Subword-Delimited Downsampling for Better Character-Level Translation
This addresses the computational bottleneck for character-level machine translation models, offering a more efficient alternative to subword models, though it appears incremental as it builds on existing downsampling techniques.
This work tackles the problem of character-level models being computationally expensive for machine translation by introducing a subword-informed downsampling method, which outperforms existing downsampling methods and shows promising performance compared to subword models without sacrificing quality.
Subword-level models have been the dominant paradigm in NLP. However, character-level models have the benefit of seeing each character individually, providing the model with more detailed information that ultimately could lead to better models. Recent works have shown character-level models to be competitive with subword models, but costly in terms of time and computation. Character-level models with a downsampling component alleviate this, but at the cost of quality, particularly for machine translation. This work analyzes the problems of previous downsampling methods and introduces a novel downsampling method which is informed by subwords. This new downsampling method not only outperforms existing downsampling methods, showing that downsampling characters can be done without sacrificing quality, but also leads to promising performance compared to subword models for translation.