Morphological Analysis Can Improve a CCG Parser for English

What it is about

The authors test whether English verb morphology, which is poor, still helps a lexicalised parser. In Combinatory Categorial Grammar the verb category often encodes both its argument structure and its inflection, so each verb form gets its own category. The paper splits the inflectional suffix (-s, -ed, -ing and so on) off the verb as a separate token with its own category, after lemmatising the verb with the WordNet lemmatiser. The CCGbank version used has 425 categories at frequency 10 or more, and the verbal categories drop from 159 to 89 after the change (with 15 inflectional categories that serve 96 argument structures). The parser gains 0.8% when gold POS tags are used, but no gain with automatic tags. The authors also correct 'inconsistent' POS tags on verbs, where the suffix and the lexical category disagree, but that correction has its own errors (for example "is" tagged as a noun). The conclusion is that morphology helps only when the POS tags it depends on are correct.

Ideas

• I1. Suffix tokens in en/src/tok.rs: split a verb's inflectional suffix into its own token only when the lemma and the suffix are both known from en/src/morph.rs, and let the parser see the suffix as a feature. Measure LAS on UD EWT dev, with gold and with automatic tags, run twice.
• I2. Make the paper's negative result a gate: the gain must hold with the tagger's own output, not only with gold tags. If it does not, the feature is not used by default.
• I3. Suffix-versus-tag check in en/src/tag.rs: when a word ends in -ing and is tagged as a noun, or ends in -ed and is tagged as an adjective, record the disagreement with its reason and count how often each pair occurs (the paper's Table 3 is this list). Keep the correction table in BTreeMap order.
• I4. Negative control: split the suffix from the wrong verb (shuffled lemmas). Accuracy must fall, or the split is not carrying the morphology.

For Mova

The chapters are 13. Verbs (verb inflection) and 10. Word classes (tag disagreements). The code is en/src/morph.rs, en/src/tok.rs and en/src/tag.rs, with en parse-eval as the measure. The idea of the paper that matters is its warning: any morphology-based gain depends on the POS tags, so measure it with the tagger's output, not only with gold tags. The paper is NC-licensed: use it as a summary and do not carry its tables into a release.

Topics: Part-of-speech tagging

Cite

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

@inproceedings{honnibal2010morphological,
  title = {{Morphological Analysis Can Improve a CCG Parser for English}},
  author = {Honnibal, Matthew and Kummerfeld, Jonathan K. and Curran, James R.},
  booktitle = {Coling 2010: Posters},
  year = {2010},
  url = {https://aclanthology.org/C10-2051/}
}

References

15 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Srinivas Bangalore and Aravind Joshi. 1999. Supertagging: An approach to almost parsing. Computational Linguistics, 25(2):237–265. find
2. Ann Bies, Mark Ferguson, Karen Katz, and Robert MacIntyre. 1995. Bracketing guidelines for Treebank II style Penn Treebank project. Technical report, MS-CIS-95-06, University of Pennsylvania, Philadelphia, PA, USA. find
3. Cem Bozsahin. 2002. The combinatory morphemic lexicon. Computational Linguistics, 28(2):145–186. find
4. Miriam Butt, Mary Dalrymple, and Tracy H. King, editors. 2006. CSLI Publications, Stanford, CA. find
5. Jeongwon Cha, Geunbae Lee, and Jonghyeok Lee. 2002. Korean Combinatory Categorial Grammar and statistical parsing. Computers and the Humanities, 36(4):431–453. find
6. Stephen Clark and James R. Curran. 2004. The importance of supertagging for wide-coverage CCG parsing. In Proceedings of 20th International Conference on Computational Linguistics, pages 282–288. Geneva, Switzerland. find our card
7. Stephen Clark and James R. Curran. 2007. Wide-coverage efficient statistical parsing with CCG and log-linear models. Computational Linguistics, 33(4):493–552. find our card
8. James R. Curran, Stephen Clark, and David Vadas. 2006. Multi-tagging for lexicalized-grammar parsing. In Proceedings of the Joint Conference of the International Committee on Computational Linguistics and the Association for Computational Linguistics, pages 697–704. Sydney, Austrailia. find
9. Julia Hockenmaier and Mark Steedman. 2007. CCGbank: a corpus of CCG derivations and dependency structures extracted from the Penn Treebank. Computational Linguistics, 33(3):355–396. find our card
10. Matthew Honnibal and James R. Curran. 2009. Fully lexicalising CCGbank with hat categories. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 1212–1221. Singapore. find
11. Yusuke Miyao and Jun’ichi Tsuji. 2008. Feature forest models for probabilistic HPSG parsing. Computational Linguistics, 34(1):35–80. find our card
12. Stepan Oepen, Daniel Flickenger, Kristina Toutanova, and Christopher D. Manning. 2004. LinGO Redwoods. a rich and dynamic treebank for HPSG. Research on Language and Computation, 2(4):575–596. find our card
13. Carl Pollard and Ivan Sag. 1994. Head-Driven Phrase Structure Grammar. The University of Chicago Press, Chicago. find
14. Stuart M. Shieber. 1986. An Introduction to Unification-Based Approaches to Grammar, volume 4 of CSLI Lecture Notes. CSLI Publications, Stanford, CA. find
15. Mark Steedman. 2000. The Syntactic Process. The MIT Press, Cambridge, MA, USA. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/honnibal-2010-morphological-analysis-can-improve-ccg/