Elamur Research⌕
☰
home / papers / 2010 / Automatic Extraction of Arabic Multiword Expressions

Automatic Extraction of Arabic Multiword Expressions

Paper · 2010 · Mohammed Attia, Antonio Toral, Lamia Tounsi et al.

Summary

The paper tackles the shortage of lexicons of Arabic multiword expressions (MWEs), meaning phrases whose meaning or behaviour cannot be fully derived from their parts. The authors focus on nominal MWEs and combine three complementary harvesting strategies, each suited to a different kind of data. First, they mine Arabic Wikipedia titles. A multiword Arabic title counts as a candidate MWE when its interwiki equivalent in any of 21 other languages is a single word, or when the title or its translations appear in WordNets or the named-entity lexicon MINELex. Second, they take English MWEs from Princeton WordNet 3.0, translate them into Arabic with Google Translate, and keep translations that search engines confirm. Third, they apply association measures (PMI and Chi-square) to bigrams and trigrams in the Arabic Gigaword corpus, after lemmatizing with MADA and filtering by POS. Together the methods yield about 33,000 MWEs, or roughly 72,000 when named entities are counted. The sources overlap little, so they enrich one another. The work matters because it shows how to build MWE resources for a less-resourced language from existing multilingual assets.

Key points

  • The authors analysed PWN 3.0: 64,331 of its 155,287 unique strings (about 41%) are MWEs, and the share is highest among nouns (about 51%). This motivates the focus on nominal MWEs. By comparison, Arabic WordNet has only 11,269 synsets, of which 2,348 are MWEs.
  • The crosslingual method widens the Zarrieß and Kuhn idea of many-to-one correspondence from one target language to 21. Its evaluation used a set of 1,100 Arabic Wikipedia multiword titles.
  • The corpus method runs on Arabic Gigaword 4th edition (LDC2009T30, 848 million words). It keeps n-grams seen at least 50 times, lemmatizes clitics with MADA, and then filters by POS. A human annotator judged 3,600 sampled candidates.
  • No single association measure won everywhere. For bigrams, PMI did best at the top 10,000 candidates; for trigrams, χ² did best at the top 5,000.
  • Yields by method:
    • crosslingual: 7,792 MWEs
    • translation-based: 13,656 MWEs (2,658 also found by other methods)
    • corpus-based: 15,000 MWEs (only 697 shared with other methods)
    • named-entity MWEs: 38,712
    • union: 33,093 without NEs, 71,805 with NEs

Interesting ideas

  • I1. If a phrase in one language corresponds to a single word in some other language, that asymmetry is cheap evidence the phrase is a lexical unit rather than a free combination. Polling many languages instead of one makes the signal more robust.
  • I2. The paper combines heterogeneous, weakly overlapping sources (encyclopedia titles, translated lexicon entries, corpus statistics) and takes their union. Each source catches a different MWE type, such as idioms, named entities or habitual collocations.
  • I3. Candidates produced by machine translation are validated by checking that they actually occur on the web, which filters out literal mistranslations without manual review.
  • I4. The authors collapse inflected and clitic variants before counting co-occurrences, since raw surface forms scatter the statistics. They also note that ignoring inflection still left matches unmerged.
  • I5. The authors explain the low overlap between corpus collocations and curated lexicons by the gap between statistical salience and what lexicographers consider worth listing. This suggests using one source as a negative control for the other.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tokens): Wikidata/Wikipedia interlanguage labels could flag English multiword spans (e.g. compounds) as single lexical units, so the tagger and lemmatizer treat them as one token-level item.
  • I2 (world): the compiled lexicon of multiword concepts could merge WordNet, Wikidata and corpus collocations, recording which source vouches for each entry so that every concept stays explainable.
  • I3 (back): when generating or translating through the meta-world, a chosen multiword rendering could be kept only if a corpus or search check attests it, rejecting word-for-word calques.
  • I4 (learn): collocation or repair-rule induction on silver data should count over lemmas and UD features, not surface strings, so that rare variants pool their evidence.
  • I5 (learn): evaluation of induced multiword units could report agreement with WordNet separately from corpus-only finds, making it visible whether a method learns lexical units or mere frequent pairs.

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

 Institutions

🎓Dublin City UniversityDublin · Ireland70 works · 9,447 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksIdioms & multiword expressions · 217 works

Cite

Cite as Mohammed Attia, Antonio Toral, Lamia Tounsi, Pavel Pecina and Josef van Genabith. 2010. Automatic Extraction of Arabic Multiword Expressions. Proceedings of the 2010 Workshop on Multiword Expressions: from Theory to Applications.

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

@inproceedings{attia2010automatic,
  title = {{Automatic Extraction of Arabic Multiword Expressions}},
  author = {Mohammed Attia, Antonio Toral, Lamia Tounsi, Pavel Pecina and Josef {van Genabith}},
  booktitle = {Proceedings of the 2010 Workshop on Multiword Expressions: from Theory to Applications},
  year = {2010},
  url = {https://aclanthology.org/W10-3704/}
}

References

21 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Attia, Mohammed. 2006. Accommodating Multiword Expressions in an Arabic LFG Grammar. In Salakoski, Tapio, Filip Ginter, Sampo Pyysalo, Tapio Pahikkala (Eds.): Advances in Natural Language Processing. Vol. 4139, pp. 87–98. Springer-Verlag: Berlin, Heidelberg. find
  2. Attia, Mohammed, Antonio Toral, Lamia Tounsi, Monica Monachini and Josef van Genabith. 2010. An automatically built Named Entity lexicon for Arabic. In the 7 th International Conference on Language Resources and Evaluation (LREC 2010), pp. 3614–3621. Valletta, Malta.
  3. Baldwin, Timothy, Colin Bannard, Takaaki Tanaka and Dominic Widdows. 2003. An Empirical Model of Multiword Expressions Decomposability. In Workshop on Multiword Expressions, the 41 st Annual Meeting of the Association for Computational Linguistics (ACL 2003), pp. 89–96, Sapporo, Japan. find
  4. Boulaknadel, Siham, Beatrice Daille, Driss Aboutajdine. 2009. A multi-word term extraction program for Arabic language In the 6 th International Conference on Language Resources and Evaluation (LREC 2008), pp. 630–634, Marrakech, Morocco.
  5. Deksne, Daiga, Raivis Skadin.š, Inguna Skadin.a. 2008. Dictionary of Multiword Expressions for Translation into Highly Inflected Languages. In the 6 th International Conference on Language Resources and Evaluation (LREC 2008), Marrakech, Morocco. find
  6. Duan, Jianyong, Mei Zhang, Lijing Tong, and Feng Guo. 2009. A Hybrid Approach to Improve Bilingual Multiword Expression Extraction. In the 13 th Pacific-Asia Conference on Knowledge Discovery and Data (PAKDD 2009), pp. 541–547. Bangkok, Thailand. find
  7. Elkateb, Sabri, William Black, Horacio Rodrı́guez, Musa Alkhalifa, Piek Vossen, Adam Pease, Christiane Fellbaum. 2006. Building a Wordnet for Arabic. In the 5 th International Conference on Language Resources and Evaluation (LREC 2006), Genoa, Italy.
  8. Habash, Nizar, Owen Rambow and Ryan Roth. 2009. A Toolkit for Arabic Tokenization, Diacritization, Morphological Disambiguation, POS Tagging, Stemming and Lemmatization. In the 2 nd International Conference on Arabic Language Resources and Tools (MEDAR 2009), pp. 102–109. Cairo, Egypt. find
  9. Hoang, Huu Hoang, Su Nam Kim and Min-Yen Kan. 2009. A Re-examination of Lexical Association Measures. In the Workshop on Multiword Expressions, the Joint Conference of the 47 th Annual Meeting of the Association for Computational Linguistics and the 4 th International Joint Conference on Natural Language Processing of the Asian Federation of Natural Language Processing (ACL-IJCNLP 2009), pp. 31–39, Suntec, Singapore. find
  10. Melamed, I. Dan. 1997. Automatic Discovery of Non-Compositional Compounds in Parallel Data. In the 2 nd Conference on Empirical Methods in Natural Language Processing (EMNLP 1997), pp. 97–108. Providence, RI. find
  11. Moirón, Begoña Villada and Jörg Tiedemann. 2006. Identifying idiomatic expressions using automatic word-alignment. In the Workshop on Multiword Expressions in a Multilingual Context, the 11 th Conference of the European Association of Computational Linguistics (EACL 2006), pp. 33–40. Trento, Italy. find
  12. Moon, Rosamund 1998. Fixed Expressions and Idioms in English: A Corpus-Based Approach. Clarendom Press, Oxford. find
  13. Nivre, Joakim and Jens Nilsson. 2004. Multiword Units in Syntactic Parsing. In Workshop on Methodologies and Evaluation of Multiword Units in Real-World Applications, the 4 th International Conference on Language Resources and Evaluation (LREC 2004), pp. 39–46. Lisbon, Portugal. find
  14. Pecina, Pavel 2010. Lexical association measures and collocation extraction. In Language Resources and Evaluation (2010), 44:137-158. ↗
  15. Ramisch, Carlos, Paulo Schreiner, Marco Idiart and Aline Villavicencio. 2008. An Evaluation of Methods for the Extraction of Multiword Expressions. In the Workshop on Multiword Expressions, the 6 th International Conference on Language Resources and Evaluation (LREC 2008), pp. 50–53. Marrakech, Morocco. find
  16. Sag, Ivan A., Timothy Baldwin, Francis Bond, Ann Copestake and Dan Flickinger. 2002. Multiword Expressions: A Pain in the Neck for NLP. In the 3 rd International Conference on Intelligent Text Processing and Computational Linguistics (CICLing-2002), volume 2276 of Lecture Notes in Computer Science, pp. 1–15, London, UK. Springer-Verlag.
  17. SanJuan, Eric and Fidelia Ibekwe-SanJuan. 2006. Text mining without document context. In Information Processing and Management. Volume 42, Issue 6, pp. 1532–1552. ↗
  18. Van de Cruys, Tim and Begoña Villada Moirón. 2006. Lexico-Semantic Multiword Expression Extraction. In P. Dirix et al. (eds.), Computational Linguistics in the Netherlands 2006, pp. 175–190. find
  19. Villavicencio, Aline, Ann Copestake, Benjamin Waldron and Fabre Lambeau. 2004. The Lexical Encoding of MWEs. In the Workshop on Multiword Expressions, the 42 nd Annual Meeting of the Association for Computational Linguistics (ACL 2004), pp. 80–87. Barcelona, Spain. find
  20. Vintar, Špela and Darja Fišer. 2008. Harvesting Multi-Word Expressions from Parallel Corpora. In the 6 th International Conference on Language Resources and Evaluation (LREC 2008). Marrakech, Morocco. find
  21. Zarrieß, Sina and Jonas Kuhn. 2009. Exploiting Translational Correspondences for Pattern-Independent MWE Identification. In the Workshop on Multiword Expressions, the Joint Conference of the 47 th Annual Meeting of the Association for Computational Linguistics and the 4 th International Joint Conference on Natural Language Processing of the Asian Federation of Natural Language Processing (ACL-IJCNLP 2009), pp. 23–30. Suntec, Singapore. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/attia-2010-automatic-extraction-arabic-multiword-expressions/