IPA Transliteration

A 1,003-example BIG-bench task that transliterates sentences between English and the International Phonetic Alphabet.

Also known as: IPA Transliterate, international phonetic alphabet transliterate

unassessed

This page is a discovery lead. Nobody has yet assessed it against the catalogue contract, so it carries no disposition. Absence of evidence here is not evidence of staleness.
Categorytranslation
SubcategoryBIG-bench sentence-level English–IPA transliteration
Page statusunknown
Metricbleu
Directionhigher_is_better
Dataset size1003
Dataset licenceOANC End User License
PublisherGoogle (BIG-bench collaboration)

What it measures

international_phonetic_alphabet_transliterate asks a model to convert a sentence from English orthography into IPA, or the reverse. The authors treat it as a probe of rare-symbol handling and of whether pretraining stored word-level pronunciation maps, not as a speech benchmark. Sentences are drawn from MultiNLI generated (non-web) hypotheses under the OANC licence, fiction excluded, then converted with the CMU Pronouncing Dictionary via eng-to-ipa. The prompt prefix gives three mixed-direction exemplars.

Task format

Free-text transliteration. task.json preferred_score is bleu; metrics also include rouge and exact_str_match. Keywords include many-shot. Canary GUID embedded. Direction is encoded in the input (IPA: … versus English: … plus an IPA: suffix).

Models reporting this benchmark

No model card in ModelSpec reports this benchmark yet.

Data

This page as JSON · Edit on GitHub