Feature overlap modulates rapid semantic but not lexical integration of novel associations by means of fast mapping

Abstract

There is evidence that rapid integration of novel associations into cortical networks is possible if associations are acquired through a learning procedure called fast mapping (FM).FM requires precise visual discrimination of sometimes highly similar pictures of a previously unknown and a known item, and linking an unfamiliar label to the unknown item. In order to shed light on the mechanisms underlying learning through FM, we manipulated feature overlap between the two items as potential modulating factor. In Experiment 1, we found that labels of the unknown items generally evoked instantaneous lexical competition when encoded through FM, indicating rapid integration into lexical networks. In Experiment 2, we observed semantic priming immediately after FM encoding but only if the items shared many features. This indicates that whereas feature overlap leaves item-level lexical integration unaffected, it might mediate semantic integration of arbitrary picture-label associations, which could explain contradictory findings in the literature. Highlights• We examined cortical integration of associations using implicit memory measures.• Fast mapping enables immediate integration of associations into cortical networks.• Semantic integration requires the discrimination between items sharing many features.• Item-level lexical integration is unaffected by feature overlap. Sharon et al. (2011) examined this learning procedure in four amnesic patients suffering from severe lesions to the medial temporal lobe, predominantly to the hippocampus. These patients did not recognize the picture-label associations above chance level if the associations were intentionally learned within a standard explicit encoding (EE) task, in which they were explicitly asked to remember an unknown item together with its label. This might be attributed to their reduction in hippocampal volume as this is a task in which the hippocampus typically would be recruited. Interestingly, when the same patients encoded novel associations within the FM paradigm, their recognition performance was as good as that of healthy controls, implying that hippocampal processing can be bypassed through learning by means of FM 1 . Despite this evidence that FM might enable successful direct integration of associations,other studies revealed contradictory findings (cf. Greve et al., 2014;Smith et al., 2014; Warren & Duff, 2014; Warren et al., 2016). In order to resolve these contradictions, factors mediating learning success in the FM paradigm yet need to be identified. Sharon et al. (2011) suggested three key determinants to be crucial for successful learning by means of FM: (1) Learning needs to be incidental. (2) The picture-label associations need to be actively discovered by the participants themselves through a process called disjunctive syllogism, that is, excluding the previously known item in order to create a link between the label and the unknown item. (3) The new associations need to be learned in the context of previously known information, activating alr...

High feature overlap reveals the importance of anterior and medial temporal lobe structures for learning by means of fast mapping

“…For example, no memory benefit from FM was observed for older adults with reduced hippocampal volume as a result of healthy aging (Greve et al, 2014) and hippocampal contribution to learning through FM in healthy young adults has been reported by Atir- Sharon et al (2015) or at least could not finally be ruled out by Merhav et al (2015). As already proposed previously (e.g., Atir-Sharon et al, 2015, and Merhav et al, 2014Zaiser et al, 2019b), it might be over-simplified to claim that FM encoding is necessarily hippocampus-independent and hippocampal contribution to FM learning should be discussed in a more differentiated manner.…”

“…There is complementary behavioral evidence that rapid semantic integration (as measured by means of semantic priming effects) through FM benefits from a high similarity between the objects that need to be discriminated at encoding (Zaiser et al, 2019b). However, despite this evidence for feature overlap as a moderating factor, it has not yet been investigated which underlying neurocognitive mechanisms and neurofunctional correlates are associated with learning by means of FM.…”

“…In particular, rapid cortical integration through FM includes the discrimination between pictures of complex objects, binding the visual features of an unknown item with an unfamiliar label to a feature conjunction that is represented as a coherent unit, and the integration of this unit into cortical memory networks. Considering the functional characteristics of the PrC and the behavioral findings of successful rapid learning through FM when high feature overlap item pairs were used (Sharon et al, 2011;Zaiser et al, 2019b), our approach was to increase the demands on the discrimination between the unknown and the known item. This should especially recruit the PrC, which might automatically trigger PrC-mediated binding mechanisms.…”

“…Traditional theories of declarative memory assume that learning of novel, arbitrary associations depends on a time-consuming consolidation process, typically based on hippocampal-neocortical interplay (e.g., Frankland & Bontempi, 2005;McClelland, McNaughton, & O´Reilly, 1995). However, there is evidence that rapid and direct cortical integration of novel picture-label associations is possible by means of an encoding procedure called fast mapping (FM; e.g., Himmer, Müller, Gais, & Schönauer, 2017;Merhav, Karni, & Gilboa, 2014Sharon, Moscovitch, & Gilboa, 2011;Zaiser, Meyer, & Bader, 2019b). Sharon and colleagues (2011) reported a clear benefit from encoding within the FM paradigm in patients with severe lesions predominantly to the hippocampus.…”

“…Hence, here we explicitly manipulated the demands on the discrimination between the previously known and the unknown item (as in Zaiser et al, 2019b) in an fMRI experiment, contrasting an FM encoding condition in which the previously unknown and the known item shared many features (fast mapping, high feature overlap; FMHO) to an FM encoding condition in which they shared few features (fast mapping, low feature overlap; FMLO). We expected that PrC engagement at encoding should be greater if the demands on perirhinal processing (i.e., the discrimination between complex objects) are higher (as in the FMHO condition) than if they are lower (as in the FMLO condition).…”