Proceedings of Machine Translation Summit XI: Papers

Bente Maegaard (Editor)


Anthology ID:
2007.mtsummit-papers
Month:
September 10-14
Year:
2007
Address:
Copenhagen, Denmark
Venue:
MTSummit
SIG:
Publisher:
URL:
https://aclanthology.org/2007.mtsummit-papers/
DOI:
Bib Export formats:
BibTeX MODS XML EndNote

The paper presents and evaluates a wide coverage, rule-governed machine translation system for Danish-English. Analysis and polysemy resolution are based on Constraint Grammar dependency trees. In its 85.000 lexeme lexicon, Dan2eng uses context-sensitive lexical transfer rules linking dependencies to semantic prototype conditions, syntactic function, definiteness etc. Dependency is further exploited instead of constituent bracketing to support syntactic movement rules. A robust derivational and compound analysis, as well as a separate NER module permit the handling of unrestricted text from a wide range of genres. The system averaged TER scores of 7 (BLEU 0.55-0.6) on student tasks, but performance varied widely against raw and edited Europarl references, respectively.