Resumen de A new score to characterise collocations: Log-r in comparison to mutual information
Itsuko Fujimura, Shigenobu Aoki
This paper proposes a new score named Log-r as a simple measure for calculating the strength of association between the constituent words of bigrams and argues that Log-r is more appropriate than Mutual Information for characterising collocation types. Arguments are based on the visualisation of one million English bigrams taken from a corpus of 1.1 billion words and of 0.4 million French bigrams taken from a corpus of 0.1 billion words. A three-dimensional analysis of each bigram will be made with its Log-r, its logarithmized frequency, and vocabulary level of its constituent words. Transparent typological study of collocations can be conducted using this procedure, which is only based on the frequency of words and bigrams, r .
