A unified approach to sentence segmentation of punctuated text in many languages

What it is about

Sentence segmentation is the first step of many NLP pipelines, yet it has few standard test sets and most tools are hand-written for one language. The authors model segmentation as a decision at each punctuation candidate, using a context window of subword tokens. The model is small: the embedding size is 32 and the vocabulary is 125 in the minimal setting. It is trained on noisily annotated data, with candidate positions labelled by a regular expression. The authors release a 23-language evaluation set and a tool, Ersatz. The model beats strong baselines on the English WSJ and Brown sets and performs well on average across the 23 languages. A single multilingual model performs nearly as well as the per-language ones, and the paper finds that the left context helps more than the right context.

Ideas

• I1. en/src/tok.rs: a context window of the tokens on both sides of each period, with a small table of weights over the token shapes, not subwords. Measure the boundary F1 on EWT test for left-only, right-only and both windows, which tests the paper's finding that left context matters more.
• I2. Use the paper's observation that the period is the most ambiguous mark: report the segmentation F1 separately for ".", "?" and "!" on EWT test, to see where the errors are before changing the model.
• I3. Build the noisy training labels for English from the regular-expression candidates of en/src/tok.rs, then check how many of them the paper's labelling rule gets wrong on the EWT test sentences. This measures label noise, which the paper says the model must tolerate.

For Mova

The code is en/src/tok.rs, and the metric is en tok-eval. The chapter is 07. Punctuation, capital letters, abbreviations and numbers. The paper's model uses subwords and is neural, so only the window idea and the per-mark report carry over into Mova's table-driven rules. The 23-language set is not English, so take numbers from EWT test only. Measure first on EWT test with a two-fold check (en folds).

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/2021.acl-long.309

@inproceedings{wicks2021unified,
  title = {{A unified approach to sentence segmentation of punctuated text in many languages}},
  author = {Wicks, Rachel and Post, Matt},
  booktitle = {Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers)},
  year = {2021},
  doi = {10.18653/v1/2021.acl-long.309},
  url = {https://doi.org/10.18653/v1/2021.acl-long.309}
}

References

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

1. Table 6: Time in seconds for 1 million English tokens in input file. F1 is score on English Test Set. We show various size encoders for our method. Linear is a linear layer with TanH activation.* Loïc Barrault, Magdalena Biesialska, Ondřej Bojar, Marta R. Costa-jussà, Christian Federmann, Yvette Gr… find
2. David D. Palmer and Marti A. Hearst. 1997. Adaptive multilingual sentence boundary disambiguation. Computational Linguistics, 23(2):241–267. find
3. Steven Bird and Edward Loper. 2004. NLTK: The natural language toolkit. In Proceedings of the ACL Interactive Poster and Demonstration Sessions, pages 214–217, Barcelona, Spain. Association for Computational Linguistics. find
4. Ines Rehbein, Josef Ruppenhofer, and Thomas Schmidt. 2020. Improving sentence boundary detection for spoken language transcripts. In Proceedings of the 12th Language Resources and Evaluation Conference, pages 7102–7111, Marseille, France. European Language Resources Association. find
5. Mauro Cettolo, Marcello Federico, Luisa Bentivogli, Jan Niehues, Sebastian Stüker, Katsuitho Sudoh, Koichiro Yoshino, and Christian Federmann. 2017. Overview of the iwslt 2017 evaluation campaign. In 14th International Workshop on Spoken Language Translation, pages 2–14, Tokyo, Japan. find
6. Michael D. Riley. 1989. Some applications of tree-based modelling to speech and language. In Speech and Natural Language: Proceedings of a Workshop Held at Cape Cod, Massachusetts, October 15-18, 1989. find
7. Dan Gillick. 2009. Sentence boundary detection and the problem with the U.S. In Proceedings of Human Language Technologies: The 2009 Annual Conference of the North American Chapter of the Association for Computational Linguistics, Companion Volume: Short Papers, pages 241–244, Boulder, Colorado. Ass… find our card
8. Eric Joanis, Rebecca Knowles, Roland Kuhn, Samuel Larkin, Patrick Littell, Chi-kiu Lo, Darlene Stewart, and Jeffrey Micher. 2020. The Nunavut hansard Inuktitut–English parallel corpus 3.0 with preliminary machine translation results. In Proceedings of the 12th Language Resources and Evaluation Confe… find
9. Tibor Kiss and Jan Strunk. 2006. Unsupervised multilingual sentence boundary detection. Computational Linguistics, 32(4):485–525. find our card
10. Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondřej Bojar, Alexandra Constantin, and Evan Herbst. 2007. Moses: Open source toolkit for statistical machine translation. In Pro… find
11. Taku Kudo and John Richardson. 2018. SentencePiece: A simple and language independent subword tokenizer and detokenizer for neural text processing. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 66–71, Brussels, Belgium. Associ… find
12. Khanh Nguyen and Hal Daumé III. 2019. Global Voices: Crossing borders in automatic news summarization. In Proceedings of the 2nd Workshop on New Frontiers in Summarization, pages 90–97, Hong Kong, China. Association for Computational Linguistics. find
13. George Sanchez. 2019. Sentence boundary detection in legal text. In Proceedings of the Natural Legal Language Processing Workshop 2019, pages 31–38, Minneapolis, Minnesota. Association for Computational Linguistics. find
14. Holger Schwenk, Vishrav Chaudhary, Shuo Sun, Hongyu Gong, and Francisco Guzmán. 2019a. Wikimatrix: Mining 135m parallel sentences in 1620 language pairs from wikipedia. CoRR, abs/1907.05791. find
15. Holger Schwenk, Guillaume Wenzek, Sergey Edunov, Edouard Grave, and Armand Joulin. 2019b. Ccmatrix: Mining billions of high-quality parallel sentences on the WEB. CoRR, abs/1911.04944. find
16. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Ł ukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in Neural Information Processing Systems, volume 30. Curran Associates, Inc. find
17. Xiaolin Wang, Masao Utiyama, and Eiichiro Sumita. 2019. Online sentence segmentation for simultaneous interpretation using multi-shifted recurrent neural network. In Proceedings of Machine Translation Summit XVII Volume 1: Research Track, pages 1–11, Dublin, Ireland. European Association for Machine… find
18. Nianwen Xue and Yaqin Yang. 2011. Chinese sentence segmentation as comma classification. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, pages 631–635, Portland, Oregon, USA. Association for Computational Linguistics. find
19. Nina Zhou, AiTi Aw, Nattadaporn Lertcheva, and Xuancong Wang. 2016. A word labeling approach to Thai sentence boundary detection and POS tagging. In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 319–327, Osaka, Japan. find
20. The COLING 2016 Organizing Committee. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/acl-2021-wicks-unified-approach-sentence-segmentation-punctuated/