Elamur Research⌕
☰
home / papers / 2015 / Post-editing a chapter of a specialized textbook into 7 languages: importance of terminological proximity with English for productivity

Post-editing a chapter of a specialized textbook into 7 languages: importance of terminological proximity with English for productivity

Paper · 2015 · Ritesh Shah, Christian Boitet, Pushpak Bhattacharyya et al.

Summary

The paper asks whether foreign students can turn free machine translations of specialized textbooks into usable versions in their native languages by post-editing them cooperatively. The authors set a practical limit of about 15–20 minutes per standard page. Researchers, mostly PhD students, post-edited chapter 21 of the English BEMbook (bioelectromagnetism) into Portuguese, Japanese, Russian, Spanish, Bengali, Hindi and Marathi. They worked in realistic, occasional sessions over several months, using the iMAG/SECTra platform, which records editing time per segment. Where target terms resemble English ones, through direct borrowing or shared Latin/Greek roots and affixes, editors could infer the terminology and stayed near the target pace. This held even for Japanese and Russian, which are structurally far from English. For the three Indic languages, editing took three to five times longer. Editors either did not know the existing technical terms or found that none existed and had to invent them. The authors argue that short snippets and easy language pairs hide this effect. Their recommendation is to gather or coin target-language terminology before cooperative post-editing begins.

Key points

  • The test text was chapter 21 of the BEMbook (Malmivuo and Plonsey, 1995): 4420 English words, about 18 standard pages of 250 words, and 374 segments averaging 22.22 words each.
  • Malayalam was planned as an eighth language but was dropped because no en-ml MT system was available at the time.
  • In SECTra/iMAG, each segment carries two labels:
    • a reliability level from one to five stars, set by the tool according to the origin of the translation (dictionary, MT, bilingual or professional post-editor);
    • a 0–20 quality score that the post-editor or any reader can change, with 12/20 counting as "good enough."
  • Typical editing time was 15–30 minutes per page. For bn, hi and mr, missing or unknown terminology added roughly an hour per page.
  • The EU Bologna translation service had no built-in collaborative post-editing and was shut down within 18 months of the project's end. The authors propose a contributive multilingual lexical base (PIVAX-like) plus SepT, a tool for collaboratively proposing pre-terminology.

Interesting ideas

  • I1. The authors separate two causes of slow editing: terms that exist but the editor does not know ("latent") and terms the language simply lacks ("absent"). This tells you whether to fix the editor's knowledge or the language resource.
  • I2. They record "primary" editing time (time spent in the edit box) separately from total time, which exposes hidden lookup and research effort.
  • I3. Each segment gets a fixed reliability level based on its origin and an editable quality score with a clear pass threshold, which keeps provenance apart from judged quality.
  • I4. The terminology effect only appeared on a whole chapter; short excerpts and close language pairs would have masked it.
  • I5. The authors recommend collecting or coining domain terms before the main work starts, and keeping that term base growing alongside use.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (back): When generating or translating from the meta-world, each concept could be tagged by whether the target lexicon has it, lacks it, or has it only weakly, so that the system flags a gap instead of producing a silent bad rendering.
  • I2 (learn): When people or LLM teachers repair silver trees, logging the time to correct separately from the time to decide would show which constructions are truly costly to annotate.
  • I3 (learn): Silver data could store a witness-agreement or teacher tier as fixed provenance plus a revisable quality score, and training could admit only items above a set cutoff.
  • I4 (answer): School-course and fable QA should include full long documents in specialized domains, because short passages may hide failures driven by vocabulary.
  • I5 (world): Before processing a new technical domain, Mova could compile a domain term layer (linked to WordNet/Wikidata) and keep updating it from corrections, so entities and frames do not break on unknown jargon.

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

 Institutions

🔬Laboratoire d'Informatique de GrenobleGrenoble · France21 works · 688 citations🎓Universidade Federal do Rio Grande do SulPorto Alegre · Brazil8 works · 522 citations🎓Indian Institute of Technology BHUVaranasi · India10 works · 548 citations🎓Banaras Hindu UniversityVaranasi · India6 works · 260 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Machine translation · 1,087 worksEnglish · 2,263 works

Cite

Cite as Ritesh Shah, Christian Boitet, Pushpak Bhattacharyya, Mithun Padmakumar, Leonardo Zilio, Ruslan Kalitvianski, Mohammad Nasiruddin, Mutsuko Tomokiyo and Sandra Castellanos Páez. 2015. Post-editing a chapter of a specialized textbook into 7 languages: importance of terminological proximity with English for productivity. Proceedings of the 12th International Conference on Natural Language Processing.

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

@inproceedings{shah2015post,
  title = {{Post-editing a chapter of a specialized textbook into 7 languages: importance of terminological proximity with English for productivity}},
  author = {Ritesh Shah, Christian Boitet, Pushpak Bhattacharyya, Mithun Padmakumar, Leonardo Zilio, Ruslan Kalitvianski, Mohammad Nasiruddin, Mutsuko Tomokiyo and Sandra Castellanos Páez},
  booktitle = {Proceedings of the 12th International Conference on Natural Language Processing},
  year = {2015},
  url = {https://aclanthology.org/W15-5947/}
}

References

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

  1. Heidi Depraetere, Joachim Van den Bogaert, and Joeri Van de Walle. 2011. Bologna Translation Service: Online translation of course syllabi and study programmes in English. In ”Proceedings of the 15th Conference of the European Association for Machine Translation”, pages 29–34, Leuven. find
  2. Spence Green, Jeffrey Heer, and Christopher D Manning. 2013. The efficacy of human post-editing for language translation. In ”Proceedings of the SIGCHI conference on Human factors in Computing systems”, pages 439–448. ACM. find
  3. Ruslan Kalitvianski, Christian Boitet, and Valérie Bellynck. 2012. Collaborative computer-assisted translation applied to pedagogical documents and literary works. In ”Proceedings of the 24th International Conference on Computational Linguistics”, pages 255–260, Mumbai. Citeseer. find
  4. Jaakko Malmivuo and Robert Plonsey. 1995. Bioelectromagnetism: principles and applications of bioelectric and biomagnetic fields. Oxford University Press. find
  5. Makoto Nagao, Jun-ichi Tsujii, and Jun-ichi Nakamura. 1985. The Japanese Government Project for Machine Translation. Computational Linguistics, 11(2-3):91–110. find
  6. Bonka Nikolova and Irina Nenova. 1982. An Automated System for Term Services. In J Horeckp, editor, ”Proceedings of the 9th Conference on Computational Linguistics”, pages 265–270, Prague. North-Holland. find
  7. J Pietrzak, A Jauregi, Joeri Van de Walle, and A Eriksson. 2013. Improving access to educational courses via automatic machine translation - New developments in post-editing. In ”Proceedings of the INTED 2013 Conference”, pages 5521–5529. IATED. find
  8. Bruno Pouliquen and Christophe Mazenc. 2011a. Automatic translation tools at WIPO. Aslib, Translating and the Computer, 33:17–18. find
  9. Bruno Pouliquen and Christophe Mazenc. 2011b. COPPA, CLIR and TAPTA: Three tools to assist in overcoming the patent language barrier at WIPO. In ”Proceedings of the 13th Machine Translation Summit”, pages 24–30. find
  10. Bruno Pouliquen, Cecilia Elizalde, Marcin Junczys-Dowmunt, Christophe Mazenc, and José Garcı́a-Verdugo. 2013. Large-scale multiple language translation accelerator at the United Nations. In ”Proceedings of the 14th Machine Translation Summit”, pages 345–352. 332 find
  11. Joeri Van de Walle, Heidi Depraetere, and J Pietrzak. 2012a. Bologna Translation Service: High-quality automated translation of study programmes into English. In ”Proceedings of the EDULEARN 2012 Conference”, pages 5831–5835. IATED. find
  12. Joeri Van de Walle, Heidi Depraetere, and J Pietrzak. 2012b. Bologna Translation Service: Making study programmes accessible throughout Europe by means of high-quality automated translation. In ”Proceedings of ICERI 2012 Conference”, pages 3910–3918. IATED. find
  13. Lingxiao Wang and Christian Boitet. 2013. Online production of HQ parallel corpora and permanent task-based evaluation of multiple MT systems: both can be obtained through iMAGs with no added cost. In ”Proceedings of the 2nd Workshop on Post-Editing Technologies and Practice at MT Summit 2013”, pages 103–110, Nice, September. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/shah-2015-post-editing-chapter-specialized-textbook/