Cognition, Vol. 7, No. 2

Cognition, 7 (1979) 99-123 @ Elsevier Sequoia S.A., Lausanne - Printed in the Netherlands The logical and empirical b...

Author: Thomas R. Shultz | Arlene Dover | Eric Amsel | Anat Ninio | Francesco Antinucci | Alessandro Duranti | Lucyna Gebert | Mark S. Seidenberg & Laura A. Petitto

22 downloads 1127 Views 8MB 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

DOWNLOAD PDF

Cognition, 7 (1979) 99-123 @ Elsevier Sequoia S.A., Lausanne

- Printed

in the Netherlands

The logical and empirical basesof conservation judgments* THOMAS

R. SHULTZ

ARLENE

DOVER

ERIC AMSEL McGill University

Abstract It isargued that conservation judgmentsare based on a particular combination of logical necessity and empirical belief The empirical belief is that a given transformation will not alter a particular quantity (Elkind’s conservation of identity). The logical aspect is a transitive deductive argument containing an initial equivalence of two quantities and the conservation of identity belief as premises which lead to the conclusion of maintained equivalence (Elkind’s conservation of equivalence). In two experiments, it is shown that conservation of identity beliefs can be manipulated in subjects who have long since developed the capacity for transitive deductive inference. Untrained IO year olds were unaware of how sublimation acts to alter certain quantities over particular transformations of shape. And untrained adults incorrectly believed that both the area and perimeter of a closed figure would be conserved over transformations which elongated the figure. Both groups of subjects could be trained in the correct conservation of identity beliefs and this affected their conservation of equivalence judgments in predicted ways. It is suggested that the locical aspect of conservation is developmentally stable and that the empirical aspect varies widely across problems and individuals because of its dependence on relevant experience.

Among the most important discoveries of Western science are the so-called “conservation laws” which state that certain basic physical properties of the world (matter, energy, and momentum) remain invariant over various trans*The research reported here was supported in part by a grant from the Canada Council. Children were recruited with the generous cooperation of Jacques Duclos of the Protestant School Board of Greater Montreal and Ron Haynes, Principal of Iona School. Robert W. B. Kits van Heyningen derived the mathematical formulas relating the area and perimeter of closed figures. Requests for reprints may be sent to Thomas R. Shultz, Department of Psychology, McGill University, 1205 McGregor Avenue, Montreal, Quebec, Canada H3A 1 Bl.

100 T. R. Shultz, A. Dover and E. Amsel

formations. Such laws serve as the virtual cornerstone of our beliefs about the structure of the world as it is generally conceded that, if these laws did not hold, the world would appear vastly different than it is currently perceived. Thus, it is not surprising to find that conservation beliefs play a central role in the systematic study of cognitive development. Piaget and Inhelder (1969), in particular, have identified the belief in the conservation of matter as a principal criterion for the period of concrete operational thought. Their findings and those of many subsequent researchers indicate that, at about 6-8 years of age, the child typically begins to construct a systematic set of beliefs regarding the invariance of material properties such as quantity, length, area, volume, and weight over various transformations of shape and position. Conservation of such material properties is probably one of the most heavily researched phenomena in developmental psychology. No aspect of Piaget’s monumental contribution seems to have captured the interest of contemporary psychologists quite as thoroughly as conservation has. Despite the hundreds of published studies of conservation, there are some very basic questions regarding the phenomenon which remain unanswered. Principal among these is the essential basis of conservation judgments. How does one come to know that the two quantities are still equivalent following the transformation of one of them? This question tends to receive two quite different sorts of answers-one holding that conservation judgments are a matter of empirical belief based on one’s accumulated experience with physical properties and their transformations, and the other holding that conservation judgments are a matter of logical necessity based on deduction from a set of premises contained in the concrete operational structures. Piaget himself has been the chief proponent of the latter point of view, arguing that “the operational conservations of the intelligence are binding and ‘necessary’ (Piaget, 1971, p. 358).” One might well question whether human cognition could ever be entirely logical to the total exclusion of empirical content and, conversely, whether it could ever be entirely empirical to the total exclusion of logical structure. The former reservation would constitute a rejection of radical rationalism and the latter a rejection of radical empiricism. A constructivist (perhaps Kantian) resolution would support both of these rejections, thus relegating the issue of the logical vs. empirical basis of conservation beliefs to the status of a pseudo-issue. Nonetheless, a good number of contemporary researchers have chosen to regard this issue as substantive, perhaps reasoning that conservation judgments could be based mainly, if not entirely, on logical considerations or, conversely, based mainly, if not entirely, on empirical consider-

The logical and empirical bases of conservation judgments

101

ations. Being psychologists, they have attacked the issue with an awesome combination of logical theorizing and empirical experimentation. Most of the studies have attempted to resolve the issue by presenting conservers with evidence which appeared to contradict a conservation conclusion, namely that the transformed quantity was no longer equivalent to some initial standard. For example, in a conservation of weight experiment, a small amount of one substance might be surreptitiously added or subtracted during the transformation of shape. If the previously conserving child simply accepts this nonconservation result, then presumably the conservation judgment was based principally on empirical belief; the child has just discovered another fact, albeit a surprising one, about the world, a fact which there is no particular reason to doubt. In contrast, if the child reacts with disbelief and concludes there has been some kind of trick or mistake, then presumably the conservation judgment was based principally on logical necessity. This socalled “extinction of conservation” paradigm has been most common, although occasionally it has been supplemented with performance on a conservation post-test. The results of these studies have been quite equivocal in that varying degrees of “resistance to extinction” have been reported. The following brief review provides some indication of the diversity of findings as well as the generally disparate nature of conclusions drawn from these findings. When Hall and Kingsley (1968) found that even college students abandoned conservation judgments when confronted with contrary evidence, they concluded that such judgments must be based on empirical belief. Miller, Schwartz, and Stewart (1973) reported a stronger, although not complete, resistance to extinction of conservation in college students. Because the college students were more resistant to extinction than were 8-l 1 year old children (Miller, 1973), it was concluded that there are developmental increases in the “certainty” with which the concept of conservation is held. These gradual increases in certainty or logical necessity were thought to extend well beyond the point at which a child is usually considered to acquire the concept of conservation. This implies that, while conservation may be initially based on empirical belief, with time it becomes increasingly based on logical necessity. Another study by Miller and Lipps (1973) found that 8,9, and 11 year olds showed greater resistance to extinction of transitivity of weight than to extinction of conservation of weight. The predicted developmental increase in resistance to extinction was found for transitivity but not for conservation. It was concluded that, although the concepts of transitivity and conservation have the equivalent logical structure, they differ in the degree to which this logical structure is expressed in feelings of necessity. Somehow, transitivity

102 T. R. Shultz,

A. Dover and E. Amsel

is more logically necessary than is conservation. Finally, Miller and Lipps warned that there is a real possibility of misdiagnosis in extinction studies and so they urged the inclusion of controls for “pseudoextinction” in future research. Strauss and Liberman (1974) reported that 6-9 year olds mostly resisted extinction of conservation of both weight and discontinuous quantity. In explaining their results, they adopted the conceptual distinction between logical necessity and nomic necessity, the latter pertaining to the necessity of physical laws. They argued that extinction studies deal essentially with nomic, and not logical, necessity; and that logical-mathematical knowledge is initially understood by young children as being nomically, but not logically, necessary. The fly in an otherwise orderly ointment is their equivocal data on conservation of weight. They conclude that, since weight is such a complex concept, children may not regard its conservation as being even nomically, much less logically, necessary. Chiseri (1975) found that administering a prior conservation questionnaire made college students less resistant to extinction of conservation than they would otherwise have been. From such context effects and from his evidence that adults gave different justifications for extinction than did young children, he concluded that the use of extinction studies with adults cannot reveal much about the basis of conservation acquisition in children. This view received some additional support from a study by Strauss, Danziger, and Ramati (1977). Using a conservation of weight task with college students, they surreptitiously removed some of the clay while transforming it. Those subjects who accepted the resulting nonbalancing of the scales (70%) tended to explain it, not in terms of nonconservation of weight, but rather in terms of the positioning of the clay on the pan as it was thought to pertain to the principle of the lever. As Strauss et al., (1977) point out, interpretation of their results depends heavily on the criteria one uses to assess conservation concepts. Emphasis on judgments alone would lead to the conclusion that conservation can be readily extinguished and hence must be based on empirical belief. In contrast, emphasizing justification (as Strauss et al., favor) leads to the opposite conclusion that conservation is extremely difficult to extinguish and thus must be based on logical necessity. In an interesting elaboration of the extinction paradigm, Miller, Brownell, and Zukier (1977) assessed the certainty with which 8 and 11 year olds expressed conservation and transitivity judgments. They reported considerably less certainty after extinction evidence than before and no developmental increases in certainty. There was some evidence that conservation of number was held with more certainty than some other conservation concepts, but no evidence that transitivity was held with more certainty than conservation. The authors conceded that their certainty data are not directly relevant to


103

the issue of the logical or empirical basis of conservation since “...a child (or an adult) may be certain of some fact without experiencing that fact as a logically necessary truth (Miller et al., 1977, p. 244)” The numerous disparities among these studies reveal a literature in considerable disarray and leave the issue of the logical vs. empirical basis of conservation unresolved. We are confronted with: (a) findings that people (both school-aged children and college students) resist extinction of conservation and other findings that people do not so resist, (b) reports of developmental increases in resistance to extinction of conservation and transitivity along with contrary evidence of no developmental changes in resistance to extinction of these concepts, (c) evidence of greater resistance to extinction of transitivity than of conservation vs. other evidence of no difference in resistance to extinction of these two concepts, (d) the claim that extinction of conservation in adults is basically similar to extinction of conservation in young children and the counter-claim that it is basically different in adults than in young children, (e) the assertion that the issue of the logical vs. empirical nature of conservation depends on the response criteria one uses, and inevitably (f) conclusions that some, but not necessarily all, conservation judgments are based on logical necessity, on nomic necessity, on empirical belief, initially on empirical belief but eventually on logical necessity, and on various combinations of these for different age groups and types of conservation. A principal reason for this impasse is that extinction studies have thoroughly confounded the logical and empirical aspects of the conservation problem, both theoretically and methodologically. In essence, the extinction procedure contradicts a logical conclusion (conservation) with an empirical outcome (nonconservation). This can be seen clearly in terms of Elkind’s (1967) analysis of the structure of the conservation problem. In this analysis, the solution to the conservation problem is based on a transitive deductive argument of the following form: A = B Initial equivalence A -+ Ai Transformation (conservation .‘.A’ = B Conservation of equivalence

of identity)

The premises of this argument include the fact of initial equivalence (A = B) and the transformation (A + A’). If the child believes that the transformation does not alter the quantity being transformed (A = A’), then the conservation conclusion (A’ = B) follows by deductive inference. In contrast, if he believes that the transformation does alter the quantity being transformed (A # A’), then a nonconservation conclusion (A’ # B) follows by deductive inference.

104 T. R. Shultz, A. Dover and fi. Amsel

Elkind has termed the second premise regarding the effects of the transformation as “conservation of identity,” the belief that the quantity of a certain substance is not altered by a transformation. He has referred to the deduced conclusion as “conservation of equivalence,” the inference that, since the two quantities were initially equal and the transformation did not alter the transformed quantity, the two new quantities must also be equal. Three points can be made concerning this analysis of conservation. The first is that it clearly distinguishes the logical from the empirical aspects of the conservation problem. Keeping in mind the extreme narrowness of either a radical empiricism or a radical rationalism as discussed above, it would seem that (a) the conservation of identity premise is primarily a matter of empirical belief; the child learns through experience which transformations alter or do not alter which substances and that (b) conservation of equivalence is primarily a matter of logical necessity given the truth of the two premises; if the premises are both true, the conclusion must also be true. The overall conservation problem is thus neither wholly logically nor wholly empirical. Second, this analysis clarifies the nature of the extinction procedure. After establishing an initial equivalence (A = B) and performing what is commonly believed to be a conserving transformation (A = A’), the examiner confronts the subject with nonconservation evidence (A’ f B) which directly contradicts a logical conclusion (A’ = B). At best this is a very odd occurrence, sufficient perhaps to disequilibrate even the most astute conserver. At worst, it is a procedure which makes it impossible to clearly separate the empirical from the logical aspects of the problem. The third implication of Elkind’s analysis is that there are two kinds of conservation: identity and equivalence. Elkind (1967) argued that, although Piaget’s theory concerned conservation of identity, his procedure assessed only conservation of equivalence. In the standard conservation procedure, the subject is not asked whether the transformed quantity is equivalent to what it was before the transformation (A’?A). Rather the subject is asked whether the transformed quantity is equivalent to the initial standard quantity (A’?B). Curiously, the major impact of Elkind’s (1967) article was to stimulate assessment of this previously unassessed conservation (conservation of identity) and research to determine which of these two conservations (identity or equivalence) emerged first in ontogenesis. Since Elkind had argued that conservation of identity was a necessary precondition for conservation of equivalence, the finding that the former preceded the latter was interpreted as support for his position. Quite a number of studies reported findings of this kind (Elkind, 1966; Elkind & Schoenfeld, 1972; Hooper,


105

1969; Papalia & Hooper, 1971; Schwartz & Scholnick, 1970). However, a nearly equal number of studies reported that identity conservation did not develop prior to equivalence conservation (see Brainerd & Hooper, 1975, 1978; and Miller, 1978, for recent reviews). Generally these studies reported basically synchronous development of the two conservations (Koshinsky & Hall, 1973 ; Miller, 1977; Moynahan & Glick, 1972 ; Murray, 1970; Northman & Gruen, 1970) but certain exceptions were also noted. For example, Moynahan and Glick (1972) found that identity conservation preceded equivalence conservation for problems involving length but not for problems involving number, quantity, or weight. And Koshinsky and Hall (1973) reported that, although 86% of their sample revealed synchronous development of the two conservations, another 10% actually passed conservation of equivalence and failed conservation of identity. As with the literature on extinction of conservation, this literature on identity and equivalence can, with hindsight, be viewed as somewhat misdirected. If it is the case that conservation of identity is primarily a matter of empirical discovery and conservation of equivalence is primarily a matter of logical deduction, then why should the two develop in any particular order? The capacity for transitive deductive inference (involving, as discussed here, two premises and three terms) appears to develop by approximately 5-7 years of age (Bryant, 1974; Halford, 1969). It is reasonable to suppose that the development of this logical ability would be relatively fixed or environmentally stable. The one cross-cultural investigation of the emergence of transitive inference, comparing Costa Rican rural and urban children with American children, suggests that cultural influence may be minimal (Youniss, 1975). Recent cross-cultural studies of other logical skills suggest that demonstrable cultural differences have more to do with how the problem is represented than how it is solved (e.g., Scribner, 1975). There is no particular reason to expect this degree of stability for the empirical discovery of what particular transformations alter or do not alter what particular substances. Children’s experiences with substances and transformations must vary a great deal in both quality and quantity. Thus, conservation of identity beliefs might well emerge at various periods in the lives of different individuals. With respect to equivalence conservation, identity conservation could therefore appear earlier, later, or at roughly the same time. Asynchronous development of identity and equivalence conservation is of special interest for understanding the basis of conservation judgments since it provides a natural separation of the empirical and logical aspects of conservation solutions. The fact that young children do sometimes possess conserva-

106 T. R. Shultz,

A. Dover and E. Anzsel

tion of identity beliefs and yet are not able to reach conservation of equivalence solutions is well established as noted above. But does the reverse also occur? Are there cases in which individuals hold inaccurate beliefs about conservation of identity well after acquiring the capacity for transitive deductive inference? The 10% of Koshinsky and Hall’s (1973) sample who passed equivalence conservation and failed identity conservation might be counted as supportive, but the possibility of such a small effect being produced by measurement error cannot be lightly dismissed. The relatively late appearance of the so-called “second-order” or “formal” conservations is well known, but this decalage is more conventionally attributed to the inherent difficulty of the concepts involved than to the degree of opportunity for empirical discovery of the effects of transformations applied to these quantities (Brainerd, 1970). It may well be that many decalage effects are empirically rather than logically based. The purpose of the present experiments was to provide a convincing demonstration that accurate conservation of identity beliefs can indeed develop sometime after the capacity for the deduction of conservation of equivalence. Individuals who had long since acquired these deductive inference skills were presented with conservation problems for which they had not yet learned the appropriate conservation of identity beliefs. In some subjects, the accurate conservation of identity beliefs were trained, and the performance of these trained subjects was compared to that of untrained subjects on new conservation of equivalence problems. It was presumed that any effects of the training procedures on conservation of equivalence judgments would be due solely to alteration of empirical beliefs about conservation of identity. Thus, there was a methodological attempt to effectively separate the logical and the empirical bases for conservation. The first experiment dealt with the conservation of continuous and discontinuous quantity and the principle of sublimation and the second concerned conservation of the area and perimeter of certain closed configurations. In both experiments, transformations altered the shape of the substance or configuration by “extending” or “elongating” it.

Experiment

It is tions tities this

1: Conservation

and Sublimation

often assumed that material properties are conserved over transformaof shape. Yet some transformations of shape do actually alter the quanof certain substances, at least over a period of time. In physical science, problem is dealt with, in part, by the principle of sublimation. This


107

principle specifies that transformations which increase the surface area of a substance serve to increase the rate of exchange between the substance and the surrounding environment. Such “extending” or “spreading” transformations thus may indeed alter quantities, and these alterations will be particularly noticeable for substances which already have a fairly high rate of interchange with their environmental surround. One of the most obvious examples of the sublimation principle is the evaporation of water; the rate of evaporation is a positive function of the amount of surface area the water has. These facts served as the basis for the present experiment. Method Subjects The subjects were 24 boys and 24 girls from the fifth grade of a school in a middle class neighborhood of urban Montreal. All of the children were competent speakers of English, with the exception of one girl who spoke Chinese and another girl who spoke French. The latter was tested in French and the former was sufficiently competent in English to complete the experiment. One-half of the boys and one-half of the girls were randomly assigned to the training condition and the others served in the control condition. Those in the training condition had a mean age of 10.2 years with a range from 9.7 to 10.8. Those in the control condition had a mean age of 10.4 years, ranging between 9.3 and 11.3. Pilot research had established that children in this age group were strong conservers when assessed by conventional Piagetian measures yet had only sketchy knowledge of the principle of sublimation and how it might apply to the conservation task. Apparatus

The containers consisted of three test tubes 2.5 cm in diameter X 20 cm in height; three pie plates 19 cm in diameter; two 100 ml beakers; and one 50 ml beaker. Substances included small plastic beads and orange colored water. Procedure

Children in the training condition received a demonstration of the principle of sublimation followed two days later by a conservation assessment. Those in the control group received only the conservation assessment. All of these sessions were conducted by the second author. The demonstration was presented in a classroom to the entire group of children assigned to the training condition. It involved a 10 minute session on each of two successive days. On the first day, the experimenter poured a standard unit amount of 30 ml


of beads or colored water from the 50 ml beaker into either a test tube or a pie plate. The 30 ml level of the 50 tnl beaker was clearly marked with a thin black line. Three test tubes and three pie plates were used. First, a test tube and then a pie plate each received 30 ml of beads. Next, a test tube and then a pie plate each received 30 ml of colored water. Finally, the remaining test tube and pie plate each received this amount of colored water and were covered-the test tube with a rubber stopper and the pie plate with a clear cellophane wrap. After re-emphasizing that the same amount of material had been poured into each of the six containers, the experimenter posed the question of whether each of the six containers would, after 24 hours, have the same amount of material or whether it would have more or less material than it presently contained. At this point, the children discussed the problem, asked questions, and ventured predictions about the changes in quantity. Then they were asked to indicate their own specific predictions by tilling out an answer sheet which contained a drawing of each of the six containers. For each container, the child was asked to place a check mark in one of three boxes to indicate how much material that container would have after 24 hours: more, the same, or less than it presently had. Each child signed his or her answer sheet and the sheets were collected by the experimenter. The six containers were placed in an out-of-the-way corner of the classroom and left alone. The children’s predictions are presented in Table 1. Generally, they expected that the amounts would remain unchanged except for the uncovered water, which they expected to decrease. This pattern of predictions reveals some understanding of sublimation (water is more likely to sublimate than beads, but only if it is left uncovered), but no understanding of how sublimation might be influenced by the shape of the container and the resultant surface area of the material.

Table

1.

Predictions by children in the training demonstration Item

Beads in Beads in Water in Water in Covered Covered

of Experiment

1

Prediction

tube plate tube plate water in tube water in plate

More

Same

LCSS

0 1 1 1 1 1

24 23 3 3 17 16

0 0 20 20 6 7


109

The next day at about the same time, the experimenter returned to discuss the results of the demonstration. She re-distributed the answer sheets to the children so that they could compare their predictions with the observed results. She emptied each of the six containers in turn into the 50 ml standard unit beaker to determine how each amount had changed, if at all. The results showed that the quantity remained unchanged in all containers except for the uncovered water in the pie plate which had completely evaporated. For each of the six containers, the child was asked to circle the correct answer on his or her answer sheet and to determine how many of his or her predictions were correct. The results of the demonstration were then discussed as were various explanations of the results proposed by the children. In the course of this discussion, the evaporation explanation was summarized and clarified by the experimenter with reference to examples from the children’s own experiences such as the drying of rain puddles. It is perhaps important to stress that the demonstration focused only on conservation of identity beliefs. As no comparisons were made among the six different containers, the demonstration did not deal explicitly with conservation of equivalence arguments. Each child in the experiment was individually assessed on a series of six conservation of equivalence problems. In the case of the training condition, this assessment occurred about two days after the demonstration sessions. For each of three substances (beads, water, and covered water), the child first established an initial equivalence between the amounts in two identical 100 ml beakers (A and B). The A beaker was then emptied into either a test tube (A’) or a pie plate (A”). Following the pouring transformation, the child was asked to say whether the two containers (B and A’; or B and A”) had the same amounts or whether one of them had more and to explain why. In the case of the covered water, the two containers (B and A’; or B and A”) were covered with a clear cellophane wrap just after the pouring transformation. After recording the child’s response, the experimenter asked the child to consider what the two amounts might be like after 24 hours-would the two containers have the same or would one of them have more and why? The amount used in the conservation assessments was always 40 ml. For each of the three materials, the test tube (A’) was used before the pie plate (A”). Each of the six possible orders of the three materials was given to two boys and two girls in each condition. Results

Conservation judgments were scored as 1 and nonconservation judgments were scored as 0. Since Lunney (1970) has shown that analysis of variance can be legitimately applied to dichotomous data when there are sufficient


degrees of freedom for error, these scores were subjected to an analysis of variance in which sex (2); order (6), and condition (2) served as between subjects factors and material (3), time of comparison (2); and shape of container (2) served as within subjects factors. It was expected that the control subjects would give conservation judgments throughout. The training subjects were expected to give conservation judgments for all three materials on the immediate question but only for the beads and covered water on the 24 hours later question. By virtue of their training, they were expected to give nonconservation judgments for uncovered water on the 24 hours later question. This amounts to a condition X material X time of comparison interaction which indeed was found to be the largest, high level effect in the analysis, F(2,48) = 57.17, p < 0.001. Regression weights were written to reflect the specific nature of the predicted interaction and the resulting I; ratio was easily the largest in the entire analysis, F( 1,48) = 709.96, I_’< 0.001, leaving a negligible residual, 1;(1,48) < 1. The mean conservation judgment scores are presented in Table 2 as a function of condition, material, and time of comparison. Inspection of these means reveals that all cells yielded a high level of conservation except for that of the training condition with the uncovered liquid for the after 24 hours question. The direction of the judgments in this particular cell are presented in Table 3. As expected, trained subjects thought the beaker (B) would, after 24 hours, have more than the pie plate (A”) and less than the test tube (A’). The former finding represents a fairly direct application of the training demonstration where water was observed to evaporate from the pie plate. The latter finding, however, suggests a generalization of the sublimation principle to a new container (B) not observed in the demonstration. Justifications for conservation judgments were classified into the following categories: (a) identity (e.g., none was taken away so it’s the same amount); (b) reversibility (e.g., if you pour it back into this one it will come up to the same place); (c) compensation (e.g., it comes up higher but it’s not as wide); and (d) sublimation, which could refer to the shape of the container (e.g., a Table 2.

Mean conservation judgments Material

Beads Water Covered

in Experiment

Control

water

1 Training

Immediate

24 hours

Immediate

24 hours

0.917 0.896 0.917

0.896 0.729 0.833

1 .ooo

1.000 0.104 0.979

1.ooo 1.000


Table 3.

111

Direction of nonconservation judgments by trained subjects in Experiment on the 24 hours later question with uncovered water Judgment

Beaker has more Beaker has less Beaker has the same

1

Shape of container Test tube

Pie plate

0 19 5

24 0 0

flat one will lose more water than a skinny one), the covering of the container (e.g., when the water is uncovered it will disappear faster than when it is covered), the nature of the material (e.g., beads can’t evaporate), or the time scale of the comparison (e.g., it’s the same now, but by tomorrow the water will be gone), or could be fairly general (e.g., some of it will go up into the air). Identity, reversibility, and compensation are considered by Piaget and Inhelder (1969) to be the essential principles underlying conservation and hence are often used as categories for conservation justifications. In the present experiment, two of the authors independently classified the justifications given by 16 children (four boys and four girls from each of the two conditions). The proportion of agreement was found to be 0.97. In most cases (542) each subject gave one justification per judgment. However, two justifications were given in 26 cases, and three justifications were given in two cases. In seven cases, no justification was offered. Although a complete tabulation of these justifications as a function of experimental conditions is too cumbersome to present here, the general pattern can be easily described. Identity was the most frequently used justification (3 18) and it was very common in all cells except for the 24 hours later question in the training condition where it was used only twice. Reversibility was rarely used in any cell (12 times in all). Compensation was used to some extent in response to the immediate questions by children in both conditions (52) but was hardly ever used in response to 24 hours later questions (2). Sublimation justifications emphasizing the shape of the containers were used in nearly all cases of uncovered water in the 24 hours later question by trained subjects (46) but were rare in all other cells (6). Sublimation emphasizing the covering of the containers was heavily used by trained subjects with the covered water on the 24 hours later question (48), and, to a lesser extent, by control subjects on the same problem (16) but were rare in all other cells (6). Sublimation focusing on the nature of the material was heavily used by trained subjects with the beads on the 24 hours later question (45), and, to a lesser extent, by control


subjects on the 24 hours later question (16) but was very rare elsewhere (1). Sublimation involving reference to the time scale of the comparison was rarely used in any problem (4 times in all). General sublimation justifications were used to some extent by control subjects on the 24 hours later question (26) but were rare elsewhere (2). Discussion The results offer strong confirmation for the idea that empirically established conservation of identity beliefs affect conservation of equivalence judgments. These 10 year olds demonstrated a strong belief that transformations of shape do not alter quantities of either beads or water (A = A’) as well as a strong capacity for conservation of equivalence deductions (A = B, A = A’, :. A’ = B). Any general intuitions they may have held naturally about the principle of sublimation were not sufficient to upset their conservation of identity beliefs and the resulting conservation of equivalence conclusions (see the control condition results in Table 2). In some cases these beliefs and conclusions were correct and in other cases they were not. Training of the correct conservation of identity beliefs clearly fostered a good understanding of how sublimation can affect the quantities of certain materials undergoing transformations of shape and, consequently, led to almost completely correct equivalence judgments (see the training condition results in Table 2). Thus, the results demonstrate that correct conservation of identity beliefs can emerge independently of the previously developed deductive inference abilities which are the basis for judgments of equivalence. Experiment 2 was designed to extend this conclusion to a different sort of conservation problem.

Experiment tions

2: Conservation

of the Area and Perimeter of Closed Configura-

The first experiment was based on the idea that some transformations of shape do alter the quantities of some substances with the passage of time. Another interesting feature of transformations of shape is that they may alter certain physical quantities of a substance even at the moment of transformation, while leaving other physical quantities of that substance unchanged. A good example of this is the effect of an “extending” or “elongating” transformation on the area and perimeter of two-dimensional closed figures. For example, the transformation might involve the re-arrangement of a maximally “centered” or “dense” configuration such as a circle or square into a more linear shape such as an ellipse or rectangle. Without an extensive


113

grounding in geometry, few people probably realize that such transformations either increase perimeter if area is held constant or decrease area if perimeter is held constant. For all but a few highly trained individuals who could perhaps deduce such outcomes from an abstract set of theorems, this knowledge is very likely a matter of empirical discovery. The purpose of the present experiment was to establish correct conservation of identity beliefs by teaching people about the various effects of these transformations and to assess the impact of this teaching on equivalence judgments. Method Subjects The subjects were 32 males and 32 females enrolled in summer Psychology and Biology courses at McGill University. One-half of the subjects of each sex were assigned to the training condition and the remainder served in the control condition. Pilot testing had established that University students could not accurately assess the effects of shape transformations on area and perimeter and that they could be trained to do so quite easily. Apparatus The apparatus for the training procedure consisted of a set of four tiles each 10 cm square. The apparatus for the assessment procedures consisted of 30 steel balls each 1.59 cm in diameter, a circular copper rim 25.3 cm in length, and two strings each 25.3 cm long with its ends tied together. Procedure Subjects in the training condition received both a training procedure and an assessment procedure. Those in the control condition received the assessment procedure only. Each subject was tested individually by the third author in a single session. The training procedure was based on the premise that the effects of shape transformations could best be grasped if the quantities involved were readily identified in standard unit measures. To facilitate identification of units, four-sided figures were used in training. The assessment procedure involved rounded figures for which standard unit measures were not easily identified. Therefore, correct solution of the assessment problems required a tirrn conceptual grasp of the effects of the transformations. Training began with the experimenter arranging the four tiles in a 2 X 2 pattern. He asked the subject to consider that the area of this configuration was four units (one unit per tile) and that the perimeter was eight units (one unit per outside edge of a tile). Then the experimenter transformed the configuration into a 1 X 4 arrangement. The subject was asked to note that this

114 T. R. Shultz, A. Dover and E. Amel

transformation (A’) kept the area constant at 4 units and increased perimeter to 10 units. Then the experimenter reconstructed the original 2 X 2 arrangement and executed a second transformation. In this second transformation, one tile was removed and the remaining tiles were placed in a 1 X 3 configuration. The subject was asked to note that this transformation (A”) kept the perimeter constant at eight units and decreased the area to three units.’ Finally, the subject was asked to summarize the effects of these two types of transformations on area and perimeter. As with the training procedure in Experiment 1, the training procedure here focused only on conservation of identity beliefs. No more than one configuration was considered at any given time and no equivalences were established between simultaneously present configurations. In the assessment phase, each subject was presented with two conservation of equivalence problems. One of them involved an A’ transformation in which area was held constant and perimeter increased while the other involved an A” transformation in which perimeter was held constant and area decreased. In the A’ problem, the subject first was asked to establish the equivalence (in terms of both area and perimeter) of two sets of 30 balls each arranged in a closed circular pattern with a perimeter of 25.3 cm and an area of 50.9 cm2. These initial equivalences (A = B) were ascertained by noting that both sets of balls fit snugly into a rigid copper circular rim. Then, with the rim removed, the experimenter re-arranged one set of balls into an elongated elliptical pattern (A’) with a perimeter of 26.7 cm. Two conservation of equivalence questions were posed for the subject who was asked to justify his or her answer to each: (a) Do the two collections of balls cover the same area or does one cover more area than the other? and (b) Do the two collections of balls have the same perimeter or does one have a longer perimeter than the other? If the subject maintained that the two quantities were different, he or she was asked to specify which was greater. For area, the correct line of reasoning entailed A = B, A = A’, :. A’ = B. And for perimeter, the correct answer emerged from A = B, A < A’, :. A’ > B.* In the A” problem, the subject first established the equivalence (in terms of both area and perimeter) of two strings arranged in circular patterns each with a perimeter of 25.3 cm and covering an area of 50.9 cm2. These initial

‘Mathematically inclined readers may be interested to know that the perimeter of a four-sided figure with a constant area equals [2(area + lengrh2)j /length and that the area of a four-sided figure with a constant perimeter equals [*&perimeter X length)] ~ length2. 2Mathematically inclined readers may be interested to know that the perimeter of a curved figure with a constant area equals (fi/radius) R’ radius4 + area’, where radius refers to the major radius.


115

equivalences (A = B) were ascertained by superimposing one circular string over the other. Then the experimenter transformed one string into an elongated elliptical pattern (A”) having an area of 47.2 cm’. As in the previous problem, two conservation of equivalence questions were posed and the subject was requested to justify his or her answers: (a) Do the two strings cover the same area or does one cover more area than the other? and (b) Do the two strings have the same perimeter or does one have a longer perimeter than the other? Again, the subject was asked to specify the direction of any differences. For area, the correct answer involved A = B, A > A”, :. A” < B. For perimeter, it was A = B, A = A”, :. A” = B.3 The orders of the assessment problems (A’ and A”) were counterbalanced as were the orders of the questions posed in each problem (area and perimeter). Each of the eight possible orders was received by two males and two females in each condition. Results

Correct judgments were scored as 1 and incorrect judgments were scored as 0. Following Lunney’s (1970) recommendation as in Experiment 1, these judgment scores were subjected to an analysis of variance in which sex (2), condition (2), and order (8) served as between subjects factors and problem (2) and question (2) served as within subjects factors. The problem variable represented the two types of transformation (A’ and A”) and the question variable represented whether the correct judgment was conservation or nonconservation. Because of the apparent general tendency to assume that quantities are conserved over transformations of shape, it was expected that the proportion of correct judgments by control subjects would be high to conservation questions and low to nonconservation questions. In contrast, it was expected that subjects in the training condition would yield a high proportion of correct judgments on both types of questions. This amounts to a condition X question interaction which was indeed found to be significant, F(1,32) = 12.74, p < 0.005. Regression weights were written to reflect the specific nature of the predicted interaction and the resulting F ratio was easily the largest in the entire analysis, F(1,32) = 56.06, p < 0.001. The mean judgment scores are presented in Table 4 as a function of condition and question. Each of these means was compared to the value of 0.5 (which would be expected on the basis of random responding) using the technique suggested by Dunnett (1955). Those means found to be significantly greater 3Mathematically inclined readers may be interested to know that the area of a curved figure with a constant perimeter equals ,_/(‘/zperimeter’) @adi&) - n2 radius’, where radius refers to the major radius.


Table 4.

Mean conservation judgment scores in Experiment 2 Question

Conservation Nonconservation

Condition Control

Training

0.906 0.391

0.884 0.672

than 0.5 were control condition-conservation questions, t(32) = 5.96, p < 0.01, training condition-conservation question, t(32) = 5.05, p < 0.01, and training condition-nonconservation question, t(32) = 2.53, p < 0.05. The mean for the control condition-nonconservation question was not significantly less than 0.5. This mean was, however, significantly less than each of the other three means 0, < 0.05) as revealed by Cicchetti’s (1972) multiple. comparison technique for interactions. Thus, the data indicated that trained subjects tended to answer both conservation and nonconservation questions correctly and that control subjects tended to answer conservation, but not nonconservation, questions correctly. Similar to Experiment 1, the justifications given for conservation judgments were classified into the following types: identity, reversibility, compensation, and other. Two of the authors independently classified the justifications given by two males and two females from each of the two conditions, and the proportion of agreement was found to be 0.95. The proportions of the total given by all of the subjects were 0.84 identity, 0.01 reversibility, 0.11 compensation, and 0.04 other. Each type of justification was represented about equally in all cells of the design. Discussion

As in Experiment 1, the results here demonstrated that empirically established conservation of identity beliefs can be separated from conservation of equivalence deductions. Results for the control condition showed that even University students have not yet developed correct conservation of identity beliefs regarding transformations of shape as they affect area and perimeter. The training condition was fairly successful in teaching the correct conservation of identity beliefs and these led rather automatically to correct equivalence conclusions. Somewhat similar conservation problems were used by Vinh Bang (1965a and b) and Lunzer (1965) in their studies of the child’s developing conception of geometry. Although they are not at all concerned with distinguishing


117

the logical from the empirical bases of conservation, their findings are not inconsistent with those of the present study. The general nature of their results suggested a three-stage developmental sequence: (a) below eight years of age, children failed to conserve either area or perimeter over transformations of shape, (b) between 8 and 12 years of age, children conserved both area and perimeter, sometimes incorrectly; and (c) subjects over 12 years of age could be taught (largely through stretching the figure to its limit) that, as perimeter remains constant, area varies, and vice versa. It seems clear that in their research, as in the present experiment, it was the empirical discovery of conservation (or nonconservation) of identity relations which led to correct performance on these problems. It is very doubtful that such beliefs were deduced from some internal logical structure.

General Discussion On the basis of the present experiments, it may now be feasible to attempt to answer the troublesome questions raised by previous researches. The primary question concerns the basis of conservation judgments-do they derive from logical necessity or from empirical discovery? We suggest that conservation judgments are neither wholly logical nor wholly empirical. Rather, a conservation of equivalence judgment derives from a combination of logical and empirical knowledge. The empirical knowledge required consists of the belief that a given quantity does not change over a certain transformation. While recognizing that knowledge cannot be totally divorced from internal cognitive structure, we feel confident in classifying this knowledge as primarily empirical because, for the vast majority of conservations studied, it is most probably developed through experience rather than deduced from internal axioms. Once so acquired, this empirically based conservation of identity belief can form a crucial premise in the deductive argument that leads to a conservation of equivalence conclusion. When coupled with the additional premise that two quantities are initially equal, the conservation of identity belief enables one to deduce that the newly transformed quantity must still be equivalent to the unchanged standard. Given the empirically established truth of the two premises, such a conclusion is more than a matter of empirical belief; it is logically necessary. Again, while recognizing that thought cannot be totally devoid of empirical content, we feel that this knowledge is primarily logical because it does not require empirical validation. For most subjects, the equivalence must be preserved-it does not have to be empirically checked. In this connection it is interesting to consider the find-


ing that even children as young as five years in a conventional conservation of liquid quantity task believe that the quantities remain equivalent as long as they are hidden behind a screen. Only when the screen is removed and empirical checking becomes possible does the conservation of equivalence conclusion begin to waver (P. Miller & Heldmeyer, 1975). The conclusion that some aspects of conservation are empirically based whereas other aspects are logically based may appear irrefutable and obvious, particularly to those who may have appreciated the full implications of Elkind’s (1967) seminal analysis of the conservation probIem. However, this conclusion does not appear to have been at all obvious to any of those researchers concerned with settling this issue with the extinction paradigm (see Hall & Kaye, 1978, for another recent review). None of these authors, for example, refer to possible differences between identity and equivalence conservation and all of them restrict their assessments to equivalence conservation only. Likewise, no researcher investigating identity vs. equivalence conservation has referred to the possible relevance of this distinction for the logical vs. empirical issue. Furthermore, the hypothesis that equivalence conservation is based on transitive deductive inference has recently been doubted (Klahr & Wallace, 1976; Miller, 1978). A secondary question has concerned the order of acquisition of the two kinds of conservation-identity and equivalence. In the light of our answer to the primary question, this question becomes much less interesting. Literature reviewed above suggests that the capacity for making transitive deductive inferences involving three terms and two premises develops by about 5-7 years. It is assumed that this ability is more the product of internal cognitive development than of any specific empirical experiences and thus is relatively environmentally stable (Piaget, 197 1, p. 307; Youniss, 1975), but this should be confirmed through further research, especially cross-cultural research. Although the logical capacity for deducing conservation of equivalence may be developmentally stable, the acquisition of conservation of identity beliefs varies considerably among materials, transformations, and probably even individuals. The present studies verified that 10 year olds and adults still had not acquired certain correct conservation of identity beliefs even though they were quite skilled at making transitive deductive inferences. Again, the notion that there is no good reason to expect any particular ordering for the acquisition of identity and equivalence conservations may appear to be irrefutable. However, none of the researchers in this area seem to have entertained this possibility. As Brainerd and Hooper (1978) point out in their recent review, “Whenever the null hypothesis was rejected, identity tests were easier than equivalence tests; the opposite result had never been observed


119

(p. 70).” The continued focus throughout the literature on this problem has been to determine whether identity conservation emerges before equivalence conservation or, alternatively, whether the two conservations emerge synchronously. The possibility of some identity beliefs developing later than equivalence conservation, simply because the two types of conservation are essentially independent, had not really been considered. One suspects that the scholarly pursuit of what particular identity beliefs emerge at what particular ages will not prove to be especially interesting or rewarding. There are too many such beliefs and they are probably influenced by too many specific experiences. However, it may be of interest to determine whether the reported delays in conservation acquisition among children in certain traditional cultures (e.g., Goodnow, 1969; Greenfield, 1966) are logically or empirically based. According to the present analysis, it would be predicted that, while these children may lack the requisite experiences for certain conservation of identity beliefs, they would possess transitive inference abilities by about 5-7 years. One of the principal differences between the current analysis and the conventional Piagetian view concerns the origin of conservation of identity beliefs. Piaget (1971, pp. 208-209) has maintained that conservation of identity beliefs follow inevitably and logically from operational reversibility, which he defines as the capacity to mentally reverse or undo a transformation and thus recreate the original configuration.4 It is perhaps intuitively appealing that conservation beliefs could derive from reversibility. The liquid is imagined to be poured back into the original beaker and reach the initial height, and so on. And indeed, a number of studies have found reversibility training to be an exceptionally effective method for inducing conservation (Brainerd & Allen, 197 1; Gladstone & Palazzo, 1974; Hamel & Riksen, 1973; Schnall, Alter, Swanland, & Schweitzer, 1972). However, it seems very doubtful that reversibility by itself could ever produce what children eventually develop, namely, the ability to distinguish conserving from nonconserving transformations. As is obvious from the problems presented in the present experiments, some transformations alter certain quantities but not other quantities while other transformations do the opposite. Both conserving and nonconserving transformations may be equally reversible. For instance, a circular string can be stretched to form an ellipse and then reformed into a 4Although identity and compensation are also considered to contribute to conservation and, indeed, are often used by children to justify conservation judgments (Piaget & Inhelder, 1969), Piaget has argued that they are not the actual source of conservation beliefs. Identity, he writes, is prior to, but not sufficient for, conservation acquisition; and compensation is better regarded as a conclusion drawn from conservation than as a source of the conservation (Piaget, 1971).

120 T R. Shultz, A. Dover and I?. A me1

circle. Reversibility alone does not reveal whether a particular quantity is conserved or altered. Yet people do come to distinguish conserving from non-conserving transformations; they do not conserve indiscriminately. Gelman (1972) for example, has shown that children as young as 354 years can distinguish transformations which alter quantities (e.g., addition) from those that do not (e.g., spatial re-arrangement). Again, reversibility would be of no help here since both sorts of transformation can be reversed-addition by subtraction, and a spreading transformation by a condensing transformation. Thus, the real mystery about conservation acquisition is not how people come to conserve, but rather how they come to distinguish those transformations which alter a quantity from those which do not. It seems clear that factors other than reversibility will have to be invoked. One very interesting possibility, proposed by Klahr and Wallace (1976) is that conservation rules emerge out of subitizing small, discontinuous quantities before and after various transformations. Such rules are presumed to eventually be extended to cases involving larger discontinuous quantities, assessed through counting and estimation, and to discontinuous quantities. Whether Klahr and Wallace’s model can successfully account for the development of all of the various conservations remains to be seen. Similar dissatisfactions with reversibility as the sole source of conservation have been expressed by others (Berlyne, 1965; Wallach, 1969) but it is noteworthy that in none of these previous efforts was there a clear distinction drawn between the deductive and the empirical aspects of the problem. Klahr and Wallace (1976), for example, assume that rules for both identity and equivalence conservations are empirically derived and emerge synchronously. Certainly there is considerable disconfirming evidence, both in the present findings and in previous research, for consistently synchronous development. Moreover, it would seem needlessly inefficient for the child to circumvent the use of a transitive deductive argument and rely instead on accumulated experience with the conservation of equivalence paradigm. Conservation observations of any kind are probably rare for most children. But conservations involving equivalences between parallel quantities must be especially infrequent. Thus, it would be much more efficient for the child to employ empirically derived conservation of identity beliefs in the service of transitive deductive arguments when dealing with two or more quantities. While many of the problems of conservation acquisition remain unsolved, it does seem likely that a clearer distinction between the essentially logical and the essentially empirical aspects will render these problems more tractable.


12 1

References

Berlyne, D. E. (1965) Structure and Direction in Thinking. New York, Wiley. Brainerd, C. J. (1970) Continuity and discontinuity hypotheses in studies of conservation. Devel. Psychol., 3. 255-229. Brainerd, C. J., & Allen, T. W. (1971) Experimental inductions of the conservation of “first order” quantitative invariants. Psychol. Bul., 75, 128-144. Brainerd, C. J.,& Hooper, F. H. (1975) A methodological analysis of developmental studies of identity conservation and equivalence conservation. Psychol. 9~1.. 82. 725-737. Brainerd, C. J., & Hooper, F. H. (1978) More on the identity equivalence sequence: an update and some replies to Miller. Psychol. 9~1.. 85. 70-75. Bryant, P. (1974) Perception and understanding in young children. London, Methuen. Chiseri, M. J. (1975) Amenability to incorrect hypotheses in the extinction of conservation of weight in college students. Merrill-Palmer, 21, 1399144. Cicchetti, D. V. (1972) Extension of multiple-range tests to interaction tables in the analysis of variance: a rapid approximate solution. Psychol. Bul., 77, 405-408. Dunnett, C. W. (1955) A multiple comparison procedure for comparing several treatments with a control. f. amer. statist. Assoc., 50, 1096-l 121. lclkind, D. (1966) Conservation across illusory transformations in young children. Acta Psychol., 25, 389-400. Elkind, D. (1967) Piaget’s conservation problems. Child Devel., 38, 15-27. Elkind, D., & Schoenfeld, E. (1972) Identity and equivalence conservation at two age levels. Devel. Psychol., 6, 529-533. Celman, R. (1972) Logical capacity of very young children: number invariance rules. Child Devel., 43, 75-90. Gladstone, R., & Palazzo, R. (1974) Empirical evidence for reversibility by inversion. Devel. Psychol., 10,942-948. Goodnow, J. J. (1969) Problems in research on culture and thought. In D. Elkind & J. H. Flavell (eds.), Studies in cognitive development. New York, Oxford University Press. Greenfield, P. M. (1966) On culture and conservation. In J. S. Bruner, R. R. Olver, & P. M. Greenfield (eds.), Studies in Cognitive Growth. New York, Wiley. Halford, G. S. (1969) An experimental analysis of the criteria used by children to judge quantities. J. exper. child Psycho/., 8, 314-327. Hall, V. C., & Kaye, D. B. (1978) The necessity of logical necessity in Piaget’s theory. In L. S. Siegel & C. J. Brainerd (eds.), Alternatives to Piaget: Critical Essays on the Theory. New York, Academic Press. Hall, V. C., & Kingsley, R. (1968) Conservation and equilibration theory. J. genet. Psychol.. 113, 195-213. Hamel, B. R., & Riksen, B. 0. M. (1973) Identity, reversibility, verbal rule instruction, and conservation. Devel. Psychol., 9, 66-72. Hooper, F. H. (1969) Piaget’s conservation tasks: the logical and developmental priority of identity conservation. J. experi. child Psychol., 8, 234-249. Klahr, D., & Wallace, J. G. (1976) Cognitive Development: An Information-Processing View. Hillsdale, New Jersey, Lawrance Erlbaum. Koshinsky, C., & Hall, A. E. (1973) The developmental relationship between identity and equivalence conservation. .I. exper. child Psychol., 15, 419-424.


Lunney,

G. H. (1970) Using analysis of variance with a dichotomous dependent variable: an empirical study. J. educat. Measure., 7, 2633269. Lunzcr. I:. (1965) Lcs coordinations et les conccrvations dans lc domainc de la geometric. Et&es d’epistefnologie Genetique, 19, 59 148. Miller, P. H., & fieldmeyer, K. H. (1975) Perceptual information in conservation: effects of screening. Child Devel., 46, 588-592. Miller, S. A. (1973) Contradiction, surprise, and cognitive change: the effects of disconfirmation of belief on conservers and nonconservers. J. exper. child Psychol., 15, 47-62. Miller, S. A., (1977) A disconfirmation of the quantitative identityquantitative equivalence sequence. J. exper. child Psychol., 24, 180-189. Miller, S. A. (1978) Identity conservation and equivalence conservation: a critique of Brainerd and Hooper’s analysis. Psychol. Bul., 85, 58869. Miller, S. A., Brownell, C. A., & Zukier, 11. (1977) Cognitive certainty in children: effects of concept, developmental level, and method of assessment. Devel. Psychol., 13, 236-245. Miller, S. A. & Lipps, L. (1973) Extinction of conservation and transitivity of weight. J. exper. child Ps.ychol., 16. 388-402. Miller, S. A., Schwartz, L. C., & Stewart, C. (1973) An attempt to extinguish conservation of weight in college students. Devel. Psychol., 8, 316. Moynahan, E., & Click, J. (1972) Relation between identity conservation and equivalence conservation with four conceptual domains. Devel. Psychol., 6, 247-251. Murray, 1:. B. (1970) Stimulus mode and the conservation of weight and number. J. educat. Psychol., 61, 2877291. Northman, J. E., & Gruen, G. E. (1970) Relationship between identity and equivalence conservation. Devel. Psychol., 2, 3 11. Papalia, D. PI., & Hooper, 1:. H. (1971) A developmental comparison of identity and equivalence conservations. J. exper. child Psychol., 12, 347-361. Piaget, J. (1971) Biology and Knowledge. Chicago, University of Chicago Press. Piaget, J., & lnheldcr, B. (1969) The Psychology ofthe Child. New York, Basic Books. SchnaU, M., Alter, E., Swanland, T., & Schweitzer, T. (1972) A sensory-motor context affecting performance in a conservation task: a closer analogue of reversibility than empirical return. Child Devel., 43, 1012-1023. Schwartz, M. M. & Scholnick, E. K. (1970) Scalogram analysis of logical and perceptual components of conservation of discontinuous quantity. Child Devel., 41, 695-706. Sceribner, S. (1975) Recall of classical syllogisms: a cross-cultural investigation of error on logical problems. In R. J. 1:almagne (Ed.), Reasoning: Representation and Process in Children and Adults. Hillsdale, New Jersey, Erlbaum. Strauss, S., Danziger, J., & Ramati, T. (1977) University students’ understanding of nonconservation: implications for structural reversion. Devel. Psychol., 13, 3599363. Strauss, S., & Liberman, D. (1974) The empirical violation of conservation laws and its relation to structural change. J. exper. child Psychol., 18. 464-479. Vinh Bang (1965) Intuition geometriquc ct deduction opcratoirc. Etudes d’episte’tnologie Genetique, 19. 1 38. (3) Vinh Bang (1965) Dc I’intuition geometriquc. Etudes d’epistPmologie GPnCtique, 19, 39-58. (b) Wallach, L. (1969) On the bases of conservation. In D. Elkind & J. H. Flavell (eds.), Studies in Cognitive Development. New York, Oxford University Press. Youniss, J. (1975) Inference as a developmental construction. In R. J. Flamagne (ed.), Reasoning: represerttation and process in children and adults. Hillsdale, New Jersey, Erlbaum.


123

L’on pr&cnd que Its jugemcnts de conservation ont pour base une combinaison particuliere dc ndccssite logique et de zroyance empirique. La croyancc empirique veut qu’unc transformation donnee ne moditie pas une quantite spdcifique (conservation dc I’identite d’l‘:lkind). L’aspcct logique est un argument deductif transitif dont Its pr6misscs sont une 6quivalcncc initiale dc deux quantites et la croyancc cn la conservation dc l’identite, lesquellcs m&rent a la conclusion dc I’equivalcnce maintenuc (conservation dc I’dquivalence dWkind). Deux cxpdricnccs nous permettent de constater quc les croyantes en la conservation de I’identitd pcuvent Ctrc manipulees chez dcs sujcts qui ont dcpuis longtemps cultivd leur capacite d’infkrcncc ddductivc transitive. Des cnfants de 10 ans non prepares n’avaicnt pas conscience le la faeon dont la sublimation agit pour modifier certaincs quantit6s apr& des transformations de formc particulierc. Des adultcs non prbpards croyaient i tort que et la surface ct le pdrimetre d’unc figure fcrmdc dcmeurdicnt inchangds apres dcs transformations visant i ailongcr la figure. Les dcux groupcs de sujcts ont pu etre familiarisds avec les conceptions exactes de la conservation dc I’idcntitd, cc qui a modifid leurs jugemcnts sur la conservation dc I’equivalence dans la direction anticipic. I1 scmblcrait que I’aspcct logique dc la conservation reste stable dans son ddveloppement aiors quc I’aspcct empirique varic fortcmcnt selon les problkmes et les individus, cela dtant fonction des experiences en la mati&re.

C’ognifion, 7 (1979) 7 (1979) 125 -144 OElsevier Sequoia S.A., Lausanne -~ Printed

2 in the Netherlands

Piaget’s theory of space perception in infancy* ANAT

NlNlO

The Hebrew

University,

Jerusalem

Abstract Piaget3 theory of space perception in infancy is presented in the format of a hypothetico-deductive system. Eleven hypotheses are defined, regarding the perception of the agent of visual change; shape and size constancy; depth, and the perception of higher-order relationships among spatial elements. The proof Piaget offers for each hypothesis is presented in the following steps. behavioral evidence, interpretation in terms of inner states; inferences and generalizations. Presuppositions underlying the arguments are explicated. Critical notes are inserted whenever appropriate. Some general conclusions are briefly discussed.

Introduction Is the infants’ perceptual space qualitatively different from that of adults? This claim has been put forward by Piaget and his collaborators in two seminal volumes (Piaget, The child’s construction of reality, 1955 : and Piaget and Inhelder, The child’s conception of space, 1956), and since then widely accepted. It is only very recently that experimental evidence directly bearing on this question has started to accumulate, and it has not been unequivocal. For instance, neuropsychological research has shown that, in animals, many higher-order perceptual forms and relations, as well as lower-order ones, are apprehended directly through “wired-in” recognition systems, which are in all probability inborn. These relations and forms include lines, angles, straightness and convexity, orientation of lines, relative depth, parallelism, perpendicularity and comers (Maturana et al., 1960; Hubel and Wiesel, 1962; Barlow et al., 1967, Pettigrew et al., 1968). It is very plausible that the same sort of inborn mechanisms operate in humans; for instance, Bower (1968) showed that newborn infants have stereopsis. This implies that the *The author wishes to thank Amia Lieblich, Dan Frankel and Zeev Klein for their helpful comments. Requests for reprints should be sent to Anat Ninio, Department of Psychology, The Hebrew University, Jerusalem, Israel.

126

A. Ninio

perceptual space of humans might not undergo that series of qualitative changes which Piaget inferred ‘from the many observations on his children. It was felt that alongside the experimental studies on infant perception there was room for a reexamination of the original Piagetian claims. This paper attempts to contribute to this effort. Piaget’s work on space is fairly central to his general theory of intelligence. and as such, it is developed in great detail. Nevertheless, neither Piaget nor any of his summarizers provide a concise and systematic rCsum6, which will at the same time contain the details necessary for a critical evaluation. The scope of the summarizers is usually too wide to fill these requirements. Flavell (1963), Ginsburg and Opper ( 1969), Hunt (196 1). and Wolff (1960), for instance, allot to the discussion of early space only about three pages, on average. This paper represents an attempt to fill this need, namely to present Piaget’s work on early perceptual space in a concise and systematic form. The method adopted is similar to that of Wolff (1960), namely, spelling out in full the steps of the Piagetian argument, and if necessary, explicating the implied. Piaget’s method of presentation is a version of the hypothetico-deductive system. He starts out with a description of some observations on infant behavior in specified situations. He then interprets the infant’s overt behavior in terms of inner states of consciousness, wishes, expectations, knowledge, perception and concepts. From the sum total of these interpretations Piaget proceeds to deduce a general statement about the infant’s cognitive capacities. In some cases he tests the limits of his generalization with additional observations, or on data collected on another child. Most inferences rest upon presuppositions, not explicit in the text. In this paper Piaget’s main statements about the infant’s space are defined as formal hypotheses. The proof he offers is followed through from observational evidence to conclusion, in the following steps: (a) Bviderzce: mostly consists of the description of observational situations (usually quasi-experiments set up by Piaget) and the infant’s responses or other behavior. The description is usually generalized across a number of similar situations, unless the differences are such as to warrant a separate presentation. (b) Ipzterprctatiorz of the observed overt behavior in terms of inner states of consciousness, abilities etc. in the infant. (c) /rzfi~-erlces: several steps of inference are usually drawn from the observations and their interpretation, each inference logically deduced from its predecessor.

Piagetk theory of space perception in infancy

127

(d) Presuppositions: the set of necessary presuppositions linking each step in the argumentto its predecessor is explicated. Presuppositions are presented here whether explicit in Piaget’s writings, or inferred from his argument. (e) Notes are inserted whenever the logic of progression from one step to its successor seems to me to be incorrect. The Notes represent the sole attempt at critical evaluation. No novel evidence whatsoever is reported; this study aims to examine the logical consistency of the Piagetian system from its own premises. However, the reader is in possession of psychological knowledge which has accumulated since the first appearance in print of the two space books, in 1937 and 1948. The critical Notes appeal to this knowledge, at least implicitly. Full reference is given for observations and statements; for the former both serial number of observation and page number are given. The names of the two main space books are abbreviated: CS refers to “The child’s conception of space” (Piaget and Inhelder, 1956), while CR refers to “The child’s construction of reality” (Piaget, 1955). Age at observation is not given. Only visual space is discussed, following Piaget’s emphasis on the visual system. Although the development of space is intimately connected to the development of sensory-motor intelligence, it was considered possible and in the best interests of clear presentation to present Piaget’s ideas regarding infant perceptual space independently of the general system. In concrete terms this means that every Piagetian hypothesis and observation on infantile perceptual space has been included here indiscriminately, regardless of the sensory-motor stage it is related to. A further simplification in presentation was achieved by omitting all reference to Piaget’s theory of groups. It was felt that this omission did not seriously hinder the achievement of this paper’s limited aims. Piage t’s Theory Before turning to the actual hypotheses, it should perhaps be recalled that for Piaget there is no clear separation between perceptual and intellectual processes. Indeed, perception is “not a primary element independent of intelligence”, but, rather, “the result of intellectual activity” (CR, p. 2 12). The way in which the infant perceives space is, therefore, directly influenced by the ways in which he conceives of space, and the two are indistinguishable. Or, in Piaget’s words, “perception implies a sensory-motor schema which brings the sum-total of previous constructions to bear on the actual situation” (CS, p. 15). Therefore, in infancy, prior to the appearance of representation, perception of space is identical to conception of space. With the development of representational and operational intelligence, it is possible to think of objects in their absence and thus perceptual and conceptual

128

A. Ninio

space can be distinguished from each other (C’S, p. 17). Such a distinction is impossible for the first period which is the subject-matter of the present exposition. What are, then, the characteristics of the infant’s spatial universe? It has four major characteristics, best expressed in the negative. The infant’s space is not objective, not populated by objects of constant shape or size; not three-dimensional and does not contain relationships of higher than topological order among its elements. These main themes might be expressed in the following more detailed and more precise form: A. ‘The infant has no means of distinguishing the source of change in the shape of objects, namely, he cannot specify whether the visual change has resulted from movements of the object or from movements of his (the infant’s) body (CS, p. 12). B. The infant has no means of distinguishing the source of change in the shape of objects, namely, he cannot specify whether it has resulted from a change of state or from a change of position of the object (CS, p. 9; CR, p. 103). C. A change in the perspective of the object is perceived by the infant as an actual transformation of the object (CS, p. 11). D. There is no shape constancy (CS, p. 6). E. There is no size constancy (CS, p. 5). F. The infant perceives space at the first period of his life as a fluid mass without depth (CR, p. 145). into two G. At a later period in infancy, space becomes differentiated zones, “near space” and “far space” (CR, pp. 143- 145). between “near space” and “far space” is not that H. The differentiation of depth (CR, pp. 143-145). I. The objects in “far space” appear to the infant diminished and distorted by perspective (CR, p. 143). into depth planes (CR, p. 145). J. “Far space” is undifferentiated of projective and euclidean (i.e. higher than topoK. Spatial relationships logical) order are not perceived (CS, p. 6). In the following the evidence Piaget marshalls for each of these hypotheses will be presented and the connection between each piece of evidence and the hypothesis made explicit. A. Hypothesis: The infant has no means of distinguishing the source of change in the shape of objects, namely, he cannot specify whether the visual change has resulted from the movements of the object or from the movements of his (the infant’s) body (CS, p. 12).

Piaget’s theory of space perception in infancy

129

Evidence. 1. Observations on the behaviour towards vanished objects or unseen parts of objects; 2. Observations on the imitation of facial gestures; 3. Observations on shaking of head and body while looking at hanging objects. Zst Evidence: Observations regarding infant behavior towards vanishing objects or parts of objects were made in ten different situations, described below. Also described is the infant’s behavior in each of the situations. Situation a: The adult moves within the visual field of the infant, then leaves it. The infant follows the movement with his eyes up to the adult’s disappearance and keeps watching the point of disappearance (CR, obs. 2, p. 13). Situation b: The adult appears and disappears repeatedly at one point in the infant’s visual field. The infant keeps watching the point of disappearance (CR, obs. 2, p. 13). Situation c: The infant looks at a stationary adult, turns away, then turns back to look at the adult several times (CR, obs. 5, p. 11; obs. 12, p. 18). Interpretation of infant behavior in situations a, b and c: The infant expects the adult (= object) to reappear if he repeats the last act of accommodation which has been accompanied by the sight of the object (CR, p. 13). (Note: It should be noted that this is a rational expectation, usually fulfilled, for instance in situations b and c, above. It is unclear what alternative rational search behaviors, if any, are possible for an infant who is confined to a bed. Searching for the adult anywhere else in the immediate visual field would be incorrect, since the adult has been observed leaving it. An immobilized adult would behave very similarly in these circumstances.) Situation d: The adult appears at successive positions A, B, and C, ordered from right to left, then disappears at extreme left (C). The infant looks at A (CR, obs. 71, pp. 118-119). Situation e: The adult opens a door, interacts with the infant, crosses the room and disappears from the infant’s sight. The infant looks at the direction of the door (CR, obs. 72, p. 119). Interpretation of infant behavior in situations d and e: The infant expects the object to reappear if he repeats the first act of accommodation which has been accompanied by the sight of the object (CR, p. 119). (Note: It should be noted that in the circumstances of situation e this seems to be a rational expectation, since the adult did in fact reappear at the door another three times (CR, obs. 72, p. 119).)

130

A. Ninio

Inference: The infant conceives of the displacements of things as extensions of his own activity (CR, p. 102) or as being at the disposal of his activity (CR, p. 20), not as external autonomous events. Moving objects are not considered as being animated by an independent motion (CR, p. 168), nor as capable of following an autonomous trajectory independent of the subject’s action (CR, p. 168). (Note: Piaget does not attempt to distinguish between those situations (such as in a to c above) in which the infant excepts the object to reappear in the last position it has been seen, and other situations (such as d and e above) in which he expects the object to reappear at the first place seen. Analysis of the situations Piaget reports reveals that the child’s guesses are anything but random. In fact he guesses perfectly (in the three situations for which the relevant information is given) the place in which the adult will reappear. As noted above, in situations b and c the adult did in fact reappear to the most recent position in which he had been seen, while in situation e he reappeared at the initial position three times. Not only do the infant’s expectations seem to be rational but, as pointed out earlier, there is no other search behavior which would be as rational as the ones he actually adopts. Piaget (CR, p. 11) claims that if the infant had an object concept, he would search around with his eyes or otherwise change his perspective. This criterion is unacceptable. Since the potential visual field of the infant is constant, restricted and defined by what he can reach with eye and head movements, and since he has just seen the adult leave this field, it would be irrational if he searched for the adult anywhere inside the same field (since he is bodily restricted, he cannot change the visual field). To summarize, the infant’s behavior in situations a to e seem to point to the following conclusions. First, that the infant can make inferences about the adult’s behavior around him and is able to predict quite accurately in a given situation where the adult will reappear in his visual field. Secondly, that he understands that his visual field can be entered only at specific “ports” of exit and entrance, such as doors, edges of furniture, etc., and not anywhere else. These expectations and pieces of knowledge are rational and ecologically sound, and seem to disprove the idea that infants lack an objective space.) Situation f: The adult moves an object horizontally at the infant’s eye level. The infant loses sight of it, continues his pursuing movement and recaptures the sight of the object (CR, obs. 11 and 12, pp. 17-18). Situation g: The adult lets an object fall from his hands, held above the infant, in such a way that the infant can see the beginning of the movement.

Piaget s themy of space perception

in infancy

13 1

The infant follows the fall of the object or searches for it on the floor beneath the adult’s hands (CR, obs. 6, 8, 9; pp. 14- 16). Situation h: The adult lets an object fall from his hands, held above the infant, in such a way that the infant cannot see the beginning of the movement. The infant tends to look at the adult’s hands or at the (empty) space where the object had been held previously (CR, obs. 6, 8, 9, pp. 14-16). Interpretation of infant behavior in situations x g and h: The infant expects a rapidly moving or falling object to be continuing its trajectory if he loses sight of it momentarily; he expects the object to reappear if he extends the last act of accommodation which was accompanied by the sight of the object (CR, pp. 19-20). (Note: It should be noted that this expectation is rational, namely, rapidly moving or falling objects can in fact be visually recaptured if the last pursuing movement is extended or continued.) Situation i: The adult covers (obscures) an object with a screen (cloth, hand). The infant does not uncover it (CR, obs. 7, p. 15; obs. 22-25, pp. 28-32; obs. 28-32, pp. 36-40). Situation j: The adult rotates an object so that the infant is presented with an unfamiliar face of it. The infant will not rotate the object so that its familiar (or desired) part faces him unless he can see some of the familiar part. For example, he will not rotate a reversed feeding-bottle (presented with the bottom facing the infant) unless some part of the nipple is visible (CR, obs. 77, p. 124- 125; obs. 78 and 78a, pp. 126-- 128). Generalization of infant behaviors in situations a to j (Ist evidence). When neither the repetition of previous accommodations nor their extension brings about the reappearance of the object, the infant is unable to search for it by inventing new procedures (CR, p. 24, p. 43, p. 168). Interpretation: The infant conceives of the displacement of things as extensions of his own activity (CR, p. 102), or as being at the disposal of his activity (CR, p. 20), not as external autonomous events. Moving objects are not considered as being animated by an independent motion (CR, p. 168), nor as capable of following an autonomous trajectory independent of the subject’s action (CR, p. 168). (Note: It is very unlikely that the infant interprets every movement in his environment as one of his own making. There must be many occasions, such as an adult leaving the room, when the object moves against the will of the infant. This point is further discussed on p. 139). Inference: The movements of objects and movements of self do not exist as two separate conceptual categories in the child’s thought. (See Note above).

132 A. Ninio

1nfcrence: Since the infant has no means of distinguishing the movements of an object from his own movements of obtaining the perception, he cannot, by necessity, distinguish the source of change in the shape of objects, namely, he cannot specify whether the visual change resulted from movements of the object or from movements of his (the infant’s) body. (Hypothesis A proven.) (Note: In all nine situations comprising the 1st evidence on Hypothesis A the infant’s behavior is observed during search for vanished objects. There is a basic incongruity in inferring from data collected exclusively in the absence of objects (i.e., search) to perceptual processes occurring in the presence of objects (i.e., specifying the agents of visual change.) There remains always the possibility that regardless of the plausibility of the argument, the two processes are governed by different laws and are essentially independent.) 2nd Evidence: The infant cannot imitate facial gestures (Piaget, Play, dreams and imitation in childhood, obs. 17- 18, p. 27-29). Inference: The infant does not have visual knowledge of his own face (CR, p. 110). (Presupposition: Visual knowledge of one’s own face is a necessary and sufficient condition for the imitation of facial gestures.) Inference: The infant cannot perceive (or imagine) the movements of his own eyes and head (CR, p. 1 IO). (Presupposition: Visual knowledge of one’s own face is a necessary and sufficient condition for the perception of the movements of one’s face.) Inference: The infant does not conceive of himself and his movements as existing independently of the perceived world (CR, p. 108). Inference: Since the infant does not differentiate between himself and the world, he cannot by necessity distinguish between sources of change in the shape of objects, namely, he cannot specify whether the visual change resulted from movements of the object or from movements of his own body. (Hypothesis A proven.) 3rd Evidence: The infant shakes his head and body and simultaneously looks at hanging objects (Piaget, The origins of intelligence in children, obs. 112, 115, 118, pp. 200-205: CR, obs. 88, p. 158-159). Interpretation: The infant wants to make the distant object move by his own movements (CR, p. 160- 16 1). (Note: Alternative interpretations are more plausible, e.g., that the infant investigates the effect of his movements on the visual scene.) Inference: The infant believes that displacements of objects are at the disposal of his activity, and an extension of this activity (CR, p. 20, p. 117).

Piaget ‘s theory of space perception in infancy

133

Inference: The infant has no means of distinguishing the source of change in the shape of objects; namely, he cannot specify whether the visual change resulted from movements of the object or from movements of his own body. (Hypothesis A proven.) (Note: The basic premise on which the proof of Hypothesis A rests is that the infant cannot and does not differentiate between perception and the perceived world, or, between his own activity and the movements of objects. This premise might be argued against by pointing out the existence of three sets of information, each providing sufficient basis for the distinction between movements of the self and movements of objects. The first set of information are non-visual, kinesthetic signals which report on the movements executed by the self, and which are obviously absent when the perceived movement was made by an agent other than the self. Piaget discussed this objection in “The Child’s Construction of Reality” (CR, p. 104), and rejected it by claiming that since the infant cannot distinguish internal from external as such, infant thought can attach any sort of body sensation to any perceived movement. Analysis of Piaget’s answer reveals it to be begging the question: as long as it is not shown that the infant cannot in fact make use of available (sensory) information to make distinctions between internal and external agency of movement, it remains unproven that he cannot distinguish internal from external as such. The second set of information is the presence or absence of volition. Apart from special circumstances, the self’s movements are voluntary, active and self-initiated. This is a powerful marker which is absent from situations where the self is quiescent. It has even been claimed (Von Holst and Mittelstadt, 1950) that the will to move (rather than the execution of the movement) is the central factor in determining whether a given visual change will be perceived as movement of the self or of the external world. The third set of information is purely visual. As Gibson (1950) has so convincingly pointed out, the visual effect of a movement of the self is usually completely different from the visual effect of a movement of an object. The movement of the self causes a complete transformation of the whole visual field, while the movement of an object only transforms that object plus a small part of its background. To recapitulate, there exists visual, kinesthetic and control information which should enable the infant to distinguish with certainty between the different agents of visual change. In order to claim, as Piaget does in Hypothesis A, that the infant cannot or does not make use of these systems of information in order to distinguish between movements of self and movements of objects, it is necessary to prove directly that his perception is dif-

134

A. Ninio

ferent or erroneous. As pointed out earlier, most of Piaget’s evidence cerns search situations, namely observations on infant behavior in absence of the (relevant) objects. There is no direct evidence to show the infant’s perception of visual change is qualitatively different from of the adult. Hypothesis A is not proven.) *

*

conthe that that

8

B. Hypothesis: The infant has no means of distinguishing the source of change in the shape of objects, namely, he cannot specify whether it has resulted from a change of state or from a change of position of the object (CS, p. 9, p. 103). Evidence. 1. Observations on the behavior towards vanished objects or parts of objects; 2. Observations on the imitation of facial gestures; 3. Observations on shaking of head and body while looking at hanging objects; 4. Observations on the sucking of desired parts of objects. Evidence l-3: The first three sets of evidence were described in detail in connection with Hypothesis A. Inference: (from the first three sets of evidence): The infant cannot distinguish between the movements of the object and the movements of his own body (Hypothesis A). Inference: The infant confuses the act of finding a displaced image with the act of recreating it (CR, p. 103). Inference: The infant cannot distinguish between changes of state and changes of position in the object (Hypothesis B proven). (Presupposition: The elementary distinction between change of position and change of state of an object is that the former can be compensated by body movements which reinstitute the former perceptual state, while the latter cannot be compensated or annulled in this way. Therefore, the infant must be able to distinguish between compensable and non-compensable changes in the visual world in order that he be able to distinguish changes of position from changes of state in objects (CR, p. 102).) (Note.- Since Hypothesis A was not proven, it cannot be assumed as true and used to prove Hypothesis B. In consequence the first three sets of evidence are irrelevant to Hypothesis B.) Fourth evidence: The infant, wishing to suck a specific part of an object, cannot turn the object immediately to the desired position, guided by vision, but has to grope around randomly with his mouth until he finds the desired place by chance (CR, obs. 67, p. 106-107; obs. 76, p. 123).

Piaget’s theory of space perception in infancy

135

Generalization: The infant can adjust his mouth to the object but not the object to his mouth (CR, p. 108). Interpretation: For the infant, space is the property of his mouth, not of objects (CR, p. 108). For him, there are no objects, only personal acts (CR, p. 35). Inference: Since, for the infant, objects do not exist, it is impossible for him to distinguish between changes of state and changes of position in objects. (Hypothesis B proven.) (Note: It might be possible to interpret the fourth evidence as showing merely that the infant can adjust his mouth but not his hands if he wishes to suck a specific part of an object he holds. This difference might have various reasons, such as better control over the movements of the lips than over the movements of the hands; or, the fact that the target is defined by the specific tactile sensations it arouses but not, yet, visually. Neither of these necessitates interpretation in terms of the non-existence of objects.) *

C. Hypothesis:

*

*

A change in the perspective of the object is perceived by the as an actual transformation of the object (CS, p. 11; CR, p. 168). Evidence: Observations on the behavior towards rotated objects. Observations: The adult rotates an object so that the infant is presented with an unfamiliar face of it. The infant does not turn the object so that its familar (or desired) part will face him unless he can see some of the familiar part. For example: he will not rotate a reversed feeding-bottle (presented with the bottom facing the infant) unless some part of the nipple is visible (CR, obs. 77, pp. 124- 125; obs. 78 and 78a, pp. 126-l 28). Interpretation: The infant considers the unseen part of the object to be “reabsorbed” into the object and to cease to exist spatially, rather than to be rotated and obscured by the visible parts of the object (CR, p. 130). Inference: A change in the perspective of the object is perceived by the infant as an actual transformation of the object (Hypothesis C proven). (Note: If hypothesis C is true, the infant should not recognize any perspective change of the object as a change of position. However, on Piaget’s own evidence there are some changes of perspective which are recognized as such by the infant. I refer here to situations in which at least a small fraction of the familiar part of the object is visible, for example, of the nipple of a feeding-bottle. Since the familiar part (the nipple) seen from an unfamiliar angle (from the direction of the bottle’s base) is a true example of a changed perspective of the object, it should not be recognized by the infant

136

A. Ninio

infant. Nevertheless, the object is recognized be concluded that Hypothesis C is not true.)

and correctly

rotated.

It might

D. Hypodresis: There is no shape constancy (CS, p. 1 1). Evidence: Observations on the behavior towards rotated objects. Observutions: The adult rotates an object so that the infant is presented with an unfamiliar face of it. The infant does not turn the object so that its familiar (or desired) part will face him unless he can see some of the familiar part (CR, obs. 77, pp. 124~-125; obs. 78 and 78a, pp. 126.--128). ftzterpretution: The infant considers the unseen part of the object to be “reabsorbed” into the object and to cease to exist spatially, rather than to be rotated and obscured by the visible parts of the object (CR, p. 130). hzfuence: The infant cannot fit the momentary perception into an organized schema of rotary movements (C’S, p. 2 16). Irzferenw; There is no shape constancy in the infant’s perceptual world (Hypothesis D proven). (Presupposition: Shape constancy is necessarily dependent on the existence of a schema for the organization rotary movements.) (Note: It should be nored that Piaget does not refer in Hypothesis D to shape constancy as the process whereby the perceived shape of an object is constant in spite of a change in its perspective (e.g., Epstein and Park, 1963). Rather, Piaget refers to the knowledge that the object possesses a permanent three-dimensional form which remains unchanged through alterations in its perceived shape (CR, p. 148). Such knowledge might affect the perception of the object even if it is seen as different from different viewpoints. That Piaget does not refer here to the constancy of perceived shape is clear from his examples, in which the test for shape constancy is not that the object would be perceived in any particular way, but that it would be rwogr2izcd as the same object through changes in viewpoint (CS, p. 5). For example, a square which appears lozenge-shaped as a result of perspective should be recognized, through shape constancy, as a square, i.e., as a figure having equal angles and equal sides (CS, p. 11); a cube seen in perspective so that none of its sides appears as a square, should be recognized as a cube (C’S, p. 15); a feeding bottle, seen from the direction of its bottom, should be recognized as the same object that contains a nipple (CR, pp. 128-130; p. 168); a toy duck, usually seen from the direction of its head and back, should be recognized if seen by its base (CR, p. 128). In some of these examples, perceptual shape constancy is actually impossible, since they

Piaget k theov of space perception in infancy

137

concern objectively different forms, such as the two opposite faces of a completely rotated object, e.g., the bottom and nipple of a feeding bottle.)

E, Hypothesis: There is no size constancy (CS, p. 5). Evidence: An experiment by Egon Brunswik and Ruth M. Cruikshank, (Perceptual size-constancy in early infancy. Psychological Bulletin, 1937,34,

713-714). Method: Infants of three to eight months of age were presented with three objects: A: a rattle presented at the distance of 25-30 cm; B: the same rattle as in A presented at a distance three times as great; C: an object of the same form and color pattern as the rattle in A and B, but three times as large, presented at the same distance as in B. (Method of presentation is not reported.) The reaching response of babies was studied. (Method of study and the definition of reaching response not reported.) Results: (Responses at three months were not reported in the paper.) At around four to five months of age there was found some relatively slight degree of size-constancy, but on the average, situation C was not responded to with complete equivalence to situation A as would be the case if there were no constancy. (Mode of response not reported either for A, for B or for C.) At about six months of age a differential response to situations A and C is developed, the response to C approaching the response typical for B. Inference: There is no size constancy in early infancy (CR, p. 5). (Hypothesis E proven.) (Note: Because of the lack of details in the Brunswik and Cruikshank paper it is impossible to infer the detailed presuppositional structure of this conclusion. Nevertheless it should be noted that the results of the experiment do not fully warrant the conclusion which was drawn from it, since at the youngest age group for which results were given, there is already a divergence of responses between the two critical situations. It might be concluded that Hypothesis E is not proven.) * F. Hypothesis:

*

*

The infant perceives space at the first period of his life as a fluid mass without depth (CR, p. 145). Evidence: Observations on the absence of systematic binocular convergence until 9 months of age (CR, obs. 68, p. 111).

138

A. Ninio

(Presupposition: Systematic binocular convergence is a necessary condition for depth perception.) (Notes: There are different and conflicting claims made by Piaget about depth perception, e.g., (CR, on p. 111) he says that the infant does perceive at various depths, but that he is not conscious of these depths. It is also possible that Piaget did not intend to offer the observations above as conclusive evidence for the absence of depth in early infancy. In that case Hypothesis F is without any proof.) *

*

*

G. Hypothesis: At a later period in infancy, space becomes differentiated into two zones, “near space” and “far space” (CR, pp. 143-145). Evidence. 1. Observations on the attempts to grasp near and far objects; 2. Observations on the approach to distant objects. Zst Eviderzce: Near objects are grasped while usually there is no attempt to grasp distant objects (CR, obs. 81-83, pp. 135Gl41). 2rzd Evidence: Distant objects are approached in order to be grasped, sucked or seen better (CR, obs. 81 and 83, pp. 136, 139). Interpretution: The infant differentiates between “near space” and “far space”, the former of objects suitable for prehension, the latter of objects which are not suitable for it. (IHypothesis G proven.) (Note.- There is no evidence that infant space at this stage, or at any other, is differentiated into two discontinuous zones, rather than perceived as a continuous whole in three dimensions. The evidence offered for Hypothesis G is consistent with adult-like perception of distances. Therefore Hypothesis G is unproven.) *

*

*

H. Hypothesis: The differentiation between “near space” and “far space” is not that of depth (CR, pp. 143Ll45). Evidence: Observations on the search for hidden or covered objects. Observations: The adult covers an object with a cloth or hides it behind a screen, in full view. The infant does not uncover it or search for it behind the screen (CR, obs. 7, p. 15, obs. 22.-25, pp. 28-32; obs. 28-32, pp. 36-40). I~ztc~rpretutio~z: The infant does not have the concepts of “in front of” and “behind” (CR, p. 11 1). /tlfi>rence: Visible objects are not perceived as arranged in sequential planes one behind the other, ordered according to the third dimension. (Hypothesis 1-Iproven.)


139

(Presupposition

I: Space in adults is perceived as a succession of two dimensional planes, ordered one behind the other according to the third dimension (CR, p. 171).) (Presupposition 2: The infant must acquire the concepts of “in front of” and “behind” as applicable to hidden objects as a necessary condition in the perception of non-hidden objects arranged in three dimensions.) (Note: There is an essential incongruity in using data collected exclusively in the absence of objects to infer on perceptual processes occurring in the presence of objects (= stimuli).) *

*

*

I. Hypothesis:

The objects in “far space” appear to the infant diminished and distorted by perspective (CR, p. 143). Evidence: Observation on the attempts to grasp distant, and near, but unfamiliar, objects. Observations: The infant does not attempt to grasp distant objects, nor near objects which are not very familiar or which are presented in abnormal circumstances (CR, obs. 83, p. 138-140). Interpretation: The infant does not attempt to grasp distant objects since they appear to him distorted, unfamiliar, and linked to a context in which direct prehension has never intervened (CR, p. 143). Znference: Distant objects appear to the infant distorted and diminished by perspective. (Hypothesis I proven.) (Note: The similarity between the infant’s behavior in two situations is not by itself a proof that the reason for that behavior is identical in both cases. It is still possible that the infant does not attempt to grasp distant objects not because they look unfamiliar but because they look distant. Therefore Hypothesis I is unproven.) *

J. Hypothesis:

“Far

space”

*

*

is undifferentiated

into depth

planes (CR, p.

145). Evidence: None. Proof I: Differentiation

into depth planes through prehension, which has just been attained for “near space”, is obviously impossible for “far space”, therefore “far space” remains undifferentiated into depth planes (CR, p. 146). (Hypothesis J proven.) (Presupposition: Near space is differentiated into depth planes through the coordination of vision and prehension (CR, p. 149.)

140 A. Ninio

(Note: Prehension cannot be a necessary condition for the differentiation of depth in “far space” since it is logically impossible ever to grasp distant objects. Since it is never a fiecessary condition, the lack of it in infancy does not imply that infants cannot differentiate depth in “far space”. Hypothesis J unproven.) Proof 2: The three necessary conditions for adult perception of distant depth are not yet realized. These are: a. The greater number of objects interposed between object and observer, the more distant the object seems; b. Objects interposed on each other are perceived in varying depth; c. The different speeds of displacements observed by moving the head or the whole body enable the observer to evaluate the parallax of distant objects (CR, p. 147). into depth planes (Hypothesis Therefore, “far space” is undifferentiated J proven). Evidence: None. (Presupposition: Depth perception in adults is necessarily based on the number of objects between observer and distant object, on interposition, and on movement parallax.) (Note: Since no evidence is offered, Hypothesis J is unproven.) *

*

*

K. H.vpot/zesi.s: The infant does not perceive spatial relations of projective and euclidean (i.e., higher than topological) order (CS, p. 6). Evidence. l-2--3: The absence of shape constancy in the infant. 2-3-4: The absence of size constancy in the infant. First Evidmce: The absence of shape constancy in the infant has been discussed as Hypothesis D (pp. 2 l--22). The main evidence has been difficulties with completely rotated objects, e.g., a feeding bottle. /rzfchozce: The infant is unable to coordinate the various perspective projections of an object (CS, p. 12). (Prcslcppositio~z: Coordination of perspective projections is a necessary and sufficient condition for shape constancy (CS, p. 12). Irlfberzce: The infant does not perceive the projective relationship between different perspective projections of an object. (Hypothesis K proven.) (Prcsuppositiorz: There exist projective correspondences between perspective projections of an object.)


141

(Note: It should be noted that there is no projective correspondence (or relation) whatsoever between the opposite sides of a completely rotated object, e.g., between the bottom and the nipple of a feeding-bottle. It might be true that between any two perspectives in most objects there exist an infinite number of intermediate perspectives, each having some projective correspondence with its neighbors, so that these correspondences are capable, as Piaget claims (in CS, p. 12) of coordinating the different perspectives into an integrated whole. Furthermore, it might also be possible that the infant has difficulty in taking advantage of these correspondences in order to recall one perspective on the basis of seeing its opposite; or that he cannot make use of the network of coordinated perspectives in order to correctly rotate an object to a favored orientation. Nevertheless, Piaget does not offer any evidence that infants are unable to take note of and make use of a direct projective correspondence, if it exists. On the contrary, his evidence points to the opposite conclusion: ;trbe,levcr thz.cz ir. ~0.~3 possibility of &ding a projective correspondence (namely in those situations when a small part of the “favored part”, e.g., the nipple of a feeding-bottle, is showing), the infant acts in a positive way, e.g., rotates the bottle. It might be concluded that Hypothesis K is incorrect as a general unqualified description of the infant’s perceptual limitations.) Second Evidence: The absence of shape and size constancies in the infant have been discussed as Hypothesis D (pp. 21--22) and Hypothesis E (pp. 23-24). Inference: The infant cannot recognize or reconstruct a fixed euclidean shape (CS, pp. lo- 12). (Presupposition I: Shape constancy is the recognition or reconstruction of euclidean shapes, namely, of shapes possessing constant euclidean properties such as length, angle, etc. (CS, p. 1 l).) Size constancy is the reconstruction of euclidean shapes, namely of shapes possessing constant size. (Presupposition 2: The absence of shape and size constancies necessarily implies an inability to recognize or reconstruct the euclidean properties of an object.) Inference: The infant does not perceive the euclidean properties of an object (Hypothesis K proven). (Presupposition: Inability to recognize or reconstruct the euclidean properties of an object necessarily implies that such properties are not perceived in general.) (Note: Piaget begs the question of topological primacy in the juxtaposition of recognized and non-recognized shapes. Whenever there is “recognition of

142

A. Ninio

shape independent of perspective” in infancy, such as of faces, he identifies the basis of it as a topological, homeomorphic correspondence. However, of all the objects the infant cannot recognize at this age he discusses only a small subset, all of which are “euclidean”, rigid objects with well-definable metric attributes, such as a feeding-bottle or a celluloid duck (CR, pp. 1266 128). Even his non-empirical examples for shape constancy are of this character; a cube; a square, a rectangle (CS, pp. 1 ll15). It might be argued that faces have higher-order invariants serving for recognition, such as the triangle of the two eyes and the nose; and that furthermore Piaget’s test for recognition is completely different for rotated objects on the one hand, and for faces, on the other hand; for example, the infant is never required to rotate a face or to manipulate it in any way in order to demonstrate recognition. For these reasons it might be concluded that Hypothesis K is unproven.) Third Evidence: The absence of shape constancy in the infant has been discussed as Hypothesis D (pp. 21 -22). The absence of size constancy in the infant has been discussed as Hypothesis E (pp. 23-24). Inference: The infant is unable to abstract a straight line from the boundaries of a complete figure, as long as the figure is not yet assumed to indicate the presence of objects with fixed shapes and sizes (CS, p. 10). (Presupposition: Abstracting straight lines from boundaries of figures is impossible if these figures are not assumed to indicate the presence of objects with fixed shape and size.) fnference: The infant cannot perceive straight lines (C’S, p. 10). (Lines are projective invariants, therefore Hypothesis K is proven.) (Presupposition: Straight lines exist perceptually only if they are first abstracted from boundaries of figures (CS, p. lo).) Fourth Evidence: The absence of size constancy in the infant has been discussed as Hypothesis E (pp. 23 -24). Itzfercrzce: The infant is unable to coordinate the various perspective projections of an object which changes in size with varying perspective (= distance) (CS, p. 11). (Presupposition: Coordination of perspective projections of various size is a necessary and sufficient condition for size constancy.) Irzfcrence: The infant does not perceive the projective relationship between perspective projections of the object which vary in size as a function of distance (Hypothesis K proven). (Presupposition: There exists a projective correspondence between perspective projections of the object which vary in size as a function of distance.)

Piaget s theory of space perception in infancy

143

(Note: The geometrical relations between receding “perspectives” of an object is that of similarity, which is not a projective but an affine relation. This in itself does not disprove Hypothesis K which refers generally to relations of higher than topological order.)

Discussion Although the orientation of this paper has been mainly descriptive, two points of criticism have emerged which warrant further discussion. The first issue concerns Piaget’s heavy use of the search paradigm in proving perceptual hypotheses, e.g., on the perception of the agency of visual change; on shape constancy; on the perception of depth. Piaget prefers this paradigm on the grounds that the child’s search behavior reflects the understanding, consciousness and structuring of sensation, rather than the sensory activity itself (CR, pp. 134-5, pp. 213-2 14). One could easily accept Piaget’s argument that perception involves more than mere sensation, and still insist that sensory activity is abolutely necessary for perception to occur. On this ground, observing the infant’s behavior in the absence of stimulation is not an appropriate way to arrive at valid conclusions about the nature of perceptual processes, although there might be other processes which this paradigm elucidates, namely representation in memory, search strategies, probability learning, etc. It might be possible to claim that these latter processes are perfectly identical to the “intelligent” component of perception, and it is even possible that this represents Piaget’s point of view, but this sort of argument cannot be accepted without strong proof, which is absent from the Piagetian corpus. The second issue is formal, and concerns Piaget’s widespread reliance on unproven claims about the nature of perception. For instance, he assumes that imitation of facial gestures depends on visual knowledge of one’s face; that near space is differentiated into depth planes through the coordination of vision and prehension; that straight lines are perceived only if they are abstracted from the boundaries of objects, given that these are attributed with constant size and shape, and so forth. Since the falsity of a presupposition falsifies the premise that depends on it, the analysis of presuppositions should take an integral part in any critical examination of Piaget’s theory.

References Barlow,

H. B., Blakemore, C., & Pettigrew, J. D. (1967) discrimination. J. Physiol., 193, 327-342.

The neural

mechanisms

of binocular

depth

144

A. Ninio

Bower,

T. G. R. (1968) Morphogcnetic problems in space perception. In: D. Hamburg & K. Pribram (L-Ids.), Proc. Assoc. for Research in Nervous and Mental Diseases. Stanford, California, Stanford University Press. Brunswick, L., & Cruickshank, R. M. (1937) Perceptual size-constancy in early infancy. Psychol. Bull., 34,713-714. Epstein, W., & Park, J. N. (1963) Shape constancy: Functional relationships and theoretical formulations. Psychol. Bull., 60, 265-288. J’lavcll, J. H. (1963) The developmentalpsychology ofJean Piaget. Princeton. Van Nostrand. Gibson, J. J. (1950) The Perception of the Visual World. Boston, Houghton-Mifflin. Ginsburg, H., & Opper, S. (1969) Piaget’s Theory of Intellectual Development. Cnglewood Cliffs, Prentice-Hall. Holst, E. von, & Mittelstldt, H. (1950) Das Reafferenz-prinzip. Naturwissenschaften, 20, 464-416. Hubel, D. H., & Wiesel, T. N. (1962) Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. .I. Physiol.. 160. 106-154. Hunt, J. McV. (1961) Intelligence andExperience. New York, Ronald. Maturana. J. R.. Lettvin. J. Y., McCulloch. W. S.. & Pitts. W. B. (1960) Anatomy and nhvsioloav of vision in the frog. j. Gen. Physiol., 43, Supplement, 129~- 171: Pettigrew, J. D., Nikara, T., & Bishop, P. 0. (1968) Binocular interaction on single units in cat striate cortex: Simultaneous stimulation by simple moving slit with receptive fields in correspondence. Rxper. Brain Res., 6, 391-410. Piaget, J. (1951) Play, Dreams and Imitation in childhood. New York, Norton. Piaget, J. (1952) The Origins of Intelligence in Children. New York, International Universities. Piagct, J. (1955) The Child’s Construction of Reality. London: Routledge and Kegan Paul. Piaget, J., & Inhelder, B. (1956) The Child’s Conception ofSpace. London: Routledge and Kegan Paul. Wolff, P. (1960) The developmental psychologies of Jean Piaget and psychoanalysis. Psvchological Issues, Monograph No. 5, New York, International Univcrsitics.

Resume On prksente la theorie de Piaget sur la perception de I&pace par les jeunes bdbes dans le cadre d’un systime hypothetico-deductif. Les auteurs definissent 11 hypotheses portant sur: I’agent du changement visuel, des constances de forme et de taille, de la distance et dc la perception des relations d’un ordre superieur entre les elements spaciaux. Les propositions de Piaget pour chacune de ces hypotheses sont present&es en plusieurs ttapes: la preuve comportementale, I’interpretation en terme d’etats inte’rieurs, les inferences et ge’neralisations. Les presuppositions qui sous-tendent les arguments sont mis en evidence et des notes critiques sont ins&&es aux moments approprits. Enfin, les conclusions g&&ales sont brievement discutees.

Cognition. @Elscvier

7 (1979) 145-176 Sequoia S.A., Lausanne

3 -- Printed

in the Netherlands

Relative clause structure, relative clause perception, and the change from SOV to SVO* FRANCESCO

ANTINUCCI

Is titu to di Psicologia, Consiglio Nazionale delle Ricerche ALESSANDRO

DURANTI

lJniversit5 di Roma LUCYNA

GEBERT

Universid di Genova

Abstract This study presents a view of diachronic change in language, according to which one of the fundamental factors motivating syntactic change is to be found in the conflicting interaction of principles determining the language organization. Specifically, it will be argued that principles of structural nature and principles of perceptual nature are in conflict in languages of the SOV type, because of the relative clause construction. The way in which a relative clause is structured in an SOV language is an obstacle to its effective perceptual processing. It will be argued that this conflict is one of the major factors determining the diachronic change of a language from an 0 V to a VO typology.

1. Introduction In this paper we shall direct our attention to what we consider to be one of the fundamental mechanisms of diachronic syntactic change: the conflict between principles that determine language structure and principles on which actual language processing is based. The specific aspect of this conflict to which we shall address ourselves is the one generated by relative clause construction in certain types of language. We shall attempt to show *Requests for reprints should be addressed Via dei Monti Tiburtini, 509, Rome, Italy.

to Francesco

Antinucci,

Istituto

di Psicologia,

CNR,

146

F. Antinucci,

A. Duranti and L. Gebert

that there is a conflict between the way in which a relative clause is structured in an SOV language and the perceptual strategies needed to effectively process a sentence in the same type of language, and that this conflict is one of the major motivations for the shift from an SOV to an SVO (or, better, a VO) typology. The paper will be organized in the following manner. First, we will examine how the relative clause is structurally organized in language. Second, we will discuss the perceptual strategies that process sentences containing a relative clause. Third, we will derive some diachronic predictions from the interaction of the two preceding factors. Fourth, we will present various kinds of evidence that seem to substantiate the diachronic predictions.

2. The Structure of Relative Clause For the purposes of the present discussion, we will consider restrictive relative clauses (RC) depending on a full (i.e., non-pronominal) head noun. A restrictive RC has the function of specifying to which individual(s) of the class denoted by the head noun the speaker is referring (see Keenan, 1972). We will assume, in accordance with various proposals (see, for example, Thompson, 1970), that on the semantic level a RC is represented as a sentential structure containing an NP identical to the head NP.’ On the surface level, there is a great deal of variability among languages as to the form in which RCs occur. Nonetheless, there are some general features that can be used to characterize them cross-linguistically. First of all, a restrictive RC is a noun-modifier and as such is syntactically part of the same NP containing the head noun. Second, since the RC has a noun identical to the head noun, the surface clause manifesting it will have a missing NP, either deleted or pronominalized. Third, the sequence manifesting a RC will include some characteristics marking it as a non-main clause. We don’t mean to imply that the above three features define a RC, in fact we don’t think that such a definition can be given in terms of surface characteristics. What we do mean is that they are typically associated with the manifestation of a RC, even if in certain languages some types of RC may lack one or another of them. Let us now consider the position occupied by the RC with respect to the head noun. It is a well known typological fact that in OV languages the RC is usually placed on the left of the head noun, while in VO languages (which

‘See below,

pp. 10-11.

The change from SOV to SVO

147

include SVO, VSO, and VOS languages) it is placed on the right of the head noun. Thus, OV languages like Turkish, Japanese, Burmese, Korean all have the RC on the left of the head noun, while VO languages like Arabic, English, Italian, Yoruba, Tagalog have the RC on the right of the head noun. There are, to be sure, languages that appear to be an exception to such a generalization. For example, although Persian is an OV language, its RC appears on the right of the head noun. But for the moment we will set aside these cases, since they will play a crucial role in the final discussion. A number of different proposals have been formulated to account for the generalization relating RC position to the basic order typology of a language. Lehmann (1973) has formulated a universal principle relating the position of noun modifiers (adjectives, genitives, relative clauses) and what he calls “verb modifiers” (like negation and causation) to the basic relative order of object complement and verb: modifiers are placed on the opposite side of a basic syntactic element from its primary concomitant. The primary concomitant of a verb (V) is its object (0), and viceversa. Therefore, in languages with a basic VO order, noun modifiers will be placed on the side of the noun opposite to the side where the verb appears. Since V appears on the left of 0, noun modifiers will be placed on the right of the noun. Conversely, with a basic OV order, noun modifiers will be placed on the left of the noun. Thus, according to this principle, since RC is a noun modifier, it will appear in postnominal position in VO languages and in prenominal position in OV languages. Kuno ( 1974) tries to offer an explanation for the generalization contained in Lehmann’s principle. He starts from the well known observation that center-embedding drastically reduces the comprehensibility of sentences. Then he takes into account an SOV language on one side and a VSO language on the other, and considers the effect of placing RC on the right or on the left of the head noun in both types of language. SOV - Prenominal (1) [RCIS 0 V (2) S [RCIO V

RC

SOV - Postnominal (3) S[RCl 0 V (4) S O[RC] V

RC

VSO - Prenominal (5) V [RCI S 0 (6) V S [RCIO

RC

VSO - Postnominal (7) V S[RC] 0 (8) V S O[ RCI

RC

In SOV languages prenominal position of RC is better than postnominal position, because it reduces the probabilities of having a center-embedded clause. In fact, while in postnominal construction both subject-modifying (3) and object-modifying (4) RCs are center-embedded, in prenominal construction this happens only to object-modifying RC (2). Conversely, in a

148 F. Antinucci, A. Duranti and L. Gebert

VSO language it is the postnominal position that minimizes the probabilities of center-embedding, as shown by (7-8) lierSU,s(5-6). Therefore, on the assumption that language patterns will tend to minimize perceptual difficulties, the different choice between prenominal and postnominal position of RC in the two types of languages is accounted for. It seems to us that both Lehmann’s and Kuno’s proposals have some internal weaknesses. Lehmann’s principle is descriptively adequate, but, as Kuno correctly observes, no explanation or justification for it is given. Why should modifiers be placed on the opposite side of a basic syntactic element from its primary concomitant, rather than, say, on the same side? Until this question is answered, Lehmann’s principle is not different from an empirical generalization. Kuno’s proposal, on the other hand, is directed toward the explanatory question. But we see two major problems with his solution. First of all, it. doesn’t account for the pattern found in SVO languages. In these languages both prenominal and postnominal position of RC give rise to exactly the same probabilities of center-embedding. Thus, if avoidance of centerembedding is what determines RC position, one would expect these languages to freely show either postnominal or prenominal RC. But this is false, since it is well known that SVO languages have postnominal RC.* A second (and more serious) problem has to do with the nature of the perceptual difficulty in center-embedding. Kuno bases his argument on the assumption that center-embedding is perceptually bad in any case. But this is clearly not true. There are no difficulties in processing sentences containing center-embedded clauses, as long as the degree of center-embedding is not higher than one. Now, in the RC case, two degrees of center-embedding are found only when there is a center-embedded RC one of whose NPs is in turn modified by a RC. This in itself is such an improbable situation in actual language use, that it is hard to think that a certain language structure is motivated by the need to avoid it. In this light, Kuno’s hypothesis loses much of its original appeal. Venneman (1973) (see also Bartsch, 1972; Bartsch and Vennemann, 1972) makes a quite different proposal. He tries to account for the position of noun modifiers in terms of a general principle underlying the whole linear organization of the sentence, rather than in terms of some principle specific to (noun) modifiers. This is called the Principle of Natural Serialization, and it accounts in a general way not only for prenominal position of noun modifiers in OV languages and for their postnominal position in VO *Actually, if one includes would expect SVO languages

intransitive sentences in the probabilities’ to have prenominal RCs.

count,

as Kuno

does,

then we

The change from SOY to S VO 149

languages, but also for the whole array of features typically present in the two types of languages. This proposal is too long to be summarized here and it will be taken for granted. We basically agree with Vennemann’s approach, though not with all the details of his solution. Therefore, we will now formulate another proposal, which, although quite similar to Vennemann’s in both its underlying logic and its predictions, makes some additional claims on the structural status of the RC construction that will be essential for our discussion. Language is a mechanism used to map meaning into sound. In very general terms, we can conceive such a mapping mechanism as a function, whose input is any of an infinite set of meanings and whose output is a uniquely specified sequence of sounds. In order to understand its nature, we have to consider the task this function has to perform. The form of its output is a linear sequence of elements, while the form of its input is a hierarchical structure. In other words, the form of the output is a series (9)

a+ b + c+ ...n

structured in terms of a “precedes” (or “follows”) relation; b”, “b precedes c”, “... precedes n”. The form of the input (meaning) is instead a structure

(lo) 1 PRED a

ARC

ARC PREflRG

i.e., “a precedes

PREiRG II!

E

where the defining relation is “E is argument of D”, “D is second argument of A”, “C is argument of B”, “B is first argument of A”.3 From this point of view, the task of the function we want to define is that of translating in a general and systematic way hierarchical structures like (10) into linear sequences like (9). Let’s consider how this task might be accomplished, starting from the simplest case. The minimal semantic structure underlying a sentence is that formed by a predicate with one argument. For example,

3As a representation of meaning, we are here assuming a kind of logical structure, but this is not essential for the proposal we are going to formulate. Any kind of hierarchical representation (for example, “dependency trees”) will work in the same way.

150

F. Antinucci, A. Duranti and L. Geherl

/A

(11)

PRED

I

CLOSE

ARC

I

DOOR

corresponding to the door is closed. The lexicon assigns the phonological segment be closed to the predicate CLOSE and door to DOOR. Once this is done, there are only two possible ways of translating a structure like (1 1) into a linear sequence. Either the lexical item corresponding to the argument is placed on the right of the one corresponding to the predicate or on the left of it. Thus, corresponding to (11) we have two possible linear sequences;4 “be closed door” “door be closed”

(12) VNo (13) No V

We will call (12) “rightward” expanded and (13) “leftward” Consider now a slightly more expanded semantic structure

expanded.

(14)

The lexicon determines the following correspondences: the encircled predicates correspond to close, JOHN toJohn and DOOR to door. Given that the lower part of (14) can be realized as either (12) or (13), the problem is what place will be assigned in the linear sequence to the item John. If the mapping mechanism of language were completely free, John could appear in six different positions:’ NA

VNo

NA

No

v v

No

NA NA

No v

v No

No

NA

VNA

4We designate the nominal argument of (11) as NO because its semantic role in structures like (11, 14, 18, 23) corresponds to that of a “Deep Object” (in Fillmore’s terms). This appears as the surface object in transitive sentences (see 14). The nominal corresponding to JOHN is designated with NA, because its semantic role in structures like (14) is that of the “Agent”.

The change from SO V to S VO

15 1

But we maintain that the function regulating the mapping mechanism of language is based on the following fundamental principle: to every logical expansion in the semantic structure there corresponds a parallel linear expansion of the surface sequence in a constant direction.6 The structure

PRED

ARC

ARC

I

BECOME

I

can be considered a logical explansion of the minimal structure (11). In fact, both (11) and (14) underlie a sentence, but (15) doesn’t, and (11) is a proper part of (14). Therefore, the principle predicts that any element manifesting (15) will be added to the sequence of elements manifesting (1 1) following the same direction of linear expansion. Thus, if a language maps (11) by means of (12) (i.e., by adding No to the right of V), John will be added to the right of sequence ( 12): (16)

VNo

“close door John”

NA

If a language maps (11) by means of (13) (i.e., by adding No to the left of V), John will be added to the left of sequence (13): “John door close”

(17) NA No V Consider now a structure

like

(18)

CAUSE

JOHN

PRED I BECOME

ARC A PRED

ARC

1 CLOSE

I DOOR

6The following presentation will be highly informal and simplified. We will give only a general idea of how the principle works, limited to those cases that are relevant to our discussion. Several problems requiring detailedjustification will be deliberately ignored. For a full treatment, see Antinucci (1977a).

152 F. Antinucci, A. Duranri and L. Gebert

underlying a sentence like John closed the door at five. Again, the encircled structure is a logical expansion of structure (14). Therefore, the principle predicts that its manifestation will be added to the sequence manifesting (14) in the same direction. Denoting at five with ADV (Adverbial), we will have, accordingly (19)

V No NA ADV

in languages showing (16) and (12), and (20) ADV NA No V in languages showing ( 17) and ( 13). However, ADV consists of two lexical items: how is their relative order determined? In both sequence (12) and (13) corresponding to the minimal structure (1 l), we can say that the predicate “precedes” its argument, since (12) grows from left to right and (13) from right to left. Therefore, also in (19) and (20) the predicate of ADV its argument in the linear sequence. Thus, we will have will “precede” “close door John at five” V No NA A N (21) “five at John door close” (22) N A NA No V (where A (adposition) is the predicate of ADV and N its argument). Finallv. consider

a structure

like

(23)

CLOSE

DOOR

underlying a sentence like John closed the door because he wus sick. Again, the encircled structure is a logical expansion of (14). It is formed by a prediARC If we denote the first with cate BECAUSE and its argument n’ PRED SI&

ARC

JOh

C (Conjunction) and the second with S (Sentence), linearizations determined by the principle will bi VNo (24) (25) S C NA No V

NA C S

the corresponding


The order internal to S will be obviously tion of the principle, giving rise to VNo

(26) (27)No

VC

N* No V

NA CVNo

determined

by a recursive

153

applica-

“closed door John because was sick John” “John was sick because John door closed”

Even from this very rough presentation, it is easy to see how the principle of expansion correctly predicts the characteristic array of features found in the basic language types. If we disregard the behavior of the so-called the principle correctly predicts the existence of two types of Subject,’ language. Those whose linear sequence is built from left to right (rightward expanding) and those whose linear sequence is built from right to left (leftward expanding). In the first type, the nominal constituents of a sentence (excluding the Subject) will follow the main verb, as in (12) and (16). Adpositions will appear as prepositions, since they will be placed on the left of their argument, as in (21). Adverbials will appear in clause final position, as in (19). Adverbially subordinated clauses will follow the main clause, as in (24). Conjunctions will appear at the beginning of the subordinate clause, as in (26).6 On the other hand, in the second type the nominal constituents of a sentence will precede the main verb, as in (13) and (17). Adpositions will appear as postpositions (or case-markers), since they will be placed on the right of their argument, as in (22). Adverbials will appear in clause initial position, as in (20). Adverbially subordinated clauses will precede the main clause, as in (25). Subordinating conjunctions will come at the end of the subordinate clause, as in (27). In short, the principle correctly predicts the characteristics of the so-called VO and OV language-types.9

‘The position and behavior of the so-called Subject is determined by a totally distinct principle (Topical Movement), whose basis is not the logical structure of sentence meaning, but rather aspects of meaning related to the distinction between Given and New information. This matter will not be pursued at all here (see Antinucci, 1977a). ‘Since we are disregarding subject position, this type includes SVO, VSO, and VOS languages. It is, in fact, well known that apart from subject position these languages share the same syntactic characteristics. ‘Notice that the English glosses of our leftward expanding formulas correspond to the typical sentence structure of an SOV language. On the other hand, if we move an NP internal to each clause at the beginning of its clause, the glosses of our rightward expanding formulas correspond to the typical sentence structure of an SVO language (which is not characterized by having the verb in medial position, but by having the verb in second position, while all the remaining sentence constituents are placed after it). As said before, this movement is accomplished by a different principle (Topical Movement).

154 F. Antinucci,

A. Durunri and L. Gebert

The motivation for the existence of a principle like that of expansion should be obvious: it offers a general solution to the basic problem faced by the mapping mechanism of language. It can translate univocally any of an infinite set of logical structures representing meaning into the appropriate linear sequence, and viceversa. Let’s now consider noun modifiers. The (simplified) structures corresponding to sentences (28) (29) (30)

The door of the house was closed The young boy left The boy who met the girl left

are, respectively, (3 1) (a> PRED

I

CLOSE

ARC

PRED

ARC

ARC

I

DOOR

(32) (a)

(b)

A

PRED

ARC

PRED

ARC

LEAVE

BOY

YOUNG

BOY b

(33) (a)

(b)

A

PRED

ARC,

PRED

ARG

I LEAVE

I BOY

I MEET

I GIRL

ARC

As we said at the beginning, noun modifiers can be considered as sentential structures containing a noun identical to the head noun (the encircled argument in (3 l-33)), associated to the main sentence structure. From this point of view, they can be considered as logical expansions of the mainclause structure, and, more specifically, as logical expansions of one of the main-clause arguments (the argument DOOR in (31a), and BOY in (32aOn this basis, the principle of expansion will determine also their 33a)).” position in the linear sequence. As any other expansion, they will “follow” the element (or sequence of elements) manifesting the structure of which ‘“Intuitivcly this is clear enough. L)ut it requires r.xpanslon, given the kind of IneaninF-representation

;L refinement used here.

of the t‘ormal definition

of logical


155

they are an expansion. Since they are an expansion of the argument appearing as the head noun (HN), they will “follow” it in the sequence. Thus, in rightward expanding languages they will be placed on the right of the head noun HN MOD

(34) In leftward noun (35)

expanding

languages

they will be placed on the left of the head

MOD HN

Notice that the principle will also determine fier itself, when this is formed by more than fact, that a lexical item corresponding to that corresponding to its argument. Thus, for (36) (37)

HN AN N A HN

the order internal to the modione lexical item. Remember, in a predicate always “precedes” (3 1) we will have

“door of house” “house of door”

In both cases the adposition manifesting the predicate of (3 lb) (genitive marker, in this case) will come between the head and the specifying noun (N). Finally, (33) will be realized as (38) (39)

HN VN N V HN

“boy [met girl] ” “[girl met] boy”

Therefore, RCs will be placed on the right of the head noun in VO languages, and on the left side of it in OV languages. To return to our problem, we must notice that both our proposal and the one by Vennemann make a specific claim concerning the position of RC. That is, such a position is not an independent variable in the syntactic organization of a language, it is instead inherently tied to the organization of the whole linear sequence. In other words, the “rule” that specifies the position of RC is not an independent rule in the grammar of a language. The placement of RC to the right or to the left of the head noun is determined by the same universal principle determining the position of verb, noun phrases, adposition, conjunctions, subordinate clause, etc. In this sense we will speak of structural factors (i.e., connected to the universal principle underlying the language mapping mechanism) affecting the RC. This does not mean that it is empirically impossible to find a language where the position of RC (or of any other constituent, for that matter) is inconsistent with, say, the relative position of NPs and verb. What our hypothesis implies is: (a) that the sequential organization of such a language cannot be imputed to structural factors only; (b) that the language will be under the pressure of structural factors to reorganize itself.

156

F. Antinucci,


At first this statement could appear to be vacuous. Shoudn’t we consider the existence of “inconsistent” languages as a disproval of the universal principle that we set up? We don’t think that this conclusion follows necessarily, given a correct understanding of the logic of scientific inquiry. We use a certain empirical basis, the relative positions of sentence constituents in a large number of languages, to build a hypothesis concerning the universal basis of the language mapping mechanism. If, at this point, we find some cases that do not conform to the prediction made by the hypothesis, there are at least two ways open to us. The first is to accept such cases as a disproval of the hypothesis, and try to formulate a new one that will include also the “deviant” cases. However, one has to remember that a scientific hypothesis is root equivalent to a generalization of empirical facts, and therefore will not be ipso facto falsified by the existence of apparent counterexamples. (The law of gravity is not ipso facto falsifed if we find some bodies that do not fall to the ground according to the parameters specified by it). The second way is to try to explain away such cases by enriching the model with additional hypotheses. And this can be legitimately done if we can show that these additional hypotheses are independently needed, and that they enable us to derive predictions that go beyond the explanation of the “deviant” cases. In this way, we will try to show that claims (a) and (b) derived from our hypothesis are correct.

3. The Perception

of Relative Clauses

Let’s now pass to our second point and consider the relative clause in relation to the perceptual mechanism that has to process it. First of all, we have to stress a general point. Psycholinguistic research carried out in the past few years has shown that the mechanisms of actual language processing are not isomorphic to the mechanisms postulated in the grammar of a language to relate underlying and surface forms (see Fodor, 197 1; Bever, 1974). Although this has been demonstrated to be true with respect to forms of grammar based on transformational operations, the same conclusion could be easily drawn for models like the one proposed above.” Notice, however, that this conclusion by no means implies that the grammar of a language isolated by a linguist is just a conventional and arbitrary systematization of empirical regularities. The grammar models a real “Notice that a processing model isomorphic to the structural model we have proposed the absurd claim that actual language perception in an SOV language proceeds from starting from the end of the sentence.

would imply right to left.


157

mental capacity, namely the speaker’s knowledge of the complex and systematic relations existing between meanings and sounds in his language (mirrored, for example, in his intuitions about the relations between sentences in his language), and the universal principles underlying grammar model the ways in which such knowledge is organized in the human mind. What we are saying is that this system of the mind is different from the one(s) that effect(s) the actual processing of language in real time. In addition to the negative findings, a consistent body of knowledge has been accumulated in the past years in the field of language perception. The common view held is that the perceptual mechanism for language operates on the basis ofstrutegies of segmentation that directly map surface sequences into underlying semantic representations.” From this point of view, RCs present two basic problems to the perceptual mechanism : 1) the segmentation of the appropriate sequence constituting the subordinate clause from the main clause; 2) the recovery of the missing NP and its function within the RC itself. We will now argue that the performance of both these tasks presents much more severe problems in an OV than in a VO language. A number of different experiments (reported in Bever, 1974) show that in English the fundamental perceptual strategy applied to the processing of a sentence is one that segments together the first N . . V . . (N) sequence as the main independent clause of the sentence, unless there is some kind of subordination marking. Bever found that this strategy is so strong that, when confronted with a sentence like (40)

The editor authors

the newspaper

hired liked laughed

subjects cannot avoid interpreting the sequence in italics as a clause, even when they are given explicit instructions that this interpretation is incorrect. “The NVN sequence - writes Bever - is so compelling that it may be described as a ‘linguistic illusion” which training cannot readily overcome”. The ground for the existence of such a strategy is obvious: it exploits in the most efficient possible way the information about the basic word order of a language. Though the formulated strategy is clearly language-particular, we will assume here that the basic principle underlying this strategy, the exploitation of basic word order to segment a clause, has a universal basis in the perceptual mechanism for language. Leaving aside, for the moment, the question of markers that can block its operation, let’s consider the ‘*For relevant

obvious reasons of space, we will deal here only with those strategies to our problems. A complete and detailed account can be found in Bever (1974).

that

are

158

F. Antinucci,


potential interaction of this strategy (henceforth, B(asic) S(trategy)) with the structure of sentences conta’ining a RC. Taking into account for the moment only RCs depending on subject and object NPs, in an SVO language we have the following four cases (41) (42) (43) (44)

Ns[Ns

Ns[VNol

VI VNo VNo

N, V No INS VI Ns VNo[VNol

With respect to our first problem, the correct segmentation of the sequence forming the RC from the main clause, sentences of type (41) create no difficulties. BS is blocked because the first sequence met is NNV. Type (42) instead creates a problem. The first sequence is NVN, and therefore potentially open to the operation of BS, thus producing an incorrect segmentation. Type (43) again poses no problem. The first NVN sequence is correctly identified as the main clause by BS. In type (44) the head noun followed by the RC constitutes a NVN sequence, but since it comes after the initial NVN and has been correctly segmented as the main clause, it does not create problems. Therefore in an SVO language only one type of sentence out of four has a potentially dangerous structure with respect to BS. The situation in an SOV language is quite different. Here the four corresponding cases are (45) (46) (47) (48)

[Ns VlNs No [No VINsNo Ns [Ns VlNo Ns [No VlNo

V V V V

First of all notice that all the four sentence types contain an initial sequence that is a potential candidate for a BS segmenting the first N. . .(N). . .V sequence as the main clause. In the absence of any other clue the strategy would be to assign a main clause status to the RC of (46) and (45), and, even worse, segment together the first NNV sequence in (47) and (48) that don’t even correspond to a clause. Therefore in an SOV language, four out of four sentences containing a RC have a structure whose initial sequence matches that of a main clause without being one. However, we have to consider that OV languages, as observed in section 3, tend to have a system of case-marking on the noun. Since nouns marked by case will appear also in a main clause, one can reasonably hypothesize that a BS in these languages will be sensitive also to this feature. Let’s consider some concrete cases where this happens. In Japanese sentences corresponding to types (45-48) will appear as


(49)

kande-iru kodomo-ga neko-o nadete-iru inu-ga child-nom cat-act patting-is dog-nom biting-is “The child whom the dog is biting is patting the cat”

(50)

nadete-iru tabete-iru kodomo-ga inu-o ringo-0 child-nom dog-act patting-is apple-ace eating-is “The child who is eating the apple is patting the dog”

(51)

tabete-iru ringo-o motte-iru kodomo-ga inu-ga apple-act holding-is child-nom dog-nom eating-is “The child is holding the apple that the dog is eating”

(52)

tabete-iru kodomo-o kande-iru ringo-0 inu-ga child-act biting-is dog-nom apple-act eating-is “The dog is biting the child who is eating the apple”

159

If we take into account case-marking, it seems that sentence types (46) and (47) are no longer a problem for BS. (46) begins with a noun marked as object (by the particle -0) and in (47) the second noun is marked by the subject particle -ga and not by the object particle. Types (45) and (48) instead will still present an initial sequence subject to the operation of BS.13 Thus, even taking into account case marking in Japanese, the possibilities that BS will incorrectly segment sentences containing a RC are still two out of four. Furthermore, it may be the case that this gain is only apparent. In fact, an initial sequence NG V, like the one in type (46), can in Japanese form a main clause by itself. This happens because Japanese can have zerosubject pronominalization (and also zero-object pronominalization). Kuno (1973) remarks that a sentence like (53)

Mary-ga

kita came

toki ai ni when see to

kita came

“can mean either ‘When Mary came, (I) came to see (her)’ or ‘When Mary also a sequence like N-ga came, (she) came to see (me)’ “. Furthermore, N-ga V, like the one found at the beginning of type (47), can form a main clause. There is in fact, in Japanese, a class of verbs that require exactly an N-ga Nga V construction (see Kuno, 1973; p. 81).14 13Notice that the interpretation of the initial sequence of (49) as a main clause will not be blocked by the fact that the verb is transitive and is not preceded by an object NP. Japanese, in fact, allows zeroobject pronominalization; see example (53). ’ The fact that in most sentences of this type the first NP is marked by +(I rather than by -ga, as in Boku-wa Mary-ga kowai am-afraid-of I is irrelevant, since the first NP of a sentence like (51) is also likely to appear with -wa. (For the alternation between -wa and -~a, see Kuno, 1973; pp. 37-78).

160

F. Antinucci,

(54) (a)

(b)

A. Durantiand

I,. Geberr

dare-ga eiga-ga suki who movie fond-of “Who likes movies?” Watakusi-ga I

eiga-ga movie

desu is?

suki fond-of

ka?

desu am

“I like movies” If BS in Japanese is flexible enough to then sentence types (46) and (47) With some minor differences, the Turkish there is instead a different appear as

be sensitive to sentences like (53354), will again cause perceptual problems. same situation holds in Korean. In system. Sentences of type (45-48)

(55)

kopeg-in isirdigi cocuk kedi-y-i oksuyor dog-gen bite child cat-act pats “The child whom the dog bites is patting the cat”

(56)

kedi-y-i isirmis kopek elma-y-i yiyor cat-act bit apple-act eats dog “The dog who bit the cat is eating the apple”

(57)

cocuk kopeg-in yedigi elma-y-i tutuyor child dog-gen apple-act holds eat “The child is holding the apple that the dog eats”

(58)

kedi elma-y-i yiyen c;ocu&u isiriyor cat apple-act eat child-ace bites “The cat is biting the child who is eating the apple”

When the relativized noun is the object of the RC verb, as in (55) and (57), the RC subject appears with a genitive case marker. Therefore the initial sequences of (55) and (57) are not subject to BS any more. On the other hand, the initial sequence of (56) is marked by the object inflection -i, which would again block the operation of BS. The initial sequence of (58) instead remains subject to the operation of BS, causing a wrong segmentation. Thus, it seems that case marking in Turkish reduces the possibilities of wrong segmentation to one out of four. However, it should be noted that the object inflection in Turkish is applied only to definite NPs. Indefinite NPs show no inflection: in these cases the noun appears unmarked as in subject NPs. Consequently, sentences of type (46) can also appear as


(59)

161

yiyor kedi isirmis kopek elma apple eats cat bit dog “The dog who bit a cat is eating an apple”

Since in this case the object noun shows up in the same form of a subject noun, the initial sequence can again trigger BS. In conclusion, even if we take into account case-marking, the interaction of sentence structures containing a RC with BS is still more problematic in OV languages than in VO languages. Furthermore, to this purely quantitative difference in the number of potentially dangerous sequences in the two types of language, should be added the effect of a qualitative difference. In VO languages the only problematic structure is (42). The application of BS to (42) interprets the initial Ns [V No I sequence as a clause and assigns a main-clause status to it. However, since in (42) the subject head noun is also semantically the subject of the RC, the interpretation derived by BS will be wrong only in assigning a main-clause status to this sequence. On the contrary, the application of BS to sentences like (48), which, as we saw, is always a problematic case in OV languages (even on the assumption of a perfectly consistent system of case-marking), causes much wider negative effects. First of all, the correct interpretation of a RC of the kind that appears in (48) is in itself a perceptual task more difficult than that of (42). This is due to the so-called “double-function” of the head noun in (48). In this case, the head noun has the function of object in the main clause and subject in the RC. This factor has been demonstrated to increase perceptual complexity (see Bever, 1974; p. 1212). A recent experiment by Cook (1975) designed to test the relative difficulty in processing different types of RC, shows that subjects’ performance is significantly lower when the head noun has a double function. 76% of all the errors were made in processing RCs whose head noun had a double function. ls Furthermore, the application of BS to (48) interprets the initial Ns[No VI as a clause. Now, contrary to what happens in (42), the first noun in this sequence, which is interpreted as the subject of the embedded verb, has no relation at all with that verb. While the semantic interpretation of the clause resulting in (42) corresponds at least to that of the RC, the one in (48) doesn’t correspond to any clause of (48). This means that the analysis of (48) derived by the initial application of BS has to be entirely discarded and redone. Up to this point we have not considered the fact that BS may be blocked by the presence of some kind of marker indicating the subordinate status of the RC. “These

results

are based on our analysis

of Cook’s data.

162

F. Antinucci,

A. Durantiand

L. Gebert

A survey of VO languages (see the relevant data in Schwartz, 1971) shows that they typically have some kind of marker situated between the head noun and the beginning of the RC. This marker is either an invariable particle (Ewe, Hebrew, Italian) or a pronoun (Latin, Greek, Russian). The effect of this marker on perceptual strategies is obvious: it blocks the operation of BS immediately after the head noun and tells the hearer that what comes next is a new, subordinate, clause. Therefore, if we assume that in sentence types (41-44) above such a marker is present, there will be no more danger of incorrect segmentation by BS. In particular, in type (42), whose initial Ns[V No ] sequence is the only potential problematic case for an SVO language, BS will be correctly blocked right after the first N. It is worth noting that English, which is fairly atypical in that it sometimes allows the deletion of the initial marker, never does so in sentences like (42) and (44), i.e., in those sentences where such deletion would leave a NVN sequence. Furthermore, Middle English allowed the deletion of the marker even in (44) but not in (42), which, as we said, is the real potential problem for BS (on this point, see Bever and Langendoen, 197 1). On the other hand, OV languages do not have any marker at the beginning of the RC. They tend to mark the subordinate status of the RC at the very end of it, i.e., after the verb. In Turkish, the RC verb takes one of a number of suffixes turning it into a participial form. Furthermore, when the relativized noun is the object, a possessive pronoun agreeing with the subject is added to the participial form (60)

adam-m bekle-dig-i misafir man-gen await-prt-his guest “The guest whom the man is awaiting”

(61)

yaz-d&lm mektup write-prt-my letter “The letter which I wrote”

However, when the relativized noun is the RC subject, some of the participial forms (the past -miq and the future -ecek) are identical to finite verb forms (62)

haber gel-ecek news come-prt “The news which will come”

(63)

haber gel-ecek “The news will come”


(64)

hazirlan-mig plan prepared-prt plan “The plan which has been prepared”

(65)

plan hazirlan-mis “I gather that the plan has been prepared”16

163

In Japanese, instead, the same verb form appears in relative and main clauses, as (49) and (52) show. There is, therefore, no marker of subordination in RCs. Only the verb to be in the present tense has a special form for RCs (66)

kinben na hito diligent is person “A person who is diligent”

Like Turkish, pial form (67)

Korean

has a set of suffixes

turning

the verb into a partici-

chayk-lil ssi-ess-nin salam book-ace write-prt man “The man who wrote the book”

The suffix-ntr? marks the verb as subordinate. Lahu shows an invariable particle ve, which is identical marker, at the end of the RC (see Matisoff, 1973)

to the genitive

(68)

ya-mi=ma 1; qha?-‘se=ma yo va?=oqG thh? cG ta ve head man’s wife pig’s_head act boiled gen woman “The woman who boiled the pig’s head is the headman’s wife

(69)

va7=oca qha?-&=ma c5 ta ve mP ja headman’s wife boiled gen pig’s_head yummy “The pig’s head that the headman’s wife boiled is yummy”

From these facts it could be argued that most of the problems deriving from the interaction of BS with the structure of sentences containing a RC in OV languages are in fact solved. There are still some problematic cases, like those of Japanese, where there is no special marking of the RC verb, and Turkish, where some verb suffixes do not univocally mark the subordinate status of the verb. But at least in many cases we do have a final marker univocally identifying the preceding clause as a subordinate clause.

16As a finite verb-form,

the past

-mi$is used inferentially.

164

F. Antinucci,

A. Durantiand

L. Geberr

Therefore, it could be argued that the status of RC with respect to BS in OV languages is not very much worse than in VO languages. However, it should be remembered that perceptual strategies process speech in real time: their operation takes place simultaneously with the temporal development of the sound sequence. In other words, they tend to segment and interpret the language input as soon as possible. From this point of view it is one thing to have a signal at the beginning of the sequence to be processed and quite another to have it at the end. What we are claiming here is that from the perceptual point of view the relative efficiency of a marker signalling the subordinate status of a sequence is much higher if this marker precedes rather than follows the sequence. Consider what happens in a VO language like English when the RC marker of a sequence like (70) . . . the pen that fell on the table . . . is perceived. On the one hand, it immediately blocks the perceptual routine that is analyzing the clause where the preceding head noun occurs; on the other hand it makes the perceiver process a new clause and simultaneously assigns a subordinate status to it. Consider now a case like (48) in an OV language: (71)

kedi elma-y-i yi-y-en cocug-u isiriyor cat child-act bite apple-act eat-prt “The cat is biting the child who is eating the apple”

When the perceiver arrives at the verb yi- he has already received a sequence formed by a noun in the (unmarked) nominative case and a noun with the accusative inflection. This creates the expectation that the coming verb will fill the missing slot and complete a main clause relating the two NPs. On reaching the verb this analysis goes into effect and is disconfirmed only by the final -en. The perceiver at this point has to look ahead to the coming NP, process it, and then reanalyze the whole sentence. Obviously, this analysis puts a much heavier load on the perceptual mechanism than a sequence like (70) ” . Cook’s experiment on RC, quoted above, offers direct experimental evidence of the dramatic effect of presence vs. absence of an initial RC marker. In this experiment, there were three sentences of type (41), i.e., all sharing the form Ns[NsVl V No. In one of them there was no

“The analysis might also be disconfirmed by the semantic nature of represent a plausible semantic relation between the first two NPs. But nature of the problem, since the expectation derived from the initial possible main-clause schema. This must be erased and the analysis of

the verb, when this does not this fact does not change the sequence has already built a the whole sequence redone.

The

change

from

marker between the head noun and the RC; in the second and in the third a what: (A) (B) (C)

SOV to SVO

165

there was a that,

Ns Ns V V No Ns that Ns V V No Ns what Ns V V No

The comprehension task involved was the same for the three sentences, and furthermore, sentences (A) and (B) were built with exactly the same lexical material. The percentages of incorrect performances for each sentence were the following: I8 (A) (B) (C)

46% 24% 30%

Performance on (A), where there is no marker, is nearly twice as bad as performance on (B), where that is present. Notice that even performance on (C) is better than performance on (A). Although (C) is an ungrammatical sentence, it has some kind of signal at the beginning of the RC. Since (A) and (B) are identical and they both contain an initial NNV sequence which is not subject to incorrect analysis by BS, these results are extremely important because they isolate and show the importance of an at the beginning of the RC. This finding is explicit “signal of interruption” confirmed by the results on the ungrammatical sentence containing what. l9 In conclusion, we have examined the first problem posed by the processing of RC to the perceptual mechanism: the segmentation of the appropriate sequence corresponding to the RC and the recognition of its subordinate status. We have argued that, for a variety of reasons, the types of structures generated by the universal principle underlying grammar for sentences containing a RC (i.e., the position of RC, the order of consitituents in clauses, the type and position of subordination markers)zo make this task much more difficult in OV than in VO languages. In view of the quantity of linguistic discussions on this topic, the nature of this difficulty is to be stressed. We are not claiming that the relative clause is difficult because it causes actual ambiguities and misunderstandings as a “This is again our analysis of Cook’s data. “Results do not change if we compare performances on all the types used in the experiment: errors when no marker is present: 44% errors when what is present: 32% errors when rhot is present: 23%, “The “structural” status of this last feature will be discussed below.

of sentences

containing

a RC

166

F. Antinucci,


result of the structure of OV languages. Although this may sometimes happen (and, apparently, does happen, as a Japanese speaker has informed us), real language use is redundant enough to minimize such inconveniences. What we are claiming is that processing a RC in an OV language constantly places a burden on the perceptual analyzer, which is much heavier than in a VO language. In other words, the trouble is not actual ambiguity but systematic processing inefficiency. Let us now examine the second problem posed by RCs: the recovery of the missing NP. In VO languages we find typically one of the following two systems: (a) the RC is marked at the beginning by an invariable particle; the missing NP is either deleted or represented by a pronoun. Usually, it is deleted in the direct cases and pronominalized in the oblique cases. The reason for this distribution is clear: the NP is retained in a pronominal form in those instances where its deletion would cause the loss of its function within the RC.21 (b) There is a special pronoun (a “relative” pronoun), placed at the beginning of the RC, which can also carry the grammatical function of the missing NP. Both these systems present no problem for perceptual analysis. In (a) subjects process the relative clause and when they reach a pronoun or a gap, recover immediately both the content of the missing NP (since it is identical to the head noun which has already been processed), and its function. In (b) the marker that also signals the beginning of the RC allows the hearer to recover content and function of the missing NP. OV languages, on the other hand, always tend to delete the coreferential NP in the RC. In fact, most OV languages do not tolerate pronominalization of the missing NP (with the exception of a very few cases). In Lahu “no relative clause may contain an NP which is coreferential with the head noun” (Matisoff, 1973; p. 473). The same holds for Turkish and Japanese. Kuno (1973; p. 237) describes the following sentence, where the coreferential NP is pronominalized, as awkward, and our informant was very reluctant to accept it: (72)

watakusi-ga so-no namae-0 wasurete-simatta I-nom that-of name-act have-forgotten “a guest whose name I have forgotten”

okyaku-san guest

This restriction is likely to create problems in the recovery of the missing NP. When the NP is in an oblique case, its deletion causes also the loss of

“If the language in every case.

allows

function

markers

to be left stranded

(as in English),

then deletion

may occur


its accompanying functional marker, which in turn makes the recovery the NP’s grammatical function more difficult. Thus in Turkish (73)

gel-dig-im come-prt-my

167

of

tren train

can mean both “the train to which I came” and “the train oy1 which I came”. The intended interpretation must be derived from the context. In Japanese, a sentence like (Kuno, 1973, p. 244) (74)

tegami-ga letters-nom

takusan many

kita came

tomodati-ga friend-nom

oozei many

ita were

can mean both “there were many friends from whom many letters came” and “there were many friends to whom many letters came”. Again, the intended meaning must be gathered from the context. Thus, a discourse like the following (75)

Tokyo

ya Osaka and

Daihyoo-ga representative Fukuoka

kara from

oozei-no many

hitori mo one-person

daihyoo-ga representative

konakatta came-not

tosi-wa city

kita. came Hiroshima

to and

dake da. only is

is interpreted as “Many representatives came from Tokyo and Osaka. Cities Jiom which no representatives came are only Hiroshima and Fukuoka”. The first sentence makes clear that the deleted NP in the RC of the second sentence must be interpreted as from which. Therefore, owing to the obligatory deletion of the missing NP, its grammatical function in the RC cannot be marked. This conclusion raises an interesting problem. If, as we saw, deletion of the coreferential NP makes the recovery of its function more difficult, then why do OV languages not pronominalize the missing NP (at least in the oblique cases) as VO languages do? The answer is again to be looked for in the conflict between structural and processing factors. In discussing constraints on the behavior of pronouns, Bever writes: “Such a complex system appears at first to be an example of a “pure” linguistic law. However, there is an intuitively clear general principle of all experience which could underlie such complex linguistic constraints. First, for one object to ‘stand for’ another, like a pronoun for a noun, a connection must already be established between them. For example, a picture of a leaf cannot be used to represent a tree unless the

168 F. Antinucci, A. Durantiand L. Gebert

viewer already knows the connection. Analogously, in He spoke to George, he cannot refer to George sin’ce the listener does not yet know who Ize is. The constraint which allows a superordinate clause noun to govern the pronominalization of a subordinate clause noun may also be interpreted as a linguistic reflection of an obvious regularity of experience: presentation of a whole includes a presentation of its subordinate part but not viceversa. For example, a picture of a tree also presents a leaf since it includes a leaf, but a picture of a leaf dbes not present a tree (without prior knowledge of the connection, as above). Analogously, a pronoun can appear, even preceding its governing noun, if it is explicitly marked as in the subordinate part of the sentence. Since every sentence has at least one main clause, the listener can predict that a pronoun in a subordinate clause will be governed by a mainclause noun. But a pronoun in an initial main clause does not necessarily have a following subordinate-clause governing noun since there may be no subordinate clause at all [ . ..I. To put it another way, the general perceptual principle is: A symbol Sl can stand for S2 if (a) the prior connection is known or (b) there is an indication that a connection is about to be established.” (Bever, 1974; p. 1195).** Now, since, for structural reasons, the relative clause has to precede the head noun in OV languages, we always have to have backward prononzinulizution in order to have a pronoun in the RC. But given the way RCs are marked in these languages, the pronoun occurs in the sequence at a point where the hearer does not yet know that the clause under processing is a subordinate one. If in processing the sequence the hearer were to hit on a pronoun, he would immediately search for an antecedent in the preceding context, and in this way miss the correct interpretation of the sentence. On the other hand, if he does not find any pronominal element to interpret. he may proceed until the end of the clause and find the head noun. Therefore, in these structures pronominalization would be worse than deletion.

**Notice that the whole range of pronominalization phenomena cannot be accounted for on perceptual grouns only (as Bever seems to imply). There are other factors involved, the most important of \vhich seem< to bc the relative degree of “topicality” of the NPs involved (see Kuno, 1972). The interplay of these different factors in dctcrminin p the behavior of pronominalization is investigated in Antinucci (1977b). However. perceptual factors will in any case prevent inter-clause backLvard pronominalization (which is the relevant point to our argument), if the clause containing the pronoun can be interpreted as a complete main clause.


4. Diachronic

169

Change

The general conclusion that we can derive from the preceding discussion is that a variety of factors concur in making the perceptual processing of RC in OV languages problematic. Therefore, languages of the OV type will be under pressure from the perceptual mechanism to reorganize the structure of RC in such a way as to eliminate these inconveniences. One could easily imagine a solution: the first group of problems is solved by having markers of subordination at the beginning of the RC, rather than at the end of it. The difficulties connected to the recovery of the missing NP would, in this case, also be solved, since pronominalization could work backwards. However, this solution is unavailable, as is shown by the fact that there are extremely few (if any) OV languages that adopt an initial RC marker. The reason is that this construction is impossible from the structural point of view. As we said in section 2 above, the fundamental principle underlying the mapping of meaning into a linear sequence (be it Vennemann’s “natural serialization” or what we called “principle of expansion”) univocally determines that the RC “follows” its head noun, since the former is a logical expansion of the latter. If this order is realized in a language that builds its linear sequence from left to right (as VO languages do) the RC will be placed on the right of the head noun, while if the linear sequence is built from right to left (as in OV languages) the RC will be placed on the left of the head noun. Therefore, in both cases the starting point of the RC is, from the structural point of view, enclosed between the head noun and the sequence corresponding to the RC; i.e., on the immediate right of the head noun when the RC is on the right of it, and on the immediate left of the head noun when the RC is on the left of it. Now, in VO languages the “temporal” beginning of the RC in the sound sequence coincides with its “structural” beginning, but in OV languages the temporal beginning of the RC has no structural status. This explains why in OV languages, if some marker is found, it comes at the temporal ending (i.e., structural beginning) of the RC. The same conclusion can be reached for those languages that treat their RC essentially as a genitive construction. For example, in Lahu the RC marker ve is the genitive marker found in nominal construction (Matisoff, 1973; p. 141) (76)

Ch-15 ve a-th? Jalaw of knife “Jalaw’s knife”

The order determined

by the structural

principle

for OV languages is Speci-

170

F. Antinucci, A. Duranti and L. Gebert

fJ)irlg noun ~ Genitive ~ Head nom and Heud noun ~- Gerlitive --- Specif19ing noun for VO languages (see (36&37) above). Since there can be no “Genitive ~~Specifying noun - Head noun order, it follows that the genitive marking the subordinate status of the RC can never be placed at the temporal beginning of the RC when this precedes the head noun. Thus, we are left with only one way to solve the problems posed by the perceptual analysis of RCs in OV languages. This is to remove the RC from its position on the left of the head noun and place it to the right of it. If this happens, then the structural beginning of the RC will coincide with its temporal beginning.23 But such a change implies a change in the direction of expansion of the linear sequence. This is constructed from right to left, and instead would have to be constructed from left to right in order to generate RCs following the head noun. But once this change occurs, the whole linear organization of the language will be inconsistent with the new building direction, and therefore it will tend to change in order to reestablish consistency. This change takes place along the time dimension and will determine the diachronic evolution of an original OV language into a VO language. To sum up, our hypothesis is that one of the main causes of the shift from an OV to a VO organization is the pressure that the processing mechanism exerts on the structure constituted by prenominal RC towards a reorganization leading to a perceptually more favorable postnominal position. But since RC position is not an independent variable in the organization of language, a change to postnominal position triggers a structural pressure to reorganize the whole linear sequence, which will push the language towards a VO construction. From this hypothesis we can derive some predictions which are subject to empirical verification. If the hypothesis is correct, then we would expect that in languages showing a drift from OV to VO, RC position (being one of the causes of such a change) will be diachronically one of the first features to change. Therefore, the change from prenominal to postnominal position of RC would have to come before most other changes connected to the general change in type; for example, before the change of object NP, indirect object NP and other complement NPs from preverbal to postverbal position. On the other hand, the processing difficulties connected with prenominal position of RC will delu.!, the unavoidable change from postnominal to pre23Notice that this solution is not “impossible” in the same sense in which we said it is impossible to have initial marking of preposed RC. The latter would radically contradict the universal principle of logical expansion. As we said before, a sequence like *f&n Specifjring noun Head noun cannot be gcneratcd in any type of language. On the other hand, a sequence like Ilead rzoun Gen Specifving now1 (which would result from the shift of modifiers to the right of the Head noun) can be generated and is inconsistent only with the directiorz of expansion of an OV language.

The change from SO V to SVO

171

nominal position in a language that is undergoing a change from the VO to the OV type.24 Therefore, in the opposite drift from VO to OV we would expect RC position to be one of the last features to change; for example, we would expect it to occur after the shift of NPs from postverbal to preverbal position. This claim can be verified in both a direct and an indirect way. The direct way is to show that for those languages whose history is recorded the sequence of changes in features is actually the one predicted. We will offer below some evidence that this in fact is the case. But since there are not many languages whose history is recorded over the wide time span needed to verify our prediction, and furthermore the data available from remote periods of the history of a language are often very limited in size and/or type and difficult to interpret, we will also rely on indirect evidence. In fact, a very strong synchronic claim can be derived from our hypothesis. Logically, there are four possible combinations of language type and RC positions: (A) (B) (C) (D)

OV OV VO VO

-

Prenominal RC Postnominal RC Prenominal RC Postnominal RC

As we said in section 2, (A) and (D) are the combinations determined by the structural principle. Now, if our hypothesis is correct, it follows that (B) will also be found among the languages of the world, but (C) won’t. In fact, according to our hypothesis, if a language is undergoing a change from OV to VO, RC will move to postnominal position before the change of NPs to postverbal position has been established. Therefore, such a language will show, synchronically, a predominant OV order of sentence constituents and postnominal RC (B). On the other hand, if a language is undergoing a change from VO to OV, the shift of NPs to preverbal position will occur before RC moves to prenominal position. Thus, such a language will also show synchronically a predominant OV order of sentence constituents and postnominal RC. A first check on this prediction can be done by looking at Greenberg’s ( 1963) 30-language sample. In this sample all the languages classified as VO 240f course, such a change is not motivated by the RC structure. Notice that our hypothesis does not imply that VO languages are in general “simpler” than OV languages and consequently that there is a constant general drift from OV to VO. There are several respects in which VO languages are more problematic than OV languages, specifically in the interaction between the construction of the linear sequence determined by the principle of logical expansion and the distribution of Given and New information (see n.7). In Antinucci (1977a), it is argued that this conflicting interaction is one of the main factors motivating the shift from a VO to an OV typology.

172

F. Antinucci,

A. Duranti and I,. Gebert

(i.e., Greenberg’s SVO and VSb) have postnominal RC. On the other hand, of all the languages classified as SOV, seven have prenominal RC, but two have postnominal RC. Two languages are said to have both prenominal and postnominal RC. One of these, Nubian, is classified as SOV and therefore is still consistent with our hypothesis. The second, Finnish, is instead classified as SVO. However, a closer examination of RC formation in Finnish (see Karlsson, 1972) shows that the predominant pattern is undoubtedly the postnominal one. Prenominal formation is limited to a few instances for which there are in any case corresponding postnominal constructions. Thus, our prediction seems to be confirmed. A second check was done by extending the language sample. However, instead of making a random selection, we deliberately chose languages whose combination of typological features offered maximum probabilities of falsifying our hypothesis. Greenberg’s Appendix II offers an extended list of languages classified into 24 groups according to all possible combinations of four features: basic order of S, V, and 0; presence of prepositions vs. postpositions; relative position of head noun and genitive; relative position of head noun and adjective (3 X 2 X 2 X 2 = 24). Since the crucial prediction of our hypothesis is the absence of VO languages with prenominal RC, we looked at languages showing an SVO order accompanied by some features inconsistent with the VO type; i.e., at languages classified in Greenberg’s groups lo-- 16. Since these languages have one or more of the three remaining features which are consistent with an OV construction, the probabilities of also finding prenominal RCs in them are in principle higher than in consistent VO languages. The results can be summarized as follows. Among the languages in group 10, where the inconsistent feature is prenominal adjective position, German, Dutch, and all the Slavic languages have postnominal RC. The three languages classified in group 11, Norwegian, Swedish and Danish, showing both prenominal adjective and prenominal genitive, all have postnominal RC. Group 12 includes only Arapesh for which we did not find data concerning RC position. Group 13 is empty. Group 14, where inconsistent features consist in the presence of postpositions and prenominal adjective positions, includes Rutulian and other Daghestan languages. Along these languages, Rutulian, Batsbi, Dido, Xwarii, all have postnominal RC (Vinogradov, 1966). On the other hand, Avar and Tabasaran have prenominal RC but their order is rigidly SOV. Among the languages of group 15, showing postposition, prenominal adjective and prenominal genitive, Finnish has already been discussed, Estonian has postnominal RC, Chinese instead has a prenominal RC construction. Finally, among languages of group 16 (postpositions and prenominal genitive), Twi, Ewe, Songhai and Guarani all have postnominal RC.


173

In conclusion, the entire group of languages checked confirmed our prediction with the only exception of Chinese.25 On the other hand, the list of OV languages having postnominal RC can be easily extended. In addition to those in Greenberg’s sample, they include, for example, Persian, Galla, Sumerian and Yaqui. The limited amount of direct diachronic evidence available seems also to confirm our hypothesis. Lehmann (1974) has argued that Proto-IndoEuropean was OV in structure but was developing into a VO language at the time of the early dialects. He has further argued convincingly that PIE had prenominal RC without a marker, in accordance with the general OV pattern. The point relevant to our discussion is whether the establishment of postnominal RC preceded that of postverbal constituent order. Even a cursory examination of the relevant data suggests that this is, in fact, what happened. Ancient Greek has a well established postnominal RC construction, while still showing a predominant SOV order (see Dover, 1960). Even the most archaic phases of Latin, where SOV order is by far the most predominant (see Marouzeau, 1922), have a postnominal RC introduced by a relative pronoun. Hittite, which in the earliest records shows even some unmarked prenominal RCs, develops a postnominal RC introduced by a form of the pronoun kuis while its pattern of sentence construction is still sov. Some direct evidence can also be gathered in favor of our second prediction; namely, that in a language changing from VO to OV the RC will move to prenominal position after the shift of NPs to preverbal position. The case in question seems to be that of Hindi and some related Indo-Aryan languages. Today Hindi is a quite rigid SOV language. However, Bloch (1934), in his history of Indo-Aryan, dealing with the verb-final pattern, says that in the modem Indo-Aryan languages we find “la fixation d’un ordre qui n’Ctait d’abord qu’habituel” (p. 306). This is shown by the fact that in Old IndoAryan (Asoka inscriptions) “les complCments indiquant la destination (infinitif, substantif au datif) se rejettent volontiers apr&s le verbe”, while these postverbal constructions are impossible today. Thus, it can be argued that the Modern Indo-Aryan languages have been drifting toward a more rigid SOV construction. This is confirmed by the history of another relevant 25Contrary to the Finnish case, where prenominal RC construction appears to be a more or less “frozen” relic of an older stage, the Chinese case looks like a real exception. At present, we are unable to offer any explanation for it; notice, however, that the question of main clause word-order in Chinese, both diachronically and synchronically, is a fairly complex and controversal one (see Li and Thompson, 1974; Tai, 1973; 1976). Since Chinese is one of the languages whose history is well attested to, a careful diachronic study of the relationship between NP and clause constituent-order should be conducted, before we can definitely assert its exceptionality.

174

F. Antinucci,


feature. In Old Indo-Aryan there were prepositions. According to Chatterji (1926) they fell into gradual disuse and lost their separate status from Late Indo-Aryan. In Middle Indo-Aryan their number is very restricted. On the other hand, “the Indo-Aryan speech began to employ the accusative, dative, ablative or locative form of some suitable noun (with the sense of location, vicinity, direction, connexion, purpose or power) along with the principal noun which retained its original inflexion. Classical Sanskrit, following the Pakrit vernaculars, took up this device. This sort of auxiliary and postpositional use was later extended to some verbal formations” (p. 766). In other words, the Indo-Aryan languages lost prepositions and formed new postpositions deriving them from original nouns and verbs. This is another typical sign of evolution toward an OV pattern. Now, if we look at the RC, we find that these languages still today allow a postnominal RC introduced by a relative pronoun. Again, our prediction seems to be confirmed. The Indo-Aryan languages have been undergoing a change toward an OV construction, as testified by the increasingly rigid verb-final order and by the loss of prepositions and formation of a system of postpositions, but they have not yet eliminated postnominal RCs. A final case in favour of our hypothesis comes from Givon’s ( 1975) reconstruction of the sequence of changes that Amharic has been undergoing in its shift form a VSO to an SOV organization. Here is his argument: “In this language the definite article is a noun suffix, but if a modifying adjective precedes the noun, the definite article is suffixed to that adjective: mikina-w car-the “the car” til i-g-u makina big-the car “the big car” In Ge’ez, the closest attestation to the VSO pre-Amharic, one finds the f‘oflowittg the noun. One modifying adjective ~~ with the definite suffix may thus conclude, as is indeed attested in Ge’ez, that the definite article -II was an NP-final morpheme. In time it became bound, and when the older N-ADJ order changed to ADJ-N. the definite article moved with the adjective to which it was bound. Now hcrc is the rub ~ in Amharic the definite article appears as a llerb srtj;fix when a noun is modified by a (preceding) relative clause:


175

mtikina yP-gazzti-w SdW that-bought-the man car “the man who bought a car” Since we know that the definitizer -u/-w was an NP-suffix in pre-Amharic, the only way of explaining its verb-suffix position in relative clauses is by assuming that there existed a stage in Amharic in which N-MOD was still the prevailing order within the noun phrase, but the VP syntax has already changed to SOV, so that in relative clauses the verb was positioned clausefinal, and therefore also NP-final” (p. 93). Thus, also in Amharic we find the sequence of changes predicted by our hypothesis: the shift of NPs to preverbal position preceded the shift of RC to prenominal position.

References Antinucci, F. (1977a). Fondamenti di una teoria tipological de1 linguaggio. Bologna, 11 Mulino. Antinucci, F. (1977b). L’intcrazione dei sistemi nella competenza linguistica: la pronominalizzazione in italiano. Riv. Grammat. Generat., 2, I. 3-- 42. Bartsch, R. (1972). Adverhialsemantik: Die Konstitution logisch-semantischer Reprtisentationen von Adverbialkonstruktionen. Frankfurt am Main, Athensurn Verlag. Bartsch, R. and Vennemann, T. (1972). Semantic Structures: A Study in the Relation between Semantics and Syntax. Frankfurt am Main, Athensurn Verlag. Bever, T. (1974). The interaction of perception and linguistic structures: a preliminary investigation of neo-functionalism. In T. Sebeek (Ed.), Current Trends in Linguistics, The Hague, Mouton, vol. 11,~~. 1159%1233. Bever, T. and Langendoen, T. (1971). A dynamic model of the evolution of language. Ling. Inq., 2, 433.-463. Bloch, J. (1934). L’Indo-aryen du veda aux temps modernes. Paris, Libraire d’Amtrique et d’orient. Chatterji, S. (1926). The Origin and Development of the Bengali Language. Calcutta, Calcutta University Press. Cook, V. (1975). Strategies in the comprehension of relative clauses. Lang. Sp., 18, 204-212. Dover, K. (1960). Greek Word Order. Cambridge, Cambridge University Press. Fodor, .I. (1971). Current approaches to syntax recognition. In D. Horton and J. Jenkins (Eds.), The Perception of Language.Merrill Publishing Company. Given, T. (1975). Serial verbs and syntactic change: Niger-Congo. In C. Li (Ed.), Word Order and Word Order Change, Austin, University of Texas Press. Greenberg, J. (1963). Some universals of grammar with particular reference to the order of meaningful elements. In J. Greenberg (Ed.), Universals of Language, Cambridge, Mass., M.I.T. Press. Karlsson, F. (1972). Relative Clauses in Finnish. In P. Perantcau, J. Levi and G. Phares (Eds.), 7?ze Chicago Which Hunt, Chicago, Chicago Linguistic Society. Keenan, E. (1972). Relative clause formation in Malagasy. In P. Peranteau et al. (Eds.), The Chicago Which Hunt, Chicago, Chicago Linguistic Society. Kuno, S. (1972) Functional Sentence Perspective. Ling. Inq., 3, 267-320. Kuno, S. (1973). The structure of the Japanese Language. Cambridge, Mass. M.I.T. Press. Kuno, S. (1974). The position of relative clauses and conjunctions. Ling. Inq., 5, 117-135. Lehmann, W. (1973) A structural principle of language and its implications. Lang., 49, 47-66. Lehmann, W. (1974). Proto-Indo-European Syntax, Austin, The University of Texas Press.

176

F. Antinucci,


Li, C. and Thomspon, S. (1974). An explanation of wordorder change SVO - SOV. Found. Lang., 12, 2OlI214. Marouzeau, J. (1922). L’ordre des mofs duns la phrase lutine. Paris, I. 1922: II, 1938; III, 1949. Matisoff, J. (1973). The Grammar ofLnhu. Berkeley, University of California Press. Schwartz, A. (1971). General aspects of relative clause formation. Working Papers on Language Universals, 6, 139-171. Tai, J. (1973). Chinese as an SOV language. In C. Corum et al. (Eds.), Papersfrom the Ninth Regional Meeting, Chicago, Chicago Linguistic Society. Tai. J. (1976). On the change from SVO to SOV in Chinese. In S. Steever et al. /Eds.l.I Diachronic Syntax, Chicago, Chicago Linguistic Society. Thompson, S. (1970). The deep structure of relative clauses. Working Papers in Linguisrics. 6, 42 68. Vennemann, T. (1973). Explanation in syntax. In J. Kimball (Ed.), Syntox and Semantics, New York, Seminar Press, Vol. 2, pp. l-50. Vinogradov, V. (1966). Yazyki narodov SSSR. Kuvkazskiye yazyki. Moskva, Izdatielstvo “Nauka”.

On propose ici une interpretation d’un changement diachronique dans le langage. Selon le point de vue adopt6 un des facteurs majeurs qui motive un changement de la syntaxe est du i une interaction conflictuelle entre des principes determinant l’organisation du langage. Plus specifiqucment on avance que la construction des propositions relatives rend conflictuels les principes d’origine structurelle et ceux d’origine perceptuelle dans les langues de type SOV. La man&e dont une proposition relative est structurde dans un langage SOV est un obstacle a son calcul perceptuel. Ce conflit serait un des facteurs majeurs i I’origine du changement diachronique d’un langage d’une typologie OV i un langage d’une typologie VO.

Discussions

Cognition, l(1979) 177 - 215 @ Elsevier Sequoia S.A., Lausanne - Printed in the Netherlands

Signing behavior in apes: A critical review*r** MARK S. SEIDENBERG Psycholinguistics Columbia

Program,

University

LAURA A. PETITTO Deafness New

Research and Training Center

York University

Introduction The recent attempts by several groups of researchers manual-visual (sign) languages to lower primates have interest. The possibility that apes might be able to through this modality was noted by the primatologist

to teach rudiments of generated uncommon learn to communicate R. M. Yerkes in 1925:

I am inclined to conclude from the various evidences that the great apes have plenty to talk about, but no gift for the use of sounds to represent individual... feelings or ideas. Perhaps they can be taught to use their fingers, somewhat as does the deaf and dumb [sic] person, and thus helped to acquire a simple, nonvocal “sign language”. (1925, pp. 179-180)

It is ten years since Beatrice and Allen Gardner first reported signing behavior by their chimpanzee Washoe. During this period, the claims on behalf of the language abilities of non-human primates have grown steadily. The Gardners initially reported establishing “two way communication” with Washoe via a “standardized system of gestures” (Gardner & Gardner, 1969). In a later paper, they claimed that Washoe’s knowledge of “sentence con*An early version of this paper appears in Papers from the 14th Regional Meeting, Chicago Linguistic Society, 1978. This paper subsumes that one. We are indebted to T. G. Bever and H. S. Terrace, and especially to U. Bellugi and E. Klima, whose pioneering work on ASL we draw upon continually. However, none of the above should be heid responsible for or identified with the contents of this paper. **This paper is concerned with those studies in which experimenters attempted to teach non-human primates to communicate in a manual-visual mode. Projects in which apes learned to arrange sequences of plastic chips (Premack, 1977) or press sequences on a computer keyboard (Rumbaugh, 1977) entail somewhat different claims and methodological issues (e.g., which essential features of language and communication are preserved in these paradigms?). For discussion of these issues, see Seidenberg and Petitto (in preparation). Requests for reprints should be addressed to: M. S. Seidenberg, Psychology Dept., Columbia University, New York, N.Y. 10027, U.S.A.

178

M. S. Seidenberg and I,. A. Petitto

stituents” and her ability to answer questions were superior to those of children at Roger Brown’s Stage III (Gardner & Gardner, 1975). They have also stated that, If the standards of experimentation in child psycholinguistics do not improve soon, we will find ourselves in the paradoxical situation of having solid experimental evidence for the syntactic abilities of chimpanzees and a complete lack of acceptable evidence for any syntactic ability in young children. (1974b, p. 735) It is clear from such statements and others that their focus has shifted from “Is it possible for non-human primates to learn aspects of the question, human languages?” to “How may we characterize the language abilities of apes and their limits?” Patterson (1978) has perhaps summarized the prevailing view of these projects in stating that the accomplishments of her signing gorilla Koko indicate that “language is no longer the exclusive domain of man”. In addition, the Gardners, Patterson, Fouts (1977) and others have described the signing of apes as American Sign Language or ASL, the language of the North American deaf. It is asserted, then, not merely that the apes learned to communicate with the experimenters via a system that is not utilized by wild apes, but also that this system shares significant features with a human language, ASL. These interpretations of the ape signing research are widely held, among both the general public and the scientific community. The projects of the Gardners, Fouts, Patterson and others appear to hold great attraction for the general public, as evidenced by the many reports on ape language on television and in the mass circulation magazines. These reports have faithfully conveyed the interpretations of the ape researchers summarized above. That these interpretations are shared by many psychologists as well is seen, for example, in the discussions of these projects in the introductory textbooks. The responses of linguists and psycholinguists to these reports show far less uniformity. The major reviews include those of Bronowski 8c Bellugi (1970), Brown (1970, 1973), Fodor, Bever & Garrett (1974), Watt (1974), Katz (1976) and Limber (1977). The attitudes in these papers range from open scepticism towards this work (e.g. Fodor et al.) to enthusiastic approval (Watt). Reviewing the work through 1973 with Washoe and Premack’s chimpanzee Sarah (Premack, 1977), Fodor et al., conclude that “The best one can say is that the present experiments provide no evidence that chimpanzees can learn a formal system in any important way similar to natural language.” (p. 450) Brown, writing somewhat earlier, concluded, in a widely quoted remark, that “... as matters now stand... the evidence that Washoe has Stage I language is about the same as it is for children.” (1973, p. 43) In his laudatory review, Watt states that

Signing behavior in apes: A critical review

179

To put it as baldly as possible, chimps can “talk”. They have now been shown to have an ability so irrefragably on a continuum with our own speech that the chimps’ ability cannot well be denied the adjective linguistic... (p. 71) The old claim that the other animals did not and could not have language has been overturned. No other terrestrial animals surely will be shown to have a capacity as developed as ours; but ours and theirs are comparable on a number of dimensions. (p. 7.5)

Bronowski & Bellugi’s early reply to this work granted that Washoe had demonstrated the ability to name objects, but noted that evidence at that time gave no indication that she had acquired the use of formal sentence structure or the ability to analyze and synthesize complex messages. Limber (1977), noting that “an organism’s use of names is surely not sufficient evidence that the organism is using human language”, also concluded that evidence supporting the hypothesis that lower primates are able to learn significant aspects of human languages does not exist. Katz (1976) has stated, Nothing in the chimp studies to date suggests that these animals can do anything more significant than a dog or cat does when it rings a bell to communicate its desire to go outside. A chimp does, of course, acquire a far more complex set of discriminations in learning to arrange chips as a signal for food than a dog or cat does in learning to ring a bell, but such a difference in degree does not amount to the difference in kind required to refute Descartes’ claim [that no animals other than humans utilize language]. (p. 33) Although the hypothesis that signing apes show linguistic abilities has been met with generally negative evaluations by linguists and psycholinguists, their criticism has been perhaps critically weakened by lack of access to the relevant data. Normally an empirical claim can be evaluated by attempting to replicate the experiment from which it followed. This is infeasible for most researchers when the subjects are apes, creating a situation in which only the persons making the strongest claims are in possession of the relevant data. In any case, it is clear that disagreements over the interpretation of ape signing behavior are substantial and continuing. Our purpose in writing this paper is a simple one. We believe that the conclusion that apes show rudimentary linguistic abilities is a mistaken on that has been drawn for two reasons. First, the procedures utilized in these projects have not been subject to the standards applied to other work in developmental psycholinguistics, and to psychological research in general. When this work is subjected to careful scrutiny, it is seen that both the data and methodology are so highly questionable as to vitiate the radical conclusions that are offered. Second, the existing literature does not take into account infor-

180 12%S. Seidenbcrg ard I,. A. Petitto

mation that has recently emerged from two sources. Substantive studies of ASL in deaf children and adults now exist (e.g., Friedman, 1977; Schlesinger & Namir, 1978; Siple, 1978; Klima & Bellugi, 1979; papers in the journal Sign Language Studies), permitting direct comparisons to the signing of apes. In addition, a replication of the Gardners’ work was attempted by an independent group of researchers (Terrace, Petitto, & Bever, 1976a and b; Terrace, in press; Terrace, Petitto, Sanders and Bever, in press. Information from these sources, combined with a careful reconsideration of the published data, yields the conclusion that the claims on behalf of ape language abilities are at best unsubstantiated, and quite probably false. In this paper, we will show that: 1. Data which are necessary in order to establish linguistic abilities have never been published;

the claim that apes show

2. The fragmentary data which do exist have been consistently interpreted, and are subject to non-linguistic interpretations;

over-

3. The apes’ signing compares unfavorably with the speech of hearing children and the signing of deaf children. In particular, this behavior does not resemble signing in ASL in any important way. Hence, we conclude that there is no evidence at this time for linguistic abilities in signing apes. These considerations do not place an upper limit on what apes are capable of learning; in fact, it may be the case that the training procedures and other aspects of the current projects underestimate their cognitive and communicative capacities. Rather, they show that the existing claims are far too strong and, in addition, that many known aspects of their signing behavior suggest that it is fundamentally non-linguistic. We will largely be concerned with the Gardners’ claims on behalf of Washoe, since they are at once the best-documented and most widely known. Our observations apply as well, however, to the very similar claims of Fouts (1972, 1977), Patterson (1978), and others. The Gardners themselves have acknowledged the importance of comparisons between Washoe’s signing and the language of deaf and hearing children, stating All of our procedures are governed by one principle: the observations obtained of the actual linguistic performance of an infant chimpanzee must be compared with the actual linguistic performance of human children... The deaf children of deaf parents comprise an essential control group for evaluating the progress of young chimpanzees that have acquired some facility in Ameslan. (1974b, p. 5)


1. Sampling:

What Do Signing Apes Actually

18 1

Say?

In presenting the accomplishments of their subjects to the scientific community, the ape sign language researchers faced an unusual problem, i.e., how to characterize complex behavior of a type never before reported. Although the Gardners place their work within the study of language acquisition, their task differed from that of the typical researcher in this area. Child language is something every adult is colloquially familiar with through direct experience; chimpanzee signing is not. In studies of the acquisition of spoken language by hearing children (e.g., Brown, 1973) and sign languages by deaf children (e.g., Klima & Bellugi, 1979; Hoffmeister, 1978l, Hoffmeister, Best & Moores; 1974*), researchers typically report corpora of child speech sampled at varying intervals. Such samples are small relative to the total corpus of a child-subject’s utterances over a given period, but are comparable in size to the samples used in other types of psychological research. Occasionally, much larger corpora are published (e.g., Bloom, 1973), so that substantial records of the language behavior of many children now exist, and detailed comparisons across subjects, ages, languages, cultures, and observers are possible (see, for example, Ferguson & Slobin, 1972). In describing Washoe’s behavior, the Gardners and Fouts did not conform to these established methods. Nor did Patterson in describing Koko’s. As a result, it is very difficult to know exactly what each ape has signed. The data that are presented in their papers are confined to (a) cumulative vocabulary counts and vocabulary lists; (b) anecdotes, frequently cited as evidence of the apes’ abilities to combine signs creatively into novel forms, e.g., Washoe signing water bird for duck, or Koko signing cookie rock for a stale sweet roll; (c) isolated summary statistics, such as Patterson’s observation that “In a l-hour dinnertime sample, Koko used a total of 2.5 1 signs” (1978, p. 78; these signs are not identified, however); and (d) the results of two tests of Washoe’s signing abilities, the double-blind vocabulary tests (Gardner & Gardner, 197 1, 1974a) and a test of her ability to answer questions (Gardner & Gardner, 1975). These sources fail to provide a full characterization of any ape’s signing behavior. The fragmentary information provided is ambiguous among several competing interpretations, including ones which do not require attributing linguistic abilities to apes.

‘Hoffmeister, R. The development of demonstrative pronouns, locatives, and personal pronouns in the acquisition of American Sign Language by deaf children of deaf parents. Unpublished Ph.D. thesis, Uqiversity of Minnesota, 1978. Hoffmeister, R., Best, B. and Moores, D. The acquisition of sign language in deaf children of deaf parents. Bureau of Education to the Handicapped Progress Report, 1974.

182


It is the absence of a substantial corpus of ape utterances which is the most serious omission. In contrast to the child language literature, there is no listing of all the ape utterances which occurred during a single period of time. By failing to provide a corpus, the Gardners and others obscure significant aspects of their subjects’ behaviors, and make it difficult to independently verify their claims. This problem is seen most clearly with respect to the anecdotes that are frequently cited (e.g., water bird). In the absence of a corpus, one cannot determine whether such sequences were synthesized through the application of linguistic rules, or merely the result of the ape acting as a random sign generator which happened to emit some interesting-looking strings. The water bird example loses much of its force if the ape also combined each of these signs with a large number of other signs (e.g., water shoe, water banana, cookie bird, etc.). This alternative is not implausible. As the Terrace et al., (in press) corpus indicates, their subject, a chimpanzee named Nim, combined signs into a very large number of permutations, most of which occurred with low frequency. Some of the resulting combinations are fortuitous, others are not. While even the most bizarre combinations could, of course, be interpreted metaphorically, it is simpler to assume that they occurred as the result of random pairings of signs or other non-linguistic combinatorial processes (see Terrace et al., in press). Determining whether or not signs were combined into meaningful strings requires extensive analyses of the structure and frequency of signs and combinations, and the contexts in which they occurred. Since neither the Gardners nor Patterson has provided such analyses, their claim that the cited examples show evidence for creative language abilities cannot be sustained. Patterson’s discussion of the cookie rock example is indicative of this samKoko has produced uninterpretable pling problem. She states, “Although strings (as do some children), most of her utterances are appropriate to the situation and some are strikingly apt”. (1978, p. 88) She then cites some “interpretable” examples, including, “cookie rock”; the “uninterpretable” strings are not described. It is the case that only “interpretable” strings are ever documented in the reports on ape signing. Only by presenting an unedited corpus of responses, however, could Patterson’s assertion be validated. Bloom (1974a) has described the fundamental weakness of anecdotal data in studies of child language: Anecdotal reports of isolated behaviors reflect what is important in a child’s behavior from an adult’s point of view. It may well be that the anecdote is important also for the child, but it may also be that it was observed in the first place because of its importance for the adult, and, in the larger scheme of things,


183

event. In analysis of behavior, one needs sufficient data to avoid emphasizing unique or only marginally important behaviors...

it is a relatively unimportant (P. 87)

We cannot help but note that writers who remark positively on the linguistic abilities of apes repeatedly rely upon a small number of these anecdotes. While they are provocative, they cannot be seriously evaluated until a large body of data becomes available.

2. Data Reduction The Gardners’ and Patterson’s claims are also seriously weakened by another aspect of their data-handling procedures. We have noted that they rely heavily upon anecdotes in describing their subjects’ performance. Where quantitative data are reported, however, they have been transformed and reduced so extensively as to obscure rather than clarify the character of the apes’ behavior. This is most obviously true of their treatment of the repetitions in ape signing. All of the ape researchers acknowledge that their subjects’ utterances included long, repetitive, continuous sequences such as me banana you banana me you give. The Gardners and Patterson present summary data from which these repetitions and other signs are eliminated, a highly questionable methodology. Patterson (1978) notes the existence of repetitions in Koko’s signing, but does not include them in her analyses. In their 1975 report on a test of Washoe’s ability to answer questions-the only paper in which extensive quantitative data on her performance appears-the Gardners reduced her replies one or more times before entering them in their analyses. Since this paper is the source of many of the strongest claims on Washoe’s behalf, and since it contains the most extensive documentation of her behavior, we will consider the Gardners’ procedures in some detail. The Gardners eliminated from Washoe’s answers all signs that repeated those in a question, signs that appeared more than once in a reply, and signs from the class termed “markers”. Signs were eliminated from 46% of her replies. Since her actual utterances are not appended, the precise number of signs which were deleted cannot be determined. From their description of this procedure, it appears likely that at least 25% of Washoe’s signs were deleted. This is an extremely large loss of information. It is difficult to recall a study of child language in which such a large proportion of utterances were deleted and left unanalyzed. An entire category of signs, termed “grammatical markers”, was also eliminated.

184

A4. S. Seidenberg and I,. A. Petitto

Signs from this group serve jmportant functions in the modulation of meaning. Nevertheless, with the exception of the temporal marker the, these signs could not alter the appropriateness or inappropriateness of the reply [to a question]. (Gardner & Gardner, 1975, p. 248)

The Gardners provide no evidence in support of this conclusion. In the absence of information on the use of “markers” such as gimme, please, want, hurry, more, can’t, and others, the opposite interpretation-that they do affect the appropriateness of replies-is equally valid, and intuitively more plausible. The Gardners are forced, for example, to defend the interpretation that Gimme Susun, Enough Susan and More Susan are as appropriate in response to the question Who she.7 as are You Susan or Susan there. The disturbing quality of these data transformations is revealed by considering how they would render a child’s utterances. A father says to his son, “Where are you standing?” The child replies, “I are you gimme hurry where here where you you are I”. By the Gardners’ rules, this would be tabulated as the presumably correct response “I here”. The other words would be eliminated as repetitions or “markers”. Yet one would require extraordinary contextual information in order to justify terming the untransformed reply “appropriate”. There are two points to be made here. The first is that we cannot determine from the Gardners’ report the exact number of Washoe’s responses which were rendered in this manner. We only know that such alterations were permitted under their rules, and that one or more signs were deleted from almost half her replies. The second is that data transformed in this manner present a substantially different picture than would the raw data. The Gardners’ few examples from Washoe’s replies illustrate this point. The response You me you out me became you me; open lollipop please became repetitions and open; Susan bite there became there. By eliminating “markers”, the Gardners transform long, redundant utterances into strings which more closely resemble human utterances in their superficial form. This yields a perhaps distorted view of the character of ape utterances. While the Gardners believe that “this scheme of simplification cannot do justice to the richness of meaning that can be found in Washoe’s replies” (1975, p. 252), its effect is to alter her utterances radically, before they have been documented. In the context of a paper containing a large corpus of actual utterances, these data reduction procedures might possess some legitimate heuristic value. In the absence of a corpus, these data are hopelessly ambiguous; rather than shedding light on Washoe’s linguistic abilities, they obscure what she actually did.3 3

See

facing

page.


185

To summarize, the Gardners, Fouts, and Patterson have failed to provide a comprehensive characterization of their subjects’ behaviors. They have not published systematic observational data or a corpus of even moderate size. The data in thzse reports are largely confined to anecdotes which describe isolated occurrences whose importance cannot be ascertained. In the few instances where quantitative data are provided, they have resulted from multiple transformations and reductions of actual utterances which are not made available. Of course, the fact that the relevant data have never been reported does not mean that the conclusions of the ape signing researchers are incorrect. In the remainder of this paper, we will examine other aspects of the apes’ known behaviors which suggest that linguistic interpretations are not merely unsubstantiated, but also in error.

3. Repetition

and syntax

As we have seen, perhaps the most serious consequence of these sampling and data reduction procedures is that they effectively suppress information concerning the structure of ape sequences, especially their characteristic redundancy. The pervasive occurrence of repetitions poses a set of questions which bear critically on the hypothesis that apes show linguistic abilities. Because these utterances have never been fully documented, these questions have never been explicitly addressed. First, how do the superficial forms of ape utterances compare with those of deaf and hearing children? Children in the early stages of language acquisition typically produce utterances that are reduced relative to the corresponding adult forms, leading to the characterization of their speech as “telegraphic” (Bloom, 1970b). In contrast, apes repetitiously expand. The occurrence of an utterance such as you me banana me banana you would be quite odd in the speech of any human. Limber (1977) notes the existence in child language of long utterances (e.g., I do pull it the way he hafta do that so he doesn’t-so the big boy doesn’t come out), but states that these typically exhibit hierarchical infrastructures that are not seen in ape sequences. Limber concludes that “Whereas virtually all children use hierarchically structured complex sentences by the beginning of their fourth year at the latest, there is little evidence that any ape ever did.” We would more strongly assert that 31t should be noted that the Gardners provide neither the actual responses ones. Their data consist entirely of the number of responses to each question from predesignated target categories (see pp. 000-000 below).

nor the transformed that contained signs

186

M. S. Scidenbcrg and IL A. Petitto

repetitive, inconsistently structured strings are in fact characteristic of ape signing (see Terrace et al., in press). The Gardners’ conclusion is quite different. They assert that “transcripts of Washoe’s spontaneous signing... show striking similarities to the speech of children”. (1974b, p. 734) However, they have never provided the transcripts that would substantiate their claim; the only data that do permit such comparisons, those of Terrace et al., appear to contradict their assertion. It is clear that, whatever their content or complexity, Nim’s utterances were very different in their superficial form from those of either deaf or hearing children. In the absence of explicit comparisons of large samples of ape and child utterances, then, the Gardners’ statement cannot be taken as fact. In the face of this superabundance of repetition, some obvious questions concerning the organization of ape utterances arise. Are the signs combined into non-random, albeit repetitious, patterns? Do they show internal structure which may be described by simple syntactic rules? Or are they merely a hash of random combinations? Although there are several discussions in the literature of the possibility that Washoe’s utterances showed evidence of syntactic structure, we take this issue to be moot until there is an accounting of a large number of utterances and their frequencies of occurrence. While the Gardners have presented examples of signs which Washoe combined in a consistent linear order as evidence for her use of syntax (e.g., clothes white, baby mine, and tickle me in Gardner & Gardner, 1974, p. 17), the status of these examples is similar to that of the cookie rock anecdotes: they cannot seriously be evaluated except in the context of a corpus and distributional analyses which have never been provided. At a minimum, what is required are statistical analyses of the frequencies of occurrence for a large number of individual signs, signs in combination with each other, and repetitions. These analyses are necessary in order to eliminate certain trivial interpretations of ape signing. Each ape’s output might be described by a finite state device, where the frequency of occurrence for a particular sign is merely a function of the frequency of the previous sign; their signs might be combined randomly, or into a small number of stereotypic patterns. Any of these outcomes would provide strong evidence against the hypothesis that the apes were signing linguistically, and until such analyses are performed, they cannot be discounted (see Terrace et al., in press). Distributional analyses would also provide an empirical basis on which to evaluate any regularities in the ordering of signs. The interpretation of such ordered strings, should they occur in substantial numbers, presents some subtle issues of interpretation. From the discussions in Gardner & Gardner (1971, 1974), Patterson (1978) and elsewhere, it appears that the authors


181

believe that a regular ordering of a subset of signs provides sufficient evidence for the use of syntax, and hence provides critical evidence in favor of a linguistic interpretation of the apes’ behavior. However, the sine qua non of syntax is not the mere fact that certain chains of responses occur in a fixed order; many lower animals show such sequencing in a wide variety of behaviors. Rather, as Wilson (1975) has stated, True syntax occurs when separate signals, say meanings when alone create new messages when CBA, CAB, and so forth. In human speech, “George hunts”, “George hunts the bear”, and very different meaning. (p. 190)

A, B, and C, that have distinct presented in various orders: AB, each of the three permutations “The bear hunts George” has a

In addition, syntactic structures underlie the combination of classes of words (nouns, verbs, etc.), rather than individual lexical items. In order to demonstrate that the apes’ signing showed syntactic structure, then, three facts must be established: 1. That signs in isolation had particular meanings; 2. That signs combined in different linear orders had different meanings, e.g., You give me # Me give you; 3. That each regular ordering was not specific to a unique combination of lexical items. As we argue below, it is questionable whether the apes’ signs have the very specific meanings which are routinely attributed to them. Furthermore, none of the ape researchers has provided the contextual analyses which could indicate whether different orderings had different meanings. Finally, the Nim data-the only available corpus of ape signs analyzed by frequenciessindicates that while some of his combinations appeared in a regular order (e.g., me Nim occurred more often than Nim me), these regularities did not extend to classes of signs. Terrace et al. (in press) argue convincingly that Nim’s sequences do not have the syntactic structure of sentences. At this time, then, there are no positive indications that the apes’ signing showed evidence of syntax or contrastive use of sign order, although a definitive judgment must be deferred pending publication of more complete data. 3.1. Functions of repetition The pervasive occurrence of repetitions in ape signing also raises the question of their function. The Gardners eliminate these signs from their analyses because they believe that they are “redundant and cannot alter the appropriateness or inappropriateness of Washoe’s replies” (Gardner & Gardner, 1975, p. 252). Yet they are so characteristic of ape signing as to demand interpretation. Although we cannot confidently ascribe a function to these repetitions in the absence of a large corpus, some preliminary observations can be made.

188

M. S. Seidenberg and L. A. Petitto

First, it is clear that the apes’ repetitions differ from those which occur in ASL. Several phenomena in ASL may be termed “repetitious”, but these have conventionally-established communicative functions, and their occurrence is bound by grammatical rules. As far as can be determined from the existing reports, the apes’ repetitions serve no communicative function and occur freely. In ASL, a fixed number of repetitions may be specified as part of the root or citation form of a sign. The grammatically-correct citation form of eat, for example, requires exactly two movements of its defining hand configuration in a specified direction (towards the mouth), with a specified hand orientation. More than the requisite number of repetitions may occur only if a signer wishes to alter the sign’s meaning. These repetitions are accompanied by changes in the movement parameter of the sign; hence, they are not identical or verbatim. The exact forms of such repetitions and the contexts in which they may occur are specified by the grammar of ASL (see Klima & Bellugi, 1979; Friedman, 1977). For example, the idea of continuous eating (as conveyed by the utterance “I was so nervous I ate non-stop for an hour”) could be communicated by drawing out, enlarging, and repeating the movement of the eat sign. This modulation must be accompanied by a contextually-appropriate facial expression (Liddell, 1975). The additional repetitions, the exact form of the movement, and the simultaneous facial expression together communicate the continuous aspect. Thus, repetition in ASL is used conventionally with other types of visual and temporal information as part of the grammar of the language. This usage is not seen in ape signing.4 In contrast, a simple, non-linguistic interpretation may suffice to explain the repetition in ape signing. One might reasonably assume that both the apes and their trainers utilized a simple rule to the effect that “more signing is better”. That is, the training conditions in these projects are designed to encourage signing which would not otherwise occur. Trainers may perceive their task as the creation of contexts in which the quantity of signing is maximized with little regard to content (e.g., the presence of repetition). Apes may learn to discriminate that longer strings will be highly rewarded. On this interpretation, then, the apes’ repetitions are similar to those which may be induced in laboratory animals such as rats and pigeons using tech-

4Another interesting use of repetition occurs in the signing of deaf children. At an early stage, they will repeat signs instead of using the system of modulations. I:or example, to communicate that one was “working very hard”, the adult signer would modulate the citation form of ~,ork, using an appropriate facial expression and altering the movement parameter. Deaf children will instead repeat the work sign several times. The point is that the repetitions are of signs which will later cary modulations; they arc a precursor to the modulation system, rather than random cvcnts.


189

niques of standard learning theory. Repetitive pecking or bar-pressing will result when anticipated reinforcement is withheld (as in the early stages of extinction). Similarly, repetitive signing would occur if reinforcement were withheld until the ape produced a lengthy sequence. No linguistic function need be attributed to the apes’ signing under this interpretation because it is not the content or structure of their utterances that is critical, but rather their length and quantity. Is is possible, of course, that the functions of repetition are related to linguistic or conceptual processes in ways which have not been explored as yet. They might facilitate the mapping of concepts onto motor responses, for example; in this way they could function as pauses or hesitations do in spoken language. This is unlikely, however, as it is clear that deaf signers do not utilize repetitions in this way. Another possibility is that repetitions are a gestural analogue of stutter or sociocentric signals such as “you know” or “really” (Duncan, 1969). As with other interpretations of repetition, the available data do not permit evaluation of this hypothesis. Rather than showing similarities to the signing of humans, the apes’ repetitions appear to place their behavior squarely in the domain of animal communication systems. As Wilson (1975) has stated, If a zoologist were required to select just one word that characterizes animal communication systems, he might well settle on “redundancy.” Animal displays as they really occur in nature tend to be very repetitious, in extreme cases approaching

the point of what seems like inanity

to the human observer. (p. 200)

Inspection of the Nim corpus reveals that his signing showed this degree of redundancy, as we believe a corpus of any of the signing apes’ utterances would as well. Whatever the functions of repetitions, it is clear that they are a fundamental characteristic of ape signing; they should be the focus of future research, rather than be deleted as they have been until now.

4. Attributions

of meaning and grammatical function

In the reports on ape signing, there is an apparent failure to confront the primary methodological issue in language acquisition research, namely, what evidence justifies the attribution of meanings and grammatical functions to a child or ape’s behavior? Clearly, the assumption underlying the anecdotes, vocabulary counts, and other fragmentary data in these papers is that the signs have similar meanings for the apes and the experimenters. What, then is

190

A4. S. Seidmberg

ard I,. A. Petitto

the evidence that it is justified? In analyzing child language, two sorts of data are cited in support of such a’ttributions-observational and experimental. Both are represented in the ape signing literature; we consider them in turn. 4. I. Observational data As an empirical methodology, observation entails the accumulation of a large corpus of utterances and documentation of the contexts in which they occur. In discussing this method, Bloom & Lahey (1978) note, Perhaps the most important single factor in obtaining evidence of child language and development is that what children do and what else goes on in the context is at least as important as what children say and what they hear... The context in which language is used, by the child and by others speaking to the child, is as important as what is actually said for understanding children’s language and making inferences about what children know. (p. 29)

As we have noted, neither a large corpus nor contextual analyses of ape signing behaviors exist. Bloom & Lahey note that this inadequacy troubles many studies of child language as well: There is very often a tendency in the literature to report lists of child utterances, perhaps categorized in some way; the assumption is that the meaning of the utterance is “transparent” and when it is not, then a translation utterance is provided. The problem is that such child utterances interpreted from the adult’s point of view. (p. 29)

or gloss of the are very often

This tendency is exemplified by the Gardners’, Pattersons’, and Fouts’ reports on their subjects. They include long lists of the apes’ vocabulary signs, assigning specific meanings and categorizing them in terms of their syntactic functions and case relations, while providing little discussion of the evidence which motivated particular attributions. In assigning meanings and grammatical functions, these researchers appeal to three general observations: first, that the apes’ signs generally appeared in “appropriate” contexts; second, that the use of individual signs generalized to new referents and situations, and “overgeneralized” to conceptually-related referents; and third, that their signs had the form of signs in ASL. Both the relevance and veracity of these observations are questionable, however. 4.1.1. Gerwralizatiorl am? overgeneralization Although the ape researchers rely heavily upon the observation that their subjects used signs in “appropriate” contexts, the notion of “contextually appropriate utterance”isnever defined in their papers. Even under an explicit definition of “contextually appropriate sign”, however, this observation would n6t itself provide compelling evidence, since the apes could in general


19 1

produce contextually-appropriate signs by non-linguistic means, for example, imitation. Since the apes’ teachers were presumably signing sequences relevant to each context, the apes would merely have to sign part or all of these inputs in return. This is reminiscent of Weitzenbaum’s (1976) computer program ELIZA, which simulated the behavior of a therapist in part by echoing the “patient’s” input. The extent to which ape signing was produced in this manner has not been evaluated. A very high correlation between teacher input and chimp response is seen in the films of Washoe (see footnote 12). The possibility that a large proportion of Nim’s output-possibly more than 80%-resulted from direct or delayed imitation is being evaluated by Sanders’. In claiming that Washoe and Koko used signs in “appropriate” contexts, both Patterson and the Gardners invoke a behavioral theory of meaning. A sign is presumed to have a particular meaning because it is associated with certain stimuli. Individual objects and actions (or classes of objects and actions) are the discriminative stimuli for particular signs. Many of the signs in the apes’ vocabularies are in fact nouns and verbs which are exemplified by simple objects or actions; the Gardners’ vocabulary tests (197 1, 1974a) were intended to demonstrate that Washoe had learned associations between signs and objects (or pictures of objects). Leaving aside the usual criticisms of behavioral theories of meaning (see Fodor, 1977, for example), a characterization of the nature of these associations would be revealing of what the apes had learned. However, little can be ascertained from the published reports. Since the signs are assigned very specific English glosses, it is implied that the associations were quite close. One might conclude, for example, that the hand movement glossed as tree was consistently formed in the presence of these objects, and much less often in the presence of others. Again, however, we are forced to rely upon the authors’ assertion that the signs were in general used “appropriately”, since the contexts in which they occurred are never systematically documented. This problem is particularly acute with respect to putative vocabulary items such as silly, good, please, and hurry, which are not exemplified by simple actions or objects. In these cases, we have no knowledge of the actual referents. The few discussions of the apes’ use of individual signs in different contexts do not lend additional support to the hypothesis that they were consistently used to refer to well-defined classes of objects or actions. For example, the Gardners observe that, In translating “Washoese”, the problem is compounded by the small size of Washoe’s repertoire of signs. She had fewer signs to use for the referents in her ‘Sanders,

R. Conversations

with

a chimpanzee.

Columbia

University

Ph.D. thesis,

in preparation.

192


world that [sic] an adult native signer would have, so that many of the signs she used had more referents than the same signs had when used by an adult native signer. In addition, we deliberately enlarged the reference of many signs in an attempt to simplify her training. For example, we taught Washoe the sign bib to refer to bibs, napkins, washcloths, handkerchiefs, facial tissues, toilet paper, etc., and we only modeled this one sign for these referents, even though there are about as many signs in ASL as there are words in English for this group of referents. (Gardner & Gardner, 1971, p. 144)

What cannot be determined from this or any other passage in the literature is this: in judging that Washoe or Koko used signs “appropriately” in referring to objects, what range of stimuli were permitted as the referents for individual signs, how many signs were accepted as the “correct” names for individual objects, and what were the nature and frequency of errors? In the absence of systematic data concerning the use of individual signs, one can make very little of the claim that the apes used signs in “appropriate” contexts, or the assertion that their use of a sign “generalized” to new objects and contexts. The Gardners and Patterson also offer the global observation that the apes’ errors were “overgeneralizations” of the sort observed in child language (Clark, 1973a). Patterson, for example, describes Koko’s use of tree, which she “overgeneralized to asparagus, green onions, and other tall thick, objects presented vertically.” (Patterson, 1978, p. 83) Similar observations are seen in discussions of Washoe’s signing. These anecdotes leave the nature of the apes’ errors opaque. Were these overgeneralizations (typically termed “overextensions” in the child language literature) in the use of signs such as bib and tree specific to objects which showed some physical or functional relationship to the denoted referents, or did the apes use the signs with respect to a wide range of stimuli, some unknown (and possibly small) proportion of which were related? The belief that the apes exhibit conceptually-based overextensions has been widely taken as important evidence that the apes’ language behavior is similar to that of children. Given the failure to document the apes’ use of any single sign, and the lack of any explicit comparisons of the over-extensions of apes and children, this claim has no empirical substance, and remains merely an intriguing possibility. Instead of systematic observations which could answer the important questions concerning the use of individual signs, the following pattern is seen repeatedly: the apes learn hand configurations which the experimenters gloss as having specific meanings on the basis of their own knowledge of ASL or their intentions in training the animals. Then it is observed that the apes’ use of a sign “generalized” to other referents and contexts, which is taken to indicate creative use of the sign. Except for anecdotal accounts, no attempt is made


193

to characterize the use of the sign, i.e., in terms of frequency, combination with other signs, contexts of occurrence, etc. In the absence of such information, one is simply left with the experimenter’s initial attribution. By assigning specific meanings to the apes’ hand shapes-when in fact they may be used with reference to heterogeneous groups of stimuli-close similarities to human language are implied when none may in fact exist. This disturbing pattern is especially clear in the case of signs whose meanings are abstract. For example, both the Gardners and Patterson attribute knowledge-of the sign please to their subjects. The grammar of please has of course interested linguists for some time. We know from Sadock (1974) and others that its use follows some very subtle constraints. Although Koko and Washoe formed hand configurations which their trainers glossed as please, no comparison of their use of these signs with the use of please in English or ASL is provided. Thus there is no evidence of any resemblance. Yet by glossing a response as please, such a correspondence is implied. The sign sorry is also attributed to Washoe and Koko without any description of its use. Nim also used a hand configuration glossed as sorry. This gesture was largely under the control of his teachers’ threats. If they appeared angry or ready to punish him, he would sign sorry. This sign appeared almost exclusively in contexts where such a threat was imminent. In common with many of the apes’ behaviors, this one is not uninteresting; it may be that Nim learned to mediate threatening interactions in a manner unavailable to apes in the wild. However, the inclusion of this sign on lists of the apes’ vocabularies-without documentation of its use-leaves the possibly disingenuous implication of a deep isomorphism with the use of sorry in English or ASL. It is unclear exactly what knowledge the Gardners and the others intend to ascribe to the ape who “knows” these signs. Does the animal who signs please understand the human’s system of social interaction and discourse, i.e., rules of the type described by Goffman (1954), Lakoff (1973), Searle (1975) Grice (1975) and others? Does it have a sense of politesse? Or is the Gardners’ attribution solely based on the observation that chimpanzees can learn hand shapes which the observers intend to be read as please? The superficiality of the Gardners’ claim is seen when compared to Bates’ (1976) study of the acquisition of polite forms of expression in Italian children. Bates cites a wide range of data which trace the child’s acquisition of several polite forms, including please; she interprets the data within a theory of the child’s developing pragmatic competence. Similarly, in signing sorry, does the ape intend to express remorse? If not, what is the sign’s meaning or function? In this case, it is instructive to consider Van Lawick-Goodall’s (197 1) observations of “apologetic” behavior in wild chimpanzees:

194


When one human being begs forgiveness or gives forgiveness to another there are moral issues involved; it is when we consider these that we get into difficulties in trying to draw parallels between chimpanzees and human behavior. In chimpanzee society, the principle involved when a subordinate seeks reassurance from a superior or when a high ranking individual calms another is in no way concerned with the right or wrong of the aggressive act. A female who is attacked for no reason other than that she happens to be standing too close to a charging male is quite as likely to approach the male and beg for a reassuring touch as is the female who is bowled over by a male while she attempts to take a fruit from his pile of bananas. (p. 244)

Simply stated, then, the fact that an ape performs a particular behavior in threatening situations does not justify labeling that behavior as the sign “sorry”. Rather, in signing sorry, the apes appear to have learned about the pragmatics of language use. Nim knew that by signing sorry in certain contexts, he could affect his teachers’ behaviors in ways beneficial to him. He learned the consequences of the act of signing sorry rather than its meaning and grammatical function. This use of the sign is similar to the child’s earliest use of words, before meanings and concepts have been mapped onto them, but after the pragmatic functions of language have begun to be understood (Bates, 1976). There is, then, an enormous problem of over-attribution in the ape signing literature. Nowhere do the researchers describe the criteria that resulted in their crediting the apes with knowledge of signs such as time, sorry, please, happy, good, bad, big, small, quite, and pound. These attributions appear to entail strong claims about the apes’ cognitive capacities-e.g., their ability to make comparative judgments or label affective states-that are vastly underdetermined with respect to the behavioral evidence offered. In place of this evidence, the researchers offer English glosses of the apes’ behaviors. It is the connotations of meaning, grammatical function, and usage which these glosses suggest to the literate observer that imply that the apes possessed linguistic skills, not their largely undocumented behaviors.

4.1.2. Arguments from the form of signing behaviors A third general observation that is weighted heavily in ascribing meanings and grammatical functions to the apes’ utterances is that they produce hand shapes and movements which resemble those of deaf children who use ASL. Since developmental psycholinguists attribute linguistic knowledge to children who exhibit these behaviors, the argument goes, we must attribute this knowledge to apes who show similar behaviors (Gardner & Gardner, 1974b, 1975).


195

We argue below that the apes’ behaviors resemble ASL very little. However, one cannot attribute linguistic functions to the apes’ behaviors simply on the basis of physical resemblance in any case. This would be to mistakenly assume that it is the form of certain behaviors which qualify them as “linguistic” rather than their conceptual bases and communicative functions (Huttenlecher & Higgins, 1972; Nelson, 1977; McNeilP). The existing reports document the apes’ ability to learn behaviors whose superficial forms resemble those of sign language behaviors. If the mere physical form of an utterance were crucial, however, it would be the case that mynah birds and telephone answering machines would exhibit linguistic abilities. In the case of the apes, one cannot abandon the distinction between speech and language simply because the channel of communication is visual. Although this point may appear trivial, it is overlooked by the ape signing researchers, who have documented the form of the apes’ signs in far more detail than their use of these signs. Although they do not provide sufficient evidence for linguistic attributions, analyses of the form of the apes’ behaviors are revealing in other respects. A significant proportion of the apes’ vocabularies consist of natural ape behaviors which are labelled as lexical items. These include pick (signed by picking a part of their anatomy), hug (signed by hugging), tickle (signed by tickling), kiss (signed by kissing activity), scratch (signed by scratching) and others. The behavioral basis of each of these “signs” is in fact an activity exhibited by apes in the wild (van Lawick-Goodall, 1968, 1971). In the sign language projects, however, they are glossed as lexical items, with attendant linguistic implications. These behaviors show almost none of the critical features of human language; estimates of the size of the apes’ vocabularies are inflated when such activities are glossed as “signs”. The fact that simple activities and gestures are glossed as signs raises another important issue. In attempting to evaluate what the apes in the sign language projects have learned, it is necessary to distinguish between natural communicative behaviors common to all apes, and those which are learned in the course of training. The ape researchers do not draw this distinction; all of the apes’ communicative behaviors (and some non-communicative ones as well, e.g., scratching) are together classified as “signs”. This generic use of the term “sign” with reference to a diverse group of behaviors contributes to uncertainty over the interpretation of their activities. It is especially misleading because the term is at the same time applied to the units of ASL, to which the apes’ behaviors show almost resemblance (see below).

6McNeill, D. Unpublished chapters from forthcoming book, 1975.

196

M. S. Scidenberg and L. A. Petitto

The Gardners mention the possibility that Washoe’s “signs” may be confounded with her naturally-occurring gestures, but they do not pursue the issue. They note the importance of determining the nature of Washoe’s “base population” of responses (Gardner & Gardner, 197 1, p. 129) and state that some of these are similar to the cheremes which Stokoe, Casterline & Croneberg (1965) have identified as the basic components of signs (similar to phonemes in spoken language). They also write that Washoe’s signing was frequently observed both “live” and on film by outside observers... Some of these outside observers were deaf or were hearing persons who were already familiar with ASL. Others were hearing persons who had to be taught the rudiments of ASL for the purpose. All of these outside observers have agreed that, in general, Washoe’s cheremic responses were quite distinct from her non-cheremic responses. Those observers who were most familiar with ASL had the least difficulty in making such discriminations. (p. 128)

Later in the same paper they state that Many of Washoe’s vocalizations were very similar to those observed in wild chimpanzees and chimpanzees raised in laboratories... Some of them are found in other primates, including man, and many of them can be interpreted by a human being the first time they are heard. We would also expect to find natural gestures, and some of these should be similar to the signs of ASL. If there were more complete information about the natural gestures of captive and wild chimpanzees it would be easier to identify them. Under the circumstances we know that a few of Washoe’s gestures could have appeared without any specific training, and we guess that this was so for some others. (p. 137)

This is followed by two examples of Washoe’s natural gestures which were glossed as signs, but the topic is not considered further (see Patterson, 1978, pp. 83-86 for similar discussion.) At this time, it is an unresolved empirical issue whether any of the apes’ naturally-occurring manual gestures resemble cheremic elements of ASL. It is clear, however, that the apes’ natural behaviors do not resemble ASL signs. The latter are highly stylized and specific in form. The citation form of each sign is defined along four parameters (hand configuration, location, movement, and orientation); with the possible exception of a small number of very simple signs, the exact configurations of elements which characterize ASL signs are not observed in the gestures of apes. These considerations suggest that an important experimental control is missing here. In order to de-confound the apes’ natural gestures and their acquired “signs”, a comparison between the behaviors of signing and nonsigning chimps is needed. Two chimps could be raised from infancy in similar environments, exposed to human contact to an approximately equal


191

extent, engaged in similar activities. The experimenters could attempt to teach sign language to only one; they could perhaps communicate with the other through gesture and other non-signing means. Longitudinal comparisons between the communicative behaviors of the two would then provide a better test of what the signing chimp had learned. A possibly more interesting group of ape signs may be indexical in the sense of Peirce (1932). This group includes eat, me, you, go come, brush, groom, up, down, give, this, that, there, and others. These may be indexical because the motion of each sign is part of its referent, or denotes it by means of pointing. The apes apparently sign brush, for example, by a brushing movement, give by a begging gesture, etc. Many of these signs purportedly involve pointing, e.g., they sign me by pointing to themselves, you by pointing to another, eat by pointing to the mouth, this, that, and there by pointing to objects or locations, etc.7 The claim that the apes utilized indexical signs, and pointing in particular, is itself a strong one. Although it is widely acknowledged that wild apes use a variety of gestures, these are not well documented. As Yerkes & Yerkes observed in 1929, the apes’ “modes of affective expression” include Position, pose, and movements, often termed gestures, of the hand, trunk, limbs, extremities... Particularly noteworthy are the so-called gestures made with the head, limbs, and extremities. Frequently these are mentioned in the literature, although seldom with sufficient precision and minuteness of description to provide the reader with a definite picture of the organism. (p. 285) Existing naturalistic studies do not make clear whether these gestures involve pointing or other types of indexical reference (cf., deVore, 1965; Jay, 1968; Goodall, 1968, 197 1; Menzel & Johnson, 1976; Chevalier-Skolnikoff & Poirer, 1977). The ape researchers’ claim, then, that their subjects used pointing to communicate several semantically-distinct concepts is very strong in light of existing primatological evidence. Again, however, the systematic

‘We will distinguish between indexical and iconic signs, although the two are usually collapsed. Indexical signs have the properties described in the text. Their meanings can be understood without special knowledge. Iconic signs contain visual information that is related to or schematically represents some aspects of the referent of the sign. This relationship varies in abstractness, but typically requires special knowledge in order to be identified. For example, the sign shoe is formed in ASL by hitting two fists together side by side exactly twice. It is iconic because the motion is thought to represent two heels clicking together. Iconicity has been defined operationally in tests such as Bellugi & Klima (1976). Typically, naive observers cannot determine the iconic basis of a sign unless they are told its meaning; then they show high levels of agreement as to the source of the iconicity. A sign such asgive is indexical (as the apes form it) because the motion of the sign is its meaning. In this way, indexical signs are an analogue of certain speech acts (Searle, 1969). The motion of an iconic sign, in contrast, schematically represents some aspect of meaning or use or the actual referent.

198

M. S. Seidenbergand L. A. Petitto

observations which would provide compelling evidence for a radical conclusion are lacking. If Washoe and Koko used indexical signs, this accomplishment would place their communication at a level beyond that of other apes and lower animals. It is important, then, that the ape researchers attempt to provide substantive supporting evidence. However, the use of indexical reference does not provide the basis for positive comparisons to human language behavior. In spoken languages, of course, words are largely symbolic rather then indexical. Speakers use arbitrary symbolic forms even for words such as me and you which could be communicated indexically. Indexical reference is utilized within American Sign Language; here the important difference is that its use is governed by the grammar of the language. As was true of repetition, this information is utilized within a complex, conventional system which is not observed in signing apes. The use of pointing provides a good example of this profound difference. The signs me and you may be signed indexically in ASL through pointing in some contexts; however, they are also expressed by non-indexical means (e.g., via incorporation of pronouns into verbs) in others. The means by which me and you are expressed depends upon a set of grammatical constraints that are only beginning to be codified (Friedman, 1975). Pointing may also be used to refer to persons, objects, and locations that are not in the immediate environment. Signers will locate such nouns at metaphoric locations in the signing space. In a conversation about two persons who are not present, for example, the signer can in effect “place” them at points on the left and right within the signing space. Among other functions, this permits pronominal reference to be accomplished by merely re-pointing to (or looking at) these locations in space. This process is termed “establishing loci”. Note that the signer points not to an actual person or place, but to an abstract locus in space. Thus, pointing is exploited in an elegant, systematic way. There is no evidence that pointing in apes follows conventions of this type, if it exists at all. Given the large proportion of indexical signs such as give and gestural or behavioral “signs” such as hug in the apes’ vocabularies, it would perhaps be a powerful test of what they had learned to attempt to teach them abstract (i.e., arbitrary) signs for messages that can be gestured or otherwise enacted. For example, they might be taught a wholly arbitrary sign for kiss or eat. On the basis of our experiences with Nim and the behavior exhibited by Washoe on film, we predict that the apes could learn such signs, but would quickly abandon them in favor of actual enactments of the behavior, unless intensive maintenance procedures were utilized. It appears that the apes will not use abstract forms to refer to activities they are themselves able to perform, ex-

Signing behavior in apes: A critical review 199

cept under duress. This would follow from a failure to exploit the symbolic relationships among an abstract sign, its meaning, and its referent.& We suggest, then that a large proportion of ape signs can be interpreted without any special knowledge of apes or ASL because they are unlearned gestures and activities that are seen in the behavior of wild apes and other animals. Although the Gardners have attached great importance to the observation that signers of ASL had no difficulty in interpreting Washoe’s behavior, we suggest that the same would largely be true of any sensitive observer. This too could be cast as an empirical question, by having both signing and non-signing observers independently interpret videotapes of Washoe signing. Although signing observers would enjoy an advantage (because the ape apparently did learn some arbitrary signs), we would expect their judgments to show high levels of agreement with those of naive observers. We do not wish to claim that all ape signs involve behaviors such as hugging or pointing, nor do we claim that the existence of such behaviors in a communicative system disqualifies it as “language” or that because the apes’

aThere are other sources of non-arbitrary visual information in ASL, in particular iconicity and the limited use of pantomime. The fact that such information is utilized in manual-visual languages such as ASL is sometimes used to draw similarities among signing in ASL, communication in wild apes, and gestural proto-languages (Hewes, 1976; Stokoe, 1978). However, the degree to which signs in ASL contain representational elements is a matter of considerable controversy (cf., Bell@ & Klima, 1976; Battison, 1978; Brown, 1978; Frishberg, 1975; Friedman, 1977; Newport & Bellugi, 1978). The essential facts are these: while it is clear that ASL embodies non-arbitrary information in its structure, the function of this information in the perception or production of utterances does not. Ionicity in particular may be a vestige of diachronic processes of sign evolution (Frishberg, 1975), rather than a perceptually salient feature of the language. Although the representational information provided by the forms of certain signs may prove to have some functional utility (perhaps in the creation of new signs, or in the acquisition of signs; Brown, 1978), much of this information is suppressed in effect by its occurrence in a discourse context where non-representational information (provided by the abstract formational parameters of signs, and by modulations, inflections, and other expressive elements) predominates (Bellugi & Klima, 1976). Identifying the sources and functions of non-arbitrary visual information in ASL remains an open empirical question, however. Roger Brown (personal communication) has suggested that iconicity may have less salience for signing apes than for deaf children. We hesitate, however, to assign much importance to the issue of iconicity until it is clear that these signs form a coherent class on the basis of either formal structure or perceptual function in native signing. Brown’s provocative hypothesis that iconicity is implicated in the sign acquisition process in deaf children is based on an experiment with hearing children who already had facility with spoken language, and who were explicitly shown the iconic bases of some signs. Note, however, that one could examine the value of iconicity to the apes by explicitly teaching both iconic and abstract forms, and comparing their acquisition and use. It would be quite interesting, for example, if abstract signs were abandoned in favor of enactments, but iconic signs were not, or if iconic signs were learned more rapidly. Given the lack of evidence concerning the functions of iconicity in child or adult signing, and the primitive nature of the apes’ behavior, however, this speculation borders on fantasy.

200


behaviors do not show the complexity of ASL they cannot learn a manualvisual language. Rather, this analysis of the form of the apes’ behaviors has several functions. First, it provides details of the apes’ performance which are obscured by merely noting that their vocabularies were of a particular size. Second, it extends the comparisons of ape signing and ASL signing, and isolates substantial differences. Third, it suggests that much of the apes’ “signing” behavior has a natural, non-linguistic basis. Finally, the demonstration that a large proportion of ape signs are natural gestures (and hence largely unlearned) increases the likelihood that the apes’ other signs were learned as individual responses to particular objects, actions, or settings through intensive and specific training. If, in fact, a large proportion of the signs are gestural, possibly modifying or extending the apes’ natural system of gestures, then the possibility that the remaining signs were learned in a rote fashion and signed using non-linguistic responding strategies increases. We should note that if our interpretations are correct, it is misleading to describe this behavior as “signing”, since this term suggests a level of abstraction that is largely absent. Nor is pantomime the correct term, since the apes’ behaviors are time-locked to on-going activities, unlike true pantomime. The correct description of this behavior was provided by the Gardners in their original (1969) paper. Nim, Washoe, Koko, and other signing pongids show evidence of having learned a “standardized system of gestures”. Although modern languages may have evolved from such gestural systems (Hewes, 1976; Wilson, 1975) these gestures do not exhibit critical features of human languages. Rather than showing similarities to human language use, many aspects of the apes’ behavior (e.g., their use of gestures, imitation, and facial expressions) show similarities to the pre- and early-linguistic behaviors of very young children. As work by Piaget (1952) Werner & Kaplan (1963) and, more recently, Bates (1976) and Shatz (1978) has indicated, the infant or young child’s gestures and sounds may be used for communicative purposes that are the precursors of language, rather than language itself. Bates has attempted to relate the development of the child’s pre-lingual communicative behavior to its cognitive development, as considered from a Piagetian perspective. Chevalier-Skolnikoff (1976) has begun to consider the communicative behaviors of apesin a similar fashion. It is likely that both the nature and the limits of ape communicative capacities will be revealed by further studies which relate their overt communicative activities to their level of cognitive development. 4.2. Experimental tests Investigations of language acquisition in children have employed experimental tests of both comprehension (e.g., Huttenlocher, Eisenberg, and


201

Strauss, 1968; de Villiers and de Villiers, 1973; Chapman and Miller, 1975) and production (e.g., Berko, 1958; Fernald, 1972). There are two widely known experimental tests of the apes’ signing abilities, the Gardners’ tests of Washoe’s vocabulary (Gardner & Gardner, 1971, 1974a) and their test of her ability to answer wh-questions (Gardner & Gardner, 1975). There have been no rigorous tests of the apes’ abilities to comprehend signs, a remarkable omission in light of recent theoretical work on the differences between comprehension and production (e.g., Bloom, 1974b; Chapman and Miller, 1975.)’ 4.2. I. Vocabulary tests Washoe was required to sign the names of pictures or objects which the experimenters could not see. Her performance on this task has led to the conclusion that she possessed the ability to name objects; however, the implied characterization of the naming process is quite sterile. The Gardners merely required of the animal the capacity to learn simple associations between individual behaviors and individual objects or classes of objects. Under this characterization, a limited ability to name objects is within the capacity of many animals, including standard laboratory subjects such as pigeons (Herrnstein, Loveland and Cable, 1977).” It is important to realize that the complete results of these vocabulary tests have never been published. Although the Gardners stated in 197 1 that “When the results of this testing program are complete, we will publish them

‘The Gardners explicitly reject all tests of comprehension because of the “experimenter errors” they are believed to involve (Gardner & Gardner, 1975). They note that comprehension tests require controls of non-linguistic cuing that are “difficult to enforce”. From the fact that such tests are difficult, however, it does not follow that they should be rejected entirely. The Gardners suggest that production tests avoid such biases; however, their own tests and that of Fouts et al. (1978; see p. 206 below) clearly show that such tests introduce other problems. “Herrnstein et al. (1977) taught pigeons to differentially respond to pictures of objects from classes such as free or body of water. If a pigeon learned to pair the presentation of a picture of a tree with the pecking of a particular colored keylight (and pair non-tree stimuli with a second light), it would be said to “name” trees under the Gardners’ characterization. It seems clear that pigeons could be taught to simulate the “overgeneralizations” reported in the ape literature. The pigeons that learned to discriminate trees might be presented with pictures of stimuli physically similar to trees, e.g., asparagus, green onions, etc. False positives to these stimuli would then represent “overgeneralizations” or “conceptuahy-based overextensions”. Varying the negative stimuli in this way would be a good way to explore what the pigeons in these “concept learning” experiments have learned. See also Wittgenstein (1953), p. 187: “If you trained someone to emit a particular sound at the sight of something red, another at the sight of something yellow, and so on for the other colours, stiII he would not yet be describing objects by their colours. Though he might be a help to us in giving a description.” Cited by Marshall (1971) in a very interesting discussion of some issues raised by chimpanzee signing.

202 M. S. Seidenberg and L. A. Petitto

in a separate article” (p. 160), they have never done so. The descriptions in Gardner & Gardner (1971, 1974a) make most aspects of their procedures clear, but Washoe’s performance is not systematically characterized.” We are again given only partial descriptions of her performance. In one account, for example, the Gardners state that Washoe’s poorest

performance on the self-paced version of the box test was obtained in a session in which 33 different items were presented with three different exemplars of each for a total of 99 trials... Washoe named 53 of the 99 examplars correctly. The following week the same test was rerun in two sessions. The 99 exemplars were presented in reverse order and a different-hence equally blind-observer served as [experimenter]. The second time around Washoe’s score was again 53. This is typical of the results that we have obtained in retests; improvement that could be attributed to practice on the original test was negligible. Most likely this was because Washoe received so much training on the items before testing, and there was so little opportunity for further training during the course of a test. (p. 160)

Again the important questions are begged. First, what stimuli were utilized in these tests? Nominals such as bird, toothbrush, or dog were apparently tested by presenting toys, the actual objects, or pictures. Were vocabulary signs such as clean, hurry, good, mine, silly, etc., which are not exemplified by simple objects, tested? If so, what was Washoe naming when she signed clean or good? Second, in scoring Washoe’s responses, was there a single prespecified correct response for each stimulus, or was any from a group of signs accepted? Since many of her signs were related (e.g., banana, fruit, eat; brush, toothbrush, comb), the latter possibility cannot be excluded. Third, how were her correct and incorrect responses distributed? If 53 of her responses were correct, and each exemplar was presented three times, then her correct responses could have been limited to as few as 18 signs, far fewer than the 1OO- 150 attributed to her elsewhere. This number would shrink even further if a sign were scored as correct to more than a single item. Finally, what was Washoe’s training for this test? How intensively was she drilled on items similar’ or identical to those used in these tests? Did she show similar levels of

“One aspect of their training and testing procedures is not clear. It can be seen by stopping individual frames of two fiis of Washoe (see footnote 12) that she was sometimes reinforced with food after naming an object. In some cases, she is seen reaching off to the left side of the screen to be handed a reinforcer; in another test, she names an object in the testing box, reaches into the box, and pulls out a morsel from inside it. The use of food or other reinforcers would not necessarily reflect negatively on Washoe’s signing abilities. However, this is an important part of the training and testing procedure which has not been documented.

Sip&g behavior in apes: A critical review 203

performance in naturalistic settings where her behavior could not be routinized? Although these vocabulary tests are the most widely cited evidence for language abilities in apes, they are of questionable interest. Certainly they cannot be taken seriously until Washoe’s performance is exhaustively characterized.

4.2.2. Question answering Although it is generally claimed that Washoe was able to form appropriate signs in the presence of particular objects or pictures, it is unclear whether her skill extended beyond the milieux of the double-blind vocabulary tests. In a somewhat different context, their 1975 test of Washoe’s ability to answer questions, the Gardners abandoned this simple requirement. Washoe’s response to a question was scored as correct if it contained a sign from a predesignated target category. For example, the question what’s that took a noun response; the question where’s that took a locative, etc. Washoe’s vocabulary signs were grouped into grammatical categories for scoring purposes; the sign glossed as listen, for example, was scored as a noun. The presumed meanings of her signs were ignored. Thus, if the Gardners held a ball in front of Washoe and asked her what’s that? the response listen come would be scored as correct because it contains listen, a member of the target category noun. Given their liberal scoring procedure, it is not surprising that the Gardners could conclude that Washoe’s responses were superior to those of children at stage III. Note that the validity of this test rests upon several issues which are not discussed. First, the judgment that Washoe answered questions with signs drawn from appropriate target categories depends entirely on the manner in which her signs were classified. The assertion, for example, that she correctly answered the question what’s that with signs from the category noun is uninteresting if signs were arbitrarily placed in this class. Since the sole criterion for a correct answer was that it belong to the appropriate target category, the method by which her signs were classified assumes great importance. The Gardners state that in performing these classifications they relied upon intuitions concerning “good usage” in Ameslan. This enterprise is of dubious validity. First, one questions the intuitions of non-native non-linguist signers of ASL. Second, many of the signs in question have multiple grammatical functions in ASL; hence intuitions do not provide a basis for placing each sign in a single category. Third, it has been unacceptable methodology for some time in developmental psycholinguistics to assume that child language is simply a reduced version of adult language. Bloom (1970a) has made this

204 M. S. Seidenberg and I‘. A. Petitto

point quite forcefully in her criticism of the generation of psycholinguistic research in which attempts were made to describe child language in terms of models developed on the basis of adult utterances. From the fact that a structure has a particular meaning or grammatical function in adult language one cannot conclude that a child employs the structure in the same way. The Gardners make this erroneous assumption in classifying Washoe’s utterances on the basis of their meaning and function in adult ASL. Finally, Washoe’s vocabulary signs could have grammatical functions only if they had meaning, which we have previously questioned. Observe that if Washoe were signing in ASL, it would be possible to classify her signs on the basis of the use of grammatical inflections and modulations. It is somewhat disingenuous for the Gardners to assert that “In ASL as in most natural languages, many signs can be used either as nouns or as verbs...” (1975, p. 25 I), since it is the case that in ASL, such differences in grammatical function are signaled by modulations and inflections on the citation forms of signs. Determining the grammatical function of a sign in ASL, then, is not dependent primarily upon intuitions about general usage in “good Ameslan”, but rather on identifying the inflectional elements and syntactic variations associated with nouns or verbs. There is no indication that the apes learned to distinguish nouns from verbs in these or any other ways, again suggesting that while the apes’ signing is termed “Ameslan”, it shows none of the characteristic structures or expressive devices of that language. It is highly questionable whether the Gardners or Patterson had an empirical basis on which to identify the grammatical function of a sign, or determine whether the apes’ signs exhibited multiple functions. Another important consideration in evaluating this test is that we cannot determine from the given information whether Washoe’s responses were rote associations to particular questions and/or objects which were learned through intensive and specific training. As was true of the vocabulary tests, her preparation for this test is not described in detail, and her actual responses are not appended. As a result, we cannot evaluate the degree of flexibility and heterogeneity seen in her replies. Given the Gardners’ description of their procedures, Washoe could have “correctly” answered the questions in this test by simply emitting signs from her vocabulary in random order until she produced one from the target category. Similarly, she could answer a question by learning to associate a single target sign with a wh-sign; since meaning was irrelevant, she could use this sign in any context where the discriminitive stimulus (wh-sign) occurred. The Gardners’ test exhibits an impoverished conception of language, since Washoe is to be credited with the ability to answer questions if she can learn to associate a narrow class of signs with each wh-sign. Thus the ability to

Signing behavior in apes: A critical review 205

answer questions is reduced to a discrimination learning task. It is not surprising to find that this task can be accomplished by apes who have experienced intensive training. In summary, then, meanings and grammatical functions are attributed to the apes’ utterances without sufficient evidence. The tests of Washoe’s vocabulary and question-answering ability are subject to non-linguistic interpretations. In addition, they are orthogonal to a critical question: can the apes use their signs to refer to objects which are not in the immediate environment, or actions which are not concurrent? The design of these tests is such that they could never provide evidence that the apes had learned to use signs in this manner. In fact, there has been no test, and thus there is no evidence, of the apes’ ability to use signs which are displaced relative to their referents. In the absence of such evidence, their behavior cannot be equated with that of humans using language.” It might be argued that in evaluating the ape research we have applied more rigorous methodological standards than those applied to child language research. Our claim, however, is that understanding the apes’ behavior requires data of the sort which are routinely reported in the child language literature: a corpus of substantial scope and detail, systematic studies which relate the

‘*There is another source of information on Washoe’s behavior, two widely seen films, the Public Broadcasting System-Nova film, and a classroom film entitled “Teaching sign language to the chimpanzee Washoe”. Although these films cannot be considered scientific records, they do provide important information concerning Washoe’s behavior, and they are perhaps the primary source of information for most persons familiar with this work. As such, they warrant serious attention. We have recently undertaken a frame-by-frame analysis of both films. They were transcribed using a system developed by Petitto for the analysis of ASL in deaf children. The transcription includes information concerning both the signing and non-signing behavior of ape and teachers, and the contexts in which it occurred. The transcription is currently being subjected to several analyses, and a complete report is in preparation. We will mention some preliminary findings here. The most striking fact emerging from these analyses is the degree to which Washoe’s signing was dependent upon that of her teachers. Signing occurred almost exclusively as a reaction to a teacher’s input, and was highly imitative. For example, the sequence baby in my drink is cited as evidence for Washoe’s ability to combine signs creatively. Washoe does not form these signs in a continuous sequence; rather, they are prompted by the teacher (see also Terrace, Petitto, Sanders, and Bever,in press). Two facts about the form of Washoe’s signs are revealed on film. First, it is clear that all of the signs which are seen were formed out of a small number of hand configurations and movements, many fewer than are used in ASL. Most of these signs are so similar in form as to be indistinguishable, and in both films, individual hand movements are assigned multiple interpretations. In these cases, the researchers appear to have glossed the movements on the basis of their own knowledge of the context. The second fact is that the small number of movements and gestures which are seen bear a striking resemblance to those of apes who have not been taught “sign language”, lending further support to our earlier observation that natural gestures are glossed as signs. Washoe’s movements are highly similar to those seen in films of Vicki, a non-signing chimp raised by Hayes & Hayes (1952). The possibility that the apes’ natural system of communicative gestures is highly unmalleable should be investigated further.

206 M. S. Seidenberg and 1,. A. Petitto

comprehension and production of utterances to the contexts in which they occur, and non-trivial experimental tests of aspects of the apes’ signing abilities. The methodology of.observational studies of both comprehension and production is described in thorough detail by Bloom & Lahey (1978), Huttenlocher (1975) and others. There is also a lively and extensive literature on experimental approaches to language acquisition research. Elegant, ingenious studies such as Clark (1973b), Huttenlocher et al. (1968), Chapman & Miller (1975), Glucksberg, Krauss, & Higgins (1975), Bates (1977), Shatz (1978) and others have demonstrated some possible strategies for experimental research; see also Bellugi-Klima (197 1). Premack (1977) has considered the issue of comprehension tests for lower primates. The ape researchers frequently demonstrate unfamiliarity with the basic methodology of language acquisition research. A particularly clear example is provided by a recent report by Fouts, Shapiro, & O’Neil (1978). Fouts et al., report a test of a signing ape’s ability to produce the signs in, on, and under in appropriate contexts. The subject, Ali, is a chimp who is reported to comprehend 130 signs and “at least as many” words in spoken English. There is an excellent study on the acquisition of in, on and under in children (Clark, 1973b). One of Clark’s major points is that children in the early stages of acquisition use non-linguistic responding strategies in following instructions containing these words (i.e., instructions of the form, “Put the X__ the Y”, where X is an object, Y is a target location and the blank is filled by in, on or under). In Clark’s experiments, successful performance depended critically upon the nature of the objects used in the test. If Y was a container, children followed most requests by placing a toy animal in it. If Y was a flat surface, they typically placed the object in it. In the former case, children appeared to “comprehend” in but not on or under; in the latter, they “comprehended” on but not in or under. Hence, properties of the testing situation and non-linguistic responding strategies controlled subjects’ performance. Failing to explore these would yield misleading conclusions concerning their comprehension abilities. Fouts et al. (1978) offer no acknowledgement that their test controlled for the use of similar strategies. In fact, it appears from their description that the subject could use a very simple non-linguistic strategy to perform at the reported levels. After an unspecified training period, Ali was tested on 80 trials in which he was requested to name the relationship between a pair of objects. The critical feature of their design is that no sign was ever the correct response on two successive trials. That is, an in trial was always followed by an orz or under trial, etc. The ape could perform reasonably well on this task by learning the rule, “respond with one of the two signs which were not used on the previous trial.” If the ape randomly selected between the two remain-


201

ing alternatives, we would expect performance to be somewhere around 50% accuracy. Ah’s reported accuracy is 49.1%. It is seen, then, that there are considerable methodological problems associated with testing a child or chimp’s production and comprehension of the words in, on, and under. These are typical of the problems involved in language acquisition research. In common with the other ape researchers, Fouts et al. (1978) show no awareness of the fact that these problems have been explored in the child language literature.” A major conclusion to be drawn from studies such as Clark’s (1973b) is that children are able to produce and respond to utterances before they have acquired knowledge of their meanings and grammatical structures, by exploiting non-linguistic contextual information and other knowledge (see also Chapman, 1977). The ape sign language researchers have failed to exploreor even acknowledge-the possibility that the apes used such non-linguistic strategies. What is required in the ape signing literature, then, is evidence that the apes’ production is not merely imitative, that they can produce or comprehend signs in non-stereotypic situations, that sign production or comprehension is not exclusively a function of cuing by non-linguistic aspects of the environment, or the teacher’s behavior, or the structure of a particular test, and that their behavior is not merely the routinized, inflexible, overlearned product of intensive and specific training. Obtaining the relevant sorts of evidence is a difficult task, whether the subject is an ape or child. Unfortunately, there is no simple test which would establish that the apes’ utterances have meaning. A case can only be built by drawing together evidence from multiple sources. If the ape is able to name objects consistently across a variety of settings and exemplars, if it is able to identify a particular property in unfamiliar objects, if it is able to identify multiple properties in a single object, if it is able to comprehend or produce signs without prompting or immediate feedback following errors, if it produces signs spontaneously without tangible reward, if it can identify relations among objects-if it can use language in some of these ways and others, then one may begin to have some confidence that linguistic attributions are appropriate. What is required, then, is a systematic documentation and characterization of the apes’ use of signs. Merely establishing through casual observation that they are able to produce behaviors whose forms superficially resemble those

‘3Comprehension tests of the signs in, on and under form of the signs could cue correct answers.

would present

other problems

because

the mere

208


of signs in ASL is inadequate, linguistic competence.14

5. Other comparisons

at least as a basis for interesting

claims about

to ASL

The frequent invocation of ASL in discussions of ape signing (e.g., Gardner & Gardner, 197 1, 1974a; Patterson, 1978; Fouts, 1977) makes it clear that these researchers believe that the apes are learning a communicative system that shares significant features with ASL. However, evidence concerning the structures and/or expressive devices common to the apes’ behavior and to signing in ASL is superficial at best. The discussions of ASL in these reports are very shallow; for example, in a recent paper by Fouts (1977), there is no mention of ASL beyond the title, “Ameslan in Pan [Troglodytes] .” That the apes use ASL is merely assumed, rather than adduced from behavioral data. It is highly questionable whether Washoe or Koko were taught ASL at all. American Sign Language is radically different from English; its grammar does not resemble English, and although some signs can be glossed as English words, its lexical structure is different as well. It is possible to sign English, that is, to use signs (embellished with fingerspelling) in English syntax; this is, in fact, termed “Signed English”, “ Straight English” or “Siglish” depending on exactly how much of spoken English is rendered. It appears that none of the apes were presented with ASL as a model. Rather, the apes were taught individual vocabulary signs, some taken directly from ASL, some reduced from ASL signs, and some invented for the ape; these were signed largely in English word order. Since neither the apes nor their teachers used the appropriate grammatical structure or expressive devices, their signing was not ASL; since they did not use grammatical morphemes (e.g., -ing, -ed, -1~) or function words (e.g., a, the, are) they were not following Signed English. The language model in each project was a pidgin sign. In light of these facts, it is misleading to term their behavior signing in ASL.

“It is interesting in this light to note Lashlcy’s (1913) observation, cited by Mowrer in a discussion of “talking birds” (Mowrer, 1950): The older literature of animal psychology abounds with anecdotes designed to display the intelligence of the parrots, but there has been no experimental study of the birds and nothing is known of the manner in which they learn to speak; whether by direct imitation, by the gradual imitative modification of instinctive notes, or by chance combinations of instinctive notes which, meeting the approval of the trainer, are rewarded and so ‘set’ in memory. (Lashley, 1913, p. 362)


209

The fact that many of the apes’ signs were non-ASL forms is acknowledged but not emphasized in the existing reports. In the Washoe films, this distinction is not observed at all. In the PBS-“Nova” program called “The first signs of Washoe”, it is asserted many times that Washoe was taught ASL. A signing, hearing person is seen forming and explaining Washoe’s signs prior to films of Washoe forming the same signs. Although the narration repeatedly describes Washoe’s signs as ASL, some of the demonstrated signs are not ASL; rather, they are idealizations of signs unique to Washoe. This fact alone does not imply that Washoe had not “acquired language”. However, it does show a remarkable lack of accuracy. Note also that the ape researchers describe their subjects’ behaviors exclusively in terms of hand configurations; the structure of signing in ASL is much more complex (Stokoe et al., 1965; Klima & Bellugi, 1979). We have noted some of the dimensions of signing already. Signs are defined along three parameters in addition to hand configuration; a change within any one of them signals a change in meaning. Signing in ASL is not merely a matter of producing the citation forms of signs in a particular order, however; an extensive system of modulations and inflections on signs is used at the same time. This system is used to signal syntactic information, tensing, subtle variations in meaning, and other aspects of the message. This is effected through the systematic use of movement and visual information provided by facial expressions, eye gaze, head and body orientation, the structured use of the signing space, and other means. A sign in ASL, then, is defined not by a unique hand configuration, but along several dimensions simultaneously. We take the fact that the apes show no ability to use these expressive devices as immediate refutation of the claim that they are using ASL. Their failure to learn to use them, despite years of intensive training, purportedly in ASL, contrasts strongly with the fact that deaf children rapidly learn to use them in the course of natural, non-intrusive interactions. Many of these expressive devices are seen in nascent or primitive form in the early utterances of deaf children. For example, deaf children progressively learn to establish loci. At early stages, they will sign on or toward an actual object or location, moving around a room if necessary, rather than placing it at a metaphoric location in the proximal signing space. Over time, they learn to stay within the signing space and establish loci through pointing, eye gaze, and body shifts. Another primary communicative device in ASL is the use of eye gaze and facial expressions in conjunction with signs. It is interesting that while lower primates use facial expressions and eye gaze as part of their natural communicative system, they do not use them to modulate the meanings of the hand shapes they learn, in contrast to the child learning ASL. It is an inter-

210


esting empirical question whether apes can learn to use this information within a conventionalized, non-natural system. A third example is provided by the deaf child’s use of the signing space in denoting time reference. This is accomplished through use of the time line, an imaginary vertical plane which passes through the signer’s body just behind his ears. A forward thrust of the signer’s hand(s), relative to this plane, at ear level denotes reference to the future. A backward thrust at the same level indicates reference to the past. The present is represented in front of the signer’s body, centered around the chest-stomach area. It would be important, of course, to determine whether lower primates can make reference to past or future activities, or learn to use this system; at present there is no evidence that they can, again distinguishing them from deaf children and adults using ASL. These examples are only illustrative. They serve to indicate some fundamental ways in which chimpanzee signing differs from signing in ASL. In their failure to even mention the basic structures and devices which characterize the language, the existing comparisons of ASL and ape signing behavior are exceedingly shallow. These examples fail to convey the dramatic differences between the signing of deaf children and the apes’ behavior. These differences cannot be overestimated. Deaf children use their language as do hearing children-in spontaneous, inventive, inquisitive conversation. The apes’ behavior is of a wholly different nature-it must be coerced, extracted, manipulated. The apes do not sign spontaneously, but rather because it is demanded of them: this is clearly seen in the Washoe films, and would be revealed, we believe, by analyses of extended discourse between an ape and its teacher. Signing must be imposed on these animals and maintained through intensive, intrusive intervention. Whatever the scope of their cognitive and communicative abilities, it cannot be claimed that their behavior resembled that of children.

6. Conclusions We have demonstrated that the conclusion that signing apes show linguistic abilities is vitiated by the absence of appropriate da.ta and analyses. The omission of information which is routinely included in psycholinguistic studies of child language obscures the essentially non-linguistic character of the apes’ behavior; the fragmentary data which are provided are consistently over-interpreted. Thus the widespread claims on the apes’ behalf are at best premature.


211

What have the apes learned? They appear to have learned about the communicative context. They knew, for example, that signing was highly valued by their teachers. They signed because the mere act of signing had positive consequences, regardless of its content. They learned to solve particular problems, such as finding the sign or signs which the experimenters were seeking in a particular context. They learned responding strategies of varying degrees of complexity. A simple strategy might be: “Sign until the experimenter terminates the trial.” A more complex strategy might be: “In an eating situation, sign any from the class of signs including eat, drink, more, banana, give, please, sweet, finish.” These may be interesting behaviors to study, but they relate only tangentially to language. We conclude with three observations. First, we await longitudinal studies of the apes’ signing. There are as yet no indications that they can retain large sign vocabularies beyond puberty. The claim that they possess the capacity to learn languages rests on demonstrations that their signing does not disappear when their intensive training is relaxed, and that they can in fact internalize rules of the sort which underlie human language capacities. Second, the source of many of the problems in the existing literature may be traced to the Gardners’ statement that their analyses “do not depend on any special theory of linguistics or psycholinguistics” (1975, p. 256). Their analyses depend upon a special theory that is created de facto by their acceptance of a simplistic set of assumptions about language structure and language learning. The Gardners simply fail to acknowledge their theory. Their antipathy to current linguistics (e.g., 1974b) has led them to embrace a theory of learning. It is possible that Washoe could have accomplished more if her trainers had possessed a richer conception of language and communication. Finally, it should be clear that there are genuinely interesting aspects of the apes’ cognitive and communicative capacities that have not been explored as yet. It is apparent from the sign language projects that the apes are extremely intelligent. Studying their natural abilities may ultimately be more revealing about the behaviors of both apes and humans than attempts to impose restricted forms of languages upon them.

References Bates, E. (1976) Language and Contexr. New York, Academic Press. Battison, R. (1978) Lexical Borrowing in American Sign Language. Washington, D.C., Linstok Press. Bellugi, U. and Klima, E. (1976) Two faces of sign: Iconic and abstract. In Harnad, Steklis, and Lancaster (Eds.).

2 12


BeUugi, U., Klima, 1:. and Siple, P. (1974/75) Remembering in signs. Cognition, 3/2, 93-125. Bellugi-Klima, U. (1971) Some language comprehension tests. In C. S. Stendler (Ed.), Lan#u~e Training in Early Childhood Education. Urbana, Ill., University of Illinois Press. Berko, J. (1958) The child’s learning of English morphology. Word, 14, 150-177. Bloom, L. (1970a) Child language, adult model. Review of Paula Menyuk, Sentences children use. Contemp. Psychol., 15, 182--184. Bloom, L. (1970b) Language Development: Form and Function in Emerging Grammars. Cambridge, Mass., MIT Press. Bloom, L. (1973) One Word at Q Time. The Hague, Mouton. Bloom, L. (1974a) Commentary. Monographs of the Society for Research in Child Development, Vol. 39, No. 3. Bloom, L. (1974b) Talking, understanding, and thinking: Developmental relationship between receptive and expressive language. In R. L. Schiefelbusch and L. Lloyd (lids.), Language Perspectives -ACquiSition, RetQrdQtiOn, and Intervention. Baltimore University Park PreSS. Bloom, L. and Lahey, M. (1978) Lank-age Development and LanAaage Disorders. New York, Wiley. Bronowski, J. and Bellugi, U. (1970) Language, name, and concept. Science, 168, 669-673. Brown, R. (1970) The first sentences of child and chimpanzee. In R. Brown (Ed.), Psycholinguistic .Papers. New York, The 1,rce Press. Brown, R. (1978) Why are signed languages easier to learn than spoken languages’! Part two. Bull. Am. ACQ~. Arts Sci., 32, 25 -44. Chapman, R. (1977) Comprehension strategies in children. In J. Kavanaugh and P. Strange (Eds.), Language and Speech in the LQborQtor_v,School, and Clinic. Cambridge, Mass., MIT Press. Chapman, R. and Miller, J. (1975) Word order in early two and three word utterances: Does production precede comprehension?J. Sp. Hear. Rex, 18, 355-371. Chevalier-Skolnikoff, S. (1976) The ontogeny of primate intelligence and its implications for communicative potential: A preliminary report. In Harnad, Steklis, and Lancaster (Fds.). Chevalier-Skolnikoff, S. and Poirier, b‘. (Eds.) (1977) Primx?te Bio-social Development: Biological, SocioloKicQl atld Ecological Determinants, New York, Garland Publishing Inc. Clark, E. (1973a) What’s in a word? In T. 1:. Moor (Ed.), Cognitive Development and the Acquisition of Language. New York, Academic Press. Clark, E. (1973b) Non-linguistic strategies and the acquisition of word meanings. Cognition, 2, 161~ 182. de Villicrs, J. and de Villiers, P. (1973) Dcvelopmcnt of the use of word order in comprehension. J. Psvcholing. Rex, 2, 331-341. de V;re, I. (1965) Primate Behavior. New York, Ilolt, Rinehart and Winston. Duncan, S. (1969) Non-verbal communication. Psj~hol. Bull., 22, 118-137. Ferguson, C. and Slobin, D. (Eds.) (1973) Studies of Child Language Development. New York: Holt, Rinehart and Winston. I:ernald, C. D. (1972) Control of grammar in imitation, comprehension, and production: Problems of replication. J. verb. Learn. verb. Behav., I I, 606-613. Fodor, J. A., Bever, T. G., and Garrett, M. F. (1974) The Ps~~9~ofogv o/‘Lan@age. New York, McGraw-Hill. I:odor, J. D. (1977) Semantics. New York, Thomas Y. Crowcll. ITouts, R. (1972) Use of guidance in teaching sign language to a chimpanzee. J. Compar. Physiol. Ps_schol.. 80, 515-522. I:outs, R. (1977) Ameslan in Pan. In G. Bourne (Ltd.), Progress in Ape Research. Neu~ York, Academic Press. Fouts, R., Shapiro, G., and O’Neil, C. (1978) Studies of linguistic behavior in apes and children. In P. Siple (Ed.).


2 13

Friedman, L. (1975) Space, time, and person reference in American Sign Language. Lang., 51, 94O961. Friedman, L. (Ed.) (1977) On rhe Other Hand. New York, Academic Press. Frishberg, N. (1975) Arbitrariness and iconicity: Historical change in American Sign Language. Lang., 51,696-719. Gardner, R. A. and Gardner, B. T. (1969) Teaching sign language to a chimpanzee. Science, 165, 664-672. Gardner, B. T. and Gardner, R. A. (1971) Two-way communication with an infant chimpanzee. In A. Schrier and F’. Stolnitz (Eds.), Behavior of Non-human Primates, Vol. 4, New York, Academic Press. Gardner, B. T. and Gardner, R. A. (1974a) Comparing the early utterances of child and chimpanzee. In A. Pick (Ed.), Minnesota Symposium on Child Psychology, Vol. 8, University of Minnesota Press. Gardner, R. A. and Gardner, B. T. (1974b) Review of Roger Brown’s “A first language”. Am. J. Psychol., 87, 729-736. Gardner, R. A. and Gardner, B. T. (1975) Evidence for sentence constituents in the early utterances of child and chimpanzee. J. exper. Psychol. Gen., 104, 244-267. Glucksberg, S., Krauss, R., and Higgins, E. T. (1975) The development of referential communication skills. In F. Horowitz (Ed.), Review of Child Development Research, Vol. 4. Chicago, University of Chicago Press. Goffman, E. (1954) The Presentation of Self in Everyday Life. New York, Anchor Books. Goodall, J. van Lawick. (1968) A preliminary report on expressive movements and communication in the Gombe Stream chimpanzees. In P. Jay (Ed.). Goodall, J. van Lawick. (1971) In the Shadow of Man. Boston, Houghton Mifflin. Crice, H. P. (1975) Logic and conversation. In J. Morgan and P. Cole (Eds.). Harnad, S., Stekhs, H. and Lancaster J. (1976) Origins and Evolution of Speech and Language. New York Academy of Sciences, Vol. 280. Hayes, K. and Hayes, C. (1952) Imitation in a home-raised chimpanzee. J. Compur. Physiol. Psycho/., 45,450-459. Herrnstein, R., Loveland, D., and Cable, C. (1977) Natural concepts in pigeons. J. Exp. Psychol.: Animal Learn. Memory. 2, 285-302. llewes, G. (1976) The current status of the gestural theory of language origin. In Harnad, Steklis, and Lancaster (Eds.). Hackett, C. (1963) The problem of universals in language. In J. Greenberg (Ed.), Universals of Language, second edition. Cambridge, Mass., MIT Press. Hoffmeister, R. (1978) The development of demonstrative pronouns, locatives, and personal pronouns in the acquisition of American Sign Language by deaf children of deaf parents. Unpublished Ph.D. thesis, University of Minnesota. Hoffmeister, R., Best, B. and Moores, D. (1974) The acquisition of sign language in deaf children of deaf parents. Bureau of Education to the Handicapped progress report. Huttenlocher, J. (1974) The origins of language comprehension. In R. L. Solso (Ed.), Theories in Cognitive Psychology. New York, Halstead. Huttenlocher, J. Eisenberg, K., and Strauss, S. (1968) Comprehension: Relation between perceived actor and logical subject. J. verbal Learn. verb. Behav., 7, 527-530. Huttenlocher, J. and Higgins, E. T. (1972) On reasoning, congruence and other matters. Psychol. Rev. 79,420-427. Jay, P. (1968) Primates: Studies and Adaptation and Variability. New York, Holt, Rinehart and Winston.

2 14

M. S. Seidcnberg and L. A. Pctitto

J. (1976) A hypothesis about the uniqueness of natural language. In Harnad, Steklis, and Lancaster (Eds.). Klima, E. and Bellugi, U. (1979) The Signs of Language. Cambridge, Mass., Harvard University Press, Lakoff, R. (1973) The logic of politeness, or, minding your p’s and q’s, In Papers from the ninth regional meeting of the Chicago Linguistic Societ)‘, University of Chicago. Lane. H., Boyes-Braem, P., and Bellugi, U. (1976) Preliminaries to a distinctive feature analysis of American Sign Language. Cog. Psychol., 8, 263-289. Lashley, K. S. (1913) Reproduction of inarticulate sounds in the parr0t.J. Animal Behavior, 3, 361& 366. Liddcll, S. (1975) Restrictive relative clauses in American Sign Language. Research report, the Salk Institute for Biological Studies, La Jolla, Cal. Limber, J. (1977) Language in child and chimp? Amer. Psychol., 32, 280-295. Marshall, J. (1971) Can humans talk? In J. Morton (Ed.), Biological and Social Factors in Psycholinguistics. London, Logos Press Ltd. Menzel, E. and Johnson, M. (1976) Communication and cognitive organization in humans and other animals. In Harnad, Steklis, and Lancaster (Eds.). Morgan, J. and Cole, P. (1975) Synfax and Semantics, Vol. 3: Speech Acts. New York, Academic Press. Mowrer, 0. H. (1950) On the psychology of ‘talking birds’-a contribution to language and personality theory. In Learning Theory and Personality Dynamics. New York, The Ronald Press Co. Nelson, K. (1977) The conceptual basis for naming. In J. McNamara (Ed.), Language, Learning and Thought. New York, Academic Press. Newport, E. and Bellugi, U. (1977) A flower is a flower is a flower. In E. Rosch and B. Lloyd (Eds.), Cognition and Categorization. Hillsdale, N. J., Lawrence Erlbaum Associates. Patterson, l;. (1978) The gestures of a gorilla: Language acquisiton in another pongid. Brain Lang. 5, 56-71. Peirce, C. (1932) Collected papers. C. Hartshorne and P. Weiss (Eds.). Cambridge Mass., Harvard University Press. Piaget, J. (1952) The Construction of Reality in the Child. New York, Balantine. Premack, D. (1971) Language in chimpanzee? Science 172, 808-822. Premack, D. (1977) Intelligence in Ape and Man. Hillsdale N.J., Lawrence Erlbaum Associates. Rumbaugh, D. (Ed.) (1977) Language Learning in a Chimpanzee. New York, Academic Press. Sadock, J. (1974) Toward a Linguistic Theory of Speech Acts. New York, Academic Press. Schlesinger, I. (1978) and Namir, P. (Eds.) Sign Language of the Deaf. New York, Academic Press. Searle, J. (1969) Speech acts. Cambridge, England, Cambridge University Press. Searle, J. (1975) Indirect speech acts. In P. Cole and J. Morgan (Eds.). Seidenberg, M. S. and Petitto, L. A. (In preparation) On the Linguistic Abilities of Apes. Hillsdale, N.J., Lawrence Erlbaum Associates Shatz, M. (1978) On the development of communicative understanding: An early strategy for interpreting and responding to messages. Cog. Psychol., IO, 271-301. Siple, P. (Ed.) (1978) Understanding Language through Sign Language Research. New York, Academic Press. Stokoe, W. (1978) Sign language vs. spoken language. Sign Language Studies, 18, 69 -90. Stokoe, W., Casterline, D. and Croneberg, C. (1975) A dicfionary ofAmerican Sign Language on Linguzktic Principles. Washington, D.C.: Linstok Press. Second edition. Terrace, H. S. (1979) Is problem solving language? J. Exper. Anal. Behau., 31, 161-175. Terrace, H. S. (in press) Nim. New York: Alfred Knopf. Terrace, H. S., Petitto, L. A. and Bever, T. G. (1976a) Project Nim: Progress report I. Manuscript. Katz,


Terrace, Terrace,

2 15

H. S., Petitto, L. A. and Bever, T. G. (1976b) Project Nim: Progress report II. Manuscript. H. S., Petitto, L. A., Sanders, D. J. and Bever, T. G. (in press) On the grammatical capacities of apes. In K. Nelson (Ed.), Children’s Language, Vol. 2. New York: Gardner Press. “Two-way communication with an infant chimWatt, W (1974) Review of Gardner and Gardner, panzee”. Behav. Sci., 19, 7 I-75. Weitzenbaum, J. (1976) Computer Power and Human Reason. San Francisco, Freeman & Co. Werner, H. and Kaplan, B. (1963) Symbol Formation: An Organismic-developmental Approach to Language and the Expression of Thought. New York, Wiley. Wilson, E. 0. (1975) Sociobiology. Cambridge, Mass., Harvard University Press. Wittgenstein, L. (1953) Philosophical Investigations. Oxford: Blackwell. Yerkes, R. M. (1925) Almost Human. New York, The Century Co. Yerkes, R. M. and Yerkes, A. W. (1929) The Great Apes. New Haven: Yale University Press.

Cognition, Vol. 7, No. 2

Cognition, Vol. 7, No. 4

Cognition, Vol. 7, No. 3

Cognition, Vol. 7, No. 1

Cognition, Vol. 2, No. 2

Cognition, Vol. 5, No. 2

Cognition, Vol. 2, No. 3

Cognition, Vol. 2, No. 1

Cognition, Vol. 2, No. 4

Cognition, Vol. 9, No. 2

Cognition, Vol. 6, No. 2

Cognition, Vol. 8, No. 2

Cognition, Vol. 3, No. 2

Cognition, Vol. 4, No. 2

Cognition, Vol. 1, No. 2-3

Textual Practice Vol 7 No 2

Cognition, Vol. 6, No. 1

Cognition, Vol. 4, No. 1

Cognition, Vol. 4, No. 3

Cognition, Vol. 5, No. 1

Cognition, Vol. 3, No. 4

Cognition, Vol. 8, No. 4

Cognition, Vol. 3, No. 1

Cognition, Vol. 3, No. 3

Cognition Vol. 8, No. 1

Cognition, Vol. 1, No. 1

Cognition, Vol. 9, No. 3

Cognition, Vol. 9, No. 1

Cognition, Vol. 8, No. 3

Cognition, Vol. 4, No. 4

Cognition, Vol. 6, No. 4

Cognition, Vol. 7, No. 2

Cognition, Vol. 7, No. 4

Cognition, Vol. 7, No. 3

Cognition, Vol. 7, No. 1

Cognition, Vol. 2, No. 2

Cognition, Vol. 5, No. 2

Cognition, Vol. 2, No. 3

Cognition, Vol. 2, No. 1

Cognition, Vol. 2, No. 4

Cognition, Vol. 9, No. 2

Cognition, Vol. 6, No. 2

Cognition, Vol. 8, No. 2

Cognition, Vol. 3, No. 2

Cognition, Vol. 4, No. 2

Cognition, Vol. 1, No. 2-3

Textual Practice Vol 7 No 2

Cognition, Vol. 6, No. 1

Cognition, Vol. 4, No. 1

Cognition, Vol. 4, No. 3

Cognition, Vol. 5, No. 1

Cognition, Vol. 3, No. 4

Cognition, Vol. 8, No. 4

Cognition, Vol. 3, No. 1

Cognition, Vol. 3, No. 3

Cognition Vol. 8, No. 1

Cognition, Vol. 1, No. 1

Cognition, Vol. 9, No. 3

Cognition, Vol. 9, No. 1

Cognition, Vol. 8, No. 3

Cognition, Vol. 4, No. 4

Cognition, Vol. 6, No. 4

Recommend Documents