B-cos LM: Efficiently Transforming Pre-trained Language Models for Improved Explainability

TL;DR

This paper introduces B-cos LMs, a method to transform pre-trained language models into explainable models with improved faithfulness and interpretability, while maintaining performance, through B-cos conversion and fine-tuning.

Contribution

The work extends B-cos networks to NLP by transforming pre-trained language models into B-cos LMs, enhancing explainability without sacrificing task accuracy.

Findings

B-cos LMs produce more faithful explanations than post-hoc methods.

B-cos LMs maintain comparable task performance to traditional fine-tuning.

Transforming decoder-only models to B-cos LMs is feasible for generation tasks.

Abstract

Post-hoc explanation methods for black-box models often struggle with faithfulness and human interpretability due to the lack of explainability in current neural architectures. Meanwhile, B-cos networks have been introduced to improve model explainability by proposing an architecture that removes bias terms and promotes input-weight alignment. Although B-cos networks have shown success in building explainable systems, their application has so far been limited to computer vision models and their associated training pipelines. In this work, we introduce B-cos LMs, i.e., B-cos Language Models (LMs) empowered for natural language processing (NLP) tasks. Our approach directly transforms pre-trained language models into B-cos LMs by combining B-cos conversion and task fine-tuning, improving efficiency compared to previous methods. Automatic and human evaluation results demonstrate that B-cos LMs produce more faithful and human interpretable explanations than post-hoc methods, while maintaining task performance comparable to conventional fine-tuning. Our in-depth analysis explores how B-cos LMs differ from conventionally fine-tuned models in their learning processes and explanation patterns. Finally, we present a first exploration of transforming decoder-only models to B-cos LMs for generation tasks. Our code is available at https://github.com/Ewanwong/bcos_lm.