Intentional control based on familiarity in artificial grammar learning
Introduction
Much of the knowledge we acquire for dealing with the world appears to be implicit. We can learn to appreciate certain styles of music, obey cultural rules, or gain perceptual motor mastery of a domain without consciously knowing the underlying regularities. Reber (1967) initially introduced the artificial grammar learning paradigm as a way of investigating such implicit learning. Typically, in artificial grammar learning experiments, subjects are asked to memorize or look at letter strings for some minutes, and only then told that a complex set of rules underlay these training strings. In the following test stage, subjects are asked to classify each test string as grammatical or not. Generally, classification performance is above chance level (typically about 65%). Thus, people can learn the structure of an artificial grammar without trying to do so and in fact in such a way that the knowledge is difficult to express (e.g., Reber, 1967, Reber, 1989, Berry and Dienes, 1993, Cleeremans et al., 1998, Pothos, 2007, Shanks, 2005.
It is a common experience to find a person or event unexpectedly familiar or unfamiliar for reasons we could not state. Further, familiarity can be acquired incidentally. Thus, it is natural to speculate that processes of familiarity play a role in implicit learning (e.g. Higham, 1997, Shanks et al., 2003, Whittlesea and Leboe, 2000, Tunney, 2007). Indeed, knowledge of specific strings or parts of strings (chunks) play a central role in artificial grammar learning (e.g., Dulany et al., 1984, Lotz and Kinder, 2006, Perruchet and Pacteau, 1990, Servan Schreiber and Anderson, 1990). In these cases people are sensitive to the presence of stimuli that are objectively familiar to them, i.e., as matter of fact they have come across those chunks before. For instance, the process of familiarity has been indicated by estimating the relationship between grammatical classification and fragment frequency (Knowlton and Squire, 1996, Meulemans and Van der Linden, 1997, Servan Schreiber and Anderson, 1990). In contrast to objective familiarity, Scott and Dienes (in press) explored the role of subjective familiarity in artificial grammar learning, that is, familiarity as a feeling (the feeling that something is objectively old). In the test phase, subjects were required to give a subjective rating of familiarity for each test string. Such subjective familiarity correlated both with the tendency to call an item ‘grammatical’ and also with objective properties of the test string, such as the frequency with which its chunks occurred in the training strings. In addition, Scott and Dienes asked subjects to indicate the basis of their grammaticality classification for each string (Dienes & Scott, 2005), with five options: guessing/random responding, intuition, familiarity, rules or memory. The most common choice was familiarity. That is, subjects often believed that their grammaticality classifications were indeed based on the relative familiarity of the strings.
Familiarity can be defined not just as an objective relation (of having been previously in mutual contact) or as a feeling, but also in terms of control. Specifically, Jacoby (1991) defined familiarity as that memorial process not affected by intentional control. Familiar items tend to be chosen regardless of one’s intentions. Consider a subject asked to look at strings from two different grammars, grammars on which the subject has been trained to an equal extent. If the subject is shown test strings from both grammars familiarity would not, on Jacoby’s definition, allow the person to choose strings from just one or other of the grammars. Dienes, Altmann, Kwan, and Goode (1995), however, confirmed that people trained on two grammars in turn could substantially control which grammar they used. When people were asked to respond to just one grammar and treat the other grammar as ungrammatical, they could do so. However, Dienes et al. did not determine on which knowledge sources it appeared to subjects they based their decisions. Maybe subjects used recollection or rules to discriminate the grammars. The results of Dienes et al. raise the question of whether subjective familiarity could be manipulated by intentions.
In the current study, we conducted two experiments to explore whether subjective familiarity could be controlled intentionally when subjects are trained on two artificial grammars. In both experiments, we replicated the Dienes et al. (1995) finding that incidentally acquired knowledge of two artificial grammars could be applied strategically and explored whether such control could be exerted when people felt they were using familiarity. We asked subjects to rate the familiarity of each string and also state the basis on which they made their grammaticality decision: guessing, intuition, familiarity, rules or memory (see Dienes, 2008, for evidence that such attributions pick out qualitatively different types of knowledge). In Experiment 1 both grammars were trained equally so should induce equal feelings of familiarity. In Experiment 2, the to-be-ignored grammar was trained for twice as long as the target grammar, to determine if intentional control could over-ride even strong training biases in determining subjective familiarity.
Section snippets
Design
We used a 2 × 2 between-subjects design: grammar (first vs. second) × test order (classification first vs. familiarity rating first). In the study stage, all the subjects were trained first on one grammar (grammar ‘A’) and then the second grammar (grammar ‘B’). In the test stage, half of the subjects were asked to check strings from the first grammar; the other half were asked to check strings from the second grammar. In addition, half of the subjects classified and gave source attributions and
Design
The design was the same as Experiment 1, with the following exceptions. We counterbalanced exposure order of grammar A and B in the study stage, and all subjects except for control groups rated familiarity last, after classifying and giving source attributions to the string. There were no detectable effects in Experiment 1 of whether familiarity was rated first or last. Also, display order of the attribution types was varied in the test stage. Furthermore, two control groups were included: in
General discussion
In the current study, we reported the results of two experiments showing subjects could intentionally control which grammar to apply while considering their responses to be based on familiarity. When subjects believed they were using familiarity, the rated familiarity of the consistent grammar was higher than that for the inconsistent grammar, even when the inconsistent grammar had been trained for twice as long. Intentions could over-ride a substantial amount of training in determining
Acknowledgments
This research was supported in part by Grants from 973 Program of Chinese Ministry of Science and Technology (Grant # 2006CB303101) and the National Natural Science Foundation of China (Grant # 60433030).
References (38)
- et al.
Can musical transformations be implicitly learned?
Cognitive Science
(2004) - et al.
Implicit sequence learning and conscious awareness
Consciousness and Cognition
(2008) A process dissociation framework: Separating automatic from intentional uses of memory
Journal of Memory and Language
(1991)- et al.
Gradations of awareness in a modified sequence learning task
Consciousness and Cognition
(2007) Implicit learning of artificial grammars
Journal of Verbal Learning and Verbal Behaviour
(1967)- et al.
Suppressing unwanted memories by executive control
Nature
(2001) - et al.
Implicit learning: Theoretical and empirical issues
(1993) - et al.
Implicit learning: News from the front
Trends in Cognitive Sciences
(1998) Subjective measures of unconscious knowledge
- et al.
Unconscious knowledge of artificial grammars is applied strategically
Journal of Experimental Psychology: Learning, Memory, and Cognition
(1995)
The role of implicit memory in controlling a dynamic system
Quarterly Journal of Experimental Psychology
Measuring unconscious knowledge: Distinguishing structural knowledge and judgment knowledge
Psychological Research
A case of syntactical learning and judgment: How conscious and how abstract?
Journal of Experimental Psychology: General
Dissociations of grammaticality and specific similarity effects in artificial grammar learning
Journal of Experimental Psychology: Learning, Memory, and Cognition
Illusory recollection and dual-process models of recognition memory
Quarterly Journal of Experimental Psychology: Section A
Beyond dissociation logic: Evidence for controlled and automatic influences in artificial grammar learning
Journal of Experimental Psychology: General
A sharper Bonferroni procedure for multiple tests of significance
Biometrika
Memory attributions
Learning artificial grammars: No evidence for the acquisition of rules
Memory and Cognition
Cited by (41)
Distinguishing the role of conscious and unconscious knowledge in evaluative conditioning
2020, CognitionCitation Excerpt :Experiments using this method demonstrated that classification of the test strings was above chance level not only when participants made memory and rules attributions (which indicates conscious knowledge of the structure of the material) but also when they made intuition, familiarity or guessing attributions (which indicates unconscious structural knowledge) (e.g., Scott & Dienes, 2008; see Dienes, 2012, for a review). The same kind of results using the structural knowledge attributions have been obtained not only with artificial grammar learning (e.g. Ivanchei & Moroshkina, 2018; Jurchiş & Opre, 2016), but also with other implicit learning paradigms, such as sequence learning (e.g. Fu et al., 2010), symmetry learning (e.g. Jiang et al., 2012; Ling et al., 2018), second language learning (e.g. Paciorek & Williams, 2015; Rebuschat, 2013; Rogers et al., 2016), probabilistic category learning (e.g., Kemény & Lukács, 2013), learning conjunctive rule sets (Neil & Higham, 2012), and learning multiple grammars (Norman, Scott, Price, Jones, and Dienes, in press; Wan et al., 2008). Hence, participants may acquire knowledge that they are aware of as well as knowledge that they are not aware of in various implicit learning paradigms.
Distinct neural networks for detecting violations of adjacent versus nonadjacent sequential dependencies: An fMRI study
2020, Neurobiology of Learning and MemoryImplicit learning of symmetry of human movement and gray matter density: Evidence against pure domain general and pure domain specific theories of implicit learning
2020, International Journal of PsychophysiologyCitation Excerpt :“Guess” indicated that the judgment could just as well be based on a toss of a coin and that you had no idea whether the given sequence was grammatical or not; “Intuition” indicated that the judgment was based on a hunch or feeling that could not be explicated further, i.e., there was confidence in the judgment but you had no idea why the judgment was right; “Familiarity” indicated that the judgment was based on familiar or unfamiliar for reasons you could not state; “Memory” indicated that the judgment was based on a recollection of part or all of one or more training sequences; “Rule” indicated that the judgment was based on a rule that could be stated if asked. The attributions of guess, intuition and familiarity indicated the unconscious structural knowledge, and those of memory and rule indicated the conscious structural knowledge (e.g., Mealor and Dienes, 2012; Rebuschat et al., 2013; Scott and Dienes, 2008; Wan et al., 2008). All participants completed the Movement Imagery Questionnaire-Revised second version (MIQ-RS) (Gregg et al., 2010), which assessed the visual and kinesthetic imagery ability.
Implicit sequence learning of chunking and abstract structures
2018, Consciousness and CognitionCitation Excerpt :Further, confidence ratings were taken for generation performance to measure awareness of knowing. Awareness of knowing often goes together with ability to control the knowledge, but awareness of knowledge and control of that knowledge can dissociate (Fu, Dienes, & Fu, 2010; Wan, Dienes, & Fu, 2008); thus, both types of measures were used for a nuanced assessment of the explicit nature of the acquired knowledge. For example, if one consciously knew that an item was legal one could include or exclude it as instructed, thereby exerting control; but one need not consciously know why it is legal.
The relationship between strategic control and conscious structural knowledge in artificial grammar learning
2016, Consciousness and CognitionCitation Excerpt :Consciousness of acquired knowledge can be assessed by various subjective awareness measures (Pasquali, Timmermans, & Cleeremans, 2010; Sandberg, Timmermans, Overgaard, & Cleeremans, 2010). If participants are trained on two different grammars, strategic control can be measured by asking them to selectively apply their knowledge of one of these grammars over a given block of test trials (Dienes, Altmann, Kwan, & Goode, 1995; Wan, Dienes, & Fu, 2008; see also Higham, Vokey, & Pritchard, 2000). An alternative procedure – argued to require a higher degree of strategic control – is to let the classification rule vary randomly between individual trials (Norman, Price, & Jones, 2011).
Subliminal understanding of negation: Unconscious control by subliminal processing of word pairs
2013, Consciousness and Cognition