Czech-English Dependency Tree-based Machine Translation

What it is about

The paper reports a first full Czech-to-English translation system built on tectogrammatical dependency trees. Czech text is tagged, parsed into analytical and then tectogrammatical trees, and each content word is translated by a lexical substitution. The substitution uses a Czech-English dictionary merged from three sources and filtered with probabilities that GIZA++ estimates on a parallel corpus of Wall Street Journal sentences translated into Czech; a translation is kept when its probability is above 0.10. The English output is then generated by five rule-based steps: contextual boundness, reordering, verb forms, insertion of prepositions and articles, and morphology. The training and test data are 11,189 translated WSJ sentences. On 490 test sentences, the system with automatic parses reaches a BLEU of about 0.16 to 0.19. Human retranslations score about 0.56, and a GIZA++ word-form baseline scores lower on the same data.

Ideas

• I1. Article insertion as a count rule: the paper uses contextual boundness, not the surface order, to choose "the" or "a". Measure this rule on the Bible pairs for the encoder: count how often the English "the" that a source pair gives matches a rule based on whether the noun was already mentioned or is postmodified (open problem 4). Report the confusion table, run twice.
• I2. Filter a concept merge by its evidence: keep a lexicon entry only if its pair probability from aligned counts is above a threshold and the part of speech of both sides agrees (the paper's two filters). This applies to the lexicon in tools/translate/src/meta.rs (open problem 2).
• I3. Reorder meta tokens by information status: place the bound adjuncts first, then the subject, verb and complements, as the paper does. Try this as an order table in en/src/lin.rs for the encoder, and compare it with the current Markov chains by exact match.
• I4. Measure how much parser noise costs: the paper compares a Czech parser with gold manually annotated trees. Do the same for the reader: run it on gold and on automatic trees, and count how many bits of the residual come from attachment alone (open problem 1).

For Mova

The code is tools/translate/src/meta.rs (dictionary filter) and en/src/lin.rs (order rules). The paper is rule-based for generation, which fits Mova, and its dictionary is a count-based table, which also fits. Its statistical part is only a filter, so no neural method is needed. Its low BLEU is also a warning: report chrF and exact match together, not BLEU alone.

Topics: Dependency parsing · Machine translation

Cite

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

@inproceedings{cmejrek2003czech,
  title = {{Czech-English Dependency Tree-based Machine Translation}},
  author = {Čmejrek, Martin and Cuřín, Jan and Havelka, Jiří},
  booktitle = {10th Conference of the European Chapter of the Association for Computational Linguistics},
  year = {2003},
  url = {https://aclanthology.org/E03-1004/}
}

References

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

1. Yaser Al-Onaizan, Jan Cuiin, Michael Jahr, Kevin Knight, John Lafferty, Dan Melamed, Franz-Josef Och, David Purdy, Noah A. Smith, and David Yarowsky. 1999. The statistical machine translation. Technical report. WS' 99, Johns Hopkins University. find
2. Alena Biihmova, Jan Hajie', Eva Hajie'ova, and Barbora Hladka. 2001. The Prague Dependency Treebank: Three-Level Annotation Scenario, In Anne Abeillê, editor, Treebanks: Building and Using Syntactically Annotated Corpora. Kluwer Academic Publishers. find
3. Alena Biihmova. 2001. Automatic procedures in tectogrammatical tagging. The Prague Bulletin of Mathematical Linguistics, 76. find
4. Eugene Charniak. 1999. A maximum-entropy-inspired parser. Technical Report CS-99-12. find
5. Ulrich Germann, Michael Jahr, Kevin Knight, Daniel Marcu, and Kenji Yamada. 2001. Fast decoding and optimal decoding for machine translation. In Proceedings of the 39th Annual Meeting of the Association for Computational Linguistics, pages 228– 235. find
6. Jan Ha* and Barbora Hladka. 1998. Tagging Inflective Languages: Prediction of Morphological Categories for a Rich, Structured Tagset. In Proceedings of COLING-ACL Conference, pages 483-490, Montreal, Canada. Jan Haji, Eric Brill, Michael Collins, Barbora Hladka, Douglas Jones, Cynthia Ku o, Lance Ramshaw, Oren Schwartz, Christopher Tillmann, and Daniel Zeman. 1998. Core Natural Language Processing Technology Applicable to Multiple Languages. Technical Report Research Note 37, Center for Language and Speech Processing, Johns Hopkins University, Baltimore, MD. find
7. Jan Hajie, Jarmila Panevova, Eva Buraliova, Zdefika Uregova, Alla Bemova, Jan ‘Cepanek, Petr Pajas, and Jill Karnfk, 2001. A Manual for Analytic Layer Tagging of the Prague Dependency Treebank. Prague, Czech Republic. English translation of the original Czech version, http://shadow.ms.mff.cuni.cz/pdt/Corpora/PDT 1. 0/References/aman en.pdf 9 6. ↗
8. Jan Haji6, Martin 'Cmejrek, Bonnie Dorr, Yuan Ding, Jason Eisner, Daniel Gildea, Terry Koo, Kristen Parton, Gerald Penn, Dragomir Radev, and Owen Rambow. 2002. Natural language generation in the context of machine translation. Technical report. WS' 02, Johns Hopkins University — in preparation. find
9. Eva Hajie'ova, Jarmila Panevova, and Petr Sgall. 2000. A Manual for Tectogrammatic Tagging of the Prague Dependency Treebank. Technical Report TR-2000-09, f_JFAL MFF UK, Prague, Czech Republic. find
10. G. Minnen, J. Carroll, and D. Pearce. 2001. Applied morphological processing of English. Natural Language Engineering, 7(3):207-223. find
11. F. J. Och and H. Ney. 2000. Improved statistical alignment models. In Proc. of the 38th Annual Meeting of the Association for Computational Linguistics, pages 440-447, Hongkong, China, October. find
12. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2001. Bleu: a method for automatic evaluation of machine translation. Technical Report RC22176, IBM. find our card
13. Adwait Ratnaparkhi. 1996. A maximum entropy part-of-speech tagger. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 133-142, University of Pennsylvania, May. ACL. find
14. Zden6k 2abokrts14, Petr Sgall, and Meroski Sago. 2002. Machine learning approach to automatic functor assignment in the Prague Dependency Treebank. In Proceedings of LREC 2002 (Third International Conference on Language Resources and Evaluation), volume V, pages 1513-1520, Las Palmas de Gran Canaria, Spain. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/cmejrek-2003-czech-english-dependency-tree-based/