Content uploaded by Grover C Gilmore
Author content
All content in this area was uploaded by Grover C Gilmore on Oct 13, 2014
Content may be subject to copyright.
Perception &Psychophysics
1979, Vol. 26 (6), 501-502
Notes
and Comment
Multidimensional letter similarity:
Areply to Mewhort and
Dow
GROVER C. GILMORE
Case Western Reserve University
Cleveland, Ohio 44106
and
HARRY
HERSH
Digital Equipment Corporation
Maynard, Massachusetts 01754
Gilmore, Hersh, Caramazza, and Griffin
(1979)
reported aconfusion matrix for uppercase letters pre-
sented on the screen of a cathode-ray tube. Mewhort
and Dow (1979) recently have questioned the validity
of this matrix and the indices of letter similarity de-
rived from it. They apparently raised their objection
because of the low correlation between the diagonal
entries of the matrices reported by Gilmore et al.
(1979)
and Townsend (1971), who generated a con-
fusion matrix using a conventional tachistoscope.
There are several reasons why this low correlation
could occur. The most obvious explanation, which
the authors mention and dismiss, is the difference in
type fonts used in the studies. Another possible
source of the difference is that the values reported
in one of the studies may be unreliable or confounded
by another variable. Since the confusion matrix
reported by Gilmore et al. (1979) was based on eight
times as many stimulus trials as that reported by
Townsend (1971), one might question the reliability
of Townsend's matrix. However, Mewhort and Dow
(1979)
choose to argue that the method used by
Gilmore et al. (1979) in presenting the stimuli is the
source of the difference and, furthermore, that this
confounding factor leaves the value of the confusion
matrix questionable.
The argument presented by Mewhort and Dow
(1979)
to support this conclusion seems to be a con-
vincing one. The weakness of the presentation, how-
ever, is that it is based on inferences about the
stimulus presentation method and not facts. The
authors infer that a low-intensity level was employed,
since Gilmore et al.
(1979)
were able to achieve a
51010
level of accuracy with a mean stimulus duration
of 33 msec. This first inference is, indeed, correct.
We regret that the stimulus luminance level was not
reported, but we were not able to obtain an instru-
ment which could reliably measure the luminance.
Reporting the luminance level, however, would not
have forestalled the major objections raised by
Mewhort and Dow (1979). First of all, they question
the stability of the equipment for displaying dots at
a low-intensity level. Rather than rely on speculation,
we checked this point with the manufacturer
of
the
equipment, Digitial Equipment Corporation,
and
were assured that there is no reason to suspect a
stability problem (Rupp, Note 1). The second objec-
tion raised by Mewhort and Dow
(1979)
concerns the
dot-brightening algorithm used in generating the
letters. They contend that it may have been very slow
and that letters containing few dots may have been
brighter than those composed of more dots. The
algorithm employed has been described by Schwartz
(1978) and is not considered slow, since a dot is
refreshed approximately every 25
usee.
Furthermore,
while the displayed dots were continuously refreshed,
possible brightness differences among letters com-
posed of an unequal number of dots were controlled
for by presenting an appropriate number
of
extra
dots at a non visible portion
of
the screen. That is
to say, the same number
of
dots were generated
and brightened regardless of the letter displayed.
This should have been an effective brightness control.
Indeed, when our subjects were questioned, none of
them reported noting any brightness differences
among the letters. The subjects did note that some
letters, such as I and T, were easier to identify
than others, but they attributed this to the fact that
there were few letters shaped as these letters.
Since the brightness
of
the letters was controlled,
how can one then account for the high correlation
(r = - .873) between the number of dots in a letter
and its identification accuracy? One reasonable inter-
pretation is that the letters with few dots are simple
figures composed
of
relatively few features. In a
limited-capacity system, simple letters would have an
advantage over more complex ones at an encoding
level. Another interpretation, suggested by our sub-
jects, is that the simpler letters are relatively unique
in their shape, thereby placing lessprocessingdemands
on the observer. Based on an inspection of the letters
used by Gilmore et al. (1979, Figure 1), one can see
that the set of letters potentially confusable with the
letter L is much smaller than the set potentially con-
fusable with the letter M.
This conclusion is supported by a recent study by
Podgorny and Garner (1979). In an investigation
of
inter- and intraletter similarity, they have reported
that in a "same-different" task there are significant
differences in
"same"
reaction times among letters.
For instance, it takes much longer for a subject to
respond
"same"
to a pair
of
Ms than to a pair
Copyright 1979 Psychonomic Society, Inc. 501 0031-5117/79/120501-02$00.45/0
502
GILMORE
AND
HERSH
of
Ls. Podgorny and Garner (1979) also have inter-
preted these differences as due to either more ef-
ficient encoding
of
certain letters, such as L, or a low
level
of
confusability
of
these letters with the other
letters
of
the alphabet.
The major point raised by Mewhort and Dow
(1979) was that the confusion matrix reported by
Gilmore et al. (1979) lacks validity because
of
a con-
founding
of
letter structure with brightness. Yet,
as we have stated, the procedures and equipment
used in generating the
letters
controlled for potentially
confounding brightness differences among the letters.
We must conclude therefore that the differences in
level
of
identification accuracy among the letters
reliably reflects true processing differences and that
the confusion matrix can be used by investigators
as a reliable and valid instrument.
REFERENCE
NOTE
I.
Rupp, C. Personal communication, August 1979.
REFERENCES
GILMORE,
G.
C.,
HERSH,
H.,
CARAMAZZA,
A., &
GRIFFIN,
J.
Multidimensional letter similarity derived from recognition
errors. Perception &Psychophysics, 1979,25,425-431.
MEWHORT,
D. J. K., &Dow,
M.
L. Multidimensional letter simi-
larity: A
confound
with brightness? Perception &Psycho-
physics, 1979, 26, 325-326.
PODGORNY,
P.,
&
GARNER,
W. R. Reaction time as a measure
of inter- and intraobject similarity: Letters of the alphabet.
Perception &Psychophysics, 1979,26,37-52.
SCHWARTZ,
S.
P.
Tachistoscope simulation package. Behavior
Research Methods &Instrumentation, 1978, 10,773-778.
TOWNSEND,
J. T. Theoretical analysis of an alphabetic confusion
matrix. Perception &Psychophysics, 1971, 2,45-53.
(Received and accepted for publication October 15, 1979.)
