1

The Effect of Prediction on Depth of Encoding Student X name University of New South Wales

2

Abstract The aim of this study was to investigate the effect of prediction on depth of encoding. Participants (N=136) were given sentence stems and instructed to think of a word to finish the sentence. The full sentence was then completed with either a predictive or a nonpredictive word, and a recognition test given to see how well each word was retained in memory. Out of the three hypotheses, the results supported stronger encoding of predictive words over non-predictive words or the prediction itself. It was suggested this could be due to predictive words forming a more elaborate memory trace, which then integrated better with existing mental categories.

3

The Effect of Prediction on Depth of Encoding When reading a sentence, the mind can form a prediction about which word should come next. Yet it is not clear how this prediction affects the encoding of the actual word that completes the sentence. A word that fulfils the prediction could be better encoded than one that violates it, simply because it forms a smoother fit, or because the mind was already prepared for it. Alternatively, the violation itself might create a stronger impact, leading to better encoding. A third possibility also exists: that the prediction overwrites the encoding of the actual word, regardless of its congruency. One of the most enduring theories of encoding is the levels of processing framework, which states that retention increases in relation to the depth at which the word is processed (Craik, 2002; Craik & Tulving, 1975; Lockhart, Craik & Jacoby, 1976; Moscovitch & Craik, 1976). This was first studied in a series of experiments by Craik and Tulving (1975), where it was found that participants encoded a word more deeply when the word was congruent with its semantic question rather than incongruent. Craik (2002) suggested that congruent questions and words combined to form a more elaborate memory trace. This trace was better remembered either due to its greater distinctiveness in memory, or because it integrated better with existing mental categories. Yet Craik and Tulving‟s study did not explicitly involve prediction of words, only judgement. In contrast, Whittlesea and colleagues have performed a number of experiments involving prediction and congruency (Whittlesea, 2004; Whittlesea & Leboe, 2003; Whittlesea & Williams, 2001). One particular study found that when completing sentence stems, retention of incongruent words was equal to or better than retention of congruent words (Whittlesea, 2004). Whittlesea believed this was because incongruent

4

words created a sense of surprise, which helped bind the word to its associated sentence stem and make it feel familiar on subsequent encounters. The current experiment attempted to clarify how prediction affects encoding of the actual presented word. Participants were shown sentence stems and instructed to think of a word to complete the stem. The sentence was then completed with a predictive or nonpredictive word, and encoding gauged by a recognition test. It was hypothesised that if predictive words were encoded more deeply, then a greater proportion of predictive words would be recognised. Alternatively, if non-predictive words were encoded more deeply, then a greater proportion of non-predictive words would be recognised. However if the prediction itself was encoded more deeply, then the results would show high recognition for predictive words combined with very low recognition for non-predictive words, as the latter would have been discarded during the encoding process. Method Participants Participants were 136 undergraduate psychology students from the University of New South Wales (M = 19.8 years, 82 female). It was completed during tutorials as a group experiment and formed part of their assessment. Design The independent variable was the type of word presented to complete each sentence, with predictive and non-predictive words counterbalanced between tutorials. The dependent variable was the proportion of „Yes‟ responses on the recognition test, where „Yes‟ referred to words presented earlier („Hits‟) and words not presented („False Alarms‟).

5

Materials A PowerPoint presentation was used to display 40 sentences in random order (20 predictive, 20 non-predictive). An example of a predictive sentence would be “she put the new picture into a frame”; a non-predictive sentence would be “she went down town to see the new frame”. The recognition test that followed contained 80 items: 20 predictive, 20 non-predictive, and 40 filler items. Procedure Each sentence was initially displayed with the final word missing for four seconds. During this time participants were asked to silently think of a word to complete the sentence. The final word was then added (either predictive or non-predictive) and the total sentence shown for a further two seconds. After the presentation participants had five minutes to complete the recognition test. The proportions of „Yes‟ responses were recorded for each item type. Results Group means were calculated and are shown in Table 1. Paired samples t-tests were used to analyse the results. A significant difference between „Yes‟ responses for predictive words and non-predictive words was found, t(135) = 12.72, p < .0001, showing predictive words were better recognised than non-predictive words. However nonpredictive words were still significantly recognised in comparison to filler words, t(135) = 27.94, p < .0001.

6

Table 1 Group means and standard deviations for the proportion of recognition responses
Proportion of Yes Responses Word Type Predictive Non-predictive Filler M 0.86 0.74 0.16 SD 0.16 0.11 0.11

Discussion The results support the hypothesis that predictive words are encoded more strongly than either the prediction itself or words that violate the prediction. As seen in Table 1, recognition of predictive words was highest, suggesting that encoding was best for either predictive words or the prediction itself. Yet the results also show strong recognition for non-predictive words. This indicates that the prediction itself was NOT encoded, as this would have caused non-predictive words to be discarded during encoding, resulting in a much lower proportion of yes responses for non-predictive words. This outcome is consistent with Craik and Tulving‟s (1975) levels of processing experiments, which suggests that predictive words and sentence stems do combine to form a more elaborate memory trace than non-predictive words and stems. Yet whether this trace is better remembered due to its distinctiveness, or due to its integration with existing mental categories, is still unclear. While these two possibilities are difficult to unravel, the strong recognition result for non-predictive words could potentially provide support for mental categories over distinctiveness. If a sense of surprise, as suggested by Whittlesea (2004), does help bind non-predictive words to their sentence stems, then presumably this word/stem combination would also create a distinctive memory trace. This would imply

7

that it was the stronger integration with existing mental categories that gave predictive words the memory advantage. While this would be an interesting issue to investigate further, there could be a methodological explanation for the high responses for both predictive and non-predictive words. The recognition test was given almost immediately after the sentences were displayed, which could have muted the differences between the two outcomes. Before future research questions are investigated in this area, it would be essential to rerun the study with a longer delay between the display phase and the testing phase. Similar results with a longer delay would provide stronger support for predictive words integrating better with existing mental categories. Assuming the results are replicated with a longer delay, then the current study has shown that we are more effective at encoding and retrieving information that fits with our expectations. If we are already thinking of a word that has sprung from past experience, then it is likely that confirmation of that word would reinforce that experience and make it even more familiar on future occasions.

8

References Craik, F. I. M. (2002). Levels of processing: Past, present...and future? Memory, 10, 305318. Craik, F. I. M. & Tulving, E. (1975). Depth of processing and the retention of words in episodic memory. Journal of Experimental Psychology: General, 104, 268-294. Lockhart, R. S., Craik, F. I. M., & Jacoby, L. (1976). Depth of processing, recognition and recall. In J. Brown (Ed.), Recall and Recognition (pp. 75-102). New York: John Wiley & Sons. Moscovitch, M. & Craik, F. I. M. (1976). Depth of processing, retrieval cues, and uniqueness of encoding as factors in recall. Journal of Verbal Learning and Verbal Behavior, 15, 447-458. Whittlesea, B. W. A. (2004). The perception of integrality: Remembering through the validation of expectation. Journal of Experimental Psychology: Learning, Memory, and Cognition, 30, 891-908. Whittlesea, B. W. A., & Leboe, J. P. (2003). Two fluency heuristics (and how to tell them apart). Journal of Memory and Language, 49, 62-79. Whittlesea, B. W. A., & Williams, L. D. (2001). The discrepancy-attribution hypothesis: II. Expectation, uncertainty, surprise, and feelings of familiarity. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27, 14-33.