The impact of progressive semantic loss on reading aloud

Abstract

The role of semantics in reading aloud remains controversial. To explore this issue, the current study examined the impact of semantic loss on reading-aloud performance in 7 patients with semantic dementia. The results revealed a heterogenous pattern of reading difficulties. Of the patients, 2 selectively made errors on inconsistent words (i.e., surface dyslexia), 4 had a generalized reading deficit with increased errors on consistent words, inconsistent words, and nonwords, while the remaining patient had relatively intact reading-aloud accuracy. All patients had longer reading latencies on real words than controls. The relationship between the reading and semantic deficits in the patients was examined at the item-specific level. This suggested that reading-aloud errors were related to the semantic impairment for inconsistent words but not consistent words. In contrast, semantic loss was related to longer latencies for both consistent and inconsistent words. These findings support models of reading that include a role for semantics in the reading-aloud process.

We are extremely grateful to the participants and their families who generously donated their time and energies to be involved in this research. We would also like to thank A/Prof. David Darby, Prof. Elsdon Storey, Prof. David Ames, Ms Anne Unkenstein, and Ms Maggie Phillips for their invaluable assistance in accessing patients for this research. Max Coltheart and two anonymous reviewers are thanked for their helpful comments on an earlier version of this manuscript.

Notes

¹ Although the age range of controls (68–86 years) was similar to that of the patients (65–83 years), as a group the controls (M = 79.3 years, SD = 4.3) were slightly older than the patients (M = 73.3 years, SD = 7.1), t(24) = 2.6, p < .025.

² All experimental stimuli are available on request from the author.

³ The first 2 patients that were tested (V.M. and T.P.) only had 10 consistent and 10 inconsistent words in the low-frequency band (for a total of 60 words). After testing these participants it was decided to increase power for the analysis of consistency effects within the critical low-frequency range, where consistency effects are most pronounced (e.g., Patterson & Hodges, 1992). The properties of this reduced set of low-frequency items (i.e., mean frequency, length, bigram frequency, etc.) did not statistically differ from the larger set used for subsequent participants (all p-values > .1).

⁴ In the analysis of low-frequency items only, patients G.M. and M.S. had impaired accuracy on inconsistent words (both p-values < .05) but intact reading of consistent words (both p-values > .1), while patients V.M., S.M., T.P., and T.W. continued to display impairments on both consistent and inconsistent low-frequency words (all p-values < .01).

⁵ Items for patient and all controls were entered into the analysis.

⁶ A factorial analysis could not be used to compare reading accuracy for known and unknown words at the same time as including consistency because such an analysis would be compromised by low power (Tabachnick & Fidell, 1996).

⁷ Tests of interactions were not conducted in the analysis of some patients using the partial and matching criteria, as the mean reading latencies being compared were based on fewer than five items. This was always associated with low numbers of “unknown” items, of which there were fewer under the more lenient partial and matching criteria. Five items was considered the minimum that was needed in order to provide a reliable mean latency.

⁸ The analysis of patient T.P. using the standard criteria did reveal an interaction between knowledge and consistency (p < .025); however, follow-up comparisons revealed no significant differences between known and unknown words for either consistent or inconsistent items (both p-values > .1).