Human-like Linguistic Biases in Neural Speech Models: Phonetic Categorization and Phonotactic Constraints in Wav2Vec2.0
This work addresses how neural speech models encode linguistic interactions, providing insights for computational linguistics and speech technology, though it is incremental in exploring model internals.
The study investigated whether Wav2Vec2 models exhibit human-like biases in phonotactic constraints by synthesizing ambiguous sounds between /l/ and /r/ in controlled contexts, finding that the models show a bias towards phonotactically admissible categories, with this effect emerging in early Transformer layers and amplified by ASR finetuning.
What do deep neural speech models know about phonology? Existing work has examined the encoding of individual linguistic units such as phonemes in these models. Here we investigate interactions between units. Inspired by classic experiments on human speech perception, we study how Wav2Vec2 resolves phonotactic constraints. We synthesize sounds on an acoustic continuum between /l/ and /r/ and embed them in controlled contexts where only /l/, only /r/, or neither occur in English. Like humans, Wav2Vec2 models show a bias towards the phonotactically admissable category in processing such ambiguous sounds. Using simple measures to analyze model internals on the level of individual stimuli, we find that this bias emerges in early layers of the model's Transformer module. This effect is amplified by ASR finetuning but also present in fully self-supervised models. Our approach demonstrates how controlled stimulus designs can help localize specific linguistic knowledge in neural speech models.