Rule-based machine translation (RbMT) refers to a machine translation engine built on algorithms that analyze the syntax of the source language and uses rules to transfer the meaning to the target language by building a sentence. Contrast this with the processes of data searching and selecting on the basis of probabilities used for statistical machine translation.

The advantage of rule-based machine translation is that a sufficiently sophisticated translation engine can translate a wide range of texts without having been trained with a large number of examples, as in statistical machine translation.

The disadvantage is that it is necessary to build custom parsing software and dictionaries for each language pair, and that it is quite “brittle”.

Rule-based translation engines don’t deal very well with slang or metaphorical texts, for example. For this reason, rule-based translation has largely been replaced by statistical machine translation or hybrid systems, though it is useful for less common language pairs (where there are often not enough parallel texts to train a statistical machine translation engine).

See also