subjects came within two semitones of the correct pitch. The subjects sang a
second song, and the findings were essentially the same (lower portion of the
figure). Subjects were also consistent across trials, that is, if they were correct
on the first song, they were likely to be correct on the second song. From these
data, it appears that these nonmusical subjects have something much like ab-
solute pitch. Instead of asking them to ‘‘sing a Caor a G,’’ we can ask them to
‘‘sing ‘Hotel California’ or ‘Papa Don’t Preach,’’’ and they produce the correct
tone. Whether or not they’ve learned the specialized vocabulary of the musi-
cian seems less important than the fact that they have learned to associate a
consistent label with a specific tone. This finding has been replicated several
times as of this writing (Ashley 1997; Levitin 1996; Wong 1996).
Anumberofpeoplehavewonderediftheseresultsmightbetheproductof
something other than long-term memory for pitch. If people sing along with
their favorite songs, the argument goes, they may have merely developed a
‘‘muscle sense’’ or ‘‘kinesthetic sense’’ from singing the song, and their knowl-
edge of the proper vocal cord tension is driving the results. However, ‘‘muscle
memory’’ is a form of long-term memory. All this argument does is specify the
subsidiary mechanism in long-term memory that is at work. In addition, it
turns out that muscle memory is not very good. W. Dixon Ward and Ed Burns
(1978) asked vocalists to sing pitches from memory while being denied audi-
tory feedback (loud white noise in headphones was used to mask the sound of
their own voice). The singers were forced to rely solely on muscle memory
to produce the requested tones. Their results showed errors as great as a major
third, indicating that muscle memory alone cannot account for the precision of
performance of the subjects in the sing-your-favorite-rock-song study.
These data support the idea that long-term memory encodes the absolute
pitch of songs, even with a group of subjects in whom AP was not thought to
exist. This finding also extends Narmour’s theory about the two components
required for musical perception, showing that both absolute and relative infor-
mation are retained in long-term memory. A form oflatentorresidueabsolute
pitch is also implied by Fred Lerdahl and Ray Jackendoff’sstrong reduction hy-
pothesis(1983).
Can a song’s tempo be accurately encoded as well? The data collected for the
pitch study were reanalyzed to test memory for tempo (Levitin and Cook 1996).
The subjects weren’t explicitly instructed to reproduce tempo during the ex-
perimental session, so to the extent that they did, they did so on their own.
Tempo would not necessarily have to be explicitly represented in memory, be-
cause a melody’s identity does not depend on its being heard at exactly the
same tempo every time. Because pitch and tempo are separable dimensions
(Kubovy 1981), it is possible that one would be preserved in memory and the
other would not.
Some interesting properties of song memory are related to the idea of sepa-
rable dimensions. When we imagine a song in our heads, most of us can easily
imagine it in different keys without changing the speed of the song. This is not
how a tape recorder works: if you speed up the tape to raise the key, you au-
tomatically speed up the tempo as well. Similarly, we can mentally scan a song
at various rates without altering the pitch. If you are asked to determine as
quickly as possible whether the word ‘‘at’’ appears in ‘‘The Star Spangled Ban-
306 Daniel J. Levitin