Quantifying ultrasound data from a tongue twister experiment using curve-to-curve distance

This study examined the production of four-word tongue twisters (e.g. top cap cop tab) by six English participants using ultrasound imaging of the tongue. Onset consonant closures for /t, d, k, ɡ/ in the contexts of /a/ and /æ/ were extracted from the video and traced using EdgeTrak software. Traces were analyzed using a mean minimum point-to-point distance measure between curves [12]. Curves were compared within allophonic group (e.g. [ka] in cop vs. [ka] in com) and across phoneme category (e.g. [ka] in cop vs. [ta] in top) to quantify similarity to the desired production and expected error. Perceived speech errors had greater within-group distance and smaller between-group distance, consistent with the production of the wrong target. Productions that deviate significantly from other productions within the group can be classified as gradient errors. By these measures both categorical errors and gradient errors were relatively rare.