Solving Thematic Divergences in Machine Translation

What it is about

A divergent predicate-argument structure is one where the main verb or its arguments take a different syntactic order in the target language, for example when the theme of a verb surfaces as the subject in Spanish but as the object in English. Dorr argues that most systems handle such cases ad hoc, and proposes a general procedure that uses the link between lexical-semantic structure and syntax. The UNITRAN system maps a lexical conceptual structure to syntax through general linking routines. Language-specific behaviour is carried by override markers on lexical entries: one marker promotes an internal argument into the head position, another demotes a head into an adjunct. The paper shows that the same routines, with these markers, handle the thematic reversal and the head-switching examples in both directions. There are no corpus counts; the evidence is worked derivations of a few sentences.

Ideas

• I1. Keep override markers as a learned table, not hand-written. For each head switch found in the aligned Bible pairs, store a promote or demote marker on the lexeme only if it lowers the reader plus encoder residual. This is the same keep-if-it-helps rule as the author delta.
• I2. Separate the head from the satellite in the meta-world. Keep a pointer to the phrasal head and a flag saying whether a token is a head or an adjunct, so that a promotion is a re-attachment in the tree and not a change of concept.
• I3. Build a small control set of thematic reversals from the verse pairs (sentences where one side has the verb as subject and the other as object). Count how many the current encoder gets right, and report the count with its date before any change.

For Mova

The derivations are a useful test set, not a model to copy: they are hand-coded per entry and the paper has no statistics. The place is tools/translate (meta.rs, profile.rs) for the markers and the encoder, en/src/lin.rs for the word order that follows the head, and tools/roundtrip for the frames. The first measure is the count of reversal cases the current encoder misses on the dev verses.

Topics: Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.3115/981823.981840

@inproceedings{dorr1990solving,
  title = {{Solving Thematic Divergences in Machine Translation}},
  author = {Dorr, Bonnie},
  booktitle = {28th Annual Meeting of the Association for Computational Linguistics},
  year = {1990},
  doi = {10.3115/981823.981840},
  url = {https://doi.org/10.3115/981823.981840}
}

References

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

1. Gretchen Brown, "Some Problems in German to English Machine Translation," MAC Technical Report 142, Massachusetts Institute of Technology, Cambridge, MA, 1974. find
2. NoRm A. Chomsky, Knowledge of Language: Its Nature, Origin and Use, MIT Press, Cambridge, MA, 1986. {Doff, 1987] Bonnie J. Dorr, "UNITRAN: A Principle-Based Approach to Machine Translation," AI Technical Report 1000, Master of Science thesis, Department Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, 1987. find
3. Bonnie J. Doff, "Lexical Conceptual Structure and Machine Translation," Ph.D. thesis, Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology, Cambridge, MA, 1990. find
4. Nell M. Goldman, "Computer Generation of Natural Language from a Deep Conceptual Base," Ph.D thesis, Computer Science Department, Stanford University, Stanford, CA, 1974. find
5. Ray S. Jackendoff, Semantics and Cognition, MIT Press, Cambridge, MA, 1983. find
6. Steven Lytinen and Roger Schank, "Representation and Translation," Technical Report 234, Department of Computer Science, Yale University, New Haven, CT, 1982. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/dorr-1990-solving-thematic-divergences-machine-translation/