Elamur Research⌕
☰
home / papers / 2002 / Using Hidden Markov Modeling to Decompose Human-Written Summaries

Using Hidden Markov Modeling to Decompose Human-Written Summaries

Paper · 2002 · Hongyan Jing

Summary

The paper looks at how professional abstractors build summaries. Rather than writing from scratch, they mostly take fragments of the source text and edit them. Jing frames this as summary sentence decomposition. For each summary sentence, the task asks whether it was built from reused material, which pieces were borrowed, and where in the original those pieces appear. An analysis of human abstracts identifies six editing operations, including reduction, combination, and paraphrasing. The key move is to recast decomposition as choosing one document position for every summary word. A hidden Markov model then scores candidate position sequences. Its transition probabilities come from hand-crafted heuristics about how people reuse text, not from annotated training data. The Viterbi algorithm finds the best sequence. The method is simple and fast and needs no tagger or parser. On sentence-level summary alignment it slightly exceeds Marcu's (1999) system in F-measure. This matters because automatic decomposition can produce large annotated corpora for extractive summarizers. It can also show how humans revise extracted text, which can guide summary generation.

Key points

  • Analysis corpus: 30 human-summarized articles. These were 15 telecommunications items (Benton Foundation), 5 medical items (CDC HIV/STD/TB Prevention News Update), and 10 legal items (New York Law Journal).
  • Six revision operations were identified: shortening individual sentences, merging multiple sentences into one, altering syntactic structure, replacing words with paraphrases, making content more general or more specific, and changing the order of material. A summary sentence counts as cut-and-paste when over half of its words come from borrowed phrases.
  • Each word position is encoded as (SNUM, WNUM). The search space grows quickly; a four-word example already yields 1,936 candidate position sequences.
  • On sentence-level alignment, the HMM scored 81.5% precision, 78.5% recall, and 79.1% F-measure. Marcu's IR- and discourse-based approach scored 77.45%, 80.06%, and 78.15% on the same test set.
  • Unlike the HMM of Brown, Lai, and Mercer (1991) for bilingual alignment, this model uses word position rather than sentence length as its feature. It also requires no aligned training corpus.

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, 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

Cite

Cite as Hongyan Jing. 2002. Using Hidden Markov Modeling to Decompose Human-Written Summaries. Computational Linguistics.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1162/089120102762671972

@article{jing2002using,
  title = {{Using Hidden Markov Modeling to Decompose Human-Written Summaries}},
  author = {Hongyan Jing},
  journal = {Computational Linguistics},
  year = {2002},
  doi = {10.1162/089120102762671972},
  url = {https://doi.org/10.1162/089120102762671972}
}

References

11 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Baum, Leonard E. 1972. An inequality and associated maximization technique in statistical estimation of probabilistic functions of a Markov process. Inequalities, 3:1–8. find
  2. Brown, Peter F., Jennifer C. Lai, and Robert L. Mercer. 1991. Aligning sentences in parallel corpora. In Proceedings of the 29th Annual Meeting of the Association for Computational Linguistics, pages 169–176, Berkeley, June.
  3. Edmundson, H. P. 1969. New methods in automatic abstracting. Journal of the ACM, 16(2):264–285. find
  4. Endres-Niggemeyer, Brigitte, Kai Haseloh, Jens Müller, Simone Peist, Irene Santini de Sigel, Alexander Sigel, Elisabeth Wansorra, Jan Wheeler, and Brünja Wollny. 1998. Summarizing Information. Springer, Berlin. find
  5. Gale, William A. and Kenneth W. Church. 1991. A program for aligning sentences in parallel corpora. In Proceedings of the 29th Annual Meeting of the Association for Computational Linguistics, pages 177–184, Berkeley, June.
  6. Harman, Donna and Mark Liberman. 1993. TIPSTER Complete. Linguistic Data Consortium, University of Pennsylvania. find
  7. Jing, Hongyan. 2001. Cut-and-Paste Text Summarization. Ph.D. thesis, Department of Computer Science, Columbia University, New York. find
  8. Kupiec, Julian, Jan Pedersen, and Francine Chen. 1995. A trainable document summarizer. In Proceedings of the 18th International Conference on Research and Development in Information Retrieval, pages 68–73, Seattle. find
  9. Marcu, Daniel. 1999. The automatic construction of large-scale corpora for summarization research. In Proceedings of the 22nd International Conference on Research and Development and Information Retrieval, pages 137–144, University of California, Berkeley, August. find
  10. Teufel, Simone and Mark Moens. 1997. Sentence extraction as a classification task. In Proceedings of the ACL/EACL’97 Workshop on Intelligent Scalable Text Summarization, pages 58–65, Madrid. find
  11. Viterbi, Andrew J. 1967. Error bounds for convolution codes and an asymptotically optimal decoding algorithm. IEEE Transactions on Information Theory, 13:260–269. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/jing-2002-hidden-markov-modeling-decompose-human/