A Bootstrapping Method for Learning Semantic Lexicons using Extraction Pattern Contexts

Summary

The paper tackles the automatic construction of semantic lexicons, meaning lists of nouns tagged with semantic classes. Such lists matter because general resources like WordNet often miss domain-specific vocabulary. The authors present Basilisk, a weakly supervised bootstrapping algorithm. It needs only an unannotated corpus and ten seed nouns per category, which are the most frequent category nouns, picked by hand.

AutoSlog is first run over the whole corpus so that every noun phrase gets an extraction pattern. Basilisk then repeats a cycle:

it ranks patterns by how reliably they pull out known category members;
it collects candidate nouns from the best-ranked patterns;
it scores each candidate by pooling evidence from all patterns that extract it;
it adds the top five candidates to the lexicon.

The authors also learn several categories at once so that the categories limit one another, and they add this ability to the earlier meta-bootstrapping algorithm too. On six categories in the MUC-4 terrorism corpus, Basilisk is substantially more precise than meta-bootstrapping. Joint learning of categories helps both algorithms, which suggests it could benefit bootstrapping methods in general.

Key points

Data and setup: experiments are run on the MUC-4 corpus, targeting six semantic classes: human, location, building, weapon, event and time.
Patterns: AutoSlog patterns capture a noun phrase in one of three roles: subject, direct object or prepositional-phrase object.
Pattern scoring: patterns are ranked with RlogF. The pattern pool starts at 20 and grows by one each iteration to prevent stagnation. Patterns whose extractions are all already in the lexicon ("depleted") are skipped.
Word scoring (AvgLog): each candidate gets the average of log2(F+1) over every pattern that extracts it, where F is the number of distinct known category members that pattern extracts.
Results: after 1000 words in the human category, 829 entries are correct for the best Basilisk variant (82.9%), versus 278 (27.8%) for meta-bootstrapping. Recall after 1000 learned words ranges from 40% to 60% against the gold standard.

Interesting ideas

I1. Pooled evidence instead of trusting one pattern: a word is judged by all the contexts it appears in, not by a single trusted pattern. This limits semantic drift in any seed-based expansion, for example growing gazetteers or terminology lists.
I2. Log-damped averaging: taking the logarithm of each pattern's count stops one very productive but unspecific context from dominating the score. The trick fits any setting where evidence comes from sources of very different strength.
I3. Slowly growing candidate pool: adding one new pattern per iteration keeps fresh candidates arriving while still favouring the strongest contexts first. It is a simple schedule against stagnation that could transfer to other iterative self-training loops.
I4. Learning competing classes together: when categories are learned jointly with conflict resolution, they block each other's errors. The effect appears mainly once the categories start to overlap. This is useful whenever several related labels are bootstrapped from small seed sets.
I5. Evaluation curves: plotting correct entries against total entries learned shows how precision degrades as the lexicon grows. This reveals more than a single cutoff number would.

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

Fields: Natural language processing

Cited by

87 papers of our library cite it (of the 140884 whose bibliographies we read); 5 of them have a page here, 82 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/1118693.1118721

@inproceedings{thelen2002bootstrapping,
  title = {{A Bootstrapping Method for Learning Semantic Lexicons using Extraction Pattern Contexts}},
  author = {Thelen, Michael and Riloff, Ellen},
  booktitle = {Proceedings of the 2002 Conference on Empirical Methods in Natural Language Processing (EMNLP 2002)},
  year = {2002},
  doi = {10.3115/1118693.1118721},
  url = {https://doi.org/10.3115/1118693.1118721}
}

References

19 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

Chinatsu Aone and Scott William Bennett. 1996. Applying machine learning to anaphora resolution. In Connectionist, Statistical, and Symbolic Approaches to Learning for Natural Language Understanding, pages 302–314. Springer-Verlag, Berlin. find
E. Brill and P. Resnik. 1994. A Transformation-based Approach to Prepositional Phrase Attachment Disambiguation. In Proceedings of the Fifteenth International Conference on Computational Linguistics (COLING-94). find
S. Caraballo. 1999. Automatic Acquisition of a Hypernym-Labeled Noun Hierarchy from Text. In Proceedings of the 37th Annual Meeting of the Association for Computational Linguistics, pages 120–126. find
M. Collins and Y. Singer. 1999. Unsupervised Models for Named Entity Classification. In Proceedings of the Joint SIGDAT Conference on Empirical Me thods in Natural Language Processing and Very Large Corpora (EMNLP/VLC-99). find our card
S. Cucerzan and D. Yarowsky. 1999. Language Independent Named Entity Recognition Combining Morph ological and Contextual Evidence. In Proceedings of the Joint SIGDAT Conference on Empirical Me thods in Natural Language Processing and Very Large Corpora (EMNLP/VLC-99). find
W. Gale, K. Church, and David Yarowsky. 1992. A method for disambiguating word senses in a large corpus. Computers and the Humanities, 26:415–439. find
Niyu Ge, John Hale, and Eugene Charniak. 1998. A statistical approach to anaphora resolution. In Proceedings of the Sixth Workshop on Very Large Corpora. find our card
M. Hearst. 1992. Automatic Acquisition of Hyponyms from Large Text Corpora. In Proceedings of the Fourteenth International Conference on Computational Linguistics (COLING-92). find
Lynette Hirschman, Marc Light, Eric Breck, and John D. Burger. 1999. Deep Read: A reading comprehension system. In Proceedings of the 37th Annual Meeting of the Association for Computational Linguistics. find
M. Marcus, B. Santorini, and M. Marcinkiewicz. 1993. Building a Large Annotated Corpus of English: The Penn Treebank. Computational Linguistics, 19(2):313–330. find our card
George Miller. 1990. Wordnet: An on-line lexical database. In International Journal of Lexicography. find
Dan Moldovan, Sanda Harabagiu, Marius Paşca, Rada Mihalcea, Richard Goodrum, Roxana Gîrju, and Vasile Rus. 1999. Lasso: A tool for surfing the answer net. In Proceedings of the Eighth Text REtrieval Conference (TREC-8). find
MUC-4 Proceedings. 1992. Proceedings of the Fourth Message Understanding Conference (MUC-4). Morgan Kaufmann, San Mateo, CA. find
E. Riloff and R. Jones. 1999. Learning Dictionaries for Information Extraction by Multi-Level Bootstrapping. In Proceedings of the Sixteenth National Conference on Artificial Intelligence. find
E. Riloff and M. Schmelzenbach. 1998. An Empirical Approach to Conceptual Case Frame Acquisition. In Proceedings of the Sixth Workshop on Very Large Corpora, pages 49–56. find
E. Riloff and J. Shepherd. 1997. A Corpus-Based Approach for Building Semantic Lexicons. In Proceedings of the Second Conference on Empirical Methods in Natural Language Processing, pages 117–124. find
E. Riloff. 1996. Automatically Generating Extraction Patterns from Untagged Text. In Proceedings of the Thirteenth National Conference on Artificial Intelligence, pages 1044–1049. The AAAI Press/MIT Press. find
B. Roark and E. Charniak. 1998. Noun-phrase Co-occurrence Statistics for Semi-automatic Semantic Lexicon Construction. In Proceedings of the 36th Annual Meeting of the Association for Computational Linguistics, pages 1110–1116. find
Stephen Soderland, David Fisher, Jonathan Aseltine, and Wendy Lehnert. 1995. CRYSTAL: Inducing a conceptual dictionary. In Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence, pages 1314–1319. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/thelen-2002-bootstrapping-method-learning-semantic-lexicons/

📄 A Bootstrapping Method for Learning Semantic Lexicons using Extraction Pattern Contexts