Can recursive neural tensor networks learn logical reasoning?

Abstract

Recursive neural network models and their accompanying vector representations for words have seen success in an array of increasingly semantically sophisticated tasks, but almost nothing is known about their ability to accurately capture the aspects of linguistic meaning that are necessary for interpretation or reasoning. To evaluate this, I train a recursive model on a new corpus of constructed examples of logical reasoning in short sentences, like the inference of "some animal walks" from "some dog walks" or "some cat walks," given that dogs and cats are animals. This model learns representations that generalize well to new types of reasoning pattern in all but a few cases, a result which is promising for the ability of learned representation models to capture logical reasoning.

Submitted for presentation at ICLR 2014. Source code and data: http://goo.gl/PSyF5u

References in corpus (2)

• Mathematical Foundations for a Compositional Distributional Model of Meaning
• Learning New Facts From Knowledge Bases With Neural Tensor Networks and Semantic Word Vectors

Cited by in corpus (11)

• Learning to Execute
• When Are Tree Structures Necessary for Deep Learning of Representations?
• Traversing Knowledge Graphs in Vector Space
• Learning Program Embeddings to Propagate Feedback on Student Code
• Probabilistic Reasoning via Deep Learning: Neural Association Models
• Dynamic Neural Program Embedding for Program Repair
• Learning to Discover Efficient Mathematical Identities
• Deep Learning for Ontology Reasoning
• Simpler Context-Dependent Logical Forms via Model Projections
• Posing Fair Generalization Tasks for Natural Language Inference
• The Uncanny Similarity of Recurrence and Depth