Numeracy for Language Models: Evaluating and Improving their Ability to Predict Numbers

TL;DR

This paper investigates how language models understand and generate numerals, introducing a novel neural architecture that significantly improves numeral prediction accuracy in scientific and clinical texts.

Contribution

The paper proposes a new neural model using a continuous probability density function for numerals, enhancing numeral handling in language models beyond existing methods.

Findings

Hierarchical models improve numeral perplexity by 2-4 orders of magnitude.

The continuous density model reduces mean absolute percentage errors by 18% and 54%.

Combining strategies yields further perplexity improvements.

Abstract

Numeracy is the ability to understand and work with numbers. It is a necessary skill for composing and understanding documents in clinical, scientific, and other technical domains. In this paper, we explore different strategies for modelling numerals with language models, such as memorisation and digit-by-digit composition, and propose a novel neural architecture that uses a continuous probability density function to model numerals from an open vocabulary. Our evaluation on clinical and scientific datasets shows that using hierarchical models to distinguish numerals from words improves a perplexity metric on the subset of numerals by 2 and 4 orders of magnitude, respectively, over non-hierarchical models. A combination of strategies can further improve perplexity. Our continuous probability density function model reduces mean absolute percentage errors by 18% and 54% in comparison to the second best strategy for each dataset, respectively.