'Relaxed' compositionality in machine translation

What it is about

The paper describes a transfer machine that keeps two principles: translation is a relation between argument structures, and it is compositional and reversible. Strict compositionality cannot handle exceptional cases, so the authors propose a relaxed version in which the translation of a whole is a function of the parts, but a rule may also cover a group of words that does not map part by part. The running case is a class of Dutch adverbs such as the one that means roughly like doing, which corresponds to an English sentential construction. The special rule sits only in the transfer rules, so the regular parts of the grammar are not affected. The paper names a risk: one structure can be translated both strictly and by a relaxed rule, and some of those double translations are wrong, for example in morphology. No numbers are given.

Ideas

• I1. Allow one meta token to map to several target tokens, as in the adverb-to-construction case, by letting a merge in meta.rs span a gap of up to four words. Record the special rule in the delta for that author only, and count how often it fires.
• I2. Rank double translations: when a strict and a relaxed path both fit a verse, prefer the one with the higher count in the standard, and log the rate of ties. This is the paper's own problem, measured on Bible pairs.
• I3. Guard the morphology: a relaxed rule must not replace a word-level form, as the Dutch agent-noun case shows. Add a control in which such rules are disabled, and check that the exact-match rate on those verses drops, to see the rule is doing work.

For Mova

The code is tools/translate/src/meta.rs (the multi-word merges) and profile.rs (the choice of chunk). The strict versus relaxed split maps onto the standard versus the delta; the paper's risk of double translation is the ranking problem in I2.

Topics: Machine translation

Cite

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

@inproceedings{arnold1988relaxed,
  title = {{'Relaxed' compositionality in machine translation}},
  author = {Arnold, Doug and Krauwer, Steven and Tombe, Louis des and Sadler, Louisa},
  booktitle = {Proceedings of the Second Conference on Theoretical and Methodological Issues in Machine Translation of Natural Languages},
  year = {1988},
  url = {https://aclanthology.org/1988.tmi-1.6/}
}

References

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

1. D.J.Arnold, S.Krauwer, M. Rosner, L. des Tombe, & G.B.Varile (1986) "The <C, A>, T framework in Eurotra" in Proceedings of Coling 86 , Bonn. find
2. Arnold,D.J. & des Tombe,L. (1987) "Basic Theory and Methodology in Eurotra" in Nirenburg, S. (ed.) pp 114-135. find
3. Appelo, L. & F. de Jong (1987) "Synonymy and Translation", Philips Research M.S. 14.269. find
4. Bresnan,J. (ed) (1982) The Mental Representation of Grammatical Relations MIT Press, Cambridge, Mass. find
5. Cresswell, M.J. (1985): Adverbial Modification Reidel Publishing Company, Dordrecht. find
6. van Eynde, Frank (1987a): "Time. A Unified Theory of Tense, Aspect and Aktionsart" (ms.), (Eurotra-Leuven). find
7. van Eynde, Frank (1987b): "Tense in Text" (ms.), (Eurotra-Leuven). find
8. Kaplan, R. & J. Bresnan (1982) "Lexical Functional Grammar: a formal system for grammatical representation" in Bresnan (ed) pp173-281. find
9. Krauwer,S. & Des Tombe,L. (1984) "Transfer in a multi-lingual system" in Proceedings of COLING 1984 Stanford, Ca. pp464-4S7. find
10. Landsbergen, J. (1985) "Isomorphic Grammars and their use in the Rosetta Translation System", in King, M (ed) Machine Translation Today Edinburgh University Press. find
11. Landsbergen, J. (1986) "Montague Grammar tion", in Whitelock et al (1987) pp 113-148. find
12. Nagao, M. (1987) "Role of Structural Transformation Translation System", in Nirenburg, S. (ed.) pp262-277. find
13. Nirenburg, S. (ed. ) (1987) Machine Translation: Theoretical and Methodological Issues CUP, Cambridge. find
14. van Noord, G. et al (1988) "A Treatment of Anaphora", ms. University of Utrecht. find
15. Petitpierre, D. et al (1986) "A model for Preference" in Proceedings of the 1986 European ACL , Copenhagen. find
16. Rigler, E. (1988) "Multilingual Application of a Theory of Tense and Aspect", Working Papers in Language Processing No. 4., Eurotra-Essex, University of Essex. find
17. Whitelock, P. et al (1987) Linguistic Theory and Computer Applications, Academic Press, London. MT and Machine Transla-in a Machine find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/arnold-1988-relaxed-compositionality-machine-translation/