klingklangklong / syllabify

Syllabify

Automatically convert plain text into phonemes (US English pronunciation) and syllabify.

Adapted from the repository set up by Anthony Evans with some key changes, itemised below:

• Ported to Python 3 from Evans' Python 2 code;
• Correction of key onset and coda rules which affect consonant clusters and involve the 'maximise onsets principle';
• Removal of all ambisyllabicity from onset and coda rules, since it's not uncontroversial;
• Removal of 'test' (demo) option from syllable script.

Please see Anthony Evans' README file for a detailed background to the project.

Set up

Requires Python 3 (Anthony Evans used Python 2: if that's what you prefer, see his repo).

Clone or download this repo and you're good to go!

Usage

One word at a time:

python3 syllable3.py linguistics

Or several (space-separated):

python3 syllable3.py colourless green ideas

Or as preprocessing for the wordseg program, `wordseg_prep' takes a CHILDES corpus (e.g. Brown) and syllabifies infant-directed speech (i.e. excluding CHI utterances) in phonemic format, with appropriate phone, syllable and word delimiters per wordseg defaults:

python3 wordseg_prep.py $CORPUSPATH

Output

If the input word is found in the dictionary, a phonemic, syllabified transcript is returned. For example, for the word linguistics:

{o: L , n: IH [st:0 ln:short], c: NG }
{o: G W , n: IH [st:1 ln:short], c: empty}
{o: S T , n: IH [st:0 ln:short], c: K S }

There's one syllable per line. Each syllable is made up of an 'o' onset, 'n' nucleus, and 'c' coda. Phonemes are space-separated and capitalized in ARPAbet format. In line with phonological theory, the nucleus must have content, whereas the onset and coda may be empty. Within the vocalic content of the nucleus there's also an indication whether the syllable is stressed ('st':0 or 1), and whether the length ('ln') is short or long.

CMU Pronouncing Dictionary

Syllabify depends on the CMU Pronouncing Dictionary of North American English word pronunciations. Version 0.7b was the current one at time of writing, but it throws a UnicodeDecodeError, so we're still using version 0.7a (amended to remove erroneous 'G' from SUGGEST and related words). Please see the dictionary download website to obtain the current version, add the cmudict-N.nx(.phones|.symbols)* files to the CMU_dictionary directory, remove the '.txt' suffixes, and update the line VERSION = 'cmudict-n.nx' in cmuparser3.py

To do

Offer the option to 'translate' US to UK pronunciations; for instance dealing with lack of rhoticity by converting AXR and ER phones to UK equivalents (I know, which UK variety!? Cross that bridge etc..)

Contact

If you have queries or feedback please contact cainesap at gmail.com

Andrew Caines, September 2017