Coming to Terms: Automatic Formation of Neologisms in Hebrew

What it is about

The system EBY proposes new Hebrew nouns for an English source word, following the Hebrew Academy's three methods: root-and-pattern, compounds and portmanteaus. Roots come from tf-idf over English definitions, translated into Hebrew and matched to roots in Hebrew Wiktionary and a dictionary; patterns come from WordNet sister terms. A character seq2seq model, trained on 3,365 root–pattern pairs from Wiktionary (0.68 accuracy), repairs the naive root-and-pattern combination, and a character 4-gram language model trained on 514K diacritized words ranks the candidates. In a user study with 177 participants on 20 source words, the algorithm wins 27 to 41% of pairwise comparisons against human suggestions, and it reaches the top quartile of the ranking for 10.3% of its outputs.

Ideas

• I1. A rule table for special roots in Mova's Modern Hebrew generator: the paper says that about two-thirds of roots need a change when joined to a pattern, and names the cases (guttural letters, a root ending in aleph, gemination, diphthong to monophthong). Write them as finite rules in Nouns, adjectives, pronouns and numerals and test them on the same 3,365 pairs; the seq2seq's 0.68 is the baseline a rule system must beat.
• I2. Patterns as semantic classes with ambiguity kept: the paper notes that one pattern marks professions and also illnesses. A table that lists every category a pattern can take, instead of picking one, keeps the choice visible to the generator.
• I3. A deterministic candidate ranker: a character n-gram model with n = 4 and add-k smoothing (lowest perplexity of the n tried), scored per word length. Prefixes are stripped by a count rule (the stem is kept only if its count rises), with a stricter threshold for the definite article. A small BTreeMap-based version fits in Rust and can rank candidate forms for 2. Script and spelling.

For Mova

I1 goes into Nouns, adjectives, pronouns and numerals, I2 into the same chapter as a category table, and I3 into a Rust tool under tools/ for ranking candidate forms. The paper's binarized rating, with two cut-offs reported side by side, is a reporting rule that Mova can reuse for any human check.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/2020.findings-emnlp.442

@inproceedings{mizrahi2020coming,
  title = {{Coming to Terms: Automatic Formation of Neologisms in Hebrew}},
  author = {Mizrahi, Moran and Seelig, Stav Yardeni and Shahaf, Dafna},
  booktitle = {Findings of the Association for Computational Linguistics: EMNLP 2020},
  year = {2020},
  doi = {10.18653/v1/2020.findings-emnlp.442},
  url = {https://doi.org/10.18653/v1/2020.findings-emnlp.442}
}

References

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

1. 2020. The Hebrew Academy official website. find
2. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2014. Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473. ↗
3. Outi Bat-El. 1996. Selecting the best of the worst: the grammar of hebrew blends. Phonology, 13(3):283– 328. find
4. Emmanuel Cartier. 2017. Neoveille, a web platform for neologism tracking. In Proceedings of the Software Demonstrations of the 15th Conference of the European Chapter of the Association for Computational Linguistics, pages 95–98. find
5. Pablo Samuel Castro and Maria Attarian. 2018. Combining learned lyrical structures and vocabulary for improved lyric generation. arXiv preprint arXiv:1811.04651. ↗
6. Morten H Christiansen and Nick Chater. 2016. Creating language: Integrating evolution, acquisition, and processing. MIT Press. find
7. Paul Cook and Suzanne Stevenson. 2010. Automatically identifying the source words of lexical blends in english. Computational Linguistics, 36(1):129– 149. find
8. Noam Ordan and Shuly Wintner. 2007. Hebrew wordnet: a test case of aligning lexical databases across languages. International Journal of Translation, 19(1):39–58. find
9. Fintan Costello. 2002. Investigating creative language: People’s choice of words in the production of novel noun-noun compounds. In Proceedings of the Annual Meeting of the Cognitive Science Society, volume 24. find
10. Gözde Özbal and Carlo Strapparava. 2012. A computational approach to the automation of creative naming. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics: Long Papers-Volume 1, pages 703–711. Association for Computational Linguistics. find
11. C. Costin-Gabriel and T. E. Rebedea. 2014. Archaisms and neologisms identification in texts. In 2014 RoEduNet Conference 13th Edition: Networking in Education and Research Joint Event RENAM 8th Conference, pages 1–6. find
12. Aliya Deri and Kevin Knight. 2015. How to make a frenemy: Multitape fsts for portmanteau generation. In Proceedings of the 2015 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 206–210. find
13. Abraham Even-Shoshan and M Azar. 2003. Even shoshan dictionary. Am Oved, Kineret Zmora Bitan, Dvir and Yediot Aaronot, Tel Aviv, page 2039. find
14. Jack Fellman. 1973. The revival of a classical tongue: Eliezer Ben Yehuda and the modern Hebrew language. 6. Walter de Gruyter. find
15. Jack Fellman. 1974. The academy of the hebrew language: Its history, structure and function. Linguistics, 12(120):95–104. find
16. Varun Gangal, Harsh Jhamtani, Graham Neubig, Eduard Hovy, and Eric Nyberg. 2017. Charmanteau: Character embedding models for portmanteau creation. arXiv preprint arXiv:1707.01176. ↗
17. Jörn Hees, Rouven Bauer, Joachim Folz, Damian Borth, and Andreas Dengel. 2016. Edinburgh associative thesaurus as rdf and dbpedia mapping. In The Semantic Web, pages 17–20, Cham. Springer International Publishing. find
18. Daphné Kerremans and Jelena Prokić. 2018. Mining the web for new words: Semi-automatic neologism identification with the neocrawler. Anglia, 136(2):239–268. find
19. Vivek Kulkarni and William Yang Wang. 2018. Simple models for word formation in english slang. arXiv preprint arXiv:1804.02596. ↗ our card
20. Polina Kuznetsova, Jianfu Chen, and Yejin Choi. 2013. Understanding and quantifying creativity in lexical composition. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 1246–1258. find
21. George A Miller. 1995. Wordnet: a lexical database for english. Communications of the ACM, 38(11):39– 41. ↗ our card
22. Chaim Rabin. 1963. The revival of hebrew as a spoken language. The Journal of Educational Sociology, 36(8):388–392. find
23. Juan Ramos et al. 2003. Using tf-idf to determine word relevance in document queries. In Proceedings of the first instructional conference on machine learning, volume 242, pages 133–142. Piscataway, NJ. find
24. Kathy Rooney and PH Collin. 2003. Easier English Student Dictionary: Over 32,000 Terms Clearly Defined, Upper intermediate level. Bloomsbury Publishing. find
25. Burr Settles. 2010. Computational creativity tools for songwriters. In Proceedings of the NAACL HLT 2010 Second Workshop on Computational Approaches to Linguistic Creativity, pages 49–57. Association for Computational Linguistics. find
26. Avi Shmidman, Shaltiel Shmidman, Moshe Koppel, and Yoav Goldberg. 2020. Nakdan: Professional hebrew diacritizer. arXiv preprint arXiv:2005.03312. ↗ our card
27. Jonathan A Simon. 2018. Entendrepreneur: Generating humorous portmanteaus using word-embeddings. find
28. Michael R Smith, Ryan S Hintze, and Dan Ventura. 2014. Nehovah: A neologism creator nomen ipsum. In ICCC, pages 173–181. find
29. Robert Speer and Catherine Havasi. 2012. Representing general relational knowledge in conceptnet 5. In LREC, pages 3679–3686. find
30. Milan Straka and Jana Straková. 2017. Tokenizing, pos tagging, lemmatizing and parsing ud 2.0 with udpipe. In Proceedings of the CoNLL 2017 Shared Task: Multilingual Parsing from Raw Text to Universal Dependencies, pages 88–99, Vancouver, Canada. Association for Computational Linguistics. find our card
31. Tony Veale and Cristina Butnariu. 2010. Harvesting and understanding on-line neologisms. Cognitive Perspectives on Word Formation, 221:399. find
32. Xiaojin Zhu, Zhiting Xu, and Tushar Khot. 2009. How creative is your writing? In Proceedings of the workshop on computational approaches to linguistic creativity, pages 87–93. A In these sections we provide more implementation details for the sake of reproducibility, some qualitative evaluations of the models and a short discussion about the choice of our metrics. We release the source, data and train-validation-test splits here. A.1 Perplexity 11.41 6.0 4.72 6.37 14.64 *Table 2: Character based n-gram language model perplexity on the validation set for different n values.* Implementation details: Seq2seq find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/findings-2020-mizrahi-coming-terms-automatic-formation-neologisms/