基于逻辑推理的机器阅读理解综述

doi:10.3778/j.issn.1673-9418.2311033

摘要/Abstract

摘要： 机器阅读理解是自然语言处理领域中的核心任务之一，该任务目标是使机器能够理解自然语言文本，并正确回答关于文本内容的问题。随着自然语言处理相关方法和模型的发展，研究者们开始关注机器阅读理解中更具挑战性的推理型问题，这些问题通常要求模型不仅理解文本中的浅层信息，还要能够在逻辑上进行思考和推理，以回答更加复杂的问题。对基于逻辑推理的机器阅读理解相关的最新成果进行全面的归纳。介绍基于逻辑推理的机器阅读理解任务。介绍该任务的相关方法，并根据侧重点的不同将这些方法分成四类：基于符号神经网络的方法、基于图神经网络的方法、基于预训练的方法和基于大模型的微调策略。重点描述四类方法的代表性工作。在LogiQA和ReClor两个逻辑推理主流数据集上探讨方法的优缺点，并总结基于逻辑推理的机器阅读理解任务的未来研究方向。

关键词: 机器阅读理解, 逻辑推理, 智能问答

Abstract: Machine reading comprehension is one of the core tasks in the field of natural language processing, the goal of which is to enable machines to understand natural language text and correctly answer questions about the content of the text. With the development of methods and models related to natural language processing, researchers have begun to focus on the more challenging inferential problems in machine reading comprehension, which often require models to not only understand superficial information in text, but also to be able to think and reason logically to answer more complex questions. This paper gives a comprehensive summary of the latest achievements of machine reading comprehension based on logical reasoning. Firstly, the task of machine reading comprehension based on logical reasoning is introduced. Secondly, this paper introduces relevant methods for this task and classifies them into four categories based on different focuses: methods based on symbolic neural networks, methods based on graph neural networks, methods based on pre-training, and fine-tuning strategies based on large models, with a focus on describing the representative work of the four methods. Finally, the advantages and disadvantages of the methods are discussed on LogiQA and ReClor datasets, and the future research directions of machine reading comprehension task based on logical reasoning are summarized.

Key words: machine reading comprehension, logical reasoning, intelligent questions and answers

李晴, 李艳玲, 董杰, 葛凤培, 林民. 基于逻辑推理的机器阅读理解综述[J]. 计算机科学与探索, 2024, 18(8): 1998-2013.

LI Qing, LI Yanling, DONG Jie, GE Fengpei, LIN Min. Survey of Machine Reading Comprehension Based on Logical Reasoning[J]. Journal of Frontiers of Computer Science and Technology, 2024, 18(8): 1998-2013.

参考文献

[1] LEHNERT W G. The process of question answering: a computer simulation of cognition[M]. Bristol: Taylor & Francis, 2022.
[2] NAUN C C. Book review: introduction to modern information retrieval[J]. Library Resources & Technical Services, 2011, 55(4): 239-240.
[3] HURLEY P J. A concise introduction to logic[M]. Cengage Learning, 2014.
[4] QIU B, CHEN X, XU J, et al. A survey on neural machine reading comprehension[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1906.03824.
[5] STORKS S, GAO Q, CHAI J Y. Commonsense reasoning for natural language understanding: a survey of benchmarks, resources, and approaches[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1904.01172.
[6] 朱斯琪, 过弋, 王业相. 基于深度交互融合网络的多跳机器阅读理解[J]. 中文信息学报, 2022, 36(5): 67-75.
ZHU S Q, GUO Y, WANG Y X. Deep interactive fusion network for multi-hop reading comprehension[J]. Journal of Chinese Information Processing, 2022, 36(5): 67-75.
[7] LIU J, CUI L, LIU H, et al. LogiQA: a challenge dataset for machine reading comprehension with logical reasoning[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2007.08124.
[8] LIU H, LIU J, CUI L, et al. LogiQA 2.0—an improved dataset for logical reasoning in natural language understanding[J]. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 2023, 31: 2947-2962.
[9] YU W, JIANG Z, DONG Y, et al. ReClor: a reading compre-hension dataset requiring logical reasoning[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2002.04326.
[10] HUANG L, BRAS R L, BHAGAVATULA C, et al. Cosmos QA: machine reading comprehension with contextual commonsense reasoning[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1909.00277.
[11] YANG Z, QI P, ZHANG S, et al. HotpotQA: a dataset for diverse, explainable multi-hop question answering[EB/OL].[2023-09-23]. https://arxiv.org/abs/1809.09600.
[12] DUA D, WANG Y, DASIGI P, et al. DROP: a reading comprehension benchmark requiring discrete reasoning over paragraphs[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1903.00161.
[13] QUINE W V O. Philosophy of logic[M]. Harvard University Press, 1986.
[14] BHAGAVATULA C, BRAS R L, MALAVIYA C, et al. Abductive commonsense reasoning[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1908.05739.
[15] DEUTSCHER G. On the misuse of the notion of ‘abduction’in linguistics[J]. Journal of Linguistics, 2002, 38(3): 469-485.
[16] PSILLOS S. An explorer upon untrodden ground: Peirce on abduction[M]//Handbook of the History of Logic. Amsterdam: North-Holland, 2011: 117-151.
[17] JIN H, LUO Y, GAO C, et al. ComQA: question answering over knowledge base via semantic matching[J]. IEEE Access, 2019, 7: 75235-75246.
[18] GU Y, PAHUJA V, CHENG G, et al. Knowledge base question answering: a semantic parsing perspective[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2209.04994.
[19] 王小捷, 白子薇, 李可, 等. 机器阅读理解的研究进展[J]. 北京邮电大学学报, 2019, 42(6): 1-9.
WANG X J, BAI Z W, LI K, et al. Survey on machine reading comprehension[J]. Journal of Beijing University of Posts and Telecommunications, 2019, 42(6): 1-9.
[20] LIU S, ZHANG X, ZHANG S, et al. Neural machine reading comprehension: methods and trends[J]. Applied Sciences, 2019, 9(18): 3698.
[21] ZENG C, LI S, LI Q, et al. A survey on machine reading comprehension—tasks, evaluation metrics and benchmark datasets[J]. Applied Sciences, 2020, 10(21): 7640.
[22] 顾迎捷, 桂小林, 李德福, 等. 基于神经网络的机器阅读理解综述[J]. 软件学报, 2020, 31(7): 2095-2126.
GU Y J, GUI X L, LI D F, et al. Survey of machine reading comprehension based on neural network[J]. Journal of Software, 2020, 31(7): 2095-2126.
[23] VARGHESE N, PUNITHAVALLI M. A comprehensive survey on machine reading comprehension: models, benchmarked datasets, evaluation metrics, and trends[C]//Proceedings of the 2022 Congress on Intelligent Systems. Singapore: Springer, 2022: 1-15.
[24] 倪艺函, 兰艳艳, 庞亮, 等. 多跳式文本阅读理解方法综述[J]. 中文信息学报, 2022, 36(11): 1-19.
NI Y H, LAN Y Y, PANG L, et al. A survey of multi-hop reading comprehension for text[J]. Journal of Chinese Information Processing, 2022, 36(11): 1-19.
[25] HABERNAL I, WACHSMUTH H, GUREVYCH I, et al. The argument reasoning comprehension task: identification and reconstruction of implicit warrants[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1708.01425.
[26] SINHA K, SODHANI S, DONG J, et al. CLUTRR: a diagnostic benchmark for inductive reasoning from text[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1908.06177.
[27] SUN K, YU D, CHEN J, et al. Dream: a challenge data set and models for dialogue-based reading comprehension[J]. Transactions of the Association for Computational Linguistics, 2019, 7: 217-231.
[28] NIE Y, WILLIAMS A, DINAN E, et al. Adversarial NLI: a new benchmark for natural language understanding[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1910.14599.
[29] GURURANGAN S, SWAYAMDIPTA S, LEVY O, et al. Annotation artifacts in natural language inference data[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1803.02324.
[30] CUI L, WU Y, LIU S, et al. MuTual: a dataset for multi-turn dialogue reasoning[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2004.04494.
[31] HUANG H Y, CHOI E, YIH W. FlowQA: grasping flow in history for conversational machine comprehension[EB/OL].[2023-09-23]. https://arxiv.org/abs/1810.06683.
[32] SEO M, KEMBHAVI A, FARHADI A, et al. Bidirectional attention flow for machine comprehension[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1611.01603.
[33] GARCEZ A A, BADER S, BOWMAN H, et al. Neural-symbolic learning and reasoning: a survey and interpretation[J]. Neuro-Symbolic Artificial Intelligence: The State of the Art, 2022, 342(1): 327.
[34] CHEN X, LIANG C, YU A W, et al. Neural symbolic reader: scalable integration of distributed and symbolic representations for reading comprehension[C]//Proceedings of the 8th International Conference on Learning Representations, Addis Ababa, Apr 26-30, 2020.
[35] GARCEZ A A, BESOLD T R, DE RAEDT L, et al. Neural-symbolic learning and reasoning: contributions and challenges[C]//Proceedings of the 2015 AAAI Spring Symposia. Menlo Park: AAAI, 2015.
[36] DONG H, MAO J, LIN T, et al. Neural logic machines[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1904.11694.
[37] GUPTA N, LIN K, ROTH D, et al. Neural module networks for reasoning over text[EB/OL]. [2023-09-23]. https://arxiv.org/abs/1912.04971.
[38] WANG S, ZHONG W, TANG D, et al. Logic-driven context extension and data augmentation for logical reasoning of text[C]//Findings of the Association for Computational Linguistics: ACL 2022. Stroudsburg: ACL, 2022: 1619-1629.
[39] LI X, CHENG G, CHEN Z, et al. AdaLoGN: adaptive logic graph network for reasoning-based machine reading comprehension[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2203.08992.
[40] SCHLICHTKRULL M, KIPF T N, BLOEM P, et al. Modeling relational data with graph convolutional networks[C]//Pro-ceedings of the 15th International Conference on Semantic Web, Heraklion, Jun 3-7, 2018. Cham: Springer, 2018: 593-607.
[41] SUN Y, CHENG G, QU Y. Reading comprehension with graph-based temporal-casual reasoning[C]//Proceedings of the 27th International Conference on Computational Linguistics. Stroudsburg: ACL, 2018: 806-817.
[42] HUANG Y, FANG M, CAO Y, et al. DAGN: discourse-aware graph network for logical reasoning[EB/OL]. [2023-09-23].https://arxiv.org/abs/2103.14349.
[43] OUYANG S, ZHANG Z, ZHAO H. Fact-driven logical reasoning[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2105.10334.
[44] CHEN J, ZHANG Z, ZHAO H. Modeling hierarchical reasoning chains by linking discourse units and key phrases for reading comprehension[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2306.12069.
[45] XU F, LIU J, LIN Q, et al. Logiformer: a two-branch graph transformer network for interpretable logical reasoning[C]//Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval. New York: ACM, 2022: 1055-1065.
[46] JIAO F, GUO Y, SONG X, et al. MERIt: meta-path guided contrastive learning for logical reasoning[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2203.00357.
[47] SANYAL S, XU Y, WANG S, et al. APOLLO: a simple approach for adaptive pretraining of language models for logical reasoning[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2212.09282.
[48] PI X, ZHONG W, GAO Y, et al. LogiGAN: learning logical reasoning via adversarial pre-training[C]//Advances in Neural Information Processing Systems 35, New Orleans, Nov 28-Dec 9, 2022: 16290-16304.
[49] XU Z, YANG Z, CUI Y, et al. IDOL: indicator-oriented logic pre-training for logical reasoning[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2306.15273.
[50] LIU H, TENG Z, CUI L, et al. LogiCoT: logical chain-of-thought instruction-tuning data collection with GPT-4[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2305.12147.
[51] BAO Q, PENG A Y, DENG Z, et al. Contrastive learning with logic-driven data augmentation for logical reasoning over text[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2305.12599.
[52] LAWRENCE J. Introduction to neural networks[M]. Nevada: California Scientific Software, 1993.
[53] ZHOU J, CUI G, HU S, et al. Graph neural networks: a review of methods and applications[J]. AI Open, 2020, 1: 57-81.
[54] WU Z, PAN S, CHEN F, et al. A comprehensive survey on graph neural networks[J]. IEEE Transactions on Neural Networks and Learning Systems, 2020, 32(1): 4-24.
[55] ZHANG S, TONG H, XU J, et al. Graph convolutional networks: a comprehensive review[J]. Computational Social Networks, 2019, 6(1): 1-23.
[56] LI R, WANG S, ZHU F, et al. Adaptive graph convolutional neural networks[C]//Proceedings of the 2018 AAAI Conference on Artificial Intelligence. Menlo Park: AAAI, 2018: 3546-3553.
[57] LIU X, ZHANG F, HOU Z, et al. Self-supervised learning: generative or contrastive[J]. IEEE Transactions on Knowledge and Data Engineering, 2021, 35(1): 857-876.
[58] ZENG A, LIU X, DU Z, et al. GLM-130B: an open bilingual pre-trained model[EB/OL]. [2023-09-23]. https://arxiv.org/abs/2210.02414.
[59] BAI J, BAI S, CHU Y, et al. Qwen technical report[EB/OL].[2023-10-15]. https://arxiv.org/abs/2309.16609.
[60] ZHANG H, CHEN J, JIANG F, et al. HuatuoGPT, towards taming language model to be a doctor[EB/OL]. [2023-10-15]. https://arxiv.org/abs/2305.15075.
[61] DAN Y, LEI Z, GU Y, et al. EduChat: a large-scale language model-based chatbot system for intelligent education[EB/OL]. [2023-10-15]. https://arxiv.org/abs/2308.02773.
[62] HOWARD J, RUDER S. Universal language model fine-tuning for text classification[EB/OL]. [2023-10-15]. https://arxiv.org/abs/1801.06146.
[63] CETTO M, NIKLAUS C, FREITAS A, et al. Graphene: semantically-linked propositions in open information extraction[EB/OL]. [2023-10-15]. https://arxiv.org/abs/1807.11276.
[64] MANN W C, THOMPSON S A. Rhetorical structure theory: toward a functional theory of text organization[J]. Text-Interdisciplinary Journal for the Study of Discourse, 1988, 8(3): 243-281.
[65] PRASAD R, DINESH N, LEE A, et al. The Penn discourse treebank 2.0[C]//Proceedings of the 2008 International Conference on Language Resources and Evaluation, Marrakech, May 26-Jun 1, 2008.
[66] COHAN A, DERNONCOURT F, KIM D S, et al. A discourse-aware attention model for abstractive summarization of long documents[EB/OL]. [2023-10-15]. https://arxiv.org/abs/1804.05685.
[67] HONNIBAL M, MONTANI I. SpaCy2: natural language understanding with bloom embeddings, convolutional neural networks and incremental parsing[J]. To Appear, 2017, 7(1): 411-420.
[68] LEVI F W. Finite geometrical systems: six public lectures delivered in February[M]. Calcutta: University of Calcutta, 1942.
[69] CUI Y, CHEN Z, WEI S, et al. Attention-over-attention neural networks for reading comprehension[EB/OL]. [2023-10-15]. https://arxiv.org/abs/1607.04423.
[70] DHINGRA B, LIU H, YANG Z, et al. Gated-attention readers for text comprehension[EB/OL]. [2023-10-15]. https://arxiv.org/abs/1606.01549.
[71] BECK D, HAFFARI G, COHN T. Graph-to-sequence learning using gated graph neural networks[EB/OL]. [2023-10-15]. https://arxiv.org/abs/1806.09835.
[72] DEVLIN J, CHANG M W, LEE K, et al. BERT: pre-training of deep bidirectional transformers for language understanding[EB/OL]. [2023-10-15]. https://arxiv.org/abs/1810.04805.
[73] DAVID B, ROSEMARY K, DAVID W, et al. Reflection: turning experience into learning[M]. London: Routledge, 2013.
[74] DI STEFANO G, GINO F, PISANO G P, et al. Making experience count: the role of reflection in individual learning[M]. Cambridge: Harvard Business School, 2016.
[75] LIU H, NING R, TENG Z, et al. Evaluating the logical reasoning ability of ChatGPT and GPT-4[EB/OL]. [2023-10-15]. https://arxiv.org/abs/2304.03439.
[76] YE S, HWANG H, YANG S, et al. In-context instruction learning[EB/OL]. [2023-10-15]. https://arxiv.org/abs/2302.14691.
[77] CHANG E Y. Examining GPT-4: capabilities, implications and future directions[C]//Proceedings of the 10th International Conference on Computational Science and Computational Intelligence, 2023.
[78] TOUVRON H, LAVRIL T, IZACARD G, et al. LLaMA: open and efficient foundation language models[EB/OL]. [2023-10-15]. https://arxiv.org/abs/2302.13971.
[79] LIU P, YUAN W, FU J, et al. Pre-train, prompt, and predict: a systematic survey of prompting methods in natural language processing[J]. ACM Computing Surveys, 2023, 55(9): 1-35.
[80] WANG Z, ZHANG Z, LEE C Y, et al. Learning to prompt for continual learning[C]//Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Piscataway: IEEE, 2022: 139-149.
[81] KOJIMA T, GU S S, REID M, et al. Large language models are zero-shot reasoners[C]//Advances in Neural Information Processing Systems 35, New Orleans, Nov 28-Dec 9, 2022: 22199-22213.
[82] KIM S, JOO S J, KIM D, et al. The CoT collection: improving zero-shot and few-shot learning of language models via chain-of-thought fine-tuning[EB/OL]. [2023-10-15]. https://arxiv.org/abs/2305.14045.
[83] LIU Y, OTT M, GOYAL N, et al. RoBERTa: a robustly optimized BERT pretraining approach[EB/OL]. [2023-10-15].https://arxiv.org/abs/1907.11692.
[84] KUBLIK S, SABOO S. GPT-3[M]. Sebastopol: O??Reilly Media, Inc., 2022.
[85] YANG Z, DAI Z, YANG Y, et al. XLNet: generalized autoregressive pretraining for language understanding[C]//Advances in Neural Information Processing Systems 32, Vancouver,Dec 8-14, 2019: 5754-5764.
[86] PENG B, LI C, HE P, et al. Instruction tuning with GPT-4[EB/OL]. [2023-10-15]. https://arxiv.org/abs/2304.03277.
[87] QIAO S, OU Y, ZHANG N, et al. Reasoning with language model prompting: a survey[EB/OL]. [2023-10-15]. https://arxiv.org/abs/2212.09597.