Improving Coherence and Consistency in Neural Sequence Models with Dual-System, Neuro-Symbolic Reasoning

Human reasoning can be understood as an interplay between two systems: the intuitive and associative (“System 1”) and the deliberative and logical (“System 2”). Neural sequence models—which have been increasingly successful at performing complex, structured tasks—exhibit the advantages and failure modes of System 1: they are fast and learn patterns from data, but are often inconsistent and incoherent. In this work, we seek a lightweight, training-free means of improving existing System 1-like sequence models by adding System 2-inspired logical reasoning. We explore several variations on this theme in which candidate generations from a neural sequence model are examined for logical consistency by a symbolic reasoning module, which can either accept or reject the generations. Our approach uses neural inference to mediate between the neural System 1 and the logical System 2. Results in robust story generation and grounded instruction-following show that this approach can increase the coherence and accuracy of neurally-based generations.

[1]  Percy Liang,et al.  Learning executable semantic parsers for natural language understanding , 2016, Commun. ACM.

[2]  Joshua B. Tenenbaum,et al.  Building machines that learn and think like people , 2016, Behavioral and Brain Sciences.

[3]  Pushmeet Kohli,et al.  RobustFill: Neural Program Learning under Noisy I/O , 2017, ICML.

[4]  Loizos Michael,et al.  Neural-Symbolic Integration: A Compositional Perspective , 2020, AAAI.

[5]  Jonathan Evans.  In two minds: dual-process accounts of reasoning , 2003, Trends in Cognitive Sciences.

[6]  Monica S. Lam,et al.  AutoQA: From Databases to Q&A Semantic Parsers with Only Synthetic Training Data , 2020, EMNLP.

[7]  J. Weston,et al.  Recipes for Safety in Open-domain Chatbots , 2020, ArXiv.

[8]  Yejin Choi,et al.  Learning to Write with Cooperative Discriminators , 2018, ACL.

[9]  A Benchmark for Systematic Generalization in Grounded Language Understanding , 2020, NeurIPS.

[10]  Artur S. d'Avila Garcez,et al.  Logic Tensor Networks: Deep Learning and Logical Reasoning from Data and Knowledge , 2016, NeSy@HLAI.

[11]  Armando Solar-Lezama,et al.  DreamCoder: growing generalizable, interpretable knowledge with wake–sleep Bayesian program learning , 2020, Philosophical Transactions of the Royal Society A.

[12]  Brenden M. Lake,et al.  Word meaning in minds and machines , 2020, Psychological review.

[13]  Armando Solar-Lezama,et al.  Learning Compositional Rules via Neural Program Synthesis , 2020, NeurIPS.

[14]  Adam Wierman,et al.  Thinking Fast and Slow , 2017, SIGMETRICS Perform. Evaluation Rev..

[15]  Chin-Hui Lee,et al.  A speech understanding system based on statistical representation of semantics , 1992, [Proceedings] ICASSP-92: 1992 IEEE International Conference on Acoustics, Speech, and Signal Processing.

[16]  Nikolaj Bjørner,et al.  Z3: An Efficient SMT Solver , 2008, TACAS.

[17]  Christina Heinze-Deml,et al.  Think before you act: A simple baseline for compositional generalization , 2020, ArXiv.

[18]  Jason Weston,et al.  Dialogue Natural Language Inference , 2018, ACL.

[19]  Chuang Gan,et al.  The Neuro-Symbolic Concept Learner: Interpreting Scenes Words and Sentences from Natural Supervision , 2019, ICLR.

[20]  Ilya Sutskever,et al.  Zero-Shot Text-to-Image Generation , 2021, ICML.

[21]  Kyunghyun Cho,et al.  Generative Language-Grounded Policy in Vision-and-Language Navigation with Bayes' Rule , 2020, ICLR.

[22]  Charles Blundell,et al.  Neural Production Systems , 2021, ArXiv.

[23]  Myle Ott,et al.  Residual Energy-Based Models for Text Generation , 2020, ICLR.

[24]  Frank Hutter,et al.  Decoupled Weight Decay Regularization , 2017, ICLR.

[25]  Leslie G. Valiant,et al.  A First Experimental Demonstration of Massive Knowledge Infusion , 2008, KR.

[26]  Dan Klein,et al.  Constrained Language Models Yield Few-Shot Semantic Parsers , 2021, EMNLP.

[27]  Y-Lan Boureau,et al.  Controlling Style in Generated Dialogue , 2020, ArXiv.

[28]  Mark O. Riedl,et al.  Event Representations for Automated Story Generation with Deep Neural Nets , 2017, AAAI.

[29]  Katherine D. Kinzler,et al.  Core knowledge. , 2007, Developmental science.

[30]  Mohit Bansal,et al.  I like fish, especially dolphins: Addressing Contradictions in Dialogue Modeling , 2020, ACL.

[31]  Jacob Andreas,et al.  Implicit Representations of Meaning in Neural Language Models , 2021, ACL.

[32]  Chen Liang,et al.  Neural Symbolic Machines: Learning Semantic Parsers on Freebase with Weak Supervision , 2016, ACL.

[33]  Omer Levy,et al.  RoBERTa: A Robustly Optimized BERT Pretraining Approach , 2019, ArXiv.

[34]  Dan Klein,et al.  Pragmatically Informative Text Generation , 2019, NAACL.

[35]  S. Frederick.  Journal of Economic Perspectives—Volume 19, Number 4—Fall 2005—Pages 25–42 Cognitive Reflection and Decision Making , 2022 .

[36]  Mark Chen,et al.  Language Models are Few-Shot Learners , 2020, NeurIPS.

[37]  Xifeng Yan,et al.  Cross-domain Semantic Parsing via Paraphrasing , 2017, EMNLP.

[38]  Hang Li,et al.  Coupling Distributed and Symbolic Execution for Natural Language Queries , 2016, ICML.

[39]  Jonathan Berant,et al.  Value-based Search in Execution Space for Mapping Instructions to Programs , 2018, NAACL.

[40]  Jürgen Schmidhuber,et al.  Learning to Reason with Third-Order Tensor Products , 2018, NeurIPS.

[41]  Xinyu Hua,et al.  PAIR: Planning and Iterative Refinement in Pre-trained Transformers for Long Text Generation , 2020, EMNLP.

[42]  Jacob Andreas,et al.  Unnatural Language Processing: Bridging the Gap Between Synthetic and Natural Language Data , 2020, ArXiv.

[43]  Omer Levy,et al.  BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension , 2019, ACL.

[44]  Jacob Andreas,et al.  Task-Oriented Dialogue as Dataflow Synthesis , 2020, Transactions of the Association for Computational Linguistics.

[45]  Jonathan Berant,et al.  Span-based Semantic Parsing for Compositional Generalization , 2020, ACL.

[46]  Lei Li,et al.  CGMH: Constrained Sentence Generation by Metropolis-Hastings Sampling , 2018, AAAI.

[47]  Yoshua Bengio,et al.  Inductive Biases for Deep Learning of Higher-Level Cognition , 2020, ArXiv.

[48]  Joelle Pineau,et al.  CLUTRR: A Diagnostic Benchmark for Inductive Reasoning from Text , 2019, EMNLP.

[49]  Xu Sun,et al.  A Skeleton-Based Model for Promoting Coherence Among Sentences in Narrative Story Generation , 2018, EMNLP.

[50]  Yejin Choi,et al.  NeuroLogic Decoding: (Un)supervised Neural Text Generation with Predicate Logic Constraints , 2020, NAACL.

[51]  Jason Weston,et al.  Towards AI-Complete Question Answering: A Set of Prerequisite Toy Tasks , 2015, ICLR.

[52]  Ilya Sutskever,et al.  Language Models are Unsupervised Multitask Learners , 2019 .