the effect of cognitive bias on children's early interpretations of locative commands

14
The Effect of Cognitive Bias on Children's Early Interpretations of Locative Commands Judith F. Duchan State University of New York A BSTRA CT This study examines the effect and the sources of nonlinguistic cognitive biases on children's early understanding of locative com- mands. Two- and three-year-old children incorrectly interpreted im- probable commands by making them probable. These same children were frequently able to correctly interpret unbiased commands which contained the same prepositions as those which they misinterpreted in the improbable commands. For example: they correctly interpreted the unbiased sentence, "Put the penny on the baby," but erred on "Put the box on the cowboy." (They put the cowboy in the box.) The sources for their cognitively biased interpretations on the im- probable commands were attributable to the perceptual features of the objects, the action schemes which children commonly use in relation to the objects, and the relative ease of motor response used to place one object in relation to another. PRELIMINARIES Preschool children, when responding to objects in their environment, have been found to behave in consistent ways (Clark 1972, Sinclair 1970). For example, when given a baby and a crib, children are likely to put the baby in the crib. These consistencies, which we will be calling cognitive biases, have been explained by Clark (1972 and 1973) as manifestations of the children's ability to recognize the perceptually salient properties of the objects in question. Sinclair (1970 and 1973), on the other hand, sees the cognitive biases as manifestations of action schemes which become activated by the objects in question. Wilcox and Palermo (1975) offer a third view which explains children's cognitive biases in terms of two - 246 -

Upload: judith-f-duchan

Post on 16-Sep-2016

212 views

Category:

Documents


0 download

TRANSCRIPT

T h e E f f e c t o f C o g n i t i v e Bias

o n C h i l d r e n ' s Ear ly I n t e r p r e t a t i o n s

o f Loca t i ve C o m m a n d s

Judith F. Duchan

S ta te Universi ty o f N e w Y o r k

A BSTRA CT

This study examines the effect and the sources of nonlinguistic cognitive biases on children's early understanding of locative com- mands. Two- and three-year-old children incorrectly interpreted im- probable commands by making them probable. These same children were frequently able to correctly interpret unbiased commands which contained the same prepositions as those which they misinterpreted in the improbable commands. For example: they correctly interpreted the unbiased sentence, "Put the penny on the baby," but erred on "Put the box on the cowboy." (They put the cowboy in the box.) The sources for their cognitively biased interpretations on the im- probable commands were attributable to the perceptual features of the objects, the action schemes which children commonly use in relation to the objects, and the relative ease of motor response used to place one object in relation to another.

P R E L I M I N A R I E S

Preschool children, when responding to objects in their environment, have been found to behave in consistent ways (Clark 1972, Sinclair 1970).

For example, when given a baby and a crib, children are likely to put the baby in the crib. These consistencies, which we will be calling cognitive

biases, have been explained by Clark (1972 and 1973) as manifestations of the children's ability to recognize the perceptually salient properties of the objects in question. Sinclair (1970 and 1973), on the other hand, sees the cognitive biases as manifestations of action schemes which become activated by the objects in question. Wilcox and Palermo (1975) offer a third view which explains children's cognitive biases in terms of two

- 246 -

COGNITIVE BIAS

"response tendencies": (1) to make the simplest motor response and (2) to put the objects in their most normal contextual relationship. While the theorists differ in their description of the processes underlying chil- dren's nonlinguistic cognitive biases, they agree that these cognitive biases strongly influence the children's early comprehension of language.

This study attempts to analyze the effect of children's cognitive biases on their early comprehension of language and to study the adequacy of various explanatory theories. In particular, it addresses itself to how cognitive biases affect the ways two- and three-year-olds interpret different locative sentences having the syntactic structure Noun-Preposition-Noun (NPN). The explanatory capabilities of the perceptual theory of Clark and the action scheme theory of Sinclair and the motor ease theory of Wilcox and Palermo are compared by using the various theories to account for different errors which the children make when carrying out the locative commands.

Several developmental studies of the processes involved in early lan- guage comprehension have revealed the influence of nonlinguistic cognitive biases on the child's interpretation of sentences (Bever 1970; Clark 1972, 1973a, b; Sinclair 1973; Strolmer and Nelson 1974; and Wilcox and Palermo 1975). Children have been shown to have difficulty carrying out commands which require that they respond in ways incompatible with their cognitive biases. For example, when Strohner and Nelson (1974) asked two- and three-year-old children to interpret The ball carried the wagon, the children responded by putting the ball in the wagon. Strohner and Nelson explained the error as due to the children's use of a "probable event strategy" leading them to interpret utterances in ways that are in keeping with their cognitive biases. Sentences whose semantic interpreta- tions are compatible with the child's (or adult's) cognitive biases have been called improbable sentences.

The two- and three-year-olds in the Strohner and Nelson study who were incorrectly interpreting improbable Noun-Verb-Noun (NVN) sen- tences were correctly interpreting another set of NVN sentences. This second set was made up of two nouns whose relationship was less pre- dictable than those in the improbable sentences. That is, for sentences such as The boy kissed the girl, the children were not likely to be as cognitively biased as they were for the improbable sentences; they were

- 247 -

LANGUAGE SCIENCES, VOLUME 2, NUMBER 2 (1980)

not so likely to derive predictable relationships between verb and noun referents. Strohner and Nelson's unbiased sentences were reversible sen- tences where noun referents could serve equally well as agents or patients. Strohner and Nelson's children had sufficient linguistic knowledge to derive a correct semantic interpretat ion for these unbiased NVN sen- tences. Since the improbable sentences were similar linguistically to the unbiased, the children must have had available to them sufficient linguistic knowledge to derive a correct semantic interpretat ion for them. The data suggest that the children constructed a eognitively biased interpretation instead of using the semantic interpretat ion derived from linguistic know- ledge that NVN = Agent, Action, Patient, which they used for the un- biased sentences. Thus, Strohner and Nelson's results suggest that children can operate within at least two specifiable frameworks when interpreting NVN utterances. The first framework might be described as one of cogni- tively biased interpretat ion and tends to occur in response to sentences

whose semantic inpretation is either congruent with or antithetical to the child's cognitive biases. This would lead correct responses to probable sentences (put the baby in thecrib), and incorrect responses to improbable ones (put the baby under the crib.) The second framework might be called semantic interpretat ion and tends to occur in responses to more neutral or unbiased sentences where the child does not have a strong cognitive bias for relating the reference objects in particular ways to one another.

Sinclair and Bronckart (1972) also studied children's responses to improbable NVN sequences. They found that children disregarded the word order cues and constructed their interpretat ion around the expected relationship between the action pattern and the agent or patient. Sinclair (1973), from this and other data, theorized that cognitive bias in sentence interpretat ion follows from prelinguistic cognitive stages with the action pattern offering the cognitive fulcrum around which the agent and patient become differentiated:

The first direct action differentiations - i.e., between the subject's own action and the object acted upon on the one hand, and the child himself as an agent or some other person or object as agent - give rise to the first grammatical functions of subject-predicate and object- action.

Clark (1973b) presented children from one and a half to three years

- 248 -

COGNITIVE BIAS

with what she regards improbable locative commands (e.g., Put the [toy animal] under the table) and probable, locative commands (e.g., Put the [toy animal] on the table). Clark's data coincides with Strohner and Nelson's and Sinclair's in that a high percentage (70%) of the errors on the improbable commands were errors predicted by the nonlinguistic cognitive bias. That is, the c lddren put objects in, on, and under the referent of the second noun depending upon what Clark called the perceptual saliency of the reference point object. Further, Clark's children showed pattern- ing which she describes in two ordered rules or strategies: (1) if the re- ference point object is a container, X goes inside it, and (2) if the reference point object is a horizontal surface, X goes on it. These rules, according to Clark, are used by the children in order - first they apply rule 1, then, if not applicable, they apply rule 2. Thus, regardless of the verbal com- mand, children put the toy animal in the objects which they perceived as containers (box, tunnel, truck, and crib); and on objects which they perceived as non-containers and as having horizontal surfaces (table and bridge).

Clark interpreted her findings as evidence that perceptual characteristics of the noun referent provide the basis for cognitive bias:

The strategies represented by rules 1 and 2 depend on some very general perceptual properties of objects in the world; namely, whether they are containers or not, and whether they have any horizontal surface or space. Tile strategies themselves are derived from the hypo- theses the child has constructed on the basis of his prior knowledge. In the case of spatial relations, the child's hypotheses depend heavily on his perception of the normal or expected spatial relations that may hold between objects in the world. (Clark 1974 : 119)

Wilcox and Palermo (1975), using the same age subjects as Clark (1973b) varies some of the stimulus items to reflect a bias in the direction of under, and found that some of the children's responses to improbable commands such as Put the boat on the bridge did not conform to Clark's perceptually defined rules but, rather, to the contextual biases (i.e., that boats go under bridges). Further, Wilcox and Palermo (1975) found their youngest children placed objects next to one another, and on rather than

in or under one another. The authors at t r ibuted next to responses to the

- 249 -

LANGUAGE SCIENCES, VOLUME 2, NUMBER 2 (1980)

fact that "some of the children did not know either the meanings of the words or the contextual relationships which usually hold between the ob j ec t s . . . [This] suggests that the youngest age groups may not have had any linguistic grasp of locatives" (Wilcox and Palermo 1975: 253).

The Clark (1973b) and Wilcox and Palermo (1975) studies clarify the importance of the contextual relationships between the referents in the construction of children's interpretation of locative commands. However, their results leave unanswered the question of whether their children were capable of making a semantic interpretation of the prepositions in, on, and under, since they did not present their subjects with unbiased utterances. That is, we do not know whether the children's cognitively biased interpretation was the only interpretive framework the children had available to them (as was suggested by Wilcox and Palermo) or whether the children understood the prepositions in more contextually neutral situations.

The present study differs from Clark's (1973b) and Wilcox and Palermo's (1975) in that children are presented with improbable as well as unbiased Noun-Preposition-Noun (NPN) commands. The improbable commands require the child to place objects in spatial relations incong- ruent with their prior experience such as Put the baby under the crib. The unbiased commands require them to relate objects that do not commonly occur together such as Put the penny on the baby. For the improbable commands, their cognitive bias should lead them to an incorrect answer via the use of a cognitively biased interpretation. For the unbiased com- mands, they should not be able to construct a cognltively biased inter- pretation and should depend on their linguistic knowledge to construct a semantic interpretation.

M E T H O D

Twenty preschool children of ages ranging from 2;0 to 3;10 with a mean age of 3 ;4 served as subjects for the study. Each child was pre~ented with 18 improbable and 18 unbiased commands. There were six improb- able commands and six unbiased commands for each of the locative prepositions in, on, and under.

The commands required the children to select objects from a group and place them in a spatial relation to one another. To reduce the search

- 250 -

COGNITIVE BIAS

l~aae for finding the objects, the objects were put into two boxes, with approximately 15 items in a box. Half of the commands required mani- pulation of objects in box 1 and half in box 2. The order of presentation of the two boxes of objects were randomized, with some children carrying out the 18 unbiased and improbable commands for box 1 objects first and others carrying out the 18 unbiased and improbable commands for box 2 objects first. The set of 18 commands for each box was presented in a different random order to each child.

The f'trst set of objects (box 1 or box 2) were presented to the child, and each object was named, either by the child or the experimenter. If the child's word for the object did not match the word used in the com- mand, he was told, "We're going to call it ," where the noun used in the command was inserted. The child was, then, presented with the first 18 commands. If he failed to fred the correct objects, the experiment- er pointed in their general vicinity and repeated the command. The com- mand was also repeated if the child requested it, if the child seemed to have forgotten but did not request repetition, or if the child's response was unclear. Once the first 18 commands were completed, the second group of objects were named and the second set of commands given.

R E S U L T S

Performance of the 20 preschool children on improbable and unbiased conditions for in, on, and under is shown in Table 1. For each preposi- tion, a one-tailed Wilcoxin Matched Pairs Signed Ranks test was used to compare improbable and unbiased commands. In the in and on conditions there were significant differences in the predicted direction with p <.005 and p <.025, respectively. However, for the preposition under the children

did no better on the unbiased commands than on the improbable com- mands. Indeed, they performed worse, and the difference was statistically significant at the .025 level. An examination of the types of errors the children made on under in the unbiased condition revealed that twenty percent of the errors involved the object "paper" in N2 position. The children igut objects on the paper rather than under it, suggesting that paper has on properties in the same way as Clark's table (1973b). Thus, the apparent reversal in the error differential between improbable and

- 2 5 1 -

LANGUAGE SCIENCES, VOLUME 2, NUMBER 2 (1980)

Table 1

in

on

under

Percentage of Errors for All Children on Improbable and Unbiased Conditions for in, on and under

Improbable Unbiased

60 (72/120) 31 (37/120)

33 (40/120) 21 (25/120)

35 (42/120) 47 (56/120)

NOTE: Numbers in parentheses are number incorrect out of a total of 120 (6 items for each of 20 children).

unbiased items on under may have been due to cognitive bias on what were originally thought to be unbiased items. The cognitive bias led the children to an on for under interpretation for these items and this led to the increase in errors for the unbiased condition.

Of the total number of errors on the improbable condition, 67% were in keeping with children's cognitive bias. For example, children put babies in cribs and bathtubs instead of under them as requested in the command. This high degree of predictability was true for all three preposi- tions and is comparable to the 70% of predictable incorrect responses for hnprobable commands in Clark's study (1973b).

The particular patterns of errors for individual items of the test provide information about the processes involved in the cognitively biased inter- pretation of utterances. That is, they offer evidence about whether chil- dren use a perceptually based interpretive strategy as advocated by Clark (1973a, b, and 1974), an action based one as suggested by Sinclair (1973), or a motor ease one advocated by Wilcox and Palermo (1975).

There were ten improbable commands to which at least five children gave identical incorrect answers. These commands and errors are shown in Table 2. Nine of the ten common errors shown in the table fit Clark's finding that mistakes children make depend upon whether they perceive a noun referent as a container or as a surface. That is, each of these com- mands designated an object which was either a container (cup, high chair, bathtub, convertible car) or surface (paper, blanket, table, chair,

- 252 -

COGNITIVE BIAS

Table 2

Common Errors made by Five or More Children

Command

1. Penny under cup

2. Highchair on baby

3. Baby under bathtub

4. Car on boy

5. Paper in scissors

6. Pencil in paper

7. Horse in blanket

8. Table on spoon

9. Baby under chair

10. Babyin rocking horse

(Horse had shelflike space on its side to place baby in for in response)

Error

Penny in cup

Baby in highchair

Baby in bathtub

Boy in car

Cup paper with scissors

Pencil on paper

Horse on blanket

Spoon on table

Baby on chair

Baby on rocking horse

Number of Children Making

Error

7

8

10

5

5

12

9

8

5

15

NOTES: The commands were given in full form to the children. The form was "Put the the _ _ ." The statement of error is a description of what the children did in response to the command.

rocking horse). The one error which does not lend itself to explanation by Clark's

rules is item 5 where children cut the paper with the scissors instead of putting the paper on the scissors. For this item, the perceptual strategy would predict that children would place the scissors on the paper since they placed the pencil on the paper in # 6 and even used paper as an on

surface for the two items on the unbiased condition. Three children did, in fact, put the scissors on the paper. Nonetheless, five children cut the paper with the scissors, a result which would argue for an action based

- 253 -

LANGUAGE SCIENCES, VOLUME 2, NUMBER 2 (1980)

origin for cognitive bias, such as that advocated by Sinclair (1973). Finally, six of the ten errors (see numbers 1, 3, 6, 7, 9, and 10 in

Table 2) were in the direction of easier motor responses. Children put objects in and on rather than under which would have required lifting the stationary object; they also put objects on rather than in which would have required manipulating an object such as wrapping a horse in a blanket or pencil in paper. The four exceptions to this motor ease tendency involved reversals or answers involving ongoing activity (cutting).

The patterns of errors for particular children were also examined. The children's overall scores ranged from 34 incorrect answers to 4 incorrect answers. This second error analysis revealed a clear relationship between error pattern and overall performance. The two worst performers tended to select the two referents and place them next to one another. This accounted for 19 of 34 errors of the worst, and I0 of 29 errors of the second worst. A better group of five children were able to perform cor- rectly on approximately half the items, with more errors made on the improbable than the unbiased condition (32 errors on unbiased; 45 on biased). Their errors on both unbiased and improbable conditions were frequently substitutions of on for in (21/29 errors or 72%) and somewhat less frequently were substitutions of on for under (10/32 or 31%). (On

errors for this analysis were defined to be not only cases where children put N1 on instead of in N2, but also in cases where they first selected N2 and proceeded to put it on N1).

A third group of children who performed in ways similar to one an- other were the ten best performers. They usually were correct on the unbiased condition. Out of a total of 60 items for each of the prepositions, they committed 5 errors on on , I0 on under , and 11 on in. Overall, they performed less well on the improbable condition with most of the dif- ference in performance for in (7/60 errors for on , 27/60 for in, and

5/60 for under). Their errors for the unbiased condition were scattered and apparently unpattemed with the exception of two of the ten chil- dren who tended to substitute on for in.

D I S C U S S I O N

The findings here for Noun-Prepos i t ion-Noun commands are com- patible with those of Strohner and Nelson (1974) for Noun-Verb-Noun

- 254 -

COGNITIVE BIAS

commands. In both this and Strohner and Nelson's study children were found, who cognitively interpreted improbable commands but were at the same time semantically interpreting unbiased commands.

While many of the children performed in this way, there were two children, the worst performers on the task, who were important excep- tions. These two children were able to identify the noun referents, but they did not place them in functional relationship with one another. Viewed from a Piagetian perspective, one might say that for these chil- dren the cognitive structures which underlie object manipulation in play are not active when the same objects are related through verbal com- mands. The children did not put the baby in or under the crib in answer to the verbal command, Put the baby under the crib, but instead placed the baby next to the crib. Perhaps, the children's conception of the task was to do no more than find each object referent, one relating to N1 (e.g., baby) and another to N2 (e.g., crib). This same "lexical additive" type of comprehension, where children treat nouns as separate entities and not in semantic relation to one another, was observed by Duchan (1972) for NVN utterances.

Wilcox and Palermo (1975) also discovered children responding to in, on and under commands by putting objects next to one another and ascribed the responses in part to the fact that file children did not know the meaning of the preposition. Their hypotheses are substantiated by these data in that the next to responders in this study performed no better on the unbiased than the improbable condition.

For those children who used a cognitively biased interpretation for constructing meaning for improbable utterances, patterns emerged which suggest that it is incorrect to use a single theory (e.g., a perceptual theory, action scheme theory, or motor ease theory) for explaining cognitive bias. Instead it appears that perceptual, action-based, and motor ease biases may be operative in children's early sentence interpretation. There were those interpretations which seemed to be almost idealized examples of Clark's perceptual strategy. Children put pencils on paper and horses on blankets even when those objects might have carried a relational history for them that would have led to other cognitive relationships (writing on the paper with a pencil, or wrapping the horse in the blanket). In these cases, cogni- tive bias seemed to reflect aClark-like perceptual strategy.

There were other interpretations which children made that are more

- 255 -

LANGUAGE SCIENCES, VOLUME 2, NUMBER 2 (1980)

readily explained by Sinclair's action-based theory. For example, children frequently cut paper with scissors rather than putting the paper on the scissors. Less frequently occurring but notable examples of an action- based strategy were seen when children put rubberbands a r o u n d the toy cupboard, not in it; wrote with the pencil instead of putting it o n t he

paper; and brushed the horse with the brush, instead of putting the horse u n d e r the brush.

Besides these apparent perceptual and action based biases, the motor ease bias was observed in the children who had a predilection for placing any two objects in an "on" relation to one another. They put the object they first picked up (referent to N1 or N2) o n the second,even when the second object (e.g., hat, shirt, flower) did not have an obvious flat support- ing surface or "on-like" property. It would be difficult to explain this behavior by Clark's perceptually based rules unless one assumed that these children regard all non-container objects as having surface or "on- like" properties. This seems an implausible explanation for putting a penny o n a flower, when the command called for in , or putting a table o n a shirt when the command called for in . Thus, the most plausible ex- planation is the one advocated by Wilcox and Palermo (1975). That is, there is a "tendency for the child to make the simplest and thus the easiest motor response" (Wilcox and Palermo 1975:251).

The easy motor explanation for the on bias may also be a way of explaining the children's n e x t to responses. However, this explanation does not account for why some children responded with o n and others with n e x t to .

The o n and n e x t to responders were different in that the children who showed on bias did well on the rest of the test, a fact which attests to their ability to form semantic relationships. That is, they were more linguistically advanced than the worst performers who responded by put- ting the noun referents n e x t to one another. The o n responders were also able to semantically interpret some of the locatives in consistent ways, as is indicated by their better performance on unbiased than improbable conditions. The differences in linguistic knowledge between the two sets of children, thus, appear to be knowledge of prepositions, as well as the fact that the better performers' o n answers derive from their ability to interpret NI and N2 as linguistic entities in semantic relationship with one another.

- 256 -

COGNITIVE BIAS

This study has shown the influence and possible sources of cognitive biases in children's early interpretation of locative commands. The in- fluence of the biases appears to be different for different children. While the study was not a developmental one, the results suggested levels of performance which might be equivalent to developmental levels. The results along with those in earlier studies suggest the following stage-like progression. First, the children, even before they can understand nouns, are probably relating objects via action schemes and perceptual configura- tions (Sinclair 1970; Clark 1973a). When they begin to understand lan- guage and begin to interpret utterances with more than one noun, they pay particular attention to nouns and may or may not assimilate these nouns into their existing perceptual configurations or action schemes. For stage 2, then, the children might be operating in a lexical-additive interpretive framework where they interpret the individual n o u n s as separate entities (e.g., when they place objects n e x t t o one another) or they may be operating in a cognitively biased interpretive framework where they assimilate the nouns to a cognitive structure and relate them in the same way they would relate those objects ordinarily in nonverbal contexts. At this stage, the prepositions are likely to be semantically empty, as is indicated by frequent errors on the unbiased commands. At a third stage, the children tend to form semantic relations between nouns even when they do not understand the prepositions (e.g., when they place N1 designated objects o n N2 designated objects). These rela- tions are expressed by placing objects in relation to one another using an uncomplicated motor activity, thereby manifesting a motor ease strategy described by Wilcox and Palermo (1975). The prepositions are beginning to be semantically interpreted at this stage.

At stage 4, the children better understand locative prepositions and respond by semantically interpreting the prepositions in more items. However, they use perceptual, action or motor ease biases to achieve contextual congruency on improbable or more difficult commands. At this stage, as well as stage 3, the children display both semantic and cognitively biased interpretations.

Finally, stage 5 children can interpret improbable locative commands correctly, perhaps checking with the person issuing the command to see if he]she is teasing. They are now comparable to adults in that they rely

- 2 5 7 -

LANGUAGE SCIENCES, VOLUME 2, NUMBER 2 (1980)

more heavily on semantic interpretations than contextually biased in- formation.

ACKNOWLEDGEMENT

The author wishes to thank Joann Kemist and Kathy Cullen who helped in the conceptualization of the study, Gayle Saffer who helped collect the data, and the teachers and children at the State University College day care center.

REFERENCES

Bever, Thomas 1970 "The Cognitive Basis of Linguistic Structure," in John Hayes

(ed.) Cognition and the Development of Language, New York: Wiley.

Clark, Eve 1972 "Some Perceptual Factors in the Acquisition of Locative

Terms by Young Children," Papers from the Eighth Regional Meeting, Chicago Linguistic Society.

1973a "What's in a Word? On the Child's Acquisition of Semantics in His First Language," in Thomas E. Moore (ed.) Cognitive Development and the Acquisition of Language, New York: Academic Press.

1973b "Non-linguistic Strategies and the Acquisition of Word Mean- ings," Cognition 2.161-82.

1974 "Some Aspects cf the Conceptual Basis for First Language Acquisition," in Richard Schiefelbusch and Lyle Lloyd (eds.), Language Perspectives-Acquisition, Retardation, and Inter- vention, Baltimore, Maryland: University Park Press.

Duchan, Judith 1972 "Three Stages in Children's Acquisition of Language," Un-

published ph. D. Dissertation, University of Illinois, Champ- aign, Illinois.

Sinclair, Hermine 1970 "The Transition from Sensory-motor Behavior to Symbolic

Activity," Interchange 1.119--26.

- 258 -

COGNITIVE BIAS

1973 "Language Acquisition and Cognitive Development," in Thomas E. Moore (ed.), Cognitive Development and the Acquisition of Language, New York: Academic Press.

Sinclair, Hermine and Bronckart, J. 1972 "SVO-a Linguistic Universal," Journal of Experimental

Child Psychology 14.329-48. Strohner, H. and Nelson, K. E.

1974 "The Young Child's Development of Sentence Comprehen- sion: Influence of Event Probability, Nonverbal Context, Syntactic Form, and Strategies," Child Development 45.567-- 76.

Wilcox, S. and Palermo, David 1975 "In, on, and under revisited," Cognition, 3.245-54.

- 259 -