MICE: Mining Idioms with Contextual Embeddings

Idiomatic expressions can be problematic for natural language processing applications as their meaning cannot inferred from constituting words. A lack of successful methodological approaches and sufficiently large datasets prevents the development machine learning detecting idioms, especially that do not occur in training set. We present an approach called MICE uses contextual embeddings purpose. a new dataset multi-word with literal idiomatic meanings use it to train classifier based on two state-of-the-art word embeddings: ELMo BERT. show deep neural networks using both perform much better than existing are capable use, even were demonstrate cross-lingual transfer developed models analyze size required dataset.