Balancing Expressiveness and Simplicity in an Interlingua for Task Based Dialogue

What it is about

The paper compares two domain-action interlinguas for speech translation in travel dialogue. The older C-star II format has 44 speech acts, 93 concepts and 117 argument names; the Nespole format is finer, with 65 speech acts, 110 concepts and 292 arguments. The authors re-tag the C-star II database (about 2,300 English and 7,100 non-English sentences) with the Nespole format and measure two things: the no-tag rate, the share of sentences that no expert can express in the interlingua, and how many distinct domain actions the data needs as it grows. The no-tag rate falls from 7.3 percent to 2.4 percent on the C-star data. Only about 50 domain actions cover 60 to 70 percent of the sentences, and the combined domains need more than 600. Cross-site reliability is weaker for Nespole: intra-site evaluations beat cross-site ones.

Ideas

• I1. Report a coverage rate for the meta-world: the share of source sentences, per author, in which every token gets a concept and a tag. Count it on the verse-aligned pairs, run it twice, and write it next to the residual.
• I2. Plot the frequency curve of meta concepts. If a small head of concepts covers most tokens, the lexicon is compact; the long tail is where the lexicon merges are most likely to be noise, so check those merges first.
• I3. Cross-site reliability as a test of merges: two independent readers (for example the Hebrew reader and a second aligner) should agree on the concept of a token before a merge is kept. Measure the agreement on the merges that were accepted.
• I4. Keep a coarse layer above the concept, like the speech-act part of a domain action: a small closed set of meta relations or verse-level roles that the encoder can read without the concept. This answers the scalability worry without growing the concept table.

For Mova

The place is tools/translate (meta.rs, the lexicon). The paper's domain actions are hand-built, while Mova would count its coarse layer from the data. It touches open problem 2 (keep a merge only if it lowers the bits) and open problem 1 (coarse relations on meta tokens). The paper gives no residual-style measure, so the coverage and merge audit above are the new parts to add.

Topics: Speech act theory · Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.3115/1118656.1118664

@inproceedings{levin2002balancing,
  title = {{Balancing Expressiveness and Simplicity in an Interlingua for Task Based Dialogue}},
  author = {Levin, Lori and Gates, Donna and Pianta, Dorcas and Cattoni, Roldano and Mana, Nadia and Peterson, Kay and Lavie, Alon and Pianesi, Fabio},
  booktitle = {Proceedings of the ACL-02 Workshop on Speech-to-Speech Translation: Algorithms and Systems},
  year = {2002},
  doi = {10.3115/1118656.1118664},
  url = {https://doi.org/10.3115/1118656.1118664}
}

References

8 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Bonnie J. Dorr. 1993. Machine Translation: A View from the Lexicon. The MIT Press, Cambridge, Massachusetts. find
2. Bonnie J. Dorr. 1994. Machine Translation Divergences: A Formal Description and Proposed Solution. Computational Linguistics, 20(4):597–633. find our card
3. Kenneth Goodman and Sergei Nirenburg. 1991. The KBMT Project: A Case Study in Knowledge-Based Machine Translation. Morgan Kaufmann, San Mateo, CA. find
4. Alon Lavie, Florian Metze, Roldano Cattoni, and Erica Constantini. 2002. A Multi-Perspective Evaluation of the NESPOLE! Speech-to-Speech Translation System. In Proceedings of Speech-to-Speech Translation: Algorithms and Systems. find
5. Young-Suk Lee, W. Yi, Clifford Weinstein, and Stephanie Seneff. 2001. Interlingua-based broad-coverage korean-to-english translation. In Proceedings of HLT, San Diego. find
6. Lori Levin, Donna Gates, Alon Lavie, and Alex Waibel. 1998. An Interlingua Based on Domain Actions for Machine Translation of Task-Oriented Dialogues. In Proceedings of the International Conference on Spoken Language Processing (IC-SLP’98), pages Vol. 4, 1155–1158, Sydney, Australia. find
7. Lori Levin, Donna Gates, Alon Lavie, Fabio Pianesi, Dorcas Wallace, Taro Watanabe, and Monika Woszczyna. 2000a. Evaluation of a Practical Interlingua for Task-Oriented Dialogue. In Workshop on Applied Interlinguas: Practical Applications of Interlingual Approaches to NLP, Seattle. find
8. Lori Levin, Alon Lavie, Monika Woszczyna, Donna Gates, Marsal Gavaldà, Detlef Koll, and Alex Waibel. 2000b. The Janus-III Translation System. Machine Translation. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/levin-2002-balancing-expressiveness-simplicity-interlingua-task/