LiTEx: A Linguistic Taxonomy of Explanations for Understanding Within-Label Variation in Natural Language Inference

TL;DR

This paper introduces LITEX, a linguistic taxonomy for categorizing explanations in NLI, to understand within-label variation and improve explanation generation by aligning more closely with human reasoning.

Contribution

LITEX provides a systematic, linguistically-informed framework for analyzing and generating explanations in NLI, addressing within-label variation and enhancing explanation quality.

Findings

LITEX annotations are reliable and align with NLI labels.

Conditioning explanation generation on LITEX produces more human-like explanations.

The taxonomy helps understand diverse reasoning behind the same labels.

Abstract

There is increasing evidence of Human Label Variation (HLV) in Natural Language Inference (NLI), where annotators assign different labels to the same premise-hypothesis pair. However, within-label variation--cases where annotators agree on the same label but provide divergent reasoning--poses an additional and mostly overlooked challenge. Several NLI datasets contain highlighted words in the NLI item as explanations, but the same spans on the NLI item can be highlighted for different reasons, as evidenced by free-text explanations, which offer a window into annotators' reasoning. To systematically understand this problem and gain insight into the rationales behind NLI labels, we introduce LITEX, a linguistically-informed taxonomy for categorizing free-text explanations in English. Using this taxonomy, we annotate a subset of the e-SNLI dataset, validate the taxonomy's reliability, and analyze how it aligns with NLI labels, highlights, and explanations. We further assess the taxonomy's usefulness in explanation generation, demonstrating that conditioning generation on LITEX yields explanations that are linguistically closer to human explanations than those generated using only labels or highlights. Our approach thus not only captures within-label variation but also shows how taxonomy-guided generation for reasoning can bridge the gap between human and model explanations more effectively than existing strategies.