Self-Rationalization in the Wild: A Large Scale Out-of-Distribution Evaluation on NLI-related tasks
This work addresses the problem of generating free-text explanations for NLP tasks like NLI, fact-checking, and hallucination detection when annotated data is limited, offering practical insights for researchers and practitioners, though it is incremental in applying existing methods to new evaluation scenarios.
The paper tackles the challenge of training models for explainable predictions when explanation data is scarce by investigating self-rationalization using existing datasets and evaluating out-of-distribution (OOD) performance. It finds that few annotated examples effectively adapt models for OOD explanation generation, with fine-tuning data source having a larger impact than sample selection strategies, and models with higher label prediction accuracy produce better explanations as measured by Acceptability scores.
Free-text explanations are expressive and easy to understand, but many datasets lack annotated explanation data, making it challenging to train models for explainable predictions. To address this, we investigate how to use existing explanation datasets for self-rationalization and evaluate models' out-of-distribution (OOD) performance. We fine-tune T5-Large and OLMo-7B models and assess the impact of fine-tuning data quality, the number of fine-tuning samples, and few-shot selection methods. The models are evaluated on 19 diverse OOD datasets across three tasks: natural language inference (NLI), fact-checking, and hallucination detection in abstractive summarization. For the generated explanation evaluation, we conduct a human study on 13 selected models and study its correlation with the Acceptability score (T5-11B) and three other LLM-based reference-free metrics. Human evaluation shows that the Acceptability score correlates most strongly with human judgments, demonstrating its effectiveness in evaluating free-text explanations. Our findings reveal: 1) few annotated examples effectively adapt models for OOD explanation generation; 2) compared to sample selection strategies, fine-tuning data source has a larger impact on OOD performance; and 3) models with higher label prediction accuracy tend to produce better explanations, as reflected by higher Acceptability scores.