Abstract
Recently, Large Language Models (LLMs) like T5 [1], GPT-3.5/4 [2], LLama-2 [3], It5 [4], and Camoscio [5] have demonstrated impressive performance across various natural language processing tasks. Despite their success, these LLMs also face limitations and risks, such as lack of factuality [6], hallucinations [7], and poor transparency [8]. As a result, there is a growing demand for ”inherent explainability,” which refers to the ability of models to provide human-like, natural language explanations for their predictions. Many studies have thus focused on natural language explanations, and numerous datasets have been created for this purpose, primarily in English [9]. However, there is a notable gap for non English languages, including Italian. To fill this void, this paper introduces the ’e-RTE-3- it’ dataset, the first Italian dataset for natural language inference enriched with free-form, human-written explanations for the relationship between two sentences. Additionally, the dataset includes alternative labels and confidence scores from annotators to account for the variability in human judgments. This aspect of the annotation scheme enhances the ’e-RTE-3-it’ dataset, making it a valuable resource for exploring s