A salient characteristic of both child language and aphasia is that target segments are often produced unfaithfully. In this study of corpora of child language and aphasic language, we show that these substitution patterns are not random; we also show to what degree both groups perform alike, and where differences occur.
Background. The Regression Hypothesis (Jakobson 1941/1963) states that language attrition in aphasic patients mirrors stages of language acquisition in children. It has been shown that this claim has to be rejected on the segmental level. For the present study, we focus on the sub-segmental substitution patterns of aphasics and children, thus allowing for the investigation of a more nuanced version of the Regression Hypothesis.
Methodology. For each subject group, every segment in the consonant inventory of Dutch (e.g. Booij 1995) was compared with all of its actual realizations. This was done separately for onset and coda positions of monosyllabic words, to begin with. This yielded target-actual confusion matrices that were subsequently analyzed to determine the relative error frequency as a function of feature-based phonemic distance. Phonemic distance was measured using the PMV metric (e.g. Bailey & Hahn 2005). The contribution to the relative error frequency of each of the three dimensions (Place of Articulation, Manner of Articulation, Voice) was subsequently measured for both positions and each group.
Results. All groups show a non-random pattern of segment-for-segment substitutions. In most cases, the relative error frequency shows a decline for increasing phonological distance. A comparison between the younger and older children reveals that the latter perform better over all. Furthermore, for both child groups, there is an important difference between single dimensional errors, i.e. errors of only PoA, MoA or Voice, and errors that involve multiple dimensions, in that the latter occur far less often. The aphasic patients pattern with the older children, but show less sensitivity to phonological distance.
Conclusions. The data reveal that neither aphasics, nor children at various developmental stages show random substitutions. Our results resemble findings by White & Morgan (2008), who showed gradient sensitivity of infants to increasing degrees of mispronunciations in perception studies. So, although the Regression Hypothesis is too strong when considered at the segmental level, new and interesting results can be found in the sub-segmental domain.