Tokenization with Factorized Subword Encoding
This addresses tokenization inefficiencies in language models, particularly for morphological tasks, but is incremental as it builds on existing subword methods.
The paper tackles the problem of simple and greedy subword tokenization in language models by proposing a novel tokenization method called the Factorizer, which factorizes subwords onto discrete triplets using a VQ-VAE model, and results show it is more appropriate and robust for morphological tasks than BPE across 7 diverse languages.
In recent years, language models have become increasingly larger and more complex. However, the input representations for these models continue to rely on simple and greedy subword tokenization methods. In this paper, we propose a novel tokenization method that factorizes subwords onto discrete triplets using a VQ-VAE model. The effectiveness of the proposed tokenization method, referred to as the Factorizer, is evaluated on language modeling and morpho-syntactic tasks for 7 diverse languages. Results indicate that this method is more appropriate and robust for morphological tasks than the commonly used byte-pair encoding (BPE) tokenization algorithm.