Is Attention Interpretable?

Abstract

Attention mechanisms have recently boosted performance on a range of NLP tasks. Because attention layers explicitly weight input components' representations, it is also often assumed that attention can be used to identify information that models found important (e.g., specific contextualized word tokens). We test whether that assumption holds by manipulating attention weights in already-trained text classification models and analyzing the resulting differences in their predictions. While we observe some ways in which higher attention weights correlate with greater impact on model predictions, we also find many ways in which this does not hold, i.e., where gradient-based rankings of attention weights better predict their effects than their magnitudes. We conclude that while attention noisily predicts input components' overall importance to a model, it is by no means a fail-safe indicator. 1

Attention in Natural Language Processing

“…Whether attention may or may not be considered as a mean to explain neural networks is currently an open debate. Some recent studies [10], [11] suggest that attention cannot be considered a reliable mean to explain or even interpret neural networks. Nonetheless, other works [6]- [9] advocate the use of attention weights as an analytic tool.…”

“…It then becomes hard if not impossible to pinpoint the reasons behind the wrong output of a neural architecture. Interestingly, attention could provide a key to partially interpret and explain neural network behavior [5]- [9], even if it cannot be considered a reliable means of explanation [10], [11]. For instance, the weights computed by attention could point us to relevant information discarded by the neural network or to irrelevant elements of the input source that have been factored in and could explain a surprising output of the neural network.…”

BERTology Meets Biology: Interpreting Attention in Protein Language Models

“…Depending on the task and model architecture, attention may have more or less explanatory power for model predictions [35,51,57,71,79]. Visualization techniques have been used to convey the structure and properties of attention in Transformers [31,40,80,82].…”

Topic classification of electric vehicle consumer experiences with transformer-based deep learning

“…This indicates evidence of super-human classification on imbalanced corpora. As deep learning models have often been criticized for their black-box nature, we suggest technical enhancements that focus on model interpretability as future work, such as through the use of rationales, 45 influence functions, 46 or sequence tagging approaches 47 that can offer deeper insights on the models and the reasons for their predictions. This is an area of active research.…”