Elamur Research⌕
☰
home / papers / 2016 / Multilingual Projection for Parsing Truly Low-Resource Languages

Multilingual Projection for Parsing Truly Low-Resource Languages

Summary

Most cross-lingual methods for POS tagging and dependency parsing were built and tested on Germanic and Romance languages. These languages have abundant parallel data, easy tokenization and similar word order, so such work may overstate what is achievable for truly low-resource languages. Agić et al. assume nothing about the target language except a translated text. They use the Bible (EBC) and Watchtower publications (WTC), which are massively multi-parallel, together with taggers and parsers for resource-rich source languages. POS tags are transferred by alignment-weighted voting across many source languages. For syntax, they propose a heuristics-free projection. Rather than projecting a single best tree from one source, they carry each source parser's standardized scores for every candidate edge through the word alignments and sum them over all sources. Only the combined target matrix is then decoded into a tree. They built models for more than 100 languages and evaluated them on 30 languages with test data. The method outperforms delexicalized transfer and standard projection heuristics, with particularly strong results on non-Indo-European languages. This makes basic syntactic tools feasible for languages without any annotated data.

Key points

  • Parallel sources: Bible excerpts exist in 1,646 languages and Watchtower texts in up to 583. No tag dictionaries, labeled data or typological information are used for target languages.
  • The dependency projection transfers dense edge-weight matrices from multiple sources and delays tree decoding (maximum spanning tree) until after aggregation. Weights are normalized per sentence.
  • Taggers and parsers were trained for over 100 languages and evaluated on 30 test languages. Training was capped at 20k sentences per language, although EBC/WTC offer up to 120k, because adding more data gave little benefit.
  • Parsers rely on projected (noisy) POS tags, yet about 40% of tokens with wrong POS predictions still receive the correct syntactic head.
  • Code and data are released at bitbucket.org/lowlands/release. The work was funded by the ERC Starting Grant LOWLANDS.

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

Topics: Part-of-speech tagging

Cited by

79 papers of our library cite it (of the 140884 whose bibliographies we read); 8 of them have a page here, 71 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1162/tacl_a_00100

@article{agic2016multilingual,
  title = {{Multilingual Projection for Parsing Truly Low-Resource Languages}},
  author = {Agić, Željko and Johannsen, Anders and Plank, Barbara and Alonso, Héctor Martínez and Schluter, Natalie and Søgaard, Anders},
  journal = {Transactions of the Association for Computational Linguistics},
  year = {2016},
  doi = {10.1162/tacl\_a\_00100},
  url = {https://doi.org/10.1162/tacl_a_00100}
}

References

17 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Željko Agić, Dirk Hovy, and Anders Søgaard. 2015. If All You Have is a Bit of the Bible: Learning POS Taggers for Truly Low-Resource Languages. In ACL. find
  2. TnT: A Statistical Part-of-Speech Tagger. In ANLP. Christos Christodouloupoulos and Mark Steedman. 2014. A Massively Parallel Corpus: The Bible in 100 Languages. Language Resources and Evaluation, 49(2). find our card
  3. Dipanjan Das and Slav Petrov. 2011. Unsupervised Part-of-Speech Tagging with Bilingual Graph-Based Projections. In ACL. find our card
  4. Chris Dyer, Victor Chahuneau, and Noah A. Smith. 2013. A Simple, Fast, and Effective Reparameterization of IBM Model 2. In ACL. find
  5. Dan Garrette, Jason Mielens, and Jason Baldridge. 2013. Real-World Semi-Supervised Learning of POS-Taggers for Low-Resource Languages. In ACL. find
  6. Rebecca Hwa, Philip Resnik, Amy Weinberg, Clara Cabezas, and Okan Kolak. 2005. Bootstrapping Parsers via Syntactic Projection Across Parallel Texts. Natural Language Engineering, 11(3). Anders Johannsen, Željko Agić, and Anders Søgaard. 2016. Joint Part-of-Speech and Dependency Projection from Mul… find
  7. Wiki-ly Supervised Part-of-Speech Tagging. In EMNLP. Zhenghua Li, Min Zhang, and Wenliang Chen. 2014. Soft Cross-lingual Syntax Projection for Dependency Parsing. In COLING. find
  8. Xuezhe Ma and Fei Xia. 2014. Unsupervised Dependency Parsing with Transferring Distribution via Parallel Guidance and Entropy Regularization. In ACL. André F. T. Martins, Miguel Almeida, and Noah A. Smith. 2013. Turning on the Turbo: Fast Third-Order Non-Projective Turbo Parsers. In ACL. find
  9. Ryan McDonald, Koby Crammer, and Fernando Pereira. 2005. Online Large-Margin Training of Dependency Parsers. In ACL. find our card
  10. Ryan McDonald, Slav Petrov, and Keith Hall. 2011. Multi-Source Transfer of Delexicalized Dependency Parsers. In EMNLP. Ryan McDonald, Joakim Nivre, Yvonne Quirmbach-Brundage, Yoav Goldberg, Dipanjan Das, Kuzman Ganchev, Keith Hall, Slav Petrov, Hao Zhang, Oscar Täckström, Claudia Bedini, Núria Be… find our card
  11. Word Order Typology Through Multilingual Word Alignment. In ACL. Slav Petrov, Dipanjan Das, and Ryan McDonald. 2012. find
  12. A Universal Part-of-Speech Tagset. In LREC. Mohammad Sadegh Rasooli and Michael Collins. 2015. Density-Driven Cross-Lingual Transfer of Dependency Parsers. In EMNLP. Rudolf Rosa and Zdeněk Žabokrtský. 2015. KLcpos3: A Language Similarity Measure for Delexicalized Parser Transfer. In ACL. find
  13. Kenji Sagae and Alon Lavie. 2006. Parser Combination by Reparsing. In NAACL. find our card
  14. Kathrin Spreyer, Lilja Øvrelid, and Jonas Kuhn. 2010. Training Parsers on Partial Trees: A Cross-Language Comparison. In LREC. find
  15. Anders Søgaard. 2011. Data Point Selection for Cross-Language Adaptation of Dependency Parsers. In ACL. Jörg Tiedemann and Željko Agić. 2016. Synthetic Treebanking for Cross-Lingual Dependency Parsing. Journal of Artificial Intelligence Research, 55. Jörg Tiedemann, Željko Agić, and Joakim Niv… find
  16. David Yarowsky, Grace Ngai, and Richard Wicentowski. 2001. Inducing Multilingual Text Analysis Tools via Robust Projection Across Aligned Corpora. In NAACL. find our card
  17. Daniel Zeman and Philip Resnik. 2008. Cross-Language Parser Adaptation Between Related Languages. In IJCNLP Workshop on NLP for Less Privileged Languages. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/agic-2016-multilingual-projection-parsing-truly-low/