The effect of conditional probability of chord progression in Western music corpus on brain response: an MEG study

THE EFFECT OF CONDITIONAL PROBABILITY OF CHORD PROGRESSION IN WESTERN MUSIC

CORPUS ON BRAIN RESPONSE: AN MEG STUDY

2010-09-11Department of Brain and Cognitive Sciences | SNU

Seung-Goo KIM

THE EFFECT OF CONDITIONAL PROBABILITY OF CHORD PROGRESSION IN WESTERN MUSIC

CORPUS ON BRAIN RESPONSE: AN MEG STUDY

2010-09-11Department of Brain and Cognitive Sciences | SNU

Seung-Goo KIM

INTRODUCTION

NEURONAL CURRENT

http://www.cnsforum.com/content/pictures/imagebank/hirespng/placement_EEG_leads.png

NEURONAL CURRENT


NEURONAL CURRENT


http://www.nmr.mgh.harvard.edu/meg/talks.php

NEURONAL CURRENT


EEG & MEG

Prof. Chang-‐Hwan Im, 2010, Class

EEG & MEG

Prof. Chang-‐Hwan Im, 2010, Classhttp://www.nmr.mgh.harvard.edu/meg/talks.php

INSTRUMENTS

http://www.calvin.edu/academic/psych/experiential/research.html

http://www.virtualmedicalcentre.com/healthinvestigations.asp?sid=11


rMusical scale§ Tonal systems within an octave such

as major or minor

rChord§ Simultaneous tones such as (C,E,G)

rChord function§ The function of a chord within a scale

• (C,E,G) is Tonic (I) in C Major and also Dominant (V) in F Major.

rChord progression§ Changes of chord functions§ “Syntactic” rules govern this progression in Western music

CHORD PROGRESSION

rMusical scale§ Tonal systems within an octave such

as major or minor

rChord§ Simultaneous tones such as (C,E,G)

rChord function§ The function of a chord within a scale

• (C,E,G) is Tonic (I) in C Major and also Dominant (V) in F Major.

rChord progression§ Changes of chord functions§ “Syntactic” rules govern this progression in Western music

CHORD PROGRESSION

PROCESS OF MUSICAL SYNTAX

rKoelsch, 2000, J. Cog. Neurosci.§ Regular (50%), N6 at the 3rd pos (25%), N6 at the 5th pos (25%)

C: I I6 IV V7 I C: I I6 N6 V7 I C: I I6 IV V7 N6

Koelsch et al., 2000, J. Cog. Neurosci.

PROCESS OF MUSICAL SYNTAX

rKoelsch, 2000, J. Cog. Neurosci.§ Regular (50%), N6 at the 3rd pos (25%), N6 at the 5th pos (25%)

C: I I6 IV V7 I C: I I6 N6 V7 I C: I I6 IV V7 N6

Koelsch et al., 2000, J. Cog. Neurosci.

NEURAL GENERATORS OF ERAN

rLocated usingMEG & fMRI

rSeveral studieslocated IFG

rSometimesbilaterallythus simply EAN(esp. in women)

Maess et al, 2001, NN; Koelsch et al., 2005, TICS; Koelsch et al., 2003, Neuroreport.

NEURAL GENERATORS OF ERAN

rLocated usingMEG & fMRI

rSeveral studieslocated IFG

rSometimesbilaterallythus simply EAN(esp. in women)

Maess et al, 2001, NN; Koelsch et al., 2005, TICS; Koelsch et al., 2003, Neuroreport.

AUTOMATICITY & ATTENTION

rAttended condition§ detect the intensity,

not the deviant chord§ (Task-‐irrelevant

-‐target condition: to avoid P300b)

rUnattended condition§ ignoring all chord

with ‘hard’ reading comprehension task

Loui et al., 2005, Cog. Brain Res.

MUSICAL TRAINING

Koelsch et al., 2002, Psychophysiol.

rERAN was greater in experts§ Not in deviant instrument MMN

MUSICAL TRAINING


rERAN was greater in experts§ Not in deviant instrument MMN

SHORT-TERM-EXPERIENCE: HABITUATION

r2 Hours of chords with silent movie» Koelsch, Jentschke, 2008, Brain Research.

§ Target(supertonic)-‐irrelevant task(timber detection)

rDeclined; but not abolished

Koelsch et al., 2008, Brain res.

OTHER IN/OUT-OF-KEY CHORDS

rNeapolitan 6th is out-‐of-‐key chord§ What if with other chords?


OTHER IN/OUT-OF-KEY CHORDS

rNeapolitan 6th is out-‐of-‐key chord§ What if with other chords?


EAN VS. MMN

rMismatch Negativity (MMN)§ Well-‐known ERP response to the rare stimuli (ex: S, S, S, D, S, S, ….)

• Similar in terms of latency, related neural generators,

§ The rarity depends to the short-‐term rarity• Same stimuli could be either a standard or a deviant

Opitz et al., 2002, Neuroimage.

EAN VS. MMN

rEAN: Invariant to the short-‐term rarity§ The irregular chords § The rarity of the progression is rather dependent§ MMN can be elicited by roving paradigm

• EAN can be evoked even with equiprobability

Garrido et al., 2008, Neuroimage.

Roving Paradigm:Firstly, it’s oddballLastly, it’s standard

SHORT-TERM PROBABILITY & MMN

Javitt et al., 1998, Biol Psy.; Haenschel et al., 2005, J. Neurosci.

NEWLY LEARNED PITCH PATTERNS

rUsing artificial scales (Bohlen-‐Pierce)§ Standard pitch pattern (70%) & Deviant pitch pattern (20%)§ 100 sequences X 10 runs = total 40 min

Loui et al., 2009, J. Neurosci.
















MOTIVATIONS

rPossible underlying mechanism of EAN has been suggested as ‘Probabilistic learning’ depending on the previous exposures.§ Unlike the MMN depending on the present situation

» Koelsch, 2009, Psychophy.

§ Learning of chord pattern could evoke EAN» Loui et al., 2009, J. Neurosci.

rHowever, the direct relation of the frequency of exposure in terms of chord progression and the physiological responses has not been investigated yet.

AIMS OF THE PRESENT STUDY

rIf the EANm responses are dependent on the probability in previous exposure

rIf this effect, if any, interacts with the musical expertise and the individual ability of discriminating chord progressions

METHODS

APPROXIMATION OF FREQUENCYrConditional Probability from J. S. Bach’s Chorale

§ Empirical probability as an approximation of frequencies in exposure of music listeners in Western classical context

P(I|V)=0.75; P(vi|V)=0.11; P(ii|V)=0.05Rohrmeier, 2007, Sound and Music Computing Conference, Greece.

TonicP(I|V)=0.75

SubmediantP(vi|V)=0.11

SupertonicP(ii|V)=0.05

I vi ii V

MATERIAL

• 1 chord=0.6 sec (100 BPM)•Modulated to 12 keys• Consecutively presented

• Note that (vi) and (ii) occurred before the last chord to control on-‐line rarity of chord itself

TonicP(I|V)=0.75



I vi ii V

MATERIAL



TonicP(I|V)=0.75



I vi ii V

MATERIAL



TonicP(I|V)=0.75



I vi ii V

MATERIAL



PARTICIPANTS

r11 Non-‐musicians & 9 Musicians (N=20)§ after exclusion for excessive MEG or EOG artifacts§ As no male musicians could be recruited, to control the sex ratio

between groups, all participants were female§ Other 2 participants were discarded for exceed EOG rejections

Mean (SD) N Age (Yr) Handedness (EI)

Musical training (Yr)

Practice (Hour/Week)

Non-‐musicians 11 24.93 (2.5) 91.49 (9.27) 3.63 (1.63) 1.0 (3.0)

Musicians 9 23.50 (3.4) 88.64 (12.78) 18.67 (4.0) 10.22 (6.74)

MEG RECORDING

r1 Block = 100 trails (6 min)§ 3 kinds of sequences x 30 = 90 trials§ 3 kinds of “staccato” sequences = 10 trials

rTask: click the left mouse button on “staccato” chord (randomly at the 2nd ~5th position)§ To preserve the arousal level of the participants

rTotal 6 Blocks with breaks (App. 1 Hour)§ 180 trails were presented for each condition

rElekta Neuromag 306-‐channel whole-‐head MEG§ Anatomical landmarks, HPI coils and points upon the head digitized§ Sampling rate=600Hz; BPF=0.1~200 Hz

MEG RECORDING

r1 Block = 100 trails (6 min)§ 3 kinds of sequences x 30 = 90 trials§ 3 kinds of “staccato” sequences = 10 trials

rTask: click the left mouse button on “staccato” chord (randomly at the 2nd ~5th position)§ To preserve the arousal level of the participants

rTotal 6 Blocks with breaks (App. 1 Hour)§ 180 trails were presented for each condition

rElekta Neuromag 306-‐channel whole-‐head MEG§ Anatomical landmarks, HPI coils and points upon the head digitized§ Sampling rate=600Hz; BPF=0.1~200 Hz

BEHAVIORAL TEST

rIdentify the last chord (Tonic/ Submediant/ Supertonic) of the sequences using a keypad in the same environment with MEG recording

r1 Block= 3 kinds x 12 keys = 36 questionsrTotal 3 blocks (App. 15 min)

rExamples for each chord progressions were presented before each block

SIGNAL PROCESSING

rSpatio-‐temporal filter applied to eliminate environmental noise (MaxFilterTM)

rAll blocks transformed to the head position of the first block

rEpoch: -‐200 ~ +500 msec to the target onset§ Epochs with EOG artifacts discarded (<10%)

rAveraged across blocks for each condition§ Mean averaged trials for each condition >150

rBand pass filter§ 1~20 Hz with zero phase for ECD analysis

EQUIVALENT CURRENT DIPOLE MODEL



≈



≈


MULTIPLE ECD ANALYSIS W/ BESA

rMultiple ECDs were fitted as followings:§ Generators of P2m activities were fitted using genetic algorithm

implemented in BESA® (Brain Electrical Source Analysis; NeuroScan)• For the latency of 180-‐190 ms in the temporal regions

§ Additional two bilateral ECDs for EANm responses were fitted• The time-‐window of 140-‐220 ms after onset• Seeding from the frontal regions bilaterally for the condition of

Supertonic as the frontal activities were shown consistently in the previous literatures (accumulated goodness of fitting was over 80%)

§ The activities of ECDs were estimated for each condition








All in-‐key chords







All in-‐key chords













Supertonic







Supertonic












MULTIPLE ECD ANALYSIS W/ BESATonic







Submediant







SubmediantSupertonic

STATISTICAL ANALYSES

rRepeated measure ANOVA§ Within-‐subject: Chord (3 levels); Between-‐subject: Group (2 levels)

rGeneral Linear Model (GLM)§ Additional quadratic model to test the effects of training

RESULTS

ECD LOCATIONS

Mean (SEM) in head coordinatex (mm) y (mm) z (mm)

lHG -‐42.43 (1.16) -‐0.52 (1.57) 46.61 (1.25)

rHG 46.06 (1.30) 6.35 (1.90) 47.69 (1.29)

lIFG -‐36.27 (2.15) 18.13 (4.97) 65.77 (3.52)

rIFG 41.99 (2.42) 19.89 (4.89) 63.56 (3.54)

Hotelling’s t2 : t2(3, 36)= 39.07, p< 0.0001); Right: t2 (3, 36)= 21.62, p< 0.001)

ECD ACTIVITIES (N=20)

Shading indicates the time-‐window of [140, 220] ms after onsetActivities didn’t differ across hemispheres (t(19)= -‐0.63, p> 0.5) thus pooled

EFFECTS OF CHORD & TRAINING

rEANm response was defined as the mean of absolute values in the time window [0.14, 0.22]

rRepeated measure ANOVA on the EANm responses rThe effect of chord functions

§ F(1.188, 21.389)= 34.636, p< 0.0001 (Greenhouse-‐Geisser’s corr.)

rThe effect of musical training with an interaction § F(1.188, 21.389)= 4.304, p= 0.0044 (Greenhouse-‐Geisser’s corr.)

CORRELATION WITH PROB. & TRAINING

The quadratic model considering the effects of musical training and interaction explained significantly better (F(2, 56)= 6.4390, p= 0.0030)

CORRELATION WITH BEHAVIORAL MEASURES

The correlation between the correct rates for Submediants and normalized EANm activities was significantly positive (r= 0.6244, p=0.0400)

DISCUSSION

EFFECT OF CONDITIONAL PROBABILITY

rEnding chord which has the lower conditional probability elicited the greater EANm response§ F(1.188, 21.389)= 34.636, p< 0.0001 (Greenhouse-‐Geisser’s corr.)

rIt is not due to acoustic deviance but syntactic§ All chord functions were played within the sequence before ending

rAlso the EANm responses correlates with the probability interacting with musical training§ Musicians: r= 0.4440; Non-‐musicians: r=0.3929

IFG AND PROBABILITY

sgKIM (2010) ThesisEAN/ MEG

Maess et al., (2001) NNERAN/ MEG

Opitz et al., (2002) NIMMN/ fMRI

Koelsch et al., (2005) NIERAN/ fMRI

RR R

VENTRAL PREFRONTAL CORTEX & PROB.

↓ Repeated “X”

Casey et al., (2001) HBM

↑ Other than “X”

INDIVIDUAL DIFFERENCES

rFor experts, the behavioral scores were not normally distributed § most of them reached the maxima (“Ceiling effect”)

rFor the normal group, Submediant scores highly correlate with the overall scores (r= 0.91)

rThe EANm could be considered as reflecting the individual competence of identifying subtle aberration§ Thus it could be suggested as the physiological index of the

probabilistic representation of individuals

LIMITATIONS

rmusicians showed enhanced responses to the short-‐term deviants & irregular chords (ERAN)

rThis might implicate the facilitated sensitivity to musical syntax or the familiarity to the corpus in the present study

Huron, 2006, Sweet AnticipationBaroque Corpora Popular music Corpora

LIMITATIONS

rEffects of the melody & chord function§ Different top voices might have confounded the effect of chord

functions§ But not likely,

rDifferent melody§ Doesn’t affect the

time-‐windowlater than 100 ms.

Koelsch & Jentschke (2010) J. Cog Neurosci.

CONCLUSION

rThe effects of the probability of chord progression on the brain responses were found§ The negative correlation between the conditional probability

and the corresponding neural response§ Also it was found to be more facilitated by the musical training

rThe current results suggest the physiological response as a reflection of the probabilistic representations on the musical syntax.

rMoreover, the results indicate that the probabilistic representation is related to the musical training as well as individual sensitivity.

CONCLUSION

rThe effects of the probability of chord progression on the brain responses were found§ The negative correlation between the conditional probability

and the corresponding neural response§ Also it was found to be more facilitated by the musical training

rThe current results suggest the physiological response as a reflection of the probabilistic representations on the musical syntax.

rMoreover, the results indicate that the probabilistic representation is related to the musical training as well as individual sensitivity.

The effect of conditional probability of chord progression in Western music corpus on brain response: an MEG study

Technology

c major

scale c

tonic i

i6 n6 v7

f major

department of brain

brain response

v7 n6koelsch