Deriving Contextualised Semantic Features from BERT (and Other Transformer Model) Embeddings
This work provides a tool for researchers to better understand the contextual semantic information captured by transformer models, addressing the opacity of these models for the NLP community.
This paper demonstrates a method to derive contextualized semantic features, specifically the 65 core semantic features proposed by Binder and colleagues, from BERT embeddings. This allows for the interpretation of semantic differences of words in context and provides insights into feature representation across BERT's layers.
Models based on the transformer architecture, such as BERT, have marked a crucial step forward in the field of Natural Language Processing. Importantly, they allow the creation of word embeddings that capture important semantic information about words in context. However, as single entities, these embeddings are difficult to interpret and the models used to create them have been described as opaque. Binder and colleagues proposed an intuitive embedding space where each dimension is based on one of 65 core semantic features. Unfortunately, the space only exists for a small dataset of 535 words, limiting its uses. Previous work (Utsumi, 2018, 2020, Turton, Vinson & Smith, 2020) has shown that Binder features can be derived from static embeddings and successfully extrapolated to a large new vocabulary. Taking the next step, this paper demonstrates that Binder features can be derived from the BERT embedding space. This provides contextualised Binder embeddings, which can aid in understanding semantic differences between words in context. It additionally provides insights into how semantic features are represented across the different layers of the BERT model.