General discussion

This study investigated the strength of cognitive tonal expectations in music perception by severely controlling for sensory expectations. Notably, our study: 1) used monophonic sequences (melodies), 2) equalized melodic contour, intervals, and target tone repetition between tonally related and less-related conditions, and 3) running the simulation with the sensory model led us to reduce the spectral richness of tones as much as possible by using pure tones (Experiment 2). Using monophonic sequences (melodies) instead of polyphonic sequences (i.e., chord sequences: Bigand & Pineau, 1997; Bigand et al., 2001, 2003; Tillmann et al., 2000) was a first step in controlling sensory expectations because it reduced the spectral quantity of sensory information that has to be controlled. Melodic contour, interval size and direction, as well as target tone repetition had all been found to play a role in musical expectations (Bigand et al., 2005; Boltz & Jones, 1986; Cuddy & Lunney, 1995; Hutchins & Palmer, 2008; Krumhansl, 1995), and had thus to be controlled. This was done by designing related and less-related conditions that differed only by a single (possibly repeated) tone. This tone difference was always at the beginning of the melodies (i.e., in the first bar), allowing us to control the local context of the target tone and to focus on global expectations. Since the related and less-related melodies were almost identical, we first tested if participants were sensitive to the remaining tonal difference by collecting completion judgements. This pretest showed that the subtle manipulation was sufficient to elicit higher completion judgements for related than for less-related melodies. Experiment 1 then focused on target tone processing and showed a tonal priming effect: facilitated processing was observed for tonally related target tones over less-related target tones, despite the difference of only one note between related and less-related melodies. However, the success of Leman’s (2000) sensory model to simulate this priming effect (i.e., simulations with piano tones) revealed the need to go beyond the use of melodies controlled for tone repetition and led us to conduct Experiment 2 with the spectral richness of tones being also controlled. The use of pure tones reduced the spectral richness of tones to the minimum, thus reducing as much as possible the amount of acoustic information differentiating the two tonal relatedness conditions. For this experimental material, the sensory model failed to predict a priming effect (see simulations for pure tones), while human listeners still showed facilitated processing for tonally related target tones over less-related target tones (Experiment 2). It is worth underlining that no effect of musical expertise was observed in the two priming experiments: processing was facilitated for both moderately experienced and less-experienced listeners. This finding suggests that even nonmusician listeners (i.e., without explicit musical training) have sophisticated knowledge about tonal relations that influences tone processing. In contrast to the priming data, an influence of musical expertise was observed for the completion judgments: the effect of tonal relatedness was weaker (but still significant) for the less-experienced participants. This is congruent with reports that an implicit investigation method like the priming paradigm is more powerful than explicit paradigms like completion judgments (Tillmann, Peretz, Bigand, & Gosselin, 2007).

Our study goes beyond previously reported harmonic priming studies investigating cognitive expectations for chord perception. In these studies, the prime context consisted of chords and the acoustic overlap with the target chord was taken into account by controlling chord and tone repetition (Bigand et al., 2003; Tekman & Bharucha, 1998; Bharucha & Stoeckig, 1987, Experiment 1) and avoiding repeated harmonics (for pairs of chords: Bharucha & Stoeckig, 1987, Experiment 2). Tekman and Bharucha (1998) noted that even avoiding the repetition of harmonics between target and prime is not sufficient to rule out sensory expectations because of the potential influence of inferred virtual pitches, which might be congruent with cognitive expectations (Parncutt, 1989). This difficulty to disentangle cognitive and sensory influences led them to use pairs of chords (i.e., a prime chord followed by one of two possible target chords), for which cognitive and sensory expectations resulted in opposite predictions for target processing. In one condition, the target chord was more strongly tonally related to the prime, but did not share any component tones whereas in the other condition the target chord shared one component tone with the prime chord but was tonally less related to it (see Bigand et al., 2003 for a similar rationale with longer contexts). However, these studies did not confront their material manipulation with predictions from a sensory-based model. The influence of sensory expectations beyond tone repetition is revealed in our simulations for piano tones with the auditory short-term memory model by Leman (2000). These simulations show that the control of sensory factors by controlling tone repetition is not sufficient to unambiguously show the contribution of cognitive components in musical expectations: facilitation of related over less-related targets can be predicted despite the equalized target tone’s frequency of occurrence in related and less-related melodies. In Leman’s model, the simulation of peripheral auditory processing, the consideration of the sound spectral complexity (which generates subharmonics and virtual pitches) and the temporal patterns of these spectral components, together with the inclusion of a short-term memory buffer, succeeded in picking up the intertwined relation between tonal relatedness and the psychoacoustic properties of sound that persisted in our melodies when played by piano tones. The additional manipulation of spectral complexity by using pure tones was found necessary to disentangle the influence of sensory expectations and to track down the contribution of cognitive expectations.

The findings of our study support the hypothesis of cognitive expectations elicited by the global tonal framework even in listeners with weak or no musical expertise. This hypothesis, which has been previously studied by Krumhansl and collaborators (Krumhansl, 1979; Krumhansl & Kessler, 1982; Krumhansl & Shepard, 1979), has been challenged by alternative explanations based only on acoustical information and without the need for tonal knowledge (Parncutt, 1994; Huron & Parncutt, 1993; Leman, 2000). For example, the sensory model used by Leman (2000) simulated the probe-tone profiles of Krumhansl & Kessler (1982) without postulating any cognitive processes. The findings of our study strengthen the cognitive hypothesis: our data on pure tones argue for tonal expectations being elicited by cognitive processes.

Since our findings were obtained for melodies, they can also be put in the light of the study by Holleran, Jones, and Butler (1995) investigating the perception of implied harmony in melodies. In this study, a pitch change detection task showed that listeners were sensitive to the harmony implied by a melodic context. However, the authors discussed the possibility that their results may come partly from a confound in their material between tonal relatedness and frequency of occurrence. This confound was an “inevitable consequence of the tonal manipulations” (p.749, Holleran et al., 1995), but limits the cognitive interpretation of the findings when considered on their own. Our study controlled for the frequency of occurrence and still found an effect of tonal relatedness (Experiments 1 and 2). Therefore, it corroborates the findings of Holleran et al. with a more controlled material.Moreover, the simulations with piano tones retrospectively highlight the importance of controlled experimental material (even beyond frequency of occurrence) since the remaining acoustical differences were still sufficient for the model to simulate the tonal relatedness effect.

It might be argued that different mechanisms are involved in the priming effects of our two experiments. Cognitive processing of tonal relatedness may be involved only as a back-up mechanism when there are not enough differences in sensory components of the musical stimulus to prime a target tone. In this view, cognitive processing of tonal relatedness would be involved in Experiment 2, while sensory expectations would be the sole factor eliciting the priming effect in Experiment 1. Since pure tones, which were necessary to ensure that sensory components of musical expectations could not result in a priming effect, are not ecological stimuli, the finding of cognitive priming in Experiment 2 would thus be of little psychological relevance. However, this hypothesis postulates that cognitive processing of tonal relatedness is mandatory whereas previous studies on harmonic priming found it to be automatic (Bigand et al., 2001; Justus & Bharucha, 2001; Tillmann, Janata, Birk, & Bharucha, 2003). Justus and Bharucha (2001), for example, manipulated veridical against schematic expectations in three harmonic priming experiments. Veridical information was given to participants in the form of prime-target previews, local transition probabilities, or valid versus invalid previews. These veridical expectations were either fulfilled or violated in the experiments, and were either congruent or incongruent with schematic expectations(linked to harmonic relatedness between prime and target chords). Independently of the manipulations of veridical expectations, an effect of harmonic priming was observed, thus suggesting that schematic expectations are elicited automatically (see Tillmann & Bigand, 2004 for a similar outcome). In addition to the argument of the automaticity of musical priming, mandatory cognitive expectations, which act as a backup to paramount sensory expectations, seem unlikely when considering the different musical instruments with varying spectral complexity to which listeners are exposed to in ecological musical recordings. Musical instruments of varying spectral complexity are likely to result in different amounts of available sensory information. This variability in sensory information suggests that sensory expectations are less reliable as a baseline mechanism for tonal expectations than are cognitive expectations, which do not depend on spectral complexity. Automatic involvement of cognitive expectations, with additional sensory expectations of varying magnitude, thus seems to be a more likely underlying process in ecological musical contexts.