Can Neural Networks Understand Logical Entailment?
This work addresses the challenge of logical reasoning in AI, which is incremental as it builds on existing sequence-processing methods with a new model class.
The authors tackled the problem of measuring neural networks' ability to understand logical entailment by introducing a new dataset and comparing architectures, finding that PossibleWorldNets outperformed all benchmarks.
We introduce a new dataset of logical entailments for the purpose of measuring models' ability to capture and exploit the structure of logical expressions against an entailment prediction task. We use this task to compare a series of architectures which are ubiquitous in the sequence-processing literature, in addition to a new model class---PossibleWorldNets---which computes entailment as a "convolution over possible worlds". Results show that convolutional networks present the wrong inductive bias for this class of problems relative to LSTM RNNs, tree-structured neural networks outperform LSTM RNNs due to their enhanced ability to exploit the syntax of logic, and PossibleWorldNets outperform all benchmarks.