How the Brain Weighs Melody Against Meaning When Songs Turn Emotionally Ambiguous

When a singer delivers joyful lyrics over a mournful melody, the listener’s brain must somehow reconcile two emotional messages arriving at once. The review, by Shuo Yang and Zhihan Wang of Herzen State Pedagogical University in Saint Petersburg, synthesizes behavioral, electroencephalographic, magnetoencephalographic, electrocorticographic, and neuroimaging findings on how the brain handles emotional agreement and conflict between what a song sounds like and what its words mean.

The authors’ central claim is deliberately modest. Lyrics and melodies, they conclude, engage partially dissociable but interacting processes during song perception, yet direct neural evidence specifically manipulating music–lyric emotional congruence remains limited. Much of what neuroscience knows about song processing comes from studies of expectancy violations, sequential priming, and memory encoding rather than from experiments that directly pit a happy tune against sad words. The review therefore proposes a provisional framework rather than an established neural architecture, and it is explicit about which conclusions rest on direct findings and which remain inferential.

The framework rests on the idea that emotional meaning in song arises from several partially separable information sources. The first is an acoustic-affective stream: musical properties such as mode, tempo, pitch contour, intensity, consonance, and timbre provide relatively direct cues to valence and arousal. Crucially, in songs this stream is not confined to the accompaniment. The singing voice itself carries emotional information through pitch variation, intensity, timing, and spectral characteristics. The second source is semantic: lexical valence, sentence-level context, and narrative meaning in the lyrics can generate affective expectations independently of how the words are sung. The same positively valenced sentence can, in principle, be delivered with a sad or anxious vocal expression, which is why the authors insist that lyrical semantics should not be equated with vocal emotion.

Within the proposed dual-stream account, vocal prosody is treated as a distinct affective cue source embedded primarily within the acoustic stream rather than as a third independent pathway. Musical and vocal-acoustic information is processed rapidly in auditory cortical regions including the superior temporal gyrus and superior temporal sulcus, and affectively relevant differentiation can emerge during early auditory processing rather than only after a separate perceptual stage. Supporting this, intracranial electrocorticographic recordings have revealed dissonance-sensitive high-gamma responses in the superior temporal gyrus within roughly 75 to 200 milliseconds, and EEG work has identified early electrophysiological differentiation between affectively distinct chords. In parallel, linguistic information engages auditory-phonological and lexical-semantic processes involving temporal and inferior frontal regions. The authors stress that these pathways should be viewed as partially differentiated but dynamically interacting, not as independent streams that converge only after acoustic analysis is complete.

Some of the most striking temporal evidence comes from magnetoencephalography. In a study of memorized songs, Yasui and colleagues reported a left-dominant M140 response to unexpected lyric substitutions and a right-dominant M130 response to unexpected melodic notes. The researchers distinguished these components from the magnetic mismatch negativity elicited in a separate frequency-deviance condition, suggesting the responses reflected expectations about memorized lyrics and melodies rather than simple acoustic mismatch. Although these findings concern expectancy violations rather than emotional congruence, they provide study-specific evidence that lyric-related and melody-related predictions can influence neural responses as early as around 130 to 140 milliseconds.

At intermediate latencies, N400 effects supply evidence of context-sensitive meaning-related processing across musical and verbal materials, though their interpretation is not limited to semantic integration. Several studies have reported relatively differentiated electrophysiological responses to semantic and musical violations in songs, and melodic differences have been shown to modulate N400 responses even when the linguistic target remained unchanged. Related music–language priming studies have found larger N400 responses to words that were conceptually unrelated rather than related to preceding musical excerpts. But the review urges caution: affective priming can arise from multiple mechanisms, including facilitated target processing, associative activation, and response-related compatibility or competition. Because participants in the influential priming experiments explicitly evaluated target words, response-level contributions cannot be ruled out, and the sequential presentation of musical primes and verbal targets differs fundamentally from the simultaneous unfolding of lyrics and melody in natural song.

Later positive components add another layer of ambiguity. In one study, affective incongruence following instrumental musical primes increased N400 amplitude at 250 to 450 milliseconds, whereas incongruence following music containing lyrics primarily elicited a larger late positive complex at 500 to 700 milliseconds. Importantly, that study used emotional faces as targets rather than directly manipulating emotional conflict between lyrics and melodies. Late positive responses to melodic changes in sung material have also been reported, but the authors note these effects do not uniquely establish emotional reappraisal. Their functional interpretations remain task-dependent, and the review warns against reading these components as a universal sequence of discrete processing stages.

On the behavioral side, the evidence complicates any simple story in which congruence always helps. Melodies generally exert a stronger influence than lyrics on perceived song emotion, while the contribution of lyrics varies with emotional category: adding lyrics tends to strengthen negative emotions such as sadness and anger but attenuate positive emotions such as happiness and calmness. Developmental work involving 172 participants aged 7 to 20 years showed that recognition of emotion in congruent popular songs improved with age, whereas judgments of incongruent songs shifted between lyrical and melodic information depending on both age and the specific emotion conveyed. Most recently, music–lyrics congruence did not reliably predict valence ratings, whereas greater affective incongruence was associated with increased arousal, a statistically significant effect with a marginal R-squared of 0.188. Emotional conflict, in other words, may sometimes heighten arousal rather than simply impose a processing cost.

Functional neuroimaging fills in part of the spatial picture while leaving the emotional question open. An fMRI adaptation study reported partially dissociable yet interacting responses to lyrics and tunes across bilateral superior temporal regions. A subsequent-memory study found that successfully binding melodies sung with lyrics was associated with activity in the left inferior frontal gyrus, bilateral middle temporal gyrus, and left motor cortex, whereas presenting melodies on the neutral syllable “la” with lyrics shown visually recruited the right hippocampus, left caudate, cerebellum, and right inferior frontal gyrus. Meanwhile, an activation likelihood estimation meta-analysis of 40 neuroimaging studies identified the amygdala, insula, striatum, hippocampus, anterior cingulate cortex, and superior temporal gyrus among regions associated with music-evoked emotional processing, though that meta-analysis excluded music containing lyrics. The authors caution that activity in these multifunctional regions does not establish any specific psychological computation, and inferring emotional salience, valuation, reward, or conflict monitoring from regional activation alone would risk reverse inference.

The review closes with a research agenda designed to move the field beyond inference. The authors call for studies that orthogonally manipulate musical emotion, lyrical-semantic emotion, and vocal prosody in a two-by-two-by-two factorial design, which would separate the three affective information sources and directly test when and where semantic–acoustic congruence becomes neurally consequential. They also propose comparing songs in familiar and unfamiliar languages while holding musical structure and vocal delivery constant, since cross-cultural work suggests listeners can infer the behavioral functions of songs from unfamiliar societies on the basis of musical form alone. Naturalistic listening paradigms combining full-length professionally performed songs with continuous valence and arousal ratings, intersubject correlation, and multimodal EEG, MEG, and fMRI recording could capture the temporal development and expressive timing that simplified laboratory stimuli strip away. Until such studies are done, the authors argue, song emotion should be understood not as the additive result of music emotion plus lyric emotion, but as a dynamic weighted integration of competing cues whose relative influence shifts with expectations, attention, expertise, and language accessibility, with incongruence sometimes acting not as a cost but as a trigger for reweighting and reappraisal.

Subject of Research: Neural and behavioral mechanisms of emotional congruence between music and lyrics in song processing

Article Title: Neural integration of musical and lyrical emotion: semantic–acoustic congruence in song processing

Article References: Yang, S., & Wang, Z. (2026). Neural integration of musical and lyrical emotion: semantic–acoustic congruence in song processing. Frontiers in Psychology, 17, Article 1973585. https://doi.org/10.3389/fpsyg.2026.1973585

Image Credits: AI Generated

DOI: 10.3389/fpsyg.2026.1973585

Keywords: affective neuroscience, music and lyrics, EEG, ERP, MEG, fMRI, emotional congruence, song processing, vocal prosody, N400, semantic integration, auditory cortex