Cognition, @Elsevier
6 (1978) 89-116 Sequoia S.A., Lausanne
1 - Printed
Perceptual
in the Netherlands
similarity of...
53 downloads
1027 Views
5MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
Cognition, @Elsevier
6 (1978) 89-116 Sequoia S.A., Lausanne
1 - Printed
Perceptual
in the Netherlands
similarity of mirror images in infancy* MARC H. BORNSTEIN** CHARLES
G. GROSS
JOAN 2. WOLF Princeton
University
Abstract Perception of mirror images by three- to four-month infants was studied in five experiments using habituation paradigms. In the first experiment, babies discriminated right profiles of two different faces but not the left and right profile of the same face. In the second, babies discriminated a 45” oblique from a vertical line, but not the oblique from its mirror image. In the third, babies discriminated oblique lines that differed by 50” and were not mirror images. In the final experiments. 90” rotations of a C-shape were discriminated but not 180” rotations that formed lateral or vertical mirror images. These results demonstrated that although babies were able to discriminate differences in orientation (even among obliques) they tended to view mirror images, especially lateral mirror images, as equivalent stimuli. We propose that the perceptual equivalence of mirror images reflects an adaptive mode of visual processing; mirror images in nature are almost always aspects of the same object, and they usually need not be discriminated. The relations of the perceptual similarity of mirror images to the ontogeny of the object concept and to the development of reading are discussed.
Introduction Orientation discrimination is critical to the perception of objects in visual space. Virtually all visual vertebrates are highly sensitive to orientation change, and the detection of orientation appears to be a relatively early *This research was supported by a grant from the Spencer Foundation to Princeton University. The authors wish to thank Kay Patterson and Barbara Cross for assistance in data collection, Joe Pylka and Joe Gnandt for technical assistance, Helen Bornstein, Eleanor J. Gibson, Betsy Ruddy, Herb Pick, Jr., Lynne Seacord, and Marian and Harold Sackrowitz for comments on an earlier draft of the manuscript, and Jacques Mehler and the anonymous reviewer who both urged us to add Experiment V. **Requests for reprints should be addressed to Marc H. Bornstein, Department of Psychology, Green Hall, Princeton University, Princeton, New Jersey 08540, U.S.A.
90
Marc H. Bornstein, Charles G. Gross and Joan Z. Wolf
stage of visual processing by the nervous system (e.g., Hubel and Wiesel, 1968). It is a curious fact, in this light, that the discrimination of a stimulus from its reflection 180” around the vertical axis (i.e., its left-right or lateral mirror image) represents an extremely difficult problem for a great variety of animals including octopuses, fishes, rats, monkeys, and human children and adults (see reviews by Bradshaw, Bradley and Patterson, 1976; Corballis and Beale, 1976; Sutherland, 1961; Tee and Riesen, 1974; Vogel, 1977). The classic demonstration of the unusual difficulty of lateral mirrorimage discrimination in children was that of Rude1 and Teuber (1963). They used a two-choice discrimination-learning paradigm in which the two stimuli were simultaneously presented and horizontally aligned. The lateral position of the stimuli was randomized from trial to trial, and the children were told whether they had chosen the “right” or “wrong” stimulus after each trial. Rude1 and Teuber found that children from four to nine years old had great difficulty in learning to discriminate mirror-image obliques (1 vs. \) and lateral mirror-image C shapes (C vs. 7) but readily learned to discriminate horizontal from vertical lines ( - vs. I) and a U-shape from its inversion or vertical mirror image (U vs. n). Rude1 and Teuber’s results have been repeatedly confirmed in both Western and non-Western cultures (e.g., Huttenlocher, 1967a; Over and Over, 1967; Sekuler and Rosenblith, 1964; Serpell, 1971), although the effect may be reduced with variants of their procedure. For example, when the stimuli are vertically aligned (z), vertical mirror images become more difficult to discriminate than horizontal mirror images vertically aligned (C,), although not as difficult as horizontal mirror images horizontally aligned (~1) (Huttenlocher, 1967a). Thus on simultaneous presentation, the presence of an orthogonal axis of symmetry between the stimuli appears critical for the discrimination-learning difficulty. The difficulty of discriminating mirror images’ seems to involve coding in memory. Animals and children can pick out the “odd” stimulus when presented with two identical patterns and their mirror image, but the same subjects find it very difficult to learn to respond consistently over a series of trials to one of two mirror images (e.g., Over and Over, 1967; Rude1 and Teuber, 1963 ; Tee and Riesen, 1974). In the Analysis of Sensations, Mach (1914, p. 110) noted that “children constantly confound the letters b and d, p and q. Adults, too, do not readily notice a change from left to right....” He ascribed this lateral mirror-image ‘We use the term “mirror images” to describe pairs of stimuli formed by reflection of an asymmetrical pattern about either the vertical axis (“lateral mirror images”) or the horizontal axis (“vertical mirror images”).
Perceptual similarity of mirror images in infancy 9 1
“confusion” to the bilateral symmetry of the body and nervous system of the perceiving organism, and his explanation is still prominent (e.g., Corballis and Beale, 1976; Noble, 1968; Orton, 1937). According to the modern verof an asymmetric stimulus in one sion of this view, the “representation” hemisphere is a lateral mirror image of its representation in the other hemisphere, and this dual representation (somehow) leads to the confusion of lateral mirror images. Successful discrimination of lateral mirror images is supposed to depend on the development of asymmetry in the organism, such as hemispheric dominance or handedness. Gross and Bomstein (1978) have argued on several grounds against such an explanation of mirror-image confusion. For example, there is no physiological evidence for “mirror representation” in the two hemispheres nor are there any known interhemispheric connections that could provide it (Allman and Kaas, 1975; Brooks and Jung, 1973; Zeki and Sandeman, 1976). Another difficulty is that when visual information does transfer from one hemisphere to the other, there is no behavioral evidence that it mirror reverses in doing so (Corballis, Miller and Morgan, 1971; Hamilton and Tieman, 1973; Lehman and Spencer, 1973; Storandt, 1974; but see Corballis and Beale, 1976). Furthermore, mirrorimage confusions persist in adulthood, even after the development of lateral asymmetries (e.g., Pomerantz, Sager, and Stoever, 1977; Wolff, 1971); indeed, in Gerstmann’s syndrome, where there is left parietal damage and consequent brain asymmetry, left-right mirror confusions are extreme (Critchley, 1953). Finally, it is unlikely that the confusion of laterally aligned lateral mirror images can have a totally different explanation from the confusion of vertical aligned vertical mirror images; yet Mach’s hypothesis would be relevant only to the confusion of lateral mirror images. If the symmetry of the body and brain cannot explain the perceptual similarity of lateral mirror images, what can? In light of the phylogenetic ubiquity of left-right confusion, the answer may lie in a consideration of the evolution of the vertebrate visual system (Gross and Bornstein, 1977). Presumably, the selective pressure of evolution made it advantageous for the visual system to be able to perform certain types of visual processing whereas other modes were irrelevant for survival. In the natural world there are rarely mirror images that would be useful for an animal to distinguish. Indeed with two exceptions there are virtually no mirror images at all. One exception is the two sides or profiles of a face or, more generally, the two sides of a bilaterally symmetrical animal. But here the two sides are two aspects of the same thing, and it would be more adaptive to treat them as the same - not to distinguish them. Another exception is that the silhouette of an object viewed from one side is the lateral mirror image of the silhouette of the same object viewed from the opposite side. Again it would be adaptive to
92 Marc H. Bornstein, Charles G. Gross and Joan Z. Wolf
treat as similar, not distinguish, these mirror images. In other words it is possible that the confusion of mirror images is not a “confusion” but an adaptive mode of processing visual information. In the natural world virtually the only mirror images that ever occur are aspects of the same thing and therefore need not be distinguished. It may be advantageous, therefore, to conceive of the difficulty of discriminating mirror images not as a “confusion” but as the perceptual similarity or equivalence of a stimulus and its 180” reflection around the vertical or horizontal axis. An implication of this evolutionary view is that the perceptual similarity of mirror images may be present early in life and may not require extensive experience or maturation. In the present study, therefore, we examined the perceptual similarity of mirror images in infancy. Our hypotheses were that infants would not discriminate mirror images of the same stimulus but that they would discriminate other and finer orientation differences. In Experiment I we tested this hypothesis with faces, stimuli that had inspired the original idea; in Experiments II and III we used line segments; and in Experiments IV and V we used geometric shapes.
EXPERIMENT
I: PROFILE
DISCRIMINATION
In the Introduction we suggested that since the only mirror images that commonly occur in the natural world are the two sides or profiles of another vertebrate and the obverse and reverse of a silhouette it would be more adaptive to equate than to discriminate them. For humans (and perhaps other animals) the two sides of a face are particularly significant mirror images. In Experiment I, we tested our hypothesis of the perceptual similarity of mirror images in a semi-realistic fashion by examining whether infants would treat the right profile and the left profile of the same person as perceptually equivalent. We predicted that babies would discriminate one person’s profile from another person’s profile but not the left and right profiles of the same person. Our results suggest that this is what infants do. Method Infants
Ten healthy, term infants participated in Experiment I. In order to obtain an N of 10, 12 babies were actually observed. One infant fretted, and one was eliminated on account of experimenter error. Table 1 gives vital statistics of the groups of infants studied in this and subsequent experiments; as may
Perceptual similarity of mirror images in infancy
93
be seen all groups are roughly comparable. The infants in all experiments were recruited by letter or phone from published birth announcements. Table 1.
Vital characteristics of the injknts in Experiments I-V
Group Number
N _-_-
1 1
1
5 6
Age (days)
-
Birth weight (kg) -----
M
F
Mean
S.D.
6
113.2
2.1
Experiment 3.60
1
4
11
11.1
Experiment 3.35
III
3.4
Experiment 3.81
IV
V
12 5
8 5
108.6 115.1
Mean
5 6 6 5
5 4 4 5
119.7 117.7 116.3 114.3
5.5 6.0 8.4 4.3
F.xpcrimcnt 3.44 3.22 3.44 3.58
5 6
5 5
115.1 115.7
3.1 4.0
Experiment 3.52 3.56
Birth length (cm) S.D.
Mean
S.D.
0.56
52.0
2.2
0.35
51.5
2.2
0.36
53.2
2.1
0.44 0.42 0.50 0.42
52.8 52.3 52.1 53.3
2.0 1.8 1.8 1.8
0.47 0.49
53.0 51.6
3.1 1.7
Apparatus
Each infant was seated in a standard infant chair approximately 60 cm from a matte-white stimulus panel, 91.5 cm X 45.7 cm, located in an observation room. The stimuli that they saw were profiles of faces of two males selected by adults as the least similar pair from a collection of male faces (Goldstein, Harmon and Lesk, 197 1). Slides of these profiles were projected through a one-way glass onto the stimulus panel by a Kodak Carousel projector (Model E12) located in an adjacent control room. The projected images, approximately 29 cm X 24 cm, subtended approximately 27.2” X 22.6” visual angle for the infants. The luminance of the stimulus was approximately 36 cd/m*, and the ambient light in the observation room was 20 fL. A signal lamp 7 mm in diameter was located centrally in the stimulus panel 3.5 cm above the infant’s eye level. The infant’s face and the projector lamp light were televised with a Panasonic TV camera (Model WV-241) whose lens was located in a 1.3 cm hole in the stimulus panel at the infant’s eye level. The video signal was displayed on a Panasonic TV monitor (Model TR-6220) to the experimenter(s) and parent(s) in an
94
Marc H. Bornstein, Charles G. Gross and Joan Z. Wolf
adjacent control room. The video signal was also recorded on a Panasonic VTR (Model NV-3 130). Approximately 60% of the 20.5 cm high monitor screen was filled with the infant’s head.
A session in Experiment I consisted of two phases: 1) a habituation phase during which infants were familiarized with the rightward-facing profile of one man (upper face, Figure 7, Goldstein clt al., 1971), and 2) a subsequent test pllasc in which each of the following stimuli was presented: the original profile and two additional ones, specifically the leftward-facing profile of the same man and the rightward-facing profile of a different man (lower face, Figure 7, Goldstein et al., 197 1). The familiarization or habituation phase in Experiment I consisted of one 60-see exposure to the right profile. In the test phase, the infants saw the three faces six times each. Each test trial was 10 set long. The schedule of presentation consisted of six triplets containing the three faces in different random orders. Orders of presentation were different for each infant. In both the habituation and test phases stimulus onset was contingent upon the infant’s forward looking. The average intertrial interval lasted approximately five to seven set; an average experimental session lasted approximately six min. The habituation-test design is based on the following rationale. If infants are exposed to a visual stimulus in an otherwise homogeneous visual environment, they will attend to that stimulus. If, however, the stimulus is presented continuously or repeatedly their visual attention to it will wane or habituate (Jeffrey and Cohen, 197 1; Kessen, Haith and Salapatek, 1970). (Habituation may represent the construction of some memory or internal representation of the stimulus.) Following habituation, presentation of a new or novel visual stimulus may elicit increased attention or dishabituation. Such dishabituation would provide evidence of the infant’s ability to discriminate the new stimulus from the original one. Duta Scoring
and Reduction
Infant looking time, the dependent measure, was judged from videotape records of the infant’s face and eyes. The camera actually photographed both the infant and the projector lamp, situated above and behind the baby; onset and offset of this lamp signaled trial onset and offset to the scorers. Interscorcr reliabilities in judging the looking of the ten infants in the study were quite high: X, = 0.96. Infants’ total looking times per trial were recorded by a digital timer-printer (Date1 DPP-7) to the nearest 0.01 sec.
Perceptual similarity of mirror images in infancy
95
Results and Discussion During the 60sec habituation phase, infants looked at the original right profile an average of 22.6 set, or 37.6% of the time. The range was 20.3% to 80.9%. Figure 1.
Experiment I. Mean percent of infant looking time at the original profile (RIGHT FAMILIAR), the mirror-image of the original profile (LEFT “FAMILIAR”), and the novel profile (RIGHT NEW) after familiarization with the original profile.
RIGHT FAMILIAR
LEFT “FAMILIAR”
RIGHT NEW
The mean percentage of time the infants looked at each of the profiles in the test phase is shown in Figure 1. The original rightward facing profile was looked at 38.9% of the time, the leftward profile of this same man was viewed 40.1% of the time, but the rightward profile of the new man was viewed 54.7% of the time. Correlated t tests indicated that infants who were habituated to the right profile failed to discriminate it from the left profile of the same face, t(9) = 0.29. Yet they easily discriminated the right profile of the new face from both profiles of the face they had seen in the habituation phase; new vs. right profile, t(9) = 2.9 1, p < 0.0 1; new vs. left profile, t(9) = 3.82, p < O.qOS. The finding that the babies looked much longer at the profile of the unfamiliar face than at the profile they had seen in the habituation phase indicated that the one-minute exposure in the habituation phase had been sufficient to familiarize them with the original stimulus. Babies remembered the original profile as evidenced by their inattention to it in the test phase; as has been shown before, infants’ recognition memory for faces is very good (Fagan, 1978). The fact that the infants looked at the leftward-facing profile (which they had never seen before) as much as at the rightward
96
Marc H. Bornstein, Charles G. Gross and Joan Z. Wolf
facing profile of the same face (which they had seen before) indicates that they treated the two profiles as equivalent. The infants, however, clearly discriminated the new face from the familiar one, independent of the orientation of the familiar face. That the infants treated the left and right profiles as equivalent is unlikely to have reflected an inability to discriminate any change in face orientation: Fagan (1976) has demonstrated that infants can discriminate smaller changes in orientation such as a full face from a threequarter view. Adults also tend to confuse or equate left and right mirror images of faces (Bartlett, 1932)‘. In summary, the results of Experiment I suggest that infants treat lateral mirror images of realistic stimuli as perceptually equivalent. This experiment was designed simply to demonstrate the mirror-image effect in young babies. It should be clear, however, that stimulus selection is key in such a demonstration, and it would be possible to choose profiles of different individuals which were not discriminable. In order to study mirror-image confusion further, we selected artificial stimuli which could be manipulated and controlled experimentally.
EXPERIMENT
II: DISCRIMINATION
OF LINE ORIENTATION
Several previous investigations have demonstrated that infants can discriminate change in the orientation of a stimulus (McGurk, 1972, 1974; McKenzie and Day, 1971; Watson, 1966; Wiener and Kagan, 1976). Experiment II used a habituation-test paradigm to assess the ability of three-month infants to discriminate a vertical line from a line tilted 45” from the vertical and to discriminate left and right 45” tilts (i.e., mirror-image obliques) from each other. The results show that the infants discriminated a 45” tilt from vertical but not the mirror-image obliques.
Method
Twenty healty, term infants participated in this study. Twenty-four infants were originally observed; four infants were eliminated for having ‘In a developmental psychology class, 31 students (mean age, 21.9 years) were asked to indicate which direction Washington’s profile on the U.S. quarter faces, to the viewer’s left or right. I:orty-two percent indicated rightward, and 58% indicated leftward which, analyzed by binomial expansion, did not differ from chance.
Perceptual similarity of mirror images in infancy 97
fallen asleep or become fretful during statistics of the experimental group.
the observation.
Table
1 gives vital
Apparatus
All infants were seated in a standard infant chair or in their mother’s lap approximately 61 cm from a stimulus panel. The panel consisted of a 42.5 cm X 32.0 cm translucent opal-glass screen mounted in the center of a 66.0 cm X 78.5 cm flat-black plywood board. The stimulus, a luminous line approximately 1.6 cm X 14.0 cm, was back-projected onto the glass during trials. For the infants, the line stimulus subtended approximately 1.5” X 12.5” on the white background which itself subtended approximately 30” X 40”. The luminance of the stimulus was approximately 10.6 cd/m*, and the ambient light in the observation room was approximately 20 fL. Procedure Experimental
Design
In general, the design of Experiment II followed that of Experiment I. Each infant was first shown a line oriented 45” to the right of vertical (the “standard” stimulus) for ten successive (habituation) trials and later tested with each of the following stimuli three times: a line oriented 45” to the right of vertical (the standard stimulus), a vertical line, and a line oriented 45” to the left of vertical. Each baby saw three randomly selected sets of the six possible permutations of these three stimuli. A two-set warning tone sounded just prior to the onset of each test trial. Both habituation and test trials were 10 set each, and the intertrial intervals were approximately 5 set; an average session lasted approximately 4.7 min. Data Scoring
and Reduction
Infant looking time, again the dependent variable, was judged in real time by a practiced, concealed observer who faced the infant and who was aware of stimulus onset and offset but was unaware of the orientation of the stimulus on each trial. Interscorer reliabilities under these conditions are high, 0.93 < r < 0.97 (Bornstein, Kessen and Weiskopf, 1976). Observer judgments, along with trial onset and offset, were recorded on an Esterline Angus event recorder. Total looking times (out of 10 set possible per trial) were reduced by a naive scorer from the records to the nearest 0.5 sec. Results and Discussion Infant looking time during the first (habituation) successive two-trial blocks. Looking time decreased
phase was averaged over from a mean of 48% of
98
Marc H. Bornstein,
Charles G. Gross and Joan Z. Wolf
the time the stimulus was available on trials 1 and 2 to a mean of 19% on trials 9 and 10. A repeated measures analysis of variance revealed that trials were a significant source of variance, t;(4,76) = 11.48, p < 0.001. Figure 2.
Experiment
II. Mean percent of infant looking time as a function of stimulus
orientation following habintation to a 45”-righ t oblique.
The mean percentage of time the infants looked at each of the three stimuli in the test phase is shown in Figure 2. The infants looked more at the vertical stimulus (35.2%) than at either of the obliques, t(19) = 3.09, I_’< 0.005 in comparison with the right oblique (25.9%) and t(19) = 2.15, p < 0.01 in comparison with the left oblique (27.1%), but they looked at the two obliques a similar proportion of the time, t( 19) = 0.11. The decline in looking during the habituation phase presumably reflected the infants’ increasing familiarity with the standard stimulus. The differential looking during the test phase suggests that the babies discriminated the vertical from the 45” obliques but not one oblique from its mirror image. Babies remembered the standard stimulus, as evidenced by their waning attention to it during the habituation phase and maintained inattention to it in the test phase. They treated the vertical stimulus but not the mirrorimage oblique as different from the original oblique. Under this interpretation, babies discriminated an angular displacement of 45” but not one of 90” when the two stimuli formed 45”-oblique mirror images. Human infants, like the other organisms cited above, appeared to treat mirror-image obliques as equivalent. A second interpretation of the results is possible, however. Looking during the test phase may have been independent of prior habituation and babies may not have discriminated orientation but simply preferred, as babies do in some situations (e.g., Bornstein, 1978), the vertical over the obliques. The
Perceptual similarity of mirror images in infancy 99
fact that the babies looked at the two obliques for a similar duration could have reflected their inability to discriminate any two oblique lines, not just 45” mirror-image obliques. Therefore in Experiment III, the infant’s ability to discriminate a 20”-oblique from a 70”oblique was investigated.
EXPERIMENT QUES
III:
DISCRIMINATION
OF NONMIRROR-IMAGE
OBLI-
Method
Ten healthy, term infants participated in this study. One additional infant was seen but eliminated when his mother inadvertently interrupted the observation. Table 1 gives vital statistics of the experimental group. Apparatus
All infants were seated in a standard infant chair approximately 96.5 cm from a matte-white stimulus panel, 76.2 cm X 91.4 cm, located in an observation room. A central window in the panel, 20.3 cm square, had been removed and replaced with a matte-white shutter. Stimulus plaques which could be fixed behind the shutter were exposed through this central window by manual removal of the shutter. The line stimuli, constructed of black tape approximately 1.8 cm X 20.3 cm, were mounted directly onto matte-white plaques. For the infants, the stimuli subtended approximately 1 .l” X 12.0” and were therefore comparable in visual angle to those used in Experiment II. A small signal lamp, attached to the back of the infant chair, was illuminated during the period the panel shutter was removed. The infant’s face and the signal lamp light were televised and recorded as in Experiment I. The video signal was also displayed on a Conrac TV monitor (Model CF 17A) to the experimenter behind the stimulus panel. Again, approximately 60% of the 20.5 cm high monitor screen was filled with the infant’s head. Ambiant light in the observation room was approximately 20 fL. Procedure Experimental
Design
The design of Experiment III was similar to that of Experiment II. During the habituation phase infants were shown one line stimulus oriented 20” to the right of vertical (the “standard” stimulus) for ten successive trials, and in
100
Marc H. Bornstein, Charles G. Gross and Joan Z. Wolf
the subsequent test phase a line stimulus 70” to the right of vertical was shown twice. An auditory prompt brought the infant’s attention to midline just prior to the onset of each trial, and as in Experiment II each trial in both the habituation and test phases lasted 10 sec. Intertrial intervals were approximately 5 set; an average experimental session lasted approximately 3.5 min. Data Scori~lg and Reduction
Total looking time, out of 10 set possible per trial, was judged from videotape records of the infant’s face as in Experiment I. The camera actually photographed both the infant and the signal lamp, situated above and behind the baby; onset and offset of this lamp signaled trial onset and offset to the scorers. Interscorer reliabilities in judging the looking of the ten infants in the study were again quite high: x, = 0.96.
Results and Discussion Infant looking time during the first (habituation) phase was averaged over successive two-trial blocks. Looking time decreased from a mean of 49% of the time the stimulus was available on trials 1 and 2 to a mean of 28% on trials 9 and IO. A repeated measures analysis of variance revealed that trials were a significant source of variance, t;(4,36) = 3.90, I_’< 0.01. Figure 3,
Experiment III. Mean percent of infant looking time as a function of stimulus orientation at the end of habituation to a 20”~right oblique. 501
/ 70’
RIGHT
The mean percentage of time the infants looked at the standard stimulus on trials 9 and 10 and at the test stimulus is shown in Figure 3. The infants looked more at the 70” stimulus (43.7%) than they had looked at the
Perceptual similarity of mirror images in infancy
10 1
standard 20” stimulus in the final two habituation trials, t(9) = 2.50, p < 0.01. Again, the decline in looking during the habituation phase presumably reflected the infants’ increasing familiarity with the standard stimulus while differential looking in the test phase indicated that the babies discriminated the line 70” from vertical from the standard at 20”. Babies clearly discriminate 50” of rotation between two obliques, as they had 45” of rotation between a line 45” left of vertical and a vertical line in Experiment II. Further support for this discriminative ability may be found in Wiener and Kagan (1976). Using a similar habituation-test design, they found that fivemonth infants distinguished 35” rotation from horizontal. In summary, Experiment III shows that babies can discriminate obliques differing by 50”. It suggests, therefore, that the babies in Experiment II were not responding simply on the basis of preference and probably did not habituate to the limited concept of “oblique” and simply generalize to another oblique. Together, Experiments II and III suggest that infants can discriminate rotational changes of 45”-50” between an oblique and an orthogonal or between two obliques; what they fail to discriminate are mirrorimage obliques.
EXPERIMENT
IV: DISCRIMINATION
OF LATERAL
MIRROR
IMAGES
In Experiment I, infants treated lateral mirror images of the same face profile as equivalent although they were able to discriminate profiles of different faces. In Experiment II, infants treated mirror-image obliques as equivalent although they could distinguish a 45” oblique from a vertical (Experiment II) and a 20” oblique from a 70” oblique (Experiment III). However, infants are known to be less sensitive to obliques than to orthogonals (the “oblique effect”, Appelle, 1972; Leehey, Moskowitz-Cook, Brill and Held, 1975; Taylor, 1963), and this may have contributed to the “confusion” of the mirror-image obliques. Therefore, in the next two experiments infants’ perception of the similarity of mirror images that were not obliques was examined. Furthermore, the 45” obliques in Experiment II were both lateral and vertical mirror images of each other and may have been confused for either or both reasons. In the present experiment the discrimination of lateral mirror images was studied, and in Experiment V the discrimination of vertical mirror images was studied. To provide converging evidence on the perceptual similarity of lateral mirror images, we used a different experimental paradigm from that of Experiments I, II, and III. Habituation of infants’ visual attention to repeat-
102
14 Marc H. Bornstein, Charles G. Gross and Joan Z. Wolf
ed stimuli is faster and more complete than is habituation to varied stimulation (e.g., Bornstein et al., 1976; Cornell, 1974; Fantz, 1964). This fact was used to study infants’ discrimination of 90” and 180” (lateral mirror-image) rotations. Four groups of children experienced different stimulus presentation conditions. For the first group a “standard” stimulus (C) was presented on every trial. The second group was shown the same standard stimulus (IT) on some trials and a 90” rotation of it (fl) on other trials. The third group was shown the standard (C) and the other 90” rotation (U). The fourth group was shown the standard (C) and its lateral mirror image (7). If the infants distinguished the rotations from the standard, then the amount of habituation should be greater in the first group than in the other three. To the extent that mirror-image stimuli are perceived as similar, as our hypothesis predicted, habituation in Group 4 should be similar to that in Group 1 and greater than that shown by Groups 2 and 3. The results suggested that babies find lateral mirror images much more similar than they do 90” rotations.
Method In fan ts
Four groups, each consisting of ten healthy, term infants, participated in Experiment IV. In order to obtain an N of 40, 55 infants were actually observed. Four infants failed to attend to the stimulus from the beginning of the experiment, nine fussed, and two were eliminated on account of experimenter error or equipment malfunction. Elimination of subjects across groups was nearly equal: three from Group 1, four from Group 2, three from Group 3, and five from Group 4. Table 1 gives vital statistics of the resulting experimental groups. Apparatus
The apparatus and recording arrangements used in Experiment IV were identical to those used in Experiment I. The stimuli were red luminous Cshapes (stem and legs 22 cm long and 4 cm wide); for the infant the C’s subtended approximately 21” on a side. The luminance of the stimulus was approximately 3.1 cd/m2, and the ambient light in the observation room was 20 fL. The infant’s face and the projector lamp light were televised with a TV camera as in Experiment I. Again, the video signal was recorded and displayed on a monitor to the experimenter(s) and the parent(s) in the control room.
Perceptual similarity of mirror images in infancy 103
Procedure Experimental
Design
Infants were randomly assigned to one of the four experimental groups. Each infant in each group saw 18 stimulus exposures. Infants in Group 1 saw a standard stimulus (C) on each of the 18 consecutive trials. Infants in Group 2 saw the standard on ten trials and the same shape rotated 90” to the right (n) on eight other trials intermixed with the standard-stimulus trials. Group 3 saw the standard stimulus on ten trials and the same shape rotated 90” left (U) on eight intermixed trials. Group 4 saw the standard stimulus on ten trials and its lateral mirror image (7) on eight intermixed trials. For all infants in Groups 1 - 4 the order of stimulus presentation followed the sequence: 11122 122 12 12 11122 1 where 1 = C, the standard stimulus, and 2 = C, n, Ll, or -J depending on the group. Each exposure was 10 set in duration, and each exposure was followed by an interstimulus interval whose duration was determined by the amount of time it took the infants to reorient to the panel (X = 5 - 7 set). Infants’ attention was redirected to the center of the stimulus panel before each trial by a blinking signal light. On three occasions, infants looked away before stimulus onset; data on these three trials (trial 2 for one baby and trial 3 for another in Group 2 and trial 2 for one baby in Group 3) were interpolated from adjacent trials. An average session lasted approximately six min. Data Scoring
and Reduction
Infant total looking time, out of 10 set possible per trial, was judged from videotape records of the infant’s face by scorers unaware of the stimulus orientation or infant group. The camera photographed both the infant and the projector lamp, situated above and behind the baby; onset and offset of this lamp signaled trial onset and offset to the scorers. Interscorer reliabilities in judging attention of 12 randomly selected infants to the stimuli in Experiment IV were quite high: x,= 0.93.
Results
and Discussion
The main purpose of this experiment was to ascertain whether or not differences in habituation might exist among groups who were shown the same stimulus (C) in the same order but for whom the context of that stimulus was varied. Since the ten trials on which the standard stimulus appeared were common for all groups they formed the basis of the analysis. Infants were assigned to the four groups in a random manner, and each group
104
Marc H. Bornstein, Charles G. Gross and Joan Z. Wolf
experienced the standard stimulus (C) on the first three trials. Each group should therefore have manifested similar amounts of looking on trials 1, 2, and 3. Unfortunately, this was not the case. Group 1 looked an average of 5.63 set; Group 2 looked 7.38 set; Group 3 looked 5.64 set; and Group 4 looked 5.73 sec. The main effect for Groups in an analysis of variance on these trials proved significant, F(3,36) = 3.16, p < 0.05, reflecting the fact that Group 2 looked more than each of the other groups. To facilitate comparisons among the babies, each baby’s data were converted to percentage scores using the initial three trials as the base. The converted scores formed the basis of all subsequent analysis. Figure 4.
Experiment IV. Mean percent decrement in infant looking time between the first three standard trials and the last three standard trialsfor each group. The standard stimulus and the context stimulus shown to each group are indicated on the abscissa.
The relative amounts of habituation shown by Groups 1 - 4 were assessed by comparing the percentage of looking time on the first three standard trials with that on the last three standard trials (14, 15, and 18) (see Figure 4). There was a decrease in looking at the standard stimulus from the initial three trials to the final three trials of 31.8% for the group seeing only the standard stimulus (Group 1) and 34.3% for the group seeing the standard and its mirror image (Group 4). Both decreases were significant by correlated t test, t(9) = 3.73, p < 0.005 and t(9) = 2.37, p < 0.01 for Groups 1 and 4, respectively. By contrast, neither of the groups which saw the standard and 90” rotations (Groups 2 and 3) showed reliable habituation from the initial to the final standard trials: Group 2 showed a decrease of 1 1.2% and Group 3 a decrease of 14.670, t(9) = 1.61 and t(9) = 1.73 for Groups 2 and 3, respectively. Thus, in a within-groups analysis, the babies treated repetition
Perceptual similarity of mirror images in infancy
105
of the standard stimulus (Group 1) similarly to repetition of the standard stimulus intermixed with its lateral mirror image (Group 4): both groups habituated. On the other hand, the babies responded in a different fashion to the standard stimulus intermixed with a 90” rotation in either direction (Groups 2 and 3): neither of these groups habituated. A between-groups analysis is made hazardous because of the groups’ unequal looking on the initial three trials in which all saw the same stimuli. However, it is suggestive that the degree of habituation did not differ between Groups 1 and 4, t(lS) = 0.15, nor did it differ between Groups 2 and 3, t(18) = 0.22. When, as a result of these analyses, Groups 1 and 4 were pooled and compared to pooled Groups 2 and 3, the former showed greater habituation than the latter, as predicted, t(38) = 1.82, p < 0.05. In summary, babies in Group 4 who saw the standard stimulus intermixed with its lateral mirror image on successive trials habituated to the standard stimulus by the final trials of the series as did the babies in Group 1 who saw only the standard stimulus. By contrast, both Groups 2 and 3 seeing the standard stimulus intermixed with its 90” rotations showed no reliable habituation to the standard stimulus. These results suggest that four-month babies view the standard stimulus and its lateral mirror image as perceptually equivalent, or more conservatively that the standard stimulus and its lateral mirror image were viewed as more similar to each other than either was to the 90” rotations. However, another interpretation of these results is possible. Each of the 90” rotations contained more vertical lines than either the standard or its lateral mirror image. Since infants may prefer to look at vertical lines (e.g., Bornstein, 1978), the absence of habituation in Groups 2 and 3 might have reflected the greater number of vertical lines in their stimuli, rather than greater discrimination of 90” rotations. In Experiment V we tested this possibility. Additionally, Experiment V was designed to assess vertical mirror-image discrimination by young infants.
EXPERIMENT
V: DISCRIMINATION
OF VERTICAL
MIRROR
IMAGES
In Experiment IV infants may have treated lateral mirror images as more similar to each other than the 90” rotations. Older children (Huttenlocher, 1967a, 1967b; Sekuler and Rosenblith, 1964; Wohlwill and Wiener, 1964), adults (Butler, 1964; Sekuler and Houlihan, 1964; Sekuler and Pierce, 1973; Wolff, 197 l), and infra-human animals (Lashley, 1938; Sutherland, 1961) also tend to confuse vertical or up-down mirror images, although usually less so than lateral mirror images. In Experiment V, discrimination of vertical
106
Marc H. Bomstein, Charles G. Gross and Joan Z. Wolf
mirror images was examined with a habituation procedure identical to that used in Experiment IV. One group (Group 5) saw a standard stimulus on every trial (D), and a second group (Group 6) saw the standard stimulus on some trials and a 180” rotation (vertical mirror image) of it (U) on other trials. If the infants perceived the standard and its vertical mirror image as similar, the two groups should show similar habituation. Furthermore, if infants, like older children and adults, find lateral mirror images more similar than vertical mirror images, then the amount of habituation under the different stimulus conditions in Experiments IV and V should fall in the following order of decreasing habituation: a) standard stimulus only (Groups 1 and 5), b) standard and lateral mirror image intermixed (Group 4), c) standard and vertical mirror image intermixed (Group 6), and d) standard and 90” rotations intermixed (Groups 2 and 3). This experiment also serves to test between the two interpretations of the results of Experiment IV. In that experiment the absence of reliable habituation in the groups that saw the 90” rotated stimuli could have been because they saw more verticals. If preference for verticals interferes with habituation, then both groups in the present experiment should show even less habituation because they saw even more verticals than the groups that received the 90” rotated stimuli in Experiment IV. Method Infants
Group 5 consisted of ten and Group 6 of eleven healthy, term infants. In order to obtain an 1%’of 2 1, 25 infants were actually observed. Two were eliminated from Group 5 (one fussed, and one failed to attend to the stimulus from the beginning of the experiment), and two were eliminated from Group 6 (one failed to attend to the stimulus from the beginning of the experiment, and one’s mother inadvertently interrupted the observation). Table 1 gives vital statistics of the resulting experimental groups. Apparatus
The apparatus, stimuli, recording, and data-collection arrangements in Experiment V were identical to those used in Experiment IV.
used
Procedure Experimental
Design
Groups 5 and 6 were run sequentially and after the completion of Experiment IV. The experimental design was identical to that of Experiment IV.
Perceptual similarity of mirror images in infancy
107
Infants in Group 5 saw a standard (n) on each of 18 consecutive trials. Infants in Group 6 saw the standard on ten trials and its vertical mirror image (U) on eight intermixed trials in the sequence used in Experiment IV (111221221212111221), where 1 =n, the standard stimulus, and 2 = U. Data Scoring
and Reduction
The data were scored and reduced as in Experiment IV. Interscorer reliabilities in judging the attention of five randomly selected infants in this experiment were again high, X, = 0.95.
Results and Discussion On the initial three trials both groups saw only the standard stimulus (n). Group 5 looked an average of 7.2 1 set, and Group 6 looked 6.82 sec. This difference was not significant, t( 19) = 0.46. Their looking durations on these initial trials (x = 7.01 set) resembled that of Group 2 in Experiment IV but w_ere significantly greater than that of the other groups in Experiment IV (X = 5.68 set), t(49) = 2.84, p < 0.01 (two-tailed), making comparison of the groups in Experiments IV and V difficult. As in Experiment IV, the data for each child in Experiment V were converted to percentage scores using the initial trials as the base, and habituation was assessed by comparing the percentage of looking time on the first three standard trials with that on the last three standard trials (14, 15, and 18). There was a decrease in looking from the initial three standard trials to the final three standard trials of 25.3% for Group 5 and 22.3% for Group 6 (see Figure 5). Both decreases were significant by correlated t tests, t(9) = 2.78, p < 0.01 and t(l0) = 2.58, p < 0.01 for Groups 5 and 6, respectively. Although the habituation shown by Group 6, which saw the standard intermixed with its vertical mirror image, was slightly less than that shown by Group 5, which saw only the standard, this difference did not approach significance; an independent t test showed t( 19) = 0.2 1. These results enable us to reject one of the two alternative interpretations of Experiment IV, namely that Groups 2 and 3 showed no significant habituation in contrast to Groups 1 and 4 because the stimuli for Groups 2 and 3 contained more vertical lines than the stimuli shown to Groups 1 and 4. Both groups in the present experiment showed reliable habituation although they saw even more verticals than Groups 2 and 3 in Experiment IV. Thus the absence of reliable habituation of Groups 2 and 3 could not have been because they saw more verticals. We can therefore conclude that the findings in Experiment IV indicate that a 180” rotation around the
108
Marc H. Bornstein, Charles G. Gross and Joan Z. Wolf
Figure 5.
Experiment V. Mean percent decrement in infant looking time between the first three standard trials and the last three standard trialsfor each group. The standard stimulus and the context stimulus shown to each group are indicat-
STANDARD; CONTEXT: GROUP:
5
6
vertical meridian is perceived as less of a stimulus change than a 90” rotation. These results support the interpretation that four-month babies viewed the standard stimulus and its lateral mirror image as perceptually similar. The main findings of the present experiment were that the group that saw only the standard stimulus (Group 5) and the group that saw the standard intermixed with its vertical mirror image (Group 6) showed similar habituation. This result suggests that babies find vertical mirror images as well as lateral ones perceptually similar. This interpretation is strengthened by the results from Groups 2 and 3 who, under identical conditions, failed to show reliable habituation when their standard stimulus was intermixed with its 90” rotation. Further support for this result is McGurk (1972). Using a habituation-test design like that used in Experiment II, McGurk also found that three-month infants failed to discriminate vertical mirror images, though six-, nine-, and twelve-month infants did. Again, a 180” rotation producing a mirror image was perceived as less of a stimulus change than a 90” rotation. Do four-month babies perceive lateral mirror images to be more similar than vertical mirror images as had often been reported for older children and adults? The group shown the vertical mirror images (Group 6) decreased 65% of the amount of decrease in looking produced by the group shown the lateral mirror images (Group 4), but the difference in decrements between the two groups was not significant, t( 19) = 0.76. However, as may be seen by comparing Figures 4 and 5, the order of decreasing habituation among the groups was, as predicted, a) standard stimulus only, b) standard and lateral mirror image, c) standard and vertical mirror image, and d) standard and 90” rotations. Applying Jonckheere’s (1954) test of order alternatives, the proba-
Perceptual similan’tyof mirror images in infancy
109
bility that this predicted order could have been obtained by change is quite low, z = 1.84, p = 0.03. In summary, human infants treated vertical mirror images as similar but the equivalence was somewhat weaker than that found for lateral mirror images. As Goldmeier (1972) observed so long ago, shapes which are symmetrical about a vertical axis, lateral mirror images, are judged more similar than shapes which are symmetrical about a horizontal axis, vertical mirror images.
General Discussion As described in the Introduction, a wide variety of species tends to confuse lateral mirror images and, to a lesser degree, vertical mirror images, although they can discriminate other orientation changes. We proposed that actually represents an adaptive perceptual equivalence this “confusion” because in nature mirror images tend to be aspects of the same object. An implication of the argument that the visual system naturally treats mirror images equivalently is that mirror images should be treated as perceptually equivalent near the beginning of life. Five experiments using infants three to four months of age were designed to test aspects of this hypothesis. Experiment I tested the original ecological hypothesis with realistic stimuli, faces. This study showed that babies discriminate faces but, as predicted, treat the left and right profiles of the same person as equivalent. Experiments II - V used lines or geometric forms to examine the hypothesis. Experiment II tested the ability of infants to discriminate a 45”-oblique line from its mirror image and from a vertical line. The infants discriminated the lines differing by 45” but failed to discriminate the mirror images, although the orientation of the latter differed by 90”. This discrimination failure could not have been ascribable to an inability to discriminate all obliques because in Experiment III, infants distinguished a 70” oblique from a 20” oblique. Furthermore, the results of Experiment IV indicated that mirror image similarity in infants is not confined to mirror-image obliques. In that experiment infants discriminated non-oblique stimuli differing in orientation by 90” but failed to discriminate lateral mirror images that differed by 180”. The combined results of the first four experiments indicate that babies as young as four months of age discriminate orientation change but perceptually equate lateral mirror images. The results of Experiment V showed that babies also confuse vertical mirror images but slightly less so than lateral ones. Of course, it is possible or even likely that infants, like older children (Hendrickson and Muehl, 1962; Jeffrey, 1958), could be trained to discrimi-
110
22 Marc H. Bornstein, Charles G. Gross and Joan Z. Wolf
nate mirror images. However, our results demonstrate that they naturally treat mirror images as more similar to each other than they do structurally identical, non-mirror stimuli differing less in orientation change. A common explanation of left-right confusion is based on the bilateral symmetry of the brain and body of the perceiving organism (Corballis and Beale, 1976; Mach, 1914; Noble, 1968; Orton, 1937). According to this view, left-right mirror-symmetric representations are confusing until some behavioral or other bodily asymmetry develops. Yet human infants are not bilaterally symmetrical. Behaviorally, they regularly tend to favor one side (Bresson, Maury, Pieraut-Le Bonniec and de Schonen, 1977; Caplan and Kinsbourne, 1976; Gardner, Lewkowicz, and Turkewitz, 1976; Glanville, Best and Levenson, 1977; Turkewitz, Gordon and Birch, 1965); anatomically humans are born with asymmetrical brains (Wada, Clarke and Hamm, 1975); and infants show electrographic asymmetries between the two hemispheres (Davis and Wada, 1977; Molfese, Freeman and Palermo, 1975). The bilateral-symmetry explanation of lateral mirror-image “confusion” is inappropriate for infants as it is for adults or other organisms (Gross and BomIn our view, the prevalence of left-right stein, 1978; see Introduction). equivalence reflects an adaptive mode of visual information processing. Lateral mirror images are strongly equivalent perceptually because in the natural world they virtually always represent twin aspects of the same object or organism. We propose that lateral mirror-image equivalence reflects bilateral symmetry, not of the perceiving organism, but of significant objects, particularly other organisms in the perceptual world. Vertical mirror images have also been found to be confusing by animals, human children, and human adults (Butler, 1964; Huttenlocher, 1967a, 1967b; Lashley, 1938; Sekuler and Houlihan, 1964; Sekuler and Pierce, 1973; Sekuler and Rosenblith, 1964; Sutherland, 1961; Wohlwill and Wiener, 1964; Wolff, 197 1); the results of Experiment V indicate that babies too tend to treat vertical mirror images as similar, The bilateral-symmetryof-the-body explanation does not explain the confusion of vertical mirror images, and it is unparsimonious to assume that vertical mirror-image confusion could have a totally different explanation from lateral mirror-image confusion. We suggest that vertical mirror images are treated as similar for the same reason as lateral mirror images: when vertical mirror images occur in the natural world they are usually aspects of the same object. However, it may be that lateral mirror equivalence is primary because lateral mirror images are more common (e.g., as two sides of a bilaterally symmetrical organism or two views of a silhouette). Vertical mirror images have been usually found to be somewhat less confusing than lateral ones (e.g., Bradshaw rt ul., 1976: Butler, 1964;
Perceptual similarity of mirror images in infancy
11 I
Huttenlocher, 1967a, 1967b; Sekuler and Rosenblith, 1964), and we found a similar tendency. The lesser similarity or confusion of vertical mirror images may reflect their derivative nature as suggested above or simply the availability of additional cues (directly or indirectly related to gravity) that are not available for left-right discrimination. A paradox remains. Throughout this paper and indeed throughout the related literature, “mirror image” is used to refer exclusively to lateral and vertical mirror images. Yet, lateral and vertical mirror images may be viewed as a special class of mirror images, namely those produced by rotation about an orthogonal axis. An infinite number of other “mirror images” may be produced by rotations about other axes. For example, horizontal and vertical line segments can be described as mirror images about a 45” axis. These other “mirror images” also rarely need to be distinguished in nature, yet they are not especially confused in discrimination tasks. As many have noted in a variety of contexts, there is something very special in perception about orthogonal orientations (Appelle, 1972; Arnheim, 1974; Bornstein, 1978; Gibson, 1966; Howard and Templeton, 1966; Olson, 1970; Pick, Yonas and Rieser, 1978). Presumably this pervasive phenomenon reflects the orthogonal orientation of our world (horizons, gravity, etc.) and in turn must be reflected in some uniqueness of the neural mechanisms that process orthogonal information. Our view of the perceptual similarity of mirror images has implications for two more general problems of development. The first is the ontogeny of the concept of the object in early infancy; the second is the development of reading skills. A common view among developmentalists has been that through perceptual experience objects in different perspectives and at different distances come to be seen as the same object (Gibson, Gibson, Pick and Osser, 1962; Oyama and Sato, 1975). But is this process wholly experiential? Since infants four months of age treat mirror images as equivalent, we would argue that left-right equivalence predisposes babies toward perceiving visual objects as invariant. Though more complex constancies of shape or size surely require considerable experience to attain a mature status, objects whose lateral halves are mirror images, like the face, might engage a very early mode of constancy perception. This primitive form of the object concept might be an incipient sign of stability amid the perceptual flux that is an infant’s visual world. So, for example, older infants will encode the face qua face rather than as a specific pattern (e.g., Cohen, 1977; Cornell, 1974; Dirks and Gibson, 1977 ; Fagan, 1978). In this way, mirror-image equivalence may serve as a core mechanism or Anlage on which more complex constanties are later built. It is possible that the perceptual invariance of the two
112
Marc H. Bornstein, Charles G. Gross and Joan Z. Wolf
halves of the face (and body) underlies the normal child’s acquisition of person constancy. In this light, Bell’s (1970) finding that person constancy develops prior to the constancy of inanimate objects is not surprising. In summary, mirror-image equivalence may form an important basis for person constancy, which in turn may be elaborated into more complex constancies of form and object permanence. Throughout this paper we have stressed the absence of necessity to discriminate mirror images in the natural world. But of course, in the unnatural, man-made world, discrimination of mirror images is crucial: it is a prerequisite to literacy. Our orthography is plagued by mirror images, and consistent left to right scanning is crucial to reading. In learning to read and write, letter reversals (e.g., h for d or I_’for y), word reversals (e.g., otz for HO), and the failure to progress consistently from left to right represent common errors for the normal child (Gibson and Levin, 1975; Orton, 1937). For example, Davidson (1935) found that 77.5% of kindergarten and first-grade children “confused” the lateral mirror images, b with (I and p with y.We suggest that letter-reversal in reading may reflect the normal child’s difficulty in overcoming a nativistic mode of visual processing. Mirror reversal of letters and other left-right difficulties have been reported to be particularly common among many children (so-called “developmental dyslexics”) who have severe difficulty in learning to read for no 1963; Orton, 1937 ; Shankweiler, known cause (Benton, 1975; Money, 1963). These reversal problems may then reflect an especial difficulty in learning to overcome the otherwise normal inclination to equate mirror dyslexics may show particularly images, that is, some developmental “strong” mirror-image equivalence. If this view is correct, this subgroup of dyslexics should demonstrate a higher incidence of mirror-image equivalence for arbitrary patterns as well as for letters. Just as mirror-image equivalence should interfere with reading acquisition, learning to read should facilitate discrimination of mirror images. Support for this possibility comes from Rude1 and Teuber’s (1963) study of U.S. children and Serpell’s (1971) study of urban Zambian children. In both cases, the greatest improvement in mirror-image discrimination occurred at the approximate age of initial reading and writing instruction, between 5% and 6% years for the U.S. children and between 7% and 10% years for the Zambians. We would predict that non-literate adults or even adults literate in languages devoid of orthographic mirror images would show greater mirror-image confusion than adults literate in a Western orthography (see Gross and Bornstein, 1978, Figure 3; Shapiro, 1970).
Perceptual similarity of mirror images in infancy
113
References Allman,
J. M., and Kaas, J. H. (1975) The dorsomedial cortical visual area: A third tier area in the occipital lobe of the owl monkey. Brain Res., 100, 473487. Appelle, S. (1972) Perception and discrimination as a function of stimulus orientation: The “oblique effect” in man and animals. Psychol. Bull., 78, 266-278. Arnheim, R. (1974) Art and Visual Perception. Berkely, CA, University of California Press. Bartlett, I’. C. (1932) Remembering: A Study in Experimental and Social Psychology. Cambridge, Cambridge University Press. Bell, S. M. (1970) The development of the concept of object as related to infant-mother attachment. Child Dev., 41, 291-311. Benton, A. L. (1975) Developmental dyslexia: Neurological aspects. In W. J. Friedlander (Ed.), Advances in Neurology (Vol. 7). New York, Raven. Bornstein, M. H. (1978) Visual behavior of the young human infant: Relationships between chromatic and spatial perception and the activity of underlying brain mechanisms. J. exp. Child Psychoi., in press. Bornstein, M. H., Kesscn, W., and Weiskopf, S. (1976) Color vision and hue categorization in young human infants. J. exp. Psychol.: Hum. Percept. Perform., 2, 115-l 29. Bradshaw, J., Bradley, D., and Patterson, K. (1976) The perception and identification of mirrorreversed patterns. Q. J. exp. Psychol., 28, 221-246. Brcsson, I:., Maury, L., Pieraut-Le Bonniec, G., and de Schonen, S. (1977) Organization and lateralization of reaching in infants: An instance of asymmetric functions in hands collaboration. Neuropsychol., 15, 311-320. Brooks, B., and Jung, R. (197 3) Neuronal physiology of the visual cortex. In R. Jung (Ed.), Handbook ofSensory Physiology (Vol VII/3B). Berlin, Springer. Butler, J. (1964) Visual discrimination of shape by humans. Q., J. exp. Psychol., 16, 272-276. Caplan, P. J., and Kinsbourne, M. (1976) Baby drops the rattle: Asymmetry of duration of grasp by infants. Child Dev., 47, 532-534. Cohen, 1. B. (1977) Concepf acquisition in the human infant. Paper presented at the Society for Research in Child Development, New Orleans, Louisiana. Corballis, M. C., and Beale, I. L. (1976) The Psychology of Left and Right. New York, Halstead. Corballis, M. C., Miller, A., and Morgan, M. J. (1971) The role of left-right orientation in interhemispheric matching of visual information. Percept. & Psychophys., IO, 385-388. Cornell, E. 11. (1974) Infants’ discrimination of photographs of faces following redundant presentations. J. exp. Child Psychol., 18, 98-106. Critchley, M. (1953) The Parietal Lobes. London, Arnold. Davidson, H. P. (1935) A study of the confusing letters b, d, p, and q. J. genet. Psycho!., 47, 458468. Davis, A. E., and Wada, J. A. (1977) Hemispheric asymmetries in human infants: Spectral analysis of flash and click evoked potentials. Brain and Lang., 4, 23-31. Dirks, J., and Gibson, E. J. (1977) Infants’ perception of similarity between live people and their photographs. Child Dev., 48, 124-l 30. Fagan, J. F. (1976) Infants’ recognition of invariant features of faces. Child Dev., 47. 627-638. Pagan, J. F. (1978) The origins of facial pattern recognition. In M. H. Bornstein and W. Kesscn (Eds.), Psychological Development from Infancy. Hillsdale, N.J., Erlbaum. Fantz, R. L. (1964) Visual experience in infants: Decreased attention to familiar patterns relative to novel ones. Science, 146, 668-670. Carder, J., Lewkowicz, D., and Turkewitz, A. (1977) Development of postural asymmetry in premature human infants. Dev. Psychobio., 10, 471480.
114
Gibson,
Marc H. Bornstein, Charles G. Gross and Joan Z. Wolf
I:. J., Gibson, J. J., Pick, A. D., and Osscr, II. (1962) A developmental study of the discrimination of letter-like forms. J. camp. physiol. Psych&, 55, 897-906. Gibson, E. J., and Levin, Il. (1975) The Psychology of Reading. Cambridge, MIT Press. Gibson, J. J. (1966) The Senses Consideredas PerceptualSysrerns. Boston, Iloughton Mifflin. Glanvillc, 8. B., Best, C. T., and Levenson, R. (1977) A cardiac measure of cerebral asymmetries in infant auditory perception. Dev. Psychol., 13, 54-59. Goldmcier, I-. (1972) Similarity in visually pcrceivcd forms. Psychol. Iss., 8, l-l 29 (Monograph No. 29). Goldstein, A. J., Harmon, L. D.. and Lesk, A. B. (1971) Identification of human faces. Proc. IEEE, 59, 748-760. Gross, <‘. G.. and Bornstcin, M. Il. (1978) Left and right in science and art. Leonardo, II, 29-38. Hamilton, C. R., and Ticman, S. B. (1973) Interocular transfer of mirror image discriminations by chiasm-sectioned monkeys. Brain Res., 64, 241-255. Ilcndrickson, L. N., and Mu&l, S. (1962) The effect of attention and motor response pretraining on learning to discriminate b and d in kindcrgartcn children. J. Ed. P.&ml., 53, 236-241. lloward. I. I’., and Templeton, W. B. (1966) Human Spatial Orienfufion. N.Y., Wiley. Flubel, D. II., and Wicsel, T. N. (1968) Receptive fields and functional architccturc of monkey striate cortex. J. Physiol., 195. 21 S-243. lluttenlocher, J. (1967a) Discrimination of figure orientation: Effects of relative position. J. camp. physiol. PsychoI., 63, 359-361. lluttcnlocher, J. (1967b) Children’s ability to orient and order objects. Child Dev., 38, 116991176. Jeffrey, W. I<. (1958) Variables in early discrimination learning: I Motor responses in the training of a Icft-right discrimination. Child Dev., 29. 269-275. Jeffrey, W. I:., and Cohen, L. B. (1971) ltabituation in the human infant. In 13. Reese (Ed.), Advances in Child Development arld Behavior (Vol. 6). New York, Academic Press. Jonckheere, A. R. (1954) A distribution-free k-sample test against ordered alternatives. Riometrika, 41, 133-145. Kcssen, W., Ilaith, M. M., and Salapatek, I’. Il. (1970) Human infxrcy: A bibliography and guide. In I’. 11. Mussen (Fd.), Carmichael’s Manual of’ Child Psychology. New York, Wiley. Lashley. K. S. (1938) The mechanism of vision: XV. Preliminary studies of the rat’s capacity for detail vision. J. ,qcn. I’sychol. IX, 123-I 93. Lcehcy, S. C., MoskowitzCook, A., Brill. S., and lleld, R. (1975) Orientation anisotropy in infant vision. Scictzce, 190. 900-902. Lehman, R. A. W., and Spcnccr, D. D. (1973) Mirror-image shape discrimination: Interocular reversal of response in the optic chiasm sectioned monkey. Bruin Rex, 52, 233-241. Mc(;urk, II. (1972) Infant discrimination of orientation. J. exp. C’hildPsychol., 14, 151-164. Mc(;urk, t1. (1974) Visual perception in young infants. In B. IToss (Pd.), New Perspectives in Child Dcvclopmerlt. Baltimore, Md., Penguin Books. McKenk, B., and Day, R. l-1. (1971) Orientation discrimination in infants: A comparison of visual fixation and operant training methods. J. exp. Child Psycho!., II, 366-375. Mnch, 1,. (19 14) The Arral.vsis of‘Serxations arld the Relation of‘ the PIlysical to the Psychical. Chicago, Open Court. blolfcse, I). I.., I rceman Jr., K. B., and Palermo, D. S. (1975) The ontogeny of brain latcruliflation for speech and nonspccch stimuli. &a+ arld Larzg., -7, 3566368. Money, J. (l:d.) (I 962) Readirig Disability. Baltimore, Johns flopkins Press. Noble, J. (1968) Paradoxical interocular transfer of mirror-image discriminations in the optic chiam scctioncd monkey. Braif Rcs., IO, 127 -15 1. Olson. 1). Ii. ( 1970) Cog?ritivc Dcw/opmo~f: The Child’s Acquisition of Diagonality. N.Y ., Academic I’rcss.
Perceptual similatity of mirror images in infancy
115
Orton, S. T. (1937) Reading, Writing, and Speech Problems in Children. New York, Norton. Over, R., and Over, .I. (1967) Detection and recognition of mirror-image obliques by young children. J. camp. physiol. Psychol., 64, 467470. Oyama, T., and Sato, K. (1975) Relative similarity of rotated and reversed figures to the original figures as a function of children’s age. J. camp. physiol. Psychol., 88, 110-117. Pick, Jr., H. L., Yonas, A., and Rieser, J. (1978) Spatial reference systems in perceptual development. In M. H. Bornstein & W. Kessen (t‘:ds.), Psychological Development from Infancy. Hillsdale, N. J., Erlbaum. Pomcrantz, J. R., Sager, L. C., and Stoevcr, R. J. (1977) Perception of wholes and of their component parts: Some configural superiority effects. J. exp. Psychol: Hum. Percept. Perform., 3, 422435. Rude& R. G., and Teuber, H.-L. (1963) Discrimination of direction of line in children. J. camp. physiol. Psychol., 56, 892-898. Sekuler, R. W., and Houlihan, K. (1968) Discrimination of mirror-images: Choice time analysis of human adult performance. Quart. J. exp. Psychol., 20, 204&207. Sekuler, R., and Pierce, S. (1964) Perception of stimulus direction: Hemispheric homology and laterality. Am. J. Psychol., 86, 679495. Sekuler, R. W., and Roscnblith, J. I;. (1964) Discrimination of direction of line and the effect of stimulus alignment. Psychon. Sci., I, 143-144. Serpcll, R. (1971) Discrimination of orientation by Zambian children. J. camp. physiol. Psychol., 75, 312-316. Shankweiler, D. P. (1963) A study of developmental dyslexia. Neuropsychol., I, 267-286. Shapiro, M. (1970) On some problems in the semiotics of visual art: Field and vehicle in image-signs. In A. J. Creimas and R. Jakobson (Eds.), Sign, Language, Culture. The Hague, Mouton. Stordndt, M. (1974) Recognition across visual fields with mirror-image stimuli. Percept. Mot. Skills, 39, 762. Sutherland, N. S. (1961) The methods and findings of experiments on the visual discrimination of shape by animals. Exp. Psychol. Sot. Mono., Whole No. 1. Taylor, M. M. (1963) Visual discrimination and orientation. J. Opt. Sot. Am., 53, 763-765. Tee, K. S., and Riesen, A. II. (1974) Visual right-left confusions in animal and man. In G. Newton and A. II. Riesen (Eds.), Advances in Psychobiology (Vol. 2). New York, Wiley. Turkewitz, G., Gordon, E. W., and Birch, H. G. (1965) Headturning in the human neonate: Spontaneous patterns. J. genef. PsychoI., 107, 143-158. Vogel, J. M. (1977) Getting letters straight: Bases for children’s mirror-image confusions. Paper presented at the Society for Research in Child Development, New Orleans, Louisiana. Wada, J., Clarke, R., and Hamm, A. (1975) Cerebral hemispheric asymmetry in humans. Arch. Neurol., 32, 239-246. Watson, J. S. (1966) Perception of object orientation in infants. Merrill-Palmer Quar?., 12, 73-94. Wiener, K., and Kagan, J. (1976) Infants’ reaction to changes in orientation of figure and frame. Perception, 5. 25-28. Wohlwill, J. F., and Wiener, M. (1964) Discrimination of form orientation in young children. Child Dev.. 35, 1113-1125. Wolff, P. (1971) Mirror-image confusability in adults. J. exp. Psychol., 91, 268-272. Ze$, S. M., and Sandeman, D. R. (1976) Combined anatomical and elcctrophysiological studies on the boundary between the second and third visual areas of rhesus monkey cortex. Proc. Royal Sot. London, 194, 555-562.
116
Marc H. Bomstein, Charles C. Gross and Joan Z. Wolf
On a 6tudie la perception d’images en miroir par des bib&s de 3 i 4 mois au tours de 5 experiences oi ont 6th utilisfs des paradigmes d’habituation. Dans une premiere experience, les bebds discriminent les profils droits de deux visages diffdrents mais ne discriminent pas le profile droit et le profil gauche d’un mEme visage. Dans une dcuxieme experience, les b&b& discriminent une oblique a 45” d’unc verticale mais pas une oblique de son image cn miroir. Dans une troisieme experience, Ies b&b&s discriminent dcs lignes obliques qui different de 50” et nc sont pas des images en miroir. Enfin dans les dcrnieres cxperienccs, Its rotations a 90” d’une forme sont discriminees, mais pas lcs rotations a 180” donnant cn miroir des images la&ales ou vcrticales. Ccs rkdtats indiquent que si les b&b&s sont capablcs de discriminer des diffdrences d’orientation (m&me parmi les obliques) ils tcndent a voir Its images cn miroir, particuliircment les images en miroir la&ales, comme des stimuli equivalents. Nous proposons que I’equivalcnce perceptuelle des images en miroir rcflete un mode d’adaptation de la procedure visuelk, Its images en miroir dans la nature sont toujours des aspects dun meme objet ct ndcessitcnt pas d’etre discriminees. On discute Ies relations de similarite perceptuclle des images en miroir pour I’ontogcni‘sc du concept d’objet et le ddveloppement dc la lecture.
Cognition, @Elsevier
6 (1978) 117-133 Sequoia S.X., Lausanne
2 - Printed
in the Netherlands
Parallel function strategy in pronoun assignment *
ELLEN H. GROBER WILLIAM BEARDSLEY ALFONSO CARAMAZZA The Johns Hopkins University
Subjects completed sentences of the form NPl aux V NP2 because (but) Pro . . . (e.g., John may scold Bill because he . ..) with a reason or motive for the action described. A basic perceptual strategy was hbpothesized to underlie the comprehension of these sentences which have a potentially ambiguous pronoun in the subject position of the subordinate clause. It was expected that listeners would interpret the pronoun as being coreferential with the subject NP of the main clause, the NP with the same grammatical function. While this strategy accounted for the major share of the results, semantic factors restricted its use, establishing an interpretation in which the pronoun was coreferential with the object NP of the main clause. The factors that influence the assignment of anaphoric pronouns to appropriate referents is a problem of considerable proportions for any model of language comprehension (Quillian, 1968; Winograd, 1972). In many cases the basis of the assignment is quite obvious, as when a pronoun is matched to an available antecedent by features marked explicitly in the surface structure of the sentence. For example, in the sentence: (1)
Mary praised the man because he was courteous.
the pronoun he is assigned to the antecedent word man because of the critical feature (male) that is common to them. In other cases, the basis of the pronoun assignment is not so obvious. Consider a situation in which the pronoun can be coreferential with either the first or second noun phrase: *Part of this 0851 to Johns an earlier draft gy, The Johns
research was supported by Biomedical Science support grant number 3505 RR07041Hopkins University. We would like to thank Rita Bemdt for her helpful comments on of this paper. Address reprint requests to Alfonso Caramazza, Department of PsycholoHopkins University, Baltimore, Maryland 21218.
118
I?. H. Grober, W. Beardsley and A. Caramazza
(2)
George telephoned
(3)
George criticized
Walter because he wanted some information. Walter because he misplaced
the file.
Although these sentences have at least two underlying representations, subjects seem to prefer one reading over another (Garvey, Caramazza and Yates, 1975). Garvey and Caramazza (1974) suggest that a property of some verb roots, direction of causality, is involved in the process of pronoun assignment in these latter sentences. This feature, implicit in certain verbs, imputes the cause of an event or situation either to the subject or object of the clause in which it serves as a main verb. Thus, the causal feature of the verb telephone in sentence (2) establishes a preferred interpretation in which the pronoun he is coreferential with the first noun phrase (NPl = George) while the causal feature of the verb criticize in sentence (3) biases the selection towards the second noun phrase (NP2 = Walter) as the antecedent of the pronoun. The direction of causality for a set of verbs was assessed by a sentence completion task that required subjects to provide a reason or motive for the action described in sentence fragments of the form NPI V NP2 because pro . . (Garvey et al., 1975). Verbs produced a consistent bias in the direction of pronoun assignment ranging from verbs which strongly determine the choice to the first noun phrase (NPl), e.g., the verb apologize, to verbs which strongly restrict selection to the second noun phrase (NP2), e.g., the verb fear. These results can be rendered more meaningful when we consider that part of what we understand about apologizing includes the knowledge that the motive arises primarily from within the person making the apology, thereby biasing the choice towards NP 1. In other words, part of the meaning of apologize is the presupposition that the person performing the act of apologizing is responsible for some prior negative action in relation to the person to whom he/she is apologizing. This is not to say, however, that the reasons for apologizing are part of the meaning of apologize. In contrast to the presuppositions of apologizing, what we understand about fearing includes the knowledge that there exists a threat of impending danger outside the fearer, thereby biasing the choice towards NP2. Not only does the implicit causality feature determine the direction of pronoun assignment in a sentence but it also facilitates the choice of an appropriate antecedent in a timed comprehension task (Caramazza, Grober, Garvey and Yates, 1977). Subjects were required to decide the coreferentiality of a pronoun in pairs of sentences such as John telephoned Bill because he withheld some information/wanted some information. Verbs were first empirically classified into those that bias assignment towards the tirst noun phrase of the main clause and those that bias assignment towards the second noun phrase. Pairs of
Parallelfunction strategy in pronoun assignment
119
sentences were constructed for each verb such that the subordinate clause in one sentence established a reading consistent with the natural bias of the verb while the others established a reading inconsistent with the bias of the verb. Time to respond was faster for the congruent sentences. This was also true for control sentences such as Sue telephoned Bill because he withheld some information in which gender differences eliminated all potential ambiguities. The results, considered together, indicate that subjects regularly make use of implicit causality relations marked by verbs in determining the selection of antecedents for ambiguous pronouns. Moreover, the influence of the implicit causality feature can be modulated by other linguistic variables (Garvey et al., 1976). For example, passivization, which reverses the surface order of the logical subject and object of the sentence, resulted in a general drift in the assignment of pronoun antecedents toward the grammatical subject of the sentence. That is, NP2 type verbs (e.g., criticize) in the active voice maintain the deep structure NP2 antecedent in the passive voice although the surface structure assignment is NPl. This can be seen in sentences (4) and (5). (4)
The director
criticized
the actor
because
(5)
The actor was criticized by the director (Surface NPl = deep NP2)
he forgot because
his lines. (NP2)
he forgot
his lines.
The introduction of negation on the assignment of pronoun antecedents had a similar though much weaker effect. There was a general shift in positive sentences from preference for NP2 assignments toward NPl assignments in negated sentences. A sentence fragment such as The doctor blamed the intern .. . elicited statistically significant NP2 responses while the fragment The doctor did not blame the intern . . . produced only a trend toward NP2 responses. In general, the effect of these syntactic factors on the direction of causality implicit in the verbs was to produce a preference for pronoun assignment to the grammatical subject or sentence initial NP. This preference is consistent with Sheldon’s (1974) speculation about the role of parallel function in pronominalization: a pronoun in the second conjunct of a complex sentence is interpreted as being coreferential with the NP that has the parallel grammatical function in the first conjunct. In sentence (6) (6)
John hit Bill and he kicked Sam. i i
the reference of the pronoun is determined by parallel function; the pronoun is assigned to the subject NP in the preceding conjunct. When NP’S
120
E. H. Grober, W. Beard&y and A. Caranlazza
which do not have the same grammatical function are interpreted as being coreferential, as in sentence (7), the result is much less acceptable. However, (7)
John hit Bill and then he kicked Sam. i i
a change in the usual interpretation can be signalled by placing contrastive stress on the pronoun as in sentence (8). The pronoun no longer refers to the subject NP in the preceding conjunct but to the object NP. (8)
John hit Bill and then he kicked Sam. i i
Further support for the role of parallel function in pronominalization can be garnered from Halliday’s distinction between theme and rheme in the analysis of clauses (Halliday, 1967). The distinction is realized by the sequence of elements in a clause: the theme comes first. The theme is the person or thing being talked about, the psychological subject. Often, the theme of the subordinate clause is presumed to be the same person who functions as the theme of the main clause. Thus, when a listener is confronted with a potentially ambiguous sentence fragment of the form used in the present study, he invokes a strategy that identifies the theme of the subordinate clause with that of the main clause. In terms of the Parallel Function Hypothesis (PFH) since the pronoun following the subordinate conjunction is the grammatical subject of that clause, it will be interpreted as being coreferential with the grammatical subject or sentence-initial NP of the main clause.
It is our contention that listeners rely on this general strategy to interpret sentences with potentially ambiguous pronouns in the first position of a subordinate clause. While the explanatory power of any general strategy may be quite adequate to account for a major share of the results, it will not work all the time. Manipulation of the semantic content of the sentence fragments may affect pronoun assignment in very regular ways. We already know that one restriction on our proposed strategy for pronoun assignment arises from the causal valence of the verb (Garvey, et al., 1975). When the verb imputes the cause of an event to the object NP, a preferred interpretation is established in which the pronoun is coreferential with the object NP, violating the proposed general strategy. In an effort to test the limits of the PFH strategy for pronominalization, we added two semantic variables to the sentence fragments subjects were asked to complete. First, the verb phrase was expanded by including a modal auxiliary verb. Second, the two clauses in half of the sentence fragments were joined by the connective but rather than because. It was expected that
Parallelfunction strategy in pronoun assignment
12 1
these manipulations would affect the proposed strategy in systematic ways and at the same time reveal other semantic variables, besides implicit causality, that could restrict its use. The specific predictions are elaborated in the following section.
Semantic Variables Main Verbs
The main verbs used in the sentence fragments were selected from the subset of English verbs that reference various types of interpersonal relationships involving judgments of worth and responsibility (Fillmore, 197 1). Several of these “verbs of judging”, criticize, scold, and praise, had previously been characterized as NP2 types (Garvey et al., 1975). Pilot. work indicated that one other verb, forgive, was also an NP2 type and two others, apologize and accuse, were NPl types. Similarities in the presuppositional and illocutionary aspects of these verbs were discussed by Fillmore (197 l), but it was uncertain whether these similarities would be implicated in pronoun assignment in the present task. It should be noted, however, that one presupposition implicit in verbs of judging - whether or not the situation or action described is favorable, (e.g.,praise) or unfavorable, (e.g.,criticize) - was not an important factor in determining response bias (Garvey et al., 1975). It was found that both criticize and praise elicited strong NP2 responses. Modal
Verbs
The main verb in each sentence fragment was modified by a modal auxiliary verb. The expansion of the verb phrase with a modal was expected to affect the PFH strategy for pronoun assignment in specific ways. Before describing these predictions, it is useful to consider a problem unique to modal verbs in English. What is the relationship between the meanings of must in (9a) and (9b)? (9a) You must be very careless. (9b) You must be very careful. (SC) You must be very sympathetic. The dominant reading of must in (9a) means approximately “it is obvious that” whereas in (9b) it means “you are required to be”. Frequently the word must can be easily interpreted in either sense as in (SC). The first of its meanings (9a) represents the speaker’s assessment of the probability of what he is saying. This use of modality derives from what Halliday (1967)
122
E. H. Grober, W. Beardsley and A. Caramazza
calls the interpersonal function of language which sets up the role relationship between speaker and hearer. The second use of modality (9b) is ideational in function, part of the meaning of the clause, and is related to the speaker’s experience of the real world. It expresses the factual conditions on the action expressed in the clause. The distinction between the two uses of modality corresponds respectively to the difference between the epistemic and root interpretations of modal verbs drawn by other linguists and philosophers (e.g., Antinucci and Parisi, 1971; Lakoff, 1972; Von Wright, 1951). In spite of the complex syntactic differences between them (see Halliday, 1967 for a thorough discussion), both senses can be subsumed under a single semantic classification system (Halliday, 1969). The epistemic notions of “probable or if not, then either possible or certain” can be equated with the root notions of “willing or if not, then either permitted or compelled”. The contrast of importance for the present study is that between strong modal auxiliary verbs (must, ought to) expressing necessity (certainty) and weak modal verbs (can, tnay) expressing permission (possibility). The two other modals used, will and should, have both a weak and a strong interpretation (Leech, 1970). Armed with this classification system, we can now speculate about the effect of modality on the PFH strategy for pronoun assignment. It was expected that the expansion of the verb phrase with a weak modal would attenuate the causal valence of the main verb. That is, reducing the likelihood of the action expressed in the main clause by including a weak modal, diminishes the importance of the antecedent event as a motive for the action. Whether the action is attributed to the subject or to the object noun should no longer restrict pronoun assignment. Instead, the subjet is free to rely on the PFH strategy in which the pronoun is assigned to the sentence initial NP. A very different result is expected when the verb phrase is expanded with a strong modal. The causal valence of the main verb should be augmented with a concommitant restriction on the application of the PFH strategy. That is, requiring that the action expressed in the main clause be carried out increases the salience of the antecedent causal event as a motive for the action. Thus, the causal valence of the verb rather than our proposed strategy, should be the primary determinant of pronoun assignment. The use of weak and strong modal auxiliaries may affect more than just pronoun assignment. It is also possible that a sentence fragment containing a strong modal may evoke an explanation for the action that is more compelling than if the sentence fragment contained a weak modal (Dakin, 1970). The rated “compellingness” of the explanations that are generated by subjects should be consistent with the semantics of the modal verbs as presented by Antinucci and Parisi (197 1) and Halliday (1967).
Parallelfunction strategy in pronoun assignment 123
But vs. Because
The other semantic variable that was expected to affect the proposed strategy for pronoun assignment was the use of the connective but, in addition to because. The use of one or the other of these subordinate conjunctions determines what will be considered an acceptable completion for a sentence fragment (Dakin, 1970). The sense of because invoked in the present task results in a description of some antecedent event or current state of affairs that, through known rules or conventions, can be considered a cause of the behavior asserted in the main clause. In contrast, but introduces a statement of what has happened which usually is the opposite of what might be expected on the basis of the behaviour described in the main clause. These differences in the nature of acceptable completions for but and because sentences may have very real consequences for our proposed strategy of pronoun assignment. Data from the Garvey et al. study demonstrate the importance of implicit causality in determining pronoun assignment for because sentences; the strategy of assigning the pronoun to the grammatical subject of the main clause will be blocked when the cause of the action is imputed to the object NP. For but sentences, however, the causal valence of the verb may have little effect on pronoun assignment. That is, the semantics of but could lead listeners to simply generate explicit denials of the action expressed in the main clause (Lakoff, 1971), maintaining a parallel construction in the subordinate clause. Thus, the pronoun would be assigned coreferential with the subject NP regardless of the direction of causality implicit in the verb, thereby supporting the proposed strategy for pronoun assignment. Method Subjects
One hundred and twenty-eight volunteer University participated in the experiment.
students
from The Johns Hopkins
Materials
of 60 sentence fragments of the format: NPI Thirty-six of these were test items in which were no grammatical cues available to assist pronoun assignment to NPI or NP2, e.g., John must scold Bill because he... The 24 distractor which were interspersed among the test items, did provide grammatical e.g., Mary must telephone Sam but she...
Test booklets NP2
because
consisted
(but)
Pro...
aux I’
there either items, cues,
124
.F H. Grober, W. Beardsley a& A. Cararnazza
Test items were constructed by combining every main verb: apologize, criticize, accuse, praise, forgive, and scold, with every modal auxiliary: must, ought to, will, should, may, and can. Each of the resulting 36 sentence fragments were then transformed to provide a set of 8 sentences that contrasted on a) voice: active vs. passive; b) verb polarity; positive vs. negative; and c) conjunction: but vs. because. Different sets of 36 test items were selected from the 288 items with the constraint that only 6 sentences for each main verb appear in a test booklet. No attempt was made to equate the 36 test sentences for voice, verb polarity, modality, or conjunction. Order of the pages in a test booklet was randomized so that no two subjects saw the same set of 60 sentences in the same order. Procedure
Subjects were tested individually and worked at their own pace. They were told to complete each sentence, writing a reason or motive that was appropriate for the action presented in the first part of the sentence.
Results Data from the passive and verb negated sentences were not included in the following analyses. The results were similar to those reported in Garvey et al., 1975. Response sheets were scored independently by two judges. Disparities in scoring between them occurred for 8% of the responses; these were then arbitrated by a third judge. Scoring consisted of judging the completed sentence as indicating either NPl or NP2 assignment of the pronoun in the second clause. Judges also scored a response as ambiguous (A) if it was not clearly interpretable as NPl or NP2; as unintelligible (U) if illegible or if it indicated lack of understanding of the test item; or as no response (NR). The last three scoring categories did not appear to be systematically distributed in relation to test items and were omitted from further analysis. Total number of possible responses was 1160: 6% of these were scored as A+U+NR responses. The NPl and NP2 type completion data are presented separately for the contrasts of conjunction and modality. But-Because
Table 1 presents the number of surface NPl and NP2 assignments for both the but and because sentences for each verb separately. As can be seen from
Parallelfunction strategy in pronoun assignment
125
the distribution of NPl and NP2 responses, type of connective influenced the assignment of pronoun antecedents (x’ = 190.45, p < O.OOl), with but sentences eliciting significantly fewer NP2 assignments than because sentences. This was confirmed in separate comparisons for each verb (p < 0.001 in all cases) with the exception of apologize, a strong NPl type verb. Moreover, the number of NP2 assignments for but sentences did not vary substantially across particular verbs. Thus, it appears that direction of causality implicit in the verbs had little effect on pronoun assignment for but sentences. Subjects simply generated an explicit denial of the action with the pronoun referring to the grammatical subject of the main clause. Table 1. Verb
apologize accuse forgive criticize praise scold Total
Completions for Because and But Sentences Because
But
PQ
NPI
NP2
NPI
NP2
-
92 64 36 34 26 21
9 35 37 55 53 66
85 19 89 93 82 14
8 6 10 I 9 15
“.S. 0.001 0.001 0.001 0.001 0.001
279
255
502
55
0.001
The pattern of pronoun assignments for because sentences is consistent with earlier results (Garvey et al., 1975). The direction of causality implicit in each verb is a salient factor in the determination of antecedent assignment. The verbs apologize and accuse, which had been identified previously as NPl type verbs, had fewer NP2 responses than the verbs criticize, praise, forgive, and scold, which had been identified previously as NP2 types. Since the semantics of causality is implicated in pronoun assignment for complex clauses joined by because, while the semantics of but appears to influence pronoun assignment for complex clauses joined by but, separate analyses were carried out for but and because sentences. Modality
In order to assess the effect of modality on pronoun assignment, we considered only those sentences containing modal auxiliaries that could be unambiguously classified as strong (must, ought to) or weak (can, rrzaJ1).
126
1:‘.H. Grober, W. Beardsley and A. Caramazza
Table 2 presents the number of NPl and NP2 assignments for each main verb when combined with a strong or a weak modal in clauses joined by because. As can be seen from the distribution, type of modal influenced the assignment of pronoun antecedents (x2 = 20.53, p < O.OOl), weak modals producing a shift toward NPl assignments. More specifically, NP2 type verbs (criticize, forgive, and scold) appeared as strong NPl’s when modified by a weak modal and as strong NP2’s when modified by a strong modal (p < 0.05 in all cases). Praise, an NP2 type, and apologize, a strong NPl type, remained unchanged when modified by either a strong or weak modal.’ Accuse, the other NPl type verb, appeared as a stronger NPl when modified by a weak modal although the effect was not significant. Table 2. Verb
Completions for Strong and WeakModal Verbs in Because Sentences Strong NPI
Weak NP2
NPI
PG NP2
apologize accuse forgive criticize praise scold
31 21 I 10 8 3
2 15 21 21 20 28
29 19 16 11 8 19
1 8 I 10 17 12
n.s. n.s. 0.005 0.05 ns. 0.001
Total
88
107
107
46
0.001
For because sentences, there is a complex interacti.on between two semantic factors, implicit causality and mood of the sentence, in determining pronoun assignment. Modality either augments or leaves unaffected NPl type verbs, but has a more dramatic effect on NP2 type verbs. Weak modal auxiliaries reverse the direction of causality implicit in these verbs, while strong modals intensify their original causal valence. The effect of modality on pronoun assignment in because sentences contrasts sharply with its lack of effect in but sentences. The number of NPI and NP2 assignments for but sentences appear in Table 3. No reliable differences emerged for individual verbs: all appeared as strong NPl’s when ‘Praise patterned like the other NP2 type verbs when the results were combined with the data from the passive and verb-negated sentences. This discrepancy was the only one which emerged when the data were separated.
Parallel function
127
strategy in pronoun assignment
combined with weak or strong modal auxiliaries. These results reinforce our earlier claim that in but sentences the semantics of causality and modality are subordinated to the parallel function strategy in determining the direction of pronoun assignment.
Table 3.
Completions for Strong and Weak Modal Verbs in But Sentences Weak
Strong Verb
NPI
apologize accuse forgive criticize praise scold Total
NP2
NPI
NP2
P
29 22 29 39 27 29
1 1 2 2 3 3
27 28 31 25 30 28
6 2 2 2 1 10
n.s. n.s. n.s. ns. ns. n.s.
175
12
169
23
n.s.
Compellingness An analysis, peripheral to the question of pronoun assignment, was performed on the explanations generated to actions described in the because sentences. Two judges rated each explanation on a five point scale depending upon whether it represented a compelling (5) or inconsequential (1) justification for the subsequent action. A product moment correlation of 0.72 indicated that the judges had substantial agreement in their ratings. Mean compellingness ratings for each modal auxiliary pooled across main verbs appear in Table 4. Sentence fragments containing strong modals (i.e., ought to, must) evoked more compelling explanations than those containing weak modals (i.e. may, can). Should appears to have been interpreted in its strong form while wiZE appears to have received both its weak and strong interpretations.
Table 4.
Mean Compellingness
ratings for modal verbs
ought to -
must
should
will
clln
may
3.82
3.71
3.64
3.22
2.75
2.57
-
128 E. H. Grober, W. Beardsley and A. Caramazza
Discussion The grammatical subject or sentence-initial NP was selected as the appropriate antecedent for the pronoun in over 70% of all the sentence fragments in the present study. This preference confirms the importance of parallel function in pronominalization: the pronoun in the subject position of the subordinate clause was interpreted as being coreferential with the NP that had the parallel grammatical function in the main clause. That listeners rely on this general strategy to interpret sentences with potentially ambiguous pronouns is consistent with the necessity for parallel constructions in the other areas of syntax. The unacceptability of the sentence “John and the hammer broke the window” (from Sheldon, 1974) results from the inability to conjoin NP’s of different cases. Similarly, any attempt to conjoin a gerund and an infinitive produces unacceptable results as in the sentence “I like to water ski and swimming”. The preference for parallel constructions in pronominalization can be evoked to explain some differences in sentences with NPl and NP2 readings. (12a) (12b)
John sold the bike to Henry because he needed the money. (NPl) John sold the bike to Henry because he could pay cash. (NP2)
The two the order sentence reversing (13a) (13b)
readings, (12a) and (12b), are not equally amenable to reversals in of their clauses. When the order is reversed for (12a), the resulting (13a) is perfectly acceptable while sentence (13b), produced by the clauses in ( 12b), is much less acceptable.
Because he needed the money, John sold the bike to Henry Because he could pay cash, John sold the bike to Henry.
According to the Parallel Function Hypothesis, the pronoun in the subject position of the first clause in (13a) is assigned coreferentially with the NP in the subject position of the second clause whereas the same strategy is blocked in (13b). A further difference is that only NP2 readings appear to be subject to passivization. Passive variants of NP2’s are accepatable whereas passive versions of NPl’s are questionable, e.g., (14a) (14b) (15a) (15b)
The prisoner shot the warden because he knew there would amnesty. (NPl) The warden was shot by the prisoner because he knew there be no amnesty. The warden shot the prisoner because he was trying to escape. The prisoner was shot by the warden because he was trying to
be no would (NP2) escape.
Parallelfunction strategy in pronoun assignment
129
The asymmetry in the acceptability of passive variants of NPl and NP2 readings is consistent with the Parallel Function Hypothesis: when the pronoun in the subject position of the subordinate clause is coreferential with the NP that has the same grammatical function in the main clause, the sentence is more acceptable than when it is coreferential with the NP that has a different grammatical function (Sheldon, 1974). Parallel function
and the semantics
of verbs
While the strategy of interpreting the pronoun as being coreferential with the grammatical subject of the main clause is adequate to account for the majority of pronoun assignments in the present study, it is limited in its applicability. One restriction on its use arises from the causal valence of the main verb. Recall that implicit causality selects one or the other of the available candidate nouns as primarily responsible for instigating the action expressed in the main clause, and unless some modulating influence is exerted by other linguistic elements, it will assign the pronoun according to the direction established by the verb. When the verb imputes the cause of the event to the object NP, the pronoun is then assigned coreferential with it, in violation of the proposed strategy. However, the generality of this result is limited to sentence fragments with main verbs unmodified by modal auxiliaries. When modal verbs are introduced, the pattern of pronoun assignment changes. Thus, in because sentences in the present study, weak modals reversed the direction of causality for NP2 type verbs (i.e., criticize, scoZd, and forgive) while strong modals intensified their original causal valence. NPl type verbs retained their sentence-initial NP assignment when combined with both weak and strong modals. This complex interaction of modality and causality can be rendered meaningful if we consider the logical structure that corresponds to the root sense of the modal verbs. When dealing with the root interpretation, two noun phrases may be present in the underlying structure of the sentence, one corresponding to the bearer of the obligation (permission) and one corresponding to the source of the obligation (permission) (Lakoff, 1972). In the sentence, “Jill must kiss Jack because he carried her pail of water up the hill”, the obligation devolves on Jill but originates from something Jack did. In the present experiment, these two roles appear to function independently of one another to determine pronoun assignment, and which one dominates depends upon whether the modal verb is a strong or a weak one. For weak modals, the bearer of the permission, in active sentences generally the sentence-initial NP, is taken to be the appropriate referent for the pronoun. For strong modals, the source
130
E. H. Grober, W. Beardsley and A. Caramazza
of the obligation, generally residing in the person to whom the cause of the action is imputed, is taken to be the appropriate referent of the pronoun. Thus, strong modals operate in conjunction with the direction of causality implicit in the main verb to determine pronoun assignment, while weak modals permit the unrestricted use of the PFH strategy for pronoun assignment. One reason why this pattern of pronoun assignments may have emerged is illustrated by the following examples. Consider the sentence fragment “John may/can scold Bill...“. John is permitted to scold Bill but whether or not he does so depends more on his own mood, feelings, physical well-being and/or disposition towards Bill than on what it was that Bill actually did to provoke John’s anger. This is in contrast to the situation expressed in the sentence, “John must/ought to scold Bill...“. Here, John is obligated to scold Bill but the source of the obligation, or alternatively the cause of the reprimand, arose out of something that Bill did rather than out of something independently conceived of by John. It seems reasonable that in the former case an explanation for the action will involve John (i.e., the bearer of the permission) while in the latter case it will involve Bill (i.e., the source of the obligation). Sentences representative of the type of completion generated in response to the presence of weak and strong modal operators serve to illustrate this point. (16) (17)
Nancy may scold Marge because she dislikes people who crack their gum. Alexander must scold Mark because he disobeys orders continuously.
Whether Nancy in (16) scolds Marge probably depends on some aversion she may have to gum crackers while Alexander’s reprimand of Mark in (17) results from Mark’s disregard of known rules or conventions. The qualitative differences inherent in violating idiosyncratic standards of behavior and violating established patterns of behavior are consistent with the qualitative differences obtained in the rated compellingness of the explanations. The presence of weak modal operators evoked much less compelling explanations of the behavior. Thus, not only does the modality of the sentence interact with the implicit causality of the main verb to determine pronoun assignment, but it also determines how compelling an antecedent event must be to “justify” a subsequent action. Purullel Function
urld the scr?urltics
of but
The combination of modality and implicit causality in became sentences can restrict the possible readings for a potentially ambiguous pronoun, thus
Parallel function
strategy in pronoun
assignment
13 1
rendering the NPl interpretation predicted by the PFH strategy less likely. The presence of the connective but, however, virtually guarantees the production of a parallel construction in the subordinate clause. In but sentences, the first noun phrase was overwhelmingly selected as the appropriate referent for the pronoun, regardless of the causal valence of the verb and the type of modal that modified it. It appears that a strategy involving parallel constructions is triggered by the presence of but. We expected that this strategy would involve the use of but referred to as “denial of expectation” (Lakoff, 197 1) and that it would result in a very specific type of completion for the situation described in the main clause. Consider the sentence “John is tall but he’s no good at basketball” (from Lakoff, 1971). This sentence consists of an assertion and a presupposition. The presupposition resides in the speaker’s knowledge of the world and involves the expectation that someone who is tall is good at basketball. The presence of but in the sentence signals the denial of this presupposition. Evidence that a strategy based on denial of expectation is operating in the present study is available from the completions that were generated to but sentence fragments. (18) (19) (20)
Joseph can scold Michael but he has no reason to. Christine must apologize to Linda but she doesn’t have to be sincere. Lawrence may praise Nick but he doesn’t have to like him.
In sentence (18), a presupposition that is part of the lexical description of the verb is being denied (Fillmore, 197 1); when a person scolds someone else, it is because he believes that the individual is responsible for some blameworthy action. In sentences (19) and (20), the presuppositions being denied are less directly a part of the semantics of the verbs than a part of what we know in general about the acts of apologizing and praising. A person apologizing to someone is usually presumed to be sincere in his request for forgiveness unless, of course, he has some ulterior motive. And, people usually do not shower other people with praise unless they like them. The strategy based on denial of expectation produced another type of completion in which the action expressed in the main clause was explicitly denied. These completions took the form of elliptical sentences, for example, “Jennifer should criticize Beverly but she won’t”.* *The semantics of but resulted in a slightly different type of completion for sentence fragments in which the verb was explicitly negated. Moreover, the alternative action was related in a very consistent way to the negated one. This pattern can be seen in some completions representative of ones that were negated. Matilda must not accuse Dorothy but she (21) (a) will spread vicious rumors. (b) will give her dirty looks.
132
E. H. Grober,
W. BeardsIt_v and A. Caramazza
Conclusion
The PFH strategy proposed for pronoun assignment is similar to the basic perceptual strategies proposed by Bever (1970) to underlie the comprehension of sentences. These perceptual strategies map external sequences of words onto internal structures. For example, Bever claims that listeners first isolate adjacent clauses in the surface structure of the sentence consisting of Noun-Verb-Noun (NVN) sequences which could potentially correspond to sentences in the underlying structure. This strategy is so compelling that subjects in an immediate comprehension task could not avoid assuming that an NVN sequence in the surface structure corresponded to a clause in the underlying structure even when explicitly instructed that this interpretation was incorrect. It is our contention that the PFH strategy proposed for pronoun assignment is a basic perceptual strategy that underlies the comprehension of sentences which have a potentially ambiguous pronoun in the subject position of a subordinate clause. Listeners readily interpret the pronoun as being coreferential with the NP that has the same grammatical function because of a predilection for parallel constructions. Just as semantic constraints may require the reassignment of internal structure to the lexical items in a NVN sequence initially assigned by a basic perceptual strategy, so too may semantic factors such as causality and modality require the reassignment of the pronoun to the NP which does not have the same grammatical function.
References Antinucci, F., and D. Parisi (1971) On English modal verbs. Chicago Linguistic Society, 28-39. Bever, T. G. (1970) The cognitive basis for linguistic structures. In J. R. Hayes (Ed.), Cognition and the development of language. New York, John Wiley and Sons, Inc. Caramazza, A., E. H. Grober, C. Garvcy and J. Yates (1977) Comprehension of Anaphoric Pronouns. J. verb. I.earn. verb. Beh., 16, 601-609. Dakin, J. (1970) Explanations. J. Ling., 6, 199-214. I:illmore, C. J. (1971) Verbs of judging: An exercise in semantic description. In C. J. Fillmore and D. T. Langendocn (Eds.), Studies in linguistic semantics. New York, Halt, Rinehart and Winston, Inc. Grvey, C. and A. Caramaiza (1975) Implicit causality in verbs. Ling. Inq., 4, #3. Garvcy, C., A. Caramazza and J. Yates (1975) Factors influencing assignment of pronoun antecedents. Cog., 3, 227m~243. tlalliday, M. A. K. (1967) Notes on transitivity and theme in English. J. Ling., 3, 199-244. Lakoff, R. (1971) Ifs, and’s, and but’s about conjunction. In C. J. I:illmore and D. T. Langendoen (Eda.). Stud&s it? lirlgrrisric semantics. New York, Halt, Rinehart and Winston, Inc. Lakoff, R. (I 972) The prapmatics of modality. Chicago Linguistic Sociefv, 229-246. Leech, G. (1970) Touwrds 0 sc/??antic description of Erzglish. Bloomington, Indiana University Press. Quillian, M. R. (1968) Semantic memory. In M. L. Minsky (Ed.), Semantic information processing. C;umbrid$c, MIT Press.
Parallel function
strategy in pronoun assignment
Sheldon, A. (1974) The role of parallel function in the acquisition of relative J. verb. Learn. verb. Beh., 13 (3), 272-281. Von Wright, G. H. (195 1) An essay in modal [ogic. Amsterdam, North-Holland. Winograd, T. (1972) Understanding natural language. Cog. PsychoL, 3, 1-191.
clauses
133
in English.
Des sujets ont complete des phrases de la forme NPI aux V NP2 because (but) Pro... (e.g., John peut reprimander Bill parce qu’il...) en apportant une raison ou un motif a I’action d&rite dans la premiere partie. On emit I’hypothkse selon laquelle une strategic perceptuelle de base ktait sous-jacente a la comprehension des phrases qui comportent un pronom potentiellement ambigu en position sujet dans la proposition subordonnee. 11 fut predit que des auditeurs interpreteraient le pronom comme ktant coreferentiel avec le NP sujet de la proposition principale, i.e., le NP possedant la msme fonction grammaticale. Alors meme que cette strategic a rendu compte de la plus grande partie des rksultats, des facteurs semantiques ont limit6 son utilisation en provoquant une interpretation dans laquelle le pronom etait coreferential avec le NP objet de ia phrase principale.
Cognition, @Elsevier
6 (1978) 135 - 153 Sequoia S.A., Lausanne
3 - Printed
in the Netherlands
Speech timing of grammatical
categories*
JOHN M. SORENSEN, WILLIAM JEANNE
E. COOPER, M. PACCIA
Research Laboratory of Electronics, Massachusetts Institute of Technology
Abstract
A series of experiments was performed to determine how the duration of a word spoken in a sentence is influenced by: (1) the grammatical category to which it belongs, and (2) the position of the word in a constituent. Experiment I contained Noun-Verb homophones (e.g. “I saw the coach...” - noun; ‘7 saw him coach...” - verb), in sentences matched for phonetic environment and stress pattern. Results support the notion that Nouns are longer than Verbs in typical sentences. Experiment II, however, demonstrated that the duration of Noun-Verb homophones in matched clause-final position is approximately equal. Experiment II1 contained transitive and intransitive Verbs in matched sentences. The results of Experiments II and III provided support for a lengthening account based on constituent-final position, while ruling out an account based on a debatable deletion site. Results of Experiment IV generalized the account of constituent-final lengthening for Nouns and Verbs to additional major categories, by comparing the duration of the phrase-initial Adjective two with the phrase-final Adverb too. Finally, Experiment V tested the distinction between minor and major categories. Results demonstrated that the Preposition to is shorter than the Adjective two by approximately 50 percent. Taken in sum, these findings indicate that it is sufficient to make a binary distinction between major and minor categories for purposes of a theory of speech timing and speech synthesis. Durational *This research was supported by NIH Grant NS-13028 and an NIH Postdoctoral Fellowship. J. M. Paccia is also at the Department of Psychology, Boston University. Reprints: J. M. Sorensen, 36-549, M.I.T., 77 Massachusetts Avenue, Cambridge, Massachusetts 02139.
136
John M. Sorensen, William E. Cooper, Jeanne M. Paccia
effects traditionally ascribed to differences within the class of major categories can be accounted for solely in terms of constituent boundaries, alreudy required in the theory to account for three other classes of phenomena
Introduction The grammatical structure of an utterance exerts a substantial influence on the duration of speech segments. Major syntactic boundaries may produce segmental lengthening (Martin, 1970; Lindblom and Rapp, 1973, Klatt, 1975; Cooper, 1977) and act to block the application of durational rules that normally operate across word boundaries (Huggins, 1974, Cooper, Lapointe, and Paccia, 1977). Furthermore, a certain class of deletions, exemplified by Verb Gapping (Ross, 1970), produce lengthening of the word preceding the deletion site (Cooper and Paccia, 1977). In addition to the effects observed at certain syntactic boundaries and deletion sites, it appears that segmental timing may be influenced by the grammar according to the type of grammatical category to which a word belongs. It is well-known that words of different grammatical categories have varying probabilities of receiving primary stress in an utterance. In particular so-called content words belonging to categories like Noun and Verb are more likely to receive primary stress than so-called function words belonging to categories like Prepositions and Conjunctions. Since primary stress is accompanied by the acoustic correlate of longer duration, in addition to higher fundamental frequency and intensity (Fry, 1957), it is reasonable to suppose that segmental timing is influenced by the grammatical category to which a word belongs. In a program to synthesize speech by rule, Coker, Umeda, and Browman ( 1973) have incorporated this principle in their rules for segmental duration, assigning nine different stress levels to lexical items. The influence of grammatical categories on speech timing is accepted as self-evident for major versus minor grammatical categories. By major categories, we mean Nouns, Verbs, Adjectives and Adverbs. In addition, it has been claimed that durational differences exist within the class of major categories. For example, it has been informally reported that Nouns are typically longer than Verbs (Lightfoot 1970; Coker, Umeda and Browman 1973). Unlike the durational effect for major versus minor categories, the differences observed within the class of major categories cannot be accepted as self-evident and have not been documented in controlled experiments. We have conducted tests in this study to determine the effects of category type in sentences matched for phonetic environment in the region surrounding the measured segment. Matching for phonetics is important because the phonetic environ-
Speech timing of grammatical categories
137
ment of a word segment also influences its duration (Klatt 1976). By conducting such experiments, we sought to determine whether timing effects traditionally ascribed to the influence of category types were in fact attributable to this source or to independent influences of constituent boundaries or the presence of a debatable deletion site. As just noted, it has been observed informally that the duration of Nouns is typically longer than that of Verbs. This difference is usually attributed to the fact that Nouns form a larger lexical class than Verbs, such that the information load carried by a given Noun is larger than that carried by a Verb, under the assumption that duration is a positive correlate of information load (Coker, Umeda, and Browman, 1973). However, another interpretation may be advanced, relying on the notion of phrase-final lengthening. Klatt (1975) has observed that major phrase boundaries such as that between the Noun Phrase and Verb Phrase (NP-VP) of a main clause are accompanied by segmental lengthening. Typically, Nouns occur in phrase-final position at the ends of Noun Phrases (NPs) in English. Verbs, on the other hand, usually occur at the beginning of Verb Phrases (VPs). Thus the influence of grammatical category type on segment duration is confounded with the influence of phrase position.
EXPERIMENT
I
In this experiment, we attempted to document the durational difference for Nouns versus Verbs in matched sentence materials. The duration of the same word segment was measured in its occurrence as either Noun or Verb in sentence pairs matched for phonetic environment and stress contour. Structurally, the Nouns occurred in phrase-final position while the Verbs occurred in phrase-initial position. As such, the present experiment is aimed at simply determining the validity of the Noun-Verb difference in duration that has been traditionally claimed.
Method Subjects
Ten M.I.T. undergraduates participated as paid volunteers in this experiment. All were native speakers of English with no history of speech or hearing impairment. The speakers had little or no training in linguistics or speech science.
138
John M. Sorensen, WilliamE. Cooper, Jeanne M. Paccia
Sentence
Materials
Eight test sentences and three fillers were constructed for this experiment. The test materials included four pairs of sentences, with each pair containing a key word as either Noun or Verb. The test sentences appear below. The key word in each pair is in italics. 1. a. b. 2. a. b. 3. a. b. 4. a. b.
I showed Marie a coach that Eve will like. I helped Maria coach the team last night. John gave Marie a tape that Christopher made. John helped Maria tape the Christmas parade. Thomas gave Eve a coat that was covered with paint. Thomas helped Eva coat the old ceiling with paint. Tom told Eve a joke about Harry. Tom and Evajoke about Harry.
In each case the key word began and ended with an obstruent to facilitate segmentation of the waveform (see Procedure). The key word was always preceded by an unstressed schwa. In addition, the key word in each pair was followed by the same phonetic segment, and the overall stress pattern of the sentences in each pair was closely matched. Procedure
Speakers were tested individually in a sound-insulated chamber. Each speaker was presented with a typewritten list containing the test sentences and fillers in a quasi-random order. They were instructed to practice reading a given sentence from the list until they could read the sentence as a unitary whole rather than word-by-word as in unpracticed reading. The speakers were told to refrain from placing emphatic or contrastive stress on any word in the sentence. Following practice, the speaker read the sentence once aloud to allow the experimenter to check recording levels and for any unusual stress contour. The speaker was then instructed to utter the sentence once for recording purposes. If the speaker departed from his intended utterance, he said repeat and then said the sentence token again. The utterances were recorded onto magnetic tape via an Altec 684A microphone and a Presto A908 tape recorder. The key segment of each sentence was measured from digitized oscillographic traces of the speechwave (sampling rate = 10 kHz). The durations were measured by manipulating a computer-controlled cursor to mark the beginning and end of the desired segment (Huggins, 1969). The duration between the two marks was displayed on the oscilloscope screen to the nearest 100 ,usec. The accuracy for each measurement was estimated to be within ?5 msec. In each case, the measured segment started at the release burst of the word-initial obstruent and ended at the termination of glottal pulsing
Speech timing of grammatical categories
139
just prior to the word-final voiceless obstruent in the key word. For example, in Pair (1) the measured segment of the key word couch included the consonant /k/ and the following vowel.
Results and Discussion The mean durations and standard deviations of the key segments for each sentence type appear in Table 1. It can be seen that the mean duration of the key segment was longer in each sentence pair when this segment was a Noun rather than a Verb. In Sentence Pairs (1) - (3), this difference was statistically significant: (Pair (1): p < 0.001, t = 6.16, df= 9; Pair (2): p < 0.01, t = 3.56, df = 9; Pair (3): p < 0.01, t = 4.37, df= 9; two-tailed t-tests for matched pairs). In Pair (4), a non-significant difference was observed in the same direction (O.lO>p>O.O5, t = 2.03, df = 9; two-tailed t-test for matched pairs). Table 1.
Mean durations and standard deviations of the key segment portion of the italicized key words in Experiment
I. X (msec)
1.
a. I showed Marie a coach that Eve will like. I helped Maria coach the team last night.
b. 2. a. b. 3. a. b. 4. a. b.
John gave Marie a tape that Christopher made. John helped Maria tape the Christmas parade. Thomas gave Eve a coat that was covered with paint. Thomas helped Eva coat the old ceiling with paint. Tom told Eve ajoke about Harry. Tom and Evajoke about Harry.
205.8 161.7 189.2 165.2 202.5 172.4 184.8 171.9
30.9 14.2 38.1 27.0 27.2 25.9 28.2 15.4
The results support the notion that the durations of Nouns are longer than the duraticns of corresponding Verbs. However, the results do not permit us to determine whether this difference in duration is attributable to grammatical category type or to phrase position. An account based on phrase position is attractive because specification of a word’s position in a constituent is already required in a theory of speech production for the application of other prosodic features (see General Discussion). EXPERIMENT
II
In the normal contexts in which Nouns and Verbs occur in English (cf., Experiment I), it is not possible to quantify separately the effects of grammat-
140
John M. Sorensen,
William E. Cooper, Jeanne M. Paccia
ical category type and phrase position. The sentence pairs in this experiment contain Noun-Verb pairs placed in constituent-final position. In this way, the position of the key word in the constituent was held constant within a pair, and any remaining effect could be attributed to grammatical category type’.
Method Subjects
Ten M.I.T. undergraduates, two of whom served in Experiment I, participated in this experiment. The eight new subjects had the same qualifications as those who had served previously. Sentence
Materials
Eight test sentences and three fillers were constructed for this experiment. The test sentences appear below, with the key words in italics. 5. 6.
a. b. a. b.
7. 8.
a. b. a. b.
John will find Eve a coach if she ever decides to sing professionally. John will help Eva coach if she ever decides to start a basketball team. At the swimming meet Paul found Marie a coach and hoped her team would win the diving contest. At the swimming meet Paul watched Maria coach and hoped her team would win the diving contest. During class Beth told Marie a joke but RenCe read the syllabus. Usually Beth and Maria joke but today they were serious. Whenever Professor Jones is late for class Jeff tells Eve a joke. Whenever Professor Jones is late for class Jeff and Eva joke.
Procedure
The procedures were identical to those in Experiment I, except that subjects were instructed to say each sentence twice for recording. The key segment of the first appropriate token of each sentence was measured for duration.
Results and Discussion Sentence Pairs (5) and (6) showed essentially no difference in the length of the key segment of coach: the duration of the Noun in (5a) averaged 2.4 msec ‘We must also assume that there constituent position. This assumption
is no interaction effect between is reconsidered in the discussion
grammatical category of this experiment.
type
and
Speech timing of grammatical categories
Table 2.
141
Mean durations and standard deviations of the key segment portion of the italicized key words in Experiment II* X (msec) 5. a. John will find Eve a coach if she ever decides to sing professionally. b. John will help Eva coach if she ever decides to start a basketball team. 6. a. At the swimming meet Paul found Marie a coach and hoped her team would win the diving contest. b. At the swimming meet Paul watched Maria coach and hoped her team would win the diving contest. I. a. During class Beth told Marie a joke but Renee read the syllabus. b. Usually Beth and Maria joke but today they were serious. 8. a. Whenever Professor Jones is late for class Jeff tells Eve a joke. b. Whenever Professor Jones is late for class Jeff and Eva joke.
226.2
20.1
223.8
25.3
239.5
20.7
236.4
19.8
219.8
24.9
241.4
24.4
247.4
18.8
260.0
21.2
*Note that the durations of the key words are longer in Experiment II than for matched words in Experiment I. The longer durations obtained in this experiment may be attributed to the fact that the key words occurred at either the boundary separating two main clauses (Sentences (4) - (7)) or at the end of the utterance (Sentence (8)). It appears that segmental lengthening is greater in these locations (Cooper and Paccia 1977).
or 1 .l% longer than the Verb form in (Sb), and the Noun in (6a) was 3.1 msec or 1.3% longer than the Verb in (6b), with p > 0.5 for both pairs (two-tailed t-tests for matched pairs). An effect in the opposite direction occurred in pairs (7) and (8). The duration of joke as a Verb in (7b) was significantly longer than as a Noun in (7a); (p < 0.05, t = 2.76, df = 9; two-tailed t-test for matched pairs), and in Sentence Pair (8), a non-significant effect occurred in the same direction (0.20> p>O.lO, t = 1.59; two-tailed t-test for matched pairs). The mean segment durations and standard deviations for all four pairs are shown in Table 2. By placing Noun-Verb homophones of the type found in Experiment I in constituent-final position, we hoped to document the independent influence of grammatical category type on segment duration. Given the results of Experiment I, it was expected that the Nouns might exhibit longer durations than the Verbs. In order to account for the results of Experiment II, two additional influences can be considered. First, the assumption made above (see Footnote on page 140) that there is no interaction effect between grammatical category type and constituent position may be incorrect. In particular,
142
John M. Sorensen,
William E. Cooper, Jeanne M. Paccia
Verbs may show more clause-final lengthening than Nouns. Verbs in clausefinal position violate the normal canonical word order (Subject-Verb-Object) of English. The speaker may lengthen a Verb in constituent-final position to allow the listener more time to process this unusual word order. This possibility will not be tested here. A second possible influence, to be tested in Experiment III, involves the debatable deletion of a direct object (Chomsky 1965, Grinder 197 1, but see Sampson 1972). As noted in the Introduction, one class of deletions, exemplified by Verb Gapping, acts to lengthen the word prior to the deletion site. It could be argued that deletion of the direct object occurs in each of the (b) versions of Sentences (5) - (8). For example, if the deleted material is reinserted in Sentence (5b), the result is the following: 5.
c.
John will help Eva coach a basketball team if she ever decides to start a basketball team.
It is conceivable that the deletion of the italicized words above may act to lengthen the Verb couch. If so, it is possible that the generally null results obtained in Experiment II may be attributed to the combination of two opposing effects: (l), the supposed inherently longer duration of Nouns versus Verbs, and (2), lengthening of the Verb just prior to the deletion site produced by Object Deletion. In any event, the present results suggest that the durational difference for Nouns versus Verbs observed in Experiment I, traditionally ascribed to their status in different grammatical categories, may rather be attributed solely to differences in position within a constituent. If so, there would be no need to specify individual category types in a theory of speech timing.
EXPERIMENT
III
This experiment was designed to test whether the debatable rule of Object Deletion (described above) acts to lengthen the duration of a Verb in constituent-final position. Sentences were constructed using Verb homophones, some of which take objects (Transitive) and some of which do not (Intransitive). Lengthening of an Intransitive Verb in constituent-final position can not be due to an influence of Object Deletion, since by definition such a Verb takes no direct or indirect object. Examples of such Verbs are sleep, shiver, and die. For purposes of constructing matched sentence pairs, we will also consider Verbs which take objects optionally, depending on the subject. One such Verb is fl.v: cf., “The pilot flew the plane”, “The bird flew”.
Speech timing of grammatical categories
143
Method Subjects Ten M.I.T. undergraduates, none of whom had served previously, participated in this experiment. All had the same qualifications as those in Experiment I. Sentence Materials Ten test sentences and three fillers were constructed for this experiment. The test materials consisted of four groups of sentences matched for key word position and stress contour. Following each sentence below, a description is given of the key Verb’s position and object relation. 9
a. b. C.
10
a. b.
11
a. b. C.
12
a. b.
If the pilot flies the plane we’ll surely crash. (PIP = Phrase Initial Position, TR = Transitive) If the pilot flies the plane will surely crash. (PFP = Phrase Final Position, TR) If the parrot flies the boy will feed him cake. (PFP, IN = Intransitive) If the baby parakeet flies to Lisa we’ll be happy. (PIP, IN) If the baby parakeet flies Teresa will be happy. (PFP, IN) If the tailor dyes the cloth we’ll refuse to buy the suit. (PIP, TR) If the tailor dyes the cloth will no longer hold a crease. (PFP, TR) If the tailor dies the cloth in his shop will all be sold. (PFP, IN) If the tailor dies in the summer his shop will be sold.* (PIP, IN) If the tailor dies in the summer his shop will be sold. (PFP, IN)
Procedure The procedure was identical to that of Experiment II. The duration of the key segment of the Verb was measured for the first appropriate token of
‘Sentences (12a) and (12b) are identical in their surface structure. Subjects were explicitly directed to consider one particular meaning when (12a) or (12b) appeared on the utterance list. In (12a), they were instructed that “the tailor passed away during the summer”, and in (12b), that “the sale of the dead tailor’s shop will occur during the summer”. (Cf., Sentence Pair (15)).
144
John M. Sorensen,
William E. Cooper, Jeanne M. Paccia
each sentence. For Pies, the measured segment began with the onset of voicing for the /I/ and ended with the offset of regular voicing of the vowel. The offset sometimes overlapped with the onset of frication noise of the /s/. In such instances, the offset was marked where the /s/ noise was no longer modulated in a semi-periodic manner characteristic of regular voicing. The key word dies or dyes was measured from the beginning of the burst of the /d/ to the offset of voicing as described forflies. The fact that the key words did not begin with clear release bursts nor end with voiceless obstruents was necessitated by the limited number of Verbs in English which fulfill the other criteria for this experiment.3 The estimated measurement error for the key segments was about t 10 msec.
Results and discussion
Comparing Transitive Verbs in phrase-initial and phrase-final position, significant lengthening occurred for the phrase-final cases: (9b) versus (9a), p < 0.001, t = 8.34,df= 9; (llb) versus (lla),p < 0.01, t =4.64,df=9;twotailed t-tests for matched pairs. Intransitive Verbs in phrase-final position were also significantly longer than the phrase-initial Transitive Verbs: (SC) versus (9a), p < 0.001, t = 8.35, df = 9; (1 lc) versus (1 la), p < 0.001, t = 4.82, df = 9; two tailed t-tests for matched pairs. The mean segment durations and standard deviations are shown in Table 3. The percent lengthening of (9b) and (SC) versus (9a) averaged 49% and 5 l%, respectively; for (1 lb) and (1 lc) versus (11 a), the percent lengthening averaged 40% and 46%. These effects must be attributed to constituent-final lengthening. The close similarity of the magnitudes of lengthening in the Transitive (b) and Intransitive (c) sentences of Groups (9) and (11) indicates no significant effect of Object Deletion on word duration. In addition, grammatical category effects were neutralized, since all the key words were Verbs. The only account consistent with the data from Experiments I - III is one based on constituent-final lengthening. Further evidence against a lengthening account based on Object Deletion can be obtained by comparison of the sentences in Pairs ( 10) and (12). Recall that in these sentences, Intransitive Verbs appear in phrase-initial versus phrase-final position. Significant lengthening was found forflies in (1 Ob) versus (lOa), (p < 0.01, t = 4.13, (If= 9) and for dies in (12b) versus (12a), 0, < 3The selection of Verbs in this experiment included criteria in addition to monosyllabalicity, required in all experiments to facilitate identification and segmentation of the waveform. Verbs in this expcrimcnt were required to either (a) take a direct or indirect object depending on the subject, or (b) be a member of a homophonous Verb Pair in which one Verb is transitive and the other is intransitive.
Speech timing of grammatical categoties
Table 3.
145
Mean durations and standard deviations of the key segment portion of the italicized key words in Experiment III
9.
a. b. C.
10.
a. b.
11.
a. b. C.
12.
a. b.
If the If the If the If the (PIP, If the (PFP, If the (PIP, If the (PFP, If the (PFP, If the (PIP, If the (PFP,
pilot jEes the plane we’ll surely crash. (PIP, TR) pilot flies the plane will surely crash. (PFP, TR) parrot fries the boy will feed him cake. (PFP, IN) baby parakeet flies to Lisa we’ll be happy. IN) baby parakeet flies Teresa will be happy. IN) tailor dyes the cloth we’ll refuse to buy the suit. TR) tailor dyes the cloth will refuse to hold a crease. TR) tailor dies the cloth in his shop will all be sold. IN) tailor dies in the summer his shop will be sold. IN) tailor dies in the summer his shop will be sold. IN)
219.7 326.3 330.7
38.0 32.3 41.7
210.0
46.5
297.4
40.8
231.7
44.9
323.9
57.6
338.6
39.5
234.3
34.1
369.9
34.9
0.001, t = 10.6, df = 9; two-tailed t-tests for matched pairs). Again, this lengthening must be attributed to the influence of constituent position, since any effects produced by grammatical category type and Object Deletion are neutralized. The magnitude of the lengthening in Pair (10) was 42% and in Pair (12) it was 58%. Two additional experiments have been conducted to extend the results of Experiment III. One study involved five test sentence pairs and ten speakers. The Verbs were placed in constituent-final and non-constituent-final positions. Two of the sentence pairs appear below with the key words in italics: 13. 14.
a. b. a. b.
If graduate students teach the class we’ll complain to the chairman. If graduate students teach the class will complain to the chairman. After we talked to Rita we went to class. After we talked Teresa went to class.
All five sentence pairs showed significant lengthening of the Verb in constituent-final position: 0, < 0.001 for all pairs, t values ranging from 4.98 to 10.5, df= 9; two-tailed t-tests for matched pairs). The average percent lengthening for key segments ranged from 21% to 49%, with a mean lengthening of 35%. A second study involved an additional four sentence pairs and ten
146
John M. Sorensen,
speakers. below: 15.
a. b.
William E. Cooper, Jeanne M. Paccia
Two of the pairs were ambiguous
sentences,
one of which is shown
If you can coaclz naturally you can join our team. (If you can coach in a natural way). If you can coach naturally you can join our team. (Of course you can join if you can coach).
All pairs showed lengthening of the Verb in constituent-final position (from 16% to 50%) that was significant (p < 0.001 for all pairs, t values ranging from 4.81 to 7.63, df = 9; two-tailed t-tests for matched pairs). Taken together, the results of Experiments I - III support the notion that words are lengthened in constituent-final position. Furthermore, this lengthening can be predicted on the basis of constituent position, without considering possible effects of (1) the inherent length of a word based on its membership in a certain major grammatical category, or (2) the deletion of an object following a Verb, leaving the Verb in constituent-final position. The findings of Experiment II also suggest that a single rule of lengthening is appropriate for constituent-final lengthening with both Nouns and Verbs. Therefore, the distinction between Nouns and Verbs need not be specified in a first-order theory of speech timing or in rules for speech synthesis.
EXPERIMENT
IV
In this experiment we extended our study to the major categories of Adjective and Adverb. The durations of Adjectives and Adverbs were compared by measuring /tu/, the phonetic form of the English homophone pair: two Adjective and too - Adverb. On the basis of the results of Experiments I - III, we suggest that no difference exists in the inherent duration of Adjectives and Adverbs4. Since typical English sentences contain two as an NP-initial Adjective and too as a constituent-final Adverb, lengthening of the Adverb can probably be accounted for by constituent-final position, without resorting to accounts of inherent length based on category type. If constituentfinal position acts to lengthen segments, too should exhibit longer segment durations than two. This prediction is in full accord with intuition. 41t is Adjectives /tu/ as an However, introduced Adjective.
conceivable, however, that there may be some smaIl difference in the inherent duration of and Adverbs. Such a difference could be assessed by examining the durational difference of Adjective and Adverb in matched constituent position, as in: I left him ~‘0; I left him too. the above sentences may not bc adequate to test this hypothesis since another variable is - specitically, speakers typically insert a pause before the Adverb, but not before the
Speech timing of grammatical categories
147
Method Subjects
Ten M.I.T. undergraduates participated as paid volunteers in this experiment. One of the subjects had participated in Experiment I. The nine new subjects had the same qualifications as those in Experiment I. Sentence
Materials
Two pairs of test sentences and five fillers were constructed for this experiment. In Pair 16, the key segment was bounded on the left by a word-final obstruent /k/ and on the right by a word-initial Is/. In Pair 17, the key segment was bounded on the left by a word-final vowel /o/ and on the right by a word-initial /p/. The test sentences appear below, with the key word in italics. 16. 17.
a. b. a. b.
Joey Joey John Mrs.
signed the check too seemingly nervous. gave Monique two slippers for Christmas. should make Kathy go too peacefully if possible. Scott offered Joe two pieces of her homemade pie.
Procedure
The testing and data analysis procedure was identical to that in Experiment II. The phonetic environment in pair 16 occasionally led to an overlapping of the vowel portion of /tu/ and the following word-initial /s/. In such cases, the offset of /tu/ was marked at a point where the /s/ frication no longer appeared to be modulated in a semi-periodic manner (cf., j7ies in Experiment III). The error estimate for these cases was f 10 msec.
Results and Discussion The mean segment durations for the key segment for each sentence are presented in Table 4. It can be seen that the duration of /tu/ as an Adverb was longer than as an Adjective. This difference was statistically significant for bothpairs: (16a)versus(l6b),p
148
Table
John M. Sorensen,
4.
William E. Cooper, Jeanne M. Paccia
Mean durations and standard deviations of the italicized key words in Experiment IV d (msec) 16. 17.
EXPERIMENT
a. b. a. b.
Joey Joey John Mrs.
signed the check too seemingly nervous. gave Monique fwo slippers for Christmas. should make Kathy go too peacefully if possible. Scott offered Joe two pieces of her homemade pie.
248.0 156.3 258.0 137.3
43.7 20.9 41.4 26.8
V
Thus far, we have studied segmental lengthening of words in major grammatical categories. In this final experiment, we extended our focus to the distinction between major and minor grammatical categories. Phonological reduction is a distinguishing property of words belonging to minor grammatical categories (Prepositions, Determiners and Conjunctions). In particular, a word belonging to any of these categories may undergo vowel reduction in casual speech. In this experiment, the major category Adjective was compared with the minor category Preposition by placing the homophone pair two and to in sentences matched for phrase position of the key word. It was expected that to would be much shorter than two, since the former is phonologically reducible (Chomsky and Halle, 1968).
Method Subjects
Ten subjects participated in this experiment. Eight of the subjects had participated previously in Experiment III. Of the two new subjects, one was an M.I.T. undergraduate and one was an M.I.T. employee. Both had the same qualifications as all other subjects. Serltetlce
Muterials
Four pairs of sentences were constructed for this experiment. They are shown below, with the key segment in italics. (See note 5, page 149.) 18.
a. b.
We heard Janice read two poems at the literary convention. We heard Janice read to poets at the literary convention.
Speech timing of grammatical categories
19.
a. b.
20. 2 1.
a. b. a.
b.
149
I saw John run two kilometers last night. I saw John run to Canaveral last night. Alice said she must write two pages this afternoon. Alice said she must write to Patty this afternoon. Ted watched the couple walk two blocks down the street. Ted watched the couple walk to Bob’s down the street.
Procedure The procedure
was identical
to that in Experiment
II.
Results and Discussion The mean segment durations for all key segments are shown in Table 5. The duration of to was significantly shorter in every case: p < 0.001 for all Pairs, t = 9.70 in Pair (18), t = 12.6 in Pair (19), t = 8.83 in Paii (20), and t = 9.72 in Pair (2 l), df = 9, two-tailed t-tests for matched pairs. The percent of shortening was approximately 50% in all pairs, as shown in Table 5. The results Table 5.
Mean durations and standard deviations of the italicized key words in Experiment V
18.
19. 20. 21.
a. We heard Janice read two poems at the literary convention. b. We heard Janice read fo poets at the literary convention. a. I saw John run two kilometers last night. b. I saw John run to Canaveral last night. a. Alice said she must write two pages this afternoon. b. Alice said she must write fo Patty this afternoon. a. Ted watched the couple walk n~o blocks down the street. b. Ted watched the couple walk fo Bob’s down the street.
X (msec)
S,
146.6
32.5
71.2 139.2 71.5 131.0 72.7
18.2 25.8 15.8 29.0 17.3
126.4
25.9
65.0
18.8
‘Sentence Pairs (18), (20), and (21) contain ambiguous sentences. Sentence (21a), for example, can take either the reading “Ted watched the couple walk a distance of two blocks down the street” or “Ted watched the couple walk and they were two blocks down the street at the moment he watched”. In general, however, it has been shown (Cooper, 1976) that ambiguities involving semantic relations do not influence speech timing, and it is reasonable to assume that any effect due to ambiguity here is minuscule in comparison with the large effect on duration due to type of grammatical category.
150
John M. Sorensen, WilliamE. Cooper, Jeanne M. Paccia
thus confirm our intuition that words belonging to minor grammatical categories are shorter in duration than words of major categories. The large difference in duration noted here indicates that the binary distinction between major and minor categories must be included in a theory of speech timing and in rules for speech synthesis.
General
Discussion
The results of these experiments indicate that segmental lengthening occurs for words in major grammatical categories in constituent-final position. The longer duration of Nouns versus Verbs was shown not to be dependent on grammatical category type per se, but rather on the difference in constituent position which these two category types typically occupy in English sentences. Constituent-final lengthening was also shown to account for durational differences between Adjectives and Adverbs. Significant differences in duration were documented for major versus minor category types, suggesting the need for a binary distinction between these two classes of categories. Informal observations of Romance languages also suggest that constituent position plays a role in determining the duration of a word belonging to a given grammatical category. For example, in French, it appears that Adjectives are longer in postnominal position, though they are also typically accompanied by a change in meaning. In Spanish, Carmen Egido has pointed out to us that a phonemic contrast exists between semantically equivalent forms of the Adjective good when these occur in prenominal versus postnominal position, as in: Juan es un buen nifio. Juan es un nmo bueno.
(John is a good boy.) (John is a good boy.)
The longer version of the Adjective occurs in constituent-final position, consistent with the results presented here. Further studies with other languages might be aimed at testing whether differences in duration among words in major grammatical categories are attributable to constituent position. By using the terms “lengthening” and “shortening” we have implied the existence of a fixed reference for durational rules. We postulate that the reference duration for a word can be defined as the duration of that word when it occurs as a major grammatical category in non-phrase-final position. For example, we considered the reference duration of /tu/ in Experiment IV to be the duration of two as it typically occurs in non-phrase-final position. Thus, the durational difference between this form and too, when it occurs as a clause-final Adverb, was described as “lengthening”. On the other hand, the
Speech timing of grammatical categories
151
durational effect for the Preposition to, a minor grammatical category word, was described as “shortening”. This classification of durational effects is in accord with our intuition about the underlying determinants of rules for speech timing. Major category words are lengthened in constituent-final position as a result of a marked slowdown of the speech processing machinery, a by-product of either general relaxation or the speaker’s planning of an upcoming constituent (Cooper and Paccia, 1977). The shortening of minor categories may be attributed to the very low information load carried by words belonging to this class (Coker, Umeda and Browman, 1973). The results of this study carry implications for speech synthesis as well as a theory of speech production. In general, stateof-the-art speech synthesis suffers from the lack of a set of timing rules which can predict timing relations based on sentence structure. Our findings indicate that rules for word durations in speech synthesis should include the specification of major versus minor grammatical categories and constituent position: In addition, our data suggest that there is no need to specify individual category labels (e.g., Noun, Verb, Adjective) for words belonging to major categories. In addition to rules covering these durational effects, special synthesis rules must be employed to generate proper duration for words receiving contrastive or emphatic stress. As demonstrated below, contrastive stress can be applied to words belonging to either major or minor categories: 22. 23.
a. b. a. b.
“Is Alice going “No, Alice said “Is Alice going “No, Alice said
to write three pages?” she must write TWO pages this afternoon.” to write a letter for Patty?” she must write TO Patty this afternoon.”
Finally, we wish to consider other data from speech production which can be accounted for in terms of constituent position. As noted earlier, an account of durational differences among major categories in terms of constituent position is particularly satisfying because such position is already a required property of a theory of speech processing. Evidence from studies of optional pausing (Downing, 1970; Stockwell, 1972; Cooper, 1977) indicates that syntactic pauses are typically inserted at major boundaries. Grosjean and Lane (1977) have also shown similar effects of pausing in American Sign Language. In addition, phonological rules which normally operate across word boundaries may fail to do so in the presence of strong intervening syntactic boundaries (Cooper et al. 1977). Furthermore, it has been shown that major syntactic boundaries within a sentence are accompanied by a fall-rise contour in fundamental frequency (Lea, 1972; Cooper and Sorensen, 1977). It appears that a unitary syntactic code of the speaker influences each of these super-
152
John M. Sorensen,
William E. Cooper, Jeanne M. Prtccia
fically unrelated phenomena, blocking, and F, inflections.
including
segmental
lengthening,
pausing,
References Chomsky, N. (1965) Aspecrs offhe Theory ofSyntax. Cambridge, Mass., MIT Press. Chomsky, N. and Halle, M. (1968) The Sound Pattern ofEnglish. New York, Harper and Row. Coker, C. H., Umeda, N., and Browman, C. P. (1973) Automatic synthesis from ordinary English text. IEEEAudio Electroacoust., AU-21, 2933297. Cooper, W. E. (1976) Syntactic Control of Timing in Speech Production. Ph.D. Thesis, M.I.T., Cambridge, Mass. Cooper, W. E. (1977) Syntactic-to-phonetic coding. In B. Butterworth (ed.) Language Production. London, Academic Press, in press. Cooper, W. E., Lapointe, S. G., and Paccia, J. M. (1977) Syntactic blocking of phonological rules in speech production. .I, acoust. Sot. Am.. 61, 131441320. Cooper, W. E. and Paccia, J. M. (1977) Syntax and Speech Coding. In preparation. Cooper, W. E. and Sorensen, J. M. (1977) Fundamental frequency contours at syntactic boundaries. J. accoust. Sot. Am., 62, 683-692. Downing, B. T. (1970) Syntactic Structure and Phonological Phrasing in English. Ph.D. Thesis, University of Texas, Austin, Texas. Fry, D. B. (1957) Duration and intensity as physical correlates of linguistic stress. J. acoust Sot. Am., 27, 765-768. Grinder, J. (1971) Chains of coreference. Ling. Znq., 2, 183-202. Grosjean, F. and Lane, H. (1977) Pauses and syntax in American Sign Language. Cog., 5, 101-117. Huggins, A. W. F. (1969) A facility for studying perception of timing in natural speech. Q. F’rog. Rep.M.I.T. R. L. E., 95, 81-83. Huggins, A. W. F. (1974) An effect of syntax on syllable timing. Q. Prog. Rep. MIT. R. L. E., 114, 179-185. Klatt, D. H. (1975) Vowel lengthening is syntactically determined in a connected discourse. J. Phonet. S, 129-140. Klatt, D. H. (1976) Linguistic uses of segmental duration in English: Acoustic and perceptual evidence. J. acoust. Sot. Am., 59, 1208-1221. Lea, W. A. (1972) Intonational Cues to the Constituent Structure and Phonetics of Spoken English. Ph.D. thesis, Purdue University, Lafayette, Ind. Lightfoot, M. J. (1970) Accent and time in descriptive prosody. Word, 26, 47764. Lindblom, B. and Rapp, K. (1973) Some temporal regularities of Spoken Swedish. Papers from the Institute of Linguistics, University of Stockholm, Publication 21. Martin, J. G. (1970) On judging pauses in spontaneous speech. J. verb. Learn. verb. Beh., 9, 75-78. Ross J. R. (1970) Gapping and the order of constituents. In M. Bierwisch and K. E. Heidolph (eds.) Progress in Linguistics. The Hague, Mouton. Sampson, G. (1972) A proposal for constraining deletion. Lingua, 29, 23-29. Stockwell, R. P. (1972) The role of intonation: Reconsiderations and other considerations. In D. Bolinger (ed.) Intonafion. London, Penguin Books. pp. 87-109.
Speech timing of grammatical categories
153
RPsumk Une s&e d’experiences a et& realisde pour determiner comment la dunk d’un mot pronon& dans une phrase est influencee 1) par la categoric grammaticale h laquelle ce mot appartient et 2) par la position de ce mot dans un constituant. Dans l’expdrience I des homophones Nom-Verbes (ex., “I saw the couch” - Nom (J’ai un roti) “I saw him couch” - Verbe (ie I’ai roti)) sont present& dans des phrases dont on a appareille le contexte phonetique et le schema d’accentuation. Les resultats indiquent que les noms sont plus longs que les verbes darts des phrases typiques. Dans l’experience II cependant, on trouve une duke approximativement &gale pour les homophones Noms et Verbes quand ceux-ci sont en positions ternkales dans des clauses appareilldes. Les resultats des experiences II et III appuient une interpretation liant l’allongement i la position terminale du constituant et &mine l’effet dfi i un emplacement ou l’effacement serait possible. Les resultats de I’experience N etendent l’interpretation don&e pour l’allongement du constituant final des Noms et Verbes i deux categories supplkmentaires s’appuyant sur la comparaison de la duke de l’adjectif du syntagme initial rwo (deux) et de l’adverbe du syntagme too (aussi). Entin I’experience V teste la distinction entre categories mineures et majeures. Les resultats montrent que la preposition to (a) est a 50% plus courte que l’adjectif two (deux). En prenant l’ensemble des resultats on voit qu’il suftit d’une distinction binaire entre les categories majeures et mineures pour ce que demande une theorie de la mesure temporelle de la parole et de sa synthese. On peut rendre compte de duke traditionnellement attribuee aux differences entre les classes de categories majeures en termes de frontieres de constituants deja requise dans une theorie qui rend compte de trois autres classes de phenomknes.
Cog&ion, @Ells&et
6 (1978) 155-168 Sequoia S.A., Lausanne
- Printed
On the acquisition
in the Netherlands
of pronouns
Ihr, dir, or mir ? in German children*
WERNER DEUTSCH** Max-Planck-Gesellschaft
THOMAS Universit&
PECHMANN Marburg/Lahn
Abstract The present study tested the hypothesis that the linguistic complexity of pronouns corresponds to the order in which children acquire them. Linguistic complexity was defined by three principles of linguistic contrast, namely the proximal-nonproximal, singular-nonsingular, and speaker-nonspeaker contrasts. In an experimental naming task, 55 German children aged .3;5 to 6;5 were asked to express the possessive relationship between different participants in a communication situation and particular objects by means of personal pronouns in the dative case, used from the speaker’s point of view, The results showed a strong correspondence between the predicted and actual order of correct use of pronouns, and provide evidence for the precedence of the proximal-nonproximal over the singular-nonsingular contrast.
1. Introduction Like lzere and there or come and go, pronouns are deictic terms, They differ from other terms, proper names or generic names for example, in their *This research was supported by the “Max-Planck-Gesellschaft, Projektgruppe Psycholinguistik” (Nijmegen, The Netherlands), granting a research scholarship to the first author at Stanford University. We wish to express grateful appreciation to Eve Clark for her encouragement, criticism and many helpful suggestions during preparation of the manuscript. Thanks are also due to Erika Barthelmey and Allmuth Weddige for their help in running the study, and to the staff and the children of the kindergarten “St. Martin”, Fritzlar (Hessen). We are indebted to an anonymous reviewer for the information concerning non-European languages. **Reprint requests should be sent to Dr. Werner Deutsch, Max-PlanckCesellschaft, Projektgruppe Psycholinguistik, Berg en Dalseweg 79, Nijmegen, The Netherlands.
156 Werner Deutsch and Thomas Pechmann
reference. As among others Jakobson (1957) and Clark (1977) have pointed out, deictic terms are characterized by shifting reference, illustrated in the following example. The possessive relation between a person and a book can be captured by “My book”, if the owner of the book produces the utterance, but it has to be changed to “Your book”, if the owner hears the utterance from someone else. However, if the relation is expressed by “Bill Smith’s book”, one can disregard the specific situation where this utterance was produced or heard because the reference to Bill Smith remains constant despite changing situational conditions. If the speaker is to avoid confusion and misunderstanding in his uses of pronouns he must take into account certain aspects of each communication situation. Which aspects are the essential ones? Fillmore (197 1) mentions three that depend on the different positions of participants in a verbal communication situation: (1) The speaker who produces the utterances containing pronouns; (2) the addressee to whom the verbal information is directed; (3) the audience consisting of one or more listeners who can hear the speaker’s utterances, but are not intentionally addressed by him. The aim of the present study is to find out how successfully children in two age-groups can use pronouns in order to refer to participants in a communication situation. Is the child able to refer to himself in the speaker’s position and his addressee by using the plural pronoun “us”, for example, before he is able to talk about some third person by using the singular pronoun “him”? Or does the child acquire all the singular pronouns before he masters any plural ones? A theoretical answer to such questions is attempted by three principles of linguistic contrast, based on the positional structure of a communication situation. From them, one can derive an order of complexity for the correct use of pronouns. Linguistic complexity has already proved a good predictor of order of acquisition in studies of dimensional adjectives (Donaldson and Balfour, 1968; Donaldson and Wales, 1970) and kinship terms (Haviland and Clark, 1974). The present study tests the hypothesis that the theoretically derived order of complexity in pronouns accounts for their order of acquisition.
2. Linguistic
complexity
in pronouns
The linguistic complexity of pronouns can be defined by three principles of contrast, illustrated in the following situation. Imagine four people: a speaker S, an addressee A, and an audience consisting of two other people,
Ihr, dir, or mir?
Figure 1.
157
Positional structire of a communication situation
0, and 02, where 0, is male and 0, female. The speaker in this situation is the source of the information being conveyed to the addressee by means of different pronouns (see Fig. 1).
Principle
I: Proximal-Nonproximal
contrast
Metaphorically speaking, this principle establishes a boundary between two areas in the positional structure with S and A on one side and 0, and 0, on the other (cf., Lyons, 1968; Fillmore, 1971). What justifies this distinction? In natural settings S and A are normally closely connected with respect to their mutual intentions and their focus on each other during communication. The relation between S and A, and their separation from 0, and 02, is often indicated by spatial proximity. The physical distance between S and A is likely to be smaller than that between either 0, or O2 and S or A. Eye contact is also more likely between S and A than between either 0 and S or A. These considerations support the assumption that from S’s point of view he and his addressee serve as the two basic reference points in a communication situation. To name himself (S), his addressee (A), or both of them together (S + A) using pronouns, requires a congruity between the content of information and the basic reference points between whom this information is conveyed. But if the speaker names 01, or 02, or both, he omits both basic reference points, since the content being conveyed is not congruent with the people immediately engaged in the exchange of information. An intermediate status can be ascribed to the naming of relations that connect either S or A and 0, or Oz. These expressions should be more complex to derive and apply than proximal terms since they include a nonbasic reference point, but less complex than the third person terms on their own, since the speaker can still rely on the basic reference points.
158
Werner Deutsch and Thomas Pechrnann
Principle 2: Speaker-nonspeaker contrast This principle introduces a distinction that makes a further differentiation within the proximal and intermediate categories and thus serves as a further specification of the first principle of contrast. Within the two basic reference points there is a preference for the speaker’s own position. This speaker bias has as its consequence that naming the speaker should be less complex than naming the addressee on the one hand, and naming a connection from the speaker’s position should be less complex than from the addressee’s position, on the other. Support for this principle comes from several sources. First, Clark (1977) found in an analysis of many diary studies that a child first learns to refer to himself: the correct use of I precedes You. By age of 3 most children use I and _VOUappropriately. Secondly, further evidence for the speaker bias is provided by a study of the deictic verbs come and go in English. As Clark and Garnica (1974) pointed out, the correct rule of application for the speaker’s use of cor?ze and go emerges before the addressee’s use of these verbs. Principle 3: Sirzgular-nonsingular contrast This principle assumes that naming a single person is less complex than naming a conjunction of two people, especially if the conjunction is or can be expressed with a single word. This assumption is supported by evidence from diverse areas. In concept formation tasks, for example, classifications based on a conjunction of attributes are harder to discover and to learn than a single attribute classification (Goede and Klix, 1971). Similarly, Piaget has found that the ability to classify simultaneously a conjunction of attributes requires concrete-operational thinking while single attribute classifications are mastered earlier, during the preoperational stage of thinking (Piaget, 1947). Finally, Cazden (1968) reported that in language acquisition, children master singular forms before plural ones. The three principles of contrast, used in combination, lead to the following order of complexity (A) (A)
[(Q + WI
> [(A + 0,); (A + @)I > [(S + 0,); (S + WI
>I@ + A)1 >
I(S)I < I(A)1 < [CO,); (WI. Arrangement (A) does not make a distinction between all the items in question as the dominance relation between proximality and singularity can hardly be clarified on the basis of the complexity hypothesis. If one would assume a hypothetical precedence of proximal/non+proximal over singular/plural, arrangement (A) could be revised into the order expressed in (B).
Ihr, dir, or mir?
(B)
[CO, + WI > [O);
159
(WI > [(A + 0,); (A + WI > [(S + 0,); (S + Odl > [(S + A)1 > [(A)1 > [WI
If one would make the reverse assumption that singularity takes precedence over proximality, arrangement (A) would have to be changed into (C) (C)
[(O, +
WI > [(A + 0,); (A + WI > [(S + 0,); (S + WI > t(S + A)1 > t(Q); (WI > [(A)1 > t(S)1
However, there seems to be no obvious a priori reason for either (B) or (C). Thus, this issue waits for an empirical solution that might be provided by the data of our experiment. This analysis is still incomplete as it covers only some of the possible factors that determine the complexity of pronouns. Other factors such as the gender of 0, and O2 as well as the nature of conjunctions involving more than two people have not been discussed, since they raise theoretical problems for which there is at present no clear solution. For example, is the singular-nonsingular contrast a qualitative distinction which needs no further differentiation with respect to plural forms, or is an additional contrast necessary within the plural category depending on how many people are involved in a particular conjunction? As it seems very difficult to answer these questions in a reasonable way, a priori, the present study will concentrate on those cases that are theoretically clear cut. Moreover, alternative forms for pronouns referring to the same participant will generally be regarded as equivalent. However, this study will include one theoretically ambiguous case which is of special interest, namely the use of an inclusive pronoun designating all the participants in the situation. One could argue that the correct use of such a pronoun might be the most difficult of all since it requires the most complex conjunction, connecting S, A, O,, and Oz. Equally plausibly, one could argue that such an expression might be the simplest of all the conjunctive terms, since it does not require any distinctions to be drawn among the different participants. In one sense, an inclusive pronoun like that seems more like a singular term than a nonsingular one. Table 1 contains all the pronouns used in the present study, together with their English equivalents. Each pronoun appears in the dative case, as required by the experimental task that was used. 3. The Experiment Subjects
The subjects were 55 children, divided into two age groups: 3;5 - 5;4 (N = 29) and 5 ;5 - 6;5 (JV = 26). All the children were attending a German kinder-
160
Werner Deu tsch and Thomas Pechmann
Table 1.
The pronouns
used in the experimental
task -
Pronourl
Referents among the participants
__.-
in the communication
German (dative case)
equivalen
mir dir ihm ihr uns 1
me
S
You him her us us
A
uns* euch ihnen unsj (inclusive)
You them Gclusive)
situation
1
01 02 S+A s+o,;s+o* A+0,;A+02 Olf02 s+‘4+0,
+02
garten in Fritzlar/Hessen. The naming task was played with each child individually by a female student familiar to all the children. She was not informed about the expected order of acquisition until all the data had been collected. Procedure The naming task had a structure similar to the one in the communication situation described above, with four people represented in the game by four dolls. The child was always responsible for the speaker doll (S), directing utterances, on the doll’s behalf, toward an addressee doll (A), for which the experimenter was responsible. The listeners were one male and female doll (0, and 0,). S and A were placed facing each other, while 0, and 0, were some way away from S and A, but looking at them (see Fig. 1). The game involved two packs of cards. The first one contained four cards, each with a different picture of an animal on it. Each participant received one card, placed face up in front of the appropriate doll. After the child had been told about the connection between each card (representing an animal) and the doll it went with, the second pack of cards was introduced. These cards had pictures of a single animal, or a combination of two, three, or four of the animals depicted on the first pack of cards. The game was played as follows: each card turned up from the second pack belonged to the person(s) whose cards showed the same picture. If the card had two animals on it, it belonged to the two dolls whose cards matched those two, and so on. During the test trials the addressee (the experimenter) showed the child one card at a time from the second pack and
Ihr, dir, or mir?
161
asked him to indicate the possessive relationship by completing a sentence like “This card belongs to...?” This procedure should result in children’s spontaneously producing personal pronouns to express the possessive relationship. Since names and other labels were not introduced earlier, children were not expected to use them. If the child used something other than a personal pronoun in the dative case, he was encouraged to change his label; he was also encouraged to change the form used if, instead of a oneword pronoun like US, he used a decomposed form like me and you. Each child received at least eight trials, one each for S, A, S + A, S + 0, A + 0, 0, + 02, O,, O2 and A + S + 0, + 0,. Alternative realizations for combinations involving one 0 were chosen randomly either from 0, or 0,. A few cards with three pictures on them were also included, but not in a systematic way.
Results Each utterance which the children used to express the relation between cards and owners was recorded. The aim of the first analysis of the data was to compare the theoretical complexity and the actual order of difficulty. Each utterance was therefore scored according to whether it contained the correct personal pronoun as one word in the dative case (scored as one) or not (scored as zero). These binary data were analysed using Bart & Krus’ (1973) theoretic ordering method. This method allows the identification of inherent structure among items (here, pronouns). The rationale for the method is as follows: within a defined set of items the relations between all pairwise combinations of items are examined and tested for whether the relation can be assessed as prerequisite, equivalent, or independent. An item i is prerequisite to an item j if the number of subjects who did not solve item i but solved item j is less than or equal to a present tolerance level of error. The zero-one response pattern for an item i and an item j is viewed as a disconfirmation that item i is a prerequisite to item j. If the response pattern zero-one as well as the pattern one-zero occurs at a frequency less than or equal to that established by the tolerance level, the two items are said to be equivalent. They are independent if more subjects solve item j than would be accepted by the tolerance level and if the same holds for the number of subjects who solved item j but not item i. A tree-diagram can be used for the simultaneous representation of all the existing relations, showing the structure of the set as a whole. A statistical significance test, however, has not yet been devised. The second step in the data analysis was
162
Werner Deutsch and Thomas Pechmann
concerned with differences between the two age-groups. The empirical structures of pronoun pair relations in each group were analyzed and compared, and then an analysis of errors in usage was undertaken in order to find out which types of errors occurred, which pronouns they were related to, and what differences there were between the two age groups. Figure 2 shows the percentage of correct uses for each pronoun collapsed over age groups. The pronouns are listed on the horizontal axis. Figure
Relative frequency
2.
of correct responses.
1.00 .90 80 f(x)
70 60
Relat we
5.
frequency 40 30 20
mar
(-5)
dir (A)
““5, 6%
uns2 CC&,
euch CA%,
Ihm
(0,)
Ihr CO+
lhnen
_.
(0, +Oz)
um3(~ncl)
/-.
(S+A+Oy02)
The results in Figure 2 support the predictions of the complexity hypothesis in general, and provide clear evidence for precedence of the proximality over the singularity principle (arrangement B) in particular. Of the two possible exceptions to the arrangement (B), only one is really unexpected. First, although the use of “mir” (S) should be less complex than the use of “dir” (A), all the children used both these pronouns entirely correctly. This is obviously due to the age of the children tested, since the diary studies reviewed by Clark (1977) provide evidence that both I and JWU are usually fully mastered by the age of three. Since the children in the present sample were all over three, the present results are not incompatible with the diary observations and should not be regarded as surprising. The second exception appeared more serious than the first, since a difference between the masculine and feminine forms of 0 was not predicted. Overall, correct use of the masculine pronoun appeared to be easier than the feminine one. The data
Ihr, dir, or mir?
163
also allowed the location of the inclusive US among the pronouns as a set. Note that the complexity of this pronoun was theoretically ambiguous. The results showed that it was more difficult than “mir” (S) and “dir” (A), but easier than all the other pronouns. As a matter of fact, inclusive US can be regarded as a holistic unit formed by the undivided set of participants in the communication situation. Figure 3 presents the tree-diagram for the actual structure of pronouns that resulted from application of the theoretic ordering method to the data. The notation -+ is to be read: all items above that sign are preceded by all items below that sign; and +--+ is to be read: two items connected by that sign are equivalent. Figure 3.
Actual ordering of pronouns (tree structure) for the whole sample.
Tolerance
Again, close: 25 the actual cribed in
Level
= 3.4 7.
the fit between arrangement (B) and empirical structures is very of the 28 predicted relations between pairs of pronouns occur in ordering of the data. The only exceptions have already been desrelation to Figure 2, namely the equivalence between “mir” (S)
164
Werner Deutsch and Thomas Pechmann
and “dir” (A), the nonequivalence (0,) with “ihm” preceding “ihr”, “euch”. Relative frequency
Figure 4.
relation between “ihm” (0,) and “ihr” and the independence of “ihm” from
of correct responses in both age groups
q age q age
100
.90
5,5 -6;5
N:26
2;5-5;L
N-29
80 70 60 f(x) .50 Relative frequency
.‘O 30 .20 10
m,r
(5)
1 ““ST
euch
Ihm
ihr
lhnen
1
uns(C inc
Figure 4 presents the relative frequencies of correct usage for each of the two age groups. The empirical order of pronouns is nearly the same in both age groups. The fact that the relative frequencies of “euch” (A + 0) and “ihm” (0,) do not differ is the only marginal exception. Both orders have a very close correspondence to the arrangement (B). In both groups the deviations are the same, namely no difference between “mir” (S) and “dir” (A), and an asymmetry between “ihr” (0,) and “ihm” (0,). Except for the pronouns “mir” and “dir”, the younger children gave fewer correct responses to each pronoun than the older ones did. Figure 5 presents the tree-diagrams for the data from each age group. The left-hand panel of Figure 5 contains the empirical structure for the younger children, the right-hand one the data for the older ones. The structure of the younger group is identical to the structure of the overall ordering (Figure 3). The structure of the older group shows a somewhat different version of the gender problem: “ihr” (0,) is independent of “ihnen” (Or+ O,), and “ihr” (0,) follows “ihm” (0,). In addition, another exception provides the equivalence relation between “unsl” (S + A) and “uns2” (S + 0).
Ihr, dir, or mir?
Figure 5.
165
Actual ordering of pronouns for the two age groups. Younger Group
Older
Group
An analysis of the incorrect responses children gave showed the following types of errors: 1. A demonstrative pronoun (e.g., dieser da (“that one”) used alone or in combination with a personal pronoun, particularly “mir” (S) or “dir” (A). In both age groups this type of error only occurred when the possessive relation included at least one of the third person participants (0, or 0,). 2. Instead of a one-word personal pronoun, a ‘decomposed’ form consisting of a conjunction of a singular pronoun. This type of response usually replaced the pronoun “unsr” (S + A). 3. Names used as singular terms or in combination with personal pronouns. In both age groups this type of error only occurred when at least one 0 was involved. 4. No verbal utterance at all, or else a misleading utterance (containing an incorrect pronoun). Although the patterns of errors were very similar in both groups, there seemed to be two differences. First, the younger age group produced more type 4 errors than the older one. Secondly, the relation between type 1 and type 3 errors was quite different in the two groups, particularly with respect to the plural forms of pronouns. The younger group obviously preferred
166
Werner Deutsch and Thomas Pechmann
type 3 with a ratio of 1.5 : 1, while the older group preferred type 1 with a ratio of 0.6 : 1 between name-type and demonstrative-type errors.
4. Discussion This study offers support for the complexity hypothesis from the domain of personal pronouns. Moreover, it gives empirical evidence for the dominance relation of the proximality over the singularity principle. The data from the experimental naming task imply the following order of acquisition: the child masters personal pronouns referring to the speaker or addressee before acquiring expressions for indicating relations between either speaker or addressee and a third person. Later still, the child acquires expressions for a third person alone. Within each ‘stage’ defined by the proximal-nonproximal contrast, the use of plural forms involves additional difficulty (the singular--nonsingular contrast). Moreover, the data also support the speaker bias, insofar as correct use of a conjunction between the speaker and a third person precedes correct use of a conjunction between the addressee and a third person. The equivalence between “mir” (S) and “dir” (A) can be regarded as an age-dependent ceiling effect. The data point to an unexpected asymmetry of the masculine and feminine forms of the pronouns for a third person. A possible explanation for this is offered by Greenberg’s (1966) claim that masculine forms are linguistically unmarked while feminine ones are marked. This explanation is compatible with our general theoretical framework as our notions of linguistic complexity are the same as the notion of nongrammatical linguistic markedness (nongrammatical in the sense of being outside the domain of formal syntax and logical form). Thus, proximal pronouns can be regarded as less marked than nonproximal ones and singulars as unmarked while plurals are marked. It would be of considerable interest if this effect of gender also leads to differences between the correct use of (S + 0,) and (S + 0,), on the one hand, and between (A + 0,) and (A + O,), on the other. This might enable one to decide whether the gender effect depends on characteristics of the verbal expression (i.e., gender differentiation of the word forms) or on characteristics of the participants involved. As these cases were not systematically varied in the present study, this question will have to wait for an answer. The theoretical framework proposed here allows one to derive predictions for other words than personal pronouns in the dative case. By using the
Ihr, dir, or mir?
167
same task, one could find out whether different results would be obtained for personal pronouns in the dative case to mark I;‘ossession versus personal pronouns in the nominative case. Of special interest would be languages which, unlike German, make a morphological clistinction between inclusive and exclusive pronouns, or distinguish three numbers (singular, dual, plural). In languages like Indonesian, for example, it should be possible to test for the different cognitive requirements assumed necessary for the different forms of ‘we’ ((S + A + 0, + O,), (S + A), (S + 0,), and (S + 0,)). In Indic languages, for example, one could test whether the dual would actually be simpler than the plural. This would be expected in terms of the most obvious translation of the singular/plural complexity relation, although that prediction is far from intuitively obvious. Moreover, the notion of proximality in our theoretical approach makes a specific prediction for those languages where pronouns have demonstrative force. In Navajo, for example, the order of acquisition for the three different terms of ‘he’ should be ‘he (here)‘/‘he(there)‘/‘he (over there)‘. But the notion of proximal need not only be taken as being a spatial concept. In Eskimo, for example, it appears as a temporal concept since referential expressions are inflected for “tense” as ‘present’, ‘not present’, and ‘not present now, but was present’. The question of what is the proper arrangement of these “tenses” deserves further research of a developmental sort. What this study has demonstrated is that children have predictable difficulties in using personal pronouns to indicate possession. The data allow the general conclusion that these difficulties do not arise from a failure to understand possessive concepts, defined here by a rule for deciding ownership through matching cards. The major difficulty rather consists of indicating a specific participant in the communication structure in which the utterance is produced. If children are not yet able to take into account the requirements of the communication situation, they will not necessarily fail to indicate the correct possessive relations. Instead, they substitute for the pronouns requested apparently easier expressions. By doing that, they make sure the addressee will be able to identify the appropriate relationship. Among the easier expressions children have recourse to are the conjunction of me and you instead of us, and the use of demonstrative pronouns or names for a third person instead of the appropriate personal pronoun. There were only a few utterances where a personal pronoun was actually used incorrectly, and even where no utterances were produced, one might expect that a child in a natural setting will use gestures instead of words to indicate a possessive relation. In summary, the correct use of personal pronouns for possession requires more than a knowledge of specific possessive relations,
168
Werner Deutsch and Thomas Pechmann
it requires rules for linking that knowledge pant structure in a communicative situation.
to specific aspects of the partici-
References Bart, W. M. and Krus, D. J. (1973) An ordering-theoretic method to determine hierarchies among items. Educ. Psychol. Measure., 33, 29ll300. Cazdcn, C. B.: (1968) The acquisition of noun and verb inflections. Child Devel., 39, 4333438. Clark, E. V. (1977) From gesture to word: On the natural history of deixis in language acquisition. In J. S. Bruner and A. Carton (Eds.), Human Growth and Development: Wolfson College Lectures Oxford. Clark, E. V. and Garnica, 0. K. (1974) Is he coming or going? On the acquisition of deictic verbs. J. Verb. Learn. Verb. Beh., 13, 559-587. Donaldson, M and Balfour G. (1968) Less is more: A study of language comprehension in children. Brit. J. Psychol., 59, 461472. Donaldson, M. and Wales R. (1970) On the acquisition of some relational terms. In R. Hayes (Ed.), Cognition and the Development of Language. New York. Fillmore, C. J. (1975) Santa Cruz Lectures on Deixis 1971. Indiana University Linguistics Club. Goede, K. and Klix, F. (1971) Strategien des Erwerbs von nichtbenannten Begriffen. Z. fiir Ps.Ycho~ogie, 17912, 149-201. Greenberg, J. H. (1966) Language Universalis. In T. A. Sebeck (Ed.), Current Trends in Lingustics, Vol. 3, Mouton, The Hague, pp. 61-112. Haviland, S. 1:. and Clark, L. V. (1974) This man’s father is my father’s son: a study of the acquisition of English kin terms. J. Child Lang., I, 23-47. Jakobson, R. (1957) Shifters, Verbal Categories, and the Russian Verb., Cambridge, Mass. Lyons, J. (1968) Introduction to Theoretical Linguistics. Cambridge. Piaget, J. (1947) Psychologie der Zntelligenz. Zurich.
R&u& Cette etude a pour but de tester l’hypothese d’une correspondance entre la complexitk linguistique des pronoms et l’ordre dans lequel lcs enfants les acquierent. La complexite linguistique a etit 6tablie sur la base de trois oppositions linguistiques: les contrastes proximal-non-proximal, singulier-nonsingulier et locuteur-non-locuteur. On a utilii une &he experimentale de designation au tours de laquelle 55 enfants, allemands, de 3,5 i 6,5 ont eu i exprimer des relations possessives entre diffhrents participants et des objets particuliers. Du point de vue du locuteur, ces relations n’expressaient pas de pronoms personnels au datif. Les resultats montrent une correspondance forte entre l’ordre predit et l’ordre obtenu pour I’utilisation correcte des pronoms, ils fournissent aussi la preuve dune pr&&dence du contraste proximalnon-proximal sur le contraste singulier-non-singulier.
Cognilion, 6 (1978) @Elsevier
Sequoia
169 - 174 S.A., Lausanne
Discussions - Printed
in the Netherlands
Anticipations,
Images,
and Introspection*
ULRIC NEISSER Cornell University * +
I welcome this opportunity to clarify the theory of mental images presented two years ago in Cognition and Reality (Neisser, 1976), and to respond to Hampson and Morris (1978). That theory of imagery is embedded in a more general account of perception, defined as the pickup of information which specifies properties of objects or events (or of the perceiver himself). Perception requires active anticipatory schemata that are attuned to this information, and can direct explorations to make more of it available. Newly-acquired information alters and sharpens the schemata themselves, thereby producing additional exploration and more information pickup. This is the perceptual cycle. Perceptual schemata are of many kinds; the information they pick up may specify meaningful objects and events as well as small details and features. As I shift my gaze across my cluttered desk, for example, I anticipate information that will specify not only edges and corners but objects and surfaces, books and papers, the draft pages of this manuscript. Schematic anticipations thus vary rather as “levels of processing” do in more conventional theories (Craik and Lockhart, 1972). There is an important difference, however. Schemata that have picked up information about local details of objects (say, edges and corners) are not passive conduits for data that some homunculus will eventually interpret. They do not simply send reports of their discoveries to higher levels, but engage in perceptual cycles of their own. “Edge schemata” anticipate more information characteristic of edges, just as “book schemata” direct explorations that may produce more information appropriate to books. Perception is cyclic at many embedded levels of meaningfulness. When an active schema (at any level) fails to find the information to which it is attuned, the character of its activity changes. The state of the perceptual *I am grateful to Elizabeth Spelke for her helpful comments on an earlier draft of this manuscript. **Reprint requests should be sent to Dr. UIric Neisser, Department of Psychology Uris Hall, Cornell University, Ithaca, New York 14853, U.S.A.
170
Uric Neisser
system then becomes a rather peculiar one, and gives rise to a correspondingly peculiar mental experience. This peculiar state certainly is not perception itself’: anticipations are going unfulfilled, and the characteristic informationdriven changes of schemata are not taking place. Nevertheless it may be said to resemble perception inasmuch as the perceptual schemata themselves are active. It is under these conditions, I submit, that we have mental images. Imagery is the inner aspect of perceptual anticipations, of readinesses to perceive. (I should add that my hypothesis applies only to the voluntary images we have when we deliberately imagine something. Hypnagogic “hallucinations” and other unbidden visible phantoms are another matter.) This account of imagining is quite different from the one offered by modern information-processing theories, whether “analogue” or “propositional”. While it resembles the analogue theories in claiming that imagining is genuinely different from other forms of thought (being based on more specifically perceptual anticipations), it does not share their assumption that images are essentially inner objects at which the homunculus can look. The entire organization of the perceptual system is conceived differently in both versions of contemporary theory than in mine. For them, perceptual awareness is not an activity of the system as a whole but depends on a final stage of processing. Earlier stages detect features, form units, and so on; their output is combined with other information from “long term memory” and forwarded to the “central processing unit”. At that point, and only then, the individual consciously perceives objects or events. Sometimes memory becomes active without any sensory input, and sends similar signals to the central processing unit on its own. Again the individual has the conscious experience of an object or event, but now it is an image and not a percept. Analogue and propositional theories differ about what is stored in long term memory, or about what sorts of signals are sent to higher centers, but agree that images are highlevel activities triggered by an internal flow of information. Cogtzitiotz and Reality offers many arguments against theories of this sort: the temporally extended character of perception, the active nature of selective attention, the apparent absence of capacity limitations in practiced subjects, the existence of object perception in infants, etc. The particular argument to which Hampson and Morris object concerns the distinction between imagining something and actually perceiving it. I contend that the information processing theories do not explain how we make this distinction so easily and accurately: they attribute both images and percepts to the same sort of process in the same central unit. Morris and Hampson claim that since even my own theory implies “knowledge of one’s own cognitive processing” (either information is being picked LIP or it is not), I should permit other theories to use this “knowledge” as well. But it is the processing theorists
Anticipation,
Images, and Introspection
171
themselves, and not I, who deny this knowledge to their perceivers. Although the difference between perceiving and imagining is defined at the periphery where receptors are (or are not) triggered by stimulation, they postulate that awareness exists only at the final stage. The central processing unit just knows what its coded inputs report; how can it tell whether an external trigger was originally responsible for those inputs? In my theory, on the other hand, perceiving is the cyclic activity of the whole visual system as it seeks and obtains information. The nature of that activity will evidently depend on whether or not the information is actually available. Hampson and Morris are also disturbed by my assertion that images are “anticipations”. As they point out, Gilbert Ryle long ago advanced this idea in The Concept of Mind (1949). (They are mistaken, however, in believing that Ryle and I reached this conclusion in similar ways. I reached it by considering the experimental evidence on such topics as perceptual set, mnemonic devices, and mental rotation. How Ryle came to it is hard to say, but he does not mention experimental results anywhere.) To show that the anticipation hypothesis is misguided, Morris and Hampson refer to a criticism of Ryle made by Hannay (197 1). When an anticipation goes unfulfilled, according to Hannay, we always experience surprise. Our images do not surprise us, however; hence they cannot be unfulfilled anticipations. But this is simply wrong: many kinds of anticipations go unfulfilled without surprising anyone. A seed is a highly structured set of anticipations - it is ready for the warmth and water and nutrients that will enable it to grow - but no one supposes that seeds are capable of surprise. The American military defense system is (supposedly) always ready for a Soviet sneak attack - it is deployed in anticipation of such an attack, one might say - but it experiences no surprise when the attack fails to materialize day after day. In short, my use of “anticipation” does not intend as much surplus meaning as Hampson and Morris suppose. What I have in mind is a specific state of readiness for specifically perceptual information. Images are anticipations, but not all anticipations are images. Just as perceptual schemata are of many kinds, so images may be of many kinds also. Some people have highly specific images: when they imagine an object, they seem to see it in rich color and fine detail. The imagery of other individuals is less sharply defined: in imagining an object they may be aware of its general size or position or potential function without any commitment to particular features. All these voluntary images are perceptual anticipations, I believe, but the kinds of information they anticipate are not equally specific. This has led to certain problems of definition. Some people prefer to reserve the term “image” for experiences that involve a high level of detail, which they may also describe as appearing especially “vivid”. I am adopting a differ-
172
Uric Neisser
ent usage, in which “image” refers to quasi-perceptual experiences of every level of meaningfulness and specificity, This broad usage is necessary if we are to avoid renaming most contemporary laboratory studies of “imagery” (mental rotation experiments, studies of mnemonics, etc.). The subjects in those experiments often manipulate only very general and non-detailed images; there is no convincing evidence that their “vividness” makes the slightest difference. I maintain, however, that all of them reflect the activation of anticipatory schemata, and would facilitate the perception of the corresponding object if they were suddenly to appear. Perhaps the sharpest of the criticisms leveled by Hampson and Morris concerns my treatment of introspection; they think I am denying what experience affirms. I hope I am not, but I do believe that introspection is a complicated affair. Because it is complicated, young children do not know how to do it. Such a child sees objects and events, and is conscious of them; he can also perceive his own position in space and his own movements. Nevertheless, he does not “know that he is perceiving”. He can see a chair - its position, its appropriateness for sitting, its distance and direction from him -- but he is not aware that he, a person with a particular history and character and probable future, is seeing the chair. He can imagine the chair as well, and does not confuse his perceptual anticipations with the real thing: some theorists believe that young children have trouble distinguishing objects from images, but I do not. Again, however, he does not “know that he is having images”, because he does not take himself and his mental life as an object of thought. That kind of self-consciousness is a much more complex activity: it may also involve anticipations of the future, but they are not so directly oriented to specific stimulus information. Genuine introspection thus involves the coordination of two separate activities: perceiving (or imagining) on the one hand and self-consciousness on the other. Like the other examples of dual tasks performance discussed in Cognition and Reality, this coordination is a difficult one and requires much practice. Eventually the child achieves it. In our culture at least, the most familiar way to do so is to divide what one experiences into the viewer and the viewed, and thereby to create the homunculus whom Morris and Hampson are so eager to face. But this separation between two aspects of experience does not correspond to any real break between earlier and later stages of information processing, at least in my view. The image is not an experience that the inner man first has and then describes by introspection. Rather, the image and the inner man are both experiences that the whole man has. Morris and Hampson insist that “the real intention of the introspector is to describe his conscious experiences”, not what he anticipates seeing. That is true, of course, but the origin and nature of experiences may not be evident
Anticipation, Images, and Introspection 173
even to the experiencer. A person who forms a mental image may be making use of his anticipatory schemata - “preparing for exterospection” -without knowing it, just as speakers often do not know that their utterances conform to rules of grammar, or dreamers that their dreams express unconscious wishes. It is true that people describe images differently than they would describe objects, but this does not refute the possibility that the images are the inner aspects of preparation to see those objects. Indeed, some difference of description would be expected on my hypothesis, since imagining is fundamentally different from perceiving. This may be a good place to say a word about “internal representations”, a term widely used in modern theories of imagery. Hampson and Morris fault me for not having any room for such entities in my theory. In fact, it would not be difficult to incorporate them somehow, but I have been reluctant to do so. Those who postulate the existence of internal representations don’t really mean the notion of “representation” seriously: images are not consensual symbols like flags or words. Rather, the term is used by theorists who want to treat images as if they were things - as if they could be manipulated, lost, found, and examined. Yet to form an image is not to lind something that was lost before, and to rotate an image is not to rotate something that might have been left stationary. These activities are more novel, and more deeply embedded in larger wholes, than such notions imply. Although I hesitate to use the terminology of “mental representations”, I would not wish to deny that we can gain access to new information through mental imagery. Indeed, we can learn by carrying out any activity. We often do not know what we can do, and how we will do it, until we have tried. The trying informs us, enabling us to provide descriptions and make predictions that were impossible before. Imagining is also a doing (in particular, it is a planning for perception), and so it can also be informative in this way. But we do not get the information by examining an internal representation; we get it by carrying out a preparatory activity and noting how it went. In conclusion, I would like to comment on several points where Hampson and Morris are quite right. First, my account of the mental rotation experiments is inadequate; I do not know why the rotation is always imagined at a particular preferred speed. The speed may be determined by some gross physiological property of the brain, or by the the same sort of limitations that determine reaction times, or by past experience with real rotations, or in some other still unknown way. No theory has yet solved this problem, or even addressed it. Second, my account of introspection cannot easily be generalized to such phenomena as pleasures and pains. Indeed it should not be; I think they are experiences of quite a different order than mental images. Third, I have offered no theory of memory in general. This is a defect that I
174 Uric Neisser
feel keenly, but for reasons explained in Cognition and Reality (pp. 141 - 142) it seems to me that the necessary data for the construction of such a theory are not yet at hand. Meanwhile, I have presented a hypothesis about the nature of mental images, or at least about the kinds of images that have been studied in psychological experiments. Since Morris and Hampson were apparently unable to find any observations which contradict it, I still believe that the hypothesis may turn out to be true.
References Craik,
F. I. M. and Lockhart, R. S. (1972) Levels of processing: a framework for memory research. J. Verb. Learn. Verb. Beh., 11. 671-684. Hampson, P. J. and Morris, P. E. (1978) Unfilled expectations: a criticism of Neisscr’s theory of imagery. Cog., 6, 79-85. Hannay, A. (1971) Mental Images: A Defence. London, George Allen and Unwin. Neisser, U. (1976) Cognifion andReaZity. San Francisco, W. H. Freeman. Ryle, G. (1949) The Concept ofMind. London, Hutchinson.