Adaptation-guided retrieval in EBMT: A case-based approach to machine translation

Abstract

In this paper we describe a methodological analysis of EBMT (Example-Based Machine Translation) based on a CBR (Case-Based Reasoning) perspective. This analysis focuses on adaptation. We argue that, just as in CBR, the overall power of an EBMT system is its ability to adapt examples retrieved to suit the new problem translation. Here we describe a technique whereby reusability is a function of the abstract "adaptability" information stored in the cases. This information is exploited during both the adaptation and retrieval stages.

EXTRA: a system for example-based translation assistance

“…Representations of text fragments are distinguished by the structure(s) employed, such as sequences or trees, and the amount and the kind of linguistic processing performed (Collins and Cunningham 1996). For instance the approach in Brown (1996) and Leplus et al (2004) stores examples as strings of words together with some alignment and information on equivalence classes (numbers, weekdays, etc.).…”

“…Further, in word-based measures, the exploitation of the information provided by the order of the words can be fundamental: word-order-sensitive approaches are demonstrated generally to outperform bag-of-words methods (Baldwin and Tanaka 2000). For instance, several papers (Collins and Cunningham 1996;Cranias et al 1994;Planas and Furuse 2000) report advanced hybrid word-and structure-based matching techniques on the TM. In particular, Collins and Cunningham (1996) and Cranias et al (1994) perform exact matching whereas Planas and Furuse (2000) adopt the edit-distance metric limiting themselves to deletions and equalities.…”

“…For instance the approach in Brown (1996) and Leplus et al (2004) stores examples as strings of words together with some alignment and information on equivalence classes (numbers, weekdays, etc.). Other approaches (Cranias et al 1994;Collins and Cunningham 1996;Planas and Furuse 2000) are more language-and knowledge-dependent and perform some text operations (e.g. POS tagging and stemming) on the sentences in order to store more structural information about them.…”

“…Such low flexibility is a big issue: consider, for instance, the frequent cases in which a translator decides to merge two sentences into a single one, or when the TM contains no whole suggestions. In Collins and Cunningham (1996) and Simard and Langlais (2001) a different approach is taken, namely, the query is first decomposed into "chunks", then the matching function makes a collection of matched chunk material. However such a subdivision into chunks is entirely static, so the approach still lacks the flexibility needed by translators.…”

“…In Brown (1996) on the other hand a rough ranking mechanism based on the freshness of the TM suggestion is proposed, while in Collins and Cunningham (1996) a more subtle ranking mechanism is exploited, taking advantage of the defined similarity measure and of adaptability scores between the source and target language of a particular translation suggestion.…”