Contributions of Processing Fluency to Repetition Effects in Masked Word Identification

Prior exposure to a word is shown to improve its later identification in a brief, masked display when a free report task is used, but not in two-alternative forced choice or single-probe matching tasks that eliminate certain bias effects and provide an assessment of discriminability. Modified forced choice and single-probe tasks were also used, in which subjects attempted to identify the target before presentation of the probe(s). This modification produced a discriminability advantage for old words, but only in the single-probe task. We argue that prior exposure does not enhance sensory processing of a target word; rather, it increases the fluency with which the target comes to mind when presented under difficult viewing conditions. In forced choice and single-probe tasks, fluency associated with processing the target may be ignored in deference to discriminating among or evaluating the processing of the probe(s). Memory for specific prior processing episodes can, under appropriate circumstances, exert an indirect or automatic influence on the performance of certain tasks. In this article, we explore how memory for specific episodes can act automatically to influence task performance. By "automatically," we mean that subjects are not attempting to consciously recollect prior events in an effort to improve performance. Rather, their performance is influenced automatically by the consequences of retrieving a representation of an earlier, relevant event. Our goal here is to investigate how automatic retrieval of such an event comes to influence current task performance. A useful guide for this investigation is the transferappropriate processing framework that Roediger and his colleagues have applied to the study of indirect influences of memory (e.g., Roediger & Srinivas, 1993; Roediger, Weldon, & Chains, 1989). According to this framework, memory for a processing episode can be beneficial for later task performance if the remembered episode and the subsequent task share processes. For example, completion of a word fragment (e.g., _l_p_a_t) is enhanced more by prior exposure to the printed word solution for the fragment (elephant) than by prior exposure to a drawing of the object denoted by that word (Weldon & Roediger, 1987). Tasks that show perceptual specificity in benefits derived from prior episodes are assumed to rely heavily on data driven processes. Application of Jacoby's (1991) process dissociation procedure to the word stem completion task (e.g., complete the stem REA to form a word) suggests that the automatic influence of prior study episodes on that task is highly dependent on perceptual processes ([acoby, Toth, & Yonelinas, 1993; Toth, Reingold, & Jacoby, 1994). For instance, study episodes that involved solving an anagram or generating a target word from a semantic cue led to no reliable increase, relative to nonstudied words, in estimates of the automatic influence of memory on stem completion, whereas reading a visually presented word substantially increased estimates of the automatic parameter. Our primary concern in this article, however, is with the effect of prior study on the masked word identification task (also known as the perceptual identification task). In this task, a target word is briefly displayed and followed by a pattern mask. The task is to identify the word. Various studies have found evidence to support the view that in this task, as in word stem and word fragment completion, effects of prior exposure are mediated by data driven processes. For example, greater enhancement has been found if target words were read as opposed to heard initially (e.g., Jacoby & Dallas, 1981; Levy & Kirsner, 1989). Moreover, Jacoby (1983b) found little or no enhanced identification of words that had been generated from their antonyms, whereas previously read words yielded marked improvement. Reinitz and Demb (1994) have argued that the masked word identification task is even more dependent on perceptual processes than the word fragment completion task. On a masked word identification task, they found no enhancement for compound words (e.g., toothache) that were conjunctions of compound words (e.g., toothpaste, headache) studied earlier. In contrast, they obtained enhanced word fragment completion and elevated false alarms on a recognition test for such items. Reinitz and Demb argued that effects of prior exposure on the masked word identification task arose from holistic perceptual memories that improved "perceptual clarity" on the identification task. Two sources of evidence run counter to the view that prior exposure effects in the masked word identification task are mediated exclusively by perceptual processes. Canadian Journal of Experimental Psychology, 1996, 50: 1, 9-21 10 Masson and MacLeod First, in an earlier report (Masson & MacLeod, 1992), we showed that enhancement due to prior exposure effects can be obtained for words that were generated from semantic clues and never seen during study. Furthermore, this enhancement can be just as strong as that produced by reading words during the study phase. Although Toth et al. (1994) argued that these results were likely due to intentional use of memory, we think there is reason for caution on this point. In particular, the Toth et al. criticism is based on results with the word stem completion task, whereas our results were obtained with the masked word identification task. It is quite conceivable that the automatic influence of memory operates differently in these two tasks. For example, the brief presentation of a word is capable of retrieving semantic information under binocular masking conditions such as those used in masked word identification (Carr & Dagenbach, 1990; Dagenbach, Carr, & Wilhelmsen, 1989; de Groot, 1983). A word stem may not provide enough constraints for this to occur, although we know of no test of this claim. In general, however, we think it is plausible that test cues in these two tasks are differentially capable of retrieving prior episodes grounded on semantic information. We note that an opposite view regarding the relative potential of word stem completion and masked word identification for retrieving semantic aspects of prior study episodes has been put forward by Keane, Gabrieli, Fennema, Growdon, and Corkin (1991) based on results from patients with Alzheimer's disease. Keane et al. found impaired effects of prior exposure to target words on word stem completion with these patients, but preserved effects in masked word identification. They concluded that prior exposure effects in word stem completion depend on conceptual or lexical processes that are at risk in Alzheimer's disease, whereas improved performance on masked word identification is mediated by perceptual processes. A difficulty with this position is that the Toth et al. (1994) results are not consistent with the supposed conceptual basis for prior exposure effects on the word stem completion task. We are not in a position to resolve this issue at the present time, but anticipate that future research will provide clarification. For current purposes, we have adopted the working assumption that a briefly presented target word in the masked word identification task is capable of recruiting semantic information from related encoding episodes. A second source of evidence for nonperceptual contributions to enhancement is relevant to the strong view that prior exposure enhances perceptual clarity in the masked word identification task. If perceptual clarity is improved by prior exposure, then variants of the masked word identification task that allow application of signal detection analysis should yield improved discriminability as a result of prior exposure. This has not proved to be the case. Ratdiff and his colleagues (Ratcliff, McKoon, & Verwoerd, 1989; Ratcliff & McKoon, 1993) used a two-alternative forced choice procedure in which masked presentation of a target was followed by two clearly presented alternatives, one of which matched the target. Prior exposure to the target failed to increase accuracy on this task. Similarly, Masson and Freedman (1990) found that when subjects made lexical decisions about masked targets, increased accuracy in responding to previously studied words was offset by increased false alarms to nonwords that resembled old words (e.g., table instead of table). Thus, discriminability for old items was not reliably different from discriminability for new items. If prior exposure does not improve perceptual clarity, what is the source of improvement on the masked word identification task? Ratcliff et al. (1989) proposed that a form of "perceptual bias" is responsible for this effect. In their forced choice experiments, they found that when the alternatives that followed the masked target consisted of one old and one new word that were orthographically similar (e.g., died, lied), subjects were more likely to select the old alternative, regardless of whether the masked target word was old or new. Ratcliff et al. attributed this bias to an early perceptual process, rather than to a deliberate response bias, because (1) subjects reported that old words seemed to "jump" out of the display, and (2) response times were generally less than 1 s, indicating that there was little strategic processing. They proposed that the bias operates when subjects must select one member of a set of candidates invoked by an impoverished stimulus (e.g., a masked target word or word fragment). Under these conditions, subjects preferentially select an old word. If response alternatives provided on a forced choice test are not orthographically similar, the bias has no grounds to operate, and no bias effect is found (Ratcliff et al., 1989, Experiment 5). Building on the Ratcliff et al. (1989) proposal, we have developed a somewhat different account of how prior exposure influences performance on masked word identification. The brief presentation of the target word serves as a retrieval cue, recruiting knowledge and episodes connected to the orthographic information in the target (see also Jacoby & Brooks, 1984; Kahneman & Miller, 1986; Whittlesea & Brooks, 1988; Whittlesea & Cantwell, 1987). For old words, the recent encoding episode in the study phase should be easily retrieved relative to other episodes, in part because of the strong temporal and contextual overlap between study and test conditions (Jacoby, 1983a). Note that, by retrieval, we do not mean that information about the study episode is necessarily available to consciousness, but rather that the target word comes fluently to mind. This fluency can be ascribed to a number of possible sources. In an identification task, subjects often attribute fluency to clarity of perception (see Jacoby, Allan, Collins, & Larwill, 1988, and Witherspoon & Allan, 1985, for other examples of attributions of memory-induced fluency to perceptual causes). It is this attribution that is responsible for the impression that a word seems to "jump" out of the display. As in the Ratcliff et al. account, nothing in the account we present here implies an improvement in sensory processing of a briefly presented target word. Thus, failure to find an effect of exposure on the forced choice task, which Ratcliff et al. took as powerful evidence against the Processing Fluency 11 "perceptual clarity" view, is not problematic from our perspective. Before rejecting the view that prior exposure improves sensory processing (perceptual clarity) in the masked word identification task, we note that there are grounds to question the generality of the Ratcliff et al. (1989) results. Target words in the Ratcliff et al. experiments either were presented in the context of sentences or were inferred from sentences without actually being read. Embedding words in text has been shown to reduce the effect of prior exposure in masked word identification (e.g., Levy & Kirsner, 1989), and there is some controversy regarding the source of prior exposure effects on such tasks for words that were not actually read (Toth et al., 1994). Given these concerns, we thought it advisable to replicate the forced choice results reported by Ratcliff et al. (1989), using the typical encoding task of reading aloud. Moreover, Ratcliff et al. demonstrated a bias to prefer an old word when presented with an old and a new word as forced choice alternatives. To rule out this particular bias, we equated forced choice alternatives with respect to whether they were old or new. We also considered the possibility that the forced choice procedure might interfere in some way with processes that would otherwise generate a prior exposure effect. For example, when required to discriminate between two available alternatives, subjects may neglect evidence, such as the fluent retrieval of the target word, that would have been used in the standard free report situation. In place of the neglected evidence, subjects might use only contrastive information that permits discrimination between the two alternatives (e.g., noticing a line feature in the second letter position of the target display permits discrimination between the alternatives list and lost). To mitigate such effects, we used a single-probe task in some of the experiments we report. In this task, only one probe word was presented following the target. The probe was always equated with the target with respect to whether it was old or new and either matched it or was unrelated to it. In each experiment, the crucial question was whether prior exposure to target words would significantly improve performance on a task designed to separate discriminability from response bias.