[Deconstructing Self Explanation to Improve Comprehensionof Text]

AIMEE CALLENDER: And so there are two basic study goals.First, we were trying to investigatehow to implement self-explanation for studentsto use during independent study.There's been a lot of research looking at self-explanationusing, for example, automated tutorial systems,but students can't use that during independent studies.We're trying to determine the best way to do that.The second goal was to understand

AIMEE CALLENDER [continued]: what kinds of processing is occurringduring self-explanation, and whatsets apart good readers and good performersfrom poor comprehenders.Just a brief background, self-explanationis basically having a learner explainwhat they're doing as they're doing the task.A lot of times, this has been done out loud,

AIMEE CALLENDER [continued]: and it's always completed concurrentlywith the other ongoing task.Initially, it was conducted with tasks such as problem solving,physics problem solving, algebra problemsolving, diagrammatic learning, and has alsobeen applied with text.Now the challenge with using self explanation with textis that you have two concurrent verbal tasks at the same time.

AIMEE CALLENDER [continued]: So researchers have tackled this problem differentlydepending on what strategies they're trying to do.Some researchers have applied self-explanationin a manner in which participants will basicallyself-explain out loud continuously while they'rereading.That's a pretty demanding sort of task to do,and requires a lot of prompting by the researcher.

AIMEE CALLENDER [continued]: So this isn't the best option to use during independent study.Other researchers have implementedself-explanation in a manner that students will self-explainafter each line of text.And essentially, what they find is that the more explanationsthat are produced, the better off comprehension is.So we sort of used that approach.And instead of implementing self-explanation

AIMEE CALLENDER [continued]: after every sentence, we implemented self-explanationafter every paragraph, because we're using longer texts.We also borrowed the research of McNamara's Self-ExplanationReading Training, in which she trained participantsin the best self-exploration strategiesto improve comprehension.And she identified five different componentsof self-explanation that are most likely

AIMEE CALLENDER [continued]: improve comprehension.These include bridging inferences, elaborations,predictions, paraphrases, and comprehension monitoring.So in her training, she did extensive training teachingparticipants how to do this.In our studies, we use very limited training.So in a previous study we presented herea couple of years ago, we examined

AIMEE CALLENDER [continued]: using prompted self-explanation after every paragraph,and we trained participants on these five strategies.As I said, it was very limited training.Basically what we did was we told themwhat each of these five strategieswere, we showed them a very short text,and what an example of each of those strategies looks like.And we compared two different self-explanation groups.

AIMEE CALLENDER [continued]: We had one where they were promptedto do all of these five strategiesafter every paragraph.So that's the PSE group that you see up there.The second group was prompted to do all fiveof these strategies, but they alsoreceived a second additional promptafter each paragraph prompting them justto do comprehension monitoring by itself in addition

AIMEE CALLENDER [continued]: to those five strategies.So basically, we really wanted them to engagein comprehension monitoring.The idea for this is that comprehension monitoringinitiates repair strategies like inferencing, elaboration,prediction.And so we thought that by adding this extra comprehensionmonitoring, we may actually improvethe other types of strategies that are being used.

AIMEE CALLENDER [continued]: So as you can see from the recall data,it's very clear that plus conditionwhere there's comprehension monitoring specifically beingprompted, performance is the highest.And it's the highest especially for the high ability readers.So this is actually a very interesting findingbecause a lot of different strategies usedto improve comprehension doesn't actually do anything

AIMEE CALLENDER [continued]: for high ability readers, or it actuallyimpairs their comprehension.So that's where we ended things previously.So we wanted to investigate are theredifferent patterns of statements thatare being produced that lead to the best comprehension.So we conducted a cluster analysison the types of statements that were produced.And two clusters were produced or resulted.

AIMEE CALLENDER [continued]: One cluster we called the paraphrasers, the other clusterwe called the IEP group, which was inferencing, elaboration,and prediction.And if you look at the number of statements producedby these two groups, it's very clear how they got those names.So the paraphrasing group made a lot of paraphrasesand didn't do very much else.And the IEP group made a lot of inferences, elaborations,and predictions.The comprehension monitoring actually

AIMEE CALLENDER [continued]: didn't go with either of these groups.And when you look at the resulting comprehensionperformance, the recall performance,there were clear differences between the two groups.One thing to note about these clustersthough is that that IEP cluster, itwas largely driven by the predictions thatwere made by readers in those groups, which is actually

AIMEE CALLENDER [continued]: sort of surprising to me.A lot of research shows that importance of inferencesand elaborations, and that predictionis sort of a later process that occurs,but it seemed to have a big impacton this particular cluster.So in the current study, we attempted to break downself-explanation even more.And so we had a group that was just performing inferences,elaborations, and predictions.

AIMEE CALLENDER [continued]: We had a group that just engaged in prompted comprehensionmonitoring.Because that seemed to be so important in the previousstudy, we wanted to see what would happen if we justprompted them to do that, if they could theninitiate the repair strategies on their own.And then a read only control.We had 200 participants in this study.It was all completely between subjects.We divided readers into low and high ability readers

AIMEE CALLENDER [continued]: based on Gernsbacher's multimedia comprehensionbattery.We conducted a median split at 31out of 48 possible questions.Then they took a final test.It was multiple choice and recall, either immediatelyor after a 48 hour delay.And they read one of two texts, either pumps or brakes.And it was taken from Mayer's workthat he's done extensively with these two texts.

AIMEE CALLENDER [continued]: Before I show you the results, I wantto show you the types of explanationsthat participants were producing, justto give you a better idea of what they were actually doing.So first, for inferences, we define thisas basically making connections between the paragraphs.A lot of times bridging inferencesare looking at connecting one sentence with another sentence.Because we were doing the self-explanationafter each paragraph, we were looking

AIMEE CALLENDER [continued]: at connecting one paragraph to another.So these are the types of inferences that were made.The difference between dynamic pumps and positive displacementpumps is the amount of fluid that each provideand how it's dispersed.So this person was looking at the differences between the twoparagraphs.Other people said things like drum brakesare also made of cast iron.So they're looking for the similarities

AIMEE CALLENDER [continued]: between the paragraphs.So there are a lot of statements like also,because, logical statements, similarities, differences,those types of statements that connected the paragraphs.The elaborations related what they werereading to prior knowledge.One of my favor elaborations that we saw,I can remember the dynamic part because dynamic people tend

AIMEE CALLENDER [continued]: to be loud in a steady flow.They never stop.So this is exactly like the pump.The pump is just a steady flow of fluid.Other elaborations, most of the elaborationswere of the following type.This pump reminds me of a boat propeller,linking it to something that's very concretely associatedwith the topic.The predictions, as I noted previously,

AIMEE CALLENDER [continued]: were very important.So the predictions were about the contentor the structure of the text.But when you look at the predictions thatwere actually made, they were very basic predictions.Statements like, I think the next sectionwill continue to talk about one of the main types of pumps.Or I predict that next we will discuss drum brakes.They are very basic predictions.So when you're looking through that,

AIMEE CALLENDER [continued]: you think it's not going to do very much.Those are predictions.And then finally, we had comprehension monitoringstatements, which are statements havingto do with understanding.We divided comprehension monitoring between whatwe call real comprehension monitoringand forced comprehension monitoring, because initiallya lot of students just weren't really putting anything down.So we wanted to make sure they were doing the task.So real comprehension monitoring statements

AIMEE CALLENDER [continued]: are the statements that are analyzed in the study.For example, what is a valve assembly?I understand what it does, but not what it is.Or the hydraulic system is hard to understand.Most of the statements that were real comprehension monitoringstatements had to do with trying to tease apart different thingsor not understanding what something is.The forced comprehension monitoring statements,

AIMEE CALLENDER [continued]: they just said, I understand.So that's comprehension monitoring.And so what we found, if you look at multiple choiceperformance, there's actually no effective condition.There was a difference between low and high reading abilityand effective delay.But if you look at the data up here,

AIMEE CALLENDER [continued]: you can start to see that for the high ability readers,that IEP group is starting to pull apartfrom the other groups.And when you look at the recall performance,that becomes very evident.For high ability readers, that IEP strategy reallyimproves their performance.And you do see that for the low ability readers as well.The comprehension monitoring alone

AIMEE CALLENDER [continued]: didn't seem to improve comprehension as much.And so we're thinking that with comprehension monitoring alone,readers aren't then initiating the repair strategiesthat they need to initiate.So in this case, there was an effective condition,reading ability delay and no interactions.So like the previous study, we wantedto investigate what was going on between the high performers

AIMEE CALLENDER [continued]: and the low performers.So we conducted another cluster analysis,and it all fell into one cluster.So instead we look at the differencebetween the low and high quality readers, and the typesof statements that were produced,or the number of statements that were produced.And here you can see that for low ability readers,they produced significantly fewer inferences, elaborations,

AIMEE CALLENDER [continued]: and predictions, even though they were specificallyprompted to engage in those same strategies as the high abilityreaders.And what's interesting is a lot of times,people would actually label their explanations.This is an inference.This is an elaboration.This is a prediction.And a lot of times they were wrong with how they labelledthings.So they really need a lot of additional trainingto understand what an inference looks like,and what an elaboration also looks like.

AIMEE CALLENDER [continued]: And even the predictions, we were surprised.The high ability readers produced a lot morepredictions, and most of them were those simple predictionsthat I showed you before.I think this is what they're going to discuss next.Now there's also evidence as we were coding the datathat other strategies are being used as well,primarily paraphrases.And this is what the low ability readers were doing.

AIMEE CALLENDER [continued]: They were simply paraphrasing the text,even though they were specificallyprompted to engage in these other strategies.So again, the low ability readersneed a lot more scaffolding to teach themhow to engage in these other strategies.When we look at comprehension monitoring,again the high ability readers areengaging in more comprehension monitoring than low abilityreaders, which is interesting because most of the statements

AIMEE CALLENDER [continued]: have to do with I don't understandthis particular concept.And so the high ability readers--you would think that low ability readers don't understandthe concepts as well as high ability readers,but the high ability readers can recognizewhen they don't understand it.So again, we wanted to see, well, do these statementsactually have anything to do with final performance.So we just did some correlations looking

AIMEE CALLENDER [continued]: at the number of statements producedand recall performance.And the only statement type that correlated with performancewere the predictions, these simple predictions about what'sgoing to come next in the text.And so based on these data, we came upwith three basic conclusions about using self-explanationfor students during independent study looking at readers

AIMEE CALLENDER [continued]: of differing abilities.First of all, comprehension monitoringshould be implemented alongside inference, elaboration,and prediction.I know sometimes when I teach introto psych, I have lots of students coming and asking me,how do I study for your tests?How do I understand what I'm reading?And we talk about sort of monitoring what you don't knowand moderating what you do know.

AIMEE CALLENDER [continued]: However, that doesn't seem to be enough.They need to be able to engage in other strategiesas well to improve their comprehension, especiallyfor low ability readers.Second, it seems that all five strategies togetherimproves performance for high ability readers,but you can break it down even more for high ability readersand just say do interesting, elaboration, and prediction,

AIMEE CALLENDER [continued]: and that will improve your comprehension.So that's actually a great findingfor high ability readers.Now we think about high ability readersin terms of what they're doing differently from low abilityreaders in that they're building a morecoherent mental representation of the text.So a better organized representation of the textleads to a better understanding and better memory.

AIMEE CALLENDER [continued]: And when I think back to those predictions thatwere highly correlated with performanceand seemed to set apart that IEP cluster,it seems that high ability readersare able to use those predictions, those verysimple predictions to plan out their mental representationand improve the coherence and the organizationof their mental representation.

AIMEE CALLENDER [continued]: So that's it.Just thanks to all the research assistants who coded the data.And any questions?