Phoneme Mapping
Predicting proper-name transliteration in Biblical texts
When translators encounter proper names in text, whether of persons, ethnicity or place they are usually rendered into the target language by a process of phonemic transliteration. Thus a name such as Abraham rendered into Tonga becomes Abulahamu. The ‘r’ phoneme in the English name rewrites as an ‘l’ and a schwa is inserted after the ‘b’ to voice the consonant correctly. Lastly, Tonga prefers open syllables and so the final closed syllable in the English ‘ham’ rewrites as two open syllables ‘ha’ and ‘mu’ with the ‘u’ added to voice the final ‘m’. The rules which govern these transformations are learned empirically by language speakers and can be surprisingly complex. Since they are typically not consciously recognised this can give rise to inconsistencies in the way names are rendered in a text. Like names, or even a single name may be rendered with small differences by different translators or even the same translator at different times. Such inconsistencies create needless problems for the reader who may fail to recognise an individual in different contexts as a consequence of inconsistent renderings.
A process is proposed by which the character phonemes of proper names can be identified in the original or model languages and then transformed into the best rendering for the target language. The transformation rules required for this operation should be derived from example data provided by the translators.
Initial explorations should include creating a proper names list from the base texts which can be rendered in the International Phonetic Alphabet (IPA). It is possible that Markov chain technology may be a helpful way of modelling phonemic shift , although it is expected that multi-dimensional Markov chains will be required to cover common transformations.
When translators encounter proper names in text, whether of persons, ethnicity or place they are usually rendered into the target language by a process of phonemic transliteration. Thus a name such as Abraham rendered into Tonga becomes Abulahamu. The ‘r’ phoneme in the English name rewrites as an ‘l’ and a schwa is inserted after the ‘b’ to voice the consonant correctly. Lastly, Tonga prefers open syllables and so the final closed syllable in the English ‘ham’ rewrites as two open syllables ‘ha’ and ‘mu’ with the ‘u’ added to voice the final ‘m’. The rules which govern these transformations are learned empirically by language speakers and can be surprisingly complex. Since they are typically not consciously recognised this can give rise to inconsistencies in the way names are rendered in a text. Like names, or even a single name may be rendered with small differences by different translators or even the same translator at different times. Such inconsistencies create needless problems for the reader who may fail to recognise an individual in different contexts as a consequence of inconsistent renderings.
A process is proposed by which the character phonemes of proper names can be identified in the original or model languages and then transformed into the best rendering for the target language. The transformation rules required for this operation should be derived from example data provided by the translators.
Initial explorations should include creating a proper names list from the base texts which can be rendered in the International Phonetic Alphabet (IPA). It is possible that Markov chain technology may be a helpful way of modelling phonemic shift , although it is expected that multi-dimensional Markov chains will be required to cover common transformations.
