Probing for Semantic Classes: Diagnosing the Meaning Content of Word Embeddings
This work addresses the challenge of evaluating word embeddings for semantic content, which is important for NLP researchers and practitioners, but it is incremental as it builds on existing diagnostic methods with a new dataset.
The paper tackled the problem of diagnosing how well word embeddings represent different word meanings, by creating a large dataset from Wikipedia annotations and using it to probe embeddings for semantic classes. The results showed that frequent senses are well-represented in single-vector embeddings, a classifier can predict single- vs. multi-sense words accurately, and rare senses' poor representation does not harm NLP applications relying on frequent senses.
Word embeddings typically represent different meanings of a word in a single conflated vector. Empirical analysis of embeddings of ambiguous words is currently limited by the small size of manually annotated resources and by the fact that word senses are treated as unrelated individual concepts. We present a large dataset based on manual Wikipedia annotations and word senses, where word senses from different words are related by semantic classes. This is the basis for novel diagnostic tests for an embedding's content: we probe word embeddings for semantic classes and analyze the embedding space by classifying embeddings into semantic classes. Our main findings are: (i) Information about a sense is generally represented well in a single-vector embedding - if the sense is frequent. (ii) A classifier can accurately predict whether a word is single-sense or multi-sense, based only on its embedding. (iii) Although rare senses are not well represented in single-vector embeddings, this does not have negative impact on an NLP application whose performance depends on frequent senses.