Applying an exemplar model to an implicit rule-learning task: Implicit learning of semantic structure

Abstract

Studies of implicit learning often examine peoples' sensitivity to sequential structure. Computational accounts have evolved to reflect this bias. An experiment conducted by Neil and Higham [Neil, G. J., & Higham, P. A.(2012). Implicit learning of conjunctive rule sets: An alternative to artificial grammars. Consciousness and Cognition, 21, 1393-1400] points to limitations in the sequential approach. In the experiment, participants studied words selected according to a conjunctive rule. At test, participants discriminated rule-consistent from rule-violating words but could not verbalize the rule. Although the data elude explanation by sequential models, an exemplar model of implicit learning can explain them. To make the case, we simulate the full pattern of results by incorporating vector representations for the words used in the experiment, derived from the large-scale semantic space models LSA and BEAGLE, into an exemplar model of memory, MINERVA 2. We show that basic memory processes in a classic model of memory capture implicit learning of non-sequential rules, provided that stimuli are appropriately represented.

Using experiential optimization to build lexical representations

“…More generally, the present work points to the importance of models capable of extracting information from large text bases, an issue that has been explored in greater detail elsewhere (e.g., Chubala et al, 2016;Hills, Jones, & Todd, 2012;Johns & Jones, 2010;Johns, Jones, & Mewhort, 2012;Johns & Jones, 2015;Johns, Taler, et al, 2017;Mewhort et al, 2017;Taler et al, 2013). Basing a model's performance on largescale environmental information provides a strong case for the model's plausibility, because it can scale to human levels of experience.…”

“…It is known that an exemplar memory system can accomplish some of the fundamental operations of language (Abbot-Smith & Tomasello, 2006). Jamieson and Mewhort (2009a, 2009b and Chubala, Johns, Jamieson, and Mewhort (2016), for example, have shown that such a model can account for several artificial-grammar and implicit-learning results. Johns and Jones (2015) extended the approach; their account explained additional results across sentence processing, grounded cognition, and the cultural evolution of language.…”

Attentional responses to stimuli associated with a reward can occur in the absence of knowledge of their predictive values

“…Prior literature in the field of Implicit learning has primarily focused on artificial grammarlearning tasks [31,32] and sequence learning [33], but few studies have used self-reports in conditioning experiments [34,35]. In these latter studies, however, no implicit conditioned responses were found.…”

The Memory Tesseract: Developing a Unified Framework for Modelling Memory and Cognition

“…In the GCM, similarity is raised to the power of c, where c is a sensitivity parameter set by the modeller. Conversely, in MINERVA 2, the exponent is always 3, though some variants of MINERVA dynamically vary the exponent (e.g., Mewhort & Johns, 2005) or use a larger exponent to minimize noise (e.g., Johns et al, 2016).…”

“…To preserve the sign of the similarity, the MINERVA 2 model uses an exponent 3 rather than 2. Larger odd numbered exponents (5, 7, 9, ...) also preserve the sign of the similarity and can be used to further reduce the amount of noise in the echo (e.g., Johns et al, 2016).…”