ORIGINAL ARTICLE E. Soetens �� A. Melis �� W. Notebaert Sequence learning and sequential effects Received: 23 July 2002/ Accepted: 15 October 2003/Published online: 25 May 2004 �� Springer-Verlag 2004 Abstract In a serial reaction time (RT) task with a probabilistic stimulus sequence, the length of the re- sponse-to-stimulus interval (RSI) and the sequence complexity was manipulated to investigate the relation- ship between sequence learning and sequential effects in serial RT tasks. Sequential effects refer to the influence of previous stimulus presentations on the RT to the current stimulus. Sequence learning is stimulus-transi- tion specific and is demonstrated as the difference be- tween practiced and unpracticed sequences within an interpolated random block of trials. There is a clear parallel between sequence learning and specific changes in sequential effect in the short RSI conditions, sug- gesting that a common mechanism may lie at the basis of sequence learning and automatic facilitation, which is responsible for sequential effects at short RSI. Impor- tantly, the changes in sequential effects accompanying sequence learning are the same as those observed with practice in random serial RT tasks, indicating that the learning process underlying sequence learning is the same as in random tasks. Sequence learning and sequential effects One of our most basic cognitive abilities is that we are able to predict the coming sequence of events on the basis of preceding events. This capability is central to learning and to the adaptation of our behavior in many everyday situations, including the skill to develop com- bined motor actions and to acquire a language (Whit- tlesea & Wright, 1997). Learning event sequences in a serial reaction time (RT) task is an interesting paradigm to study the basic characteristics of such learning pro- cesses, because it is devoid of confounding influences of overlearned knowledge such as language. In a typical sequence-learning task, participants are asked to re- spond as rapidly as possible to the location of each of several stimuli. They are not informed about the struc- tured nature of the stimulus sequence. Learning of the regularities in the sequence is derived from a larger de- crease in response latencies compared with a random stimulus sequence (Cleeremans & McClelland, 1991 Nissen & Bullemer, 1987) and from a slowing in re- sponse speed when an unannounced random sequence is introduced. In the present paper we want to demonstrate that sequence learning is based on the frequency of oc- currence of particular transitions between successive stimuli and that this learning process is not different from practice effects in random serial RT tasks. More- over, we will search for a parallel between sequence learning and particular changes in sequential effects in conditions with a high rate of stimulus presentation, which may indicate that both effects have a common underlying origin. There is an abundance of studies showing that RTs change as a function of the preceding sequence of stimuli or responses (e.g., Bertelson, 1963 Kirby, 1976). Two processes have been distinguished to cause sequential effects. The stimulus presentation rate is the most important factor that determines which process domi- nates. When the succession of stimuli is fast, a type of automatic process, called automatic facilitation has been found, and with a slower succession of stimuli a kind of subjective expectancy process develops. There have been claims that these local sequential dependencies may lie at the basis of some type of learning process, but the out- come of these studies are equivocal (Pashler & Baylis, 1991a, 1991b Rabbitt, 1968 Soetens, Boer, & Hueting, 1985). The sequence-learning paradigm offers a unique opportunity to find out whether the sequential dependencies may be related to this type of learning. Sequence learning too is substantially mediated by the rate of stimulus presentations. It is inversely related E. Soetens (&) �� A. Melis �� W. Notebaert Cognitive and Physiological Psychology, University of Brussels (VUB), Pleinlaan 2, 1050 Brussels, Belgium E-mail: esoetens@vub.ac.be Tel.: +32-2-6292873 Fax: +32-2-6292489 Psychological Research (2004) 69: 124���137 DOI 10.1007/s00426-003-0163-4

to the length of the interval between response initiation and the presentation of the next stimulus, called the response-stimulus interval (RSI Frensch & Miner, 1994 Willingham, Greenberg & Thomas, 1997). Participants trained with a 1,500-ms RSI showed a smaller RT in- crease with the introduction of a random transfer se- quence than participants exposed to a 500-ms RSI (Frensch & Miner, 1994). According to Frensch and Miner a short RSI is beneficial for learning, because, in comparison to slower presentation rates, it allows more stimuli to be simultaneously active in short-term mem- ory, where an associative mechanism may detect sequential regularities between subsequent stimuli. Maybe it is no coincidence that RSI also determines the underlying mechanism of sequential effects. Se- quence learning and sequential effects may originate from a common underlying mechanism. If a relationship can be found, the study of sequential effects could be promising to provide a better explanation for the RSI effect on sequence learning. It is also possible that sequential effects may lie at the basis of some learning process, in the sense that short-term decreases in RT as a consequence of preceding events, may lead to more permanent facilitation effects. We proceed with a brief overview of the most important findings on sequential effects, including the influence of practice, and the mechanisms they are reflecting in choice serial RT tasks with random stimulus sequences. Sequential effects Sequential effects in serial RT tasks with random stim- ulus sequences have attracted considerable attention since the very beginning of information processing the- ories in the 1950s (e.g., Hyman, 1953) and are important for many paradigms in mental chronometry (for a re- view, see Sanders, 1998). They describe variations in response latency as a function of transitions between stimuli or responses. In two-choice tasks, transitions are coded as repetitions or alternations. The influence of the transition between the current trial n and the previous trial n-1 is referred to as a first-order effect. When rep- etitions are reacted to faster than alternations, it is called a repetition effect (Bertelson, 1961). Alternatively, when RTs to alternations are faster than those to repetitions it is called an alternation effect. Repetition and alternation effects have been studied more closely with many-to-one stimulus-response map- pings, where the repetition of a response does not nec- essarily involve a repetition of the stimulus. When two or more stimuli are mapped onto the same response, it is possible to separate stimulus and response factors con- tributing to the overall repetition effect (Bertelson, 1965). Obviously, we talk about stimulus repetitions when the same stimulus is presented in successive trials. Response-only repetitions involve the presentation of a different stimulus, while the response is repeated. Fi- nally, there are also response alternations. A stimulus repetition effect is obtained when RTs to stimulus repe- titions are faster than to response-only repetitions, and a response repetition effect is obtained when RTs to re- sponse-only repetitions are faster than to response alternations. Stimulus and response repetition effects have been found to co-exist, indicating that priming in repetition trials can be attributed to facilitation in the processing of both stimulus and response characteristics (Campbell & Proctor, 1993 Pashler & Baylis, 1991a, 1991b Soetens, 1998). Higher order effects refer to RT differences that are caused by transitions between stimuli earlier in the se- quence. Third-order transitions, for instance, refer to the successive transitions between stimuli n-3 and n-1 rela- tive to the current stimulus n. In two-choice tasks, two different patterns emerge depending on RSI duration. The first-order and higher order RT patterns are asso- ciated with an automatic facilitation mechanism when RSI is short, and with subjective expectancy when RSI is long (Kirby, 1976 Kornblum, 1973 Soetens, 1998). The transition from automatic facilitation to subjective expectancy is around 100 ms for a spatially compatible two-choice serial RT task (Soetens, Boer, & Hueting, 1985). Higher order effects of automatic facilitation display a benefit-only pattern. The more repetitions there are, and the more recent the repetitions are in the pre- ceding higher order sequence, the faster the RTs are to the coming stimulus, irrespective of which stimulus it is. The first-order repetition effect of automatic facilita- tion relies on the re-use of activation left by previous stimulus-response cycles for speeding up reactions to subsequent stimuli. It has been linked to a kind of priming mechanism (Campbell & Proctor, 1993 Pashler & Baylis, 1991b, Soetens, Deboeck, & Hueting, 1984). The benefit-only pattern is thought to result from the monitoring of response accuracy (Kirby, 1980), which is assumed to be more important for alternations than for repetitions. The more monitoring a response needs, the slower the RTs to the next stimulus will be. Subjective expectancy is characterized by a first-order alternation effect and a higher order pattern of the cost- benefit type. The alternation effect indicates that parti- cipants display expectancy for alternations. The effect is also found in random generation tasks, where it is known as the ������gambler���s fallacy������ (Rapoport & Budescu, 1997 Wagenaar, 1972). Expectancy needs time to de- velop and influences RTs only when RSI is long. The higher order cost-benefit pattern has also been explained as a type of expectancy. Participants seem to favor a continuation of a run of repetitions or alternations, be- cause they yield the fastest RTs. Disruptions of such a run, for instance when a series of alternations is followed by a first-order repetition, produce long RTs. In two- choice tasks these subjective expectations lead to an exchange relationship (Audley, 1973) or cost-benefit pattern. Sequential effects in serial RT tasks with random stimulus sequences change with practice, suggesting that the mechanisms causing sequential effects may be related 125