Extracting Semantic Transfer Rules from Parallel Corpora with SMT Phrase Aligners

What it is about

Transfer rules for the Japanese-to-English rule-based system Jaen are learned from a parallel corpus of about 20 million words, using SMT phrase aligners. There are two routes. In the shallow route the text is lemmatised, the aligned lemmas are mapped to semantic predicates through the grammars' lexicons, and the predicate pairs are matched against 16 rule templates, giving about 127,000 rules. In the deep route both sides are parsed first, the parsed predicates are aligned, and 22 templates give about 113,000 rules. The shallow route is more robust, the deep one more precise. Each route alone translates about a fifth of the test sentences, and the union of both rule sets raises coverage to 26.6%. A human judge preferred Jaen to Moses on 58 of 100 sampled sentences. A pre-selection step loads only about 15,000 of the 190,000 rules for a batch of 1,500 sentences, which the paper reports as a threefold speed-up.

Ideas

• I1. Run two alignment routes, one on surface tokens and one on the parsed meta tokens, keep the pairs that both routes agree on and the pairs that only one route finds, and measure each route's precision and coverage on dev. This is a concrete form of open problem 7 (Hebrew alignment).
• I2. Give each extracted entry a template name (for example noun-to-noun, or compound-to-adjective-plus-noun) and keep only templates with a counted minimum of support. An error can then be traced to its template instead of to a single entry.
• I3. Load only the lexicon entries whose concepts occur in the input verse. Measure speed and check that the output is byte-identical with and without the pre-selection, twice on the same input.

For Mova

The code is tools/translate/meta.rs (concept merges) and hebrew.rs (alignment). I1 is a direct test of open problem 7, and I2 suggests a counted, template-named form of the lexicon that open problem 2 asks for. The paper's rules are learned from parsed predicates, which is the direction open problem 1 proposes for the meta-world.

Topics: Machine translation

Cite

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

@inproceedings{haugereid2012extracting,
  title = {{Extracting Semantic Transfer Rules from Parallel Corpora with SMT Phrase Aligners}},
  author = {Haugereid, Petter and Bond, Francis},
  booktitle = {Proceedings of the Sixth Workshop on Syntax, Semantics and Structure in Statistical Translation},
  year = {2012},
  url = {https://aclanthology.org/W12-4208/}
}

References

16 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Peter Adolphs, Stephan Oepen, Ulrich Callmeier, Berthold Crysmann, Dan Flickinger, and Bernd Kiefer. 2008. Some fine points of hybrid natural language parsing. In European Language Resources Association (ELRA), editor, Proceedings of the Sixth International Language Resources and Evaluation (LREC’08), pages 1380–1387. Marrakech, Morocco. find
2. Francis Bond, Hitoshi Isahara, Kiyotaka Uchimoto, Takayuki Kuribayashi, and Kyoko Kanzaki. 2010. Japanese WordNet 1.0. In 16th Annual Meeting of the Association for Natural Language Processing, pages A5–3. Tokyo. find
3. Francis Bond, Stephan Oepen, Eric Nichols, Dan Flickinger, Erik Velldal, and Petter Haugereid. 2011. Deep open source machine translation. Machine Translation, 25(2):87–105. URL http://dx.doi.org/10.1007/s10590-011-9099-4, (Special Issue on Open source Machine Translation). ↗
4. James W. Breen. 2004. JMDict: a Japanese-multilingual dictionary. In Coling 2004 Workshop on Multilingual Linguistic Resources, pages 71– 78. Geneva. find
5. Ann Copestake, Dan Flickinger, Carl Pollard, and Ivan A. Sag. 2005. Minimal Recursion Semantics. An introduction. Research on Language and Computation, 3(4):281–332. find
6. Bart Cramer and Yi Zhang. 2010. Constraining robust constructions for broad-coverage parsing with precision grammars. In Proceedings of COLING-2010, pages 223–231. Beijing. find
7. Dan Flickinger. 2000. On building a more efficient grammar by exploiting types. Natural Language Engineering, 6(1):15–28. (Special Issue on Efficient Processing with HPSG). find our card
8. Eva Forsbom. 2003. Training a super model lookalike: Featuring edit distance, n-gram occurrence, and one reference translation. In In Proceedings of the Workshop on Machine Translation Evaluation. Towards Systemizing MT Evaluation. find
9. Philipp Koehn, Wade Shen, Marcello Federico, Nicola Bertoldi, Chris Callison-Burch, Brooke Cowan, Chris Dyer, Hieu Hoang, Ondrej Bojar, Richard Zens, Alexandra Constantin, Evan Herbst, Christine Moran, and Alexandra Birch. 2007. Moses: Open source toolkit for statistical machine translation. In Proceedings of the ACL 2007 Interactive Presentation Sessions. Prague. URL http://www.statmt.org/moses/. ↗
10. Taku Kudo, Kaoru Yamamoto, and Yuji Matsumoto. 2004. Applying Conditional Random Fields to Japanese Morphological Analysis. In Dekang Lin and Dekai Wu, editors, Proceedings of EMNLP 2004, pages 230–237. Association for Computational Linguistics, Barcelona, Spain. find
11. Adrien Lardilleux and Yves Lepage. 2009. Sampling-based multilingual alignment. In Proceedings of Recent Advances in Natural Language Processing (RANLP 2009), pages 214–218. Borovets, Bulgaria. find
12. Stephan Oepen, Erik Velldal, Jan Tore Lønning, Paul Meurer, and Victoria Rosen. 2007. Towards hybrid quality-oriented machine translation. on linguistics and probabilities in MT. In 11th International Conference on Theoretical and Methodological Issues in Machine Translation: TMI-2007, pages 144–153. find
13. Lluís Padró, Miquel Collado, Samuel Reese, Marina Lloberes, and Irene Castellón. 2010. Freeling 2.1: Five years of open-source language processing tools. In Proceedings of 7th Language Resources and Evaluation Conference (LREC 2010). La Valletta. (http://nlp.lsi.upc.edu/freeling. ↗
14. Melanie Siegel and Emily M. Bender. 2002. Efficient deep processing of Japanese. In Proceedings of the 3rd Workshop on Asian Language Resources and International Standardization at the 19th International Conference on Computational Linguistics, pages 1–8. Taipei. find
15. Yasuhito Tanaka. 2001. Compilation of a multilingual parallel corpus. In Proceedings of PACLING 2001, pages 265–268. Kyushu. (http://www.colips.org/afnlp/archives/pacling2001/pdf/tanaka.pdf). find
16. Kiyotaka Uchimoto, Yujie Zhang, Kiyoshi Sudo, Masaki Murata, Satoshi Sekine, and Hitoshi Isahara. 2004. Multilingual aligned parallel treebank corpus reflecting contextual information and its applications. In Gilles Sérasset, editor, COLING 2004 Multilingual Linguistic Resources, pages 57–64. COLING, Geneva, Switzerland. URL http://acl.ldc.upenn.edu/W/W04/W04-2208.bib. ↗

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/haugereid-2012-extracting-semantic-transfer-rules-parallel/