Multiword Expressions in the wild? The mwetoolkit comes in handy
What it is about
The paper is a demonstration of mwetoolkit, a tool that extracts multiword expression (MWE) candidates from a monolingual corpus. It first generates candidates from surface n-grams, then filters them with association measures such as Dice and PMI computed from n-gram counts. The authors run it in its standard setting on general English and discuss its pros and cons with selected examples. Its strength is that it is language independent and needs no parser. Its limit is that it uses only shallow information, so the lists need language tools for better precision. The paper gives no precision or recall figures. It cites the estimate that specialised lexica contain 50% to 70% multiword entries, and that 56.7% of the terms in the GENIA corpus are multi-word.
Ideas
- I1. Candidate extraction for
en/src/lexgen.rsfrom the Mova raw collection: count n-grams with BTreeMap, score them with Dice and PMI, and list the top candidates that are not yet indata/lemmas.tsv. Measure: precision of the top 500 candidates by hand, with two identical runs. - I2. Recursive filtering as the paper suggests: apply the association measure over increasing n-gram length, and keep a longer candidate only if it beats its parts. Measure: candidate precision against the unrecursive version on the same corpus.
- I3. Use the candidate list as a coverage probe for
en tok-eval: how many EWT test multi-word units are absent from the lexicon. Measure: the share covered, before any lexicon change.
For Mova
The chapter is 19. The lexicon: vocabulary, multiword expressions and register. The code is en/src/lexgen.rs, with a small Rust counter under tools/ rather than a shell script. The paper's data is not named, and its numbers are not for Mova. Any precision claim must come from EWT test.
Cite
@inproceedings{ramisch2010multiword,
title = {{Multiword Expressions in the wild? The mwetoolkit comes in handy}},
author = {Ramisch, Carlos and Villavicencio, Aline and Boitet, Christian},
booktitle = {Coling 2010: Demonstrations},
year = {2010},
url = {https://aclanthology.org/C10-3015/}
}References
- Evert, Stefan and Brigitte Krenn. 2005. Using small random samples for the manual evaluation of statistical association measures. Comp. Speech & Lang. Special issue on MWEs, 19(4):450–466. find
- Jackendoff, Ray. 1997. Twistin’ the night away. Language, 73:534–559. find
- Pecina, Pavel. 2008. Reference data for czech collocation extraction. In Proc. of the LREC Workshop Towards a Shared Task for MWEs (MWE 2008), pages 11–14, Marrakech, Morocco, Jun. find
- Ramisch, Carlos, Helena de Medeiros Caseli, Aline Villavicencio, André Machado, and Maria José Finatto. 2010a. A hybrid approach for multiword expression identification. In Proc. of the 9th PROPOR (PROPOR 2010), volume 6001 of LNCS (LNAI), pages 65–74, Porto Alegre, RS, Brazil. Springer. find
- Ramisch, Carlos, Aline Villavicencio, and Christian Boitet. 2010b. mwetoolkit: a framework for multiword expression identification. In Proc. of the Seventh LREC (LREC 2010), Malta, May. ELRA. find
- Ramisch, Carlos, Aline Villavicencio, and Christian Boitet. 2010c. Web-based and combined language models: a case study on noun compound identification. In Proc. of the 23th COLING (COLING 2010), Beijing, China, Aug. find
- Sag, Ivan, Timothy Baldwin, Francis Bond, Ann Copestake, and Dan Flickinger. 2002. Multiword expressions: A pain in the neck for NLP. In Proc. of the 3rd CICLing (CICLing-2002), volume 2276/2010 of LNCS, pages 1–15, Mexico City, Mexico, Feb. Springer. find our card
- Seretan, Violeta. 2008. Collocation extraction based on syntactic parsing. Ph.D. thesis, University of Geneva, Geneva, Switzerland. find
- Smadja, Frank A. 1993. Retrieving collocations from text: Xtract. Comp. Ling., 19(1):143–177. find
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2010-ramisch-multiword-expressions-wild-mwetoolkit-comes/