Numeracy for Language Models: Evaluating and Improving their Ability to Predict Numbers
This addresses the problem of numeracy in language models for technical domains like clinical and scientific documents, representing an incremental improvement with specific gains.
The paper tackled the problem of improving language models' ability to predict numbers by exploring strategies like memorization and digit-by-digit composition, and proposed a novel neural architecture using a continuous probability density function. The results showed improvements in perplexity by 2 and 4 orders of magnitude on clinical and scientific datasets, and reduced mean absolute percentage errors by 18% and 54% compared to the second best strategy.
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.