Improving Word Translation Disambiguation by Capturing Multiword Expressions with Dictionaries
Lars Bungum, Björn Gambäck, André Lynum, Erwin Marsi · 2013
The paper describes a method for identifying and translating multiword expressions using a bi-directional dictionary. While a dictionarybased approach suffers from limited recall, precision is high; hence it is best employed alongside an approach with complementing properties, such as an n-gram language model. We evaluate the method on data from the English-German translation part of the crosslingual word sense disambiguation task in the 2010 semantic evaluation exercise (SemEval). The output of a baseline disambiguation system based on n-grams was substantially improved by matching the target words and their immediate contexts against compound and collocational words in a dictionary. 1