Summary. Forty‐nine children (aged 8:4 to 10:4) were grouped according to reading ability and were tested on lexical decision, naming and spelling of words and non‐words which differed in orthographic neighbourhood size. Friendly words with many neighbours caused significantly fewer errors on all these tasks. Item analysis demonstrated that these effects were robust when corrected for frequency, regularity, word length and age of acquisition. Good readers showed less reliance on neighbourhood size and more evidence of accurate use of grapheme‐phoneme correspondence rules. The results indicate that children find common orthographic sequences easier to read and spell before they have learned to use grapheme‐phoneme correspondences consistently.

Research on children's reading has shown that words high in imageability are easier to read than words low in imageability. It has been suggested that this occurs because low imageability words are acquired later in life than are high imageability words. The effects of age of acquisition and imageability were studied in two reading tasks. In Expt 1, nine-year-old children had to read aloud words differing in age of acquisition (but qatched for imageability, length and frequency). A highly significant effect of age of acquisition on reading accuracy was obtained. The children also had to read aloud words differing in imageability (but matched for age of acquisition, length and frequency). Although imageability affected accuracy, the difference in high and low imageability words was significant only for the poorer readers in the sample. Thus, it appears that age of acquisition of words is a major determinant of reading accuracy and that the imageability effect is attributable to this variable, except for poor readers, who do show a genuine effect of imageability upon reading accuracy. In Expt 2 adult naming latencies were obtained for the same tasks. These latencies were significantly lower for acquired words early than for words acquired later. Word imageability did not affect adult naming latencies.In an early study, Stoke (1929) found that children aged 10-11 years and college students recalled concrete words of higher imageability better than they did low imageability abstract words. Many later experiments reviewed by Paivio (1971) confirmed this finding when word frequency and other variables are controlled. Word lists high in imageability are more memorable in a range of episodic memory tasks and, furthermore, highly imageable sentences are more easily comprehended and remembered than sentences low in imageability (e.g. Holmes & Langford, 1976). Although most of the research on memory has studied adult performance, Vellutino & Scanlon (1985) found that children who were poor readers had special difficulty in recalling low imageability words under conditions of auditory presentation.In addition to the pervasive effects on memory tasks, word imageability has also been found to affect reading performance: it appears to affect reading accuracy in certain types of acquired dyslexics. Marshall & Newcombe (1973) and others (e.g. M. Coltheart et af., 1980) have reported that deep dyslexics, who are notable for the fact that they make semantic errors when reading aloud words presented in isolation, also find highly imageable words easier to read than low imageability words. The deep dyslexic's semantic errors and deficit on low imageability words have been attributed to a disordered semantically mediated reading routine. Interestingly, a 1 P S Y 79

We thank Max Coltheart, Karalyn Patterson, and Derek Besner for comments on a draft of this article. We are also grateful to Philip Smith and Estelle Doctor with whom we had useful discussions.

Two groups of-children with mean ages of 7.47 and 9.04 years and with reading ages of 7.27 and 9.48 years respectively were asked to read lowfrequency words which differed in the consistency and regularity of their endings and which had many or few orthographic neighbours. Both groups found words with many neighbours easier to read than those with few. Only the better readers were affected by the type of ending; regular-consistent words were easier than both regular-inconsistent words and exception words which were irregular and inconsistent. However, there was no difference between the latter two types of word. It is suggested that consistency affects children's reading before regularity per se and some regularity effects may be attributable to consistency. The pattern of performance observed is most easily interpreted within connectionist models of reading in conjunction with a distributed model of phonological production such as that of Dell (1986).

No abstract