Elamur Research⌕
☰
home / papers / 2008 / Part-of-Speech Tagging for English-Spanish Code-Switched Text

Part-of-Speech Tagging for English-Spanish Code-Switched Text

Summary

The paper addresses POS tagging of conversational text where speakers alternate between English and Spanish ("Spanglish"), a setting with no dedicated tagger and almost no annotated data. The authors recorded about 39 minutes of conversation among three bilingual university staff members and transcribed it into 922 sentences (~8k words). Each token was manually tagged with a merged inventory built from the English and Spanish Tree Tagger tag sets. Instead of training a tagger from nothing, they reuse the two monolingual Tree Taggers and compare three ways of combining them: simple selection heuristics, choosing a tagger according to word-level language identification, and a machine-learning (SVM) classifier whose features come from both taggers' outputs. Language identification beat every heuristic. The learned combination did best and even surpassed an oracle that always picks the better of the two monolingual tags. The SVM was also notably more robust on sentences containing intrasentential switches. This matters because it shows that strong monolingual tools can be adapted to mixed-language text with only a small in-domain corpus.

Key points

  • Dataset: a spoken Spanglish corpus with these properties:
    • 922 sentences, about 8k tokens, 1,516 distinct word forms;
    • English-dominant: 6,020 English tokens versus roughly 2k Spanish tokens;
    • 239 manually identified switches, of which 129 are intrasentential.
  • Base taggers: the English and Spanish versions of the Tree Tagger. The English model was trained on the Penn Treebank and reports 96.36% accuracy. The Spanish model uses 75 tags and was trained on the CRATER corpus.
  • Oracle errors: the oracle still made 445 errors, and about half of them (233) fell in sentences containing code-switches.
  • Language-identification errors: the dictionary-based method made 787 errors, nearly 70% of them in code-switched sentences. A large share stemmed from words that exist in both English and Spanish, for example "con", "a", "no" and "me".
  • SVM and CRF comparison:
    • The SVM fixed 223 of the oracle's errors and 481 of the language-identification method's errors, but introduced 105 errors of its own.
    • A CRF trained from scratch in MALLET reached only 81% average accuracy, which is below the language-identification heuristic.

Interesting ideas

  • I1. Use monolingual taggers as feature generators. Their tags, tag probabilities, foreign-word tags and "unknown" lemma markers can be fed to a classifier rather than trusting either tagger directly. This suits any low-resource mixed-language task, such as Ukrainian-Russian surzhyk.
  • I2. Benchmark against a pick-the-best oracle. An oracle that always chooses the correct monolingual output shows how much combining can gain. Beating it reveals that a learner can fix errors present in both base systems.
  • I3. Treat lemmatization failure as a language signal. When a tagger cannot lemmatize a word, that is cheap evidence the word comes from the other language. This is useful for token-level language identification without extra resources.
  • I4. Break errors down by sentence type. Separating errors in switched versus monolingual sentences exposes exactly where a method struggles. It is a simple diagnostic that transfers to evaluating any code-mixed NLP system.
  • I5. Compare reuse against a from-scratch baseline. Training a CRF on the small corpus confirms that reusing existing resources beats learning from scratch when in-domain data is scarce. This check is worth repeating whenever a new tiny corpus is built.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Natural language processing

TopicsPart-of-speech tagging

Institutions

🎓The University of Texas at DallasRichardson · United States34 works · 3,985 citations

Cite

Cite as Thamar Solorio and Yang Liu. 2008. Part-of-Speech Tagging for English-Spanish Code-Switched Text. Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing.

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

@inproceedings{solorio2008part,
  title = {{Part-of-Speech Tagging for English-Spanish Code-Switched Text}},
  author = {Thamar Solorio and Yang Liu},
  booktitle = {Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing},
  year = {2008},
  doi = {10.3115/1613715.1613852},
  url = {https://doi.org/10.3115/1613715.1613852}
}

References

39 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ardila, A. (2005) Spanglish: an anglicized dialect. Hispanic Journal of Behavioral Sciences, 27(1): 60–81. find
  2. Brants, T. (2000) TnT - a statistical part-of-speech tagger. In Sixth Applied Natural Language Processing Conference ANLP-2000 Seatle, WA. our card
  3. Brill, E. (1990) A simple rule-based part-of-speech tagger. In 3rd Conference on Applied Natural Language Processing, pp. 152–155. Trento, Italy. our card
  4. Carreras, X. and Padró, L. (2002) A flexible distributed architecture for natural language analyzers. In Third International Conference on Language Resources and Evaluation, LREC-02, pp. 1813–1817. Las Palmas de Gran Canaria, Spain. find
  5. Charniak, E. (1993) Equations for part-of-speech tagging. In 11th National Conference on AI, pp. 784–789. find
  6. Ervin, S. and Osgood, C. (1954) Second language learning and bilingualism. Journal of abnormal and social phsychology, Supplement 49, pp. 139–146. find
  7. Espinosa, A. (1917) Speech mixture in New Mexico: the influence of English language on New Mexican Spanish. In H. Stevens and H. Bolton, (eds.), The Pacific Ocean in History, pp. 408–428. find
  8. Franco, J.C. and Solorio T. (2007) Baby-steps towards building a Spanglish language model. In 8th International Conference on Computational Linguistics and Intelligent Text Processing, CICLing-2007, pp. 75–84. Mexico City, Mexico. find
  9. Friedman, J., Hastie, T., and Tibshirani, R. (1998) Additive logistic regression: a statistical view of boosting. Technical Report, Stanford University. find
  10. Godfrey, J., Holliman, E. and McDaniel, J. (1992) Switchboard: Telephone speech corpus for research development. In ICASSP, pp. 517–520, San Francisco, CA, USA. find
  11. Goyal, P., Mital, Manav R., Mukerjee, A., Raina, Achla M., Sharma, D., Shukla, P. and Vikram, K. (2003) A bilingual parser for Hindi, English and code-switching structures. In Computational Linguistics for South Asian Languages –Expanding Synergies with Europe, EACL-2003 Workshop, Budapest, Hungary. find
  12. Grosjean, F. (1982) Life with Two Languages: An Introduction to Bilingualism. Harvard University Press. find
  13. Gumperz, J. J. and Hernandez-Chavez, E. (1971) Cognitive aspects of bilingual communication. Oxford University Press, London. find
  14. Gumperz, J. J. (1964) Linguistic and social interaction in two communities. In John J. Gumperz (ed.), Language in social groups, pp. 151–176, Stanford. Stanford University Press. find
  15. Gumperz, J. J. (1971) Bilingualism, bidialectism and classroom interaction. In Language in social groups, pp. 311–339, Stanford. Stanford University Press. find
  16. Joshi, A. K. (1982) Processing of sentences with intrasentential code-switching. In Ján Horecký (ed.), COLING-82, pp. 145–150, Prague. find
  17. Kucera, H. and Francis, W. N. (1967) Computational analysis of present-day American English. Brown University Press. find
  18. Lafferty, J. and McCallum, A. and Pereira F. (2001) Conditional random fields: Probabilistic models for segmenting and labeling sequence data. In 18th ICML, pp. 282–289. MA, USA. find
  19. Lipski, J. M. (1978) Code-switching and the problem of bilingual competence. In M. Paradis (ed.), Aspects of bilingualism, pp. 250–264, Columbia, SC. Hornbeam. find
  20. McCallum, A. (2002) MALLET: A Machine Learning for Language Toolkit. Retrieved January 7, 2008 from http://mallet.cs.umass.edu ↗
  21. Paul, D. B. and Baker, J. M. (1992) The design of the Wall Street Journal-based CSR corpus. In HLT’91: workshop on speech and Natural Language pp. 357– 362, Morristown, NJ, USA. find
  22. Poplack, S., Sankoff, D. and Miller, C. (1988) The social correlates and linguistic processes of lexical borrowing and assimilation. Linguistics, 26(1): 47–104. find
  23. Poplack, S. (1980) Sometimes I’ll start a sentence in Spanish y termino en español: toward a typology of code-switching. Linguistics, 18(7/8): 581–618. find
  24. Poplack, S. (1981) Syntactic structure and social function of code-switching. In R. Duran (ed.), Latino discourse and communicative behavior, pp. 169–184, Norwood, NJ. Ablex. find
  25. Quinlan, J. R. (1986) Induction of decision trees. Machine Learning, 1: 81–106. find
  26. Ratnaparkhi, A. (1996) A maximum entropy model for part-of-speech tagging. In EMNLP, pp. 133–142, Philadelphia, PA, May. our card
  27. Rosner, M. and Farrugia, P. (2007) A tagging algorithm for mixed language identification in a noisy domain. find
  28. In INTERSPEECH 2007, pp. 190–193, Antwerp, Belgium. find
  29. Sankoff, D. (1968) Social aspects of multilingualism in New Guinea. Ph.D. thesis, McGill University. find
  30. Sankoff, D. (1981) A formal grammar for codeswitching. Papers in Linguistics: International Journal of Human communications, 14(1): 3–46. find
  31. Sankoff, D. (1998a) A formal production-based explanation of the facts of code-switching. Bilingualism, Language and Cognition, 1: 39–50. Cambridge University Press. find
  32. Sankoff, D. (1998b) The production of code-mixed discourse. In 36th ACL, volume I, pp. 8–21, Montreal, Quebec, Canada. find
  33. Schmid, H. (1994) Probabilistic part-of-speech tagging using decision trees. In International Conference on New Methods in Language Processing, Manchester, UK. find
  34. Schölkopf, B. and Smola, A. J. (2002) Learning with Kernels: Support Vector Machines, Regularization, Optimization and Beyond. MIT Press. find
  35. Toribio, A. J. (2001a) Accessing Spanish-English code-switching competence. International Journal of Bilingualism, 5(4):403–436. find
  36. Toribio, A. J. (2001b) On the emergence of bilingual code-switching competence. Bilingual Language and Cognition, 4(3):203–231. find
  37. U.S. Census Bureau. (2003) Language use and English speaking ability: 2000. Retrieved October 30, 2006 from http://www.census.gov/prod/2003pubs/c2kbr-29.pdf. ↗
  38. Witten, I. H. and Frank, E. (1999) Data Mining, Practical Machine Learning Tools and Techniques with Java Implementations. Morgan Kaufmann. find
  39. Wolpert, D. H. (1992) Stacked Generalization. Neural Networks, 5(2):241–259. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/solorio-2008-part-speech-tagging-english-spanish/