Using Multilingual Topic Models for Improved Alignment in English-Hindi MT

Parallel corpora are often injected with bilingual dictionaries for improved Indian language machine translation (MT). In ab-sence of such dictionaries, a coarse dictio-nary may be required. This paper demon-strates the use of a multilingual topic model for creating coarse dictionaries for English-Hindi MT. We compare our ap-proaches with: (a) a baseline with no ad-ditional dictionary injection, and (b) a cor-pus with a good quality dictionary. Our re-sults show that the existing Cartesian prod-uct approach which is used to create the pseudo-parallel data results in a degrada-tion on tourism and health datasets, for English-Hindi MT. Our paper points to the fact that existing Cartesian approach using multilingual topics (devised for European languages) may be detrimental for Indian language MT. On the other hand, we present an alter-nate ‘sentential ’ approach that leads to a slight improvement. However, our sen-tential approach (using a parallel corpus injected with a coarse dictionary) outper-forms a system trained using parallel cor-pus and a good quality dictionary.