Questions: Information Theory and Entropy in Musical Structure

Question 1

A composer designs a piece in which every possible chord transition is equally probable — a maximally diverse harmonic vocabulary. How does information theory predict listeners will experience this piece?

• A. As highly engaging, because maximum harmonic variety creates maximum interest
• B. As difficult to parse, because high entropy means the next chord is almost unpredictable — closer to noise than music
• C. As pleasantly surprising, because surprise is the main driver of musical engagement
• D. As technically complex but emotionally neutral, because entropy and emotion are unrelated

Answer: (B)

Maximum harmonic entropy means every chord is equally likely regardless of what came before. There is no pattern to learn, no expectation to satisfy or violate, no trajectory. This is perceived as random noise rather than music — the same way white noise is acoustically rich but perceptually meaningless. Optimal engagement lies in the middle range of entropy: enough predictability for listeners to build expectations, enough uncertainty to satisfy and occasionally violate them. Maximum entropy is the worst case for meaningful engagement, not the best.

Question 2

Why does conditional entropy H(Xₙ₊₁ | Xₙ) better capture perceived musical predictability than marginal entropy H(Xₙ₊₁)?

• A. Conditional entropy is always smaller than marginal entropy, so it is more precise
• B. Conditional entropy measures how much uncertainty remains about the next event given the current event, which is what the listener actually experiences moment to moment
• C. Marginal entropy requires more data to compute and is less reliable for short pieces
• D. Conditional entropy captures the tonal hierarchy more accurately than marginal entropy

Answer: (B)

As music unfolds, the listener's uncertainty about the next note is not the abstract probability over all notes in the piece — it is the uncertainty given what has just happened. A scale melody in C major might use all seven scale degrees (moderate marginal entropy), but if you're on the leading tone (B), the conditional entropy is very low: the next note is almost certainly the tonic. Conditional entropy H(Xₙ₊₁|Xₙ) captures this context-dependent predictability that is actually heard, while marginal entropy ignores sequential structure entirely.

Question 3

A serial (twelve-tone) melody intentionally avoids repeating pitch classes and therefore has higher conditional pitch entropy than a tonal melody.

• T. True
• F. False

Answer: True

True. In tonal music, scale degree tendencies and voice-leading conventions strongly constrain which notes follow which — the conditional entropy is low because knowing the current pitch substantially narrows the probable next pitches. Serial technique is specifically designed to break these expectation patterns: once a pitch class is stated, it cannot repeat until all twelve are used, which distributes probability more evenly and raises conditional entropy. This higher uncertainty is one reason serial music sounds 'less predictable' to trained listeners — it structurally removes the tonal constraints that produce low conditional entropy.

Question 4

Music with the highest possible entropy — where most note is largely unpredictable given any prior context — provides the richest aesthetic experience.

• T. True
• F. False

Answer: False

False. Maximum entropy corresponds to maximum unpredictability — the musical equivalent of white noise. Since listeners engage with music partly through expectation and anticipation, music with zero predictability provides nothing to anticipate, no patterns to learn, and no satisfying or surprising resolutions. The psychoacoustic evidence suggests optimal engagement occurs at intermediate entropy levels: enough structure to form expectations, enough uncertainty to sustain interest. Composers like Haydn are admired precisely for their mastery of controlled entropy — not for maximizing it.

Question 5

Why does conditional entropy provide a better measure of perceived musical predictability than marginal entropy, and how does this connect to the entropy profile of a piece over time?

Model answer: Marginal entropy measures how evenly distributed all events are across the entire piece — a global statistic that ignores sequential context. But a listener experiences music moment-to-moment: their uncertainty is about the next event given what just happened. Conditional entropy H(Xₙ₊₁|Xₙ) captures this local, sequential predictability. A piece's entropy profile — how conditional entropy varies across time — maps directly onto its formal structure: high-entropy passages correspond to tension and ambiguity, low-entropy passages to resolution and stability.

This distinction matters analytically because two pieces can have identical marginal entropy (same distribution of pitches overall) but very different conditional entropy profiles. A tonal piece and a random permutation of the same notes have nearly identical marginal distributions but radically different conditional entropy: the tonal piece has low conditional entropy shaped by harmony, while the permutation has nearly maximal conditional entropy. The entropy profile is thus a tool for distinguishing genuine compositional structure from accidental statistical similarity.