Elamur Research
☰
home / papers / 2014 / Semi-supervised learning of morphological paradigms and lexicons

Semi-supervised learning of morphological paradigms and lexicons

What it is about

The paper turns inflection tables into abstract paradigms, by replacing every common substring of a table with a variable. The result stays readable, so a linguist can inspect and correct it. Paradigms are then used to inflect unseen base forms, and to collect lemmas from unannotated text. The evaluation uses Wiktionary tables for German, Spanish and Finnish: without corpus information, per-form accuracy on unseen base forms runs from 87.81% (German nouns) to 99.52% (Spanish verbs). With a Wikipedia-based frequency tie-break it runs from 91.81% to 99.58%. In a Swedish lexicon task, the first 100 collected words are 100.0% correct and the first 1000 are 96.4% correct. On the Durrett and DeNero (2013) tables, the number of output paradigms is below 10% of the number of input tables in every case.

Ideas

For Mova

The code is en/src/morph.rs and en/src/lexgen.rs, with en lexicon and en morph-eval for checks. The chapters are languages/en/11-nouns.md and 14-irregular-verbs.md, where the paradigm view fits best. The paper's tables are Wiktionary-derived and its Swedish list is not English, so use it for ideas and tests only. Measure first: per-form accuracy on held-out English lemmas, two runs identical, before any claim.

Library summary

The Research library's own summary of the paper, in our words — not the paper's text.

Written by Mova (claude-opus-5-5) from the paper text, in our own words: the license "CC BY-NC-SA 3.0 (ACL Anthology, materials before 2016)" does not allow redistributing the text itself. Original — aclanthology.org/E14-1060.pdf

Summary

The authors take on a labor-intensive task: building a large morphological lexicon. They propose a semi-supervised, language-independent approach. The system takes a few ready-made inflection tables as input and generalizes them into abstract paradigms. An abstract paradigm is an ordered set of strings. In it, the fragments shared by all forms are replaced with variables x1, x2, etc., while the invariant parts encode affixes and alternations. The shared material is defined as the longest common subsequence (LCS) of all forms in the table. It is computed with finite-state automata in foma, and identical paradigms are then merged. An unannotated corpus helps select the paradigm for new words more accurately. The method was evaluated in two ways. The first is reconstructing inflection tables from Wiktionary for German, Spanish and Finnish. The second is a simulation of building a Swedish lexicon based on SALDO. The results match the best ones known at the time and often exceed them. At the same time, the model remains transparent: a linguist can read every paradigm and correct it if needed. The method is therefore suitable as a practical tool for rapidly expanding lexical resources.

Key points

  • Without an additional corpus, per-word-form accuracy for unseen base forms ranges from 87.81 % (German nouns) to 99.52 % (Spanish verbs). When unannotated texts are added, it rises to 91.81–99.58 %.
  • The same Wiktionary dataset as in Durrett and DeNero (2013) was used. In full-table accuracy, the new system outperforms their result in four of the five tests: from 76.50 % (German verbs) to 98.00 % (Spanish verbs).
  • In the Swedish experiment, the 20 most frequent noun paradigms out of 346 were taken, with one table for each. Based on them, all 79.6 thousand SALDO nouns were ranked. Among the first 100 words, 100 % received the correct paradigm; among the first 1000, 96.4 %.
  • The paradigm p_akademi performs worst. It is most often confused with the defective p_akribi, which has no plural (334 cases out of 1000). The second source of errors is the unseen paradigm parti, which differs in gender (210 cases).
  • Finding the LCS of many strings (MLCS) is an NP-hard problem, but for real inflection tables it can be solved with heuristics. Språkbanken has made the code, data and evaluation procedure openly available.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.3115/v1/e14-1060

@inproceedings{hulden2014semi,
  title = {{Semi-supervised learning of morphological paradigms and lexicons}},
  author = {Hulden, Mans and Forsberg, Markus and Ahlberg, Malin},
  booktitle = {EACL 2014 (Gothenburg)},
  year = {2014},
  doi = {10.3115/v1/e14-1060},
  url = {https://doi.org/10.3115/v1/e14-1060}
}

References

27 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Lars Borin, Markus Forsberg, and Lennart Lönngren. 2013. SALDO: a touch of yin to WordNet’s yang. Language Resources and Evaluation, May. Online first publication; DOI 10.1007/s10579-013-9233-4. — [doi:10.1007/s10579-013-9233-4](https://doi.org/10.1007/s10579-013-9233-4) find
  2. Erwin Chan. 2006. Learning probabilistic paradigms for morphology in a latent class model. In Proceedings of the Eighth Meeting of the ACL Special Interest Group on Computational Phonology and Morphology, pages 69–78. Association for Computational Linguistics. find
  3. Lionel Clément, Bernard Lang, Benoı̂t Sagot, et al. 2004. Morphology based automatic acquisition of large-coverage lexica. In LREC 04, pages 1841– 1844. find
  4. Mathias Creutz and Krista Lagus. 2007. Unsupervised models for morpheme segmentation and morphology learning. ACM Transactions on Speech and Language Processing (TSLP), 4(1):3. find
  5. Grégoire Détrez and Aarne Ranta. 2012. Smart paradigms and the predictability and complexity of inflectional morphology. In Proceedings of the 13th EACL, pages 645–653. Association for Computational Linguistics. find
  6. Markus Dreyer and Jason Eisner. 2011. Discovering morphological paradigms from plain text using a Dirichlet process mixture model. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 616–627. Association for Computational Linguistics. find
  7. Greg Durrett and John DeNero. 2013. Supervised learning of complete morphological paradigms. In Proceedings of NAACL-HLT, pages 1185–1195. find our card
  8. Ramy Eskander, Nizar Habash, and Owen Rambow. 2013. Automatic extraction of morphological lexicons from morphologically annotated corpora. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1032–1043. Association for Computational Linguistics. find
  9. Markus Forsberg, Harald Hammarström, and Aarne Ranta. 2006. Morphological lexicon extraction from raw text data. In Advances in Natural Language Processing, pages 488–499. Springer. find
  10. John Goldsmith. 2001. Unsupervised learning of the morphology of a natural language. Computational linguistics, 27(2):153–198. find
  11. Harald Hammarström and Lars Borin. 2011. Unsupervised learning of morphology. Computational Linguistics, 37(2):309–350. find
  12. Charles F Hockett. 1954. Two models of grammatical description. Morphology: Critical Concepts in Linguistics, 1:110–138. find
  13. Mans Hulden. 2009. Foma: a finite-state compiler and library. In Proceedings of the 12th Conference of the European Chapter of the European Chapter of the Association for Computational Linguistics: Demonstrations Session, pages 29–32, Athens, Greece. Association for Computational Linguistics. find
  14. Kimmo Koskenniemi. 1991. A discovery procedure for two-level phonology. Computational Lexicology and Lexicography: A Special Issue Dedicated to Bernard Quemada, 1:451–46. find
  15. Kimmo Koskenniemi. 2013. An informal discovery procedure for two-level rules. Journal of Language Modelling, 1(1):155–188. find
  16. Krister Lindén. 2008. A probabilistic model for guessing base forms of new words by analogy. In Computational Linguistics and Intelligent Text Processing, pages 106–116. Springer. find
  17. David Maier. 1978. The complexity of some problems on subsequences and supersequences. Journal of the ACM (JACM), 25(2):322–336. find
  18. Peter H. Matthews. 1972. Inflectional morphology: A theoretical study based on aspects of Latin verb conjugation. Cambridge University Press. find
  19. Christian Monson, Jaime Carbonell, Alon Lavie, and Lori Levin. 2008. Paramor: finding paradigms across morphology. In Advances in Multilingual and Multimodal Information Retrieval, pages 900– 907. Springer. find
  20. Sylvain Neuvel and Sean A Fulop. 2002. Unsupervised learning of morphology without morphemes. In Proceedings of the ACL-02 workshop on Morphological and phonological learning-Volume 6, pages 31–40. Association for Computational Linguistics. find
  21. Robert H Robins. 1959. In defence of WP. Transactions of the Philological Society, 58(1):116–144. find
  22. Patrick Schone and Daniel Jurafsky. 2001. Knowledge-free induction of inflectional morphologies. In Proceedings of the second meeting of the North American Chapter of the Association for Computational Linguistics on Language technologies, pages 1–9. Association for Computational Linguistics. find our card
  23. Gregory T. Stump. 2001. A theory of paradigm structure. Cambridge University Press. find
  24. Tzvetan Tchoukalov, Christian Monson, and Brian Roark. 2010. Morphological analysis by multiple sequence alignment. In Multilingual Information Access Evaluation I. Text Retrieval Experiments, pages 666–673. Springer. find
  25. Pieter Theron and Ian Cloete. 1997. Automatic acquisition of two-level morphological rules. In Proceedings of the fifth conference on Applied natural language processing, pages 103–110. Association for Computational Linguistics. find
  26. Antal van den Bosch and Walter Daelemans. 1999. Memory-based morphological analysis. In Proceedings of the 37th Annual Meeting of the Association for Computational Linguistics, pages 285–292. Association for Computational Linguistics. find
  27. David Yarowsky and Richard Wicentowski. 2000. Minimally supervised morphological analysis by multimodal alignment. In Proceedings of the 38th Annual Meeting on Association for Computational Linguistics, pages 207–216. Association for Computational Linguistics. find