Dictionary, corpus and CAT tool use in legal translation: A comparative pilot study of student translations

ABSTRACT Theoretical background In an early study conducted by Bowker (1998) it was shown that the use of monolingual original corpora in the target language, viz. corpora containing texts in one language produced by native speakers, leads to an increase in translation quality, including understanding of the subject field, selecting correct terminology and idiomatic expressions (Bowker, 1998, p. 648). This may be due to the fact that these monolingual original corpora do not contain translated language, and are therefore uninfluenced by so-called translationese. However, since Bowker (1998) the impact of monolingual original corpora on translation quality has not been explored further to any great extent. In addition, survey research has indicated a low level of corpus awareness among student and professional translators (Zaretskaya et al., 2015, p. 250), which can be attributed to the lack of training among future translators for the use of corpus tools (e.g. WordSmith, AntConc, Sketch Engine), the unavailability of ready-made specialized corpora (Wilkinson, 2010) or unawareness among translators about the potential benefits of corpora “as a supplement to other resources and references” (Frankenberg-Garcia, 2015, p. 354). The use of computer-assisted translation (CAT) tools is currently much more widespread than corpus use in translation practice. Therefore, user-friendly integration of (monolingual original) corpora into CAT tools, particularly in their translation memory (TM) component, might allow and encourage (future) translators to use, construct and query such target language corpora better as part of their workflow (cf. Bernardini & Castagnoli, 2008). Furthermore, these corpora can complement the CAT tool translation approach, as TMs also entail disadvantages: TM segments, like most conventional dictionary entries, are decontextualized and Jiménez-Crespo (2009) has shown that inconsistency is higher in translations executed with TMs than in original texts. As translation quality depends on the content of the TM, TMs of poor quality will also negatively impact overall translation quality. To address these drawbacks original language corpora may be incorporated in translation practice. Aims and methods In order to assess whether the use of monolingual original corpora in the target language proves more beneficial with regard to translation quality than the use of conventional bilingual resources (bilingual dictionaries) and CAT tools (TMs), a pilot study was conducted among 11 master students taking a specialized legal translation course. We chose specialized texts as these are characterized by subject-specific language or Language for Specific Purposes (LSP). The use of corpora benefits translation of such texts greatly (Kübler, 2003, p. 25). The students were asked to translate two legal texts fragments. Their translation activity was also logged using keylogging software (Inputlog) and screen recording software (ActivePresenter). Text fragment one was translated with either a bilingual dictionary or a specialized TM in SDL Trados Studio 2017. Subsequently, a corpus compilation phase took place. During this phase, which lasted 30 minutes at most, the students were asked to compile a monolingual corpus of texts originally produced in the target language using pre-defined keywords. In order to find resources which were originally written in the target language, website country codes could be adjusted. For the translation of the second text fragment, the students either used a bilingual dictionary in combination with their self-compiled monolingual corpus in the target language or a specialized TM and their self-compiled corpus. The corpus was queried using the software AntConc. Translation quality assessment was initially error-based and conducted using error typologies based on MeLLANGE (Kübler et al., 2016) and on a set of annotation guidelines defining different error categories (Daems & Macken, 2013). In a later stage the translations will also be assessed from a linguistic perspective using tagging and parsing procedures. Preliminary pilot test results As revealed by the data from the screen recording software ActivePresenter, the number of corpus files gathered varied greatly, but the more corpus files the students had at their disposal in their self-compiled corpus, the more they consulted it. This led to longer translation times, but also enabled these students to maintain the same quality translation standard they achieved without corpus resources (text fragment 1). An increased number of corpus consultations also appeared to correlate with lower error rates in comparison to students with fewer corpus files and fewer corpus consultations. No major difference could be seen in the number of errors made between TM only translations and TM-corpus translations: the number of errors made was only slightly higher. Looking at error frequencies it was established that most frequently words and phrases were translated correctly in TM only and dictionary-corpus translations, but the translations did not fit the context (word sense disambiguation). This implies that nor a (decontextualized) TM, nor a (contextualized) corpus, when used separately, provide enough context to determine a correct context-specific translation. However, in TM-corpus translations the error category word sense disambiguation occurred much less often than in TM only translations: words and phrases were translated correctly more often and the translations used fitted the context. This suggests that the combination TM-corpus accounts for an increased number of adequate translations: by consulting the TM/corpus, the context-specific translation is found. However, the error category mistranslation (viz. when a word or phrase is not a possible translation of that word or phrase) also ranks high under the different translation conditions (including TM-corpus translations): the translation is unknown, nor can it be retrieved from the aids allowed depending on the translation condition. On the one hand, the above suggests that the combination TM-corpus may positively influence translation quality with regard to error frequencies (word sense disambiguation occurring less frequently). On the other hand, there are still other factors at play which may increase the error rate (e.g. the number of mistranslations), such as limitations of the TM/corpus, and inadequate corpus compilation and searching skills. The results to be presented for this paper will also include test results from a replication of the experiment described above with other student groups, including 11 master students taking a specialized scientific/medical translation course. References Bernardini, S. & Castagnoli, S. (2008). Corpora for translator education and translation practice. In Yuste Rodrigo, E. (ed.) Topics in Language Resources for Translation and Localisation. Amsterdam/Philadelphia: John Benjamins, 39-65. Bowker, L. (1998). Using specialized monolingual native-language corpora as a translation resource: a pilot study. Meta: Journal des traducteurs / Meta: Translators' Journal, 43(4), 631-651. Daems, J. & Macken, L. (2013). Annotation Guidelines for English-Dutch Translation Quality Assessment, version 1.0. LT3 Technical Report-LT3 13.02. Retrieved 27 October, 2017 from https://www.lt3.ugent.be/media/uploads/publications/2013/Technical%20Report%20TQA%20Annotation.pdf Frankenberg-Garcia, A. (2015). Training translators to use corpora hands-on: challenges and reactions by a group of thirteen students at a UK university. Corpora, 10(3), 351-380. Jiménez-Crespo, M. (2009). The Effect of Translation Memory Tools in Translated Web Texts: Evidence from a Comparative Product-Based Study. Linguistica Antverpiensia, 8, 213-232. Kübler, N. (2003). Corpora and LSP translation. In F. Zanettin, S. Bernardini, & D. Stewart (eds.) Corpora in Translator Education. Manchester: St. Jerome Publishing, 25-42. Kübler, N., Mestivier, A., Pecman, M. & Zimina, M. (2016). Exploitation quantitative de corpus de traductions annotés selon la typologie d’erreurs pour améliorer les méthodes d’enseignement de la traduction spécialisée. Actes des 13èmes Journées internationales d’analyse statistique des données textuelles (JADT 2016). Nice, France, 7-10 June 2016 (pp. 731-741). Wilkinson, M. (2010). Quick Corpora Compiling Using Web as Corpus. Translation Journal, 14(3). Retrieved January 8, 2016 from http://translationjournal.net/journal/53corpus.htm Zaretskaya, A., Corpas Pastor, A., & Seghiri, M. (2015). Translators’ requirements for translation technologies: a user survey. In Proceedings of the AIET17 Conference New Horizons in Translation and Interpreting Studies. Malaga, Spain, 29-31 January 2015 (pp. 247-254). Geneva: Tradulex.edition: CECL Papers 1: Using Corpora in Contrastive and Translation Studies Conference (5th edition)status: publishe