One Size Does Not Fit All: Finding the Optimal Subword Sizes for FastText Models across Languages

TL;DR

This paper investigates how optimizing subword sizes in fastText models across multiple languages improves word representation quality, proposing a simple n-gram coverage model that enhances performance with minimal tuning.

Contribution

It identifies optimal subword sizes for several languages and introduces a straightforward n-gram coverage model that predicts effective subword sizes, reducing the need for costly parameter tuning.

Findings

Optimized subword sizes improve word analogy accuracy by up to 14%.

The n-gram coverage model predicts near-optimal subword sizes within 1% accuracy.

Model improvements are consistent across multiple languages.

Abstract

Unsupervised representation learning of words from large multilingual corpora is useful for downstream tasks such as word sense disambiguation, semantic text similarity, and information retrieval. The representation precision of log-bilinear fastText models is mostly due to their use of subword information. In previous work, the optimization of fastText's subword sizes has not been fully explored, and non-English fastText models were trained using subword sizes optimized for English and German word analogy tasks. In our work, we find the optimal subword sizes on the English, German, Czech, Italian, Spanish, French, Hindi, Turkish, and Russian word analogy tasks. We then propose a simple n-gram coverage model and we show that it predicts better-than-default subword sizes on the Spanish, French, Hindi, Turkish, and Russian word analogy tasks. We show that the optimization of fastText's subword sizes matters and results in a 14% improvement on the Czech word analogy task. We also show that expensive parameter optimization can be replaced by a simple n-gram coverage model that consistently improves the accuracy of fastText models on the word analogy tasks by up to 3% compared to the default subword sizes, and that it is within 1% accuracy of the optimal subword sizes.