|
CMU implemented
an initial version of a corpus-driven decompounder for German,
based on cognates between the German and English (or dictionary
translations of English) halves of a parallel corpus. In
early tests, the system achieved 22% recall (of all German
compounds) at greater than 99% precision. As a side effect
of the decompounding process, a bilingual term lexicon from
German compound words to English phrases is generated. |
|