We propose a method for automatic term extraction based on a statistical measure that ranks term candidates according to their semantic relevance to a specialised domain. As a measure of relevance we use term co-occurrence, defined as the repeated instantiation of two terms in the same sentences, in indifferent order and at variable distances.
In this way, term candidates are ranked higher if they show a tendency to co-occur with a selected group of other units, as opposed to those showing more uniform distributions. No external resources are needed for the application of the method, but performance improves when provided with a pre-existing term list.
We present results of the application of this method to a Spanish-English Linguistics corpus, and the evaluation compares favourably with a standard method based on reference corpora.