Generation from Lexical Conceptual Structures

What it is about

The paper describes a generator that turns Lexical Conceptual Structures (LCS), a compositional, language-independent interlingua, into English sentences. An LCS is a graph whose nodes have a type (event, state, path, manner, property, thing), a primitive and a field. The composed structure is matched against the lexicon of the target language, and a lexical item can cover several nodes or only some of them. The example is a Chinese verb that covers a whole structure, where the English lexicon splits it into "cut", "US", "quota" and "down". The generator linearises the result with bigram statistics and uses parts of an existing nitrogen generator. The authors note an open question about numbers, which can be stored in the source-language form or as canonical numerals. The evaluation is small and subjective: about 80 sentences, judged by quality. The system was also used to generate thousands of simple template sentences for cross-language retrieval.

Ideas

• I1. Choose target chunks by cover: pick the set of standard chunks that covers the most meta tokens at the lowest cost, and report the cover ratio per sentence. This is the lexicalisation step the paper describes, and the counts come from the standard.
• I2. Combine the bigram order statistics with a second, fixed score from the standard, as the paper proposes for its future work. Measure exact-match on dev, run twice before any claim.
• I3. Canonical numerals: keep numbers in the meta-world as canonical values, with the source form on the token. Test on verses with numerals and count the ones that break under the two choices.
• I4. Attribute errors to a layer. The paper says it could not tell analysis, lexicon and generation apart. The residual by layer that Mova already measures answers that question, and the paper shows why it is needed.

For Mova

The places are en/src/lin.rs (the linearizer and its Markov chains), profile.rs (the chunk cover) and measure.rs (the residual by layer). I1 touches open problem 3 (the encoder writes trees, not strings); I3 touches the number features of open problem 5. The paper's evaluation is too small to use as a number, so the new measures must be taken on the full pairs.

Topics: Machine translation

Cite

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

@inproceedings{traum2000generation,
  title = {{Generation from Lexical Conceptual Structures}},
  author = {Traum, David and Habash, Nizar},
  booktitle = {NAACL-ANLP 2000 Workshop: Applied Interlinguas: Practical Applications of Interlingual Approaches to NLP},
  year = {2000},
  url = {https://aclanthology.org/W00-0207/}
}

References

12 references, 0 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, James Hendler, Scott Blanksteen, and Barrie Migdaloff. 1993. Use of Lexical Conceptual Structure for Intelligent Tutoring. Technical Report UMIACS TR 93-108, CS TR 3161, University of Maryland. find
2. Bonnie J. Dorr, Nizar Habash, and David Traum. 1998. A Thematic Hierarchy for Efficient Generation from Lexical-Conceptal Structure. In Proceedings of the Third Conference of the Associatio.n for Machine Translation in the Americas, AMTA-#8, in Lecture Notes in Artificial Intelligence, 15~9, pages 333-343, Langhorne, PA, October 28-31. find
3. Bonnie J. Dorr. 1993. Machine Translation: A View from the Lexicon. The MIT Press. find
4. Bonnie J. Dorr. 1997. Large-Scale Acquisition of LCS-Based Lexicons for Foreign Language Tutoring. In Proceedings of the A CL Fifth Conference on Applied Natural Language Processing (ANLP), pages 139-146, Washington, DC. find
5. Ray Jackendoff. 1983. Semantics and Cognition. The MIT Press, Cambridge, MA. find
6. Ray Jackendoff. 1990. Semantic Structures. The MIT Press, Cambridge, MA. find
7. Ray Jackendoff. 1996. The Proper Treatment of Measuring Out, Telicity, and Perhaps Even Quantification in English. Natural Language and Linguistic Theory, 14:305-354. find
8. Irene Langkilde and Kevin Knight. 1998a. Generation that Exploits Corpus-Based Statistical Knowledge. In Proceedings of COLING-A CL '98, pages 704-710. find
9. Irene Langkilde and Kevin Knight. 1998b. The Practical Value of N-Grams in Generation. In International Natural Language Generation Workshop. find
10. Beth Levin and Malka Rappaport Hovav. 1995. Un-aecusativity: At the Syntaz-Lezical Semantics Interface. The MIT Press, Cambridge, MA. LI Monograph 26. find
11. Gina Levow, Bonnie J. Dorr, and Dekang Lin. 2000. Construction of chinese-english semantic hierarchy for cross-language retrieval, forthcoming. find
12. Randolph Quirk, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik. 1985. A Comprehensive Grammar of the English Language. Longman, London. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/traum-2000-generation-lexical-conceptual-structures/