Toward Semantic Machine Translation

This thesis presents a novel approach to interlingual machine translation using λ-calculus expressions as an intermediate representation. It investigates and extends existing algorithms which learn a combinatorial category grammar for semantic parsing, and introduces two new algorithms for generation out of logical forms inspired by that semantic parser. The results of a set of new experiments for generation and parsing are described, as well as an evaluation of the performance of a semantic translation system created by joining the semantic parser and generator together. Experimental results demonstrate that under certain conditions, this semantic model achieves better performance than a standard phrase-based statistical MT system in both an automated evaluation of translation output and a manual evaluation of adequacy and fluency