Attention Interpretability Across NLP Tasks
This work addresses the confusion in the NLP community about attention interpretability, offering systematic insights for researchers and practitioners, though it is incremental in clarifying existing debates.
The paper tackles the contradictory viewpoints on the interpretability of attention weights in neural networks by providing a comprehensive explanation that justifies when attention is interpretable and when it is not, validated through experiments on diverse NLP tasks and manual evaluation.
The attention layer in a neural network model provides insights into the model's reasoning behind its prediction, which are usually criticized for being opaque. Recently, seemingly contradictory viewpoints have emerged about the interpretability of attention weights (Jain & Wallace, 2019; Vig & Belinkov, 2019). Amid such confusion arises the need to understand attention mechanism more systematically. In this work, we attempt to fill this gap by giving a comprehensive explanation which justifies both kinds of observations (i.e., when is attention interpretable and when it is not). Through a series of experiments on diverse NLP tasks, we validate our observations and reinforce our claim of interpretability of attention through manual evaluation.