About the Tutorial

Understanding natural language interactions in realistic settings requires models that can deal with noisy textual inputs, reason about the dependencies between different textual elements and leverage the dependencies between textual content and the context from which it emerges. Neural-symbolic approaches, which combine highly expressive neural representations with symbolic reasoning capabilities, directly fit these settings. Despite that fact, most current NLP work focuses on learning neural representations.

In this tutorial, we will motivate neural symbolic modeling as a general approach for a wide range of natural language scenarios. We will review several recently proposed approaches designed around different NLP domains, such as knowledge-base completion, quantitative reasoning, question-answering, relation extraction, and grounding text in images. We will propose a general formulation for neuro-symbolic modeling and discuss the key research challenges and opportunties for NLP tasks that can drive future research directions. Finally, we will provide an interactive hands-on demonstration showing how to model a complex natural language domain using a declarative neural-symbolic framework.


Introduction Symbolic vs. Distributed Representations, Representing Context and Structure, Challenges and Opportunities, Tutorial Goals
Technical Overview Statistical Relational Learning, Distributed Representations for Relational Data, Neural-Symbolic Frameworks, Recent Neural-Symbolic Techniques for NLP Domains
Neural-Symbolic Modeling for Natural Language Discourse Parsig the Landscape of Opinions and Perspectives, Declarative Modeling, Learning and Inference with Deep Relational Learning, Case Study: Understanding Debate Networks
Symbolic Explanations for Discourse Analysis Emotions, Intent and Motivation, Character-Driven Narratives, Decoding Political Messaging, Political Issue Stance and Framing, Moral Foundations, Modeling Social and Behavioral Information, Morality Frames
Demo Modeling Morality Frames with Deep Relational Learning