Priming in melody perception: tracking down the strength of cognitive expectations

Marmel, F.1, Tillmann, B.1, Delbé, C.2

1Université de Lyon, Université Lyon 1, Laboratoire Neurosciences Sensorielles Comportement Cognition, CNRS UMR 5020, IFR 19, Lyon

2Université de Bourgogne, Laboratoire de l’Etude de l’Apprentissage et du Développement, CNRS UMR 5022

Journal of Experimental Psychology : Human Perception and Performance
(revised version submitted)
Correspondence concerning this article should be addressed to

Frédéric Marmel

Université Claude Bernard - Lyon I

CNRS UMR 5020

Neurosciences Sensorielles Comportement Cognition

50 Av. Tony Garnier

F-69366 Lyon Cedex 07

France

Tel : +33 (0) 4 37 28 74 90

Fax : +33 (0) 4 37 28 76 01

frederic.marmel@olfac.univ-lyon1.fr

The musical priming paradigm has shown facilitated processing for tonally related over less-related targets. However, the congruence between tonal relatedness and the psychoacoustical properties of music challenges cognitive interpretations of the involved processes. Our goal was to show that cognitive expectations (based on listeners’ knowledge about the Western tonal system) elicit tonal priming in melodies independently of sensory components (e.g., spectral overlap). A first priming experiment minimized sensory components by manipulating tonal relatedness with a single (possibly repeated) tone change in the melodies. Processing was facilitated for related over less-related target tones, but an auditory short-term memory model succeeded in simulating this tonal relatedness effect, thus suggesting a sensory-based explanation. When the same melodies were played with pure tones (instead of piano tones), the sensory model failed to differentiate between related and less-related targets, while listeners’ data continued to show a tonal relatedness effect (Experiment 2). The here observed tonal priming effect in melodies thus provides strong evidence for the influence of listeners’ tonal knowledge on music processing. The overall findings point out the need of controlled musical material (and notably beyond tone repetition) to study cognitive components in music perception.

Keywords: priming, cognitive, sensory, tonal expectations

A context allows perceivers to generate expectations for future events, leading to facilitated processing (i.e., priming) of expected events. Expectations can be either sensory-driven, generated by the physical similarity or identity between events (i.e., repetition priming), or schemata-driven, based on perceivers’ knowledge about the world’s structure (i.e., contextual probability of occurrence or conceptual relations between events). For language, cognitive priming involves, for example, the strength of association between words (e.g., spider and web) or their semantic relatedness (e.g., plane and car). Cognitive and sensory components of expectations are not mutually exclusive, as reflected in semantically related words being physically related (e.g., nurse and nursery). Similarly, Western tonal music contains regularities that can generate sensory- and cognitive-based expectations. Our present study focused on the cognitive components of musical expectations in melody perception by controlling the sensory components of expectations in the melodic material. We investigated the influence of these expectations (referred to hereafter as cognitive and sensory expectations) on the speed of tone processing. While the influence of cognitive expectations on processing speed has been previously studied for chords (Tekman & Bharucha, 1998; Bigand, Poulin, Tillmann, Madurell, & D’Adamo, 2003), our study investigated the processing of tones and, for the first time, takes in consideration sensory influences beyond tone repetition, notably by considering the contribution of the spectral richness of the sound.

Regularities in Western tonal music that generate sensory expectations include tone repetition (Tekman & Bharucha, 1998; Justus & Bharucha, 2001; Bigand et al., 2003; Bigand, Tillmann, Poulin-Charronnat, & Manderlier, 2005; Hutchins & Palmer, 2008), whereas regularities responsible for cognitive expectations are linked to tonal structure (Bigand et al, 2003; Dowling & Harwood, 1986; Krumhansl, 1990, 2004). Tonal structure refers to a system of conceptual relations between musical events (i.e., tones and chords), which can be described in terms of frequencies of co-occurrence between events (association strengths) and frequencies of occurrence of these events. A growing set of research provides evidence that listeners without formal musical training have acquired implicit knowledge about tonal structures by mere exposure to music in everyday life thanks to implicit learning processes (Bigand & Poulin-Charronnat, 2006; Francès, 1958; Tillmann, Bharucha, & Bigand, 2000).

In the Western tonal system, twelve pitch classes (C, C#/Db, D, D#/Eb, E, F, F#/Gb, G, G#/Ab, A, A#/Bb, B) are organized in subsets of seven tones, defining tonalities (or keys). A chord (i.e., at least three simultaneously sounding tones) can be built on each of these seven tones. The tones and chords of a key are hierarchically organized to reflect different tonal functions. For tones, the tonal function of the first degree is called the tonic, and it represents the most referential member of a key having the highest tonal stability and defining a reference point to which other tones are anchored (Bharucha, 1984; Krumhansl, 1990). In the tonal hierarchy, the tonic is followed by the dominant (5th degree), and the mediant (3rd degree). The leading tone (7th degree) is the lowest degree among the in-key tones. A similar hierarchy can be described for chords, with the chord built on the first degree (i.e., the tonic chord) being at the top, followed by the chords built on the 5th and 4th degrees (referred to as dominant and subdominant, respectively). Since the tonal system is based on a restricted set of twelve tones, context dependency is one of its critical features. The same tone (or chord) occurs in several keys and its tonal function depends on the other tones of the context and its underlying key. For example, the tone C functions as the tonic in the key of C major (when the subset of used tones is C, D, E, F, G, A, B), but as the leading tone in the key of Db major (used subset: Db, Eb, F, Gb, Ab, Bb, C).

The seminal probe-tone studies by Krumhansl and collaborators showed that listeners perceive final tones (and chords) differently depending on their tonal function in the currently instilled key (see Krumhansl, 1990 for a review). A short tonal context (i.e., a scale, a chord sequence) was followed by one of the 12 tones as probes, and participants’ ratings about how well each probe tone fitted with the preceding context resulted in key-specific profiles reflecting tonal hierarchy: in-key tones provided a better fit with the context than out-of-key tones, while among in-key tones, the highest ratings were observed for the tonic tone, followed by the tones of the 5th and 3rd scale degrees, with the lowest values among in-key tones being observed for the leading tone.

Tonal functions are related to statistical regularities in the use of musical events: in a musical piece, tones and chords belonging to a key are frequently associated, and those with highly referential tonal functions (e.g., the tonic) are used more often than less referential ones (Francès, 1958; Krumhansl, 1990). Listeners become sensitive to these regularities via implicit learning processes and gain implicit knowledge about the tonal functions based on these regularities. Listeners’ perception of tonal hierarchy, as revealed by the key-profiles of Krumhansl and Kessler (1982), correlates partly with the tones’ frequencies of occurrence (see Huron, 2006). For example, the tonic is more stable than the subdominant tone in tonal hierarchy and it is also occurring more frequently in musical pieces. Thus, tonal function is entwined with sensory components in musical contexts: tonally stable events tend to be used more often than less stable (or unstable, out-of-key) tones.

To study the cognitive components of music perception (i.e., listeners’ tonal knowledge of tonal function), it is thus necessary to control sensory components in the musical material. Music cognition research has provided evidence for listeners’ tonal knowledge either by post-hoc analyses or controlled experimental materials. Regression analyses, which separated variance explained by sensory and cognitive components, revealed the influence of listeners’ tonal knowledge on tone and chord perception, in addition to the influence of sensory components like tone repetition (Hébert, Peretz, & Gagnon, 95) and chord repetition (Bigand & Pineau, 1997). Other studies controlled sensory components directly in the experimental material (Krumhansl, 1979; Bigand, 1997). For example, pairs of melodies were constructed in such a way that melodies differed only by some tones, and listeners’ judgments of musical tension showed that the same tones were interpreted differently depending on the key instilled by the manipulated tones (Bigand, 1997).

Experimental data on tone and chord perception have thus provided some evidence for listeners’ knowledge about tonal functions. For musical expectations, more specifically, studies on harmonic priming have investigated the respective contributions of cognitive and sensory components in local contexts (i.e., pairs of chords: Bharucha & Stoeckig, 1987; Tekman & Bharucha, 1998) and global contexts (i.e., chord sequences: Bigand & Pineau, 1997; Bigand et al., 2003). In their harmonic priming studies, Bigand and collaborators (Bigand et Pineau 1997; Bigand et al., 1999, 2003) presented a prime context (i.e., a chord sequence) followed by the to-be-processed target event (i.e., a chord). The relation between the prime and the target chord was manipulated by changing the prime and keeping the target identical, so that the same target chord was either the most important chord in tonal hierarchy (i.e., the tonic) or a less-important chord in tonal hierarchy (i.e., the subdominant). This resulted in the same target being more expected (i.e., the tonic, being higher in the tonal hierarchy of the prime context) or less expected (i.e., the subdominant, being lower in the tonal hierarchy of the prime context). The sequences with the target functioning as the tonic defined the related condition and the sequences with the target functioning as the subdominant defined the less-related condition. To keep local sensory influences constant and to focus on global context and global relatedness of the target, the penultimate chord was the same in related and less-related conditions. The penultimate chords and targets were chosen to keep the distance between the tonal degrees of these chords on the circle of fifths (i.e., 1 step) and the commonality of the resulting local chord progression (Piston, 1978) constant between the two conditions. These priming studies showed facilitation processing for the more expected, related tonic targets than for the less-expected, less-related subdominant targets. The harmonic priming paradigm has the advantage to be an indirect investigation of the context’s influence on event processing: participants are not required to make direct judgments on the relation between prime context and target, but the task requires the judgment of a perceptual feature of the target chord. The perceptual task can be based on sensory consonance/dissonance judgments (Bharucha & Stoeckig, 1986, 1987; Bigand, Madurell, Tillmann, & Pineau, 1999), timbre identification (Tillmann, Bigand, Escoffier, & Lalitte, 2006) or phoneme identification for sung music (Bigand, Tillmann, Poulin, D’Adamo, & Madurell, 2001). This indirect investigation method is particularly interesting for the investigation of implicit tonal knowledge of nonmusician listeners, as in our present study (Bigand & Poulin-Charronnat, 2006; Tillmann, 2005).

To focus on cognitive priming, sensory differences between related and less-related conditions (i.e., repetition priming) have been controlled in the experimental material. After one-chord primes, related targets were processed faster than unrelated targets, even when prime and target chords did not share any component tones (Bharucha & Stoeckig, 1987), and when they were separated by long silent intervals or a white noise burst (Tekman & Bharucha, 1992). After longer prime contexts (i.e., chord sequences), processing of related targets was facilitated even when less-related targets shared more tones with the context than related targets (Bigand et al., 2003) or when immediately preceded by the same chord (Bigand et al., 2005).

To track down the strength of cognitive components in music perception, our present study controlled sensory components by 1) adapting the priming paradigm to melodies, 2) manipulating the target’s tonal function with a single tone change, and, most importantly, 3) varying the spectral richness of the used sound (piano vs. pure tones), and thus controlling sensory influences beyond tone repetition. Since melodies are monophonic tone sequences (i.e., only one tone is played at the same time), they provide less sensory information than chord sequences (i.e., at least three tones played at the same time). The used melodic material consisted of pairs of nearly identical melodies where the target (i.e., the last tone) functioned either as a tonic (i.e., the first degree) or as a subdominant (i.e., the fourth degree). This change of the target’s tonal function was instilled by a semitone change of one (possibly repeated) tone in the first part of the melodies. This manipulation controlled tone repetition so that the target pitch occurred equally often in related and less-related conditions (and thus leading to equally strong expectations for the target tone on the basis of tone repetition). This manipulation controlled also other local influences such as melodic contour as well as interval size and direction⁶. The remaining sensory components of the melodic contexts were tested with a sensory model, leading us to manipulate timbre complexity (piano tones in Experiment 1, pure tones in Experiment 2).

Our study includes two priming experiments and simulations with an auditory short-term memory model (Leman, 2000). Experiment 1 showed that the melodies described above were able to elicit a tonal relatedness effect, thus pleading for cognitive expectations. However, the simulations with the sensory model predicted the tonal relatedness effect observed in Experiment 1. This observation challenges the cognitive interpretation of the observed priming effect and highlights the importance of the redundancy between acoustic and syntactic information in music. Experiment 2 thus set an additional control of sensory components by playing the same melodies with pure tones. This reduced the periodicity pitch information provided by the context melodies, and the sensory model failed to predict a difference between related and less-related target tones. However, behavioral data of Experiment 2 still showed a tonal priming effect with pure tones: This tonal relatedness effect for human listeners together with the absence of a tonal relatedness effect for the sensory model provides evidence for the role of cognitive processes in musical expectations.