A nonlinear model of word length effects in apraxia of speech

Previous studies of length effects in normal and disordered spoken language production were based on a linear view of words as strings of phonemes or syllables. This approach was limited by an inherent interaction of numbers of phonemes and syllables with syllable complexity. The present investigation was therefore based on a metrical tree model of word form representations, which was implemented by a nonlinear regression model predicting accurate word production in apraxia of speech. One hundred data samples from 20 apraxic speakers were used to estimate the probabilties of accurate production for a data corpus of 72 carefully controlled words and nonwords of varying lengths. The nonlinear regression model explained almost 80% of the variance in these data. The shape of the model indicated a preserved cohesion of phonetic representations at the rime and the foot level in apraxia of speech.