Learning to pronounce as measuring cross lingual joint orthography-phonology complexity

Recent work has demonstrated that machine learning models allow us to compare languages by showing how hard each language might be to learn under specific tasks. Following this line of investigation, we investigate what makes a language "hard to pronounce" by modelling the task of grapheme-to-phoneme (g2p) transliteration. By training a character-level transformer model on this task across 22 languages and measuring the model's proficiency against its grapheme and phoneme inventories, we show that certain characteristics emerge that separate easier and harder languages with respect to learning to pronounce. Namely that the complexity of a languages pronunciation from its orthography is due to how expressive or simple its grapheme-to-phoneme mapping is. Further discussion illustrates how future studies should consider relative data sparsity per language in order to design more fair cross lingual comparison tasks.