AN EVALUATION OF THE APPLICABILITY OF ADDITIVE FACTOR MODELS TO SENTENCE-SENTENCE VERIFICATION TASKS Thesis for the Degree of M. A. MICHIGAN STATE UNIVERSITY ELIZABETH ANNE MAIER. 1977 rummagnmwfl x. ‘.7, . j T “ 2‘- 4 9 ' £- 9" ' l .' ' A, u ‘ . L. ‘ ..’t _'. _' ‘j " '1'. a ‘ , ABSTRACT AN EVALUATION OF THE APPLICABILITY OF ADDITIVE FACTOR MODELS TO SENTENCE-SENTENCE VERIFICATION TASKS By Elizabeth Anne Maier Three experiments were conducted investigating rules of processing in sentence-sentence verification tasks. In each experiment, the content of the sentences referred either to the midpoint of the dimen- sion of time, that is, On Time, or the endpoint, Late. Experiment I examined subjects' verification of both explicitly conveyed information (Assertions) and implicit information (Inferences) over three testing sessions using a within-subjects design. Subjects heard a set of two sentences. The task required them to indicate whether the two sentences conveyed the same meaning responding as quickly and as accurately as possible. In general, Inference conditions, conditions containing an On Time or a negative in Sentence 2, or in which an inner string mismatch occurred, were most difficult for the subjects to process. A special analysis suggested that the effects of On Time and the Inference con- dition were restricted to the encoding phase of processing, whereas the effects of negation were seen in both encoding and comparison opera- tions. No evidence was obtained to support the idea that in more com- plex situations subjects adopt conversion operations in processing. Elizabeth Anne Maier Experiment 2 investigated subjects' verification of explicitly relayed information (Assertions) in the auditory modality. As in Exper- iment 1, conditions in which an inner string mismatch occurred, or in which an 0n Time2 or Negative appeared were difficult for the subjects 2 to process. The role of Sentence 1 was more important in this experi- ment, as indicated by the Term 1 main effect and the Term 1 x Negation 1 x Term 2 interaction. The latter result suggested that when a negative occurred in Store 1, the subjects encoded both affirmative and negative representations of the sentence. Which representation was chosen in comparing Sentence 1 and Sentence 2 was dependent upon what had been encoded in Sentence 2. Experiment 3 was designed to determine whether differences in pro- cessing occur between the auditory and visual modalities. As in Ex- periment 2, only the verification of Assertions was examined. On Timez, Negativez, and inner string mismatch conditions were more time consuming for the subjects to process than their counterparts. The role of Sen- tence l was more pronounced in this study in that both the Term 1 and Negation 1 main effects were significant. RT orderings suggested that after the subjects encoded Sentences 1 and 2, they went back and re- encoded Sentence 1 before commencing comparison operations. The results of all three experiments were interpreted in terms of additive factor models. The three models constructed adequately repre— sented the obtained results in each case. Although some problems arose within the encoding stages of processing in the model of Experiment 2, it was the conclusion of the study that additive factor models which make distinctions between the encoding and comparison stages of proces- sing are appropriate frameworks in which to interpret the results of sentence-sentence verification tasks. AN EVALUATION OF THE APPLICABILITY OF ADDITIVE FACTOR MODELS TO SENTENCE-SENTENCE VERIFICATION TASKS By Elizabeth Anne Maier A THESIS Submitted to Michigan State University in partial fulfillment of the requirements for the degree of MASTER OF ARTS Department of Psychology 1977 ii ACKNOWLEDGMENTS I would like to thank Rose Zacks for her concern, guidance and assistance; and Lauren Harris, Gordon WOod, and Gary Olson for their thoughtful comments on this thesis. iii TABLE OF CONTENTS LIST OF TABLES O O O O O O O O O O O O O O O O O I O O O O O O O O O O O O O O O O O O O O O O O O I O O O 0 LIST OF FIGURES O O I I I O O O O O O O O O O O O O O O O O O O O O O O O O O I O O O O O O O O O O O O O O CMTER 1 - IntIOduction O O O O O O O O O O O O O O O O O O I O O O O O O O O O O O I O O O O O 0 Pilot Study Results Discussion The Model Experimental Extensions Summary CHAPTERz-Experiment 1 O00.0.0000...OOOOOOOOOOOOOOOOOOOOOOOO Method Results and Discussion Summary CEAPTER3-Experiment2 OOOOIOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO Method Results and Discussion Summary CHAPTERa-Experiment3000OOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOOO Method Results and Discussion Summary CHAPTERS-General DiSCUSSion OOOOOOOOOOOOOOOOOOOOOO0.0.0.... The Psychological Reality of Processing Variables Usefulness of the Additive Factor Model Approach Other Issues APPENDIX A.- Mean RT in Msec. Per Condition on Day 1 - Experiment 1 00......0....OOOOOOOOOOOOOOOOOOOOOI. APPENDIX B - Mean RT in Msec. Per Condition on Day 2 - Experiment 1 O0....O..0.0...OOOOOOOOOOOOOOOOOOOOO iv vi viii 24 50 69 89 96 97 APPENDIX C APPENDIX D — APPENDIX E APPENDIX F APPENDIX G APPENDIX H APPENDIX I BIBLIOGRAPHY Significant (p_< .05) Results of the 2x2x2x2x2 Within-Subjects ANOVA on Days 1 and 2 in Experiment 1 OOOCCOQOCOOOCOOCCOOOCOOOCOOOOOOOOOOO Mean RT in Msec. Per Condition on Day 1 - Experimentz I.OIOOCOOOCOOOOIOCCOOOO'COOOQOOOOOCO Mean RT in Msec. Per Condition on Day 2 - Experimentz OOOOOOOOOOOOOIOOOOOOOOO'COOOOOOOOOO. Significant (p_< .05) Results of the 2x2x2x2 Within-Subjects ANOVA on Days 1 and 2 in EXPerimen-tz 0.0..0..O...OOOOOOOOOOOOOOOOOOOOOOO. Mean RT in Msec. Per Condition on Day 1 — Experiment3 .00....0...OOOOOOOOOOOOOOOOOOOOOOOOO Mean RT in Msec. Per Condition on Day 2 - Experiment3.00000000000000000000000.0.0.0000... Significant (p_< .05) Results of the 2x2x2x2 Within-Subjects ANOVA on Days 1 and 2 in Experiment3 0....00......COOOOOOOOOOOOOOOOOOO... 98 99 100 101 102 103 104 105 Table 1.1 1.2 1.3 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.1 3.2 LIST OF TABLES Examples of Experimental Conditions Presented in the Pilot Study O...O000......O......OOOOOOOOOOOOOOOOOO Mean RT in Msec. Per Condition, Estimated and Observed .00....0.000.00.0000.........OOOCOCOOCOOOOOOOO Condition Predictions According to the True and Conversion Models of Negation (Clark, 1970; Clark&Chase’ 1972) ......OOOOOOO......OOOOOOOOOOOOOOO Conditions of Experiment 1 ............................ Significant Main Effects and Interactions of the Analyses of Experiment 1 .............................. MEan RT in Msec. for the Condition x Negation 1 Interaction OO......0.00.0.00.........OOOOOOOOOOOOOO0.0 Mean RT in Maec. for the Condition x Negation 2 Interaction OO................OOOOOOOOOOOCOOO00.0.00... Mean RT in Msec. for the Negation 1 x Negation 2 Interaction .....OOOOOOOOOOOOOOO......OOOOCOOOOOOOOOOOO Mean RT in Msec. for the Term 1 x Term 2 x Negation 2 Interaction .0...O......OOOOOOO......OOOOOOOCOOIOOOCOOO Pilot Model Predictions and Obtained Values in Msec. for Selected Conditions of Days 1 and 3 ............... Predictions of the Clark and Chase Mbdel and Obtained Values in Msec. for Selected Conditions Of Days land3 OO.I0.0....0.........OOOOOOOOOOOOOOOOOO Mean RT in Msec. Per Condition, Estimated, Observed and Deviations for Experiment l-Day 3 Data ... Conditions of Experiments 2 and 3 ..................... Significant Results of the Analyses of Variance in Experimentz I.I.O0.0.0.000.........IOOCOOOOOOOOOOIO vi 10 11 26 29 30 30 30 32 36 39 47 51 53 Table 3.3 3.4 3.5 3.6 3.7 3.8 4.1 4.2 4.3 4.4 4.5 4.6 Mean RT in Msec. to Verify Inner String Matches and MismatChes O.I.0..........OOOOOOIOOOOOOOOO0.0.0.... Mean RT in Msec. to Verify Inner String Matches and Mismtct‘es over Days 0.00.00.00.000.000.000.0000... The Effect of Negation on RT in Msec. in On Time2 and Latez conditions 0......O........OOOOOOOOOOOOCOOOOO The Effect of Negation on RT in Msec. to Conditions Containing Inner String Matches and Mismatches ........ Pilot Model Predictions and Obtained Values in Msec. for Selected Conditions of Day 3 ...................... Mean RT in Msec. Per Condition, Predicted Scores, and Their Deviations for Day 3 Performance ............ Significant Results of Analyses of Experiment 3 ....... Mean RT in Msec. to Inner String Identities and MismatChes ....0.0.00.0...I0............OIOOOOOOOOOOOO. Effect of a Negative on RT in Msec. to Inner String Match and Mismatch Conditions .................. Effect of the Level of Term 1 on the Verification of Outer String Matches and Mismatches in Maec. ....... Predictions of the Young and Chase (1971) Conversion Model and the Actual Results of ExperimentB 00.0.0.0...00............OOOOOOOOOOOOOOOOO Mean RT in Msec. Per Condition, Predicted Scores and Their Deviation for Day 3 Performance ............. vii 54 56 56 60 66 73 75 75 77 79 86 Figure 1.1 2.1 2.2 3.1 4.1 LIST OF FIGURES Process Model for Sentence-Sentence Verification ..... Mean RT in Msec. to Term 2 as a function of Term 1 ... Processing Alternatives of the Model of Experiment 1 ......OOOOOOOO0.0......OOOOOOOOOOOOCOOOOO Processing Alternatives of the Model of Experimentz ......OOOOOO......OOOIOOOOOOCOOOOOO...... Processing Alternatives of the Model of Experiment3......OOOOOOOOO0............OOOOOOOOOOOOO viii 46 62 83 CHAPTER 1 Introduction Early investigations of negation (Cough, 1965, 1966; Slobin, 1966; Wason, 1959, 1961; wason & Jones, 1963) uniformly reported that negative sentences took longer to comprehend than affirmative sentences whether the sentences were true or not. More recent investigations of the com- prehension of negation have had as their aim the discovery of specific underlying components of true—false verification processes, components which would adequately represent and accurately predict information processing (Carpenter & Just, 1975; Chase 8 Clark, 1972; Clark, 1970, 1974; Clark & Chase, 1972; Trabasso, Rollins & Shaughnessy, 1971). In such experiments the tasks are generally quite simple: subjects are shown sentence-picture stimuli simultaneously, and are instructed to specify by pressing a true or false button as quickly as possible whether the illustration correctly represents the sentence. From such a task, Clark and Chase (1972) have constructed a four-stage model of negation characterizing sentence-picture comprehension based upon Sternberg's (1969a,b) additive factor approach to mental operations. The four-stage sentence-first model assumes that during Stage 1 the subject will encode sentences in a linguistic representation which pre- serves all meaningful aspects of the original sentence. For example, sentence (1) is presumed to be encoded as representation (2). The 2 (1) The star is above the plus. (2) (star above plus) internal representation of an affirmative is assumed to involve a con— stant amount of processing time. Its negative counterpart, sentence (3), would be represented as (4). The actual representation of the negative (3) The star is not above the plus. (4) false (star above plus) as a falsehood indicator is adopted from wason's (1965) proposition that the negative naturally functions to deny a proposition, as in (5). Based upon previous studies of negation comprehension diffi- (5) It is false that the star is above the plus. culties (Cough, 1965, 1966; Slobin, 1966; wason, 1959, 1961), Clark and Chase (1972) posited that the encoding of a negative would be a more difficult operation that the encoding of an affirmative, re- quiring an additional amount of processing time, time 2, Just as time measurement is sensitive to variations in encoding time between affirmatives and full explicit negatives (Clark, 1970, 1974), Clark and Chase (1972) further hypothesized that less explicit forms of negation (Clark, 1969, 1970, 1974) also would increase encoding time. In the Clark and Chase (1972) paradigm, bglgw_was considered to be less neutral than its contrary above, The encoding of sentence (6) as (7) was predicted to exceed that of sentence (1) as (2) by the (6) The star is below the plus. (7) (star below plus) constant §_amount of time. Similarly, since the model proposed by Clark and Chase (1972) was an additive factor model, the encoding of sentence (8) as (9) was expected to exceed (1) encoding by (§'+ 2) time. 3 (8) The star is not below the plus. (9) false (star below plus) During Stage 2, the subject focuses his attention on the picture stimulus, presumably encoding its meaning in a representation like that used in Stage 1 sentence encoding. Clark and Chase's (1972) principle of congruence asserts that for any comparison of the meanings of sen- tence and picture to be made, sentence and picture must be encoded in similar formats. A final assumption underlying the second stage of processing is that the encoded picture representation employs the same proposition appearing in the previous sentence. Thus, if sentence (8) appeared with picture stimulus (10), (10) would be encoded as (11); (10) + (11) (plus below star) while the same picture (10) would be encoded as (12) if paired with (12) (star above plus) sentence (6). An interesting feature of Stage 2 is that a picture is 23325 assumed to be encoded in a negative format. Thus, the possibili- ty of two explicit full negatives co-existing in the same comparison operation is consistently averted. Finally, the Clark and Chase model makes no prediction of differ- ential time encoding of pictures. It assumes that picture encoding will take a constant amount of time regardless of the proposition; picture encoding time was a component of the base processing time, 30' The major processing to be accomplished during the verification task occurs during Stage 3, the comparison stage. Assuming common representations of picture and sentence encodings, the model proposes that the processor will first examine the innermost propositional strings 4 seeking identity and subsequently check the identity of the outer strings. The purpose of Stage 3 is to determine the validity of the sentence picture pair. Hence, for each mismatch the subject will convert the value of the truth index of the pair to its polar opposite. The model assumes that the value of the truth index is set at true at the onset of processing. A single mismatch, of either inner or outer strings, would convert the truth index from true to false, whereas concurrent inner and outer string mismatches would result in two reversals of the truth index value, yielding a true response. Each change in the value is expected to be accompanied by an increase in processing time, times g_and g, for inner and outer string mismatches, respectively. The outcome of Stage 3 is a true or false response input to Stage 4, in which the subject executes the response by pressing the appro- priate response button. Clark and Chase did not elaborate further on this stage. It is assumed, however, that response execution time (time taken to mobilize the hand and press the appropriate button) adds a constant amount of time to the base reaction time (RT), 50. Experimental manipulations by Clark and Chase (1972) using 22222. and bglgg_provided strong support for the model. Parameters were es- timated by means of a least squares analysis to be: Below time, a, 93 msec., negation time, (b_+-g), 685 msec., and falsification time, 2, 187 msec. Collectively, the five parameters accounted for 99.8% of the total between conditions variance, and each parameter proved to be highly significant (p < .05). No interactions were statistically sig- nificant, lending further credence to the independence of stages notion. 5 Clark (1970, 1974; Clark & Chase, 1972) referred to this conceptu- alization as the "true" model of negation, being "true" is so far as the linguistic representations of sentence and picture were assumed to pre- serve all semantic and syntactic features of the actual stimuli. Another model of negation has been proposed (Clark, 1970), the conversion model, accounting for those situations in which the negative is converted to its affirmative equivalent prior to the onset of Stage 3 comparison operations (Trabasso et a1., 1971; wason & Jones, 1963; Young & Chase, 1971). Clark (1970, 1974) felt that the conversion model was subor- dinate to the true model conceptualization, given that by the conversion of a proposition, some of the information contained in the original representation was forfeited. Furthermore, in order for conversion operations to be executed, either the use of contradictory content words or an experimental design strictly defining binary operations is required. For example, the conversion of sentence (13) to (14) is true (13) The plus is not above the star. (14) The plus is below the star. only of a particular spatial configuration (10). Conversion becomes possible, however, when the subject is informed that only binary situ- ations will be under consideration. Use of a conversion strategy was found (Clark, 1970) to yield a pattern of RT's similar to the sentence construction subjects of wason (1961): True Affirmatives (TA)‘< False Affirmatives (FA)‘< True Negatives (IN)‘< False Negatives (FN). Both the true and converstion models of negation have proven to be powerful instruments in the measurement of psychological processes. The former model, using five parameters, accounted for 99.8% of the total between conditions variance (Clark & Chase, 1972); the latter, using four 6 parameters t , g, g, and k, a conversion parameter) was responsible for 952 of the between conditions variance (Clark, 1974). Clark (1974) also successfully interpreted earlier studies of negation (Gough, 1965, 1966; Green, 1970a,b; Slobin, 1966; Trabasso et a1., 1971; wason, 1959, 1961; wason & Jones, 1963) according to either the true or conversion models. Thorough examination and evaluation of the true and conversion models of negation reveal serious experimental shortcomings, however. Simplistic tasks have been used to demonstrate principles of understanding which are theoretically applicable to any information processing task. With the exception of Just and Clark's (1972) work with presuppositions and implications, validation of this model in more complex situations is lacking. Furthermore, the restriction to simplistic paradigms overlooks cases in which negatives occur in both stimulus inputs (i.e., Stage 1, Stage 2, or Stages 1 and 2). Neither model would predict a significant effect on processing in the case of the double negative, but no investigation has tested this premise. Finally, virtually all studies have been in- vestigations of negation using solely the visual modality. Serious inattention to the auditory modality has resulted, although most ver- bal information is heard rather than seen (e.g., read). Processing limitations may exist within the auditory modality prompting the sub- ject to use a different comprehension strategy from those used for visually presented information. In an attempt to resolve some of these issues, a study was con- ducted of the effects of negation (Stage 1, Stage 2, and Stages 1 and 2) on the verification of Assertions and Inferences. Assertions were 7 defined as simple declarative statements about an individual's activity on a particular day; Inferences referred to the individual's usual pattern of behavior, derived from information in the sentence. Sen- tences were restricted to the dimension of time (On Time vs. Late). Pilot Study Subjects. Twenty-five male and female undergraduates participated as a requirement in an introductory course at Michigan State University. Stimuli. TWO hundred fifty-six sentences, recorded by a female speaker, were presented in two sentence test sequences. Sentence 1 con- structed a context for an individual's activity, ending with a clause indicating what the particular actor did today, i.e., whether he was On Time or Not On Time. Sentence 2 was a simple statement containing the proposition Late or On Time; it also varied as to whether is was an As- sertion or an Inference, and whether it was a negative or an affirmative sentence. Half the sentences represented each factor. The subjects re- ceived eight examples of each of the 16 possible conditions (see Table 1.1 for a complete list of the possible sentence combinations). Apparatus. Stimuli, recorded on a Teac stereo tape recorder, were presented through stereo headphones. Concurrent with the cessation of speech a click (inaudible to the subject) automatically started Hunter times recording subjects's RT's. A three button panel was located directly in front of the subject: a central button for comprehension responses, one button for true responses, the other for false. De- pression of any button stopped the Hunter timers and triggered the pre- sentation of the next stimulus. Procedure. The subjects were tested individually. Each was in- tructed to listen carefully to the stimulus presentations. Following 8 TABLE 1. 1 Examples of Experimental Conditions Presented in the Pilot Study 81: Mr. Jones rides the train downtown to work everyday; surprisingly, he was . . . . . . gg_time today. . . . not 22_time today. 82: A. Mr. Jones was 22_time today. B. Mr. Jones was la£g_today. C. Mr. Jones was g££_gg_timg_today. D. Mr. Jones was Eg£_la£g_today. E. Mr. Jones is usually gn_£img, F. Mr. Jones is usually late. G. Mr. Jones is not usually gg_£img, H. Mr. Jones is not usually late. Sentence 1, they were required to depress a button marked "C" (Come prehension) as quickly as possible indicating that they had understood the sentence. Sentence 2 immediately followed. A button marked "T" (True) or "F" (False) was to be depressed indicating the truth value of the match. The position of the T and F buttons was randomly varied between subjects. Both accuracy and speed were emphasized in decision making. The intertrial interval between each test comparison was five seconds. The subjects received eight practice trials before the exper- iment began. If more than two errors occurred, the subjects were in- formed of their inaccuracy in an attempt to improve their performance. At the completion of 128 sentence comparison trials, the subjects were asked to describe what strategies, if any, they had used during the experiment. The experimenter recorded all of the subjects' comprehension and verification RT's. Results Mean RT data per treatment for 25 subjects is contained in Table 1.2. Four factors were manipulated in the experiment: Sentence 1 Nega- tion (presence vs. absence), Sentence 2 Negation (presence vs. absenCe), Sentence 2 Term (Late vs. On Time), and Sentence 2 Statement (Asser- tion vs. Inference). A 2 x 2 x 2 x 2 within subjects analysis of variance was performed on mean RT data. All main effects as well as all two-way interactions involving the Stage 2 Term factor were significant, where p_‘<.05. In general, therefore, comparisons involving Late were more time consuming than On Time verifications; likewise, negatives as well as Inferences involved more processing time than affirmatives and Asser- tions, respectively. The overall error rate was 8.25%. Discussion If the true model of negation were applied to the present paradigm, the RT's (adding only an i parameter for the encoding of the Inference in Stage 2) represented by the parameter components in Table 1.3a would be predicted to occur. Similar extension of the conversion model would produce the parameter groupings appearing in Table 1.3b. Comparison of these predictions with the findings in Table 1.2 shows that neither model can account for the subjects' processing. While it is difficult to conceptualize all treatments in terms of True and False Affirmatives and True and False Negatives because of the presence of the false double negative condition, we still can exa— mine these features in those conditions most comparable to the original 10 TABLE 1.2 Mean RT in Msec. Per Condition, Estimated and Observed Conditions Projected Parameters Estimated Observed On Time- tO 284 310 On Time On Time- to+c 653 626 Late on Time- to+b+d 488 477 Not On Time- On Time- to+b+c+d 857 900 Not Late Not On Time- to+d 428 373 On Time Not On Time- t +k 597 652 0 Late Not On Time- to+b 345 329 Not On Time Not On Time- to+b+2k+c 1338 1321 Not Late Not Usually On Time- t +i+d 608 585 Usually On Time 0 Not Usually On Time- to+i+n 776 678 Usually Late Not Usually On Time- to+i+n 821 796 Not Usually On Time Not Usually On Time- to+i+n+k+d 1277 1261 Not Usually Late Usually On Time- to+i 464 529 Usually On Time Usually On Time- to+i+k+d 920 976 Usually Late Usually On Time- to+i+n+d 965 969 Not Usually On Time Usually On Time to+i+n+k 1133 1170 Not Usually Late 11 TABLE 1.3 Condition Predictions According to the True and Conversion Models of Negation (Clark, 1970; Clark & Chase, 1972) Conditions True (3a) Conversion (3b) On Time- to t On Time ° On Time- to+c to+c Late On Time- to+b+d to+k+c Not On Time On Time- to+b+c+d t +1: Not Late 0 Not On Time- t0+d to+k+c On Time Not On Time- to+c+d t +k Late 0 Not On Time- to+b to+2k Not On Time Not On Time- to+b+c t +2k+c Not Late 0 Not Usually On Time- to+i+d t +i+k+c Usually On Time ° Not Usually On Time- to+i+c+d t +i+k Usually Late 0 Not Usually On Time- to+i+b t +i+2k Not Usually On Time ° Not Usually On Time- to+i+b+c to+i+2k+c Not Usually Late Usually On Time- to+i t +1 Usually On Time 0 Usually On Time- t +i+c t +i+c Usually Late 0 0 Usually On Time- to+i+b+d t +i+k+c Not Usually On Time 0 Usually On Time- to+i+b+c+d to+i+k Not Usually Late 12 Clark and Chase (1972) paradigm: On Timel-On Timez; On Timel-Latez; On Time -Not On Time ~Not Latez. Recall that corresponding 1 2 1 to the true model of negation, TA < FA < FN < TN, or that On Timel-On < - - - Time2 On Time1 Late2 < 0n Time1 Not On Time2 < On Time1 Not Latez. The conversion model predicts the inversion of False and True Negatives. ; and On Time Examination of Table 1.2 again reveals that neither model accounts for the results found that TA.< FN < FA.< TN. This pattern of results was also found in the comparable more complex Inference conditions in which Usually On Time -Usually On Time < Usually On Time -Not Usually On Time 1 < Usually On Time 2 < Usually On Time 1 -Not Usually Late 2 -Usually Late 1 2 1 2' Furthermore, neither model would predict the false double negative Asser- tion as the most difficult condition, which, in fact, was the finding. The inversion of FN and FA was a surprising result. It appears that inner string mismatches were more difficult to resolve than outer string mismatches, contrary to earlier findings (Clark, 1970, 1974; Clark 6 Chase, 1972). This could result, in part, from the higher order processing required by the task (the verification of Assertions and In- ferences) or from the different demands required of the subject in an auditory task. The interactions yielded by the 2 x 2 x 2 x 2 within-subject anal- ysis of variance were also unexpected. Additive factor models, as out- lined by Sternberg (1969a,b), should produce no interactions, as each stage of processing is assumed to be independent of all others. Hence, any model generated using the factors suggested by this design would be an inaccurate reflection of subjects' processing. The findings suggest that an additive factor approach is inappropriate in the present paradigm. 13 Certain aspects of the data indicated, however, that subjects were operating according to some type of routine. For example, in less com? plicated conditions, why should a FN treatment consistently be easier than a FA? Of greater interest is the fact that neither model is able to account for the extreme difficulty of the false double negative Asser- tion condition. Furthermore, no differences in mean RT patterning were found between subjects reporting using some sort of strategy and those not. Whatever subjects were doing, they all seemed to be acting in the same manner. In an attempt to explain these issues and characterize the results of the pilot study according to a stage theory of information processing, it was hypothesized that the data reflect a mixture of the true and con- version models. A four-stage, seven-parameter additive factor model, abstracting features from both accounts of negation has been projected to account for the subjects' performance (see Table 1.2 for treatment parameters and estimates; Figure 1.1 for a representation of the model itself). millage; It is proposed that in Stage 1, Sentence 1 is encoded in some pro- positional format, similar to that suggested by Clark (1970; Clark & Chase, 1972). Encoding is expected to take a fixed amount of time; how- ever, since this time was recorded separately, and since an underlying premise of any additive factor model is independence of stages, Stage 1 is not expected to interfere with future processing or encoding times. It became clear to subjects shortly after the onset of testing that they would be required to verify information directly encoded in Stage 1, or an abstraction (Inference). Thus, a further assumption of Stage 1 14 madam vaonmou and away sauna nanny 0:u no 05Hw> omawsu 117 madam vacuum“ coauoowufinm> monouaomloonuudmm you Hung: oouooum .~.~ shaman ouamoamo mug ou Honda nuauu mo ovao> mwawnu omnwuuo Hausa many a, ...... 1| Am~+ no awry anon o>aumsuammo ou muuao o>wuawoa may uuo>aoo no» madam upmaoou was AVITV unmva.“ IO . sunny «nu mo 05Hw> «mango mmnwuun umu:o sue ououm ca -oaouomnH no no He ououm ca m>wuowmc a muonu mH On» «myv o=Hw> ouwoonao muw ou nova“ nusuu on» no u=Ho> on» uuo>noo .fl u wanna mH _F+ .ooaouomom can o>auuwod a HH a+ .aoauuomma new o>wuuwoa o «H «+ .ooaouomnH on ma “N ovooam — H ovooam 4‘ 15 is that upon encoding Sentence 1, the comparable Inference will also be encoded. During Stage 2, the processor encodes Sentence 2. As in the Clark and Chase (1972) model, the encoding of a negative is expected to involve a given amount of time longer than an affirmative. It was apparent from the data, however, that the encoding of a negative in the presence of an Inference took considerably longer than in the presence of an Assertion. This was probably because of differences in their scope (Clark, 1970, 1974; Klima, 1964). Consider the following statements (15a,b). The latter negates a larger, more complex proposition which (15) a. Mr. Jones was not 22_time today. false (on time) b. Mr. Jones is not usually 22_time. false (usually (on time)) should in fact require more time. It was thus proposed that parameter b_represents the additional time needed to encode a negative in an Assertion, and E_be the negation encoding time in an Inference. The nature of Sentence 2 also varied as to whether it was an Asser- tion or an Inference. Since an Assertion was assumed to be cognitively less complex than an Inference, dealing with information directly re- layed in Sentence 1, the encoding of an Inference was predicted to re— quire additional processing time, time i. In Stage 2, Clark and Chase (1972) hypothesized that EElEELwOUld take a time longer to encode than.§bggg, This extra time was inter- preted as being the result of the inherent negativity of 2213! along the vertical dimension. Implicit negation or lexical marking was a prin- ciple introduced by Clark (1969; Clark 8 Chase, 1972) which accounted for the asymmetrical features of certain dimensional contraries, e.g., 16 high-log, above-below, and deephshallow. Applied to the present study, it would be reasoned that should marking effects be found, Late would be marked with respect to On Time. Along a dimension of timeliness, On Time is the focal point, that is, the most neutral point along that di- mension. Proceeding outward in either direction, one becomes more Early or more Late, acquiring a negative:sense. Since Clark (1969) posited that lexically unmarked items would be stored in and retrieved from memory more rapidly than marked items, one would predict that On Time would be verified more rapidly than Late. Although the analysis of variance reported a significant main effect for Late, which could reflect Late's being inherently negative with res- pect to On Time, it cannot be determined from the present design whe- ther the effect stems from lexical marking or to simply a change in the stimulus. Hence, no marking parameter has been included. The occurrence of Late in Stage 2, however, does play a significant role in the processing of infomration in the present experiment. Upon encoding Late, it appears that the subject will search his internal rep- resentations for the presence of either a negative qualifier in Repre- sentation l, or an Inference in Representation 2. If either or both of these conditions is found, the subject will automatically convert the negative representation to its affirmative equivalent. Since there are two negatives in the false double negative Assertion treatment, two con- versions would be predicted. Each conversion is expected to involve an additional time, 5, At this point, all encodings and elaborations are presumed complete. There are two important features to note about this particular substage of processing. First, this stage is a critical pivot point on 17 which all other processing hinges. A "no" response to question (16) (16) "Is there a Late in Representation 2?" results in a simple outer string identity check, at most involving 1 (outer string mismatch) additional time. A "yes" response, however, will lead either to a semantic truth value change (f2) or at least one conversion (f5), two in the case of the false double negative Assertion condition (+2k), with the possibility of inner (f2) or outer (fig) string mismatches also occurring. Following the "yes" route at this initial decision point would ac- count for the difficulty of the false double negative Assertion Condi- tion. It is the only condition in which two conversion operations may occur. In addition, the outcome of the two conversion executions is an inner string mismatch, incurring additional time 2, Hence, the pre- sence of these three operations (ZEjE), each quite timely, within one treatment greatly increases the processing load of the subject. The processor is now able to commence comparison operations in Stage 3 in order to determine the truth value of the paired represen- tions, as in Clark and Chase (1972). If the subject followed the "no" route at the initial decision point (16), all that remains for him to do is a check of the identity of the outer embedding strings. At most, this will take more time, d, if a mismatch is detected. Subjects following the "yes" route must undergo a more complex set of operations. If no Inference in Store 2 or negative in Store 1 is found at decision point (17), in Stage 3 the subject will convert the (17) "Is there a negative in Store 1 or an Inference in Store 2?" value of the truth index to its opposite, an operation involving time 5, (18). The subject will then proceed to check the identity of the outer 18 (18) (on time) (late) +£- string entries; a mismatch and change of the truth index to its op— posite will involve time d_(19). If, however, a conversion operation (19) (on time) false (late) +(ng) has occurred, as a result of an Inference in Store 2 or a negative in Store 1, the subject, finding identity of the outer strings (all nega- tive entries are eliminated by model specifications), will recheck the identity of the inner strings. Finding identity, he will respond true, incurring no additional time. It is only in the case of the false double negative Assertion (20) that a mismatch will be found, incur- (20) false (on time) +c false (late) —- ring additional time 2, Stage 3 operations aid in the explanation of why the FN treatment is less difficult than the FA. All FN's traverse the "no" path. Only one operation is required (outer string identity check) before the sub- ject is able to leave the system. FA's, however, require the subject to follow the "yes" route. Before changing the value of the truth index to its opposite and checking the identity of the outer strings, the sub- ject must conduct a search of his internal representations. This extra operation would account for the increased difficulty of the FA's. The outcome of Stage 3 is then input to Stage 4 where a response, true or false, is executed. Any time remaining has been allotted to parameter 50’ reflecting base processing time common to all treatments. A least squares analysis of mean RT's yielded the following para- meter estimates: 50, 284 msec.; b, 60 msec.; 2, 357 msec.; i, 180 msec.; k, 312 msec.; g, 368 msec.; and g, 143 msec. The root mean squared 19 deviation (RMSD) was calculated to be 43 msec. Since the RMSD was lower than the smallest parameter, b_(See Sternberg, l969a,b), the model proposed was interpreted as being a reliable description of subjects' processing. In addition, related measures Eftests of each parameter all proved highly significant, where p_< .05. Despite its proficiency at estimating treatment response times, several criticisms must be directed at the proposed model. Failure to completely counterbalance Late and On Time in Stage 1 has previously been cited as a design deficiency resulting in inability to interpret slower RT's in the presence of Late. Furthermore, the change in stim— ulus (i.e., Late in Sentence 2) is theoretically responsible for more complex, time consuming processing. Whether this is due to variation in inner string propositions (Inner String 1 # Inner String 2) or is simply the result of the presence of Late cannot be determined from the present experiment, again due to the failure to counterbalance. Examination of within-subject RT's variability in each condition also raised some doubts as to subject processing. Specifically, there were indications that individual subjects may have been differen- tially responding to a given condition. That is, response to a particu- lar treatment was not necessarily constant across eight presentations. Finally, the proposed model yields an incomplete factorial design. Examination of the model reveals that there is no condition charac- terized by (so + 2 + _k_), (50 + 11 + 213), (£0 + E + g), or any combina- tion of (£0 +n+£+ ...), of (£0 +E+11+ ...), for example. Fur- thermore, if the model is an accurate representation of subject pro- cessing, conditions such as these could not be constructed manipula- ting On Time and Late. It is possible, therefore, that the model 20 produced is the outcome of a host of interacting artifactual vari- ables. While these deficiencies have made it more difficult to analyze and interpret the findings of the pilot study, the ability of the model to characterize trends in the data, particularly in the case of the false double negative Assertion treatment, leads one to believe that if such deficiencies were minimized or eliminated, the model would be representativeof the pattern of RT's yielded, for complex processing (e.g., the abstraction of Inferences) necessitates the use of more complex strategies for successful task completion. To test this hy- pothesis, the following experiments have been conducted. Experimental Extensions A major criticism of the Clark and Chase (1972) approach to infor- mation processing is that very simple tasks have been used to illustrate principles of analysis theoretically applicable to any task. Yet, lit- tle research has been carried out to validate this generalization. The results of the pilot study indicate that in more complex situations, only peripheral aspects of the pure true or conversions models can be found. It would seem that processors adapt the complexities of their strategies to meet the demands of the task at hand. It is important to determine whether subjects in fact do use highly complex strategies in difficult problems, or whether all verification strategies can be reduced to the true or conversion strategies. Experiment 1 examines this problem. An auditory sentence-sentence verification of Assertions and In- ferences task is used in an attempt to replicate the experimental fin- dings of the pilot study; specifically, that a highly complex form of 21 conversion (i.e., subject to two conditions of occurrence) operates in the verification of Assertions and Inferences; and that the false double negative Assertion, involving two conversion operations, is the most difficult verification to make. As in the pilot study, negative entries have been input to Stage 1, Stage 2, and Stages 1 and 2 encodings. Experiment 1 differs from the pilot study in three ways: first, there is complete counterbalancing of On Time and Late in Sentences l and 2. This will allow one to determine whether lexical marking exists along the dimension of time. Secondly, all the subjects are tested in three individual sessions. It is hypothesized that the problem of within-subject variability will be remediated by multiple session tes- ting of subjects. It is thought that an individual's performance pat- tern will stabilize across three sessions of testing such that, by ses- sion 3, a subject will be processing a given condition consistently on all presentations. It is predicted that Experiment 1 will reproduce the findings of the pilot study using eight parameters: to, base processing time; a, Stage 2 Late encoding time; 2, Stage 2 negation encoding time (Asser- tions); 29 negation encoding time in Stage 2 (Inferences); 3, Stage 2 Inference encoding time; k, conversion operation in Stage 2 substage; 2: change in value of truth index in Stage 3 due to the presence of an inner string mismatch; and g, change in the value of the truth index in Stage 3 due to the occurrence of an outer string mismatch. Since in the pilot study, no lexical marking effects were predicted for Late, in Experiment 1 prediction of parameter 5.18 expected to decrease the pilot study values of both g_and 5, Conversion strategies are expected to be used in conditions in which inner string mismatches occur with 22 a negative in Representation 1 or an Inference in Representa- tion 2. It is quite possible, however, that complex operations are used in any higher order language comprehension task. Experiments 2 and 3 are designed to test this hypothesis in the auditory and visual modali- ties, respectively. Both experiments investigate subjects' verifica- tion of simple Assertions (21), that is, information directly relayed in a sentence. (21) He was 22_time today. He was on time today. It is predicted that in both experiments the effects of encoding a negative and a Late in Sentence 2 will be seen. Furthermore, it is predicted that inner string mismatches will be more time consuming than inner string identities. If in simple language comprehension tasks the same rules are used to verify information as in the simple sentence—picture verification tasks, one would expect the results of Experiments 2 and 3 to meet the predictions of either the true model of negation or the conversion model. It may be, however, given the complexities of language, that different rules (which may be characterized by another serial stage processing model, e.g., the pilot model, or which may be executed in parallel operations) are used in processing. Finding an answer to this question is one of the purposes of Experiments 2 and 3. The final aim is to determine whether differences exist in how the same information is processed in different modalities. It is possible that conversion strategies would not be used in a visual modality sentence-sentence verification task, as conversion does not seem to be a necessary routine (since the stimuli are always available); that is, 23 the routine would not seem to maximize processing time. Conversion with subsequent mismatch checks may be less economical than simple immediate inner and outer string checks when stimuli are always available. On the other hand, if conversion operations were executed in an auditory task, they could serve to minimize the number of elements being held in the short term store at any one time, thus easing the load of the processor. Hence, conversion in auditory tasks may be an optimal strategy. Summary The experiments are investigations of negative conversion opera- tions in a sentence-sentence verification of Assertions and/or Infer- ences task. Their basic premise is the determination of whether con- version operations or complex rule systems are automatically used in higher order processing as a function of the task being a language comprehension problem. Experiment 1 addresses itself to the problems of lexical marking and conversion operations in a complex task (the verification of Asser- tions, and the construction and verification of Inferences). Experiment 2 examines whether complex operations are simply a function of higher order processing (in situations in which Inferences must be made, as in Experiment 1) or are rather a function of the complexity of the task used. Finally, Experiment 3 attempts to determine whether come plex comprehension operations are equally employed in both auditory comparison tasks and visually presented sentence~sentence verifications. CHAPTER 2 Experiment 1 Experiment 1 is a strict test of the model generated from the pilot study's results. Essentially, the experiment purports to test the hyh pothesis that complex conversion operations are employed in more de- manding problem solving settings. Investigation of negative conver- sion operations will be made using an auditory sentence-sentence veri- fication task. Verification of both directly conveyed information (Assertions) and abstracted information (Inferences) will be examined. Method Subjects. The subjects were 12 male and female undergraduates participating for credit in an introductory psychology class at Michigan State University. All the subjects were also paid $2.00 for their participation. Stimuli. Two hundred fifty-six sentences recorded by a female speaker were presented in two-sentence test sequences. Sentence 1 con- structed a context concerning an individual's activity, closing with a phrase indicating what a particular actor did today. One-half of the phrases contained the term On Time, the remaining half contained the term Late. In addition, one-half of the sentences were affirmaitve and one-half negative. Sentence 2 was a simple verification statement; one- half of the sentences contained the term On Time; one-half, Late; one 24 25 half were affirmative statements, with the remaining half being negative. Sentence 2 also varied as to the nature of the information tested: one— half of the sentences referred to information directly conveyed in Sen- tence 1 (Assertions), while the remaining half tested Inferential judgments. Table 2.1 contains the 32 possible stimulus combinations. Apparatus. Sentence stimuli, recorded on a TEAC 23008 stereo recorder, were presented to the subjects through stereo headphones. A click inaudible to the subject was placed on the tape approximately 200 msec. after the beginning of clause two of Sentence 1 (...sugpri- singly...) and approximately 200 msec. after the onset of speech in Sentence 2. These clicks automatically activated Hunter timers which recorded the subjects' RT's. A three button panel was located directly in front of the subjects: a button marked "C" (Comprehension) was cen- trally positioned, with a "T" (True) or "F" (False) button to either side. Depression of any button automatically stopped the Hunter timers and triggered the presentation of the next stimulus. Procedure. All the subjects were individually tested in three performance sessions. The order of stimulus presentation was randomly varied across the three sessions, so that no subject receiVed the same input order twice. The participants were instructed via the tape recor- der to listen carefully to the contents of each sentence, for they would be asked to make true-false judgments of each sentence-sentence pair. Following Sentence 1, the subjects were required to press the comprehen- sion ("C") button as quickly as possible, indicating that they had understood the sentence. Sentence 2 immediately followed. The subjects were then required to depress either the "T" (True) or "F" (False) but- ton, depending upon their evaluation of the comparison. The position of 26 TABLE 2.1 Conditions of Experiment 1 Sentence 1 surprisingly, he was . . . 22_time today. late today. not 22 time today. not late today. Sentence 2 smith 0 O O . was Ln time today. . was late today. . was Lot Ln time today. . was Lot late today. . is usuallT_ on time. . is usually_late. . is not usually 22_time. . is not usually late. . was Ln time today. . was late today. . was Lot Ln the today. . was Lot lLte today. . is usually_ on the. . is usually late. . is not usually gg_time. . is not usually late. . was 22_time today. . was late today. . was Lot Ln the today. . was Lot lLte today. . is usuallT_ on the. . is usually late. . is not usually'gg_time. . is not usually late. . was gg_time today. . was late today. . was 3515111 time today. . was Eg£_late today. . is usually 22_time. . is usually late. . is not usually'22_time. . is not usually late. 27 the "T" and "F" buttons was randomly varied between subjects. All the subjects were instructed to answer each comparison as quickly and as accurately as possible. A five second interval occurred between each test comparison. The experimenter recorded all response times. All the subjects received eight practice trials prior to actual testing, and were advised of their performance. In addition, after each day of testing, the subjects were advised of their overall per- formance for the day. Upon completion of the third day of testing, the subjects were asked to specify strategies used during processing, if any. The subjects were also asked to indicate whether they had changed their strategy across the three testing sessions, to describe this change, and to explain it. Results and Discussion In this section, performance over all three sessions, as well as performance in Session 3 alone, will be discussed. In addition, an attempt will be made to interpret the Day 3 results in terms of an ad- ditive factor (independent, serial stage) model of processing. The rationale for focusing on the Day 3 results was as follows. It was felt that if the subjects did choose a common strategy which directed their information processing, it wuold be most obvious and stable by the Day 3 performance. Thus, it was hoped that analysis of the Day 3 data would yield the most clear—cut insights into the processes underlying perfor- mance in the conditions of Experiment 1. It was also hoped that the within-subject variability in responding seen in the pilot study would decrease over the three testing sessions. 28 Since the major results to be reported occurred in both the Day 1 and Day 2 performances, these data will not be discussed. Mean RT per condition for Days 1 and 2 are contained in Appendices A and B, res- pectively. '§_va1ues of the significant results of within-subject analy- ses of variance performed on these data can be found in Appendix C. Overall. Only correct responses were included in the means; in- dividual data were eliminated if the were i;3 standard deviations from the mean. The error rate for the overall performance in Experiment 1 was 7.2%. A 2 x 2 x 2 x 2 x 2 x 3 within-subjects analysis of variance was performed on the data. The factors were: Condition (Assertion vs. Inference), Term 1 (On Time vs. Latel), Negation 1 (Affirmative vs. 1 1 Negativel), Term 2 (0n Time2 vs. Latez), Negation 2 (Affirmative2 vs. Negativez), and Day (1, 2, or 3). Significant results of this analysis, where p ‘<.05, in comparison with the significant results of Days 1, 2, and 3, are indicated in Table 2.2. In general, Inferences were more difficult than Assertions, Condition, {(1,11) = 231.94, MSe - .410; the presence of a negative in either sentence significantly increased processing time, Negation 1, 3K1,11) - 9.415, MSe - .143, and Negation 2, £(1,11) - 158.987, MSe - .217; and the subjects were able to perform the verifications more ra- pidly over time, Day, {(2,22) = 126.54, MSe - .201. Contrary to pre- dictions, comparisons containing an 0n Time2 were more difficult than Late2 comparisons, Term 2!.Efi1’11) - 77.981, MSe - .140. The presence of an Inference served in several cases to inflate the effect of other factors paired with it. As indicated in Tables 2.3 and 2.4, the effect of a negative in either Sentence 1 or 2 was greater for 29 TABLE 2.2 Significant Main Effects and Interactions of the Analyses of Experiment 1 Condition Overall _I_)_§y__l_ 23.2.2. 23.2.}. Negation 1 * * * M Term 2 t * t * Negation 2 * * * * Condition * * * * Day * Condition x Negation 1 * * * Condition x Term 2 * * M Condition x Negation 2 * Negation 1 x Negation 2 * * M Term 1 x Negation 1 M Term 1 3 Term 2 * * * * Term 1 x Term 2 x Negation 2 * * M Condition x Term 1 x Negation l x Negation 2 * Condition x Term 1 x Term 2 x Negation 2 * Condition x Term 1 x Negation 1 x Term 2 x Negation 2 * * * Condition x Term 1 x Negation 1 x Negation 2 x Day * Condition x Term 1 x Negation 1 x Term 2 x Day * Where * denotes a significant result (p_< .05); and M denotes a marginally significant result. 30 TABLE 2.3 Mean RT in Msec. for the Condition x Negation 1 Interaction Condition Negation 1 Affirmative1 Negativel Mean Assertion 1627 1647 1637 Inference 2154 2270 2213 Mean 1891 1958 TABLE 2.4 Mean RT in Msec. for the Condition x Negation 2 Interaction Condition Negation 2 Affirmativez Negative2 Mean Assertion 1489 1786 1638 Inference 2015 2410 2213 Mean 1752 2098 TABLE 2.5 Mean RT in Msec. for the Negation 1 x Negation 2 Interaction Negation 2 Negation 1 Affirmative1 Negativel Mean Affirmative2 1702 1800 1751 Negative2 2078 2116 2097 Mean 1890 1958 31 Inferences than for Assertions, Condition x Negation 1,'§(l,11) I 12.965, MSe I .050, and Condition x Negation 2, §(1,11) I 7.71, MSe I .090. Also, the effect of an On Time in Sentence 2 was greater among Inferences (mean RT of On TimeZ—Late2 I 245 msec.) than among Assertions (mean RT difference between On TimeZ-Late2 {(1.11) - 27.547, use - .027. I 144 msec.), Condition x Term 2, As can be seen in Table 2.5, the effects of two negatives were not additive, Negation 1 x Negation 2, E(1,11) I 8.031, MSe I .032. That is, the effect of a Negative2 was greater in the presence of an Affir- mative1 than a Negativel. It could be the case that negation in Sen- tence 1 primed the subject for another negative entry, thus facilitating a response. An affirmative entry, on the other hand, would not prepare the subject for a negative, thus resulting in a greater slowdown of processing. Consistent with the predictions, inner string mismatches were more time consuming than inner string identities, Term 1 x Term 2, £(1,11) I 12.459, MSe I .171. That is, Latel-On Time2 comparisons had slower RT's than On Timel-On Time2 comparisons (2069 msec. vs. 1974 msec., respectively) and On Timel-Late2 comparisons were longer than Latel- Late2 comparisons (1866 msec. vs. 1790 msec., respectively). This effect is illustrated in Figure 2.1. As seen in Table 2.6, the effect of a Negative2 was greater in the presence of an inner string mismatch than in the presence of a match, Term 1 x Term 2 x Negation 29.3fi1’11) I 7.535, MSe I .036. When inter- preted in light of the Term 1 x Term 2 interaction reported above, one might conclude that as the processing demands of a comparison are in- creased (for example, due to the occurrence of an inner string 32 Term 2: 2100i ,3} 900i .3 l ~ ~ - ~ ~ ‘52 18004 "- 17004 7 I On Time1 Late Term 1 Figure 2.1 Mean RT in Msec. to Term 2 as a function of Term 1 TABLE 2.6 On Time Late Mean RT in Msec. for the Term 1 x Term 2 x Negation 2 Interaction Negation 2 Match Mismatch Mean On Timel- Latel- Latel- On Timel- On Time2 Late2 On Time2 Late2 Affirmative2 1815 1633 1888 1671 1751 Negative2 2134 1945 2251 2061 2098 Mean 1975 1789 2070 1866 33 mismatch), the effects of the Negative2 will be exaggerated, resulting in slower RT's. The remaining significant interactions of the overall analysis were the following five-way ones: Condition x Term 1 x Negation 1 x Term 2 x.Negation 2, £(1,11) I 18.608, MSe I .072; Condition x Term 1 x Negation 1 x Negation 2 x Day, 2(2,22) I 5.393, MSe I .028; and Condition x Term 1 x Negation 1 x Term 2 x Day, E(2,22) I 3.632, MSe I .042. Basically, these indicated that comparisons requiring an Infer- ence judgment, and which contained a Negativez, an 0n Timez, and in which the inner strings did not match were most time consuming. Fur- thermore, it appeared that while the subjects' performance did improve over days, Assertion judgments showed greater improvement over time than did Inference conditions. A}; Assertion conditions were executed more rapidly on Day 3 than on Day 1; however, only certain of the Inference conditions improved over time. 931;. Mean RT per condition for the Day 3 performance is con- tained in Table 2.9 (see p. 47 ). The average error rate for session 3 was 5.5%. A 2 x 2 x 2 x 2 x 2 within-subject analysis of variance was per- formed on the data. The factors investigated were: Condition (Asser— tion vs. Inference), Term 1 (On Time1 vs. Latel), Negation 1 (Affirma- tivel vs. Negativel), Term 2 (On Time2 vs. Latez), and Negation 2 (Affirmative2 vs. Negativez). The significant results of this analysis, where p < .05, are indicated in Table 2.2. As in the overall analysis, main effects of Term 2, £(1,11) I 60.006, MSe I .058; Negation 2, Efil,11) I 270.205, MSe I .042; and Condition, §(1,11) I 133.133, MSe I .281 were significant. However, 34 the main effect of Negation 1 was only marginally significant, §(l,11) I 4.681, MSe I .061. Why this was so is not readily understood. It is possible that subjects can more efficiently encode and operate on a Negative1 (i.e., the Negative makes fewer processing demands) after 1 three days of testing. Only two interactions were significant in the Day 3 data: Term 1 x Term 2, §(1,11) I 5.513, use I .158, and Condition x Term 1 x Negation l x Term 2 x Negation 2, §(1,11) I 14.313, MSe I .201. As in the over- all analysis, mismatches were more difficult than matches, and negative Inferences containing inner string mismatches and an On Time2 were the most difficult comparisons to make. We will now turn to an evaluation of how well the data of Experi— ment 1 fit the model derived in the pilot study. Evaluation _o_£ the fillot S tudy Model. The predictions of the pilot study model are not supported by the results of the present exper- iment. It was predicted that Late would be marked with respect to On Time; however, the opposite effect was obtained. In each Experiment 1 analysis, On Time2 comparisons took significantly longer than Late2 comparisons. Whether this effect was due to the lexical marking of On Time or was an artifact of the stimulus itself (On Time2 stimuli were approximately 297 msec. longer than Late stimuli) cannot be determined 2 at present. Furthermore, recall that in the pilot study, Affirmative Assertion 2 conditions ordered themselves in the following manner: TA < FN < FA < TN. Inclusion of Negativel Assertions yielded the following ordering: TA < FA “ TN < FN. In the present experiment for Day 3 data, when only Affirmative1 Assertion comparisons were considered, it was found 35 that TA (1286 msec.) < FA (1447 msec.) < EN (1632 msec.)‘< TN (1771 msec.). When Negative1 Assertion conditions were included in the analysis, the following ease of processing ordering was produced: TA (1286 msec.)‘< FA.(1447 msec.) < TN (1603 msec.) I FN'(1609 msec.). So on either count, the results of the pilot study do not match those of Experiment 1. The pilot model encounters more trouble when one compares the predicted and actual orderings of conditions on Days 1 and 3. The most noticeable violation of the pilot model's predictions was the fin- ding that the false double negative Assetion was not the most time consuming condition on either day. Examination of Table 2.9 reveals that by Day 3, 13 of the 16 Inference conditions were as or more time consuming than the false double negative Assertion. In order to more fully assess the inadequacies of the pilot model, ordinal predictions of the model were compared with the obtained orders of the conditions. A tabulation of the comparisons between the pre- dicted and obtained data is shown in Table 2.7. In each section of this table are listed two or more conditions which should significantly in- crease in difficulty as we move from left to right according to the pilot model because of an increase in the number of parameters. For instance, according to the pilot model, Latel-On Time2 should take sig- nificantly longer to process than 0n Timel-On Time2 because one more parameter is required (to + a + 3 vs. £0 + a). It should be noted that in virtually every ordering, the model makes an error in prediction for one of the conditions. It can also be seen that the misrepresentation of the data usually occurs when parameter d_(outer string mismatch) is predicted to be the only factor 36 TABLE 2.. 7 Pilot Model Predictions and Obtained Values in Msec. for Selected Conditions of Days 1 and 3 Day Day Day Day Day Day Day Day Day Day H On Timel- Not On Time1 0n Timel- Latel- On Time2 On Time2 Not On Time2 Not On Time2 t +a t +a+d t +a+b+d t +a+b+c+d o o o o 1654 1661* 1986 1993* 1383 1480 1683 1759 On Timel- Latel- On Time2 On Time2 t +a t +a+c o o 1654 1703 1383 1519 On Timel- Not On Timel- 0n Timel- Latel- On Time2 Not On Time2 Not On Time2 Not On Time2 t +a t +a+b t +a+b+d t +a+b+c+d o o o o 1654 1862 1986 1993* 1383 1681 1683* 1759 On Timel- Not Latel- On Time2 On Time2 t +a t +a+k o o 1654 1787 1383 1517 Not On Timel- On Timel- Usually On Usually On Time Time 2 2 t +i+a t +i+a+d o o 2234 2056* 2116 2064* 37 TABLE 2.7(cont'd.) Day Day Day Day Day Day Day Day H UH Not On Timel— Usually On Time2 t +i+a o 2234 2116 Not On Timel— Usually On Time2 to+i+a 2234 2116 Not Latel- Usually On Time2 to+i+a+k+d 2307 2195 Not Latel- Not Usually On Time2 to+i+a+n+k 2747 2471 On Timel- Not On Timel- Not Usually Not Usually On Time2 On Time2 to+i+a+n to+i+a+n+d 2397 2599 2397 2518 Latel- Usually 0n Time2 to+i+a+k 2043* 2070* L - atel Not Usually On Time2 to+1+a+k+d+n 2721 2462 Late1 Not Usually On Time2 to+i+a+n+k+d 2721* 2462* Latel- Not Usually On Time2 to+i+a+n+d+k 2721 2462* Where * indicates an inaccurate prediction 38 differentiating two conditions. This was true of both Day 1 and Day 3 performances. In conclusion, the pilot model is £g£_a good predictor of either Day 1 or Day 3 performance. It is clear that practice is not responsible for the dramatic change of events between the pilot study and Experiment 1. It is possible that the change in the point from which one measured RT in the pilot study and in Experiment 1 is in part responsible. Since RT in the pilot study was measured from speech offset, a great deal of the subjects' encoding and processing could have been lost. RT measurement from Sentence 2 onset in Experiment 1 is undoubtedly a more sensitive measure of processing. Evaluation.g£_the True Model of Negation. Recall from Chapter 1 that the true model of negation (Clark & Chase, 1972) predicts that TA < FA < FN < TN. Thus, the true model of negation was able to account for the true—false, affirmative-negative orderings cited above (p. 35). for the Affirmative -Assertion conditions of Experiment 1. However, as 1 can be seen in Table 2.8 (which is similar to Table 2.7), this model has just as many problems as the pilot model in predicting overall per- formance on Days 1 and 3. For this model, both parameters £_(inner string mismatch) and d_(outer string mismatch) appear to have question- able roles. Thus, the true model must also be discarded as an accurate representation of the subjects' processing. Development 9_f_ 2 MM g Information Processing. In developing a model of information processing in the present paradigm, it would be most efficient to assume that the verification of Assertions and Infer- ences are subject to the same set of comparison rules, that is, en— tailing no extra processing time beyond encoding due to the nature of 39 TABLE 2.8 Predictions of the Clark and Chase Model and Obtained Values in Msec. for Selected Conditions of Days 1 and 3 Latel- Late2 t O 1381 1189 Day 1 Day 3 Latel- Late2 t 0 Day 1 1381 Day 3 1189 Latel- Late2 t o p... 1381 1189 Day Day 3 Latel- Late2 t o 1381 1189 Day 1 Day 3 Not Latel- Usually Late2 t +1 0 2057 1817 Day 1 Day 3 On Timel- Not On Timel- Late2 Late2 t +c t +c+d o o 1569 1550* 1374 1323* On Timel- Not On Timel- Late2 Not Late2 t +c t +b+c o o 1569 1970 1374 1635 Not On Timel- -On Timel- Not On Time2 Not On Time2 t +b t +b+d o o 1862 1986 1681 1683* lot 0n Timel- Not On Timel- On Time2 Late2 t +d t +c+d o o 1661 1550* 1480 1323* Latel- On Timel- Usually Usually Late2 Late2 t +i+d t +i+c+d o o 1958* 1821* 1822* 1802* On Time1 Not Late t +b+c+d o 1839 1785 2 . i _ Cn T me1 Not Late t +b+c+d o 1839* 1785 2 On Timel- Not Late to+b+c+d 2 1839* 1785 Not On Timel— Not Usually Late2 to+i+b+c+d 2359 2353 TABLE 2.8(cont'd.) Not Latel- Usually Late2 t +1 0 2057 1817 Day 1 Day 3 Not Latel- Usually Late2 t +1 o 2057 1817 Day 1 Day 3 Not Latel- Usually Late2 t +i o Latel- Not Usually Late2 t +i+b o 2102 2102 Not On Time - 1 Usually Late2 to+i+c 1964* 1980 Latel- Not Usually Late2 t +i+b o On Timel- Not Usually Late2 t°+i+b+c 2481 2308 On Timel- Not Usually Late2 t +i+b+c o 2481 2308 Not Latel- Not Usually Late2 t +i+b+d 0 Not On Timel- Not Usually Late2 to+i+b+c+d 2359* 2353 Not On Timel- Iot Usually Late2 t +i+b+d o Where * denotes an inaccurate prediction 41 their being an Assertion or an Inference. This, of course, would make processing most economical. In order to determine whether the differences between Inferences and Assertions, and negatives and affirmatives, are confined to the encoding stage of processing or whether time consuming differences in processing occur in both encoding and comparison stages, six subjects participated in a language perception experiment. It was reasoned that time to recognize a stimulus is a fundamental part of the encoding process. Therefore, if one parcels out the time it takes to identify a stimulus from the overall processing time, one should be able to identify processes affecting the comparison stage. It was also hoped that this experiment would shed some light on the Latez-On Time2 dif- ference which was not predicted. Six college students participated in a language perception ex- periment in fulfillment of an introductory psychology course require- ment. The stimuli were Assertion and Inference test sentences appearing in Experiment 1. Subjects were instructed to depress a button as soon as they recognized the target phrase for which they were primed: On Time, Late, Not On Time, Not Late, Usually On Time, Usually Late, Not Usually On Time, or Not Usually Late. All the subjects were tested in blocks under all priming conditions. In each condition, one-half of the stimu- li contained the term On Time, one—half, Late; one—half were affirmative, the remaining half, negative; finally, one-half of the stimuli were Assertions, and the remaining half tested Inference constructions. Under Assertion instructions, the subjects made 80 judgments; 96 judg— ments were made under Inference instructions. 42 Mean RT per condition for correct identifications were tabulated: On Time I 1094 msec.; Late + 959 msec.; Not On Time I 1227 msec.; Not Late I 1145 msec.; Usually On Time I 1670 msec.; Usually Late I 1407 msec.; Not Usually On Time I 1846 msec.; and Not Usually Late I 1679 msec. The overall error rate was 2.02. Mean RT's were then subtracted from the overall condition mean RT's of Experiment 1 and the Day 3 data were reanalyzed according to the analysis of variance described earlier. The effect of Term 2 was no longer significant (£(1,11) I 1.373, ‘p_> .05), nor was the Condition main effect (F(1,11) I 2.198, p_> .05). It must be concluded then, that the effects of an Inference and an On Time2 are effects which must be restricted to the encoding stage of processing. The effects of negation (Negation 2, £(1,11) I 52.824, £_< .005) as well as the effects of inner string mismatch (Term 1 x Term 2, E(1,ll) I 5°5139.2.< .039) were found to exceed the bounds of the encoding stage. In conclusion, any model which purports to represent the data must posit encoding parameters for On Time2 and the Inference, and both en- coding and comparison parameters for negation. The following model is proposed as an adequate characterization of the Day 3 data. Of the models tested, it was found to be the most representative of actual per- formance; furthermore, it was able to restrict the effect of the In- ference to the encoding stage alone, as mandated by the last reported analysis. As in the previously discussed models, processing will be broken down into four successive stages. In Stage 1, subjects will encode Sentence 1. The following representation (22) is suggested but not (22) . . . surprisingly, he was not 22 time today. false (on time) 43 necessarily the only possible one. As mentioned earlier, however, time to encode this sentence is not part of the measured and analyzed RT. In Stage 2, the subject will encode the second entry in a repre- sentation similar to that of Stage 1 (23). The encoding of this sentence (23) Sentence: Mr. Smith is not usually 22 time. Representation: false (usually (on time)) is part of the overall RT recorded, therefore certain encoding para- meters must be posited. Given that On Time sentences took consistently more time to verify 2 than Late2 sentences, parameter §_will be the time taken to encode an On Timez, parameter b will reflect the time taken to encode a Negativez, and i will be the time needed to encode an Inference. Finally, since in both the Assertion and Inference Conditions two negative entries ap- peared to be most difficult, it is proposed that parameter 2 will re- flect the extra processing time due to the occurrence of a second negative. It is only during Stage 2 that parameters a, _b_, i, and 2 function. If example (23) were encoded in the presence of (22), one would predict that parameters a, b, i, and _11 would all be operating. Once the second entry has been entirely encoded, processing enters the comparison phase, or Stage 3. It is during this stage that the E) g, and 5 parameters operate. It is proposed that in both Assertion and Inference comparisons, sub- jects will verify first the contents of the inner strings, and then pro- ceed to verify outer strings. Finding mismatches in either of these two comparisons will result in an increase in processing time by fg_ and.fg, respectively. 44 Following inner and outer string comparisons, the Inference or Assertion condition will determine the final decision rule. It is be- lieved that subjects will not automatically make changes in the value of the truth index upon finding a mismatch. Rather subjects will tempor- arily store the fact of the mismatch in a hypothetical memory buffer. Where Inference and Assertion operations differ will be in their interpretation of the contents of the memory buffer. In the Assertion condition, if the number of mismatches recorded in the buffer is an even number (0 or 2), the subject will respond True. If an odd number of mismatches has been found, however, the subjects will change the value of the truth index to its opposite value (False), requiring additional processing time, fx. As in other models, the value of the truth index is assumed to be set at True at the outset. In the Inference comparisons, decision making is governed by a converse set of rules. In this instance, if the value stored in the buffer is an even number, the subject will respond false. It is only when the buffer is set at an odd value that the subject will respond True. As in the Assertion comparison, the changing of the truth index to its opposite polarity will increase processing time by fig, Upon completion of comparison operations, the output of Stage 3 is executed in the response phase, Stage 4. The additive factor model assumes that each stage is independent of all others, and that the sum of all operations in a given condition will reflect the total processing time. In the example cited below (24), the response time of that comparison is predicted to be the sum (24) false (on time) false (usually (on time)) t°+i+a+b+n+x of to, a base processing time, and the necessary parameters appearing in 45 Stages 2 and 3. Figure 2.2 displays a flowchart of the possible al- ternatives one has in such a paradigm. A least squares analysis was performed using subjects' mean RT per condition in order to determine parameter estimates. The eight parameters were estimated to be: , 1169 msec.; i, 625 msec.; g, 191 ED msec.; b, 285 msec.; g, 95 msec.; d, 72 msec.; g 118 msec.; and _x_, 52 msec. Observed scores, predicted values, and their deviations from the actual scores per condition are contained in Table 2.9. In order to test the goodness of fit of the model, the RMSD was obtained, RMSD I 53 msec. Since parameters £0, a, _b_, _i_, g, d, and _n were all greater than the RMSD, these parameter estimates were inter- preted as being reliable features of information processing (see Stern- berg, l969a,b). Parameter x is questionable, however, since it is equivalent to the RMSD. Parameter a, as represented by the Term 2 main effect was highly significant. Parameter 2, represented by the Negation 2 main effect was also significant. In addition, parameter i_was shown to be a reli- able effect through the main effect for Condition, and parameter g_sig— nificant by the Term 1 x Term 2 interaction. Unfortunately, significant effects corresponding to the remaining three parameters, d, g, and x could not be isolated. No two conditions could be found which differed only in terms of 223_of these parameters. However, it is the case that unlike other models, the proposed model is able to predict the T-F, A-N ordering found in the data, that is, that TA< FA< TN: FN. 46 # uneawuoexm mo Home: one no mo>wuoeuouad wewmmoooum .~.~ shaman o, _ a. license Q on 11 no» awash vaommum ~£ouua «cosmos mwowuum Houao waaon nowuuomm< mmhv on may on on so mH mMFV no» onus ouumoneo voodoo“ mu“ ou f 1|: «nouns News“ nupuu mwsuuum Hosea on» no oaao> mM¥v on may on one swoono @ n+3. .Nofiuuwuz was at new aofiuuwoz «H on .m¥ .moaououon wH 08.... .m+ .Nofiuumoz «H maoemom . .I .N acouoa o+ mafia no NH «pounce mwswuum Houso “N ovooom swan doauuoomw ooh 2: on no mH _ u mucosa — J 47 TABLE 2.9 Mean RT in Msec. Per Condition, Estimated, Observed and Deviations for Experiment l—Day 3 Data Conditions Projected Estimated Observed Deviation Parameters On Time- to+a 1360 1383 - 23 On Time On Time- to+c+x 1316 1374 - 58 Late On Time- to+a+b+d+x 1769 1683 + 86 Not On Time On Time- to+b+c+d 1621 1785 -164 Not Late Late- to+a+c+x 1507 1519 - 12 On Time Late- to 1169 1189 - 20 Late Late- t +a+b+c+d 1812 1759 + 53 Not On Time 0 Late- to+b+d+x 1578 1580 - 2 Not Late Not On Time- to+a+d+x 1484 1480 + 4 On Time Not On Time- t +c+d 1336 1323 + 13 Late o Not On Time- to+a+b+n 1763 1681 + 82 Not On Time Not On Time- to+b+c+n+x 1719 1635 + 84 Not Late Not Late- to+a+c+d 1527 1517 + 10 On Time Not Late- to+d+x 1293 1334 - 41 Late Not Late- to+a+b+c+n+x 1910 1945 - 35 Not On Time Not Late- to+h+n 1572 1552 + 20 Not Late 48 TABLE 2.9(Cont'd) Conditions ' Projected Estimated Observed Deviation Parameters On Time- to+i+a+x 2037 2064 - 27 Usually On Time On Time- to+i+c 1889 1802 + 87 Usually Late On Time- t +i+a+b+d 2342 2397 - 55 Not Usually On Time ° On Time- t +i+c+b+d+x 2298 2308 - 10 Not Usually Late Late- t +i+a+c 2080 2070 + 10 Usually On Time Late- to+i+x 1846 1822 + 24 Usually Late Late- t +i+a+b+c+d+x 2489 2462 + 27 Not Usually On Time Late- t +i+b+d 2151 2102 + 49 Not Usually Late Not On Time- to+i+a+d 2057 2116 - 59 Usually On Time Not On Time- to+i+c+d+x 2013 1980 + 33 Usually Late Not On Time— to+i+a+b+n+x 2440 2518 - 78 Not Usually On Time Not On Time- t +i+b+c+n 2292 2353 - 61 Not Usually Late, Not Late- t +i+a+c+d+x 2204 2195 + 9 Usually On Time Not Late- to+i+d 1866 1817 + 49 Usually Late Not Late- to+i+a+b+c+n 2483 2471 + 12 Not Usually On Time Not Late- to+i+b+n+x 2249 2254 - 5 Not Usually Late 49 Summary It is clear from Experiment 1 that Inference verifications take considerably more time than Assertion judgments. Furthermore, the increase in processing time in this case is restricted to the encoding stage. Consistent with earlier studies, our findings confirmed that negatives are more difficult than affirmatives, their effects being seen in both the encoding (fb) and the comparison (fig) stages; and that inner string mismatches are psychologically more difficult than iden- tities. Furthermore, despite the subjects' variation in processing across the three sessions, these features are characteristic of each day's performance. Thus, we have been able to isolate some critical features of processing in a sentence-sentence verification task of Inferences and Assertions. CHAPTER 3 Experiment 2 Experiment 2 investigates whether subjects use complex strategies in a simple sentence-sentence verification task. In the present study, all verifications of Inferences have been eliminated; only simple de- clarative sentences counterbalanced for Term (On Time vs. Late) and sentence polarity (affirmative vs. negative) will be examined. If, the data cannot be adequately represented by simple models, it would appear that the comprehension processes involved in a strictly linguistic task, either in higher order situations (formulating Inferences) or simple Assertion comprehension, are more complex than those involved in pre- vious sentence-picture matching experiments. Method Subjects. The subjects were 12 male and female undergraduates participating for credit in an introductory psychology at Michigan State University. Stimuli. Two hundred fifty—six declarative sentences recorded by a female speaker were presented in two-sentence test sequences. Sen- tence 1 was a simple statement of fact. One-half of these sentences con- tained the term On Time, the remaining half, Late. One-half of the sen— tences were affirmative, one-half, negative. Sentence 2 was a simple ver- ification sentence, one-half being affirmaitve, and one-half, negative. One-half of the sentences contained the term On Time, the remaining 50 51 ones, Late. The design yielded 16 possible combinations, as seen in Table 3.1. Apparatus. Identical to that used in Experiment 1. Procedure. All the subjects were subject to the same experimental procedures used in Experiment 1. Results and Discussion The significant results of an overall analysis of the Experiment 2 data, and of a separate analysis of the Day 3 data will be presented and discussed in the following section. As in Experiment 1, an at- tempt will be made to interpret the Day 3 data in terms of an additive factor model. TABLE 3. 1 Conditions of Experiments 2 and 3 Sentence 1 Sentence 2 Mr. Smith was on time today. . . . on time today. "" . . . Tate today. . . . 22£_22 time today. . . . £££_late today. . . . . . . . late today. . . . 22 time today. . . . late today. . . . 22£_22_time today. . . . £2£_late today. . . . . . . . n2£_22_time today. . . . g§_time today. . . . late today. . . . 22£_on time today. . . . Eg£_late—230ay. . . . . . . . 22£_late today. . . . gg_time today. . . . late today. . . . “—0222. time today. . . . 32£_late today. 52 No specific discussion will be made of the results of Day 1 and Day 2 performance analyses, as all the significant results of these two per- formances were also significant on Day 3. A complete listing of the mean RT's per condition on Days 1 and 2 can be found in Appendices D and E, respectively. The significant main effects and interactions in the analyses of variance for these two days are indicated in Appendix F. Overall. Only correct responses were included in the analysis; individual data;: 3 standard deviations from the mean were interpreted as errors. The overall error rate was 4.12. A 2 x 2 x 2 x 2 x 3 within-subjects analysis of variance was per- formed on the data. The factors were: Term 1 (On Time vs. Latel); 1 Negation 1 (Affirmative1 vs. Negativel); Term 2 (On Time2 vs. Latez); Negation 2 (Affirmative2 vs. Negativez); and Day (1, 2, or 3). The significant results of this analysis, where p_< .05, in relation to the significant results of Days 1, 2, and 3, are indicated in Table 3.2. In general, On Time2 conditions were more difficult than Late2 judgments, Term 2, F(1,11) I 43.230, MSe I .036; and, conditions con- taining negative second sentences were more time consuming than those containing affirmative ones, Negation 2, F(1,11) I 214.667, MSe I .060. The subjects' execution of responses also became more rapid over time, Day, 2(2,22) I 38.195, MSe I .062. Consistent with the findings of Experiment 1, inner string mis— matches were more difficult to process than inner string matches, Term 1 x Term 2, Ffil,11) I 14.958, MSe I .185. This interaction can be seen in Table 3.3. While this interaction was significant in all three days analyses (as indicated in Table 3.2), the differences between match and mismatch conditions on Day 1 (189 msec.) was greater than on Day 3 53 TABLE 3.2 Significant Results of the Analyses of Variance in Experiment 2 Condition Overall 'Day;l_ EEELJE IEDLL: Term 1 M * Negation 1 M Term 2 * * * * Negation 2 * * * * Day * Term 1 x Term 2 t * * * Negation 1 x Day * Term 2 x Negation 2 * * M _Term 1 x Negation 1 x Term 2 * Term 1 x Term 2 x Negation 2 * * * * Term 1 x Term 2 x Day * Term 1 x Negation 1 x Term 2 x Negation 2 * * t * Term 1 x Negation 1 x Term 2 x Day * Where * denotes a significant result, p_‘<.05; and M denotes a marginally significant result 54 TABLE 3.3 Mean RT in Msec. to Verify Inner String Matches and Mismatches Term 2 Term 1 On Time1 Late1 Mean On Time2 1550 1713 1632 Late2 1585 1470 1527 Mean 1568 1470 TABLE 3.4 Mean RT in Msec. to Verify Inner String Matches and Mismatches Over Days Day Day 1 Day 2 Day 3 Mean Match 1606 1490 1435 1510 Mismatch 1795 1626 1527 1649 Mean 1705 1558 1481 55 (92 msec.), Term 1 x Term 2 x Day, §(2,22) I 10.862, MSe I .010, as seen in Table 3.4. While the main effect of Negation was not significant in the over— all analysis, [(1,11) I 1.015, MSe I .035, there was a significant Nega- tion 1 x Day interaction, 2(2,22) I 4.056, MSe I .008. The effect of Negation 1 was marginally significant on Day 1, F(1,11) I 4.145, MSe I .02, but decreased over days. The two-way Negation 2 x Term 2 interaction was also significant, 2K1,11) I 14.132, MSe I .010. As can be seen in Table 3.5, the presence of a Negative decreased the difference between On Time and Late 2 2 2 which existed in the affirmative conditions (On Time -Late2 I 136 msec.; 2 vs. Not On Time ~Not Late2 I 73 msec.). 2 Negation 2 was also found to interact with inner string congruity, Term 1 x Term 2 x Negation 2,IE(1,11) I 21.047, MSe I .024. As indicated in Table 3.6, the difference between Affirmative2 and Negative2 con- ditions was greater when inner strings did not match (358 msec.) than when they were identical (241 msec.). The remaining significant interactions were the following fourIway ones: Term 1 x Negation 1 x Term 2 x Negation 2, E(1,11) I 39.599, MSe I .883; and Term 1 x Negation 1 x Term 2 x Day, F(2,22) I 9.611, MSe I .009. As can be seen in Table 3.2, the latter interaction repre- sents the fact that the three-way interaction, Term 1 x Negation l x Term 2, was significant in the daily analyses only on Day 3. In summary, comparisons which contained an 0n Timez, a Negativez, an inner string mismatch, or some combination of these factors were more time consuming for the subjects to process than their counterpart conditions (Latez, Affirmativez, inner string match, etc.). 56 TABLE 3.5 The Effect of Negation on RT in Msec. in On Time2 and Late2 Conditions Negation 2 Term 2 On Time2 Late2 Mean Affirmative2 1498 1362 1430 Negative2 1766 1693 1730 Mean 1632 1528 i TABLE 3.6 The Effect of Negation of RT in Msec. to Conditions Containing Inner String Matches and Mismatches Negation 2 Terms 1 and 2 Match Mismatch Mean Affirmaitve2 1390 1470 1430 Negative2 1631 1828 1730 Mean 1511 1649 57 '2 lug. Mean RT's per condition for Day 3 performance is contained in Table 3.8. Since the subjects's protocols did not provide information as to what kinds of strategies they may have adopted throughout pro- cessing, all the subjects' data were pooled. The overall error rate for Day 3 performance was 2.4%. A 2 x 2 x 2 x 2 within-subjects analysis of variance was performed on the data. The variables were: Term 1 (On Time1 vs. Latel), Nega- tion 1 (Affirmative1 vs. Negativel), Term 2 (0n Time2 vs. Latez), and Negation 2 (Affirmative2 vs. Negativez). Significant results of this analysis are contained in Table 3.2. As in the overall analysis, conditions containing On Time2 and a Negative2 were more time consuming than Late2 and Affirmative2 judg- ments, Term 2, {(1,11) I 22.378, MSe I .032, and Negation 2, F(1,ll) I 156.468, MSe I .032, respectively. The main effect of Term 1 was also significant, £(1,11) I 14.231, MSe I .008, indicating that On Time1 conditions were more difficult to process than Late1 conditions. Inner string mismatches required significantly more processing time than inner string identities (Mean RT difference for Mismatch- Match comparisons I 92 msec.), Term 1 x Term 2, F(1,11) I 9.09, M58 I .044. The effect of a Negative2 was also found to be greater among inner string mismatch conditions (Mean RT difference between Negativez- Affirmative2 I 396 msec.) than among inner string identities (Mean RT difference between Negatviez-Affirmative2 I 246 msec.), Term 1 x Term 2 x Negation 2,.§(1,11) I 21.409, MSe I .013. The fourIway Term 1 x Negation 1 x Term 2 x Negation 2 interaction, E(1,11) I 15.832, MSe I .018, was also significant. 58 The Term 1 x Negation 1 x Term 2 interaction was the remaining significant effect, 211,11) I 14.784, MSe I .013. In those conditions in which the inner strings matched, Negative1 conditions were longer than Affirmative conditions by 107 msec. However, in conditions in 1 which the inner strings did not match, Affirmative conditions were 1 148 msec. longer than Negative1 judgments. This is a most interesting finding. To begin with, it is an aspect of the subjects' processing that is seen only in Day 3 performance. Furthermore, it suggests that when a negative appears in Sentence 1, ‘ggd_an inner string mismatch occurs, a subject may opt for a strategy that would ease his processing load, perhaps a conversion strategy. Consider the following encoded comparison (25). The Term 1 x (25) false (on time) (late) Negation l x Term 2 interaction suggests that "false(on time)" would be converted to "(late)"; the inner strings would then be compared and finding no mismatch, a true response would be executed (thus, no addi- tional time for a truth index change would accrue). This operation is to be contrasted with the true model of negation's comparison of both inner and outer strings, resulting in two changes in the truth index and an increase in processing time. It is important to note, however, that the Term 1 x Negation 1 x Term 2 interaction does not specify 2222 the conversion operation takes place. It is possible that after both sentences are encoded, the subject will convert Sentence 1. It would be equally plausible to assume that the subject encodes a Negativel sentence as an affirmative and then continues processing. It may also be the case that a subject has both 59 affirmative and negative representations equally available, the one chosen being dependent upon the second representation. We shall now examine the results of the Day 3 analysis in light of the pilot study's findings. Evaluation.g£ the Pilot Model. Recall that in the pilot study data TA < FN < FA < TN in Affirmativel—Assertion conditions. In the present experiment when Negative1 entries were excluded from analysis, TA (1255 msec.) < FA (1412 msec.) < EN (1562 msec.) < TN (1715 msec.). The inclusion of Negative1 conditions yielded the following ordering: TA (1255 msec.) < FA (1412 msec.) < TN (1519 msec.) < FN (1567 msec.). This finding was also not consistent with the results of the pilot study (TA < FA = TN < FN). Thus a discrepancy exists between the pattern of RT's to equivalent conditions of the pilot study and Experiment 1. Table 3.7 contains predictions of the pilot model regarding the ordering of certain conditions. Within each row of comparison, as one moves from left to right, RT is predicted to increase by virtue of adding one more parameter. As can be seen in the table, the pilot model encounters some difficulty in predicting conditions containing either a d_or a §_parameter. The model errs in predicting an increase in pro— cessing due to g_when the preceeding condition contains a double nega- tive. As for parameter 3, when k_is associated with an On Timez, errors in prediction occur. Furthermore, the pilot model predicts that false double negative Assertions will be the most difficult condition. However, in Day 3 performance, the Latel-Not On Time condition (1811 msec.) was more time 2 consuming than either false double negative Assertion (Not On Timel- Not Latez, 1683 msec.; Not Latel-Not On Timez, 1784 msec.). Pilot Model Predictions and Obtained Values in Msec. 60 TABLE 3. 7 for Selected Conditions of Day 3 Latel-Late2 t o 1189 Latel-Late2 t o 1189 Latel-Late2 t o 1189 Latel-Late2 t o 1189 On Time t +a 0 1320 On Timel- On Timez t +a 0 1320 On Timel- On Time t +a 0 1320 On Timel— On Time2 t +a 0 1320 2 2 Latel-On Time2 t +c o 1499 Not Latel-On Time t +k o 1354 Not Latel-Not Late2 t +b 0 1528 Not Latel-Late2 to+d 1315 Not On Timel- On Time2 t +a+d o 1424 Latel- On Time2 t +a+c o 1499 Not On Timel- Not On Time t +a+b o 1581 Not Latel- On Time2 t +a+k o 1354* 2 2 Latel-Not Late2 to+b+d 1564* Latel-Not Late to+b+d 1564 On Timel- Not On Time to+a+b+d 1559 2 On Timel- Not On Time t +a+b+d o 1559* 2 2 Latel- Not On Time2 to+a+b+c+d 1811 Latel- Not On Time2 to+a+b+c+d 1811 Where * denotes an inaccurate prediction 61 Given these considerations, the model constructed from the results of the pilot study was rejected as a candidate for Experiment 2 data representation. An attempt was made to characterize the data using the Young and Chase (1971) conversion model. However, this also proved fruitless. The model stipulates that all negatives would be converted to their equivalent affirmative form. As such, one would predict that, for example, Latel-Not On Time2 < Latel-Not Latez, or that On Timel- Not Late < On Time -Not On Time . Consulting Table 3.8 one will see 2 1 2 that this is not the case (1811 vs. 1564 msec.; and 1619 vs. 1581 msec., respectively). Therefore, this approach was also abandoned. Development 2£;§_New Model 2f_Information Processing, It was decided to test a modified version of the Experiment 1 model in which i_is eliminated (no encoding time for Inferences) and in which §_is eliminated (since there is no i, it would not be necessary to have a s parate truth index changer for the Assertions alone). This version is pictured in Figure 3.1. Six parameters are proposed: Eo’ a base processing time common to all comparisons; in accordance with the main effect of Term 2, parameter a will represent the difference between Late2 and On Time2 encoding times; parameter b will represent the amount of time needed to encode a Negativez; parameter §_will represent the additional time taken to encode a Negative2 in the presence of a Negativel (that is, fhfg); parameter g_will represent the amount of time taken to detect an inner string mismatch and change the value of the truth index; and parameter d_will represent the amount of time needed to change the value of the truth index upon finding an outer string mismatch. 62 N unusauoexm mo Home: ecu mo mo>wuoshmuad wofloooooum @ Av+v ouHooeao muH on wound guano mas «o osHmp use owsoso On— omHoMJ‘ vooeoom_ no» —1 o Idtwoon sue 0M ensues mwowuuo nouso one on «myv ouHa loano new on novaH sauna may we osHo> one swoono .H.m ouawwm _ Amyv ouwmomno muH on Movow soon» mas «o 0=Hu> use swoonu q saunas Nausea mwsHHum Houno one on ooh mmowuum Hosea one on meme .Hm>Huaamz wow o>Huowoz m my .H0>Huamuz «H _mr .NaaHH no «H “N ovooom mooHuousomoueom o>Huowoz use ueHuuauaww< neon "H mucosa 63 Processing time will be divided into four stages: Encode Sentence 1; Encode Sentence 2; Compare the encoded representations; Execute a response. According to an additive factor model, each stage is presumed to be independent of each other. In Stage 1, subjects will encode Sentence 1. This particular stage is not usually the subject of detailed discussion; rather, the proposed internal representation is generally the only feature of this stage which is mentioned, for example (26). (26) . . . not 22 time today. false (on time) Recall that the Term 1 x Negation 1 x Term 2 interaction suggested that subjects were using some sort of conversion strategy to ease pro- cessing demands. However, the rejection of a complete conversion model (that of Young and Chase) indicates that subjects do not convert nega- tives in every case. The following interpretation of these data was drawn. When encoding a Negative sentence, it is proposed that two 1 representations are constructed: the negative and its affirmative equivalent. Both representations are proposed to be equally available. Thus, one need not propose a conversion parameter to account for pro- cessing time. Which representation will be used in the comparison will ‘ be entirely determined by the nature of Sentence 2. Post—hoc examina- tion of the Day 3 data indicated that subjects opted for the affirmative representation in the following two conditions: Not On Timel-Late2 and Not Latel—On Timez. It thus appeared that whenever an affirmative term was encoded in Sentence 2 which matched the affirmative represen- tation in Store 1, subjects opted for the affirmative representation. 64 Sentence 2 will be encoded immediately following the end of Stage 1. It is only during this stage that the effects of parameters 3, b, and‘g_are seen. It is also important to note that the polarity of Sentence 1 differentially affects the encoding of Sentence 2. If Sentence 1 is affirmative, encoding a Negative2 will take f2 msec.; if a Negative1 is stored in representation 1, encoding a Negative2 will require fbfg msec. more. With the completion of the encoding phases, comparison operations to determine truth validity are commenced. It is proposed that the subject will first compare the inner strings of the representations. Finding an inner string mismatch and changing the value of the truth index will increase processing by time f5. The outside strings will then be examined, with a mismatch and index change increasing processing time by time fig, The value of the truth index at the end of Stage 3 is then input to Stage 4 at which point a response (pushing a true or false button) is executed. There are three major differences between the models of Experiments 1 and 2. First of all, with the elimination of the Inference condition, there was no longer any need to include a fi_encoding parameter; simi- larly, there was no longer any need to include f3 which operated dif- ferentially on Assertions and Inferences. The major difference between the models, and one which cannot readily be interpreted as the result of the exclusion of Inference conditions, is the generation of two Sentence 1 representations in Stage 1. It could be the case, however, that a lighter processing load (i.e., no Inferences) increased the likeli- hood that the subjects would generate more than one representation. 65 A least squares analysis using mean RT per condition was employed to estimate each of the proposed parameters. Parameter estimates to the nearest msec. were: _t_o, 1210; _a_, 111; b, 191; _c_, 156; _d_, 104; and E» 110. Observed scores, condition estimates and their deviation can be found in Table 3.8. The RMSD was calculated in order to estimate the goodness of ift of the model. Since the RMSD was equal to 32 msec., and all parameter estimates were found to be well above this value, the model was interpreted as being a reliable characterization of subjects' processing (see Sternberg, l969a,b). As a further test of the parameters' reliability, an attempt was made to relate analysis of variance effects to the proposed parameters. Related measures tftests were also performed where necessary. Parameter a, the time to encode an On Timez, was significant according to the Term 2 main effect, as well as parameter b_(Negation 2 main effect), parameter 2 (Term 1 x Term 2 interaction), and paramter d (5(11) I 4.200, p.< .01). Unfortunately, the effect of parameter 2 could not be isolated in an interaction or by a related measures tftest. Upon finding that parameter 2_could not be isolated and tested for its significance, a least squares analysis using only parameters 30’ g, b, g, and _d was performed on the data. This is, of course, the Clark and Chase (1972) model of negation. However, elimination of'g resulted in a negative d_value, a totally unacceptable result. Thus, the Clark and Chase model was abandoned in favor of maintaining 2, While the model is able to represent the data of Experiment 2, there is a flaw in this approach. The model itself violates a basic assumption of the additive factor framework in at least two instances. The additive factor model has as its root the notion that each stage of Mean RT in Msec. Per Condition, Predicted Scores, and Their Deviations for Day 3 Performance 66 TABLE 3.8 Conditions Predicted Predicted Observed Deviation Parameters 3: ¥:::' to+a 1321 1320 + 1 gztzime‘ to+c 1366 1325 + 41 SgtTéEeE1me to+a+b+d 1616 1559 + 57 gth:::; t°+b+c+d 1661 1619 + 42 git;;me to+a+c 1478 1499 - 21 :222’ to 1210 1189 + 21 figtegn Tiwe to+a+b+c+d 1772 1811 - 39 fizfieiate to+b+d 1505 1564 - 59 gthggeTime- to+a+d 1425 1424 + 1 52:60“ Time- to 1210 1220 - 10 :3: 8: $1::' to+a+b+n 1621 1581 + 40 is: fiztzime" to+b+c+n 1667 1683 - 16 EZ‘TEZEE‘ to+a 1321 1354 - 33 EZEeLate' to+d 1314 1315 - 1 :3: 33t;;me to+b+c+d+n 1778 1784 - 6 :2: t:::’ to+b+n 1510 1528 - 18 67 processing is independent of every other stage. However, in order to represent the data, the model posits that the final representation of Stage 1 will not be chosen until Stage 2 is completed. Thus, Stage 1 is dependent upon Stage 2. Furthermore, the encoding of a Negative is dependent upon the 2 contents of Stage 1. If an affirmative is in Store 1, encoding Negative2 requires additional time fb_msec. If, however, a Negative1 appears in Store 1, encoding is slowed down by f2 msec. (that is, Negative2 requires fbfg_msec.). In this experiment, we have been able to represent the data by a model. However, given that we have violated a basic tenet of the serial process position, one must be cautious in concluding that the additive factor approach is an adequate framework in which to interpret the results of all sentence-sentence verification studies. Summary Using an auditory sentence-sentence verification task to test people's comprehension of Assertions, it was found that On Time2 compari~ sons were more difficult than Late2 comparisons, and that comparisons containing a Negativez, or an inner string mismatch were more time con- suming than Affirmative2 or inner string identity comparisons, respect- ively. These findings, in conjunction with the findings of Experiment 1, provide support for the psychological reality of these factors. It was also the case that aSpects of Sentence 1 played a larger role in subjects' processing in Experiment 2. It was suggested that as processing capacity was freed during the task, subjects' RT's may have become more sensitive to new stimulus parameters. 68 An attempt was made to model the data of Experiment 2 according to an additive factor model. While a model was constructed that was able to capture the overall condition orderings, problems inherent in the model force one to interpret with caution the modeling attempts of Experiment 2. CHAPTER 4 Experiment 3 The results of Experiment 2 indicated that even simple auditory language comprehension tasks involve more complex operations than the sentence—picture tasks. A feature of the preceding experiments which may be responsible for those findings is the nature of the task used. In both Experiments 1 and 2, auditory presentations were employed. Recall, however, that earlier sentence-picture models (Clark & Chase, 1972; Young & Chase, 1971) were based on RT to visual stimuli. There- fore, it is possible that the differences in processing that were found in Experiments 1 and 2 are the result of modality differences rather than the result of the use of complex language materials. Experiment 3 was designed to investigate sentence-sentence veri- fication processing using a visual RT task. With the exception of the change in stimulus modality, Experiment 3 is essentially a replication of Experiment 2. Method Subjects. The subjects were 12 male and female undergraduates participating for credit in an introductory psychology class at Michigan State University. Stimuli. Two hundred fifty-six declarative sentences in pica-size type print (black print, white background) appearing on slides served as 69 70 stimuli, and were presented in two sentence test sequences. Sentence 1 was a simple declarative statement of fact. One-half of the sentences contained the term On Time, the remaining sentences included Late. Sentence 1 also varied as to the presence or absence of negation: one- half of the sentences were affirmative, one-half, negative. Sentence 2 was a simple verification sentence varying with respect to the same fac- tors manipulated in Sentence 1. The design yields 16 possible combina- tions (see Table 3.1, page 51). Apparatus. Two slide projectors arranged vertically were used to project slides onto a screen located four feet in front of the subjects. Projector 1 projected Sentence 1 onto the top of the screen; Projector 2 projected the second sentence onto the lower portion. Sentences were separated by 18 inches. Automatically operated Hunter timers were used to record the subjects' RT's, recordings beginning concurrently with each stimulus onset. A three button panel was located directly in front of the subject: a button marked "C" (Comprehension) was centrally located, with a "T" (True) or "F" (False) button positioned to either side. Depression of the "C" button activated Projector 2. Depression of any button stopped the Hunter timers. Procedure. All the subjects were individually tested in three performance sessions. Order of the stimulus presentations was randomly varied over the three sessions so that no subject received the same input order twice. All instructions were administered verbally by the Experimenter. Participants were instructed to carefully read each sentence presented, in order to make accurate true-false judgments of each sentence-sentence pair. The Experimenter initiated Trial 1. The subjects were required to depress the "C" button as quickly as possible 71 following the presentation of Sentence 1, indicating that they understood the sentence. Sentence 2 immediately followed. The subjects were in- structed to depress either the T or F button, depending upon their eval- uation of the comparison. The position of the T and F buttons was ran- domly varied between subjects. Both the speed and accuracy of the res- ponse were emphasized. A five second interval occurred between each test comparison. The Experimenter recorded all responses. Eight practice trials were administered prior to actual testing. The subjects were advised of their accuracy before proceeding to the actual test. In addition, prior to each session, the subjects were advised of their previous performance. Upon completion of the third session of 128 sentence comparisons, the subjects were asked to specify strategies used during processing, if any. The subjects were also asked to indicate whether they were aware of any change in the strategy selected across the three sessions of testing, to state the nature of this change, and the proposed reason for it. Results and Discussion The significant results of an analysis of the Experiment 3 data across all three days, and of a separate analysis of the Day 3 data will be presented and discussed in the following section. As in the previous experiments, an attempt will be made to interpret the Day 3 data in terms of an additive factor model. Day 3 data was specifically examined because it was felt that if the subjects were processing information according to a common strategy, this pattern would most likely be seen, and be most stable, in the most practiced Day 3 data. No specific discussion will be made of the results of the Day 1 and Day 2 performance analyses, as all the significant results of these two 72 performances were significant on Day 3. However, the mean RT's per condition for Day 1 and Day 2 can be found in Appendices G and H, res- pectively. The significant §_values of the within-subject analyses of variance are reported in Appendix I. Overall. Only correct responses were included in the analysis; individual data i_3 standard deviations from the mean were interpreted as errors. The overall error rate was 6.2%. A 2 x 2 x 2 x 2 x 3 within-subject analysis of variance was per- formed on the data. The factors were: Term 1 (On Time1 vs. Latel); Iegation 1 (Affirmative vs. Negativel); Term 2 (On Time vs. Latez); 1 2 Negation 2 (Affirmative2 vs. Negativez); and Day (1, 2, or 3). The significant results of this analysis, where p_< .05, are indicated in Table 4.1. The significant results of the analyses on the Day 1, Day 2, and Day 3 data are also included in this table. In general, conditions which contained a negative in either Sentence 1 or 2 were more time consuming than their affirmative counterpart con- ditions, Negation 1, F(1,11) I 24.476, MSe I .163; and Negation 2, 211,11) I 47.816, MSe I .186. Although the Negation 2 main effect was significant in each separate day analysis, the difficulty of a Nega- tive2 was greater on Days 1 and 2 than on Day 3, Negation 2 x Day, 2(2, 22) I 4.958, MSe I .070 (Mean RT difference between Negative IAffirma- 2 tive2 for Day 1 I 328 msec.; for Day 2 I 260 msec.; and for Day 3 I 158 msec.). As found in Experiments 1 and 2, On Time2 conditions were more time consuming than Late conditions, Term 2, F(1,11) I 47.552, MSe I .037. 2 The subjects' performance was also found to become significantly faster over days, Day, §(2,22) I 36.180, MSe I .249. 73 TABLE 4.1 Significant Results of Analyses of Experiment 3 Condition Overall M PM P_ay_3_ Term 1 * Negation 1 * e * * Term 2 * * e * Negation 2 * s * * Day * Term 1 x Term 2 * * * * Negation 1 x Negation 2 M Term 2 x Day M Negation 2 x Day * Term 1 x Negation 1 x Term 2 M Term 1 x Negation 1 x Negation 2 * Term 1 x Term 2 x Negation 2 * * * * Term 1 x Negation 1 x Term 2 x Negation 2 * t t * Term 1 x Term 2 x Negation 2 x Day * Where * denotes a significant result (p_< .05); and M denotes a marginally significant effect 74 As indicated in Table 4.2, inner string mismatches were more dif- ficult than inner string identities, Term 1 x Term 2, {(1,11) I 62.373, EMSe I .628. Furthermore, the effect of a Negative2 was greater among the mismatch conditions than in identity conditions, Term 1 x Term 2 x Negation 2, Ffil,1l) I 21.501, MSe I .200. This effect is illustrated in Table 4.3. While the Term 1 x Term 2 x Negation 2 interaction was significant in all performances (see Table 4.1), the effect of a Negative2 on the difference between Match and Mismatch conditions was greater on Day 1 (Mean RT difference of Mismatch-Match I 499 msec.) than on either Day 2 or Day 3 (280 msec., and 256 msec., respectively), Term 1 x Term 2 x Negation 2 x Day, §fi2,22) I 4.696, MSe I .045. The remaining significant result of the overall analysis was the fourIway Term 1 x Negation 1 x Term 2 x Negation 2 interaction, .§(1,11) I 31.822, MSe I .285. This effect was also observed in the analyses of the Experiment 2 data. .Eézuér Examination of the subjects protocols did not yield a suitable criterion by which to divide the subjects' performances ac- cording to strategies employed. Therefore, all the subjects' data were pooled for the Day 3 analysis. Mean RT's for each condition are con- tained in Table 4.6. The overall error rate for the Day 3 performance was 5.12. A 2 x 2 x 2 x 2 within-subjects analysis of variance was performed on the data. The factors were: Term 1 (On Time1 vs. Latel); Negation 1 (Affirmaitve 1 vs. Negativel); Term 2 (On Time vs. Latez); and Nega- 2 tion 2 (Affirmative2 vs. Negativez). The significant results of this analysis, where p_< .05, are indicated in Table 4.1. 75 TABLE 4.2 Mean RT in Msec. to Inner String Identities and Mismatches Term 2 Term 1 On Time1 Late1 Mean On Time2 1555 2042 1799 Late2 1966 1410 1688 Mean 1761 1726 TABLE 4.3 Effect of a Negative on RT in Msec. to Inner String Match and Mismatch Conditions Negation 2 Inner Strings Match Mismatch Mean Affirmative2 1445 1794 1620 Negative2 1521 2215 1868 Mean 1483 2005 76 As in the overall analysis, conditions containing either a Negative1 or a Negative were more time consuming than Affirmaitve and Affirma-. 2 1 tive2 conditions, respectively, Negation 1,'§(1,11) I 14.930, MSe I .096; and Negation 2, Ffi1,11) I 23.438, MSe I .051. On Time2 conditions were also found to be more difficult than Late conditions, Term 2, Efil,11) I 2 7.660, MSe I .033. In the Day 3 analysis, however, conditions con- taining an On Time in Sentence 1 were also more time consuming than those containing Late Term 1, E(1,11) I 9.581, MSe I .042. 1’ Inner string mismatches were more time consuming than inner string identities, Term 1 x Term 2,'E(1,11) I 47.339, MSe I .271. Once again, the effect of a Negative2 was greater upon the mismatch conditions conditions I 287 (Mean RT difference between Negative -Affirmative 2 2 msec.) than the inner string identities (30 msec.), Term 1 x Term 2 x Negation 2, §(1,11) I 11.461, MSe I .069. The Term 1 x Negation 1 x Negation 2 interaction was also signifi- cant, §(1,11) I 5.572, MSe I .023. As indicated in Table 4.4, the difference in mean RT between outer string mismatches and identities was greater when an On Time appeared in Sentence 1 (Mean RT difference between identities and mismatches I 131 msec.) than when a Late1 ap- peared there (Mean RT difference I 27 msec.). The four-way Term 1 x Negation 1 x Term 2 x Negation 2 interaction was the remaining significant result, §(1,11) I 10.806, MSe I .211. The two-way Negation 1 x Negation 2 interaction was also marginally significant, £(1,11) I 4.112, MSe I .072. Comparison pf_Experiment'§ with Experiments l_and.2, In all three experiments which we have discussed, certain aspects of processing were constant across a change in task demands and a change in stimulus 77 TABLE 4.4 Effect of the Level of Term 1 on the Verification of Outer String Matches and Mismatches in Msec. Term 1 Outer Strings Match Mismatch Mean On Time1 1531 1662 1597 Late1 1492 1519 1506 Mean 1512 1591 modality: On Time2 conditions were more difficult than Late2 conditions; conditions containing a Negative2 were more difficult than those con- taining an Affirmativez; and inner string mismatches were more diffi- cult than inner string identities. we saw, however, that as the processing load of the subject was reduced, in terms of the task demands of Experiment 1 and Experiment 2 (i.e., the removal of Inference conditions), the polarity of the first sentence came to play an important role in processing. With the change from auditory to visual stimuli, it is now the case that 2252 the term and the polarity of Sentence 1 have important effects on subsequent processing. In Experiment 3, Sentence 1 was present during the presentation of Sentence 2. Therefore, it is possible that the effect of Term 1 was really the result of its continuous presence, rather than the result of it being a visual stimulus. Contrasting the results of the present experiment with a visual RT experiment in which the first sentence is removed from view following comprehension would 78 provide some insight to this question. In terms of this study, however, the key feature to note is that aspects of Sentence 1 play a greater role in determining the ultimate processing time. Another interesting feature of the change from an auditory to a visual modality task is the increased importance of pattern matching. In Experiment 3, by Day 3 311 conditions in which Sentence 1 and Sentence 2 matched were faster than any other condition. This was not the case in the previous experiments. In fact, in the earlier studies On Timel-Late2 mismatches were generally as fast or faster than On Timel-On Time2 identities. (Applicability;2£.Past Models: The Pilot Study. Recall that in the pilot study the following ordering of data among Affirmativel-Asser- tions was found: TA < FN < FA < TN. In the present study on Day 3, TA (1083 msec.) < FN (1409 msec.) < FA (1611 msec.) < TN (1758 msec.). This same ordering was obtained with the Day 1 data. Inclusion of Negative1 conditions produced the following ordering of the pilot study conditions: TA < FA I TN < FN. Among the Experi- ment 3-Day 3 data, TA (1083 msec.) < TN (1562 msec.) < FA (1611 msec.) < EN (1677 msec.). The ordering was also found among the Day 1 data. While the latter comparison did not match the results of the pilot study, it must be noted that for at least the Affirmativel-Assertion conditions, subjects in the pilot study and subjects in Experiment 3 seemed to be responding in the same manner. It should also be noted that the false double negative Assertion which was the most difficult condition of the piLot study was also the most difficult condition of the Experiment 3, Day 1 and Day 3 data. 79 The Young and Chase Conversion Model. Recall that this conversion model predicted that a negative entry would automatically be encoded as its affirmative equivalent. As such, if a negative were in Store 1, it should automatically be changed to an Affirmativel; for example, Not On Time1 would be converted to Latel. Therefore, one would predict that there would be no difference in RT between Not On Timel-Late2 comparisons and Latel-Late2 comparisons. Table 4.5 contains some of the predictions that this model would make. As can be seen from the table, in none of the cases cited do the results of Experiment 3 match the predictions. Therefore, the Young and Chase conversion model was abandoned as a candidate for characterizing the results of Experiment 3. TABLE 4.5 Predictions of the Young and Chase (1971) Conversion Model and the Actual Results of Experiment 3 Prediction Obtained Times (Msec.) Eztegn Timel- . £2222 1754 # 1011 1:232; = 2:: 312:; * Eztelatel 3 0n Timel 1379 7 1541 2 Late2 80 The True Model. The true model of negation predicts that TA < FA < FN < TN. As indicated in the evaluation of the pilot model, this or— dering was not obtained. Furthermore, the true model predicts that con- ditions which contain both an inner string and an outer string mismatch will be more time consuming than false double negative Assertions. This was not the case in Day 1, 2, or 3. Thus, the true model of negation was also rejected. The Models pf Experiments l_and.2, Recall from Experiment 1 that Affirmativel-Assertion verifications were ordered in the following manner: TA < FA < FN < TN, while the inclusion of Negativel-Assertions yielded: TA < FA < TN I FN. In Experiment 2, TA < FA < FN < TN among Affirmative -Assertions and TA < FA < TN < FN when Negativel-Assertions 1 were included in the comparison. Since none of these orderings match the obtained condition orderings of Experiment 3, it must be concluded that neither model (constructed to characterize each experiment's pat- tern of responses) will be able to accurately account for the data of the present experiment. Thus, both models were rejected as possible can- didates. The Development of _a_ New M25131. Although the pilot study model proved to be a better characterization of Experiment 3 performance than any of the other models considered, its failure to adequately charac- terize condition orderings including Negative -Assertions indicated that 1 certain aspects of processing were being overlooked. Hence, an attempt was made to construct a model which would more accurately reflect features of the subjects' processing. Past models of information processing have all concluded that it takes one significantly longer to encode a negative than an affirmative; 81 it should, therefore, take longer to encode "Not On Time" than "On Time", and encoding "Not Late" should take more time than "Late". In the present experiment since RT was recorded from the onset of Sentence 2, it should be the case that: On Timel-Not On Time2 RT > Not On Timel- > _ 1 2 RT Not Late1 Late2 2 RT > Not On Timel-Late2 RT; and that On Time -Not Late RT 1 2 > Not Latel—On Time2 RT by at least a factor of encoding a negative, or, On Time RT; that Late -Not Late RT; that Latel- 2 Not On Time as referred to in most models, fp_(that is, the second comparison of each pair should be longer than the first due to the presence of a nega- tive in Sentence 2). Inspection of Table 4.6 will reveal that this ordering was only true of the On Timel-Not Latez, Not Latel-On Time2 judgments. Thus, any model which attempts to characterize the data of Experiment 3 will have to deal with this result. The finding that three pairs of the conditions were the reverse of what was predicted or that members of the pairs were equivalent is counterintuitive. The only explanation which seems feasible is that since Sentence 1 remained in view during the entire course of processing, the subjects encoded Sentence 2 and then returned to Sentence 1 to carry out further operations. Thus, if a negative appeared in Sentence 1, additional time would be reflected in the overall RT for the re- encoding of Negativel. If this assumption is correct, then if one were to contrast a condition in which Sentence 1 is affirmative and Sentence 2 is negative, with the reverse condition where Sentence 1 is negative and Sentence 2 is affirmative, there would be little difference in processing time, as the Negative2 of the former condition would cancel out the Negative1 82 of the latter. The major problem with this explanation is that it does not explain why On Timel-Not Late2 > Not Latel-On Timez. That is, there is no a priori rule by which to explain subjects' processing. Keeping these findings in mind, as well as the failure of other models to predict the results of Experiment 3, an attempt was made to construct a model which would characterize the Day 3 results. A four- stage, five parameter model was devised. The processing alternatives of the model are featured in Figure 4.1. Information processing is divided among two encoding stages (a separate stage for each sentence), a comparison stage, and a response execution stage. In Stage 1, the subjects encode Sentence 1. It is assumed that the subjects do not perform any conversion operations during this stage, given that the term and polarity values of Sentence 1 may be rechecked at any time during the verification process. Further- more, the three—way Term 1 x Negation 1 x Term 2 interaction, which suggested alternative representations in Experiment 2, was not sig- nificant in the Experiment 3-Day 3 results. Upon completion of Sen- tence l encoding, the subject will begin Stage 2 processing. Since Sentence 1 encoding is not part of the RT of interest in the present study, no parameters accounting for its time course are proposed. In Stage 2, Sentence 2 is encoded. It is proposed that sentences containing On Time2 will take f§_msec. longer to encode than Late2 sentences, as suggested by the Term 2 main effect. In trying to fit the data to particular models, it was noticed that the Not On Timer-Late2 and Not On Timel—Not Late2 conditions were usual- ly poorly predicted. In each case, the error in the predictions was enough to suggest that another parameter was involved. In order to 83 m unusauoaxm mo Homo: one mo mo>HuosHoua< msawmoooum awash voonmom at 38. Ioneo new on NovsH guano oeu mo ooHo> osu swooeu psouwe wwnwuum Hausa sea on «mama woodman _r||._HIL mun ouwooamo muH ou Macaw sunny onu mo osHo> one owomnu Av+v ouamomno mow ou nova“ Sasha can «0 o=Ho> onu owoosu .H.s ouomHH 9. onus vooamom no» Sousa f1 2,3953% ou amuse o>Huowoo woo uuo>cou «may a.Ho« ~o>Huowoo o mowuuoo was «0 mac Ofl mmsHHum House on on» on ooh psouwa mwcauum Hosea use on Gun s+ Nessa t NsHoH «H was mafia no 002 HH .mv .HoaHH no «H “N ocooom ‘ 4 _1H ovoosu _ 84 accommodate this problem, paramter'p_is proposed. If a subject has stored both a negative and an On Time in Store 1, and if the visual pro- perties of the Term 2 factor are different from Term 1, second stage encoding is believed to be prolonged by {a msec. The current model does not propose a negation encoding parameter. As suggested earlier, the effects of negation encoding tend to wash out over comparable conditions, for example, Not On Time -On Time2 and 1 On Timel-Not On Timez. However, an attempt was made to model the data using a negative encoding parameter which would capture Negative en- 2 coding and the re-encoding of Negativel. The effort was unsuccessful, resulting in negative estimates of both £0, the base processing time, and d, the outer string mismatch parameter. A model showing no dif- ferential effects of negation was therefore decided upon. Once the encoding of the second sentence is completed, the subject will begin Stage 3 or the comparison process. The inner strings of the two representations will first be compared. If a match is found, the values of the outer strings will be compared. If a match is again found, the subject will respond True. This comparison, for example (27), is (27) (late) t (late) 0 proposed to take‘_1_:O time, a base processing time. If, however, a mis- match of the outer strings is detected, as in (28), it is proposed that (23) (late) false (late) to+d the subject will change the value of the truth index to False, a process requiring‘fd,time. At the time of the inner string comparison, if the subject had found a mismatch, it is proposed that he would have then checked to see 85 if one of the entries was a negative. A.no response (29) would result (29) (on time) (late) to+c in the subject changing the value of the truth index to False, an operation requiring f2 msec. If a negative had been found, it would have been converted to its affirmative equivalent (fk_time) and the outer strings would then be checked. A match in this case (30) would (30) (on time) false (late) to+k result in a true response. A mismatch (31), however, would result in a (31) false (on time) false (late) to+n+k+d change of the truth index value, adding fd_msec. Finally, in Stage 4, the value of the truth index is indicated by the depression of a True or False button. A least squares analysis was performed on the actual data in order to arrive at parameter estimates. The analysis yielded the following estimates to the nearest msec.: 30, 1076; a, 103; g, 483; _c_1_, 329; lg, 598; ande, 121. Observed scores, predicted values for each condition and their deviation from the actual case can be seen in Table 4.6. The RMSD was obtained to test the goodness of fit of the model, RMSD I 66 msec. Since the RMSD was lower than any of the predicted parameters, the parameters and their condition estimates were believed to be accur- ate reflections of the subjects' processing. An attempt was made to test the significance of the individual parameters. Parameter g_corresponded to the significant Term 2 main effect, and parameter p was reflected in the significant Term 1 x Term 2 interaction. While parameters g_and‘§_were not reflected directly in the analysis of variance effects, related measures pftests did show Mean RT in Msec. Per Condition, Predicted Scores and Their Deviation 86 TABLE 4.6 for Day 3 Performance Condition Predicted Observed Predicted Deviation Parameters On Time- On Time to+a 1154 1179 + 25 On Time- Late to+c 1541 1559 + 18 On Time- + .— Not On Time to a+d 1513 1508 5 On Time- + - Not Late to k 1814 1674 140 Late- On Time to+a+c 1680 1662 — 18 Late- Late to 1011 1076 + 65 Late- Not On Time to+a+k 1701 1777 + 76 Late- .1. Not Late to d 1305 1405 +100 Not On Time- On Time to+a+d 1566 1508 - 58 N“ 0“ Time" 1: +k+n 1754 1795 + 41 Late 0 Not On Time- Not On Time to+a 1265 1179 - 86 Not On Time- + + - Not Late to d k+n 2166 2125 41 N“ Lam" t +a+k 1690 1777 + 87 On Time 0 N“ Late‘ 1: +d 1379 1405 + 26 Late 0 Not Late- Not On Time to+a+d+k 2131 2107 - 24 N°t Late" 6 1146 1076 - 70 Not Late 0 87 each to be significant effects (2: _t_(11) I 4.205, p < .01; 1:: _t_(11) I 7.953, p < .001). A problem arose in the test of parameter p, .E was not directly reflected in the significant interactions of the analysis of variance, and while four test conditions were predicted to contain parameter 3, its significance could only be tested using a related measures pftest contrasting pofkfp (Not On Timel-Latez) and Eofk (On Time -Not Latez). 1 Parameter p was not significant, 5(11) I .58, p_> .05. Examination of Table 4.6 reveals that of all of the conditions, the On Time INot Late 1 2 condition was the one most poorly predicted by the model (being under- estimated by 140 msec.). It would appear that some other parameter is l-Late2:0n Timel-Not Late2 contrast inappropriate for the testing of 2, operating in this condition, thus making the Not On Time However, elimination of paramter p_is not an acceptable solution. This procedure resulted in parameters 2_and d becoming unreliable predictors of behavior; furthermore, estimations of conditions become in error by as much as 290 msec. in some cases. The existence of parameter .2 is indeed questionable; however, since its elimination resulted in a poorer prediction of performance, the decision was made to allow parameter p_to remain as an encoding parameter of the model. Summagy Important aspects of visual sentence processing have been demon- strated in Experiment 3: encoding a negative does consume a significant portion of a subjects' processing time and mismatches of either term or polarity are more difficult than identities. This last point was es- pecially interesting in this experiment for all identities (regardless of whether they contained a negative) were processed more rapidly than 88 any mismatch. This was not always the case in the two previous ex- periments. Thus, the subjects may be able to make better use of pattern matching strategies in the visual than the auditory modality. Failure to find strong support for the true or conversion models of negation in the results of Experiment 3 suggests that sentence- sentence verification regardless of modality is a more complex task than the sentence-picture verification task. As an alternative, 8 four-stage, five-parameter model was proposed to account for the data of Experiment 3. The model employed conversion operations, and was able to give a reasonably good representation of the results of the present study. CHAPTER 5 General Discussion An attempt will now be made to integrate the findings of Experiments 1, 2, and 3, so as to better understand sentence processing in general, and to appreciate the issues which underlie its study. The Psycholggical Reality p£_Processing Variables It is clear that the results of Experiments 1, 2, and 3 lend credence to the psychological reality of certain psychological pheno- mena, given the variety of experimental manipulations in which they are found. In all three experiments, the paramters of Sentence 2 had drama- tic effects on processing times. On Time2 comparisons were more time consuming than Late2 comparisons, and Negative2 comparisons required more processing time than Affirmative2 judgments. While it cannot be directly shown in the present findings, the former result is probably largely the result of the presentation time parameters of the Term 2 stimuli (On Time2 being approximately 300 msec. longer than Latez). That an effect of negation was seen in Sentence 2 was a most en- couraging result. Clark and Chase (1972) had posited separate en— coding and comparison stage effects of negation, but the paradigm emr ployed in that study did not permit their isolation. In the present series of experiments, this problem was avoided; yet, the effects of negation in encoding and/or comparison stages remained. Thus, negation regardless of which stage of the verification process it affects, is psychologically more complex to comprehend that its affirmative counterpart. 89 90 In addition, Experiment 1 of the series provided evidence for the psychological reality of Inferences. It took time to verify information which was not directly presented in a story, even when that information could be readily and easily deduced from the facts presented. The results of the pilot study indicated that the majority of subjects generated Inferences appropriate to a particular context, and then executed True-False judgments on the basis of that deduction. Experiment 1's results, on the other hand, argued for a decision rule in which generation of the actual Inference was not necessary. Rather, decision making depended upon the total number of mismatches encoun- tered in a comparison and upon whether an Inference or an Assertion judgment was being called for. While there were methodological differences between the two studies, both strategies are viable options for successful completion of the task. Thus, it would be interesting to determine under what con- ditions a particular strategy would be chosen, and what advantages (in terms of time gained, memorability of items, etc.) exist in opting for one strategy as opposed to the other. At the very least, the existence of two very different strategies as a means of solving the task at hand emphasizes the complex nature of the language comprehension task. Another phenomenon reported by Clark and Chase (1972) and Young and Chase (1971) consistently found in the present experiments was the Term 1 x Term 2 interaction, or the inner string matchImismatch effect. Simply stated, recognizing an inner string mismatch (that On Time1 f Latez) is a more time consuming process than recognizing an inner string match (that On Time1 I On Timez). The results of the present series of experiments were interesting in that this effect was found to be 91 significant in both the auditory and visual modalities. One can con- clude, therefore, that in general, difference relationships are cog- nitively more complex to the subject (or the information processor) than sameness relationships. The results of the experiments yielded other interesting findings. Despite the similarities in responding across the two modalities tested (e.g., Term 2 and Negation 2 main effects, the Term 1 x Term 2 interaction), there were some strong differences. In the visual mo- dality, the effects of pattern recognition were clearly seen: all true comparisons in which Sentences 1 and 2 were identical were res- ponded to more rapidly than any other comparisons. This was not true of the auditory modality, in which some FA comparisons (On Timel- Latez) were responded to more rapidly than True identical comparisons (On Time -On Timez). 1 It is possible that this difference was not the result of a change in modality, but rather was due to a change in stimulus presen- tation co-occurring with the modality switch. Visual presentation lends itself to either a simultaneous or successive framework, and in Experiment 3, simultaneous presentation of the stimuli was used. Auditory presentation in sentence—sentence verification tasks, on the other hand, demands that the stimuli occur successively, thus, the first stimulus could never co-occur with the second. Research is currently underway to determine whether the pattern recognition effect observed in Experiment 3 was the result of the manner of stimulus pre- sentation or the result of the modality employed. If we examine the task demands of Experiments 1, 2, and 3, it ap- pears that with each successive experiment, the processing demands made 92 upon the subject decreased. Experiment 1 required that the subjects make verification judgments of both explicit and implicit information, while Experiment 2 eliminated the verification of all implicit infor- mation. Experiment 3, while in spirit a replication of Experiment 2, was constructed so that subjects were not required to store 222. information in their heads, that is, the stimuli were always available to them. An interesting corollary of this trend was the growing effect of Sentence 1 parameters on processing time from Experiment 1 to Experiment 3. In the first experiment, no parameters of Sentence 1 significantly affected processing on Day 3, whereas in Experiment 2, Term 1 was sig- nificant. In Experiment 3-Day 3 performance, both Term 1 and Negation 1 significantly affected processing time. Furthermore, in Experiment 1, only two higher order interactions were significant on Day 3, while in the other experiments, twice that number were significant. It is possible that as the processing demands of an experimental paradigm are lightened, freed capacity is then directed to intricate aspects of the stimuli. Experiments which by their very nature consume all of the subjects' available capacity (e.g., Experiment 1) would not be expected to be as sensitive to all stimulus parameters. Further experimentation manipulating task demands would shed light upon this hypothesis. Usefulness pf_the Additive Factor Model Approach It was the case in the experiments described herein that while the true and conversion models of negation were able to capture certain as- pects of the data in each instance, they were not as accurate in their predictions as they were in the sentence-picture verification 93 experiments. In each experiment, new models of information processing were generated which yielded better representations of the condition orderings. There was a problem with the violation of the independent stage principle in the model of Experiment 2. Recall that it was proposed that before one could choose between an affirmative and negative repre- sentation in Stage 1, Sentence 2 encoding had to be completed. Thus, Representation 1 was found to be dependent upon Representation 2. This could be a serious problem depending upon how intricate an additive factor one wishes to construct. If one wishes to account for every aspect of a subject's processing, then this finding would raise some doubt concerning the applicability of additive factor models to sentence-sentence verification tasks. However, if one is interested in capturing basic features of processing, then this finding does not affect additive factor model applicability. The four models proposed herein have one thing in common: they all can be represented in terms of an encoding phase, a comparison phase, and a response execution phase. The intricate differences be- tween the models arose from an attempt to capture the details of pro- cessing corresponding to different task demands. If one examines the nature of the independent stage principle violation, one sees that it occurs within a stage, thus preserving the overall serial structure of processing. It would appear, therefore, that between stage pro- cessing occurs in a serial stage fashion, whereas the possibility exists for either serial or parallel processing within a stage. In conclusion, additive factor models are useful tools by which to interpret the results of sentence-sentence verification tasks in so far 94 as they are able to capture the most fundamental features of processing: encoding and comparing. $1151.: Issues The issue of the nature of the internal representation of stimu- li is always a concern in verification studies, but the problems asso- ciated with this issue were particularly relevant in the consideration of the results of Experiment 2. It is clear that considerably more research must be devoted to the problem of internal representation. At present, we have no rule system available which defines a priori how a subject will encode a sentence. Indeed, we do not even know the vari— ety of ways a sentence may be encoded, and whether experimental par- amters, if any, would favor one format over another. How one encodes information, however, is a critical factor in the subsequent processing of that information. Thus, in order to gain insight to the operations involved in sentence comprehension, knowledge of encoding formats must be obtained. Another issue which the series of experiments raises is what is meant by language comprehension. Clark and Chase (1972) assumed that the operations reported in the sentence-picture studies were basic to any language comprehension situation. The results of the present study support this assumption to a certain degree: negatives and inner string mismatches were more difficult than affirmatives and inner string identities, respectively; and processing could be divided into encoding and comparison stages. If the modeling interpretations of the pilot study and Experiments 1, 2, and 3 are correct, however, it must be the case that sentence-sentence processing is a more complex task than sentence-picture verification, as evidenced by the variety of operations 95 and combinations of operations employed. At the very least, in terms of the alternative operations available to the processor, sentence- sentence processing appears to be quite sensitive to particular task demands. What is needed is a delineation of those processes which are necessary for comprehension to occur. The issue now becomes one of whether or not the complexities of the individual sentence-sentence task schemes (i.e., models) were essential for comprehension. In other words, one must determine whether comprehension in the sentence- picture task can be defined in the same manner (or equated with) comprehension in the sentence-sentence task. APPENDICES APPENDIX A Nwoz HHoN HoHN onH Has” sme qsmH onH ammH Hausa HmoH wmmH soaH HHmH «HoH HmmH ommH amnH N««< HsHH HHHH Nst Hana mmaH mamH HowH smmH Hwoz NoaHH Homm meow «mam omou HwHH moHH HooH smoH N««< no Hmoz H««< Hmoz H««< Hmoz H««¢ Hmoz H««< mood HmEHH :0 wood HmEHH so oocoHomcH :oHuHommd H ucoaHHoawm H hon so coHunaoo Hum .oomz oH Hm coo: < NHszmm¢ 96 APPENDIX B Nwoz «mum mHHN HHHH eHHH NHsH oaoH smHH msHH N NouoH ammH mmHH wHou ommH sHsH aNmH HHeH sHsH ««¢ Hana osmu Homm coca oaaH HHmH SNHH mmHH mez HoaHH mama soHN osHH omom mamH HmmH samH quH ~««< no Hmoz N««< Hmoz H««< Hwoz H««< Hwoz H««< Hooaa HoaHH so moaH Hosea no OUCOHQHHHH HHOHOHUDQ¢ H udoaauomxm N mom co coHuHmcoo pom .oomz sH am new: m NHszmm< 97 APPENDIX C APPENDIX C Significant (p.< .05) Results of the 2x2x2x2x2 Within-Subjects ANOVA on Days 1 and 2 in Experiment 1 Condition x Term 1 x Negation 1 x Term 2 Negation 2 3(1,11)=15.409 Condition M M Negation 1 2K1,11)I10.981 E(l,ll)I6.824 Term 2 _F_(1,11)I29.624 F(1,11)I67.488 Negation 2 E(I,11)I54.893 FK1,11)I104.366 Condition Efil,1l)=363.120 IKI,11)I147.442 Condition x Term 2 E(1,11)I14.526 Condition x Negation 1 ‘§(1,11)I10.532 {(1,11)I6.698 Negation 1 x Negation 2 _F(1,11)5.410 £11,11)I4.768* Term 1 x Term 2 .E(1,11)I10.19l E(1,11)I14.041 Term 1 x Term 2 x Negation 2 Ff1,11)=5.178 Condition x Term 1 x Negation 1 x Negation 2 ‘§(1,11)I6.169 Condition x Term 1 x Term 2 x Negation 2 §(1,11)I5.136 g<1,11)=4.181* *denotes a marginally significant result 98 APPENDIX D 8.: 3: .38 82 930862 ouoH 82 $2 32 82 N45345:: V 38 $2 $3 82 .3382 63H 2: 23 .33 8.: «6506332 8 Ho>Huowoz Ho>Humauwmw< Ho>Hummoz Ho>HuoaHHmw< H33 H82. so H Hon so :oHquooo Hem .ommz :H Hm cmox N uaoaHHomxm Q xHszmm< 99 APPENDIX E mHmH mHsH meH asHH No>Hoamoz NoumH mHmH meH mmmH NamH NopHoaaoH««< mawH HNmH HHoH wmoH ~6>Hoamoz NoaHH HomH oHHH momH HHmH Ho>HomaoH««< so Ho>Humwoz Hoeaumauamm< H0>Hummoz Ho>HumEHHmm< HouoH HoeHH so N uamBHHomxm N Ion so soHuHmooo Hum .oomz ca am now: m NHDzmmm< 100 APPENDIX F APPENDIX F Significant (p_< .05) Results of the 2x2x2x2 Within-Subjects ANOVA on Days 1 and 2 in Experiment 2 Condition Day 1 Day 2 Term 2 Efil,11)I30.686 Ffil,11)I23.932 Negation 1 .E(1,11)=4.145* Negation 2 §(1,11)=312.941 F(1,11)I108.022 Term 1 x Term 2 E(l,ll)I23.679 E(1,11)I9.924 Term 2 x Negation 2 FK1,11)I7.612 Term 1 x Term 2 x Negation 2 E(1,11)=9.343 2(1,11)I8.059 Term 1 x Negation 1 x _ Term 2 x Negation 2 .F(1,11)=30.649 Ffll,11)=7.308 *denotes a marginally significant result 101 APPENDIX C N HHeH HoNH amoN MONN o>Huowoz N 33 cowH HmeH oowH HomH o>HumsHHmm< N 33 2mm 8: $2 653862 as: 82 NHHN 32 22 «65345:: no Ho>Hummoz Ho>HumaHHmw¢ Ho>Humwoz Ho>HumaHme< H33 H25 8 H mom so coHqucou Hum .oomz cw Hm coo: m uaoafiummxm U NHszmm< 102 APPENDIX H No>Humwwz HmmH HHnH mama oamH oumH unsH NoHH msoH «HHH ~6>HoaaoH««< mesa samH mmsH HHsH H6>Hoamoz maHH HHwH moHH Hoes HHNH ~6>H04a4H««< so Ho>Huowoz Ho>HusaHHmm< Ho>Huowoz Hm>HuoaHwa< Hanan HoaHH no N man so dowufivcoo Mom .ommz cw Hm coo: m uaoEHHomxm m Nanmmm< 103 APPENDIX I APPENDIX I Significant (p < .05) Results of the 2x2x2x2 Within-Subjects ANOVA on Days 1 and 2 in Experiment 3 Condition Day 1 Day 2 Negation 1 _F_(1,11)I35.293 _F_(1,11)I8.067 Term 2 _F(1,11)I33.613 _F_(1,11)I22.182 Negation 2 _F_(1,11)I45.041 F(1,11)=20.152 Term 1 x Term 2 F_(1,11)I50.493 F(1,11)I59.654 Term 1 x Term 2 x Negation 2 _F(1,11)I31.706 _F_(1,11)I7.487 Term 1 x Negation 1 x Term 2 x Negation 2 £(1,11)=25.739 _1:(1,11)I15.177 104 BIBLIOGRAPHY BIBLIOGRAPHY Carpenter, P. A., 8 Just, M. A. Sentence comprehension: A psycholin— guistic processing model of verification. Psychological Review, 1975, 82, 45-73. Chase, W. G., & Clark, H. H. Mental operations in the comparison of sentences and pictures. In L. Gregg (Ed.), Cogpition‘ip_learning and memogy. New York: Wiley, 1972. Clark, H. H. Linguistic processes in deductive reasoning. Psychologi- cal Review, 1969, 12, 387-404. Clark, H. H. How we understand negation. Paper presented at Workshop on Cognitive Organization and Psychological Processes, University of California, Irvine, 1970. Clark, H. H. Semantics and comprehension. In T. A. Sebeok (Ed.), Current trends ip_linguistics, V01. 13} Linguistics and —- _adjacent arts and sciences. The Hague: Mouton, 1974. Clark, H. H., & Chase, W. G. On the process of comparing sentences against pictures. Cognitive Psychology, 1972,.3, 472-517. Cornish, E. R. Pragmatic aspects of negation in sentence evaluation and completion tasks. British Journal p£;Psychology, 1971, 62, 505-511. Gough, P. B. Grammatical transformations and speed of understanding. Journal pf Verbal Learning and Verbal Behavior, 1965, A; 107-111. Gough, P. B. The verification of sentences: The effects of delay of evidence and sentence length. Journal pf Verbal Learnipg_and Verbal Behavior, 1966, 2a 492-496. Greene, J. M. Syntactic form and semantic function. Qparterly Journal p£_Experimental Psychology, 1970, 22, 14-27. (a) Greene, J. M. The semantic function of negatives and passives. British Journal pf Psycholggy, 1970, 61) 17-22. (h) Just, M. A., 8 Clark, H. H. Drawing inferences from presuppositions and implications of affirmative and negative sentences. 'Journal pf Verbal Learning and Verbal Behavior, 1973, 12, 21-31. 105 106 Klima, E. S. Negation in English. In J. A. Fodor 8 J. J. Katz (Eds.), The structure pf languagg: Readingpuip_thephilosophy'2£_langpage. Englewood Cliffs, N. J.: Prentice Hall, 1964. Slobin, D. I. Grammatical transformations in childhood and adulthood. Journal 2£_Verbal Learnigg and Verbal Behavior, 1966, 2, 219-227. Sternberg, S. Memory scanning: Mental processes revealed by reaction time experiments. American Scientist, 1969, 21, 421-457. (a) Sternberg, S. The discovery of processing stages: Extensions of Donders' method. In W. G. Koster (Ed.), Attention and_performance II, Amsterdam: North Holland, 1969. (b) Trabasso, T., Rollins, H., 8 Shaughnessy, E. Storage and verification stages in processing concepts. Cognitive Psychology, 1971, 2, 239-289. Wason, P. C. The processing of positive and negative information. Quarterly Journal 2£_Expgrimental Psychology, 1959, 11, 92-107. Wason, P. C. Response to affirmative and negative binary statements. British Journal 2£_Psychology, 1961, 22’ 133-142. Wason, P. C. The contexts of plausible denial. Journal pf_Verbal Learning and Verbal Behavior, 1965, 4) 7-11. Wason, P. C., 8 Johnson-Laird, P. N. Psychology pf reasonipg: Structure and content. Cambridge, Massachusetts: Harvard University Press, 1972. Wason, P. C., 8 Jones, S. Negatives: Denotation and connotation. British Journal 2£_Psychology, 1963, 53, 299-307. Young, R., 8 Chase, W. G. Additive stages in the comparison of sentences and pictures. Paper presented at Midwestern Psychological Association meetings, Chicago, 1971. MIC 1113171 11! 1111111111711, 1111111111.“ I