How Incidental Sequence Learning Creates Reportable Knowledge: the Role of Unexpected Events

How Incidental Sequence Learning Creates Reportable Knowledge: The Role of Unexpected Events

Dennis Ru¨nger Peter A. Frensch Berlin–Brandenburg Academy of Sciences and Humanities Humboldt-Universita¨t zu Berlin

Research on incidental sequence learning typically is concerned with the characteristics of implicit or nonconscious learning. In this article, the authors aim to elucidate the cognitive mechanisms that contribute to the generation of explicit, reportable sequence knowledge. According to the unexpected- event hypothesis (P. A. Frensch, H. Haider, D. Ru¨nger, U. Neugebauer, S. Voigt, & J. Werg, 2003), individuals acquire reportable knowledge when they search for the cause of an experienced deviation from the expected task performance. The authors experimentally induced unexpected events by disrupting the sequence learning process with a modified serial reaction time task and found that, unlike random transfer sequences, a systematic transfer sequence increased the availability of reportable sequence knowledge. The lack of a facilitative effect of random sequences is explained by the detrimental effect of random events on the presumed search process that generates reportable knowledge. This view is corroborated in a final experiment in which the facilitative effect of systematic transfer blocks is offset by a concurrent secondary task that was introduced to interfere with the search process during transfer.

Keywords: sequence learning, serial reaction time task, reportable knowledge, explicit knowledge, unexpected events

Individuals can learn about the sequential structure of environ- able to provide a verbal description of the sequential regularity mental events incidentally, that is, without prior intention to ap- after the training phase. This and related findings beget the ques- prehend a sequential regularity. For the most part, empirical re- tion of which cognitive mechanisms might be responsible for the search on sequence learning has been concerned with the generation of explicit knowledge during incidental sequence learn- characteristics of implicit learning. Sequence learning is said to be ing. However, researchers have only recently begun to approach implicit when it occurs in the absence of conscious or explicit this particular issue theoretically; pertinent empirical research is knowledge about the sequential regularity (cf. Erdelyi, 2004). still largely absent from the literature. Little theoretical value has been attached to the ubiquitous finding The purpose of this report is to test a central prediction of a that incidental sequence learning also creates explicit knowl- theoretical framework advanced by Frensch et al. (2003) to explain edge—at least in some participants. Moreover, a review of the the acquisition of explicit, reportable knowledge in incidental literature suggests that the generation of explicit sequence knowl- learning situations. The framework was dubbed the unexpected- edge varies systematically across experimental conditions (Fren- event hypothesis in reference to the central theoretical notion that sch et al., 2003). For example, Frensch, Lin, and Buchner (1998; unexpected events can trigger the generation of reportable knowl- Experiments 2a and 2b) manipulated the amount of training, the edge. Before we delineate the hypothesis in greater detail, we type of training (i.e., single or dual task training), and the type of review some alternative theoretical accounts that have been of- sequence to be learned with the serial reaction time (SRT) task fered in the literature. (Nissen & Bullemer, 1987). On each trial with the task, a partic- It is possible to identify two broad classes of theories on the ipant responds to a target that appears in one of several screen generation of explicit sequence knowledge—single-system and positions by pressing a spatially compatible response key. Re- multiple-system accounts. According to the single-system view, sponse locations on consecutive trials conform to a fixed sequence implicit and explicit knowledge are rooted in the same set of that is continuously recycled throughout the training phase. All learning mechanisms (e.g., Cleeremans, 2006; Cleeremans & three factors were found to influence whether participants were Jime´nez, 2002; Kinder & Shanks, 2003; Perruchet & Vinter, 2002; Shanks, Wilkinson, & Channon, 2003). In its most stringent form, the single-system view rejects the notion of separable knowledge bases altogether, that is, the distinction between implicit and Dennis Ru¨nger, Berlin–Brandenburg Academy of Sciences and Human- explicit knowledge. It is assumed that all markers of learning, be ities, Berlin, Germany; Peter A. Frensch, Department of Psychology, it faster responses to sequentially structured stimuli or verbal Humboldt-Universita¨t zu Berlin, Berlin, Germany. descriptions of the sequential regularity, provide different expres- This research was supported by Federal Ministry of Education and Research Grant 01GWS061. sions of the same underlying memory representations. Learning Correspondence concerning this article should be addressed to Dennis increases the quality of representations, which, in turn, leads to Ru¨nger, Berlin–Brandenburg Academy of Sciences and Humanities, Inter- improved performance in all available measures of learning (e.g., disciplinary Research Group “Functions of Consciousness,” Ja¨gerstr. 22/ Perruchet & Amorim, 1992; Perruchet, Bigand, & Benoit-Gonin, 23, D-10117 Berlin, Germany. E-mail: [email protected] 1997).

1011 1012 RU¨ NGER AND FRENSCH

Weaker versions of the single-system view allow for a partial then might, or might not, lead to discovery and, subsequently, to decoupling of alternative measures of sequence learning. For ex- verbal report of the regularity (cf. Dienes & Perner, 1999). ample, according to Shanks and collaborators (Shanks, 2005; According to the unexpected-event hypothesis, the search for Shanks & Perruchet, 2002; Shanks et al., 2003), response time the cause of an observed unexpected event will likely lead to (RT) priming and recognition, two indicators of sequence learning, discovery of the regularity when the observed unexpected event is reflect different transformation processes of the same underlying related to the incidentally experienced regularity and when no knowledge representations. The transformation of memory repre- other immediately attributable causes for the unexpected event sentations into manual responses might be affected less by random exist. Consider, for example, a typical experiment with the SRT noise, for instance, than the transformation of the same represen- task. At some point during the training phase, the regular response tations into recognition judgments (e.g., Shanks et al., 2003). sequence is briefly replaced by random sequences, so that se- Therefore, it is conceivable that experimental manipulations that quence learning can be assessed by contrasting average RTs to affect the quality of memory representations influence different regular and random response locations. When viewed in light of measures of learning to different extents. In this way, empirical the unexpected-event hypothesis, this standard indirect test of dissociations may emerge that are typically interpreted as evidence sequence learning takes on a new significance: Over the course for the existence of separable implicit and explicit knowledge of training, implicit sequence learning improves task performance bases. by reducing both average RT and error rate. When the participant According to Cleeremans (2006; Cleeremans & Jime´nez, 2002) is transferred to random response locations, responses are likely to sequence learning is a mandatory consequence of performing a become slower and more error-prone due to the unpredictability of sequentially structured task such as the SRT task. Learning pro- target locations. We surmise that this sudden deterioration in task duces, over time, increasingly strong, stable, and distinct repre- performance, if registered by the participant, might constitute an sentations of the underlying sequential regularity. However, ensu- unexpected event that can trigger the generation of explicit se- ing high quality of a representation is not a sufficient condition for quence knowledge. conscious awareness of its content. For knowledge to be explicit, If interpolated random sequences can indeed function as a it has to be re-represented in a metarepresentation (Cleeremans, trigger for the generation of explicit knowledge, one should expect to find more explicit knowledge in this situation compared with an 2006; cf. Dienes & Perner, 1999). Cleeremans’ framework thus experimental condition in which sequence learning proceeds with- demarcates explicit and implicit knowledge according to the pres- out disruption. Preliminary support for this hypothesis comes from ence or absence of relevant metaknowledge. Nevertheless, the a study by Buchner, Steffens, Erdfelder, and Rothkegel (1997). framework qualifies as a single-system account of sequence learn- Buchner and colleagues used an auditory version of the SRT task ing because metarepresentations are produced by the same learn- in which participants discriminated between four different tones ing mechanisms in the same representational systems as their instead of target locations. Sequences of tones (and their corre- first-order counterparts (Cleeremans, 2006). sponding response locations) conformed to a repeating 10-trial Proponents of the multiple-system view of sequence learning pattern. In Experiment 1, half of the participants were alerted to the argue that explicit and implicit learning are supported by different existence of the sequential regularity (intentional learning condi- memory systems. In its most stringent form, this view holds that tion), whereas the other half received no such clues (incidental the memory systems are “encapsulated,” that is, functionally in- learning condition). After six trial blocks of training with the SRT dependent and reliant on different brain systems (Reber & Squire, task, participants took a recognition test that required them to 1994, 1998). Along similar lines, Willingham (1998; Willingham distinguish between the training sequence, a novel systematic & Goedert-Eschmann, 1999) proposed that explicit and implicit sequence, and random distractor sequences. The recognition test sequence learning can proceed in parallel. Explicit learning re- data were used to identify a multinomial measurement model quires a strategic process—akin to high-level problem solving— whose parameters reflected the different cognitive processes that that selects and sequences spatial targets that are represented in presumably support sequence identification—recollection, syste- allocentric space. In contrast, implicit learning relies on target maticity detection, perceptual-motor fluency, and guessing. representations in egocentric space that are, in principle, inacces- Buchner et al. (1997) reported that the intentional learning sible to consciousness. Implicit learning is achieved through tuning condition and the incidental learning condition differed only with of a sequencing process that is engaged whenever a sequence of respect to the model’s recollection parameter, whose estimate was spatial targets is executed. larger for the intentional learners. We assume that the recollection Frensch and collaborators (2003; Haider & Frensch, 2005) con- parameter is closely linked to participants’ ability to verbally cur with the core assumption of the multiple-systems view that describe the training sequence. Consequently, the result is consis- explicit and implicit learning are carried out by dedicated memory tent with the much-replicated finding that participants acquire systems. According to the unexpected-event hypothesis, learning more explicit sequence knowledge in an intentional learning con- in the implicit memory system precedes explicit learning and dition than in an incidental one. generates memory representations that directly control behavior. Experiment 2 used the same learning orientation manipulation Unexpected events refers to discrepancies at the behavioral level but differed from Experiment 1 in one crucial respect: After six that occur when the learner performs a task in a way that deviates blocks of practice with the systematic training sequence, partici- from the performance she expects of herself. If registered, discrep- pants were exposed to two consecutive pseudorandom blocks in ancies between actual and expected behaviors trigger attributional which the tones were unpredictable. The training sequence was then processes in the explicit memory system (i.e., hypothesis testing, reinstated for another two blocks of trials. After the practice phase, attributing a cause to the observed behavioral discrepancy) that participants received the same recognition test as in Experiment 1. SEQUENCE LEARNING AND REPORTABLE KNOWLEDGE 1013

The model-based analysis of the recognition data revealed that the trial were determined by a comparison of colored rectangles and two learning conditions did not differ on any of the model param- not by the appearance of an asterisk at different screen locations. eters. Most importantly, the estimates of the recollection parameter Second, we employed six response locations instead of four. Third, for both learning conditions closely matched the estimate for the successive response locations in the training phase were governed intentional learners in Experiment 1. It appears that the additional by a repeating six-element sequence (e.g., 1-3-5-2-4-6) that con- random blocks increased the contribution of recollective processes, tained only unique associations among sequence elements. In Reed but only so in the incidental learning condition. Buchner et al.’s and Johnson’s (1994) terminology, such a sequence is a first-order (1997) results are thus fully consistent with the central assumption conditional (FOC) sequence because the response location on any of the unexpected-event hypothesis that unexpected events precip- given trial is sufficient to predict the response location on the next itate the generation of explicit knowledge. trial. FOC sequences, rather than second-order conditional (SOC) The main goal of the experiments summarized in this article was sequences, were chosen for the following reasons: Pilot studies to test the core assumption of the unexpected-event hypothesis showed that the use of a six-element FOC sequence together with more directly than had been done in the Buchner et al. (1997) the color-matching version of the SRT task allows one to observe study. Specifically, using a color-matching version of the SRT the full range of levels of reportable sequence knowledge in a task, we experimentally induced unexpected events by disrupting given sample of participants. In this way, we ensured that potential the sequential regularity in two different ways: In separate exper- effects of our experimental manipulations were not masked by iments we interpolated either random sequences or an alternate ceiling or floor effects due to sequential regularities that were too systematic sequence during the training phase with the SRT task. easy or too difficult to uncover. Moreover, we assumed that Either way, the unexpected-event hypothesis predicts an increase induced unexpected events would yield greater levels of reportable in the availability of reportable sequence knowledge compared sequence knowledge only if the sequential regularity were simple with a control condition in which training with the SRT task enough to be discovered and explicated by the postulated search proceeded without interruption. process. Our choice of participants’ verbal reports of the sequential Experiment 1a and its replication, Experiment 1b, make up the regularity as the measure of explicit sequence knowledge was baseline condition. Participants were trained on a repeating se- motivated by the following considerations. Research in the past quence of response locations without interruption. In Experiments decade has demonstrated that the validity of the concept of “im- 2a/b, we interpolated random transfer blocks in the training phase plicit” sequence learning depends on which particular direct test of with our modified SRT task. In Experiments 3a/b/c, participants explicit knowledge is employed (Frensch & Ru¨nger, 2003). Al- were switched from the systematic training sequence to a different though participants are often unable to verbally describe the se- but isomorphic transfer sequence. Thus, contrasting the available quential regularity of the SRT task (e.g., Curran & Keele, 1993; reportable sequence knowledge in Experiments 1a/b with report- Frensch, Buchner, & Lin, 1994; Frensch et al., 1998; Frensch & able knowledge generated in Experiments 2a/b and 3a/b/c allows Miner, 1994), they are typically capable of deploying sequence us to decide whether the disruption of the regular training sequence knowledge on other direct tests such as forced choice recognition affects the acquisition of reportable sequence knowledge. Finally, or generate tasks (e.g., Perruchet & Amorim, 1992; Shanks & with Experiments 4a/b, we examined whether the expected effects Johnstone, 1999; Shanks et al., 2003; Wilkinson & Shanks, 2004). of transfer blocks could be offset by having participants perform a Frensch and Ru¨nger (2003) concluded that only if verbal report is secondary task during the transfer blocks. chosen as the direct test of explicit sequence knowledge does implicit sequence learning constitute a valid concept. Secondly, verbal report arguably is the most valid measure of Experiments 1a/b explicit sequence knowledge. In contrast, other measures of explicit knowledge such as generate tasks or recognition tests have Experiment 1a and its replication, Experiment 1b, provide a been criticized repeatedly for also being sensitive to nonconscious baseline measure for reportable sequence knowledge when learn- sources of knowledge (for a concise summary of the main argu- ing proceeds without disruption. All subsequent experiments are ments, see Cohen & Curran, 1993). For example, Norman, Price, compared against this baseline condition. Participants practiced the and Duff (2006) argued that nonconscious sequence knowledge modified SRT task in which successive response locations con- induces conscious feelings of rightness or wrongness associated formed to a six-element FOC sequence. Immediately after the with the execution of response sequences that participants are able practice phase, we assessed the available reportable sequence to deploy on recognition tests. A more comprehensive theoretical knowledge by requesting participants to verbally describe the argument for the use of verbal report as the principal measure of serial order of response locations in the training phase. In order to conscious sequence knowledge can be found in Ru¨nger and Fren- quantify the amount of sequence knowledge expressed in a verbal sch (2007). report, we devised a new scoring procedure that allows one to integrate different levels of statistical structure contained in a Overview of the Experiments sequence. The procedure assigns a score to a verbal report depending on the degree of structural overlap between the verbalized We report four experiments that describe the effects of different sequence and the actual training sequence. The magnitude of the types of disruption of the training sequence on the generation of score is determined by the likelihood of achieving a particular reportable sequence knowledge with a modified SRT task. Our degree of overlap by mere guessing, with low guessing probabil- version of the task differs in three main ways from the original ities affording larger verbal knowledge scores (see Method section version of Nissen and Bullemer (1987). First, responses on each for details). 1014 RU¨ NGER AND FRENSCH

Method warm-up trials with the SRT task, during which response locations were determined randomly with the constraint that a response Participants. Forty-four participants, predominantly students location could not be used on consecutive trials. The warm-up at Berlin universities, were recruited to take part in Experiment 1a. trials were repeated if a participant committed mistakes more than They were paid 8 € (about $9) for participation. Three participants 20% of the time. had to be excluded from the main analyses. One student reported Experiments 1a/b each comprised 10 training blocks during that he had previously participated in a similar experiment. A which participants performed the six-choice color-matching ver- second participant claimed that she had already expected to en- sion of the SRT task. Each block consisted of 120 trials, for a total counter some form of hidden regularity before she even started to of 1,200 trials, or 200 repetitions of the six-element FOC sequence. perform the SRT task. Data from a third participant were elimi- On each trial, participants had to determine which of the six target nated because he deviated from the finger-to-response-key map- rectangles at the bottom of the screen matched the color of the ping that was described in the instructions. The remaining participants, 20 women and 21 men, ranged in age from 18 to 38 years large rectangle on top and to press the response key that was (M ϭ 24.98, SD ϭ 4.52). assigned to that target rectangle. They responded to target Rect- Fifty-three women and 45 men, predominantly university stu- angles 1, 2, and 3 with the ring, middle, and index fingers of their dents in Berlin, participated in Experiment 1b. They were 18 to 33 left hands, respectively, and to target Rectangles 4, 5, and 6 with years old (M ϭ 24.52, SD ϭ 3.17). Payment varied between 7 € the index, middle, and ring fingers of their right hands, respec- and 9 € (approx. $8 to $10), depending on individual performance tively. The first target location in each trial block was determined in a recognition test in the final phase of the experiment. randomly with the constraint that the response location had to Apparatus. Stimulus presentation, RT measurement, and re- differ from the response location on the final trial of the previous sponse recording were implemented on IBM-compatible PCs with block. Thereafter, response locations were chosen according to the 33-cm color monitors and standard German QWERTZ keyboards. sequential regularity. A trial ended when a participant pressed one The viewing distance was about 60 cm. A large colored rectangle of the six response keys. In the case of an erroneous response, (8 cm wide and 6 cm high) and six small colored rectangles (2.5 participants heard a beep for a duration of 100 ms. When the cm wide and 2.5 cm high) were displayed simultaneously on a response key was released, the screen blanked after 200 ms, and light gray background. The large rectangle was centered in the top the next trial began 200 ms later. The total response-stimulus half of the display, 3 cm below the top of the monitor. The six interval was therefore 400 ms. Response latencies were measured small rectangles—subsequently referred to as target Rectangles 1 from the onset of a trial to the depression of the response key. to 6—appeared 3.5 cm from the bottom of the monitor and 9 cm Participants received feedback about their mean RTs and error below the top rectangle. They were separated horizontally by 2 cm, rates after Blocks 1 to 9. If the error rate exceeded 10% for the first except for the third and fourth rectangles, which were spaced 3 cm time, participants were prompted to make fewer mistakes. After a apart. Each target rectangle was mapped to a spatially compatible second block with more than 10% errors, participants were warned response key on the computer keyboard (X, C, V, B, N, M). The that the experiment would be discontinued if they did not lower their response keys were labeled 1 to 6 from left to right. The same six error rates. Termination of the experiment occurred after a third block colors (green, red, cyan, dark gray, magenta, and blue) were used with more than 10% errors. on every trial, but each rectangle changed its color pseudoran- Upon completion of the final block of trials, the experimenter domly from one trial to the next. returned to the testing cubicle and assessed participants’ reportable Materials. Response locations during the training phase with knowledge about the training sequence in a semistructured inter- the modified SRT task were governed by a repeating six-element view. The experimenter presented a cue card with six boxes FOC sequence. Each of the six possible response locations oc- labeled 1 to 6 and told the participant that the boxes represented curred once in the sequence (e.g., 1-5-2-6-4-3). Consequently, the the six response keys that corresponded to the six target rectangles. response location on any given trial was predictive of the response The experimenter then declared that responses in the training location on the next trial. The modified SRT task contained no phase followed a regular pattern and asked the participant to further sequential regularities other than the repeating sequence of verbally describe the serial order of response locations by referring response locations. Each participant was randomly assigned to a to the labels on the cue card. In order to prevent any spontaneous different six-element sequence that was drawn from a pool of 70 typing activity, we had participants cross their arms in front of sequences. The sequences were permutations of the six response locations that satisfied the following conditions: First, “runs” of three or more adjacent response locations (e.g., 1-2-3, 1 In Experiments 1b, 3a/b/c, and 4a/b, we excluded all sequences with 2-3-4-5, 6-5-4) were not permitted. Second, adjacent response one within-hand transition between adjacent response keys (1-2, 2-1, 2-3, locations (e.g., 1-2, 3-4, 6-5) could not occur more than twice 3-2, 4-5, 5-4, 5-6, 6-5) and employed only sequences with either zero or in a sequence.1 two such transitions. In Experiments 3a/b/c and 4a, a participant’s transfer Procedure. Participants were told that they were taking part in sequence contained zero within-hand transitions if his or her training a simple-choice RT experiment designed to see how practice sequence contained two such transitions and vice versa. There is no indication that the number of within-hand transitions in a training sequence affects the ability to discriminate colors. They were not informed affects the acquisition of reportable sequence knowledge. We conducted of the fact that correct response locations during the training phase two separate analyses of variance on verbal knowledge scores for Exper- followed a repeating pattern. Learning of the sequential regularity iments 1a and 2a/b, and for Experiments 1b, 3a/b/c, and 4a/b with “Number was thus incidental. Instructions for the SRT task were presented of Transitions” and “Experiment” as between-subjects factors. All Fs for onscreen in the presence of the experimenter and followed by 40 effects involving “Number of Transitions” were equal to or smaller than 1. SEQUENCE LEARNING AND REPORTABLE KNOWLEDGE 1015 their upper body and hold a pencil in each hand while they Structural overlap reflects the degree of correctness of a verbal- attempted to report the sequence of response locations. ization and is defined in terms of n-tuples of successive response Note that we deliberately deviated from the common strategy of locations that are shared between the verbalized sequence and the opening the assessment of verbalizable sequence knowledge with training sequence. Take, for example, a participant who received general questions about the training phase (e.g., Frensch et al., the training sequence 1-3-5-2-6-4. The sequence can be parsed into 1998; Willingham, Greeley, & Bardone, 1993). In our view, a clear six consecutive pairs, triplets, quadruples, and quintuples, respec- focus on the relevant serial-order information ensures maximum tively. Let’s assume the participant reported the sequence 5-2-4-6-1-3 sensitivity of the verbal report measure (Ru¨nger & Frensch, 2007). after the training phase. This sequence contains one quadruple (1-3- A final question prior to debriefing concerned any preexisting 5-2), two triplets (1-3-5, 3-5-2), and three pairs (1-3, 3-5, 5-2) from the notions regarding the purpose of the experiment and, in particular, training sequence. regarding hidden regularities. If a participant indicated a priori Next, one needs to determine how likely it is that a particular expectations about a hidden regularity, he or she was excluded pattern of structural overlap could be obtained by mere guessing. from further analyses. To do so, we pseudorandomly generated 2 ϫ 106 sequences of six For reasons external to the present investigation, participants in in length with the single constraint that immediate repetitions of Experiment 1a performed a free generation task after the assess- response locations were not allowed, and for each sequence we ment of reportable sequence knowledge, whereas participants in determined the number of n-tuples shared with the reference Experiment 1b received a recognition test and filled out a person- sequence 1-2-3-4-5-6.2 We found that each sequence generated ality questionnaire. Due to the theoretical and methodological corresponds to 1 of 12 patterns of structural overlap that are reasons outlined above, participants’ verbal reports were the only displayed in Table 1 together with the probabilities for each measure of declarative sequence knowledge that was recorded pattern. In general, a greater match with the reference sequence is consistently in all experiments. We therefore do not report the associated with a lower guessing probability. results for other direct tests of sequence knowledge, whose appli- It is now possible to assign a score to a verbalization based on the cation varied across experiments. n-tuples shared between the verbalized sequence and the training Quantifying verbal reports. In order to quantify the sequence sequence and the known probability for that particular pattern of knowledge contained in participants’ verbal reports, we adopted a structural overlap. We decided to scale the probabilities according novel scoring procedure. Conventionally, verbal reports are scored ϭ ϫ to the equation Scorepattern i 100 (1 – Probabilitypattern i), so with respect to the number of correctly recalled sequence segments that low guessing probabilities afford high verbal knowledge of a specific length (i.e., pairs for FOC sequences, triplets for SOC scores. For example, a verbalized sequence that contains two pairs sequences) or with respect to the length of the single largest and no higher-order tuple receives a score of 88.64. sequence fragment included in the report. However, these scoring So far, we have assumed that a verbal report consists of a single procedures suffer from a significant shortcoming: They do not sequence of six response locations, and this is indeed true for the fully capture a participant’s knowledge about the structure of the majority of participants’ verbalizations in our study: Across the four repeating sequence. Consider, for example, a participant who is experiments, 83% of verbal reports consisted of a single six-element found to have recalled three out of six pairs of an FOC sequence. sequence. However, some participants reported more than one se- What is missing from this particular analysis is any information quence and/or fewer than six elements per sequence (16% of all about higher-order statistical structure—whether the three pairs verbal reports). Several of these reports consisted of a single se- form a single triplet accompanied by a pair, two triplets, a single quence with fewer than six elements (e.g., “1-6”). In this case, we quadruple, or no higher-order tuple. Conversely, let’s assume that pseudorandomly extended the verbalization to a six-element se- the longest segment of the FOC sequence contained in a partici- 6 times, scored each sequence, and calculated the mean pant’s report is a triplet. This, however, can mean any of three quence 10 things: that the participant recalled a single triplet, two triplets, or verbal knowledge score. a triplet accompanied by a pair. If a participant reported more than one sequence, we first The example illustrates that the sequential regularities that are checked whether the verbalizations could be recombined into a most commonly used in research on sequence learning (i.e., FOC and SOC sequences) contain different levels of statistical structure 2 Constraints during random sequence generation define chance perfor- that are partially independent. Conventional scoring procedures mance in the absence of sequence knowledge and thus reflect a priori focus on a particular level of structure, thereby discounting infor- assumptions about the type of knowledge generated during sequence mation that is contained at higher or lower levels. Consequently, it learning proper. For example, an additional constraint could be to include would be desirable to have a verbal report measure that takes into all six response locations in a randomly generated six-element sequence. account all levels of sequence structure simultaneously. In the By so doing, one would subscribe to two additional assumptions: First, following, we describe a two-step scoring procedure that meets having learned that in six consecutive trials during the training phase each this goal. response location occurred exactly once does not constitute sequence The first step in scoring a verbal report is to determine the learning (cf. Lee, 1997). Second, differences in the latter type of knowledge between verbalizers and nonverbalizers are negligible. Since we do structural overlap of the verbalized sequence with the participant’s not endorse either assumption, we forwent this additional constraint. More- actual training sequence. In a second step, we assign a score to the over, our investigation focuses on relative differences in reportable se- verbalization based on the probability of generating this particular quence knowledge. Hence, the precise definition of chance performance is pattern of structural overlap by random guessing. The score thus of somewhat lesser importance than it would be if we sought to determine reflects the basic notion that a low guessing probability of a the absolute level of reportable sequence knowledge in a given sample of particular structural pattern indicates high verbal knowledge. participants. 1016 RU¨ NGER AND FRENSCH

Table 1 We believe that conflicting intuitions are generally misleading Patterns of Structural Overlap With Reference Sequence, because, unlike the proposed verbal knowledge score, we fail to Associated Probabilities, and Verbal Knowledge Scores take into account different levels of sequence structure simultaneously. To illustrate this point with yet another example: It is a Pairs Triplets Quadruples Quintuples Probability Score difficult to decide, intuitively, which of the following two verbal- 0 0 0 0 .2717 72.83 izations indicates more sequence knowledge: three independent 1 0 0 0 .4307 56.93 pairs or two pairs that conjointly make up a triplet. Our verbal 2 0 0 0 .1136 88.64 knowledge score provides a clear-cut answer: As one is less likely 2 1 0 0 .1129 88.71 to report three independent pairs by chance, the corresponding 3 0 0 0 .0045 99.55 verbal knowledge score is higher (see Table 1). 3 1 0 0 .0281 97.19 3 2 0 0 .0019 99.81 In sum, the key feature of the proposed verbal knowledge score 3 2 1 0 .0288 97.12 is its ability to integrate different levels of sequence structure. 4 2 0 0 .0003 99.97 Because of its comprehensiveness, the score is preferable to con- 4 2 1 0 .0006 99.94 ventional measures of reportable sequence knowledge, which are 4 3 2 1 .0064 99.36 6 6 6 6 .0003 99.97 limited to the analysis of a specific level of sequence structure. Nevertheless, in this article we juxtapose our verbal knowledge a The verbal knowledge score for a particular pattern is derived from the score with two traditional measures of reportable knowledge— probability according to this equation: Score ϭ 100 ϫ (1 – Probability). “percentage of pairs recalled correctly” (PPC) and “length of the longest sequence segment” (LSS) contained in a report—in order to establish a continuity with the common practice of analyzing single six-element sequence. If this was possible (e.g., reporting verbal reports. “4-6” and “1-5”), we pseudorandomly generated 106 six-element sequences that contained all verbalizations (e.g., 4-6-2-3-1-5, 4-6- Results and Discussion 3-1-5-2, etc.), scored each sequence, and used the mean score as the participant’s verbal knowledge score. Mean verbal knowledge scores for all experiments reported in If it was not possible to recombine the verbalizations into a this article are shown in Figure 1. Since Experiment 1a and its single sequence (e.g., reporting “2-5-3-6-1-4” and “5-4-3”), we replication, Experiment 1b, yielded highly similar results, we determined the verbal knowledge score for each verbalization pooled the two experiments in all subsequent analyses. In the separately. If a verbalization contained fewer than six response context of the present investigation, the precise numerical value of locations, we again pseudorandomly extended the sequence 106 the mean verbal knowledge scores in Experiments 1a/b, 81.46, has times and calculated the mean score for that verbalization. The little significance by itself, because the key question is whether participant’s final score corresponded to the mean verbal knowl- experimentally induced unexpected events in subsequent experi- edge score across all verbalizations. ments will increase the available reportable knowledge. However, A small fraction of verbal reports (1%) contained sequences it is informative to consider two anchor points on our verbal with seven or eight elements. In these rare cases, we parsed the sequences into two or three six-element sequences, respectively, and applied the scoring procedure for six-element sequences described above. Finally, it is necessary to determine the verbal knowledge score for participants who were unable to provide a verbal description of the serial order of response locations, typically because they were unaware of the mere existence of a systematic response sequence. It seems reasonable to require that the complete lack of reportable sequence knowledge afford a score that is lower than the score assigned to any type of verbalized sequence, independently of its length and correctness. However, it is difficult to vindicate a precise numerical value for such a verbal knowledge score below the lowest score for a verbalized sequence (i.e., below the score for a sequence with a single correct pair). Therefore, we chose to employ the same score that other participants received for a single correct pair in a verbalized six-element sequence (i.e., 56.93). A close examination of the scores summarized in Table 1 may create the impression that some scores are intuitively wrong. For example, two independent triplets receive a higher score than do Figure 1. Mean verbal knowledge scores for Experiments 1a/b (baseline two triplets that conjointly make up a quadruple; the score is condition), 2a/b (interpolated random sequences), 3a/b/c (interpolated iso- higher for reporting nothing correct than for reporting one pair morphic regular sequence), and 4a/b (concurrent secondary task on trial correct. However, all of the scores strictly conform to the basic blocks with random sequences or isomorphic regular sequence). Error bars principle that sequence fragments (or combinations of fragments) depict Ϯ1 standard error of the mean. interpol. ϭ interpolated; sequ. ϭ that are harder to guess receive a higher verbal knowledge score. sequence. SEQUENCE LEARNING AND REPORTABLE KNOWLEDGE 1017 knowledge scale: If each participant correctly reported the entire during the training phase was motivated by the following consid- six-element sequence, the mean verbal knowledge score would be erations: Participants are likely to differ in the efficiency of their 99.97. Contrariwise, if each participant were merely guessing, the sequence-learning processes and, consequently, in the rate of expected score would be 71.33 (because chance would have it that learning. Effects of random blocks might be contingent on the 27% of the participants would not report a single correct pair, 44% amount of sequence learning prior to the transfer blocks. Conse- one pair, 11% two pairs, and so on). quently, the two consecutive random blocks in Experiment 2a For comparison purposes, Figure 2 displays two conventional might come “too early” for some participants to be effective. By measures of reportable sequence knowledge. The first measure, distributing random transfer blocks over the training phase, we PPC, ranges from 0% to 100%. The second measure, LSS, ranges aimed to maximize their effect on reportable sequence knowledge. from 0 (no verbal report)to6(entire sequence reported correctly). In sum, we expected to obtain a qualitatively similar result to To anticipate: The two alternative measures fully support the Buchner et al. (1997). The addition of random blocks of trials in pattern of results that we obtained with our novel verbal knowl- Experiments 2a/b should increase the availability of reportable edge score. sequence knowledge relative to Experiments 1a/b, in which sequence learning proceeded uninterruptedly. Experiments 2a/b Experiments 2a/b were designed to assess the effect of random Method transfer blocks on the availability of reportable sequence Participants. Forty-two participants initially took part in Ex- knowledge. According to the unexpected-event hypothesis, in- periment 2a. For one participant, the experiment was automatically terpolated random sequences should lead to unexpected perfor- discontinued after the third block of trials in the training phase mance decrements, which can trigger the generation of report- because the person’s error rate was greater than 10%. The remain- able knowledge. Consequently, we expected to find greater ing participants were 28 women and 13 men who ranged in age amounts of reportable sequence knowledge in Experiments 2a/b from 18 to 32 years (M ϭ 23.98, SD ϭ 3.7). Sixteen women and than in Experiments 1a/b. 26 men participated in Experiment 2b. One participant had to be Participants received the same amount of training with the replaced because she reported great difficulties in pressing the systematic response sequence as did participants in the baseline response keys and took an unusual amount of time to complete the condition. In Experiment 2a participants were exposed to two training phase with the SRT task. Participants ranged in age from additional random blocks after the first five blocks of trials with 18 to 35 years (M ϭ 23.4, SD ϭ 3.41). The great majority of the systematic training sequence. They were then switched back to participants in Experiments 2a/b were university students. They the training sequence for another five blocks of trials. Experiment were paid 8 € (approx. $9). 2a thus mimics the standard procedure for the indirect assessment Materials. The same FOC sequences as in Experiments 1a/b of sequence learning. were used. During transfer blocks, response locations were deter- In Experiment 2b, we inserted single random blocks after two, mined pseudorandomly with the constraint that the same response four, six, and eight blocks of training with the systematic sequence. location could not be used on consecutive trials. Interpolating random blocks of trials at different points in time Procedure. Participants in Experiment 2a completed 12 blocks of 120 trials with the modified SRT task. Blocks 1 to 5 contained the training sequence. During Blocks 6 and 7, response locations were determined pseudorandomly. The training sequence was reintroduced on Block 8 for the remainder of the training phase. Thus, participants in Experiment 2a performed a total of 1,200 trials with the training sequence and 240 random trials. Experiment 2b comprised 14 training blocks, 4 of which were random (Blocks 3, 6, 9, and 12). The remaining blocks of trials contained the systematic training sequence. The total numbers of systematic and random trials in Experiment 2b were 1,200 and 480, respectively. After the assessment of reportable sequence knowledge, which was identical to the assessment in Experiments 1a/b, participants in Experiments 2a/b completed a free generation test. However, we only report the results for reportable sequence knowledge.

Results and Discussion

A significance criterion of ␣ϭ.05 was chosen for all statistical Figure 2. Alternative measures of sequence knowledge derived from tests reported here and in the following experiments. Figure 3 participants’ verbal reports. Empty circles stand for the mean percentage of compares mean RTs obtained over the training phase in the base- pairs recalled correctly (PPC) in a report. Filled circles depict the average line condition (see Experiments 1a/b) and Experiments 2a/b. A length of the longest segment of the training sequence (LSS) contained in noticeable increase in RTs on random blocks of trials relative a report. Error bars depict Ϯ1 standard error of the mean. to regular blocks indicates that participants learned about the 1018 RU¨ NGER AND FRENSCH

participants were quite surprised when they learned from the experimenter about a regular response sequence in the training phase, presumably because they never engaged in a search for a regularity. We argued that unexpected events trigger the search process that generates reportable sequence knowledge. It follows that the experimental induction of unexpected events should re- duce the number of participants who completely lacked reportable sequence knowledge (nonverbalizers) because they never engaged in a search for a regularity. However, it can be seen in Table 2 that the percentage of nonverbalizers in Experiments 2a/b was similar to the percentage of nonverbalizers in the baseline condition. In order to formally compare the proportion of nonverbalizers in Experiments 1a/b with the proportion in Experiments 2a/b, we pooled the two random groups and computed a chi-square test on the raw frequencies of participants with and without reportable sequence knowledge. We found that the proportion of nonverbal- Figure 3. Mean of individual median response times (RTs) in ms per trial izers in Experiments 1a/b and 2a/b did not differ significantly from block in the training phase with the serial reaction time task. Filled dots each other, ␹2(1, N ϭ 222) ϭ 0.26, p Ͼ .61. represent the baseline condition (see Experiments 1a/b). Participants in In summary, there is no indication that the interpolation of Experiment 2a received an additional two random transfer blocks, and random blocks in the training phase with the modified SRT task participants in Experiment 2b an additional four random transfer blocks. affected the availability of reportable sequence knowledge. Many Error bars depict Ϯ1 standard error of the mean. Exp. ϭ experiment. factors can be held responsible for our failure to conceptually replicate Buchner et al.’s (1997) finding. Among these are various procedural differences between Buchner et al.’s and our experi- systematic response sequence. In order to determine whether se- ments, for example, the use of a six-choice color-matching version quence learning expressed in participants’ RTs differed between of the SRT task in conjunction with FOC training sequences Experiments 1a/b and Experiments 2a/b, we performed an analysis instead of the four-choice tone version together with 10-trial of variance (ANOVA) with block (10 levels) as a within-subject ambiguous sequences. As a first step toward clarification, it would variable and experiment (3 levels) as a between-subjects variable be helpful to replicate Buchner et al.’s experiments with our verbal on individual median RTs in the 10 systematic blocks of trials in report measure of conscious sequence knowledge instead of a each group. Due to violations of sphericity, multivariate Fs for all recognition test. effects involving within-subject variables are reported. The ANOVA revealed significant main effects of block, F(9, 211) ϭ Experiments 3a/b/c 183.07, p Ͻ .001, and experiment, F(2, 219) ϭ 4.11, MSE ϭ 215636.91, p ϭ .018. Figure 3 shows that particularly slow re- With the next experiments we investigated the effects of a sponses in Experiment 2a were responsible for the significant main different type of disruption of the training sequence on reportable effect of experiment. Since this difference was already apparent in sequence knowledge. When viewed in light of the unexpected- the initial blocks of training, we attribute it to random sampling event hypothesis, our failure to observe an effect of interpolated error. More importantly, a significant Block ϫ Experiment inter- random blocks in Experiments 2a/b might have two reasons: First, action would indicate that the RT speed-up over training and, thus, random blocks may not have created unexpected events of suffi- sequence learning was more pronounced in one of the experimen- cient strength to induce the generation of reportable knowledge. tal groups. This interaction, however, was not significant, F(18, Note that during random transfer blocks, on average one sixth of 422) ϭ 1.36, p Ͼ .149. Consequently, we did not obtain any the response transitions still conformed to the training sequences. RT-based evidence that the interpolation of random blocks in The second explanation focuses on the putative search process. Experiments 2a/b affected sequence learning. Perhaps random transfer sequences did elicit unexpected events Figure 1 shows that mean verbal knowledge scores in Experi- and triggered a search for their proper cause. However, if the ments 2a/b did not differ from the verbal knowledge score in search process was initiated and carried out during random transfer Experiments 1a/b. A planned contrast in an ANOVA3 that com- blocks, a participant necessarily had to conclude that there was no pared the two groups of participants that were transferred to hidden task regularity. Consequently, the search did not lead to random blocks with the baseline condition confirmed this impres- verbalizable sequence knowledge about the training sequence. sion, F(1, 431) ϭ 0.01, MSE ϭ 359.26, p Ͼ .93. The two In Experiments 3a/b/c, we addressed both issues by interpolat- alternative indexes of reportable knowledge displayed in Figure 2, ing an alternate isomorphic sequence instead of random sequences PPC and LSS, provide convergent evidence that the two random ϭ groups did not differ from the baseline condition, F(1, 431) 3 ϭ Ͼ ϭ The error term for this and all subsequent planned contrasts is derived 0.46, MSE 1993.31, p .49, for PPC, and F(1, 431) 0.57, from the omnibus ANOVA for the seven experiments reported in this ϭ Ͼ MSE 6.65, p .45, for LSS. article. The one-way ANOVA with experiment (seven) as between- In a supplemental analysis of participants’ verbal reports, we subjects variable yielded a nonsignificant effect of experiment, F(6, 431) ϭ focus on the percentage of participants who were unable to provide 1.78, MSE ϭ 359.26, p ϭ .1. The error terms for the alternative verbal any verbal description of the training sequence. Typically, these knowledge scores, PPC and LSS, were computed accordingly. SEQUENCE LEARNING AND REPORTABLE KNOWLEDGE 1019

Table 2 prior knowledge of a hidden regularity. The remaining partici- Percentage of Participants in Experiments 1–4 Without pants, 29 women and 15 men, were 18 to 38 years old (M ϭ 23.55, Reportable Sequence Knowledge (Nonverbalizers) SD ϭ 4.95). The great majority of participants in Experiments 3a/b were Percentage of students at Berlin universities who were paid 8 € (approx. $9). Experiment nonverbalizers SE Participants in Experiment 3c were first-semester psychology stu- 1a/b 30.9 3.9 dents at Humboldt-Universita¨t zu Berlin, who received course 2a 26.8 7.0 credit. 2b 28.6 7.1 Materials. The same FOC training sequences were used as in 3a/b 13.1 3.7 Experiments 1a/b. For each subject, an alternate FOC sequence 3c 15.9 5.6 4a 31.8 7.1 was randomly chosen from the same pool as the training sequence 4b 38.6 7.4 with one major constraint: The alternate FOC sequence could not share any transitions between adjacent sequence elements with the training sequence. For example, if Response Location 2 preceded (cf. Reed & Johnson, 1994). A participant’s alternate sequence was Response Location 1 in the training sequence, then Response chosen from the original pool of sequences so that each response Location 2 had to be followed by a location other than 1 in the transition differed from the primary training sequence. An impor- transfer sequence. tant difference between shifting participants to random sequences Procedure. Participants in Experiments 3a/b completed 14 and shifting them to a different systematic sequence is that in the blocks of 120 trials with the modified SRT task. During Blocks 3, latter case, sequence learning is possible on transfer blocks. Thus, 6, 9, and 12, response locations were determined by an alternate whenever an unexpected event triggered a search for its cause FOC sequence. The remaining blocks of trials contained the FOC during the training phase, the search could, in principle, lead to training sequence. Thus, participants received a total of 1,200 trials discovery of a sequential regularity. We expected that some par- with the training sequence and a total of 480 trials with the transfer ticipants would acquire reportable sequence knowledge about the sequence. systematic transfer sequence before they became aware of the In Experiment 3c participants performed 10 blocks of 144 trials training sequence. However, once alerted to the existence of task with the modified SRT task. The first and the final block were regularities, they would likely search for additional regularities composed of trials that were structured according to the training when the known regularity no longer applied and thus also dis- sequence. Blocks 2 to 9 contained 114 trials that conformed to the cover the primary training sequence. training sequence and 30 consecutive trials that followed the To test our hypothesis, we repeated Experiment 2b with one transfer sequence. The onset of the five repetitions of the transfer crucial procedural modification. In Experiment 3a, we replaced the sequence within a block varied randomly between Trial 1 and Trial four random transfer blocks with transfer blocks during which 115. Shifts from the training sequence to the transfer sequence and response locations were determined by an alternate FOC sequence. back to the training sequence were realized in such a way that Experiment 3b was an exact replication of Experiment 3a. Finally, consecutive response locations always conformed to either the with Experiment 3c we implemented the shift from the training training sequence or the transfer sequence. All in all, participants sequence to an alternate FOC sequence in a different way. Instead in Experiment 3c performed 1,200 trials with the training sequence of employing transfer blocks, we inserted 30 consecutive trials and 240 trials with the transfer sequence. with the alternate FOC sequence (i.e., five sequence repetitions) at Since participants were now able to learn two sequential regu- random positions in 8 out of 10 blocks of trials with the training larities (the training sequence and the transfer sequence), we sequence (cf. Curran, 1997; Perruchet et al., 1997). introduced three minor procedural changes in the assessment of reportable sequence knowledge in order to disambiguate partici- Method pants’ verbal reports. First, the experimenter started the interview by remarking (in German) that “most of the time, key presses Participants. Forty-three participants initially took part in Ex- during the experiment followed a rule” instead of saying that “key periment 3a. Due to technical errors we had to discard the data of presses during the experiment followed a rule.” We deemed this 2 participants. A 3rd participant had to be excluded from further change of wording necessary in order to avoid confusing those analyses because he had expected a hidden regularity before he participants who were aware of both the training and the transfer started with the training phase of the modified SRT task. The sequence. Second, if the participant reported more than one remaining participants were 20 women and 20 men who ranged in sequence (e.g., “There was the sequence 1-3-5-2-4-6, and some- age from 18 to 32 years (M ϭ 23.35, SD ϭ 3.53). times I also typed 1-2-5-6-3-4”), the experimenter asked which Twenty-seven women and 18 men initially participated in Ex- of the sequences was in effect at the end of the training phase. periment 3b. The data from one participant had to be discarded This sequence was then taken as the verbal report for the because she claimed to have expected a hidden regularity. The training sequence. Third, if, and only if, a participant described remaining participants ranged in age between 18 and 30 years a single sequence that was identical with the transfer sequence, (M ϭ 22.48, SD ϭ 3.22). the experimenter pointed out the existence of yet another se- Forty-six participants were recruited for Experiment 3c. The quence and asked for a verbal report of that sequence. In all other data for 2 participants had to be excluded from further analyses. aspects, the assessment of reportable sequence knowledge about One participant did not follow the finger-to-key assignment as the training sequence was identical to the procedure employed in described in the instruction, and the other reported having had the previous experiments. One might argue that the procedural 1020 RU¨ NGER AND FRENSCH changes provided additional retrieval cues for the training ments 3a/b and 3c, we performed an ANOVA with block (10 sequence that were not available to participants in Experiments levels) as a within-subject variable and experiment (3 levels) as 1a/b and 2a/b. However, in those experiments verbal recall of the a between-subjects variable on median RTs to targets that training sequence was also not subject to interference from a followed the training sequence. The analysis revealed signifi- systematic transfer sequence. Therefore, we do not believe that the cant main effects of block, F(9, 256) ϭ 60.38, p Ͻ .001, and slightly modified verbal report measure in Experiments 3a/b/c experiment, F(2, 264) ϭ 4.32, MSE ϭ 869481.36, p ϭ .014. overestimates the available reportable sequence knowledge rela- The Block ϫ Experiment interaction was also significant, F(18, tive to the earlier experiments. 512) ϭ 2.22, p ϭ .003. After the assessment of reportable sequence knowledge, par- To further explore the significant interaction, we contrasted ticipants in Experiments 3a/b/c completed a recognition test. As Experiment 1a/b with Experiments 3a/b and 3c in separate before, we report the results for only reportable sequence ANOVAs. Experiments 3a/b were indistinguishable from Experi- knowledge. ments 1a/b both with respect to mean RT levels, F(1, 221) ϭ 2.16, MSE ϭ 218039.03, p Ͼ .14, and with respect to the RT speed-up Results and Discussion over training, F(9, 213) ϭ 1.47, p Ͼ .15. Contrariwise, RTs in Experiment 3c were generally slower than in Experiments 1a/b, Experiment 3a and its replication, Experiment 3b, yielded F(1, 181) ϭ 4.09, MSE ϭ 184121.61, p ϭ .045, and the RT highly similar results and were therefore pooled in all subsequent speed-up was less pronounced, F(9, 173) ϭ 2.93, MSE ϭ 7474.8, analyses. Figure 4 compares mean RTs during practice in Exper- p ϭ .003. iments 3a/b, 3c, and 1a/b. Increased RTs on transfer Blocks 3, 6, In summary, our RT analyses did not provide any evidence for 9, and 12 in Experiments 3a/b indicate that learning of the primary improved learning of the training sequence when participants were training sequence was more pronounced than learning of the shifted to a systematic transfer sequence. In fact, RT savings transfer sequence. In Experiment 3c, participants were shifted to across blocks with the training sequence were even smaller in the transfer sequence and back to the training sequence on 8 out of Experiment 3c than in the baseline condition. This may mean one 10 blocks of trials. Since the onset of the transfer sequence in each of two things: Interpolating the systematic transfer sequence either block varied randomly between participants, we plotted RTs sep- reduced learning of the training sequence or interfered with the arately for the primary training sequence and for the transfer behavioral expression of what participants had learned about the sequence. Again, mean response latencies were higher for re- training sequence (Frensch et al., 1998; Frensch, Wenke, & sponse locations conforming to the transfer sequence, indicating Ru¨nger, 1999). superior learning of the primary training sequence. Figure 1 shows that participants in Experiments 3a/b and 3c In order to determine whether sequence learning expressed in acquired, on average, significantly more reportable sequence participants’ RTs differed between Experiments 1a/b and Experi- knowledge about the training sequence than did participants in the baseline condition, F(1, 431) ϭ 7.13, p ϭ .008, and participants in the two groups with interpolated random blocks, F(1, 431) ϭ 65.81, p ϭ .016. As can be seen in Figure 2, the same pattern of results emerged with PPC and LSS as measures of reportable sequence knowledge. The three systematic transfer groups differed in their PPC scores from the baseline condition, F(1, 431) ϭ 4.39, p ϭ .036, and from the random groups, F(1, 431) ϭ 6.17, p ϭ .013. The same was true for LSS scores, F(1, 431) ϭ 5.3, p ϭ .021, for the comparison with the baseline condition and F(1, 431) ϭ 7.51, p ϭ .006, for the comparison with the random groups. It is interesting to compare reportable sequence knowledge in Experiments 3a/b with reportable knowledge acquired in Experi- ment 2b because the only difference between the former experiments and the latter one lies in the randomness versus systematic- ity of trials during the four interpolated transfer blocks. Although reportable knowledge was numerically higher in Experiments 3a/b, this planned contrast failed to reach significance, F(1, 431) ϭ 3.04, p ϭ .082, for our primary verbal knowledge score; F(1, 431) ϭ Figure 4. Mean of individual median response times (RTs) in ms per trial 1.9, p ϭ .168, for the PPC score; and F(1, 431) ϭ 2.29, p ϭ .13, block. Filled dots represent the baseline condition (see Experiments 1a/b). for the LSS score. In Experiments 3a/b, Blocks 3, 6, 9, and 12 were transfer blocks that Table 2 reveals the origin of the increase in mean levels of contained an alternate systematic sequence. On the remaining blocks, reportable sequence knowledge when participants were transferred response locations followed the systematic training sequence. In Experi- to a systematic sequence: The percentage of nonverbalizers ment 3c we interpolated a series of 30 trials that conformed to a systematic transfer sequence on random positions within Blocks 2–10. RTs are plotted dropped from close to 30% in the previous experiments to about separately for trials that followed the training sequence (empty diamonds 15% in Experiments 3a/b and 3c. We pooled the experimental with solid line) and trials that followed the transfer sequence (empty groups with interpolated systematic sequences and compared the diamonds with dashed line). Error bars depict Ϯ1 standard error of the frequency of participants with and without reportable sequence mean. Exp. ϭ experiment. knowledge to the corresponding frequencies in Experiments 1a/b SEQUENCE LEARNING AND REPORTABLE KNOWLEDGE 1021 and Experiments 2a/b. In both cases, we found that the proportion Method of nonverbalizers was significantly lower in Experiments 3a/b/c, ␹2(1, N ϭ 267) ϭ 10.76, p ϭ .001, for the comparison with the Participants. Forty-six participants were recruited for Exper- baseline condition and ␹2(1, N ϭ 211) ϭ 5.99, p ϭ .014, for the iment 4a. We had to exclude the data of 2 participants from further comparison with participants who were exposed to random trans- analyses. One participant claimed to have expected a hidden reg- fer blocks. ularity. Another participant reported “1” for three of the eight We also contrasted Experiments 3a/b (four interpolated system- counts of the secondary task, which indicates that he had great atic transfer blocks) with Experiment 2b (four interpolated random difficulties performing the two tasks concurrently. The remaining transfer blocks). Even though the two groups did not differ sig- participants were 21 women and 23 men who ranged in age from ϭ ϭ nificantly in our verbal knowledge score, they can be distinguished 18 to 35 years (M 23.75, SD 3.78). on the basis of the proportion of nonverbalizers in each group, Initially, there were 48 participants in Experiment 4b. Four ␹2(1, N ϭ 126) ϭ 4.49, p ϭ .034. participants were excluded from further analyses. One participant In summary, the interpolation of systematic transfer sequences, had expected to encounter a hidden regularity right from the start. rather than random blocks of trials, proved to be crucial in obtain- A second participant deviated from the finger-to-key assignment ing a facilitative effect of transfer sequences on the acquisition of as described in the instructions. A third participant was distracted reportable sequence knowledge about the training sequence, as by his ringing mobile phone. Finally, a fourth participant left the predicted by the unexpected-event hypothesis. testing cubicle during the first dual task block and approached the experimenter because she had forgotten the instructions. The remaining participants, 28 women and 16 men, were 18 to 35 years Experiments 4a/b old (M ϭ 23.95, SD ϭ 4.05). The results presented so far bear out our contention that disrup- Procedure. Except for the addition of a secondary task during tions of the training sequence with the modified SRT task consti- transfer blocks, Experiment 4a was identical to Experiments 3a/b, tute unexpected events that promote the acquisition of reportable and Experiment 4b was the same as Experiment 2b. That is, sequence knowledge—if these disruptions do not prevent the suc- participants completed 14 blocks of 120 trials with the modified cessful completion of the search for their cause, that is, discovery SRT task. During Blocks 3, 6, 9, and 12, successive response of the sequential regularity. By contrast, disruptions that derail the locations were determined either randomly (see Experiment 4b) or search process (i.e., random blocks of trials) appear to be ineffec- by an alternate FOC sequence (see Experiment 4a). Response tual. With our final two experiments, we intended to provide locations on the remaining blocks of trials conformed to the FOC further evidence for the postulated relation between the type of training sequence. disruption of the incidental learning process and the acquisition of For the secondary task during transfer blocks, participants had to reportable sequence knowledge. count separately the number of times the large rectangle in the top With Experiments 4a/b, we replicated two of the previous ex- half of the stimulus display changed its color to red or blue. To periments in which four interpolated transfer blocks consisted of increase the difficulty of the color-counting task, we gave partic- either random trials (see Experiment 2b) or trials structured ac- ipants different predetermined starting values for the two counts cording to an alternate systematic sequence (see Experiments that varied between 17 and 35. Upon presentation of a red or blue 3a/b). However, for the duration of the four transfer blocks, we top rectangle, participants were to increment the respective count introduced a second task that participants had to perform concur- by 1. At the end of a transfer block, they were prompted to key in rently with the SRT task—maintaining separate counts of the the final values for the two counts. Immediate feedback was given number of times the top rectangle turned red or blue. From the on the accuracy of the counts. If one or both counts were wrong, perspective of the unexpected-event hypothesis, participants are participants were told to count more accurately. Detailed instruc- less likely to execute a search for a regularity during transfer tions for the secondary task were provided together with the SRT blocks when attentional resources are deployed on a taxing sec- task instructions before the training phase. During the training ondary task (cf. Frensch & Miner, 1994). Consequently, the pre- phase, prior to a transfer block, participants received the message vention of the search process in Experiment 4a should render that they were to perform the additional color-counting task on the ineffectual the systematic transfer blocks that were previously next block of trials, together with the initial values for the two shown to increase the amount of reportable sequence knowledge. counts. Before a trial block began that was not a transfer block, We therefore expected to find levels of reportable knowledge participants were informed that color counting was not required on about the training sequence that were similar to those in the the following block of trials. With this procedure we intended to baseline condition. remove any uncertainty that participants might have concerning Since a concurrent secondary task is, in itself, a type of disrup- whether a trial block was a color-counting block. tion, we needed to show that the secondary task does not affect the The assessment of reportable sequence knowledge in Experi- availability of reportable sequence knowledge per se, but only if it ments 4a/b proceeded in exactly the same way as in Experiments disrupts the search process that would otherwise lead to discovery 3a/b/c. The results for a recognition test administered after the of the transfer sequence. In Experiment 4b, we therefore combined assessment of reportable sequence knowledge are not reported in the secondary task with four random transfer blocks. Since no this article. sequence learning is possible on random blocks, we expected to Results and Discussion obtain the same amount of reportable sequence knowledge after the training phase as in the baseline condition and in the two An analysis of participants’ RTs allows us to check whether our groups with interpolated random blocks without a secondary task. experimental manipulation was successful in preventing learning 1022 RU¨ NGER AND FRENSCH of the systematic transfer sequence by administering a secondary levels) as a within-subject variable and transfer sequence (2 levels: task. Moreover, it is possible to determine whether the addition of random vs. systematic) and secondary task (2 levels: yes vs. no) a secondary task during transfer had any effects on RTs on blocks as between-subjects variables. All main effects were signifi- that contained the training sequence. To achieve these goals, we cant: F(3, 208) ϭ 97.12, p ϭ .001, for block; F(1, 210) ϭ 6.22, found it most informative to contrast mean RTs obtained during MSE ϭ 90354.38, p ϭ .013, for transfer sequence; and F(1, the training phase in Experiments 4a/b with mean RTs from 210) ϭ 162.43, MSE ϭ 90354.38, p Ͻ .001, for secondary task. Experiments 2b and 3a/b (see Figure 5). The only difference The Block ϫ Transfer Sequence interaction was significant, between the former two experiments and the latter lies in the F(3, 208) ϭ 4.46, p ϭ .005, as was the Secondary Task ϫ presence or absence of a secondary color-counting task during Transfer Sequence interaction, F(1, 210) ϭ 9.96, MSE ϭ transfer blocks. 90354.38, p ϭ .002, whereas the Block ϫ Secondary Task Our first RT analysis is concerned with SRT task performance interaction failed to reach significance, F(3, 208) ϭ 2.54, p Ͼ during blocks of trials that contained the systematic training se- .057. Finally, the three-way interaction was not significant, F(3, quence. In order to investigate the effects of different types of 208) ϭ 1.89, p Ͼ .132.4 transfer sequence and the presence or absence of a secondary task To further investigate how the secondary task impinged on during transfer blocks on participants’ RTs on nontransfer blocks, learning of the systematic transfer sequence, we conducted sepa- we performed an ANOVA with block (10 levels) as a within- rate ANOVAs for conditions with and without the secondary task. subject variable and transfer sequence (2 levels: random vs. sys- Participants who did not receive a secondary task showed a sig- tematic) and secondary task (2 levels: yes vs. no) as between- nificant effect of block, F(3, 122) ϭ 47.52, p Ͻ .001. More subjects variables. We obtained a significant effect of block, F(9, importantly, Figure 5 shows that participants responded more ϭ Ͻ ϫ 202) 64.43, p .001, and a significant Block Secondary Task slowly on random transfer blocks than on systematic transfer ϭ ϭ interaction, F(9, 202) 2.06, p .034. As can be seen in Figure 5, blocks, F(1, 124) ϭ 22.72, MSE ϭ 72111.07, p Ͻ .001. This participants who performed the color-counting task during transfer finding constitutes clear evidence that participants in Experiments blocks typically showed faster SRT task performance on blocks 3a/b also learned the systematic transfer sequence. Moreover, following a transfer block than did participants who did not receive learning continued across the four transfer blocks, as evidenced by a secondary task. Thus, the additional tone-counting task during an increasing difference between RTs to random response loca- transfer facilitated learning of the training sequence or, alterna- tions and those to systematic response locations, F(3, 122) ϭ 8.44, tively, the behavioral expression of what was learned, when learn- p Ͻ .001. ing is indexed by the speed of manual responses during blocks Participants who performed a secondary task on transfer blocks with the training sequence. All remaining effects in this analysis showed an effect of block, F(3, 84) ϭ 46.66, p Ͻ .001. Impor- were not significant, all Fs Ͻ 1. tantly, there was no effect of transfer sequence, F(1, 86) ϭ 0.15, The main goal of our second analysis of participants’ RTs is to MSE ϭ 116658.69, p Ͼ .69, and no Transfer Sequence ϫ Block determine how the addition of the tone-counting task affected interaction, F(3, 84) ϭ 0.21, p Ͼ .89. Consequently, there was no learning of the systematic transfer sequence. We submitted RT evidence that participants learned the systematic transfer sequence data generated during transfer blocks to an ANOVA with block (4 when they concurrently performed the color counting task. Alter- natively, if any learning of the systematic transfer sequence did occur, then participants were unable to express their knowledge in the presence of the tone-counting task. The RT data presented so far are fully consistent with the intent of our experimental manipulation—to prevent a search during transfer blocks that would otherwise lead to discovery of the systematic transfer sequence. We now turn to our central prediction regarding the effect of the tone-counting task during transfer blocks on the generation of reportable knowledge about the training sequence. The results for Experiments 4a/b are straightforward. Mean levels of reportable sequence knowledge (see Figure 1) are of the same magnitude as in the baseline condition, F(1, 431) ϭ 0, p Ͼ .97, and in the two groups with interpolated random blocks, F(1, 431) ϭ 0.01, p Ͼ .93, and they are significantly lower than in the three groups with interpolated systematic sequences, F(1, 431) ϭ 5.79, p ϭ .017. A comparison between the two identical Experiments 3a/b with four Figure 5. Mean of individual median response times (RTs) in ms per trial block. Blocks 3, 6, 9, and 12 were transfer blocks during which response locations were determined randomly (see Experiments 2a and 4b) or by an 4 Figure 5 and the results from separate ANOVAs for conditions with alternate systematic sequence (see Experiments 3a/b and 4a). On the and without the secondary task suggest a significant three-way interaction. remaining blocks, response locations followed the systematic training It is not clear why we failed to obtain this effect in our multivariate sequence. In Experiments 4a/b, participants performed a secondary color- analysis. Notably, the three-way interaction was significant in a mixed counting task during transfer blocks. Error bars depict Ϯ1 standard error of univariate ANOVA, F(3, 630) ϭ 3.46, MSE ϭ 5787.49, Greenhouse– the mean. Exp. ϭ experiment. Geisser adjusted p ϭ .029. SEQUENCE LEARNING AND REPORTABLE KNOWLEDGE 1023 interpolated systematic transfer blocks and Experiment 4a, which unsuited to triggering explicit learning with the SRT task (cf. differed only with respect to the additional color-counting task Buchner et al., 1997). For example, it is conceivable that an during transfer blocks, failed to reach significance, F(1, 431) ϭ experimental design that requires brief shifts from the training 2.93, p ϭ .088. sequence to a random response sequence and back to the training Once again, we obtained the same pattern of results for the two sequence (cf. Experiment 3c) would yield the expected facilitative alternative measures of reportable sequence knowledge displayed effect on the generation of reportable sequence knowledge. in Figure 2. Neither the PPC scores nor the LSS scores in Exper- We argued that the main reason for the lack of an effect of iments 4a/b differed from the respective scores in the baseline random transfer blocks in Experiments 2a/b was that interpolated condition and in the two random transfer conditions, all Fs Ͻ 1. By random events interfered with the search process that, according to contrast, a significant difference in LSS scores occurred for the the unexpected-event hypothesis, produces reportable sequence comparison with participants who received systematic transfer knowledge. Two findings support this conclusion. First, in order to sequences, F(1, 431) ϭ 3.96, p ϭ .047. However, the same allow for the discovery of a sequential regularity on transfer contrast failed to reach significance for the PPC scores, F(1, blocks, we shifted participants to an alternate response sequence 431) ϭ 2.47, p Ͼ .11. Finally, both measures yielded nonsignifi- and observed the predicted increase in reportable sequence knowl- cant effects for the comparison between Experiments 3a/b and edge about the training sequence (see Experiments 3a/b/c). Sec- Experiment 4a, F(1, 431) ϭ 2.39, p Ͼ .12, for PPC and F(1, ond, when a taxing secondary task was administered during trans- 431) ϭ 2.73, p Ͼ .09, for LSS. fer blocks (see Experiments 4a/b) in order to keep participants We also compared the proportion of nonverbalizers in Experi- from performing a search for a sequential regularity, reportable ments 4a/b with the proportion of nonverbalizers in the baseline knowledge about the training sequence remained at baseline level. condition and in the two groups with random transfer blocks (see Interestingly, the unexpected-event hypothesis offers yet another Table 2). In neither case did the proportion of nonverbalizers differ explanation for the observed effect of the secondary task: With the significantly, ␹2(1, N ϭ 227) ϭ 0.45, p Ͼ .5, for the comparison introduction of the tone-counting task, participants exhibited a with the baseline condition and ␹2(1, N ϭ 171) ϭ 1.12, p Ͼ .29, dramatic deterioration in SRT task performance (see Figure 5). for the comparison with the pooled random groups. We obtained, However, these performance decrements were not unexpected. To by contrast, a significant difference for the comparison with the three the contrary, participants had a good explanation for their slow groups that received systematic transfer sequences (see Experiments responses—the burden of having to perform the secondary color- 3a/b/c), ␹2(1, N ϭ 216) ϭ 13.32, p Ͻ .001. Moreover, the counting task. Thus, participants likely did not experience unex- proportion of nonverbalizers in Experiment 4a (four systematic pected events during transfer blocks with a concurrent secondary transfer blocks with a secondary task) by itself also differed task and, consequently, did not even attempt to initiate a search from the proportion of nonverbalizers in Experiments 3a/b (four that could lead to discovery of a sequential regularity. systematic transfer blocks without a secondary task), ␹2(1, N ϭ The results for Experiments 4a/b are also compatible with 128) ϭ 6.44, p ϭ .011, even though this specific contrast failed to alternative accounts of the effects of a secondary task on sequence reach significance with the continuous measures of reportable learning that do not postulate the existence of a search process. For sequence knowledge. example, several authors proposed that the secondary task draws In summary, the findings from Experiments 4a/b, taken together on the same attentional resources that are needed for sequence with the results of the previous experiments, strongly support our learning to occur (e.g., Cohen, Ivry, & Keele, 1990; Nissen & hypothesis that only disruptions that do not themselves obstruct the Bullemer, 1987). Others provided evidence that a secondary task search for their proper cause facilitate the acquisition of reportable has a disruptive effect, if, and only if, it cannot be integrated with knowledge about the training sequence. By contrast, when a sec- the primary SRT task (e.g., Jimeńez&Meńdez, 1999; Schmidtke ondary task interfered with the search process, the facilitative & Heuer, 1997). Frensch and collaborators (2003) suggested that it effect of interpolated systematic blocks disappeared. is not sequence learning per se that is affected but only the behavioral expression of what is learned (Frensch et al., 1998, General Discussion 1999). Clearly, additional experiments are needed to characterize in In four experiments, we tested the core assumption of the greater detail the search process and the conditions under which it unexpected-event hypothesis, a theoretical framework advanced operates. The same holds true for the investigation of unexpected by Frensch and collaborators (2003) to explain the generation of events as the triggering condition for explicit learning. Current explicit, reportable knowledge in incidental-learning situations. efforts at our lab are directed at understanding precisely which According to the framework, the generation of reportable knowl- aspects of tasks and task behaviors contribute to participants’ edge is triggered by the observation of unexpected events. We expectations and their violations. therefore induced unexpected events experimentally by disrupting Since we used a modified version of the SRT task in our the sequence-learning process with a color-matching version of the experiments, it is sensible to ask whether our findings can be SRT task and compared the available reportable knowledge with generalized to the standard version of Nissen and Bullemer (1987). knowledge generated in a baseline condition in which sequence Even though our color-matching SRT task contained an additional learning proceeded undisruptedly. Of the two types of disruptions perceptual matching component, sequence learning was confined that were used—interpolated random and systematic transfer se- to series of response locations. In this respect, it is fully compatible quences—only the latter raised reportable sequence knowledge with the standard SRT task. In our view, a more important issue above the level observed in the baseline condition. However, it concerns the complexity of the to-be-learned sequence. We pro- would be premature to conclude that random events are principally posed that observed unexpected events trigger a search for their 1024 RU¨ NGER AND FRENSCH proper cause, but this search need not be successful. In particular, sequence knowledge. Depending on the particular type of indirect when being trained on a more complex SOC sequence, many test (interpolated systematic or random sequences with or without participants may be able to generate only fragmentary reportable subsequent reinstatement of the regular training sequence), partic- knowledge. We surmise that particularly salient segments of the ipants may be more or less likely to generate explicit knowledge. sequence such as a succession of neighboring response locations, Since implicit sequence learning is commonly defined as learning also referred to as “runs” (e.g., 1-2-3-4 in the SOC sequence in the absence of conscious or explicit sequence knowledge, em- 4-2-4-3-1-2-3-4-1-3-2-1 used by Shanks & Channon, 2002) are pirical evidence for implicit learning may vary with the particular most likely to be included in participants’ verbal reports. indirect test employed in a hitherto unforeseen way. Sequence learning is carried to extremes when participants are The final issue we would like to address is how the results trained on probabilistic instead of deterministic sequences. For summarized in this report appear in light of theoretical accounts example, Jimeńez, Meńdez, and Cleeremans (1996) generated other than the unexpected-event hypothesis. Our theoretical frame- sequential contingencies on the basis of a noisy finite-state gram- work is fully compatible with multiple-systems accounts of se- mar, with 15% of the systematic stimuli being replaced by random quence learning that postulate dedicated memory systems for the stimuli. Particular care was taken in choosing a grammar that did generation of explicit (i.e., reportable) and implicit sequence not produce any salient patterns of response locations. Importantly, knowledge (e.g., Reber & Squire, 1994, 1998; Willingham, 1998). one group of participants was informed that stimuli were generated However, in empirical investigations of explicit sequence learning, according to a complex set of rules, which they were urged to participants are typically informed of the existence of a sequential discover and use to improve their performance. In a second group regularity and urged to memorize the sequence and use it to of participants, sequence learning was incidental. After 20 sessions improve their performance (e.g., Reber & Squire, 1998; Willing- of training with the SRT task, participants were given a generate ham & Goedert-Eschmann, 1999). Consequently, these studies do task to assess the amount of explicit sequence knowledge. Jimeńez not provide any insight into how an explicit learning process might et al. found no effect of participants’ learning orientation on be invoked in an incidental-learning situation. The unexpected- generate task performance. Although the availability of reportable event hypothesis fills this gap in existing theoretical accounts by sequence knowledge was not directly assessed in this experiment, postulating that explicit learning is triggered by the observation of we believe Jimeńez et al. created an experimental situation that unexpected events in incidental-learning situations. rendered completely ineffectual the search process that is respon- Single-system accounts of sequence learning assume that the sible for the generation of reportable knowledge. same set of learning mechanisms underlies both behavioral and In summary, we consider the occurrence of unexpected events a verbal expressions of sequence learning. Typically, such learning general phenomenon in incidental-learning situations. However, is assumed to be an automatic, associative process that is sensitive whether or not the occurrence of unexpected events is associated to the statistical properties of the learning situation (e.g., Cleer- with an increase in reportable knowledge about underlying task emans & Jimeńez, 2002; Perruchet & Vinter, 2002). It is difficult regularities depends on the specific conditions under which the to envision how a single-system account could accommodate ensuing search process is invoked. Whether similar results can be our central finding that an interpolated transfer sequence can obtained with other direct tests of sequence learning such as improve learning of the training sequence. From a connectionist generate tasks or recognition tests is an open question. In our modeling perspective, a transfer sequence is a source of inter- study, we focused on verbal reports to assess participants’ explicit ference that should weaken existing associations between ele- sequence knowledge because we consider verbal reports the most ments of the training sequence. Consequently, single-system valid among the available measures of explicit learning (Ru¨nger & accounts should predict a negative effect on learning of the Frensch, 2007). training sequence instead of the facilitative effect that we The main methodological contribution of the present investiga- observed in our experiments. tion to the sequence learning literature consists of a novel measure Perhaps one could argue that transfer blocks do not affect of explicit sequence knowledge derived from participants’ verbal sequence learning per se but more general aspects of task perfor- reports. The development of the new measure was motivated by mance. A participant who is shifted to a transfer sequence is likely the insight that conventional scoring procedures analyze verbal to experience performance decrements. He or she might respond reports at a particular level of sequence structure (e.g., at the level with increased effort to return to previous levels of SRT task of triplets for SOC sequences), thereby potentially neglecting performance. If this additional effort is carried over and main- information at higher or lower levels. Although several studies tained on trials that conform to the training sequence, we might took a more comprehensive approach and examined SRT task indeed observe a facilitative effect of the interpolated transfer performance at different levels of statistical structure (e.g., Per- sequence. However, several of our findings speak against this ruchet & Amorim, 1992; Shanks & Channon, 2002; Stadler, 1992), interpretation. First, random transfer blocks produced even greater our novel measure is, to the best of our knowledge, the only one to performance decrements than did systematic transfer blocks, but integrate knowledge at different levels of sequence structure into a we did not observe increased levels of reportable sequence knowl- single score. edge. Second, in those experiments that yielded superior levels of Our findings also have some relevance for research on sequence reportable knowledge, SRT task performance on nontransfer learning that focuses on the characteristics of implicit learning. As blocks was either identical to (see Experiment 3a/b) or even worse noted above, interpolating random sequences or a systematic trans- than (see Experiment 3c) in the baseline condition. Third, although fer sequence is a standard procedure for the indirect assessment of we did find that participants who performed a taxing secondary sequence learning. Our results suggest the administration of an task during transfer blocks (see Experiments 4a/b) responded indirect test can affect the subsequent assessment of reportable faster on subsequent nontransfer blocks than did participants SEQUENCE LEARNING AND REPORTABLE KNOWLEDGE 1025 without the secondary task (see Experiments 2b and 3a/c), the counting task suppresses expression of knowledge in the serial reaction administration of secondary tasks was not associated with in- task. Journal of Experimental Psychology: Learning, Memory, and creased reportable knowledge about the training sequence. On the Cognition, 25, 260–274. basis of these results, the alternative single-system explanation of Haider, H., & Frensch, P. A. (2005). The generation of conscious aware- our central finding can be ruled out. ness in an incidental learning situation. Psychological Research, 69, 399–411. In summary, the unexpected-event hypothesis places incidental Jimeńez, L., & Meńdez, C. (1999). Which attention is needed for implicit sequence learning at the junction of automatic perceptual-motor sequence learning? Journal of Experimental Psychology: Learning, learning, attributional processes, and a search for the cause of Memory, and Cognition, 25, 236–259. observed unexpected events. Whether this admittedly complex Jimeńez, L., Meńdez, C., & Cleeremans, A. (1996). Comparing direct and picture of incidental sequence learning can be upheld against more indirect measures of sequence learning. Journal of Experimental Psy- parsimonious single-system accounts of sequence learning (e.g., chology: Learning, Memory, and Cognition, 22, 948–969. Shanks et al., 2003) is an issue that awaits further debate. Kinder, A., & Shanks, D. R. (2003). Neuropsychological dissociations between priming and recognition: A single-system connectionist account. Psychological Review, 110, 728–744. References Lee, Y. S. (1997). Learning and awareness in the serial reaction time task. In Proceedings of the 19th Annual Conference of the Cognitive Science Buchner, A., Steffens, M. C., Erdfelder, E., & Rothkegel, R. (1997). A Society. Hillsdale, NJ: Erlbaum. multinomial model to assess fluency and recollection in a sequence Nissen, M. J., & Bullemer, P. (1987). Attentional requirements of learning: learning task. Quarterly Journal of Experimental Psychology: Human Evidence from performance measures. Cognitive Psychology, 19, 1–32. Experimental Psychology, 50(A), 631–663. Norman, E., Price, M. C., & Duff, S. C. (2006). Fringe consciousness in Cleeremans, A. (2006). Conscious and unconscious cognition: A graded, sequence learning: The influence of individual differences. Conscious- dynamic perspective. In Q. Jing, M. R. Rosenzweig, G. d’Ydewalle, H. ness and Cognition, 15, 723–760. Zhang, H.-C. Chen, & C. Zhang (Eds.), Progress in psychological Perruchet, P., & Amorim, M. (1992). Conscious knowledge and changes in science around the world: Vol. 1. Neural, cognitive, and developmental performance in sequence learning: Evidence against dissociation. Journal of issues (pp. 401–418). Hove, United Kingdom: Psychology Press. Experimental Psychology: Learning, Memory, and Cognition, 18, 785–800. Cleeremans, A., & Jimeńez, L. (2002). Implicit learning and conscious- Perruchet, P., Bigand, E., & Benoit-Gonin, F. (1997). The emergence of ness: A graded, dynamic perspective. In R. M. French & A. Cleeremans explicit knowledge during the early phase of learning in sequential (Eds.), Implicit learning and consciousness (pp. 1–40). Hove, United reaction time tasks. Psychological Research, 60, 4–13. Kingdom: Psychology Press. Perruchet, P., & Vinter, A. (2002). The self-organizing consciousness. Cohen, A., & Curran, T. (1993). On tasks, knowledge, correlations, and dissoci- Behavioral and Brain Sciences, 25, 297–388. ations: Comment on Perruchet and Amorim (1992). Journal of Experimental Ru¨nger, D., & Frensch, P. A. (2007). Defining consciousness in the context Psychology: Learning, Memory, and Cognition, 19, 1431–1437. of incidental sequence learning: Theoretical considerations and empir- Cohen, A., Ivry, R. I., & Keele, S. W. (1990). Attention and structure in ical implications. Manuscript submitted for publication. sequence learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 16, 17–30. Reber, P. J., & Squire, L. R. (1994). Parallel brain systems for learning Curran, T. (1997). Effects of aging on implicit sequence learning: Account- with and without awareness. Learning & Memory, 1, 217–229. ing for sequence structure and explicit knowledge. Psychological Re- Reber, P. J., & Squire, L. R. (1998). Encapsulation of implicit and explicit search, 60, 24–41. memory in sequence learning. Journal of Cognitive Neuroscience, 10, Curran, T., & Keele, S. W. (1993). Attentional and nonattentional forms of 248–263. sequence learning. Journal of Experimental Psychology: Learning, Reed, J., & Johnson, P. (1994). Assessing implicit learning with indirect Memory, and Cognition, 19, 189–202. tests: Determining what is learned about sequence structure. Journal of Dienes, Z., & Perner, J. (1999). A theory of implicit and explicit knowl- Experimental Psychology: Learning, Memory, and Cognition, 20, 585–594. edge. Behavioral and Brain Sciences, 22, 735–808. Schmidtke, V., & Heuer, H. (1997). Task integration as a factor in Erdelyi, M. H. (2004). Subliminal perception and its cognates: Theory, secondary-task effects on sequence learning. Psychological Research, indeterminacy, and time. Consciousness and Cognition, 13, 73–91. 60, 53–71. Frensch, P. A., Buchner, A., & Lin, J. (1994). Implicit learning of unique Shanks, D. R. (2005). Implicit learning. In K. Lamberts and R. Goldstone and ambiguous serial transitions in the presence and absence of a (Eds.), Handbook of cognition (pp. 202–220). London: Sage. distractor task. Journal of Experimental Psychology: Learning, Memory, Shanks, D. R., & Channon, S. (2002). Effects of a secondary task on and Cognition, 20, 567–584. “implicit” sequence learning: Learning or performance? Psychological Frensch, P. A., Haider, H., Ru¨nger, D., Neugebauer, U., Voigt, S., & Werg, Research, 66, 99–109. J. (2003). Verbal report of incidentally experienced environmental reg- Shanks, D. R., & Johnstone, T. (1999). Evaluating the relationship between ularity: The route from implicit learning to verbal expression of what has explicit and implicit knowledge in a sequential reaction time task. been learned. In L. Jimeńez (Ed.), Attention and implicit learning (pp. Journal of Experimental Psychology: Learning, Memory, and Cogni- 335–366). Amsterdam: Benjamins. tion, 25, 1435–1451. Frensch, P. A., Lin, J., & Buchner, A. (1998). Learning versus behavioral Shanks, D. R., & Perruchet, P. (2002). Dissociation between priming and expression of the learned: The effects of a secondary tone-counting task recognition in the expression of sequential knowledge. Psychonomic on implicit learning in the serial reaction task. Psychological Research, Bulletin & Review, 9, 362–367. 61, 83–98. Shanks, D. R., Wilkinson, L., & Channon, S. (2003). Relationship between Frensch, P. A., & Miner, C. S. (1994). Effects of presentation rate and priming and recognition in deterministic and probabilistic sequence individual differences in short-term memory capacity on an indirect learning. Journal of Experimental Psychology: Learning, Memory, and measure of serial learning. Memory & Cognition, 22, 95–110. Cognition, 29, 248–261. Frensch, P. A., & Ru¨nger, D. (2003). Implicit learning. Current Directions Stadler, M. A. (1992). Statistical structure and implicit serial learning. in Psychological Science, 12, 13–18. Journal of Experimental Psychology: Learning, Memory, and Cogni- Frensch, P. A., Wenke, D., & Ru¨nger, D. (1999). A secondary tone- tion, 18, 318–327. 1026 RU¨ NGER AND FRENSCH

Wilkinson, L., & Shanks, D. R. (2004). Intentional control and implicit a serial response time task using a recognition measure: Comment on sequence learning. Journal of Experimental Psychology: Learning, Perruchet and Amorim (1992). Journal of Experimental Psychology: Memory, and Cognition, 30, 354–369. Learning, Memory, and Cognition, 19, 1424–1430. Willingham, D. B. (1998). A neuropsychological theory of motor skill learning. Psychological Review, 105, 558–584. Willingham, D. B., & Goedert-Eschmann, K. (1999). The relation between implicit and explicit learning: Evidence for parallel development. Psy- Received July 17, 2006 chological Science, 10, 531–534. Revision received April 20, 2008 Willingham, D. B., Greeley, T., & Bardone, A. M. (1993). Dissociation in Accepted April 23, 2008 Ⅲ

Call for Papers: Special Section titled “Spatial reference frames: Integrating Cognitive Behavioral and Cognitive Neuroscience Approaches”

The Journal of Experimental Psychology: Learning, Memory, and Cognition invites manuscripts for a special section on spatial reference frames, to be compiled by Associate Editor Laura Carlson and guest editors James Hoffman and Nora Newcombe. The goal of the special section is to showcase high-quality research that brings together behavioral, neuropsychological, and neuroimaging approaches to understanding the cognitive and neural bases of spatial reference frames. We are seeking cognitive behavioral studies that integrate cognitive neuroscience findings in justifying hypotheses or interpreting results and cognitive neuroscience studies that emphasize how the evidence informs cognitive theories regarding the use of spatial reference frames throughout diverse areas of cognition (e.g., attention, language, perception and memory). In addition to empirical papers, focused review articles that highlight the significance of cognitive neuroscience approaches to cognitive theory of spatial reference frames are also appropriate. The submission deadline is February 28, 2009. The main text of each manuscript, exclusive of figures, tables, references, or appen- dixes, should not exceed 35 double-spaced pages (approximately 7,500 words). Initial inquiries regarding the special section may be sent to Laura Carlson ([email protected]). Papers should be submitted through the regular submission portal for JEP:LMC (http:// www.apa.org/journals/xlm/submission.html) with a cover letter indicating that the paper is to be considered for the special section.