You are on page 1of 11

Developmental Psychology 1996. Vol. 32, No. 2.

269-279

Copyright 1996 by the American Psychological Association, Inc. 0012-1649/96/S3.00

Number-Based Expectations and Sequential Enumeration by 5-Month-Old Infants


Richard L. Canfield and Elliott G. Smith
Cornell University
Two studies used the visual expectation paradigm to determine whether 5~month~old infants spontaneously use the number of pictures appearing in one location (left) to predict when a stimulus wil! appear in a second location (right). Infants' eye movements were recorded by using infrared corneal reflection photography while they viewed either a numerically predictable or a numerically unpredictable repeating sequence of pictures. Neither stimulus timing parameters nor stimulus identity predicted future stimulus location. Analyses of anticipatory saccades provided strong evidence that infants rapidly recognize and use the number of sequentially presented pictures (2 in Experiment 1, 3 in Experiment 2) to predict the location of the next picture in the sequence. Results are discussed in relation to process models of counting and subitizing and emphasize the need for these models to accommodate early sequential enumerative capacities.

Piaget (1952) was among the earliest developrnentalists to examine the foundations of mathematical knowledge. Using children's performance on the number conservation task, he viewed number as a complex logical concept and believed that it was not acquired until middle childhood. However, Piaget's views on number development have been the subject of increasing critical scrutiny. For example, Gelman and Gallistel (1978, 1985) argued that conventional definitions of numerical competence are biased against the preverbal or newly verbal child, and they cited evidence that although young children have difficulty representing numerosity, they possess competence when reasoning with magnitudes they are able to represent. As a result, researchers began to investigate the foundations of numerical competence in preverbal human infants and have found consistent evidence for perceptual sensitivity to the dimension of number. In the present article, we investigate the numerical capacities of young infants by using a recently developed perceptual-cognitive paradigm, the visual expectation paradigm (Canfield & Haith, 1991; Haith, Hazan, & Goodman, 1988). After reviewing the principlefindingsof infant numerical com-

petence, we present data from two studies that demonstrate the capacity of 5-month-oid babies to rapidly acquire visual expectations when number is the predictor of future stimulus location.

Numerical Abilities Revealed by Fixation Duration Measures


Thefirststudies of infant numerical competence used procedures that rely on measures offixationduration. An important consequence of these procedures is that the items to be enumerated are presented simultaneously in a single display. For example, habituating an infant to "two'* involves the repeated presentation of displays, each containing two items. Starkey and Cooper (1980) were the first to report numerical abilities in young infants. By measuring duration of first fixation during the dishabituation trials of an infant-controlled habituation procedure, they showed that 4- to 7-month-oId infants can discriminate two- from three-item arrays containing equal-sized dots but could not discriminate four from six items. Although Starkey and Cooper (1980) varied array length and density in their experiment, they did not control for several alternative explanations for the source ofthe discrimination, such as background brightness, contour, or background area. Subsequent studies using multiple habituation sets provided controls for these concerns ( Strauss & Curtis, 1981). A converging body of evidence indicates that with small numerosities, infants less than 1 year of age, possibly at birth, have a basic perceptual ability to discriminate sets of objects using number as the only cue (Antell & Keating, 1983; Treiber & Wiicox, 1984; van Loosbroek & Smitsman, 1990). It has been less common to study infant numerical competence using sequentially presented items. In one such study, Starkey, Spelfce, and Gelman (1983, 1990)foundthat infants make intermodal numerical matchesa reliable tendency to look more at the static visual display that corresponds in number to a sequence of drumbeats heard at the beginning of a trial.

Richard L. Canfield and Elliott G. Smith, Department of Human Development and Family Studies, Cornell University. This research was supported by National Institute of Mental Health Grant I-R03-MH45298-01 and by the special projects fund of the Cornell University College of Human Ecology. Portions of the data were presented in April 1993 at the meetings of the Society for Research in Child Development, New Orleans, Louisiana, and in June 1994 at the International Conference on Infant Studies, Paris, France. We acknowledge the contributions of Mike Brezsnyak, Linda Kletzkin, Lauren Marder, Natalie Sikka, Kyle Snow, Rebecca Wilson, and Emily Vacherforexpert data collection and coding, Barbara Koslowski for suggestions on experimental design, and Steve Robertson for comments on an earlier draft of this article. We owe a special debt of gratitude to the many infants and parents who visited our laboratory. Correspondence concerning this article should be addressed to Richard L. Canfield, Department of Human Development and Family Studies, MVR Hall, Cornell University, Ithaca, New York 14853. 269

270

CANHELD AND SMITH

Success on this task requires infants to extract number from the si multaneously presented visual display and associate it with the number of sequentially presented drumbeats. Recent work by Wynn (1992) can also be interpreted as evidence for infants' ability to enumerate sequentially presented items. Infants watched as one or two identical objects were displayed and then occluded by an opaque barrier. While the barrier was up, the infant either watched one object being pulled from behind the barrier (if there had originally been two objects) or watched as one object was added to what was behind the barrier (if there was originally only one object). After either of these manipulations, the screen was lowered and infants saw either a possible or an impossible outcome. For the possible outcome, the number of objects remaining behind the screen agreed with the result expected after the manipulation (i.e., if an object had been added, then there were two objects). For the impossible outcome, the number of objects remaining did not agree with what should be expected (i.e., an object was added, but there is only one object). Wynn found that 5-month-old infants looked longer at the impossible outcomes than at the possible ones, which she interpreted as evidence for the recognition of and surprise at the impossible outcome. Wynn concluded from her results that infants perform simple arithmetic with small sets of items, suggesting that they enumerate items over time. For several reasons, the results of Starkey et al.'s (1983, 1990) and Wynn's (1992) research do not yet provide conclusive evidence for the existence of sequential enumeration in young babies. First, thefindingsof both research teams need to be replicated in independent laboratories. No published replication of Wynn (1992) exists at the present time, and there exists one published study that could be considered a nonreplication of Starkey et al.'s (1983) intermodal correspondence finding (Moore, Benenson, Reznick, Peterson, & Kagan, 1987; also see the response by Starkey et at, 1990). Second, the studies by Starkey et al. (1983, 1990) and Wynn (1992) rely on lookinglime measures. Converging results using an independent measure would strengthen these findings (see Davis & Perusse, 1988). Third, the studies of both teams require numerical discriminations that compare the result of the sequential enumerative process to the number of items in a set of simultaneously presented stimuli. It is possible that the numerical representation was derived solely from the simultaneously presented item displays and only then mapped onto the sequentially presented items. Therefore, it is not known whether infants enumerate items solely on the basis of sequential information. Visual Expectations and Sequential Enumeration In the visual expectation paradigm, infants are presented stimuli in a sequential manner, and only one stimulus is displayed at a time. Using this approach, researchers have begun to address the issue of sequential enumeration (Canfield, 1988; Canfield&Haith, 1991;Haith, Wentworth, & Canfield, 1993). By using asymmetrical sequences in which two or three homeside (e.g., left) pictures were presented before a target-side (e.g., right) picture, Canfield and Haith (1991) found that at the age of 3 months infants tend to accurately predict the occurrence of the target-side picture. It is possible that infants in this study

were using a sequential enumeration strategy to predict when the target picture would appear, although this is impossible to determine because number of home-side pictures was completely confounded with trial duration and stimulus display duration. Thus, infants could have used a nonnumerical strategy, such as time estimation, to anticipate the targets. The studies in the present investigation controlled for these alternative hypotheses. The stimulus sequences used by Canfield and Haith (1991) were modified so that the duration of stimuli and their interstimulus intervals (ISIs) were varied, preventing reliance on trial duration as a predictor for the appearance of the target picture. The number of home-side pictures, however, was retained as a perfect predictor, making the use of a sequential enumeration strategy possible. If time was eliminated as a reliable cue and infants developed an expectancy for the target picture (evidenced by anticipatory saccades to future picture locations), then the case for a sequential enumeration capacity would be strengthened. A comparison condition was used to evaluate anticipatory behavior when neither time nor number was a reliable predictor. This condition was identical to the number condition (the probability of finding a picture on the left or right was identical), except for the predictive value of numerical information. If infants in both conditions were more likely to anticipate the appearance of the target picture after the second rather than the first home-side picture, then a nonnumerical basis to the visual expectancy would be likely. In addition, the use of this comparison condition was used to gather evidence for a specific effect of numerical predictability on saccade reaction lime (RT). That is, even if no effect of number on visual anticipatory behaviors was found, the between-subjectsfindingof differentially speeded RTs would provide evidence for number-based expectations. Experiment 1

Method Participants
Forty-four healthy, full-term infants who were within i week of their 5-month birthday served as study participants. The large majority of infants originated from dual-career White families with two collegeeducated parents in their early 30s. One half the infants were assigned to the number condition and one half to the irregular condition. Because of the differential availability of male and female infants, each group of 22 infants consisted of 14 girls and 8 boys. Remarkably, there was no participant attrition in this study; data from all 44 infants were included in thefinalanalyses.

Apparatus
Visual stimuli were generated on-line using GRASP graphics animation software (Bridges, 1990) installed on a 386 computer and presented on a 19-in. (48.25 cm) SVGA NEC (Multisync XL) color video monitor. Infants were seated on their parent's lap in an opaque foursided chamber measuring 1.22 m wide by 1.22 m deep by 1.67 m high. The inside of the chamber was painted with an infrared-light absorbent paint(Mooreblack215 80). The infants' faces were illuminated by a Bausch and Lomb illuminator (Cat. No. 31-32-42) fitted with niters that restricted its emissions to a narrow band of wavelengths in the invisible near-infrared range.

INFANT SEQUENTIAL ENUMERATION Recordings of eye movements were made using a Cohu (4810) camera fitted with a chip sensitive to the same narrow band of invisible wavelengths as that produced by the filtered light source. (For additional details on infrared recording techniques, see Canfield & Haith, 1991; Haith, 1980.) An image of the infant's eyes together with a 1 / 100-s time code generated by a Panasonic (WJ810) time-date generator, and a digit indicating stimulus location and duration were recorded onto VHS format videotape by a Panasonic (AG73OO) video deck. Saccade latencies were coded from these tapes off-line by human observers using a combination of slow-motion and stop-frame analyses. Interrater agreement for categorizing pictures as anticipations, reactions, or missing data was maintained at or above 95%, as were judgments of absolute saccade latency (within one video frame).

271

The stimuli subtended approximately 8" of visual angle and appeared approximately 15 to the left or right of visual center. During the 2- to 4-min period of data collection, the only visible light available to the infant camefromthe video monitor. Filming continued until the experimenter determined that the infant had become disinterested or fretful. If, at the end of 4 min, the infant was still engaged by the task, filming was discontinued. On average, infants saw 95 pictures during a session, 70 of which were classified as usable data (i.e., 2.5 min or 23 trials). Because the visual expectation procedure is afree-lookingprocedure, there was wide variability in the number of pictures seen by individual infants (SD for usable pictures = 24). Infants in the irregular condition had an average of 23 more usable data points than infants in the number condition, Dependent measures. Following Canfield and Haith (1991), we were interested in whether the infants would develop an expectation for the appearance of if based on the number of pictures appearing on the left. It is important to distinguish the future-oriented cognitive construct ofexpectation from its behavioral indexes anticipation and facilitation (Candid & Haith, 1991; Haith etal., 1988). Definition and empirical validation of the anticipation interval. A saccade is considered anticipatory if the eye begins to move toward the location of the next picture and occurs either before the next picture appears or so quickly after its appearance that the reaction latency is lower than the minimum possible RT for infants that age. (In this context, RT is defined as the latency of thefirsteye movement and does not include movement time.) The time between the offset of the previous picture and the minimum possible RT is referred to as the anticipation interval. In previous studies, the lowest possible RT for infants has been defined to be 200 ms after stimulus onset. However, this cutpoint was determined by inference from existing adult saccade data (Canfield & Haith, 1991; Haith et al., 1988) and has not been empirically validated for infants. Given that adults and children sometimes react much more quickly than 200 ms when stimuli are partially predictable (Cohen & Ross, 1977; Findlay, 1981; Ottes, Van Gisbergen, & Eggermont, 1985) and given that reaction latency has been found to decrease dramatically during infancy (Canfield & Haith, 1991; Canfield, Wilken, Schmerl, & Smith, 1995), it is necessary to establish an empirical foundation for defining the anticipation interval for 5month-old infants. Figure I shows a histogram of latencies for every saccadic eye movement measured in Experiment 1 (2,168fixationshifts by 44 participants). Displaying the data in this way allows one to identify the anticipation distribution and the reaction distribution and examine their shapes. In addition, the method provides us with information helpful for determining the latency cutpoint between anticipations and reactions. Because reactions are elicited by environmental input, the reaction distribution should appear positively skewed with the minimum value corresponding to the fastest possible RT (Luce, 1986). Anticipations, on the other hand, are self-selected rather than elicited, and should therefore have a more uniform distribution in which the maximum value corresponds to the latency bin just before the minimum RT. Therefore, we would expect to see a discontinuity in the frequency histogram of saccade latencies for all participants at the point at which infants switchfromthe anticipation process to the reaction processthe point at which the frequency histogram of saccades increases rapidly. Inspection of the latency histogram in Figure 1 reveals a marked discontinuity at 200 ms following picture onset. Therefore, we defined shift latencies greater than or equal to 200 ms as belonging to the RT distribution, and shift latencies less than 200 ms are considered to belong to the anticipation distribution. Previous research with younger infants (Canfield & Haith, 1991; Haith etal., 1988) classified 200-ms latencies as anticipatory, whereas we include them in the reaction distribution. Chance-corrected anticipation. As a method for characterizing infant anticipatory behavior in the context of an asymmetric sequence, we intro-

Design
Infants were randomly assigned to one of two stimulus conditions. In the number condition, infants saw a repeating, predictable sequence of pictures alternating left (L) and right (R) of visual center in a left-leftright (L1-L2-R) asymmetrical pattern. (Because previous research [e.g., Canfield & Haith, 1991] has found no difference in performance between infants given a L1-L2-R sequence or a R1-.R2-L sequence, a R1-.K2-L sequence was deemed unnecessary for the present study.) Each repetition of L1-L2-R is referred to as a trial. The stimulus durations and the durations of the ISIs varied among trials. Within trials, the left pictures appeared according to one of three different timing schedules: fast, average, and slow. For the fast trials, L1 and L2 appeared for 500 ms each and were separated by 500-ms ISIs. The average trials used 1,000-ms stimulus durations and ISIs, and the slow trials used 1,500-ms durations and ISIs. The target picture (R) in each trial was always preceded by a 1,000-ms ISI and remained on for 1,500 ms. Trial durations were randomized across the sequence, which contained equal numbers of fast, average, and slow trials during each minute. Thus, although time was varied, infants could treat all trial types as similar with respect to the number of pictures appearing on the left, enabling them to correctly anticipate R. Infants in the irregular condition saw the same set offivepictures in a sequence that alternated left and right of visual center in an unpredictable manner. One third of the pictures appeared on the right side and two thirds appeared on the left, as in the number condition. Similarly, left pictures varied among the fast, average, and slow timing parameters used for the number condition. Unlike the number condition, however, trials consisted of one of three patterns, L1 -R, L1-L2-R, or L1 L2-L3-R, which were randomly presented in the sequence shown to the infants. There were equal numbers of the three patterns, and each pattern was shown an equal number of times as fast, average, and slow trials. As in the number condition, R was always preceded by a 1,000ms ISI with a 1,500-ms stimulus duration.

Procedure
Once judged to be in a quiet or active alert state, infants were placed on their parent's lap in a chair located in the experimental chamber and were positioned so that their eyes were approximately 50 cm from the video monitor. Parents were cautioned that when looking at the pictures themselves, they were not to move their head or body in concert with the picture side shifts. While the baby was being positioned, the monitor displayed large, high-contrast, black-and-white stimuli filling each quadrant of the screen. These stimuli served to orient the infant's eyes to the video screen and allow the experimenter to focus the camera. When the experimenter had the infant's eyes in focus, the already dim room lights were extinguished, the VHS deck was started, and the stimulus sequence was begun. Stimulus content consisted of a randomly ordered, heterogeneous set offiveanimated stimuli (smiling face, turning head, running puppy, rotating wheel, and jumping stick figure).

272
200
n

CANF1ELD AND SMITH

-1000 -800

-600

-400

-200

200

400

600

800

1000

Saccade latency bin


Figure I. Frequency histogram of all saccade latencies for all participants in Experiment 1 (N - 44 participants, N = 2,168 shifts). There is a marked discontinuity in the distribution at 200 ms. This discontinuity was located at the same point regardless of stimulus condition and was likewise reflected in individual participant data. Furthermore, the discontinuity confirms our model of the independence of reaction and anticipation, as well as recommending a minimum reaction time (RT) value of 200 ms. RT bin widths are 33.33 ms(one video frame).

duce a new measure which we call chance-corrected anticipation (CCA). The calculation of CCA is best described with an example. Assume that during the presentation of an L1-L2-R sequence, an infant sees 30 repetitions of the sequence (i.e., a total of 90 pictures). Assume further that the infant anticipates 27 times, or 30% ofthe pictures. If the infant's frequency of anticipation does not differ as a function of picture position, that is, anticipations occur at random times, then we would expect there to be 9 anticipations after L1,9 after L2, and 9 after R. By subtracting 9, the estimate of chance-level performance, from the observed number of anticipations at a particular picture position, CCA can be determined. Continuing with the example, assume that, the infant anticipated after LI 5 times, after L2 13 times, and after R 9 times. By subtracting 9 from each of these values, we see that CCA is - 4 after LI, 4 after L2, and 0 after R. Negative CCA val ues indicate that the infant anticipated less than would be expected by chance, and positive CCA values indicate that the infant anticipated more than expected by chance. Facilitation. The second measure of expectation in this study is facilitation. As noted by Canfield and Haith (1991), infants might have an expectation for a picture side shift and yet not actually anticipate the change in location with an appropriate eye movement. Evidence for this expectation might be found in infants' reaction latencies. If the latencies are speeded in relation to a baseline measure, then those times are facilitated by the infants' expectation for the upcoming picture side shift. Thus, facilitation, in the form of reduced latency to respond to the picture after it appears, serves as the second measure of expectation. Reduced latency is denned as the amount of change in median RT from the baseline block (first six pictures presented) to the median RT of the post baseline block (the subsequent pictures fixated). A large drop in median RT from the baseline block indicates a high degree of facilitation of the infant's saccadic behavior, providing evidence for expectation.

Hypotheses
For infants in the number condition, CCA is expected to be positive after L2 and negative after L1, indicating that the infants anticipate to the right more than would be expected by chance after two left pictures and less than would be expected by chance after one left picture. Infants in the irregular condition are expected to show CCA no different from zero after LI and L2. It is important to point out, however, that the irregular condition does have some numerical predictability. Namely, L3 is always followed by R. Furthermore, in both conditions, R is always followed by LI, so it is expected that CCA will be greater than zero after R in both conditions. Within the number condition, infants who detect the numerical regularity should have higher CCA scores after L2 than after LI, but this should not be true within the irregular condition. When comparing between conditions, we expect that CCA scores after L2 will be higher for infants in the number condition. There may also be suppression of anticipation, that is, lower CCA scores after LI in the number condition compared with the irregular condition. As for facilitation, it is expected that, in both conditions, there will be an overall decrease in median RT from baseline to postbaseline. Such a finding would reflect infants' perception of some of the invariant properties common to both conditions, for example, that pictures only appear on the left or right, and LI appears after R. However, if the numerical regularity adds specific predictability to the sequence, then we expect infants in the number condition to show a greater drop in RT from baseline to postbaseline compared with infants in the irregular condition.

Results
As hypothesized, only infants in the number condition shifted fixation during the anticipation interval in a manner indicating

INFANT SEQUENTIAL ENUMERATION

273

they learned the structure of the picture sequence. Infants in the irregular condition showed no tendency to differentially anticipate as a function of picture position. Also as expected, infants' RTs to postbaseline pictures were faster than baseline, but, contrary to expectations, there was no differential facilitation as a function of numerical predictability. We found no significant sex differences on any dependent variable, and thus data were collapsed across sex for the analyses that follow. An alpha level of .05 was used for all statistical tests. Chance-Corrected A nticipation We analyzed CCA in two ways. First, within each condition, the CCA value at each picture position was tested against a null value of zero, or chance. Second, an overall analysis of variance (ANOVA) was conducted and followed by various planned comparisons. Difference from chance levels. Figure 2 summarizes the results when CCA at each position is tested against a null value of zero, the expected value when infants shift fixation in a random manner. Within the number condition, CCA after LI was significantly less than zero, /(21) - 5.55, p = .000, whereas CCA after L2 was significantly greater than zero, t(21) = 2.58, p - ,017. However, CCA after R was not different from zero. Within the irregular condition, CCA scores after LI, after L2, and after R were not significantly different from zero. However, CCA after L3, which was always followed by R, was significantly greater than zero, t{2\) = 2. H , p = . 047. Analysis ofvariance. Table 1 provides the means and standard deviations for CCA as a function ofcondition and picture position. We conducted a Condition X Position mixed factorial ANOVA using CCA as the dependent measure. Condition was a betweensubjects variable with two levels (number and irregular), and position was a within-subject variable with three levels (After LI, After L2, and After R). Note that After L3 was not included in the model because this position does not exist in the number condition. A planned interaction contrast comparing After LI to After L2 revealed a significant Condition X Position interaction, F( 1, 42) = 8.88, p = .005. Within the number condition, infants anticipated more after L2 than after LI, F( 1, 42) = 27.34, p = .000, but within the irregular condition, there was no significant difference in CCA between After LI and After L2, F{ 1, 42) = 1.03, p -.316. For both conditions, within-condition contrasts involving After R revealed no significant pairwise differences. Analyses of CCA between conditions at each picture position did not reveal any significant differences, although the means were in the predicted direction. Facilitation Analysis of RT data indicated facilitation of rapid responding by infants in both conditions. For each infant, a baseline RT was calculated as the median RT for fixation shifts greater than or equal to 200 ms for the first six opportunities to react. Postbaseline RT consisted of the median of all RTs greater than or equal to 200 ms after the baseline. Mean of median baseline RT was 442 ms (SD - 124) for the number condition and 426 ms {SD = 94) for the irregular condition. Mean of median post-

baseline RT was 383 ms (SD = 76) for the number condition and 378 ms (SD = 52) for the irregular condition. In both conditions, infants showed a significant decrease of about 50 ms from baseline to postbaseline RT. A 2 X 2 repeated measures ANOVA with condition (number or irregular) as a betweensubjects variable and block (baseline or postbaseline) as a within-subjects variable revealed a significant effect of block, F{ 1, 42) = 15.75, p = .000), but no evidence for a condition effect or for a Condition X Block interaction. Although infants in the number condition consistently showed a greater amount of change from baseline to postbaseline at each picture position, in no case did the difference approach statistical significance. Discussion Results of Experiment 1 indicate that even when the timing of sequentially presented stimuli is varied, 5-month-old infants
Experiment 1
Number Irregular

After L1

After L2

After L3

After R

Picture Position

Experiment 2
Number Irregular

After L1

After L2

After L3

After L4

After R

Picture Position

Figure 2. Chance-corrected anticipation as a function of picture position and condition for Experiment 1 and Experiment 2. Note that chance-corrected anticipation after L2 for the irregular condition of Experiment 1 is .00. Also note that the picture position after L3 exists only for the irregular condition of Experiment 1, and the position after L4 exists only for the irregular condition of Experiment 2. Single asterisks indicate values that are significantly different from zero at/> < .05; double asterisks indicate values that are significantly different from zero at p <.01. L = left; R = right.

274
Table I

CANFIELD AND SMITH

able to correctly discriminate between positions L2 and L3 beMean and Standard Deviation Chance-Corrected Anticipation cause both are "more than 1." as a Function ofCondition and Picture Position for Experiments 1 and 2 Method Condition
Number Picture position Experiment I 8 After LI After L2 After L3 After R Experiment 2b After LI After L2 After L3C After L4 After R M SD M -0.42 0.00 0.42 -0.01 -0.44 -0.20 0.35 -0.40 0.69 Irregular SD 2.11 2.22 0.94 4.04 1.09 1.48 1.02 0.72 1.37

Participants
The large majority of the families who agreed to participate in the study consisted of two-parent families in which over 50% of the fathers had at least some graduate education and over 50% of the mothers had at least an undergraduate degree. Thirty-two healthy, full-term 5month-old infants (M = 155 days; SD = 6 days) served as study participants. One half of the infants were assigned to the number condition and one half to the irregular condition. Each group of 16 infants consisted of 8 girls and 8 boys. Data from six additional infants, who were fussy during the procedure, were not included in the analyses.

0.9 l b 0.34b,c -0.93a -O.55 a 1.54b -0.06 a

1.06 0.94 1.93 0.81 0.94 1.50 1.68

Design
The design was identical to Experiment 1 except that trials in the number condition consisted of three left pictures followed by one right picture (L1-L2-L3-R). Infants in the irregular condition were shown three different trial types: L1-L2-R, L1-L2-L3-R, and L1-L2-L3-L4-R. The various stimulus durations and ISIs were the same as those used in Experiment I. As in Experiment 1, the ratio of left pictures to right pictures was equivalent for both conditions. In this study, 75% of the pictures were on the left and 25% were on the right.

Note. For each experiment, means in the same column that do not share a common subscript differ at/7 < .05. L = left; R = right. 8 n = 22 in each condition b n = 16 in each condition. c Betweenconditions comparison, p< .05.

use numerical information to anticipate the appearance of the target picture. After one left picture in the number condition, infants inhibit anticipations to the right, but after two left pictures, they increase their frequency of anticipation in relation to that expected by their base rate over all positions. Infants in the irregular condition, on the other hand, did not show differential anticipatory responding after one or two left pictures, providing further support for the conclusion that infants in the number condition were responding on the basis of the number of home-side events. In further support of infants' ability to extract numerical information from sequentially presented stimuli, infants in the irregular condition discovered that a right picture always appeared after three left pictures. This suggests the possibility that 5-month-olds can enumerate at least three sequentially presented stimuli, a possibility we explored in Experiment 2. Experiment 2 Experiment 2 was designed to determine whether 5-monthold infants can correctly use numerical information when viewing a sequentially presented stimulus sequence of the form LlL2-L3-R. This study will provide information about the range of numerical capacities at this age while also providing important information about the processes underlying early numerical abilities. Considerably more numerical competence is required to correctly isolate the dimension of number for a 3 / 1 sequence. This can most easily be seen by considering the competence required to correctly inhibit an anticipation after L2. The second picture cannot be correctly identified within a simple "more" versus "less" scheme because, in the context of an L1-L2-L3-R sequence, L2 represents both "more" than one and "less" than three. Also, if infants of this age are limited to conceptions of " 1 " and "more than 1," then they would be un-

Apparatus and Procedure


The apparatus and procedure were identical to that used in Experiment 1. Infants saw an average of 101 pictures during a session, of which 80 provided usable data (i.e., 2.8 min or 20 trials). Unlike Experiment 1, the mean number of usable pictures for the irregular condition (78) did not differ from the number condition (82 ),t( 15) = .65, p = .523.

Dependent Measures
The same two dependent measures from Experiment 1 were used: CCA and facilitation. In addition, the same method for confirming the anticipation cutoff at 200 ms was used in the present study. As was true for Experiment I, the latency histogram for Experiment 2 showed a sharp discontinuity at 200 ms.

Hypotheses
Results in support of sequential enumeration of three events would parallel the findings of the first experiment. In the number condition, infants would show positive CCA after L3 and negative CCA after LI and L2. Infants in the irregular condition would have CCA after L2 and L3 that is no different from zero. As in Experiment 1, there is some numerical information in the irregular sequence, namely, LI is always followed by a left picture and L4 is always followed by a right picture. If infants detect this information, then CCA in the irregular condition will be negative after LI and positive after L4. Once again, R is aiways followed by a left picture in both conditions, so detection of this regularity would result in positive CCA after R, although this was not found in Experiment 1. Within the number condition, infants who detect the numerical regularity should have higher CCA scores after L3 than after LI or L2, but this should not be true within the irregular condition. When comparing between conditions, we expect that CCA scores after L3 will be higher for infants in the number condition. Suppression of anticipation would be indicated by lower CCA scores after LI and after L2 in the number condition in relation to the irregular condition.

INFANT SEQUENTIAL ENUMERATION Experiment 1 did not indicate that infants' RTs are facilitated differentially as a function of the number of left pictures, although there was an overall drop in RT from baseline to postbaseline in both conditions. Nevertheless, under the hypothesis that numerical regularity will facilitate faster reactions, we predict a greater drop in RT from baseline to postbaseline in the number condition.

275

Results
Results indicate that infants in the number condition distinguished the third left picture from thefirstand second and used this information to anticipate preferentially after L3, whereas infants in the irregular condition did not. Analyses of RT indicated no evidence of overall facilitation or of differential facilitation as a function of condition. As in Experiment 1, an alpha level of .05 was used on all statistical tests. Chance-Corrected Anticipation Difference from chance levels. Figure 2 summarizes the results when CCA at each position is tested against a null value of zero. Within the number condition, CCA after LI and after L2 was significantly less than zero, t(15) = 4.55,p = .000 and r( 15) = 2.33, p = .034, respectively. CCA after L3 was significantly greater than zero, /(15) = 4.10, p = .001, but CCA after R was not significantly different from zero. Suprisingly, within the irregular condition, CCA was significantly less than zero at position L4, ?(15) = 2.20, p = .044. However, CCA did not differ from zero at any other picture position. Analysis of variance. Table 1 provides the means and standard deviations for CCA as a function of condition and picture position. A Condition X Position mixed factorial ANOVX was conducted using CCA as the dependent measure. Condition was a between-subjects variable with two levels (number and irregular), and position was a within-subjects variable with four levels (After LI, After L2, After L3, and After R). Note that After L4 was not included in the modei because this position has no counterpart in the number condition. A planned interaction contrast in which After LI and After L2 were compared with After L3 revealed a significant Condition X Position interaction, F( 1,30) = 8.16, p = .008. The nature of this interaction is revealed in that infants in the number condition, but not the irregular condition, anticipated more after L3 than after either LI or L2, F{ 1, 30) = 42.84, p = .000 and F( 1, 30) = 15.71, p = .000, respectively. CCA for After LI and After L2 in the number condition did not differ from one another, F( 1, 30) = .70, p - .409. Finally, in a comparison of the two positions with predictable side shifts, infants in the number condition anticipated less after R than after L3, F{ 1, 30) = 6.57, p = .016. Between-conditions analyses of CCA scores revealed a significant difference only after L3, indicating that infants in the number condition were more likely to make anticipatory fixation shifts after the third left picture than were infants in the irregular condition, F{ 1, 30) = 6.87,p = .014.

irregular), and block was a within-subjects variable with two levels (baseline and postbaseline). Mean baseline RT was 411 ms (SD = 91) for the number condition and 397 ms (SD = 117) for the irregular condition. Mean postbaseline RT was 403 ms(SD = 70) for the number condition and 375 ms(SD = 73) for the irregular condition. No main effect of condition or block was found, and there was no significant Condition X Block interaction. There was also no significant change from baseline to postbaseline as a function of picture position in either condition.

Discussion
Similar to the results of Experiment 1, infants in the number condition appeared to use the number of left pictures to anticipate the appearance of the right picture. Specifically, they refrained from anticipating after thefirstand second left pictures, but they anticipated at greater than chance rates after the third left picture. Infants in the irregular condition did not show the same pattern: Their anticipatory behavior differed from chance only at picture position L4. In addition, infants in the number condition were significantly more likely to shift fixation after L3 than were infants in the irregular condition. Unlike Experiment 1, no significant change in RT from baseline to postbaseline was found. Because infants in the number condition increased their anticipation rate relative to chance only after the third left picture and because CCA after L2 was not significantly different from that after LI, thesefindingsare congruent with the hypothesis that infants in the number condition were correctly distinguishing LI and L2 from L3. In addition, the fact that infants in the irregular condition did not respond in a similar manner to those in the number condition argues against a nonnumerical explanation. Therefore, these results confirm and extend the findings from Experiment 1 and suggest that 5-month-old infants can enumerate up to three sequentially presented events. General Discussion Taken as a whole, our findings suggest that at the age of 5 months infants can rapidly detect the numerical relation to spatial location in a complex sequence of visual stimuli. Infants go beyond mere detection: They develop an expectation for where and when the next picture will appear, and they act on this expectation by making an appropriate anticipatory eye movement. In both experiments, infants in the number condition, but not in the irregular condition, differentially shifted fixation to the right after the appropriate number of pictures on the left. Thesefindingsprovide important confirmation for the type of sequential numerical capacity suggested by the findings of Starkey et al. (1983, 1990) in their studies of intermodal numerical correspondence and by Wynn (1992). However, our findings extend those previously reported. First, we used a recently developed procedure, visual expectation, which has not previously been used to examine infants' numerical competence and does not use traditional fixation duration measures. Instead, the infants are required to actively anticipate events in the absence of any sensory stimulation. We believe interpretations of behavior from anticipation to be more direct than

Facilitation
A Condition X Block mixed factorial ANOVA was computed with median RT serving as the dependent variable. Condition was a between-subjects variable with two levels (number and

276

CANFIELD AND SMITH

interpretations from fixation duration. For example, in the cross-modal matching paradigm, it is not clear whether the extraction of numerical information from the auditory event sequences should produce a familiarity response (i.e., looking longer at the numerically corresponding array) or a novelty response (i.e., more looking at the numerically noncorresponding display). Either response could be used to support numberbased responding, even after the results are in. In fact, in Moore et al.'s (1987) attempt to replicate the cross-modal effect, the only reliable trend they reported was for infants to look longer at the noncor responding display, whereas the original finding reported by Starkey et al. (1983) was based on infants showing a preference for the numerically corresponding display. Second, by using a paradigm in which infants can demonstrate partial competence, we believe our findings provide a more detailed picture of infants' early grasp of number. For example, our infants clearly demonstrated an ability to use numerical information to inform their behavioral choices, but they were far from perfect in their predictions. This is indicated by the relative number of correct to incorrect anticipatory shifts. Visual preference methods seem less well-suited to the demonstration of a partial competence than to demonstrations of existence or nonexistence. As long as the relevant group fixates the relevant stimulus more, the case for competence, no matter how primitive, can be made. Using a procedure that reveals partial competence may facilitate discoveries of age-related changes in the infant's grasp of number and allow one to assess the effects of specific procedural changes on the degree of competence shown. Finally, we demonstrated infants' ability to enumerate events in a completely sequential task. The numerical discriminations required in our application of the visual expectation paradigm do not involve comparing a representation of number from a sequential set with the number of items in a simultaneous set. Instead, infants use number to connect perception to action in real time.

sequence, the trial begins with a 500-ms ISI. For fast trials in a L1-L2-R sequence, R always follows 2.5 s after the onset of LI. By associating the 500-ms ISI with the 2.5-s interval during the fast trials while concurrently forming similar associations for average and slow trials, infants would be able to use a timing strategy to produce the results we observed. If infants used a timing strategy in our experiments, we would expect their anticipatory behavior to show scalar variance. In this situation, scalar variance means that as the duration to be timed increases, the variability in estimating the end of the interval increases. Scalar variance is commonly found in studies involving the synchronization of a motor response with an anticipated sensory event (Bartlett & Bartlett, 1959). Our present data do not enable us to speak definitively to the scalar variance issue because infants in our experiments do not show sufficient numbers of anticipations for all three trial types to allow for powerful hypothesis tests. Nevertheless, the scalar variance property predicts that infants' anticipations for R should show progressively greater variability for short, average, and long trial types. If Experiments 1 and 2 are combined, there are 6 infants with two or more anticipations for each trial type, allowing the computation of a latency range for anticipations to the right picture. Of these 6 infants, only 1 obeyed the scalar variance property. Thus, there appears to be no consistent relation between variability of anticipation latency and trial duration and, therefore, no support for the complex timing strategy.

The Tempo Hypothesis


Related to timing is the tempo hypothesis. Because the time until the appearance of the next right picture can be predicted from the duration of the ISI following the previous right picture offset, perhaps infants detect the different tempos of the fast, average, and slow trials, rather than their precise timing parameters. That is, maybe infants learn three approximate durationshift associations of the form: Fast pictures mean shift to right soon after the onset of LI, slow pictures rnean shift right in a more leisurely fashion, and average pictures mean shift right at an intermediate tempo. Presumably, if infants are more highly "energized" or stimulated by the fast trials (faster tempo), then they would make their anticipations earlier in the ISI than they would for average and slow trials (slower tempos). Similarly, as the trial tempos become slower, infants should anticipate later in the ISI following the second (L-L-R) or third (L-L-L-R) left picture. By combining those infants from Experiments 1 and 2 who anticipated the right picture at least one time for each trial type, we can test this prediction. Nineteen of 38 infants could be used for the analysis. A repeated measures ANOVA revealed no significant difference in anticipation latency as a function of trial type, F(2, 36)= 1.87,p = .168. In fact, anticipation latencies were longest for average trials and shortest for long trials. Therefore, the tempo hypothesis is not supported, as we found no difference in average anticipatory shift time as a function of sequence duration. In addition to the lack of empirical support for the tempo and timing hypotheses, both strategies seem implausible on intuitive grounds. The demonstration that 5-month-old infants can form expectations for three temporal durations, which are randomly presented approximately seven times each over a 2- to

Alternative Hypotheses
Our conclusion that infants can enumerate sequences of visually presented stimuli must withstand the challenge presented by several alternative explanations for our results. Use of a complex timing strategy is one such alternative hypothesis. Responding on the basis of tempo is another possible strategy. Both of these possible hypotheses can be addressed by our data.

The Timing Hypothesis


From the design of our experiments, it can be argued that infants used a complex timing strategy to anticipate the occurrence of the right picture. Because picture durations and ISTs were constant within a trial, in both experiments, infants could have learned the temporal parameters associated with each sequence type (fast, average, and slow) and used the consistent within-trial timing information to inform their anticipations. Such a strategy would involve estimating the time from the onset of LI to the onset of R within each sequence type and associating this time interval with the duration of the ISI following the right picture of the previous trial. For example, in a fast

INFANT SEQUENTIAL ENUMERATION

277

4-min period, would indicate substantially more sophisticated information-processing ability than would the much less effortful strategy of detecting the numerical invariant that predicts the appearance of R. Therefore, given the weight of both the rational and empirical evidence, wefindit difficult to entertain either the timing or tempo hypotheses as plausible alternatives to the number hypothesis. Processes Involved in Sequential Enumeration To this point, we have addressed the question of whether infants can enumerate a purely sequential set of events. We have not, however, addressed the important question of enumerative process. Traditionally, there have been two hypothesized processes involved in enumeration: counting and subitizing. It is generally agreed that counting involves the one-to-one correspondence of a stably ordered set of numerons, or mental tags, with items in a to-be-counted set. The final assigned numeron represents the cardinal quantity of the set (Gelman & Gallistel, 1978, 1985). To support a claim for infant counting using a conservative definition (e.g., Fuson, 1988; Thomas & Lorden, 1993), it would be necessary to show that babies have access to a stably ordered set of unique numerons that are applied to each picture within a trial, with the final tag having a direct cardinal or ordinal association or both. (Our design does not allow us to distinguish a response that is based on an appreciation of cardinal number ["three" ] from one based on an appreciation of ordinal position ["third"].) By requiring direct evidence for each component of counting, our definition is more restrictive than the somewhat more liberal definitions that are often used (e.g., Gallistel, 1993; Gelman & Gallistel, 1978, 1985), but we prefer it because it allows for greater potential differentiation between various enumeration processes. In the case of our studies, a counting hypothesis could be rejected if it were found that infants scanned the contents of working memory to reaccumulate the entire set of events that were being enumerated. Scanning would indicate an appreciation of "numerousness" rather than "numerosity" (Kaufman, Lord, Reese, & Volkman, 1949; Stevens, 1951;Taves, 1941). Having to reaccumulate items by scanning working memory would demonstrate that infants were not using the final assigned numeron to directly infer cardinal number or ordinal position. An important difference between enumeration in infants and children may be related directly to the availability of such direct associations. Without them, infants' counting range would be limited by available working-memory capacity, which may explain the consistent finding that infants do not enumerate sets greater than four (e.g., Strauss & Curtis, 1981). If infants use a working-memory scan in our visual expectation task, one would expect the latency to anticipate R after L2 in the two-event sequence to be shorter than the latency to anticipate R after L3 in the three-event sequence, under the assumption that it takes less time to scan two events in working memory than three. To explore the plausibility of this hypothesis, we compared the time of occurrence of anticipatory fixation shifts to R from the two-event sequence of Experiment 1 to the same measure derived from the three-event sequence of Experiment 2. Our analysis provided no support for the scanning hypothesis. The mean anticipation latency for infants from Exper-

iment 1 was -138 ms (138 ms before the onset of R), and the mean latency for infants in Experiment 2 was only 10 ms longer, -148 ms, r(30) = . 14, p = .89. Thus, on the basis of this analysis, we cannot rule out the possibility that infants are making a direct association between the final item and the number of items in the set (i.e., that they are truly counting). We believe, however, a more parsimonious explanation for our findings involves an enumerative process that is simpler than counting. This process has long been known as subitizing (Kaufman et ah, 1949). The characterization of subitizing as a process has changed substantially since it wasfirstintroduced, but presently it seems to refer to two distinct types of enumeration: numerical subitizing and nonnumerical (orfigural)pattern-based subitizing. In addition, independent of whether subitizing is numerical or pattern-based, some models assume that subitizing is possible only for sets of items that are simultaneously displayed, whereas others allow for the possibility of sequential enumeration. Thus, existing models of subitizing fall into one of four classes. Mandler and Shebo's (1982) model considers subitizing to be a nonnumerical process of pattern perception. A second model, that of von Glasersfeld (1982), also considers subitizing to be a predominantly nonnumerical process, but his model is meant to explain the enumeration of both simultaneously and sequentially presented items in terms offiguralpatterns. Unlike Mandler and Shebo's and von Glasersfeld's models, two other models consider subitizing to be a truly numerical process. Meek and Church (1983) and Dehaene and Changeux (1993) viewed subitizing as a numerical process enabled by the availability of the functional equivalent of a number line in the brain of the rat or in the computer model. These two models can be distinguished, however, in that the Meek and Church model] is designed to enumerate both simultaneous and sequential item arrays, whereas Dehaene and Changeux's model has not yet been applied to sequential enumeration. Despite the different classes these models represent, they all satisfy certain formal criteria for an adequate model of enumeration. First, each provides a means for unitizing perceptual experience. Unitizing requires two steps: (a) segmentation of perceptual experience into those things to be enumerated and (b) either active stripping away of or more passive disregard for any sensory attributes associated with the segments. The second step is important because it prevents an especially intense or salient unitfrombeing mistaken as "more numerous" than a less-intense unit. Each of the models listed above has its own means of unitizing perceptual experience. Mandler and Shebo considered small collections of items to correspond to geometric patterns. The patterns themselves are abstract and have no sensory attributes. Von Glasersfeld offered the view that people release "attentional pulses" as they scan their environment. Each pulse is equivalent regardless of the nature of the stimulus that caused its release. Meek and Church have a similar unitizing mechanism to von Glasersfeld's. They proposed a pacemaker that constantly releases pulses, and as with von Glasersfeld's model, each pulse (unit of neural activity) is equivalent. R>r the Meek and Church model, when a countable

1 Meek and Church's (1983) experiments only used sequential presentation of stimuli, but Gallistel and Gelman {1992) expanded Meek and Church's model to include simultaneous displays. Because Meek and Church developed the model that Gallistel and Gelman later adopted, we refer to it as the "Meek and Church model."

278

CANFIELD AND SMITH sionals (Version 3.5) [Computer software]. Ashland, OR: Paul Mace Software. Canfield, R. L. (1988, April). Counting skills in early infancy? Poster session presented at the biennial International Conference for Infant Studies, Washington, DC. Canfield, R. L., &Haith, M. M. (1991). Young infants'visual expectation for symmetric and asymmetric stimulus sequences. Developmental Psychology, 27, 198-208. Canfield, R. L., Wilken, J., Schmerl, L., & Smith, E. G. (1995). Agerelated change and stability of individual differences in infant saccade reaction time. Infant Behavior and Development, 18, 351-358. Cohen, M. E., & Ross, L. E. (1977). Saccadic latency in children and adults: Effects of warning interval and target eccentricity. Journal of Experimental Child Psychology, 23, 539-549. Davis, H., & Perusse, R. (1988). Numerical competence in animals: Definitional issues, current evidence, and a new research agenda. Behavioral and Brain Sciences, 11, 561-615. Dehaene, S., & Changeux, J. P. (1993). Development of elementary numerical abilities: A neuronal model. Journal of Cognitive Neuroscience, 5, 390-407. Findlay, J. M., (1981). Spatial and temporal factors in the predictive generation of saccadic eye movements. Vision Research, 21, 347354. Fuson, K. C. (1988). Children s counting and concepts of number. New York: Springer-Verlag. Gallistel, C. R. (1993). A conceptual framework for the study of numerical estimation and arithmetic reasoning in animals. In S. T. Boysen & E. J. Capaldi (Eds.), The development of numerical competence (pp. 211-223). Hillsdale, NJ: Erlbaum. Gallistel, C. R., & Gelman, R. (1992). Preverbal and verbal counting and computation. Cognition, 44, 43-74. Gelman, R., & Gallistel, C. R. (1978). The child's understanding of number. Cambridge, MA: Harvard University Press. Gelman, R., & Gallistel, C. R. (1985). The child's understanding of number (2nd ed.). Cambridge, MA: Harvard University Press. Haith, M. M. (1980). Rules that babies look by. Hillsdale, NJ: Erlbaum. Haith, M. M., Hazan, C , & Goodman, G. (1988). Expectation and anticipation of dynamic visual events by 3.5-month-old babies. Child Development, 59, 467-479. Haith, M. M., Wentworth, N., & Canfield, R. L. (1993). The formation of expectations in early infancy. In C. Rovee-Collier & L. P. Lipsitt (Eds.), Advances in infancy research (pp. 251-297). Norwood, NJ: Ablex. Kaufman, E. L.. Lord, M. W., Reese, T. W.. & Volkman, J. (1949). The discrimination of visual number. American Journal of Psychology, 62, 498-525. Luce, R. D., (1986). Response times. New York: Oxford University Press. Mandler, G., & Shebo, B. J. (1982). Subitizing: An analysis of its component processes. Journal of Experimental Psychology: General, 111. 1-21. Meek, W. H., & Church, R. M. (1983). A mode control model of counting and timing processes. Journal of Experimental Psychology: Animal Behavior Processes, 9, 320-334. Moore, D., Benenson, J., Reznick, J. S., Peterson, M., & Kagan, J. (1987). Effect of auditory numerical information on infants' looking behavior: Contradictory evidence. Developmental Psychology, 23, 665-670. Ottes, F. P., Van Ginsbergen. J. A. M., & Eggermont, J. J. (1985). Latency dependence of colour-based target vs. nontarget discrimination by the saccadic system. Vision Research, 25, 849-862. Piaget, J. (1952). The child's conception of number. New York: Basic Books.

item is encountered, a pulse is allowed to pass through a gate and enter an accumulator. Finally, Dehaene and Changeux's model unitizes perceptual experience by passing visual stimulation through a normalizing field before passing it to an accumulator. The normalizingfieldtransforms all stimuli into equivalent units. Once the relevant percept has been unitized, the units must be combined so that number can be perceived. The combining of equivalent units is the second formal feature of process models of enumeration. For nonnumerical models of subitizing, the units are combined into either some geometric or figural pattern that is mapped to a certain numerosity in much the same way that something is recognized as being a certain color. Numerical models, on the other hand, assert that units enter a graduated accumulator. The gradations on the accumulator indicate the number of items in the set in the same way that height corresponds to number for a stack of coins. By responding on the basis of number in the context of our anticipation task, infants demonstrated their capacity (a) to treat pictures as units whose sensory attributes are irrelevant to the prediction of the future stimulus position, (b) to combine these units over time, and (c) to use the number of items accumulated to predict the occurrence of R. The nature of the relation between the accumulated items and either a cardinal or ordinal representation of number cannot be determined given the design of our study; however, we are not able to rule out the possibility of a direct association. Nevertheless, our results suggest the need for a model of infant enumerative processes that is truly numerical in nature and that can accommodate sequential presentation of items. Because infants never see more than a single stimulus at any moment during the session and yet still anticipate R more than would be expected by chance, we conclude that they are combining the events over time to arrive at an appreciation of quantity, either of numerousness or possibly of numerosity. This process of combining sequential events is numerical because the items were presented in a manner that was neither rhythmic nor temporally predictable. These facts serve to constrain existing models of infant numerical competence. We find Meek and Church's (1983) model, which Gallistel and Gelman (1992) have used to explain infant numerical competence, to be most consistent with our data. However, we have chosen to adopt a more conservative definition of counting than Gallistel and Gelman. Until we have more direct evidence that infants are responding on the basis of the unique numeron assigned to thefinalelement (numerosity) and not the entire collection of units (numerousness), we are not prepared to argue that the number-based responding in our study constitutes counting. Nonetheless, we agree with Gallistel and Gelman's view that the numerical abilities of infants is likely to serve as an important seed from which later symbolic counting grows. References
Antell, S. E., & Keating, D. P. (1983). Perception of numerical invariance in neonates. Child Development, 54, 695-701. Bartlett, N. R., & Bartlett, S. C. (1959). Synchronization of a motor response with an anticipated sensory event. Psychological Review, 66, 203-226. Bridges, J. (1990). GRASP: Graphics Animation Software for Profes-

INFANT SEQUENTIAL ENUMERATION Starkey, P., & Cooper, R. G., Jr. (1980). Perception of numbers by human infants. Science, 210,1033-1035. Starkey, P., Spelke, E. S., & Gelman, R. (1983). Detection of intermodal numerical correspondences by human infants. Science, 222, 179-181. Starkey, P., Spelke, E. S., & Gelman, R. (1990). Numerical abstraction by human infants. Cognition, 36, 97-127. Strauss, M. S., & Curtis, L. E. (1981). Infant perception of numerosity. Child Development, 52, 1146-1152. Stevens, S. S. (1951). Mathematics, measurement, and psychophysics. InS. S. Stevens (Ed.), Handbook of experimental psychology (pp. I49). New York: Wiley. Taves, E. H. (1941). Two mechanisms for the perception of visual numerousness. Archives ofPsychology, 37, 1-47. Thomas, R. K.., & Lorden, R. B. (1993). Numerical competence in

279

animals: A conservative view. In S. T. Boysen & E. J. Capaldi (Eds.), The development of numerical competence (pp. 127-147). Hillsdale, NJ: Erlbaum. Treiber, E, & Wilcox, S. (1984). Discrimination of number by infants. Infant Behavior and Development, 7,93-100. van Loosbroek, E., & Smitsman, A. W. (1990). Visual perception of numerosity in infancy. Developmental Psychology, 26, 916-922. von Glasersfeld, E. (1982). Subitizing: The role offiguralpatterns in the development of numerical concepts. Archives de Psychologie, 50, 191-218. Wynn, K. (1992). Addition and subtraction by human infants. Nature, 358, 749-750.

Received September 7, 1994 Revision received February 6,1995 Accepted June 5, 1995

Low Publication Prices for APA Members and Affiliates


Keeping you up-to-date. All APAFellows, Members, Associates, and Student Affiliates receiveas part oftheir annual duessubscriptions to ih&American Psychologist andAPA Monitor. High School Teacher and International Affiliates receive subscriptions to the APA Monitor, and they may subscribe to the American Psychologist at a significantly reduced rate. In addition, all Members and Student Affiliates are eligible for savings of up to 60% (plus a journal credit) on all other APA journals, as well as significant discounts on subscriptions from cooperating societies and publishers (e.g., the American Association for Counseling and Development, Academic Press, and Human Sciences Press). Essential resources. APA members and affiliates receive special rates for purchases of APA books, including the Publication Manual of the American Psychological Association, and on dozens of new topical books each year. Other benefits of membership. Membership in APA also provides eligibility for competitive insurance plans, continuing education programs, reduced APA convention fees, and specialty divisions. More information. Write to American Psychological Association, Membership Services, 750 First Street, NE, Washington, DC 20002-4242.

You might also like