A Comparative Study of Embedding Models in Predicting the Compositionality of Multiword Expressions