Comparing zero-shot self-explanations with human rationales in text classification

2024-10-04

AI Summary

This work investigates the quality of zero-shot self-explanations generated by instruction-tuned large language models (LLMs) for text classification, focusing on plausibility (human interpretability) and faithfulness (model alignment). We evaluate across English, Danish, and Italian sentiment classification, as well as cross-lingual forced labor risk detection, systematically comparing LLM-generated explanations against human-annotated rationales and XAI baselines—including Layer-wise Relevance Propagation (LRP). Our key contribution is the first empirical demonstration that zero-shot self-explanations substantially outperform LRP across multilingual and multitask settings: they achieve significantly higher plausibility while maintaining comparable faithfulness. Crucially, this improvement requires no additional training, fine-tuning, or architectural modification—ensuring lightweight deployment and broad applicability. The findings establish a novel paradigm for scalable, trustworthy model interpretation grounded in instruction-tuned LLMs’ inherent reasoning capabilities.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models
• Humans and AI: Explainable AI (XAI) for Human Understanding

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

Instruction-tuned LLMs are able to provide an explanation about their output to users by generating self-explanations. These do not require gradient computations or the application of possibly complex XAI methods. In this paper, we analyse whether this ability results in a good explanation. We evaluate self-explanations in the form of input rationales with respect to their plausibility to humans as well as their faithfulness to models. We study two text classification tasks: sentiment classification and forced labour detection, i.e., identifying pre-defined risk indicators of forced labour. In addition to English, we include Danish and Italian translations of the sentiment classification task and compare self-explanations to human annotations for all samples. To allow for direct comparisons, we also compute post-hoc feature attribution, i.e., layer-wise relevance propagation (LRP) and analyse 4 LLMs. We show that self-explanations align more closely with human annotations compared to LRP, while maintaining a comparable level of faithfulness. This finding suggests that self-explanations indeed provide good explanations for text classification.

Problem

• Evaluate self-explanations in text classification tasks
• Compare self-explanations with human annotations and LRP
• Analyze plausibility and faithfulness of self-explanations

Innovation

• Instruction-tuned LLMs generate self-explanations
• Self-explanations align with human annotations
• Post-hoc feature attribution compared with LRP