Exploring BERT's Sensitivity to Lexical Cues using Tests from Semantic Priming
This work addresses the problem of understanding contextual word prediction in NLP models for researchers, but it is incremental as it applies existing methods to analyze BERT.
The study investigated how BERT uses lexical cues in context by testing it with semantic priming stimuli, finding that BERT shows priming effects similar to humans, with the effect decreasing as context informativeness increases, and it becomes distracted by related primes in more informative contexts.
Models trained to estimate word probabilities in context have become ubiquitous in natural language processing. How do these models use lexical cues in context to inform their word probabilities? To answer this question, we present a case study analyzing the pre-trained BERT model with tests informed by semantic priming. Using English lexical stimuli that show priming in humans, we find that BERT too shows "priming," predicting a word with greater probability when the context includes a related word versus an unrelated one. This effect decreases as the amount of information provided by the context increases. Follow-up analysis shows BERT to be increasingly distracted by related prime words as context becomes more informative, assigning lower probabilities to related words. Our findings highlight the importance of considering contextual constraint effects when studying word prediction in these models, and highlight possible parallels with human processing.