The Impact of Word Splitting on the Semantic Content of Contextualized Word Representations
This addresses a practical issue in NLP for researchers and practitioners using language models, but it is incremental as it focuses on evaluating existing methods.
The study investigated how splitting out-of-vocabulary words into subwords affects the quality of contextualized word representations, finding that these representations are often worse than those of in-vocabulary words but require careful interpretation.
When deriving contextualized word representations from language models, a decision needs to be made on how to obtain one for out-of-vocabulary (OOV) words that are segmented into subwords. What is the best way to represent these words with a single vector, and are these representations of worse quality than those of in-vocabulary words? We carry out an intrinsic evaluation of embeddings from different models on semantic similarity tasks involving OOV words. Our analysis reveals, among other interesting findings, that the quality of representations of words that are split is often, but not always, worse than that of the embeddings of known words. Their similarity values, however, must be interpreted with caution.