Mechanisms Of Contextual Predictive Effects In Sentence Reading

Contextual predictability refers to the probability that the upcoming words can be predicted by the reader（Rayner ＆ Well,1996）.Numerous studies have shown that there were contextual predictability effects in reading,and readers processed predictable words faster than unpredictable words.For the performance of the contextual predictability effects,a body of previous studies have obtained stable and consistent results.However,the contextual predictability effects measured in reading might originated from the effects of bottom-up integrative processing,or from the combined effects of bottom-up integrative processing and top-down predictive processing.There is no clear consensus on exactly how the contextual predictability effect arises.Therefore,on the basis of previous research,the current research mainly focused on the predictive processing in the predictability effects of context,and discussed the mechanism of the predictability effect of context in reading through 3 studies with 7experiments.Firstly,Study 1 explored whether the effect of top-down predictive processing was really included in the predictability effects of context.The key to explore this question was to examine the dependence of contextual predictability effects on bottom-up information.Experiment 1a and Experiment 1b explored the influence of bottom-up preview information on early contextual predictability effects by manipulating preview type and contextual predictability of target words.Experiment 1a was 2（preview type:valid preview,no preview）× 2（contextual predictability: predictable words,unpredictable words）within-subject design.Experiment 1a set no preview condition by the novel revealing text paradigm.The results of Experiment 1a showed that the main effects of preview type and contextual predictability were both significant on the early processing,but there was no any interaction between them.The contextual predictability effects on early target word processing were not affected by bottom-up early bottom-up preview information.In order to set a more natural reading environment,Experiment 1b replaced no preview（blank preview）with meanings ※ preview and also was a 2（preview type:valid preview,※ preview）× 2（contextual predictability: predictable words,unpredictable words）within-subject design.The results again found that the main effects of preview type and contextual predictability were significant on the early eye movement measures,but there was not any significant interaction.The results of Experiment 1a and Experiment 1b both showed that the early contextual predictability effects did not depend on bottom-up preview information.Contextual predictability effects can occur without the bottom-up preview information,which suggested the contextual predictability effects involved in the influence of predictive processing.Experiment 1a and Experiment 1b examined the influence of early bottom-up preview information on the early contextual predictability effects of context.However,reading not only includes early bottom-up parafoveal preview information,but also the foveal information.Therefore,to further investigate the influence of bottom-up visual information integration processing on contextual predictability effects.Experiment 1c manipulated the contextual predictability types of target nouns（predictable nouns,unpredictable nouns）and conducted a single factor within-subject experiment.Importantly,experiment 1c measured the difference of quantifiers located in front of the nouns between different experimental conditions.The processing difference of quantifiers directly tested whether the contextual predictability effects involved up-bottom predictive processing.The results showed that compared with the unpredictable noun condition,quantifiers had more skipping,shorter fixation times,and more regressive processing on predictable noun condition.These results suggested before the noun appears,readers have already performed up-bottom predictive processing based on the previous contextual information.It exactly occurs predictive processing in sentence reading.To sum up,the results of Study 1 showed that readers do predict the upcoming content in advance according to the contextual information in natural sentence reading.The up-bottom predictive processing had a significant impact on the word processing.Based on the fundings of study 1,study 2 examined the prediction error cost effects which was the key argument about two models of predictive processing.Prediction error cost was the interference on unpredictable word in high contextual constraint,compared to low contextual constraint.One model assumed that readers first made a prediction for the most likely specific word in series processing model,which suggested there was prediction error cost in sentence reading.The other model assumed that readers parallelly grade pre-active a series of upcoming context that was given different weight,which indicated there was no prediction error cost.Study 2controlled the context constraint（high contextual constraint,low contextual constraint）and the type of contextual predictability of the target word（predictable words,unpredictable words）.Experiment 2a examined the effect of prediction error cost in natural sentence reading with a single factor and three levels（predictable words in high contextual constraint,unpredictable words in high contextual constraint,unpredictable words in low contextual constraint）within-subject design.The results found that there was a significant main effect of contextual predictability,contributing to more skipping,shorter fixation time,and less regressive processing on unpredictable words in high contextual constraint compared with predictable words in high contextual constraint.However,the results not only showed no effect of contextual prediction errors cost,but high contextual constraint promoted the processing of unpredictable words.Compared with unpredictable words in low contextual constraint,unpredictable words in high contextual constraint had more skipping,shorter fixation times,and less regression.Experiment 2b ruled out the influence of semantic relevance between predictable words and unpredictable words on the effect of prediction error cost by using semantic violation paradigm.It was a 2（contextual constraint: high contextual constraint,low contextual constraint）× 2（type of target word: semantic identity,semantic violation）within-subject experimental design.Semantic identical words were predictable words in high contextual constraint and unpredictable words in low contextual constraint.Semantic violated words were unpredictable words in both high and low contextual constraint.The results found that semantic identical words had more skipping,shorter fixation times,and less regressions in high contextual constraint condition than low contextual constraint condition.For semantic violated words,there was no significant difference between the two conditions.The results of study 2 showed that prediction mainly promoted the processing of predictable words,but did not cause any interference on unpredictable words.The results supported grade pre-activation.To examined the effect of lexical coding duration on predictive processing during sentence reading,Study 3 extended the lexical coding time by manipulating the presentation（blur）of the target word.Experiment 3a conducted 2（type of presentation:normal,blur）×2（contextual predictability: predictable words,unpredictable words）to examine the contextual predictability effects of different lexical encoding durations.Contextual predictability effects mainly represented contextual facilitation for word processing in Experiment 3a.It was found that there was a significant interaction between the type of target word presentation and contextual predictability.Under the condition of blur presentation,the difference between predictable words and unpredictable words was greater significant.The results indicated that the lexical encoding time was longer,the contextual facilitation was stronger.Experiment 3b investigated the influence of lexical encoding duration on the prediction error cost performance.It was a 2（type of presentation: normal,blur）×2（contextual constraint: high contextual constraint,low contextual constraint）.It was found that there was no interaction between type of presentation and contextual constraint.There was no any significant prediction error cost effect on both normal and blur presentation condition.The results of Study 3 showed that lexical encoding duration did not change the prediction processing during Chinese sentence reading.In conclusion,the current study explored the mechanism of the predictive effect of context in Chinese sentence reading.The results showed that,according to the previous contextual text,readers had already performed up-bottom prediction processing for the upcoming words.Importantly,prediction occurred in the form of parallel model,which resulted in graded pre-activation of potentially many words.