Elamur Research⌕
☰
home / papers / 2011 / Translationese and Its Dialects

Translationese and Its Dialects

Summary

The paper asks whether translated English differs from native English because of general effects of translating, or because each source language leaves its own traces ("interference"). The authors test both views with text categorization. They use Europarl translations into English from Finnish, French, German, Italian and Spanish, plus original English, split into equal chunks. Each chunk is represented only by frequencies of 300 LIWC function words, which avoids topical artifacts, and classified with Bayesian logistic regression. Classifiers identify the source language with high accuracy. They also separate translated from original text very well when training and testing on the same source language. Training on one source language and testing on another works only as well as the two languages are related. A second corpus (IHT) with very different source languages and a combined eight-language, two-genre corpus show that general translationese markers still allow reliable detection. The conclusion is that translationese is graded. Some of its traits are universal, and others behave like dialects tied to particular source languages. This matters for translation studies, typology, and building language models for MT.

Key points

  • Data and setup: Five Europarl components of just under 500,000 words each, plus original English, cut into chunks. 1000 translated and 1000 original chunks were used for the experiments, with extra chunks held out for development. Features were 300 LIWC function words, the learner was Bayesian logistic regression, and evaluation used 10-fold cross-validation.
  • Source-language identification: Five-way classification reached 92.7% accuracy. In the confusion matrix, errors cluster among the Romance languages, are fewer involving German, and are fewest for Finnish.
  • Cross-language transfer: When the T-vs-O classifier is trained on one Romance source language and tested on another, accuracy stays between 84.5% and 91.5%. Transfer to less related languages is much weaker. Some markers even flip direction. For example, me signals original English against Finnish translations but signals translation from German.
  • IHT corpus: On IHT, which includes Hebrew and Korean, within-corpus T-vs-O accuracy is 86.3%. Transfer between corpora is weak: Europarl→IHT reaches 64.8%, and IHT→Europarl reaches 58.8%.
  • Combined corpus: With eight source languages and two genres (one spoken, one written), classification reaches 90.5%. The word the is over-represented in translations (15.36% vs 13.31% in IHT), and so are most cohesive adverbs (e.g., therefore, thus, however). Actually is the exception in both corpora.

Interesting ideas

  • I1. Function words only: Limiting features to function words lets a classifier capture style while ignoring topic. This could carry over to authorship work or to detecting machine-translated Ukrainian text, where content differences would otherwise confound the results.
  • I2. Cross-condition train/test matrix: The authors train on condition x and test on condition y, then read the accuracy matrix as a measure of how similar the conditions are. The same design could quantify distances between dialects, genres, or data sources.
  • I3. Translation as a probe of typology: Translations from different languages into one shared target can be compared to estimate how far apart the source languages are. This gives an empirical complement to traditional typological classifications.
  • I4. Unit-agnostic chunking: Corpora are cut into equal-size chunks regardless of natural document boundaries. This controls for length and yields many balanced samples, a simple trick for any stylometric experiment.
  • I5. Language models from translations: The authors suggest that translated text, especially from related source languages, may yield better language models for MT than native text. This is worth testing when choosing training data for translation into Ukrainian.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Natural language processing

Cited by

102 papers of our library cite it (of the 140884 whose bibliographies we read); 15 of them have a page here, 87 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{koppel2011translationese,
  title = {{Translationese and Its Dialects}},
  author = {Koppel, Moshe and Ordan, Noam},
  booktitle = {Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies},
  year = {2011},
  url = {https://aclanthology.org/P11-1132/}
}

References

19 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Mona Baker. 1993. Corpus linguistics and translation studies: Implications and applications. In Gill Francis Mona Baker and Elena Tognini Bonelli, editors, Text and technology: in honour of John Sinclair, pages 233-252. John Benjamins, Amsterdam. find
  2. Marco Baroni and Silvia Bernardini. 2006. A new approach to the study of Translationese: Machine-learning the difference between original and translated text. Literary and Linguistic Computing, 21(3):259-274. find
  3. Shoshan Blum-Kulka. Shifts of cohesion and coherence in translation. 1986. In Juliane House and Shoshana Blum-Kulka (Eds), Interlingual and Intercultural Communication (17-35). Tübingen: Günter Narr Verlag. find
  4. William Frawley. 1984. Prolegomenon to a theory of translation. In William Frawley (ed), Translation. Literary, Linguistic and Philosophical Perspectives (179-175). Newark: University of Delaware Press. find
  5. Maria Antonietta Fusco. 1990. Quality in conference interpreting between cognate languages: A preliminary approach to the Spanish-Italian case. The Interpreters’ Newsletter, 3, 93-97. find
  6. Martin Gellerstam. 1986. Translationese in Swedish novels translated from English, in Lars Wollin & Hans Lindquist (eds.), Translation Studies in Scandinavia (88-95). Lund: CWK Gleerup. find
  7. Iustina Ilisei, Diana Inkpen, Gloria Corpas Pastor, and Ruslan Mitkov. Identification of translationese: A machine learning approach. In Alexander F. Gelbukh, editor, Proceedings of CICLing-2010: Computational Linguistics and Intelligent Text Processing, 11th International, volume 6008 of Lecture No… find
  8. Kenneth Katzner. 2002. The Languages of the World. Routledge. find
  9. Grzegorz Kondrak and Tarek Sherif. 2006. Evaluation of several phonetic similarity algorithms on the task of cognate identification. In Proceedings of the Workshop on Linguistic Distances (LD '06). 43-50. find
  10. David Kurokawa, Cyril Goutte, and Pierre Isabelle. 2009. Automatic detection of translated text and its impact on machine translation. In Proceedings of MT-Summit XII. find
  11. Sara Laviosa: 1997. How Comparable can 'Comparable Corpora' Be?. Target, 9 (2), pp. 289-319. find
  12. Sara Laviosa-Braithwaite. 1998. In Mona Baker (ed.) Routledge Encyclopedia of Translation Studies. London/New York: Routledge, pp.288-291. find
  13. Sara Laviosa. 2002. Corpus-based Translation Studies. Theory, Findings, Applications. Amsterdam/New York: Rodopi. find
  14. David Madigan, Alexander Genkin, David D. Lewis and Dmitriy Fradkin 2005. Bayesian Multinomial Logistic Regression for Author Identification, In Maxent Conference, 509-516. find
  15. James W. Pennebaker, Martha E. Francis, and Roger J. Booth. 2001. Linguistic Inquiry and Word Count (LIWC): LIWC2001 Manual. Erlbaum Publishers, Mahwah, NJ, USA. find
  16. Helmut Schmid. Probabilistic Part-of-Speech Tagging Using Decision Trees. 2004. In Proceedings of International Conference on New Methods in Language Processing. find
  17. Larry Selinker.1972. Interlanguage. International Review of Applied Linguistics. 10, 209-241. find
  18. Gideon Toury. 1995. Descriptive Translation Studies and beyond. John Benjamins, Amsterdam / Philadelphia. find
  19. Hans van Halteren. 2008. Source language markers in EUROPARL translations. In COLING '08: Proceedings of the 22nd International Conference on Computational Linguistics, pages 937-944. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/koppel-2011-translationese-its-dialects/