Decoding Layer Saliency in Language Transformers
This work addresses the challenge of interpreting saliency in transformer-based language models, which is less studied than in visual networks, providing a computationally efficient solution for classification tasks.
The paper tackled the problem of identifying textual saliency in language transformers for classification tasks by adapting gradient-based methods and evaluating layer semantic coherence, resulting in consistent improvements over other methods on multiple benchmark datasets.
In this paper, we introduce a strategy for identifying textual saliency in large-scale language models applied to classification tasks. In visual networks where saliency is more well-studied, saliency is naturally localized through the convolutional layers of the network; however, the same is not true in modern transformer-stack networks used to process natural language. We adapt gradient-based saliency methods for these networks, propose a method for evaluating the degree of semantic coherence of each layer, and demonstrate consistent improvement over numerous other methods for textual saliency on multiple benchmark classification datasets. Our approach requires no additional training or access to labelled data, and is comparatively very computationally efficient.