Multiple linear regression analysis of errors in deaf speech
Harvey Stromberg, Harry N. Levitt · The Journal of the Acoustical Society of America · 1979
A multiple linear regression analysis was performed relating percent intelligibility to segmental and suprasegmental errors in the speech of the deaf. Segmental errors were categorized in several ways according to intended phoneme, produced phoneme (or omission), location in work, and articulatory features involved. Suprasegmental errors were subdivided into 26 basic types within five broad categories (source characteristics, pitch, stress, timing, and rate). The analysis was repeated several times using different error groupings. Relatively small weightings were obtained for the majority of error types. The ranking of errors with respect to their effect on intelligibility was found to depend on how the errors were categorized. Both of these results are a consequence of the high multiple correlations between errors in the speech of the deaf. Nevertheless, a few error types were found to be relatively independent of other errors and these showed extremes in weighting, i.e., either having a marked or comparatively small correlation with reduction in intelligibility. [Work supported by NINCDS.]