spellingtophoneme

Spelling-to-phoneme, commonly known as grapheme-to-phoneme (G2P) conversion, is the process of converting a sequence of written letters (graphemes) into a corresponding sequence of phonemes, the units of sound in speech. The output may include stress marks and syllable boundaries. This task is foundational for many speech technologies, pronunciation lexicons, and language learning tools. In English, spelling-to-phoneme is challenging due to irregular spellings and multiple pronunciations for some spellings.

Languages differ in difficulty for spelling-to-phoneme mapping. Shallow orthographies yield relatively consistent grapheme-to-phoneme correspondences, while deep orthographies like English require context-sensitive rules and handling of irregularities, digraphs, and morphology. Some languages map each grapheme to a single phoneme, whereas others involve multi-letter sequences, allophony, and tone or stress information that affects pronunciation.

Approaches to spelling-to-phoneme include rule-based systems that encode linguistic knowledge, data-driven methods that learn mappings from word-pronunciation pairs, and end-to-end neural models that map grapheme sequences to phoneme sequences. Outputs are typically phoneme sequences aligned to a chosen phoneme inventory, such as IPA or a language-specific set (for example ARPABET in English). Resources for development include pronunciation dictionaries (e.g., the CMU Pronouncing Dictionary), lexicons, and open-source tools like Phonemizer and Sequitur-G2P. Applications span text-to-speech, pronunciation assessment, language learning software, and preprocessing for speech recognition. Evaluation uses metrics such as phone error rate and G2P accuracy.

Challenges include irregular pronunciations, homographs, allophony, and dialectal variation. Ongoing work aims to improve coverage, cross-language transfer, and efficiency across languages.