Construction of English MWE Dictionary and its Application to POS Tagging
What it is about
The authors build an English multiword-expression (MWE) dictionary from the English Wiktionary. They take fixed functional expressions (adverbs, prepositions, conjunctions, determiners, prepositional phrases and pronouns, such as "after all", "according to", "no one") and leave out compound nouns, phrasal verbs and multiword adjectives. They then find each entry in the Penn Treebank: 1,501 adverb types in Wiktionary, of which 468 occur in the PTB, and so on. Occurrences are split into MWE and literal uses, first by clustering the phrase-structure trees and then by hand for the ambiguous trees. The PTB is also corrected for 1,084 inconsistent POS tags inside MWEs. A CRF tagger is trained on four versions of the PTB (original, revised, BIO-marked MWE, and MWE as one token). Per-token accuracy is 97.54% for the original, 97.56% for the revised, 97.32% for BIO and 97.62% for the MWE-as-one-token version.
Ideas
- I1. Multiword units in
en/src/tok.rs: treat a closed list of function-word MWEs ("about to", "in spite of") as one token before tagging, as the fourth version does. Measure per-token accuracy on UD EWT dev with and without the merge, run twice. - I2. Count-based majority correction: for each MWE, take the majority POS of its internal words over the corpus and fix outliers, as the paper does with the RB/NN inconsistencies. Report how many tags change and which ones, so the change is traceable.
- I3. Negative control: merge a random set of word pairs that are not MWEs. Accuracy must fall; otherwise the gain comes from the merge rule's coverage and not from the MWE sense.
- I4. Keep the MWE-literal decision explainable: for each candidate, store the tree cluster that decided it (the paper's cluster-first step), so a literal or MWE reading can say why it was chosen.
For Mova
The chapter is 19. The lexicon: vocabulary, multiword expressions and register, with 07. Punctuation, capital letters, abbreviations and numbers for tokenisation of spaced forms. The code is en/src/tok.rs (token merge), en/src/tag.rs and en/src/ptag.rs (tagging). The measure is per-token accuracy on UD EWT dev, twice, plus the negative control. The source Wiktionary data are used only as raw input and must not be carried into the build as an edited release.
Cite
@inproceedings{shigeto2013construction,
title = {{Construction of English MWE Dictionary and its Application to POS Tagging}},
author = {Shigeto, Yutaro and Azuma, Ai and Hisamoto, Sorami and Kondo, Shuhei and Kose, Tomoya and Sakaguchi, Keisuke and Yoshimoto, Akifumi and Yung, Frances and Matsumoto, Yuji},
booktitle = {Proceedings of the 9th Workshop on Multiword Expressions},
year = {2013},
url = {https://aclanthology.org/W13-1021/}
}References
- Matthieu Constant and Anthony Sigogne. 2011. MWU-Aware Part-of-Speech Tagging with a CRF Model and Lexical Resources. In Proceedings of the Workshop on Multiword Expressions: from Parsing and Generation to the Real World, MWE ’11, pages 49–56. find
- Matthieu Constant, Anthony Sigogne, and Patrick Watrin. 2012. Discriminative Strategies to Integrate Multiword Expression Recognition and Parsing. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics, ACL ’12, pages 204– 212. find
- Mark Alan Finlayson and Nidhi Kulkarni. 2011. Detecting Multi-Word Expressions improves Word Sense Disambiguation. In Proceedings of the Workshop on Multiword Expressions: from Parsing and Generation to the Real World, MWE ’11, pages 20–24. find
- Spence Green, Marie-Catherine de Marneffe, John Bauer, and Christopher D Manning. 2011. Multiword Expression Identification with Tree Substitution Grammars: A Parsing tour de force with French. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, EMNLP ’11, pages 725–735. find
- Nicole Grégoire. 2007. Design and Implementation of a Lexicon of Dutch Multiword Expressions. In Proceedings of the Workshop on a Broader Perspective on Multiword Expressions, MWE ’07, pages 17–24. find
- Ioannis Korkontzelos and Suresh Manandhar. 2010. Can Recognising Multiword Expressions Improve Shallow Parsing? In Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the Association for Computational Linguistics, HLT ’10, pages 636–644. find
- Taku Kudo, Kaoru Yamamoto, and Yuji Matsumoto. 2004. Applying conditional random fields to japanese morphological analysis. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, EMNLP ’04, pages 230–237. find
- Eric Laporte and Stavroula Voyatzi. 2008. An Electronic Dictionary of French Multiword Adverbs. In Language Resources and Evaluation Conference. Workshop Towards a Shared Task for Multiword Expressions, MWE ’08, pages 31–34. find
- Eric Laporte, Takuya Nakamura, and Stavroula Voyatzi. 2008. A French Corpus Annotated for Multiword Nouns. In Proceedings of the Language Resources and Evaluation Conference. Workshop Towards a Shared Task on Multiword Expressions, MWE ’08, pages 27–30. find
- Joakim Nivre and Jens Nilsson. 2004. Multiword Units in Syntactic Parsing. In Workshop on Methodologies and Evaluation of Multiword Units in Real-World Applications, MEMURA ’04, pages 39–46. find
- Ivan A Sag, Timothy Baldwin, Francis Bond, Ann A Copestake, and Dan Flickinger. 2002. Multiword Expressions: A Pain in the Neck for NLP. In Proceedings of the Third International Conference on Computational Linguistics and Intelligent Text Processing, CICLing ’02, pages 1–15. find our card
- Kosho Shudo, Akira Kurahone, and Toshifumi Tanabe. 2011. A Comprehensive Dictionary of Multiword Expressions. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, HLT ’11, pages 161– 170. find
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/shigeto-2013-construction-english-mwe-dictionary-its/