Verb Conjugation in Transformers Is Determined by Linear Encodings of Subject Number
This provides insight into linguistic feature representation for NLP researchers, though it is incremental as it builds on existing interpretability work.
The study tackled the problem of interpretability in Transformer models by demonstrating that BERT's verb conjugation ability relies on a linear encoding of subject number, which can be manipulated to affect conjugation accuracy.
Deep architectures such as Transformers are sometimes criticized for having uninterpretable "black-box" representations. We use causal intervention analysis to show that, in fact, some linguistic features are represented in a linear, interpretable format. Specifically, we show that BERT's ability to conjugate verbs relies on a linear encoding of subject number that can be manipulated with predictable effects on conjugation accuracy. This encoding is found in the subject position at the first layer and the verb position at the last layer, but distributed across positions at middle layers, particularly when there are multiple cues to subject number.