Elamur Research
☰
home / papers / 2010 / A Monolingual Tree-based Translation Model for Sentence Simplification

A Monolingual Tree-based Translation Model for Sentence Simplification

What it is about

The paper treats sentence simplification as translation from complex to simple sentences, with the complex parse tree as the source. Its Tree-based Simplification Model (TSM) covers four operations in one model: splitting, dropping, reordering and substitution. Each operation is driven by a learned feature table over tree nodes, such as a node's constituent label and its children, and the probabilities of all operation sequences that yield the same output are summed. To train it, the authors build PWKP, a parallel corpus of Wikipedia and Simple Wikipedia sentences, by aligning article pairs at sentence level with TF-IDF similarity. Evaluation uses BLEU, NIST and a readability formula; the model scores better on readability than the baseline systems. The work is the reference point for later tree-based simplifiers in this group of papers.

Ideas

For Mova

The dropping and reordering tables are the counted form of the operations the translator needs when an author removes or moves material. The split-and-complete step fits tools/roundtrip and the encoder's chunking (open problem 3). The model itself is not for Mova; the counted tables and the verse-free alignment fallback are.

Library summary

The Research library's own summary of the paper, in our words — not the paper's text.

Written for the Research library (claude-opus-5-5) from the paper text, in our own words: the license "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)" does not allow redistributing the text itself. Original — aclanthology.org/C10-1152.pdf

Summary

The paper frames sentence simplification as monolingual translation, where a complex sentence is the source and its simpler version is the target. Earlier systems usually handled only one operation, such as splitting or compression. The authors propose the Tree-based Simplification Model (TSM), a statistical model that works on parse trees. It combines four operations in one probabilistic framework: splitting, dropping, reordering and word/phrase substitution. Training uses an EM algorithm, sped up by monolingual word mapping, and a greedy decoder with language models produces the output.

To get training data, they built PWKP, a corpus of more than 108k complex–simple sentence pairs. It was aligned automatically from English Wikipedia and Simple English Wikipedia using sentence-level TF*IDF. On a 100-sentence test set, TSM scored lower on BLEU than Moses but had the best Flesch, Lix and perplexity scores. The authors conclude that BLEU is a poor guide for judging simplification. PWKP became an early large resource for this task.

Key points

  • PWKP corpus: 108,016 sentence pairs drawn from 65,133 paired articles. Average sentence length is 25.01 tokens for complex sentences and 20.87 for simple ones. 1-to-n alignments are allowed so that sentence splits are captured.
  • Choice of alignment measure: Three similarity measures were tested on 120 manually annotated pairs, with recall held near 55%. Precision was 91.3% for sentence-level TF*IDF, 50.5% for word overlap and 13.9% for MED.
  • How operations are scored: Each operation draws its probabilities from a learned feature table:
    • SFT for segmentation;
    • BDFT for dropping the border word;
    • CFT for copying;
    • DFT for dropping;
    • RFT for reordering.
  • BLEU and NIST results: Moses scored 0.55 BLEU and TSM 0.38, while the unchanged complex sentences already reached 0.50. Moses copied 25 of 100 inputs verbatim; TSM did so for only 2.
  • Readability results: TSM achieved Flesch 67.4, Lix 36.7 (school grade 5) and the lowest perplexity (353). It ranked second on the share of words outside the BE850 list.

Interesting ideas

  • Splitting as two steps: Splitting is handled as segmentation (where to cut) followed by completion (dropping the connecting word and copying the missing subject phrase). Separating these decisions makes each one easier to learn and to debug in any rewriting system.
  • Relative length feature: Instead of raw sentence length, the model uses sentence length divided by the average simple-sentence length, rounded up (iLength). Bucketing a continuous feature against a corpus reference like this reduces sparsity in count-based models.
  • Adjusted constituent: A word's syntactic context is found by climbing the tree while the node is the leftmost child, so "which" maps to SBAR rather than WHNP. This cheap heuristic captures the clause a word opens, which helps identify boundaries.
  • Copy-input baseline: The authors score the unmodified source sentences against the references and count outputs identical to the input. This exposes metrics that reward systems for doing almost nothing, and it is useful for any task where output and input overlap heavily.
  • Readability-oriented evaluation: Alongside BLEU, they report readability indices, out-of-vocabulary rate against a basic word list, and perplexity under a language model trained on simple text. This combination could transfer to evaluating simplification in other languages, including Ukrainian.

Topics: Machine translation

Cite

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

@inproceedings{zhu2010monolingual,
  title = {{A Monolingual Tree-based Translation Model for Sentence Simplification}},
  author = {Zhu, Zhemin and Bernhard, Delphine and Gurevych, Iryna},
  booktitle = {Proceedings of the 23rd International Conference on Computational Linguistics (Coling 2010)},
  year = {2010},
  url = {https://aclanthology.org/C10-1152/}
}

References

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

  1. Barzilay, Regina and Noemie Elhadad. 2003. Sentence alignment for monolingual comparable corpora. In Proceedings of the 2003 Conference on Empirical Methods in Natural Language Processing, pages 25–32. find
  2. Carroll, John, Guido Minnen, Darren Pearce, Yvonne Canning, Siobhan Devlin, and John Tait. 1999. Simplifying text for language-impaired readers. In Proceedings of the 9th Conference of the European Chapter of the Association for Computational Linguistics (EACL’99), pages 269–270. find
  3. Chandrasekar, R., Christine Doran, and B. Srinivas. 1996. Motivations and methods for text simplification. In Proceedings of the Sixteenth International Conference on Computational Linguistics (COLING’96), pages 1041–1044. find
  4. Filippova, Katja and Michael Strube. 2008a. Dependency tree based sentence compression. In International Natural Language Generation Conference (INLG’08), pages 25–32. find
  5. Filippova, Katja and Michael Strube. 2008b. Sentence fusion via dependency graph compression. In EMNLP ’08: Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 177–185. find
  6. Graehl, Jonathan, Kevin Knight, and Jonathan May. 2008. Training tree transducers. In Computational Linguistics, volume 34, pages 391–427. MIT Press. find
  7. Heilman, M. and N. A. Smith. 2009. Question generation via overgenerating transformations and ranking. Technical Report CMU-LTI-09-013, Language Technologies Institute, Carnegie Mellon University. find
  8. Inui, Kentaro, Atsushi Fujita, Tetsuro Takahashi, Ryu Iida, and Tomoya Iwakura. 2003. Text simplification for reading assistance: A project note. In Proceedings of the 2nd International Workshop on Paraphrasing: Paraphrase Acquisition and Applications (IWP), pages 9–16. find
  9. Jonnalagadda, Siddhartha and Graciela Gonzalez. 2009. Sentence simplification aids protein-protein interaction extraction. In Proceedings of the 3rd International Symposium on Languages in Biology and Medicine. find
  10. Jurafsky, Daniel and James H. Martin. 2008. Speech and Language Processing (2nd Edition). Prentice Hall, 2 edition. find
  11. Klein, Dan and Christopher D. Manning. 2003. Fast exact inference with a factored model for natural language parsing. In Advances in Neural Information Processing Systems 15 (NISP’02), pages 3–10. find
  12. Knight, Kevin and Daniel Marcu. 2000. Statistics-based summarization - step one: Sentence compression. In AAAI, pages 703–710. find
  13. Levenshtein. 1966. Binary code capable of correcting deletions, insertions and reversals. In Soviet Physics, pages 707–710. find
  14. Nelken, Rani and Stuart M. Shieber. 2006. Towards robust context-sensitive sentence alignment for monolingual corpora. In Proceedings of 11th Conference of the European Chapter of the Association for Computational Linguistics, pages 161–168. find
  15. Petersen, Sarah E. and Mari Ostendorf. 2007. Text simplification for language learners: a corpus analysis. In Proc. of Workshop on Speech and Language Technology for Education, pages 69–72. find
  16. Schmid, Helmut. 1994. Probabilistic part-of-speech tagging using decision trees. In International Conference on New Methods in Language Processing, pages 44–49. find
  17. Schwarm, Sarah E. and Mari Ostendorf. 2005. Reading level assessment using support vector machines and statistical language models. In ACL’05: Proceedings of the 43rd Annual Meeting on Association for Computational Linguistics, pages 523–530. find
  18. Siddharthan, Advaith. 2002. An architecture for a text simplification system. In Proceedings of the Language Engineering Conference (LEC’02), pages 64–71. find
  19. Siddharthan, Advaith. 2006. Syntactic simplification and text cohesion. In Research on Language & Computation, volume 4, pages 77–109. Springer Netherlands, June. find
  20. Stolcke, Andreas. 2002. SRILM - An Extensible Language Modeling Toolkit. pages 901–904. find
  21. Vickrey, David and Daphne Koller. 2008. Sentence simplification for semantic role labeling. In Proceedings of ACL-08: HLT, pages 344–352, June. find
  22. Watanabe, Willian Massami, Arnaldo Candido Junior, Vinı́cius Rodriguez Uzêda, Renata Pontin de Mattos Fortes, Thiago Alexandre Salgueiro Pardo, and Sandra Maria Aluı́sio. 2009. Facilita: reading assistance for low-literacy readers. In SIGDOC ’09: Proceedings of the 27th ACM international conference on Design of communication, pages 29–36. ACM. find
  23. Yamada, Kenji and Kevin Knight. 2001. A syntax-based statistical translation model. In ACL’01: Proceedings of the 39th Annual Meeting on Association for Computational Linguistics, pages 523–530. find our card
  24. Yamada, Kenji and Kevin Knight. 2002. A decoder for syntax-based statistical mt. In ACL’02: Proceedings of the 40th Annual Meeting on Association for Computational Linguistics, pages 303–310. find
  25. Zesch, Torsten, Christof Müller, and Iryna Gurevych. 2008. Extracting Lexical Semantic Knowledge from Wikipedia and Wiktionary. In Proceedings of the Sixth International Language Resources and Evaluation (LREC’08), pages 1646–1652. find
  26. Zhao, Shiqi, Xiang Lan, Ting Liu, and Sheng Li. 2009. Application-driven statistical paraphrase generation. In Proceedings of ACL-IJCNLP, pages 834–842, Suntec, Singapore, August. find