Elamur Research⌕
☰
home / papers / 2014 / Unsupervised Construction of a Lexicon and a Repository of Variation Patterns for Arabic Modal Multiword Expressions

Unsupervised Construction of a Lexicon and a Repository of Variation Patterns for Arabic Modal Multiword Expressions

Summary

The authors fill a gap in Arabic NLP: no lexicon or pattern resource existed for Arabic Modal Multiword Expressions (AM-MWEs). These multiword constructions convey seven senses of modality drawn from Palmer's typology, namely volition and ability, commitment, permission and obligation, plus evidentiality and (un)certainty. Because many of these expressions inflect, accept inserted modifiers, or allow their head word to be swapped, a plain list of strings is not enough for recognising them. The authors therefore built two linked resources with a four-stage unsupervised pipeline:

  1. Select corpus words that may derive from known modality roots, then extract frequent n-grams scored with Log-Likelihood Ratio.
  2. Separate modal from non-modal candidates with k-means.
  3. Abstract the surviving expressions into POS-based variation patterns.
  4. Reuse those patterns to recover rare and unseen expressions.

After manual checking, the XML lexicon holds 10,664 entries. Each entry points to a pattern in a repository encoded in three MADAMIRA tagsets. Both resources are public and can support work on uncertainty, politeness and power relations.

Key points

  • Five AM-MWE types are defined by how productive they are: fully idiomatic, morphologically productive (quasi-modals / phrasal verbs), lexically productive, syntactically productive, and productive on all three levels.
  • Candidates come from four corpora (Ajdir, LDC ISI, YADAC, Tashkeel). They were preprocessed, normalised, and tokenised and POS-tagged with MADAMIRA.
  • Candidate words were found by matching modality roots from the Arabic Modality Lexicon (Al-Sabbagh et al. 2013) with RegExps.
  • n-grams were limited to quadrigrams, with frequency thresholds of 20, 15 and 10 for bigrams, trigrams and quadrigrams. Pairs were scored with LLR.
  • Pattern-based bootstrapping added 358 AM-MWEs. In the final lexicon, epistemic (4,233) and volitive (3,330) are the largest senses. By size, there are 4,806 bigrams, 3,244 trigrams and 2,614 quadrigrams.

Interesting ideas

  • I1. Turn each multiword expression into a pattern: keep function words as literal tokens, drop inflectional features, mark the content word as HEAD and mark insertion slots as GAP. One pattern then covers many surface variants. — For Mova (tokens): English modal chunks like "be able to", "have to" or "there is a chance that" could be stored as HEAD/GAP patterns, so that "was hardly able to" is still lemmatised as one modal unit.
  • I2. Once frequent instances yield a pattern, run the pattern in reverse to find rare inflections and new heads that fit the same frame. Every harvested item is then checked by a person. — For Mova (learn): self-induced repair or construction rules could be generalised from frequent cases and applied to the long tail. A verification gate would keep the extra recall from damaging precision.
  • I3. An expression counts as a single unit if one word can replace it in context, for example "there is a possibility that" becoming "possibly". — For Mova (frames): this substitution test could decide when a periphrastic construction should become one modal or attitude operator on a frame instead of separate predicate nodes.
  • I4. Bootstrapping surfaced expressions like "it is clear that" and "it is normal that", which assume their clause is true rather than weigh its truth. The authors kept them but marked them apart from true epistemic modals. — For Mova (frames): speech-act and implicature handling should give factive or evaluative frames a presupposition flag, separate from uncertainty markers.
  • I5. Each lexicon entry carries a pattern ID linking it to the repository. Entries added by hand get the tag MANUAL in place of a corpus frequency. — For Mova (world): compiled knowledge entries could record provenance and the rule that generated them in the same way, so any answer can be traced back to its source.

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: Linguistics & languages

Topics: Semitic languages · Idioms & multiword expressions

Institutions

🎓University of Illinois Urbana-ChampaignUrbana · United States76 works · 63 460 citations

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/v1/w14-0820

@inproceedings{alsabbagh2014unsupervised,
  title = {{Unsupervised Construction of a Lexicon and a Repository of Variation Patterns for Arabic Modal Multiword Expressions}},
  author = {Al-Sabbagh, Rania and Girju, Roxana and Diesner, Jana},
  booktitle = {Proceedings of the 10th Workshop on Multiword Expressions (MWE)},
  year = {2014},
  doi = {10.3115/v1/w14-0820},
  url = {https://doi.org/10.3115/v1/w14-0820}
}

References

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

  1. Rania Al-Sabbagh and Roxana Girju. 2012. YADAC: Yet another Dialectal Arabic Corpus. Proc. of LREC'12, Istanbul, Turkey, May 23-25 2012 find
  2. Rania Al-Sabbagh, Jana Diesner and Roxana Girju. 2013. Using the Semantic-Syntactic Interface for Reliable Arabic Modality Annotation. Proc. of IJCNLP'13, Nagoya, Japan, October 14-18 2013 find
  3. Mohammed Attia, Antonio Toral, Lamia Tounsi, Pavel Pecina and Josef van Genbith. 2010. Automatic Extraction of Arabic Multiword Expressions. Proc. of the Workshop on MWE 2010, Beijing, August 2010 find
  4. Elsaid Badawi, M.G. Carter and Adrian Gully. 2004. Modern Written Arabic: A Comprehensive Grammar. UK: MPG Books Ltd Satanjeev Banerjee and Ted Pedersen. 2003. The Design, Implementation, and Use of the Ngram Statistic Package. Proc. of CiCling'03, Mexico City, USA Ibrahim Bounhas and Yahya Slimani. 2009. A Hybrid Approach for Arabic Multi-Word Term Extraction. Proceedings of NLP-KE 2009, Dalian, China, September 24-27 2009 find
  5. Kristen E. Brustad. 2000. The Syntax of Spoken Arabic: A Comparative Study of Moroccan, Egyptian, Syrian and Kuwaiti Dialects. Georgetown Uni. find
  6. Press, Washington DC, USA Tim Buckwalter. 2002. Arabic Morphological Analyzer. find
  7. Technical Report, Linguistic Data Consortium, Philadelphia Cristian Danescu-Niculescu-Mizil, Moritz Sudhof, Dan Jurafsky, Jure Leskovec and Christopher Potts. 2013. A Computational Approach to Politeness with Application to Social Factors. Proc. of the 51 st ACL, , Sofia, Bulgaria, August 4-9 2013 find our card
  8. Mona Diab, Lori Levin, Teruko Mitamura, Owen Rambow, Vinodkumar Prabhakaran, and Weiwei Guo. 2009. Committed Belief Annotation and Tagging. Proc. of the 3rd LAW Workshop, ACL-IJCNLP'09, pp. 68-73, Singapore Carla Parra Escartín, Gyri Smørdal Losnegaard, Gunn Inger Lyse Samdal and Pedro Patiño García. 2013. Representing Multiword Expressions in Lexical and Terminological Resources: An Analysis for Natural Language Processing Purposes. Proc. of eLex 2013, pages 338-357, Tallinn, Estonia, October 17-19 2013 find
  9. Abdelati Hawwari, Kfir Bar and Mona Diab. 2012. Building an Arabic Multiword Expressions Repository. Proc. of the 50 th ACL, pages 24-29, Jeju, Republic of Korea, July 12 2012 find
  10. Andrew Lampert, Robert Dale and Cecile Paris. 2010, Detecting Emails Containing Requests for Action. Proc. of the 2010 ACL, pages 984-992, Los Angeles, California, June 2010 find
  11. F. Mitchell and S. A. Al-Hassan. 1994. Modality, Mood and Aspect in Spoken Arabic with Special Reference to Egypt and the Levant. London and NY: Kegan Paul International Ola Moshref. 2012. Corpus Study of Tense, Aspect, and Modality in Diglossic Speech in Cairene Arabic. PhD Thesis. University of Illinois at Urbana-Champaign Dragos Stefan Munteanu and Daniel Marcu. 2007. find
  12. ISI Arabic-English Automatically Extracted Parallel Text, Linguistic Data Consortium, Philadelphia Malvin Nissim and Andrea Zaninello. 2013. A Repository of Variation Patterns for Multiword Expressions. Proc. of the 9 th Workshop of MWE, pp. 101-105, Atlanta, Georgia, June 13-14 2013 find
  13. Frank R. Palmer. 2001. Mood and Modality. 2 nd Edition. find
  14. Cambridge University Press, Cambridge, UK Silvia Pareti, Tim O'Keefe, Ioannis Konstas, James R. Curran and Irena Koprinska. 2013. Automatically Detecting and Attributing Indirect Quotations. Proc. of the 2013 EMNLP, pages. 989-1000, Washington, USA, October 18-21 2013 find
  15. Arfath Pasha, Mohamed Al-Badrashiny, Ahmed El Kholy, Ramy Eskander, Mona Diab, Nizar Habash, Manoj Pooleery, Owen Rambow and Ryan Roth. 2014. MADAMIRA: A Fast, Comprehensive Tool for Morphological Analysis and Disambiguation of Arabic. Proc. of the 9th International Conference on Language Resources and Evaluation, Reykjavik, Iceland, May 26-31 2014 find our card
  16. Vinodkumar Prabhakaran and Owen Rambow. 2013. Written Dialog and Social Power: Manifestations of Different Types of Power in Dialog Behavior. Proceedings of the 6 th IJCNLP, pp. 216-224, Nagoya, Japan, October 14-18 2013 find
  17. Ivan Sag, Timothy Baldwin, Francis Bond, Ann Copestake and Dan Flickinger. 2002. Multiword Expressions: A Pain in the Neck for NLP. Proceedings of CiCling 2002, pages 1-15, Mexico City, Mexico Michael Stubbs. 2007. An Example of Frequent English Phraseology: Distributions, Structures and Functions. Language and Computers: Corpus Linguistics 25 Years on, pages 89-105, (17) find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sabbagh-2014-unsupervised-construction-lexicon-repository-variation/