Phonological recoding and lexical access

MAX COLTHEART 0 1 2 3 4 0 DEREK BESNER University ofReading , Reading, England RG6 2AL 1 Birkbeck College , Malet Street, London, England 2 EILEEN DAVELAAR University of Reading , Reading, England RG6 2AL 3 Requests for reprints should be sent to Eileen Davelaar, Department of Psychology, Reading University , Reading, England RG6 2AL 4 JON TORFI JONASSON University of Iceland , Reykjavik, Iceland Four experiments are reported that examine the effects of homophony (e.g., SAIL/SALE) on response latency in a lexical decision task. The results indicated that an effect of homophony was evident only if the nonword dis tractors consisted of legal, pronounceable strings (e.g., SLINT), but that this effect disappeared if the nonwords sounded like English words (e.g., BRANE). An optional encoding strategy is proposed to account for this differential effect. It is suggested that while both graphemic and phonemic encoding occurred simultaneously, naive subjects tended to rely on the outcome of the phonological route. However, when such reliance produced a high error rate (i.e., when the nonwords sounded like English words), these subjects were able to abandon a phonological strategy and rely on the graphemic encoding procedure instead. Two further aspects of the results are of interest. First, the less frequent member of a homophone pair was slower when compared with a control item if the nonword distractors were of the SLINT type, but not different if they were of the BRANE type. The high-frequency members did not differ from their controls in either nonword environment. Second, in a homophone "repetition" experiment, the frequency order of presentation within pairs of homophones (i.e., the high-frequency member followed by the low-frequency member, or vice versa) had a substantial effect. A spelling recheck procedure and a response-inhibitory mechanism are postulated to incorporate these effects into a dual-encoding direct-access model of word recognition. - view comes from Bloomfield (1942), Gough (1972), and Rubenstein, Lewis, and Rubenstein (1971), among others. A second suggestion proposes that, on the contrary, no phonological recoding of the printed word is required and that lexical access proceeds directly using the visual representation. This contrasting view has been supported by Baron (1973), Bower (1970), and Kolers (1970). A third view, incorporating both visual and phonological access, has been characterized as a race model by Meyer, Schvaneveldt, and Ruddy (1974a) and as a model of cooperation by Coltheart, Davelaar, J onasson, and Besner (1977). While a good deal of evidence has been advanced in support of each of the above viewpoints, it is difficult to evaluate the relevance of much of this evidence. It is our view that a number of the experimental tasks from which the supporting data have been drawn are not logically appropriate and therefore cannot necessarily address the question concerning the nature of the lexical access code. We have advanced this argument elsewhere (Coltheart et al., 1977) and make the point here again briefly. It is our objection that a variety of the tasks employed can be successfully performed by the subject without involving the use of his or her lexical store. Such tasks as same-different judgments, rhyming judgments, tachistoscopic recognition, and naming-latency experiments do not logically require the subject to make use of lexical knowledge, and therefore, data obtained from these tasks cannot be considered as admissible evidence in the debate. One task that we feel is not subject to the above objections, and that provides the data to be discussed below, is the lexical decision task. Here, the subject is required to discriminate letter strings that are English words from letter strings that are not. To perform this task successfully, it is logically necessary that the subject consult his store of English words, that is, his internal lexicon, in order to discover whether a given letter string is contained in it or not. We would like to restrict this argument, however, to cases where the letter string under consideration complies with the orthographic rules of English. A subject presented with the string BRHND can reject this item as an English word on the basis of orthographic legality alone, since English does not allow such a combination. If presented with a string like SLINT, however, a subject cannot reject such an item on the basis of the rules of English orthography alone and must consult his lexicon to determine its presence or absence. An examination of the data obtained from this task is therefore valid in order to determine the nature of the lexical access code. While this task has been used extensively to investi gate such phenomena as context effects (Meyer, Schvaneveldt, & Ruddy, 1974b), repetition effects (Scarborough, Cortese, & Scarborough, 1977), hemis pheric effects (Cohen & Freeman, in press; Marcel & Patterson, in press), and the lexical access procedure (Coltheart et aI., 1977; Rubenstein et aI., 1971), only a few papers in the literature deal with the lexical access code per se. A paper by Rubenstein et al. (1971) was the first of these. Rubenstein et al. (1971) compared the decision times of subjects to two types of words and to two types of nonwords. The words consisted either of items that were homophones, such as SAIL (SALE), or of items that were not homophones, such as TREE. Similarly, the nonwords were either homophonous with an English word, for example, BRANE, or they were not, for example, SLINT. The data indicated that the subjects took longer to decide that homophones were words than to decide that nonhomophones were words; also, the subjects took longer to reject pseudohomophones such as BRANE than to reject nonpseudohomophones such as SLINT. On the basis of these results, Rubenstein et al. (1971) suggest that the lexical access code is a phonological one. Their argument runs as follows. A printed letter string is first recoded into a phonological representation and this representation is then used to discover whether or not a corresponding entry exists in the internal lexicon. The procedure by which this investigation is carried out is characterized as a search process, proceeding from high- to low frequency items. When an entry has been successfully located, the search terminates and a positive response results. If no entry is found, the search terminates only after every entry has been examined. Since pseudo homophones such as BRANE will result in an entry's being located, only a spelling recheck with the stimulus will allow this item to be correctly rejected. Since the spelling recheck operation results in some time cost before the search can be recommenced, these items present slower decision times than nonpseudohomo phones for which no corresponding entries exist. Similarly, in the case of homophones, sometimes the located entry will be the incorrect one and again only a spelling (...truncated)