Extraction automatique de collocations : Peut-on étendre le test exact de Fisher à des séquences de plus de 2 mots?

In textual statistics, as in natural language processing and corpus linguistics, the study of sequences of contiguous words that occur together more often than by chance is a major topic of interest. In the case of pairs of words, Fisher’s exact test is becoming the reference index to identify them. The objective of this study is to propose a generalization of this index to the analysis of trigrams and longer sequences using a Monte-Carlo procedure. The results of an initial evaluation suggest that this approach could complement other indices, but also that it has a major drawback: a large number of trigrams get a maximum score of collocation.