Skip to content

Latest commit

 

History

History
14 lines (8 loc) · 1.65 KB

README.md

File metadata and controls

14 lines (8 loc) · 1.65 KB

Multilingual IPA and romanization data

The data is presented as tab-delimited text files (separately for IPA transcriptions, romanizations, and ASCII romanizations). Multiple transcriptions and/or romanizations for the same word are given as separate entries of the word. Fully duplicate entries are removed.

File names are Wiktionary language codes.

Wiktionary data is collected partly using a Wiktextract dump of the English Wiktionary and partly with a custom tool by Tamila Krashtan. Clipped transcriptions (such as /-səɹi/ in /ˈdʒænəˌzeɹi/, /-səɹi/) are skipped. The data is provided in a canonically decomposed form.

ASCII romanizations are identical to the romanizations found in Wiktionary, except for being additionally normalized (using AnyAscii) to only contain lowercase Latin letters (a-z) and spaces.

Both ASCII and non-ASCII romanizations are currently filtered to be at most 20 (non-combining) characters long, which helps make the data much cleaner. This constraint applies to the words being romanized as well.

CMU Dictionary transcriptions were converted into IPA with a straightforward1 algorithm: see conversion chart. Note that AH in unstressed syllables is represented as ə, and ER as ɚ. These are the only instances of vowel reduction applied.

Footnotes

  1. That is, apart from the syllabification bit.