Abstraction and underspecification in semantic transfer

What it is about

The paper presents MinT, a semantic transfer module in the Verbmobil speech translation system for German to English. It is unification-based and lexicalist. Its input is Underspecified Minimal Recursion Semantics, where operator scope and modifier attachment are left open as sets of possible sites. Transfer relates source and target descriptions at the most abstract level that still works. Predicates are typed in a hierarchy, and a bilingual abstract type such as "suits" covers the German "passt" and "klappt" and the English "suits" and "works for". The generator then chooses among the subtypes. Ambiguities that hold in both languages are passed on unresolved, and only the ones that differ are resolved. The gain claimed is fewer transfer rules and more freedom in lexical choice. The paper gives worked examples and no measured counts in the parts read.

Ideas

• I1. Abstract types for meta concepts: group concepts that the standard can realize in several ways under one type, with one subtype per realization. The encoder picks the subtype from the author delta. Measure the rule count in profile.rs and the exact-match result on dev before and after, run twice.
• I2. Underspecified attachment in the reader: where the parse head is uncertain, keep the candidate heads on the meta token rather than one head. The encoder picks the candidate its target standard supports. Count how many meta tokens have more than one plausible head across two translations of a verse.
• I3. A type hierarchy for concepts: when an unseen concept has no entry, fall back to its parent in the hierarchy instead of a bare form. The profile already backs off from concept and tag to concept; a hierarchy would make that back-off principled.
• I4. Preserve shared ambiguity: count how often the encoder must resolve an ambiguity that the target language also has. Those cases are the ones to keep unresolved.

For Mova

The places are tools/translate/profile.rs (the encoder back-off and rules) and meta.rs (the reader). I1 and I3 touch open problem 3 (the encoder should write trees, not strings) and the lexicon. I2 touches open problem 1, the reader's attachment noise. The paper's numbers are not in the parts read, so the counts must come from Mova's own data.

Topics: Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.22028/d291-25319

@inproceedings{abb1996abstraction,
  title = {{Abstraction and underspecification in semantic transfer}},
  author = {Abb, Bernd and Buschbeck-Wolf, Bianka and Tschernitschek, Christel},
  booktitle = {Conference of the Association for Machine Translation in the Americas},
  year = {1996},
  doi = {10.22028/d291-25319},
  url = {https://doi.org/10.22028/d291-25319}
}

References

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

1. B. Abb and B. Buschbeck-Wolf. 1995. A Minimal Transfer Conception for Verbmobil. Verbmobil-Report, 84. find
2. B. Abb and C. Maienborn. 1994. Adjuncts in HPSG. Tagungsband KONVENS 94- Verarbeitung natürlicher Sprache, pages 13-22. find
3. H. Alshawi, D. Carter, M. Rayner, and B. Gambäck. 1991. Translation by Quasi Logical Form Transfer. Proceedings of the ACL, Berkeley. find our card
4. B. Buschbeck-Wolf and R. Nübel. 1995. Die Behandlung übersetzungs-ambiger Präpositionen im Transfer des VERBMOBIL-Demonstrators. Verbmobil-Report, 87. find
5. A. Copestake and T. Briscoe. 1995. Semi-productive Polysemy and Sense Extension. Journal of Semantics, 12(1). find
6. A. Copestake, D. Flickinger, R. Malouf, S. Riehemann, and I. Sag. 1995. Translation using Minimal Recursion Semantics. Proceedings of the 6th International Conference on Theoretical and Methodological Issues in Machine Translation, Leuven. find our card
7. A. Copestake. 1995. Semantic Transfer in Verbmobil. Verbmobil-Report, 93. find
8. J. Dörre, M. Dorna, and J. Junger, 1994. The CUF User's Manual. Insititut für Maschinelle Sprachverarbeitung, Universität Stuttgart. find
9. M. Egg and K. Lebeth. 1995. Semantic Underspecification and Modifier Attachment. Integrative Ansätze in der Computerlinguistik. Beiträge zur 5. Fachtagung für Computerlinguistik der DGfS. find
10. W.J. Hutchins and H.L. Somers. 1992. An Introduction to Machine Translation. Academic Press, Cambridge. find
11. S. Jekat, A. Klein, Maier E., I. Maleck, and J. Quantz. 1995. Dialogue Acts in Verbmobil. Verbmobil-Report, 65. find
12. M. Kameyama, R. Ochitani, and S. Peters. 1991. Resolving Translation Mismatches with Information Flow. Proceedings of the A CL, Berkeley. find
13. M. Kay, J.M. Gawron, and P. Norwig. 1994. VERBMOBIL - A Translation System for Face-to-Face Dialogue. Number 33 in CSLI Lecture Notes. University of Chicago Press. find
14. K. Lebeth and F. Schilder. 1996. On Scope Ambiguities and Aspect Composition. Verbmobil-Report, 114. find
15. C. Pollard and I. Sag. 1994. Head-driven Phrase Structure Grammar. Chicago University Press, Chicago. find
16. U. Reyle. 1993. Vagueness and Ambiguity in DRT. Journal of Semantics. find
17. B. Vauquois. 1975. La Traduction Automatique à Grenoble. Dunod, Paris. find
18. P.J. Whitelock. 1992. Shake-and-Bake-Translation. Proceedings of the I4th International Conference on Computational Linguistics, Nantes. find our card

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/abb-1996-abstraction-underspecification-semantic-transfer/