LOGICINFERENCE

LOGICINFERENCE是一个专为评估序列到序列模型逻辑推理能力而设计的新数据集。该数据集主要关注命题逻辑和一小部分一阶逻辑，数据内容包括半形式逻辑表示和自然语言表示。数据集的创建旨在解决机器学习模型在逻辑推理任务上的局限性，特别是Transformer和LSTM模型。LOGICINFERENCE包含多种任务类型，如语言到逻辑的转换、一步或多步推理，并要求模型展示推理链。数据集还特别设计了边缘案例，如矛盾前提，以测试模型的处理能力。此外，数据集提供了多种数据分割和配置，以支持更全面的评估。

LOGICINFERENCE is a novel dataset specifically designed for evaluating the logical reasoning capabilities of sequence-to-sequence models. This dataset primarily focuses on propositional logic and a small subset of first-order logic, with its content comprising semi-formal logical representations and natural language representations. The dataset was created to address the limitations of machine learning models on logical reasoning tasks, particularly Transformer and LSTM models. LOGICINFERENCE includes multiple task types, such as language-to-logic translation, single-step or multi-step reasoning, and requires models to demonstrate reasoning chains. The dataset also specially designs edge cases such as contradictory premises to test the processing capabilities of models. Additionally, the dataset provides multiple data splits and configurations to support more comprehensive evaluations.