Using corpora in contrastive and translation studies

1
Using corpora in contrastive and translation
studies

• Corpus Linguistics
• Richard Xiao
• lancsxiaoz_at_googlemail.com

2
Aims of this session

• Lecture
• Corpora in contrastive and translation studies
• Use of comparable and parallel corpora
• Case study Translation universals, do they
  really exist?
• Lab session
• CUC paraconc and Babel parallel corpus
• Closing
• Shedding of valedictory tears

3
Types of corpora Some distinctions

• Monolingual versus multilingual corpora
• Parallel versus comparable corpora
• Comparable versus comparative corpora

4
Monolingual versus multilingual corpora

• Monolingual corpora
• A corpus that only involves one language
• Multilingual corpora
• A corpus that involves texts of more than one
  language
• A corpus covering two languages is conventionally
  known as bilingual
• Multilingual corpora, in a narrow sense, must
  involve more than two languages
• Multilingual and bilingual are often used
  interchangeably
• Parallel and comparable corpora

5
Parallel versus comparable corpora

• Terminological confusion centres around the terms
• For some scholars (e.g. Aijmer and Altenberg
  1996 Granger 1996 38)
• Corpora composed of source texts in one language
  and their translations in another language (or
  other languages) are translation corpora while
  those comprising different components sampled
  from different native languages using comparable
  sampling techniques are called parallel corpora
• For many others (e.g. Baker 1993 248, 1995,
  1999 Barlow 1995, 2000 110 Hunston 2002 15
  McEnery and Wilson 1996 57 McEnery, Xiao and
  Tono 2006)
• Corpora of the first type are labelled parallel
  corpora while those of the latter type are
  comparable corpora

6
Parallel versus comparable corpora

• In classifying corpora, the criteria used must be
  consistent and logical ways of doing things
• - We can say a corpus is a translation or a
  non-translation corpus if the criterion of corpus
  content is used
• - But if we choose to define corpus types by the
  criterion of corpus form, we must use the
  criterion consistently
• - We can say a corpus is parallel if the corpus
  contains source texts and translations in
  parallel, or it is a comparable corpus if its
  components or subcorpora are comparable by
  applying the same sampling techniques and similar
  balance and coverage
• - It is simply inconsistent and illogical to
  refer to corpora of the first type as translation
  corpora by the criterion of content while
  referring to corpora of the latter type as
  parallel corpora by the criterion of form!

7
Multilingual vs. monolingual comparable corpora

• A common practice in TS is to compare a corpus of
  translated texts (translational corpus) with a
  corpus consisting of comparably sampled
  non-translated texts in the same language
• The two sub-corpora form a monolingual comparable
  corpus for translation research, as opposed to a
  multilingual comparable corpus composed of
  comparable texts for different languages for
  cross-linguistic contrast

8
Comparative corpora

• Corpora containing different regional varieties
  of the same language are not comparable corpora
• E.g. the International Corpus of English (ICE),
  the Brown family of corpora
• All corpora, as a resource for linguistic
  research, have always been pre-eminently suited
  for comparative studies (Aarts 1998 ix), either
  intralingually or interlingually
• Corpora of this kind are comparative corpora

9
Use of parallel comparable corpora

• Parallel and comparable corpora offer specific
  uses and possibilities for contrastive and
  translation studies (Aijmer Altenberg 1996 12)
• - they give new insights into the languages
  compared insights that are not likely to be
  gained from the study of monolingual corpora
• - they can be used for a range of comparative
  purposes and increase our knowledge of
  language-specific, typological and cultural
  differences, as well as of universal features
• - they illuminate differences between source
  texts and translations, and between native and
  non-native texts
• - they can be used for a number of practical
  applications, e.g. in lexicography, language
  teaching and translation.

10
Use of parallel comparable corpora

• Used primarily for translation and contrastive
  studies
• The two types of corpora have their own
  characteristics, and serve different purposes
• Parallel corpora are useful in translation
  studies, but they alone serve as a poor basis for
  cross-linguistic contrast, because translations
  cannot avoid the effect of translationese
• Comparable corpora are well suited for
  contrastive research, but are less useful in
  translation studies

11
Using corpora in translation studies

• Parallel corpora
• Useful in exploring how an idea in one language
  is conveyed in another language, thus providing
  indirect evidence to the study of translation
  processes
• Indispensable for building statistical or
  example-based machine translation (EBMT) systems,
  and for the development of bilingual lexicons and
  translation memories
• Parallel concordancing is a useful tool for
  translators
• Comparable corpora
• Useful in improving the translators
  understanding of the subject field and improving
  the quality of translation in terms of fluency,
  correct term choice and idiomatic expressions in
  the chosen subject field
• Can also be used to build terminology banks

12
Using corpora in translation studies

• Translational corpora
• Provide primary evidence in product-oriented
  Translation Studies, and in studies of
  translation universals
• If corpora of this kind are encoded with
  sociolinguistic and cultural parameters, they can
  also be used to study the sociocultural
  environment of translations (e.g. functions of
  translation in DTS)
• Monolingual corpora (source / target language )
• Raising the translators linguistic and cultural
  awareness in general
• Providing a useful and effective reference tool
  for translators
• In combination with a parallel corpus to form a
  so-called translation evaluation corpus that
  helps translator trainers or critics to evaluate
  translations more effectively and objectively

13
Corpus-based translation studies

• Laviosa (1998a)
• the corpus-based approach is evolving, through
  theoretical elaboration and empirical
  realisation, into a coherent, composite and rich
  paradigm that addresses a variety of issues
  pertaining to theory, description, and the
  practice of translation.
• Hypotheses that translation universals can be
  tested by corpus data (Baker 1993, 1995)
• Rapid development of corpus linguistics, esp.
  multilingual corpus research in the early 1990s
• Increasing interest in Descriptive Translation
  Studies (Toury 1995)
• Tymoczko (1998)
• Corpus Translation Studies is central to the way
  that Translation Studies as a discipline will
  remain vital and move forward.
• Meta 43/4 (1998) Kenny (2001) Laviosa (2002)
  Granger et al (eds.) (2003) Olohan (2004)
  Mauranen et al (eds.) (2004) Kruger Munday
  (ed.) (2011) Hu (2011), Wang (2011), Xiao (2012)

14
The Holmes-Toury map

• Applied Translation Studies
• Descriptive Translation Studies
• Theoretical Translation Studies

15
Applied Translation Studies

• Three major contributions of corpora
• Corpus-assisted translating
• Bowker (1998 631) corpus-assisted translations
  are of a higher quality with respect to subject
  field understanding, correct term choice and
  idiomatic expressions.
• Corpus-aided translation teaching and training
• Bernardini (1997) large corpora concordancing
  (LCC) can help students to develop awareness,
  reflectiveness and resourcefulness, which are
  said to be the skills that distinguish a
  translator from those unskilled amateurs
• Development of translation tools
• Corpora, and especially aligned parallel corpora,
  are essential for the development of translation
  technology such as machine translation (MT)
  systems, and computer-aided translation (CAT)
  tools

16
Descriptive Translation Studies

• Characterized by its emphasis on the study of
  translation per se
• It is to answer the question of why a translator
  translates in this way instead of how to
  translate
• Baker (1993) predicted that the availability of
  large corpora of both source and translated
  texts, together with the development of the
  corpus-based approach, would enable translation
  scholars to uncover the nature of translation as
  a mediated communicative event

17
Descriptive Translation Studies

• Three focuses (Holmes 1972/1988)
• Translation as a product
• Concerned with describing translation as a
  product by comparing corpora of translated and
  non-translational native texts in the target
  language
• Attempting to uncover evidence to support or
  reject the so-called translation universal
  hypotheses
• Translation as a process
• Aims at revealing the thought processes that take
  place in the mind of the translator while she or
  he is translating
• One possible way for corpus-based DTS is to
  investigate the written transcripts of these
  recordings off-line, which is known as
  Think-Aloud Protocols (or TAPs)
• Translation as product providing indirect
  evidence to translation as process
• The function of translation
• The study of contexts rather than texts function
  or impact of a translation
• Relatively few function-oriented studies that are
  corpus-based

18
Theoretical Translation Studies

• Aims to establish general principles by means of
  which these phenomena can be explained and
  predicted (Holmes 1988 71)
• Closely related to, and often reliant on the
  empirical findings produced by Descriptive
  Translation Studies
• One good battleground of using DTS findings to
  pursue general theory of translation is the
  hypothesis of so-called translation universals
  (TUs) and its related sub-hypotheses
• Sometimes referred to as the inherent features of
  translational language, or translationese

19
TU A focus of CBTS

• An important area of corpus-based TS over the
  past decade
• Baker (1993, 1996) Chesterman (2004) Kenny
  (1998, 1999, 2000, 2001) Laviosa (1998b)
  Mauranen Kujamaki 2004) McEnery Xiao (2002,
  2007) Olohan (2004) Olohan Bakers (2000)
  Øverås (1998) Pym (2005) Xiao and Yue (2008),
  Xiao (2010), Xiao Dai (2010), Xiao (2010, 2011,
  2012)
• The Translational English Corpus (TEC)
• Manual
• http//www.llc.manchester.ac.uk/ctis/research/engl
  ish-corpus/
• Software
• http//ronaldo.cs.tcd.ie/tec2/jnlp/

20
Features of translated English

• Laviosa (1998b) Four core patterns of lexical
  use in translational English
• - A relatively low proportion of lexical words
  over function words
• - A relatively high proportion of high-frequency
  words over low-frequency words
• - A relatively great repetition of the most
  frequent words
• - Less variety in most frequently used words

21
Features of translated English

• Beyond the lexical level
• Simplification tendency to simplify the
  language used in translation (Baker 1996
  181-182)
• simpler language than target native language
  lexically / syntactically / stylistically
• Normalization tendency to exaggerate features
  of the target language and to conform to its
  typical patterns (Baker 1996 183)
• more normal than the target native language
• Explicitation tendency in translations to spell
  things out rather than leave them implicit
  (Baker 1996 180)
• more frequent use of conjunctions, and increased
  cohesion in translated text
• Sanitization translated texts are somewhat
  sanitized versions of the original (Kenny
  1998 515)
• Lost or reduced connotational meaning in
  translation
• TU hypotheses

22
TU A target of debate

• Is translational language different from target
  native language?
• Translational language is at best an
  unrepresentative special variant of the target
  language because translations cannot possibly
  avoid the effect of translationese
• e.g. Baker 1993 Gellerstam 1996 Hartmann 1985
  Laviosa 1997 McEnery Wilson 2001 McEnery
  Xiao (2002, 2007) Teubert 1996

23
TU A target of debate

• Are the features uncovered on the basis of
  translational English generalizable to other
  translated languages?
• Existing evidence has largely come from
  translational English and related European
  languages
• If such features are to be generalized as
  translation universals, the language pairs
  involved must not be restricted to English and
  closely related languages
• Cheongs (2006) study of English-Korean
  translation contradicts even the least
  controversial explicitation hypothesis
• Evidence from genetically distinct language
  pairs such as English and Chinese is undoubtedly
  more convincing, if not indispensable

24
The ZCTC corpus

• Created with the explicit aim of studying the
  features of translated Chinese
• A translational counterpart of the Lancaster
  Corpus of Mandarin Chinese (LCMC), a
  one-million-word balanced corpus of native
  Chinese (McEnery Xiao 2004)
• www.ling.lancs.ac.uk/corplang/lcmc/
• Five hundred 2,000-word text samples taken
  proportionally from fifteen written text
  categories published in China in the 1990s
• www.ling.lancs.ac.uk/corplang/ZCTC/

25
LCMC / ZCTC corpus design
26
ZCTC vs. LCMC
27
Corpus markup and annotation

• CES-compliant XML
• CES www.cs.vassar.edu/CES/
• Tokenization and POS tagging
• ICTCLAS2008 www.ictclas.org
• A precision rate of 98.54 for tokenization
• Paragraph, sentence, word token
• Encoded in Unicode (UTF-8)

28
Core patterns of lexical use

• Do the core patterns of lexical use Laviosa
  (1998b) observes in translational English also
  apply in translated Chinese?
• Same criteria and parameters as in Laviosa
  (1998b)
• Lexical density
• Frequency profiles
• Mean sentence length

29
Lexical density

• The Stubbs-style lexical density the ratio
  between the number of lexical words (i.e. content
  words) and the total number of words (Stubbs
  1986 33 1996 172)
• Measure of informational load
• Adopted in Laviosa (1998b)
• Lexical density measured by TTR or Standardized
  TTR (STTR) (Scott 2004)
• Measure of lexical variability
• Commonly used in Corpus Linguistics

30
Stubbs-style lexical density

• Mean LD is significantly greater in native than
  translational corpus (66.93 vs. 61.59, t
  -4.94, plt0.001)
• In addition, the native Chinese corpus displays a
  greater LD score in all of the 15 genres and
  significant for nearly all genres (except for M)
• Translations make more frequent use of function
  words

31
Standardized TTR

• Mean STTR is slightly greater in native than
  translation corpus (46.58 vs. 45.73) not
  significant (t -0.573, p0.571)
• The differences in most genres are also marginal
• Greater STTR scores can be found in both native
  (e.g. A) and translated (C) Chinese genres

32
Lexical-function ratio Stubbs LD

• Mean ratio between lexical and function words is
  significantly greater in native than
  translational corpus (2.08 vs1.64, t -4.88,
  plt0.001)
• Also, native Chinese has a greater ratio in all
  genres, and the differences are significant in
  nearly all genres (except for M)
• In line with Laviosas (1998b) initial hypothesis
  that translational language has a relatively low
  proportion of lexical words over function words

33
Frequency profiles of LCMC/ZCTC

• Laviosas (1998b) list head or high frequency
  words
• Wordlist items which individually account for at
  least 0.10 of the total number of tokens in a
  corpus
• The same criterion for high frequency words in
  this study to ensure comparability

34
Frequency profiles

• The numbers of high frequency words are very
  similar in the two corpora
• High frequency words account for a considerably
  greater proportion of tokens in the translational
  corpus (40.47 vs. 35.70)
• High frequency words display a much greater
  repetition rate in translated Chinese (3154.37
  vs. 2870.37)
• Also the ratio between high- and low-frequency
  words is greater in the translational corpus
  (0.6988 vs. 0.5659)

35
Mean sentence length vs. simplification

• Conflicting observations of mean sentence length
  as an indicator of simplification (e.g. Laviosa
  1998b vs. Malmkjaer 1997)
• The native Chinese corpus (LCMC) shows a
  marginally greater mean sentence length not
  significant (t - 1.41, p 0.17)
• Mean sentence length is sensitive to genre
  variation and may not be reliable as an indicator
  of simplification in translational Chinese
• (Mean sentence segment length)

36
Lexical use in translational Chinese

• Summary
• - Analysis of lexical density and frequency
  profiles shows that the four core patterns of
  lexical use in translational English are
  essentially also applicable in translated Chinese
• - But mean sentence length is less reliable as
  an indicator of simplification in translational
  Chinese

37
Explicitation Connectives as a device?

• Perhaps the most studied topic in TU research and
  also the least controversial hypothesis
• Chen (2006)
• Connectives are a device for explicitation in
  English-Chinese translation of popular science
  books
• Xiao and Yue (2008)
• Connectives are significantly more frequent in
  translational than native Chinese fiction
• Question
• Can we generalize this finding from these
  specific genres to Mandarin Chinese in general?

38
Conjunctions in ZCTC and LCMC

• Mean frequency of conjunctions is significantly
  greater in translational than native corpus
• 306.42 and 243.23 instances per 10,000 tokens,
  LL723.12 for 1 d.f., plt0.001
• In addition, genre-based distribution shows that
  most genres covered in the corpora display a
  significantly more frequent use of conjunctions
  in translational Chinese in spite of some
  genre-based subtleties (e.g. F, J)

39
Conjunctions of different frequency bands

• More conjunction types of high frequency bands
  (0.01 or above) are used in translational corpus
• There are an equal number of conjunction types
  (56 types) of medium frequency band (0.005) in
  translational and native corpora
• Beyond this balance point, the native corpus
  displays a greater number of conjunction types of
  low frequency band (0.001 or below)
• In line with observations about high vs. low
  frequency words

40
Conjunctions of different styles

• A closer comparison of the lists of frequent
  conjunctions (0.001 or above) in their
  respective corpus also sheds some new light on
  the simplification hypothesis
• There are 91 and 99 types of frequent
  conjunctions in the two corpora 86 items
  overlap in the two lists
• Conjunctions on the translational but not native
  list are all informal, colloquial, and simple ,
  which usually have more formal alternatives (e.g.
  ?? for ??,?? for ????)
• Conjunctions on the native but not translation
  list are typically formal, literate and archaic
  (e.g. ????????????????????????)
• These results provide evidence for the
  simplification hypothesis but against the
  normalization hypothesis

41
Conclusions

• Results based on two comparable Chinese corpora
  suggest that the core patterns of lexical use in
  translational English are generally also
  applicable in translated Chinese
• Beyond the lexical level
• Mean sentence length is sensitive to genre
  variation and may not be reliable as an indicator
  of simplification
• A comparison of frequent conjunctions in native
  and translated Chinese shows that simpler forms
  tend to be used in translations
• In spite of some genre-based subtleties,
  conjunctions are more frequently used in
  translational Chinese, which provides evidence in
  favour of the explicitation hypothesis
• Corpus Translation Studies is a promising area of
  research
• ??????????????????????,?????????,2012

42
CUC ParaConc

• Software demo

43

• Shedding of valedictory tears
• lancsxiaoz_at_googlemail.com
• xiaoz_at_zju.edu.cn