Regional vowel patterns as shown by discrete cosine transforms

This project presents a new approach to analyzing geographical patterning in vowel variation that combines discrete cosine transforms (DCTs) with cluster analysis. It offers a means of reducing bias from analyses of geographical patterning in vowel variation. DCT0 and DCT1 capture the overall position of vowels in the vowel envelope and DCT2 adds information about curvature. To mitigate anomalies, these metrics are based on numerous tokens and measurement points. Cluster analysis can then be applied to the DCT data to indicate which speakers are most similar without predetermined groupings. The procedure suggests how vowel realizations are correlated with geographical divisions, if at all, within the area covered by a dialect survey. Here, we apply DCTs and hierarchical clustering to a corpus of speakers born 1970 or later and covering eastern Ohio, West Virginia, and western North Carolina. The results align only partially with isophones from earlier dialect surveys. Analyses of individual vowel phonemes typically exhibit considerable intermixture of forms; clearer geographic patterns emerge primarily when multiple vowels in named chain shifts are considered together. Recent dialect leveling appears to play a role in the paucity of distinguishable regional patterns.