Temporal effects in sequence learning

Through the use of double task conditions, the sequence learning (SL) paradigm offers unique opportunities to study the relationships between learning and attention. In their original study, Nissen & Bullemer (1987) argued that a secondary tone-counting task prevents SL because it exhausts participants’ attentional resources. Other authors have instead suggested that the detrimental effects of tone-counting are due to scheduling conflicts between performing the main and secondary tasks rather than to attentional load. Frensch & Miner (1994), for instance, suggested that the secondary task impairs sequence learning because it lengthens the response-to-stimulus interval (RSI) and hence makes it less likely for relevant contingencies to be represented together in short-term memory, — a condition for learning. Stadler (1995), on the other hand, argued that the secondary task introduces variability in the RSI and disrupts the organization of the sequence into chunks. Further, according to Willingham, Greenberg & Cannon Thomas (1997) manipulation of the RSI influences performance but not sequence learning per se. The goal of this paper is to further explore and clarify the role of the RSI in the SL paradigm. To do so, we systematically manipulated the RSI, and assessed performance through different objective and subjective measures. In contrast to previous results, we found that increasing the RSI improves explicit SL. We further show how a neural network model based on the Simple Recurrent Network can account for our data, even though the model neither uses decay nor develops chunked, declarative representations of the sequence. These findings suggest that RSI effects in SL are rooted in the temporal dynamics of learning