Between lexical meaning and valency : towards a method for extracting equivalents based on a parallel corpus
Abstrakt
Some entries in translation dictionaries (even in those for closely related languages) offer multiple target equivalents with disparate meanings even when the meaning of the source word is not considered ambiguous within the source language. The translator's task of choosing the best fitting text unit is made even more difficult by sparse exemplification and missing or sketchy valency information.
Using a parallel corpus, we explore to what extent valency (seen in a broad sense both as a syntactic and semantic property) can help to cope with a few difficult cases of Czech psych verbs in relation to Polish and how the lexical caps can be filled. Next, we compare the manually compiled lists of translation equivalents in context, annotated and sorted by their valency properties, with results of an automatic method, using word-to-word alignments in a lemmatised parallel corpus to build a list of equivalent pairs, annotated by their frequency.
We conclude that valency is an important (though not omnipotent) discriminant for the choice of the target equivalent and that the list of translation pairs extracted automatically from a parallel corpus can be a useful supplement to a standard translation dictionary.