Presentation on theme: "Using Corpora in Translation Studies Research"— Presentation transcript:
1Using Corpora in Translation Studies Research Dorothy KennySALIS/Centre for Translation and Textual Studies, Dublin City University
2Outline Part One: General introduction to corpora Part Two: What they’re good forHow to get your hands on onePart Two:Corpora in translation studies research
3To warm you up: What does ‘cause’ (verb) mean? What does ‘provide’ mean?How often is the passive voice used in English?Handouts
4Tell me something I don’t know… Not all facts about language are accessible through native-speaker intuitionIt’s often impossible to describe meanings of words without reference to the contexts those words occur inSometimes you need a lot of data to see patternsSometimes you need to count
5What a corpus can help us see… cause has negative ‘semantic prosody’provide has positive ‘semantic prosody’(see Louw 1993; Stubbs 1995)in general, passive is used around 10% of the time in English (see Halliday 1991)
6Corpusa collection of naturally-occurring texts that are the object of literary or linguistic studyheld in electronic formthus amenable to (semi-)automatic processingusually assembled in some principled wayoften highly structured, containing very carefully selected textscompiled for a purposeeg linguistic analysis, lexicography, Natural Language Processing, translation research…or a by-product of another activityeg parliamentary debates (and their translations in bilingual or multilingual parliaments)
7Well-known monolingual corpora British National Corpus (100m words)The Bank of English (524m words)free access to a sample of 56m words atThe entire World Wide Web
8Corpora and corpus processing software Provide:a (relatively) objective basis for commentary on linguistic phenomena, or linguistic realizations of, eg, social phenomenaa resource for quantitative and qualitative researchthe ability to access and manipulate (sort, display, annotate) vast quantities of data, thus facilitating analysis by humans
9Access to corpora web access purchase own copy (eg BNC) Bank of English, BNC,IDSfree access to ‘sampler’access to larger corpus by subscription/permissionpurchase own copy (eg BNC)create your own corpus
10Web access to corpora: corpus processing software Web-accessible corpora have dedicated interfaces that allow users, eg, to:choose texts/sub-corporasearch the corpus for instances of a particular word (results in KWIC concordance format)display and sometimes sort results
13Web access to corpora: corpus processing software More sophisticated interfaces, eg, Cosmas I* (IDS in Mannheim), allowed searches for:all forms of a lemmagehen gets gehe, ging, gehst, geht, etc.all compounds formed from a search wordMöbel gets Gartenmöbel, Möbellager, etc.all forms derived from a search wordKind gets Kindchen, kindlich, etc.*replaced in 2003 by Cosmas II
14Building your own corpus - basic design issues Written vs spoken vs written and spokenStatic vs dynamic (monitor)Synchronic (time of production?) vs diachronicGeneral reference vs specialisedMonolingual vs bilingual vs multilingualDomains to be coveredText types to be coveredLevel of annotation: raw vs annotation with extra-textual info (headers to indicate text title, author, speakers, etc) vs detailed linguistic annotation
15Raw corpus/clean text Example: As he weakened, Moran became afraid of his daughters. This once powerful man was so implanted in their lives that they had never really left Great Meadow, in spite of jobs and marriages and children and houses of their own in Dublin and London. Now they could not let him slip away.from John McGahern’s Amongst Women
17Building your own corpus - Getting the Texts Full texts vs text samplesSamplingrandom Sampling or handpicking?Copyright PermissionRepresentativenessdifficult concept to apply to textual dataonus is on researcher to document corpus contents very carefully
18Conversion to electronic form? Text not available in electronic form:Scanning + OCR vs KeyboardingText available in electronic formWeb pages (see Kilgarriff et al 2006)Downloads from websitesFull-text databasesDonations from authors/translatorsFormat texts should be saved in?.txt? .html? .xml? (know what your software can handle!)Is alignment necessary?
19Doing Research Using Corpora Corpora can help us answer questions, but they also generate questionsIn general, passive is used around 10% of the time in English; but nearly 20% of the uses of cause as a verb in my scientific corpus are in the passive voiceDoes this tell me something aboutthe verb cause?scientific English?my sample of scientific English?
20Hypothesis generation and testing Hypothesis – a tentative claimeg cause tends to be used more in the passive than other English verbshypotheses sometimes emerge from corpus datacorpus-driven researchhypotheses sometimes formed before we look at corpus datacorpus-based researchnewly generated hypotheses can be tested against other (often bigger) corporacycles of hypothesis formation, testing, refinement, testing… common in corpus research
21Some early descriptive hypotheses in translation studies Translations tend to be simpler/more explicit/more conventional than:other texts in the same languagetheir source textsBut how do you operationalize notions like simplification, explicitation, normalization in corpus-based research?Eg, what concrete features of a text show it to be ‘simpler’ than another text?(See especially, Baker 1993, 1995, 1996)
22Early phase: Developing methodologies and resources Eg Question: Are translations really ‘simplified’?Eg Answer: Compared to what?source texts?=> parallel corpus methodologyother texts in the target language?=> comparable corpus methodology
23Parallel Corpusset of source texts in language A alongside their translations into language B (and perhaps languages C, D, E …)can be bilingual or multilingualeg English-Norwegian Parallel Corpus (ENPC)eg European Parliament Proceedings Parallel Corpuscan be unidirectional or bidirectionaleg German-English Parallel Corpus of Literary Texts de->eneg ENPC en->no and no->en
24Comparable Corpus Monolingual Comparable Corpus set of texts translated into a language A alongside texts originally written in that same languageTranslational English Corpus (TEC)+ comparable subset of the BNCFinnish Comparable Corpus (Mauranen 2004)
25Early Corpus-based TS: Quantitative Bias? Attempt to approach translation objectivelyReliance on properties of text that can be measuredAverage word lengthAverage sentence lengthLexical densityType-token ratio, etcFocus on monolingual comparable corporaFocus on general tendencies (or ‘universals’) in translation (Laviosa 2002; Mauranen and Kujamäki 2004)
26Lexical Densitythe ratio of content words to the total number of words in a textI made the chemicals hotter so they changed.(4/8 = 50%)Raising the temperature produced a chemical change.(5/7 = 71%) (from Gibbons 2003:20)the lower the lexical density, the ‘simpler’ the text.
27Type-Token Ratio the ratio of types to tokens in a corpus eg the TTR for The cat sat on the mat near the log fire is 8/10 (or 80%)assumed to measure the (lexical) variety in a text:the higher the TTR, the more varied the text’s vocabulary
28TTR: An ExampleIn this piece I rail against the tendency of linguists to write about the philosophy of science as applied to their subject field instead of writing about what languages are like, which is what linguists are supposed to be good at. Unsympathetic critics will no doubt charge that by doing so I instantiate the very kind of behavior that I am railing against.From Pullum 1991:123TTR = 49/63 = 78%
29TTR continuedIn this piece I rail against the tendency of linguists to write about the philosophy of science as applied to their subject field instead of writing about what languages are like, which is what linguists are supposed to be good at. Unsympathetic critics will no doubt charge that by doing so I instantiate the very kind of behavior that I am railing against. This is not so. I am complaining about unproductive metalevel discussion, which consists of linguists talking about doing linguistics instead of doing it. By offering a critique of such work, I am operating at a meta-metalevel, talking about linguists talking about doing linguistics instead of doing it. There is a difference.TTR = 66/115 = 57%
30Type/Token Ratio is extremely sensitive to text length the longer the text is, the lower the TTR is (normally)solution: calculate the TTR for successive chunks of texts (eg every 100 or 1,000 words), and then take an average count at the end (standardized TTR)but even standardized TTRs are problematic as a marker of simplicity/complexity as they can’t capture the difference between hard words and easy wordseg fire vs conflagration
31Operationalizing ‘Simplification’ If translated texts are somehow ‘simpler’ than original texts in the same language, then they might have:shorter average word lengthshorter average sentence lengthlower lexical densitylower standardized type-token ratioscompared to originals…
33Initial Results: Investigating Simplification (Laviosa 1998a) Corpus: newspaper articles translated into English, and newspaper articles originally written in EnglishTool: WordList in WordSmith ToolsResultslexical density: lower in translated articlesaverage sentence length: lower in translated articlesstandardized type/token ratio: no significant difference
34Initial Results: Investigating Simplification (Laviosa 1998b) Corpus: fiction translated into English, and fiction originally written in EnglishTool: WordSmith Tools WordListResultslexical density: lower in translated articlesaverage sentence length: higher in translated articlesstandardized type/token ratio: no significant differencebut translations make greater use of high frequency words than do non-translations = core patterns of lexical use
35Combining Quantitative and Qualitative Research Investigating Explicitation (Olohan and Baker 2000)Hypothesis: Translators’ (unconscious?) tendency to be more explicit than other writers in the same language will be visible in their greater use of optional explicitating grammatical elementsEg: that after ‘He says that Mozart was a wonderful composer’Compare: ‘He says Mozart was a wonderful composer’Corpus: TEC and BNC
36Optional that as seen in concordance lines Jeux à deux is the circle, which is to say that the duration of the game has nenocuous actions in reality, which is to say that the distance must also be maintsped you hideously. Rock of branches, say that it roars. The fields, as the fue ink, Yevgenia Wdowin, but it doesn't say that the eggs in the basement are gosk what smells so awful, and it doesn't say that my mother hasn't uttered a wordad in mind for you, then the people who say that Russians stink are right." Dearupstairs to wait for a moment. He is to say that the stairs are blocked becausen his cigar. It occurred to Vogtmann to say that an apt mode of expression avoidcompany with her share. Today she could say that she had made the right decisionmust change its owner. I believe I can say that Herr Urban's thoughts are alson to express our concern. I think I may say that he shares our view.' 'I've gotcking off immediately, I just wanted to say that you would get all the support ye was not there, but had left a note to say that she had driven to the hospitalhappened to Franz prompts me to lie. I say that he has gone off to work in hisorm of institution. What I am trying to say, though, is that this institution istecting suspicion, although he couldn't say that he made a deliberate habit of
37Investigating Explicitation (Olohan and Baker 2000; Olohan 2003, 2004) Optional that in reporting structures built around SAY and TELLthat far more common in translated English than in original Englishfactors conditioning inclusion of that:formality; matrix verb; potential ambiguity;long adverbials; cognitive complexityattempt not just to describe, but also to explain the specificities of translation
38But what about the Source Texts? (Kenny 2005) Can patterns of inclusion/exclusion of optional that be influenced by patterns for dass in German source texts?Oberflächlich zufriedengestellt sagte er, dass er sowieso Zeitungen kaufen und telefonieren wolle.Superficially satisfied, he said that he wanted to buy newspapers and make a telephone call anyway.Corpus – German-English Parallel Corpus of Literary Texts (GEPCOLT)Looked at SAY + zero connective vs SAY+ that
39SAY + that vs zero-connective: Results (Kenny 2005) optional that used with SAY in 157 cases78 optional uses of that with SAY coincide with a dass in the ST79 coincide with a zero-connective in the STIf you see that, there’s a 50/50 chance that dass was in the STzero-connective used with SAY in 219 cases46 coincide with a dass in the ST173 coincide with a zero-connective in the STIf you don’t see that, then eight out of ten times there was no dass in the ST either
40SAY + that vs zero-connective: Results (Kenny 2005) patterns of omission of that follow patterns of omission of dass; but patterns of inclusion of that cannot be explained by the STdass to 0 shifts do happen (46), but they’re not as common as 0 to that shifts (79)overall tendency is towards grammatical explicitation
41From the General to the Particular Much CTS focuses on general patterns in translated text (see Mauranen and Kujamäki 2004)Search for generalizations inevitably leads to the recognition of particularities, eg the distinctive behaviour of individual translatorsTranslators’ individual styles come into focusBaker (2000), Kenny (2001), Saldanha (2004, 2005)Winters (2004, 2005), Olohan (2004:Chapter 8), Bosseaux (2007)
42General Features of Translation vs Features of Translator Style Eg Kenny (2001)focuses on lexical creativity/normalisationuses German-English Parallel Corpus of Literary Textsuses frequency-ranked word list (WordSmith) to findcreative once-off words in the STs, andwords or clusters of words peculiar to one writerUses concordancer (WordSmith) to find creative collocations (node=Auge ‘eye’)Uses parallel concordancers (Multiconcord) to find translations in the TTs
43Clusters: Sample Results for John Brownjohn (from Natascha Wodin’s (1983) Die Gläserne Stadt) Und plötzlich im Parterre das Schrillen des Telefons. Helmut. Um diese Zeit konnte nur er es sein.Suddenly the phone rang—Helmut--it couldn't be anyone else at this hour.Das Schrillen des Telefons. Nur zwanzig Minuten diesmal. Ihre Anmeldung Moskau. L's russisches Allo ... die rauhe Oberfläche einer Birne.The phone rang. Only a twenty-minute delay this time. Your call to Moscow, L's Russian "Allo?"--husky as only his voice could be.Das Schrillen des Telefons unter der Bettdecke.,The telephone rang beneath the bedclothes..Das Schrillen des Telefons unter der Bettdecke.Again the telephone rang beneath the bedclothes.
44Writer-specific forms: Sample Results for Malcolm Green (from Gerhard Roth (1986) Am Abgrund) Da kommt der Staatsbeamte herein und verlangt den Herrn Irrenwäscher zu sprechen.Then the government official enters and asks to speak to the head madmen-washer.Der Herr Irrenwäscher kommt auf einem Fahrrad angeradelt und hat eine Fetten.The head madmen-washer arrives by bicycle, he's as pissed as a fart.»Das weiß ich selber«, gibt der Herr Irrenwäscher frech zurück"You're telling me!" the madmen-washer replies impudently.Der Herr Irrenwäscher ist ein fülliger, behäbiger Mann.The head madmen-washer is a portly, comfort-loving man.
45Homing in on the translator’s voice Whose voice do we hear in translation? Whose point of view do we share? The author’s? The translator’s? Both?Can we develop methodologies that allow us to show systematic differences between translators, and thus the specific voice of each individual translator? (handout)Bosseaux (2007),Saldanha (2005), Winters (2004, 2005)integrate narratological structure, deixis, modality, typography, etc into data- and theory-rich studies of the translator’s presence in translation.B. and W. look at different translations of the same source text(s)S. looks at several translations by two translators of different source texts, but from the same source langauge(s)
46Winters 2005 Bilingual parallel corpus One ST Two TTs F Scott Fitzgerald’s (1922)The Beautiful and DamnedTwo TTsboth into Germanboth published in 1998TT1 by Hans-Christian OeserTT2 by Renate Orth-GuttmannFocuses on aspect of the TL that does not have a counterpart in the SL – German modal particlesResearch initially data-driven
47Winters 2005FSF: He considered, nevertheless, that he had given her an object-lesson and that the matter was closed…ROG: Immerhin, sagte er sich, habe er ihr eine Lektion erteilt und damit sei der Fall wohl erledigt.HCO: Nichtsdestoweniger war er der Meinung, daß er ihr einen Denkzettel verpaßt hatte und daß die Sache damit erledigt war…
48Bennett (1994) distinguishes: Interplay between corpora and theoretical translation studies: Translation Units (Kenny 2004 & forthcoming)Bennett (1994) distinguishes:translation atomstranslation focustext as macro-unitFor technologists working in Natural Language Processing translation units are segments of ST aligned with segments of their TT
49Translation Units DTS approach (Toury 1995) Coupled pairs of ST and TT segments whose boundaries are identified by the analyst, according to ‘no-leftovers principle’‘Thus, the analyst will go about establishing a segment of the target text, for which it would be possible to claim that – beyond its boundaries – there are no leftovers of the solution to a translation problem which is represented by one of the source text’s segments, whether similar or different in rank and scope.’ (Toury 1995:78-79).
50Translation Units in NLP (Kraif 2003) Lexical correspondencesStable semantic equivalentsPairing in a corpus a regular occurrenceUseful for bilingual lexicon extractionTranslation EquivalentsContext-bound, possibly once-off pairingsConsidered as ‘noise’ in bilingual lexicon extraction
51Looking for Translation Units in a Parallel Corpus Interested in :extended units of meaning and translation unitscontext ‘independence’ and context dependencetext as macro-translation unitharnessing potential of corpora to uncover groups of related instancesAssume:TUs are mutually defining ST-TT segmentssearch will begin in one language but shuttle back and forth between the SL and the TLvariable TUs (though not normally exceeding clause)
52Concordance of mit aller Kraft in Gepcolt ff angefroren sein mußte. Sie versuchte mit aller Kraft, bei klarem Verstand zun aus den Gedanken, in denen er steckt, mit aller Kraft herausziehen, doch selbse Stopfnadel aus meinem Handarbeitszeug mit aller Kraft, und wenn ich sage Kraftahrtendolch in Pension ist, und posiert mit aller Kraft wie auf Sophies Bruderfohließend ersticht er die tote Schwester mit aller Kraft. Dann ist er endlich damder Kurt Lukas war, schien sein Messer mit aller Kraft in die Hündin zu treibenhle Bäume vor dem Feld, der Mann rannte mit aller Kraft, das sah ich an den gewötarre der Meereisdecke ziehen die Hunde mit aller Kraft nach der nächstgelegenenultern gegen die Matratze, stemmte mich mit aller Kraft gegen seinen Amoklauf, ehte lebenden Embryo-Gesichter, scheinen mit aller Kraft von der Sonne angezogeneren Glas-Röhrchen auf ihr Blut warten, mit aller Kraft und Empörung auf den Fuß
53Mit aller Kraft examples (Tables 1 and 2) What can we say about the:boundaries of TUs?No left-overs? Attentional focus on ST? Macro-unit?stability of translation equivalents/lexical correspondences?context (or co-text) dependence
54Context DependenceEri wehrte sich zwar mit aller Macht dagegen, verurteilt zu werdenhei resisted being convicted with all hisi mightSiej stößt mit dem Stuhl mit aller Gewalt gegen die Türe.Shej beats the chair against the door with all herj mightman hieb auf ihn von der Seite ein, mit aller Wuchthe was being lashed by someone at his side with the utmost weight and force
55Text as Macro-Unit (from Wodin 1983/1986) Sein gebräuntes Gesicht lacht in den Falten. Das Lachen eines Apfeldiebes. (Par. 125)His tanned face was creased with laughter, the laughter of a boy stealing apples.L mit seinem neuen, schlanken, verjüngten Körper, mit seinem alten Apfeldieblachen im Gesicht. (Par. 582)I couldn’t take my eyes off L, with his new, slim, rejuvenated body and his old, inimitable, mischievious smile.
56Translation Units - Conclusions ST segment subject to attentional focus, translation atoms and translation macro-unit can all differ, even on a single occasionContext dependence can’t be assumed to indicate a once-off idiosyncratic translation; sometimes it’s the norm
57ConclusionsCorporaprovide a more objective basis for studies of translation, statements about translator behaviourallow us to zoom in and out, from general to particular and back againallow synthesis of different approaches (cognitive, empirical, etc)help us to refine key notions in translation theory
58The Future?Better contextualization of studies/results, integrating other sources of dataMore sophisticated methodologies and analysesMore varied corpus typesEg multimodal corpora – audiovisual corpora, sign language corpora, etcMainstreaming of basic corpus processing
59References.Baker, Mona (1993) ‘Corpus Linguistics and Translation Studies. Implications and Applications’, in Mona Baker, Gill Francis and Elena Tognini-Bonelli (eds), Text and Technology: In Honour of John Sinclair Amsterdam and Philadelphia: John BenjaminsBaker, Mona (1995) ‘Corpora in Translation Studies: An Overview and Some Suggestions for Future Research’, Target 7(2):Baker, Mona (1996) ‘Corpus-based Translation Studies: The Challenges that Lie Ahead’, in Harold Somers (ed) Terminology, LSP and Translation: Studies in Language Engineering, in Honour of Juan C. Sager Amsterdam and Philadelphia: John Benjamins,Baker, Mona (2000) Towards a Methodology for Investigating the Style of a Literary Translator’ in Target 12(2):Bennett, Paul (1994) ‘The Translation Unit in Human and Machine’ Babel 40:1,Bosseaux, Charlotte (2007) How Does it Feel? Point of View in Translation Amsterdam/New York: Rodopi.Gibbons, John (2003) Forensic Linguistics. An Introduction to Language in the Justice System Oxford: Blackwell Publishing.Haller, Helmut, Roeland Van Hout and Jeanine Treffers-Daller ‘Lexical Richness in the Spontaneous Speech of Bilinguals’ in Applied Linguistics 24(2):Kenny, Dorothy (2001) Lexis and Creativity in Translation: A Corpus-based Study Manchester: St. Jerome.Kenny, Dorothy (2004) ‘Die Übersetzung von usuellen und nicht unusuellen Wortverbindungen vom Deutschen ins Englische’ in Kathrin Steyer (ed) Wortverbindungen – mehr oder weniger fest. Institut für Deutsche Sprache Jahrbuch 2003 Berlin/New York: Walter de Gruyter,Kenny, Dorothy (2005) ‘Parallel Corpora and Translation Studies: old questions, new perspectives? Reporting that in Gepcolt. A case study’ in Geoff Barnbrook, Pernilla Danielson and Michaela Mahlberg (eds) Meaningful texts: the extraction of semantic information from monolingual and multilingual corpora London/New York: Continuum,Kenny, Dorothy (forthcoming) ‘Translation Units and Corpora’ in Alet Kruger and Kim Wallmach (eds) Corpus-based Translation Studies: More Research and Applications Manchester: St. JeromeKilgarriff, Adam, Michael Rundell and Elaine Uí Dhonnchadha (2006) ‘Efficient corpus development for lexicography: building the New Corpus for Ireland’, Language Resources and Evaluation 40(2):Kilgarriff, Adam and Gregory Grefenstette (2003) ‘Web as Corpus’, Computational Linguistics 29(3):Kraif, Olivier (2003) ‘From translational data to contrastive knowledge’ International Journal of Corpus Linguistics 8:1, 1-29.
60References continuedLaviosa, Sara (1998a) ‘The English Comparable Corpus: A Resource and a Methodology’ in Lynne Bowker, Michael Cronin, Dorothy Kenny and Jennifer Pearson (eds), Unity in Diversity? Current Trends in Translation Studies Manchester: St. Jerome,Laviosa, Sara (1998b) ‘Core patterns of lexical use in a comparable corpus of English narrative prose’ Meta 43(4):Laviosa, Sara (2002) Corpus-based Translation Studies. Theories, Findings, Applications Amsterdam/New York: Rodopi.Mauranen, Anna (2004) ‘Corpora, universals and interference’, in Anna Mauranen and Pekka Kujamäki (eds) Translation Universals. Do they exist?, Amsterdam and Philadelphia: John Benjamins,Mauranen, Anna and Pekka Kujamäki (2004) (eds) Translation Universals. Do they exist? Amsterdam and Philadelphia: John Benjamins.Olohan, Maeve (2003) ‘How frequent are the contractions? A study of contracted forms in the Translational English Corpus’ Target 15:59-89.Olohan, Maeve (2004) Introducing Corpora in Translation Studies London and New York: Routledge.Olohan, Maeve and Mona Baker (2000) ‘Reporting that in Translated English: Evidence for Subconscious Processes of Explicitation?’ Across Languages and Cultures 1:Saldanha, Gabriela (2004) 'Accounting for the Exception to the Norm: a Study of Split Infinitives in Translated English', Language Matters 35(1):39-53.Saldanha, Gabriela (2005) The Translator's Style: A Corpus-based Exploration Unpublished PhD thesis. Dublin: Dublin City University.Toury, Gideon (1995) Descriptive Translation Studies and Beyond Amsterdam and Philadelphia: John Benjamins.Winters, Marion (2004) ‘German Translations of F. Scott Fitzgerald’s The Beautiful and Damned – A Corpus-based Study of Modal Particles as Features of Translators’ Style’. In: Ian Kemble (ed.) Using Corpora and Databases in Translation Portsmouth: University of Portsmouth,Winters, Marion (2005) A corpus-based study of translator style: Oeser’s and Orth-Guttmann’s German translations of F. Scott Fitzgerald’s The Beautiful and Damned Unpublished PhD thesis. Dublin: Dublin City University.