Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

The effect of context on the role of imagery in language processing Sutherland, Brian Ross 1975

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-UBC_1975_A8 S98_4.pdf [ 5.11MB ]
Metadata
JSON: 831-1.0093469.json
JSON-LD: 831-1.0093469-ld.json
RDF/XML (Pretty): 831-1.0093469-rdf.xml
RDF/JSON: 831-1.0093469-rdf.json
Turtle: 831-1.0093469-turtle.txt
N-Triples: 831-1.0093469-rdf-ntriples.txt
Original Record: 831-1.0093469-source.json
Full Text
831-1.0093469-fulltext.txt
Citation
831-1.0093469.ris

Full Text

THE EFFECT OF CONTEXT ON THE ROLE OF IMAGERY IN LANGUAGE PROCESSING by BRIAN ROSS SUTHERLAND B. Sc., U n i v e r s i t y of B r i t i s h Columbia, 1971 A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS i n the Department of PSYCHOLOGY We accept t h i s t h e s i s as conforming to the req u i r e d standard THE UNIVERSITY OF BRITISH COLUMBIA A p r i l , 1975 In presenting th i s thes is in pa r t i a l fu l f i lment of the requirements for an advanced degree at the Univers i ty of B r i t i s h Columbia, I agree that the L ibrary shal l make it f ree ly ava i l ab le for reference and study. I further agree that permission for extensive copying of th is thes is for scho lar ly purposes may be granted by the Head of my Department or by his representat ives. It is understood that copying or pub l i ca t ion of th is thesis fo r f inanc ia l gain shal l not be allowed without my written permission. Department of The Univers i ty of B r i t i s h Columbia Vancouver 8, Canada Date dfAJJ 29;rf7S~ Abstract The r o l e of imagery i n language processing has received much recent attention. Paivio's two-process theory implies that concrete and abstract material are represented d i f f e r e n t l y i n memory. S p e c i f i c a l l y , concrete language i s assumed to be retained i n terms of an image which summarizes the meaning of the material i n an integrated form. Abstract language, on the other hand, i s assumed to be represented i n a verbal-sequential form, with l i t t l e i n t e g r a t i o n of meaning. A review of the evidence on the e f f e c t s of concreteness on memory for meaning indicates that Paivio's theory does not provide a completely adequate account of the r e s u l t s . In many of the studies which have pro-duced findings consistent with the theory, an explanation i n terms of differences i n comprehensibility seems as appropriate as one based on d i f f e r e n t modes of storage. In addition, some studies have shown a sub-s t a n t i a l degree of semantic i n t e g r a t i o n i n abstract language. In an attempt to provide some c l a r i f i c a t i o n of t h i s issue, a Levels of Processing model was proposed as an a l t e r n a t i v e to Paivio's theory. The model i s based on the assumption that while abstract and concrete language t y p i c a l l y d i f f e r i n access to semantic processing, memory for meaning may be independent of concreteness under c e r t a i n conditions. In p a r t i c u l a r , context may increase semantic processing i n abstract language and thereby reduce the s u p e r i o r i t y of concrete language i n memory for meaning which would otherwise be predicted. Two experiments were c a r r i e d out to evaluate the model. In the f i r s t , the e f f e c t of concreteness, context, and presentation time on recog-nition of meaning and wording changes i n sentences was investigated. The i i i r e s u l t s supported the model i n that memory f o r meaning r e l a t i v e to wording increased as a f u n c t i o n of concreteness only when the sentences were pre-sented i n i s o l a t i o n from context. When the sentences were presented i n the context of meaningful paragraphs, r e l a t i v e memory f o r meaning was equal i n concrete and a b s t r a c t m a t e r i a l . The second experiment assessed the image-evoking c a p a c i t y , c o m p r e h e n s i b i l i t y , and degree of meaning change i n t e s t items, f o r the m a t e r i a l s used i n the f i r s t experiment. The r e s u l t s allowed s e v e r a l a l t e r n a t i v e i n t e r p r e t a t i o n s of Experiment I to be discounted. In p a r t i c u l a r , P a i v i o ' s two-process theory was shown to be incapable of accounting f o r the e l i m i n a t i o n of the concreteness e f f e c t on memory f o r meaning as a f u n c t i o n of context. I t was concluded that the Lev e l s of Processing model provides a more v i a b l e account of the r o l e of imagery i n language processing than P a i v i o ' s two-process theory. TABLE OF CONTENTS Page ABSTRACT i 1 TABLE OF CONTENTS . i i f LIST OF TABLES . . v LIST OF FIGURES v i i ACKNOWLEDGEMENT v i i i INTRODUCTION 1 TWO-PROCESS THEORY AND THE ROLE OF IMAGERY IN LANGUAGE PROCESSING . Paivio's two-process .theory 2 Two-process theory and language processing 3 Evidence on two-process theory and language processing 4 SUMMARY AND IMPLICATIONS OF THE EVIDENCE 25 Imagery and levels of comprehension in language processing . . . . 26 THE EFFECT OF CONTEXT ON THE ROLE OF IMAGERY IN LANGUAGE PROCESSING 52 EXPERIMENT I . . . . 56 METHOD 56 RESULTS 64 EXPERIMENT II . . . . . 85 METHOD 89 RESULTS 93 GENERAL DISCUSSION 107 FOOTNOTES 113 BIBILIOGRAPHY> 114 APPENDIX , Examples of materials used in Experiment I 117 V LIST OF.TABLES page TABLE I Mean Proportion of Hits as a Function of Type of Change, Presentation Time, Concreteness, and Context 66 TABLE II Summary of the Analysis of Variance on Hits 67 TABLE III Mean Proportion of Hits Minus Mean Proportion of 69 False Alarms as a Function of Type of Change, Pre-sentation Time, Concreteness, and Context. TABLE IV Summary of the Analysis of Variance on Hits-False Alarms 70 TABLE V Mean Proportion of Semantic Hits as a Function of Presentation Time, Concreteness and Context. 74 TABLE VI Summary of the Analysis of Variance on Proportion of Semantic Hits. 75 TABLE VII Mean Proportion of Lexical Hits as a Function of Presentation Time, Concreteness, and Context 77 TABLE VIII Summary of the Analysis of Variance on Proportion of Lexical Hits 78 TABLE IX Mean High Confidence Hits as a Function of Type of Change, Presentation Time, Concreteness, and Context. 79 TABLE X Summary of the Analysis of Variance on High Confidence Hits 80 TABLE XI Mean Confidence Rating per Item as a Function of Type of Change, Presentation Time, Concreteness, and Context. 82 TABLE XII Summary of the Analysis of Variance on Confidence Ratings. 83 TABLE XIII Summary of the Analysis of Variance on Imagery Ratings for a l l Sentences 95 TABLE XIV Summary of the Analysis of Variance of the Imagery Ratings for C r i t i c a l Sentences 96 V 1 . LIST OF TABLES (cont'd) TABLE XV TABLE XVI TABLE XVII TABLE XVIII TABLE XIX TABLE XX Summary of the Analysis of Variance on Compre-hensibility Ratings for a l l Sentences Summary of the Analysis of Variance on Compre-hensibility Ratings for C r i t i c a l Sentences Summary of the Analysis of Variance on Compre-hensibility Ratings for Whole Sets Summary of the Analysis of Variance on Compre-hensibility Ratings for Test Sentences Summary of the Analysis of Variance on Similarity Ratings Mean Similarity of Meaning as a Function of Concreteness, Type of Change and Context page 98 99 101 102 103 105 LIST OF FIGURES Figure 1 - Design of Experiment I Figure 2 - Interaction Between Context,Concreteness and Type of Change in Experiment I v i i . page 60 72 v i i i ACKNOWLEDGEMENT My gratitude goes f i r s t of a l l to Dr. Dennis Foth for the friendship and scholarship that he has provided throughout this project. I also thank Dr. Jim Johnson, Dr. John Yuille, and Dr. Nancy Suzuki for serving on my thesis committee and providing a great deal of valuable advice. To Gloria Lachmanec, who typed the entire thesis on very short notice, goes my sincere appreciation. Finally, a special note of thanks to my wife Carol and the rest of my family, to whom this thesis is dedicated. Through the many d i f f i c u l t periods that were associated with this project, they never failed to give me their complete support. Having done so, they 1 ---<• share a great deal of the responsibility for the paper which follows. 1 INTRODUCTION In the past ten years, research in human memory has undergone a number of significant changes. Areas of investigation which had previously been neglected, due primarily to a lack of appropriate experimental tech-niques, are now receiving increased attention. Two of the most important examples of recently developed research topics are imagery and language. As indicated by the recent review by Paivio (1971), imagery has become, in a very short time, one of the most widely investigated concepts in cog-nition. Similarly, since the pioneer work of Chomsky (1957), the f i e l d of psycholinguistics has had increasing influence on research in human information processing. The purpose of this paper is to investigate the relationship between imagery and language processing. The most extensive theoretical treatment of this topic is provided by Paivio's two-process theory of meaning and mediation. After presenting the basic assumptions of the theory and the primary predictions i t makes about language processing, a c r i t i c a l review of the research in this area is provided. An alternative to the two-process approach is then proposed and a study is described which was carried out to evaluate this alternative model. ( The goal of the discussion i s to provide a more integrated account of the role of imagery in language processing than is presently available. In light of the degree of interest expressed in these topics recently, such a discussion seems most appropriate at this time. 2 TWO-PROCESS THEORY AND THE ROLE OF IMAGERY IN LANGUAGE PROCESSING P a i v i o ' s Two-Process Theory P a i v i o ' s t h e o r e t i c a l p o s i t i o n on the r o l e of imagery i n c o g n i t i o n i s summarized by what he c a l l s the two-process theory of meaning and mediation. The b a s i c assumption of t h i s theory i s that there are two q u a l i t a t i v e l y d i f f e r e n t coding systems a v a i l a b l e f o r the processing of inf o r m a t i o n : the imaginal system and the v e r b a l system. These two systems are d i s t i n g u i s h e d by three f u n c t i o n a l d i f f e r e n c e s i n t h e i r processing c h a r a c t e r i s t i c s , only two of which are r e l e v a n t here. F i r s t , the imaginal system i s assumed to be s p e c i a l i z e d f o r the processing of concrete i n -formation w h i l e the v e r b a l system i s more s u i t e d f o r a b s t r a c t information."*" Second, the imaginal system has the c a p a c i t y f o r p a r a l l e l processing w h i l e the v e r b a l system i s c h a r a c t e r i z e d by s e q u e n t i a l processing. These t h e o r e t i c a l d i s t i n c t i o n s provide the b a s i s f o r examining the i m p l i c a t i o n s of P a i v i o ' s theory w i t h regard to the f u n c t i o n of imagery i n c o g n i t i o n . The ab s t r a c t / c o n c r e t e d i s t i n c t i o n d e f i n e s the r e l a t i v e a v a i l a -b i l i t y of the two coding systems as a f u n c t i o n of stimulus c h a r a c t e r i s t i c s . S p e c i f i c a l l y , w i t h regard to v e r b a l m a t e r i a l s , two-process theory i m p l i e s that the v e r b a l system i s e q u a l l y a v a i l a b l e to both concrete and a b s t r a c t i n f o r m a t i o n , w h i l e the a v a i l a b i l i t y of the imaginal system increases w i t h the concreteness of the m a t e r i a l . The second d i s t i n c t i o n between the imaginal and v e r b a l systems, that of s p a t i a l l y - p a r a l l e l versus s e q u e n t i a l p r o c e s s i n g , d e f i n e s the nature of 3 the o r g a n i z a t i o n a l processes which are assumed to operate i n the two systems. In p a r t i c u l a r , the imaginal system i s assumed to organize m a t e r i a l s p a t i a l l y such that s e v e r a l b i t s of i n f o r m a t i o n can be processed simultaneously. The s e q u e n t i a l processing c h a r a c t e r i s t i c of the v e r b a l system, on the other hand, i s assumed to produce o r g a n i z a t i o n of a temporal nature which allows only successive processing of i n f o r m a t i o n . As o u t l i n e d here, P a i v i o ' s two-process theory provides the f o l l o w i n g general i m p l i c a t i o n f o r the r o l e of imagery i n processing i n f o r m a t i o n . Concrete m a t e r i a l w i l l be processed i n p a r t by the imaginal system and t h e r e f o r e be subject to p a r a l l e l o r g a n i z a t i o n . A b s t r a c t m a t e r i a l , on the other hand, i s handled almost t o t a l l y by the v e r b a l system and w i l l be processed s e q u e n t i a l l y . From t h i s t h e o r e t i c a l statement, t e s t a b l e pre-d i c t i o n s a r i s e about a wide v a r i e t y of m a t e r i a l s i n an e q u a l l y wide v a r i e t y of t a s k s . The s p e c i f i c i m p l i c a t i o n s of the theory f o r the processing of language m a t e r i a l s (phrases, sentences and paragraphs) w i l l now be considered. Two-Process Theory and Language Processing As w i t h a l l kinds of v e r b a l m a t e r i a l , P a i v i o assumes that concrete language can be processed by the imaginal code, w h i l e a b s t r a c t language only has access to the v e r b a l code. P a r a l l e l processing w i l l t h e r e f o r e occur w i t h concrete language. This s p a t i a l o r g a n i z a t i o n may r e s u l t i n the simultaneous r e p r e s e n t a t i o n of the elements of a concrete phrase, sentence, or paragraph i n terms of a complex image. Thus the meaning of an e n t i r e concrete sentence, f o r example, may be summarized by a s i n g l e organized u n i t . A b s t r a c t m a t e r i a l , on the other hand, w i l l undergo s e q u e n t i a l 4 processing. This presumably results in l i t t l e unitization and integration, and so the meaning of abstract language is assumed to remain closely tied to the word-sequence i t s e l f . The main implication of two-process theory is that concrete language gives rise to an image which involves more integration and unitization of meaning than does the verbal-sequential representation evoked by abstract language. It should be noted that these processing differences do not necessarily predict that concrete language w i l l be recalled better than abstract language. According to Paivio, this i s because the complex image representation evoked by a concrete sentence, for example, may be subject to more decoding errors than the verbal-sequential representation of an abstract sentence. The emphasis then is on processing differences rather than simply recall performance differences. Evidence on Two-Process: Theory and Language Processing The evidence reviewed here is concerned with testing the implications of Paivio's two-process theory with regard to the processing of language. Although relatively few studies have investigated this issue, those that have been done cover a wide range of materials and tasks and therefore provide at least a basis for further investigation of the role of imagery in language. Imagery and Thematic Organization Studies by Lachman and his associates (Pompi and Lachman, 1967; Lachman and Dooling, 1968; Dooling and Lachman, 1971) have been concerned 5 with the nature of organizational processes involved i n the retention of language. These studies have implicated the "theme" of a passage as an important factor i n the retention of connected discourse. They found that the order i n which the words of a paragraph are presented has a large effect on the free r e c a l l of those words. In p a r t i c u l a r , i f the words are presented one at a time i n their natural syntactic order, r e c a l l i s vastly improved compared to presentation of those same words i n random order (Pompi and Lachman, 1967). Lachman and his co-workers have concluded that the theme of the message i s evoked by syntactic presentation and i t provides an organizational framework for the storage and r e t r i v a l of the words of the passage. More s p e c i f i c a l l y , syntactic by not random presen-tation i s said to give r i s e to "surrogate processes" which organize the material around i t s theme. The unified representation of the theme i s then stored and on r e c a l l i t i s evoked and used i n the reconstruction of the words of the message. The surrogate processes proposed by Lachman are d i r e c t l y analogous to the p a r a l l e l processing c h a r a c t e r i s t i c of the image code i n Paivio's theory. Because th i s type of organization i s assumed to be limited to highly concrete material, two-process theory predicts that the order effects observed i n Lachman's studies should not occur with abstract and connected discourse. Y u i l l e and Paivio (1969) tested this prediction by varying both order of presentation of words from paragraphs and the concreteness of the content. Concreteness was determined by judges who categorized the material as concrete or abstract. The words were presented one at a time at a 6 constant rate and performance was measured by free r e c a l l over two t r i a l s . Y u i l l e and Paivio suggested that presenting the words i n syntactic order would improve r e c a l l r e l a t i v e to random presentation only for the concrete material. Abstract material would not be affected by order of presentation i f i t i s t y p i c a l l y processed sequentially with l i t t l e integration and organization of meaning. The free r e c a l l r esults were exactly as predicted. Syntactic order increased r e c a l l r e l a t i v e to random order for concrete material, and there was no effect of presentation order for abstract material. Recall under the random presentation did not d i f f e r as a function of concreteness. The authors concluded that imagery plays an important part i n the storage of the theme of a passage. S p e c i f i c a l l y , they proposed that thematic ( i . e . , syntactic) presentation of the words of a concrete passage leads to "integrated meaningful imagery" that contains the essential ideas of the passage and can mediate the r e t r i e v a l of the o r i g i n a l verbal information. Abstract material, on the other hand, gives r i s e to a verbal-sequential representation which w i l l not be unified and integrated around a theme and therefore r e c a l l i s not affected by order of presentation. Y u i l l e and Paivio also did an analysis of the kinds of intrusion errors made i n r e c a l l i n g material. Judges c l a s s i f i e d these errors as either theme-related or not theme-related. The proportion of theme-related errors to t o t a l intrusion errors was then computed i n each condition. The only s i g n i f i c a n t effect here was that there were more theme-related errors made with syntactic order than with the random presentation order. This effect was confined to the f i r s t r e c a l l t r i a l and there was no difference i n the 7 proportion of theme-related errors as a function of concreteness. In-spection of the means i n each group reveals that on the f i r s t free r e c a l l t r i a l the mean proportion of theme-related errors i n the Abstract-Syntactic Order condition was i n fact s l i g h t l y greater than for the Concrete-Syntactic Order condition: .55 and .33, respectively. The corresponding values i n the Random Order conditions were .18 and .16. Therefore, i f the proportion of theme-related errors can be taken as an index of the degree to which an integrated representation of the theme i s retained, then there i s no evidence here that concrete and abstract materials d i f f e r s i g n i f i c a n t l y i n t h i s respect. This of course argues against the conclusion which was made on the basis of the free r e c a l l r e s u l t s . I f concrete material evokes a theme which i s stored i n an integrated fashion and abstract passages do not, then surely there should have been a higher proprotion of theme-re-lated errors i n the Concrete-Syntactic Order condition, which was not found. Indeed the difference was i n the opposite d i r e c t i o n . The analysis presented here suggests that the conclusion drawn by Y u i l l e and Paivio may have to be modifed somewhat. Recently, Philipchalk (1972) replicated the Y u i l l e and Paivio (1969) free r e c a l l r e s u l t s and found that they continued to hold after a two week delay i n testing. Philipchalk does not report any s i g n i f i c a n t effects due to concreteness or presentation order i n an analysis of theme-related intrusion errors. Indeed, he does not report the results of t h i s analysis at a l l , and therefore i t i s not possible to see whether there was any s i g n i f i c a n t departure from the free r e c a l l results as was found i n the Y u i l l e and Paivio study. 8 Thus the significant findings concerning the relationship between imagery and thematic organization are as follows. The words from concrete paragraphs are recalled significantly better when presented in natural, thematic order than when order of presentation is random. The re c a l l of abstract material i s not affected by presentation order. Furthermore, under thematic presentation, r e c a l l increases with concreteness. The results of these studies do not, however, provide unequivocal support for Paivio's two-process theory which implies that only concrete material gives rise to an integrated representation of meaning. The effect of concreteness on re c a l l of thematically-presented passages might be explained by an increased probability of meaning-related decoding errors as the material becomes more abstract. That i s , abstract paragraphs may be rep-resented in memory in terms of integrated meaning, but there may be greater d i f f i c u l t y than with concrete material in transforming this represen-tation into the exact words which were presented. This interpretation gains some support from the finding in Yuille and Paivio (1969) that there was a slightly higher proportion of theme-related errors in the Abstract-Syntactic Order condition than in the corresponding Concrete condition. In any event., there was no concreteness effect found in the error analysis which argues agains the conclusion that different degrees of thematic organization occur for concrete and abstract material. The discussion now turns to some studies which have investigated more directly the issue of differential integrative organization i n absract and concrete language. 9 Imagery as an Integrating Process Begg (1972) sought to provide d i r e c t evidence that concrete and abstract language materials are subject to different organizational pro-cesses. He argued, b a s i c a l l y from Paivio's two-process theory, that concrete phrases can be represented i n mamory i n terms of a single, highly i n -tegrated image. Abstract phrases, on the other hand, are assumed to be processed sequentially i n the verbal system and represented i n memory as verbal strings. Applying these implications to meaningful adjective-noun phrases, Begg made the following prediction. I f a two-word, concrete phrase i s stored as a single image, then there should be no increase i n memory load compared to the processing of a single concrete word. Abstract phrases, a l t e r n a t i v e l y , are assumed to be represented as a two-word verbal s t r i n g and should require, roughly, twice as much processing as a single abstract word. Begg tested t h i s prediction by comparing r e c a l l of adjective-noun phrases with r e c a l l of single-words l i s t s , as a function of concreteness of materials. He predicted that for concrete materials, r e c a l l of nouns and adjectives from the phrase l i s t s , would not d i f f e r from r e c a l l of these words presented i n single word l i s t s , even though presentation time was equal for the two l i s t s . For abstract materials, the phrase l i s t s should require twice as much processing as the single-word l i s t s , and with equal presentation time, should result i n s i g n i f i c a n t l y poorer r e c a l l (about half) than the single-word l i s t s . The results were as predicted. For abstract materials, the percentage 10 of a d j e c t i v e s and nouns r e c a l l e d was about twice as high when presented i n single-word l i s t s as when presented i n phrases. For concrete m a t e r i a l s there was no d i f f e r e n c e i n percent r e c a l l i n the two l i s t c o n d i t i o n s . I t appears then that only f o r concrete m a t e r i a l s can two-word phrases be processed as e f f e c t i v e l y as s i n g l e words, given the same amount of time. Begg a l s o t e s t e d another p r e d i c t i o n a r i s i n g from h i s assumptions. He reasoned that i f a concrete phrase can be represented i n memory as an i n -tegrated u n i t , then r e t r i e v a l of one of the words from the phrase should arouse the appropriate image, and lead to r e c a l l of the other word. A b s t r a c t phrases, i f they are represented i n a s e q u e n t i a l f a s h i o n , should not b e n e f i t as much from r e t r i e v a l of one component of the phrase. Begg argued that i n order f o r a cue to be f a c i l i t a t i v e , the cue and the to-be-remembered item must be represented i n memory as par t of the same i n t e g r a t e d u n i t . According to two-process theory, t h i s property i s c h a r a c t e r i s t i c of the imaginal r e p r e s e n t a t i o n of concrte m a t e r i a l but not the v e r b a l - s e q u e n t i a l r e p r e s e n t a t i o n of a b s t r a c t m a t e r i a l . To t e s t t h i s p r e d i c t i o n , Begg presented a b s t r a c t and concrete phrases, and t e s t e d r e c a l l under e i t h e r cued or non-cued c o n d i t i o n . Cued r e c a l l i n v o l v e d presenting one of the words of each phrase and r e q u i r i n g r e c a l l of the other word. Non-cued r e c a l l was simply a standard f r e e r e c a l l task. The r e s u l t s were completely i n l i n e w i t h h i s p r e d i c t i o n s . , Cueing s i g n i f i -c a n t l y increased r e c a l l of concrete m a t e r i a l s but had no e f f e c t on r e c a l l of a b s t r a c t m a t e r i a l s . On the b a s i s of t h i s evidence, Begg concluded that concrete and a b s t r a c t phrases d i f f e r i n terms of the o r g a n i z a t i o n a l processes that are 11 a v a i l a b l e t o t h e m . A c o n c r e t e p h r a s e a r o u s e s a n i n t e g r a t e d i m a g e w h i c h r e p r e s e n t s t h e c o m p o n e n t s o f t h e p h r a s e i n t e r m s o f a s i n g l e u n i t . A b s t r a c t p h r a s e s , o n t h e o t h e r h a n d , a r e p r o c e s s e d s e q u e n t i a l l y a n d a r e r e p r e s e n t e d a s n o n - i n t e g r a t e d v e r b a l s t r i n g s . B e g g a n d R o b e r t s o n ( 1 9 7 3 ) t e s t e d t h e r e l a t i o n s h i p b e t w e e n c u e i n g a n d c o n c r e t e n e s s i n a s l i g h t l y d i f f e r e n t s i t u a t i o n . H e r e t h e m a t e r i a l s w e r e n o u n p a i r s p r o d u c e d b y r a n d o m p a i r i n g s o f a b s t r a c t o r c o n c r e t e n o u n s , r a t h e r t h a n m e a n i n g f u l a d j e c t i v e - n o u n p h r a s e s . I n a d d i t i o n , m e d i a t i o n i n -s t r u c t i o n s w e r e g i v e n ( i m a g i n a l o r v e r b a l ) a l o n g w i t h f i v e s u b j e c t - p a c e d l e a r n i n g t r i a l s ( a s o p p o s e d t o o n e , e x p e r i m e n t e r - p a c e d t r i a l i n t h e B e g g , 1972 s t u d y ) . A l s o , r e c a l l p e r f o r m a n c e w a s t e s t e d b o t h i m m e d i a t e l y f o l l o w i n g t h e f i v e p r e s e n t a t i o n t r i a l s a n d a g a i n s i x d a y s l a t e r . A s i n t h e e a r l i e r B e g g s t u d y , c u e i n g h a d a g r e a t e r e f f e c t o n c o n c r e t e t h a n o n a b s t r a c t m a t e r i a l s . T h i s i n t e r a c t i o n w a s s i g n i f i c a n t a t b o t h r e t e n t i o n i n t e r v a l s , b u t was m o r e p r o n o u n c e d a f t e r s i x d a y s . U n l i k e t h e e a r l i e r s t u d y , h o w e v e r , c u e i n g d i d s i g n i f i c a n t l y i n c r e a s e r e c a l l o f t h e a b s t r a c t p a i r s o n t h e i m m e d i a t e r e c a l l t e s t . T h i s r e s u l t w o u l d s u g g e s t a s l i g h t m o d i f i c a t i o n o f B e g g ' s o r i g i n a l c o n c l u s i o n . I f o n e a c c e p t s h i s a s s u m p t i o n t h a t c u e i n g i s o n l y f a c i l i t a t i v e i f t h e t o - b e - r e m e m b e r e d i t e m a n d t h e c u e a r e r e p r e s e n t e d i n memory a s a n i n t e g r a t e d u n i t , t h e n o n e w o u l d h a v e t o c o n c l u d e t h a t t h e a b s t r a c t n o u n p a i r s i n t h e a b o v e s t u d y w e r e s t o r e d a s a s i n g l e r e p r e s e n t a t i o n . I t i s n o t c l e a r why t h i s s h o u l d b e t h e c a s e f o r a b s t r a c t p a i r s s e l e c t e d r a n -d o m l y a n d n o t t h e c a s e f o r m e a n i n g f u l a d j e c t i v e - n o u n p h r a s e s , a n d i n d e e d B e g g d o e s n o t d i s c u s s t h i s d i s c r e p a n c y . 12 The most reasonable conclusion from Begg's work i s that concrete material undergoes a greater degree of integration than abstract material. Contrary to the strong form of Paivio's two-process theory, however, ab-stract materials do appear to undergo a s i g n i f i c a n t degree of integrative processing under certain conditions. Franks: and Bransford (1972) have provided additional evidence on this point. They had e a r l i e r (Bransford and Franks, 1971) reported evidence suggesting that p a r t i a l information expressed i n separate concrete sentences tends to be integrated into complete ideas. B a s i c a l l y , their paradigm involves presentation of a series of sentences each expressing part of a complex idea. Sentences r e l a t i n g to several different ideas are pre-sented i n the same l i s t with no two items r e l a t i n g to the same idea being presented contiguously. They found (Bransford and Franks, 1971) that on a subsequent recognition task, the degree of confidence expressed i n per-formance was an increasing li n e a r function of the number of ideas contained i n the test item. Test items contained one, two, three or a l l four of the ideas r e l a t i n g to a complete idea, and highest confidence was expressed i n the l a t t e r which had never been presented as one sentence. Thus there was more confidence i n errors than i n correct recognitions. As the o r i g i n a l result was found with concrete material, Franks and Bransford (1972) decided to re p l i c a t e the procedure to test the implication of two-process theory that integration of meaning does not occur i n abstract language. As the findings using abstract sentences were v i r t u a l l y i d e n t i c a l to those produced by concrete materials, the authors concluded that no support was shown for Paivio's theory. S p e c i f i c a l l y , the process 13 of i n t e g r a t i o n of p a r t i a l meanings into h o l i s t i c representations had been demonstrated for both concrete and abstract sentences. Subsequent i n v e s t i g a t i o n s of the Bransford and Franks paradigm (Katz, 1973; Singer and Rosenberg, 1973) have suggested that the l i n e a r r e l a t i o n s h i p between the number of ideas expressed i n test sentences and recognition confidence ratings may be an a r t i f a c t of either the p a r t i c u l a r materials used to test recognition or the recognition i n s t r u c t i o n s . Singer and Rosenberg (1973) found that the l i n e a r r e l a t i o n s h i p was best accounted for by the semantic importance of the t e s t item to the complete idea rather than simply the number of ideas contained i n the item. Katz (1973) found that the increase i n confidence with the number of ideas expressed i n the test items was s u b s t a n t i a l l y reduced when i n s t r u c t i o n s emphasized recog-n i t i o n of meaning rather than exact wording of the presentation materials. The r e s u l t s of both studies, however, did support the general conclusion of the Bransford and Franks (1971) study, that p a r t i a l meanings are i n -tegrated into u n i f i e d representations. The evidence r e l a t i n g to imagery as an i n t e g r a t i n g process i s con-f l i c t i n g . The findings of Begg (1972) strongly suggest that i n t e g r a t i o n of verbal material into semantic units i s r e s t r i c t e d to concrete material. The Begg and Robertston (1973) and Franks and Bransford (1972) studies, however, demonstrate i n t e g r a t i o n of meaning i n abstract material. C l e a r l y , further work i s needed i n t h i s area i n order to account for the discrepancy. The next section reviews the evidence regarding memory for meaning and wording i n abstract and concrete sentences. This issue and i t s r e -l a t i o n to Paivio's theory has received much recent a t t e n t i o n . 14 Imagery and Memory for Meaning and Wording Sach (1967a; 1967b) carried out a series of studies designed to investigate whether the meaning of sentences or their precise wording was better retained. She presented paragraphs and tested recognition for various kinds of changes in the presented material. The changes introduced into test sentences were either semantic or syntactic in nature. Semantic changes significantly altered the meaning of the sentences, but l e f t the wording basically unchanged (e.g., interchanging the subject and object, negation, etc.). Syntactic changes, alternatively, changed the wording but l e f t the meaning intact (e.g., active to passive, substitutions of synonyms, etc.). Sachs' results showed that memory for the precise wording of sentences, indicated by the correct identification of a syntactic change, faded rapidly. Indeed with a large amount of interpolated material between presentation and test, recognition of syntactic changes was l i t t l e better than chance. Memory for semantic features of sentences, on the other hand, remained realtively high across a l l levels of interpolated material. Sachs concluded that a sentence is maintained in i t s original sequential form for only a short time, during which i t i s organized and coded for meaning. Following this process, the wording is lost from memory and the meaning i s stored and used to reconstruct the words of the sentence. These results are precisely i n line with the prediction from two-process theory with regard to concrete sentences. A concrete sentence can undergo parallel processing which leads to a unified representation of the meaning of the sentence in terms of an image. Abstract sentences, however, are 15 assumed to be processed i n a sequential manner c l o s e l y corresponding to the word sequence i t s e l f . Paivio's theory then, predicts that Sachs' r e s u l t s should hold only f o r concrete materials. With abstract sentences, memory for wording should be better than memory for meaning. Begg and Paivio (1969) c a r r i e d out a study to test these p r e d i c t i o n s . They noted that Sachs' materials were r e l a t i v e l y concrete and therefore that her r e s u l t s were i n accord with two-process theory. To tes t the pre-d i c t i o n about abstract sentences they r e p l i c a t e d the Sachs procedure using materials which varied i n concreteness. Semantic changes i n the test sentences involved reversing the subject and object, while s y n t a c t i c or l e x i c a l changes, as they were c a l l e d , were produced by s u b s t i t u t i n g synonyms for e i t h e r the subject or the object i n the o r i g i n a l sentence. One important modification of the Sachs procedure was that the "paragraphs" used as study materials consisted of unrelated sentences a l l having the same grammatical form. The p r e d i c t i o n of two-process theory i s that there should be an i n t e r -a c t i o n between the type of change and concreteness i n recogntion performance. S p e c i f i c a l l y , f o r concrete materials semantic changes should be detected at a higher rate than l e x i c a l changes, while the reverse should be true for abstract sentences. The analysis of recognition of performance revealed support f o r these p r e d i c t i o n s . Semantic changes were recognized better than l e x i c a l changes i n concrete sentences and the r e s u l t was reversed i n abstract sentences. Furthermore, semantic changes i n concrete sentences were detected better than s i m i l a r changes i n abstract sentences. The conclusion was that concrete 16 sentences are processed by the Imaginal code which leads to the represen-tation of the essential meaning of the sentence in a single organized unit. Changes in meaning alter the nature of this representation and are recog-nized more easily than lexical changes which leave the representation intact. Abstract sentences, however, are processed in a verbal-sequential manner which gives rise to a representation which is closely tied to the original wording. Changes in wording produced by synonym substitutions presumably alter this presentation more than semantic changes involving subect-object, reversals, and are therefore recognized more easily. Johnson, Bransford, Nyberg, and Cleary (1972) replicated the major finding of the Begg and Paivio (1969) study that changes in meaning are detected more readily in concrete than in abstract sentences. They sug-gested, however, that these results may have been due, in part, to d i f -ferences in comprehensibility between the abstract and concrete sentences. Comprehensibility ratings on a scale from 'very hard' to 'very easy' to understand revealed that significantly higher ratings were given to the concrete sentences.used by Begg and Paivio, than to their abstract sentences. Furthermore, they found that the semantic changes (reversing the subject and object) in the concrete sentences resulted in a greater change in meaning than these changes in abstract sentences. The authors concluded that either one of these factors could have produced the greater recognition of semnatic changes in concrete as compared to abstract sentences found by Begg and Paivio. A recent study by Moeser (1974) sought to test this interpretation of the Begg and Paivio results. In three experiments (I, II, and III), 17 Moeser used a subset of the Begg and P a i v i o (1969) sentences which had been r a t e d equal i n c o m p r e h e n s i b i l i t y . A l s o , semantic changes i n these experiments were produced by s u b s t i t u t i n g one, two, or three words i n the sentences which a l t e r e d the meaning. L e x i c a l changes i n v o l v e d s u b s t i t u t i o n of words which l e f t meaning i n t a c t . Moeser reasoned that i f the f a c t o r s mentioned by Johnson et a l . (1972)—:comprehensibility d i f f e r e n c e s and d i f f e r e n c e s i n the degree of meaning change produced by s u b j e c t - o b j e c t r e v e r s a l s — h a d produced the Begg and P a i v i o r e s u l t s , then there should be no d i f f e r e n c e s between concrete and a b s t r a c t sentences i n d e t e c t i o n of meaning changes i n her experiment. The r e s u l t s f a i l e d to support t h i s hypothesis. Recognition of meaning changes was greater i n concrete sentences than i n a b s t r a c t sentences. In a d d i t i o n , d e t e c t i o n of l e x i c a l changes was su p e r i o r i n the concrete m a t e r i a l . There wereno other e f f e c t s revealed i n the a n a l y s i s of r e c o g n i t i o n per-formance. Thus the Begg and P a i v i o (1969) r e s u l t w i t h respect to d e t e c t i o n of meaning changes was confirmed when ra t e d c o m p r e h e n s i b i l i t y was equal f o r concrete and a b s t r a c t sentneces and when s u b j e c t - o b j e c t r e v e r s a l s were not used to t e s t f o r r e t e n t i o n of meaning. However, u n l i k e the Begg and P a i v i o r e s u l t , there was no i n t e r a c t i o n between the type of change and concreteness. Both meaning changes and wording changes were recognized b e t t e r f o r the more h i g h l y concrete m a t e r i a l and w i t h i n each l e v e l of concreteness, memory f o r meaning was not s i g n i f i c a n t l y d i f f e r e n t from memory f o r wording. In an e f f o r t to account f o r t h i s discrepancy, Moeser c a r r i e d out another experiment ( I V ) . In t h i s study, the procedures used by Begg and 18 Paivio were followed more c l o s e l y . L e x i c a l changes were produced by s u b s t i t u t i n g only one synonym i n the o r i g i n a l sentences and subject-object reversals were used to produce semantic changes. Abstract and concrete sentences were equated i n the degree to which these changes a l t e r e d meaning. The r e s u l t s r e p l i c a t e d those found by Begg and Paivio exactly. For meaning changes, concrete ;sentences: produced better recognition performance than abstracts and the reverse was true for l e x i c a l changes. The r e s u l t suggested that the Begg and Paivio f i n d i n g i s s p e c i f i c to the type of test sentence used, and therefore that the generality of i t s conclusions may be l i m i t e d . In p a r t i c u l a r , memory for meaning appears to increase with concreteness regardless of the type of test sentence used, but for detection of wording changes the r e s u l t appears to depend on how many words are substituted i n the t e s t items. With only one word sub-s t i t u t e d , l e x i c a l changes i n abstract sentences are detected better than i n concrete sentences and the reverse i s true when test items involve sub-s t i t u t i o n of more than one word. F i n a l l y , there i s no diffe r e n c e between detection of meaning changes and wording changes for either concrete or abstract material when both of these changes are produced by s u b s t i t u t i n g one, two, or three words i n test sentences. Two f i n a l experiments involved only l e x i c a l changes produced by sub-s t i t u t i n g one synonym. In one (Vb), Moeser found that recognition p e r f o r -mance was better with concrete than with abstract sentences. In t h i s study, unlimited study time was given for each of the presented sentences. Experiment Va, which had a constant f i v e second presentation rate, revealed no e f f e c t of concreteness on recognition performance. 19 Moeser's c o n c l u s i o n was, f i r s t of a l l , t hat n e i t h e r c o m p r e h e n s i b i l i t y d i f f e r e n c e s nor d i f f e r e n c e s i n the degree of meaning change produced by subj e c t - o b j e c t r e v e r s a l s could account f o r the concreteness e f f e c t s ob-served by Begg and P a i v i o . Her reasoning was that i f these f a c t o r s were in v o l v e d there should have been no e f f e c t of concreteness i n d e t e c t i o n of meaning changes i n experiments I , I I , or I I I of her study. This con-c l u s i o n i s not r e a l l y warranted, however. While i t i s true that comprehen-s i b i l i t y r a t i n g d i f f e r e n c e s could not have produced Moeser's r e s u l t s i n experiment I to I I I , they could s t i l l have been a s i g n i f i c a n t f a c t o r i n the Begg and P a i v i o study, as performance measures were not based on e x a c t l y the same sets of sentences i n the two s t u d i e s . D i f f e r e n c e s i n ra t e d c o m p r e h e n s i b i l i t y could a l s o be a f a c t o r i n the r e s u l t s of Moeser's e x p e r i -ment IV which r e p l i c a t e d the Begg and P a i v i o f i n d i n g , because t h i s f a c t o r was not equated i n a b s t r a c t and concrete sentences used here. S i m i l a r l y , thedegree of meaning change f a c t o r remains a p o s s i b l e c o n t r i b u t o r to the Begg and P a i v i o r e s u l t , as semantic changes i n Moeser's study were not produced i n e i t h e r the same sentences nor i n the same way. I t could a l s o be a f a c t o r i n experiments I to I I I . Subject-object r e v e r -s a l s were not used, but meaning could have been changed more i n concrete sentences than i n a b s t r a c t s by the word s u b s t i t u t i o n s that were made. No measures of the degree of meaning change i n t e s t sentences used i n these three experiments are reported to discount t h i s p o s s i b i l i t y . Only i n experiment IV can t h i s f a c t o r be dismissed as a p o s s i b l e c o n t r i b u t o r to the concreteness e f f e c t i n memory f o r meaning. Moeser's r e s u l t s , then, do not d i s c o n f i r m the Johnson et a l . (1972) 20 suggestions regarding the Begg and Paivio study. Indeed they cannot be directly disconfirmed, as rated comprehensibility and degree of meaning change in test items are confounded with concreteness in that study. Her findings do support the conclusion that memory for meaning increases with concreteness when rated comprehensibility is controlled (as in experiments I to III) or when the degree of meaning change in test sentneces i s equated (as in experiment IV). However, in none of the studies were the two factors controlled simultaneously and therefore one of them could have contributed to the result in each case. Moeser's other main conclusion is that some of her results cannot be handled by Paivio's two-process theory. Specifically, she found in ex-periments I, II, and III and also in Vb that memory for wording increased with concreteness. Paivio's theory assumes that concrete and .abstract material have equal access to the verbal code, but that specific wording may be lost in the imaginal representation of concrete language. This argument was used by Begg and Paivio to explain the superiority in abstract sentences in detection of lexical changes that they observed. Moeser offers an explanation of the Begg and Paivio result, which she replicated in experiment IV, in terms of response bias. The suggestion here i s that when there is uncertainty with regard to the type of change in a given test sentence, there w i l l be a tendency to identify i t as a wording change i f i t is abstract and as a meaning change i f i t i s concrete. This explanation has no foundation whatsoever as an explanation of either the Begg and Paivio result or the result of experiment IV where the finding was replicated. In both studies, the test consisted of marking either "changed" or "identical" 21 for each item. Even i f response bias operated in the way that Moeser claims, i t should be obvious that, under these test conditions, the factor could not produce a concreteness effect in detection of wording changes. The response bias explanation is based on the premise that the wording change is noted equally well in concrete and abstract test items, but that i t is more lik e l y to be judged as a wording change in abstracts. No such judgement was required in either study. The factor could only operate when an identification of the type of change must be made. This was the case in experiments I to III, where the reverse of the prediction arising from the response bias explanation was found: wording changes were identi-fied better in concrete sentences. In any event, Moeser reports that there was no support for the response bias explanation found in experiment Va. The question of whether or not Paivio's theory can account for the superiority of concrete sentences in detection of wording changes can now be discussed. In experiment I, II, and III the concreteness effect could have been due to a greater change in meaning produced by word substitutions in concrete test items. No ratings were obtained to control for the degree of meaning change in the test sentences used in these studies. Judges simply agreed that semantic test items changed meaning and lexical changes did not. An examination of the examples provided by Moeser lends support to this possibility: Concrete: Original - The f a l l i n g rock k i l l e d a sinful captive. Lexical - The f a l l i n g rock crushed a sinful captive. Abstract: Original The careful study resolved an open question. Lexical - The careful report resolved an open question. 22 Surely " k i l l e d " and "crushed" do not have the same meaning. I f these examples are r e p r e s e n t a t i v e of the r e s t of the t e s t items used i n the study, then there would appear to be a greater change i n meaning produced by l e x i c a l t e s t items f o r concrete m a t e r i a l than f o r a b s t r a c t m a t e r i a l . Otherwise, the r e s u l t would suggest that concrete sentences have greater access to the v e r b a l code than a b s t r a c t sentences. This i s the con c l u s i o n reached by Moeser and i t i s co n t r a r y to P a i v i o ' s theory. Why the opposite was revealed i n the Begg and P a i v i o study i s not at a l l c l e a r . Experiment Vb a l s o showed that d e t e c t i o n of wording changes increased w i t h concreteness, and i n t h i s study concrete and a b s t r a c t sentences were equated on the degree to which l e x i c a l t e s t items changed meaning. The major procedural, d i f f e r e n c e between t h i s experiment and the Begg and P a i v i o study was i n the amount of study time given f o r each sentence. Begg and P a i v i o presented sentences at a constant f i v e second r a t e w h i l e i n experiment Vb, u n l i m i t e d study time was allowed. I f concrete sentences have access to both the imaginal and the v e r b a l code and a b s t r a c t sentences are only coded v e r b a l l y , then i n c r e a s i n g p r e s e n t a t i o n time should increase memory f o r wording i n a b s t r a c t m a t e r i a l at l e a s t as much as i n concrete m a t e r i a l , according to P a i v i o ' s theory. I t could be that P a i v i o ' s theory must be modified when there i s no r e s t r i c t i o n on processing time. In par-t i c u l a r , i t could be that concrete m a t e i r a l undergoes l i t t l e v e r b a l processing only when study time i s l i m i t e d . When there i s no such r e s t r i c t i o n , the p r o b a b i l i t y of l e x i c a l coding may increase s i g n i f i c a n t l y , l e a d i n g to the r e s u l t found by Moeser. To conclude t h i s d i s c u s s i o n of Moeser's study, one f u r t h e r r e s u l t 23 should be mentioned. Although concrete and abstract sentences were equated on rated comprehensibility in experiments I to III, an analysis of en-coding time in experiment I and II showed that concrete sentences were encoded faster than abstract sentences. Encoding time was measured from the i n i t i a l presentation of the sentence to when the subject indicated that he had studied the item long enough to remember i t and pressed a button in order to change the slide. The finding here suggests some dis-crepancy between the encoding time and rated comprehensibility as measures of how d i f f i c u l t sentences are to understand. This point w i l l be returned to later. Two other studies of the relationship between imagery and memory for meaning and wording have failed to c l a r i f y this issue. Begg (1971) used a continuous recognition paradigm and found memory for meaning to be superior to, and independent of, memory for wording in concrete sentences. Abstract materials were not used in the study, however, and so the c r i t i c a l implications of two-process theory could not be tested. Goldfarb, Wirtz, and Anisfeld (1973) investigated false recognitions in concrete and ab-stract phrases. They found that more errors were made to test phrases synonymous with previously presented material, than to antonymous phrases, and this result did not change as a function of concreteness. The effect was larger for the concrete phrases, however, which the authors attributed to the relative indistinctiveness of the abstract phrases. This possibility was supported by a second study which showed a significant increase in the superiority of antonym detections to synonym detections with increasing rated 24 distinctiveness of abstract phrases. The authors concluded that both ab-stract and concrete phrases are coded for meaning rather than wording and that concreteness effects on this relationship can be explained by d i f -ferences in distinctiveness. The studies reviewed here do not provide clear-cut support for the implications of Paivio's theory with regard to memory for meaning and wording. The discussion of the Begg and Paivio (1969), Moeser (1974) and Goldfarb et a l . (1973) experiments suggests that in each case, factors other than concreteness could haveled to the obtained results. In the section which follows, an attempt i s made to arrive at a parsimonious explanation for these results, and those described earlier that relate to the effect of imagery in language processing. 25 SUMMARY AND IMPLICATIONS OF THE EVIDENCE The evidence reviewed in the last section revealed many conflicting findings with regard to Paivio's two-process theory and i t s implications for the relationship between imagery and language processing. The studies by Yuille and Paivio (1969), Begg (1972) , ..and Begg and Paivio (1969), provide at least partial support for the theory. They indicate that concrete language tends to be intergrated into semantic units to a greater degree than abstract language which appears to be processed in a manner more closely tied to a verbal sequential representation. The results of studies by Begg and Robertson (1973), Franks and Bransford, (1972), and Goldfarb, Wirtz, and Anisfeld (1973) indicate, on the other hand, that abstract language can be integrated into a semantic representation rather than being retained in terms of wording. In addition, the Yuille and Paivio (1969) analysis of intrusion errors lends further support to the notion that an organized representation of meaning results from the pro-cessing of abstract language material. Two of the studies reviewed suggested that factors other than con-creteness may have led to results of those studies supporting Paivio's theory. Johnson, Bransford, Nyberg, and Cleary (1972) found differences in rated comprehensibility and degree of meaning change in test items in the concrete and abstract materials used by Begg and Paivio. Goldfarb, Wirtz, and Anisfeld (1973) suggested that concrete and abstract language materials differ in distinctiveness which might account for the superiority of concrete materials in retention of meaning reported in some of the 26 above studies. Finally, Moeser (1974) showed that memory for meaning and wording was superior in concrete sentences. With regard to retention of meaning, one of the factors suggested by Johnson et a l . (1972) could have been involved as both were never controlled simulataneously. The concreteness effect on memory for wording was only clearly established when unlimited encoding time was given. In the other demonstrations of the result, the degree of meaning change produced by lexical changes in test sentneces was not con-trolled and could have contributed to the effect. In this section, an attempt w i l l be made to resolve some of this conflicting evidence. Specifically, the proposal w i l l be made that d i f -ferences in level of comprehension rather than mode of storage can best account for this wide variety of results. Imagery and Levels of Comprehension .in Language Processing A number of recent studies have investigated the relationship between concreteness and comprehensibility. Paivio and Begg (1971) presented the concrete and abstract sentences used in the Begg and Paivio (1969) study and had subjects press a button when they f e l t that they understood the sentence. An analysis of reaction times indicated that there was no sig-nificant difference between concrete and abstract sentences. The authors concluded that abstract sentences are no more d i f f i c u l t to understand than concrete sentences. Three other studies, however, have produced findings which argue against this conclusion. Johnson et a l . (1972), as mentioned earlier, 27 found that concrete sentences used by Begg and P a i v i o (1969) were r a t e d as s i g n i f i c a n t l y e a s i e r to understand than the a b s t r a c t sentences used there. Klee and Eysenck (1973) a l s o found evidence f o r d i f f e r e n c e s i n co m p r e h e n s i b i l i t y i n a b s t r a c t and concrete sentences. In t h i s study, subjects were r e q u i r e d to judge whether sentences were meaningful or anomolous and these judgements were made s i g n i f i c a n t l y f a s t e r f o r concrete sentences. F i n a l l y , Moeser (1974) found that even when rated c o m p r e h e n s i b i l i t y was equated f o r concrete and a b s t r a c t senteneces, there were d i f f e r e n c e s i n the speed of encoding as a f u n c t i o n of concreteness. S p e c i f i c a l l y , when p r e s e n t a t i o n time was c o n t r o l l e d by the s u b j e c t , l e s s time was spent studying concrete sentences than a b s t r a c t sentences. This suggests that concrete m a t e r i a l was f u l l y understood at a f a s t e r r a t e than a b s t r a c t m a t e r i a l . The r e s u l t s of these s t u d i e s suggest a very complex r e l a t i o n s h i p between comprehension and concreteness. The i s s u e may be s i m p l i f i e d somewhat, however, by n o t i n g the d i f f e r e n c e s i n the task demands of the s t u d i e s . Moeser (1974) suggested that some measures of c o m p r e h e n s i b i l i t y may i n d i c a t e the ease of understanding the i n d i v i d u a l words of the sentence w h i l e others r e v e a l the c o m p r e h e n s i b i l i t y of the whole sentnece as an i n t e g r a t e d u n i t . In p a r t i c u l a r , she proposed that the speed of encoding, because i t i s l i n k e d to a memory task, emphasized i n t e g r a t i o n to a greater extent than some of the other estimates of c o m p r e h e n s i b i l i t y . A reasonable r e s o l u t i o n of the r e s u l t s of these s t u d i e s , then, i s that they measured the c o m p r e h e n s i b i l i t y of d i f f e r e n t aspects of sentences. S p e c i f i c a l l y , the P a i v i o and Begg (1971) study may have measured only word comprehension. In the Johnson et a l . (1972) study, there may have been a 28 greater tendency to make judgements based on the whole sentence, but i t is not clear that ease of integration of the sentence elements i s stressed in the rating procedure. Finally, both the speed of encoding measure used by Moeser (1974) and the speed of categorization procedure used by Klee and Eysenck (1973) would seem to reflect the ease of understanding of the whole sentence as an integrated semantic unit. On this basis, i t would appear that concrete and abstract sentences do not differ in the ease of understanding their individual words, but that concrete sentences can be understood as an integrated semantic unit more easily than abstract sentences. This analysis i s consistent with the recent views of comprehension as a multi-level process. A primary exponent of this approach is Mistier-Lachman. In a recent study (Mistler-Lachman, 1972) she gave subjects three different comprehension tasks: (1) Meaningfulness Judgements - cate-gorizing sentences as meaningful or anomolous , (2) Context Integration -judging whether or not sentences follow from context (3) Production - making up a sentence which follows from the presented sentence. When sentences were ambiguous, there was an increase in the time i t took to complete the Context Integration and Production tasks, but no change in the speed with which the Meaningfulness Judgement could be carried out. Mistler-Lachman suggested that Meaninfulness Judgements could be completed without ambiguity resolution and therefore represented a "shallower" form of comprehension than the other two tasks. Using this terminology, understanding a sentence as an integrated semantic unit could be regarded as a deeper form of comprehension than 29 understanding the i n d i v i d u a l words. Thus concrete and abstract sentences would seem to d i f f e r i n comprehensibility only at t h i s deeper l e v e l . I t seems reasonable, then, that i n each of the studies which show concrete sentences to be superior to abstracts i n terms of r e t e n t i o n of integrated meaning (e.g., Y u i l l e and P a i v i o , 1969; Begg, 1972; Begg and P a i v i o , 1969; Moeser, 1974), the r e s u l t could be explained by d i f f e r e n c e s i n comprehensibility at the l e v e l of integrated semantic meaning. Compre-h e n s i b i l i t y i n t h i s sense was not c o n t r o l l e d i n any of these studies. Moeser (1974) c o n t r o l l e d for rated comprehensibility, but found differences i n speed of encoding which, as she suggests, i s probably a better estimate of comprehension at the l e v e l of being considered here. As discussed e a r l i e r , concrete and abstract sentences do not appear to d i f f e r i n terms of the ease of understanding i n d i v i d u a l words. This would suggest that there should be no d i f f e r e n c e i n memory for wording as a function of concreteness. In opposition to t h i s p r e d i c t i o n are the findings of Begg and Paivio (1969) where memory for wording decreased with concreteness, and the r e s u l t of experiments I, I I , I I I and Vb of Moeser's (1974) study which found the reverse. A p a r t i a l r e s o l u t i o n i s suggested here by emphasizing that the notion of depth of processing implies a continuum i n time rather than a dichotomy of d i f f e r e n t types of processing. In terms of the analysis which was pre-sented above, t h i s conceptualization suggests that l e x i c a l processing occurs equally quickly for concrete and abstract language but that deeper processing i n terms of integrated meaning occurs f a s t e r when the material i s more highly concrete. By s t r e s s i n g the a c c e s s i b i l i t y of the l e v e l s of 30 processing in this way, i t becomes clear memory for wording may vary as a function of conreteness within the conceptualization being offered here. Consider the Begg and Paivio result f i r s t . In this study a fixed (five seconds) amount of time was available for processing each sentence, and memory for wording decreased with concreteness. It could be argued that after five seconds, concrete material is already being processed at the deeper semantic level and that the results of the prior lexical analysis are no longer completely available. With abstract material, the entire five seconds may be spent on lexical analysis due to the relative inac-ce s s i b i l i t y of the deeper semantic level. This is essentially the arguement used by Begg and Paivio. In terms either of total time spent on lexical processing, or by the assumption that specific wording may be lost when meaning is being processed (Sachs, 1967a) the decrease in memory for wording with increasing concreteness is predicted. In experiment III of Moeser's study, the same presentation rate was used and the opposite result was found. Two points are important here. F i r s t , the superiority of concrete over abstract in memory for wording was small in comparison to the effect in memory for meaning, although the interaction was not significant. Also, as indicated by Moeser's examples (given in the last section), lexical changes in concrete sentences may have resulted in larger changes in meaning than the lexical changes in abstract sentences. Thus the greater detection of wording changes in concrete sentences could have been based on the increased amount of semantic processing that is assumed for the more concrete material. 31 The other demonstrations of the superiority of concrete material in memory for wording in Moeser's study, occurred when unlimited study time was given (experiments I, II, and Vb). The mean encoding time for concrete sentences was about 10 seconds and the corresponding value for abstract sentences was approximately 12 seconds. The analysis above suggests that after five seconds, concrete sentences are being coded semantically and that lexical features may be relatively unavailable., while abstract sen-tences are s t i l l being processed lexically. The extra five seconds of encoding time with concrete sentences may be used primarily for further lexical analysis, as meaning may already be represented in memory and therefore further semantic processing i s unnecessary. For abstract sentences however, the extra seven seconds of encoding time may be used in gaining access to the semantic level of processing, which could result in some of the lexical information being made unavailable. This analysis would predict that lexica l changes would be detected at a higher rate in concrete sentences in which lexical analysis may be ongoing for a greater total amount of time and has been more recently activated at the time of the test. Note that this analysis can s t i l l account for the greater memory for meaning in concrete material when unlimited encoding time i s given, as in Moeser's experiment I and II. Although the extra time taken to encode abstract sentences may allow processing to reach an equivalent depth of comprehension, the c r i t i c a l parameter in the conceptualization which has been presented i s the accessibility of the levels. That i s predictions about the quality of the memory representation for a particular level are based primarily on the ease or speed with which that level i s attained 32 rather than whether or not i t i s attained at a l l . Therefore, increasing presentation or encoding time w i l l not in general overcome accessibility differences. However, i f accessibility to a particular level i s equal, then increasing the time spent at that level may increase memory for the elements processed there. This was the assumption used to account for the concreteness effects in memory for lexical features (where accessibility of concrete and abstract is equal) discussed above. The persistence of the concreteness effect despite unlimited pro-cessing time could also be explained within this general framework by assuming that comprehensibility differences correspond to different depths of semantic processing, rather than simply the speed at which a particular depth i s attained. This explanation would seem to apply more to materials differning greatly in semantic content, such as random strings of words and meaningful sentences. When only meaningful language materials of the linguistic 'size' (i.e., only phrases; only sentences; only paragraphs) are being considered, the accessibility explanation would seem to be more valid. The conceptualization of comprehension that has been proposed can now be summarized with respect to i t s implications for memory of semantic and lexical features of language. Comprehension is viewed as a multi-level process with lexical analysis being shallower than semantic analysis. Language processing i s assumed to proceed in time through deeper and deeper levels, with the f i n a l level determined in general by the degree of semantic content. If semantic content (which i s assumed to increase from nonsense syllables to words, from words to phrases, phrases to sentences, and 33 sentences to paragraphs) i s approximately equal, then i t i s assumed that different levels w i l l be reached only when processing time i s limited, and in this case, the accessibility of the various levels to particular kinds of material becomes the key determinant of the f i n a l level of processing. Within a given type of material, the kinds of features which are rep-resented best in memory w i l l be those which are processed at the f i n a l level reached by that material (except when processing time is unlimited in which case memory for the features of lower levels may be increased). For materials which differ in their ease of access to a particular level, memory for the features of that level w i l l be greater for the material with the greater (i.e., faster) access. When the access of two kinds of material to a particular level of processing i s equal, memory for the features of that level w i l l be greater for the material which i s processed for the longest time at that level. The general framework for the model i s based on Mistler-Lachman's conceptualization of comprehension. Specific assumptions about the charac-teri s t i c s of the model derive from the relationships between concreteness, comprehensibility, and memory for semantic and lexical features of language which have been reviewed here. If the assumption is made that abstract and concrete language have equal access to lexical processing but that concrete language has greater access to integrative semantic processing, then the above model appears to provide a reasonable account of many of the studies which have been described. In particular, i t accounts for a l l of those studies which show a greater degree of semantic processing in concrete as compared to abstract language (Yuille and Paivio, 1969; ; 34 Begg, 1972; Begg and Paivio, 1969; Goldfarb, Wirtz and Anisfeld, 1973; Moeser, 1974). In addition, i t appears to provide a plausible explanation for the descrepancy between the results of the Begg and Paivio (1969) study and those found by Moeser (1974) with regard to concreteness effects on memory for wording. This account, thus far, bears much resemblence to Paivio's two-process theory. Both models assume that access to verbal-lexical coding does not vary as a function of concreteness, while concrete material has greater access to integrative-semantic coding. Indeed, the predictions arising from the two approaches are vir t u a l l y identical with respect to those studies cited in the last paragraph. There i s a major conceptual difference however. Paivio regards abstract and concrete language as being processed by two separate coding systems and although there i s some interplay be-tween the systems, concrete and abstract material are assumed to undergo qualitatively different kinds of processing and are therefore represented in memory in qualitatively different forms. This difference i s assumed to hold as long as there i s a difference in imagability and does not depend on comprehensibility per se. The Levels of Processing model, however, assumes that the degree of access to integrative semantic processing i s the c r i t i c a l factor involved in the relationship between concreteness and memory for language. Although typically concreteness and access to semantic processing covary, the possibility exists in this model for concrete and abstract language to undergo exactly the same kind of processing and therefore to be represented in memory in exactly the same form. This possibility exists, because 35 semantic and lexical processing are viewed as different levels along a continuum, rather than functional characteristics of separate coding systems as in Paivio's model. The two models, then, make identical predictions only when abstract and concrete language materials differ in ease of access to integrative semantic processing. Paivio's theory assumes that this i s always the case whereas the Levels of Processing model allows the possibility of abstract and concrete material being equal in this respect. If so, then the present model would predict that abstract and concrete language would be processed in the same way and represented in memory in the same way. This possibility i s dealt with in the next section. Context and the Relationship of Imagery to Language Processing The obvious way to test the differences between the two models would be to try to equate concrete and abstract language materials on some measure of access to integrative semantic processing. As suggested earlier, speed of encoding or speed of judgement as meaningful or anomolous might be appropriate measures. However, a l l of the available evidence suggests that typical examples of concrete and abstract sentences differ signficantly on these measures and indeed this was a basic assumption of the Levels of Processing model. The implication is that equating access to the semantic code on the basis of some measure of comprehensibility would result in the selection of sets of concrete and abstract sentences which are not 36 representative of the populations from which they were drawn. Thus i f a generalizable test i s to be made of the two models, some other method of trying to control the access of concrete and abstract sentences to inte-grative semantic processing must be found. One possibility i s suggested by the work of Bransford and his associates. As a result of a series of studies (e.g., Bransford and Franks, 19.71; Bransford and Johnson, 1972; Bransford, Barclay, and Franks, 1972; see Bransford and Johnson, 1973 for a review.), Bransford has proposed that the effective processing of language depends on "access to a substrate of additional information." (Bransford and Johnson, 1973, p. 434). The basic implication is that comprehension and retention of language depend more on pre-existing knowledge about the events described than on the specific characteristics of the linguistic input. That i s , a paragraph or sentence is never treated in isolation. Instead of the f i n a l representation of language in memory is a "semantic product" (Bransford and Johnson, 1973, p. 384) of input information and extra-linguistic information which can be characterized by the term knowledge. This substrate of additional information or knowledge is typically provided by the processor of language during comprehension. In this res-pect, Bransford, Barclay, and Franks, (1972) showed that inferences of linguistic input are retained along with the input i t s e l f . This result has been supported in a number of recent studies (Johnson, Bransford, and Soloman, 1973; Kintsch, 1972; Of f i r , 1973). In addition, however, the knowledge may be supplied directly by the input information presented prior to the processing of a particular 37 paragraph or sentence. For this kind of 'substrate of additional information' the term context may be used. The context for a particular language unit is typically provided by previous linguistic input which then provides the basis for comprehension. A study by Bransford and Johnson (1972) investigated the role of context ava i l a b i l i t y on comprehension and r e c a l l . They constructed para-graphs which, although made up of meaningful, related sentneces, had l i t t l e overall meaning in the absence of additional contextual information. In one experiment, this information was supplied by a picture which integrated the events described in the paragraph, and in a second study context was provided by the appropriate topic for the message. In both cases, re c a l l and rated comprehensibility of the paragraph were increased by the presence of the context prior to (but not after) presentation. It was concluded from the results of these studies that the picture and the topic acted in the same way as prior linguistic input acts in more typical language processing situations. That i s , they provided additional contextual information that adds to the information being processed and may increase the comprehension of the material and perhaps, therefore, enhance recall of the input. Context, as described above, may be a c r i t i c a l factor in the relation-ship of imagery to retention of linguistic meaning. This possibility arises out of a study by Bransford, Johnson, and Raye (1973) which varied concreteness with topic availability in materials similar to those used in the Bransford and Johnson (1972) study described above. The results showed that there was an interaction between these two variables, with the 38 reca l l of the main ideas of the concrete paragraph being unaffected by topic-availability while recall of the abstract material was significantly improved by providing the appropriate topic. Comprehensibility ratings by the same subjects who had recalled the material revealed precisely the same effect. The study indicates that concrete and abstract material bene-fited differentially from the presence of contextual information prior to input. Semantic processing of abstract material was increased by the addition of context. Comprehensibility and re c a l l of concrete paragraphs was unaffected by additional context with performance on both measures being superior to that for the abstract material. The Levels of Processing model presented earlier proposed that concrete and abstract material have equal access to lexical processing, while con-crete language typically has greater access to integrative semantic pro-cessing than abstract material. On this basis, the prediction may be made that memory for meaning in concrete language w i l l be superior, generally, to that in abstract language. In addition, however, the model suggests that i f concrete and abstract language are equated on the ease of access to integrative semantic processing, then memory for meaning may be shown to be equal regardless of variation in concreteness. The Bransford, Johnson, and Raye (1973) study showed that context increased memory for meaning in abstract material and had no effect, on concrete material, which was superior to abstract in rated comprehensibility and rec a l l under both context and no context conditions. The implication for the model, then, is that context can increase the ease of access of abstract language to integrative semantic processing and thereby reduce the superiority of 39 of concrete material in memory for meaning which would otherwise be predicted. This predicted change as a function of context is not implied by Paivio's two-process theory which contends that imagability i s the c r i t i c a l factor involved in semantic processing. The mechanism by which context may have this effect i s suggested by Bransford and Johnson (1972). They proposed that context acts primarily during comprehension of language. Specifically, i t i s thought to provide a semantic framework to aid in the organization of the material and enhance the ease with which i t can be understood as a semantic unit. The 'semantic product' of the input and the contextual framework then provides the basis for the representation which i s retained in memory. In the next section, the results of studies reviewed earlier are re-examined in light of the proposed effect of context on the relationship of imagery to semantic processing. Evidence for the Role of Context in the Relationship  of Imagery to Language Processing The studies reviewed earlier which investigated semantic processing as a function of imagery can now be re-examined i n light of the proposed effect of context on this relationship. Context may be defined for this purpose as the presence of semantically related language units in keeping with the general use of the term which, as pointed out earlier, i s con-sistent with Bransford's basic conceptualization. Thus, i t w i l l be assumed 40 that context is present when the material to be retained is embedded in an integrated set of semantically related materials (e.g., a word in a meaningful sentence; a sentence in a meaningful paragraph) and absent when there is no overall semantic relation between the presentation materials. The conditions used in the Yuille and Paivio (1969) study can be analyzed in this way, with the Syntactic Order of presentation providing context, and the Random Order condition providing no context. In the free rec a l l measure, only concrete material was affected by context as defined here which would seem to be contrary to the present proposal. However, i t could be argued that free recall of words from a paragraph does not constitute an. appropriate measure of the degree to which integrated meaning of the whole passage is retained. Indeed, i t would seem to related more closely to a measure of retention of lexical features. Presentation in syntactic order may have led to an increase in the degree of semantic integration in the abstract material, but re c a l l of exact words may not have reflected this due to a large proportion of semantically-related decoding errors which would not be included in this measure. This inter-pretation i s supported by the analysis of intrusion errors which showed no overall effect of concreteness on the proportion of theme-related errors. There was a significant effect of presentation order, however, suggesting that context, as provided by syntactic presentation, significantly improved integration of meaning in both concrete and abstract material relative to the no context, random order condition. Thus there was no difference in the degree of semantic integration, as measured by the proportion of theme-related errors, with variation in concreteness of contextual materials. 41 There was also no concreteness effect in the no context condition where the words were presented in random order. While in general this might have been considered i n opposition to the present argument, i t i s i n no way a c r i t i c a l finding here, as the proportion of theme-related errors in this condition was very low in a l l cases and most l i k e l y reflects guessing behavior. The Yuille and Paivio study, then, provides some pre-liminary support for the suggestion that integrative semantic processing does not vary as a function of concreteness when contextual information is present. The studies by Begg and Paivio (1969), Begg (1972), Moeser (1974), and Goldfarb, Wirtz, and Anisfeld (1973) a l l used what are being referred to here as no context materials. In each case the language units being tested for retention were not semantically related to toher presented units. Under these conditions, the Levels of Processing model predicts that the concrete material should have greater access to semantic coding and therefore should exceed abstract materials in memory for meaning. The studies by Begg and Paivio (1969) and by Moeser (1974) used sentences and the above prediction was confirmed. Begg's (1972) experiment involved adjective-noun phrases and revealed a greater amount of integrative pro-cessing in the concrete material, which i s again in line with the present model. Goldfarb, Wirtz, and Anisfeld (1973) also used phrases and found that memory for meaning was greater i n concrete material but this superiority was not significant. It could be that the phrases used here did not differ greatly in the degree of access to the semantic code, even 42 though ..they differed in distinctiveness according to the authors. In the Bransford and Franks (1971 and the Franks and Bransford (1972) studies, groups of related sentences, each expressing part of a complex: idea were presented. Sentences relating to the same idea were not pre-sented consecutively, and yet i t could be argued that this procedure is more closely related to context presentation than when completely un-related sentences are used as in the Begg and Paivio (1969) and Moeser (1974) studies. One would expect, therefore, that on the basis of the previous discussion, concreteness should have less effect on integrative semantic processing in this study as compared to the others. Indeed, this was the case, with both abstract and concrete sentences being inte-grated into h o l i s t i c representations. The evidence reviewed here lends at least tentative support to the proposed effect of context within the Levels of Processing model. In general, concrete language is easier to understand at the level of semantic integration than abstract language. When context is unavailable, this difference leads to a greater degree of semantic processing in the more concrete material. ^Providing context increases the ease of access of •abstract language to integrative semantic coding, and may reduce the difference between concrete and abstract language in memory for meaning. It is worthwhile to speculate how great this reduction might be. The Bransford, Johnson, and Raye (1973) study showed that when retention of meaning was measured by re c a l l of the main ideas, concrete material produced higher performance than abstract, even with context provided. Similarly, the free recall results of Yuille and Paivio (1969) revealed . 43 better performance with the concrete words under the contextual (syntactic) presentation. On the other hand, the Yuille and Paivio analysis of proprotion of theme-related errors showed no concreteness effect and a comparison of the studies using the Bransford and Franks paradigm reveals no difference in the degree of integrative semantic processing as measured by confidence in recognition performance. The implication here i s that i f r e c a l l i s used to measure retention of meaning, then even though there may be a significant reduction in the concreteness effect when context i s present (as shown by Bransford, Johnson, and Raye), the superiority of the more concrete material may s t i l l be demonstrated. However, i f recognition or error analysis i s used, the effect of context may be shown to produce equivalence of concrete and abstract language with respect to semantic processing. This suggested variation according to the nature of the performance measure may be due to differences in retrievability and in the probability of meaning-related decoding errors as a function of concreteness. If the memory representation of abstract language is less retrievable and more subject to decoding errors than that of concrete language, then re c a l l measures (especially free r e c a l l of words) would favour the more concrete material. Specifically, even i f meaning were represented equally well for the two types of material, recall performance might s t i l l show a superiority in the more concrete material. Recogntiion measures or error analysis, on the other hand, are less subject to the effects of retrievability and de-coding errors, and might then reveal the equivalence of the representations. The predictions regarding the effects of concreteness, context, and memory for meaning that have been developed on the basis of the Levels of 44 Processing model relate to the nature of the semantic representation (rather than retrievability per se) and therefore would apply mainly to recognition performance. They can be summarized as follows: Typical examples of concrete langauge are assumed to have greater access to integrative semantic processing than typical examples of abstract language. Providing contextual information significantly increases the access of abstract language (relative to concrete) to semantic processing. The Levels of Processing model outlined earlier predicts, then, that (i) Under no context conditions (e.g., unrelated sentences) memory for meaning w i l l be superior in concrete material as compared to abstract, ( i i ) When context is available (e.g., sentences embedded in meaningful paragraphs) memory for meaning w i l l be equivalent in concrete and abstract material. These predictions were tested in a recent study by Pezdek and Royer (1974). The authors suggested that embedding sentences in paragraphs might increase their comprehensibility. In particular, given the Johnson, Bransford, Nyberg, and Cleary (1972) finding that the Begg and Paivio (1969) concrete sentences were rated as easier to understand than their abstract sentences, Pezdek and Royer proposed that the comprehensibility of abstract sentences might be increased by providing the contextual information of a paragraph. They predicted on this basis that detection of meaning changes in abstract sentences would be increased by embedding the sentences in short, meaningful paragraphs. 45 The procedure used was very similar to that used by Begg and Paivio. Retention of semantic and lexical information in concrete and abstract sentences of the same grammatical form was tested by recogntion of meaning and wording changes. Lexical changes were produced by substituting synonyms for the subject noun of the presented sentence, and semantic changes involved subject-object reversals. Concrete sentences contained high imagery words from the Paivio, Yuille and Madigan (1968) norms, and abstract sentences contained low imagery words. The mean imagery rating for these words in the concrete and abstract sentences was 6.13 and 3.23 respectively. These ratings do not, however, reflect the imagability of the whole sentences. No ratings were available for many of the words used, and the sentences themselves were not assessed on their imagery value. In one group (E) the to-be-tested sentences were presented as the last sentence of a three sentence paragraph. The materials were pre-sented at about a 15 second rate. Two control groups were included that received only the to-be-tested sentences. In C-l, 15 seconds were provided to study each sentence while in C-2, the sentences were presented at a 5 second rate, the latter being therefore the same as the Begg and Paivio procedure. Performance was measured by d'. A comparison of the two control groups showed no significant differences between them, and so a f u l l analysis comparing groups E and C-l was carried out. There was a marginally (p_ < .06) significant interaction between groups, concreteness, and the type of change in the .test sentence. From the figure which was provided, i t 46 appears that for C-l, there was an interaction between concreteness and type of change. Detection of meaning changes was better than detection of wording changes, for both concrete and abstract material, but the difference was much larger for the concrete sentences. Also, performance was better with concrete material in detection of both meaning and wording changes, but the superiority over the abstract material was much larger in detection of meaning changes. In the E group the pattern was much different. There appeared to be no interaction between concreteness and type of change, and no effect of concreteness in either the detection of wording changes or meaning changes. For :both concrete and abstract sentences, meaning changes were detected much better than wording changes and the difference was virtua l l y identical for both types of material. Only one simple effect was analysed, which revealed that detection of meaning changes in abstract sentences was significantly better in group E than in group C-l. In a second study, which included only the E and the C-l groups, this general pattern was replicated. The interaction between concreteness groups and type of change was highly significant this time, as was the increase in detection of meaning changes in abstract sentences in group E relative to C-l. On the basis of these results, Pezdek and Royer concluded that em-bedding abstract sentences in the context of meaningful paragraphs sub-stantially improves detection of semantic changes in these sentences. Furthermore, the graph of their results indicates that detection of wording changes in abstract and detection of both meaning changes and wording 47 changes i n concrete m a t e r i a l was unaffected by paragraph embedding. The authors suggested that the increase i n memory f o r meaning i n the a b s t r a c t sentences as a f u n c t i o n of context was due to an increase i n the com-p r e h e n s i b i l i t y of these sentences. There were two major d i f f e r e n c e s i n the r e s u l t s of t h i s study as compared to the Begg and P a i v i o (1969) study. F i r s t , memory f o r meaning was b e t t e r than memory f o r wording i n a l l c o n d i t i o n s i n the Pezdek. and Royer study. In p a r t i c u l a r , t h i s was tru e f o r both a b s t r a c t and concrete sentences i n the C-2 group which r e c e i v e d e s s e n t i a l l y the same treatment as the Begg and P a i v i o groups. Begg and P a i v i o found that memory f o r meaning was superi o r to memory f o r wording i n the concrete sentences, but the reverse was tru e f o r a b s t r a c t sentences. Pezdek and Royer suggested that t h e i r f i n a l pool of m a t e r i a l s c o n s i s t e d of sentences which were more comprehensible than those used by Begg and P a i v i o . They d i d not o b t a i n any measures of c o m p r e h e n s i b i l i t y to confirm t h i s i n t e r p r e t a t i o n however. Their study d i d r e p l i c a t e the i n t e r a c t i o n between concreteness and type of change i n the C-2 group. The s u p e r i o r i t y i n d e t e c t i o n of meaning changes over wording changes was greater w i t h the concrete m a t e r i a l than i n the ab s t r a c t sentences. The second major discrepancy w i t h the Begg and P a i v i o r e s u l t was found i n the E group. Begg and P a i v i o had found that memory f o r meaning increased s i g n i f i c a n t l y w i t h concreteness w h i l e there appeared to be no e f f e c t of concreteness on d e t e c t i o n of meaning changes i n group E. As noted e a r l i e r , t h i s was i n t e r p r e t e d as being due to an increase i n co m p r e h e n s i b i l i t y of the a b s t r a c t sentences as a f u n c t i o n of paragraph embedding. 48 The results of the Pezdek and Royer study are completely consistent with the predictions of the Levels of Processing model as outlined earlier. Under the no context conditions (C-l and C-2) memory for meaning increased with concreteness. .When context was available (E) there was no difference in memory for meaning as a function of concreteness. Thus context was shown to increase memory for meaning in abstract sentences. Brief mention should be made about the Pezdek and Royer finding that memory for wording was not affected substantially by concreteness. In the i n i t i a l discussion of the Levels of Processing model, i t was suggested that concrete and abstract material have equal access to the lexical code but that differences in memory for meaning might arise due to differences in the amount of processing time spent at the lexical level. This hypothesis was used to account for the difference between the Begg and Paivio (1969) study and the Moeser (1974) study with respect to memory for wording. Specifically, the suggestion was made that at a 5 second presentation rate, more time might be spent on lexical coding i n the abstract material because concrete .language has faster access to the semnantic code which might therefore be reached within the 5 second interval. On this basis, the prediction was made that at a 5 second presentation rate, memory for wording might be better in abstract as compared to concrete sentences, which was the result of the Begg and Paivio study. The C-2 group in Pezdek and Royer's experiment also received a 5 second presentation rate and therefore the same prediction would be made. Inspection of the graph of the results of the d' analysis reveals that there was very l i t t l e difference between abstract and concrete sentences in memory for wording 49 i n the C - l group. In f a c t the performance f o r concrete m a t e r i a l was s l i g h t l y higher than f o r a b s t r a c t sentences and as no s i g n i f i c a n t d i f -ferences were found between C - l and C-2 i t must .be assumed that the same was true i n the C-2 group. This f i n d i n g would argue against the i n t e r p r e t a t i o n o f the Begg and P a i v i o study which had been proposed. A c t u a l l y , however, the p r e d i c t e d e f f e c t was obtained when p r o p o r t i o n of h i t s r a t h e r than d' was used. As the former was the p r i n c i p a l measure used by Begg and P a i v i o , t h i s would seem to be the appropriate comparison. Pezdek and Royer provide a f i g u r e comparing t h e i r r e s u l t on t h i s measure w i t h that of Begg and P a i v i o . I t shows memory f o r wording i n a b s t r a c t sentences to be s u p e r i o r to that i n concrete sentences and t h i s d i f -ference i s v i r t u a l l y i d e n t i c a l i n the two s t u d i e s . Although the Pezdek and Royer study provides complete support f o r the Levels of Proce s s i n g model, i t i s d e f i c i e n t i n many respects as an adequate t e s t of t h i s model i n r e l a t i o n to P a i v i o ' s two-process theory. F i r s t , and most important, i t i s not c l e a r that the 'concrete' and 'abs t r a c t ' sentences used i n the study were s i g n i f i c a n t l y d i f f e r e n t i n ease of access to the imaginal code, e s p e c a i l l y i n the paragraph-embedded c o n d i t i o n . Imagery values of the complete sentences were not measured. The concrete and a b s t r a c t sentences simply d i f f e r e d i n the imagery value of some of the nouns i n some of the sentences. Furthermore, i n the E group, the c o n t e x t u a l m a t e r i a l , i . e . , the f i r s t two sentences o f each paragraph, were designed to be uniformly concrete, which, to use P a i v i o ' s term, could have c o n c r e t i z e d the a b s t r a c t sentences and increased t h e i r ease of access to the image code. Thus, an obvious e x p l a n a t i o n of the r e s u l t s obtained by Pezdek and Royer, i s that there may have been 50 no significant difference in the ease of access of the 'concrete' and 'abstract' materials to the image code in the context group E. If both sets of materials were high and equal in imagability, then on the basis of Paivio's theory, meaning should be retained better than wording and there should be no difference i n this superiority across sets. This of course was the result obtained in the E group. Another problem caused by using concrete context throughout, was that Pezdek- and Royer were unable to reject completely an interpretation (offered by a reviewer) that subjects in the context group who received 'abstract' sentences to remember, may simply have been retaining the concrete contextual sentences, which were presented prior to i t . Thus detection of meaning changes may have been made on the basis of the degree of congruence between the test sentence and this recalled con-text. A further problem in the study concerns Pezdek and Royer's conclusion that the source of the context effect i s an increase in the comprehensibility of the abstract sentences. This is a reasonable suggestion and is in fact congruent with Bransford's conceptualization of the effect of context. Their interpretation however, lacks any empirical support, as no measures of comprehensibility were obtained in the study. A l l sentences used were simply judged as 'sensible and plausible' by a panel of judges. Similarly, no estimates were obtained of the degree of meaning change produced by subject-object reversals which might have contributed to the results i n the same way as Johnson et a l . (1972) suggested in their discussion of the Begg and Paivio (1969) study. 51 One final point concerns the way i n which context availability was manipulated i n this study. This variable is completely confounded with the amount of information which had to be processed by the groups. The control groups had either 5 or 15 seconds to study one sentence, while the context group (E) had only 15 seconds to study three sentences. Thus i t i s not clear that the only difference between the groups was the availability of context, as the variation in processing demands produced by differences i n the amount of information could have contributed to the results. In order to provide a more adequate test of the implications of the Levels of Processing model which has been developed here, two experi-ments were carried out. The next section presents a description of the methods, results, and conclusions of each. 52 THE EFFECT OF CONTEXT ON THE ROLE OF IMAGERY IN LANGUAGE PROCESSING The experiments presented here were designed to evaluate the Levels of Processing model as an alternative to Paivio's two-process theory. In particular, they were carried out to provide a more adequate test of the u t i l i t y of the present approach than did the study by Pezdek and Royer (1974). In Experiment I of the present study, recognition of meaning changes and wording changes was assessed in concrete and abstract material both when context was present and when i t was absent. Context availability was manipulated so that the same '.amount of information was presented under both the context and no context conditions. This was done to ensure that the amount of information could not contribute to any observed effects of context, as may have been the case in the Pezdek and Royer study. Presentation time was also varied in order to assess effects of processing time on the major conditions. The Levels of Processing model makes the following c r i t i c a l pre-dictions about the results of Experiment I: (1) When context is not available, memory for meaning w i l l be superior in concrete material relative to performance with abstract material. (2) When context is provided, concrete and abstract material w i l l be equivalent with respect to memory for meaning. Actually, (2) above is the strong form of the prediction from the model. A weaker version is that the superiority of concrete material in memory 53 f o r meaning should be s i g n i f i c a n t l y reduced i n the context c o n d i t i o n r e l a t i v e to the no context c o n d i t i o n . The model a l s o makes s e v e r a l other p r e d i c t i o n s about performance under the major c o n d i t i o n s of Experiment I . F i r s t , memory f o r meaning should be s u p e r i o r to memory f o r wording f o r both concrete and a b s t r a c t m a t e r i a l when context i s a v a i l a b l e . This p r e d i c t i o n d e r i v e s from the assumption that both kinds of m a t e r i a l have s u b s t a n t i a l access to the semantic l e v e l of processing under t h i s c o n d i t i o n , and there f o r e the primary component of the memory re p r e s e n t a t i o n w i l l be semantic i n f o r m a t i o n . When context i s u n a v a i l a b l e , the p r e d i c t i o n i s l e s s c l e a r . A b s t r a c t m a t e r i a l w i l l have l e s s access to the semantic code than w i l l concrete m a t e r i a l . However, i f the a b s t r a c t sentences are, i n general, more meaningful than those used by Begg and P a i v i o (1969) then memory f o r meaning may be b e t t e r than memory f o r wording i n both a b s t r a c t and con-c r e t e m a t e r i a l , j u s t as Pezdek and Royer (1974) found i n t h e i r no context c o n d i t i o n s . The p r e d i c t i o n s regarding the e f f e c t s of p r e s e n t a t i o n time r e l a t e to the r e l a t i v e importance of a c c e s s i b i l i t y and amount of pro c e s s i n g time. The Levels of Pro c e s s i n g model assumes that i f a c c e s s i b i l i t y to a given l e v e l of processing i s not equal f o r two kinds o f m a t e r i a l , then simply i n c r e a s i n g processing time f o r the m a t e r i a l w i t h l e s s access should have l i t t l e e f f e c t . Thus, f o r the no context c o n d i t i o n , no e f f e c t of pr e s e n t a t i o n time on the r e s u l t s r e l a t i n g to memory f o r meaning i s p r e d i c t e d . The model a l s o assumes, however, that when a c c e s s i b i l i t y i s equal f o r two kinds of m a t e r i a l , then r e t e n t i o n of the features processed at that l e v e l w i l l be greater f o r the m a t e r i a l which i s processed longer 54 at that l e v e l . I f context increases the access of abstract material to semantic processing, to the point where i t i s equal to the access of concrete material to this l e v e l , then memory for meaning may be greater at a longer presentation time regardless of the type of material. S t r i c t l y speaking, however, the model does not assume that context w i l l r esult i n equal a c c e s s i b i l i t y of the semantic code to concrete and abstract material, only that i t w i l l s i g n i f i c a n t l y improve the access of abstract material. The predictions regarding memory for wording relate again to the processing time hypotheses above. Under the context condition, no effects are predicted as meaning should be the primary component of the memory representation of both concrete and abstract material. At the no context condition, the result predicted w i l l depend on the general meaningfulness of the abstract sentences. I f i t i s found that abstract material i s coded primarily for meaning, as indicated by a superiority of memory for meaning over memory for wording, then, again, no effects would be predicted. I f this i s not found to be the case, then i t i s predicted that memory for wording may be greater i n abstract than i n concrete material, due to more time being spent at the l e x i c a l processing l e v e l . Experiment I I was designed to evaluate several alternative hypotheses regarding the results of Experiment I . Ratings were obtained to assess the imagery value, comprehensibility, and the degree of meaning change produced by subject-object reversals, i n the materials used i n Experiment I . I f there i s no change i n the r e l a t i v e imagability of abstract and con-crete sentences as a function of context, then Paivio's two-process theory 55 w i l l not explain any observed effects of context on the relative memory for meaning in the two kinds of material. Similarly, i f there are no effects of context on the degree of meaning change produced by subject-object reversals in concrete relative to abstract material, then sensi-t i v i t y of the test items can be discarded as an explanation of any context effects. -Finally, the relationship between context and comprehensibility ratings may provide information about the source of any effects of context. Pezdek and Royer (1974) and Bransford (e.g., Bransford and Johnson, 1972) have suggested that context increases memory for meaning by increasing the comprehensibility of the material. If so, an effect of context may appear i n the comprehensibility ratings. 56 EXPERIMENT I METHOD Subjects F o r t y - e i g h t unpaid volunteers (24 of each sex) from Summer Session Psychology and Education courses a t the U n i v e r s i t y of B r i t i s h Columbia served as subjects i n the experiment. M a t e r i a l s . . . . (a) P r e s e n t a t i o n M a t e r i a l s Twenty-four sentences of the f o l l o w i n g form were constructed: The ( a d j e c t i v e ) (noun) (past tense verb) a(n) ( a d j e c t i v e ) (noun). H a l f of the sentences were concrete and h a l f were a b s t r a c t . On a seven p o i n t s c a l e from low to high image-evoking c a p a c i t y , the mean r a t i n g (based on 30 subjects) f o r the concrete sentences was 5.07 and the mean r a t i n g f o r the a b s t r a c t sentences was 2.65. Each sentence was designed so that the subject and objec t could be reversed, producing a s u b s t a n t i a l change i n meaning, but l e a v i n g the sentence e s s e n t i a l l y un-changed i n o v e r a l l ease of understanding. In a d d i t i o n , e i t h e r the subje c t o r the objec t of the sentences was a word which had a common synonym that could replace i t i n the sentence without a l t e r i n g the meaning. One of the three p o s s i b l e forms of each sentence (which w i l l be r e f e r r e d to as the C r i t i c a l sentnece - CS) was randomly chosen to be p a r t of the p r e s e n t a t i o n m a t e r i a l s . A short paragraph was const r u c t e d around each CS. Each paragraph c o n s i s t e d of f i v e s h o r t sentences, w i t h the CS as the f o u r t h i n each case. A l l were of s i m i l a r form, and an attempt was made to have each d e s c r i b e a s i n g l e event i n an i n t e g r a t e d 57 fashion. In add i t i o n , the CS was always a statement of ;significance to the o v e r a l l theme of the paragraph. The themes of the paragraphs were kept as d i s t i n c t as pos s i b l e and care was taken not to use the subject and object noun of a p a r t i c u l a r CS i n e i t h e r the paragraph containing i t or other paragraphs. Those paragraphs with concrete CSs (which w i l l be c a l l e d Concrete Sets) were made up of highly d e s c r i p t i v e sentences. Those which contained abstract CSs ( c a l l e d Abstract Sets) were made up of sentences judged to be d i f f i c u l t to represent by imagery. The mean r a t i n g on a seven point image-evoking capacity scale (30 subjects, 1-Low Imagery and 7-High Imagery) f o r the Concrete Sets was 6.00 and for the Abstract Sets was 2.91. An e f f o r t was made to equalize the t o t a l number of words i n each paragraph, and the mean for the Concrete and Abstract Sets was 63 and 65 words, r e s p e c t i v e l y . The 12 Concrete and the 12 Abstract paragraphs made up the Context materials. To form the No Context materials, the 60 sentences from the paragraphs within each l e v e l of concreteness were regrouped to form 12 Concrete and 12 Abstract Sets of f i v e sentences each which did not have any i n t e -grated meaning or theme. The CS remained the fourth sentence i n each of the Sets. The positions of a l l the other sentences however, d i f f e r e d from t h e i r p o s i t i o n i n the Context Set. No two sentences from the same Context paragraph appeared i n the same No Context Set. The basic c r i t e r i o n f o r the choice of these groupings was that the Sets so formed should have no cohesiveness or thematic content and that the CS i n each Set should bear no semantic r e l a t i o n s h i p to the other sentences i n the Set. Imagery 58 ratings were also obtained for these No Context Sets with the mean for the Concrete Sets being 5.07 and for the Abstract Sets, 2.91. Thus there were four kinds of presentation materials: 12 Concrete-Context Sets, 12 Abstract-Context Sets, 12 Concrete-NoContext Sets, and 12 Abstract-No Context Sets. Examples of each type of Set are presented in the Appendix. (b) Test Materials For each of the 12 abstract and 12 concrete CSs, there were three kinds of test sentences: (1) the unchanged CS (U); (2) the CS with the subject and object reversed producing a Semantic change (S); and (3) the CS with a synonym substituted for either the subject or the object which l e f t the meaning intact but produced a Lexical change (L). Half of the L sentences at each level of concreteness involved subject substitutions, and half involved object substitutions. Care was taken that the S sentences, i f substituted in the appropriate Context Set, would disrupt the integrative nature of the Set, while the L sentence could be as easily integrated into the appropriate Context Set as could the CS. Furthermore, care was taken that the S sentences were, not semantically related to any of the presentation materials. Finally, the synonyms substituted in the L sentences did not appear in any of the presented sentences. Examples of the S and L sentences appear in the Appendix. Design and Procedure There were four major variables involved i n the design and each consisted of two levels; Concreteness (Concrete and Abstract), Context 59 (Context and No Context), Type of Change (Semantic and L e x i c a l ) , and P r e s e n t a t i o n Time (Short and Long). Concreteness and Context were be-tween subjects v a r i a b l e s w i t h four groups of 12 subjects (Ss) eahh (6 males) r e c e i v i n g one of the f o l l o w i n g types of p r e s e n t a t i o n m a t e r i a l s described above: Concrete-Context,Abstract-Context, Concrete-No Context, or Abstract-No Context. Type of Change and P r e s e n t a t i o n Time were w i t h i n subjects w i t h each ^ r e c e i v i n g four Semantically changed t e s t sentences and four L e x i c a l l y changed t e s t sentences. W i t h i n each type of t e s t sentence, two were presented a t the Short time and two were presented at the Long time. There were four sub-groups w i t h i n each of the four major groups. H a l f of the jSs r e c e i v e d P r e s e n t a t i o n Order A and the other h a l f r e c e i v e d P r e s e n t a t i o n Order B. Wi t h i n each P r e s e n t a t i o n Order c o n d i t i o n , h a l f of the Ss re c e i v e d P r e s e n t a t i o n Time Order 1 and the other h a l f r e c e i v e d P r e s e n t a t i o n Time Order 2. F i n a l l y , w i t h i n each P r e s e n t a t i o n Time Order group, one r e c e i v e d Test Booklet ay one r e c e i v e d Test Booklet b_, and one re c e i v e d Test Booklet c^ A schematic diagram o f the design i s shown i n Fi g u r e 1. The two P r e s e n t a t i o n Orders were r e l a t e d i n the f o l l o w i n g ways: ( i ) I f a Set was presented i n the f i r s t s i x Sets i n A, then i t was presented i n the l a s t s i x i n B. ( i i ) No two Sets t h a t were presented c o n s e c u t i v e l y i n A, were presented c o n s e c u t i v e l y i n B i n the same order, ( i i i ) I f a Set was presented i n e i t h e r the f i r s t three o r the l a s t three Sets i n A, then i t was presented i n the middle s i x Sets i n B. Note that the 60 2 3C ul < o r tu o 2 o a i a i m r-t O l U r * * *! g o o (0 t» § vn oi U J UJ LU a i a c * O u l 5 8 9 o o o o o a o o o CO COcO c0 <5 <£> SSS U) IU Ul t" r H fei 8 < •z I •a UJ IS cc Ui @ O •z vA 8 Ui h 8 o o i> 2 VJ c 3 § -u i o CD <c < 5b-lis CO Q «* o Ul ON < da 8 . O >_> 2 FIGURE 1. DESIGN OF EXPERIMENT I 61 order of the C r i t i c a l Sentences in A and in B was the same for the Context and No Context groups. Time Order 1 was a random ordering of six Short and six Long presentation times, and Time Order 2 was the 'complement' of this order so that i f the nth item in order 1 was Short, then i t was Long in order 2. Each S_ received a test booklet containing one sentence from each of the Sets presented: four (U) sentences, four (S) sentences, and four (L) sentences. For Test Booklet a., the choice of which of the pre-sented sets would be tested by which type of test sentence was random except for the constraints that two of the Sets had to have been pre-sented at the Short time and the other two at the Long time, and this had to apply to both of the Time Order. Test Booklets b_ and c_ were then the 'complements' of _a, such that i f a CS from Set (n) was tested with an S test sentence in Test Booklet a_, then i t was tested with an _L sentence in b_ and a U sentence in c_. By counterbalancing in this way, each of the Sets in each major group was presented i n both the f i r s t and the last half of the presentation order, both in the middle and at the ends of the order, under both the Long and the Short presentation times, and with each type of test sentence. This was done i n order to control for any possible main effects of the particular Sets that were used within each group. For example, i f a given Set was tested only at the Short time and another was presented only at the Long time, then any effect of Presentation'* Time that might appear could be due to some unforseen difference between the Sets, which would then severely restr i c t the generalizability of the effect. 62 Each received a booklet containing the 12 Sets of presentation materials appropriate to his group. There was one set per page and a blank sheet of heavy paper between each set. The sets were presented auditorily on a tape recorder, each Set being read at a slow to moderate rate with approximately a three second pause between each sentence. Care was taken not to provide any emphasis in the reading to ensure that the Context and No Context material were presented i n the same fashion. Subjects were told that a number of sets of sentences, each consisting of five sentences, would be read to them on the tape and that they were to read along with the recording s i l e n t l y . Just prior to each set the recording announced "Turn the page" and after a five second delay the sentences i n that set were read. Three seconds after the last sentence was read, the tape was stopped and one of two instructions was given. For those six Sets presented at the Short presentation time, the i n -struction was given to turn the page exposing the blank sheet concealing the next Set. For jfche other six Sets presented for the Long time, the instruction was given to continue to study the material and after exactly 15 seconds, the order was given to turn the page to the blank. Five seconds after the pages were turned, the recorder was turned on and i t gave the instruction to turn the page and expose the next set. After five seconds, the sentences in this set were read and the above procedure repeated. Thus approximately 10 seconds elapsed between the time when one Set was removed from view and the next Set was exposed. Each Set took approximately 15 seconds to read and thus the total presentation took about 6 to 7 minutes (15 sec/short X 6 plus 30 sec/Long X 6 plus 10 sec/ interval X 6). Subjects were instructed to try to remember as much as 63 they could about each Set, and to pay a t t e n t i o n to both the meaning of the sentences and the s p e c i f i c words that were used. F o l l o w i n g p r e s e n t a t i o n of the 12 Sets, Ss were provided w i t h Test Booklets, e i t h e r a, b_, or c. They were t o l d that the booklet contained a sentence derived from each of the Sets that were presented. I t was s p e c i f i e d that some of the sentences i n the booklet would be i d e n t i c a l to those that were presented, and some would be changed s l i g h t l y . They were t o l d that b a s i c a l l y two types of changes could occur, e i t h e r one or more words may be s u b s t i t u t e d i n t o the sentence, or the same words may appear but be rearranged to a l t e r the meaning. They were r e q u i r e d to c i r c l e 'changed' o r 'unchanged' under each sentence i n the t e s t booklet and to r a t e how c o n f i d e n t they were i n t h e i r judgment on a f i v e - p o i n t s c a l e from low to High. The Ss paced themselves through the task but were t o l d to do the items i n the order given and not to go back and change t h e i r choice a f t e r a f i n a l d e c i s i o n was made. They were t o l d to do each and every item and i f they were guessing completely, to simply mark a '1' on the confidence s c a l e . I t was s t r e s s e d that they should mark 'changed' i f the t e s t sentence was i n any way d i f f e r e n t from what was presented, and to mark 'unchanged' only i f i t was i d e n t i c a l . The Ss were t o l d that there may be any number of examples of each pos-s i b i l i t y i n t h e i r b o o k l e t , and that they should not worry about making any one choice too o f t e n as long as i t was t h e i r true judgment. Approximately 2 to 2% minutes elapsed between p r e s e n t a t i o n of the l a s t sentence and when jSs opened t h e i r t e s t b o o k l e t s . The t o t a l time f o r the experiment was approximately 20 minutes and jSs were test e d i n groups of three. 64 RESULTS The principal aim of the anlayis of results was to evaluate the effects of concreteness, context, and presentation time on the recognition of meaning changes and wording changes. In order to gain a maximum of information from the data, a number of dependent measures were used. This method of evaluating results seems particularly appropriate in the present study in that no single dependent measure seems to be completely sufficient in assessing recognition performance i n the Sachs (1967) paradigm. Begg and Paivio (1969), for example, used five different per-formance measures. Pezdek and Royer (1974) used d' as the major dependent variable i n evaluating detection of meaning and wording changes. While this measure of recognition performance has gained wide acceptance in a number of areas, i t was considered inappropriate for use in the present study, due to the samll number of detections that each was required to make here. Theoretically, d' measures the 'distance'between two normal curves, a signal curve and a noise curve. To assign such a value to each j> on the basis of only 12 observations would seem high inappropriate. In addition, the recognition paradigm used here produces many scores of 0% and 100%. As pezdek and Royer point out, these values must be given arbitrarily large /Z/ scores i n computingd' which can lead to a distortion of the data. The Pezdek and Royer study, involving 32 ob-servations per S_, showed that this distortion • was not major in their case and an analysis based on proprotion of hits produced equivalent results. 65 Despite this finding, the arguements raised here s t i l l apply to the general validity of the d' measure in recognition studies using so few observations per jS. Indeed, the equivalence of d' to a measure which is not even corrected for guessing, as shown in the Pezdek and Royer experiment, was probably due to the equivalence of guessing behavior across conditions. This is not l i k e l y to be the general case and when i t i s , d' i s completely superfluous and adds nothing to the analysis. Five different dependent variables were used in the analysis. They are described separately below, along with the results of an analayis of variance carried out on each. Hits Hits are defined here as the correct detection of a changed test item. Therefore, every had four Hit scores, one for each combination of Type of Change (Semantic or Lexical) and Presentation Time (Short or Long). The maximum number of Hits i n each category was two. The mean proportions of Hits for each of the four independent groups (n=12) are presented in Table I. A 2 X 2 X 2 X 2 analysis of variance was carried out on the Hit scores, with" Context and Concreteness between j>s \and Type of Change and Presentation Time within Ss. The analysis, as shown in Table II, revealed only one significant effect, namely the main effect of Type of Change, F_ (1,44) = 25.58, p_ < .001". Semantic changes were detected at a significantly higher rate than Lexical changes at a l l levels of Concreteness, Context, and Presentation Time. TABLE I MEAN PROPORTION OF HITS AS A FUNCTION OF TYPE OF CHANGE, PRESENTATION TIME, CONCRETENESS, AND CONTEXT. Semantic L e x i c a l Short Long Short Long Concrete-Context Concrete-No Context Abstract-Context Abstract-No Context 0 . 8 7 5 1 . 0 0 0 0 . 8 7 5 0 . 7 9 2 0 . 8 7 5 0 . 9 1 7 0 . 9 1 7 0 . 8 3 3 0 . 7 0 8 0 . 6 6 7 0 . 6 6 7 0 . 6 6 7 0 . 6 6 7 0 . 5 4 2 0 . 5 0 0 0 . 7 0 8 67 TABLE I I SUMMARY OF THE ANALYSIS OF VARIANCE ON HITS SOURCE df SS MS F Between Ss 47 16.62 - -Concreteness (A) 1 0.26 0.26 * Context (B) 1 0.01 0.01 * A X B 1 0.00 0.00 * Ss/A X B 44 16.35 0.37 -Within Ss 144 67.25 - -Type of Change (C) 1 11.51 11.51 25.58 Presentation Time (D) 1 0.26 0.26 * C X D 1 0.25 0.25 * C X Ss 47 21.24 - -C X A 1 0.08 0.08 * C X B 1 0.00 0.00 * C X A X B 1 1.55 1.55 3.44 C X Ss/A X B 44 19.61 0.45 -D X Ss 47 14.49 - -D X A 1 0.12 0.12 * D X B 1 0.00 0.00 * D X A X B 1 0.43 0.43 1.34 D X Ss/A X B 44 13.94 0.32 -C X D X Ss 47 19.50 _ C X D X A 1 0.10 0.10 * C X D X B 1 0.13 0.13 * C X D X A X B ; i 0.09 0.09 * C X D X Ss/A X B 44 19.18 0.44 -Total 191 83.87 <.001 <.10 note: * F < 1 68 Hits-False Alarms Hit scores for each S_ were corrected for guessing by substracting False Alarms which are defined here as marking 'changed' for an unchanged item. As there were four unchanged sentences i n each test booklet, each j> received two False Alarm scores, one for the Short presentation time and one for Long, with a maximum score of two i n each case. These False Alarm scores were substracted from the Hit scores described above, within the appropriate Presentation Time condition. Thus, the False Alarm score (for a given S) under the Short condition was subtracted from the Hit score for both the Semantic Short condition and the L e x i c a l Short condition. S i m i l a r l y , each S/sFalse Alarm score under the Long condition was subtracted from both the Semantic and the Lexical Hit scores under that condition. This correction for guessing does not, therefore, affect the difference between the Semantic and Lexical Hit scores for a given :j> under each time condition. I t could, however, affect the relationship between the levels of other variables i f guessing behavior varied s i g n i f i -cantly across the conditions. Table I I I shows the mean proportion of Hits minus the mean proportion of False Alarms i n each of the major groups. The analysis of variance on the (Hit-False Alarm) scores revealed the same result as the previous analysis of uncorrected Hit scores. Only Type of Change was s i g n i f i c a n t , _F(1,44) = 25.55, p_ < .001), with Semantic changes being detected better than L e x i c a l changes. The results of the analysis of variance are presented i n Table IV. TABLE I I I MEAN PROPORTION OF HITS MINUS MEAN PROPORTION OF FALSE ALARMS AS A FUNCTION OF TYPE OF CHANGE, PRESENTATION TIME, CONCRETENESS, AND CONTEXT SEMANTIC SHORT LONG .. LEXICAL SHORT: LONG Concrete-context .500 Concrete-no context .792 Abst r a c t - c o n t e x t .583 Abstract-no context .542 ,667 ,833 ,708 ,375 .333 .458 .375 .417 .458 .458 .292 .250 70 TABLE IV SUMMARY OF THE ANALYSIS OF VARIANCE ON HITS-FALSE ALARMS SOURCE df SS MS F P Between Ss 47 45,74 Concreteness (A) 1 2.75 2.75 3.02 <.l Context (B) 1 0.13 0.13 A A X B 1 2.76 2.76 3.03 <.l Ss/A X B 44 40.10 0.91 — Within Ss 144 113.25 — — Type of Change (C) 1 11.50 11.50 25.55 <.001 Presentation Time (D) 1 0.00 0.00 * C X D 1 0.26 0.26 * C X Ss 47 21.25 C X A 1 0.14 0.14 * C X B 1 0.01 0.01 * C X A X B 1 1.49 1.49 3.31 <.l C X Ss/A X B 44 19.60 0.45 — D X Ss 47 60.75 — — D X A 1 1.18 1.18 A D X B 1 1.17 1.17 * D X A X B 1 0.04 0.04 * D X Ss/A X B 44 58.36 1.33 C X D X Ss 47 19.49 — , C X D X B 1 0.04 0.04 * C X D X B 1 0.13 0.13 * C X D X A X B 1 0.15 0.15 A C X D X Ss/A X B 44 19.18 0.44 Total 191 158.99 Note: * <1 71 Proportion of Hits While the previous analyses based on Hits and Hits-False Alarms produced no significant interactions, the Concreteness X Type of Change X Context effect was very close to significance in both cases: Hits: F(l,44) = 3.44, p_ < .10; Hits-False Alarms: F(l,44) = 3.31, p_ < .10. Figure 2 presents the relationship between these variables for the pro-portion of Hits - proportion of False Alarm scores contained in Table III. Inspection of this figure suggests that the principal contribution to the interaction was a much more pronounced Concreteness X Context interaction for Semantic changes than for Lexical changes. Indeed, an analysis of simple interaction effects showed that this interaction was significant for the Semantic change condition, J?(l,44) =6.15, p_ < .05, and not significant for Lexical changes. The source of the interaction here appears to be a much larger effect of Concreteness under the No Context condition than at Context. These considerations suggest that an analysis be done using a dependent measure which mainly reflects the performance in detection of Semantic changes, (a) Proportion of Semantic Hits The dependent measure chosen was the Proportion of Semantic Hits, defined here as the ratio of Semantic Hits to the total number of Hits (Semantic plus Lexical Hits). The Hit scores making up the ratio were f i r s t corrected for guessing. For each j> the number of Semantic Hits (max. 2), the total number of Hits (Semantic plus Lexical, Max. 4) and the number of False Alarms (max. 2) were tabulated under each of the two Presentation Time conditions. Then the following formula was applied: 72 FIGURE %> INTERACTION BETWEEN CONTEXT, CONCRETENESS, AMD TYPE OF CHANGE IN EXPERIMENT I 73 (Semantic Hits),,, - 1/2. (False Alarms)^ i. 1. T - Txme l l i - Short or (Total Hits) - 1/2 (False Alarms) L ° n g i i Each S_ therefore received two proportion scores, one for each level of Presentation Time. The correction factor of l/2(False Alarms) was used in order to include a l l 48 Ss in the data. If the False Alarm score i t s e l f had been used, then S_s who had Total Hits equal False Alarms at any level of Presentation Time would have had to be eliminated because the denominator would be zero. (Compared to Hit-False Alarm scores, using this correction does not produce any distortion in the data as each j5's score i s simply increased by one-half his False Alarm score. An analysis done using this alternate correction and the results were vir t u a l l y identical to those summarized in Table IV). Table V presents the mean Proportion of Semantic Hit scores for the major groups (n=12). A 2 X 2 X 2 analysis of variance was carried out on these proportion scores, with Concreteness and Context between Ss and Presentation Time within Ss. The only significant effect was an inter-action between Concreteness and Context, F(l,44) = 4.11, p< .05). Analysis of simple effects revealed that Concreteness was significant at the No Context condition, F_(l,22) = 5.00, p_ < .05), but not at the Context condition, F_< 1. Specifically, there was a higher proportion of Semantic Hits in Concrete sentences than in Abstract sentences in the No Context condition and no effect of Concreteness on this proportion for the Context condition. The results of the analysis are presented in Table VI. (b) Proportion of Lexical Hits A similar analysis was done on the Proportion of Lexical Hits, again TABLE V MEAN PROPORTION OF SEMANTIC HITS AS A FUNCTION OF PRESENTATION TIME, CONCRETENESS AND CONTEXT Short Long Concrete-Context 0.51 0.57 Concrete-No Context 0.61 0.68 Abstract-Context 0.54 0.63 Abstract-No Context 0.44 0.44 J 75 TABLE VI SUMMARY OF THE ANALYSIS OF VARIANCE ON PROPORTION OF SEMANTIC HITS SOURCE df SS MS F Between Ss 4_7 4. 63 — — Concreteness (A) 1 0. 14 0. 14 1.55 Context (B) 1 0. 01 0. 01 A A X B 1 0. 37 0. 37 4.11 Ss/A X B 44 4. 11 0. 09 — Within Ss 48 3. 97 — — Presentation Time (C) 1 0. 08 0. 08 * C X Ss 47 3. 89 — — C X A 1 0 0 * C X B 1 0 0 •k C X A X B 1 0. 02 0. 02 * C X Ss/A X B 44 3. 87 0. 09 Total 95 8. 60 Note: *F<1 76 corrected for guessing using the 1/2(False Alarm) factor. The means are shown in Table VII and the results of the analysis of variance, which revealed no significant effects, are presented in Table VIII. Confidence Ratings (a) High Confidence Hits Two analyses were carried out to assess :the effects of Concreteness, Context, Presentation Time, and Type of Change on the confidence that jSs had in their recognition performance. In the f i r s t , each S_ was given a score of 1 for each correct identification of a change in which a confidence rating of 5 was given. A l l other detections with lower confidence ratings were eliminated from the analysis. Each S^, therefore, had four High Confidence Hit scores, one at each combination of Type of Change and Presentation Time. No correction for guessing was made here, as this i s provided for by the elimination of the low confidence detections. In Table IX i s shown the mean High Confidence Hit score (max. 2) for each of the groups (n=12). A 2 X 2 X 2 X 2 analysis of variance, with Concreteness and Context between and Type of Change and Presentation Time within, was carried out on these scores. The results as shown in Table X were very similar to those from the proportion of Semantic Hits analysis. In addition to a significant effect of Type of Change [F(l,44) = 38.41, p_< .001; Semantic greater than Lexical), there was a significant interaction between Con-creteness and Context, F_(l,44) = 4.29, p_ < .05. The analysis of simple main effects carried out on this interaction, showed a significant effect TABLE V I I MEAN PROPORTION OF LEXICAL HITS AS A FUNCTION OF PRESENTATION TIME, CONCRETENESS, AND CONTEXT SHORT LONG C o n c r e t e - c o n t e x t .34 .34 Concrete - no context .32 .33 A b s t r a c t - context .33 .27 A b s t r a c t - no context .42 .35 TABLE'VIII SUMMARY OF THE ANALYSIS OF VARIANCE ON PROPORTION OF LEXICAL HITS Source Between Ss Concreteness (A) Context (B) A X B Ss/A X B Within Ss Presentation Time (C) C X Ss C X A C X B C X A X B C X Ss/A X B Total df SS MS 47 2.80 1 0.00 0.00 1 0.03 0.03. 1 0.05 0.05 44 2.72 0.06 48 2.73 1 0.02 0.02 47 2.71 1 0.03 0.03 1 0.00 0.00 1 0.00 0.00 44 2.68 0.06 95 Note: *F<1 TABLE IX MEAN HIGH CONFIDENCE HITS AS A FUNCTION OF TYPE OF CHANGE, PRESENTATION TIME, CONCRETENESS, AND CONTEXT SEMANTIC LEXICAL SHORT . LONG SHORT LONG Concrete context Concrete no context A b s t r a c t context A b s t r a c t no context 1.08 1.42 1.00 0.67 1.33 1.50 1.33 0.67 0.50 0.75 0.58 0.33 0.58 0.50 0.58 0.42 80 TABLE X SUMMARY OF THE ANALYSIS OF VARIANCE ON HIGH CONFIDENCE HITS Source df SS MS F Between Ss 47 40.6 — — Concreteness (A) 1 3.2 3.20 4.16 < .05 Context (B) 1 0.4 0.40 * A X B 1 3.3 3.30 4.29 < .05 Ss/A X B 44 33.7 0.77 Within Ss ,144 74.7 — — Type of Change (C) 1 16.9 16.90 38.41 < .oo: Presentation Time (D) 1 0.2 0.20 * C X D 1 0.5 0.50 1.72 — C X Ss 4_7 21.3 — — C X A 1 1.1 1.10 2.50 — C X B 1 0.1 0.10 * C X A X B 1 0.7 0.70 1.59 — C X Ss/A X B 44 19.4 0.44 — D X Ss 47 23.0 — • . D X A 1 0.1 0.10 A D X B 1 0.5 0.50 A D X A X B 1 0.0 0.00 A D X Ss/A X B 44 22.4 0.51 C X D X Ss 47 12.8 — C X D X A 1 0.1 0.10 A C X D X B 1 0.0 0.00 A C X D X A X B 1 0.1 0.10 A C X D X Ss/A X B 44 12.6 0.29 Total 191 115.3 — — Note *T< \ 81 of Concreteness at No Context, F_(l,22) = 6.64, p_ < .05, and no effect of Concreteness for the Context condition (P[=0) . In particular there were more High Confidence Hits with Concrete material than with Abstract at No Context, and no difference on this measure as a function of Concreteness at Context. (b) Confidence Ratings The second analysis of the confidence data was carried out to assess the degree of confidence expressed in recognition performance independent of correctness. For each of the eight changed items, Ss were given a score equal to their confidence rating for that item (1-low and 5-high). Thus, each S^  had four Confidence Rating scores, one at each combination of Type of Change and Presentation Time, where the maximum score i s 10, i.e., a rating of 5 given to both items in that category. Table Xi shows these scores expressed in terms of the mean rating per item (max. 5) i n each group. The 2 X 2 X 2 X 2 analysis of variance showed a significant effect of Type of Change, with higher confidence expressed in Semantic test items than i n Lexical test items, F(l,44) = 12.64, p_ < .01. There was also a significant interaction between Concreteness and Context [F(l,44) = 4.32, p_ < .05) with the simple effects analysis revealing higher confidence in Concrete test items than in Abstract test items under the No Context con-dition [F_(l,22) = 5.50, p_ < .05) and no effect of Concreteness in the Context condition (J? < 1). Table XII shows the results of this analysis. TABLE XI MEAN CONFIDENCE RATING PER ITEM AS A FUNCTION OF TYPE OF CHANGE, PRESENTATION TIME, CONCRETENESS, AND CONTEXT SEMANTIC LEXICAL SHORT LONG SHORT LONG Concrete context 4.17 Concrete no context 4.54 Abstract context 3.96 Abstract no context 3.92 4.42 3.33 3.67 4.50 3.92 4.25 4.25 4.08 3.96 3.96 3.71 3.63 83 TABLE XII SUMMARY OF THE ANALYSIS OF VARIANCE ON CONFIDENCE RATINGS. SOURCE •df SS MS F Between Ss 41 244.81 - -Concreteness (A) 1 5.33 5.33 1.08 Context (B) 1 1.02 1.02 * A X B 1 21.34 21.34 4.32 <.0l Ss/A X B 44 217.12 4.94 -Within Ss 144 427.00 - -Type of Change (C) 1 30.08 30.08 12.64 <.0J Presentation Time (D) 1 3.00 3.00 A C X D 1 0.02 0.02 * C X Ss 47 117.92 - -C X A 1 9.19 9.19 3.86 < .1 C X B 1 0.34 0.34 A C X A X B 1 3.55 3.55 1.44 -C X Ss/A X B 44 104.85 2.38 -D X Ss 47 160-. 00 - -D X A 1 1.69 1.69 A D X B 1 0.75 0.75 A D X A X B 1 0.01 0.01 A D X Ss/A X B 44 157.55 3.58 -C X D X Ss 47 115.98 - -C X D X A 1 3.00 3.00 1.17 -C X D X B 1 1.02 1.02 A C X D X A X B 1 0.00 0.00 A C X D X Ss/A X B 44 112.98 2.57 -TOTAL 191 671.81 _ _ Note: * F< 1 84 Summary of Re s u l t s Two c o n s i s t e n t f i n d i n g s were revealed i n the analyses of r e s u l t s . F i r s t , Type of change was s i g n i f i c a n t on a l l of the dependent measures (except the p r o p o r t i o n measures where i t was not inc l u d e d as a v a r i a b l e ) . Semantic changes produced by s u b j e c t - o b j e c t r e v e r s a l s were detected b e t t e r than L e x i c a l changes produced by s u b s t i t u t i n g a synonym f o r e i t h e r the subject or the obje c t of the presented sentence. This was true f o r H i t s , H i t s - F a l s e Alarms, and High Confidence H i t s . In a d d i t i o n , greater confidence was expressed i n Semantic t e s t items than i n L e x i c a l t e s t items. The second major f i n d i n g was the i n t e r a c t i o n of Concreteness w i t h Context that emerged i n the P r o p o r t i o n of Semantic H i t s , High Confidence H i t s and Confidence Ratings. Thus Concrete sentences had a higher pro-p o r t i o n of Semantic H i t s , a greater number of High Confidence H i t s and higher Confidence Ratings given to them, than A b s t r a c t sentences when both were presented a t the No Context c o n d i t i o n . On the other hand, there was no Concreteness e f f e c t on these measures when the m a t e r i a l s were presented i n the Context c o n d i t i o n . 85 EXPERIMENT I I The s i g n i f i c a n t e f f e c t of Type of change found i n Experiment I suggests that memory f o r meaning i s s u p e r i o r to memory f o r wording f o r both concrete and ab s t r a c t sentences, presented w i t h or without c o n t e x t u a l i n f o r m a t i o n , and given e i t h e r one reading or 15 seconds of e x t r a study time. In terms of the Lev e l s of Processing model, the i m p l i c a t i o n i s that there i s some degree of access to the semantic l e v e l of processing under each of these c o n d i t i o n s . The s u p e r i o r i t y of memory f o r meaning over wording i n the a b s t r a c t sentences presented without contextual i n f o r m a t i o n i s c o n s i s t e n t w i t h the r e s u l t reported by Pezdek and Royer, (1974) but i s the reverse of the f i n d i n g i n the Begg and P a i v i o (1969) study. Pezdek and Royer suggested that t h i s d i f f e r e n c e was due to the greater c o m p r e h e n s i b i l i t y of t h e i r a b s t r a c t sentences as compared to the Begg and P a i v i o a b s t r a c t m a t e r i a l s . This might a l s o be the case i n Experiment I , and the i n t e r p r e t a t i o n can be assessed i n t h i s case by o b t a i n i n g r a t i n g s of the comprehensi-b i l i t y of the a b s t r a c t m a t e r i a l s used there and comparing, them w i t h those reported by Johnson et a l . (1972) obtained on the Begg and P a i v i o m a t e r i a l s . Another p o s s i b l e explanation of the d i f f e r e n c e i s that s u b j e c t -object r e v e r s a l s produced greater changes i n meaning i n the ab s t r a c t sentences used i n Experiment I compared to the e f f e c t of such changes i n Begg and P a i v i o ' s a b s t r a c t sentences. Again, t h i s p o s s i b i l i t y can be evaluated by comparing r a t i n g s of the s i m i l a r i t y i n meaning between the o r i g i n a l sentences and those w i t h the subject and object reversed, w i t h the r a t i n g s on t h i s measure reported by Johnson et a l . 86 The general s u p e r i o r i t y i n d e t e c t i o n of Semantic changes over L e x i c a l changes i s open to another i n t e r p r e t a t i o n besides r e t e n t i o n of meaning being b e t t e r than memory f o r wording. . I t could be that Semantic and L e x i c a l t e s t items d i f f e r e d i n c o m p r e h e n s i b i l i t y , and that performance was based on t h i s , r a t h e r than t r u e r e c o g n i t i o n of meaning. I f , f o r example, the Semantic t e s t items were harder to understand than the L e x i c a l t e s t items, then Ss might have a tendency to mark 'changed' f o r the Semantic items to a greater extent than f o r the L e x i c a l items. That i s , choices might be based on how meaning-f u l the t e s t sentences are, r a t h e r than whether or not they were a c t u a l l y presented. This might occur i f the p r e s e n t a t i o n sets are remembered as g e n e r a l l y easy to understand, and t h e r e f o r e any t e s t sentence which i s not, i s given a 'changed' score. In l i g h t of t h i s p o s s i b i l i t y , the t e s t sentences used i n Experiment I should be rated on c o m p r e h e n s i b i l i t y . B r i e f mention should be made of the f a c t that no e f f e c t s of pr e s e n t a t i o n time were found i n Experiment I . In the Levels of Processing model, processing time i s assumed to be a s i g n i f i c a n t f a c t o r i n memory, only when ease of access to a given l e v e l of processing i s equal f o r the m a t e r i a l s under c o n s i d e r a t i o n . Thus, the e f f e c t of concreteness on memory f o r wording was assumed to depend on the amount of study time given to the m a t e r i a l . This e x p l a n a t i o n was used to account f o r the d i f f e r e n c e s between the Begg and P a i v i o (1969) study and the Moeser (1974) study, w i t h regard to memory f o r wording i n concrete and ab s t r a c t sentences. Begg and P a i v i o found that memory f o r wording decreased w i t h concreteness when a f i v e 87 second presentation rate was used, whereas Moeser found the reverse when Ss were given unlimited study time. In the framework of the Levels of Processing model, the implication i s that more time is spent on lexical processing in abstract than in concrete material when study time is limited, but less time is spent on lexical coding of abstract relative to concrete sentences when study;:time is unlimite In Experiment I, the Short presentation was about 15 seconds for each set of five sentences, while the Long presentation condition allowed an additional 15 seconds of study time for the set. This extra time (about thee seconds per sentence) i s much less than the unlimited study time given in the Moeser (1974) study, and therefore i t is not surprising that there was no effect of presentation time in the present study. Furthermore, even in the Short presentation condition, memory for meaning was superior to memory for wording which would tend to minimize the effects of processing time discussed above. Although the interaction between concreteness and context was present in the analysis of three of the dependent measures, the effect i s c r i t i c a l in i t s implications for memory processes only in the case of the Proportion of Semantic Hits measure. The two variables related to confidence (High Confidence Hits and Confidence Ratings) are primarily supportive evidence and do not relate directly to questions of memory performance. Further discussion of the interaction w i l l therefore concentrate on the Proportion of Semantic Hits and mention of the other measures w i l l be made only when issues other than performance are being considered. 88 Experiment I showed that there was a greater p r o p o r t i o n of Semantic H i t s i n concrete sentences as compared to a b s t r a c t sentences at the No Context c o n d i t i o n , and no e f f e c t of concreteness at the Context c o n d i t i o n . This f i n d i n g suggests that concreteness s i g n i f i c a n t l y improves the r e l a t i v e amount of semantic processing only when con-t e x t u a l i n f o r m a t i o n i s u n a v a i l a b l e . When context i s provided, by embedding sentences i n paragraphs, there i s no d i f f e r e n c e between concrete and a b s t r a c t m a t e r i a l i n the p r o p o r t i o n of semantic coding. This f i n d i n g i s completely c o n s i s t e n t w i t h the p r e d i c t i o n of the L e v e l s of processing model presented i n the i n t r o d u c t i o n . The r e s u l t i s open to a number of a l t e r n a t i v e i n t e r p r e t a t i o n s , however. F i r s t , i t could be the case that embedding ab s t r a c t sentences i n paragraphs increases t h e i r ease of access to the imagery code, and i f so, the r e s u l t would be accounted f o r by P a i v i o ' s two-process theory. In order to t e s t t h i s h ypothesis, i t i s necessary to o b t a i n measures of the i m a g a b i l i t y of the concrete and a b s t r a c t m a t e r i a l s used i n Experiment I . I f there proves to be a s i g n i f i c a n t d i f f e r e n c e i n the ease of imagery between these sentences at both the Context and No Context c o n d i t i o n s , then P a i v i o ' s theory would be without support i n t h i s experiment. A second p o s s i b l e e x p l a n a t i o n of the i n t e r a c t i o n between Con-creteness and Context r e l a t e s to the degree of meaning change produced by sub j e c t - o b j e c t r e v e r s a l s . I f there was a d i f f e r e n c e on t h i s v a r i a b l e between the concrete and a b s t r a c t m a t e r i a l s at the No Context c o n d i t i o n such that these changes a l t e r e d the meaning of concrete sentences more than a b s t r a c t sentences, but no such d i f -89 ference at Context, then the interaction would be predicted. Thus the degree of s i m i l a r i t y i n meaning between the o r i g i n a l and Semantic changes must be assessed for the abstract and concrete material at both Context and No Context. If both of these alternatives proved to be without support, then the Levels of Processing model w i l l emerge as a viable explanation of concret eness effects i n language processing. With the model, context i s assumed to increase the degree of access of abstract material to the semantic l e v e l of processing. A possible mechanism for t h i s effect would be an increase i n the comprehensibility of the abstract language r e l a t i v e to concrete. This interpretation might be assessed by obtaining measures of the comprehensibility of concrete and abstract sentences from Experiment I. under the Context and No Context condition. Support for t h i s suggested mechanism for the effect of context would be gained i f an increase i n the comprehensibility of Abstract r e l a t i v e to Concrete material occurred as a function of the Context manipulation i n Experiment I. Experiment I I was carried out i n order to evaluate the various interpretations of Experiment I that are presented above. Method A t o t a l of 114 ^s took part i n the study, which involved three different tasks: Imagery Rating, Comprehensibility Rating and S i m i l a r i t y of Meaning Rating. The methods involved i n each of the tasks are described separately below. 90 Imagery Rating  Subj ects Thirty unpaid volunteers (15 of each sex) from second and third year Psychology courses at The University of British Columbia served as subjects. Materials The materials to be rated were a l l taken from Experiment I. Two sets of material were used. The No Context Set consisted of the 12 Concrete and 12 Abstract No Context Sets from Experiment I and the Context materials consisted of the 24 Sets presented to the Context groups in that study. For each type of material booklets containing the 24 Sets were made up. There was one Set per page and the pages were randomly ordered i n each booklet. On each page appeared the five sentences of the Set and opposite each was a rating scale (7 point) with 1 labelled Low Imagery and 7 labelled High Imagery. Design and Procedure Two independent groups of 15 Ss performed the Imagery Rating task. The Context group;(7 males) received the Context booklets and the No Context group received the No Context booklets as described above. A l l Ss received the same instructions to read. They emphasized that only ease of imagery was to be considered and that a l l of the sentences on a given page were to be read before any ratings were done on that page. This was to ensure that any effect of context 91 would be reflected in the ratings. The study was done in groups of 15 Ss at a time. Comprehensibility Ratings  Subj ects Fifty-four volunteers (30 males) from second and third year Psychology courses at The University of British Columbia served in the experiment. Materials (a) Presentation materials from Experiment I. Context and No Context booklets identical to those used in the Imagery Rating task were compiled. The only difference was that the rating scale opposite each sentence was labelled 1 - hard and 7 -easy, and at the bottom of each of the 24 pages was printed "Whole Set" with a similar rating scale opposite. (b) Semantic and Lexical test sentences from Experiment I. The 48 test sentences (12 Concrete Semantic; 12 Concrete Lexical; 12 Abstract Semantic; 12 Abstract Lexical) from Experiment I were arranged into two types of booklets, A and B. Each booklet contained two pages one with 12 Concrete (6S and 6L) sentences and the other with 12 Abstract (6S and 6L) sentences. In half the booklets the Concretes were f i r s t and in the other half they were on the second page. Each booklet contained either the Semantic or the Lexical test sentence for a given c r i t i c a l sentence. Thus no one booklet contained both versions of a given sentence. There were two random orders of S and L sentences on the page within each type of booklet Opposite each sentence was a 7 point scale with 1 labelled 92 'hard' and 7 labelled 'easy'. Design and Procedure Four independent groups performed the Comprehensibility Rating task. Two groups of 15 Ss received the Presentation materials, one getting the Context Sets and the other the No Context sets. The other two groups of 12 Ss each received the Test sentences, one getting the A booklets and the other the B booklets. The instructions were the same for a l l Ss and were similar in form to the Imagery Rating instructions. It was stressed that only the ease of understanding the material should be considered. Again in order to ensure that context might have i t s maximum effect, a l l of the sentences on a given page were to be read before rating any of the sentences on that page. Subjects were also instructed to rate the whole set on a given page as to how easy i t is to understand as an integrated paragraph. The study was run in groups of 20 to 30 Ss at a time. Similarity of Meaning Ratings  Subj ects Thirdy unpaid volunteers from third year Psychology courses at The University of British Columbia served as Ss in the study. Materials Two types of materials were used. The Context materials contained.the 12 Abstract and 12 Concrete Sets presented to the Context groups in Experiment I, and the No.Context materials 93 contained the Sets given to the No Context groups in that experiment. Each type of material was put in booklet form with one Set per page and the order of the 24 Sets randomized in each booklet. On each page, the fourth sentence i.e., the c r i t i c a l sentence, was underlined and below the set were printed the Semantic and the Lexical test sentences corresponding to that c r i t i c a l test sentence. Opposite each was a seven point scale with 1-different and 7-similar. Within each booklet, half of the Concrete sets had the Semantic test sentence printed f i r s t and the same was true for the Abstract sets. Design and Procedure Two independent groups of 15 !5s each took part in the task. One received the Context material and the other received the No Context material. A l l JSs received the same instructions to read. They were told to be concerned only with the similarity in meaning between the underlined sentence and each of the sentences printed at the bottom of the page. Again, a l l of the sentences on the page were to be read before rating the two sentences. Results Imagery Ratings (a) A l l Sentences The rating given by Each S^  to each of the 120 sentences presented in the Context and the No Context groups was recorded. The mean rating for the 60 Concrete and 60 Abstract sentences was then tabulated for S,' The overall mean imagery ratings for the 2 groups (n = 15) were as follows: Concrete-Context 5.89, Concrete-No Context 5.13, Abstract-Context 2.99 and Abstract-No Context 3.07. 94 A 2 x 2 analysis of variance with Concreteness within and context between, was carried out on the mean rating for each The results shown i n Table XIII revealed a signigicant effect of Concreteness, _F (1,28) = 174.47, p_ <.001, with Concrete sentences receiving higher imagery ratings than Abstract sentences there was also a s i g n i f i c a n t interaction between Concreteness and Context, F_ (1,28) = = 5.06, p_ < .05. Analysis of simple main effects showed that higher ratings were given to Context materials than to No Context sentences when the material was Concrete (F 1,28) = 6.41, £ < .05) but there was no effect of Context for Abstract material. A Scheffe's test showed higher ratings were given to Concrete sentences, than to Abstracts, i n both Context (F 3,56) = 43.36, £ <.01) and No Context (F (3,56) = 21.80, £ <.01) conditions, (b) C r i t i c a l Sentences A sim i l a r analysis was carried out on the mean rating given by Each S^  to the 12 concrete and 12 Abstract Sentences which were tested for retention i n Experiment I. The ove r a l l mean ratings for the two groups (n = 15) were: Concrete-Context 5.49, Concrete-No Context 4.65, Abstract-Context 2.61, Abstract-No Context 2.68. Table XIV presents the results of the 2 x 2 analysis of variance on these mean ratings for each S^. The results were i d e n t i c a l to those obtained when a l l of the sentences were considered. Concrete sentences received higher ratings than Abstract sentences (F_ (1,28) = 152.53, p_ <.001), which was shown by Scheffe's test to hold for both the Context (F 3,56) = 35.31, £ <.01) and the No Context (F (3,56) = 16.68, £ <.01) conditions. The s i g n i f i c a n t interaction between 95 TABLE XIII SUMMARY OF THE ANALYSIS OF VARIANCE ON IMAGERY RATINGS FOR ALL SENTENCES Source df. SS MS Between Ss Context (A) Ss/A 29 1 28 27.64 1.73 25.91 1.73 0.93 1.86 Within Ss Concreteness (B) BXSs BXA BXSs/A 30 1 29 1 28 110.11 92.47 17.64 2.68 14.96 92.47 174.47 2.68 0.53 5.06 <.001 <.05 Total 59 137.75 96 TABLE XIV SUMMARY OF THE ANALYSIS OF VARIANCE OF THE IMAGERY RATINGS FOR CRITICAL SENTENCES Source df SS MS Between Ss Context (A) Ss/A 29 1 28 35.35 2.21 33.14 2.21 1.18 1.87 Within Ss 30 Concreteness (B) 1 BXSs 29 BXA 1 BXSs/A 28 107.71 88.47 19.24 3.05 16.19 88.47 152.53 3.05 0.58 5.25 <.001 <.05 Total 59 143.06 97 Concreteness and Context (F(l,28) = 5.25, p < .05) was shown to be pro-duced by higher ratings given to material presented in the Context group when i t was concrete (F(l,28) = 5.07, p < .05) and no difference between Context and No Context for Abstract sentences. Comprehensibility Ratings (a) A l l sentences. For each the mean rating given to the 60 Concrete and 60 Abstract sentences was tabulated. The overall means for the two groups (n=15) were: Concrete-Context 6.54, Concrete-No Context 6.30?Abstract-Context 6.12, Abstract No Context 5.98. Table XV shows the results of the 2 X 2 analysis of variance carried out on the mean ratings for each J3. The only significant effect was Con-creteness, J?(l,28) = 17.42, p_ < .001, with Concrete sentences receiving higher comprehensibility ratings than Abstract sentences. (b) C r i t i c a l sentences The analysis of mean ratings when only the C r i t i c a l Sentences used in Experiment I were considered produced identical results. The analysis of variance, shown in Table XVI showed that Concrete sentences recieved higher ratings than Abstract sentences, F(l,28) = 8.96, p_ < .01. The overall mean ratings for the two groups (n=15) were: Concrete Context 6.60, Concrete No Context 6.18, Abstract Context 6.10, and Abstract No Context 5.96. (c) Whole Sets For each S^  the mean rating given to the whole set was tabulated for the 12 Concrete and 12 Abstract Sets. The means for the two groups were: 98 TABLE XV SUMMARY OF THE ANALYSIS OF VARIANCE ON COMPREHENSIBILITY RATINGS FOR ALL SENTENCES SOURCE df SS MS Between Ss Context (A) Ss/A 29 1 28 29.09 0.53 28.56 0.53 1.02 Within Ss Concreteness (B) BXSs BXA BXSs/A 30 1 29 1 28 5.09 2.09 3.00 0.01 2.99 2.09 0.01 0.12 17.42 <.01 Total 59 34.18 Note * F < 1. TABLE XVI SUMMARY OF THE ANALYSIS OF VARIANCE ON COMPREHENSIBILITY RATINGS FOR CRITICAL SENTENCES SOURCE df Between Ss 29 Context (A) 1 Ss/A 28 Within Ss 30 Concreteness (B) 1 BXSs 29 BXA 1 BXSs/A 28 Total 59 SS MS F 29.24 1.17 1.17 1.17 28.07 1.00 8.40 1.97 1.97 8.96 6.43 0.28 0.28 1.27 6.15 0.22 37.64 100 Concrete Context 6.12, Concrete No Context 4.46, Abstract Context 5.65, Abstract No Context 4.42. The analysis of variance, summarized in Table XVII, on these Whole Set ratings showed only an effect of Context, with Context Sets receiving higher ratings than No Context Sets, J?(l,28) = 6.72, _p_< .05. (d) Semantic and Lexical Test Sentences. For each ^ the mean rating over the 6 Concrete Semantic, the 6 Concrete. Lexical, the 6 Abstract Semantic and the 6 Abstract Lexical Sentences was recorded. Subjects receiving a_ booklets were grouped with those receiving b_ booklets, and a 2 X 2 analysis of variance, with both Concreteness and Type of Change within Ss, was carried out. Table XVIII shows the results of the analysis of variance. Only Concreteness was significant, with Concrete test sentences receiving higher comprehensibility rating than Abstract sentences, F(l,23) = 10.56, p<.01. The means over the 24 Ss were: Concrete Semantic 5.73, Concrete Lexical 5.85, Abstract Semantic 4.98, Abstract Lexical 5.37. Similarity Ratings For each S_, the mean rating over 12 Concrete-Semantic, 12 Concrete Lexical, 12 Abstract Semantic and 12 Abstract Lexical test items was com-puted. A 2 X 2 X 2 analysis of variance, with Context between Ss, and Concreteness and Type of Change within Ss was carried out. The results, as shown in Table XIX, revealed that the type of test sentence was highly significant (F(l,28) = 655.0, p< .001). Lexical test items were judged 101 TABLE XVII SUMMARY OF THE ANALYSIS OF VARIANCE ON COMPREHENSIBILITY RATINGS FOR WHOLE SETS SOURCE df SS MS Between Ss Context (A) Ss/A 29 1 28 159.95 30.97 128.98 30.97 4.61 6.72 <.05 Wit h i n Ss 30 Concreteness (B) 1 BXSs 29 BXA 1 BXSs/A 28 8.83 0.92 7.91 0.65 7.26 0.92 0.65 0.26 3.54 2.50 <.10 T o t a l 59 168.78 102 TABLE X V I I I SUMMARY OF THE ANALYSIS OF VARIANCE ON COMPREHENSIBILITY RATINGS, FOR.TEST SENTENCES SOURCE df SS MS F p_ Concreteness (A) 1 9.10 9.19 10.56 < .01 Type of Change (B) 1 1.55 1.55 3.61 < .10 A X B 1 0.44 0.44 1.07 Subjects 23 62.55 2.72 A X Ss 23 20.08 0.87 B X Ss 23 9.95 0.43 A X B X Ss 23 9.37 0.41 TOTAL 95 113.13 103 TABLE XIX SUMMARY OF THE ANALYSIS OF VARIANCE ON SIMILARITY RATINGS SOURCE df SS MS Between Ss 29 22.24 Context (A) 1 0.57 0.57 Ss/A 28 21.67 0.77 Within Ss 90 620.60 Concreteness (B) 1 0.00 0.00 Type of Change(C) 1 589.50 589.50 6.55 <.001 B X C 1 1.31 1.31 13.94. <.001 B X Ss _29 1.92 B X A 1 0.00 0.00 * B X Ss/A 28 1.92 0.07 C X Ss 29 25.25 C X A 1 0.07 0.07 * C X Ss/A 28 25.18 0.90 B X C X Ss 29 2.62 B X C X A 1 0.00 0.00 * B X C X Ss/A 28 2.62 0.09 TOTAL 119 642.84 -Note: * F < 1 104 as more similar in meaning to the c r i t i c a l sentences than were semantic test items- The interaction between concreteness and Type of Change was also significant, jF(l,28) = 13.94, p_ < .001). An analysis of simple main effects showed that Concreteness was significant for both Semantic (F(l,29) = 7.44, £ < .01) and Lexical (F(l,29) = 9.05, _p_ < .01) test sentences. For Semantic test sentences, Abstract material was rated as more similar in meaning to the original than concrete sentences. For Lexical test sentences, the reverse was true, with Concrete materials being judged as more similar to the CS in meaning than Abstracts. The mean similarity ratings for the two groups (n=15) are presented in Table XX. Summary of Results For the materials used in Experiment I, the following results were found: Imagery Ratings - Considering either a l l of the sentences or only the c r i t i c a l sentences, the Concrete material received higher Imagery ratings than the Abstract material. This was true for both the Context and No Context presentation. In addition, the Context materials received higher imagery ratings than the No Context materials when they were Concrete, but there was no difference for Abstract material. Comprehensibility Ratings - Again considering either the c r i t i c a l sentences only, a l l of the sentences, or the S and L test sentences, Concrete materials received higher comprehnsibility ratings than abstract material. When the Whole Set ratings were considered, Context Sets received higher comprehensibility TABLE XX MEAN SIMILARITY OF MEANING RATINGS AS A FUNCTION OE CONCRETENESS, TYPE OF CHANGE AND CONTEXT Concrete Abstract Semantic Lexical Semantic Lexical Context 1.68 6.39 1.91 6.16 No Context 1.61 6.18 1.79 5.99 106 ratings than No Context Sets. S i m i l a r i t y of Meaning Ratings - Le x i c a l test sentences were rated as more similar i n meaning to the o r i g i n a l l y presented sentences than were Semantic test sentences. For the Semantic test sentences, Abstract materials were judged as more similar i n meaning to the or i g i n a l s than Concrete materials. For the Lex i c a l test sentences, the reverse was true, with Concrete sentences being judged more similar i n meaning to the presented material than Abstracts. 107 GENERAL DISCUSSION The various hypotheses about the results of Experiment I can now be evaluated. With respect to the superiority of the detection of Semantic changes over detection of Lexical changes, the explanation in terms of judgements being based on comprehensibility differences in these test items was not supported. Comprehensibility of the test sentences varied only as a function of Concreteness and not as a function of the Type of test sentence. Thus i t would appear that the basis of the Type of Change effect is that memory for meaning was better than memory for wording in a l l of the conditions of Experiment I. This result, in the case of Abstract materials presented without Context, is in opposition to the result reported by Begg and Paivio (1969), where the reverse was found. It was suggested that this difference in findings may have been due to either greater comprehensibility or greater changes in meaning produced by Semantic changes in Abstract sentences used in Experiment I, as compared to those used by Begg and Paivio. These suggestions can be evaluated by comparing the results of Experiment II with those reported by Johnson, Bransford, Nyberg and Cleary (1972) which were carried out on the Begg and Paivio sentences. The mean comprehensibility rating for the No Context presentation of Abstract C r i t i c a l Sentences used in Experiment I was 6.03, while that re-ported by Johnson et a l . was 4.41. Thus there i s some support for the suggestion that the difference in results for abstract material between Experiment I and the Begg and Paivio study was due to the higher 108 comprehensibility of these sentences in the present study. There was also support for the interpretation relating to the degree of meaning change in Semantic test items. The ratings obtained by Johnson et a l . were done on a scale from 1-similar to 7-different, while the reverse of these labels were used in Experiment II Johnson et a l . found that the mean rating for the Begg and Paivio Abstract sentences was 4.47, and Ex-periment II reported a value of 1.85 for the No. :Context presentation. This latter value converts to 6.15 on the Johnson et a l . scale, and there-fore i t appears that the Semnatic test sentences in Experiment I changed the meaning of the Abstract sentence more than these test items altered meaning in the Begg and Paivio sentences. These two factors then provide a reasonable explanation of the d i f -ference between the Begg and Paivio study and Experiment I with respect to the relative memory for meaning and wording. This interpretation must remain highly speculative, however, as the two sets of ratings were obtained from different Ss, and under different circumstances. In terms of the Levels of Processing model, the result suggests that a l l of the materials in a l l conditions of Experiment I had some degree of access to the semantic processing level. This i s derived from the assumption that the primary component of a memory representation w i l l be delivered, generally, by the features of the deepest level of processing that the material has access to in a given presentation situation. The interaction between Concreteness and Context in the proportion of semantic hits i s more c r i t i c a l to an evaluation of the Levels of Processing model. The significant effect of Context in eliminating the difference 109 between Concrete and Abstract material i n memory for meaning i s predicted by the model. However, as mentioned e a r l i e r , the result i s open to other interpretations. F i r s t , i f the ease of access to the image code increased i n the Abstract material as a function of Context, then Paivio's two-process theory would predict the obtained interaction. There was no support for t h i s interpreation i n Experiment I I . The Concrete material from Experiment I was rated higher i n image-evoking capacity than the Abstract material i n both the Context and the No Context conditions. Also, Context increased the rated imagability of the Concrete material, but had no effect on the ratings for Abstract sentences. Thus Experiment I would seem to provide evidence against Paivio's theory. A second interpretation for the obtained interaction related to the degree of meaning change i n Semantic test sentences. I f Semantic test items altered the meaning of Concrete material more than Abstract at No Context, but there was no difference at Context, then the interaction would again be predicted. There was also no support for t h i s interpretation found i n Experiment I I . There were no effects of Context on the s i m i l a r i t y of meaning ratings, as Semantic test items altered meaning of Concrete sentences more than Abstract materials at both Context and No Context. Of the interpretations of the data which have been considered then, the Levels of Processing model seems to provide the most adequate account of the interaction between Concreteness and Context that was found i n Experiment I. The model assumes that concrete and abstract material have equal access to l e x i c a l processing, but that access to semantic processing generally increases with concreteness. In addition, however, context i s 110 :assumed to increase the access of Abstract information to the semantic processing level. Thus the model predicts that without context, memory for meaning w i l l be better in the more concrete material, and that this difference w i l l be significantly reduced when context is available. This was precisely the finding for the proportion of semantic hits analysis. Furthermore, the result suggests that the strong form of the prediction concerning the magnitude of the reduction was supported, as the concrete-ness effect was eliminated when context was available. The implication is that context may produce equivalence in the ease of access of concrete and abstract material to the semantic processing level. Unfortunately, the present study did not provide any direct infor-mation about the source of the context effect. Both Pezdak and Royer (1974) and Bransford and Johnson (1972) suggest that contextual information i n -creases the comprehensibility of language material. The comprehensibility ratings obtained in Experiment II, however, did not reveal any effect of context in the individual sentence ratings. Only the.Whole Set ratings were influenced by context with higher ratings being given to the Context Sets than to No Context sets. This finding simply indicates that the regrouping of sentences to form the No Context materials, significantly reduced the integrative nature of the Sets as planned. In light of the previous discussion of comprehension as a multi-level process, the failure to find an effect of context on comprehensibility ratings, does not eliminate the possibility that contextual information may have primarily affected comprehension in the present study. Ratings might provide a relatively poor estimate of how easily sentences can be understood. I l l Indeed, Moeser (1974) equated concrete and abstract sentences on rated comprehensibility, and found s t i l l a significant decrease in speed of  encoding as a function of increasing concreteness. The implication i s , then, that in order to assess the relationship between context and comprehension, a more sensitive measure of ease of understanding (than ratings) must be used. The speed of encoding measure used by Moeser (1974) or the Klee and Eysenck (1973) measure of the speed of categorizing sentences as meaningful or anomolous, would seem to reasonable candidates. Mistler-Lachman's (1972) Context Integration and Production tasks would also be useful here, as they seem to provide more adequate estimates of 'deep' comprehension than tasks which emphasize rapid judgments on isolated sentences. The study, then, leaves open the question of the specific relationship between imagery, context and memory for meaning. The evidence presented here, indicates that, contrary to Paivio's two-process theory, memory for the meaning of language may be independent of ease of access to the image code. This situation occurs when contextual information is provided for the processing of language. As to why i t occurs, one can only speculate. Bransford and Johnson (1972) suggest that semantic context may be a 'pre-requisite for understanding certain language materials. In particular they present the proposal that j5s do not simply interpret and store the meanings of isolated sentences. Instead, they create 'semantic products' which are a combination of the input and prior knowledge. Thus the f u l l understanding of some language material (i.e., the formation of semantic products) may require the availability of prior knowledge. 112 In this framework, the results of the present study may imply that abstract language requires contextual knowledge for understanding, to a greater extent than does concrete language. When the sentences were presented in isolation from semantic context, concrete material produced much better memory for meaning than abstract. This difference was eliminated, however, when the sentences were embedded in a semantic context. The subjective experience of imagery, then, may indicate the spontaneous provision of semantic knowledge, which can aid comprehension and memory for meaning.v.. When imagability of material i s relatively low, such knowledge must be added to the input i n order to gain a substantial degree of semantic processing. The above discussion is of course highly speculative and requires further research. The levels of processing approach which has been developed here, combined with the work of Bransford and his associates (e.g., Bransford and Johnson, 1973), would seem to provide a more adequate framework for this investigation than would Paivio's two-process theory. Imagery has been shown to play a v i t a l role in language processing, but further specification of that role must go beyond the notion of different modes of storage. Hope-fu l l y , the present paper has made a significant contribution towards the research effort in this direction. 113 FOOTNOTE Paivio, Yuille and Madigan (1968) make an operational distinction between concreteness and image-evoking capacity. For the purposes of this discussion, however, the term concrete w i l l refer to material receiving a high rating on either scale, and abstract to material with correspondingly low ratings. 114 BIBLIOGRAPHY Begg, I. Recognition memory for sentence meaning and wording. Journal  of Verbal Learning and Verbal Behavior, 1971, 10, 176-181. Begg, I. Recall of meaningful phrases. Journal of Verbal Learning and  Verbal Behavior, 1972, 11, 431-439. Begg, I., and Paivio, A. Concreteness and imagery i n sentnece meaning. Journal of Verbal Learning and Verbal Behavior, 1969, j3, 821-827. Begg, I., and Robertston, Imagery and long-term retention. Journal of Verbal  Learning and Verbal Behavior, 1973, 12_, 689-700. Bransford, J. D., Barclay, J. R., and Franks, J. J. Sentence memory: a constructive versus interpretive approach. Cognitive Psychology, 1972, 3, 193-209. Bransford, J. D., and Johnson, M. K. Contextual preprequisites for under-standing: some investigations of comprehension and r e c a l l , Journal  of Verbal Learing and Verbal Behavior, 1972, 11, 717-726. Bransford, J. D., and Franks, J. J. The abstraction of l i n g u i s t i c ideas. Cognitive Psychology, 1971, _2, 331-351. Bransford, J. D., and Johnson, M. K. Considerations of some problems of comprehension. In W. G. Chase (Ed.), Visual Information Processing. New York: Academic Press, 1973. Pp. 383-462. Bransford, J. D., Johnson, M. K., and Raye, C. Unpublished experiment. Re-ported i n W.G. Chase (Ed), Visual Information Processing. New York: Acadmeic Press, 1973. Pp. 383-462. Chomsky, N. Syntactic Structures. The Hague: Mouton, 1957. Dooling, D. J . , and Lachman, R. Effects of comprehension on retention of prose. Journal of Experimental Psychology, 1971, j88, 216-222. Franks, J. J . , and Bransford, J. D. The acquisition of abstract ideas. Journal of Verbal Learning and Verbal Behavior, 1972, 11, 311-315. Goldfarb, C , Wirtz, J . , and Anis f e l d , M. Abstract and concrete phrases i n f a l s e recognition. Journal of Experimental Psychology, 1973, j)8, 25-30. Johnson, M. K., Bransford, J. D., Nyberg, S. E., and Cleary, J. J. Comprehension factors i n interpreting memory for abstract and concrete sentences. Journal of Verbal Learning and Verbal Behavior, 1972, 11, 451-454. Johnson, M. K., Bransford, J. D., and Solomon, S. K. Memory f o r t a c i t implications of sentences. Journal of Experimental Psychology, 1973, 98, 203-205. 115 Katz, S. Procedural factors in the abstraction of linguistic ideas. Journal of Experimental Psychology, 1973, 98_, 79-84. Kintsch, W. Notes on the structure of semantic memory. In E. Tulving and W. Donaldson (Eds.), Organization of Memory, New York: Academic Press, 1972. Klee, H., and Eysenck, M. W. Comprehension of concrete and abstract sentences. Journal of Verbal Learning and Verbal Behaviour, 1973, 12, 522-529. Lachman, R., and Dooling, D. J. Connected discourse and random strings: Effects of number of inputs on recognition and r e c a l l . Journal of  Experimental Psychology, 1968, _77, 517-522. Mistler-Lachman, J. L. Levels of comprehension in processing normal and ambiguous sentences. Journal of Verbal Learning and Verbal Behavior, 1972, 11, 614-623. Moeser, S. D. Memory for meaning and wording in concrete and abstract sentences. Journal of Verbal Learning and Verbal Behavior, 1974, 13, 682-697. Offir, C. E. Recognition memory for presuppositions of relative clause sentences. Journal of Verbal Learning and Verbal Behavior, 1973, 12, 636-643. Paivio, A. Imagery and Verbal Processes. New York: Holt, Rinehart, and Winston, 1971. Paivio, A.., and Begg, I. Imagery and comprehensibility latencies as a function of sentence concreteness and structure. Perception and  Psychophysics, 1971, 10, 408-412 Paivio, A., Yuille, J. C. and Madigan, S. A. Concreteness, imagery and meaningfulness values for 925 nouns. Journal of Experimental Psychology  Monograph Supplement, 1968, _76, No. 1, Part 2. Pezdek, K., and Royer, J. M. The role of comprehension in learning concrete and abstract sentences. Journal of Verbal Learning and Verbal Behavior, 1974, 13, 551-558.' Philipchalk, R. P. Thematicity, abstractness, and the long erm recall of connected discourse. Psychonomic Science, 1972, 2_7, 361-362. Pompi, K. F., and Lachman, R. Surrogate processes in the short term re-tention of connected discourse. Journal of Experimental Psychology, 1967, 75, 143-150. 116 Sachs, J. S. Recognition memory for syntactic and semantic aspects of connected discourse. Perception and Psychophysics, 1967, 2_, 437-442. (a). Sachs, J. S. Recognition of semantic, syntactic, and lexical changes in sentences. Paper presented at Psychonomic Society Meetings, Chicago, 111., October, 1967 (b). Singer, M., and Rosenberg, S. T. The role of grammatical relations in the abstraction of linguistic ideas. Journal of Verbal Learning  and Verbal Behavior, 1973, 12, 273-284. Yuille, J. C. and Paivio, A. Abstractness and re c a l l of connected discourse. Journal of Experimental Psychology, 1969, 82, 467-471. APPENDIX 117 PRESENTATION MATERIALS CONCRETE - CONTEXT (i) The walls of the elegant ballroom were decorated with hundreds of large paintings. ( i i ) Rich merchants from a l l over the country strolled around the room and nodded with approval, ( i i i ) In his sunny studio over the gallery, the young painter talked happily of his good fortune, (iv) The cheerful a r t i s t entertained a lonely damsel, (v) The gi r l ' s simple clothes were a sharp contrast to the superbly embroidered sofa on which she sat. CONCRETE - NO CONTEXT (i) The mariner hurriedly raised his anchor and set s a i l for land, ( i i ) Suddenly, the man released the heavy iron chain and pointed to a thicket of trees, ( i i i ) The women and children huddled together, while the men whispered nervously to one another, (iv) The cheerful a r t i s t entertained a lonely damsel, (v) On that special day, the king was dressed in his finest red si l k s , and was adorned with a diamond-studded crown. 118 ABSTRACT - CONTEXT (i) The kidnapping case received much attention because of the mysterious nature of the charges, (ii ) An obscure religious sect was accused of holding converts against their w i l l and forcing them to perform ancient rituals, ( i i i ) The court of appeals delivered a guilty verdict and recommended l i f e imprisonment for the founders of the movement, (iv) The unexpected conviction reversed a prior decision, (v) The original t r i a l had produced an aquittal on the grounds of insufficient evidence. ABSTRACT - NO CONTEXT (i) For almost a year the dispute continued and hope of any resolution dwindled, ( i i ) The tiny monarchy faced a serious energy c r i s i s due to an overconsumption of domestic reserves, ( i i i ) Products containing the chemical were allowed to be sold again provided they carried an explanation of these results, (iv) The unexpected conviction reversed a prior decision, (v) Many humanitarian causes had been aided by the association and i t s reputation was thought to be above reproach. 1 1 9 TEST MATERIALS CONCRETE. C r i t i c a l Sentence: The cheerful Semantic Change: The cheerful Lexical Change: The cheerful a r t i s t entertained a lonely damsel, damsel entertained a lonely a r t i s t , a r t i s t entertained a lonely maiden. ABSTRACT. C r i t i c a l Sentence: Semantic Change: Lexical Change: The unexpected conviction reversed a p r i o r decision. The unexpected decision reversed a prior conviction. The unexpected conviction reversed a p r i o r judgment. 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0093469/manifest

Comment

Related Items