Grammar-Based Concept Alignment for Domain-Specific Machine Translation

What it is about

The paper proposes concept alignment (CA) from parallel sentences in a grammar-based way, to build the translation lexicon of a domain-specific GF system. It takes UD trees for both sides of a sentence pair, and aligns subtrees recursively: a pair is aligned if its roots have the same dependency label, with a backup rule on matching open-class POS multisets, and with the heads of matched subtrees also added so that one-word correspondences are not lost. Known translation divergences are handled explicitly, such as an adverb translated as a prepositional phrase. The aligned pairs are converted into GF lexicon rules by the gf-ud tool. On 100 sentences of the Parallel UD corpus, the alignments were correct for 77 percent (English-Italian) and 85 percent (English-Swedish) of distinct pairs, against 42 and 64 percent for the fast_align statistical aligner trained on 1,000 sentences. On raw course-plan text, parsed with UDPipe, correctness was 63 to 66 percent. A small MT test on generated sentences then used the lexicon.

Ideas

• I1. Align the trees of the two languages for each parallel verse, with priority-ordered criteria (same label first, then POS match, then head alignment). This is an alternative to the longest-common-subsequence anchors of meta.rs, and it can be measured by counting merges and checking a hand-labelled sample against the current lexicon.
• I2. Keep a table of known divergences as data: adverb against oblique, direct object against oblique. Each rule has a hit count, and an author's constructions (such as "it came to pass" against "it happened") can be added as one more row instead of as code.
• I3. Separate pair-specific alignments from alignments usable for a standard. The paper's example, an idiom that mentions a different object in each language, is correct for the pair but must not enter the lexicon. Mova's merge needs the same label, so an idiomatic merge stays with its author.
• I4. Use an open-class POS-multiset check as a cheap test on each proposed merge, as the paper does. It would catch merges of the kind be with have, where function words are removed before the comparison.

For Mova

The code is tools/translate (meta.rs for merges and anchors, hebrew.rs for the tree alignment with MACULA trees) and tools/roundtrip for the trees. The alignment criteria and the divergence table fit open problems 1 and 2: the lexicon keeps a merge only if the trees agree. The paper's precision figures are for a different task and are not a target for Mova; the criterion that a merge should be usable in a standard is the part to copy.

Topics: Machine translation

Cite

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

@inproceedings{masciolini2021grammar,
  title = {{Grammar-Based Concept Alignment for Domain-Specific Machine Translation}},
  author = {Masciolini, Arianna and Ranta, Aarne},
  booktitle = {Proceedings of the Seventh International Workshop on Controlled Natural Language (CNL 2020/21)},
  year = {2021},
  url = {https://aclanthology.org/2021.cnl-1.2/}
}

References

14 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Peter F. Brown, Stephen A. Della Pietra, Vincent J. Della Pietra, and Robert L. Mercer. 1993. The mathematics of statistical machine translation: Parameter estimation. Computational Linguistics, 19(2):263– 311. find
2. Danqi Chen and Christopher Manning. 2014. A fast and accurate dependency parser using neural networks. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 740–750, Doha, Qatar. Association for Computational Linguistics. find our card
3. Bonnie J. Dorr. 1994. Machine translation divergences: A formal description and proposed solution. Computational Linguistics, 20(4):597–633. find our card
4. Chris Dyer, Victor Chahuneau, and Noah A. Smith. 2013. A simple, fast, and effective reparameterization of IBM model 2. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 644–648, Atlanta, Georgia. Association for Computational Linguistics. find
5. Eriksson, Gabrielsson, Hedgreen, Klingberg, Vestlund, and Ödin. 2020. Grammar-based translation of computer science and engineering terminology. find
6. Prasanth Kolachina and Aarnte Ranta. 2016. From abstract syntax to Universal Dependencies. In Linguistic Issues in Language Technology, Volume 13, 2016. CSLI Publications. find
7. Franz Josef Och and Hermann Ney. 2000. Improved statistical alignment models. In Proceedings of the 38th Annual Meeting of the Association for Computational Linguistics, pages 440–447, Hong Kong. Association for Computational Linguistics. find
8. Slav Petrov, Dipanjan Das, and Ryan McDonald. 2012. A universal part-of-speech tagset. In Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC’12), pages 2089– 2096, Istanbul, Turkey. European Language Resources Association (ELRA). find
9. Alexandre Rademaker and Francis Tyers, editors. 2019. Proceedings of the Third Workshop on Universal Dependencies (UDW, SyntaxFest 2019). Association for Computational Linguistics, Paris, France. find
10. Aarne Ranta. 2011. Grammatical Framework: Programming with Multilingual Grammars. CSLI Publications, Stanford. find
11. Aarne Ranta, Krasimir Angelov, Normunds Gruzitis, and Prasanth Kolachina. 2020. Abstract syntax as interlingua: Scaling up the grammatical framework from controlled languages to robust pipelines. Computational Linguistics, 46(2):425–486. find our card
12. Aarne Ranta and Prasanth Kolachina. 2017. From Universal Dependencies to abstract syntax. In Proceedings of the NoDaLiDa 2017 Workshop on Universal Dependencies (UDW 2017), pages 107–116, Gothenburg, Sweden. Association for Computational Linguistics. find
13. Milan Straka, Jan Hajič, and Jana Straková. 2016. UD-Pipe: Trainable pipeline for processing CoNLL-U files performing tokenization, morphological analysis, POS tagging and parsing. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC’16), pages 4290– 4297, Portorož, Slovenia. European Language Resources Association (ELRA). find
14. Jörg Tiedemann. 2011. Bitext alignment. Synthesis Lectures on Human Language Technologies, 4(2):1– 165. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/masciolini-2021-grammar-based-concept-alignment-domain/