Introducing example-based machine translation
In the era of RBMT systems, it became apparent that a new paradigm in MT was necessary. The reliance on linguistic rules presents many shortcomings. As we saw previously, using a corpus of already-translated examples could serve as a model to base the translation task on. This is the basic idea behind example-based machine translation (EBMT) systems; keep track of well-translated fragments and use this information to facilitate the translation of new sentences. Humans often process short sentences this way; first, they split the source into smaller fragments, then translate the pieces by analogy into previous examples, and, finally, recombine those translations into the target sentence. Deep linguistic analysis is not necessary, and the more examples that are available, the more the translation accuracy improves. Figure 6.14 shows an example:
Figure 6.14 – Using existing translated fragments in MT
The primary...