Elamur Research⌕
☰
home / papers / 2019 / Towards Handling Verb Phrase Ellipsis in English-Hindi Machine Translation

Towards Handling Verb Phrase Ellipsis in English-Hindi Machine Translation

Summary

English–Hindi machine translation systems often misinterpret sentences with verb phrase ellipsis (VPE), where the rest of the predicate is omitted after an auxiliary verb or the particle "to". The authors propose a rule-based system that processes the English sentence before translation. It builds on the Hindi phenomenon of VVPE (verb-stranding VP ellipsis): in Hindi, the verb's arguments can be omitted while the verb itself stays at the ellipsis site and keeps all possible readings. So there is no need to find the exact boundaries of the omitted phrase, which is difficult and often leads to errors. It is enough to find the main verb of the antecedent and insert it at the ellipsis site. The system handles phrasal, serial and compound verbs separately. It works in four steps: detect the ellipsis, find the head of the antecedent, select the required verbal segment, and rewrite the sentence. On the BNC corpus, the system detects ellipsis on par with the best known solutions. Manual evaluation of the translations after processing shows a moderate improvement. According to the authors, this is the first attempt to tailor VPE resolution specifically to the English–Hindi pair. The work shows that knowledge of the target language's grammar can simplify the task for MT.

Key points

  • VPE detection: precision 94.83 % and recall 83.04 % on 3900 sentences from the BNC corpus.
  • Antecedent head identification and verbal segment selection were evaluated jointly: accuracy of 91 % on 200 sentences from the WSJ corpus.
  • Motivation for the work: in a preliminary test, Google Translate correctly translated only 6 of 50 sentences with VPE taken from WSJ.
  • The effect on translation was evaluated manually on 100 sentences using the authors' own scale from 1 to 5, where 5 means a drastic improvement. The average score with the Anusaaraka system was 3.55, and with Google Translate 3.7. The least benefit was seen for sentences that MT translates poorly even without ellipsis, and for constructions of the "as did" type.
  • The system distinguishes five ellipsis classes by the auxiliary element: to_be, to_have, to_do, modal verbs and the particle to. "Do so" anaphora is deliberately not handled.

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: Linguistics & languages

Topics: Coreference resolution · Machine translation

Cite

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

@inproceedings{bafna2019handling,
  title = {{Towards Handling Verb Phrase Ellipsis in English-Hindi Machine Translation}},
  author = {Bafna, Niyati and Sharma, Dipti},
  booktitle = {Proceedings of the 16th International Conference on Natural Language Processing},
  year = {2019},
  url = {https://aclanthology.org/2019.icon-1.18/}
}

References

10 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. J. Bos and J. Spenader. An annotated corpus for the analysis of vp ellipsis. Language Resources and Evaluation, 45(4):463–494, 2011. find
  2. D. Hardt. An algorithm for vp ellipsis. In Proceedings of the 30th annual meeting on Association for Computational Linguistics, pages 9–14. Association for Computational Linguistics, 1992. find
  3. K. Kenyon-Dean, J. C. K. Cheung, and D. Pre-cup. Verb phrase ellipsis resolution using discriminative and margin-infused algorithms. In Proceedings of the 2016 Conference on Empirical (Score: 4) find our card
  4. Methods in Natural Language Processing, pages 1734–1743, 2016. find
  5. G. N. Leech. 100 million words of english: the british national corpus (bnc). 1992. find
  6. Z. Liu, E. G. Pellicer, and D. Gillick. Exploring the steps of verb phrase ellipsis. In Proceedings of the Workshop on Coreference Resolution Beyond OntoNotes (CORBON 2016), pages 32–40, 2016. find our card
  7. E. Manetta. Verb-phrase ellipsis and complex predicates in hindi-urdu. Natural Language & Linguistic Theory, pages 1–39, 2018. find
  8. L. A. Nielsen. A corpus-based study of Verb Phrase Ellipsis Identification and Resolution. PhD thesis, King’s College London, 2005. find
  9. D. B. Paul and J. M. Baker. The design for the wall street journal-based csr corpus. In Proceedings of the workshop on Speech and Natural Language, pages 357–362. Association for Computational Linguistics, 1992. find
  10. E. Voita, R. Sennrich, and I. Titov. When a good translation is wrong in context: Context-aware machine translation improves on deixis, ellipsis, and lexical cohesion. arXiv preprint arXiv:1905.05979, 2019. find our card

This page in Markdown → https://md.research.elamur.ai/free/papers/bafna-2019-handling-verb-phrase-ellipsis-english/