Elamur Research⌕
☰
home / papers / 2008 / The Generation of Textual Entailment with NLML in an Intelligent Dialogue system for Language Learning CSIEC

The Generation of Textual Entailment with NLML in an Intelligent Dialogue system for Language Learning CSIEC

Paper · 2008 · Jiyou Jia

Summary

The report tackles generating textual entailment (GTE): given one sentence, produce other sentences that follow from it. This differs from the better-known recognition task (RTE), which only checks whether a given pair holds. The work sits inside CSIEC, a web chatbot for English learners. There, derived sentences help in four ways: they keep the user-facts store free of duplicates, shorten scripted dialogues, support question answering, and let replies be simplified to a learner's level. The approach copies how teachers present sentence patterns. Hand-written rules in NLML, a markup mixing phrase-structure and dependency information, contain pseudo variables such as <person> or <name>. An input sentence is parsed into NLML and then into the NLOMJ object model. It is matched against rules by mood and clause structure, and the bound values are inserted into the rule's output side, with the verb re-inflected for tense and subject. A scaling test with 10,000 rules showed roughly constant time but memory growing linearly with rule count. The work matters as an early, transparent, rule-based take on entailment generation.

Key points

  • The paper separates two tasks. GTE means producing what a text T implies, while RTE means deciding whether T entails H. The author found almost no earlier work on GTE, including in the PASCAL RTE proceedings.
  • The pipeline runs in this order:
    • preprocess the input,
    • parse it to NLML and then to NLOMJ,
    • compare it with the rule database,
    • transform the matched rule's output into new NLML and text,
    • repeat on the outputs to get deeper entailments.
  • If no rule fits, a fallback called "logical entailment" applies common-sense steps such as hypernym substitution (e.g. a dog → an animal).
  • TEE (Textual Entailment Editor) is a Java GUI that builds a rule from one example pair through button-guided choices. Annotators working with it wrote about 10 rules per hour.
  • In the scaling test, 10,000 rules (only 20 distinct) gave about 100 ms per input and used about 300 MB of memory. Planned external resources are Sekine's Paraphrase Database and TEASE (136 templates), set against DIRT's 231,000 unique rules.

Interesting ideas

  • I1. Rules are written as templates with typed slots (<person>, <name>), and matching binds those slots so the output side can be filled in. This keeps every inference traceable to one named rule.
  • I2. Before slot matching, a candidate rule is first filtered by sentence mood and clause shape. This cheap pre-filter keeps matching fast and avoids applying a statement rule to a question.
  • I3. The rule's output side is stored in base form, and tense and agreement are copied over from the input at generation time. One rule therefore covers "is", "was" and "has been".
  • I4. Each rule records the ID of its reverse rule, so chained ("deeper") entailment does not loop between equivalent forms.
  • I5. Entailment is used to spot repeated user facts: a new statement that follows from a stored one is not saved again.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (frames): Paraphrase and entailment rules could be stated over frame slots with role types, so each derived frame records which rule and which bindings produced it.
  • I2 (frames): Rules could be indexed by speech act and the predicate's dependency skeleton, so only structurally compatible rules are tried on a given frame.
  • I3 (back): The generator could take tense, aspect and person features from the source frame and realize them on the entailed clause, rather than storing inflected strings.
  • I4 (world): When the world model closes over entailments, recording each rule's inverse would prevent endless A↔B rewriting and duplicate states.
  • I5 (world): Before adding a state or holding to the world of the text, Mova could check whether existing facts already entail it, and if so record it as a confirmation instead.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Peking UniversityBeijing · China244 works · 39,359 citations🎓Korea Advanced Institute of Science and TechnologyDaejeon · South Korea51 works · 3,167 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Reasoning · 1,426 worksDialogue systems · 624 works

Cite

Cite as Jiyou Jia. 2008. The Generation of Textual Entailment with NLML in an Intelligent Dialogue system for Language Learning CSIEC. arXiv:0802.4326.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.0802.4326 · arXiv 0802.4326

@misc{jia2008generation,
  title = {{The Generation of Textual Entailment with NLML in an Intelligent Dialogue system for Language Learning CSIEC}},
  author = {Jiyou Jia},
  year = {2008},
  doi = {10.48550/arxiv.0802.4326},
  eprint = {0802.4326},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.0802.4326}
}

References

24 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. R. Bar-Haim, I. Szpecktor, and O. Glickman. 2005. Definition and analysis of intermediate entailment levels. Proceedings of the ACL Workshop on Empirical Modelling of Semantic Equivalence and Entailment, pp. 55–60. Association for Computational Linguistics. find
  2. R. Bar-Haim, I. Dagan, B. Dolan, L. Ferro, D. Giampiccolo, B. Magnini and I. Szpektor. 2006. The Second PASCAL Recognising Textual Entailment Challenge. Proceedings of the Second PASCAL Challenges Workshop on Recognising Textual Entailment. find
  3. J. Bos & K. Markert. 2005. Recognising Textual Entailment with Logical Inference. Proceedings of EMNLP 2005.
  4. R. Braz, R. Girju, V. Punyakanok, D. Roth, and M. Sammons. 2005. Knowledge Representation for Semantic Entailment and Question-Answering. IJCAI-05 Workshop on Knowledge and Reasoning for Answering Questions. find
  5. I. Dagan, O. Glickman and B. Magnini. 2005. Proceedings of the PASCAL Challenges Workshop on Recognising Textual Entailment. LNCS, Vol. 3944. find
  6. C. Fellbaum. 1998. Ed.: WordNet: An Electronic Lexical database. MIT Press. find
  7. M. Geffet and I. Dagan. 2004. Feature Vector Quality and Distributional Similarity. Proceedings of the 20th International Conference on Computational Linguistics (COLING). find
  8. M. Geffet and I. Dagan. 2005. The Distributional Inclusion Hypotheses and Lexical Entailment. ACL 2005, Michigan, USA.
  9. U. Hahn and K. Schnattinger. 1998. Towards text knowledge engineering. Proceedings of AAAI-98. pp. 524-531. Menlo Park, California. find
  10. A. Haghighi, A. Y. Ng, C. D. Manning. 2005. Robust Textual Inference via Graph Matching. HLT-EMNLP 2005. find
  11. T Hasegawa, S. Sekine, R. Grishman. 2004. Discovering Relations among Named Entities from Large Corpora, ACL-04.
  12. J. Jia. 2004, a. CSIEC (Computer Simulator in Educational Communication): A Virtual Context-Adaptive Chatting Partner for Foreign Language Learners (in English). In Kinshuk, etc. (Eds.), Proceedings of ICALT 04, P.690-692. New York: IEEE Press. find
  13. J. Jia. 2004, b. NLML - a Markup Language to Describe the Unlimited English Grammar. CoRR cs.CL/0404018: (2004). — NLML — a Markup Language to Describe the Unlimited English Grammar
  14. J. Jia, Y. Ye, K. Mainzer. 2004. NLOMJ — Natural Language Object Model in Java, in Z. Shi, Q. He (Eds.), Intelligent information processing II. Proceedings of ICICP 04, P.201-209. New York: Springer. find
  15. C.H.A. Koster. 1991. Affix Grammars for natural languages. Attribute Grammars, Applications and Systems, International Summer School. Lecture Notes in Computer Science, volume 545. Springer. find
  16. D. Lin. 1998. Dependency-based evaluation of MINIPAR. Proceedings of the Workshop on Evaluation of Parsing Systems at LREC. find
  17. D. Lin, P. Pantel. 2001. Discovery of inference rules for question-answering. Natural Language Engineering, 7: 343-360 Cambridge University Press. ↗
  18. B. MacCartney, T. Grenager, M. de Marneffe, D. Cer and C. D. Manning. 2006. Learning to Recognize Features of Valid Textual Entailments. HLT-NAACL 2006. find
  19. V. Pekar. 2006. Acquisition of Verb Entailment from Text. HLT-NAACL 2006. find
  20. R. Raina, A. Y. Ng, and C. Manning. 2005. Robust textual inference via learning and abductive reasoning. Twentieth National Conference on Artificial Intelligence (AAAI-05). find
  21. L. Romano, M. Kouylekov, I. Szpektor, I. Dagan and A. Lavelli. 2006. Investigating a Generic Paraphrase-based Approach for Relation Extraction. EACL 2006. find
  22. V. Rus, A. Graesser and K. Desai. 2005. Lexico-Syntactic Subsumption for Textual Entailment. RANLP 2005. find
  23. S. Sekine. 2005. Automatic Paraphrase Discovery based on Context and Keywords between NE Pairs. IWP-05. Y. Shinyama, S. Sekine, K. Sudo and R. Grishman, Automatic Paraphrase Acquisition from News Articles, HLT-02. find
  24. M. Stefik. 1995. Introduction to Knowledge Systems. San Francisco, CA: Morgan Kaufmann. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/jia-2008-generation-textual-entailment-nlml-intelligent/