Towards Functionally Similar Corpus Resources for Translation

Proceedings of Recent Advances in Natural Language Processing, pages 583–592, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_069 583 Towards Functionally Similar Corpus Resources for Translation Maria Kunilovskaya University of Tyumen Tyumen, Russia m.a.kunilovskaya@utmn.ru Serge Sharoff University of Leeds Leeds, UK s.sharoff@leeds.ac.uk Abstract The paper describes a computational ap- proach to produce functionally compa- rable monolingual corpus resources for translation studies and contrastive analy- sis. We exploit a text-external approach, based on a set of Functional Text Dimen- sions to model text functions, so that each text can be represented as a vector in a multidimensional space of text functions. These vectors can be used to find reason- ably homogeneous subsets of functionally similar texts across different corpora. Our models for predicting text functions are based on recurrent neural networks and traditional feature-based machine learning approaches. In addition to using the cat- egories of the British National Corpus as our test case, we investigated the func- tional comparability of the English parts from the two parallel corpora: CroCo (English-German) and RusLTC (English- Russian) and applied our models to de- fine functionally similar clusters in them. Our results show that the Functional Text Dimensions provide a useful description for text categories, while allowing a more flexible representation for texts with hy- brid functions. 1 Introduction Comparable corpora are an important prerequisite for translation studies (TS) and contrastive analy- sis. One wants to make sure that the corpus re- sources used to explore differences between lan- guages or aspects of translational specificity in several target languages (TL) are comparable in the first place. One of the common approaches to corpus com- parability is to define it as the domain similarity and to rely on the vocabulary overlap as the mea- sure of comparability. A brief summary of possi- ble interpretations of the concept and comparabil- ity measures can be found in Li et al. (2018). The authors give a domain-based definition to cross- linguistically comparable corpora: “document sets in different languages that cover similar topics”. While lexical similarity is an important factor in linguistic variation, we would argue that it does not capture all the translationally relevant features of texts. Neumann (2013), Kruger and Van Rooy (2010) and Delaere (2015) have also highlighted the importance of register and genre in study- ing translations by showing that different registers produce different types of translationese. More- over, functional theories within translation studies (TS) insist that what matters in translation is func- tional adequacy. The target text (TT) is expected to fulfill the same communicative functions as the source text (ST) and meet the TL conventions ex- pected in the situation of TL communication of the message (Nord, 2006). Both Reiss and Ver- meer (1984) and Neubert (1985) build their theory of translation around genres or text types, while Shveitzer (1973) underlines the impact of the text functions hierarchy on the translator’s linguistic choices. The above suggests that translational compara- bility of corpus resources should take into account social and situational constrains of the communi- cation and …

Content truncated to 500 words