The purpose of this study is to investigate the acoustic correlates of secondary articulation in the Abzhywa dialect of Abkhaz. Data for this study were collected in Sukhum in September of 2014 from four speakers of Abzhywa Abkhaz, two male and two female. Each speaker read a list of 144 words three times, resulting in a total of 1728 words. From these data 151 tokens of word-initial voiceless fricatives were selected for comparison. Tokens belong to one of eight types: retroflex, plain palato-alveolar, labialized palto-alveolar, plain uvular, labialized uvular, palatalized uvular, plain pharyngeal, or labialized pharyngeal. Each token was measured for duration, average intensity, and center of gravity in Praat.

Using the 'hclust' clustering function in R, tokens from all eight types were clustered based on the three acoustic measures. Then, fricatives with shared primary places of articulation but different secondary articulations were compared along each of the three acoustic measures. The 'k-means' clustering function was used to cluster tokens sharing a primary place of articulation based on two of the three acoustic measures determined by the function to be the best predictors of token type.

This study found that the three acoustic measures used were not sufficient to distinguish the fricatives across places of articulation. For fricatives sharing a primary place of articulation, center of gravity and duration were selected by k-means as the two most relevant components in all three cases. However, the clustering in all cases was not homogenous, indication that further acoustic features are necessary for accurately distinguishing fricatives that share a primary but not a secondary place of articulation.