Deep Grammars in Hybrid Machine Translation

old_uid5248
titleDeep Grammars in Hybrid Machine Translation
start_date2008/09/08
schedule14h30-16h30
onlineno
summaryThe topic of the talk is a proof-of-concept machine translation demonstrator translating tourism-related texts from Norwegian into English. The system is developed within a project called LOGON (http://www.emmtee.net/) in cooperation between groups at the universities of Bergen, Oslo and Trondheim. It is a hybrid transfer system in which the backbone is rule-based, with statistical processing interspersed. The Norwegian analysis is performed by a comprehensive Norwegian Lexical-Functional grammar developed on the Xerox Linguistic Environment (XLE) platform, in conjunction with morphological analysis and other modules. The grammar is augmented with a Minimal Recursion Semantics (MRS) component, whose representations are the input to transfer. An unusual feature of the system is that it combines two grammatical models: the English target grammar is the English Resource Grammar (ERG), which is based on the HPSG framework. Statistical processing is responsible for parse selection, ranking of transfer outputs, and ranking of generator outputs. The presentation will focus on the analysis component of the system, including LFG grammar development, the development of the MRS component, and the development of a treebank tool called a 'Parse Banker', which is a discriminant-based system for selecting the desired analysis efficiently in treebank construction. The Parse Banker was used to produce training material for the derivation of the statistical model for parse selection.
responsiblesInformation non disponible, Crabbé