Automatic extraction of collocations from large corpora : alternative techniques for trigrams and higher grams.
Jean-Pierre Colson · DIAL (Catholic University of Leuven) · 2011
The automatic extraction of collocations from lare corpora has so far been restricted to binary structures (bigrams). This is mainly due to statistical issues, as it is not easy to extend the traditional scores (dice, log-likelihood, z-score, t-score, mutual information, etc.) to higher grams. There has also been criticism on the great variation in the results obtained by these scores. This paper proposes an alternative approach to the automatic extraction of collocations from corpora, that can be applied to higher grams and is based on the average proximity between the grams.