The Transfer of Finite Verb Forms in a Machine Translation System

What it is about

Work done within the EUROTRA project on how finite verb forms should be transferred between the European languages. Rather than map forms language pair by language pair, the paper defines a shared set of tense values that every language is read into and out of. The values are an axis (present or past anchoring), prospective and retrospective markers, aspect (continuous, emphatic, progressive) and an Aktionsart (durative or not), with a few extra values suggested for English. A rule set then says, for each language, which combinations occur and how they are expressed, for example that English does not combine the perfect with a when-adverbial. The author reports a simulation on Community texts that gave good results. The part read has no table of numbers.

Ideas

• I1. Tense values as features: store the tense and aspect features of the UD trees (the FEATS on verbs) as a small enum on each meta token, with the anchor, prospective, retrospective and aspect values. The encoder then maps by values rather than by word pairs, which is the multilingual design the paper argues for.
• I2. Language rules as constraints with counts: keep a table of forbidden combinations for the target (the perfect with a time adverbial in English) in en/src/morph.rs, and count how often the encoder or the linearizer hits a forbidden combination on the Bible pairs.
• I3. Error by tense value: split the morphology residual by the tense value of the meta token, so that a weak value (for example the progressive) shows up as a row rather than a total.

For Mova

The paper is the origin of a shared tense space, and it supports open problem 5 (number and person from UD FEATS on meta tokens, so a merge does not lose what its members differ in). The same idea applied to tense is a feature check in the encoder. The constraint table is a count-based version of the rules, and the per-value residual fits the measure split by class. Each step is to be measured on held-out verses.

Topics: Machine translation

Cite

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

@inproceedings{maegaard1982transfer,
  title = {{The Transfer of Finite Verb Forms in a Machine Translation System}},
  author = {Maegaard, Bente},
  booktitle = {Coling 1982 Abstracts: Proceedings of the Ninth International Conference on Computational Linguistics Abstracts},
  year = {1982},
  url = {https://aclanthology.org/C82-2046/}
}

References

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

1. Bruce: A Model for Temporal References and Its Application in a Question Answering Program. Artificial Intelligence 1972, vol 3, PP 1-25. find
2. Greenbaum, Quirk, Leech, Svartvik: A Qramm~r of Contemporary E~lish. Longman. London 1976. find
3. Maegaard, Ruus, Spang-Hanssen: Investigation of the Verbal S2stemfor.EUROTRA, Report for the EEC, 1981. find
4. Reichenbach: Elements of S~mbolic LoKic. Macmillan. New York 1947, pp 287-298. - - Charbon find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/maegaard-1982-transfer-finite-verb-forms-machine/