{"title":"Are Large Language Models Really Good Logical Reasoners? A Comprehensive Evaluation and Beyond","authors":"Fangzhi Xu;Qika Lin;Jiawei Han;Tianzhe Zhao;Jun Liu;Erik Cambria","doi":"10.1109/TKDE.2025.3536008","DOIUrl":null,"url":null,"abstract":"Logical reasoning consistently plays a fundamental and significant role in the domains of knowledge engineering and artificial intelligence. Recently, Large Language Models (LLMs) have emerged as a noteworthy innovation in natural language processing (NLP). However, the question of whether LLMs can effectively address the task of logical reasoning, which requires gradual cognitive inference similar to human intelligence, remains unanswered. To this end, we aim to bridge this gap and provide comprehensive evaluations in this paper. First, to offer systematic evaluations, we select fifteen typical logical reasoning datasets and organize them into deductive, inductive, abductive and mixed-form reasoning settings. Considering the comprehensiveness of evaluations, we include 3 early-era representative LLMs and 4 trending LLMs. Second, different from previous evaluations relying only on simple metrics (e.g., <italic>accuracy</i>), we propose fine-level evaluations in objective and subjective manners, covering both answers and explanations, including <italic>answer correctness</i>, <italic>explain correctness</i>, <italic>explain completeness</i> and <italic>explain redundancy</i>. Additionally, to uncover the logical flaws of LLMs, problematic cases will be attributed to five error types from two dimensions, i.e., <italic>evidence selection process</i> and <italic>reasoning process</i>. Third, to avoid the influences of knowledge bias and concentrate purely on benchmarking the logical reasoning capability of LLMs, we propose a new dataset with neutral content. Based on the in-depth evaluations, this paper finally forms a general evaluation scheme of logical reasoning capability from six dimensions (i.e., <italic>Correct</i>, <italic>Rigorous</i>, <italic>Self-aware</i>, <italic>Active</i>, <italic>Oriented</i> and <italic>No hallucination</i>). It reflects the pros and cons of LLMs and gives guiding directions for future works.","PeriodicalId":13496,"journal":{"name":"IEEE Transactions on Knowledge and Data Engineering","volume":"37 4","pages":"1620-1634"},"PeriodicalIF":8.9000,"publicationDate":"2025-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE Transactions on Knowledge and Data Engineering","FirstCategoryId":"94","ListUrlMain":"https://ieeexplore.ieee.org/document/10870148/","RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
引用次数: 0
Abstract
Logical reasoning consistently plays a fundamental and significant role in the domains of knowledge engineering and artificial intelligence. Recently, Large Language Models (LLMs) have emerged as a noteworthy innovation in natural language processing (NLP). However, the question of whether LLMs can effectively address the task of logical reasoning, which requires gradual cognitive inference similar to human intelligence, remains unanswered. To this end, we aim to bridge this gap and provide comprehensive evaluations in this paper. First, to offer systematic evaluations, we select fifteen typical logical reasoning datasets and organize them into deductive, inductive, abductive and mixed-form reasoning settings. Considering the comprehensiveness of evaluations, we include 3 early-era representative LLMs and 4 trending LLMs. Second, different from previous evaluations relying only on simple metrics (e.g., accuracy), we propose fine-level evaluations in objective and subjective manners, covering both answers and explanations, including answer correctness, explain correctness, explain completeness and explain redundancy. Additionally, to uncover the logical flaws of LLMs, problematic cases will be attributed to five error types from two dimensions, i.e., evidence selection process and reasoning process. Third, to avoid the influences of knowledge bias and concentrate purely on benchmarking the logical reasoning capability of LLMs, we propose a new dataset with neutral content. Based on the in-depth evaluations, this paper finally forms a general evaluation scheme of logical reasoning capability from six dimensions (i.e., Correct, Rigorous, Self-aware, Active, Oriented and No hallucination). It reflects the pros and cons of LLMs and gives guiding directions for future works.
期刊介绍:
The IEEE Transactions on Knowledge and Data Engineering encompasses knowledge and data engineering aspects within computer science, artificial intelligence, electrical engineering, computer engineering, and related fields. It provides an interdisciplinary platform for disseminating new developments in knowledge and data engineering and explores the practicality of these concepts in both hardware and software. Specific areas covered include knowledge-based and expert systems, AI techniques for knowledge and data management, tools, and methodologies, distributed processing, real-time systems, architectures, data management practices, database design, query languages, security, fault tolerance, statistical databases, algorithms, performance evaluation, and applications.