Semantic Inference at the Lexical-Syntactic Level for Textual Entailment Recognition

What it is about

The paper proposes a proof system for textual entailment that works directly on dependency trees. Entailment rules, manual or acquired from WordNet and from DIRT-style paraphrase data, rewrite the text tree step by step towards the hypothesis, and a cost function measures the remaining distance when no full proof is found. Proofs are limited to depth four, and the parser is Minipar. On the RTE-3 test set the full system reaches 62.4% accuracy with an optimal threshold, and the lexical-only configuration 62.9%, so the richer rules give no consistent gain. On the development set exact proofs have 92% precision but only 4.3% recall; lexical-syntactic rules fire in about 3% of attempted proofs, and missing world knowledge is the main cause of failure.

Ideas

• I1. en/src/morph.rs and the tree code: a small rule base in the paper's form, a left template matched on a UD tree, a right template generated, and alignment rules that copy the modifiers. Start with the passive-to-active rule and check it on tree pairs built from the school lessons.
• I2. Negation and modality as annotation rules: they set features on nodes and block later rewrites (for example under "never"). Measure how many false entailments appear when the blocking is switched off, on a small negative control set.
• I3. Report exact-proof precision and recall separately, and use a cost threshold only when no proof is found, as the paper does. This keeps every "yes" traceable to the rules that produced it.
• I4. A polarity column for complement-taking verbs: "know that p" licenses p in a positive context, "believe that p" does not. Put the list in the complementation chapter and test it on paired sentences.

For Mova

The chapters are languages/en/28-voice-and-passive.md (the passive rule), languages/en/27-negation.md (negation and polarity) and languages/en/23-complementation.md (complement polarity). The code is in prag/src/infer.rs and world/src/qa.rs. Measure exact-proof precision and recall on a held-out set of pairs, and run it twice to check determinism.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.3115/1654536.1654563

@inproceedings{barhaim2007semantic,
  title = {{Semantic Inference at the Lexical-Syntactic Level for Textual Entailment Recognition}},
  author = {Bar-Haim, Roy and Dagan, Ido and Greental, Iddo and Szpektor, Idan and Friedman, Moshe},
  booktitle = {Proceedings of the ACL-PASCAL Workshop on Textual Entailment and Paraphrasing},
  year = {2007},
  doi = {10.3115/1654536.1654563},
  url = {https://doi.org/10.3115/1654536.1654563}
}

References

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

1. Roy Bar-Haim, Ido Dagan, Iddo Greental, and Eyal Shnarch. 2007. Semantic inference at the lexical-syntactic level. In AAAI (to appear). find
2. Christiane Fellbaum, editor. 1998. WordNet: An Electronic Lexical Database. Language, Speech and Communication. MIT Press. find
3. Dekang Lin and Patrik Pantel. 2001. Discovery of inference rules for question answering. Natural Language Engineering, 4(7):343–360. find
4. Dekang Lin. 1998a. Automatic retrieval and clustering of similar words. In Proceedings of COLING/ACL. find
5. Dekang Lin. 1998b. Dependency-based evaluation of minipar. In Proceedings of the Workshop on Evaluation of Parsing Systems at LREC. find
6. C. Macleod, R. Grishman, A. Meyers, L. Barrett, and R. Reeves. 1998. Nomlex: A lexicon of nominalizations. In EURALEX. find
7. Ruslan Mitkov, Richard Evans, and Constantin Orasan. 2002. A new, fully automatic version of Mitkov’s knowledge-poor pronoun resolution method. In Proceedings of CICLing. find
8. Rowan Nairn, Cleo Condoravdi, and Lauri Karttunen. 2006. Computing relative polarity for textual inference. In Proceedings of ICoS-5. find our card
9. Jeffrey C. Reynar and Adwait Ratnaparkhi. 1997. A maximum entropy approach to identifying sentence boundaries. In Proceedings of ANLP. find
10. Tal Ron. 2006. Generating entailment rules based on online lexical resources. Master’s thesis, Computer Science Department, Bar-Ilan University, Ramat-Gan, Israel. find