A study on non-adjacent dependency learning in Yélî Dnye with Rebecca L. A. Frost and Marisa Casillas. Data collected August-October 2019.
Note to self (2019-01-08): Manuscript and analyses under development in separate repo for the time being
Replicate prior non-adjacent dependency learning (NAD) findings with a population that is not Western, not industrialized, and (primarily) non-literate in the native language.
We will follow https://pure.mpg.de/rest/items/item_2498714_3/component/file_2498722/content as closely as possible, but low-tech.
The basic instructions for using the set of scripts in the stimuli folder are as follows:
- Record the syllables of interest w/ a head-mounted mic and recording device MC has access to during fieldwork (in case we need comparable stuff later/last-minute changes)
- Carrier prhase: X, aX (e.g., to ato) multiple times in a few different orders
- Use Praat to manually annotate candidate syllables and then automatically pull them out and save them as individual word clips
- For this use save_labeled_intervals_to_wav_sound_files.praat
- Pick the best example of each syllable, that is:
- Good modal voicing on the vowel, accurate consonant production, flat pitch at the target tone, not too long/breathy)
- All else being equal, pick the token that is the best match to the others already chosen in its overall gestalt
- Add 225ms of "closure" silence to the voiceless stop consonants if they were cut at the release burst
- Trim all word clip files to have accurate onset and offset boundaries and add a TextGrid for each clip to mark the boundary between the consonant and the vowel
- "C" in the consonant interval and "V" in the vowel interval
- Use a Praat script to read in the word clips and their TextGrids, equalize the duration of the parts across words, and write out the new syllables
- For this use equalize_segment_duration.praat
- Note that we used: 150ms for consonants and 300ms for vowels
Still to come.
Still to come.