Using a Bi-Lingual Dictionary in Lexical Transfer.
Lars Nygaard, Jan Tore Lønning, Torbjørn Nordgård, Stephan Oepen · 2006
This paper reports on experiences with fully automatic lexical transfer rule acquisition in the domain of rule-based machine translation with deep syntactic and semantic processing. We demonstrate that if comprehensive grammars with broad lexical coverage exist for the source and target languages, then open POS class lexical transfer rules can be derived automatically if access to some bilingual dictionary is provided. We show the results in terms of extended coverage and BLEU scores. 1 Project Background LOGON is a machine translation project (Oepen et al., 2004) which makes use of Lexical-Functional Grammar in parsing (on the XLE platform) and Head-Driven Phrase Structure Grammar in generation (via the LKB platform) to translate texts about hiking in the backcountry from Norwegian to English. A training data set containing approximately 5000 sentences has been collected and translated by professional translators (three translations for each sentence). The project has access to a comprehensive machine-readable bilingual dictionary (Kunnskapsforlaget, 2001) with more than 200.000 translation pairs (approximately 80.000 from Norwegian to English). The Norwegian analysis component is based on an existing LFG resource grammar, NorGram, under development on the Xerox Linguistic Environment (XLE), see {larsnyg | jtl | torbjorn |