Paper: XMEANT: Better semantic MT evaluation without reference translations

ACL ID P14-2124
Title XMEANT: Better semantic MT evaluation without reference translations
Venue Annual Meeting of the Association of Computational Linguistics
Session Main Conference
Year 2014
Authors

We introduce XMEANT?a new cross-lingual version of the semantic frame based MT evaluation metric MEANT?which can cor- relate even more closely with human ade- quacy judgments than monolingual MEANT and eliminates the need for expensive hu- man references. Previous work established that MEANT reflects translation adequacy with state-of-the-art accuracy, and optimiz- ing MT systems against MEANT robustly im- proves translation quality. However, to go beyond tuning weights in the loglinear SMT model, a cross-lingual objective function that can deeply integrate semantic frame crite- ria into the MT training pipeline is needed. We show that cross-lingual XMEANT out- performs monolingual MEANT by (1) replac- ing the monolingual context vector model in MEANT with simple translation probabilities,...