Cognition, Vol. 2, No. 2

1

Non-linguistic

strategies and the acquisition of word meanings*

EVE ‘J. CLARK Stanford University

Abstract The present study proposes that children’s apparent comprehension of certain words is atjrst dependent on a combination of their linguistic hypotheses about a word’s meaning and certain non-linguistic strategies. Children aged 1;6-5;O were given instructions requiring comprehension of the locative terms in, on and under. The results showed that children go through three stages: At first, they consistently use certain non-linguistic strategies that can be characterized by two ordered rules; next, they apply these rules to only one or two of the locative instructions; and$nally, they exhibit full semantic knowledge of the three word meanings. Because of these non-linguistic strategies, the younger children always appear to understand in correctly, sometimes appear to understand on and never understand under. It is argued, nevertheless, that these non-linguistic strategies determine the order of acquisition of the three locative terms. Recent work on the acquisition of word meanings suggests that children begin by using certain non-linguistic strategies based on their perceptual knowledge. Indeed, they have obvious recourse to perceptual information in the numerous over-extensions that have been observed during the first year or so of language use (between about 1; 0 and 2; 6). For example, the earliest meaning ascribed to a word like apple is often something like ‘[SMALL X] & [ROUND Xl’. The child’s interpretations of these percept-based features, size and shape, are then used criterially in applying the word * This research was supported in part by the National Science Foundation, GS-30040. Deborah Rosenblatt assisted in running the first two experiments, and Carol B. Farwell helped with Experiment 3. I am grateful to the stat7 of Bing Nursery School, the Stanford Child Care Center and the parents of the children for all their cooperation. I would like

to thank Herbert H. Clark for making detailed comments on the manuscript at various stages, and lastly I wish to acknowledge an anonymous reviewer whose remarks substantially improved this paper. Requests for reprints should be sent to the author at Committee on Linguistics, Stanford University, Stanford, California 94305, U.S.A. Cognition Z(2), pp. 161-182

162

Eve. V. Clark

apple to other

objects such as door-knobs, rubber balls, round light switches and paperweights. These over-extensions, characteristically based on features of shape, size, movement, sound, texture and taste, are well documented in diary studies of the acquisition of a large number of different languages (E. Clark, 1973b). These over-extensions might be best explained as the outcome of the young child’s hypotheses about word meaning. His hypotheses would seem to take the general form of ‘A word refers to some identifiable [perceptual] attribute of the object pointed to’. Thus, the child is relying on the perceptual knowledge that he already has of the things around him when he begins to form hypotheses about the meanings of new words. These linguistic hypotheses then lead him to act on the assumption that the feature or attribute he has picked out is what that particular word designates. Therefore, any other object that matches those features can be ‘named’ by the same word (E. Clark, 1974). It is therefore the child’s interpretations of his percepts, his knowledge about the properties of objects, that provide a basis for some linguistic hypotheses for assigning meaning to ‘new’ words. Children, however, also show certain biases in their treatment of the world around them that bear no direct relation to their linguistic hypotheses. For example, if a young child is shown a piece of chocolate and a pebble and is allowed to choose one, he will probably always choose the chocolate. This choice would then be independent of the meanings of words used in the instructions to choose; his choice is the outcome of a non-linguistic strategy. This form of behavior is traditionally referred to as a preference or a response bias and may be present in the child (or the adult) for a variety of reasons. If the child relies consistently on such preferences, though, it is important to identify them. This is because they could make it appear that the child had understood something when, in fact, his response was simply due to a nonlinguistic strategy. This question is a particularly important one where children’s comprehension of word meanings is concerned. The child might appear to have grasped the adult meaning of some complex word when he was actually only responding on the basis of a non-linguistic strategy. There are therefore two ways of looking at many comprehension studies. First of all, the child’s responses, including his errors, could be treated as if they were the outcome of his linguistic hypotheses about the meanings of particular words. Secondly, the child’s responses could be regarded as the outcome of some non-linguistic strategy. However, these two approaches are often impossible to separate. For example, there have been several studies of the relational terms more and less in which children appeared to interpret the word less as if it meant more. For instance, whenever children were asked to choose from two trees either the one with more apples or the one with less, they always chose the one with the greater number (Donaldson and Balfour, 1968; Donaldson and Wales, 1970; Palermo, 1973). There are two possible explana-

Non-linguistic

strategies

and meaning

I63

tions that might be offered for these data, explanations that have not been clearly distinguished in the discussions of what less actually means to the child. The first explanation assumes that the child is relying on his linguistic hypothesis about the meanings of more and less, namely that both words refer to amount or quantity [ +Amount], and to the positive pole, i.e., the greater of two or more amounts [ +Polar]. This is the correct meaning for more, but not for less. This analysis will be called the full semantics hypothesis. An alternative explanation for the same data makes a weaker assumption about the child’s meanings for the two words. Instead of assuming that more and less are synonymous, both with the meaning of more, one could suppose that both word meanings are actually incomplete and that the child’s responses are based on the partial meaning that he has for both more and less, in combination with certain nonlinguistic strategies. Thus, the child might know only that both more and less refer to amount [+Amount]. This tends to be supported by the children who, when asked to show which tree had more apples on it, replied Both of them, That one does an’ that one, Both the trees, They two ones, Each tree, and so on. Other children, asked to make the amount less on one tree (the one with more) objected: But it is less on that tree (Donaldson and Wales, 1970, p. 248). In addition to this partial semantic knowledge, the child at this stage would be assumed to have a non-linguistic strategy of usually choosing the greater of two or more amounts or the more extended object on a dimension such as length or height. H. Clark (1970) appealed implicitly to such a notion in suggesting that the first stage in the acquisition of more and less involves only a nominal sense of the words, i.e., both words refer only to amount, and that one had to assume that the best exemplars of amount for the young child are those objects with greater amount. This assumption has since been tested experimentally by Klatzky, Clark and Macken (1973). They showed that children found it significantly easier to learn the meanings of nonsense syllables that referred to relatively greater extent along several dimensions than to learn their polar opposites. Thus, in the case of more and less, the child would know that both words meant [+Amount], but would not yet have learnt the feature [&Polar]. It only appears that he has [t-Polar as well as [+Amount] because of the strategy of choosing the greater amount. This explanation, based on the combination of partial semantic knowledge and a nonlinguistic strategy, will be called the partial semantics hypothesis. The full semantics hypothesis and the partial semantics hypothesis should be distinguishable if one could find a situation in which the child treated WORD, as if it meant the same as WORD2 in some contexts, while at the same time treating WORD1 as if it meant WORD, in another set of contexts. One would thus be able to demonstrate that the two incompatible senses consistently given to WORD1 were not random but, in fact, could be ascribed to the child’s use of certain non-linguistic

164

Eve. V. Clark

strategies. This has not actually been shown in the case of more and less, although it has been shown, in effect, by the over-extended uses of early words like apple (E. Clark, 1973b). If the child’s responses to any particular words can be shown to be the outcome of a little semantic knowledge (if that) combined with certain non-linguistic strategies, then it would be clear that the full semantics hypothesis would make too strong a claim about the extent of the child’s semantic knowledge. The partial semantics hypothesis, in contrast, is concerned with precisely that interaction, the way in which the child relies on non-linguistic strategies prior to, and during, the acquisition of word meanings. These two hypotheses are not true or false in general, but they may be contrasted for any particular word or set of words. In the present paper, I shall present some comprehension studies of locative prepositions in an attempt to explore one set of strategies and their relationship to the young child’s semantic knowledge. During the acquisition of the terms in, on and under, children appear to treat the word under as if it means in in some contexts, but as if it means OIZin others. This suggests that this is a case where the partial semantics hypothesis, with its non-linguistic strategies, might provide a better account of the data than an explanation based on the full semantics hypothesis.

Experiment

1

Method Each child was asked to follow instructions containing the words in, on and under in contexts that allowed both correct and incorrect responses. The instructions were of the form Put the x in [on, under] the y, where x was one of eight small toy animals, and y one of six reference point objects (henceforth RPs). The child’s task was to correctly locate one object, X, with respect to the RP, y.l The six RPs, shown in Figure 1, were chosen because each allowed two of the three spatial relations being named in the instructions. Thus, the box on its side and the tunnel each allowed an object to be placed either in or on the RP. The dump truck and the crib each allowed either in or under, and the table and the bridge each allowed either on or under. There were 24 instructions in all, with eight for each of the three prepositions in, on and under. Each preposition occurred twice with each of the four RPs that allowed 1. The children in these experiments always picked up X, rather than y, and then placed it in relation to y. Thus, the RPs were always treated as reference points. Other research has shown that young children find it impossible to

place B when told ‘A is on top of B’; they can always do the task, though, if told to place A (Bern, 1970; Huttenlocher, Eisenberg and Strauss, 1968; Huttenlocher and Strauss, 1968).

Non-linguistic

Figure

1.

strategies

and meaning

165

The reference point objects (RPs) used in Experiment I: Items I and 2 allow the relations IN and ON; items 3 and 4 allow the relations IN and UNDER; items 5 and 6 allow the relations ON and UNDER

the relation, e.g., in occurred twice each with the box, the tunnel, the truck and the crib. The names of the eight animals each occurred three times in all, once each with each preposition. The instructions were divided into two identical blocks of 12, and the order of instructions within each block was randomized separately for each subject. The subjects were 70 children (31 male and 39 female) taking part in a larger study of the acquisition of spatial terms in English. All the children were native speakers of English and had no contact with other languages in the home. The older children (2; 6-5;O) were attending Bing Nursery School, Stanford; the younger ones (I ; 6-2; 5) were contacted through Bing Nursery School and the Stanford Child Care Center. The children were divided into seven age-groups at six-monthly intervals, with ten childrenin each group: 1. 1;6-I;11 (mean age 1;9); II. 2;0-2;5 (mean 2;3); III. 2;62;ll (mean 2;9); IV. 3;0-3;5 (mean 3;3); V. 3;6-3;ll (mean 3;9); VI. 4;0-4;5 (mean 4;2); VII. 4;6-4; 11 (mean 4;8). Each subject was interviewed separately in a small experimental room at the nursery school. (Five subjects from Group II were seen under comparable circumstances at the child care center.) The children in the two youngest groups (aged 1;6-2; 5) generally had one parent present throughout, seated slightly behind the child and out of his line of sight. The parent made no comments during the experimental trials. Each child was first asked to name each of the RPs and the animals. The child’s own names for the objects were used throughout the session wherever these differed from the ones originally assigned. Then the experimenter gave the child the two blocks of

166

Eve. V. Clark

12 instructions each. Some of the younger subjects had a short break between the two blocks; all completed the task, with no responses omitted, in a single session that lasted between ten and 20 minutes.

Results

The data will be analyzed first in terms of how the children carried out the instructions they were given where a correct response is a response that appears to reflect adultlike knowledge of the word meaning in question. Following this, the kind of errors made will be used as the basis for analyzing the data in terms of possible non-linguistic strategies where a ‘correct’ response is defined as consistent, predictable use of a particular strategy. For the first analysis, the percentage of correct responses to each instruction is shown for all seven age groups in Table 1. Because of the low error rate and absence of any significant differences between age groups for subjects over 3 ;O years (Groups IV-VII), an analysis of variance was performed only on the data from the three youngest groups. These three groups differed from each other significantly overall, F (2,27)= 16.00, p < 0.001. The interaction between age groups and instructions, F (4,54)= 12.66, p
Instruction IN ON UNDER Mean: a.

Age group IV

v

VI

VII

96 80 98

100 96 89

100 94 100

100 100 100

100 100 100

91

95

98

100

100

I

11

111

94 61 4

98 72 57

53

76

:

Each percentage based on 80 data points.

The three instructions, in, on and under, also differed significantly overall, F(2,54) = 30.45, p < 0.001. Within Group I (1; 6-l ; 1l), instructions containing in produced significantly more correct responses than those containing on, t(54) = 3.50, p < 0.005, and instructions containing oiz produced significantly more correct responses than under, t(54) = 5.93, p (0.001. Within Group II (2;0-2;5), in was still significantly better than on, t(54) = 2.43, p < 0.025, and than under, t(54) = 4.04, p < 0.001. But the difference between on and under was no longer significant. Within Group III

Non- linguistic strategies and meaning

(2 ; 6-2 ; 1l), there were no significant

differences

167

between the three instructions.

A further analysis was carried out to see if the RPs had any effect on the responses given. For each instruction, there were two pairs of RPs that differed in their properties. For example, with in, one pair of RPs allowed on as the alternative relation (box, tunnel) and the other allowed under as the alternative relation (truck, crib). For the in instructions as a whole, those used with the truck and the crib were significantly easier than those with the box and the tunnel, 1(54) = 2.19, p < 0.05. The on instructions were also easier where the alternative relation was under (bridge, table) rather than in (box, tunnel), t(54) = 9.29, p < 0.001. For the under instructions, there were slightly fewer errors with the bridge and the table than with the truck and the crib (44% versus 50 %), but this difference was not significant. These data are shown in Table 2. A correct response therefore seems to depend in part on the particular properties of the RP. Table 2.

Percentage of correct responses to each instruction for each RP by age group’ Instruction On

In Age group:

Under

I

II

III

1

II

III

I

II

III

80 95 100 100 -

100 90 100 100 -

85 100 100 100 -

50 1.5 80 100

50 50 95 95

75 60 90 95

-

-

-

RP: bOX

tunnel truck crib bridge table

0

0 10 5

55 45 70 60

100 100 90 100

a. Each percentage based on 20 data points.

The errors made reveal a very consistent pattern. If the kind (i.e., box, tunnel, truck, crib), the younger children the instructions contained in. Whenever the RP was not porting surface (i.e., bridge, table), these children tended as though they contained on. The error data can therefore where correct responses are shown in (a) and errors in (b) (1) RP allowing in or under [truck, crib] (2) RP allowing in or on [box, tunnel] (3) RP allowing on or under [bridge, table]

(a) (b) (a) (b) (a) (b)

instruction instruction instruction instruction instruction instruction

RP was a container of any tended to act as though all a container but had a supto treat all the instructions be characterized as below, for each type of RP :

in elicits ‘in’ under elicits ‘in’ in elicits ‘in’ on elicits ‘in’ on elicits ‘on’ under elicits ‘on’

168

Eve. V. Clark

Thus, in appears nearly always to be interpreted correctly, and OIZis also interpreted correctly unless the RP is a container. Under is never correct because the options are always either ‘in’ or ‘on’. While this pattern of error could simply reflect the order of acquisition of the three prepositions, according to some measure of complexity, it could also be the result of applying two simple, ordered rules based on the child’s non-linguistic knowledge of the usual or expected spatial relationship between x and the RP: RULE I : If the RP is a container, x is inside it RULE 2: If the RP has a horizontal surface, x is on it These two rules are strictly ordered in the sense that Rule 1 is always applied. In the event that it fails, Rule 2 is applied next. The use of Rule 1 predicts the error of on going to ‘in’, and of under going to ‘in’, while Rule 2 predicts the error of under going to ‘on’. The use of two such ordered rules accounts for 92 % of the errors made by the youngest group (1; 6-l ; 11) and for 91% of the errors made by Group II. The error rate for Group III is less than 10% overall, but the two rules still account for 71% of these errors. The data presented in Table 1 above were scored with the assumption that the child had some linguistic hypotheses about, and even adult-like knowledge of, the meanings of in, on and under. Thus, a response appropriate to the instruction was assumed to indicate knowledge of the word meaning. This knowledge will be referred to as the adult schema. The percentage of responses accounted for by such semantic knowledge is shown by the mean correct for each age group. In the youngest group, the adult schema accounts for only 53 % of the responses given. Since each RP offers a choice of two locative relations, e.g., in and on with the box, there is a 50 % probability that the child could pick the ‘semantically correct’ relation by chance. Thus, the data from Group 1 appear to indicate almost total absence of adult-like semantic knowledge. In contrast, if the child does know the meanings of all three prepositions, the adult schema should account for 100 % of the responses given, as indeed it does for the older groups shown in Table 1. The means in Table 1 indicate, therefore, that the percentage of responses accounted for by the adult schema (semantic knowledge) increases with age from a chance 50 o/oto 100 o/0of the responses given. However, the responses scored as errors according to the adult schema criteria suggest that the child actually begins by using non-linguistic strategies like Rules 1 and 2, and only relinquishes these strategies when he learns what the words in, on and under mean. Responses based on the use of the two ordered rules (I and 2) will be called the child schema. The percentage of data accounted for in each age group by use of the child schema is shown in Table 3. In the case of in, the use of Rule 1 cannot be distinguished from use of semantic knowledge (the adult schema). With on,

Non-linguistic strategies and meaning

169

however, the use of the child schema can be differentiated from that of the adult schema in those instances where on contrasts with in, i.e., with the box and the tunnel. This is because Rule 1 will always apply in these instances, giving the response ‘in’, until the adult schema takes over and Rule 1 ceases to be applied. Where on contrasts with under, there is again no way to distinguish between use of Rule 2 in the child schema and semantic knowledge in the adult schema. Thus, the appearance of the adult schema with on will signal a drop to 50% of the on responses accounted for by the child schema. In the case of under, the child schema should account for all the responses until the child learns what under means. Once the adult schema takes over, the child schema will be unable to account for any of the responses to under (see Table 3). The mean percentage of responses accounted for overall by the child schema in each age group should therefore decrease from 100% at the earliest stage (Group I and younger) to 50% as the child acquires the meanings of in, on and under. The means of the data accounted for by the child schema appear in Table 3. The decrease between Groups II and III, down to 54 %, coincides with a large increase in the data accounted for by the adult schema, up to 91% (Table 1). There is therefore a cross-over between the child schema and the adult schema in terms of the amount of data accounted for at different stages. Table 3.

Percentage of responses accounted for by use of ordered Rules I and 2 by instruction and age group”

Instruction IN ON UNDER Mean

:

I

II

III

94 79 96

95 72 43

89

69

Age group IV

v

VI

VII

96 62 3

100 54 11

100 54 0

100 50 0

100 50 0

54

55

51

50

50

:

a. Each percentage based on 80 data points.

It is evident, therefore, that the data presented in Table 1 for Groups I and II are not a direct reflection of how well children of this age ‘understand’ the spatial prepositions in, on and under. Instead, they appear to reflect the children’s use of prior non-linguistic strategies (the child schema) that are used in the virtual absence of comprehension.

170

Eve. V. Clark

Experiment 2 The next experiment was designed to provide independent evidence for the nonlinguistic strategies that seemed to be being used by the younger children in Experiment 1. If these particular strategies - the child schema - are in fact independent of the linguistic input using locative words (the instructions containing in, on and under), they should also appear in situations where the young child hears no locative words in the task. It is necessary to show that these non-linguistic strategies are really present in the child because there is practically no data available on the kinds of non-linguistic strategies that the young child might have in his repertoire to rely on prior to, and during, the process of acquiring language. Children were therefore asked to copy a configuration of x and y modelled by the experimenter, a procedure adapted from Laurendeau and Pinard (1970). It was hypothesized that if a young child was using a non-linguistic strategy such as that described as the child schema, he should ‘correctly’ copy any configuration that conformed to Rules 1 and 2. However, those configurations not conforming to the child schema should produce predictable errors in the form of a relationship that does conform to the two rules.

Method Both the experimenter and the child, seated side by side, began with one each of a pair of RPs in front of them on the table. The experimenter modelled a configuration by placing x in a particular spatial relationship to his own RP, then handed the child an identical x and asked him to do the same or to make it like the experimenter’s. The child received no other verbal instructions and the spatial relationship was never named. Each child was presented with ten different configurations (shown in Table 4) of an RP together with a small animal, x. The experimenter’s and the child’s RP and x were always identical. Six of the ten configurations in Table 4 conformed to the rules given in the child schema, and four did not. If the children used Rules 1 and 2, then they should make errors on the four starred items. The subjects were the 20 children from Groups I and II of Experiment 1. The second experiment was run within three weeks of the first, and, as in the first, one parent generally accompanied the younger children into the experimental room. Each child received the set of ten configurations in a different random order and took between five and ten minutes to complete the task of copying.

Non-linguistic strategies and meaning

Table

4.

171

Co/$gurations used in Experiment 2 Applicable strategy”

Configuration

I. x on upside-down glass 2. x in upright glass 3. x beside upright glass, with no contact 4. x on block 5. x beside block, with no contact 6. x on bridge 7. x under bridge 8. x beside bridge, with no contact 9. x in crib 10. x under crib a. Prediction of which rule will be applied in making response; starred items are predicted to produce errors because of the use of either Rule 1 or Rule 2 where neither actually should be applied. b. This is assuming that the bridge space may be treated as a container since there is contact

Rule Rule *Rule Rule *Rule Rule Rule *Rule Rule *Rule

2 1

I 2 2 2 lb 2’ 1

1

and an enclosure-like space involved. It is possible that this should also apply in the case of item 8. c. This prediction is based on the fact that the bridge has a surface and no real containerspace. However, it is possible that Rule 1 may apply here, as with item 7.

Results Fifteen of the 20 subjects in this experiment made a total of 40 errors. Of these, 38 were predicted errors, occurring on items 3, 5, 8 and 10 (starred in Table 4), and only two were not predicted. Since six of the ten items in Table 4 conformed to the child schema, and only four did not, the a priori probability of making a predicted error was lower than that of making an unpredicted error. Yet 13 of the 15 subjects making an error made only predicted errors, while the remaining two subjects, with one unpredicted error each, made more predicted than unpredicted errors (p < 0.001, by Sign Test). Overall, the percentages of errors accounted for by the four starred items in Table 4 was 93 % for Group 1 (because of the two unpredicted errors), and 100% for Group II. In other words, whenever the experimenter’s configuration was one that could be achieved simply through use of the child schema (ordered Rules 1 and 2), the child made no errors. Nonetheless, it is clear that the task, on the whole, was treated as a copying task because reliance on the child schema alone would have produced a total of 80 errors on the four starred items. The actual error total therefore represents an error rate of 48 %, not one of 100 %. The predicted errors on the four items not conforming to the rules are shown in Table 5. In general, the particular errors made can be attributed to the use of Rules 1

172

Eve. V. Clark

and 2. Rule 1 would account for the errors in (i) and (iv) in Table 5, while Rule 2 would account for those in (ii) and (iiia). However, eight of the children making an error on Item 8 [shown in (iii) of Table 51 placed x underneath the bridge rather than on it. One possible explanation for this is that the children were treating the hollow rounded space below the lower surface of the bridge as a container (see Figure 1). If this were the case, the errors in (iiib) could be attributed to use of Rule 1. This proposal is backed to some extent by several children in Group II who spontanously described placements under the bridge as ‘in here’. However, there were no errors of this kind on item 6 (Table 4) where the model configuration has x on the bridge. Nor were there any responses of this kind from the two younger groups in Experiment 1.

Errors made on critical configurations in Experiment 2

Table 5.

Item

Experimenter’s model

Subject’s erroneous copy

Number of Instances

(0

3

x beside upright glass with no contact

x inside glass

14

(ii)

5

x beside block with no contact

x on block

I

(iii)

8

x beside bridge with no contact

(a) x on bridge (b) x under bridge

2 8

(iv)

10

x under crib

x in crib

7

There were only two unpredicted errors, made by two subjects in the youngest group. Both errors occurred on item 1 (x on upside-down glass), and both consisted of the child’s turning up the glass and placing x inside it. One of the children also turned up the experimenter’s glass and did the same thing with it. A similar manipulation of the experimenter’s RP also occurred with item 3 where four children who made the predicted error of placing x in the glass then proceeded to ‘correct’ the experimenter’s configuration also. Again, these four children all came from the youngest group (1;6-1; 11). The manipulation of the glass so its opening was upwards, i.e., in the normal orientation for a glass, is similar to the manipulation of the box on its side by several of the younger children in Experiment 1; it, too, was consistently turned so that its opening was upwards. This suggests that young children may have certain expectations about the proper orientation of objects based on their experience of what the usual orientation is. This point will be taken up further in the final discussion. Lastly, among the responses that were counted as correct copies of the configurations in items 3, 5 and 8, there were a number that maintained the spatial relation between

Non- linguistic strategies

and meaning

173

x and the RP in all respects save one: x was juxtaposed to the RP so that there was contact between them. This occurred five times with item 3 (X placed against the side of the glass), eight times with item 5 (X placed against the side of the block), and ten times with item 8 (X placed against the side of the bridge). Juxtaposition with contact, therefore, also seemed to play a role in making locative placements. This use of juxtaposition suggests that contact may be necessary for the child to attribute a relation in space to the pair formed by x and the RP. The results of Experiment 2, therefore, provide strong independent evidence for the presence of the non-linguistic strategies - the child schema - that appeared to be being used in Experiment 1. Young children rely on the same non-linguistic strategies in a copying task as they do in a comprehension task using the words in, on and under.

Experiment 3 Semantic knowledge (the adult schema) and non-linguistic strategies (the child schema) may coincide in the kind of response given by the child. In the case of the word in, it may be impossible to find out which basis the young child is relying on for his response. However, both on and under in Experiment 1 allowed some separation of the child schema and adult schema (Tables 1 and 3). The present experiment uses a somewhat different technique from that of Experiment 1 in order to look at the transition from use of the child schema to use of the adult schema with different materials. In a pilot study, children were asked to manipulate an RP so that it would conform to the instruction given. For example, if the RP was an upright glass but the instruction contained the word on, the child was to turn the glass upside-down so as to be able to place x on a surface. This task proved to be very difficult, even for considerably older children, so a different technique was devised in which the child had to choose one from a pair of RPs in placing x.

Method Each child was presented with a pair of RPs, identical in all respects save orientation: One had its opening upwards, and the other its opening face down on the table. With them, the child was given an instruction of the form Put the x in [on, under] a y. The pairs of RPs used were two opaque plastic tubs 3.7 cm high and 6.2 cm in diameter, two transparent plastic glasses 7.4 cm high and 6.2 cm in diameter at the rim, and two opaque plastic boxes 7.4 cm on a side, with no lids. The objects to be placed were six toys small enough to fit inside the RPs. The RPs were always presented

174

Eve. V. Clark

in pairs, with left-right positions counterbalanced with opening-up versus openingdown orientations. Each of the three prepositions, in, on and under, was used six times, twice each with each pair of RPs, for a total of 18 instructions. The instructions were presented in two identical blocks of nine each; within each block the instructions were randomized separately for each subject. The subjects were 20 children (6 male and 14 female) in two age groups, corresponding as nearly as possible to Groups II and III from Experiment 1: A. 1; lo-2;2 (mean age 2; 1) B. 2;6-3; 1 (mean age 2; 10) All the children were native speakers of English and had no exposure to any other languages in the home. Each child was interviewed separately at Bing Nursery School as in the previous experiments. The child was told he was to choose the correct RP of the two in front of him after he heard each instruction. He was also told he could move the RPs if he wanted to and was shown various manipulations by the experimenter. If the child completed the first block of instructions with no sign of fatigue, he was given the second immediately. Otherwise, he received the second block the following day. (Only three children from Group A required two visits.) The task as a whole took between 10 and 15 minutes to complete.

Results The percentage of correct responses according to the adult schema for each of the three instructions is shown in Table 6. The older group, B, made significantly more correct responses overall than the younger group, F (1,18) = 10.19, p < 0.001. There was also a significant difference between the three instructions, F (2,36) = 10.08, p < 0.001; those instructions containing in elicited significantly more correct responses than either on [t (36) = 2.92, p < 0.011 or under [t (36) = 4.41, p < 0.011. The difference between on and under failed to reach significance. When the two groups are considered separately, the younger children, Group A, produced significantly more correct responses to in than to on [t (36)= 2.75, p
comparable to the original subjects Groups 11 and 111in Experiment 1.

from

Non-linguistic

strategies

and meaning

175

Errors in this analysis consisted of choosing the (upright) container RP with on or under, or of choosing the uppermost surface of the upside-down RP with in or under. Because of the nature of the RPs,~ under was necessarily more difficult in this experiment for the younger children because a correct response always involved some manipulation. Whichever RP was chosen had to be tilted or lifted up in order to place x underneath. In other words, with manipulation, either RP could be correct; without manipulation, both were wrong. There was a significant decrease in the number of errors from Group A to Group B with both OIE[t (36) = 2.99, p < 0.011 and under [t (36)= 3.59, p
Table 6.

Percentage of correct responses by instruction with choice of an appropriate and inappropriate RP for each age group (adult schema) Instruction Group

Mean Age

In

On

Under”

Mean

A B

2:l 2;lO

87 100

43 77

23 63

51 80

a. Each percentage based on 60 data points. b. Either RP was an inappropriate choice for under unless the child manipulated it so as to

insert x underneath, was appropriate.

in which case either RP

Table 7 presents the data scored according to use of the non-linguistic strategies, represented by ordered Rules 1 and 2 of the child schema. In the case of in, the child schema accounts for the same amount of data as the adult schema. It is never possible, therefore, in this type of experiment to separate the responses due to the child schema from those due to the adult schema for instructions with in. The data elicited by on, in contrast, can be differentially accounted for by use of the child schema versus the adult schema. Since the child is always presented with two RPs, one with opening up

3. A protracted attempt was made to find pairs of RPs that could be inverted that would allow on/under and in/under as well as in/on as possible relations, without any direct manipula-

tion being necessary. Since we were unable to find any suitable objects, the present RPs were chosen as a compromise.

176

Eve. V. Clark

Table 7.

Percentage of responses accountedfor (ordered Rules 1 and 2)

Instructions On

Group

MeaIl Age

In

A B

2;l 2;lO

81 100

a. Each percentage

in Experiment 3 by the child schema

5-l 23

Under

Mean

48 17

64 47

based on 60 data points.

and one with supporting surface up, the situation is unlike that of Experiment 1. Here the child has to choose between two RPs. Therefore, if he is using the child schema, he should always chose the RP with its opening up, i.e., the container, since Rules 1 and 2 are strictly ordered. As a result, the child schema should account for 100% of the responses to begin with and later decrease to 0%. However, many of the children in Group A had evidently already learnt something about the meaning of on (the adult schema) because the child schema only accounts for 57% of the data. For Group B, this percentage has decreased still further to 23 % (Table 7). The same form of argument holds in the case of under. Use of the child schema with pairs of RPs should again account for 100 % of the data initially, with the percentage decreasing to zero as the child acquires the meaning of the word. For Group A, the child schema accounted for 48% of the data on under, while for Group B, it accounted for only 17%. Overall, the child schema accounts for slightly more data than the adult schema for Group A (64% versus 51 ‘Z,), but by Group B, the child schema accounts for considerably less than the adult schema (47% versus 80%) (see Tables 6 and 7). By the time children know the meanings of all three words perfectly, the maximum amount of data that the child schema could account for with these materials would be 33 ‘%; this is because the child and adult schemas cannot be separated in the responses for in, although they can be completely separated for both on and under. These data provide still further support for the existence of the non-linguistic strategies embodied in Rules 1 and 2. The results also provide additional evidence that the transition from use of the child schema to use of the adult schema is fairly near completion for all three locative terms, in, on and under, at age 2; 6 to 3 ;O.

Discussion The children in these three experiments linguistic strategies prior to, and during,

appeared to rely heavily on certain nonthe acquisition of the locatives in, on and

Non-linguistic strategies and meaning

177

under. The children can best be viewed as progressing through three stages: A stage where they rely on non-linguistic strategies plus partial semantic knowledge of the words; a transition stage; and a stage where they rely on full semantic knowledge of the words. According to this view, the children at the first stage (Group I in Experiment 1) do not understand in, on and under at all, except perhaps that they are locative. Instead, they base their responses to in, on and under on a non-linguistic strategy characterized by two ordered rules: RULE I: If the RP is a container,

x is inside it RULE 2: If the RP has a supporting surface, x is on it So when presented with RPs that are containers, the children always treat in as if they understood it correctly, but they treat on and under as if they meant the same as in. When presented with RPs that have supporting surfaces but are not containers, the children always treat on as if they understood it correctly but treat under as if it meant the same as OIZ.Thus, the children at this stage appear to use in correctly all the time, on correctly some of the time and under correctly none of the time, even though they understand all three words identically. The children do, however, appear to know at least that in, on and under refer to spatial location [+Locative], rather than to some other relation. All the children in Experiments 1 and 3, even the youngest, consistently constructed locative relations, placing x at least in contact with the RP, y. None of the children, for example, treated the relation as an agentive one and made x push y, or move y in any way. These children apply the same two rules in copying tasks in which no locative terms are mentioned at all. The data from Experiment 2, then, constitute independent evidence for the existence of the non-linguistic strategies proposed to account for the data from Experiments 1 and 3. The children in the transition stage to full semantic knowledge (Groups 11 and III in Experiment 1 and Groups A and B in Experiment 3) show partial, but not complete, competence with the adult meanings of in, on and under. In the present data, the children gave semantically appropriate responses to most of the on instructions and up to half of the under instructions. Just as for the children at stage one, these children were always correct for in instructions, but it is not possible to separate responses to in that were based on Rule 1 from those that were based on semantic knowledge. However, there is some other evidence that suggests that these children may already know the full meaning of in by this stage. Brown and his colleagues have observed that when children begin to produce their first two prepositions, in and on, they always appear to use the two terms correctly (Brown, 1973; Brown, Cazden and Bellugi, 1969). These two prepositions are the first to be used spontaneously and first appear sometime after the age of 2;0. In the present study also, several of the older children

178

Eve. V. Clark

spontaneously used the preposition in to describe (correctly) some of the placements in containers in the copying task (Group II from Experiment 2). These data, and in particular the data on under, therefore, support the partial semantics hypothesis over the full semantics hypothesis for the child’s earliest responses to in, on and under. At first, the young child does not know anything about the meaning of these words, except, perhaps, that they refer to location. The child’s responses are the result of his dependence on certain non-linguistic strategies - Rules 1 and 2 which operate both in comprehension tasks (Experiments 1 and 3) and in tasks involving no locative instructions (Experiment 2). So when the child treats on as if it meant in, he is doing so not because he thinks on means [$-Locative] and [+Containment], but because he thinks on means [$-Locative] and because he is using a non-linguistic strategy in conjunction with this partial meaning. Instead of assuming that the younger children know that in, on and under refer to location, though, one could suppose that they do not pay any attention to the instructions. This would still be fully consistent with the use of non-linguistic strategies. If the child did not pay attention and did not have any non-linguistic strategies to fall back on, he should respond randomly, or even not respond at all. However, the data showed that the children were all responding in a very consistent manner even if they were not attending to the instructions. The main difficulties with this approach are that one still has to explain why children should apparently begin to attend after age 2;0, and why they consistently treated the relation between x and y as a locative one. The non-linguistic strategies themselves would presumably have to be derived from the child’s conceptual knowledge about the objects and events around him. This general knowledge is based on his interpretations of his percepts. While the particular strategies examined in the present study appear to depend on his knowledge of containers and supporting surfaces, the children also appeared to make use of their knowledge about the usual or canonical orientations of some objects (H. Clark, 1973). For example, many of the younger children manipulated the box so its opening faced upwards (Experiment 1); they often righted the upside-down glass in Experiment 2, and three children subsequently righted the experimenter’s as well; and they showed a general preference for putting objects in the crib rather than under it. (A few of the children in Group III explicitly objected to the under instruction although they then carried it out correctly.) Although there was not a sufficient range of RPs in these experiments to assess the extent to which the young child might depend on such knowledge, other investigators have also found some evidence of reliance on the usual or canonical orientations of objects. For example, Ghent (1960) found that three-year-olds had significantly greater difficulty in recognizing pictures of realistic objects when the pictures were rotated 90” left or right, or 180” from their normal upright position. The younger the children, the more dependent they were on the

Non- linguistic strategies and meaning

179

expected or familiar orientation for recognition. This dependence was no longer noticeable by age 5 ;O. If reliance on such non-linguistic strategies is widespread during the acquisition of language, then it is possible that they play an important role in determining the order of acquisition of certain linguistic distinctions. Slobin (1971) proposed that the main determinant of order of acquisition for linguistic (semantic) distinctions was their relative cognitive complexity, but it is not obvious how to measure the cognitive complexity of different linguistic forms. The child’s reliance on non-linguistic strategies, though, suggests one possible approach. These non-linguistic strategies may provide the basis for the child’s linguistic hypotheses about the meanings of words. If so, it becomes very important whether or not the responses based on strategies coincide with those based on semantic knowledge. One could argue that, wherever such responses coincide, the child actually has very little to learn, but that where they do not coincide, he has much more to learn before he will acquire that meaning. The distance between the two types of response could be treated as a measure of the cognitive complexity of the semantic distinctions involved in the meanings of the words in question. For example, in should be cognitively simpler than either on or under because it requires minimal adjustment of the child’s hypothesis about its meaning, where this hypothesis is derived from the non-linguistic strategy, namely, in this instance, Rule l.4 On should be more complex because the child has to learn first that only the second of the two ordered rules coincides with the meaning of the word. Under should be more complex still because neither Rule 1 nor Rule 2 will produce any responses that coincide with the actual meaning of the word. This approach would argue that the order of acquisition of these three locatives should be in first, then on, and lastly under. The data from the present study are all quite consistent with this proposed order of acquisition, but it is difficult to tell whether in, on and under are actually acquired in that order. This is because the present experiments do not fully separate the responses based on a non-linguistic strategy from those based on semantic knowledge for the words in and on. Theoretically, it is possible to separate these two accounts for the responses. For example, the young child could be shown an RP that had a supporting surface but was not a container and asked to put x in the RP. If the child balked at such an instruction, he could be credited with the knowledge that in means containment, and not support. Indeed, this technique was tried in a pilot study, but even many of the older children were unwilling not to comply, and they always tried to place x somewhere, even when they showed full semantic knowledge in a task like

4. The use of in where there are no obvious containers will eventually lead to some re-

adjustment here also, e.g., The man in the photograph, The ball in the middle.

180

Eve. V. Clark

Experiment 3. Such judgments of inappropriateness are extremely hard to elicit from children under age 3 ;0.5 It has been argued that children begin with non-linguistic strategies and then use these as the basis for their first linguistic hypotheses; because of this, certain meanings can be regarded as cognitively simpler than others. This approach could also be taken to a number of other studies of comprehension. In the acquisition of more and less, for instance, one could argue that it is a non-linguistic strategy (‘Pick the greater amount’) that provides the basis for the child’s linguistic hypothesis about the feature [+Polar]. The positive term, more, should be cognitively simpler than less because responses based on the strategy coincide with the real meaning of more. The same argument could be made with respect to the positive members of all the dimensional adjective pairs, e.g., big, tall, long, wide, etc. The strategy of choosing the object with the greater extent or amount coincides with the meanings of the positive terms (Donaldson and Wales, 1970; Klatzky, Clark and Macken, 1973). There are also other phenomena that may be explained by the child’s use of nonlinguistic strategies, but it is difficult to know if this explanation is the right one. For example, Bever (1970a, 1970b) found that very young children appeared to interpret reversible passives such as The cat was chased by the dog as if they were actives, i.e., The cat chased the dog (c$ also Slobin, 1966). Bever argued that both children and adults use a parsing ‘strategy’ of identifying noun-verb-noun sequences and then assigning to those elements the semantic roles of actor-action-object in that order. The question that remains is where the order implicit in the underlying semantic relations comes from. It could be derived from the child’s non-linguistic knowledge: The nonlinguistic strategy in this instance would be based on the child’s knowledge that certain objects (people?) could do some action, and this action resulted in a change in a further class of objects at a subsequent time. On the other hand, it could be argued that the child has generalized the actor-action-object pattern from the typical use of active sentences in English. At this point, it is impossible to tell which of these two accounts is the more viable. A similar argument could be put forward to account for the child’s reliance on order of mention in reconstructing the actual order in which a sequence of events occurred. When young children are asked to act out the events described in a sentence like The boy jumped the fence after he patted the dog, they consistently treat the first clause as a description of the first event, and the second clause as a description of the second event (E. Clark, 1971; Ferreiro, 1971). Young children also tend to describe a series of events in time in the order in which they happened (E. Clark, 1973a). The child could have an a priori non-linguistic strategy of acting out and talking about 5. It is possible that a technique such as that used by Donaldson and Lloyd (1974) might

prove more viable for this type of experiment.

Non-linguistic

strategies

and meaning

18 1

events in the order in which he remembers them, with the assumption that they occurred in that order. This non-linguistic strategy then leads the child to make the linguistic hypothesis that the order of mention always reflects actual order. Alternatively, of course, children may have realized that adults normally talk about events in the order in which they occur, and it is this that accounts for their initial order of mention strategy. To sum up, the present study has shown that in their attempt to comprehend in, on and under, young children rely on a combination of linguistic hypotheses about the word meanings and certain non-linguistic strategies. Furthermore, it has been argued that these non-linguistic strategies probably form the basis for their hypotheses about the meanings of new words. Thus, the degree of coincidence between responses based on a non-linguistic strategy and responses based on semantic knowledge may determine the relative cognitive complexity of different linguistic forms and hence determine their order of acquisition.

REFERENCES Bern, S. L. (1970) The role of comprehension in children’s problem-solving. Dev. Psychol.,

-

2, 351-358.

Bever, T. G. (1970a) The cognitive basis for linguistic structures. In J. R. Hayes (Ed.)

development and the acquisition of language. New York, Academic Press. Pp. 65

Cognition and the development of language. New York, Wiley. Pp. 279-352.

-

(1970b) The integrated study of language behavior. In J. Morton (Ed.) Biological and

social

factors

in psycholinguistics.

Urbana, Ill., University of Illinois Press. Pp. 158-209. Brown, R., Cazden, C. B., and Bellugi, U. stages. Cambridge, Mass., Harvard University Press. Brown, R., Cazden, C. B. and Bellugi, U. (1969) The child’s grammar from I to III. In J. P. Hill (Ed.) Minnesota symposium on child psychology, Vol. 2. Minneapolis, Minn., University of Minnesota Press. Pp. 28-73. Clark, E. V. (1971) On the acquisition of the meaning of before and after. J. verb. Learn. verb. Beh., 10, 266-275. (1973a) How children describe time and order. In C. A. Ferguson and D. I. Slobin (Eds.) Studies of child language development. New York, Holt, Rinehart &

Winston. Pp. 585-606. (1973b) What’s in a word? On the child’s acquisition of semantics in his first language. In T. E. Moore (Ed.) Cognitive

-

110. (1974) Some aspects of the conceptual basis for first language acquisition. In R. L. Schiefelbusch and L. L. Lloyd (Eds.) Language perspectives-Acquisition, retardation and intervention. Baltimore, Md.,

University Park Press. Clark, H. H. (1970) The primitive nature of children’s relational concepts. In J. R. Hayes (Ed.) Cognition and the development language. New York, Wiley. Pp. 269-278. (1973) Space, time, semantics and the child. In T. E. Moore (Ed.) Cognitive development and the acquisition of language. New York, Academic Press. Pp. 28-

63. Donaldson, M. and Balfour, G. (1968) Less is more: A study of language comprehension in children. Brit. J. Psychol., 59, 461-472. and Lloyd, P. (1974) Sentences and situations: Children’s judgments of match and mismatch. In Proceedings of sym-

182

Eve. V. Clark

posium on current problems linguistics. Paris, CNRS.

-

and Wales, R. J. (1970) On the acquisition of some relational terms. In J. R. Hayes (Ed.) Cognition and the development of language. New York, Wiley. Pp. 235-268. Ferreiro, E. (1971) Les relations temporelles duns le langage de l’enfant. Geneve, Droz. Ghent, L. (1960) Recognition by children of realistic figures in various orientations. Can. J. Psychol.,

14, 249-256.

Huttenlocher, J., Eisenberg, K., and Strauss, S. (1968) Comprehension : Relation between perceived actor and logical subject. J. verb. Learn. verb. Beh., 7, 527-530. __

polar adjectives: Linguistic or conceptual?

in psycho-

and Strauss, S. (1968) Comprehension and a statement’s relation to the situation it describes. J. verb. Learn. verb. Beh., 7, 300-304.

Klatzky, R. L., Clark, E. V. and Macken, M. (1973) Asymmetries in the acquisition of

J. exp. child Psychol.

Laurendeau,

16, 32-46.

M. and Pinard,

development child. New

A. (1970) The of the concept of space in the

York, International Universities Press. Palermo, D. S. (1973) More about less: A study of language comprehension. J. verb. Learn.

verb. Beh., 12, 211-221.

Slobin, D. I. (1966) Grammatical transformations and sentence comprehension in childhood and adulthood. J. verb. Learn. verb. __

Beh., 5, 219-227. (1971) Developmental

psycholinguistics. In W. 0. Dingwall (Ed.) A survey oj’linguistic science. Linguistics Program, University of Maryland. Pp. 298-400. [Reprinted (1973) as ‘Cognitive pre-requisites for the development of grammar’. In C. A. Ferguson and D. I. Slobin (Eds.) Studies of child Ianguage development. New York, Holt, Rinehart & Winston. Pp. 175-208.1

Rksumt

Cet article a pour but de demontrer que dam un premier temps la comprehension apparente de certains mots chez les enfants depend autant d’hypotheses linguistiques qu’ils font sur le sens dun mot que de certaines strategies d’ordre non-linguistique. Dans l’experience decrite par l’auteur des enfants ages de 18 mois a 5 ans ont recu des directives concemant la comprehension des mots ‘dam’, ‘sur’ et ‘sous’. Les resultats de cette experience ont demontre que la comprehension chez les enfants passe par trois Btapes consecutives. D’abord ils utilisent certaines strategies nonlinguistiques caracterisees par deux series de regles bien dtterminees; dam un deuxieme

temps ils appliquent ces regles a une ou deux seulement des instructions locatives donnees et finalement ils dimontrent une connaissance semantique totale de la signification des trois mots en question. Grace a leurs strategies non-linguistiques, les enfants plus jeunes paraissent comprendre toujours ‘dans’ correctement, ne comprennent ‘sur’ que dans certains cas, et ne comprennent jamais le mot ‘sous’. L’auteur soutient que ce sont les strategies d’ordre non-linguistique qui en fait determinent l’ordre d’acquisition des mots ‘dans’, ‘sur’ et ‘sous’ chez les jeunes enfants.

2

Acceptability,

interpretation and knowledge of the world Remarks on the verb PLANTER (to plant)*

JEAN-PAUL

BOONS

Universities

of Paris VII and VIII

E.R.A.

(C.N.R.S.)

247

The reader may please to observe that, in the last article of the recovery of my liberty, the emperor stipulates to allow me a quantity of meat and drink sufficient for the support of 1,728 Lilliputians. Some time after, asking a friend at court how they came to fix on that determinate number, he told me that his majesty’s mathematicians, having taken the height of my body by the help of a quadrant, and finding it to exceed theirs in the proportion of twelve to one, they concluded, from the similarity of their bodies, that mine must contain at least 1,728 of theirs, and consequently would require as much food as was necessary to support that number of Lilliputians. By which the reader may conceive an idea of the ingenuity of this people, as well as the prudent and exact economy of so great a prince. Jonathan Swift, Gulfiver’s Travels

Abstract Transformational methods have made of the acceptability test the instrument of measure and class$cation par excellence in linguistics. Of the various a prioripossible parameters which can injuence the result of this test, it is especially the factor of semantic interpretation which is considered here. We would like to show the necessity of dissociating an extralinguistic part of this interpretation (including the universe of discourse adopted by the informant, the culture to which he belongs, and the importance of certain factors of * Translated by C. Perdue. A French version of this article, Acceptabilite, Interpretation et Connaissance du Monde: A propos du verbe planter, will appear at a later date in Actes du

colIoque franco-allemand

de grammaire

trans-

formationnelle, Christian Rohrer and Nicolas Ruwet (Ed%), Vol. 2, Etudes de Se’mantique et Autres, Niemeyer, Linguistische Arbeiten. Cognition

Z(2),

pp.183-211

184

Jean-Paul Boons

knowledge - or of ignorance - of the world) from a linguistically pertinent hypothetical interpretation which correlates very closely with questions of syntax or morphology. Extra-linguistic semantic concepts are ‘natural’, come immediately to the mind and screen the research of much more abstract, unfamiliar and probably unconscious linguistic semantic concepts. This is a boundary which it is necessary to draw in semantics between what is linguistic and what is not, i.e., between what must and what cannot appear in a generative grammar. We insist here on the necessity in linguistics of controlling, and thus of varying, the universes of discourse which can modify the acceptability of a given sentence. The argument centres around several uses of the verb planter (‘to plant’). I. Judging the acceptability of a sentence - i.e., deciding whether or not it belongs in a natural language - has come to be the privileged tool of classification and of measurement in linguistics. The advent of transformational methods made this procedure decisive, since they usually consist in comparing at least two sentences or sentence-types whose syntax is different but which are closely related both by the nature of the lexical items they contain and by their ‘synonymy’, with a view to establishing whether or not they belong to a language. ‘Synonymy’ in this case generally turns out to be a difference of specific meaning within a basis of relative synonymy. The linguist’s choice of sentence is not a random one but rather determined by a hypothesis predicting the degree of acceptability of one sentence from the other. Any definition of acceptability by the presence or absence of sentences in some empirically limited corpus (a written text, a recorded conversation) is thus precluded by a method whose constant preoccupation is the study of relations between at least two sentences. Even if one of these sentences could have been taken from some corpus, the fact that the other occurs in the corpus or not is not significant, since the probability of finding it in the corpus is nearly zero. Furthermore, in the event that it were in the corpus, its featuring in another context would prevent the linguist from postulating the basis of relative synonymy which backs up the syntactic relationship. Its nonoccurrence in the corpus does not allow the linguist to draw any conclusion as to its belonging or not in the language. The criterion of a sentence’s being part of a language necessitates therefore an acceptability test, in which the sentences under consideration are presented to a subject (or informant) whose task it is to give a judgment on them. The use of a corpus can be only an auxiliary aid or corrective in relation to the acceptability test, in the sense that the repeated occurrence of a sentence-type in a certain type of corpus can be taken as a pointer to, or a proof of, its acceptability, given an appropriate choice of corpus. As the validity of the acceptability test is a necessary condition on the validity of

Acceptability, interpretation and knowledge of the world

transformational

methods,

it is worth examining

the parameters

involved,

185

which are,

a priori, five in number: - the linguist’s construction and choice of a battery of sentences S to be judged by informants as belonging or not belonging to a natural language L. - the value j of this judgment. - the informants i of whom the judgment is asked: The parameter is liable to interinformant variability (the informants disagree amongst themselves) and to intrainformant variability (the informant disagrees with himself over a certain length of time). - a semantic interpretation according to which the informant i may attribute j to S. - the order in which the sentences of S are presented to each informant i. In this paper we shall be chiefly concerned with the effect of the semantic interpretation on the acceptability judgment. The parameters informant and order of examples are considered merely from the point of view of their connection with this effect. The term ‘interpretation’ will be used at the start in its widest sense: It includes a hypothetical interpretation in the strict, i.e., linguistically relevant, sense, as well as extralinguistic factors such as the world of discourse the informant places himself in, or the informant’s culture or knowledge of the world in which the interpretation itself takes place. What we shall try to do here is examine examples where these extra linguistic factors have an incidence on acceptability, taking as our starting point some constructions containing the verb planter (to plant).

II.

Consider

(1)

(a) Pierre (Peter (b) Pierre (Peter (c) Pierre (Peter (a) Pierre (Peter

(2)

the constructions

with planter in (l), (2) and (3).

a plan&? une sapinihe dans sa propri&d. has planted a fir plantation on his estate.)l a plant& de nombreux sapins dans sa propri&C. has planted lots of fir trees on his estate.) a plant& un sapin dans sa propriht. has planted a fir tree on his estate.) a plantk sa proprih! d’une sapini&e. has planted his estate with a fir-plantation.)

1. ( ), [ I: brackets are (i) used to indicate the product of factors, (ii) placed round the English translation of examples. + : the plus sign indicates a logical ‘or’. E: capital E indicates a sequence composed of zero elements.

?: doubtful examples are preceded by a question-mark. * : unacceptable examples are preceded by an asterisk. (Judgments of acceptability are given only for the examples in French. The translation is not graded.)

186

(3)

Jean-Paul Boons

(b) Pierre a plant2 sa propri&t de nombreux sapins. (Peter has planted his estate with lots of fir trees.) (c)? Pierre a plant6 sa propri&C d’un sapin. (Peter has planted his estate with a fir tree.) (a) Sa propri&k Ctaitplant&e d’une sapinihe. (His estate was planted with a fir plantation.) (b) Sa proprihk Ctaitplantte de nombreux sapins. (His estate was planted with lots of fir trees.) (c)? Sa proprikt? &ait plant&e d’un sapin. (His estate was planted with a fir tree.)

The construction in (1) contains, let us say, a prepositional locative complement on his estate. This complement has the semantic role of ‘goal’ in relation to the fir tree(s) in the process of being planted. The fir tree(s) have, in relation to the agent (Peter) and the goal, the role of a moved object, a ‘theme’. (4) gives us another example of a verb taking the construction of (1): (4)

Pierre met des pavPs dans le camion. (Peter puts paving stones into the lorry.)

The constructions of (2) and (3) have an obvious semantic relationship with (l), and differ syntactically in a straightforward way. The actant ‘goal’ appears in direct object position and the actant ‘theme’ in indirect object position, preceded by the preposition de. Construction (2) contains, let us say, a non-prepositional locative complement. Another example of a verb taking construction (2) is : (5)

Pierre remplit le camion de pavts. (Peter fills the lorry with paving stones.)

In (2) and (5), the non-prepositional locative complement appears in direct object position, but this complement can also appear in subject position, as in: (6)

La propriM pullule de sapins. (The estate is thick [pullulates] with fir trees.)

Planter has the characteristic of being able to take constructions characteristic is not proper to planter; thus we have: 2. Actunt is a term first used by Tesnibe (1959) to designate NPs which are involved in the

process described by the verb of the sentence. It may be used syntactically or semantically. The verb rain in It is rainingis said to have no actants; run in Bill is running has one actant, Bill; put, in Bill put the book on the table has

(1) and (2). This

three actants, Bill, the book and the table. Theme, goal are terms used by Gruber (1965) to describe the net of relationships between the verb and its actants which he calls ‘thematic relations’. Here, they are used uniquely to express semantic relationships vis-d-vis the verb and each other.

Acceptability, interpretation and knowledge of the world

(7)

(a) Pierre (Peter (b) Pierre (Peter

Neither (8) (9)

charge des paves darts le camion. loads paving stones into the lorry.) charge Ie camion de paves. loads the lorry with paving stones.)

is it common

*Pierre (Peter *Pierre (Peter

187

to all verbs, as we see in:

met Ie camion de paves. puts the lorry with paving stones.) remplit des paves dans le camion. fills paving stones into the lorry.)

Mettre and remplir take exclusively constructions (1) or (4) and (2) or (5), respectively, whereas charger (like planter) takes both. From now on we shall call the type of construction in (1) or (4) ‘canonical’, and that of (2) or (5) ‘permuted’. These terms are merely a convenient means of description; the term ‘permuted’, in particular, should not be taken to indicate that (2) or (7a) are derived transformationally or otherwise from (1) or (7a). The verbs taking one or other of these constructions will also be called canonical or permuted. Mettre is thus exclusively canonical, remplir exclusively permuted, while planter (or charger) are both canonical and permuted. There are about 600 canonical verbs and 1,0003 permuted verbs in French. These two sets have an intersection (where we find planter and charger) whose size is difficult to ascertain, since it depends on the limits of acceptability one is prepared to go up to for each construction. We may, however, say that the intersection contains at least a hundred verbs which incontestably take both constructions provided that the noun heads of the goal and theme NPs are chosen appropriately. The condition on the choice of nouns is essential, as the following examples show: (10)

Pierre charge (Marie de cette t&he + cette t&he [a + sur] Marie).4 (Peter entrusts [Mary with this job + this job (to + on) Mary].)

3. These numerical estimations have been made possible by work on the tabulation of syntactic properties of French verbs undertaken by the Laboratoire d’Automatique Documentaire et Linguistique. A study of verbs taking completive constructions has been completed (Gross, 1969, 1974); a second study is concerned with verbs taking simple, i.e., noncompletive constructions (Boons, Guillet and Leclere, 1974). In this article, general atflrmations as to the domain or correlations of syn-

tactic properties in a lexis of verbs are based on the results of these studies. 4. Example (10): Pierre charge (Marie de cette t&he + *cetfe t&he [Li+ sur] Marie). should be read: (a) Pierre charge Marie de cette t&he. (b) *Pierre charge cette t&he d Marie. (c) *Pierre charge cette t&he sur Marie. The fact that the translation of (b) Peter entrusts thisjob to Mary is grammatical should not be taken into account.

188

Jean-Paul Boons

(11)

(a) On a attele six chevaux d la diligence.

(Someone harnessed six horses to the stagecoach.) (b) On a attele la diligence de six chevaux. (Someone harnessed the stagecoach with six horses.) (c) On a attele six personnes d cette t&he. (Someone harnessed six people to the job = six people were put on the job.) (d) *On a attele cette t&he de six personnes. (Someone harnessed the job with six people.) Strictly speaking, therefore, it is not so much the verbs charger or atteler which we find in the canonical/permuted intersection but certain usages of these verbs. Differences of usage similar to those of (10) and (11) but of a finer nature, exist for planter. They come to light in the hesitations informants have about the degree of acceptability to be attributed to the nine sentences (l)-(3). When different informants are asked to give an acceptability judgment for these sentences, a wide range of reactions is observed. At the conservative end of the range, some informants accept only (lb), (lc), (3a) and (3b); at the liberal end, all the sentences are accepted. A more moderate position is that which does not accept the sentences of (2), and which only accepts the permuted use of planter if the construction contains an adjectival past participle, as in (3). However, most informants do not accept (2~) and (3~): Sapin can be singular only in the canonical construction and is obligatorily plural in the permuted construction. The complement de DET N of the permuted use of planter would seem therefore to refuse a semantic singular. De DET N would necessarily be plural if N denotes someone or something countable, like sapin, and can only be in the singular if N is collective, denoting a set of countables like sapinidre. There is a widespread tendency to accept the permuted constructions only if they contain the past participle. Many informants accept exclusively as adjectives lexical items which others accept as verbs. Indeed, the views of some informants are more conservative than those of a dictionary like Le Petit Robert, where we find for the root of a noun N a verbal entry V = N - er, whereas the only form accepted by some informants is the adjective N - P. This is the case for balafrer (to cut), mater (to mast a ship), pointiller (to dot). Dictionaries do not agree amongst themselves over this, moreover. Thus we find in Bescherelle’s list, some verbs of the form V = N - er which appear in Le Petit Robert only as adjectives of the form A = N- 6, marqueter (to inlay) and persiller (to sprinkle with parsley). Conversely, some possible verbs are not found in Bescherelle and appear as the adjective N - t; in Le Petit Robert, which defines for example the adjectif melonne as ‘having the shape of a melon’. Nothing prevents the verbal form melonner NP1 meaning ‘to give the form or appearance of a melon to NP,’ - except the absence in our culture of the appropriate technology.

Acceptability,

interpretation

and knowledge

of the world

189

The verb/past participle status of an item is thus a widespread problem in the French lexicon. We shall return to it in this article, considering however only a limited number of verbs.

111. Let us return to the question of the obligatory plural in the de DET N complement of permuted structures. We have seen that verbs taking the permuted construction tend not to accept the specification ‘semantically singular’ for this complement. Thus, (12)

(a) Pierre a constelle le plafond (*d’une + d’) ttoile (s) en papier dare. (Peter spangled the ceiling with [a silver paper star + silver paper stars].) (b) Pierre a (boise + herisse) la colline (*d’un + de) sapin( (Peter [afforested the hill + made the hill bristle] with [a fir tree + fir trees].) Pierre a (entourt + cerne + encercle) sa maison (*d’un + de) sapin( (c) (Peter [surrounded + shut in + encircled] his house with [a fir tree+fir trees].) (d) Pierre a (mouchett + tachete +piquett) sa toile (*d’une + de) Pclaboussure(s). (Peter [spotted + flecked + dotted] his canvas[with a splash + with splashes] of paint.) Pierre a (hachure + nervurd + strie + sillone) sa toile (*d’une + de) trace(s) (e) zigzaguante(s). (Peter [hachured + striped + streaked + scored] his canvas [with a zigzag line + with zigzag lines].)

Notice that (12), as well as (2~) and (3c), where de DET N is in the singular, becomes more acceptable (in varying degrees) if DET N is interpreted generically, i.e., as signifying (one + this) (type + kind+ etc.) of (fir tree + star + splash + line). This way of making doubtful singulars acceptable has some generality and will not be taken into account from now on. The restriction ‘obligatory plural’ for the permuted constructions of certain verbs must either appear in the lexical entry of these verbs or be predictable from the entry. Thus for boiser, we can predict this restriction from the noun bois, from which the verb is derived morphologically, since the noun carries the same restriction for its own complement: Un bois de sapins (a wood of fir trees) necessarily has sapins in the plural, and the sentence (13)

Si je comprends bien, le bois dont tu me parlais se compose d’un seul sapin. (If l’ve got you right, this wood you were telling me about consists of one ti tree.)

can only be a joke.

We can make a similar

prediction

for herisser from the noun

190

Jean-Paul Boons

herisson (hedgehog), and for consteller, constellation from the prefix con-. The restriction is equally easy to predict for the verbs of (12~) from the semantic relationship ‘around’. The noun heads of de DET N will denote either collectives or lengthy objects thus : (14)

Pierre a entourt l’arbre (d’une ficelle f d’un serpent d sonnette f ? d’un crocodile + ? *d’une vache). (Peter has put [a length of string + a rattlesnake + a crocodile + a cow] round the tree.)

The existence of the restriction for the verbs of (12d) is doubtless related to the morphological structure N-t-er of these verbs. The acceptability of the singular for the complement in (12e) is partly dependent on the possibility of using the nouns hachure, nervure, strie, sillon in the singular. The nouns are ordered here for increasing possibility. Many factors can cause the obligatory plural of the complement de DET N in the permuted constructions. We shall examine in particular this plural in relation to planter in sentences (2~) (3~). We set out to show here that the presence of this restriction in the lexical entry of planter would not be justified, since sentences (2~) and (3~) are in fact perfectly grammatical, or, more precisely, since they are in no way semantically anomalous; there should be no question-mark in front of these sentences, as was the case above. We shall try to show that their apparent non-acceptability is an artifact explainable in terms of a genuine semantic property of planter interacting with certain extra-linguistic factors. The semantic property in point is shared by numerous permuted constructions. It has been described in various ways in linguistic writings, especially in Matthews (1968), Chomsky (1972) and Hall (1971), who discussed the canonical and permuted constructions of to swarm: (15)

(a) The bees are swarming in the garden. (b) The garden is swarming with bees.

Whereas Matthews simply distinguishes the canonical from the permuted construction, remarking that the latter includes an aspectual trait not present in the former and which he calls ‘perfective’, Chomsky points out that the sense of the permuted construction seems to be that there are bees swarming everywhere in the garden, while the canonical construction does not. Barbara Hall Partee qualifies this interpretation by saying that the permuted construction supposes that there are only bees in the garden, while the canonical construction allows one to imagine other types of insects or creatures swarming away as well. These attempts at explanation are only partly satisfying. It seems to us that it is quite possible to accept sentences (3b) or (7b) without necessarily supposing that the

Acceptability, interpretation and knowledge of the world

191

fir trees or paving stones fill all the available space (Chomsky), nor that there is, or will be, no plantation or load in the estate or lorry other than fir trees or paving stones (Hall). We shall leave this hunting after sense (which could well turn out to be a hunting of the Snark), remarking merely that (3b) or (7b) must contain some idea of occupation of the estate or the lorry by the fir trees or paving stones. This semantic property of the permuted construction can be made more precise, not by postulating a ‘semantic trait’ of the form : /j-occupation of the goal by the theme/, which would only prolong the ‘hunting of the sense’ but rather by saying that the process described by sentences (2) and (3), (7b) and (12) implies, if they are judged to be acceptable, that the following sentences (16)

(Une sapiniere + (unt de nombreux) sapin occupe(nt) la propriett. ([A fir-plantation + (a + several) fir tree(s)] occupy[ies] the estate.) (Une + des) (&claboussure[s] + traces zigzaguante[s]) occupe(nt) la toile. ([A + E] [splash(es) + zigzag line(s)] occupy[ies] the canvas.), etc.

are acceptable and true, in whatever sense occupy is taken; non-metaphorically (i.e., all the space is literally filled up) or metaphorically (i.e., the available space is taken up by the presence of somebody or something). Perhaps there is some other verb than occupy which characterises better the semantic relationship existing between the ‘goal’ actant and the ‘theme’ actant of the permuted as opposed to the canonical constructions, but occupy will serve our purpose. This semantic analysis seems to correlate (statistically at least) with the characteristics of the permuted constructions, i.e., the presence of an actant in subject or object position, not preceded by a preposition, having the role of ‘goal’. Notice that paraphrases with occupy such as we find in (16) adapt badly to (12c), entourer, especially since in this case it is not the house itself that the fir trees occupy. It is necessary in the case of entourer to allow in paraphrase (16) that the ‘theme’ occupies, not the ‘goal’ noun N, but some Ni of N, where N* (surrounds or neighbourhood) would be morphologically linked to the verb. Thus we would have (17)

Les sapins occupent les (entours + alentours) de la maison. (The fir trees occupy the [surrounds + surroundings] of the house.)

Another example of this variant is NI = dessus (top), from the prefix sur- (on, on top of) of surmonter (surmount) : The result of the process described in the sentence (18)

Pierre a surmonte I’armoire a glace d’une potiche. (Peter surmounted the wardrobe with a vase.)

is more acceptably (19)

described

if the Ni dessus appears

in the sentence with occuper :

Une potiche occupait (? E+ le dessus de) I’armoire a glace. (A vase occupied [E + the top of] the wardrobe.)

192

Jean-Paul Boosts

(19) does not mean that the vase covers the greater part of the top of the wardrobe. It illustrates why a paraphrase using occuper is useful, since we thus avoid searching for an absolute meaning. We notice, however, taking a possible ‘absolute’ meaning into account, that (19) favours Hall’s interpretation, since it does not seem to permit the presence on top of the wardrobe of another object or other objects of equal or greater size than the vase; the vase is the ‘principal’ object to be found on top of the wardrobe.

IV.

Let us turn to one of the extra-linguistic

factors we mentioned.

(20)

Un pot en terre cuite planti d’un jeune sapin garnissait agrkablement l’appui de la fenhre. (A terra-cotta pot planted with a young fir tree provided a pleasant decoration for the window ledge.)

seems perfectly acceptable. It could come out of a descriptive Take away the flourishes and we get: (21)

The sentence

passage

in a novel.

Lepot ttaitplantt? d’un sapin (The pot was planted with a fir tree.)

(21) is necessarily acceptable if we accept (20) and is almost identical with (3c), where we findpropriett instead ofpot. Obviously, the fir tree can occupy the place denoted by the direct object only if this place is of appropriate (relatively small) size. The acceptability judgments of an informant can thus be influenced by certain elementary notions he has of the world of discourse evoked by the examples. In the case of planter, examples (1) to (3), these notions have to do (above all) with the relative size of objects in our everyday world. The choice of propriM in (1) to (3) was designed to make the non-acceptance of (2~) and (3~) as widespread as possible. Had the word jardin (garden) been chosen, certain informants would certainly have accepted (2~) and (3~) by picturing to themselves a very small garden and a very large fir tree. This incursion of the knowledge of the relative size of objects into informants’ judgments will be considered as an extra-linguistic factor. We can nevertheless imagine that the lexical entry of an item denoting a concrete object could carry a numerical estimate of the mean and standard deviation of the distribution of the size of that object in our everyday world. A generative grammar would then include a calculator comparing the size of different objects denoted by nouns appearing in a sentence, and the result of the calculation would lead to the non-acceptability of (3c), the acceptability of (21) and the borderline acceptability of the equivalent sentence

Acceptability, interpretation and knowledge of the worEd

193

with jardin on condition that the garden was only little. The task is a priori possible. It does not however seem to us to be a linguistic task, since the work involved has to do with information storage rather than linguistics; indeed, the set of numerical particulars to be collected and the ensuing calculation seems devoid of all relevance to the structure of a natural language. It is moreover impossible for a grammar to make do with a calculation of the relative size of objects in our everyday world of discourse alone. Many cultures have given birth to myths wherein the everyday world of discourse is not respected, where men, animals and objects may have different sizes depending on their origins. One thinks immediately of Gulliver and the Lilliputians, of Ulysses and the Cyclops, of Gargantua and the Parisians, or indeed of any tale in which there are giants or ogres. Sentence (3~) is immediately acceptable if the estate belongs to the King of Lilliput and the fir tree is a present brought by Gulliver from England. Up until now, the problem of size relationships and their effect on a generative grammar has been considered within the framework of information storage and artificial intelligence. The grammar is programmed, and the programme might include a perfectly ad hoc accessory component which would calculate size relationships and would exist to avoid the grammar’s generating sentences such as (2b) and (3~). Moreover, in the event of the language describing the Lilliputian world of discourse, this component would carry a fixed coefficient allowing it to calculate the size of the Lilliputians and of objects from their country. Swift gives this coefficient: It is l/12. This framework is scarcely exciting, especially as nothing allows us to say that such applications of information science will ever exist. But there are more interesting perspectives. It is worth demonstrating that it is at the level of the construction of his examples that the linguist is led to make the size relationships amongst objects vary and to use, for example, a Lilliputian world of discourse. There exists in French a set of syntactic structures having in common the fact that an NP of the form Dd Npb (where Dd represents the definite article and NDb is a noun representing a part of the human body) is interpreted as referring to one of the ‘human’ actants of the sentence. The ‘human’ actant may be represented by the pre-verbal pronouns lui ([appertaining] to him, her) or the reflexive se. Consider the sentences: (22)

Pierre (Peter (II) Pierre (Peter (b) (I) Pierre (Peter

(a) (I)

a regx Ie pa& sur sa t&e. copped [Lit. = ‘received’] the paving stone on his head.) a reg4 le pave sur la tete. copped the paving stone on the head.) Iave les mains de Marie. washes Mary’s hands.)

194

Jean-Paul Boons

(Ii) Pierre lui lave les mains h Marie. (Peter washes the hands of Mary.)5 We notice a systematic difference of meaning between sentences (I) and (II). In sentences (I) sa t&e or les mains de Marie allow the interpretation of sculptures representing a head or hands and belonging to Peter or Mary (Mary’s hands= the hands that Mary has = alienable possession); but it is much more difficult to give this interpretation to sentences (II). Here the NPs la t&e or les mains are normally interpreted as forming part of the body of Pierre or of lui (i.e., Marie: Mary’s [inalienable possession] hands).6 Notice incidentally that in (22 a I) the interpretation of sa tete as a sculpture is only possible if Pierre is holding it close to him, but this is a property of a small number of verbs - recevoir, prendre, attraper - and is not of direct concern to us. here. These syntactic structures are extremely varied in French, as much through the position of the actant containing the Npb as through that of the actant to which it refers. In order not to overburden this text, we shall take just two of these structures, which appear in (23): (23)

(24)

(a) (1) Pierre embrasse le front de Marie. (Peter kisses Mary’s forehead.) (11) Pierre embrasse Marie sur le front. (Peter kisses Mary on the forehead.) (b) (1) Le sang coule sur le bras de Marie. (Blood is running down Mary’s arm.) (II) Le sang lui coule sur le bras h Marie. (Blood is running down the arm of Mary.) (a) (I) D’Artagnan respecte les bras de Porthos. (D’Artagnan respects Porthos’ arms.) (11) *D’Artagnan respecte Porthos aux bras. (D’Artagnan respects Porthos in the arms.) (b) (1) La lave coulait vers les pieds de Marie. (Lava flowed towards Mary’s feet.) (II)? La lave lui coulait vers les pieds, d Marie. (Lava flowed towards the feet of Mary.) (c) (I) Le briquet est sur le ventre de Marie. (The lighter is on Mary’s stomach.) (11) *Le briquet lui est sur le ventre, a Marie. (The lighter is on the stomach of Mary.)

5. Literally ‘Peter to her washes the hands, to Mary’. The comma represents a pause in speech; Mary is detached from the sentence and

thrown

into relief; cf. English Mary,

Peter

washed her hands.

6. Chomsky (1970) discusses similar examples.

Acceptability,

(25)

(a) (I)

(b)

(26)

(a)

(b)

(27)

(a)

(b)

(28)

(a)

(b)

(29)

(a)

(b)

interpretation

and knowledge of the world

I95

*Pierre expire la main de Marie. (Peter breathes out Mary’s hand.) (II) *Pierre expire Marie h la main. (Peter breathes out Mary on the hand.) (I) *Pierre decoule vers la main de Marie. (Peter is implied towards Mary’s hand.) (II) *Pierre lui decoule vers la main, a Marie. (Peter is implied towards the hand of Mary.) (I) ?Le voilier a accost& le genou de Pierre. (The sailing-ship tied up alongside Peter’s knee.) (II)? Le voilier a accoste Pierre au genou. (The sailing-ship tied up alongside Peter at the knee.) (I) ? Le planeur a atterri sur le bras de Pierre. (The glider landed on Peter’s arm.) (II)? Le planeur lui a atterri sur le bras, h Pierre. (The glider landed on the arm of Peter.) (I) ?Arthur regagne l’estomac de Marie. (Arthur is getting back into Mary’s stomach.) (II)?Arthur regagne Marie h l’estomac. (Arthur is getting back into Mary in the stomach.) (I) ?Le voilier cinglait vers le genou de Pierre. (The sailing-ship was making towards Peter’s knee.) (II)? Le voilier lui cinglait vers le genou d Pierre. (The sailing-ship was making towards the knee of Peter.) (I) Le voilier lilliputien a accoste le genou de Gulliver. (The Lilliputian sailing-ship tied up alongside Gulliver’s knee.) (II) Le voilier lilliputien a accost6 Gulliver au genou. (The Lilliputian sailing-ship tied up alongside Gulliver at the knee.) (I) Le planeur lilltputien a atterri sur le bras de Gulliver. (The Lilliputian glider landed on Gulliver’s arm.) (II) Le planeur lilliputien lui a atterri sur le bras, h Gulliver. (The Lilliputian glider landed on the arm of Gulliver.) (I) Jonas regagne l’estomac de la baleine. (Jonah is getting back into the whale’s stomach.) (II) *Jonas regagne la baleine a l’estomac. (Jonah is getting back into the whale in the stomach.) (I) Le voilier lilliputien cinglait vers le genou de Gulliver. (The Lilliputian sailing-ship was making towards Gulliver’s knee.) (II)?Le voilier lilliputien lui cinglait vers le genou, d Gulliver. (The Lilliputian sailing-ship was making towards the knee of Gulliver.)

196

Jean-Paul Boons

In the structure shown in (a It), there appears a complement Lot Dd Npb (where Lot represents a locative preposition such as h [to], dam [in], sur [on], contre [against]) which refers to the direct object and does not belong to the normal structure (a 1) taken by the verb. In the structure shown in (b II) the complement Lot Dd Npb corresponds to the normal complement Lot NP of the verb, and refers to the preverbal pronoun hi. The structures of (II) are not always acceptable. Their acceptability depends in the main on the nature of the verb or of the preposition, as the examples of (24) show. In (24 a) and (24 c) the non-acceptability of sentences (II) is due to the verb, and to the preposition in (24 b), as we see if we compare (24 b II) and (23 b II). The question mark preceding (24 b II) does not necessarily indicate hesitation on the part of one informant ; it can also indicate a variation of judgment between informants. It is not necessary to postulate a transformational derivation of (II) from (I), with (I) being closer to some deep structure, for it to be interesting to present the two structures together. In fact with everything seeming to be against such a derivation (cf: Kayne, 1974), this parallelism nevertheless allows us to distinguish (24) from (25). The type of non-acceptability present in the sentences (25 II) is very different from that of the sentences (24 II). Whereas in (25 II) the non-acceptability is banal, involving selectional restrictions on the direct object and locative complements of expirer and dkouler, which are already violated in the examples of (25 I), this is not so in (24 II) since the sentences of (24 1) are acceptable. It is the acceptability of sentences of type (I) that governs the linguistic interest of the possible non-acceptability of sentences of type (II). These remarks lead us therefore to distinguish between three types of verb, shown here in examples (23), (24), (25), and which can be characterised as follows: (23) I- II (24) I-*11 (25) *I-*11 If we want to find out why it is that some verb takes a ‘part of the body’ structure verbs taking the structures of (23) - verbs belonging to (25) are not revealing, whereas verbs belonging to (24) give us crucial evidence, since it is these verbs that will allow us to falsify any hypothesis as to the extent of this structure amongst the verbs of the lexicon. It is therefore vital to be able to distinguish between (24) and (25). Let us now look at examples (26) and (27). The reason for the bizarre nature of (26 I) and (27 I) is evident: The size relationship of subject and complement does not conform to the norms of the everyday world of discourse. After all, we only have to imagine that the ship or the glider are particularly sophisticated models, or that

Acceptubility,

Arthur

interpretation

is in fact the name of a tapeworm,

and knowledge

for the sentences

of the world

to become

197

acceptable.

But the same thing is not true for sentences (II). The unusual character of the examples is all that is needed to create a doubt, to blur the limits of acceptability so that we do not know which sentences are to be assimilated to (23) (24) or even (25). It turns out that this vagueness can be overcome by putting (26) and (27) into the form of (28) and (29) respectively. We consider that the sentences (28 J) and (29 I) are perfectly acceptable, especially if one imagines that in (28 a) and (29 b), Gulliver, ‘the man-mountain’, is lying in (for him) shallow water with his knee sticking out. We consider the sentences (28 11) also to be perfectly acceptable, (29 a II) not to be acceptable and (29 b II) to be doubtful. Given this judgment, then the sentences of (28) are to be assimilated to those of (23) and those of (29) to (24). This assimilation goes further than a mere similarity in the distribution of degrees of acceptability. Embrasser and accoster are both verbs having the structure NP, V NP, and expressing in the classic formulation a ‘surface contact’ between the referents of NP,, and NPI (c$ Fillmore, 1970). Caresser (to caress), fouetter (to whip), toucher (to touch) and fiapper (to strike) are other examples of verbs of this class, which can be defined by two syntactic properties. The first is the impossibility of using the past participle as an adjective: If the structure NP, est V-P (NPI is V - ed) is grammatical, then it is an agentless passive. Thus the sentence (30)

Marie est fouettee

depuis dix minutes.

means that Mary has been being whipped for ten minutes and not that Mary has been in a state of ‘whipped-ness’ since someone stopped whipping her ten minutes ago. The second property of these verbs is the ‘locative part of the body’ structure as in example (23 a). These two properties together are the equivalent of the semantic property ‘surface contact’. Now, the verb accoster, in (31)

Le canot a accost& la berge. (The dinghy has tied up alongside

the bank.)

is a good example of the semantic intuition of ‘surface contact’ between the referents of the subject and direct object. We see that there is indeed no possible adjectival past participle, since the sentence (32)

La berge est accost&e depuis dix minutes.

must mean, in order to be judged acceptable in have not stopped tying up alongside the bank; in passive. If in order to test accoster for the other ‘locative part of the body’ structure - we choose

French, that for ten minutes boats other words, it must be an agentless property of this class of verbs - the a sentence such as (26 a), then the

198

Jean-Paul Boons

result is negative, or at best doubtful, and we are forced to treat accoster as an exception, since it behaves more like expirer (25 a) or respector (24 a) than like embrasser. If, on the other hand, a Lilliputian world of discourse is proposed, we find accoster in a sentence such as (28 a), accoster no longer constitutes an exception to the ‘surface contact’ class of verbs, and we have made an important linguistic generalization. The same type of generalisation is obtained if we compare the examples (b) of (23), (26) and (28); when NP1 of the locative structure NP,, V Lot NP, represents the ‘goal’ of the ‘theme’ NPO, the structure lui I/ Prep Dd Npb is acceptable. If the verb signifies static location, the structure is not acceptable (24~). Here again, the Lilliputian context (28b) of atterrir is preferable to (26b), since it allows us to group atterrir and couler (23b) together. A converse but equally interesting generalisation is obtained if we consider the ‘directional’ complements introduced by the preposition vers (towards) in sentences (b) of examples (24), (27) and (29). The difficulty one has in accepting (24 b II) versus (24 b I), which is acceptable, is not peculiar to couler but rather to all complements introduced by vers. Here again, the context of (29 b) is preferable to that of (27 b); the strangeness of (29 b II) is linguistically more interesting when extraneous conditions are excluded (29 b I) than when they are not (27 b I). Closer study of this opposition between ‘contact’ and ‘direction’ (i.e., ‘non-contact’) is necessary, both for complements introduced by a preposition and those not so introduced. Thus the verbs of (23) and (28) seem to be ‘contact’ verbs, accepting ‘part of the body’ structures, and those of (29) to be ‘directional’ verbs, which do not readily take ‘part of the body’ structures. This remains to be verified, however.

V. ‘Part of the body’ structures are not the only syntactic phenomana in French to vary in acceptability according to the relative size of objects. Consider the examples: (33)

Pierre (Peter (II) Pierre (Peter (b) (I) Pierre (Peter (II)?Pierre (Peter (a) (I)

a co1l.Gla page 136 d la page 137. stuck page 136 to page 137.) a co&+(lapage 136 et lapage 137 + les deuxpages) (E + ensemble). stuck [page 136 and page 137 + both pages] [E + together].) a collt+le timbre sur I’enveloppe. stuck the stamp on the envelope.) a collk le timbre et l’enveloppe (ES_ ensemble). stuck the stamp and the envelope [E + together].)

The verb caller is said to be symmetrical since together with the prepositional constructions of (I), we find constructions where the direct object is either an NP with a plural head, or conjoined NPs. Construction (II) is the one that is properly called

Acceptability,

interpretation

and knowledge of the world

199

‘symmetrical’; the referents of the conjoined NPs in object position have the same thematic role in relation to each other and to the verb. In sentence (II), ensemble (together) adds weight to the symmetrical interpretation, which is nevertheless perfectly natural in French when ensemble is not present in the sentence. Notice that there is a restriction on the symmetrical construction which is absent from the prepositional construction, namely, that the referents of the conjoined NPs must be of similar nature and, specifically, of similar size. Once again we are confronted with a problem; what is the grammatical status of sentences like (33 b II)? Obviously, this particular example is straightforward, but it is this straightforwardness that will allow us better to grasp the problem. We postulated (cJ $1) that five parameters are likely to be pertinent to an acceptability test. These were the sentences (S), value ofjudgment (j) informant (i), semantic interpretation and order of presentation of sentences. Our problem is the experimental control of these parameters. S is controlled by definition; jis free by definition; i and order ofpresentation may in theory be controlled but in practice virtually never are. Indeed they become uncontrollable when, as is frequently the case, the linguist is his own informant. Semantic interpretation is generally badly controlled, the more so since it intersects with and indeed hides the supposedly traceable border between what is linguistic and what is not. Notice that if S is unacceptable for any possible interpretation, control of the three other parameters is not essential. Thus in (34)

(a) Marie confie ses soucis ci Paul. (Mary confides her worries to Paul.) (b) Marie lui conze ses soucis. (Mary confides her worries to him. Lit. : Mary to him confides her worries.) (c) Marie se con$e h Paul. (Mary confides herself to Paul.) (d) *Marie (se lui+ lui se) conjie. (Mary confides herself to him. Lit. : Mary [herself to him + to him herself] confides.)

(34 d) is unacceptable whatever the order of presentation, whatever the interpretation and whoever the subject, and it is assumed (but by what right?) that there is no acceptable sentence in French where the pre-verbal pronouns lui and se can appear together. But for a great many sentences that the linguist must seek a judgment on, the facts are nowhere near as clear. The study of any syntactic problem leads inevitably to his constructing borderline examples whose acceptability is of necessity uncertain. This is particularly true when a sentence’s acceptability may depend on the inter-

200

Jean-Paul

Boons

pretation given and, specifically, on the extra-linguistic motivations behind the interpretation. In the traditional acceptability test, the only controlled parameter is S and the response asked for is a judgment. The linguist cannot know for which interpretation, for which world of discourse or for which set of cultural habits the judgment is given, unless of course the informant spontaneously accompanies his judgment with a phenomenistic explanation. But as these explanations are infinitely variable, and, as their homogeneousness is difficult to control, they are of little help. There can appear differences of judgment between informants that are all too easily put down to ‘idiolect’ whereas they are due quite simply to an uncontrolled interpretation. Thus (33 b II) is not acceptable for a stamp and an envelope of ordinary size but is assumed to be acceptable for a (large) stamp and a (small) envelope whose sizes are about the same. Thus differences of judgment for (33 b 11) would mean that one group of informants i, will not go beyond the everyday size of objects in interpreting a sentence whereas another group i2 has no problem in doing so. Suppose that intra-informant judgments were constant and that these two groups il and i2 have a similar difference of opinion over (26 a II), then even in these ideal conditions (which are rarely, if ever, achieved in linguistics) - i.e., with a perfect informant-judgment correlation -- we are not dealing with idiolects but rather with loyalty of a particular informant to a particular world of discourse. Whilst many linguists are happy with the recourse to the notion of idiolect when confronted with a good informant-judgment correlation for a series of examples, Gross (1974) thinks this notion can only be used if it is first defined in geographical and sociological terms, or, more generally, in terms that are independent of linguistic habits. Sentences (33 b) and (26 a) led us to present an elementary and idealised example of a plausible informant-judgment correlation where it would be absurd to resort to the hypothesis of two different idiolects. It is of course very easy to realize that one is faced with a problem such as the relative sizes of objects. One has merely to present (33 b II) in a different form (35)

(a) Gulliver a co& son timbre et l’enveloppe du Lilliputien. (Gulliver stuck his stamp and the Lilliputian’s envelope [together].) (b) Gulliver a collk son timbre et I’enveloppe du Lilliputierz avec de la colle arabique.

(Gulliver stuck gum arabic.)

his stamp

and the Lilliputian’s

envelope

[together]

with

or to present (26 a IL) as (28 a It), or (3 c) as (20) or (21) for many informants belonging to iI to change their judgments to those of is. Notice that it is possible for informants to continue not to accept (35 a) since even though the difference between the size of the envelope and the stamp has disappeared, the two NPs are asym-

Acceptability, interpretation and knowledge of the world

201

metrical in relation to colier and to its derived noun colle (glue); in our everyday world the glue is already exclusively attached to the stamp and not to the envelope. This is the reason why (35 b) will certainly be preferred; the Lilliputian world of discourse allows Gulliver’s stamps to have been shipwrecked and immersed with their owner. He must therefore use glue, whose syntactic presence in (35 b), in the form of an instrumental adverbial avec de la gomme arabique, ensures the sentence’s acceptability by reinforcing the idea of ‘symmetry’. The (relative) clarity of the above examples, and, in general, of sentences where the verb and its actants denote a concrete, or ‘technical’, process allows us to see some way into the extra-linguistic motivations which may be brought to bear on an acceptability judgment. We should assume that most test situations where subjects are presented with sentences that are syntactically or semantically more complex (completive constructions, sentences with verbs, such as ‘psychological’ verbs, denoting abstract processes, constructions with several human actants) also call on these motivations. In this case, however, they remain, temporarily perhaps, inaccessible to us, since they have their origin in cultural beliefs or conceptions of psychological verisimilitude rather than in the nature of concrete, inanimate things or in technology. (36) gives some idea of what we mean by ‘psychological verisimilitude’. (36)

Pierre (Peter (II) Pierre (Peter (b) (I) Pierre (Peter (II) Pierre (Peter (c) (I) Pierre (Peter (I I) Pierre (Peter

(a) (I)

a h&e de devoir partir. is eager to have to leave.) a h&e de pouvoir partir. is eager to be able to leave.) a soin de devoir partir. is careful to have to leave.) a soin de pouvoir partir. is careful to be able to leave.) a horreur de devoir partir. is horrified at having to leave.) a horreur de pouvoir partir. is horrified at being able to leave.)

These six sentences give rise to hesitations and differences of opinion on the part of informants which we may explain as follows: (36 a I) may appear bizarre in an ordinary world of discourse where one supposedly cannot be eager to do something that one does not wish to do and where one does not wish to do something that one has to do. Conversly (36 c II) does not conform to the convention whereby one cannot hate having the possibility or permission to do something. Thirdly, both sentences (36 b) are strange; we feel that the infinitive avoir soin cannot take a ‘passive’ semantic interpretation. This is confirmed by:

202

Jean-Paul Boons

(37)

Pierre a eu soin (? d’&re battu + de se faire battre) par Paul. (Peter was careful [to be beaten + to get himself beaten] by Paul.)

The only truly natural sentences of (36) are (36 a 11) and (36 c I). The world of discourse evoked by these judgments of bizarreness is somewhat difficult to define. We might suggest that it is the world of a conformist and rosetinted literature where it is assumed in particular that the character called Pierre is never torn between his present wishes and future aims, avoids situations where he would have to turn his back on his present desires or take up an opportunity he did not bargain for, has suppressed any masochistic element in his character, could not be dismayed by his own capabilities, etc. . . . One could very possibly search through all such literature without finding one work to fit these conditions. On the other hand, nothing allows us to suppose that we will find in this literature the bizarre sentencetypes of (36) and (37). It is therefore possible that judgments on these sentences are linguistically pertinent. The alternative would be to suppose that at some, probably unconscious, level, the subject’s view of the world is more conformist than the view reflected by the most insipid type of romantic fiction. This is of course far from impossible, but as nothing allows us to assert the extra-linguistic character of the possible effects of this level on his judgment, we shall pursue the matter no further. The hesitant nature of the above discussion reinforces the view that sentences describing some concrete process in which, with the exception of an agent, the actants are inanimate, are particularly well-suited to our purpose. With them, it is possible to control the reactions of the informant much more closely than with sentences describing some abstract, ‘psychological’ process. Indeed, we are dealing with no less a problem than that of the extra-linguistic existence of the referent, of its independence vis-d-vis the sentence in which it is evoked and, more generally, vis-ri-vis the linguistic structures and the ideologies of linguist and informant. When a sentence describes a psychological event, the very question of this independence takes us into metaphysics. But when the event described is concrete or technical, ideology and linguistic ‘code’ weigh much less heavily since the event may be ‘transcribed’ into terms that are reasonably language-independent: A series of topological schemata, for instance. Linguists and logicians are usually in agreement in saying that what is linguistically pertinent in semantics is that which remains constant for all possible worlds of discourse. The trouble with this definition is that the notion ‘all possible worlds’ is not procedurally valid. As the value of an acceptability judgment can depend on the world of discourse, (experimental) procedures are necessary in order to define certain of these with sufficient precision. In the typical test situation, where sentences are merely set up for judgment, the linguist does not control the selection on the part of the subject of a value for each of

Acceptability,

interpretation

and knowledge

of the world

203

the parameters judgment and interpretation. If he wishes to have some chance of acceding to certain worlds of discourse whose effect is relevant to acceptability, then he must also control one of these parameters. It is impossible to set up a test situation where the instructions read: ‘Make a sentence which is acceptable for such and such an interpretation’ since an interpretation cannot be stated in the absence of the sentence it applies to. It is dangerous to give instructions of the type: ‘Ts sentence S acceptable for such and such an interpretation?’ This type of instruction is frequently given, but only in simple cases where the paraphrase alluding to the interpretation does not pose a problem, and where the informant, who by now has become a linguist if he was not already one, understands what is being asked of him. There remains therefore only one type of test where two of the parameters are controlled: This is a test we may describe as the ‘forced acceptability test’ and which is defined by its instructions: ‘Find a context for which the sentence S is acceptable.‘7 The forced acceptability method is obviously not a panacea, since the fact that none of the subjects consulted succeeded in inventing a context would not necessarily mean that no context exists. Moreover, for the subjects to agree on a context favouring the acceptability of a sentence S would not imply that this were the only possible context but rather that it was the easiest to imagine. Thus in (33 b) the size factor was easier to find than the fact that the subject sees the instrumental colle as applying uniquely to the actant timbre. This method nevertheless presents the interest of allowing the linguist to present in subsequent tests conducted along traditional lines examples including the actual contexts discovered, for example, the introduction of the proper nouns Gulliver and L&put in certain examples so as to guide the subject into the world of discourse one is interested in. It is worth pointing out that the forced acceptability test is much nearer to listening or reading that the traditional test; it is the person the sentence is directed at who supplies the necessary elements for its acceptability. This point is particularly striking in the case of Gulliver’s Travels, and shows the absurdity of envisaging the calculation of the relative size of objects within the grammar. The reader automatically adapts the size of Lilliput and its inhabitants to the scenes Swift describes, without consciously recalling Swift’s ‘proportion of twelve to one’. Indeed, many critics have pointed out that Swift himself does not remain faithful to this proportion in at least one episode of Gulliver’s travels in Lilliput. When Gulliver 7. Nicolas Ruwet (1972) considers the use of the forced acceptability test but in a different, although related, perspective. This is the control of the validity of a semantic hypothesis elaborated independently of the test. Moreover,

he proposes questionably

including in the test such unungrammatical sentences as:

Pierre a tribuchi Paul. (Peter tottered Paul.)

204

Jean-Paul Boons

swims back to Lilliput towing behind him fifty of the largest men-of-war of the fleet of Blefescu, he accomplishes an impossible task. It is easy to calculate that each of these ships is one twelfth as long as a normal eighteenth century warship, i.e., twelve to sixteen feet long. This observation is however typical of critics who are over-fastidious as to the proportion of one to twelve, and who seem oblivious to the fact that Swift only gives this proportion in order better to poke fun at the Lilliputians’ calculations, and at those of his future critics of course. The Blefescu episode does not seem at all incongrous to the casual reader. The incongruity would only appear in an attempt to adapt the tale to a film or cartoon; it would be necessary to respect the proportion much more strictly for visual media. We may point out that in the case of Rabelais, this kind of criticism is in any case unthinkable since the size relationship between Gargantua, Pantagruel and the ‘humans’ varies from one to ten from one episode to another.

VI. We saw with colfer in examples (33) and (35) that two modifications of the everyday world of discourse could be necessary for a single sentence to become acceptable. Returning to planter, we shall now see that one of differing types of modification may stice in making a sentence acceptable. Variations in normal size relationships are not essential to the acceptability of (2~) and (3~). We have already discussed the paraphrase with occuper [example (16)]. The sentence (38)

Une tapisserie occupait tout le mur du fond.

(A tapestry occupied the whole of the far wall.) is most naturally interpreted if we assume that the size of the tapestry is about the same as that of the wall. We nevertheless find: (39)

Ce minuscule tableau a une faGon admirable d’occuper d hi tout seul l’immense mur du fond. (This minute painting succeeds admirably in filling the immensity of the far wall.)

The wall is ‘occupied’ by the painting thanks to some aesthetic, decorative effect, achieved for example by the judicious choice of the wall where it hangs. If the choice of occuper to characterize the semantic property of the permuted constructions is correct, then we would expect that the informant’s seeing in (2 c) and (3 c) an attempt at the description of an aesthetic effect would be all that was required to make these sentences acceptable, without changing everyday relationships of size between objects. Thus we could have:

Acceptability,

(40)

interpretation

Le talentueux architecte-paysagiste sapins dont il disposait. (This gifted landscape-gardener trees he had at his disposal.)

Then, by reducing (40), we get (41)

a littdralement plante la propriM has literally

of fir trees and by eliminating

now that the gardener’s

des quelques

the estate with the few fir

the decorative

parts of

name was Pierre, we would then have

(c) Pierre a plant& sa propriCtP d’un sapin. (Peter has planted his estate with a fir tree.)

or, for those who only accept the adjectival (3)

planted

205

L’architecte-paysagiste a plantt la propriM d’un unique sapin. (The landscape gardener has planted the estate with just one fir tree.)

Suppose (2)

the number

and knowledge of the world

version

of the permuted

sentence

(c) Sa propriM Ptait plantke d’un sapin. (His estate was planted with a fir tree.)

Seeing an attempt at the description of an aesthetic effect is therefore left up to the informant when the sentence does not contain elements which impose that interpretation. We considered the ‘size relationship of objects’ factor to be extra-linguistic. It is more difficult to consider the ‘aesthetic effect’ factor in the same way; it seems to be a semantic possibility which is widespread in language and which we find intervening in syntactic (42)

problems

constructions.

Thus we find:

Pierre a (fait + *mis) un trou dans ie sol. (Peter [made + put] a hole in the ground.)

We may nevertheless (43)

other than permuted

admire a sculpture

thus:

Le sculpteur a eu la bonne id&e de (faire + mettre) un trou ci cet endroit-li. (It was a good idea of the sculptor’s to [put + make] a hole in that particular place.)

Where cet endroit-lh refers to some part of the sculpture. Although the ‘aesthetic effect’ factor cannot be deemed extra-linguistic, the progression from (40) to (41) to (2~) demonstrates the importance of a parameter which really is extra-linguistic, the order of presentation of the examples. The progression from (20) to (21) shows the same thing. This parameter can be considered from at least two angles. Firstly, it should be possible to prove that the presentation of a series of examples from the most to the least acceptable brings down the acceptability

206

Jean-Paul Boons

threshold;

secondly,

that the reverse order pushes up the acceptability

threshold.

The

phenomenon is well-known in psycho-physics and will not be dealt with here. What does interest us, on the other hand, is the fact that a series of examples containing the same nouns and the same proper names can inculcate a story effect in the informant’s mind, in that as one example follows another, he seems unconsciously to fix in his mind the semantic presuppositions of the first examples, and refers to these presuppositions to judge the later examples. Thus, the presentation of the planter examples in the canonical-permuted order (44) (45)

Pierre (Peter Pierre (Peter

a plank un sapin dans son jardin. planted a fir tree in his garden.) a plant& son jardin d’un sapin. planted his garden with a fir tree.)

would increase the likelihood of the permuted constructions not being accepted since it would be judged in function of the semantic presuppositions accompanying the canonical construction. The canonical construction tends to evoke the idea of a garden which is distinctly larger than the fir tree and in no way suggests an ‘aesthetic occupation’ of the former by the latter. (45) is more likely to be accepted if it is presented first, the informant then being freer to imagine to himself interpretations and contexts which would favour acceptability. Similarly, it seems to us that the sentence (46)

Pierre a badigeonnk le mur d’un slogan. (Peter plastered

the wall with a slogan.)

will for the same reasons be more likely to be accepted if it is presented in isolation, (or after other, non-canonical sentences suggestive of a slogan filling up the whole of a wall) rather than after: (47)

Pierre a badigeonnk un slogan sur le mur. (Peter plastered a slogan on the wall.)

VII. Two unexplained sources of variation of acceptability appear in examples (I)-(3). The first concerns sentences (a) in which appears the collective noun sapini&e. Any sub-set of sentences (a) is non-acceptable to certain informants. We have no comment to make on this (nor indeed on other constructions with planter, such as ? Le trottoir ktaitplantk de badauds[The pavement was planted with people strolling up and down], une pelouse plant&e d’une estrade [a lawn planted with a platform], ? Le caur plantP d’un poignard [the heart planted with a dagger], etc.). The other source has to do with

Acceptability,

the preference

often expressed

interpretation

by informants

and knowledge

for sentences

of the world

(3) rather

207

than sentences

(2). We have seen in $11 that the tendency to accept an adjective N-C and to refuse the verb N-er is very general and not very stable. It would seem that every native speaker sets a limit on his stock of verbs in a more or less arbitrary manner. This tendency is particularly evident in the case of permuted constructions. The explanation we shall now propose for this fact should not be taken to be generally applicable, since it concerns only a small number of verbs. It calls on our extra-linguistic understanding of how plants grow, and of tree-growing techniques. Consider the examples : (48)

(a) Son visage trait balafrk d’une longue (estaflade + cicatrice.) (His face was cut by a long [gash + scar].) (b) D’Artagnan hi balafra le visage d’une longue (estafilade + “cicatrice). (D’Artagnan cut his [i.e., someone else’s] face with a long[gash + scar].)

Balafre is one of those nouns for which certain informants refuse the derived verb, accepting only the adjective. A gash and a scar are phenomena whose existence is a ‘cutting’ must have taken place for the dependent on the process of ‘cutting’; gash, then the scar, to appear. It seems that the permuted verbal construction contains an aspectual constraint which is absent from the corresponding adjectival construction; the N head of the de NP complement in the verbal construction (48 b) may denote the trace left by cutting as it appears during or immediately after this process. Estaflade (gash) fulfills these conditions; cicatrice (scar) does not; it is common knowledge that scar formation is a slow process. This could explain why (3) is preferred to (2); trees, and plants in general, grow at a slow rate. For the permuted construction to be accepted, it is necessary for the trees or the tree somehow to occupy the garden or the estate. In order that this occupation be sufficiently majestical, it is easiest to suppose that the trees (or the tree) are large, therefore old. As one normally imagines that at the moment of its planting the tree is young and small - literally a plant - the acceptability of sentences (2) becomes improbable through this contradiction. It is interesting from this point of view to consult the entry of plant in a dictionary such as Le Petit Robert. We find that plant has two senses. The first is ‘assemblage of vegetation of the same type growing in one place’. This definition gives us the idea of plural and the idea of a plot of land, which might well explain, incidentally, why for many informants the complement de NP of (2) and (3) must be plural. These informants treat planter like boiser (cJ [12 b]), but with some hesitation, since plant has a non-‘collective’ sense which bois does not have. This sense, the one that interests us here, is: ‘Vegetable at the beginning of its growth, which will be or has just been

208

Jean-Paul Boons

planted out’. The specification ‘has just been planted out’ goes along with the explanation given here for the possible non-acceptability of sentence (2) due to the difficulty we have in imagining a plot of land occupied by small plants. We can see how this explantation also fits the acceptability of sentences (1) and (3). The sentences of (3) are acceptable since in the adjectival construction an arbitrary length of time can have elapsed between the process of planting and the moment that interests us (always assuming the process of planting took place, since this is in no way guaranteed by the adjectival past participle). The trees can therefore be imagined with the height and age required to satisfy the semantics of ‘occupation’. Sentences (I) are acceptable in any case since the canonical construction does not appear in the semantic reading we have been discussing: The fir tree planted in (I) can be as small as one cares to imagine. The effect of this constraint appears clearly in the examples with balafrer (48), where cicatrice cannot be the metonymy of balafre fraiche (fresh cut), it is less clear in (2), where sapin is seen as a young plant by some informants, as a fully grown tree by others. This would account for the variations of judgments we have for (2).

VIII. Examples (I)-(3), containing the verb planter have enabled us to took at some extra-linguistic factors capable of intluencing an acceptability judgment. Leaving aside the factor of the order of presentation of examples (cf. §VI), which in any case is by definition a parameter intervening in an acceptability test, these factors are: - knowledge (or ignorance) on the part of the subject of elementary facts about the world of discourse evoked by the examples (relative size of objects, growth of plants, tree-growing techniques or, in the case of [35], the presence of glue on a stamp). - the informant’s seeing in the example an attempt at the description of some aesthetic effect. We have seen that the ‘relative size’ factor is relevant to the acceptability of permuted constructions since it ties in with the semantic property of ‘occupation’ of the place represented by the direct object. Similarly, knowledge of the way objects change in time (growth of plants, healing of cuts) is relevant to the aspectual constraint which differentiates the verbal from the adjectival constructions in permuted structures. Or again, the presence of glue on a stamp is in contradiction with the necessity for both NPs in the symmetrical structure (35 a) to have identical roles vis-d-vis the verb. In these three cases, a manifestly extra-linguistic ‘semantic’ factor influences acceptability thanks to its connection with a genuine semantic property which is linguistic and directly implied by the syntax of certain structures. We may assume in the practice

Acceptability,

interpretation

and knowledge

of the world

209

of what we called ‘the hunting of the sense’, that extra-linguistic concepts defined by notions about the physical nature of things or by socio-cultural categories will come more quickly to mind than concepts necessary to linguistic description, or than the system which binds them. In research, the extra-linguistic will screen off the discovery of what is linguistic. Thus the ‘aesthetic effect’ factor proposed in paragraph VI in relation to the semantics of ‘occupation’ is suggestive more of a sociocultural than a linguistic category and should make way for some more abstract category which would allow us to describe a larger number of phenomena. Indeed, in (19) there is nothing necessarily ‘aesthetic’ in the occupation of the top of the wardrobe by the vase. Similarly, expressions such as mettre des trous[put holes into: C$ (45)] can appear in contexts which are in no way aesthetic; we can have for example: (53)

I1 me faudra diplacer cette boutonni&e. (I’ll have to move that buttonhole )

This sentence which is bizarre if taken literally, seems to have been arrived at by some sort of abstraction of or retreat from the concrete process it suggests: ‘Moving’ a buttonhole can only entail practically speaking our making it disappear (by sewing it up, for example) and by creating a new one near the place where the original one was; but if the article of clothing and its buttons are considered as one whole we are dealing with the same buttonhole since it must correspond to one and the same button. The terms ‘abstraction’ and ‘retreat’ (related to aesthetic: ‘Perspective’) constitute a provisional means of grasping a hazy intuition. Whereas the ‘relative size’ factor is extra-linguistic and affects acceptability the ‘aesthetic effect’ factor is probably in no way extra-linguistic. It is probably quite simply a badly defined concept consisting in reality of the freedom the speaker has, which, depending on the sentence, is more or less limited, to give himself a certain perspective in apprehending what is described. But replacing the local, socio-cultural notion of ‘aesthetic effect’ by a more general concept is not self-evident. It appears that the concepts to be discovered are in general extremely abstract, not well-known and such that a language or culture does not contain nouns or succinct noun phrases allowing us unequivocally to ‘represent’ them. In the fields of thematic categories or of aspect, the ‘semantic’ concepts to be unearthed probably belong to the unconscious. Their part is played less in the meaning of the sentence than in what we may call the meaning of the syntax of the sentence. We can ask ourselves two very different questions concerning meaning and ‘semantics’ (i propos of the structure of simple sentences. The first is about what the sentences ‘mean’. It presupposes a semantics of ‘communication’ and is of little interest to us here; even assuming that it is possible to construct such a semantics, which is by no

210

Jean-Paul Boons

means sure, it would doubtless be largely socio-cultural. The second is about the way the incredibly diverse constraints which define the structure of simple sentences may be acquired by a child. This question englobes the meaning of syntax: It postulates a body of ‘meaningful’ and unconscious concepts which are supposed to facilitate language acquisition. This body of concepts would mediate between syntax and knowledge of the world. Knowledge of the world is considered to be extra-linguistic not in the sense that its organisation is independent of linguistic structure but in the sense that it is excluded from the grammar. The fact that some factor has been judged to be extra-linguistic does not mean that the linguist should regard it as being of no interest. On the contrary, it is knowledge of correlations between certain aspects of the way different speakers view the world and their judgments of sentences that allow them to approach the semantic factors which characterize the syntax of those sentences. An extra-linguistic factor can be disregarded only if its impact on syntax is understood. We are dealing with a frontier, and we must place this frontier not between syntax and morphology on the one hand and semantics on the other hand (between ‘form’ and ‘content’) but - given what is meaningful within syntax and morphology - between what is linguistic and what is not, i.e., between what should and what should not appear in the grammar. This frontier is none other than the frontier of the object of linguistic science. To draw it precisely, if only in places and piece by piece, supposes that we are certain of positions which are ‘near’ the frontier to be drawn and sure of their relationship to it. It is no less important to occupy positions just this side of the linguistic frontier than to occupy those in the linguistic world beyond.

Acceptability,

interpretation

and knowledge

of

the world

211

REFERENCES Boons, J. P. Guillet, A. and Leclere, C. (1974) La structure des phrases simples en francais (constructions non completives). Doctoral thesis, Paris. Chomsky, N. (1970) Remarks on Nominalisation. In R. Jacobs and P. S. Rosenbaum (Eds.), Readings in English transformational grammar. Waltham, Mass., GinnBlaisdell. -(1972) Some empirical issues in the theory of transformational grammar. In P. S. Peters (Ed.), Goals of linguistic theory. Englewood Cliffs, N.J., Prentice-Hall. Fillmore, C. J. (1970) The grammar of hitting and breaking. In R. Jacobs and P. S. Rosenbaum (Eds.). Readings in English trunsformutionul grammar. Waltham, Mass., Ginn-Blaisdell. Gross, M. (1969) Lexique des constructions complttives. Laboratoire d’Automatique Documentaire et Linguistique, C.N.R.S., Paris, mimeographed.

(1974) Me’thodes en syntuxe. Paris, Hermann. Gruber, J. S. (1965) Studies in lexical relations. Doctoral thesis, M.I.T. (distributed by the Linguistics Club of Indiana University). (Eds.), Readings in English trunsformutionul grammar. Waltham, Mass., Ginnt Hall-Partee, B. (1971) On the requirement that transformations preserve meaning. In C. J. Fillmore, and D.T. Langendoen (Eds.), Studies in linguistic semantics. New York, Holt, Rinehart and Winston. Kayne, R. S. (1974) French syntax - The transformational cycle. Cambridge, Mass. M.T.T. Press. Matthews, G. H. (1968) Le cas tcheant. M.I.T., mimeographed. Ruwet, N. (1972) Les constructions factitives en francais. In ThPorie syntuxique et syntuxe du fruncais. Paris, Editions du

__

Seuil.

R&urn6

Grace aux methodes transforrnationnelles, l’acceptabilite est devenue la mesure et le systeme de classification par excellence de la linguistique. De tous les parametres possibles qui peuvent influencer les resultats de ce test, le probltme soumis au lecteur darts cet article est celui de I’interpretation semantique. L’auteur essaie de dtmontrer la necessitt de dissocier tout facteur extra-linguistique (y compris I’univers du discours de celui qui parle, la culture a laquelle il appartient, l’importance de certains facteurs de connaissance ou ignorance du monde) dune interpretation hypothetique linguistiquement pertinente Ctroitement Ii&e a la syntaxe et a la morphologic. Les concepts semantiques extra-linguistiques sont ‘naturels’, peuvent Ctre facilement evoques

et filtrent la recherche des concepts semanticolinguistiques beaucoup plus abstraits, moins connus et probablement inconscients. Ceci represente la limite qu’il faut dtablir dans la semantique entre ce qui appartient au domaine de la linguistique et ce qui n’y appartient pas, c’est-a-dire entre ce qui doit, et ce qui ne doit pas &tre considere comme appartenant a la grammaire generative. II faut, d’ailleurs, insister a ce sujet sur le besoin qui existe dans la linguistique de contrbler et done de varier les univers du discours qui peuvent modifier I’acceptabilite dune phrase donnee. Dans cet article, la discussion est centree sur les divers usages du verbe ‘planter’.

Meanings and concepts: A review of Jerrold J. Katz’s Semantic theory*

HARRIS B. SAVIN University of Pennsylvania

theory is the latest version of the theory of meaning that Mr. Katz has elaborated and defended in numerous publications, beginning with The structure of a semantic theory (1963, in collaboration with Jerry A. Fodor). With respect to almost every detail, the present theory is different from the earliest and best-known versions, but the changes are changes of detail, in response to detailed criticisms; in spirit the theory has remained much the same. During the middle 196Os, Katzian semantics was one of the central topics in generative grammar. More recently most generative grammarians have interested themselves in other questions and, as the center of linguistic interest has shifted away from him, he has increasingly come to seem to psychologists a peripheral figure in linguistics. Nonetheless, his work remains the only serious, sustained attempt to elaborate a theory of meaning within the framework of generative grammar.l For this reason, it deserves careful study by students of cognition - and especially by those psychologists who have embraced Chomskian syntax with enthusiasm but who, unless they appeal to Katz, have no answer to the question What does all this syntactic machinery have to do with language as a medium of expression and communication? In this review, I will be concerned primarily with the relationship of Katzian semantics to cognitive psychology. In order to discuss this question, however, I will first present Katz’s conception of the subject matter of semantics and then ask whether Semantic

* 1972, New York, Harper and Row.

1. There is a widespread but mistaken belief among psychologists that the generative grammarians who concern themselves especially with questions of meaning are the proponents of a theory called generative semantics. In fact, generative semantics is not a doctrine about meaning at all but about the best way to write

grammatical rules that will relate semantic representations to surface structures. In their discussions generative semanticists make vague but more or less Katzian assumptions about the nature of the semantic representations and disagree with Katz about the rules that relate the uncontroversial semantic representations to uncontroversial surface structures. Cognition (2)2, pp. 212-238

214

Harris B. Savin

his theory is successful by his own criteria. Finally, we turn to the question interest - the relationship of Katzian semantics to cognition.

of central

1. The subject-matter of semantics Meaning, for Katz, is a theoretical construct which is supposed to explain a certain range of semantic facts (or, not to prejudge one of the important arguments between Katz and his critics, a certain range of purported facts). The data of semantic theory ambiguity, contradiction, for Katz are ‘phenomena’ - his term - of paraphrase, implication and so forth. In all, Katz gives 15 such kinds of semantic phenomena in a list (pp. 5-6) which is meant to be merely illustrative, not exhaustive. To begin with an example, it is a datum for Katz that (la) is a paraphrase of (means the same thing as) (1 b). The example requires some discussion. (la) (lb)

John bought the book from Mary. Mary sold the book to John.

In saying that (la) means the same thing as (I b), Katz does not claim either that they are used in the same circumstances or that they have the same effect on the listener. For example (la), unlike (lb), is a natural answer to (2). (2)

What did John do?

Such observations show that sameness of meaning in Katz’s sense cannot be identified with sameness of usage. Nor can sameness of meaning be identified with sameness of reference; (3a) and (3b) refer to the same creatures but do not have the same meaning. (3a) (3b)

creature creature

with a heart with a kidney

The identity of the referents of (3 a) and (3 b) is a matter of zoological fact. If someone says that he saw a creature with a heart but without a kidney, he is not talking nonsense but claiming to have discovered a new zoological species. If, on the other hand, someone reports that (la) was true but (lb) false, then, assuming him to be a fluent speaker of English and assuming that John and Mary denote the same individuals in both sentences, we cannot understand this as the report of a new economic phenomenon. Rather, the speaker has contradicted himself by denying in one clause the same thing that he affirmed (in different words) in another. What, then, does Katz claim when he says that (1 a) and (1 b) have the same meaning? Roughly, that, by virtue of the linguistic structure of the sentences, they have the

Meanings and concepts

215

same cognitive content. Their differing forms give them different uses, for example, as answers to different questions, or as lines in different bad poems, but Katz claims to be able, by reflecting on the two sentences, to determine that, these differences in usage notwithstanding, the meanings are the same. Two points about this conception of meaning. First, it is concerned with a much narrower range of phenomena than are the conceptions of many psychologists, who identify meaning with use. Second, the only source of data about the sameness of meaning is subjective judgment. It is evidently not possible to find any sort of behavioral criterion for sameness of meaning, nor for most of the other semantic phenomena of Katz. We postpone until later the question whether a theory of meanings, in this sense, is interesting. The first question that arises is whether the data derived from subjective judgments are reliable enough to be taken seriously for any purpose. The data are, for the most part, reliable. Once one sees, from a few examples, what Katz means by synonymy, by contradiction, or whatever, one can identify new cases of the phenomenon without much difficulty. There are, to be sure, borderline cases about which one is uncertain, but there are plenty of clear cases of each of his types of phenomena. That these judgments are reliable should come as no surprise. They are mostly judgments that we all make confidently every day when we decide that two people are saying the same thing in different words or that an author has contradicted himself or that a statement fails to answer a question. It is hard to see how any kind of scientific or philosophical discussion could take place without such judgments. If they are legitimate in discussions of physics and philosophy, why not in linguistics and psychology? Katz’s work cannot, therefore, be dismissed because the data are methodologically suspect - not unless we reject the data out of hand on the grounds of subjectivity without regard to either their reliability or their inherent interest. We turn next to the theory he proposes to account for his data.

2. The semantic theory Katz supposes, as surely everyone must, that the meaning of a phrase or sentence depends upon the meanings of its constituent words and upon the relationships among those words. In broad outline, then, a semantic theory must have two components: A dictionary, which gives the semantic representation (the meaning) of each word, and a system of ‘projection rules’ for determining the meanings of phrases and sentences from the meanings of their constituent parts, taking into account the semantically relevant relationships between the words. As for the semantically relevant relation-

2 16

Harris B. Savin

ships (which will be the basis for distinguishing between the meanings of The man bit the dog and The dog bit the man, in which the same words are related differently to one another), they are, according to Katz, the grammatical relationships specified in the deep structures of a grammar of the kind described by Chomsky (196Qp The test of such a theory will be its success at explaining the semantic data. Katz requires for example that the theory assign the same semantic representation to sentence (la) and (lb), that it associate two distinct semantic representations with a sentence that is two ways ambiguous, and that it provide the basis of a definition of contradiction meeting the requirement that two sentences are mutually contradictory if and only if their semantic representations bear a certain well-defined relationship to one another. These requirements are a good deal more stringent than was apparent from the examples in Katz’s early work. For example, John is an only child is contradicted not only by John is not an only child and John has two brothers, but also by Bill did not know John’s sister’s address. The network of logical relationships among sentences is extremely rich, and it is the richness of this network that is Katz’s best answer to the charge of triviality that was often raised against his early work. In broad outline, then, this was Katz’s conception of semantic theory in 1963, and it remains his conception today. The broad outline is plausible enough but hardly specific enough for serious discussion. In addition to the broad outline, Katz gives us specific examples of dictionary entries, projection rules and definitions of logical relationships and the like. These are easy enough to discuss, but detailed discussion of them is often beside the point. Katz has no particular stake in any of his proposed dictionary entries, rules or definitions. He responds to counter-examples by inventing new dictionary entries, rules and definitions (sometimes not merely new in detail but with new formal properties) and by pointing out that he had offered the old examples only as examples and not even necessarily as typical ones. After a decade of such modifications, scarcely any details of the original theory remain. It is, therefore, not very interesting to discuss the details of the latest version as if they were meant to be part of a definitive theory. Katz himself points out various facts that he cannot yet explain, and the careful reader can discover many more. In an attempt to give a more informative picture of the evolution of Katz’s semantics, I will first discuss some dictionary entries and projection rules that he and Fodor pro2. Exactly how the deep structures do this will become clear when we consider categorized variables. For the present, we call the reader’s attention, to the fact that this assumption provides the connection between Katz’s semantics and Chomskian syntax. This assumption is rejected outright, along with Chomsky’s syntax, by proponents of generative semantics,

and is accepted only in part by Chomsky himself. But all these linguistic positions have a great deal more in common than such, capsule summaries could suggest. With respect to the issues of principal concern in this review, there is far more agreement than disagreement among Katz, Chomsky and the generative semanticists.

Meanings and concepts

2 I7

posed in 1963 and then show how the most recent theory is able to account for various semantic facts that embarrassed the earliest version. Consider first the dictionary entry for bachelor given by Katz and Fodor (Fig. 1; their Fig. 5). Each path from the top of this tree to the bottom is supposed to correspond to a sense of the word bachelor. The symbols in parentheses are known as semantic markers, and the bracketed symbols are called distinguishers. Figure 1. From Katz and Fodor, 1963, p. 190

noun I

I

(Human) I

-

[who has the first or lowest academic degree]

(Male) ,I, [who has never married]

I

(Animal)

(Young) I

[knight serving under the standard of another knight]

I

(Male) I

1 Oroung) I

[fur seal when without a mate during the breeding time]

Postponing for a moment the distinction between semantic markers and distinguishers, ‘The semantic markers and distinguishers are used as the means by which we can decompose the meaning of a lexical item (in one sense) into its atomic concepts, thus enabling us to exhibit the semantic structure in a dictionary and the semantic relations between dictionary entries’ (Katz and Fodor, 1963). The projection rules operate on such representations of the meanings of words, giving as a result the semantic representations of phrases and sentences. As for the distinction between semantic markers and distinguishers, the formulation given by Katz and Fodor was, as Katz now recognizes (section 3.5), internally inconsistent, but the general idea was that semantic markers mark conceptual distinctions which play a systematic part within semantic theory, but that two non-synonymous wordsmight have precisely the same semantic markers and yet differ in denotation. We will not explore the rationale for the distinction thoroughly because it plays no part in the theory-neither in early versions nor in late. ‘For present purposes, we make no distinction between semantic markers and distinguishers. The latter are treated

2 18

Harris B. Savin

as special cases of the former’ (Katz, 1972, p. 73). In all definitions, rules and discussions throughout his latest book, semantic markers and distinguishers are treated alike, except in chapter 3, section 5, where Katz criticizes his earlier treatment of the difference between the two kinds of constructs and proposes instead that semantic markers mark conceptual distinctions while distinguishers mark perceptual distinctions. Katz’s views of the relationship between perception and conception will concern us in section 3, where we discuss the relationship of his semantics to cognitive psychology. Within the semantic theory itself, the distinction plays no part. Although the formal difference between semantic markers and distinguishers plays no part in the semantic theory, the particular distinguishers Katz used in his early work were quite unsatisfactory, as many of his critics have pointed out. The problem with these early distinguishers is worth exploring because it is a particularly clear instance of the way in which criticism of his examples often fails to touch his important theoretical claims. Consider the definition of bacheZor which has the semantic markers (human) and (male), and the distinguisher [who has never married]. Distinguishers are supposed to be unanalyzable, but it is obvious that this proposed distinguisher has a complex conceptual structure and that a theory which does not analyze who has never married cannot claim to have given a very interesting analysis of bachelor. Granting, then, that we have been given a bad definition of bachelor, the question is whether its badness testifies to a bad semantic theory or merely to bad lexicography. If the objectively bad definition is good enough for Katz’s theoretical purposes, then the theory is no good. But in fact, by Katz’s own criteria, what we have here is a piece of bad lexicography. In order to account for such semantic phenomena as the contradictoriness of married bachelor and bachelor’s wife, it will be necessary to analyze the allegedly unanalyzable distinguisher. Indeed, it is a point in favor of the general framework that it has caught its authors at a great deal of bad lexicography. In addition to semantic markers and distinguishers, which give the meanings of words, many dictionary entries contain information about the semantic properties of other words that the given word may combine with. ‘For example, consider the Shorter Oxford English Dictionary entry for the word honest: . . 3. of persons: of good moral character, virtuous, upright, . . . of women: chaste, “virtuous”, . . . . The “of persons” and “of women” are intended to indicate that the senses that follow them apply only under the conditions they specify’ (Katz and Fodor, 1963, p. 191). Katz and Fodor formalize this notion in the obvious way by including in their dictionary entry for the first of these two senses of honest the selection restriction that this sense is possible only when honest modifies a noun which itself has the semantic marker (human), and by restricting the second sense of honest to nouns having both the markers (human) and (female).

h1eaning.v and concepts

219

A complete dictionary entry, then, consists or phonological and syntactic markers and, for each sense of the defined word, suitable semantic markers, distinguishers and, possibly, selection restrictions. The projection rules operate on the meanings (i.e., semantic markers and distinguishers) of words to give the meanings of phrases and sentences. For example, one projection rule combines the readings of an adjective with the readings of the noun it modifies. The reading for unmarried man, for example, includes all the semantic markers of unmarried together with all those of man. All the projection rules discussed in Katz and Fodor are of this combinatorial type. The meaning of any phrase is a set of semantic markers formed by taking the settheoretic3 union of the semantic markers of the constituents of that phrase. Katz and Fodor suggested that other types of projection rules might be required if the theory were to account for the semantic characteristics of certain grammatical transformations. However, Katz and Postal (1964) argued persuasively that the syntactic theory which gave the best account of purely syntactic phenomena was one in which all transformations were meaning preserving. It was therefore possible to let projection rules apply only to deep structures (phrase markers which have not been operated upon by any syntactic transformations). All the projection rules proposed by Katz and Postal in fact combined readings of constituents of a phrase to give readings of the whole phrase. (I do not accuse them of having enunciated the principle that all projection rules were of this type.) To summarize, each dictionary entry proposed in Katz’s early papers consisted of sets of markers. Furthermore, the semantic markers in these examples had no internal structure. Each projection rule simply combined sets of markers, when the combination was permitted by the grammatical structure and by the selection restrictions. Many readers took these characteristics of Katz’s examples to be essential characteristics of the theory. Various papers were written arguing that, if the dictionary entries and projection rules were of this form, a satisfactory theory could not result. Katz replied to each of these papers, accusing his critics of having misread him by taking his examples, rather than his text, as defining the theory. The result is a seem-

3. To be quite accurate, all the projection rules in Katz and Fodor refer to strings, rather than sets, of semantic markers, the difference being that theelements of a string, unlike the elements of a set, are ordered. The ordering, however, is never used in the definition of any semantic concept, and, indeed, it is impossible to see how they could define such notions as synonymy except by assuming that the order of semantic markers does not matter. Unfortunately, if we ignore order altogether, then

the semantic representations of The man bit the The dog bit the man become indistinguishable. There are, of course, alternatives to the two extremes of always ignoring order and always regarding changes of order of semantic markers as changing meaning. We do not explore any of these possibilities here because the entire problem is eliminated by the introduction of categorized variables, which we consider shortly. dog and

220

Harris B. Savin

ingly endless debate (it occupies more than a chapter of the latest book) between Katz and various critics - not a debate about the linguistic data or about what constitutes an adequate explanation of the data, but a debate about whether a theory that handles certain semantic facts is essentially different from or essentially the same as the theory of Katz and Fodor. The reader who wants to read this polemical literature can fmd it by pursuing the references Katz gives in chapter 3 of Semantic theory. In this review, I will not attempt to decide how different a theory can be from the original theory of Katz and Fodor and yet count as essentially the same theory. Rather, I will simply survey the kinds of semantic facts - some first discussed by Katz and some first by his critics - that require dictionary entries and projection rules of other kinds than the ones originally used by Katz and Fodor.

2.1

Selection restrictions and syncategorematicity

A loud noise is one that excites the auditory system by more than a certain (contextually determined) amount. A loud necktie is one with many bright colors. The obvious way to account for these facts is to have two readings for loud, each with a suitable selection restriction - the formal device Katz and Fodor used to explain parallel facts about honest. Consider, however, such phrases as good knife, good watch, good screwdriver, and good screwdriver to open paint cans with. A good knife is one that cuts (something unspecified) easily; a good watch keeps time accurately; a good screwdriver drives screws well; a good screwdriver to open paint cans with opens paint cans easily. Obviously, the device that worked for honest and loud does not work here. If we write four definitions of good, and four selection restrictions restricting the application of each definition to the right sort of context (respectively, cutting implements, timepieces, tools for driving screws and tools for opening paint cans), we will not be able to stop. Nearly every new context of good will require a new selection restriction, and there will be an infinite number of different contexts. In short, a theory with no other resources for constructing dictionary entries cannot work. The problem is that good does not contribute to the meaning of good knife (say) in the same way that red contributes to the meaning of red knife. Instead, it seems to be a property of the meaning of knife that its normal evaluation is with respect to ease of cutting. Good contributes to the phrase only the detail that the evaluation is positive, and the evaluated attribute is determined by knife itself (unless there is an explicit statement of another evaluated attribute, as in good knife to drive screws with. Katz (1964) proposed the following formalization of this account of good: In the dictionary entry for each word like knife and watch - for each word, that is to say,

Meanings

and concepts

for which there is a normally evaluated attribute - there is a semantic sort, for example (4), which contains the information that (4)

221

marker of a new

(eval.,,, :(cutting))

knives are normally evaluated according to how well they cut. The dictionary entry for good specifies only that the evaluation is positive. A new projection rule operates on the markers for good knife to produce (5). (5)

(eval.,,,:(cutting))

(positive)

This treatment of good differs in two important ways from anything in earlier works of Katz: First, semantic markers like (eval.,,, : (cutting)), have internal structure and, second, the projection rule for good does not merely form the set-theoretic union of two sets of markers but rather operates on a marker, changing it into a different marker. In Semantic theory, Katz proposes numerous semantic markers with complex internal structure and numerous semantic markers which are changed by projection rules (and by various other rules, in chapter 7, which Katz, for obscure reasons, formulates not as projection rules but as parts of dictionary entries). As a result of these changes the theory as a whole takes on a completely different appearance. The sets of unstructured semantic markers and the combinatorial projection rules of Katz and Fodor were, depending on one’s point of view, either engagingly simple or embarrassingly simple-minded. Readers who accept the Gestalt psychologists’ critiques of atomistic perceptual theories might well have seen in the examples of Katz and Fodor the linguistic counterpart of the simple-minded atomism of Titchener (without, of course, Titchener’s theory of meaning). Such readers, whatever else they may think of Semantic theory, will not find it simple-minded. The most important formal device for solving a variety of problems that could not have been solved with rules of the earlier types is the categorized variable. Consider again sentences (1 a) and (1 b), which are paraphrases. Because the semantic representations of sentences, in earlier versions of the theory, were in large part determined by the syntactic structures of the sentences, the synonymy of (la) and (lb) seemed inexplicable unless they had identical deep syntactic structures - unless either John was the subject of (lb) or Mary was the subject of (la). Since there are very good syntactic grounds for rejecting any such syntactic solution, an explanation of the synonymy of (la) and (lb) required a change in the semantic theory. Katz’s new explanation, somewhat simplified, runs as follows. Both buy and se0 describe exchanges between two people - exchanges which can be described by specifying the states of affairs at two times - call them tl and tz. At tr, person, possesses a certain thing, and person, possesses a sum of money. At t2 (which is later than t,) person, has the money and person, the thing. So far, buy and sell are identical. Now to the

222

Harris B. Savin

difference between them: The grammatical subject of buy is person, and sell is person, ; the indirect object of buy is person, and the indirect is person,. The only difference between buy and sell, that is to say, semantic roles (viz., buyer, seller) are paired with syntactic functions indirect object). Somewhat more formally, we may represent the part entry for buy which shows the state of affairs at t I as follows in (6).

(6)

[direct object]

[object of prepositionfrom]

X

possesses {(physical

X

of object))

((human)

and (infant))

[object of prepositionfir] possesses (i

the subject of object of sell is in the way (viz., subject, of the lexical

[subject]

X ((mow)>

X

of 1

((human)

and (infant))

The new formal device - the categorized variable -consists of a variable - for example, each occurrence of x in (6) with a grammatical function written above, in square brackets, and a selection restriction, below, in angled brackets. A projection rule interprets such a variable as an instruction to replace the variable x with the semantic representation of the part of the sentence which meets the conditions stated in square and angled brackets. For example, the last categorized variable in the definition of buy, [subject] is to be replaced by the semantic representation of the X in (6), ((human) and (infant)} subject of buy, provided that the subject contains the semantic marker (human) but not the marker (infant). (If the subject noun phrase does not meet these selection restrictions, the sentence is semantically anomalous and gets no interpretation.) Definition (6) is a highly simplified version of Katz’s definition, and it contains many symbols (e.g. [direct object], [object ofprepositionfrom]) which must be given a further analysis. The definition given by Katz fills a whole page. As an explanation of the synonymy of (la) and (lb) - and a paradigm for explaining numerous other paraphrastic pairs which presented analogous difficulties for earlier theories - the device illustrated in (6) clearly succeeds. Note that this definition of buy is only possible because of the internal structure of semantic markers in the new theory. The atomistic definitions and rules of Katz and Fodor could not combine the reading for John, say, in (la) with the rest of the sentence while continuing to show that John was the possessor of the book at time, and the possessor of the money at time,. The internal structure (represented in (6) by the nesting of semantic markers within other semantic markers) preserves all this information in the structures which result after all the categorized variables are replaced by

Meanings and concepts

223

the appropriate readings of noun phrases. Space does not permit thorough discussion of other details of Katz’s theory. Semantic theory includes an extensive, although incomplete, treatment of the semantics of tenses and adverbs of time, a chapter on questions and more or less detailed analyses of several other problems within semantics. Although it is easy to point out shortcomings of many of Katz’s specific proposals, none of them appears to be irremediable within his general framework. Critics of Katz and Fodor attempted to show that the sort of theory they had in mind could not possibly accommodate what they themselves regarded as semantic data. Such demonstrations always relied on the extremely simple nature of their rules and dictionary entries. It is unlikely that anyone will make such arguments against Katz’s latest book because it is evident that Katz is making only very weak and plausible assumptions about the form of a semantic theory and, within the framework defined by those assumptions, he has enormous freedom to revise details in order to handle new evidence. Criticism of specific dictionary entries, then, is too petty to be very interesting. The most important question is whether the domain of semantics as conceived by Katz is of any general importance. We turn to this in section 3, after concluding the present section by attempting to indicate the distance that remains between Katz’s present theory and his objective - a theory that would explain all the semantic phenomena, as he conceives them. It goes without saying that the dictionary is woefully incomplete. Of the 100,000 or so words in a good desk dictionary, Katz has given careful consideration to perhaps a dozen. As for projection rules, Katz says (p. 116) that only one rule seems to be necessary, given the use of categorized variables. But other rules which do the same sort of thing as the projection rule appear in dictionary entries (see especially footnote 13 on p. 314). Another projection rule-like principle appears (on p. 107) ‘as part of the definition of “categorized variable”.’ It is unclear, therefore, what content the claim has that only one projection rule is needed. But more is required to complete the theory than mere diligent elaboration of a lexicon and projection rules. Theoretical problems of many sorts remain, and unless they are solved the theory will be unsatisfactory. To give some sense of the kinds of theoretical problems that remain unsolved, I will give three examples, none of which is necessarily unsolvable but all of which will require for their solution something more than the mere working out of details along the lines already begun. a. Transitivity. The ancestor of one’s ancestor is one’s ancestor. This property of certain relations is known by logicians as transitivity. More precisely, a relation R is transitive if and only if (7) is true. (7)

For all x, y, and z, xRy and yRz imply xRz.

It is easy enough, obviously, to include a statement like (7) in a dictionary entry and to

224

Harris B. Savin

formulate new projection rules and new sub-clauses of definitions of concepts like analyticity which will interpret (7) correctly. But a solution of that sort would be no solution at all. An intransitive relation with the meaning of ancestor is a conceptual impossibility, like a square circle or a married bachelor. To say the same thing the other way round, the transitivity of ancestor is a consequence of its meaning, and a dictionary entry which treats its transitivity as an independent property, even if that entry can be used to explain all the semantic phenomena is unsatisfactory. Katz (p. 197) remarks that the problem is difficult and not amenable to the kind of explanation he gives (pp. 193-197) for reflexivity and symmetry, but he does not suggest an alternative approach. b. The relationship between selection restrictions and readings. Honest, on one reading, means innocent of illicit sexual intercourse. This sense is applicable only to women, a fact that Katz marks with the selection restriction ((human) and (female)). The restriction to females is quite clearly arbitrary and without any semantic explanation. Similarly, addIed means rotten but can be said only of eggs. The selection restriction is unrelated to any other semantic property of addled and is rightly represented in the lexicon as a separate fact. Compare waterproof, which has the selection restriction ((physical object)). Here, the selection restriction expresses not a conventional restriction of the possible contexts of waterproof but a restriction which is related in some important but obscure way to the meaning of waterproof. A waterproof integer is a conceptual impossibility (assuming both words have their most common senses), not a mere breach of an arbitrary conventional restriction on usage. The overwhelming majority of selection restrictions appear to be of this second kind, related intrinsically to the meanings of the words they restrict. Yet Katz treats all selection restrictions the same way, as separate, independent features of lexical entries. As far as the theory is concerned, any selection restriction might be combined with any meaning, so that there might be an adjective with the meaning of waterproof and the selection restriction ((abstract idea)). This defect of the theory does not make it incapable of explaining the ‘semantic phenomena’ that Katz takes as defining his subject matter. But a failure of this kind does seriously limit the possible relevance of such a theory to broader questions of cognitive psychology. We return to this subject in section 3. c. The syntax and semantics of verbs. Verbs differ greatly from one another in their syntactic properties - in whether or not they take direct objects, indirect objects, prepositional phrases and adverbial modifiers of various types; also, in what syntactic transformations they undergo. Indeed, there are scarcely two verbs with identical syntactic properties. More important, verbs that have similar syntactic properties are as a rule similar also in meaning, as one can see by looking at practically any list of verbs compiled by students of syntax because of their syntactic similarities. There are,

Meanings

and concepts

225

evidently, rules of syntactic-semantic correspondence, and they must be of great help to the child in mastering the several dozen syntactic properties of each of some three thousand verbs of his language (cf Gross, 1971; Gleitman, 1970). Katz’s theory leaves him completely free to mate any semantic features with any syntactic properties. In this respect, his theory is at the opposite extreme from Fillmore’s (1968) case grammar. Fillmore claimed that the syntactic properties of verbs were completely determined by their meanings. This is too strong, as shown by buy and sell, which have the same meaning but different assignments of semantic roles to syntactic functions, and certain other minor syntactic differences. (See Dougherty, 1970, for an extensive discussion of the problems of case grammar.) The truth lies somewhere between. Syntactic properties of verbs are neither completely determined by their semantic properties, as Fillmore claimed, nor completely independent, as Katz’s formalism allows them to be. See Vendler (1972) for some syntactic-semantic correspondences which are particularly interesting because they concern overt syntactic correlates of such elusive mental phenomena as believing, wondering, doubting, and the like. Observe that all three of these unsolved problems concern the general properties of lexical entries. Semantic features in fact cannot be freely combined with each other (e.g., the transitivity of ancestor is not an independent feature of its meaning), with selection restrictions, or with syntactic features. To date, Katz has formulated no principles concerning the two latter types of restrictions, and nothing but some very simple ‘redundancy rules’ for the first type - a rule, for example, to the effect that every lexical entry with the semantic feature (human) also has the feature (animate). Clearly there remains to be developed a theory of the structure of well-formed lexical entries. And this as yet undeveloped part of Katz’s semantics is the part that, if it can be developed successfully, will have the most important bearing on psychological questions, both for what it will suggest about language acquisition and, more important, what it will suggest about the structure of thought. There are many reasons to doubt Katz’s thesis that whatever can be conceived can be said in a natural language. (Try to describe your wife’s face well enough for a stranger to pick her out of a crowd.) But, whether or not there are unspeakable thoughts, the structure of meaning is intimately related to the structure of thought. The lexical impossibilities we have been considering here - for example, a word that differs in meaning from ancestor only in being intransitive, or a word meaning waterproof but predicable of integers-are surely the linguistic counterparts of conceptual impossibilities. Thus we come to the issue of central concern in this review: The relationship of Katz’s semantics to cognitive psychology.

226

Harris B. Savin

3. But what if I don’t like lemonade?

Katz’s first chapter, The scope of semantics, attempts to define his subject matter by giving various examples of semantic phenomena. He introduces the chapter, disarmingly, with the quotation (from Dale Carnegie, he says), ‘If you have a lemon, make lemonade’. Fair enough, but if one doesn’t like lemonade, why bother? If one doesn’t find Katz’s semantic phenomena intrinsically interesting, is there any other reason to admire his theory? In defending his work against the accusation (which has been frequently made by empiricist philosophers) that he is squeezing an imaginary lemon, Katz argues at length that there are intuitively clear cases of semantic phenomena, and that they are as legitimate a subject for study as are syntactic phenomena. This argument, although it has not persuaded the philosophical opposition, ought to persuade psychologists who are prepared to accept the sorts of subjective judgments that are treated as data by students of sensory psychology, for example. One can ask which of two differently colored lights is the brighter, and one can construct a rich theory to explain a network of such subjective judgments -judgments which are unverifiable except in the sense that most other observers will give the same ones and the few non-conformist subjects themselves show internally consistant patterns of abnormal response which suggest different kinds of abnormal color vision. Granting that the lemon exists, though, psycholinguists may well wonder what is the point of squeezing it. Is Katz’s so-called semantics relevant to any psychologically interesting aspect of meaning, or is it really a sort of over-refined syntax? The ‘semantic phenomena’, after all, seem to be concerned wholly with relationships between linguistic entities. Semantics, on the other hand, is supposed to have to do with relationships between linguistic forms and some non-linguistic realm. Katz’s theory associates sentences of, say, English with configurations of semantic markers. In what way is this more semantic than, say, a theory that associates sentences of English with sentences of Latin, or sentences of English with their phonetic transcriptions? Katz’s answer, in part, is that semantic representations, unlike Latin sentences or phonetic transcriptions, have the formal structure necessary to explain the various semantic phenomena. But this is evidently no answer to those who question the importance for semantics of his phenomena. The more interesting part of Katz’s defense of his claim that his theory is about semantics is to be found in what he says about the relevance of his semantics to various non-semantic questions (non-semantic, at least, in Katz’s sense). His (extremely sketchy) discussions of the use of language in normal situations, of the theory of reference and of the structure of concepts suggest the role that Katzian semantics might play in answering questions psychologists might have supposed were semantic questions.

Meanings and concepts

3.1

227

Semantic theory and the use of language

Katz’s theory, by itself, quite obviously says nothing about the normal use of language. Synonymous sentences, like (la) and (1 b), may have very different uses and very different effects on their hearers. Semantically anomalous sentences have their uses, especially in poetry and in various kinds of word-play. The semantic ambiguities identified by the theory are not usually noticed in conversation. None of these facts is, in Katz’s mind, a defect of the theory. Rather, Katz supposes that all these aspects of language use depend in part on meaning, but also in part on questions of style, of the use of context to resolve linguistic ambiguities and so forth. This position seems eminently reasonable. Consider ambiguity first. Certain sentences, out of context, admit of several interpretations. I left my money at the bank, for example, is ambiguous according to Katz and, out of context, it is ambiguous in fact. But suppose you and I, while swimming in a river, encounter a floating ice cream vendor. I call to you, ‘Can you lend me ten cents; I left my money at the bank’. Now, you may well fail to notice the ambiguity. Does this show that the theory of ambiguity-out-of-context is irrelevant to what happens in this conversation, or does it show only that one of the linguistically possible interpretations is excluded by non-linguistic factors? The latter interpretation, surely, is the plausible one: The utterance out of context admits of certain interpretations. The non-linguistic context excludes certain of these and may introduce still others (for example, by suggesting that the utterance is to be interpreted metaphorically or ironically). A complete account of the interpretation of speech in context will have to explain this interplay of linguistic and non-linguistic considerations. But surely there are linguistic considerations, and the semantic properties of speech out of context are part of the explanation of interpretations in context. Katz’s claim is only that his theory includes the linguistic part of the explanation. The relationship between synonymy and the use of language is somewhat analogous. Synonymous sentences - for example, (8) and (9) - may differ enormously in their style, in their persuasiveness, in what they suggest about their speaker’s social class and educational background, and in a thousand other ways, and, notwithstanding these differences, they have the same meaning. What they literally assert, that is to say, is the same. (8) (9)

These are times that try men’s souls. Soulwise, these are trying times.

Indeed, it is impossible to see how literary styles could exist if there were not different ways to say the same thing, and it is impossible to see how one could make much progress in characterizing styles and rhetorical effects without distinguishing between

228

Harris B. Savin

meaning and form. If this is granted, then a theory of meaning becomes a contribution, at least indirectly, to the theory of rhetoric and the theory of style. 3.2

Meaning and reference

Reference is the relationship that the noun phrase my cat bears to the cat that I keep. Unlike meanings, which have often been condemned as useless theoretical fictions, reference is everywhere recognized as a genuine and important aspect of natural language. Those who believe in meanings usually believe that reference is a consequence of meaning; those who deny the existence of meanings usually attempt to show that reference, truth conditions and whatever other properties of language they believe in can be accounted for perfectly adequately without appeal to meanings. The issue is an important one in the philosophy of language, and much of Katz’s writing (including a large part of Semantic theory) is more concerned with defending meanings against the attacks of such empiricist philosophers as Quine than with developing his own theory. The extensionalist position, as the dominant anti-meaning view may be called, is in spirit analogous to methodological behaviorism. That is not to say that all extensionalists are behaviorists (although Quine, the most influential of them, is, at least sometimes) but that the two positions have strong affinities and, moreover. that extensionalists and behaviorists are similar in that they like to regard themselves as toughminded and rigorous, while their opponents accuse them of having closed their eyes to some of the most important phenomena in order to make their theories conform to arbitrary a priori criteria of methodological purity. The basic semantic phenomena, for extensionalists, are facts about reference and truth conditions: What things are referred to by expressions like the cat; under what conditions is it true that the cat is on the mat? Such believers-in-meaning, or intensionalists, as Katz, regard these as interesting and important questions, but they have two important points of disagreement with extensionalists: First, they believe that there are numerous semantic phenomena in addition to reference and truth; second, they suppose that reference and truth depend, among other things, on meaning. That is to say, they suppose that the theory of meaning includes the theories of reference and truth conditions. For example, such semantic features as (unmarried) and (male) in the definition of bachelor determine the reference of bachelor. The intensionalist position, of course, is a form of the common-sense mentalistic theory that assumes that we have a concept of bachelorhood and that we determine which individuals are bachelors by determining which ones have the properties that define our concept. Hence arguments about meaning and reference lead directly to a consideration of concepts. Psychologists talk a good deal about concepts, and the

Meanings and concepts

229

textbooks often say that a concept is simply a class of objects. This is sound extensionalist doctrine, but no psychologist has ever embraced its consequences: Any set of objects, for an extensionalist, is as natural, psychologically, as any other, and it is a mere cultural accident that we have no word in English that means either-a-man-whohas-never-married-or-a-non-returnable-bottle. There is nothing about human cognition, on the extensionalist view, that makes this an intrinsically complex, unnatural concept.4 Again, for an extensionalist, two classes are identical if they have the same members - as do creature with a heart and creature with a kidney, for example. These phrases, according alike to common sense and Katzian semantics, refer to the same animals, as a matter of zoological fact, but the property that makes Fido a creature with a heart is different from the property that makes him a creature with a kidney. Such properties, evidently, are in the minds that conceive them, and are therefore unacceptable to extensionalists.6 To be sure, one can imagine a creature with a heart but no kidney, but a theory that bases a distinction between the two properties upon the possibility of this feat of imagination is not discernibly less mentalistic than a theory that deals frankly in meanings and concepts. To summarize, there are serious difficulties with the program of accounting for reference and truth conditions without appealing to meanings, and, from a psychological point of view, no obvious reason not to let meanings or concepts determine reference, unless one regards the proposal as methodologically unacceptable on a priori grounds. There is, however, some reason to believe that the meanings of many words do 4. The claim that every class is as natural as any other, so far as man’s innate cognitive abilities are concerned, is the basis of an extreme form of the Who&n doctrine that language gives the concepts in terms of which we think. Any collection of otherwise heterogeneous objects, on this view, becomes a concept as natural as any other if all its members come to be called by the same name. There are, in this theory, no intrinsic properties of cognition that determine which classes of things are cohesive enough so that someone might find them worth referring to with a single term. 5. In order to see the strengths and limitations of this view, consider one of the most interesting attempts to make it seem plausible: Goodman (1952) notes that unicorn and centaur have the same extension. He does not want to reify meanings, but neither does he want to claim that unicorn and centaur can be used interchangeably, like eye doctor and oculist.

He proposes as a criterion sameness of ‘secondary extension’; to say that two terms have the same secondary extension is to say that all compound terms containing them have the same extension. Since picture of a unicorn differs in extension from picture of a centaur, unicorn and centaur differ in secondary extension. This device acknowledges a difference between centaurs and unicorns and does it without invoking meanings. But such an approach, from the point of view of most psychologists and linguists, is precisely backward. Pictureof-a-unicorn is not a semantically unanalyzable idiom like how-do-you-do. It has an internal structure, and we know how to tell what is in its extension by knowing the semantic principles according to which it is compounded. A theory that ignores the internal structure and the principles of compounding may have its uses, but it is not promising as a theory of people’s linguistic abilities.

230

Harris B. Savin

not by themselves suffice to determine reference. Katz himself raises this possibility in another connection but seems not to have pursued its consequences for the relationship of meaning to reference. Consider the reference of the word meaning itself. According to Katz, meaning refers to the semantic representations characterized by a correct semantic theory. The meaning of the word meaning, on the other hand, is a particular semantic representation in such a theory and this semantic representation does not contain within itself the entire theory of meaning, and therefore it does not determine the reference of meaning. Katz distinguishes between the ‘narrow concept’ of meaning, which determines the semantic properties of the word meaning but does not determine its extension, and the ‘broad concept. of meaning which is specified by the entire semantic theory and which does determine the extension of meaning. This is Katz’s argument against philosophers who have assumed that they could answer the question ‘what is meaning’ by studying the meaning of the word meaning. Such an investigation yields conclusions only about the narrow concept, although the original question concerned the broad concept (p. 451). What is true of meaning would appear to be true of electron, logical argument and many other terms. There is, in each case, a narrow and a broad concept. The narrow concept gives the meaning, and it may in some cases suffice to determine reference, but it equally well may not in any given case. Katz’s theory does not require that reference always be determined by linguistic considerations, and it is simply not possible, at present, to say whether this theory, when completed, will contain nearly all of the constructs necessary to determine reference, or none of them. We return to the question of broad concepts, narrow concepts and the concepts of advanced sciences below, when we take up Katz’ claim that his meanings represent the structure of concepts.

3.3

Psychological verbs and the reiJication of meaning

Extensionalists ask what is the use of a swarm of semantic markers or concepts since one can give the reference of bachelor directly, without appealing to its meaning. Katz’s reply is that, in addition to determining reference, meanings are required to explain many other semantic phenomena that the extensionalist account leaves unexplained for example, that bachelors are unmarried is true by virtue of the meanings of the words, while bachelors are taxed at higher rates than married people is a matter of empirical fact. It is clear enough how to use Katzian meanings to systematize such semantic phenomena, but extensionalists like Quine deny the very existence of the phenomena to begin with and think the only hard facts of semantics are the facts about reference and truth conditions (and the logical properties of connectives and quantifiers). The

Meanings

and concepts

23 1

linguistic intuitions that are the source of most of Katz’s data are as disreputable among extensionalist philosophers of language as they are among behavioristic psychologists. There is by now a rather extensive polemical literature in which the two camps mostly take each other to task for having wrong-headed notions about what the data of semantics are - viz., either all of Katz’s semantic phenomena or just reference and truth. The argument is, to a large extent, an argument about mentalism, for the main extensionalist objection to meanings is that they are mental entities and as such lack the tangibility of the physical. A particularly interesting issue concerns the verbs philosophers call ‘psychological’ verbs of knowing, wanting, doubting and the like. Note first of all that in most contexts, replacing a term with another having the same reference does not change the truth value of a sentence. Thus, if it is true that John saw Richard Nixon, then, it is true that John saw the Present of the United States. The psychological verbs (and some other words as well) create contexts that do not allow such substitutions. Clearly, (10) and (11) need not have the same truth value. (10) (11)

John wants to be Richard Nixon. John wants to be the President of the United

States.

Katz finds nothing mysterious about these facts, as long as Richard Nixon has not the same meaning as the President of the United States. The fact that these two nounphrases happen, for the time being, to denote the same individual, is an irrelevant fact of political history. The wish that is expressed in (10) differs from the wish in (11) in a way that is related directly to the difference in meaning between the two nounphrases. If one denies the existence of meanings, one has to explain the difference in truth conditions in some other way. There is nothing available except the difference in form between Richard Nixon and the President of the United States, and it is pretty clearly not possible to explain very much very plausibly by appeal to the difference in form. (See Quine, 1971, for an explanation along these lines of the fact that (10) and (11) differ in truth conditions. Although he can explain the bare existence of a difference, he cannot say anything very helpful about the nature of the difference between the two desires, as indeed he need not, given his behavioristic conviction that desires are very murky things anyhow.) Chisholm (1957) has argued - and Katz, curiously, does not adopt the argument that it is not possible to eliminate the mentalism from the psychological verbs, that in particular there are no paraphrases of those verbs that mention only behavior or tendencies to behave in particular ways. Suppose, to take his example, Jones expects to meet his aunt at the railroad station within twenty-five minutes. We cannot

232

Harris B. Savin

analyze his expectation as a tendency to react in one way if his aunt does appear and in another if she does not, for he might meet his aunt but fail to recognize her, or meet someone else and mistake her for his aunt. His expectation, in short, is a psychological state, reflected in behavior not directly but only indirectly, in ways that depend not on who gets off the train but on who he thinks he sees getting off the train. Such facts about expecf raise no special problems for Katz. Quine, ever the consistent behaviorist, accepts the correctness of this argument of Chisholm’s but takes the unavailability of a translation of expect into behavioral terms to show the emptiness not of behaviorism but of such mental notions as expectation (Quine, 1960, p. 221)! It is interesting to note that Irwin (1971, p. 145), in what is surely the most careful attempt ever to translate expectation, desire and related notions into behavioral language, has pointed out a similar difficulty in connection with, for example, the desire to square the circle. Irwin’s interest is not in semantics but in the psychology of motivation. His problem about desiring to square the circle arises from the fact that while it is, psychologically, a perfectly possible desire, it is a desire of an impossible state of affairs, and, if he cannot allow a desirer-to-square-the-circle to succeed, the definition that the proposes does not work. The impossibility of circle-squaring interferes with his program of getting the desire out of the realm of the mental and defining it in behavioral terms. Irwin, like Quine, does not hesitate to expose the worst problems of his own theory, and, also like Quine, he maintains the theory in the face of these problems because of his methodological convictions. But readers who are not irrevocably wedded to methodological behaviorism may find in Quine’s and Irwin’s discussions of expecting, desiring and the like persuasive reasons for acknowledging the existence of irreducibly mental constructs, and, in particular, meanings. 3.4

Meanings and concepts

We turn finally to the most important (a) (b) (c)

of Katz’s claims:

‘The logical form of a sentence is identical with its meaning’ (Katz, 1972, p. xxiv). ‘The logical form of a sentence consists of all the grammatical features that determine valid inferences into which the sentence enters’ (p. xx). ‘. . . a semantic marker is a theoretical construct which is intended to represent a concept that is part of the sense of morphemes and other constituents of natural languages. By a concept in this connection we do not mean images, ideas, or particular thoughts . . . . Cognitions are datable . . . . Concepts, on the other hand, are abstract entities. They do not belong to the experience of anyone, though they may be thought about, as in thinking about the concept of a circle. They are

Meanings and concepts

233

not individuated by persons: you and I may think about the same concept. They are not, as Frege urged, elements in the subjective process of thinking, but rather the objective content of thought processes’ (p. 38). If Katz cannot sustain these claims, he cannot claim that his semantic representations are, in any interesting sense, meanings. Pairing an English sentence with a sentence of some newly invented artificial language does not give the meaning of either. The basis of Katz’s claim that translations into his notation give the meaning is his psychological thesis that semantic structures are conceptual structures. There is, of course, no extant theory of conceptual structure to compare with Katz’s semantic representations in order to decide whether he is right. The theory of concepts that would sustain Katz would be a sort of mental chemistry-with some unanalyzable atomic concepts and some rules giving the possible configurations of atomic concepts. On the face of it, such a theory of concepts is plausible. Most of the concepts we occupy ourselves with can obviously be analyzed into simpler constituents. Consider, for example, visiting assistant professors, or non-returnable bottles. Indeed, up to a certain point, the overt structure of language suggests such conceptual analyses. Katz began one of his earlier books with the quotation from Frege: ‘It is astonishing what language can do. With a few syllables it can express an incalculable number of thoughts, so that even a thought grasped for the first time can be put into a form of words which will be understood by someone to whom the thought is entirely new. This would be impossible, were we not able to distinguish parts in the thought corresponding to parts of the sentence, so that the structure of the sentence serves as an image of the structure of the thought.’ But, if this sort of mental chemistry is plausible, it also bears the onus of having been tried once, by the British empiricists of the eighteenth and nineteenth centuries, and then abandoned. In fact, the sort of cognitive psychology Katz needs, although it resembles British empiricism in being atomistic, is free of the characteristics that seem to have been primarily responsible for the failure of that traditional psychology. Where the empiricists assumed that the atomic elements were all perceptual, Katz assumes they are mostly non-perceptual. Thus he does not face the great difficulty they had in accounting for abstract and relational concepts. Rather than attempting the apparently impossible reduction of abstractions and relations to perceptual things, he is free to postulate abstract and relational semantic elements whenever the facts seem to require them. Second, whereas traditional ‘compound ideas’ were mere aggregates of simple ideas, with no organization (or, where they were anything else, were described in very mysterious terms), Katz has the formal apparatus for representing different configurations of the same elements. For example, the semantic representations of The man bit the dog and The dog bit the man have the same elements

234

Harris B. Savin

arranged differently. Finally, the traditional analyses were analyses of conscious experience, and surely the most important psychological advance in the twentieth century has been the discovery that most of what is important about the mental processes is not directly available to introspection. In particular, the introspective analyses of traditional empiricism yield results that have little or no relevance to the logical properties of concepts. The failure of traditional British empiricism is often laid to its use of an allegedly bad method - introspection; modern linguists are sometimes suspected by psychologists of having revived the same bad method. It is therefore interesting to compare the two methods and the uses made of their respective results. Consider Bachelors are unmarried, which is a necessary truth. The traditional explanation of its necessity is that the predicate unmarried is already part of the meaning of bachelors. On this account, which Katz adopts, Bachelors are unmarried is vacuous in essentially the same way as Bachelors are bachelors. They are both examples of vacuous predication. But this account does not work if idea means what it meant in traditional introspective psychology. One has different conscious experiences on different occasions while thinking about bachelors and about being unmarried; it is simply not true that, on hearing Bachelors are unmarried one can distinguish a part of the imagery which is evoked by bachelor and which is identical to that evoked by unmarried. (See Frege, 1893, for a general discussion of the impossibility of relating logical notions to this kind of psychology.) Katz, of course, makes no such implausible claim. He consults his linguistic intuition only to decide that, in point of fact, Bachelors are unmarried is a necessary truth. His semantic representation of bachelor, which includes a feature (unmarried), is not a direct result of reflection on bachelor but a theoretical construct, intended to help explain judgments about analyticity, contradiction, synonymy, implication and the like -judgments of a very different kind from the introspective analyses of the eighteenth and nineteenth centuries. To summarize, a mental chemistry of concepts is an attractive program because most concepts are obviously composite. And the sort of mental chemistry Katz proposes differs in crucial ways from the unsuccessful attempt of the British empiricists. Katz’s claim to have a theory of the structure of concepts cannot therefore be rejected out of hand. On the other hand, there are no very persuasive grounds for accepting his claim. His theory of meaning is based on data exclusively about a small class of logical properties and relationships. Such data are relatively easy to come by, and it would be very pleasant to discover that they are all we need to answer questions that have far-reaching consequences for psychology. Unfortunately, such a prospect is as unlikely as it is alluring. Although it is plausible enough that semantic representations are a part of the structure of corresponding concepts, there is good reason to suppose

Meanings and concepts

235

that there is more to the structure of concepts than Katz will turn up while studying his semantic phenomena. Indeed, in developing his own program, Katz needs to make two distinctions - between broad and narrow concepts, and between immediate and mediate inference - which raise doubts about the thesis that semantic representations are concepts. We turn now to these distinctions. ‘We may illustrate the difference between the narrow notion of a concept and the broad notion of a concept, as we shall refer to them, in terms of an example. The dictionary sense of “Martian”, the narrow concept of a Martian, is that of a rational creature who is an inhabitant of the planet Mars. But the average person’s conception of a Martian, the broad concept of a Martian, might be that of a little green humanoid creature with a plasticlike skin, luminous eyes, bobbing antennae, telepathic powers, and an intelligence vastly superior to ours. The former concept contains the properties that we as speakers of English commonly use to identify the primary, secondary, etc., extension of the term “Martian”. The latter is a frequent stereotype, containing properties culled from our experience with science fiction. This difference in the status of the properties that make up the concept emerges clearly when we use them to form explicative judgments about Martians. Thus, (12) is analytic but (13) is not. (12) (13)

Martians are creatures who inhabit Mars. Martians are telepathic (green, etc.)’ (Katz, 1970, pp. 450-451).

Now it is no argument against Katz’s psychological claims that various parts of the broad concept are not parts of the corresponding narrow concept. If the broad concept of a Martian can be expressed in English, it can be given a semantic representation within his theory. For example, the broad concept of a Martian is the same as the narrow concept of a ‘little green humanoid creature with plasticlike skin . . .‘. But the distinction raises difficulties nonetheless. Consider the sorts of conservationproperties studied by Piagetian developmental psychologists. For example, adults and older children believe that the quantity of a liquid is independent of the shape of its container, the very compelling perceptual illusions to the contrary notwithstanding. Katz has to say that this invariance is a part of the broad concept of liquid quantity, not the narrow concept because (14) is not anomalous semantically but merely empirically false. (14)

There was only a little water when it was in John’s glass, but its quantity increased greatly when he spilled it over Mary’s dress.

Such conservation principles are doubly difficult for Katz’s psychological claims. First, they are apparently not reflected in any data about semantic phenomena. Second, they cannot be incorporated into lexical entries with formal properties anything like the ones that Katz has so far used. If Katz claims only to have a psychologically

236

Harris B. Savin

uninteresting theory of some linguistic phenomena, then obviously these two problems cancel each other: Conservation is neither in the data nor in the theory. But conservation principles embarrass his claim to have a theory of concepts. Consider now the distinctions between immediate inferences and inferences requiring an argument. Katz needs the distinction to distinguish between pairs like (15a-b) and (16a-b) where T, and T2 are theorems that are logically equivalent, but the proof of their equivalence is long and difficult. (15a) (15b) (16a) (16b)

John John John John

knows knows knows knows

that that that that

Bill is an oculist. Bill is an eye doctor. T,. T,.

(16a) and (16b), then, cannot be paraphrases. Indeed, one of them may be true and the other false. (15a) and (15b), on the other hand, are paraphrases. Synonymy, therefore is not the same as logical equivalence since the synonymy of oculist and eye doctor has different semantic consequences from the mere logical equivalence of T1 and T,. For the present, Katz has no theory of logical arguments, but if he wants to claim that meaning is logical form, he has to show that the semantic representations of any two logically equivalent statements contain enough structure to sustain a proof of their equivalence. Now there is no reason to assume that semantic representations that were devised to account for a relatively limited variety of semantic phenomena will have the properties required to explain a far more varied set of types of arguments. This might turn out to be the case, but, if it should, Katz would appear to be the beneficiary of an astonishing stroke of good luck. Finally, a theory of the structure of concepts ought to say something about impossible concepts. I have already remarked that Katz so far has had nothing to say about the principles determining the relationships between selection restrictions and meanings and therefore cannot account for the impossibility of (for example) a verb that has the meaning of eat but takes abstract nouns (like knowledge) as its subject. In short, Katz’s semantics lacks various characteristics that a theory of the structure of concepts ought to have, and its limitations are not particularly remarkable. Why should a theory of the ‘semantic phenomena’ turn out to be just what is needed to explain all manner of cognitive phenomena? On the other hand, it seems perfectly plausible to suppose that Katzian semantic structure is a part - and an important part - of conceptual structure. Implication, analyticity, contradiction and the like are cognitive phenomena, which the cognitive psychologist owes some sort of explanation for. The semantic features that figure in Katz’s explanations of these phenomena often turn out to be psychologically salient properties of the concepts they define.

Meanings and concepts

237

And even such behavioral manifestations of cognitive structure as word-associations are in part explained by Katz’s semantic analyses. For example, a great many responses on word-association tests are words that share many semantic features with the presented word. These sorts of arguments for the psychological importance of Katz’s semantics would be a good deal more persuasive if the theory itself were more nearly complete. In fact, of course, he has scarcely begun to work out details, and, in the foregoing discussion, I have simply assumed that no insurmountable problems will arise in the completion of his program. But, whatever the ultimate fate of Katz’s theory, for the present it commands the attention of cognitive psychologists because it constitutes the only systematic attempt to reduce to some semblance of order the immense variety of concepts. Psychologists have not attempted anything of the sort since the end of the nineteenth century, when they gave up trying to analyze all of the contents of the mind into perceptual constituents. There are today two main types of psychological investigation of concepts. The study of ‘concept attainment’, in the manner of Bruner, Hovland and others, takes for granted the attainment of various concepts and investigates the conditions under which people compound new concepts out of this material together with such logical connectives as and and or. The Piagetians, on the other hand, have given us careful and illuminating studies of the development of a handful of particularly interesting concepts but have nothing remotely like a general theory of the structure of concepts. For the only attempt at such a theory one must turn to Katz, and most especially to his latest book. Having urged the reader to study Semantic theory, I must warn him that he will not find it easy. Katz’ prose is difficult, and his book is badly organized, with many of the most important parts buried in replies to critics and omitted from the index. But, notwithstanding the problems both of form and content, it is too important to be ignored. REFERENCES Chishohn,

Roderick

M. (1957) Perceiving: study. Ithaca, Cornell. Chomsky, Noam (1965) Aspects of the theory ofsyntax. Cambridge, M.I.T. A philosophical

Dougherty, Ray C. (1970) Recent studies on language universals. Found. Lang., 6, 505561.

Fillmore, Charles, J. (1968) The case for case. In E. Bach and R. T. Harms (Eds.), Universals in linguistic theory. New York, Holt, Rinehart and Winston. Frege, Gottlob (1893) Grundgesetze der Arithmetik, Vol. 1. (English translation by

Montgomery Furth, University of California, Berkeley, 1967). Gleitman, Lila R. (1970) Semantic constraints on surface structure, or the great verb game. Eastern Pennsylvania Psychiatric Institute Technical Report 13. Goodman, Nelson (1952) On likeness in meaning. In Leonard Linsky (Ed.), Semantics and the philosophy of language. Chicago, University of Illinois. Pp. 67-76. Gross, Maurice (1971) Remarques sur les processes d’apprentissage de la syntaxe dune premiere langue. Mimeographed,

230

Harris B. Savin

Universite de Paris VIII et Laboratoire d’Automatique Documentaire et Linguistique. Irwin, Francis W. (1971) Zn&ntionaZ behavior and motivation: A cognitive theory. Philadelphia, Lippincott. Katz, Jerrold J. (1964) Semantic theory and the meaning of ‘good’. 3. Philo., 61 (235). and Fodor, Jerry A. (1963) The structure of a semantic theory. Language, 39, 170-210.

-and

Postal, Paul (1964) An integrated theory of linguistic descriptions. Cambridge, M.I.T. Quine, W. V. 0. (1960) Word and object. Cambridge, M.I.T. (1970) Quantifiers and propositional attitudes. In Leonard Linsky (Ed.), Reference and modality. Cambridge, Oxford. Pp. 101-111. Vendler, Zeno (1972) Res cogitans: An essay in rationalpsychology. Ithaca, Cornell.

Discussions

A short

note

on Evans’ criticism of reasoning experiments and his matching response hypothesis*

P. C. VAN DUYNE Rijksuniversiteit

Groningen.

Netherlands

The aim of this short paper is to reformulate some of the problems in the psychology of reasoning, which Evans (1973) in his paper has left open or for which he proposes solutions which seem to lead to as many pitfalls as he himself detects. First of all, the distinction which he makes between logical, illogical and nonlogical behaviour remains dubious in the absence of a more precise definition of ‘logical’. Since Piaget (1928) it has become a tradition to take the propositional calculus as a standard for correct reasoning. However, this is a serious misunderstanding of the calculus. From a logical point of view (and also linguistically) most reasoning experiments have no connection with any kind of logical calculus. In fact, the logical status of these empirical investigations is purely semantic (Carnap, 1958). It concerns the assignment of truth values to certain sentence types expressed in natural language. Such sentences are typically constructed with connectives like ‘or’, ‘and’, ‘if . . . then . . .’ etc., the interpretation of which happens to correspond to their definitions in symbolic logic. But this is an historical accident which is regretted by almost all logicians. The acceptance of the propositional calculus as a baseline is theoretically more than misleading. In this respect Evans is correct when he argues that comparison of subjects’ performance with formal logic leads to misinterpretation of the data, but for the wrong reasons. What we really want in these investigations is a theory about the logic of natural languages. This means more investigations of actual linguistic behaviour, rather than of ‘non-logical operational variables’ although it is of course important to explore and control the latter. However, Fodor (1970) has shown that the gap between linguistics and formal logic is still unsurmountable l Grateful acknowledgements are due to the British Council for a grant for scientific research and to Dr. P. C. Wason for his encouragement and critical reading of an earlier

draft of this paper. This paper was written while the author was resident at University College London.

cognition 2(2), pp. 239-242

240

P. A. Van Duyne

(But see Lakoff, 1971). As long as that gap remains, the only thing that can be done is to compare the responses of subjects at different levels of linguistic behavior, e.g., in the Wason et al. experiments which compared subjects’ interpretation of sentences with their selection of verifying and falsifying instances, and the reasons given for selecting them (Wason and Johnson-Laird, 1970; Wason and Goodwin, 1972; Wason and Johnson-Laird, 1972; Legrenzi, 1970; Legrenzi, 1971). Only in this way can the interpretations, inconsistencies and deviations from intuitively correct ways of reasoning be detected and described. Evans objects to a formal interpretation of the subjects’ interpretation of a rule and argues that this does not have an explanatory value. This is, of course, correct: Every description does not aim at an explanation. Investigations in reasoning have not yet reached that level. However, an accurate description of the linguistic behaviour, in whatever terms, is psychologically of value in its own right. Introspective evidence clearly demonstrates that subjects are very often inconsistent and self-contradictory, in spite of their efforts to grasp the problem and to explain their own responses. In these reasoning experiments the subjects do evidently really try to reason correctly. Hence it would seem obligatory to describe such reasoning in formal terms as well as in terms of non-logical variables. Introspective reports could reveal the ways in which aspects of linguistic behaviour are interrelated. I: is unfortunate in this respect that Evans does not include in his ‘matching response’ experiments any introspective, or ‘thinking aloud’ protocols, in spite of the importance he attributes to such data. Evans’ objection to the investigation of typical errors has little justification. By comparing the conditions in which errors do occur (e.g., with abstract material) with conditions in which subjects show insight, we can hope to detect the interaction between sentence types and ‘non-logical operational’ variables (Legrenzi, 1971; Wason and Shapiro, 1971; Johnson-Laird, Legrenzi and Legrenzi, 1972; Van Duyne, 1973). In addition, the comparison between performance under abstract and concrete conditions offers the best opportunity to obtain information about the semantics of different sentence structures and connectives (Johnson-Laird, 1969; Wason and Johnson-Laird, 1969; Legrenzi 1970; Evans, 1972). Evans complains correctly about the paucity of theory in this field, but his own hypothesis of ‘matching response bias’ does not appear to satisfy this defect (Evans, 1972; Evans and Lynch, 1973). It tries to explain the selection of thep and q values in sentences like ‘if p then q’ in terms of matching bias, which induces subjects to select the values mentioned rather than to alter them. However, this sort of explanation emphasizes too much the so-called ‘non-logical operational variables’, without taking into account the fact that in this task subjects actually do reason. For example, when presented with an implicational rule in the linguistic guise of disjunction (not-p or q), subjects choose p and q very infrequently (Johnson-Laird and Tagart, 1969; Wason

A short note on Evans’ criticism of reasoning experiments

241

and Johnson-Laird, 1969; Legrenzi, 1970; Van Duyne, 1973). Another experiment performed by the author confirmed the hypothesis that matching was primarily confined to the ‘ifp then q’ sentence type. In this experiment a group of 18 subjects performed the ‘selection task’ on four abstract sentences in the form: ‘Whenp then q’; ‘if not-p then q’; ‘p or q’; ‘not-p and not-q’. The stimulus cards were represented as drawings on sheets of paper and the subjects had to place a tick under the cards which they thought had to be turned over. Table 1 summarizes the results. Table 1. Frequency of the selected values for the four sentence types Valuesselected

‘When p then q’

‘If not-p then q’

‘p or q’

‘not-p and not-q’

P, 4 -P. --4 P> -P* 49 --4 -P, 4 -P, 49 --4 P9 --4

Miscellaneous N

10 0 2 0 0 1* 5 18

1 1* 5 3 5 0 3 18

2 10* 3 0 0 0 3 18

5 6’ 4 1 0 1 1 18

Note:

with an * indicate

The numbers

the frequency of the correct selections.

Only the ‘when p then q’ sentence yielded significantly more matching responses than the other sentences. The remaining sentence types, with a double mismatching between the appropriate cards and the sentence, produced much less ‘matching response’. It will be seen from Table 1 that much more mismatching occured in the ‘p or q’ and ‘not-p and not-q’ sentence types. And this is inconsistent with the matching response hypothesis. Even with the implication sentence ‘if not-p then q’, the infrequency of the p and q selection is completely incompatible with matching bias. Indeed, the matching response hypothesis does not offer any explanation, in whatever terms, of the effect of underlying operational variables: It only says that subjects choosep and q because they are inclined to choosep and q. But that is almost a tautology. Evans does not indicate anywhere the nature of the interaction between the ‘non-logical operational variables’ (sic) and the interpretation of the sentences in question, i.e., the other source of information which he postulates. A more precise theory is really needed to explain how people think about propositions as well as why they reason in such typical ways.

242

P. A. Van Duyne

REFERENCES Catnap, R. (1958) Introduction to symbolic logic. New York, Dover. Evans, J. St. B. T. (1972) Interpretation and ‘matching bias’ in a reasoning task, Q,J. exp. Psychol., 24, 193-199. (1972) On the problems of interpreting reasoning data: Logical and psychological approaches., Cog. 1 (4), 373-384. and Lynch, J. S. (in press) Matching bias in the selection task, Brit. J. Psychol. Fodor, J. D. (1970) Formal linguistics and formal logic. In J. Lyons (Ed.), New horizons in linguistics. Middlesex, Hanmondsworth. Goodwin, R. Q., and Wason, P. C. (1972) Degrees of insight, Brit. J. Psychol., 63, 205-212. Johnson-Laird, P. N. (1969) On understanding logically complex sentences. Q. J. exp. Psychol., 21,1-13. and Tagart, J. (1969) How implication is understood. Amer. J. Psychol., 82, 367-373. Legrenzi, P. and Legrenxi, M. S. (1972) Reasoning and a sense of reality, Brit. J. Psychol., 63, 395-400. Lakoff, G. (1971) The role of deduction in grammar. In C. J. Fillmore and D. T.

Langendoen (Eds.), Studies in linguistic semantics. New York, Holt, Rinehart and Winston. Legrenzi, P. (1970) Relations between language and reasoning about deductive rules. In G. B. F. D’Arcais and W. J. M. Levelt (Eds.) Advances inpsycholinguistics. Amsterdam, North-Holland. __ (1971) Discovery as a means to understanding. Q. J. exp. Psychol., 23,417-422. Piaget, J. (1928) Judgment and reasoning in the child. London, Routledge. Van Duyne, P. C. (in press) Realism and linguistic complexity in reasoning, Brit. J. Psychol. Wason, P. C. (1968) Reasoning about a rule. Q. J. exp. Psychol., 20, 273-281. and Johnson-Laird, P. N. (1969) Proving a disjunctive rule. Q. J. exp. Psychol., 21, 14-20. (1970) A conflict between selecting and evaluating information in an inferential task, Brit. J. Psychol., 61, 509-515. (1972) Psychology of reasoning: Structure and content. London, Batsford. and Shapiro, D. (1971) Natural and contrived experience in a reasoning problem, Q. J. exp. Psychol., 23, 63-71.

On the ethics of intervention With special reference

in human psychological research: to the Stanford prison experiment

PHILIP G. ZIMBARDO Stanford University

Research was conducted recently (August 14-21, 1971) in which subjects assumed the roles of ‘prisoner’ or ‘guard’ for an extended period of time within an experimentally devised mock prison setting on the Stanford University campus. The projected twoweek study had to be prematurely terminated when it became apparent that many of the ‘prisoners’ were in serious distress and many of the ‘guards’ were behaving in ways which brutalized and degraded their fellow subjects. In addition, the emerging reality of this role-playing situation was sufficiently compelling to influence virtually all those who operated within it to behave in ways appropriate to its demand characteristics, but inappropriate to their usual life roles and values; this included the research staff, faculty observers, a priest, lawyer, ex-convict, and relatives and friends of the subjects who visited the prison on several occasions (for details see Zimbardo, Banks, Haney and Jaffe, 1973; Haney, Banks and Zimbardo, 1973). This research represents one of the most extreme experimental demonstrations of the power of situational determinants in both shaping behaviour and predominating over personality, attitudes and individual values. As such it extends the conclusions from Stanley Milgram’s research on obedience to authority (1974). But the ethical concerns voiced over Milgram’s treatment of placing subject-teachers in a conflict situation where they believed (incorrectly) that they were hurting another person are even more pronounced in the present case. Volunteer prisoners suffered physical and psychological abuse hour after hour for days, while volunteer guards were exposed to the new self-knowledge that they enjoyed being powerful and had abused this power to make other human beings suffer. The intensity and duration of this suffering uniquely qualify the Stanford prison experiment for careful scrutiny of violations of the ethics of human experimentation. The plan of this article is to: (a) Give a synopsis of the experiment to familiarize the reader with its basic features; (b) summarize one set of critical arguments levelled Cognition 2(2), pp. 243-256

244

Philip G. Zimbardo

against the experiment (which invited my reply in this particular journal); (c) analyze the sense in which the mock prison study can be considered to be unethical, and (d) present a body of information relevant to passing judgment on its ethicality from a legal, pragmatic, utilitarian or relativistic model of ethics. My intention is less to assume a defensive stance in support of this particular study than to use it as a vehicle for delineating the enormous complexity of making ethically based decisions about interventions in human experimentation.

Synopsis of mock prison study Interpersonal dynamics in a prison-like environment were studied experimentally by designing a functional (rather than literal) simulation of a prison. Environmental, structural, institutional and social variables were manipulated in an effort to create a ‘psychology of imprisonment’ in a group of subjects who role-played being guards (for eight hours a day over three shifts) and a group who acted as prisoners (for twentyfour hours a day). To assess the strength of the social, situational forces on the behaviour of these volunteer subjects, alternative explanations in terms of pre-existing dispositions were eliminated through subject selection and random assignment to treatments. A homogeneous sample of about two dozen normal, average, healthy American college males was chosen after extensive interviewing and diagnostic testing of a large group of applicants recruited through newspaper advertisements. The subjects were from colleges throughout the United States and Canada who volunteered to be in ‘a study of prison life’ in return for receiving a daily wage of tifteen dollars for a projected two-week period. Half of these pre-selected subjects were randomly assigned to role-play prison guards, the others to the mock-prisoner treatment. Neither group received any formal training in these roles - the cultural mass media had already provided the models they used to define their roles. The mock guards were impressed with the ‘seriousness’ of the experiment and by the demeanor of the research staff; the prospective prisoners began to take their roles seriously when they were subjected to an unexpected arrest by the city police. After being processed and temporarily detained at the police station, they were escorted to the experimental setting. Uniforms and differences in power further served to differentiate the two groups of subjects. Continuous, direct observation of all behavioural interactions was supplemented by video- and audio-taped recordings, questionnaires, self-report scales and interviews. All of these data sources converge on the conclusion that this simulated environment was sufficiently realistic and forceful to elicit intense, personal and often

On the ethics of intervention

in human psychological

research

245

pathological reactions from the majority of the participants. Many of the prisonersubjects exhibited behaviors characteristic of the learned helplessness syndrome described by Seligman (1973) in his research on traumatic, avoidance conditioning. The guard-subjects displayed a behavioral profile which was marked by its verbal and physical aggressiveness, arbitrariness and dehumanization of the subjects in the prisoner condition. None of these (and other) group or individual behaviour patterns was predictable from the medical, social or educational histories of the subjects, nor from a battery of personality test scores.

Summary of critique of the Stanford prison experiment

One critic of this research (Savin, 1973) begins his commentary by raising the fundamental question of ‘under what conditions is research justifiable which subjects people to injury?‘. Unfortunately, the promise of suggested guidelines for meaningfully dealing with this vital and difficult ethical issue is never realized, as his argument is first trivialized (by asserting that one would be justified to embarrass a few self-confident volunteers if this ‘injury’ resulted in a cure for schizophrenia), and then descends into a series of direct and implied ad hominem attacks. Some choice instances are: ‘One cannot make a prison a more humane institution by appointing Mr. Zimbardo its superintendent’; ‘professors in pursuit of their own academic interests and professional advancement’ are subverting the teacher-student relationship; there are some psychologists who are ‘as obnoxious as the law allows’, who show ‘a morally obtuse zeal in the pursuit of their careers, and who can be likened to used-car salesmen; finally, ‘on occasion there is a hell like Zimbardo’s’. The major substantive points raised in this commentary appear to be: (4 ‘The roles of prisoners and guards have been much discussed of late, and their characteristics are reasonably well known’, therefore, no new worthwhile knowledge was derived from this study. A study which presents results that are not surprising to the scientific research W community is not justified in subjecting volunteers to harm. If the outcome could have been predicted in advance by an already existing body of knowledge, there is no reason to conduct a study whose treatment of human subjects is questionable. (4 Some researchers, blinded by their own ambition, are unable to identify with the noxious experiences to which they expose their research subjects. Cd) Researchers and academicians resist any objective evaluation and appraisal of the conduct of their research. (4 Subjects would be better protected if there were more ‘law and order’. If only

246

Philip G. Zimbardo

the police were not so ‘apt to be somewhat lax in upholding the laws that might otherwise protect subjects’, we can assume guilty psychological experimenters would be brought to justice [and imprisoned??].

Absolute ethical principles to guide research

Ethics embody individual and communal codes of conduct based upon adherence to a set of principles which may be explicit and codified or implicit and which may be abstract and impersonal or concrete and personal. For the sake of brevity, we may say that ethics can be dichotomized as ‘absolute’ and ‘relative’. When behaviour is guided by absolute ethical standards, a higher-order moral principle can be postulated which is invariant with regard to the conditions of its applicability - across time, situations, persons and expediency. Such principled ethics allow no degrees of freedom for ends to justify means or for any positive consequences to qualify instances where the principle is suspended or applied in an altered, watered-down form. In the extreme, there are no extenuating circumstances to be considered or weighed as justifying an abrogation of the ethical standard. To search for those conditions which justify experiments that induce human suffering is not an appropriate enterprise to anyone who believes in the absolute ethical principle that human life is sacred and must not in any way be knowingly demeaned physically or mentally by experimental interventions. From such a position, it is even reasonable to maintain that no research should be conducted in psychology of medicine which violates the biological or psychological integrity of any human being regardless of the benefits that might, or even would definitely, accrue to the society at large. Many people feel that the world is already too polluted with human pain, alienation and the cynicism and mistrust of Watergates, known and suspected, to allow any further increments. This is so, even if they are in the name of science, for the sake of knowledge, ‘national security’ or any other high flying banner. Within psychology, some of those identified with the humanist tradition have been most vocal in urging that these basic concerns for human dignity take precedence over the stated goals of the discipline, namely, to predict and control behaviour. To the Christian Scientist, for example, the principle of the integrity of the human body and God’s design in the ebb and flow of life does not allow therapeutic drugs and internal treatments for illnesses even though death may result in the absence of such interventions. On the basis of such an absolute ethic, the Stanford prison experiment must certainly be ajudged unethical because human beings did suffer considerable anguish, yet it was possible to terminate the experiment once that was apparent.

On the ethics of intervention

in human psychological

research

247

Relative ethical principles in human experimentation When an ethical principle admits of any contingent applications, it is a relative standard to be judged by pragmatic criteria according to some weighted, utilitarian model of ethics. Obviously it is such a model which guided the research in question, as well as most other psychological experimentation. But what are the elements to be considered in this loss-gain equation? How are they to be proportionately weighted? Who is to judge whether the gains offset the losses? These are some of the starter questions which must be faced if a position of relative ethics is to be, nevertheless, still an ethical position. Some of the solutions are resolved on the basis of conventional wisdom, the present state of relevant knowledge, precedents in similar cases, social consensus, the values and sensitivity of the individual researcher and the prevailing level of consciousness in the given society, Let us, however, be more specific than this in our analysis of the ethics of intervention. Every act of intervention in the life of an individual, a group or an environment is a Therapists, behavior matter of ethics (R. D. Laing would say is a political decision). modifiers, industrial psychologists, counselors, experimentalists, educators, urban planners, architects, reformers, and, yes, even used-car salesmen, all have as one of their prime goals cure, modification, recommendation for action, alteration, manipulation, control, change, destruction, allocation, construction - in sum, intervention. Even the hard-headed cognitive psychologist studying reaction time as a function of contextual cues in information retrieval must be accountable to the ethics of intervention by virtue of taking an hour out of a person’s life to engage in a task typically boring to the subject, with little or no personal value to the student -and with no redeeming value to anyone if the experiment does not yield statistically significant results. Our ethical sense is usually not disturbed when a person seeks help for a problem; some expert gives the help in return for a fee, and the problem abates. Everyone gains, and, as they say in the Bronx, ‘A fair exchange is no loss’. But suppose the person seeks help not from choice but by a court order (as might be the case of a pederast, homosexual, extremely assaultive individual, etc.)? And suppose the agent of change demands a pound of flesh for a fee? And suppose the hyper-aggressive person happens to be a member of the Black Panthers, or for democracy in Greece? And suppose the treatment is castration or lobotomy? Then, perhaps, some people gain a little more than others, while some lose more. In this case, the expert-helper gains personally, and society-at-large may gain by reducing the potential ‘danger’ posed by the person, but at the expense of that individual’s sense of self-determination, freedom and integrity.

248

Philip G. Zimbardo

Similarly, we take for granted the powerful socializing influence parents exert upon their children to shape them to their image and toward a socially imposed ideal. Consider, however, the following actual case of a father who wrote for advice to an advice-dispensing medical doctor in a nationally circulated magazine (McCall’s). He wanted to know if sleep-learning was an effective technique to use in teaching his young son to be a patriotic American. Was it o.k. to give him a little ‘pep talk’ now and then while he was asleep, in order to instill a firm belief in him of the importance of being a good American? Is it ethical for dear old dad to indoctrinate his defenseless child in this way? Would it be ethically better if he did so when the child was awake, or if he used monetary reinforcers or social approval instead of the more ineffective technique? Is it his goal or his means that are ethically offensive? Is it preferable instead for this anxious father to rely upon the more subtle indoctrination devices disguised as ‘education’ in the classroom - in the form of national flags, pictures of national leaders, singing of national anthems, chauvinistic history, geography and civics’ textbooks, and the like? The point here is the necessity for increasing our collective sensitivity to the broad range of daily situations where interventions occur as a ‘natural’ process of social life and where the violation of ethics goes unnoticed because of its very prevalence. These few examples begin to make apparent that the loss-gain equation involves more than just the ‘agent of change’ and the ‘target of change’. We must add the sponsoring institution of the change agent, and the society which typically is alleged to be the ultimate repository of the positive consequences of that intervention. Let us examine each in turn with reference to their loss-gain in the Stanford prison experiment. The gain to the principal investigator (P.G.B.) was primarily from four sources. There was the new knowledge of a social-psychological, academic nature; there was a consciousness raising regarding prison conditions and the need for prison reform; there were several publications (which, incidentally, have no effect upon my professional advancement); and there was a considerable amount of publicity in the mass media. On the other hand, there were negative counterparts, or losses, to be added in. There was an unusual amount of personal stress experienced during this around-theclock study because of the continual need to make immediate decisions of both a scientific nature (in my role as principal investigator), as well as of a custodial or administrative nature (in the role of superintendent of the prison). During this time, for instance, I lost ten pounds, had difficulty eating and sleeping and developed a chronic headache. Furthermore, in our psychology department, such ‘flashy’ research is rewarded with benign indifference by colleagues, which in some cases was rather embarrassing. Other mixed blessings include extensive, time-consuming (nonremunerated) activities related to prison reform, such as talks to Chamber of Com-

On the ethics of intervention in human psychological research

249

merce groups, prison groups, correctional personnel, high school students and others not to mention personally answering several hundred letters from these and other sources. Such a sudden, unexpected demand has had a deleterious effect on meeting other prior professional and personal commitments. What about the subjects? They did suffer in the ways described previously and that was a substantial loss which should not be minimized. There was physical pain, psychological humiliation, anxiety, perhaps a loss of innocence and extremely unpleasant memories. Did they gain anything? From intensive individual interviews and encounter group experiences with them, as well as from a series of follow-up surveys, 1 would conclude they did. The suffering appears to have been restricted to the situation, constrained by its novelty, artificiality and differentness from their prior or subsequent ‘real’ life. There were no reported persisting negative reactions during any time since the end of the study. Moreover, most of the subjects report that it was a valuable didactic experience for them. They learned many new and valuable lessons about themselves because the situation elicited reactions they believed to be ego-alien. These behaviors were never previously manifested because normally they could arrange their lives to avoid unfamiliar situations or ones where they could not predict how they would behave and also have considerable control over their behaviors and the consequences. During our day-long debriefing sessions, the moral conflicts posed by this experiment were made explicit as part of our ‘moral reeducation’ training (according to Kohlberg, this is the primary or only way to raise an individual’s level of moral development). Some subjects have since volunteered their summer vacation time to work in local prisons and most have become advocates for penal reform. Other gains to be mentioned are the money paid the subjects, term papers and speeches they gave in subsequent college courses about their unique experience, as well as, for some who wanted it, the notoriety of a picture and their name in a Life magazine story about the experiment. There were two sponsors of this research, Stanford University and the Office of Naval Research (ONR). The latter funded the research as part of a general contract to me for the investigation of conditions which facilitate antisocial behavior; the university provided the space and many other resources. The university gains financially from the conduct of this research through overhead paid by the granting agency, while ONR gains from having supported basic research which has gained much publicity and yields new information. In addition, this research and my direct efforts have stimulated the Group Effectiveness branch of ONR to begin to support research into military prisons, especially problems of racism in the military correctional system - from which they will derive additional benefits. Since the university’s sponsorship involves legal responsibility for negligence of its staff, or injuries suffered by citizens on its premises, it assumes the potential of un-

250

Philip G. Zimbardo

favorable publicity (and negative alumni reaction) and costs of law suits (which have occurred as a consequence of other research, although not in the present case). At the other end of the loss-gain function stands the ubiquitous ‘society’ which allegedly benefits from the sacrifices of some unfortunate research subjects. But because ‘society’ is rarely defined in any denotative sense, it is impossible to meaningfully assess what, if anything, it really gains from any one research endeavor. This is especially true of psychological research which does not produce a consumable product, a vaccine, or technological innovation, but only knowledge or new ways of conceptualizing the relationships between individuals and their environment. But to be seriously concerned about the effects of research requires subsequent evaluation research. How many studies have evaluation procedures as part of their overall research design? How many researchers have the facilities and resources to begin to assess the long-term, pervasive impact of their research on even a small sub-group of society, let alone on ‘The Society’? We are just now being forced to acknowledge the unexpected, eventual negative consequences of environmental interventions which initially were hailed as beneficial to society. Consider DDTpollution, the increased crime, vandalism and alienation produced by urban renewal projects or the debilitating, dehumanizing effects on workers of technologically efficient work procedures (see Sommer, 1969; Toffler, 1970; Craik, 1973). It cannot be proven that the Stanford prison study has had a profound, positive, socially redeeming value on society. Indeed, some prison authorities, such as the Warden and Associate Warden at San Quentin, have publically derogated the study, believing it casts a bad light on corrections. On the other hand, they seem to be in a decided minority, since a great many prisoners, former inmates, legislators, criminal lawyers and parole officers have gone on record endorsing the findings and implications of our study. One initial measure of the impact of any research is in the dissemination of its ideas to relevant interest groups. Another would be extensions of the ideas in subsequent research as well as implementation of its social recommendations with new procedures, policies and laws.

Indirect impact evaluation of the Stanford prison study If this research were saying nothing new, and if its findings could all have been predicted in advance by everyone, then it would amount to nothing more than just another exercise in ‘Bubba Psychology’. As such, it would have joined the legions of other studies instantly buried in the So-what-who-cares? Archives of Psychology. Detailed records kept of the dispersion of information about this experiment over

On the ethics of intervention

in human psychological

research

251

the past two years reveal the following pattern: 215 Phone calls to office requesting information about the study. Correspondence requesting written information, confirming findings, questioning implications (reports sent to each one) 202 Prisoners 70 Correctional personnel 30 Criminal justice organizations 25 Legislators and politicians 210 Social scientists 90 Students 230 General public Mass media coverage has included: 1. A 20-minute feature on NBC-TV Chronolog (to an estimated audience of 3 million). 2. A lo-minute feature on NET-Public Television of AAAS panel on Prisons. 3. 12-Minute feature on CBS-TV (Los Angeles). 4. A variety of local TV and radio programs. 5. Featured story in Life magazine (10/15/71). 6. One of the few reports of a single experiment ever published in The New York Times Magazine (4/8/73) and syndicated in 14 other newspapers throughout the USA and Canada. 7. Articles and editorials in over 100 other newspapers in the USA and throughout the world. Interest by the U.S. Congress in the form of an invitation to appear as a witness before a special sub-committee investigating prison reform (October 1971), with a statement published in the Congressional Record. Invited and appeared as witness before Senate sub-committee studying problems 1973). Presented of juvenile justice and juvenile detention procedures (September audio-visual show of our research to the committee. The Chairman, Senator Birch Bayh, stated that the research and presentation had a significant impact on his thinking. Public speaking engagements by principal investigator and research associates (Haney, Banks, Jaffe, Prescott, White and Phillips) at national psychological and science conventions, to over two dozen colleges and high schools, to prisoner groups, Sheriff’s deputies, judges, parole officer’s units, several Chamber of Commerce meetings, law school faculty groups and others. A slide-sound, self-contained presentation of this experiment was prepared by the author and Greg White and has been distributed to colleges, high school and correctional groups. The feedback of its emotional impact and discussion-

252

Philip G. Zimbardo

generating appeal has been unequivocally and unanimously enthusiastic. (is) The above material has been requested and submitted (along with an affidavit by me) as supporting evidence in a class action law suit currently being brought against the New York State Department of Corrections by a team of lawyers and social scientists (Wallace et al., vs. Kern et al., U.S. District Court, Eastern District of New York). @I The results of the study have been used by a citizen’s group in a legal action to prevent the construction of a new, large, impersonal prison in Contra Costa County, California, in favor of more costly, smaller, more personal facilities and community-based facilities. 6) The President of the Louisiana State Senate is developing a coordinated research-, social-legal action program to reform the state’s juvenile delinquency facilities, and to change attitudes of the public toward the need for prison reform in general. This program was in part stimulated by this author’s research on attitude change and on the psychology of imprisonment. I shall serve as one of his senior consultants on this innovative project. (j) Of the several experiments (and Ph.D. dissertations) which have been specifically designed to extend the ideas contained in our study, the most’noteworthy is a mock-psychiatric ward study done at Elgin State Mental Hospital in Illinois (Orlando, 1973). Using 58 staff members in the roles as mock-patients and roleplaying staff, this experiment substantiated many of our findings about the dehumanizing effects of institutionalized roles and rules. The participants in Orlando’s study have since formed an action group to improve the social psychological treatment of patients at their hospital. How can we account for the unprecedented amount of publicity and broad appeal this study has generated? We have been told by people who are using the results of this research in a variety of social-legal actions that it is one of the most convincing demonstrations of the pathological impact of a prison-like environment on human behaviour. Not that anyone ever doubted the horrors of prison, but rather, it had been assumed that it was the predispositions of the guards (‘sadistic’) and prisoners (‘sociopathic’) that made prisons such evil places. Our study holds constant and positive the dispositional alternative and reveals the power of social, institutional forces to make good men engage in evil deeds. In contrast to the observations of generations of criminologists, and first-person accounts of prison life, this study is unique in its appropriate utilization of intervention methodology: A diagnostic selection procedure, random assignment of subjects to treatments and a careful recording of the process and chronology of the psychology of becoming imprisoned. In the words of a team of social psychologists recently reviewing the status of small group research (Helmreich, Bakeman and Scherwitz, 1973), the heuristic value of

On the ethics of intervention in human psychological

research

253

this research is similar to that of Milgram’s. ‘The upset generated by a Milgram or Zimbardo, both from the public and from their colleagues, in part stems from ethical concerns. But another part of their power lies precisely in their demonstration of how strong situational determinants are in shaping behavior. No resort to a correlation between “those” people who do “evil” things is allowed: the subjects were randomly assigned. It is the experimental method, not a fascination with ‘the artificial, that convinces. ‘Milgram’s and Zimbardo’s studies evoke public outcry in part because, through shaming dramatizations, they remind us just how fragile our ethical independence and integrity really are’ (p. 343). To be sure, we believe we have discovered other things of academic and scientific value (e.g., the function of explicit and implicit rules in behavior control, conditions which promote preferences for the use of punishment over reward by training agents, as well as clues to investigating subtle forms of psychological prisons, such as shyness). But the social value of this study is in demonstrating what a mock-prison environment could do to healthy, law-abiding, middle-class young men in less than a week. Moreover, we demonstrated it to those middle-class people who make the laws, enforce them and pay the bulk of the taxes which finance the operation of prison systems everywhere. It is these people who have hitherto been ignorant of, or actively unconcerned about, what prisons are or what they are doing to too many men, women and children subjected to their treatment procedures. If the roles of prisoner and guard are well known and being considerably discussed of late, it is not unduly immodest to believe that our research has contributed to that increased concern and public discussion. The major recent prison riots in America occurred just after our study was completed and had been reported in the media. The murder of George Jackson, several guards and inmates at San Quentin took place August 21, 1971. Attica was less than one month later. These dramatic events forced prisons into the awareness of the public; our study had already made all of us who had spent a week in the Stanford prison aware of the hell of prison. In some small way we believe our analysis of the ingredients which create such hells may help to change them.

A note on precautions and postcautions The final ethical point to be raised in this paper is the extensive precautions we took prior to putting the subjects in our prison, and the follow-up activities we have conducted to ensure that any possible chronic, negative after-effects were identified and adequately treated.

254

Philip G. Zimbardo

We did consider whether there was an alternative methodology to use which might avoid the possible distress to the subjects and still yield the information we sought. Since we were interested in the general psychology of imprisonment and not just in role-playing, or anonymity, or other specific variables and processes, there seemed to be no suitable alternative. The legal counsel of Stanford University was consulted, drew up a formal ‘informed consent’ statement and told us of work, fire, safety and insurance requirements we had to satisfy (which we did). The ‘informed consent’ statement, signed by every participant, specified that there would be an invasion of privacy, loss of some civil rights and harassment. Neither they nor we, however, could have predicted in advance the intensity and extent of these aspects of the prison experience. We did not, however, inform them of the police arrests, in part, because we did not secure final approval from the police until minutes before they decided to participate and, in part, because we did want the mock arrests to come as a surprise. This was a breach, by omission, of the ethics of our informed consent contract. The staff of the university’s Student Health Department was alerted to our study and prior arrangements made for any medical care which might be required. Approval was officially sought and received in writing from the sponsoring agency, ONR, the Psychology Department and the University Committee of Human Experimentation. The members of this committee, like the subjects themselves, did not predict the impact a mock prison could have on a group of carefully screened, healthy students. Since that time, however, armed with the hindsight from the knowledge of our results, they have ‘withheld approval’ for a subsequent study designed to investigate the positive effects of different types of prison guard training on the prisonerguard interaction. Their reasoning is, if the training variable does not make a difference, the outcome may again be as negative as we have shown it to be previously. Similarly, our former subjects will think twice and demand more knowledge before they again situational sign away their ‘informed consent’, since they no longer underestimate control of behavior nor so overvalue dispositional dominance. Following the study, we held group and individual debriefing sessions, had all subjects return post-experimental questionnaires several weeks later, several months later, and at yearly intervals. Many submitted retrospective diaries and personal analyses of the effects of their participation. We have met with most of the subjects since the termination of the study singly or in small groups, or where that was not possible, have discussed their reactions in telephone conversations. We are sufficiently convinced that the suffering we observed, and were responsible for, was stimulusbound and did not extend beyond the confines of that basement prison. In the future, we will insist that students, or representatives of the population being studied, be part of the University Committee to pass on the ethics of human experi-

On the ethics of intervention

in human psychological

research

255

mentation. We will encourage that committee to send an observer to, or secure recordings from, a pilot session of any ‘potentially unethical’ research. We will ourselves in the future incorporate a meta-experimenter in the role of unbiased monitor with ‘detached concern’. His/her task will be to assess the impact of the treatment on the subjects as well as the impact of the progress of the experiment on the researchers. Such a person should also have the authority to intervene on behalf of the subjects if necessary. Finally, I wish to contend that responsible and competent teachers and/or researchers welcome assessment and evaluation of their work by their peers and by those affected by their teaching and experimentation. The academic, scientific Ivory Tower is fast becoming a cliche as professors and social scientists become more directly involved in the life of their community and want meaningful feedback on the worth and impact of what they are professing and doing. The need for an aware, enlightened consciousness among psychologists about the ethics of intervention is more critical than ever, since we are at a time when our research findings and techniques are being ‘given away to the public’ in the form of therapies, remedial practices, selection procedures and cures for a host of social and personal problems. It would be ironic to limit our concern to the ethics of what happens to volunteers in a mock prison experiment and not also to what is happening daily to untold numbers of people suffering in social-political prison ‘experiments’ being conducted in our own cities and nations. But at the very least psychologists should themselves be vigilant of, and willing to take public exception to, research and practices of their colleagues which appear to violate basic principles of human rights, dignity and ethics. The proper initial course of action in such cases is to register a complaint with the American Psychological Association Ethics Committee, which has an investigatory body that thoroughly reviews such ethically questionable practices by its members. They have the power to bring considerable social pressure against offending members. Censure or ostracism from the APA carries serious professional penalties and personal shame. I have been informed by the head of the APA Ethics Committee, Dr. Brenda Gurel, that as of July 24, 1973, only one inquiry had been directed to her committee to investigate the ethics of the Stanford prison experiment. Curiously, it was not from Professor Harris Savin (who may have been too involved with the professional activity of publishing several critiques of the ethics of this research). Rather it was my recommendation to review the ethics of this research that was acted upon about a year ago by the APA. A final decision by that committee awaits the last follow-up evaluation being undertaken at this time.

256

Philip G. Zimbardo

REFERENCES Craik, K. H. (1973) Environmental psychology. In P. H. Mussen and M. R. Rosenzweig (Eds.). Annual review of psychology, 1973. Palo Alto, Armual Reviews, Inc. Pp. 403422. Haney, C., Banks, C., and Zimbardo, P. G. (1973) Interpersonal dynamics in a simulated prison. Intern. J. Criminol. Penal., 1,69-97. Helmreich, R., Bakeman, R., and Scherwitz, L. (1974) The study of small groups. In P. H. Mussen and M. R. Rosenzweig (Eds.) Annual review of psychology, 1974. Palo Alto, Annual Reviews, Inc. Pp. 337354. Milgram, S. (1974) Obedience to authority. New York, Harper & Row. Orlando, N. J. (1973) The mock ward: A study in simulation. In 0. Milton and

R. G. Wahler (Eds.), Behavior disorders: Perspectives and trends. Philadelphia, J. B. Lippincott. Savin, H. B. (1973) Professors and psychological researchers: Conflicting values in conflicting roles, Cog. 2 (I), 147-149. Seligman, M. E. P. (1973) Fall into Helplessness. Psychology Today, 7,43-48. Sommer, R. (1969) Personal space: The behavioral basis of design. Englewood Cliffs, N.J., Prentice-Hall. Toffler, A. (1970) Future shock. New York, Random House. Zimbardo, P. G., Haney, C., Banks, W. C., and JatIe, D. (1973) The mind is a formidable jailer : A Pirandellian prison. The New York Times Magazine, April 8, Section 6, 38-60.

6

Ethics

for gods and men

HARRIS 8. SAWN University of Pennsylvania

Professor Zimbardol has joined with the U.S. Navy to stage a happening that purported to be a scientific experiment but whose primary purpose was apparently to generate publicity for a political objective (prison reform). In order to stage the happening it was necessary for the professor to deceive students and then to degrade and brutalize them. He urges us to balance the possible good results of his political program against the harm he has done, but in his reckoning of the harm, he has attached no importance to the effects on the universities of students coming to regard professors as deceitful and dangerous, nor to the effects on the larger society of scientists collaborating with the military to disguise political publicity as scientific experiment. Utilitarian ethics are all very well for omniscient beings. The rest of us, if we try to live our lives on such ethical principles, run the risk of overlooking much unintended harm and, whatever our professed beliefs, acting on the theory that the end justifies the means.

1. Seepreceding article, On the ethics of intervention in human psychological research: With special reference to the Stanford prison ex-

periment, editors].

pages 243-256 of this issue [The

Cognition 2(2), p. 257