aligned HP-corpus is shown in table 4.1,
in which the second sentence pair is an example of how the translator has chosen
to translate one sentence as two sentences.

Source sentence Target sentence

He sat up and Hagrid’s heavy Han satte sig upp och Hagrids
coat fell off him. tunga rock föll av honom.
The hut was full of sunlight, the Rucklet var fyllt av solljus, stor-
storm was over, Hagrid himself men var över, Hagrid själv sov
was asleep on the collapsed sofa, på den nersjunkna soffan och det
and there was an owl rapping its var en uggla som knackade med
claw on the window, a newspaper klon på fönstret. I näbben höll
held in its beak. den en tidning.

Table 4.1: An excerpt of the sentence aligned corpus.

For small corpora like the HP-corpus, sentence alignment can be done quite
easily using basic word processing software such as Microsoft Word. For larger
collections of text, automatic tools are necessary.

4.3 Part-of-speech Tagging

Once the sentence alignment is done, the corpus can be classified on a more fine-
grained level. Part-of-speech tagging (henceforth POS-tagging) of the words in
the corpus is one way to proceed that is often used in translation studies. POS-
tagging is done because keeping track of the structural information of words
and other text components is relevant. In translation, words and segments
of a source text will sometimes change word class or have another function
in the target text. The voice can also change, from passive to active or vice
versa. These small linguistic changes can be indicators of more wide-spanning
changes done to the text, which makes them liable for further investigation.
Modern language processing tools such as the Machinese Syntax by Connexor
4.4. WORD ALIGNMENT 21

uses functional dependency grammar (henceforth FDG) to POS-tag corpora

automatically. For a description of FDG, see Tapanainen and Järvinen (1997).

4.4 Word Alignment

To be able to discover when the corresponding word is of a different word class
in the TT than in the ST, the texts must be aligned on the word level. The ST
word (or words) must be linked with the corresponding TT word (or words),
and for this task, specialised software tools are required.
Traditionally, word alignment is done automatically and the performance of
the software that is used is evaluated on both precision and recall. Precision is
“the accurateness of the links relations” and recall is “the number of possible
links that are retrieved” (Sågvall Hein 2002, p. 68). The automated systems
tend to have precision figures ranging from 80 to 95 percent (Merkel, Petterstedt
& Ahrenberg 2003). As for recall, automatic alignment systems do well if the
texts contain only one-to-one correspondences, but have severe difficulties in
identifying multi-word units, “especially those that are discontinuous or have a
low frequency; it is more or less impossible to know exactly how many multi-
word units there are in a text” (Ahrenberg, Merkel, Sågvall Hein & Tiedemann
2000, p. 2). This causes problems for the recall measure, which can “therefore in
practice only be made on samples of a bitext” (ibid., p. 2). Since very few texts
contain only one-to-one correspondences, the performance of automated systems
is simply not good enough if a full investigation of all words and tokens in a
corpus is to be carried out. However, for very large corpora, manual alignment is
not an option because of the workload involved, and in such cases, it is necessary
to use an automated system.

4.4.1 Guidelines for Manual Word Alignment

When aligning a corpus manually, it is important to link the material as con-
sistently as possible, which is difficult to achieve when several annotators work
together on one project (Merkel 1999). But even with just one annotator, it
is important to work with consistency in mind. In my opinion, the task of
achieving consistency becomes more complex the larger the corpus. Not only
specific terms, names and other lexical items need to be consistently aligned,
but also syntactic structures, and remembering exactly how one treated a word
or construction 1000 sentence pairs ago is not always easy. In this sense, the an-
notator’s job is reminiscent of the translators, and the same challenges face both
the one producing the target text, and the one studying that very translation.
The general guidelines used in the annotation of the HP-corpus were via
Merkel (1999) adopted from Véronis:

1. Mark as many words as necessary on both the target and source side.

2. Mark as few words as necessary on both the target and source side.
22 CHAPTER 4. STUDYING TRANSLATIONS

Following the guidelines is supposed to ensure that all links have a two-way
equivalence between the source and target segments.

4.5 Non-1-to-1-operations
When aligning a corpus it becomes evident that some segments of the ST do
not have a one-to-one correspondence with a TT segment, and the annotator
is forced to link together segments in (usually) larger chunks. These non-1-to-
1-operations include additions, deletions, convergences and divergences (Merkel
1999).
The focus of this study is on the segments of both source and target texts that
do not have a corresponding segment in the other language, namely additions
and deletions. These are significant changes to the text made by the translator,
and in the aligning process, they lead to the annotator marking the segments
as NULL-links, i.e. segments without corresponding segments. This does not
apply to divergence and convergence, and they will only be mentioned briefly
below for completeness. All examples below are taken from the Harry Potter
corpus.

Divergence and Convergence

Divergence is when a construction spans more segments in the target text than
in the source text. Remember in the example below corresponds to komma
ihåg, and a one word construction in the source text has become a two word
construction in the target text.

Example:
He rolled onto his back and tried to remember the dream he had been
having.
Han rullade över på rygg och försökte komma ihåg drömmen han hade
haft.

Convergence is the opposite, when the TT equivalent of an ST-expression

spans fewer segments. In this example, the two word construction in the source
text corresponds to the one word construction in the target text.

Example:
At last.
Äntligen.

Divergence and convergence are oftentimes necessary operations that are

motivated by differences between the languages that need to be accommodated
for. Additions and deletions, however, are rarely completely motivated by dif-
ferences in the languages, but rely more heavily on the choices of the particular
translator.
4.5. NON-1-TO-1-OPERATIONS 23

Additions
Translators sometimes add information to the text, and those additions are
elements of the TT that are not present in the ST. The effect an addition
has on a text is to a great extent dependent on the linguistic nature of the
addition. It is reasonable to expect that added verbs, nouns and adjectives add
actual information, where as added pronouns can indicate that the translator
has in fact grammaticised the text. In the ideal case, the translator only makes
additions when it is absolutely necessary. However, this is not always the case,
as can be seen in the example below, where Fries-Gedin has added the equivalent
of long, a piece of information that is not motivated by the meaning of the source
word cloaks.

Example:
People in cloaks.
Folk i långa mantlar.

Deletions
Deletions occur in the aligned material when the translator has chosen not
to include some piece of information from the ST. The effect of a deletion is
usually that the text has been simplified. In the example below, around has
been deleted.

Example:
He looked around at Harry and Hermione.
Han såg på Harry and Hermione.

Should the source sentence contain a deletion and the target sentence an
addition, it can be reasonable to suspect that there might be a relationship
between the two.

Studying Additions and Deletions

Deletions and additions are structural changes that are easily detectable with
the tools and methods used in this study. Thus the interest in these particular
changes is twofold, partly motivated by the ease of structuring and studying
them with the available tools, and partly by the fact that they are rarely com-
pletely necessary operations. Additions and deletions tend to a great extent to
be based on subjective judgements made by the isolated translator, and there-
fore depend heavily on the individual translator.
As a general recommendation for translators, Newmark emphasises the nat-
uralness of the target text (1988). Accuracy, however, is even more important
and “you have no licence to change words that have plain one-to-one transla-
tions just because you think they sound better than the original, though there
24 CHAPTER 4. STUDYING TRANSLATIONS

is nothing wrong with it” (ibid., p. 36). Specifically, “mind particularly your
descriptive words: adjectives, adverbs, nouns and verbs of quality” (ibid., p.
36). Consequently, the use a translator makes of adding or deleting descriptive
words and segments to or from the text can be seen as a part of his or her style
of translating, and will be the focus of the investigation into how Fries-Gedin
uses addition and deletion in the samples.

4.6 Lexical Shifts

In translations, the meaning of some segments is sometimes changed between
the source and target texts. These lexical shifts can be of three different types,
according to Merkel (1999). The translated lexical item can be:

1. less specific, i.e. more general, than the source item.

2. more specific than the source item.

3. neither less nor more specific and not equivalent, i.e. it has a different
meaning than the source item.

These definitions can also be termed a less specific shift, a more specific
shift, and an other lexical shift (ibid.). Examples of the different types of lexical
shifts are given below. The bold faced words are the source item and its cho-
sen translation. Gloss translations of the actual meaning of the chosen target
segments are given in the square brackets in the English sentences, illustrating
the lexical shifts (whelk has in Swedish been generalised into [seafood], it has
been specified as [The stench], and darkly has been changed into [quietly]).

Example of a less specific lexical shift:

Ate a funny whelk [seafood].
Åt nåt konstigt skaldjur.

Example of a more specific lexical shift:

It [The stench] seemed to be coming from a large metal tub in the sink.
Stanken verkade komma från en stor plåtbalja i diskhon.

Example of an other lexical shift:

The giant chuckled darkly [quietly].
Jätten skrockade tyst.

Like additions and deletions, lexical shifts are significant changes made to
the text, and they are rarely necessary to make. Consequently, analysing trans-
lations in terms of lexical shifts can illustrate the influence of the translator on
the text.
4.7. PARAPHRASING AND LEXICAL CHOICE 25

4.6.1 Strategy for Lexical Shifts

In the word alignment system used in this study, it is not possible to mark
segments where lexical shifts have occurred as lexical shifts. As a result, the
choice is either to accept lexical shifts as regular translations, or to mark the
segments as additions and deletions.
In this study, where the influence of the translator is measured in significant
changes made to the text, it is important to solve the dilemma of how to mark
lexical shifts. Some lexical shifts are perhaps necessary to make, due to differ-
ences in vocabulary between the source and target languages. Such necessary
shifts do not depend as heavily on the choices of the particular translator, and
can in this study therefore be linked as regular translations.
For the lexical shifts that the translator has made voluntarily, my solution
is to focus on the degree of change each specific lexical shift implies. If a small
change has been made, like when a pronoun has been changed into the noun it
refers to, as in the example showing a more specific lexical shift on the opposite
page, I have chosen to somewhat reluctantly accept the segments as a regular
translation. This is because although these lexical shifts do imply that the
meaning of the segment has been changed voluntarily, they do not change the
meaning of the reference, they only make it more explicit. Above all, they are
not as significant changes as additions and deletions.
Where the target segment is farther removed from the meaning of the source
segment, however, as for most lexical shifts, I have opted to mark these segments
as additions and deletions. This includes the examples for less specific lexical
shift and other lexical shift on the opposite page.
The advantage of the chosen strategy is that it at least makes small changes
distinguishable from significant changes, that depend more heavily on the choices
of the translator. The disadvantage is that smaller more and less specific lexical
shifts cannot be distinguished from regular translations, and more significant
more and less specific lexical shifts, as well as other lexical shifts, cannot be dis-
tinguished from additions and deletions. The implications of this will be further
dealt with in section 8.1.2 in the discussion chapter.

4.7 Paraphrasing and Lexical Choice

When aligning a corpus, the passages that are the most problematic tend to
be those that paraphrase the meaning of the source words. It is very difficult
indeed to draw a line between what is a working paraphrase and what is too far
from the original sentence to be accepted as a natural and accurate translation.
Paraphrases sometimes border on errors in lexical choice, and it is not always
easy to determine whether the translator has made a mistake or not. In such
cases, the annotator must trust his or her own resources, both in the form of
personal knowledge about a word, concept or activity, as well as dictionaries
and other sources of linguistic information. The annotator must, in the end,
make a choice and either accept or reject the choice of the translator.
26 CHAPTER 4. STUDYING TRANSLATIONS

4.7.1 Examples of Rejected Lexical Choices in the HP-

corpus
The focus of this study is on patterns that differ between the source and target
texts, but in order to briefly explain my strategy in the alignment process, a few
examples of dubious lexical choices and how I chose to treat them are needed.
In any case where I was reluctant to accept the choice made by the translator, I
consulted one or more dictionaries, both English/Swedish and English/English.
One example of a lexical choice I rejected was the choice the translator had
made for sherbet lemons, the Muggle sweet that Albus Dumbledore eats in the
first chapter of the first Harry Potter book, Harry Potter and the Philosopher’s
Stone (Rowling 1998). This is translated as citronisglassar, and the literal
translation in English of this is lemon ice lollies. This particular lexical choice
is not equivalent to the source segment, i.e. it is an instance of an other lexical
shift. Furthermore, it is semantically impossible even within the context of the
story, as Dumbledore explicitly states that they are a Muggle sweet kept in a
bag in his pocket, which is an impossible way to store an ice lolly. I chose to
not accept this link and thus treated the former as a deletion and the latter as
an addition.
One other recurring lexical choice that was treated as a deletion/addition
pair was don’t ask questions and its chosen translation, kom inte med några
frågor. This is because kom inte med appears dated and is not common Swedish
usage, whereas the English equivalent is common usage in the source language.
Consequently, kom inte med was marked as an addition, and don’t ask as a
deletion.
Chapter 5

Methodology

This chapter outlines how the HP-project was carried out, and describes the
specialised software tools that were used in the process. In addition, some
advantages of using these new types of alignment tools are explained.

5.1 The Sequence of Work

The sequence of work in this study can be summarised as follows.
1. The texts to be included in the corpus were chosen and read, and a decision
was made on a suitable size of the samples.
2. The sample texts were transferred to electronic form, in this case by scan-
ning, and the texts were proof-read.
3. The samples were aligned manually on the sentence level.
4. Machinese Syntax by Connexor was used to supply part-of-speech-tags to
all tokens in the samples.
5. The POS-tagged samples were word aligned using two different tools,
I*Link and I*Trix. The two tools were combined in four different strate-
gies. Each strategy was used for one sample only, in order to enable an
evaluation of the chosen tools and strategies.
6. The word aligned samples were studied using LinkInspector and LinkRe-
porter, tools included in I*Link, and the results were analysed.
7. A small scale case study was performed on the treatment of the dialects
of the characters Rubeus Hagrid and Stan Shunpike.
8. A close investigation of the last 150 sentence pairs of HP4 was made in
order to investigate possible relationships between additions and deletions
and to search for manifestations of the translation universals in more de-
tail.

27
28 CHAPTER 5. METHODOLOGY

5.2 A Presentation of the Tools

In the word alignment, two different tools were used, I*Link and I*Trix. Both
were developed at NLPLAB, the natural language processing division of the
computer department of Linköping University.

5.2.1 I*Link
The word alignment system used in this study, I*Link, is interactive in that it is
used in collaboration with a human annotator in order to increase the efficiency
and performance of the tool. In collaboration with a human annotator, the
precision figure of I*Link is more or less 100 percent, which is necessary in this
study. In order to study the entire samples and search for patterns, the entire
samples including complex structures that are sometimes very difficult to align
must be as fully aligned as possible.
I*Link is a semi-manual alignment tool that uses information from bilin-
gual resources and built-in heuristics to suggest correspondence candidates for
alignment, which the user accepts, revises or rejects (Merkel et al. 2003). Any
element the tool cannot suggest a match for, the user chooses a match for man-
ually by clicking on the matching word, should one exist, and then presses the
“Match”-button. If no matching word exists, the user marks the element as
a NULL-link. I*Link uses machine learning techniques to store the choices of
the user in dynamic resources that are built during and used directly in the
linking process. Thus “the accuracy of the proposed word links is continuously
improved during and across word alignment sessions, which in turn means in-
creased efficiency” (ibid., p. 2). This is, however, dependent on the ability of the
user to be consistent in his or her chosen links. If the choices are inconsistent,
it will harm the learning effect and I*Link will not perform optimally.
In addition to the built-in resources, I*Link can be fed with user-specific dy-
namic resources. If the user has worked with the tool previously, the resources
collected from those sessions can be used as an additional knowledge base for
the system, which should enhance the performance of the system. I*Link auto-
matically collects statistical data on the performed translational actions.

The Graphical Interface of I*Link

The graphical interface of I*Link consists of four windows: the Link Panel, the
Link Table Panel, the Resource Panel and the Settings Panel. The Link Panel
in figure 5.1 is the window in which the current sentence pair is presented, the
source sentence in the upper half and the target sentence in the lower half. It is
in this window that the user can accept or reject the automatic proposals and
select links manually. Chosen links are marked using corresponding colours, and
are also shown in the Link Table Panel in figure 5.2. Additions and deletions
can be marked as NULL-links by right-clicking with the mouse on the word or
words, and choosing NULL.
5.2. A PRESENTATION OF THE TOOLS 29

Figure 5.1: Screenshot of the Link Panel in I*Link.

Figure 5.2: Screenshot of the Link Table Panel in I*Link.

30 CHAPTER 5. METHODOLOGY

In the centre of the Link Panel, directly below the windows where the source
and target sentences are shown, some important pieces of information are dis-
played. The box in the middle contains the number of the current sentence
pair, in this case number 1258. The green pieces of text on both sides of this
box says “Source completed” and “Target completed” when both sentences are
fully aligned and the “Done”-button is pressed. This is significant since the
advantage of this system is that full and complete alignment can be achieved,
and it is thus important to be able to verify that all tokens in each sentence
have been aligned before moving on to the next sentence.
The eight fields in the lower left corner of the Link Panel window show
linguistic data on the current link on four levels: word form, base form, POS
and the function the word or words have in the sentence.
The Resource Panel and the Settings Panel were not used actively in this
project. Descriptions of these panels are available in Merkel et al. (2003).

Tools Included in the I*Link System

I*Link also features two possibilities to search the corpus material, in the rather
similar tools LinkInspector and LinkReporter. Both can be used to search for,
among other things, occurrences of different word classes, constructions, words,
aligned pairs and added and deleted elements.

5.2.2 I*Trix
Another word alignment tool that was used in the study is I*Trix, which differs
from I*Link by being a tool with which fully automatic alignment can be done.
The sample to be aligned is run through I*Trix, which links whatever it can
in the sample. The output can then be manually post-edited in I*Link by the
user, in order to correct mistakes and achieve a complete alignment where all
tokens in the sample are aligned. Like I*Link, I*Trix can be fed with user-
specific resources built up in previous sessions using I*Link in order to enhance
the performance of the tool.

5.2.3 New Tools, New Possibilities

The big difference between the tools used in this project and traditional word
alignment tools is the possibility for interaction between I*Link and I*Trix. Tra-
ditional tools tend to include only the automatic part, corresponding to I*Trix.
With I*Link, it is possible to align samples manually or semi-manually, thereby
creating user-specific resources that in turn can be used to train either I*Link
or I*Trix. This training will increase the performance of the tools. Traditional
tools are usually not possible to train, meaning that the researcher cannot affect
the performance of the tool or the quality of the output.
Using I*Link and I*Trix in combination represents a new way of studying
translations, and they comprise a more powerful resource in comparison with
traditional corpus tools. However, the fact that this is a new set of tools means
5.2. A PRESENTATION OF THE TOOLS 31

that the old framework for analysis is less useful, which entails that this study
differs somewhat from traditional studies.
Traditionally, other measurements were used, such as type-token ratio and
lexical density (Baker 1996). The main purpose of these measures is to investi-
gate translation in a broader perspective and to describe general principles that
can be found in translations. In contrast, the tools used in this study makes
it possible to systematically analyse particular translations in a more power-
ful way than was possible with traditional tools. Consequently, the methods
used in this study are not suitable for investigating translations in general, but
are very well suited for making a more thorough investigation of one or more
translations.
32 CHAPTER 5. METHODOLOGY
Chapter 6

The Making of the

HP-corpus

In this chapter, the building and aligning of the HP-corpus are described. The
four different alignment processes are described and discussed in some detail, so
as to explain how the different strategies affect the process.

6.1 The Corpus

The parallel translational corpus produced and analysed in this project consists
of samples of the first four Harry Potter books in English, and their respective
translations in Swedish (Rowling 1998, Rowling 2000b, Rowling 1999, Rowling
2000c, Rowling 2000a, Rowling 2001b, Rowling 2001a, Rowling 2002). The
relationship between each book and the corresponding sample can be seen in
table 6.1.

Book Sample name

Harry Potter and the Philosopher’s Stone HP1
Harry Potter and the Chamber of Secrets HP2
Harry Potter and the Prisoner of Azkaban HP3
Harry Potter and the Goblet of Fire HP4

Table 6.1: The names of the books the samples are taken from, and the names
of the corresponding samples.

Samples of the first 20000 words in each ST were chosen, rounded to the
nearest chapter. There were several reasons as to why only whole chapters were
used in the samples, the perhaps most important one being that in order to
study the translations contrastively, the semantic integrity of the texts needed
to be preserved. This was also why the samples all contain the beginnings of

33
34 CHAPTER 6. THE MAKING OF THE HP-CORPUS

the four books, as it was deemed more difficult to track the translator’s change
unless the same part of the different books were being studied. In addition,
the beginning and ending of chapters have specific characteristics. The extent
of the resulting samples in the number of tokens and chapters included can be
seen in table 6.2. The total token count in the corpus is 189116 tokens.

Sample Sample size ST Sample size TT Chapters

HP1 26121 26642 5
HP2 24780 25359 5
HP3 20036 20671 4
HP4 22546 22961 6

Table 6.2: The respective sizes (in number of tokens) of the samples in the
HP-corpus, and the number of chapters in each sample.

The selected samples were transferred to electronic versions by scanning,

proofread, and sentence aligned manually using Microsoft Word. This manual
process enabled the author to form an initial idea of what translational phenom-
ena might be interesting to investigate further. The production of the sentence
aligned corpus required 7 weeks. The corpus contains in total 5816 sentence
pairs, and how these are divided over the four samples can be seen in table 6.3.

Sample Number of sentence pairs

HP1 1768
HP2 1545
HP3 1233
HP4 1270

Table 6.3: The number of sentence pairs in the samples.

After sentence aligning the samples, they were part-of-speech-tagged auto-

matically using Machinese Syntax by Connexor.

6.2 Word Aligning the Corpus

The POS-tagged data was used as input to I*Link and I*Trix, in which the
word alignment took place using different strategies for each sample. All four
samples were aligned sequentially, starting with the first sentence of the first
chapter, and finishing with the last sentence in the sample. Table 6.4 shows a
summary of the different strategies and resources used.
Before starting the alignment of HP1, I spent 4 hours learning how to use
I*Link. This was necessary because otherwise, the aligning of HP1 would take
an unproportional amount of time compared to the other samples, because too
much time would have been devoted to understanding the system and how to use
6.2. WORD ALIGNING THE CORPUS 35

Sample Tool(s) used Resources

HP1 I*Link Built-in
HP2 I*Trix, I*Link Built-in
HP3 I*Trix, I*Link Dynamic from HP1 and 2
HP4 I*Link Dynamic from HP1, 2 and 3

Table 6.4: The different strategies used in the alignment of the samples.

it. In addition, trying to learn the system before starting to align the samples
was, of course, positive both regarding the attempt to be consistent and the
over-all quality of the chosen links.

6.2.1 Aligning HP1

The strategy for the first sample was to semi-manually align HP1 in sequence,
using only the built-in resources of I*Link. The system suggested links, which
were continuously accepted or rejected. The alignment of HP1 took 18 hours to
complete1 .
When I first started aligning HP1, it became clear that I was not entirely
familiar with the system, and I was very preoccupied with trying to be consis-
tent, although that failed somewhat. In retrospect, more time should have been
spent familiarising myself with the system.
Unfortunately, two sentence pairs were flawed in HP1, namely 1042 and
1043. The base of the problem is that 1042 was completely deleted, i.e. had no
corresponding sentence in the target text, which means that the opposite cell in
the sentence aligned corpus was marked &&&NO CORRESPONDENCE&&&.
This was done in the corresponding empty cell of all completely added and
deleted sentences in order to prevent the empty cells from disappearing in the
POS-tagging. For some unknown reason, this failed to work in this instance, and
in the output from Machinese Syntax the first sentence of 1043 had migrated
to 1042. This meant that neither of these sentence pairs were possible to link.
However, the sample includes 1768 sentence pairs, and though it is unfortunate
that two sentence pairs were marred, it is in no way catastrophic as they were
such a minute part of the sample. Therefore, the fact that the two sentences
were left unaligned has been disregarded, and they have not been subtracted
from the corpus, neither in the number of sentence pairs, nor in the number of
tokens.
It also became evident during the aligning of HP1 that despite the meticulous
proof-reading of the scanned texts, some small imperfections prevailed in the
samples. These were few and far between, however, and any effect they might
have had would have been marginal. They were therefore not corrected, as it
was judged that correcting them might simply take too much time in proportion
1 This number does not include the 3 hours required for the subsequent post-editing of

HP1, see section 6.4.

36 CHAPTER 6. THE MAKING OF THE HP-CORPUS

to the marginal effect it would produce. Such imperfections also exist in the
other samples, and were treated in the same way.

6.2.2 Aligning HP2

HP2 was automatically aligned using the built-in resources of I*Trix, and the
data from this second session was post-edited using I*Link. The alignment of
HP2 required 16.5 hours.
The difference in the strategies used for HP1 and HP2 had some interesting
implications. When only using the buttons Match, Accept, Reject and Done in
I*Link, as in HP1, the links are presented in turn, and with enough pushing
of the buttons, either a match can be found or the word is treated as a NULL
link. This means that the links get coloured one by one, and it is therefore easy
for the annotator to keep track of the segments that are linked together. HP2,
however, was aligned automatically using I*Trix, which means that in the post-
editing session using I*Link, the links already matched by I*Trix were already
coloured. Sometimes, words that stand next to each other in the sentences but
are not a part of the same link can have very similar colours. This means that
there is a risk of accepting links made by I*Trix that should not be accepted,
because the eye does not pick up on the slight difference between the colour of
the matched links. To me, this meant that in aligning HP2, I had to be very
careful, and quite a few sentence pairs were aligned before I discovered this, and
so the sentences that I aligned while still ignorant of this had to be post-edited.
This was done in the same session, as soon as it was discovered, and the required
time is included in the 16.5 hours it took to align the whole sample.
Another consequence of the colour scheme was that in linking HP2, I learned
to use the Link Table Panel. In this window, all linked pairs appear after the
Accept-button is pressed, and it is possible to check the links as you go along. In
HP1, I did not use this as I did not need it, but for the samples pre-aligned with
I*Trix, it became indispensable as it diminished the problem with neighbouring
links of almost the same colour.
One idea that occurred to me after aligning the first 350 sentence pairs
of HP2 was that if one wanted to use the same guidelines that the heuristics
of I*Link and I*Trix are based on, it might have been better to start with
the strategy that was used for HP2, i.e. using I*Trix to pre-align HP1. The
matches in the system’s output quite often differed from my own choices for
matches, and this made me start to doubt my reasons for not using the exact
same heuristics as the system. My theory is that using the same guidelines
and starting with a pre-aligned text might have ensured the consistency of the
chosen links, as it is much easier to just accept what the system suggests and
simply correct the mistakes and link what the system has not been able to link.
For the unexperienced annotator, this could be used as a way to simplify the
process and ensure a greater consistency.
6.3. COMMENTS ON THE ALIGNMENT PROCESS 37

6.2.3 Aligning HP3

The dynamic resources created during the first two sessions were used as ad-
ditional learning material for I*Trix in the third strategy. Using the created
resources from HP1 and HP2 as input data to I*Trix, HP3 was automatically
aligned and then revised in I*Link. The third strategy is thus basically the same
as the second, apart from the fact that the dynamic user-specific resources were
used. Aligning HP3 took 10.5 hours.
As was the case with HP1, some sentence pairs in HP3 are flawed. In pair
19, one sentence on the target side had somehow disappeared already in the
scanning, which was unfortunately not discovered in the proofreading of the
text. In pair 199, the sentence on the target side is repeated. In pairs 343
and 854 spelling mistakes that were present in the actual printed texts were
marked with a [sic!] in the proofreading of the scanned texts, and they were
unfortunately not removed before the texts were POS-tagged. Both [sic!]s were
marked as NULL links. As with the flawed sentences in HP1, these sentences
and tokens were not removed from the corpus, but left disregarded.
Although the dynamic resources were used in this session, I*Trix built-in
heuristics still overrode the dynamic resources in some cases. For example,
following guideline number two in marking as few words as possible in each
link, I chose to link name and name, surname and surname, i.e. Harry was
linked with Harry, and Potter with Potter. I*Trix aligns name and surname as
one link, Harry Potter with Harry Potter, and this was retained in HP3.

6.2.4 Aligning HP4

The dynamic resources from the first three sessions were fed to I*Link, which
was used to semi-manually align HP4 using the Match button in I*Link. In
other words, this is the same strategy as for HP1, with the exception that the
dynamic resources from the earlier sessions were used. The alignment of HP4
required 16 hours.
No sentences were flawed in HP4, and the used strategy did not pose any
significant problems. However, in the process of aligning HP4, I got the impres-
sion that there were more differences between the source and target texts in this
sample than in the previous ones.

6.3 Comments on the Alignment Process

The time it takes to align different parts of a sample is heavily dependent on
the nature of the sentences included in that segment. If there are many short
sentences, I*Link needs to do little work and few calculations and the aligning
takes little time. If there are longer, more complex sentences and sentence
pairs, there is a delay in the system and the aligning takes longer time, simply
because there are more possible matches for every token. This is of course also
dependent on the platform used, i.e. the capacity of the PC. In aligning the
HP-corpus, two different PCs with different capacities were used. To balance
38 CHAPTER 6. THE MAKING OF THE HP-CORPUS

the effect of the PC-capacity, a simple time-test was done on the PCs, and the
summary of the time required for aligning the different samples in table 6.5 has
been modified to accommodate for that difference.

Sample Tool(s) used Resources Required time

HP1 I*Link Built-in 18 hours
HP2 I*Trix, I*Link Built-in 16.5 hours
HP3 I*Trix, I*Link Dynamic from HP1 and 2 10.5 hours
HP4 I*Link Dynamic from HP1, 2 and 3 16 hours

Table 6.5: The different strategies and the time required for aligning each sam-
ple.

6.3.1 Problems Common to the Samples

During the POS-tagging of the sentence aligned texts using Machinese Syntax,
almost all of the quotation marks that surround dialogue on the source side
disappeared, while those on the target side remained. This was unfortunate,
but as there generally is a very high correspondence between the source and
target use of quotation marks, it was deemed irrelevant. Furthermore, quotation
marks are of no particular interest in this study, so all instances on both source
and target sides were simply subtracted from the token-count.
Another problem was the frequent use of the three periods in sequence (“...”)
construction on the target side, in most cases corresponding to the use of a
dash (“-”) on the source side. This construction was also damaged in the POS-
tagging, which proved to be a little more problematic as the use of periods is
of interest to the focus of this study. The reason behind this problem is, in all
likelihood, that when the files are saved in text format, these constructions are
not converted properly.
In addition, there are as already mentioned some small problems with imper-
fections in the form of misspellings etc. that were overlooked in the proofreading
and has since prevailed in the samples. All tokens with such minor imperfections
have been left in the corpus and disregarded in the analysis, because of the very
limited effect they might have on the results in proportion to how much time it
would take to correct them.

6.4 Post-editing HP1

When all samples were aligned, it was possible to reflect on the consistency of
the chosen links. It was discovered that in the beginning of HP1, the alignment
was inconsistent regarding the aligning of names, i.e. how name and surname
were treated. The chosen strategy for constructions like for example Harry
Potter, was to link Harry in the ST with Harry in the TT, and Potter with
Potter in the same manner. This, however, had not been done consistently in
6.4. POST-EDITING HP1 39

the beginning of HP1, probably due to the fact that a heuristic used in I*Link
always suggests that such constructions should be aligned as one link, not two.
At the very start of the aligning I was apparently too preoccupied with handling
the system to notice that I was not following my own guidelines. Because of
this, HP1 was post-edited in order to make the links conform to the patterns
used in HP2, 3 and 4. The post-editing session, in which all 1768 sentence pairs
were checked and mistakes corrected, required 3 hours.
40 CHAPTER 6. THE MAKING OF THE HP-CORPUS
Chapter 7

Results

The first part of this chapter briefly states that the HP-corpus is a result in
itself. The second part describes the translational results of the analysis of the
corpus and some complications that occurred during the analysis. I have chosen
to give a more detailed description of the analysis because there is no ready-to-
use framework for studies like this, as mentioned in the introduction. The third
part of the chapter presents the methodological results that concern the tools
and strategies used in the project.

7.1 The HP-Corpus

The most tangible result of this study is the HP-corpus in itself. Fully aligned
on a word level, this corpus of 189116 tokens in 5816 sentence pairs is ready to
be used in other descriptive translation studies.

7.2 Translational Results

The translational results are divided in three main sections. The first section
presents the results of the investigation into additions and deletions, and ex-
plains the consequences of the somewhat coarse FDG-tagging. Furthermore,
one subsection is devoted to showing examples of additions and deletions in
order to illustrate the effect these operations have on the actual texts. This is
followed by the results of a closer investigation of the last 150 sentence pairs of
HP4, giving a description of how additions, deletions and translation universals
can be manifested if studied in more detail.
The second section contains the results concerning the translation universals.
Manifestations of explicitation, normalisation and simplification are presented,
both on a lexical level, and on a more general, pattern-oriented syntactical level.
The third section describes the effect of some translational choices made
by the translator. Lexical choice is investigated, and one specific translation
difficulty in the Harry Potter domain is used to exemplify a translational choice.

41
42 CHAPTER 7. RESULTS

Finally, a way of organising and presenting the data obtained in this study using
semantic mirroring is presented.

7.2.1 Additions and Deletions

The starting point for this analysis is the advice to translators from Peter New-
mark, cited in section 4.5, to be careful not make unnecessary changes to the
text, and “mind particularly your descriptive words: adjectives, adverbs, nouns
and verbs of quality” (Newmark, p. 36). Consequently, addition and deletion
were studied for verbs, adjectives, adverbs, nouns and pronouns. The last cat-
egory was included because addition of pronouns can be a sign of explicitation
and normalisation.

Analysing Verbs and Adjectives

As already mentioned, one of the most important advantages with I*Link and
similar alignment tools is the possibility to distinguish additions and deletions.
However, at the start of the analysis process, a problem was discovered con-
cerning the analysis of verbs and adjectives. The basis of the problem is that
not only the tokens tagged by Machinese Syntax as verbs, V, and adjectives, A,
are verbs and adjectives. The parts of speech tagged EN and ING in the source
samples, and AD and NDE in the target samples can have both adjective and
verb function, and the functional dependency grammar (FDG) that was used
in the POS-tagging does not distinguish between the two different uses.
In English, ING is either the present participle that has an adjectival func-
tion, or the gerund form of a verb, as in sentence pair 1503 of HP1: “They
were going even deeper now and gathering speed”. EN is the past participle as
used in sentence pair 19 of HP1: “Little tyke, chortled Mr Dursley as he left
the house”. In Swedish, the problem is manifested in the words tagged AD and
NDE. AD is a participle form, such as in brutit in sentence pair 290 of HP1:
“Ett lågt mullrande ljud hade brutit tystnaden runt dem”. An example of a
NDE is mullrande in the same sentence. The corresponding source sentence
reads: “A low rumbling sound had broken the silence around them”. Brutit in
the equivalent of broken and mullrande corresponds to rumbling.
When analysing the corpus in terms of what has happened to verbs and
adjectives, the words marked EN, ING, AD and NDE must in some way be
accounted for. One way to do that is to manually go through and analyse every
single token with such a mark to ascertain if it has verb or adjective function.
However, I opted not to do so, since it would take an unproportional amount
of time. Instead, a brief examination of approximately 100 sentence pairs was
made, and the result of this was that the majority of these particles functioned
as verbs.
Because of the somewhat uncertain classification of these particles, the deleted
and added verbs are represented by two columns in tables 7.1 and 7.2. The first
verb column, V, contains only the added or deleted tokens tagged V for verb.
The second, V2, contains the same added or deleted tokens tagged V, plus the
7.2. TRANSLATIONAL RESULTS 43

EN, ING, AD and NDE marked tokens. In table 7.1, V2 thus contains nulled
tokens in the TTs marked V, AD and NDE, because all additions are, naturally,
made in Swedish and cannot be ING or EN. Consequently, V2 in table 7.2 thus
includes nulled tokens from the STs tagged V, EN or ING.

Results for Additions and Deletions

The percentage of additions and deletions can be seen in table 7.1 and table 7.2.
As is evident in the tables, there is a large and steady increase in the number of
both additions and deletions made in the four different samples. The tendency is
that HP1 has the lowest numbers of additions and deletions and HP4 the highest
numbers, while HP2 and HP3 fall in between. In other words, the numbers grow
sequentially and consistently, which is true for all the investigated word classes
and both possibilities for analysing verbs.

Sample V1 V2 Adjectives Adverbs Nouns Pronouns

HP1 6.2 5.3 6.3 15.7 3.7 10.5
HP2 9.4 7.8 9.3 23.2 4.8 15.2
HP3 12.4 10.5 9.6 22.2 6.3 16.2
HP4 16.0 13.4 16.9 25.3 7.1 20.5

Table 7.1: The percentage of additions in the samples. For each investigated
word class, the number shown is in relation to the total amount of words of that
word class in the source texts.

Sample V1 V2 Adjectives Adverbs Nouns Pronouns

HP1 3.2 2.8 2.1 3.7 1.6 3.1
HP2 4.9 4.3 4.1 7.7 2.6 5.5
HP3 5.9 5.3 3.8 7.7 2.7 5.0
HP4 9.6 8.0 6.9 14.4 4.1 9.3

Table 7.2: The percentage of deletions in the samples. For each investigated
word class, the number shown is in relation to the total amount of words of that
word class in the source texts.

The results are not homogeneous, however, as there are differences in the
distributions of additions and deletions. Figure 7.1 and figure 7.2 are included
to show the results in a directly perceptible way 1 . In studying these, it is
obvious that there are few deletions for all the word classes in HP1, but for
additions, the results have a much greater range for the different word classes.
1 The lines in the figures are the straight lines that minimise the sum of the squared distances

between each line and its four corresponding dots. The purpose of the lines is to visualise
trends in the data, mainly to illustrate the sequential differences between the samples. The
underlying data is displayed in tables 7.1 and 7.2.
44 CHAPTER 7. RESULTS

30%
+
Pronoun
⊕ ×
25% Noun
⊕
⊕ ⊕
Adverb
+
20% ♦
Adjective
♦ ◊
⊕ + ∆ Verb 2
+
15% ∆
◊ Verb
∆
+ ◊
10% ∆
♦ ♦
◊
×
∆
♦ ×
◊ ×
5%
×

0%
HP1 HP2 HP3 HP4

Figure 7.1: Plot showing the percentage of additions.

Despite this difference in distribution of additions and deletions in HP1, the

tendency of steady and sequential increase in the percentages from HP1 to HP4
is the same for both additions and deletions. For both addition and deletion,
the noun category has the lowest percentage of change, and the adverb category
the highest.
The implications of the results evident in the tables and figures is that there
have been significant changes made to the texts in terms of additions and dele-
tions, and the tendency of Fries-Gedin to alter elements of text has increased
over time.

Additions and Deletions in Context

Although the percentages of additions and deletions clearly show that there are
differences between the source and target versions of the Harry Potter books,
they do not say anything about the actual effect of the additions and deletions
in the texts. As an illustration of what additions and deletions can look like,
one example from each investigated word category is shown in table 7.3 for
additions and table 7.4 for deletions. These are basic additions and deletions
that have no corresponding element in the corresponding sentence at all. This
7.2. TRANSLATIONAL RESULTS 45

30%
+
Pronoun
×
25% Noun
⊕
Adverb
♦
20% Adjective
◊
Verb 2
⊕ ∆
15% Verb

10% ∆
+
⊕ ⊕ ◊
♦
+ ∆
◊
∆
◊ +
5% ⊕ ♦ ♦ ×
∆
+
◊ ×
♦ ×
×
0%
HP1 HP2 HP3 HP4

Figure 7.2: Plot showing the percentage of deletions.

type of addition or deletion tend to be manifested in small pieces of information

that have been added or omitted.
But oftentimes both addition and deletion are present in one sentence pair.
In some such cases, there is no semantic correspondence at all between the added
and deleted elements, as in pair 797 in HP4. The source sentence reads: “A bag
of sweets had spilled out of Fred’s pocket and the contents were now rolling in
every direction - big, fat toffees in brightly coloured wrappers”. In the Swedish
sentence, fat has been omitted, but the Swedish blanka, equivalent of shiny, has
been added to the description of the toffee wrappers: “En godispåse hade trillat
ut ur Freds ficka, och innehållet rullade nu åt alla håll - stora kolor i blanka,
färgglada papper”.
However, addition and deletion are also present in those sentence pairs where
there is an element in the corresponding sentence that vaguely corresponds to
the added or deleted element, but is not a direct equivalent. These are instances
where a closer translation is possible than the option Fries-Gedin has chosen,
meaning that a lexical shift has been made voluntarily. In such sentence pairs,
the affected words are marked as deletions in the source sentence and additions
in the target sentence. One example of this is sentence pair 1108 in HP4. The
46 CHAPTER 7. RESULTS

Word class Source sentence Target sentence

Verb [Are] All these yours, Är alla de här dina,
Arthur? Arthur?
Adjective Who did he know who sent Vem kände han som skick-
letters by the [regular] post- ade brev med en vanlig
man? brevbärare?
Adverb Empty your pockets [imme- Töm genast fickorna! Sätt
diately], go on, both of you! igång, båda två!
Noun War turned him funny [in Kriget gjorde honom kon-
the head], if you ask me, stig i huvudet, om ni
said the landlord. vill veta vad jag tror, sa
krogvärden.
Pronoun [They were] Lying there De låg där med ögonen
with their eyes wide open! vidöppna!

Table 7.3: A representation of additions for the different word classes. The
added information is shown in bold face in the target sentence column. A gloss
translation of the information that has been added is given in the square brackets
in the source sentence column.

Word class Source sentence Target sentence

Verb Can you take this to Sirius Vill du ta med dig det här
for me? he said, picking up till Sirius åt mig? Han tog
his letter. upp sitt brev.
Adjective It took Harry several days Det tog flera dagar för
to get used to his strange Harry att vänja sig vid sin
new freedom. nya frihet.
Adverb The sky lightened very Himlen ljusnade långsamt
slowly as they made their medan de gick genom byn,
way through the village, its den bläcksvarta färgen
inky blackness diluting to späddes ut till djupblått.
deepest blue.
Noun There came the chink of a Det kom ett klirr från en
bottle being put down upon flaska som sattes ner på
some hard surface, and then någon hård yta och sedan
the dull scraping noise of ett lågt skrapande från en
a heavy chair being dragged tung stol som släpades över
across the floor. golvet.
Pronoun Don’t you lie to me! Ljug inte för mig!

Table 7.4: A representation of deletions for the different word classes. The
deleted information is shown in bold face in the source sentence column.
7.2. TRANSLATIONAL RESULTS 47

source sentence reads: “He was wearing what appeared to be a golfing jumper
and a very old pair of jeans, slightly too big for him and held up with a thick
leather belt”. The target is: “Han var iförd något som såg ut som en golftröja
och ett par slitna jeans, som var lite för stora för honom och därför hölls uppe
av ett tjockt läderbälte”. Here, very old, which should be mycket gamla in
Swedish, has instead become slitna, the equivalent of worn. In this example, it
is evident that the Swedish and English constructions are not equivalent as closer
translations are possible. However, there is still some degree of correspondence
between the deleted and added elements, as the meaning of the Swedish word
at least has some kind of semantic relationship with the meaning of the source
words. Consequently, this combination of addition and deletion is in fact a
lexical shift.

A Close Investigation of Additions and Deletions

In order to search for possible patterns of change that could not be detected with
the search tools in I*Link, the last 150 sentence pairs of HP4 were subjected to
a closer scrutiny. These sentence pairs were chosen because they are the most
recently translated part of the HP-corpus, and were deemed most indicative of
what the translations will be like in later parts of Harry Potter and the Goblet
of Fire (Rowling 2001a), as well as in later Harry Potter-books not included in
the corpus. The close investigation had two main purposes; to try to establish
how common it is that additions and deletions in the same sentence pair are
related, and to search for translation universals in more detail.
Bearing in mind that combinations of addition and deletion can indicate a
lexical shift, the subsample was thoroughly investigated manually, and all addi-
tions and deletions, as well as their POS-tags, were recorded for further analysis.
Table 7.5 shows the distribution of additions and deletions on a general surface
level, i.e. how many sentences contained only an addition or a deletion, both
addition and deletion, or neither. The numbers show that if this subsample of
HP4 is representative of the translational style of Fries-Gedin, about 45 per-
cent of the sentences in her current translating style have both additions and
deletions in them, whereas 26 percent are translated without any addition or
deletion being made to them. This cannot, naturally, be generalised to the rest
of the corpus, but nevertheless gives an idea of the distribution of change in the
translated material.

Only add. Only del. Both add. and del. Neither add. nor del.
25 18 68 39

Table 7.5: Results of the close investigation of the last 150 sentence pairs of
HP4.

However, this surface relationship does not say anything about whether or
not there is a relationship between the additions and deletions in the sentence
pairs that contain both. It would be reasonable to expect that if there is a
48 CHAPTER 7. RESULTS

relationship between an addition and a deletion, the words involved will often be
of the same word class. A detailed presentation of the distribution of additions
and deletions for each word class is given in table 7.6.

Word class Add. Del. Both add. and del. Neither add. nor del.
Verb 45 32 24 97
Adjective 19 8 3 126
Adverb 37 24 11 100
Noun 26 8 3 119
Pronoun 43 27 12 92

Table 7.6: Distribution of additions and deletions for the investigated word
classes in the subsample.

On the surface, it appears that there is a higher degree of correlation between

added and deleted verbs than between additions and deletions in the other word
classes. Because of this, the verb category was further investigated. In the
sentence pairs where both additions and deletions were made, a general pattern
of relationship between the two was indeed observable. In almost all these 24
sentence pairs, there was an obvious relationship between one or more of the
added and deleted verbs, but some type of unnecessary change had been made
in the translation. In these cases, Fries-Gedin paraphrases the meaning of the
source segment or does not use the closest possible translation. One example
of this is sentence pair 1161, where the verb said in “’We told you to destroy
them!’ said mrs Weasley [...]” is deleted as the target sentence instead uses
the verb skrek, the equivalent of shouted. The target reads: ““Vi sa ju åt er
att förstöra dem!” skrek mrs Weasley [...]”. This is a rather typical example of
a word pair that are semantically related, but where a much closer translation
could have been chosen, wherefore the pair is treated as an addition/deletion
pair.
This close investigation also made it possible to investigate manifestations
of the translation universals that were not feasible to investigate for the entire
corpus, due to the extent of the corpus and the lack of automated possibilities for
searches of this kind. Signs of explicitation, as well as patterns of simplification
and normalisation were obtained in this way.
Explicitation was manifested in a way that could possibly be a pattern.
In English, characters are sometimes affectionately called short forms of their
names. Cedric Diggory, a boy Harry knows from Hogwarts, is repeatedly called
Ced by his father, as in sentence pair 1241 that reads: “’Ced’s talked about
you, of course’, said Amos Diggory”. This is translated as “”Cedric har förstås
pratat om dig”, sa Amos Diggory”. In cases like this where the short form of
the name is used in the source text, the translation either reads Cedric, or the
passage containing the name is deleted in the target text. I argue that this is
a way of explicitating the text, but whether or not it is a pattern cannot, of
course, be stated without a more thorough investigation of the entire corpus.
7.2. TRANSLATIONAL RESULTS 49

For simplification, one possible manifestation is that there are oftentimes

more sentences in the target text than in the original, which proved to be true
for the 150 sentence pair subsample of HP4. The source side consisted of 150
sentences, and the target of 176. This is an increase by 17 percent in the
subsample. Although this cannot be generalised to the whole corpus, it is at
least an indication that there are indeed more sentences in the target texts.
In translating between English and Swedish, some verb constructions will
inevitably lead to differences between the source and target texts, which was
apparent in a few cases. For example, when describing something a person does
in Swedish, a construction based on stood and, sat and or similar combinations
of verb plus and is often used as an equivalent to the gerund form in English in
order to signal ongoing action. This is exemplified in the first words of sentence
pair 1177, where the source sentence reads: “Harry, having been thinking about
thousands of wizards [...]”. The target sentence, however, is: “Harry, som gick
och tänkte på alla de tusentals trollkarlar [...]”. The literal translation of this
back to English would be Harry, who was walking and thinking about all the
thousands of wizards [...]. The target sentence contains other changes as well,
but it still illustrates the point of how a gerund construction can be adapted to
fit Swedish usage, which is an instance of normalisation.

7.2.2 Translation Universals

Manifestations of the translation universals were also investigated for the entire
corpus.

Explicitation

For the core purpose of this study, explicitation was expected to be manifested
primarily in two ways. The first was that if there were more additions than
deletions in the samples, this could be seen as an indication of explicitation.
Looking at the combined effect of table 7.1 and table 7.2, there is indeed more
added than deleted information in the samples. However, addition of nouns
is, as mentioned in section 3.4.1, considered to be a typical manifestation of
explicitation. In the HP-corpus, the number of additions is lower for the noun
category than for any of the other categories of words. This fact is an indication
that explicitation is not so strongly manifested in the HP-corpus, at least not
for the traditional type of lexical explicitation by the addition of nouns.
Notwithstanding this, the second expected manifestation of explicitation in
the corpus follows logically from the first, in that the translated texts were likely
to be longer than the original texts. As early as during the sentence alignment, it
became clear that the samples seemed to conform to the tendency of translations
to be longer than their originals. As can be seen in table 7.7, there are more
tokens in all the TT-samples, compared to their STs. This increase, although
consistent, struck me as being smaller than expected, as the impression during
the sentence alignment was that the Swedish texts were noticeably longer. A
50 CHAPTER 7. RESULTS

possible explanation for this will be given in section 8.1.3 in the discussion
chapter.

Sample Sample size ST Sample size TT Difference

HP1 26121 26642 + 521
HP2 24780 25359 + 579
HP3 20036 20671 + 635
HP4 22546 22961 + 415

Table 7.7: The respective sizes (in number of tokens) of the samples in the
HP-corpus, and the difference in number of tokens.

Another way of explicitating translations is to add lexical elements that

serve to explain events and notions in the source texts. Although this type of
manifestation of explicitation is not in focus in this study, two examples from
the corpus are brought up nevertheless. The point of doing so is partly to show
that there are indeed examples of lexical explicitation in the corpus, and partly
to give the reader a better grasp of what such an addition might look like.
The first example is in sentence pair 1206 in HP4, which reads: “Hermione
came over the crest of the hill last, clutching a stitch in her side”. This has
been translated as (added information in italics): “Hermione kom upp sist över
branten. Hon höll sig i sidan, där hon hade fått håll av den ansträngande
klättringen”. A gloss translation of this would be something like: Hermione
came up last over the steep. She was holding her side, where she had gotten a
stitch from the strenuous climb. This is a clear example of explicitation, as the
fact that the stitch was caused by the strenuous climb has been added in the
Swedish text.
Another example of explicitation is sentence pair 588 in HP4. The source
sentence reads: “He had never seen anything that looked less like a pig”. The
“anything” refers to Harry’s friend Ron’s owl, which Ron calls both Pig and
Piggy. To explain the semantic meaning in English of the name Pig and its
relation to the animal pig, Fries-Gedin has added information. The target sen-
tence reads “Piggy, det var ju en liten gris, och han hade aldrig sett någonting
som mindre liknade en gris”. The first part of this sentence, Piggy, det var ju en
liten gris, och is an addition roughly corresponding to Piggy, that was a small
pig, and. Apart from these two examples, there are many other instances of
lexical explicitation in the corpus, but these sentence pairs suffice to illustrate
that lexical explicitation is manifested, as well as some of its possible effects on
a text.

Simplification
One of the possible manifestations of simplification investigated for the HP-
corpus is the punctuation, and whether or not it has been strengthened. As can
be seen in table 7.8, there is indeed evidence of strengthened punctuation in the
7.2. TRANSLATIONAL RESULTS 51

samples. Commas and semicolons have been changed to full stops, and signifi-
cantly, there is a change over time concerning the strengthening of punctuation
markers. The implication of this is that the texts have become more simplified
from HP1 to HP4.

Sample ,→. ;→. ;→, ;→;

HP1 8 0 1 28
HP2 44 17 13 6
HP3 95 13 8 5
HP4 88 31 12 13

Table 7.8: The changes in punctuation in the samples.

In a few instances, other punctuation markers than commas have been

strengthened, but in proportion to the total number of punctuation markers
of each kind, commas dominated among the strengthened elements. The treat-
ment of semicolons will be further discussed in the normalisation section below.
The complexity of the texts has also been decreased by the way Fries-Gedin
has handled the regional dialects of the characters Rubeus Hagrid and Stan
Shunpike (see table 7.9 for examples of the dialects). That the dialects were
more prominent in the source texts became evident already during the word
alignment. In order to investigate if this impression was correct, a small-scale
case study was performed on subsamples of the corpus where the two characters
make utterances. The case study had to be focused on different samples for the
two characters, as there is only dialogue from Stan Shunpike in HP3, and Hagrid
makes utterances in only HP1 and 2.
The case study revealed that Fries-Gedin has treated these two very different
accents in the same way, to all intents and purposes. The dialogue of both
Hagrid and Stan is translated using simple markers, i.e. spelling the words as
they sound when spoken. Both characters, in Swedish, say å instead of och
for the source use of an (which would be and in standard English spelling),
mej and dej instead of mig and dig for me and you, and e instead of är for is.
These expressions, that are used profusely for Hagrid and Stan, are not used
for any of the other characters (except for å, which can also be an interjection
corresponding to the English ah, or oh). The simple markers are the only
elements that separate the utterances of Hagrid and Stan from the rest of the
text in the TTs, and thus these two characters seem to speak in the same manner
in Swedish. Furthermore, there is only very little contrast in dialect between
Hagrid and Stan and the rest of the characters in Swedish, which also simplifies
the dialogue.
Thus simplification is indeed manifested, but the treatment of dialects does
not only mean that the texts have been simplified, but also that they have been
normalised. Removing coarse dialects in dialogue adapts the text to a Swedish
audience because in written Swedish, it is very uncommon for authors to use
regional dialects. Further implications of the use of simple markers will be dealt
52 CHAPTER 7. RESULTS

Character Source utterance Target utterance

Hagrid ’Best be off, Harry, lots ter do “Bäst å ge oss iväg, Harry,
today, got ta get up ter Lon- massor å göra idag, måste
don an buy all yer stuff fer fara opp till London å köpa
school.’ alla dina saker till skolan.”
Hagrid ’Suppose the myst’ry is why “De konstiga e väl varför Du-
You-Know-Who never tried Vet-Vem aldrig försökte få
to get ’em on his side över dom på sin sida tidigare
before...probably knew they visste nog att dom stog för
were too close ter Dumble- nära Dumbledore för å vilja
dore ter want anythin’ ter do ha nånting med Den Mörka
with the Dark Side.’ Sidan att göra.”
Stan ’Very close to You-Know-Oo, “Han stod väldigt nära Du-
they say anyway, when lit- vet-vem, säjer dom. Men den
tle Arry Potter put paid to där gången då lille Harry Pot-
You-Know-Oo’ - Harry ner- ter gav Du-vet-vem på nöten
vously flattened his fringe så” Harry slätade nervöst
down again - ’all You-Know- till luggen över pannan igen
Oo’s supporters was tracked “spåra dom opp alla Du-vet-
down, was n’t they, Ern?’ vems anhängare, visst va de
så, Ern?”
Stan ’Eleven Sickles, said Stan, but ”Elva siklar”, sade Stan,
for firteen you get ot choco- “men för fjorton fåru varm
late, and for fifteen you get an choklad å för femton fåru en
ot water bottle an a toofbrush varmvattensflaska å en tand-
in the colour of your choice.’ borste i vicken färg du vill
ha.”

Table 7.9: The dialects in English and Swedish.

with in section 8.1.3.

In addition, there are indications of other manifestations of simplification
in the HP-corpus. Long sentences have in some cases been divided into several
shorter ones, as was revealed in the close investigation of the 150 last sentence
pairs of HP4. In other instances, information that increases the complexity of
a sentence has been removed. However, these manifestations have not been
investigated for the full corpus in any structured way, due to the extent of the
corpus and the lack of tools suitable for studying these aspects.

Normalisation
As mentioned above, normalisation is manifested in the way the translator has
chosen to treat the dialects of Rubeus Hagrid and Stan Shunpike. In addition
to this, two other kinds of manifestations of normalisation have been found in
the HP-corpus.
7.2. TRANSLATIONAL RESULTS 53

Firstly, there are sentences in the corpus that have been made more gram-
matically correct in the translations than they were in the source texts. One
example of this is sentence pair 1225 in HP4. The original sentence reads:
“’Long walk, Arthur?’ Cedric’s father asked”. The Swedish sentence has been
grammaticised by completing the sentence. The Swedish equivalent of did you
have a has been added before long walk, as in “”Hade ni långt att gå, Arthur?”
frågade Cedrics far.” Another example of normalisation through grammaticis-
ing an ungrammatical utterance is in the top row of table 7.3, where the verb
has been added, making the Swedish sentence complete.
Secondly, normalisation can also be manifested in the translation of punc-
tuation markers. Translators tend to adapt the usage of punctuation markers
to fit better with the target language usage, and evidence of this have been
found in the HP-corpus. Particularly interesting is the treatment of semicolons
in the translations. Semicolons are not very common in original Swedish texts,
especially not in children’s literature, but as can be seen in table 7.8, many
semicolons are kept in the target versions of the HP-samples. In comparing
the numbers for the respective samples, it is evident also that there has been a
change over time in the treatment of semicolons.
In HP1, 28 semicolons have been retained, and very few other changes have
been made to this particular punctuation marker. In HP4, only 13 semicolons
have been retained. Moreover, 31 semicolons have been changed into full stops
and 12 into commas. For HP1, the corresponding figures are much lower, as no
semicolon has been changed into a full stop, and only one semicolon has become
a comma.
Also concerning commas there are indications of normalisation. As is evident
in table 7.10, many commas have been omitted in the target texts, which is also
an indication of normalisation because commas are used much more frequently
in English, compared to Swedish. The conclusion I draw from this is that
in the usage of syntactic markers, the texts are normalised through adapted
punctuation, and the tendency for the translator to normalise the texts in this
way has increased over time, at least regarding semicolons.

Sample Nr. of commas in ST Deleted commas Added commas

HP1 1504 404 57
HP2 1639 519 83
HP3 1331 439 66
HP4 1461 528 170

Table 7.10: The changes in commas in the samples.

7.2.3 Investigating Translational Choices

In order to show that tools like I*Link have enormous potential in the investi-
gation of lexical choices in the corpus, some examples of different translational
54 CHAPTER 7. RESULTS

choices and situations are presented below. These examples are also meant to
illustrate that the choice of translator does have a real effect on the produced
translation, an effect that is manifested in the particular translational choices
of that translator.

Lexical Choice
During the manual word alignment, I noticed that Fries-Gedin used two alterna-
tive translations for wand, namely trollstav and trollspö. Fries-Gedin has opted
for using trollstav when the carrier of the wand is male, and trollspö when the
carrier is female (see examples from HP2 in table 7.11).

Source sentence Target sentence

Ron let go of the steering wheel com- Ron släppte ratten helt och hållet
pletely and pulled his wand out of och drog fram sin trollstav ur bak-
his back pocket. fickan.
She was wearing a flowered apron Hon var iförd ett blommigt förkläde
with a wand sticking out of the och ur en av fickorna stack ett
pocket. trollspö upp.

Table 7.11: A representation of the translator’s choices in translating wand.

There is no obvious explanation as to why Fries-Gedin has done this, as

wand is gender-neutral in English, as is both trollstav and trollspö in Swedish.
Moreover, trollspö is a compound in Swedish, and the second part, spö is an
instrument used mainly by fairies, not witches and wizards. In Swedish a spö is
quite different from a stav, and the difference is not based on the gender of the
carrier, but rather on the context in which the instrument is used. Possibly, it
could be argued that a stav somehow gives the impression of being longer and
more sturdy than a spö, which in general corresponds to the properties that
Rowling describes in the wands women and men use. However, since spö and
stav are generally not used in the same context, the difference between them
is emphasised. The consequence is, perhaps, that a Swedish reader interprets
them as being different instruments, which they are not supposed to be.
Moreover, in the political climate of today, the act of gender-differentiating
when it is not absolutely necessary is somewhat dubious, especially since chil-
dren’s literature plays an important role in shaping children’s social and cultural
identity (Puurtinen 1998). Interestingly, this gender-differentiating is present in
the translation of the first two books, but not in the fourth. Wand is not
mentioned in the third sample, so unfortunately no comparison can be made
with the third book. Whether using trollstav consistently in HP4 is a conscious
choice by Fries-Gedin or just inconsistency is, of course, impossible to say just
by investigating the texts.
Apart from wand, there are other examples of expressions specific to the
Harry Potter books where Fires-Gedin has changed her translation in the later
7.2. TRANSLATIONAL RESULTS 55

parts of the series. The cupboard under the stairs is a well-known concept to
any Harry Potter reader, as it is what functions as Harry’s room in the Dursley
house in the beginning of the series. Later on, it is where Harry’s Hogwarts
things are kept when he is home for the holidays. The cupboard under the
stairs is not translated consistently throughout the corpus. The construction
can be found in all four samples, and in HP1 the full construction is translated
as skrymslet under trappan. In other instances, where the source consists of
only cupboard, the Swedish translation is krypin. Both skrymsle and krypin are
in some cases modified with the adjective trånga, denoting narrow in English.
In HP2, HP3 and HP4, the full construction is translated as skrubben under
trappan, and shorter versions as skrubben. Why this change has been made is,
naturally, impossible to say without asking the translator, but it is an indication
that she is not averse to change, if it is called for. In this case, I argue that it
is a change for the better, as skrymsle denotes a very small and narrow space,
generally impossible to close off with a door, corresponding more closely to the
English nook than to cupboard. Skrubb, on the other hand, is a more likely
description, since it denotes a rather small, closed-off space, but still giving the
impression of being large enough to hold an eleven year old boy.

A Translation Difficulty in the Harry Potter World

The fact that the very specific world of the Harry Potter books and its asso-
ciated neologisms causes some problems to translators becomes obvious when
the translation for the neologist use of apparition is studied. Rowling uses Ap-
parition with a completely new meaning, to describe the way in which wizards
can teleport themselves to another location instantly. Also the verbs used in
connection with this activity are neologisms. To Apparate is to appear in the
new location by Apparition and to Disapparate is to disappear by Apparition.
The difficulties in translating this activity is illustrated in sentence pairs 1188
and 1189 of HP4 in table 7.12.
Sentence pair Source text Target text
1188 Some Apparate, of Några använder
course, but we have to set sig förstås av
up safe points for them spöktransferens, men
to appear, well away from vi måste välja ut säkra
Muggles. ankomstställen som de
kan dyka upp på, väl
dolda för mugglarna.
1189 I believe there’s a handy Jag tror att de har hit-
wood they’re using as the tat en lämplig skog för
Apparition point. ändamålet.

Table 7.12: Examples of the translation of Apparate and related constructions.

Fries-Gedin has chosen the neologist compound noun spöktransferens as an

56 CHAPTER 7. RESULTS

equivalent of both Apparition and the infinitives of the associated verbs, which
has some interesting implications. The first half, spök- is derived from the
Swedish word for ghost, spöke, which is very close in meaning to the original
meaning of apparition. Transferens is a neologism that Fries-Gedin has probably
built on the word transferering, denoting a transfer of some resource, usually
money. Spöktransferens works as an equivalent of the infinitives of the verbs to
Apparate/Disapparate, but it does not work in the active sense, when somebody
Apparates, or Disapparates. The chosen translation for the active senses is
använda sig av spöktransferens, the equivalent of to use ghost transferal, which
is a cumbersome construction. In table 7.12, sentence pair 1188 shows the
relationship between Apparition and spöktransferens. In 1189 the difficulties in
translating constructions containing Apparition is illustrated. Fries-Gedin has
in this case chosen to paraphrase and simplify by changing the Apparition point
to the Swedish equivalent of for the purpose.

Lexical Patterns
From the resources built in the alignment, it is possible to create alphabetical
lists of the words and their translations. By investigating these lists, lexical
patterns of how words are translated can be discovered. In instances where
words have many and diverse translations, this is generally a sign that they have
been difficult to translate. Because the HP-books portray a complex, magical
environment with quite detailed vocabulary, it would be reasonable to expect
that words specific to this world might have been especially challenging for the
translator. Therefore, a closer investigation was made into the translation of
vocabulary typical to this domain.
Whether a person is wizard or Muggle is paramount in the Harry Potter
world. Muggle is consistently translated with the neologism mugglare. In the
cases where it is a part of a longer noun construction, such as Muggle clothes,
this, equally consistently, becomes a compound noun in Swedish, mugglarkläder,
with the stem mugglar-.
Similarly, wizard is translated as trollkarl in the absolute majority of the
cases. When it is a part of a longer noun construction, such as in the wizarding
bank, this becomes a compound noun in Swedish with the stem trollkarls-, in this
particular case trollkarlsbanken. One rare exception is wizard gold, translated as
trollmynt, which changes the meaning of the word, since troll in Swedish means
exactly what troll does in English. The second part of the word is also changed,
as gold is translated into the equivalent of coin.
Patterns of consistency, as well as patterns of inconsistency, become apparent
when investigating the resources in this way. For example, noun constructions
with magic are oftentimes translated into a compound with trollkarls- as the
stem, i.e. the same stem as used for translating compounds containing forms
of wizard. The translation equivalent of magic is magi, but the word has eleven
different translations (see table 7.13). Magical, however, has fewer translations,
but they are built both around the magi and the troll stems. Judging from
the amount of translation equivalents for words about wizards and magic, the
7.2. TRANSLATIONAL RESULTS 57

magical element of the Harry Potter world seems to have caused a problem in
the translation process.
An interesting lexical pattern of inconsistency apparent in the HP-corpus
is the translation of the Dursleys. Perhaps surprisingly, it has ten different
translations ranging from similar constructions equivalent to the Dursley couple
and the Dursley spouses to equivalents of his uncle or aunt, them and the others
(see table 7.13). This illustrates a difference between English and Swedish; in
Swedish a plural form of a family name is not used as consistently to describe
the unit of that family as it is in English, which could account for the many
different translation alternatives.

Source word Target translations

magic magi, magiska, trolla, trolldom,
trolldomskraft, trolleri, trollkarlar,
trollkarlsvärlden, trollkonst, trol-
lkonster, trollkunskap
magical förtrollande, magisk, magiska,
magiskt
the Dursleys de, de andra, Dursleys, familjen
Dursley, familjen Dursleys, hans
morbror eller moster, makarna
Dursleys, mr och mrs Dursley, paret
Dursley, paret Dursleys

Table 7.13: The patterns of translations for certain typical Harry Potter related
words.

Semantic Mirroring
The resources built during the word alignment can also be used to create more
powerful resources than the alphabetical lists, such as semantic mirrors. With
semantic mirrors, it is possible to extract information about the translations
that is not available in I*Link in itself.
Semantic mirroring of the resources built up in the HP-project was made by
Helge Dyvik at the University of Bergen, Norway, and this resulted in two addi-
tional means for studying the material. One is a thesaurus-like file that shows
all the words in the corpus that have many different translation alternatives,
and does not contain words with few translation alternatives. As mentioned
above, a large number of translation alternatives for a certain word indicates
that it has been difficult to translate, as there is not only a few possibilities for
equivalence. Thus the thesaurus shows words that have been translated incon-
sistently. The other is a search tool that makes it possible to search for specific
words in the thesaurus. For a full description of semantic mirrors, see Dyvik
(2003).
58 CHAPTER 7. RESULTS

The semantic mirrors do not show direct correspondences, that is source

word and its translation or translations in the corpus. Instead, they attempt
to regroup the data so that semantic information such as hyperonyms and syn-
onyms can be extracted from the corpus. These indirect correspondences can
be very interesting to investigate closer if the object is to study language use in
a particular semantic field, such as the fantasy genre of literature. One possible
use of the thesaurus-listing is that it functions as a Harry Potter dictionary,
and could be paired with other resources from other studies in order to create
a genre dictionary for fantasy books.
The semantic mirrors are also practical tools for obtaining an overview of
the words in the corpus, as they make the material more accessible and easier
for a layman to read than the output from I*Link. A listing in the thesaurus can
look like this, with the listed word in bold face, its translation or translations,
synonym or synonyms, and where applicable, related words:

grinning
(Translation: log. )
Synonyms: grinned—1—
gripped
(Translation: tog. )
Synonyms: withdrew.
grow
(Translation: blev. )
Synonyms: became.
growled
(Translation: röt, brummade. )
Synonyms: roared—1—, snarled—1—
Related words: barked, bellowed.
grudgingly
(Translation: motvilligt. )
Synonyms: resentfully.

For the particular purpose of this study, the semantic thesaurus and search
tool were used to some extent in the investigation of lexical choice. However,
the main contribution of these two resources are perhaps to simplify searching
the material and browsing the thesaurus for those with a particular interest, be
it in semantic relationships of translations or in the Harry Potter books.

7.3 Methodological Results

The methodological analysis focuses on the different strategies used in the word
alignment.
7.3. METHODOLOGICAL RESULTS 59

7.3.1 Evaluation of the Different Strategies

The four different strategies that were used for the four samples constitute the
basis of a brief evaluation of I*Link and the strategies themselves. A summary
of the strategies can be seen in table 7.14.

Sample Tool(s) used Resources Required time

HP1 I*Link Built-in 18 hours
HP2 I*Trix, I*Link Built-in 16.5 hours
HP3 I*Trix, I*Link Dynamic from HP1 and 2 10.5 hours
HP4 I*Link Dynamic from HP1, 2 and 3 16 hours

Table 7.14: The different strategies used in the alignment of the samples.

The basis for the evaluation of the strategies presented above is their effi-
ciency, measured in the time it took to align each sample. This is because manual
aligning is, as already mentioned, despite its advantages, very time-consuming.
Consequently, it is relevant to consider if any one strategy decreases the time
required by the aligning more than the others. As is evident in table 7.14,
HP3 required the least time to align, and HP1 the most. Running a sample
through I*Trix takes only a few minutes, so this time can be disregarded in the
comparison between the different strategies.
One of the two most obvious explanations to why HP1 required so much
time is that it was the largest sample in terms of number of tokens. The other
explanation is the fact that it was the first sample to be aligned, and a certain
amount of the time was spent dealing with insecurities in using I*Link and
trying to maintain consistency.
However, in investigating the efficiency of the strategies, it is of course pivotal
to take the exact sizes of the samples, i.e. the token-count, into consideration.
In table 7.15 below, such a comparison is made.

Sample Seconds/sentence pair Seconds/token Words/sentence

HP1 36.6 2.46 14.77
HP2 38.4 2.40 16.04
HP3 30.6 1.86 16.25
HP4 45.6 2.58 17.75

Table 7.15: The efficiency of the different strategies, in relation to sample size.
The words/sentence count is the mean number of words per sentence on the
source side.

The fact that there is such a great difference between the time required to
align HP3 in comparison with the other samples is of course the most relevant
finding of the strategy evaluation. The implication of the results is that it
appears to be most efficient to align a subsample or a part of the corpus, use the
60 CHAPTER 7. RESULTS

dynamic resources from that session to automatically align the next subsample
in I*Trix, and then revise those results in I*Link.
Chapter 8

Discussion

First of all, I would like to point out that this thesis is in no way intended to
be a value-judgement of the translations or the translator. As Newmark points
out (1988), translations must be discussed as they are always made subjectively,
and this thesis is a mere discussion of the translation of Harry Potter, also made
from a subjective viewpoint.
That being said, the results unanimously indicate that there are indeed
significant changes between the translations in relation to their respective orig-
inals, and that these changes increase sequentially. Additionally, the universals
of translation are manifested in the corpus. Moreover, the different strategies
used in the alignment process gave different results concerning their efficiency,
and in summary, the third strategy seems to be the most efficient, at least in
the case of this annotator. In other words, all the hypotheses stated in the
introduction have been verified by the results, which is highly encouraging.
Following the format of the last chapter, the discussion will be divided into
three main sections. The first section will discuss the translational results. The
second will deal with the methodological results, and will also include some
additional discussion about my experiences of I*Link and using this set of tools
for projects of this kind. The third and final section contains suggestions for
further research.

8.1 Discussion on the Translational Results

8.1.1 FDG Imperfections
Apart from the difficulties in obtaining exact word class labels for all lexical
tokens in the samples, the functional dependency grammar standard that was
used for this study has unfortunately flawed the material in other ways. As
already described, the construction where three dots (...) indicate a pause in a
character’s dialogue, disappeared in the tagging process. This caused delays in
the aligning process, and made it impossible to fully study punctuation markers
to the intended extent.

61
62 CHAPTER 8. DISCUSSION

In the final hours of analysis carried out, another FDG-related problem was
discovered. In sentence pair 1125 in HP4, the Swedish råkade, a verb equivalent
to happened to, has not, as expected, been tagged V, but instead A, for adjective.
Due to this discovery, all word class tags for all tokens of the last 150 sentence
pairs of HP4 used in the close investigation of additions and deletions were
checked manually. It was found that for both the source and the target sides,
less than 2 percent of the tokens carried a faulty tag. If this is indicative for the
whole HP-corpus, the FDG precision rate is approximately 98 percent. Based
on this, I conclude that the few flaws concerning FDG-tags present in the HP-
corpus are not likely to have affected the reliability of the results to any great
extent.

8.1.2 The Relationship between Additions, Deletions and

Lexical Shifts

Because the focus of this study is in part on the influence of the translator on
the translated text, I have tried to distinguish between small and significant
changes, which ties into the concepts of lexical shifts, additions and deletions.
In my definition of necessary lexical shifts, it is only natural that they have
no closer or more accurate translations. In other words, these can be treated as
regular translational equivalents, because to all intents and purposes, they are,
as long as the meaning of the source words could not have been more preserved
in any other target construction.
Concerning the nature of unnecessary, or voluntary, changes made to the
texts, the close investigation revealed that at least regarding verbs, many addi-
tions and deletions are used in combination, as pairs. They are in fact lexical
shifts or parts of paraphrases, not regular additions and deletions.
Because of the fact that significant lexical shifts seem to be rather common,
I feel that it would be positive to be able to distinguish them from regular
additions and deletions, at least if the focus is on the degree of change. In
my opinion, this could be done by making it possible to mark lexical shifts in
alignment programs such as I*Link. In the analysis, all types of lexical shifts
could then be treated as indicators of change in the translation, and other lexical
shifts could be analysed as having a similar effect to additions and deletions.
Implementing lexical shifts in I*Link would enrich the analysis of the corpus
as an even more fine-grained analysis could be made into the nature of the
changes the translator has made to the target text. Especially, more specific
lexical shifts could be seen as clear indicators of explicitation. In addition, the
changes that are simply more free in relation to the source text than necessary
could be distinguished, and this could in turn be used to measure how free the
translation is in relation to the original text. One possible disadvantage could
be that it might make the aligning more time-consuming due to the time that
would be spent distinguishing the different kinds of lexical shifts.
8.1. DISCUSSION ON THE TRANSLATIONAL RESULTS 63

Lexical Shifts and Translation Universals

My strategy for aligning lexical shifts, as described in section 4.6, has some
implications for the results on translation universals.
Regarding more specific lexical shifts, different types are aligned differently,
some as regular links and some as addition and deletion pairs. This makes it
more difficult to search for these manifestations in the HP-corpus. Explicitated
references as manifestations of explicitation are to some extent hidden among
the regular links and cannot easily be counted. Equally, less specific lexical
shifts can indicate simplification, but they are also difficult to find, as they are
treated as additions and deletions.
However, the intent in this study was never to be able to count the number
of specific manifestations, as it is not feasible in the current version of I*Link.
The goal concerning translation universals was to search for manifestations of
the three different types, and such manifestations have indeed been found.

8.1.3 Translation Universals

Explicitation
The fact that there are much more additions than deletions in the target texts
imply that they have been explicitated, and that they have been more explic-
itated over time. However, the reasons to why there are more additions and
deletions after time can only be hypothesised about here. Perhaps the transla-
tor “knows” the characters better after translating many hundreds of thousands
of words written by Rowling; she is more familiar with them and therefore in-
terprets the texts in ways she did not in the beginning, reading between the
lines and adding things that perhaps are implicit in the source texts, making
them explicit in the target. In some lexical choices, it is possible to see her
interpretations.
Another manifestation of explicitation was evident in that the target texts
contained more tokens than the original texts. The difference in number of
tokens was smaller than I expected, however, and one reason for this can possibly
be the difference in how nouns are constructed in English and Swedish. In
Swedish, one single compound noun is used as the equivalent of a string of
nouns in English, which of course affects the token count as one of these tokens
in Swedish can be the translational equivalent of two or more tokens in English.
On the other hand, the reverse is true for Swedish equivalents of English verb
constructions of the type was doing and was thinking etcetera, which become
one token in Swedish, in this case gjorde and tänkte. This study does not include
any further investigation into how these convergences and divergences due to
differences between source and target languages have affected the token count,
but it is reasonable to assume that they have indeed had an effect.
In addition to this, changes in the use of commas between the source and
target texts can have contributed to the fact that the target texts were perceived
as being longer in comparison to the source texts than the token count implies.
Many commas have been removed from the ST’s and few added to the TT’s (as
64 CHAPTER 8. DISCUSSION

is obvious in table 7.10). Each comma is counted as one token, just like each
word is counted as one token. This is of course true for all punctuation markers.
In some instances, there might be a relationship between the added and
deleted commas, like between added and deleted words. Some of the added
commas could be replacements for deleted ones. If commas were moved within
the same sentence or sentence pair, I tried to be consistent in linking them as
each other’s equivalent, notwithstanding the fact that the comma was moved.
If the number of added commas is subtracted from the number of deleted
commas, the difference equals the number of tokens that were commas in the
source text, but are not commas in the target text. This way of analysing the
use of commas using the data in table 7.10 indicates that even if there are only
521 more tokens in the TT of HP1, there might, in effect, be many more words,
since 404 commas have been removed and only 57 added. In other words, the
target texts might have been more explicitated by containing more words than
a simple comparison of token counts between the samples reveals.

Normalisation and Simplification in the Treatment of Dialects

Though the case study on the dialects of Hagrid and Stan Shunpike reveals that
they have been normalised and simplified, it can be argued that Fries-Gedin has
tried to retain part of the effect of the dialects. In marking the speech of Hagrid
and Stan with unconventional spelling, she still signals something of a dialect in
these characters. However, the fact that they are two very different dialects has
not been retained, which means that the texts have been simplified. In addition,
keeping in mind that the Harry Potter series are children’s books, a substantial
part of the intended audience will perhaps not actually read the text, but get
it read to them, by adults. For this group, any effect of the simple markers will
most likely be completely lost, as there will be no audible difference between
the characters with dialect and those without, unless of course the person that
does the reading picks up on the simple markers and in some imaginative way
acts out the perceived difference.

8.1.4 The Development of the Translator

The fact that there are such large differences between the number of additions
and deletions for the target texts of HP1 and HP4 appears to be a very clear
sign of development in the translating style of the translator. In addition to
this, the treatment of punctuation markers also indicates that the translating
style has changed from HP1 to HP4. However, there is a serious threat to the
validity of the results attained in this study, concerning the sequential difference
between the samples.

8.1.5 Sources of Error for the Translational Results

One possible source of error is that the clear indication of change between HP1
and HP4 can have two different explanations. It could either be that there
8.2. DISCUSSION ON TOOLS AND METHODOLOGICAL RESULTS 65

is an actual difference between the samples, caused in the process of transla-

tion. However, the difference could also possibly depend not on the samples in
themselves, but on the aligning and the annotator. Unintentionally, I may have
changed my way of linking between the samples, thereby causing the sequential
effects myself. However, that I did post-edit HP1 after finishing HP4, and only
found it necessary to correct a few links, indicates that perhaps I was rather
consistent and did not change my way of choosing links significantly during the
project.
Trying to maximise the use of the built-in resources meant that the samples
had to be aligned in sequence in the HP-project. Another reason for doing
so was to, with I*Link, mimic the process of the translator. However, taking
into account what is said above about possible sequence effects, it would have
been better, perhaps, not to align the samples sequentially. One idea for future
studies of this nature is to divide each sample into subsamples, and randomising
the order in which they are aligned, to prevent sequence effects in the alignment
process from influencing the results.

8.2 Discussion on Tools and Methodological Re-

sults
8.2.1 Using the Alignment Tools
Word alignment with software tools such as I*Link as a method for studying
translations has many benefits compared to manual inspection and other low-
tech methods. The fact that the texts are POS-tagged is absolutely pivotal,
as it provides the researcher with so much readily available information which
would otherwise have been very cumbersome to extract. The material becomes
searchable, and most words and constructions can be closely examined from a
linguistic perspective.
Notwithstanding this, I*Link, like alignment systems in general, needs to be
developed further. One general issue traces back to Borin’s point I brought up in
section 4.1, that the annotator is in the hands of the tools (2002). This is a valid
point, both regarding the alignment and the subsequent analysis. Especially the
analysis does become heavily reliant on what the tools allow for and simplify.
In my case, the analysis of the results was very cumbersome, mainly perhaps
because of the lack of a framework for this type of study. The results, however,
are encouraging enough to motivate further research of this type.

8.2.2 Advantages and Disadvantages of Using I*Link

In relation to this study, the most significant advantage I*Link has in comparison
with automatic alignment systems is that in I*Link, additions and deletions can
be distinguished, whereas in the latter, such cases cannot be separated from
instances where the system just does not find a suitable link. Naturally, if
additions and deletions cannot be distinguished, they cannot be studied.
66 CHAPTER 8. DISCUSSION

Above and beyond all, what a system such as I*Link provides to the field
of translation studies is the ability to fully align and investigate a large corpus
of texts in a structured way, using the tools integrated in the system. It makes
it possible to search the material and get structured output in mere seconds,
once the alignment is done. In my opinion, the challenge is to find a way to
analyse the material and the outputs so that scientifically interesting results can
be presented. A risk with all systems that provide a lot of results in the shape
of numbers and statistics is that it is tempting to over-use the possibilities for
making calculations. Consequently, the researcher must be very focused on the
scope of the study and avoid presenting all figures that can be calculated on the
aligned material.
In my personal experience, the built-in heuristics can also cause problems
in some cases, because if the user chooses another strategy than the one pre-
programmed in I*Link, the system does not respond to this as quickly as perhaps
desirable, but continues to suggests links that are preferable according to the
heuristics. This can be very frustrating to the user and the risk is that the
system by continuously working against the user dominates the choice process
and convinces the user to adapt to I*Link, which means that the links will be
less consistent than necessary.
The specific situation in which this caused a problem to me was described
in section 6.2.3, and relates to the linking of proper names. Had I instead of
my own strategy chosen to make one link of a character’s name and surname,
as I*Link is built to do, the automatic heuristic would have been very helpful,
and could have aided me in keeping my links consistent.

8.2.3 Specifics of I*Link as Sources of Error

One of the biggest problems with I*Link is that it does not support the alignment
of discontinuous phrases, which is a cause of frustration to the annotator, but
more importantly, the quality of the alignment is affected. The implications of
this for this particular study is that sometimes additions and deletions cannot
be marked as NULL links, because they are, for example, surrounded by an
auxiliary verb and its main verb, as in sentence pair 626 in HP1: “He dodged
the Smeltings stick and went to get the post.” In Swedish, the information
specifying that the stick is from Smeltings is deleted: “Han hoppade åt sidan
för käppen och gick ut för att hämta posten.” Here, käppen is the equivalent
of the stick, and the annotator is forced either to include Smeltings in the link,
or mark the Smeltings as deleted and align just stick with käppen. This is a
situation that the annotator often faces, and it is basically a matter of choice
whether to delete the whole segment, or to align it and include the word or
words that are in fact added or deleted.
In defence of I*Link can be said that the creators of the system are well
aware of this problem (Merkel et al. 2003), and it is possible to mark links as
discontinuous. However, in the current version of I*Link, this does not have any
practical effect as the system lacks possibilities to treat the included segments in
a link marked discontinuous any differently than if the link was simply accepted
8.2. DISCUSSION ON TOOLS AND METHODOLOGICAL RESULTS 67

as one large continuous segment. In other words, a discontinuous link must be

nulled or accepted in its entirety. In the alignment of the HP-corpus, I confess
that I did not mark discontinuous segments as discontinuous, partly because I
discovered the possibility to do so at a very late date in the alignment process,
and partly because doing so would have no practical effect on the results of this
study.
In the sentence pair above in which Smeltings is omitted, the problem could
also have been solved if lexical shifts were implemented. The Smeltings stick
and käppen could then be linked together as a less specific lexical shift.
Another issue with I*Link is that words cannot be divided and aligned in
subsegments, which can be a problem with the tendency in Swedish to form
compound nouns. Sometimes, a piece of information is added to the compound,
but it cannot be marked as an addition. One example of this is sentence pair
385 in HP1: “The only thing Harry liked about his own appearance was a very
thin scar on his forehead that was shaped like a bolt of lightning.” The target
sentence reads: ”Det enda Harry gillade i sitt eget utseende var ett mycket smalt
ärr i pannan som hade formen av en sicksackblixt.” Sicksack is an addition
that gives additional information about the shape of the scar, but it cannot be
marked as an addition in I*Link. However, if lexical shifts were implemented
in the program, sicksackblixt could be linked together with bolt of lightning as
a more specific lexical shift, which would solve this problem.
I find it highly likely that the problems with discontinuous segments and not
being able to divide words have affected the results of this study. How much
and in precisely what way is, unfortunately, not possible for me to say.

8.2.4 Suggestions for Improvements of the Tools

Drawing on the experience of aligning and working with I*Link and I*Trix
accumulated in this project, a few suggestions for augmentations that can be
made to the tools does not seem out of place. The suggested changes could be
of use both in projects like this one, and in working with the tools in general.
No so-called usability evaluation of I*Link has been made during this project,
and the suggested improvements are primarily motivated by either being able to
shorten the time required by the alignment, or by making it possible to perform
a more fine-grained and automated analysis of the material.
Although I*Link gives the researcher an opportunity to study translations
in a structured and accessible manner, more could be wished for in terms of
search options. For any study focusing on the differences between source and
target texts, it would among other things have been very positive to be able to
search for elements of a particular word class that are translated into a different
word class, for example verbs that are translated as anything but verbs. This
can be done to some extent, but since one actively has to search on every
possible change separately (verbs translated as adjectives, nouns, etc.), it is
either necessary to do a great deal of work, or to know exactly what to look for
before the start of the analysis of the material. This can be a pity in explorative
studies such as this one, where the results are sometimes not entirely predictable.
68 CHAPTER 8. DISCUSSION

Another possible feature that would be very useful in I*Link would be to

be able to select exactly what sentence pairs, or string of sentence pairs, a
particular search covers. In this project, a feature like that would have come to
much use for example in the case study made on the treatment of the dialects
of Hagrid and Stan Shunpike. Was such a search option available, structured
exact investigations of the characters’ specific ways of talking would have been
very easy to conduct.
In addition, an algorithm of some sort that would make sure that links that
are in close proximity of each other in the sentence would have more differenti-
ated colours would be helpful.
It could possibly save time in the alignment process if I*Link included one
button that, if pressed, performed some action like “Mark everything that is
unmarked as a NULL link”. This could be useful in studies like this, because
much time is devoted to nulling the links that have no match in the correspond-
ing text. A button that marks all unmarked elements NULL might speed up
the process, although using it would mean that the annotator must be very
careful and always check that all unmarked elements are indeed without corre-
spondences.
Lastly, as I have already hinted at, implementing lexical shifts would have a
number of positive effects. The analysis could be made more fine-grained and
more powerful searches of the material could be made automatically. This could
simplify and strengthen the possibilities for investigating both the influence of
the translator on the text and manifestations of translation universals.

8.3 Suggestions for Further Research

The results of this study seem promising, and gives much food for thought
on further research in the field of descriptive translation studies. The first
suggestion for future research is specific to the word aligned corpus used in this
study. The other two suggestions are given in an order of increasing general
applicability in studies of this kind.
Firstly, as I have reason to question if the seemingly clear sequential devel-
opment of the translator stems from the translations or the word alignment, a
study that deals with this question could be fruitful. On a more general level, I
suggest research into possible sequence effects and how to avoid them in studies
of this kind.
Secondly, one issue that needs to be paid attention to is the consistency
of annotators, and ways of controlling and investigating consistency. I believe
that this is necessary to ensure the reliability of the results of studies of this
kind, as means to control consistency will hopefully make word alignment less
subjective.
Lastly, fantasy, and even fiction, are more or less unexplored genres of lit-
erature when it comes to translation studies. I think that this is a pity, since
much of the material that is translated and read is fiction. Moreover, the fan-
tasy genre should prove to be particularly interesting to study as it contains
8.3. SUGGESTIONS FOR FURTHER RESEARCH 69

much domain specific language use and many neologisms. Consequently, more
research into translation of fantasy and fiction is needed.
70 CHAPTER 8. DISCUSSION
Bibliography

Ahrenberg, L., M. Merkel, A. Sågvall Hein & J. Tiedemann (2000), Evaluation

of word alignment systems, in ‘Proceedings of the Second International
Conference on Linguistic Resources and Evaluation’, Vol. III: 1255-1261,
Athens, Greece.

Baker, M. (1995), ‘Corpora in translation studies: An overview and some sug-

gestions for future research’, Target 7(2), 223–243.

Baker, M. (1996), Corpus-based translation studies: The challenges that lie

ahead, in H.Somers, ed., ‘Terminology, LSP and Translation’, Benjamins,
Amsterdam.

Baker, M. (2000), ‘Towards a methodology for investigating the style of a literary

translator’, Target 12(2), 241–266.

Bergius, H. (2003), ‘Hon tar sig friheter med Harry Potter’, Dagens Nyheter,
July 27 2003 .

Borin, L. (2002), ...and never the twain shall meet?, in L.Borin, ed., ‘Parallel
Corpora, Parallel Worlds’, Rodopi B.V., Amsterdam.

Davies, E.E. (2003), ‘A goblin or a dirty nose? The treatment of culture-

specific references in translations of the Harry Potter books’, The Transla-
tor 9(1), 65–100.

Dyvik, H. (2003), Translations as a semantic knowledge source. Draft.

Hatim, B. & I. Mason (1990), Discourse and the translator, Longman, London.

Holmes, J.S. (2000), The name and nature of translation studies, in L.Venuti &
M.Baker, eds, ‘The Translation Studies Reader’, Routledge, London.

Ingo, R. (1991), Från källspråk till målspråk, Studentlitteratur, Lund.

Malmkjær, K. (1997), Punctuation in Hans Christian Andersen’s stories and

their translations into English, in F.Poyatos, ed., ‘Nonverbal communica-
tion and translation : new perspectives and challenges in literature, inter-
pretation and the media’, Benjamins, Amsterdam.

71
72 BIBLIOGRAPHY

Merkel, M. (1999), Understanding and enhancing translation by parallel text

processing, PhD thesis, Linköpings Universitet.
Merkel, M., M. Petterstedt & L. Ahrenberg (2003), Interactive word alignment
for corpus linguistics, in ‘Proceedings from Corpus Linguistics 2003’, Lan-
caster, UK.
Newmark, P. (1988), A Textbook of Translation, Prentice Hall, Hemel Hemp-
stead.
Nida, E. (2000), Principles of correspondence, in L.Venuti & M.Baker, eds, ‘The
Translation Studies Reader’, Routledge, London.
O’Connell, E. (1999), Translating for children, in G.Anderman & M.Rogers, eds,
‘Word, Text, Translation, Liber Amicorum for Peter Newmark’, Mulitlin-
gual Matters, Clevedon.
O’Connell, E. (2003), ‘What dubbers of children’s television programmes can
learn from translators of children’s books?’, Meta 48(1-2), 222–232.
Puurtinen, T. (1998), ‘Syntax, readability and ideology in children’s literature’,
Meta 43(4).
Rowling, J.K. (1998), Harry Potter and the Philosopher’s Stone, Bloomsbury,
London.
Rowling, J.K. (1999), Harry Potter and the Chamber of Secrets, Bloomsbury,
London.
Rowling, J.K. (2000a), Harry Potter and the Prisoner of Azkaban, Bloomsbury,
London.
Rowling, J.K. (2000b), Harry Potter och De Vises Sten, Tiden, Stockholm.
Rowling, J.K. (2000c), Harry Potter och Hemligheternas kammare, Tiden,
Stockholm.
Rowling, J.K. (2001a), Harry Potter and the Goblet of Fire, Bloomsbury, Lon-
don.
Rowling, J.K. (2001b), Harry Potter och Fången från Azkaban, Tiden, Stock-
holm.
Rowling, J.K. (2002), Harry Potter och Den flammande bägaren, Tiden, Stock-
holm.
Sågvall Hein, A. (2002), The PLUG project: parallel corpora in Linköping, Upp-
sala, Göteborg: aims and achievements, in L.Borin, ed., ‘Parallel Corpora,
Parallel Worlds’, Rodopi B.V., Amsterdam.
Shavit, Z. (1981), ‘Translation of children’s literature as a function of its position
in the literary polysystem’, Poetics Today 2(4), 171–179.
BIBLIOGRAPHY 73

Stolze, R. (2003), ‘Translating for children - world view or pedagogics?’, Meta

48(1-2), 208–221.
Tabbert, R. (2002), ‘Approaches to the translation of children’s literature. A
review of critical studies since 1960’, Target 14(2), 303–351.
Tapanainen, P. & T. Järvinen (1997), A non-projective dependency parser, in
‘Proceedings of the 5th Conference on Applied Natural Language Process-
ing’, pp. 1:49–52.
Toury, G. (1995), Descriptive Translation Studies and Beyond, Benjamins, Am-
sterdam.
Wyler, L. (2003), ‘Harry Potter for children, teenagers and adults’, Meta 48(1-
2), 5–14.
74 BIBLIOGRAPHY
 Avdelning, Institution Datum
Division, Department Date
2005-05-24

Institutionen för datavetenskap

581 83 LINKÖPING

Språk Rapporttyp ISBN

Language Report category
Svenska/Swedish Licentiatavhandling ISRN LIU-KOGVET-D--05/09--SE
X Engelska/English Examensarbete
C-uppsats Serietitel och serienummer ISSN
X D-uppsats Title of series, numbering
Övrig rapport
____

URL för elektronisk version

http://www.ep.liu.se/exjobb/ida/2005/009/

Titel Att spåra översättningsuniversalier och översättarutveckling genom att ordlänka en Harry
Title Potter-korpus

Tracing Translation Universals and Translator Development by Word Aligning a Harry Potter
Corpus
Författare Sofia Helgegren
Author

Sammanfattning
Abstract
For the purpose of this descriptive translation study, a translation corpus was built from roughly the first
20,000 words of each of the first four Harry Potter books by J.K. Rowling, and their respective translations
into Swedish. I*Link, a new type of word alignment tool, was used to align the samples on a word level and
to investigate and analyse the aligned corpus. The purpose of the study was threefold: to investigate
manifestations of translation universals, to search for evidence of translator development and to study the
efficiency of different strategies for using the alignment tools.

The results show that all three translation universals were manifested in the corpus, both on a general pattern
level and on a more specific lexical level. Additionally, a clear pattern of translator development was
discovered, showing that there are differences between the four different samples. The tendency is that the
translations become further removed from the original texts, and this difference occurs homogeneously and
sequentially. In the word alignment, four different ways of using the tools were tested, and one strategy was
found to be more efficient than the others. This strategy uses dynamic resources from previous alignment
sessions as input to I*Trix, an automatic alignment tool, and the output file is manually post-edited in
I*Link.

In conclusion, the study shows how new tools and methods can be used in descriptive translation studies to
extract information that is not readily obtainable with traditional tools and methods.

Nyckelord
Keyword
word alignment, translation universals, translator development, corpus, additions, deletions