Ballesteros and Croft (1998) pointed out that ‘‘the correct translations of query terms should co-occur in
target language documents and incorrect translations should tend not to co-occur.’’ Suppose that two
terms, ‘‘Mercury’’ and ‘‘planet,’’ are included in the source language (English) query. Although the term
‘‘Mercury’’ (or ‘‘mercury’’) has multiple senses, it is clear that ‘‘Mercury’’ is unrelated to mythology or
chemistry in the context of the given query due to an existence of the other term ‘‘planet.’’ Thus it is reasonably
expected that a correct translation equivalent of ‘‘Mercury’’ will tend to co-occur with a correct
equivalent of ‘‘planet’’ in the target document collection. In order to detect a set of correct translations
from the target collection, Ballesteros and Croft (1998) employed a variation of mutual information (MI)
computed from co-occurrence of a pair of two terms. According to the values for all possible pairs of
two translation equivalents, we may choose a correct combination of translations. In Ballesteros and Croft
(1998), this technique was actually applied to the disambiguation of phrasal translation (see below)