Journal of the International Neuropsychological Society (1996), 2, 340-349. Copyright © 1996 INS. Published by Cambridge University Press. Printed in the USA. Reading lexically without semantics: Evidence from patients with probable Alzheimer's disease ANASTASIA M. RAYMER* AND RITA SLOAN BERNDT Department of Neurology, University of Maryland Medical School, Baltimore, Maryland (RECEIVED June 23,1995; ACCEPTED October 27,1995) Abstract Recent modifications of the lexical model of oral reading make the prediction that under conditions where sublexical reading processes alone cannot achieve the target pronunciation (i.e., when words have exceptional spellings or when sublexical processes are impaired), patients with severe semantic impairment should have more difficulty reading aloud semantically impaired words than semantically retained words. In a battery of lexical-semantic and reading tasks, two neurologically normal control subjects and two subjects with probable Alzheimer's disease (AD) and only moderate semantic impairment read aloud all words accurately. One AD subject with severe semantic impairment was impaired in word reading but demonstrated no difference in reading words with regular and exceptional spellings. Another AD subject with severe semantic impairment read aloud without error virtually all regular and exception words. Neither severely impaired AD subject demonstrated any relationship between oral reading accuracy and semantic knowledge of exception words. These findings support a model of word reading incorporating lexical, nonsemantic processes by which lexical orthographic input representations directly activate lexical phonological output representations without the necessity of semantic mediation. (JINS, 1996, 2, 340-349.) Keywords: Oral reading, Acquired dyslexia, Dementia INTRODUCTION The study of reading deficits in brain-damaged individu- als provides evidence for the structure of cognitive mech- anisms involved in word reading. For example, individuals with "surface dyslexia" have difficulty reading aloud real words with exceptional spelling-sound correspon- dences (e.g., bread, sweater), but are much better at reading regularly-spelled words (e.g., bead) and plausible non- words (e.g., quint, vace) (Marshall & Newcombe, 1973; Shallice & Warrington, 1980; Patterson et al., 1985). Con- versely, individuals with "phonological dyslexia" read fa- miliar regular and exception words well but have difficulty when reading nonwords (Beauvois & Derouesne, 1979; Fun- nell, 1983). The contrasting patterns of surface and phono- logical dyslexia have been used to support a dual-route reading model (see Figure 1) including a semantically me- *Now affiliated with the Department of Neurology, University of Flor- ida, Gainesville, Florida. Reprint requests to: Anastasia Raymer, Ph.D., Box 100236, Depart- ment of Neurology, University of Florida, Gainesville, FL 32610-0236. diatcd lexical route for reading familiar words (irrespective of the regularity of their spellings) and sublexical grapheme- to-phoncme conversion procedures for pronouncing unfamil- iar words (i.e., nonwords) and regularly spelled familiar words (Colthcart, 1985). The reading abilities of other patients have led some re- searchers to propose a third process for reading words. Some individuals with progressive dementia are able to read many words without difficulty, including those with exceptional spellings, in spite of severe semantic impairment (Schwartz et al., 1980; Kremin, 1986). It is not likely that these sub- jects read exception words through lexical-semantic means because knowledge of word meanings is impaired. It is also unlikely that they read exception words through the use of sublexical reading procedures because incorrect regular- ized pronunciations would result (e.g., /brid/ for bread ). Therefore, to account for accurate exception word reading in these cases, some have proposed a lexical-nonsemantic reading route that associates lexical orthographic input rep- resentations directly with lexical phonological output rep- resentations without the requirement of semantic mediation (Morton & Patterson, 1980; Schwartz et al., 1980) (Fig- 340