Maintenance of semantic information in capacity-limited item short-term memory

Abstract

We report a semantic effect in immediate free recall, which is localized at recency and is preserved under articulatory suppression but is highly reduced when recall is delayed after an intervening distractor task. These results are explained by a neurocomputational model based on a limited-capacity short-term memory (STM) store, consisting of activated long-term memory representations. The model makes additional predictions about serial position functions in semantically cued recall, indicating capacity limitations caused by a displacement type mechanism, which are confirmed in a second experiment. This suggests that in addition to the phonological component in verbal STM, there is an activation/ item-limited component with semantically sensitive representations.

Early dynamics of the semantic priming shift

“…In the network, the longer Prime 2 duration increases the duration of its perceptual response leading to strong inhibitory feedback, which in turn deactivates associates to the first prime in the RU condition. This interference on semantic priming is in accordance with the model's prediction, showing that retroactive interference generated by the sensory activity during the processing of an item (i.e., here the Prime 2) is stronger than retroactive interference generated by an item but after its offset Brunel & Wang, 2001;Haarmann & Usher, 2001). Results of Experiment 3 and the model behavior indicate that the effects of retroactive interference on RU priming are stronger during Prime 2 processing (Experiment 3) than during the ISI2 (Experiment 2).…”

“…During longer ISIs, the primes are off and generate retrospective activity only and hence also weak retroactive interference. When durations of Primes 1 and 2 are identical, we expect retroactive interference to be stronger than proactive interference (see Cowan, 2001;Haarmann & Usher, 2001), leading to UR effects stronger than RU effects. In the experimental literature, cases of equal values of short SOA1 and SOA2 show results varying from significant RU and UR priming (e.g., Angwin, Chenery, Copland, Murdoch, & Silburn, 2005;Balota & Paul, 1996;Chenery, Copland, McGrath, & Savage, 2004) to significant UR priming only (e.g., Masson, 1995;Thérouanne & Denhière, 2002).…”

“…Therefore, in multiple priming, processing of the sequential primes and their activated associates generates unselective inhibition proportional to the overall level of activation. Such inhibitory feedback limits the number of simultaneously activated items (Haarmann & Usher, 2001), and thereby constrains the capacity of the network's working memory (Cowan, 2001). As a consequence, the activity of a target increases with the SOA following a related prime, but also decreases with the duration of the strong inhibitory feedback by an unrelated prime (Lavigne et al, 2011).…”

“…Table 4). In addition, the non-selective inhibitory feedback by Prime 1 limits the activation of further items due to proactive interference Cowan, 2001; see Haarmann & Usher, 2001, for a review). This limiting inhibition by any Prime1 implies that Prime 2 activation is subject to proactive interference by Prime 1.…”

“…A consequence of the limited capacity of the working memory system (see Cowan, 2001;Haarmann & Usher, 2001) is that all the associates to the words read in a sequence cannot be activated, leading to only partial activation of the meaning of a given word. Selection processes are therefore involved that keep activated only the associates that best correspond to the sequence of words (Whitney, Grossman, & Kircher, 2009;Whitney, Jefferies, & Kircher, 2010).…”