Just Add Functions: A Neural-Symbolic Language Model

TL;DR

This paper introduces a neural-symbolic language model that incorporates simple functions to encode deterministic relationships, improving language modeling especially for rare tokens and specific relationships like numbers and geography.

Contribution

It proposes a novel hierarchical neural-symbolic architecture that explicitly encodes symbolic relationships into neural language models, enhancing their inductive bias and performance.

Findings

Significant perplexity reduction on small-corpus language modeling.

Performance improvements persist for rare tokens in larger corpora.

The approach is simple, general, and applicable to various word classes.

Abstract

Neural network language models (NNLMs) have achieved ever-improving accuracy due to more sophisticated architectures and increasing amounts of training data. However, the inductive bias of these models (formed by the distributional hypothesis of language), while ideally suited to modeling most running text, results in key limitations for today's models. In particular, the models often struggle to learn certain spatial, temporal, or quantitative relationships, which are commonplace in text and are second-nature for human readers. Yet, in many cases, these relationships can be encoded with simple mathematical or logical expressions. How can we augment today's neural models with such encodings? In this paper, we propose a general methodology to enhance the inductive bias of NNLMs by incorporating simple functions into a neural architecture to form a hierarchical neural-symbolic language model (NSLM). These functions explicitly encode symbolic deterministic relationships to form probability distributions over words. We explore the effectiveness of this approach on numbers and geographic locations, and show that NSLMs significantly reduce perplexity in small-corpus language modeling, and that the performance improvement persists for rare tokens even on much larger corpora. The approach is simple and general, and we discuss how it can be applied to other word classes beyond numbers and geography.