Neuro-symbolic Natural Logic with Introspective Revision for Natural Language Inference

Neuro-symbolic Natural Logic with Introspective Revision for Natural Language Inference Yufei Feng∗ Ingenuity Labs Research Institute & ECE, Queen’s University, Canada feng.yufei@queensu.ca Xiaoyu Yang∗ Ingenuity Labs Research Institute & ECE, Queen’s University, Canada xiaoyu.yang@queensu.ca Xiaodan Zhu Ingenuity Labs Research Institute & ECE, Queen’s University, Canada xiaodan.zhu@queensu.ca Michael Greenspan Ingenuity Labs Research Institute & ECE, Queen’s University, Canada michael.greenspan@queensu.ca Abstract We introduce a neuro-symbolic natural logic framework based on reinforcement learning with introspective revision. The model sam- ples and rewards specific reasoning paths through policy gradient, in which the intro- spective revision algorithm modifies interme- diate symbolic reasoning steps to discover reward-earning operations as well as lever- ages external knowledge to alleviate spurious reasoning and training inefficiency. The framework is supported by properly designed local relation models to avoid input entangling, which helps ensure the interpretability of the proof paths. The proposed model has built-in interpretability and shows superior capability in monotonicity inference, systematic gener- alization, and interpretability, compared with previous models on the existing datasets. 1 Introduction In the past decade, deep neural networks have achieved impressive performance on modeling natural language inference (NLI) (Dagan et al., 2005; MacCartney, 2009; Bowman et al., 2015; Chen et al., 2017a,b), which aims to determine the entailment relations between a premise sen- tence and its corresponding hypothesis. Progress in NLI has greatly benefited from the models’ capabilities at approximating complex under- lying functions, discovering and utilizing rich (true and/or spurious) patterns, and exhibiting robustness to noise and ambiguity. However, the black-box models inherently lack interpretability, and still fail to capture many aspects of hu- man reasoning, including monotonicity inference (Yanaka et al., 2019a,b, 2020), systematic ∗Equal contribution. compositionality and generalization (Fodor and Pylyshyn, 1988; Aydede, 1997; Yanaka et al., 2020), and negation (Geiger et al., 2020), among others. A recent research trend has attempted to ad- vance the long-standing problem of bringing together the complementary strengths of neural networks and symbolic models (Garcez et al., 2015; Yang et al., 2017; Rockt¨aschel and Riedel, 2017; Evans and Grefenstette, 2018; Weber et al., 2019; De Raedt et al., 2019; Mao et al., 2018). Specifically for natural language, natural logic has long been studied to model reasoning in human language (Lakoff, 1970; van Benthem, 1988; Valencia, 1991; Van Benthem, 1995; Nairn et al., 2006; MacCartney and Manning, 2009; MacCartney and Manning, 2009; Icard, 2012; Angeli and Manning, 2014). However, the work of investigating the joint advantage of neu- ral networks and natural logic is sparse (Feng et al., 2020) (see Section 2 for more details) and understudied. In this paper, we present a neuro-symbolic framework that integrates natural logic with neu- ral networks for natural language inference. At the local level, we explore appropriate transformer networks to model the local relations between the constituents of a premise and hypothesis, in order to prevent attention from fully entangling the input, which otherwise can seriously impair the interpretability of proof paths built on lo- cal relations. We then construct natural logic programs and use reinforcement learning to re- …

Content truncated to 500 words