Music & Language Studies

SERAFINE, Mary Louise, Janet Davidson, Robert G. Crowder, and Bruno Repp. “On the Nature of Melody-Text Integration in Memory for Songs.” Journal of Memory and Language 25 (1986): 123-35.

Reports on experiments intended to follow up on possible theories explaining away the integration effect described in Serafine, Crowder & Repp 1984, above. The first new experiment tested an hypothesis which posited that apparent integration was the result of subtle semantic assignments of meaning to the melodies involved, in ways familiarly termed word-painting in musical fields. The target songs (original presentations) however, had their words replaced by nonsense syllables, which nonetheless sounded like words (i.e. “Cape Cod girls they have no combs” became “Tade top berf shey jaze mo tong”). Thus, any effect of semantics would be destroyed. The experiment however failed to support this hypothesis.

Melodies were recognized better when they were paired with their original text than when paired with another, even if equally familiar text. Since this effect held when nonsense texts were used, the semantic hypothesis must be ruled out as an explanation for the integration effect.(129)

A second hypothesis was that the integration effect was not due to a strengthening of memory on the basis of pairing the melody with words, but rather the converse of a decrement effect due to the distracting influence of wrong words.

Perhaps the melody by itself could be recognized well without the original words, but adding new or mismatched words somehow disguises the retained melodic information. (129)

This hypothesis was tested by counterposing a hummed version of the song with properly matched and mismatched stimuli. The original meaningful words were retained in this experiment. The argument is that a hummed version should be recognized more easily than a mismatched version, and equally well as the properly matched target (Experiment 2). The findings were somewhat inconclusive. Although they failed to support the decrement hypothesis, they did not strongly disconfirm it. However,

melodies were better recognized in the presence of their original words than on their own, without words. (131)

Likewise, the words on their own should be more easily recognized without a novel melody, and equally so with the original melody (Experiment 3). Experiment 3 returned to the nonsense syllables, counterposing the sung originals with spoken versions. The findings for Experiment 3 were strong, since recognition of the text without a melody (whether original or mismatched) was near chance. That is, memory for nonsense words was better in the context of the melody with which they were first presented, than it was without a melody. In discussion of these findings, the experimenters note, however:

Integration of melody and text in memory for songs is an experimental result, not an explanation. (133)

They suggest certain avenues for further investigation. Specifically, they identify that the subtle influences that melody and text have on one another result in minute changes in the acoustic signal, which may likely be perceptually salient to a listener.

we suggest that integration in memory may result from other, more subtle effects that melody and text have on each other. These may be thought of, broadly, as prosodic effects in that they concern the nonsemantic sound pattern of either melody or text,. For example, a text’s consonant pattern, vocal timbres, and accents may affect the attack and decay patterns, stresses, or other aspects of tones in a melody….What this means is that a melody is physically different depending on the words to which it is sung. (133)

Although the authors argue that the hummed stimuli rule out the possibility that a melody could be equally recognized without its text, it would be interesting to test this hypothesis not with a hummed variation, but rather by pairing an instrument with the voice in the first presentation (say, a flute or violin, doubling the voice part), and playing only the instrumental version of the melody in the test condition. Would subjects be equally able to recognize the melody presented without a voice, and thus without the original words?

It has been shown that various stimulus conditions alter the mental mechanism by which sounds are processed. Perhaps the human voice (even without linguistically meaningful sounds) provides a certain degree of distraction which alters a listener’s strategy in analyzing the sound signal. If that were the case, the decrement hypothesis might prove correct. However, rather than it being the deleterious influence of improper words, it is the presence of the human voice that alters the strategies employed to make sense of the input signal.