Copyright 1999 by the American Psychological Association, Inc. 0894-4105/99/S3.00
Neuropsychology 1999, Vol. 13, No. 3,404-414
Allocation of Attention in Dichotic Listening: Differential Effects on the Detection and Localization of Signals RoxanneInch
Merrill Hiscock University of Houston
University of Saskatchewan Marcel Kinsbourne New School University
In 2 dichotic listening experiments, 96 normal right-handed adults attended selectively to the left and right ear and divided their attention equally between both ears. Participants listened for specified targets and reported the ear of entry. The material consisted of pairs of consonant-vowel syllables in Experiment 1 and pairs of rhyming consonant-vowelconsonant words in Experiment 2. Both experiments yielded a right-ear advantage for detection and for localization. Attention instructions had no effect on detection. However, focusing attention on 1 ear increased the number of targets attributed to that ear while decreasing the number of targets attributed to the opposite ear. The dissociation between detection and localization indicates that volitional shifts of attention influence late (response selection) processes rather than early (stimulus identification) processes. Selective-listening effects can be accounted for by a 2-stage model in which a fixed input asymmetry is modulated by a biased selection of responses.
representation, as suggested by clinical data (Carter, Hohenegger, & Satz, 1980; Rasmussen & Milner, 1975), an 80% frequency of REA is significantly discrepant from the prevalence of the language asymmetry it is thought to represent. Satz (1977) has used a Bayesian analysis to show that inferences about anomalous language lateralization are highly misleading when based on a laterality measure that substantially underestimates the prevalence of left-lateralized language in the population. The inaccuracy of dichotic listening in classifying individuals as left dominant or right dominant for language might be attributed to several factors, including unreliability of eardifference scores (Blumstein, Goodglass, & Tartter, 1975; Teng, 1981), use of suboptimal statistical criteria to classify participants (Wexler, Halwes, & Heninger, 1981), and individual differences in peripheral auditory sensitivity (Borod, Obler, Albert, & Stiefel, 1983). One of the shortcomings of commonly used dichotic listening methods is the researcher's lack of control over the participant's processing and reporting strategies (Bryden, 1978). Although especially salient when multiple stimuli are presented to each ear (e.g., Kimura, 1961a, 1961b), strategy effects remain evident when stimuli are presented in single pairs (Bryden, 1982, 1988; Hiscock, Lin, & Kinsbourne, 1996; Spellacy & Blumstein, 1970). Of particular concern are questions about how participants allocate attention in dichotic listening and the effect of selective attention on left- and right-ear performance. These questions lead us back to the origin of dichotic listening as a method for studying selective attention (Broadbent, 1954, 1958). Auditory selective attention has been studied in two contexts. One line of studies has been focused on the degree to which signals from an unattended channel are processed. The results have led to various attempts to specify the stage
Dichotic listening is a method that presents simultaneous auditory messages, one message to each ear. When the stimulus material is linguistic (words, nonsense syllables, digit names, etc.), the message presented to the right ear is more likely to be reported than the message presented to the left ear. The right-ear advantage (REA) has been associated empirically with clinical evidence of left-sided language representation (Geffen & Caudrey, 1981; Geffen, Traub, & Stierman, 1978; Kimura, 1961a, 1961b; Strauss, Gaddes, & Wada, 1987; Zatorre, 1989). As observed by Bryden (1988), the REA has "proved to be a very robust effect no matter what [procedural] variations were introduced" (p. 3). The robustness of the REA across studies, especially studies of normal right-handers, raises the possibility that dichotic listening procedures might be useful in determining the side of language representation at the level of the individual. However, the ubiquity of the REA at the level of group is seldom mirrored in individual data. Despite an occasional report of REA in more than 90% of right-handed participants (e.g., Grimshaw, McManus, & Bryden, 1994), the typical frequency of REA among normal right-handers is about 80% (Bryden, 1988). Assuming that 95% to 99% of right-handers have left-sided speech Merrill Hiscock, Department of Psychology, University of Houston; Roxanne Inch, Department of Psychiatry, University of Saskatchewan, Saskatoon, Saskatchewan, Canada; Marcel Kinsbourne, Department of Psychology, New School University. This work was supported in part by a grant from the Medical Research Council of Canada. We thank Marilynn Mackay for her assistance. Correspondence concerning this article should be addressed to Merrill Hiscock, Department of Psychology, University of Houston, Houston, Texas 77204-5341. Electronic mail may be sent to
[email protected]. 404
ATTENTION IN DICHOTIC LISTENING
of processing at which attention has its effect (e.g., J. A. Deutsch & Deutsch, 1963; N. L. Wood & Cowan, 1995). The other line of studies investigates the effect of attending to one or the other ear on the REA (e.g., Asbjornsen & Hugdahl, 1995). Interest in selective auditory attention originated in a 1953 study by Cherry in which participants shadowed (repeated) prose heard at one ear while disregarding irrelevant stimuli at the other ear. Selective attending was so effective that participants failed to notice transitions at the unattended ear unless the transitions entailed a change in a fundamental acoustic property, such as pitch. Considerable subsequent work has focused on the degree or level to which taskirrelevant stimuli are processed and recalled (Broadbent, 1958; J. A. Deutsch & Deutsch, 1963; Moray, 1959; Treisman, 1960; N. L. Wood & Cowan, 1995). Recent evidence indicates that there is some automatic processing of information from the unattended channel, but unattended stimuli are not processed to a level that allows awareness. When input from the unattended channel is recalled and reported, its availability apparently reflects a shift of attention to that channel (N. L. Wood & Cowan, 1995). The second line of studies has shown that it typically is easier to attend to speech stimuli from the right ear than from the left ear (Treisman & Geffen, 1968). This asymmetry is thought to be a manifestation of the REA that is commonly observed in dichotic listening tasks that have a free-report format (e.g., Studdert-Kennedy & Shankweiler, 1970). Although the REA is thought to be determined by neurological factors (see Studdert-Kennedy, 1975), ear asymmetry may be influenced by volitional shifts of attention (Bryden, Munhall, & Allard, 1983; Treisman & Geffen, 1968). Under some circumstances, focusing attention on the left ear in a dichotic listening task will nullify the REA (Andersson & Hugdahl, 1987; Hiscock & Beckie, 1993) or even produce a left-ear advantage (LEA; Asbjornsen & Hugdahl, 1995; Bryden et al., 1983; Hugdahl & Andersson, 1986). Nonetheless, selective attention instructions in dichotic-listening tasks typically do not have as dramatic an effect as do similar instructions in shadowing tasks. Whereas the shadowing literature suggests very limited processing of stimuli in the unattended channel (N. L. Wood & Cowan, 1995), the dichotic-listening literature indicates that unattended stimuli, especially unattended stimuli in the right ear, often are processed to the point of being available for verbal report (Bryden et al., 1983). In children, signals from the unattended right ear may be processed to a degree comparable to, or even exceeding, that of stimuli in the attended left ear (Hiscock & Beckie, 1993; Hugdahl & Andersson, 1986; Obrzut, Boliek, & Obrzut, 1986). The differential strength of selective attention effects in the shadowing and dichotic-listening studies may stem from either of two salient differences between the respective methods. First, the shadowing task probably entails a more effective manipulation of attention. Repeating the material heard on one channel demands a high level of effort (Underwood & Moray, 1971), and the fact that shadowing performance is observable presumably enhances participants' compliance with the selective attention instructions.
405
In contrast, as suggested by Mondor and Bryden (1991), the selective attention instructions used in dichotic-listening studies may constitute a rather weak means of manipulating attention. Second, the attended material in the shadowing task, being of longer duration, may be more likely to capture the participants' attention. The material to be shadowed typically consists of connected prose that is presented over a period of at least a few minutes, whereas dichotic stimuli typically consist of single pairs of brief sounds or lists of no more than three or four pairs presented in rapid succession. The stimuli used in dichotic-listening experiments may not provide the participants with sufficient time in which to focus attention optimally (see Mondor & Bryden, 1991; Mondor & Zatorre, 1995). Alternatively, the brief duration of the stimuli may allow attention to be switched to the unattended ear before information in that channel has decayed from preattentive (echoic) storage (Darwin & Baddeley, 1974). Few studies have focused on the mechanism or mechanisms of attentional effects in dichotic listening. The available evidence is largely restricted to descriptions of how listening asymmetry is altered by focused attention, and these descriptions are not entirely consistent across studies (see Mondor, 1994, for a review). The specific consequences of attention shifts on performance at the attended and unattended ears also vary across studies. Bryden et al. (1983) and Asbjornsen and Hugdahl (1995) reported that selective attention instructions produce both an increase in the number of stimuli reported from the attended ear and a decrease in the number of stimuli reported from the unattended ear. A study of children by Hiscock and Beckie (1993) supported that conclusion, but only for consonantvowel (CV) stimuli. When children were asked to detect targets from lists of dichotic words and they attended to either ear, the number of hits from the unattended ear decreased without the number of hits from the attended ear increasing. Obrzut et al. (1986) found that selective attention instructions had no effect on children's perception of CV stimuli at either ear and had diverse effects on the perception of other dichotic stimuli (words, digit names, and melodies). Thus, attention shifts typically increase the number of stimuli reported from the attended ear or decrease the number of stimuli reported from the unattended ear. Sometimes both effects are obtained. Outcomes appear to vary according to the nature of the dichotic stimuli and, in some instances, according to the ear that is being attended (Obrzut et al., 1986). In the absence of analyses based on signaldetection theory, it is impossible to know whether the effects of selective attention represent changes in sensitivity or in decision criteria. The primary purpose of this study is to determine more precisely the effect of selective attention instructions on adults' perception of dichotic stimuli from two categories: single pairs of CV nonsense syllables and single pairs of fused rhyming words. By combining selective attention instructions with a detection task, we are able to assess the effects of selective attention on (a) the sensitivity of signal detection at each ear, (b) the sensitivity of signal localization at each ear, and (c) the criterion for responding to signals at
406
HISCOCK, INCH, AND KINSBOURNE
each ear. In addition, by manipulating the time at which the participant learns the identity of the target, we can vary the memory load imposed by the task. If the identity of the target is not known until after the dichotic stimuli have been presented (i.e., in the case of a postcued target), a response must be made on the basis of a stored representation of the input. On the other hand, advance knowledge of the target's identity (i.e., in the case of a precued target) may allow the participant to choose a response without fully processing or storing all of the input (Kahneman & Treisman, 1984). Accordingly, precuing serves to minimize the effects of memory and associated variables such as order of retrieval and differential decay.
Experiment 1 Pairs of CV nonsense syllables were selected as stimuli for Experiment 1. Since 1970, when Studdert-Kennedy and Shankweiler reported a strong REA for the initial stop consonants in dichotic pairs of consonant-vowel-consonant (CVC) syllables, single pairs of CV nonsense syllables have been used frequently as stimuli in dichotic-listening studies (Bryden, 1988). The use of single pairs of stimuli serves to reduce or to eliminate many of the input and output organization factors that confound the interpretation of dichotic-listening studies involving lists of two or more stimulus pairs per trial. Because the paired CV stimuli differ only with respect to the initial consonant's voice onset time, place of articulation, or both, the stimuli frequently "fuse" perceptually so that participants report hearing only one sound (Wexler & Halwes, 1985). Nonetheless, as noted by Bryden (1982), using single pairs of stimuli does not preclude attention being allocated in various ways nor does it prevent right- and left-ear stimuli from being reported in different orders.
Method Participants. Forty-eight undergraduate volunteers (24 women, 24 men) were recruited from introductory psychology classes to participate in a study of auditory perception. Ages ranged from 18.1 to 33.6 years (M = 22.3, SD = 3.4). All participants were righthanded by self-classification, by use of their right hands for writing, and by their scores on the Edinburgh Handedness Inventory (Oldfield, 1971). The mean laterality quotient from the handedness questionnaire was 77.1 (SD = 21.0) on a scale ranging from —100 (extreme left-handedness) to +100 (extreme right-handedness). All participants were native speakers of English. Criteria for exclusion from the study were (a) known hearing impairment, (b) history of seizures or other neurological disorder, and (c) history of speech therapy. Materials. The dichotic stimuli were tape-recorded CV nonsense syllables produced by combining each of the six English stop consonants with the vowel /a/: /ba/, /pa/, /da/, /ta/, /ga/, and /ka/. Dichotic pairs of these syllables were prepared and were recorded at the Kresge Hearing Research Laboratory of the South (New Orleans, LA). Computer processing of digitized natural speech tokens (male voice) resulted in alignment of consonant onsets within 2.5 ms and matching of vowel segment amplitudes within 2.5 dB. The duration of each syllable was 296 ms. The tape consisted of three random sets of stimuli, each of which contained
30 counterbalanced CV pairs with an intertrial interval of 6 s. Playing the three sets twice yielded a total of 180 test trials. In addition, a set of 30 binaural practice trials was available to familiarize participants with the material. Apparatus. Stimuli were presented via a Revox A77 twochannel tape recorder (Willi Studer AG, Regensdorf, Switzerland) and Koss K/6 stereophonic headphones (Koss Corporation, Milwaukee, WI). The average signal intensity for each channel was set to 80 dB (A), as measured using a 1000 Hz steady-state calibration tone, and the channels were balanced within 1 dB. The ambient noise level was 32 dB (A) in the sound-attenuated chamber used for testing. Procedure. Each participant was tested individually. The participant first completed the Edinburgh Handedness Inventory and a personal history questionnaire, which were used to confirm information that had been obtained via telephone during an initial screening. The experimenter then described the CV stimuli that the participant would be hearing. Before the first set of test trials, the participant donned the headphones and listened passively to 30 binaural presentations of the CV stimuli. Following these familiarization trials, three blocks of 60 test trials were administered. One block of trials (left) entailed attending selectively to the left ear, another block (right) entailed attending selectively to the right ear, and the remaining block (divided) entailed dividing attention equally between the left and right ears. The order in which the blocks were administered was counterbalanced completely across participants within each sex. A target CV syllable, printed on a 10 X 15-cm card, was presented either 1 s before the onset of the dichotic stimuli (precue condition) or 1 s after the offset of the dichotic stimuli (postcue condition). Half of the participants within each sex were assigned randomly to the precue condition and half were assigned to the postcue condition. Irrespective of cue condition, the participant was instructed to report his or her perception of the target by writing "L" (left), "R" (right), or "N" (neither) onto a form devised for that purpose. Within each block of 60 trials, the target arrived at the left ear on 20 trials, at the right ear on 20 trials, and at neither ear on 20 trials. A list of targets was assigned randomly to each participant from a set of six lists. Within each list, targets were distributed randomly across possible locations (left ear, right ear, neither ear). The initial position of the headphones (normal vs. reversed) was counterbalanced within each combination of sex, cue condition, and target list. The headphones were reversed after every 30 trials. Scoring. Three measures of sensitivity and one measure of response bias were computed using procedures similar to those described previously (Bryden, 1976; Bryden et al., 1983; Hiscock, Hampson, Wong, & Kinsbourne, 1985). Detection performance can be represented in terms of a single 2 X 2 decision matrix in which the target is present or absent and the response is either "yes" or "no." In computing the index of detection sensitivity, a response was counted as correct if the target was identified when actually present, regardless of whether the ear of arrival was specified correctly. Localization performance can be represented in terms of a 3 X 3 decision matrix in which the target is present at the left ear, present at the right ear, or absent from both ears, and the response is "left," "right," or "neither." A response is counted as a localization hit only if the target is detected and localized correctly. Following Bryden (1976), we assumed that decisions about left- and right-ear stimuli occur along two separate decision axes. Thus, estimates of sensitivity and bias are based on four cells of the matrix, that is, localization hits for each ear as well as false alarms attributed to each ear. The index that we refer to as detection-plus-localization sensitivity reflects the number of localization hits. The index
ATTENTION IN DICHOTIC LISTENING referred to as localization-only sensitivity is based on the ratio of localization hits to detection hits. Localization-only sensitivity indicates the likelihood that a target, once detected, will be localized correctly. As in previous detection studies with dichotic stimuli, nonparametric measures of sensitivity and bias were used (Grier, 1971; Hodos, 1970; McNicol, 1972; Pastore & Scheirer, 1974). The area under the receiver-operating characteristics (ROC) curve, p(A), was selected as the measure of sensitivity, and an arcsine transformation was applied to_ p(A) scores before analysis of variance (ANOVA). Higher p(A) values represent greater sensitivity. A measure of displacement from the negative diagonal of the ROC curve, p', was used as the index of bias. We used a computational formula that accommodates points falling on either side of the diagonal (Grier, 1971, Expression 11). All mean P' values were positive, which indicates that the points fall below the negative diagonal. Greater magnitudes represent greater displacement from the diagonal and thus more stringent decision criteria. All data analyses were repeated with sex of participant as an additional factor. Unless otherwise indicated, no significant sex differences were obtained.
Results Raw scores. Table 1 shows the mean proportion of detection hits, localization hits, localization errors, and false alarms (out of a possible 20) for each cue condition, ear, and attention condition. Detection sensitivity. Arcsine-transformed p(A) scores for detection were analyzed in a 2 X 2 X 3 ANOVA with cue condition (precue or postcue), ear, and attention condition (left, right, or both) as the respective independent variables. There were repeated measures on the second and third variables. The ANOVA yielded no significant findings except a main effect for ear, F(l, 46) = 21.73, p < .0001, which reflected an REA. There was neither a significant main effect for attention (F < 1), nor an Ear X Attention interaction (p > .10). Mean p(A) scores for each ear and attention condition are shown in the upper left panel of Figure 1. Detection-plus-localization sensitivity. The Cue X Ear X Attention ANOVA for arcsine-transformed p(A) scores yielded significant main effects for ear, F(l, 46) = 11.59,
407
p < .005, and attention, F(2, 92) = 3.99, p < .025. The ear effect indicated an REA, and the attention effect was attributable primarily to poorer overall performance with left attention than with right attention, F(l, 46) = 4.40, p < .05. A significant Ear X Attention interaction, F(2, 92) = 25.99, p < .0001, was attributable entirely to the Ear X Left Versus Right Attention component, F(l, 46) = 37.60, p < .0001. Leftward attention yielded an LEA, whereas rightward attention increased the REA above the magnitude observed in the divided attention condition. There were no significant effects for time of cuing. The results are shown in the upper right panel of Figure 1. A supplemental ANOVA with sex of participant as a design variable yielded a significant main effect for sex, F(l, 40) = 4.17, p < .05, which did not interact with any other variables. Women achieved higher scores than did men (M = 2.14 and 2.04, respectively). Localization-only sensitivity. Analysis of this index, which removes the effects of differential detection from localization scores, yielded results similar to those described in the previous paragraph. Again, there were significant main effects for ear, F(l, 46) = 6.32, p < .025, and attention, F(2, 92) = 4.28, p < .025, and the attention effect was attributable to the left versus right component, F(l, 46) = 5.34, p = .025. The Ear X Attention interaction was significant, F(2, 92) = 33.28, p < .001, as was the Ear X Left Versus Right Attention component, F(l, 46) = 54.16, p < .0001. The pattern of results is shown in the lower left panel of Figure 1. Response bias. The lower right panel of Figure 1 shows the mean (3' for each attention condition. An ANOVA yielded neither a significant main effect for ear (F < 1), nor a significant Ear X Attention interaction (F < 1). Analysis of false-alarm rates also yielded negative results. Frequency data. The number of participants with an REA, LEA, or no ear advantage (NEA) for the various dependent variables is shown in Table 2 as a function of attention condition. (For false alarms, the advantaged ear is the ear receiving the greater number of responses. For 3', the advantaged ear is the ear at which the criterion is more lax.) Cochran's Q test (Siegel, 1956) indicated that for the three
Table 1 Mean Proportion of Detection Hits, Localization Hits, Localization Errors, and False Alarms for Each Ear and Attention Condition in Experiment 1 Divided
Left Cue condition
Detection hits Precue Postcue Localization hits Precue Postcue Localization errors Precue Postcue False alarms Precue Postcue
Left ear
AJV^IIL
ivl£XJI L VIU
t^CU
Right Left ear
j.x»fe.
i^U£l I I V^MJ.
M
SD
M
SD
M
SD
M
SD
M
SD
M
SD
0.77 0.78
0.16 0.15
0.85 0.83
0.12 0.13
0.78 0.82
0.17 0.12
0.83 0.85
0.14 0.13
0.72 0.79
0.15 0.13
0.85 0.87
0.13 0.13
0.56 0.51
0.18 0.15
0.48 0.38
0.19 0.16
0.47 0.51
0.19 0.18
0.59 0.58
0.17 0.16
0.36 0.45
0.18 0.19
0.66 0.67
0.15 0.15
0.21 0.27
0.10 0.13
0.37 0.45
0.16 0.18
0.31 0.32
0.17 0.18
0.24 0.27
0.12 0.13
0.36 0.35
0.15 0.10
0.19 0.21
0.11 0.12
0.20 0.21
0.12 0.11
0.17 0.20
0.10 0.14
0.16 0.25
0.11 0.12
0.18 0.22
0.11 0.14
0.16 0.26
0.12 0.11
0.20 0.16
0.11 0.12
408
HISCOCK, INCH, AND KINSBOURNE
l3ft v_/
.a < V
9 d ^•H
Detection
2.3
2 .4 A
I
T^^
2.2
I
i3 V-X
2.2
2 .S
2.0
