Using residualization to separate phonological and articulatory complexity in speech production

The extent to which phonological and articulatory processes can be separated in speech production models, theoretically and neurologically, remains debated. One way to study their roles is by measuring phonological and phonetic (articulatory) complexity. A primary challenge is the lack of tools that separate linguistic structure from motor demands. Clinically, accurate complexity measures are essential for diagnosing and treating speech disorders like apraxia of speech and aphasia. However, common measures such as the Word Complexity Measure (WCM) blend phonological markedness and articulatory difficulty, limiting their specificity. To address this, we applied a residualization approach to isolate these dimensions. Fifty-three individuals with left-hemisphere stroke completed a pseudoword repetition task using 58 pseudowords designed with the WCM. Items were re-scored using Kent’s articulatory-motor index, and WCM was regressed on Kent and syllable count. Residuals (WCMres) were interpreted as phonological complexity with articulatory load removed. Kent scores represent articulatory complexity. Preliminary analysis revealed moderate correlation (R² = 0.33), and residual diagnostics showed no linear association with articulatory complexity. Ongoing mixed-effects modeling examines how these two metrics predict reaction time, accuracy, duration, and fluency. Future work will investigate whether WCMres and Kent are linked to phonological and motor control regions via white matter connectivity.