Implicit learning of melodic structure Martin Rohrmeier, Patrick Rebuschat University of Cambridge
Implicit learning Definition Reber (1965) Acquisition of a complex, rule-based system Without intention to learn (incidental learning) Without awareness of the learning outcome Characteristics (Reber 1989, Frensch & Rünger 2003) Abstract Tacit knowledge Independent of IQ and age Unconscious Root process for interaction with environment
Implicit learning Major research paradigms Artificial Grammar learning & sequence learning Reber 1965 Nissen & Bullemer 1987 Dienes et al. 1995 Control of complex, dynamic systems Berry & Broadbent 1984; Broadbent, Fitzgerald & Broadbent 1986 Language acquisition Rebuschat & Willams 2006 Relevance for music perception
Experiment Investigating musical structure acquisition under implicit learning experimental paradigm Learning phase with distraction task Testing phase Basic framework: Acquisition of a rule based sequential, monophonic musical system Modelled as finite state grammar (standard paradigm in implicit learning experiments)
Experiment 1 Finite-State Grammar 13 states, 9 tone pairs 33 sequences (between 8 and 30 tones) Learning phase: 17 sequences presented 3 times in randomised order Testing phase: 33 grammatical sequences (17 old, 16 new) 33 ungrammatical sequences
Experiment 1 Ungrammatical stimuli Applying different error types Error type 1: random order of lexical elements Error type 2: randomised sequence, bigrams intact Error type 3: anchor positions correct, middle randomised Error type 4: upper and lower paths switched Error type 5: grammatical stimulus with one order exchange
Experiment 1 Experimental setup overview Learning phase Testing phase 3 blocks of 17 grammatical stimuli in randomised order total: 51 sequences Tone counting distraction task Testing phase 66 stimuli: 33 grammatical 17 old-grammatical (from learning phase) 16 new-grammatical 33 ungrammatical Forced choice familiarity and confidence judgments
Results experimental group control group musicians 72,00% 62,04% nonmusicians 69,97% 61,16% average 70,98% 61,59% n=59 subjects (22 experimental, 37 control group) Learning effect (both conditions) No difference between musicians and nonmusicians
Results Performance across stimulus types Experimental group outperforms control group in all conditions No difference between new-gram and old-gram for control group (confirms control condition)
Results Performance across error types Experimental group above chance for Error types 1-4 Control group above chance for Error type 1 and 2 Implications for experimental design
Results Control group performs significantly above chance Performance increases over time (displayed at every 6 responses) Online-learning effect
Experiment 2 Learning effect: surface or underlying structure? Experiment 2: Cross-lexical transfer paradigm Different lexicon, but same underlying grammar in learning and testing phase Lexicon B: 9 different diatonic tone pairs Experimental Control Cross-lexical Learning phase Lexicon A - Lexicon B Testing phase
Results N=13 subjects for cross-lexical condition No sig. difference between cross-lexical and control group for overall performance, across all stimulus and error types
Results Learning not enhanced by structural transfer Cross-lexical group performance replicates control group Conclusion Learning may be based on surface features
Conclusions Experiment 1: Finite state-grammar, tone sequencees Evidence for implicit learning found Strong online learning effect Experiment 2: Cross-lexical transfer No effect found No performance difference for musicians – nonmusicians Performance varies across error types Implications for experimental design Acknowledgements This study has been carried out under the funding of the AHRC, and the Microsoft European PhD Scholarship Programme.