Neural Grapheme-to-Phoneme Conversion with Pre-trained Grapheme Models
This addresses the challenge of G2P conversion for low-resource languages, though it is incremental as it adapts existing pre-training methods to a specific domain.
The paper tackles the problem of grapheme-to-phoneme conversion for languages with limited pronunciation dictionaries by proposing a pre-trained grapheme model (GBERT) and integrating it into Transformer-based models, achieving effective results on multiple languages under medium- and low-resource conditions.
Neural network models have achieved state-of-the-art performance on grapheme-to-phoneme (G2P) conversion. However, their performance relies on large-scale pronunciation dictionaries, which may not be available for a lot of languages. Inspired by the success of the pre-trained language model BERT, this paper proposes a pre-trained grapheme model called grapheme BERT (GBERT), which is built by self-supervised training on a large, language-specific word list with only grapheme information. Furthermore, two approaches are developed to incorporate GBERT into the state-of-the-art Transformer-based G2P model, i.e., fine-tuning GBERT or fusing GBERT into the Transformer model by attention. Experimental results on the Dutch, Serbo-Croatian, Bulgarian and Korean datasets of the SIGMORPHON 2021 G2P task confirm the effectiveness of our GBERT-based G2P models under both medium-resource and low-resource data conditions.