Fig. 5: Contribution of distinct diphone categories to phonemic VM.
From: Phonemic segmentation of narrative speech in human cerebral cortex
In order to elucidate whether the unique explanatory power from diphone features could be a reflection of responses to high-frequency short words, top panel shows that diphones are split into three categories: short words, word beginnings and diphones that do not belong to these two categories called residual. Bottom panel shows that the average normalized prediction performance of short words is significantly higher than word beginnings and the diphone residual. The inset shows the fraction explained before normalization (lighter box) and the proportion of diphones in each group (solid bar). The statistics have been derived from all the significantly diphone voxels (n = 15,396 to 16,001 voxels) examined across 11 independent subjects. All tests are two sided and corrected for multiple comparisons (*p < 0.05, **p < 0.01 and ***p < 0.001). The exact p-values can be found in the Diphone segmentation for identification, expectations or lexical retrieval section. The box plots are defined the same way as in Fig. 4. Data used to generate this figure has been provided in source data.
