Work

Ecologically Valid Explanations for Label Variation in NLI

Year: 2023

Type: conference-paper

Abstract: Human label variation exists in many natural language processing (NLP) tasks, including natural language inference (NLI).To gain direct evidence of how NLI label variation arises, we build LIVENLI, an English dataset of 1,415 ecologically valid explanations (annotators explain the NLI labels they chose) for 122 MNLI (Williams et al., 2018) items (at least 10 explanations per item).The LIVENLI explanations confirm that people can systematically vary on their interpretation and highlight within-label variation: annotators sometimes choose the same label for different reasons.This suggests that explanations are crucial for navigating label interpretations in general.We few-shot prompt language models (LMs) to generate explanations but the results are inconsistent: the models sometimes produce valid and informative explanations, but they also generate implausible ones that do not support the label, highlighting directions for improvement.

Source: Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)

DOI: 10.18653/v1/2023.findings-emnlp.712

Authors: Nan-Jiang Jiang, Chenhao Tan, Marie-Catherine de Marneffe

Institutions: University of Chicago, The Ohio State University, UCLouvain

Language: English

Topic: Natural Language Processing Techniques

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: hybrid

Funders: National Science Foundation, University of Chicago, Fonds De La Recherche Scientifique - FNRS

Award: CAREER: Drawing inferences for human-like language understanding