Pangloss: A Knowledge-based Machine Assisted Translation Research Project - Site

are developing a Translator's Workstation to assist a user in the translation of newspaper articles in the area of finance (mergers and acquisitions) in one language (Spanish initially) into a second language (English). At its core is a multilingual, knowledge-based, interlingual, interactive , machine-assisted translation system consisting of a source language analysis component, an interactive aug-mentor, and a target language generation component. In the initial phase, the CRL's objectives are to develop tools for constructing lexical items and ontological entries automatically from on-line resources, to develop the initial Spanish analysis component, and, jointly with CMT and ISI, to establish the infrastructure for the three site project, develop the formats and initial content of the interlingua, the ontology, and the knowledge base, and to prepare design documents for the second phase versions of the analysis and generation components, the augmentor, and the translator's workstation. With respect to developing tools for extracting information from on-line resources, during the first year we are focusing on providing general classificatory information for use in constructing the lexicons of the Spanish analysis and English generation components. Here we are building on work on automatically constructing interlin-gual word sense specifications from Longman's which we are extending and adapting to language particular lexical entries. With respect to the Spanish analysis component , the objective is to modify and extend the Spanish component of the CRL's multilingual machine translation system both in terms of coverage and robustness. At this point, the three sites have established the central infrastructure for the Pangloss project. For the first year system, the formats of the interlingua, the ontology, and the knowledge base have been set and initial design of the interlingua for the second phase version is underway. As a preliminary to the work on extracting information from on-line resources, we continue to gather resources in the form of monolingual and bilingual dictionaries and monolingual and bilingual corpora. We have obtained Collins English Dictionary, Collins Bilingual Spanish-English Dictionary and are looking into a Spanish mono-lingual dictionary and Japanese monolingual and bilingual dictionaries. We have identified a source for Span-ish texts in the financial domain and are seeking further sources as well as sources of Japanese texts. The central resource to date, however, is the Longman Dictionary of Contemporary English and the information being extracted relates to word formation, syntactic category, syntactic subcategorization, and semantic selection restrictions. Procedures are currently being developed to provide …