Learning a POS tagger for AAVE-like language

What it is about

Newswire POS taggers do poorly on subtitles, hip-hop lyrics and tweets in African-American Vernacular English. The authors train on 1,827 annotated tweets plus about 4.5 million unlabelled sentences. They partly label the unlabelled text with tag dictionaries mined from Wiktionary, a slang glossary, a list of names and Urban Dictionary. Two methods use the dictionaries: ambiguous supervision in a cost-sensitive structured perceptron, where the mined tags carry no cost, and self-training with the dictionary as type constraints. The best system reaches 89% accuracy on subtitles, 85% on lyrics and 83% on tweets. The abstract reports up to 55% error reduction over a newswire tagger and 15 to 25% over a Twitter tagger. Most of the gain is on unseen words (ambiguous supervision cuts their error by 14% on average), while known words gain about 1%.

Ideas

• I1. en/src/ptag.rs: an ambiguous-supervision update for the existing perceptron. If the predicted tag is inside a word's mined tag set, make no update; otherwise update with cost 1. Measure on EWT dev and on a small hand-annotated AAVE-like set, run twice.
• I2. Type constraints at tagging time: for known words, allow only the tags in the lexicon of en/src/dict.rs. Report accuracy on known and unseen words apart, because the paper's gains are almost all on unseen words.
• I3. The sources matter: Urban Dictionary scraping is not usable under our rules. A Wiktionary dump is CC BY-SA, so quote only and do not carry it into a release.
• I4. Stop self-training by a rule the paper uses: stop when dev accuracy falls over three iterations. Keep the stopping point in the log.

For Mova

The tagger chapters are 10. Word classes and 37. Register and variation. Spelling variants such as g-dropping belong in 06. Spelling rules: inflection, derivation, compounds and national standards, and slang in 19. The lexicon: vocabulary, multiword expressions and register. The code is en/src/ptag.rs, en/src/tag.rs and en/src/dict.rs. Measures: accuracy split into known and unseen words, the run-twice check, and the gain from the ambiguous update over the plain perceptron on the same data.

Topics: Part-of-speech tagging

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/n16-1130

@inproceedings{jorgensen2016learning,
  title = {{Learning a POS tagger for AAVE-like language}},
  author = {Jørgensen, Anna and Hovy, Dirk and Søgaard, Anders},
  booktitle = {Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies},
  year = {2016},
  doi = {10.18653/v1/n16-1130},
  url = {https://doi.org/10.18653/v1/n16-1130}
}

References

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

1. Hanan Aldarmaki and Mona Diab. 2015. Robust part-of-speech tagging of Arabic text. In Proceedings of the Second Workshop on Arabic Natural Language Processing, Beijing, China. find
2. Delphine Bernhard and Anne-Laure Ligozat. 2013. Hassle-free pos-tagging for the Alsatian dialects. In Marcos Zampieri and Sascha Diwersy, editors, Non-Standard Data Sources in Corpus Based-Research, pages 85–92. ZSM Studien. find
3. Rahma Boujelbane, Meriem Ellouze Khemekhem, and Lamia Hadrich Belguith. 2013. Mapping rules for building a Tunisian dialect lexicon and generating corpora. In International Joint Conference on Natural Language Processing, pages 419–428. find
4. Michael Collins. 2002. Discriminative Training Methods for Hidden Markov Models: Theory and Experiments with Perceptron Algorithms. In EMNLP. find our card
5. Dipanjan Das and Slav Petrov. 2011. Unsupervised part-of-speech tagging with bilingual graph-based projections. In ACL, pages 256–263. find
6. Kevin Duh and Katrin Kirchhoff. 2005. Pos tagging of dialectal Arabic: A minimally supervised approach. In Proceedings in the ACL Workshop on Computational Approaches to Semitic Languages. find
7. Dan Garrette and Jason Baldridge. 2013. Learning a part-of-speech tagger from two hours of annotation. In Proceedings of NAACL-HLT, pages 138–147. find
8. Kevin Gimpel, Nathan Schneider, Brendan OConnor, Dipanjan Das, Daniel Mills, Jacob Eisenstein, Michael Heilman, Dani Yogatama, Jeffrey Flanigan, and Noah A. Smith. 2011. Part-of-speech tagging for twitter: Annotation, features, and experiments. In ACL. find
9. Stephen Gouws and Anders Søgaard. 2015. Simple task-specific bilingual word embeddings. In Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 1386– 1390. find
10. Bo Han and Timothy Baldwin. 2011. Lexical Normalisation of Short Text Messages: Makn Sens a #twitter. In ACL. find
11. Dirk Hovy and Eduard Hovy. 2012. Exploiting partial annotations with em training. In Proceedings of the NAACL-HLT Workshop on the Induction of Linguistic Structure, pages 31–38. Association for Computational Linguistics. find
12. Dirk Hovy, Barbara Plank, Héctor Martı́nez Alonso, and Anders Søgaard. 2015a. Mining for unambiguous instances to adapt part-of-speech taggers to new domains. In Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 1256–1261. find
13. Dirk Hovy, Barbara Plank, Héctor Martı́nez Alonso, and Anders Søgaard. 2015b. Mining for unambiguous instances to adapt pos taggers to new domains. In NAACL-HLT. find
14. Anna Jørgensen, Dirk Hovy, and Anders Søgaard. 2015. Challenges of studying and processing dialects in social media. In Proceedings of the ACL Workshop on Noisy User-generated Text. find
15. Shen Li, João V. Graça, and Ben Taskar. 2012. Wikily supervised part-of-speech tagging. In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning. find
16. T. Mikolov, K. Chen, G Corrado, and J. Dean. 2013a. Efficient estimation of word representations in vector space. ArXiv e-prints. find
17. T. Mikolov, W.T. Yih, and G. Zweig. 2013b. Linguistic regularities in continuous space word representations. In Proceedings of NAACL-HLT. find
18. Olutobi Owoputi, Brendan O’Connor, Chris Dyer, Kevin Gimpel, Nathan Schneider, and Noah A. Smith. 2013. Improved part-of-speech tagging for online conversational text with word clusters. In Proceedings of NAACL-HLT. find
19. Slav Petrov, Dipanjan Das, and Ryan McDonald. 2011. A universal part-of-speech tagset. In Proceedings of LREC. find
20. Barbara Plank, Dirk Hovy, Ryan McDonald, and Anders Søgaard. 2014a. Adapting taggers to twitter with not-so-distant supervision. In Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1783– 1792. find
21. Barbara Plank, Dirk Hovy, and Anders Søgaard. 2014b. Learning part-of-speech taggers with inter-annotator agreement loss. In EACL. find
22. Anders Søgaard. 2013. Semi-supervised learning and domain adaptation for NLP. Morgan & Claypool. find
23. Ian Stewart. 2014. Now We Stronger Than Ever: African-American syntax on Twitter. In Proceedings of the Student Research Workshop to the 14th Conference of the European Chapter of the Association for Computational Linguistics, pages 26–30, Gothenburg, Sweden, April. find
24. Oscar Täckström, Dipanjan Das, Slav Petrov, Ryan Mc-Donald, and Joakim Nivre, 2013. Token and Type Constraints for Cross-Lingual Part-of-Speech Tagging, pages 1–12. Association for Computational Linguistics. find
25. Guillaume Wisniewski, Nicolas Pécheux, Sophir Gahbiche-Braham, and François Yvon. 2014. Cross-lingual part-of-speech tagging through ambiguous learning. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/jorgensen-2016-learning-pos-tagger-aave-like/