When word identification gets tough, retrospective semantic processing comes to the rescue

https://doi.org/10.1016/j.jml.2012.02.002Get rights and content

Abstract

Semantic priming is typically enhanced by target degradation in both lexical decision and pronunciation tasks. Using these tasks, we examined this priming × target degradation interaction when the prime and target were related via symmetrical (SYM) associations (e.g., east west), as in previous research, or for the first time via forward asymmetrical (FA) prime-to-target associations (e.g., keg beer) or backward asymmetrical (BA) target-to-prime associations (e.g., small shrink) of the same strength as those in the SYM pairs. For both clear and degraded targets, SYM priming was equal to the sum of FA and BA priming, with BA priming being equivalent to FA priming except for clear targets in pronunciation, which as in previous research failed to yield significant overall BA priming. An analysis of the full reaction time (RT) distribution showed that (a) except for BA priming for clear targets in pronunciation, all priming effects were significant even for the fastest RTs, (b) priming effects were greater for degraded than clear targets for the BA and SYM pairs for the fastest RTs but moreso for the longest RTs, and (c) priming effects were equal for degraded and clear targets for FA pairs throughout the RT distribution. For RT bins for which the RTs for degraded and clear unrelated targets were matched, RTs in the related priming condition were faster for degraded related targets than clear related targets. These findings suggest that the Priming × Target Degradation interaction typically observed for SYM related pairs is mediated by a retrospective prime-retrieval mechanism that depends on a backward target-to-prime association and is strategically engaged for degraded targets so that semantic information can compensate for the deficient bottom-up processing of a degraded target. They also provide strong evidence for the psychological reality of the general distinction between retrospective and prospective processing mechanisms, a distinction that has been receiving considerable recent interest in research on attentional control and memory.

Highlights

► We test forward or backward asymmetric or symmetric associative priming. ► We use masked or clear targets in lexical decision and pronunciation tasks. ► Priming is greater for masked targets, except for forward asymmetric priming. ► Priming for masked targets enhanced more for longer reaction times. ► Enhanced priming for masked targets due to strategic, retrospective mechanisms.

Introduction

An issue receiving considerable recent attention in cognitive psychology is the distinction between prospective and retrospective processing mechanisms. This distinction has been made in long-term memory paradigms and has been of special interest in the cognitive aging literature (see McDaniel & Einstein, 2007, 2008) and in the realm of working memory and attentional control (e.g., see Braver, Gray, & Burgess, 2007). The distinction can be understood in the context of a short-term memory task in which people are given a few (e.g., 4) items, one at a time, as a memory set and after a brief delay are given a probe item to which they respond whether or not it was in the memory set (Sternberg, 1966). There are two processing modes a person could use in this task. In the prospective processing mode, the person could store along with each memory set item the instruction “press the ‘yes’ key if this item appears as the probe” and then when the probe appears execute that instruction or press the ‘no’ key. In the retrospective processing mode, the person would store only the memory items and when the probe appears use it to search the memory set items to determine if there is a match and then retrieve the response instructions and execute the correct response.

For nearly 40 years, a retrospective account of performance in the Sternberg (1966) short-term memory paradigm was universally accepted and the prospective account was given virtually no consideration. However, to test whether subjects will under certain conditions adopt a prospective processing mode in the Sternberg paradigm, Speer, Jacoby, and Braver (2003) cleverly manipulated the average size of the memory set in different blocks of trials. Specifically, they embedded critical 6-item memory sets, from which they reported the data, in a block of mostly small memory sets or mostly large memory sets. (Because memory sets were presented one item at a time, memory-set size could not be known until the probe appeared.) In the block of mostly smaller memory sets, subjects could effectively use the prospective processing mode during encoding and presumably would apply it to all trials in that block. However, in the block containing mostly larger memory sets, it presumably would be too burdensome to store and retain the response instruction with so many memory-set items. Thus, subjects would utilize a retrospective processing mode.

The results were that reaction times (RTs) to probes from the critical 6-item memory sets were faster and more accurate in the small-memory-set block (prospective processing). The authors argued that the generation of the response code for each memory set item during encoding in the prospective processing mode allowed for a rapid target-detection process when the probe appeared, whereas when subjects were in the retrospective processing mode, they had to adopt a slower retrospective retrieval process that could not be initiated until after the probe appeared. More compellingly, the putative prospective and retrospective processing modes produced a double dissociation in fMRI-measured brain activity. In the putative prospective processing block, activity in the dorsolateral prefrontal cortex steadily increased during memory set encoding and through the probe delay whereas in the putative retrospective processing block it increased only during the encoding of the first few items in the memory set and then showed a small decrease until the probe appeared. In contrast, during probe processing, activity in the anterior prefrontal cortex (which has been linked to retrospective episodic retrieval processes, e.g., Lepage, Ghaffar, Nyberg, & Tulving, 2000) decreased and markedly increased, respectively, in the putative prospective and retrospective processing blocks.

Single and double dissociations such as those reported by Speer et al. (2003) do not provide compelling direct evidence that the qualitatively distinct processes are indeed prospective vs. retrospective because there typically is no direct manipulation of a variable that is necessarily tied to prospective vs. retrospective processing (for an exception using pigeons, see Roitblat, 1980). To remedy this, in the present experiment we manipulate such a variable in the semantic priming paradigm, which in contrast to performance in the Sternberg paradigm, was for many years assumed to be mediated by only prospective processing mechanisms (with a couple of early exceptions being Forster, 1979, Forster, 1981 and Stanovich and West (1983)).

Semantic priming is the observation that in the lexical (word/nonword) decision task (LDT) and pronunciation task, people respond to a target word (e.g., west) more quickly and accurately when it is immediately preceded by a semantically or associatively related prime (east) rather than an unrelated prime (table). (See McNamara, 2005, Neely, 1991 for extensive reviews.) To constrain theories of semantic priming, researchers have investigated how the effects of other variables influence (interact with) priming. Two such variables are the target’s frequency of occurrence in print (word frequency) and the visual quality of the target (target degradation). Results have consistently shown that priming is greater for low-frequency than high-frequency targets in both the LDT (e.g., Becker, 1979, Stone and Van Orden, 1992, Tse and Neely, 2007, Yap et al., 2009) and pronunciation (e.g., Stanovich & West, 1979) and for visually degraded targets than for clear targets in both the LDT (e.g., Balota et al., 2008, Becker and Killion, 1977, Meyer et al., 1975) and pronunciation (e.g., Balota et al., 2008, Besner and Smith, 1992). Specifically, the impaired performance observed for low-frequency targets relative to high-frequency targets and for visually degraded targets relative to clear targets is greater following unrelated primes than following related primes. These interactions are called overadditive because the detrimental effects of lower word frequency and of target degradation are amplified by the adverse effect of the target being presented in an unrelated priming context rather than in a related priming context. Viewed through the lens of priming effects, this means that related primes mitigate the detrimental effects that lower word frequency and target degradation have on word recognition.

The present research focuses on the overadditive Priming × Target Degradation interaction (hereafter P × TDI), which occurs for both the LDT and pronunciation, but only when the relatedness proportion (the proportion of word-prime/word-target pairs that are related) is high (i.e., ⩾.5) and the prime–target associative strength is high (Besner and Smith, 1992, Borowsky and Besner, 1993, Robidoux et al., 2010, Stolz and Neely, 1995). When the relatedness proportion is low or the prime–target associative strength is low, priming is of the same magnitude for clear and degraded targets (Stolz & Neely, 1995). It is noteworthy that when the forward prime-to-target associative strength is high, the backward target-to-prime associative strength is also typically high (e.g., catdog, eastwest), which means there are strong symmetrical (SYM) associations between the prime and target. Because priming also occurs in the LDT when there is a backward asymmetric (BA) association between the prime and target (prime: small, target: shrink) (Koriat, 1981; see Kahan, Neely, and Forsythe (1999) for a summary of backward priming effects), an interesting issue, which the present research addresses, is whether the overadditive P × TDI obtained with symmetrically related primes and targets is being mediated by forward associations, backward associations or both.

This issue is of general theoretical interest because it has the potential to provide evidence for the distinction between prospective vs. retrospective processing mechanisms (Neely & Keefe, 1989) that is more direct than the prior indirect evidence for that distinction. Prospective theories of priming all assume that a prime’s facilitatory effect on related target processing is mediated by a mechanism that is initiated before the target is presented and gives the related target a processing “head start” relative to an unrelated target (Balota et al., 2008, Posner and Snyder, 1975). In activational prospective accounts, facilitation is produced either by the prime’s automatically preactivating the related target’s representation via spreading activation (e.g., Neely, 1977, Posner and Snyder, 1975) or by automatically preactivating the semantic features it shares with the prime in a parallel distributed memory network (e.g., Masson, 1995, Plaut and Booth, 2000). This preactivation occurs within less than 250 ms of the prime’s onset (e.g., Hutchison et al., 2001, Neely, 1977). Activation-based prospective priming theories can account for the P × TDI by postulating that preactivation of the target’s representation or semantic features either (a) speeds the rate at which the target’s visual letter features are extracted and to a greater degree when the target is visually degraded or (b) reduces the threshold for the amount of visual information required to identify the target. Because visual information is extracted more slowly for degraded than clear targets, the selective lowering of the threshold for related targets will result in degradation slowing recognition less for related than unrelated targets, with the result being a larger priming effect for degraded than clear targets.

One currently accepted prospective account of the P × TDI includes several subsystems, two of which are the orthographic input lexicon and the semantic system (Besner and Smith, 1992, Borowsky and Besner, 1993, Stolz and Neely, 1995). When the subject encounters a visually presented word, activation begins in the orthographic input lexicon and then proceeds to the semantic system. Activation in the semantic system produces activation of the prime’s associates which then feeds back into the orthographic input lexicon, thereby lowering the threshold required for identification of targets associatively related to the prime. Presumably, this feedback of activation from the semantic system to the orthographic input lexicon is under strategic control, which explains why the P × TDI occurs only with a high relatedness proportion.

In expectancy-based prospective priming theories, the prime’s presentation leads to the generation of an expectancy set containing items related to the prime (Becker, 1980, Becker and Killion, 1977). The generation of this expectancy set (a) requires attentional resources, (b) is under strategic control and more likely to occur the higher is the relatedness proportion and (c) requires 300 ms or more to complete (Hutchison et al., 2001, Neely, 1977). Becker and Killion’s (1977) expectancy-based prospective priming theory accounts for the P × TDI by assuming that related targets that are in the expectancy set can be identified with a more cursory analysis of their letter features than can words unrelated to the prime. Thus, as in the activation-based accounts, target degradation slows recognition less for related targets than for unrelated targets.

Unlike prospective priming theories, retrospective accounts assume that RTs are faster for related than unrelated targets because of processes that are initiated after the target is presented (e.g., Balota et al., 2008, Bodner and Masson, 1997, Bodner and Masson, 2001, Forster, 1979, Forster, 1981, Neely and Keefe, 1989, Neely et al., 1989, Ratcliff and McKoon, 1988). All of these retrospective accounts assume that the target’s presentation leads to the prime’s representation being retrieved and its semantic properties being matched to or combined with the target’s semantic properties. The most compelling evidence for a retrospective priming mechanism is the aforementioned BA priming effect (Kahan et al., 1999), in which RTs are faster for related than unrelated prime–target pairs when there is a BA association between the prime and its related target (prime: small, target: shrink). Clearly, prospective theories of priming cannot account for backward priming. For now, the details of exactly how and when retrospective prime retrieval leads to faster RTs for BA related prime–target pairs relative to unrelated pairs need not concern us. What is important is how these retrospective accounts could in general accommodate the P × TDI.

Most retrospective accounts of prime retrieval are based on intentional strategies that require attentional resources. (The one exception is Ratcliff and McKoon’s (1988), compound-cue theory, but the role of strategies need not concern us at this point.) When a target is degraded the extraction of its letter features is slowed. This could set into motion a consciously controlled compensatory mechanism that accesses contextual information provided by the prime that could aid in deciphering the degraded target (Balota et al., 2008, Stanovich and West, 1983). Although this mechanism presumably operates even when the target is not degraded (because BA priming occurs for clear targets in the LDT), as targets become increasingly degraded the subject’s reliance on retrieving information from the prime increases. Because of this, the contextual facilitation produced by the retrieval of a related prime will be greater when the target is degraded, such that the detrimental effects of target degradation will be diminished even more by a related prime, which is what is observed in the overadditive P × TDI.

Based on analyses of the full RT distribution, Balota et al. (2008) suggested that the P × TDI that occurs in the LDT and in pronunciation may indeed be mediated by retrospective priming mechanisms. The impetus for RT distributional analyses stems from the fact that RT distributions are often positively skewed. This creates a situation in which a focus on only differences between overall means can potentially be misleading. For example, prospective priming mechanisms such as spreading activation or expectancy could shift the overall RT distribution downward by a constant amount throughout the entire RT distribution. However, at the same time, retrospective priming mechanisms could also affect the skew of the distribution by exerting a greater effect in the tail containing the longest RTs. To assess this possibility, Balota et al. (2008) examined the effects of priming and target degradation throughout the full RT distribution with a prime–target stimulus onset asynchrony (SOA) of 800 ms. They found that the P × TDI was present for the fastest RT bin but increased in magnitude for the slowest RT bins for both the LDT and pronunciation. Balota et al. interpreted the finding that the P × TDI was significant even for the bin containing the shortest RTs as indicating that this interaction is being mediated by prospective priming mechanisms that begin to operate before the target appears. Balota et al. further argued that the increase in the magnitude of the P × TDI as RTs increased was due to the longer RTs having provided enough time for a slower acting retrospective priming process initiated after the target’s appearance to become operative, such that the related prime could further alleviate the detrimental effects of target degradation through retrospective priming mechanisms. As will now be discussed, by manipulating a variable (i.e., the direction of the prime–target association) that is directly linked to whether prospective or retrospective priming mechanisms can produce priming, the present experiment has the potential to allow us to isolate more clearly the contributions that prospective and retrospective mechanisms make to the P × TDI.

Section snippets

The present experiment

In the present experiment we sought to answer a previously unanswered simple empirical question. Will the robust overadditive P × TDI obtained in the LDT and in pronunciation when there are strong SYM associations between the prime and target (e.g., east west) also be obtained when there is only a strong forward asymmetric (FA) association from the prime to the target (e.g., keg beer) or only a strong BA association from the target to the prime (e.g., small shrink). This question remains

Results

Only data from correct responses were included in the RT analyses. For the LDT, data were discarded from 12 subjects whose overall accuracy was less than 90%. Statistical analyses were performed on the data from the remaining 168 subjects, with 42 subjects having been tested on each of the four counterbalancing lists. Pronunciation data for all subjects were analyzed, as accuracy exceeded 91% for all subjects. Because we wanted to retain as many RTs (especially long ones) as possible for the

General discussion

Our experiment has yielded six empirical contributions, five of which are entirely novel, that we believe have important theoretical implications. They are as follows:

(1) For both degraded and clear targets, in both the LDT and pronunciation SYM priming was equivalent to the sum of the priming effects produced by FA and BA pairs that had forward and backward associative strengths matched with those present in the SYM pairs. This new finding extends Balota and Paul’s (1996) finding from two

Conclusion

Contrary to most theoretical accounts of the P × TDI, which are based on prospective priming mechanisms (e.g., Becker and Killion, 1977, Besner and Smith, 1992, Borowsky and Besner, 1993, Plaut and Booth, 2000, Robidoux et al., 2010, Stolz and Neely, 1995), the present data clearly demonstrate that the nearly ubiquitous overadditive P × TDI previously reported for SYM pairs in the LDT and in pronunciation is mediated by a backward (target-to-prime) association and is not produced nor enhanced by a

Acknowledgments

This research was presented at the 49th Annual Meeting of the Psychonomic Society in November, 2008. We thank Thomas Dansereau, Sarah Deane, and Saskia Smeele for their assistance with data collection and David Balota, Derek Besner, Kit Cho, Keith Hutchison and Sachiko Kinoshita for their very helpful comments on an earlier draft of this manuscript.

References (61)

  • D.A. Balota et al.

    Summation of activation: Evidence from multiple primes that converge and diverge within semantic memory

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (1996)
  • D.A. Balota et al.

    The English Lexicon project

    Behavior Research Methods

    (2007)
  • C.A. Becker

    Semantic context and word frequency effects in visual word recognition

    Journal of Experimental Psychology: Human Perception and Performance

    (1979)
  • C.A. Becker

    Semantic context effects in visual word recognition: An analysis of semantic strategies

    Memory & Cognition

    (1980)
  • C.A. Becker et al.

    Interaction of visual and cognitive effects in word recognition

    Journal of Experimental Psychology: Human Perception and Performance

    (1977)
  • Besner, D. (2006). Visual language processing and additive effects of multiple factors on timed performance. A...
  • D. Besner et al.

    Models of visual word recognition: When obscuring the stimulus yields a clearer view

    Journal of Experimental Psychology: Learning, Memory and Cognition

    (1992)
  • G.E. Bodner et al.

    Beyond binary judgments: Prime validity modulates masked repetition priming in the naming task

    Memory & Cognition

    (2004)
  • R. Borowsky et al.

    Visual word recognition: A multistage activation model

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (1993)
  • D.C. Bourassa et al.

    When do nonwords activate semantics? Implications for models of visual word recognition

    Memory & Cognition

    (1998)
  • T.S. Braver et al.

    Explaining the many varieties of working memory variation: Dual mechanisms of cognitive control

  • M. Brown et al.

    Dissociative effects of stimulus quality on semantic and morphological contexts in visual word recognition

    Canadian Journal of Experimental Psychology/Revue canadienne de psychologie experimentale

    (2006)
  • J. Cohen

    Statistical power analysis for the behavioral sciences

    (1988)
  • W.P. Dunlap et al.

    Meta-analysis of experiments with matched groups or repeated measures designs

    Psychological Methods

    (1996)
  • M.E. Faust et al.

    Individual differences in information-processing rate and amount: Implications for group differences in response latency

    Psychological Bulletin

    (1999)
  • K.I. Forster

    Levels of processing and the structure of the language processor

  • K.I. Forster

    Priming and the effects of sentence and lexical constraints on naming time: Evidence for autonomous lexical processing

    Quarterly Journal of Experimental Psychology

    (1981)
  • S.W. Greenhouse et al.

    On methods in the analysis of profile data

    Psychometrika

    (1959)
  • K.A. Hutchison et al.

    Predicting priming at the item level

    Quarterly Journal of Experimental Psychology

    (2008)
  • K.A. Hutchison et al.

    With great expectations, can two “wrongs” prime a “right”

    Journal of Experimental Psychology: Learning, Memory, and Cognition

    (2001)
  • Cited by (35)

    View all citing articles on Scopus
    View full text