Generalization and source memory in acquired equivalence

Abstract

<p>Memory allows us to remember specific events but also combine information across events to infer new information. New inferences are thought to stem from integrating memories of related events during encoding but can be also generated on-demand, based on separate memories of individual events. Integrative encoding has been argued as dominant in the acquired equivalence paradigm, where people have a tendency to assume that when two faces share one preference, they also share another. A downside may be a loss of source memory, where inferred preferences are mistaken for observed ones. Here, we tested the predictions of integrative encoding across five data sets collected using small variations of the acquired equivalence paradigm. Results showed a statistically reliable but numerically small tendency to generalize preferences across faces, with stronger evidence for on-demand inferences at retrieval rather than spontaneous integration during encoding. Newly included explicit source memory test showed that participants differentiated learned from inferred preferences to a high degree, irrespective of whether they generalized preferences across faces. Overall, the results indicate that representations of individual events and retrieval-based processes may play a larger role in acquired equivalence than previously thought, informing current theories of generalization and knowledge representation.</p>

Building integrated representations through interleaved learning

“…This is in line with existing evidence that encoding-based mechanisms support inference after interleaved exposure (Shohamy & Wagner 2008;Zeithamova et al, 2012), whereas inference may rely on sequential, retrieval-based processes when related pairs are shown in a blocked manner (Barron et al, 2020;Banino et al, 2016) or with limited exposure (Banino et al, 2016;Koster et al, 2018). In both our explicit and implicit assessments, we found faster RTs for direct than indirect pairs, which has sometimes been interpreted as evidence for retrieval-based mechanisms (Shohamy & Wagner 2008;de Araujo Sanchez & Zeithamova 2020). However, models learning distributed representations at encoding also predict stronger associations among directly learned associates, so we do not view these RT differences as diagnostic between the mechanisms considered here.…”

“…Several prior imaging and behavioral studies speak to the different strategies and representations that can underlie associative inference. Some have argued that the hippocampus constructs overlapping representations of related AB and BC pairs at encoding to support later inference, as in integrative encoding (Shohamy & Wagner 2008;Zeithamova et al, 2012;Schlichting et al, 2014;, whereas others have argued that inference is supported by sequential activation at retrieval (Banino et al, 2016;Koster et al, 2018;Barron et al, 2020;de Araujo Sanchez & Zeithamova 2020). This dichotomy mirrors the distinction in decision-making between model-free and model-based behavior (Daw et al, 2005), where information is either directly stored in representations during learning or computed on the fly.…”

Neural representation of abstract task structure during generalization

“…The lengthy RTs during the first trials of generalization suggest that this inference required additional online processing, and this abstract task structure may not have been fully constructed until it was probed. These data are consistent with accounts arguing that generalization during acquired equivalence can also occur at retrieval ( De Araujo Sanchez and Zeithamova, 2020 ) and depends on the demand to make new inferences. It is likely that linking of contexts to a LS occurs at both encoding and retrieval, with the timecourse of this process varying across individuals.…”

Neural representation of abstract task structure during generalization

“…The lengthy RTs during the first trials of generalization suggest that this inference required additional online processing, and this abstract task structure may not have been fully constructed until it was probed. These data are consistent with accounts arguing that generalization during acquired equivalence can also occur at retrieval (De Araujo Sanchez and Zeithamova, 2020), and depends on the demand to make new inferences. It is likely that linking of contexts to a latent state occurs at both encoding and retrieval, with the timecourse of this process varying across individuals.…”