Relative Importance in Sentence Processing
This work addresses the problem of interpreting neural language models for researchers in NLP and cognitive science, though it is incremental as it builds on existing saliency and attention methods.
The study compared how humans and neural language models determine the relative importance of elements in English sentences, finding strong correlations between human processing patterns and saliency-based importance in models, but not with attention-based importance.
Determining the relative importance of the elements in a sentence is a key factor for effortless natural language understanding. For human language processing, we can approximate patterns of relative importance by measuring reading fixations using eye-tracking technology. In neural language models, gradient-based saliency methods indicate the relative importance of a token for the target objective. In this work, we compare patterns of relative importance in English language processing by humans and models and analyze the underlying linguistic patterns. We find that human processing patterns in English correlate strongly with saliency-based importance in language models and not with attention-based importance. Our results indicate that saliency could be a cognitively more plausible metric for interpreting neural language models. The code is available on GitHub: https://github.com/beinborn/relative_importance