A simple psychophysical procedure separates representational and noise components in impairments of speech prosody perception after right-hemisphere stroke

Abstract

After a right hemisphere stroke, more than half of the patients are impaired in their capacity to produce or comprehend speech prosody. Yet, and despite its social-cognitive consequences for patients, aprosodia following stroke has received scant attention. In this report, we introduce a novel, simple psychophysical procedure which, by combining systematic digital manipulations of speech stimuli and reverse-correlation analysis, allows estimating the internal sensory representations that subtend how individual patients perceive speech prosody, and the level of internal noise that govern behavioral variability in how patients apply these representations. Tested on a sample of N = 22 right-hemisphere stroke survivors and N = 21 age-matched controls, the representation + noise model provides a promising alternative to the clinical gold standard for evaluating aprosodia (MEC): both parameters strongly associate with receptive, and not expressive, aprosodia measured by MEC within the patient group; they have better sensitivity than MEC for separating high-functioning patients from controls; and have good specificity with respect to non-prosody-related impairments of auditory attention and processing. Taken together, individual differences in either internal representation, internal noise, or both, paint a potent portrait of the variety of sensory/cognitive mechanisms that can explain impairments of prosody processing after stroke.

Keywords: Internal noise; Perception; Prosody; Reverse-correlation; Stroke.

Figure 1

The representation + noise model. Patients were presented with 150 successive trials consisting of pairs of manipulated prosodies (A) and asked to judge, within each pair, which sounded most interrogative (B). Patient responses in each trial were fitted with a 2-stage psychophysical model (C), consisting, first, of a prosodic template (or “internal representation”) to which sound stimuli are compared and, second, of a level of “internal noise” which controls how consistently this representation is applied to incoming stimuli. See main text for details about the model-fitting procedure. In this work, we estimate the two model parameters (representation and noise) for each patient individually and compare them with patient records to test their value as markers of receptive aprosodia.

Figure 2

Patient parameters (internal representations and internal noise) estimated by reverse-correlation separate controls from patients above and below the pathological cut-off on the MEC prosody comprehension scale (9/12). Left: Internal representations of interrogative prosody computed from control group responses exhibited a typical final-rise contour, with a marked increase of pitch at the end of the second syllable. In contrast, patients’ internal representations had both lower amplitude and more variable shape across individuals. The bottom waveform illustrates the shape of the base sound used to generate stimuli (a male-recording of the word vraiment/really). Right: control participants were able to apply these representations remarkably consistently across trials, with internal noise values < 1 standard deviations of stimulus noise. In contrast, patients’ internal noise levels were larger and more variable, and scaled with prosodic difficulties measured by MEC.

Figure 3

The representation + noise model captures a rich diversity of sensory/cognitive mechanisms underlying impairments of prosody processing after stroke. Center: Distribution of representation typicality and internal noise for controls and patients (considering all 4 sessions), overlaid with by kernel density estimate. Histograms on the marginal axes show univariate distributions for each variable in the patient group. Corners: Corner boxes show internal representations (top) and behavioral series of responses (bottom) for 4 illustrative patients. Patients in top corners have internal representations (blue) that are similar to controls (orange), but vary in amounts of internal noise (e.g. showing excessive response perseveration; top-right). Patients in bottom corners have atypical representations (blue), but some nevertheless retain healthy levels of internal noise (e.g., being normally consistent in wrongly expecting question phrases to decrease rather than increase in pitch; bottom-left). The estimation of internal noise was limited to the range [0; + 5std]; data points in the upper side of that range may either correspond to true internal noise values, or to larger values for which we could not provide an exact estimate, as illustrated here with a dotted line in the central panel (see Appendix for details).