Human Guided Exploitation of Interpretable Attention Patterns in Summarization and Topic Segmentation

TL;DR

This paper presents a human-in-the-loop approach to identify and inject interpretable attention patterns into transformer models, improving their accuracy and efficiency in summarization and topic segmentation tasks.

Contribution

It introduces a novel pipeline combining pattern discovery and injection into models, enhancing interpretability and performance in specific NLP tasks.

Findings

Injected patterns improve model accuracy

Patterns lead to more efficient models

Case studies validate approach effectiveness

Abstract

The multi-head self-attention mechanism of the transformer model has been thoroughly investigated recently. In one vein of study, researchers are interested in understanding why and how transformers work. In another vein, researchers propose new attention augmentation methods to make transformers more accurate, efficient and interpretable. In this paper, we combine these two lines of research in a human-in-the-loop pipeline to first discover important task-specific attention patterns. Then those patterns are injected, not only to smaller models, but also to the original model. The benefits of our pipeline and discovered patterns are demonstrated in two case studies with extractive summarization and topic segmentation. After discovering interpretable patterns in BERT-based models fine-tuned for the two downstream tasks, experiments indicate that when we inject the patterns into attention heads, the models show considerable improvements in accuracy and efficiency.