Machine translation from deep-syntax, dependency parsing, and generation
Speaker:Keith Hall
Institution: Johns Hopkins University
Abstract:
Transfer-based syntactic machine translation is a relic of natural language processing that reappears regularly. I will talk about the inherent inadequacy of traditional transfer-based systems and then introduce a transfer-based, deep-syntactic translation system which we have been exploring. In order to overcome the aforementioned shortcomings, we propose a technique which retains ambiguity in the analyses produced at each step of transfer (surface analysis, deep analysis, lexical transfer, and generation). Finally, I will present an overview of the current n-best techniques we are developing for our translation system.