A framework for interactive and automatic refinement of transfer-based machine translation

EAMT 2005 Conference Proceedings 87 A Framework for Interactive and Automatic Refinement of Transfer-based Machine Translation Ariadna Font Llitjós, Jaime G Carbonell and Alon Lavie Language Technologies Institute, Carnegie Mellon University 5000 Forbes Ave. Pittsburgh PA 15213 USA aria@cs.cmu.edu Abstract. Most current Machine Translation (MT) systems do not improve with feedback from post-editors beyond the addition of corrected translations to parallel training data (for statistical and example-base MT) or to a memory database. Rule based systems to date im- prove only via manual debugging. In contrast, we introduce a largely automated method for capturing more information from human post-editors, so that corrections may be performed automatically to translation grammar rules and lexical entries. This paper introduces a gen- eral framework for incorporating a refinement module to rule-based transfer MT systems. This framework allows to generalize post-editing efforts in an effective way, by identifying and correcting rules semi-automatically to improve coverage and overall translation quality. 1. Introduction Although Machine Translation (MT) has ad- vanced recently for language pairs with large amounts of parallel data, translation quality has not yet reached satisfactory levels, especially not for resource-poor languages with little if any parallel text to train statistical or example- based MT systems. Examples of resource poor languages are Quechua and Mapudungun, which contrast with languages that have more eco- nomic, and therefore also electronic, resources, such as Spanish and English. Rule-based transfer MT systems are the only feasible solution for resource-poor scenarios. Developing and expanding such systems manu- ally can, however, prove very costly and time consuming. On the other hand, finding trained computational linguists with knowledge of re- source-poor languages is a real challenge. More- over, if the translation rules are written manu- ally, no matter how many rules there already are, coverage and accuracy can always be in- creased. If they are automatically learned, they might be either too general or too specific. In both cases, the translation rules can be refined to account for new data. The goal of our re- search is to generalize post-editing efforts in an effective way, by identifying and correcting rules semi-automatically in order to improve cover- age and overall translation quality. In this paper, we introduce a novel approach that proposes an MT module for automatically refining translation rules based on the feedback provided by bilingual speakers. There are two main challenges in this ap- proach. First, the elicitation of accurate correc- tion information from non-expert bilingual speakers. Second, the automatic refinement of existing translation rules, given a corrected and word-aligned translation pair, and information about the MT errors. The approach described in this paper auto- matically determines the appropriate rule re- finement operations that need to be applied to a grammar and a lexicon in order for the system to output the correct translation, as given by the native speaker. The resulting refinements and extensions can therefore apply not only to the translation instance corrected by the user, but also to other similar cases where the same error would be encountered. Font …

Content truncated to 500 words