Some competence-process relationships in noun phrases with prenominal and postnominal adjectives

Some competence-process relationships in noun phrases with prenominal and postnominal adjectives

JOURNAL OF VERBAL LEARNING AND VERBAL BEHAVIOR, 8, 4 7 1 - 4 8 0 (1969) Some Competence-Process Relationships in Noun Phrases with Prenominal and Po...

838KB Sizes 0 Downloads 14 Views

JOURNAL OF VERBAL LEARNING AND VERBAL BEHAVIOR, 8, 4 7 1 - 4 8 0

(1969)

Some Competence-Process Relationships in Noun Phrases with Prenominal and Postnominal AdjectivesX J. E. MARTIN The Pennsylvania State University, University Park, Pennsylvania 16802

The view that the syntactic description of a sentence can provide a model of certain habits involved in sentence encoding has not been extensively tested. A theory of encoding was proposed in which some aspects of order in the base were taken to correspond to the psychological order of choice of morphemes for production. The theory was tested for English noun phrases with prenominal adjectives and for Indonesian noun phrases with postnominal adjectives. The results were consistent with predictions based on the theory. An analysis-by-synthesishypothesis of decoding was proposed, which contained the proposed encoding model as a subcomponent. Experimental results did not confirm the predictions based upon the decoding hypothesis. that the morphemes associated with the scanned "idea" are chosen for production. Lashley also suggested an ordering and integrating mechanism which arranged the morphemes in their proper order for production. Notice the two essential components of Lashley's theory as interpreted here. First, there is a scanning process during which relevant morphemes are chosen. The order of such scanning may be a function of the structure of the representation or of some situational factor which directs attention to a particular aspect of the representation. Second, there is an integrating and ordering process which orders morphemes accoraing to the rules of grammar. The output of this process would be the planned sequence of speech acts. The present theory is a n attempt to make the kind of theory Lashley sketched explicit in the context of a particular class of psycholinguistic phenomena. The phenomena in question are English nounphrases with prenominal adjectives and Indonesian nounphrases with postnominal adjectives. The grammatical descriptions of t This paper is based on some of the results reported these phrases will be given first. Second, it in a doctoral dissertation completed at the University will be shown how the descriptions are of Illinois. The author gratefully acknowledges his debt to Charles E. Osgood, whose invaluable advice hypothesized to model the psycholinguistic encoding of these phrases. In English, the was a constant source of help during this work. 471

One of the fundamental issues in psycholinguistics concerns the relationship between competence and process. Chomsky (1965) has emphasized that grammars are theories of competence, or underlying linguistic knowledge; not descriptions of process. Fodor and Garrett (1966) pointed out that the linguistic rules used in describing a sentence may not be directly represented in the psychological processing of that sentence and they propose that the grammatical generation of a sentence merely associates a given structural description to the sentence and does not necessarily model the psycholinguistic process. Despite these disclaimers, the author wishes to present a theory of the encoding of certain types of noun phrases in which aspects of grammatical descriptions are taken as models of psycholinguistic process. The theory proposed here is similar to that sketched by Lashley (1951). In that paper, Lashley proposed that the "idea" to be expressed is given nontemporal (perhaps spatial) representation. The representation was said to be scanned so

472

MARTIN

derivation of these phrases is: {N+VP, N + copula + A} => N(N + copula + A) + VP ~ NA + VP => A N + VP. 2 In Indonesian, the derivation is {N + VP, N + copula + A} ~ N(N + copula + A) + VP => NA + VP. These rules are simple and easily established. In this context, the hypothesis that the grammar provides a model of performance takes the following form. First, it is hypothesized that the left-to-right order of morphemes in the bases corresponds to the habitual real time order of choice of these morphemes. Second, it is hypothesized that the left-toright reordering of morphemes by transformations corresponds to the habitual reordering of those morphemes in real time for production. It is therefore proposed that speakers of English and Indonesian habitually choose adjectives after choosing the modified nouns. There are two kinds of evidence for this hypothesis. First, it seems evident that speakers generally choose nouns before modifying adjectives on common sense grounds. One must usually decide what one is going to talk about before one can decide what to say about it. Second, there are certain adjectives which must be chosen after the choice of the noun on logical grounds. The choice of many adjectives is a contextsensitive procedure; i.e., in order to appropriately choose these adjectives, it is necessary to consider the relation between the modified noun and the property denoted by the adjective. For example, one may describe a silver dollar as a large coin or as a small artifact. It is clear that the proper choice of the adjective depends upon the consideration of the meaning of the modified noun and the denoted property. These arguments support the view that the base structures of the phrase types under consideration correspond to the habitual psychological order of choice of morphemes for encoding. Consider the extension of the hypothesis to cases of multiple prenominal or multiple postnominal adjectives. It has often been noticed 2 N = noun, A = adjective, VP = verb phrase.

that such strings of adjectives are under strong constraints with respect to their preferred order. Martin (1968) has argued that there is no well-motivated syntactic account of adjective ordering; i.e., preferred adjective ordering has not been shown to be a function of syntax. Both English and Indonesian adjectives are subject to such constraints; e.g., in English, one preferred surface order is large red house and in Indonesian the corresponding phrase is literally translatable as

house red large. The stipulated syntactic analysis of these structures in English is, {N + VP, N + copula + A1 A2---A,} => N(N + copula + AI ---A,) + VP => NA1---A, + VP => A,---A1 N + VP. In Indonesian the analysis is, {N + VP, N + copula + AI Az---A,} => N(N + copula + A1---An) + VP => NA1---A, + VP. This analysis of the English construction is implied by Smith (1961). Note that the analyses imply that the base order of adjectives in English and Indonesian are the same if and only if the surface orders are opposite; i.e., in the two languages the base orders are implied to be the same, house red large, since the surface orders are different. This fact suggests that the constraints on the base order of adjectives may be the same in English and Indonesian. Martin (1968) has argued that the preferred order of adjectives in the English and Indonesian base structures are constrained by the definiteness of adjective denotation. Adjectives which denote the same property regardless of the meaning of the modified noun are said to be more definite in denotation than adjectives which denote different properties in the context of different nouns. Martin argued that adjectives preferred closer to the noun (further to the left) in English and Indonesian base structures are more definite in denotation than adjectives which are preferred further from the noun (further to the right). According to the proposed psycholinguistic theory, therefore, adjectives which are more

COMPETENCE-PROCESSRELATIONSHIPS

definite in denotation should be chosen for encoding prior to adjectives which are less definite in denotation. This is because the left-to-right order in the base is said to correspond to the first-to-last order of choice for encoding. This prediction can be made on other grounds. Adjectives which are low in definiteness require more nominal context for their appropriate choice than do adjectives which are high in definiteness. For example, consider the phrase the large red house. Given the color of the described object it is not necessary to take into account any of the meaning of the noun in order to choose appropriately the adjective denoting that color. Given the size of the described object it is necessary to take into account the most specific aspects of the noun's meaning in order to appropriately describe the size of the object. Adjectives of high definiteness of denotation should be more readily chosen than adjectives of low definiteness of denotation because less consideration of nominal context is required for their correct choice. Thus, in English, preferred adjective order should correlate negatively with the order of adjective choice and, in Indonesian, the correlation should be positive. A second argument that adjectives preferred closer to the noun are chosen earlier is the following. Lockhart and Martin (in press) showed that adjectives which are preferred closer to the noun are more easily learned as responses to the noun than are adjectives which are preferred further from the noun. It may be inferred that adjectives preferred close to the noun are generally capable of stronger associations with the noun than adjectives preferred further from the noun. Shapiro (1968) showed response latency to be a sensitive indication of association strength. Therefore, it may be predicted that those adjectives which are preferred closer to the noun will generally be of higher temporal accessibility than those preferred further from the noun. This expectation would be consistent with the view that adjectives preferred closer

473

to the noun are chosen prior to the choice of those preferred further from the noun. ENCODING STUDIES

In the experiments described below there was an attempt to gather evidence relevant to the proposed encoding hypothesis. It was thought that the order of choice of adjectives for encoding would be related to the accessibility of those adjectives for encoding. Therefore inferred adjective accessibility was assessed as a function of preferred adjective order and preferred adjective order was assessed as a function of inferred adjective accessibility. The first question to be examined was that concerning the relationship between adjective order as the independent variable and adjective accessibility, defined as response latency, as the dependent variable. Six experiments were conducted in order to test this relationship. The general outline of all six experiments was as follows. The Ss were presented with an array of four figures which were numbered one, two, three, and four. These figures were such that they could be cross-classified in terms of two two-valued dimensions. For example, in the first experiment, the figures were four colored circles--one large red circle, one small red circle, one large yellow circle, and one small yellow circle. In order to specify the independent variable, the Ss were presented with the arrays described above and then given adjectives appropriate for the description of the figures in the arrays. These Ss were requested to observe the figures in the arrays and describe each of them with adjective-adjective-noun phrases, e.g., that is a large red circle. The Ss were requested to give the order of adjectives which seemed most appropriate in the context of the arrays. In this way adjectives were assigned orders with respect to the array. In order to test the dependent variable, Ss were instructed to attend to one of the figures

474

M~TI~

and then were given a word representing one o f the t w o d i m e n s i o n s , e.g., the w o r d size o r color. T h e i r t a s k was to p r o d u c e the a p p r o p r i a t e v a l u e f o r t h e d i m e n s i o n p r e s e n t e d to t h e m as q u i c k l y as possible. T h u s , if t h e y h a d b e e n i n s t r u c t e d to a t t e n d to L A R G E R E D C I R C L E a n d h a d b e e n g i v e n t h e w o r d size, it was t h e i r t a s k to r e s p o n d w i t h t h e w o r d large as q u i c k l y as possible. By r e c o r d i n g t h e latencies o f t h e s e e n c o d i n g r e s p o n s e s , it w a s p o s s i b l e to d e t e r m i n e t h e r e l a t i v e accessibilities o f the a d j e c t i v e s i n v o l v e d . T h e r e l a t i v e accessibility o f t h e a d j e c t i v e s was t h e n c o m p a r e d w i t h t h e i r order. I n this w a y , t h e relationship between adjective order and a d j e c t i v e accessibility was tested. S u b j e c t s were run individually.

Experiment 1 Assessment of the independent variable. The first experiment involved the dimensions of size and color. The independent variable in this study was the relative ordering of the adjectives large or small and red or yellow when used in describing the figures of the stimulus array. Eighteen undergraduate volunteer Ss produced preferred orderings of the relevant adjectives in the context of the array figures. They were required to give 48 orderings, 12 responses to each figure. The ordering given by Ss was completely consistent both within and across Ss. Size adjectives were ordered further from the noun than color adjectives. It was therefore predicted that color adjectives would require shorter encoding times than size adjectives. This result would be taken as evidence that color adjectives are more accessible than size adjectives for these stimulus materials. Materials. The materials used in this study consisted of a 3 x 3 ft. white background on which were four figures, one in each quadrant. The figures were solid colored circles which varied both in color and size. Each figure was a different combination of size and color. The colors presented were red and yellow. The sizes presented were large (6-inch diameter) and small (3-inch diameter). Each figure was numbered by a number appearing below it. Procedure. In assessing the dependent variable, latency of response, five undergraduate volunteers from an introductory psychology course at the Pennsylvania State University were used. They were tested individually. They were seated facing the stimulus

array at a distance of 4.5 ft. The center of the array was at their eye-level. The Ss were given two types of auditorily presented cues on a tape recorder: pointing cues and dimensional cues. Pointing cues were designed to instruct Ss to attend to a particular figure on the array. A typical pointing cue would consist of the following command: "Ready! Number one!" These pointing cues were followed by dimensional cues. Dimensional cues were designed to call attention to a particular dimension of the figure specified by the pointing cue. The two dimensional cues utilized in this experiment were the words "Size" or "Color." After having received each pair of such cues, Ss responded simply by saying the adjective which characterized the figure pointed out by the pointing cue in terms of the dimension referred to by the dimensional cue. The dependent variable was the interval of time between the dimensional cue and S's response. There were three conditions which were defined in terms of the interval between the pointing cue and the dimensional Cue. In the first condition, the interval was .25 sec. In the second condition the interval was 1 sec. In the third, the interval was 3 sec. These conditions, each tested twice, were counter-balanced over the six sessions in order to reduce the possibility of practice being confounded with conditions. The reason for varying the interval between the pointing cue and the dimensional cue was that it was hoped that the effects could be interpreted as being independent of time required for perception. It was thought that in a longer interval, the time required for perception would be less likely to be confounded with other factors than in a shorter interval. If it were the case that time required for the perception of the item designated by the pointing cue is an important factor, then the effects might interact with the temporal interval. Each S was tested six times, once on each of six different days, with 48 observations being made in each test session. Within each session, 12 observations were made with each of the four figures as the stimulus. Within each of these sets of 12 observations, six were made in the context of the dimensional cue size and six were made in the context of the dimensional cue color. Within the above-mentioned constraints, the order of observations was randomized. Apparatus. The apparatus used for recording latency included a four-channel tape recorder, a voice key, and a standard electric clock timer. Channels one and three of the taped recording contained the pointing cues. Channels two and four, which contained the dimensional cues, were wired to the voice key and initiated the timer.The Ss wore a throat microphone which was connected to the voice key in such a way that their vocal response would stop the timer.

COMPETENCE-PROCESS RELATIONSHIPS

Results. The critical c o m p a r i s o n is that between the response latencies associated with the d i m e n s i o n a l cue size a n d the response latencies associated with the d i m e n s i o n a l cue color. It was predicted that the former w o u l d be significantly longer t h a n the latter. The m e a n s were 695 a n d 646 msec., respectively. A n analysis o f variance indicated this difference to be significant, F(1, 4) = 9.23, p < .05. The prediction is thus confirmed. Adjective r a n k is related to adjective accessibility in the way that the proposed e n c o d i n g model w o u l d require. There were n o effects due to interval. Experiment 2 Materials. The figures were hand drawn pictures of four male faces which varied in age and mood. The degrees of age represented were young and old age. The moods represented were happiness and sadness. Otherwise, the array was the same as that used in Exp. 1. In testing the independent variable it was found that of 18 Ss, 14 consistently ordered age adjectives closer to the noun than mood adjectives. A sign test showed the tendency to order adjectives in this way to be significant, p < .03. The adjectives used were happy or sad and young or old. Procedure. The method of assessing the dependent variable was equivalent to that used in Experiment 1, with the exception that the materials and adjectives were as specified abox,e. There were five Ss tested individually. Results. It was predicted that adjectives d e n o t i n g degree of age would be more quickly encoded t h a n those d e n o t i n g mood. The m e a n s were 731 msec. for age a n d 920 msec. for m o o d . Analysis o f variance showed the difference to be significant, F(1, 4 ) = 21.40, p < .01. The predictions derived f r o m the model were confirmed. There was a m a i n interval effect. The m e a n s were 749 reset, for .25 see., 774 msec. for 1 sec., a n d 953 reset. for 3 sec. Analysis of variance gave F(2, 8) = 10.00,p < .01. There was also a n interval b y d i m e n s i o n interaction, F(2, 8) = 4.90, p < .05. This i n t e r a c t i o n was such that the differences in reaction times between dimensions increased with the interval between the pointing a n d d i m e n s i o n a l cues.

475

Experiment 3 Materials. The materials used in this study were the same as in Exp. 2, except for the following differences. The drawings in this experiment were intended to depict the dimensions of intelligence and mood. The moods represented were denoted by happy and sad. The degrees of intelligence represented were denoted by smart and dumb. Intelligence was represented by relative sharpness of features. The purpose of this experiment was to assess the relative latencies associated with adjectives which are not ordered, i.e., whose ordering within phrases is optional. In testing the independent variable the adjectives were not found to be consistently ordered across the group of 18 Ss. From this group, five Ss were selected who showed no tendency to give a consistent ordering as individuals. These Ss were then used in the latency task. Procedure. The method of assessing the dependent variable was the same as in Exp. 1, with the exception of the materials and adjectives described above. Results. It was n o t expected that there w o u l d be significant difference between e n c o d i n g times required for adjectives d e n o t i n g m o o d a n d adjectives d e n o t i n g intelligence. M e a n latencies were 855 a n d 836 msec., respectively. This difference was n o t significant, F ( I , 4) < 1. There were n o effects due to interval. Experiment 4 Materials. The materials used in this study were similar to those used in Exp. 3. The drawings were intended to depict degrees of intelligence and mood. However, the drawings used were not the same drawings used in Exp. 3. The difference between these figures and those used in Exp. 3 was that the figures used in this experiment elicited an ordering of adjectives from Ss when Ss used happy, sad, smart, and dumb in describing them. This difference in ordering appeared to be due to some unspecified differences in the drawings. In testing the independent variable, 14 out of 18 Ss ordered the adjectives denoting intelligence further from the noun than the adjectives denoting mood. The tendency to order adjectives in this way was significant, p < .03. Procedure. The method of testing the dependent variable was equivalent to that used in Exp. 1~with the exception of the materials and adjectives specified above. There were eight randomly selected Ss in this case, run individually. Results. It was predicted that adjectives den o t i n g m o o d would be encoded more quickly

476

MARTIN

t h a n adjectives d e n o t i n g intelligence. The m e a n latencies were 749 a n d 845 msec., respectively. A n analysis of variance showed this difference to be significant, F(1, 7 ) = 25.13, p < .01. The p r e d i c t i o n s derived from the model were again confirmed. There were no effects due to interval.

Experiment 5 Subjects. This experiment was performed with speakers of Indonesian as Ss. Subjects were graduate students at the Pennsylvania State University. It should be noted that with most of these Ss, Indonesian was a second language. This is because Indonesian is only the official language of Indonesia. Most of the Ss learned Indonesian in schools. However, in every case, the dialect that was their first language was equivalent to Indonesian with respect to the grammar of the adjective. Adjectives appeared after the noun and with respect to preferred order in reverse of that found in English. Materials. The stimulus array used in this experiment was the same as that used in Exp. 2. The words used in pointing and dimensional cues were the Indonesian translations of the English words used in Exp. 2. The independent variable was established by questioning the Ss as to the preferred order of the adjectives involved. All five Ss agreed that adjecitves denoting age normally appeared closer to the noun than adjectives denoting mood. Given the structure of the Indonesian grammar, this implies that age adjectives are introduced into the grammatical derivation of noun phrases with multiple adjectives prior to mood adjectives. Therefore, the model under test requires that in Indonesian, as well as in English, adjectives denoting age should be more accessible than adjectives denoting mood. It was predicted that the Indonesian adjectives denoting age would be encoded more quickly than the Indonesian adjectives denoting mood. Procedure. One small difference in design between this experiment and Exp. 2 is that here the number of sessions was three instead of six. Each session was run under a different time condition. This means that the number of observations was one-half that of Exp. 2. The reason for this was that it was becoming increasingly clear that the data in all these experiments were sufficiently stable to warrant this reduction of observations.

Results. The results were clear a n d i n the expected direction. The m e a n time required to encode I n d o n e s i a n words d e n o t i n g age was 9 t 8 msec. as c o m p a r e d to 1,036 required

for d e n o t i n g m o o d . Analysis of variance showed this difference to be significant, F(1, 4 ) = 44.43, p < .01. There was a m a i n interval effect. The m e a n s were 907 msec. for .25 sec., 975 msec. for 1 sec. a n d 1,049 msec. for 3 sec. Analysis of variance s h o w e d this effect to be significant, F(2, 8) = 4.46, p < .05. There was n o i n t e r a c t i o n involving interval.

Experiment 6 Subjects and materials. This experiment was also performed with Indonesian speakers and Indonesian verbal materials. Five Ss participated, being run individually. The array was once again a set of four figures representing human faces. The dimensions utilized were translatable into English as honesty and intelligence. The adjectives used in the experimental procedure were translatable as honest, dishonest, intelligent, and unintelligent. These dimensions were chosen because the Indonesians reported no preferred order for the adjectives when the array was present. Therefore, the independent variable as assessed indicated no differential ranking of adjective.

Results. It was n o t expected that there w o u l d be a significant difference between adjectives d e n o t i n g honesty a n d those d e n o t i n g intelligence. The m e a n time required to encode adjectives d e n o t i n g honesty was 1,076 msec. The m e a n time required to encode adjectives d e n o t i n g intelligence was 1,046 msec. This difference was n o t significant, F(1, 4 ) = 4.50, p > .05. There were n o effects due to interval. Experiment 7 The experiments discussed above were designed to assess the relative accessibility of adjectives as a f u n c t i o n of preferred adjective order. The present experiment was designed to test preferred adjective order as a f u n c t i o n of relative adjective accessibility. Shapiro (1968) showed that the frequency of association of a pair of words in word-association n o r m s correlates with the accessibility of one of the pair when it is being learned as a response to the other m e m b e r of the pair. Therefore, it was considered possible to affect differentially adjective accessibility by varying

477

COMPETENCE-PROCESS RELATIONSHIPS the association strength between the adjectives a n d t h e m o d i f i e d n o u n s . It was p r e d i c t e d t h a t a given adjective would tend to appear closer to t h o s e n o u n s w h i c h elicit it in a r e s t r i c t e d a s s o c i a t i o n t a s k t h a n to t h o s e n o u n s w h i c h d o n o t elicit it in s u c h a task. T h i s p r e d i c t i o n was based on the hypothesis that adjective order w a s a f u n c t i o n o f t h e accessibility o f a d j e c t i v e s during encoding.

Subjects. Forty-seven undergraduates from an introductory course in psychology at Pennsylvania State University served as Ss. Materials. Twenty-six pairs of adjective-adjectivenoun triplets were constructed. The members of a pair had the same adjectives but different nouns. The relationship between the nouns and adjectives was such that each adjective was an associate to only one of the nouns in restricted association norms (collected by R. S. Lockhart for another purpose) with N = 100. For example, one pair of triplets consisted of tall, bright, lamp, and tall, bright, mountain. In this case, the norms had shown that tall was a strong associate of mountain and that bright was a strong associate of lamp. The norms did not give any evidence that tall was an associate of lamp or that bright was an associate of mountain. Furthermore, there was no evidence that either tall or bright elicited mountain or lamp. Procedure. Each adjective-adjective-noun triplet was presented such that the two adjectives could be written in front of the noun in any order the Spreferred. Triplets were presented randomly, with the constraint that the second member of a pair of triplets was not presented until all the first members of each pair of triplets had been presented. It was predicted that the ordering would be systematically related to the noun-adjective association strengths. For example it was predicted that tall bright lamp should be more frequent than tall bright mountain. Results. T h e r e w e r e 26 c o m p a r i s o n s . F o r o n e p a i r o f triplets t h e o r d e r i n g d i d n o t differ b e t w e e n n o u n s . F o r t h r e e p a i r s o f triplets t h e difference in o r d e r i n g b e t w e e n n o u n s w a s n o t in t h e e x p e c t e d d i r e c t i o n . I n t h e r e m a i n i n g 22 p a i r s o f triplets t h e difference in o r d e r i n g b e t w e e n n o u n s w a s in t h e e x p e c t e d d i r e c t i o n . By a sign test, this r e s u l t is significant, p<.0001. T h e m e a n difference b e t w e e n o r d e r i n g s p e r p a i r o f t r i p l e t s w a s 2.77 Ss. A l t h o u g h , r e l a t i v e l y s m a l l in a b s o l u t e m a g n i t u d e , this f i n d i n g w a s c o n s i s t e n t w i t h t h e

hypothesis that adjectives influenced by association control adjective order.

accessibility as strength could

DECODING STUDIES

Experiment 8 The materials used in this experiment were the same as those used in Exp. 1. The procedure of the decoding experiments was as follows: The Ss were seated facing the array. After a specified interval, they were first given a pointing cue which directed their attention to one of the members of t h e array. After a specified interval, they were given another cue. Instead of a dimensional cue, Ss in the decoding experiments were given an adjective cue. This adjective cue denoted one of the four properties relevant to the array. For example, in Experiment 8, the following would be adjective cues: large, small, red, and yellow. The Ss were then required to respond simply with either "yes" or "no," depending on whether or not the adjective presented in the cue could be said to denote the property foundia the figure which was being attended to. The latency between the adjective cue and S's response was the dependent variable. The apparatus and procedures for recording latencies were the same as in the encoding experiments. As in the case of the encoding experiments, the independent variable was the relative ranks of the adjective relevant to the array. Therefore, in Exp. 8, the independent variable was the same as had been establishedin Exp. 1. It was predictedthatthedependent variable would co-vary with the order of the adjective. Specifically, it was predicted that adjectives closer to the noun in preferred order would be associated with shorter reaction times than adjectives of higher rank. There were five Ss. The results did not confirm the prediction. There was no significant difference in mean latency, F(1, 4) = 2.25, p > .05. The mean reaction time associated with adjectives denoting size was 639 msecs.; the mean reaction time associated with adjectives which denote color was 651 msec. This is actually the opposite of the direction predicted. There was a main interval effect. The means were 720 msec. for .25 sec., 590 msec. for 1 sec., and 624 msec. for 3 sec. Analysis of variance showed this effect to he significant, F(2, 8)= 4.94, p < .05.

Experiment 9 This experiment involved the same materials and dimensions as were used in Exp. 2. There were six Ss. The independent variable was as specified in the second encoding experiment. The prediction was not confirmed. The mean decoding time of adjectives denoting age was 744 msec., while the mean decoding time for adjectives

478

M~TIN

denoting mood was 768 msec. The difference between these means was not significant F(1, 5) = 3.26, p > .05. There was no interval effect.

Experiment 10 The materials and dimensions involved in this experiment were the same as were used in Exp. 3. Since there was no difference in order between adjectives denoting intelligence and those denoting mood, it was predicted that the mean latencies associated with the two dimensions would not be significantly different. There was no significant difference between dimensions, F(1, 5) = 1.66, p > .05. The mean latency for adjectives denoting intelligence was 764 msec. and the mean latency for adjectives denoting mood was 752 msec. There was a main effect due to intervals. The means were 796 msec. for .25 sec., 740 msec. for 1 sec., and 738 msec. for 3 sec. Analysis of variance showed this effect to be significant, F(92, '10) = 6.80, p < .01.

Experiment 11 The materials and dimensions involved in this experiment were the same as were used in Exp. 4. The independent variable was the same as in that experiment. The prediction was not confirmed. The mean latency for adjectives denoting intelligence was 775 msec. and that for adjectives denoting mood was 796 msec. ; this difference is not significant, F(1, 5)= 3.88, p > .05. The direction of the difference is opposite to that predicted. There was no interval effect.

DISCUSSION I n summary, two main hypotheses were tested in these experiments. The first hypothesis, concerned with the encoding process, was that adjective order is closely related to adjective accessibility. This hypothesis derives f r o m the more general assumption that the relationship between description o f competence and psychological process is relatively direct. A set o f experiments (1 t h r o u g h 6) were conducted which tested adjective accessibility as a function o f adjective order. A n o t h e r experiment (7) was run which tested adjective order as a function of inferred adjective accessibility. The results were consistent with the view that the necessary and sufficient conditions for differential adjective order was differential adjective accessibility.

The second main hypothesis, concerned with the decoding process, was that adjective order is closely related to the speed o f adjective decoding. It derives f r o m the more general notion that the decoding process is best construed as an analysis-by-synthesis procedure o f which the encoding system supported above is a component. This h3~pothesis was not supported. The argument supported the view that adjectives are habitually chosen after the noun. In adjective-noun phrases in English and noun-adjective phrases in Indonesian, the g r a m m a r m a y be taken to be a model o f performance if the left-to-right dimension in the base is taken to correspond to the real time order of m o r p h e m e choice in the performance model. Both the arguments and the evidence reported here are consistent with the view that the order of choice o f adjectives for encoding corresponds to the order o f adjective introduction into the base of the associated grammatical derivations. In the case o f multiple as well as single adjectives, therefore, it appears that the grammatical descriptions provide a reasonably g o o d model o f habitual psychological process. Whether this result m a y be generalized to other constructions is a question to be answered by further research. It is also appropriate to note the limitations of the experiments discussed above. Perhaps the main limitations of the encoding latency studies was that the n u m b e r o f comparisons in relation to the n u m b e r of ordered and unordered adjective pairs was relatively small. Because of the novelty of the experimental design there were doubtless m a n y u n k n o w n factors which might have affected these results. Even t h o u g h each o f the six experiments produced the expected results, it is possible that some confounding factor was responsible for them. There is no d o u b t that further comparisons of adjective accessibilities would increase confidence in the findings. It is conceivable that the differential reaction times were a function o f the time required for the decoding o f the dimensional cues.

COMPETENCE-PROCESS RELATIONSHIPS

However, the negative results of the decoding studies argue against this possibility. Differential reaction times may have been a function of coding of items in terms of the adjective appearing closest to the noun. This could have been caused by the representation of only one kind of object on each array. Thus, the Ss were influenced to code the objects in terms of the adjectives closest to the noun instead of the noun itself. This is a possible interpretation of the data, but it need not be contradictory to the hypothesis proposed in this paper. The relative importance of dimensions for the coding of objects may influence the accessibility of adjectives. If so, and if that relative importance correlates with adjective order, then the adjective order should correlate with adjective accessibility. With respect to the study showing adjective order to be a function of association strength between adjectives and nouns, an alternative explanation can be given. It can be argued that (a) adjectives strongly associated with the nouns are preferred closer to the noun because they have been associated closely with those nouns in the past, having frequently modified them as single adjectives in the Ss' past experience, and (b) adjectives ordering was not affected by strength of association in some underlying structure, but rather by strength of association in the surface or produced structure. This explanation is possibly correct. Both of the above qualifications are worthy of note. Whenever there are as many unknown variables, as there certainly are in this area, it is only appropriate that any results be interpreted from a conservative perspective. The qualificatiohs only point out our limited understanding at present. They do not invalidate the claim that the results are consistent with the encoding hypothesis. The results were not consistent with the analysis-by-synthesis hypothesis considered here. As such, they gave indirect support to the views presented by Fodor and Garrett (1966). It is their theory that decoding involves the

479

use of heuristic procedures for directly inferring the base without considering the derivation of the surface structure from the base. It may be asked why such a reversal of order should appear in a natural language such as English; i.e., why should adjectives, which are chosen after nouns, be expressed before nouns and why should those adjectives which are chosen last be the first adjectives to be expressed. It would seem that the Indonesian construction would be psychologically simpler and therefore preferable. A possible answer to this question lies in the fact that, as Lockhart and Martin (in press) have shown, adjectives are generally less available for recall than are nouns. Furthermore, Lockhart and Martin have shown that the adjectives which are expressed first in English are less available for recall than those which are expressed later. These facts indicate that there may be a tendency to formulate the grammar so that those morphemes associated with the least stable memory trace are produced first. Such a formulation would aid the encoder by allowing him to express such items soon after their choice, thus freeing storage capacity required for further computation. It might aid the decoder by facilitating the retention of difficult to remember items. Due to the serial position effect it would be an advantage to the decoder to receive the adjective associated with the least stable memory trace first. Finally, if the constraints on the order of adjective choice are independent of particular syntactic constraints and dependent upon psychological constraints affecting accessibility, the order of choice of adjectives may well be universal for speakers of all languages. This suggests the hypothesis that adjective order in the base may be a linguistic universal. Such a hypothesis must be supported on linguistic grounds. If support were found, a very deep relationship between the linguistic and psychological analyses would have been established. The universal linguistic account could be taken to model certain universal

480

MARTIN

psychological processes involved in adjective choice. REFERENCES CHOMSKY, N. Syntactic Structures. 's-Gravenhage: Mofiton, 1965. FODOR,J., & GARRETT,M. Some reflections on competence and performance. Psycholinguistic Papers. Proceedings of the 1966 Edinburgh Conference. Edinburgh: University Press, 1966. FODOR, J., & GARRETT, M. Some syntactic determinants of sentential complexity. Massachusetts Institute of Technology. Unpublished paper, 1967. LASHLEY, K. S. The problem of serial order and behavior. In L. A. Jeffress (Ed.), Cerebral Mechanisms and Behavior. New York: Wiley, 1951.

LOCKHART,R. S., d~;MARTIN,J. E. Adjective order and the recall of adjective-noun triplets. Journal of Verbal Learning and Verbal Behavior, 1969, 8, 272-275. MARTIN,J. E. A study of the determinants of preferred adjective order in English. Unpublished doctoral dissertation, University of Illinois, 1968. SHAPIRO, S. I. Response latencies in paired-associate learning as a function of free association strength, hierarchy, directionality, and mediation. Research Bulletin No. 66, Pennsylvania State University, 1966. S~Tn, C. S. A class of complex modifiers in English. Language, 1961, 37, 342-365.

(Received December 16, 1968)