|Semantic satiation as a function of initial polarity and scale relevance. 1|
The University of Illinois
Semantic satiation refers to the decrease in the intensity of meaning of a sign which comes about as a result of repeated presentation (Lambert and Jakobovits, 1960). Some recent investigations have focused on two variables which may affect the amount and reliability of semantic satiation obtained when pre- and post-repetition ratings on the semantic differential are used to index the phenomenon. One study (Yelen and Schulz, 1963) attempted to manipulate the initial polarity of the ratings by selecting a set of semantic differential scales for specific words which received low polarity ratings in a previous experiment and another set of scales which received high polarity ratings for the same words. The results of that experiment led the authors to conclude that the semantic changes that occur as a result of verbal repetition are a function of initial (pre-repetition) polarities. They argued for a regression interpretation of semantic satiation such that decreases in the intensity of meaning occur when the initial ratings are high in polarity, and increases in the intensity of meaning when initial polarities are low. These conclusions were called into question (Amster, 1964; Jakobovits and Lambert, 1967) on methodological grounds related to the scoring system used by Yelen and Schulz as well as to the design used for the manipulation of initial polarities. Accordingly, one purpose of the present study is to reinvestigate the relation of initial polarity and semantic satiation by using a more adequate design.
A second purpose of this study relates to the contention by Madigan and Paivio (1967) that the definition of the mid-point of the semantic differential as "meaningless" given to subjects in the instructions facilitates the occurrence of the semantic satiation effect. This point is of some importance since its acceptance would lead to the conclusion that the semantic differential method of indexing semantic satiation is not any more objective., as claimed by Lambert and Jakobovits (1960), than previously employed subjective methods (Severance and Washburn, 1907; Smith and Raygor, 1956; Wertheimer, 1960). The design used in the present study to investigate this instructions effect is simpler than that used by Madigan and Paivio and is not complicated by an interaction factor with "task orientation" which they included in their design and which turned out to be significant in the results.
A third purpose of this study is to investigate the effect of verbal repetition on changes in an aspect of meaning not readily measured by the semantic differential scale. On logical grounds., it is possible to distinguish between the intensity and direction of a rating on the semantic differential on the one hand, and on the other, the degree of relevance of the scale when used to rate a particular concept. Thus, the words FATHER and STONE may both receive a rating of +2 on the scale "heavy-light" (viz.: "very heavy") and yet the scale "heavy-light" may be judged more relevant to STONE than to FATHER. Or., the scale "hard-soft" may be judged equally relevant to the words PEACH and BASEBALL and yet PEACH may receive a rating of -1 (i.e. "slightly-soft") and BASEBALL may be rated as +2 (i.e. "very hard"). In general, polarity and relevance are logically separable but it should be pointed out that their relatedness in fact for any sample of words and scales is an empirical question. The interest in the present study lies in an assessment of the changes in the relevancy of scales when the words to be rated are verbally repeated.
The fourth and final purpose of the study is to determine the persistence over time of the semantic satiation effect. Lambert and Jakobovits (1960) have reported that no significant recovery of semantic satiation occurs when subjects are retested one half-hour after repetition. Here., the recovery test period was extended to one full week.
In summary, the present study has the following purposes: (a) to determine the effect of initial polarity on the amount of semantic satiation obtained; (b) to examine whether two different definitions of the mid-point of the semantic differential has differential effects on the reliability of the semantic satiation phenomenon; (c) to investigate the changes in relevancy of the scales when the rated words are verbally repeated; and (d) to assess the persistence of the semantic satiation effect over a fairly long time period (one week).
The subjects were 40 college undergraduates enrolled in the psychology department's main introductory course. Half were males and half were females.
The subjects were assigned randomly to each experimental condition and were tested individually by the same male experimenter. The design was a 2 x 2 x 3 factorial with repeated measures. "Instructions" varied two ways; "Word Sample" varied two ways; and "Polarity and Relevance" in combination varied three ways. The variation in Instructions consisted in the definition given to the middle position on the semantic differential: in one condition it was defined as "equally both" (e.g. "equally hard and soft," "equally good and bad,," etc.)., while in the other condition it was defined as "meaningless." No other variations in the instructions were introduced (see below). Two different samples of six words were used. These are shown in Table 1.
Polarity was determined by initial semantic differential ratings. Relevance was determined by ratings on a relevance scale that took the following form: FATHER: good-bad __0__:__1__:__2__:__3__with the subject (s) checking off one of the four scale positions accordIng to the following instructions:
When S first reported for the experiment, he (or she) was asked to rate the six concepts (either sample A or sample B) on 20 semantic differential scales, and then to rate the relevance of each scale for each concept. S was then given a rest period' while the experimenter (E) selected five scales for each concept according to the following requirements: (a) a combination of High Polarity (HP) and High Relevance (HR), that is five scales which S had rated "2" or "3" in relevance for that concept and at the same time rated that concept in a highly polarized fashion on the semantic differential task (i.e. 2 or 3, "very" or "extremely"); (b) a combination of Low Polarity (LP) and Low Relevance (LR), that is "0" or "1" in relevance and in polarity ("equally both" [or "meaningless"), or "slightly"); (c) E attempted to choose a combination of middle Polarity (MP) (i.e. 1 and 2) and Middle Relevance, but this turned out to be impossible with the present sample of words and scales. Since the primary emphasis of this study was an investigation of the effect of polarity,, the decision was made to opt for a combination of Middle Polarity (MP) (I and 2) and relatively High Relevance (HR) ( $list , "2". and "3"). As a result of this limitation, this study permits the assessment of only two levels of relevance (High and Low) on semantic changes with repetition.
The five scales chosen for each concept for an individual S were then used for the post-repetition ratings (P). Thus scale selection was completely variable across subjects, words,, and conditions.
Upon completion of scale selection, the repetition procedure was begun. S was shown a word printed on an index card and asked to repeat it aloud for four fifteen-second periods (two or three repetitions per second) with three five-second rest intervals in between. Following the fourth repetition, S rated the word on only the five semantic differential scales preselected for that word for that subject. S was given the following repetition and re-rating instructions.
Upon completion of repetition and re-rating of all words, Ss were asked to rerate the relevancy of the same five scales for the concept along the same relevancy dimension used initially.
At the conclusion of the experiment the Ss were given a take-home test booklet containing all six concepts with instructions to rate them on-all 20 semantic differential scales and twenty relevancy scales used in the initial pre-repetition task. They were instructed to fill out the booklet one week later at approximately the same time of day.
Polarity-difference scores. The mean polarity-difference score per scale (see Lambert and Jakobovits, 1960) 2 between initial (1) and post semantic differential ratings (P) was calculated for each S for each of the three Polarity conditions. The group means are presented in Table 2. A minus sign indicates a decrease in polarity following repetition; a plus sign indicates an increase (semantic generation). Ten of the 12 means show a decrease. The results of the analysis of variance shown in Table 3 indicate, however, that there is no significant difference between Word Samples, Instructions, or Word Conditions, although the third variable and its interaction with Word Sample both approach significance (p <.10). When the two instruction groups are combined under each Word Sample, the means for HP, MP, and LP for Word Sample A are -.22, +.05, and -.22 and for Word Sample B they are -.34, -.31, and -.27. When each of these mean differences is tested for significance for zero, all are highly significant (p <.01) except for the MP condition of Word Sample A. With the reservation that the absence of a reliable change observed in that condition might be due to a specific word effect, one can conclude that satiation effects are generally independent of initial polarity. Furthermore, the difference in definition of the mid-point as either "meaningless" or "equally both" resulted in no difference in the semantic changes that occurred.
Relevancy-difference scores. The mean relevancy-difference score per scale between initial and post relevancy ratings was calculated for each S for all three word conditions. The means are presented in Table 4 and the results of the analysis of variance in Table 5. Since the scale selection was made with both the polarity and relevancy criteria in mind, the present analysis maintains the three-condition design even though, in fact, only two levels of relevance were achieved. Similarly, although the Instructions variable is mostly relevant to the polarity measure on the semantic differential scale, it is retained for the relevancy analysis since it is conceivable that relevancy ratings of semantic differential scales may be affected by the definition given for the mid-point of the scale. As can be seen in Table 5, however, the Instructions variable is not significant. Word Condition and its interaction with Word Sample are both highly significant (p <.01). When the two instruction groups are combined, the mean differences for HR, HR, and LR are .43, .39, and +.17 for Sample A and .36, .75, and +.15 for Sample B. The significant interaction effect is accounted for by the very large decrease in the HR condition in Sample B. All six means are significantly different from zero. Therefore, the conclusion is that scales initially highly relevant become significantly less so following repetition and that scales initially low on relevance become more so after repetition. This increase in scale relevancy occurs despite the fact that these scales also show a significant decrease in polarity following repetition.
Recovery. Recovery scores were obtained by selecting out only those scales used during the post-test for each concept-subject combination. Both polarity and relevancy recovery scores were calculated and were compared in separate analyses of variance with the initial and post-polarities and relevancy ratings. The means for the three sets of polarity scores appear in Table 6 and the results of the analysis of variance in Table 7. Time (Initial, Post, Recovery) was a significant effect as was also the interaction of Time , Word Condition, and Word Sample. An examination of this triple interaction by a comparison of individual means yielded the following pattern:
where RC = mean Recovery polarity, i = mean Initial polarity, P = mean Post-repetition polarity,, and the comparison sign indicates either significantly higher or lower polarities. As can be seen., the recovery data are equivocal when assessed in this fashion. There is a curious effect to be noted with the MP data for word sample A. It will be recalled that this condition did not yield significant polarity changes with repetition (i.e. I = P). The mean polarity change showed a slight and insignificant generation effect of +.05. During the recovery test, this slight generation effect became magnified over time to a significant generation effect of +.16.
Another manner of estimating the persistence of semantic satiation is to calculate the probabilities involved in either a satiation type or generation type change during the Recovery test relative to the initial and post-repetition polarities. This comparison is given in Table 8 which shows the distribution of probabilities associated with three types of changes for P minus I and RC minus I differences. Overall X2 and individual column X2 values are all highly significant except for the No change column. As can be seen, Satiation is the most likely type of change to occur immediately after repetition (114/221) and this type of change is still the most probable one week later (96/221). Nevertheless, in the few cases where generation occurs after repetition (48/221), the effect is likely to persist one week later (with 32 out of the 48 still showing generation).
Similar analyses were parried out for the relevancy data. Table 9 presents the relevant means and the analysis of variance is given in Table 10. Comparisons among means reveal the nature of the significant triple interaction of Time, Word Condition and Word Sample:
where RC = mean Recovery relevancy, P = mean Post-repetition relevancy and I - Initial relevancy. It will be recalled that initially HR ratings decreased in relevancy after repetition while initially LR ratings increased in relevancy. It is clear from the pattern of results that these effects are still present during the recovery test and that only partial or no recovery took place. Table 11 which presents the distribution of changes in ratings confirms this conclusion.
In summary, verbal repetition causes a significant semantic satiation effect which is not related to initial polarity and which recovers only partially over a one-week interval period. Changes in relevancy ratings, on the other hand, do depend on initial scores in such a way that initially highly relevant scales decrease in relevance whereas scales initially low in relevance increase in relevance as a result of verbal repetition. These relevancy changes show no recovery after one week.
The results of this 'study are completely incompatible with those of Yelen and Schulz (1963) and support the original position of Lambert and Jakobovits (1960). The overall polarity-difference score for the 12 words used in this study shows a highly significant semantic satiation effect (X = -.21, p <.001) despite the fact that the words and scales were different from those used in the Lambert and Jakobovits study and despite the fact that substantial variations were introduced in the repetition and rating procedures. Also, no evidence was found for the contention (Madigan and Paivio, 1967) that the definition of the mid-point oil the semantic differential as "meaningless 11 facilitates the satiation effect. Furthermore, the semantic satiation effects obtained cannot be explained by a regression hypothesis (Yelen and Schulz, 1963) in view of the present demonstration that the effect is independent of initial polarity and the specific scales employed. This conclusion is further supported by recent evidence from another source (League, 1966).
On the other hand,, the present study does point to the possibility of a word specificity effect in the sense that some words may show a distinct resistance to semantic satiation. It will be recalled that two of the 12 words used in this study (CITY and CONFLICT) did not show significant polarity changes after verbal repetition. Wertheimer (1960) and Wertheimer and Gillis (1958) using a subjective method of self-report of meaning lapse during verbal repetition did find some evidence of differential susceptibility to satiation of words differing in certain pre-determined characteristics (e.g. length, abstractness, sound quality). Further work along these lines is needed to elucidate this problem.
The results on changes in the relevance of scales to repeated words are clear but no unequivocal interpretation can be offered at the present time. On the one hand, semantic satiation of a word shows decreases in the relevancy of scales when these scales are initially high in relevance to the word and are accompanied by a decrease in polarity. On the other hand, when the scales are initially low both in relevance and in polarity, they show an increase in relevance despite their decreased polarity. Since only two levels of relevance were employed - high vs. low - and since the changes obtained Were opposite with these two levels, a regression hypothesis cannot be rejected as the simplest explanation.
The results of the recovery test also point to a word specificity effect. In some cases, complete recovery of the semantic satiation effect occurs after the one week delay period, but in other cases the effect persists. Overall, the evidence clearly indicates that one minute of verbal repetition of a word results in satiation effects that are discernible one week later. The evidence suggests that the persistence effect is specific (Table 8) but since no unsatiated control words were included this point remains in doubt. The persistence effects of relevancy changes are more conclusive than the data on the polarity changes, corroborating the general conclusion that a relatively short period of verbal repetition has long lasting effects.
Amster., Harriett. Semantic satiation and
generation: Learning? Adaptation? Psychological Bulletin 1964, 62, 273-286.
1 An earlier version of this paper was presented at the Eastern Psychological Association annual convention in Boston, April 1967. This research was supported by NSF grant GS 953 awarded to the senior author.
2 In this study, the average change in polarity is given per rating (i.e. per word per scale).
Applied Psycholinguistics in Social Psychology
Effects of Mere Exposure: A Comment
Rhetoric and Stylistics: Some Basic Issues in the Analysis of Discourse
Semantic Satiation and Cognitive Dynamics
Semantic Satiation as a Function of Initial Polarity and Scale Relevance
Some Potential Uses of the Cross-cultural Atlas of Affective Meanings
An Evaluation of People's Attitudes Toward Technostress and Techniques on How to Overcome it.
To Fear or not to Fear the Computer
More Articles by Leon James