The Coefficient of Synonymy

The measurement of synonymy between words is an essential function required by numerous NLP tasks. However, similarity measures often give word similarity values that are very difficult to interpret and compare when applied to popular word embeddings. We introduce a coefficient of synonymy C s , which uses a new method based on word distance and the density of words around those being compared for measuring synonymy. This method provides consistent, comparable values indicating the level of synonymy between words. We compare measurements of C s with cosine using four off-the-shelf word embeddings created by Word2Vec and GloVe and two 350,000 word pair datasets.