Differential roles for implicit and explicit knowledge in distributional learning

Cylcia Bolibaugh, Patrick Rebuschat

Research output: Contribution to conferencePosterpeer-review


A number of recent studies have demonstrated that adults exposed to an artificial language can derive abstract categories on the basis of distributional information in the input alone (e.g. Reeder et al 2013). However, findings also suggest (e.g. Mintz 2002, 2014) that learning is often driven by endorsement of distributionally supported strings rather than reliable rejection of non-conforming strings. This pattern of results mirrors observations from natural and artificial second language learning studies in which adults incidentally exposed to linguistic regularities develop knowledge of attested patterns and are able to generalise these to novel grammatical instances but do not discriminate ungrammatical items at levels above chance. On the basis of these and similar findings, it has been proposed that implicit second language knowledge results in similarity based prototype effects whilst categorical discrimination requires explicit knowledge.

Although most experiments investigating distributional learning in adults are meant to simulate the implicit inductive mechanisms assumed to operate in infants and children, it is presently unclear what type of knowledge results from adult distributional learning, and how the conscious status of this knowledge interacts with grammaticality. In the present study, we investigate the extent to which distributional learning in incidental conditions results in implicit and explicit knowledge, and further distinguish the contribution of each to transfer (generalisation) and rejection (discrimination) of novel distributionally consistent and distributionally atypical strings. Subjects (n=32) were incidentally exposed to auditory strings of pseudowords generated by a (Q)AXB(R) grammar. In an auditory exposure condition, adults (n=16) passively listened to the language. In a semantic referent condition, participants (n=16) listened to the language while viewing line drawn illustrations of scenes in which the AXB elements were mapped to agents, actions and objects. Thus the auditory condition provided only distributional information based on the patterning of the words in sentences, whilst participants in the semantic condition benefitted from a correlated cue to word class category structure. All participants were then asked to rate trained and novel sentences based on their confidence that the strings were part of the language they had been exposed to. After each rating, participants additionally reported the source of their decision, whether recollection, rule, intuition or guess.

Findings suggest that distributional learning in adults relies on explicit memory based processes which bias learners to generalise more widely than the statistical evidence warrants, and rule based knowledge appears to be necessary to appropriately restrict generalisations. A secondary cue not only changes the overall distribution of the type of knowledge which is generated, in that it enables more rule-based judgments, but also promotes discrimination (as seen in reliable rejection of ungrammatical strings). Where the stimulus environment enables greater confidence in the systematicity of the underlying structure (via correlated semantic cues), learners more readily formulate rules and concomitantly more accurately restrict their generalisations.

Mintz, T. H. (2002). Category induction from distributional cues in an artificial language. Memory & Cognition, 30(5), 678-686.

Mintz, T. H., Wang, F. H., & Li, J. (2014). Word categorization from distributional information: Frames confer more than the sum of their (bigram) parts. Cognitive psychology, 75, 1-27.

Reeder, P. A., Newport, E. L., & Aslin, R. N. (2013). From shared contexts to syntactic categories: The role of distributional information in learning linguistic form-classes. Cognitive psychology, 66(1), 30-54.
Original languageEnglish
Publication statusPublished - 23 Jun 2016
EventFifth Implicit Learning Seminar - Lancaster University, Lancaster, United Kingdom
Duration: 23 Jun 201625 Jun 2016


ConferenceFifth Implicit Learning Seminar
Country/TerritoryUnited Kingdom


  • distributional learning
  • categorisation
  • implicit and explicit learning

Cite this