ABSTRACT
Methodology-wise, example-based machine translation (EBMT) shares many important stages of processing with its co-paradigms, rule-based machine translation (RBMT) and statistical machine translation. In EBMT, a similarity score is given to phrasal fragments of the input sentence that match the text units in the database of examples. EBMT obtains matching phrasal fragments, while RBMT obtains a deep semantic graph or an intermediate representation like the dependency tree. EBMT and translation memory are often compared. Both use a repository of example translations. Current EBMT systems make use of statistical alignment during the analysis phase to discover matching candidates. EBMT sought to adapt existing parallel sentences to the new input to get the translation. Since case-based reasoning stressed learning by analogy, so did EBMT by introducing translating by analogy. Computing similarity between a pair of sentences is at the heart of EBMT.
