What's Wrong with Hebrew NLP? And How to Make it Right
What it is about
The paper presents ONLP (YAP), an open system for Modern Hebrew that does segmentation, POS tagging, lemmas, features and labelled dependencies in one package. The argument is that pipelines fail on Hebrew because an early morphological decision is wrong and no later step can recover from it. YAP therefore keeps the whole morphological lattice of each sentence (every possible segmentation with its arc-level analyses) and runs a joint transition-based search, so the morphological choice and the dependency tree are made together. The paper does not give new accuracy figures; it points to earlier results and reports that the system outperforms standard pipelines such as UDPipe and CoreNLP on Hebrew. It describes the output formats, a web demo that shows the lattice for debugging, a 6 GB memory requirement, and the lexicon (about 500K entries) as the limit for out-of-domain text.
Ideas
- I1. Keep the lattice: let
tools/hbooutput every segmentation candidate of a token, with its tag and lemma per arc, and pass the whole lattice togparse, so that the parser can choose the segmentation that fits the tree. The paper's own error analysis is the reason. - I2. A debugging view like the ONLP demo: show the candidate lattice next to the parse for each sentence. It makes each choice visible, which is the transparency Mova needs.
- I3. Out-of-vocabulary policy: for an unknown token, offer the most likely analyses seen in training (the paper uses ten), and measure a domain lexicon extension as an ablation, reporting the change without claiming a gain before two-run checks.
For Mova
I1 and I2 are design for tools/hbo (segmenter output and the debug view) and for 13. Biblical Hebrew in Mova: the tools. I3 is a test for The lexicon and for the Modern Hebrew plan in 12. Mova and Modern Hebrew: where it stands, what it would need. The paper's SPMRL-to-UD conversion repository is a source for treebanks/he/.
Cite
@inproceedings{tsarfaty2019what,
title = {{What's Wrong with Hebrew NLP? And How to Make it Right}},
author = {Tsarfaty, Reut and Sadde, Shoval and Klein, Stav and Seker, Amit},
booktitle = {Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP): System Demonstrations},
year = {2019},
doi = {10.18653/v1/d19-3044},
url = {https://doi.org/10.18653/v1/d19-3044}
}References
- Meni Adler and Michael Elhadad. 2006. An unsupervised morpheme-based hmm for Hebrew morphological disambiguation. In ACL. The Association for Computer Linguistics. find our card
- Reut Tsarfaty. 2006. Integrated morphological and syntactic disambiguation for modern Hebrew. In Proceedings ACL-CoLing Student Research Workshop, pages 49–54, Stroudsburg, PA, USA. ACL. find our card
- Roy Bar-haim, Khalil Sima’an, and Yoad Winter. 2008. Part-of-speech tagging of Modern Hebrew text. Natural Language Engineering, 14(2):223–251. ↗
- Reut Tsarfaty. 2010. Relational-realizational parsing. Ph.D. thesis. find
- Sabine Buchholz and Erwin Marsi. 2006. Conll-x shared task on multilingual dependency parsing. In Proceedings of CoNLL, pages 149–164. find our card
- Reut Tsarfaty. 2013. A unified morphosyntactic scheme for stanford dependencies. In Proceedings of ACL. find
- Yoav Goldberg and Michael Elhadad. 2009. Hebrew dependency parsing: Initial results. In Proceedings of the 11th International Conference on Parsing Technologies, IWPT ’09, pages 129–133. find our card
- Reut Tsarfaty, Djamé Seddah, Yoav Goldberg, Sandra Kübler, Marie Candito, Jennifer Foster, Yannick Versley, Ines Rehbein, and Lamia Tounsi. 2010. Statistical parsing of morphologically rich languages (spmrl): What, how and whither. In Proceedings of the NAACL HLT 2010 First Workshop on Statistical Parsing of Morphologically-Rich Languages, SPMRL ’10, pages 1–12, Stroudsburg, PA, USA. Association for Computational Linguistics. find
- Yoav Goldberg and Michael Elhadad. 2011. Joint Hebrew segmentation and parsing using a PCFGLA lattice parser. In Proceedings of ACL. find our card
- Yoav Goldberg and Reut Tsarfaty. 2008. A single framework for joint morphological segmentation and syntactic parsing. In Proceedings of ACL. find
- Yue Zhang and Stephen Clark. 2011. Syntactic processing using the generalized perceptron and beam search. Computational Linguistics, 37(1):105–151. find
- Spence Green and Christopher D. Manning. 2010. Better Arabic parsing: Baselines, evaluations, and analysis. In Proceedings of COLING. find our card
- Yue Zhang and Joakim Nivre. 2011. Transition-based dependency parsing with rich non-local features. In Proceedings of the ACL, HLT ’11, pages 188–193, Stroudsburg, PA, USA. ACL. find
- Matthew Honnibal and Ines Montani. 2017. spaCy 2: Natural language understanding with Bloom embeddings, convolutional neural networks and incremental parsing. To appear. find
- A Alon Itai and Shuly Wintner. 2008. Language resources for Hebrew. Language Resources and Evaluation, 42(1):75–98. Resources 1. Paper Appendix and Supplementary Materials: https://arxiv.org/abs/1908.05453 2. Github: https://github.com/OnlpLab/yap 3. Demo - Website: http://onlp.openu.org.il 4. Demo - Screencast: https://www.youtube.com/watch?v= H6pvh1x20FQ 5. API Docker Image: https://hub.docker.com/r/onlplab/ yap-api 6. SPMRL-to-UD Conversion: https://github.com/OnlpLab/Hebrew_UD 7. ONLP Lab Website: http://onlp.openu.org.il/home ↗
- Christopher Manning, Mihai Surdeanu, John Bauer, Jenny Finkel, Steven Bethard, and David McClosky. 2014. The stanford corenlp natural language processing toolkit. In Proceedings ACL: system demonstrations, pages 55–60. find
- Amir More, Amit Seker, Victoria Basmova, and Reut Tsarfaty. 2019. Joint transition-based models for morpho-syntactic parsing: Parsing strategies for MRLs and a case study from modern Hebrew. Transactions of the Association for Computational Linguistics, 7:33–48. find our card
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/aclbulk-2019-tsarfaty-what-s-wrong-hebrew-nlp/