MANTa: Efficient Gradient-Based Tokenization for Robust End-to-End Language Modeling

TL;DR

MANTa introduces a differentiable, end-to-end trainable tokenizer that enhances robustness and speed in language models by combining the benefits of byte-level and subword tokenization.

Contribution

This work presents MANTa, a novel neural tokenizer that is trainable alongside language models, improving robustness and efficiency over traditional static subword tokenizers.

Findings

MANTa improves robustness to character perturbations and out-of-domain data.

It performs comparably to existing models on the GLUE benchmark.

MANTa is significantly faster than byte-level models.

Abstract

Static subword tokenization algorithms have been an essential component of recent works on language modeling. However, their static nature results in important flaws that degrade the models' downstream performance and robustness. In this work, we propose MANTa, a Module for Adaptive Neural TokenizAtion. MANTa is a differentiable tokenizer trained end-to-end with the language model. The resulting system offers a trade-off between the expressiveness of byte-level models and the speed of models trained using subword tokenization. In addition, our tokenizer is highly explainable since it produces an explicit segmentation of sequences into blocks. We evaluate our pre-trained model on several English datasets from different domains as well as on synthetic noise. We find that MANTa improves robustness to character perturbations and out-of-domain data. We then show that MANTa performs comparably to other models on the general-domain GLUE benchmark. Finally, we show that it is considerably faster than strictly byte-level models.