Investigating semantic subspaces of Transformer sentence embeddings through linear structural probing
This addresses the problem of understanding sentence-level linguistic encoding in Transformers for the NLP community, but is incremental as it extends existing probing methods.
The paper investigates what linguistic information is encoded in different layers of Transformer language models by applying semantic structural probing to study sentence-level representations across model families and sizes. They found that model families differ substantially in performance and layer dynamics, but results are largely model-size invariant.
The question of what kinds of linguistic information are encoded in different layers of Transformer-based language models is of considerable interest for the NLP community. Existing work, however, has overwhelmingly focused on word-level representations and encoder-only language models with the masked-token training objective. In this paper, we present experiments with semantic structural probing, a method for studying sentence-level representations via finding a subspace of the embedding space that provides suitable task-specific pairwise distances between data-points. We apply our method to language models from different families (encoder-only, decoder-only, encoder-decoder) and of different sizes in the context of two tasks, semantic textual similarity and natural-language inference. We find that model families differ substantially in their performance and layer dynamics, but that the results are largely model-size invariant.