Response-based Learning for Grounded Machine Translation

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 881–891, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Response-based Learning for Grounded Machine Translation Stefan Riezler and Patrick Simianer and Carolin Haas Department of Computational Linguistics Heidelberg University, 69120 Heidelberg, Germany {riezler,simianer,haas1}@cl.uni-heidelberg.de Abstract We propose a novel learning approach for statistical machine translation (SMT) that allows to extract supervision signals for structured learning from an extrinsic re- sponse to a translation input. We show how to generate responses by grounding SMT in the task of executing a seman- tic parse of a translated query against a database. Experiments on the GEO- QUERY database show an improvement of about 6 points in F1-score for response- based learning over learning from refer- ences only on returning the correct an- swer from a semantic parse of a translated query. In general, our approach alleviates the dependency on human reference trans- lations and solves the reachability problem in structured learning for SMT. 1 Introduction In this paper, we propose a novel approach for learning and evaluation in statistical ma- chine translation (SMT) that borrows ideas from response-based learning for grounded semantic parsing. In this framework, the meaning of a sen- tence is defined in the context of an extrinsic task. Successful communication of meaning is mea- sured by a successful interaction in this task, and feedback from this interaction is used for learning. We suggest that in a similar way the preser- vation of meaning in machine translation should be defined in the context of an interaction in an extrinsic task. For example, in the context of a game, a description of a game rule is translated successfully if correct game moves can be per- formed based only on the translation. In the con- text of a question-answering scenario, a question is translated successfully if the correct answer is returned based only on the translation of the query. We propose a framework of response-based learning that allows to extract supervision signals for structured learning from the response of an extrinsic task to a translation input. Here, learn- ing proceeds by “trying out” translation hypothe- ses, receiving a response from interacting in the task, and converting this response into a supervi- sion signal for updating model parameters. In case of positive feedback, the predicted translation can be treated as reference translation for a structured learning update. In case of negative feedback, a structural update can be performed against transla- tions that have been approved previously by pos- itive task feedback. This framework has several advantages: • The supervision signal in response-based learning has a different quality than super- vision by human-generated reference transla- tions. While a human reference translation is generated independently of the SMT task, conversion of predicted translations into ref- erences is always done with respect to a spe- cific task. In this sense we speak of ground- ing meaning transfer in an extrinsic task. • Response-based learning can repeatedly try out system predictions by interacting …

Content truncated to 500 words