Rethinking the graphemic buffer?

Abstract

One of the processes involved in spelling is the short-term buffering of graphemes prior to serial production of their corresponding letter forms in written spelling and letter names in oral spelling. Recent cognitive neuropsychological and computational work has been directed at developing a more detailed understanding of this buffering process (Rapp & Kong, 2002; Sage & Ellis, 2004). Rapp and Kong (2002) argued that graphemic buffering consists of (at least) two major operations: the activation of a word’s constituent graphemes, followed by the serial selection of the graphemes for the temporally ordered production of their names or forms (see also Houghton, Glasspool, & Shallice, 1994). Contrary to a commonly-held position, Sage and Ellis (2004) argued that representations at the level of the graphemic buffer (GB) are sensitive to lexical factors such as lexical frequency, age of acquisition, imageability and neighborhood size. They also questioned the robustness of two common manifestations of GB deficits—length effects, and the bow-shaped accuracy function across letter positions. Sage and Ellis specifically claimed that length effects are largely attenuated when word sets are matched for lexical factors, and that the bowshaped accuracy function may be an artifact of limiting analyses to words with single errors. In the work presented here, we: (1) integrate these two sets of claims by showing that activation deficits are sensitive to lexical frequency, while selection deficits may not be; and (2) report, contra Sage and Ellis, robust length effects and bow-shaped accuracy functions even under conditions they predicted would cause their attenuation or elimination.