Share PDF

Search documents:
  Report this document  
    Download as PDF   
      Share on Facebook

Journal of Literacy Research

Where Do Their Answers Come from? A Study of the Inferences which Children Make when Answering Questions about Narrative Stories

Tom Nicholson and Robert Imlach

Journal of Literacy Research 1981 13: 111

DOI: 10.1080/10862968109547400

The online version of this article can be found at:

Published by:

On behalf of:

Literary Research Association

Additional services and information for Journal of Literacy Research can be found at:

Email Alerts:





>> Version of Record - Jun 1, 1981

What is This?

Downloaded from at Airlangga University on March 10, 2013

Journal of Reading Behavior

1981, Volume XIII,No. 2




Tom Nicholson

Universityof Waikato, Hamilton, New Zealand

Robert Imlach

FruitvaleSchool, Auckland, New Zealand

Abstract. Young readers often seem to overlook explicitly stated causal statements in narrative texts and instead give their own versions of why a text event occurred. Some researchers would agree with Smith (1979) that children do this because they read for meaning rather than word-by-word. This is an "inside-out" (or, "schema-based") view of text comprehension. Other researchers, however, agree with Thorndike (1917) that "errors" occur because "the mind is assailed by every word in the paragraph." This is an "outside-in" (or, "text-based") view of the com- prehension process. The purpose of this study was to find out the relative influence of text data and prior knowledge on the kinds of inferences which children make when answering questions about stories. In Experiment 1, text structure was altered by embedding either predictable or unpredictable reasons for events in the text, and also by varying the position and distance of these reasons from the text event being asked about. Some of these stories were familiar; others less so. Text accessibility was also varied. In all, the design was a 24 × 3 factorial, using repeated measures. In Experiment 2, a causal "preference" factor was added, to take account of the fact that children seemed predisposed toward certain kinds of inferences, whether they are predictable or not. The results provide support for the notion that text data and background knowledge compete for priority in question-answering. They suggest that children may benefit from instruction which helps them to arbitrate between plausible, yet competing explanations for important text events.

Children often seem to overlook explicitly stated causal statements in narrative stories and instead make their own inferences about why certain text events occur.

Special thanks are due to the children, principals and staff of Richmond Park, Glenview, Hillcrest, Hamilton East and Woodstock schools for helping us with this study. We are also grateful to the principal of Fruitvale School, and to the children there who allowed us to talk to them at length about our questions and their answers. The commentsDownloadedf JournalfromofjlrReading.sagepub.comBehavioat AirlanggareviewersUniversity onwereMarchalso10, 2013of considerable assistance to us in revising this article.


Journal of Reading Behavior

Very often these inferences are sensible, and even creative, yet are not what is stated in the text itself.

Some researchers, such as Smith (1979), argue that this happens because children read for meaning rather than word-by-word. According to this view, the meaning which children get from a story is determined mainly by the meaning which they bring to it. In other words, reading is seen as an "inside-out" process in which comprehension is only "incidentally textual."

Researchers who support this view argue that stories have their own "syntactic structure" (Rummelhart 1975, Mandler &Johnson 1977, Kintsch &van Dijk 1978). They argue that children use their prior knowledge of the "grammar" of stories in order to map surface text data onto more abstract, "deep structure" representations. As Stein and Glenn (1978, p. 65)putit—

"When reading or listening to a story, subjects expect certain patterns of in- formation, attend to informational sequences which match these patterns and organize incoming information into similar patterns..''

Support for this view also comes from researchers who have tried to account for both the syntax and the semantics of the "behind the eyeballs" view of text com- prehension. These researchers have used Bartlett's (1932) notion of "story scheme"as a way of conceptualizing the meaning which the reader brings to stories. Their studies have shown that our presuppositions (or "schemata") about story meaning can strongly influence our understanding (Pichert & Anderson, Note 1; Spiro 1977, Thorndyke 1977, Bower 1977). In fact, readers may not accept text in- formation which is incongruent with their own presuppositions (Meyer &Freedle, Note 2).

According to the "inside-out" view, text data may only play a secondary role in story comprehension. Instead, comprehension involves "filling the slots of the ap- propriate schemata in such a way as to jointly satisfy the constraints of the message and the schemata" (Anderson et al., Note 3). When all the slots are filled, then the schemata are said to be "instantiated"—that is, the story is understood.

If we accept the "inside-out" view of text comprehension, then it easy to see how some text details are likely to be lost or ignored by the reader, especially if those details do not fit in with an already pre-constructe"d story grammar, or if they are in- congruent with a reader's existing schema (or schemata).

Yet the notion that we have schemata in our heads, which text data must instan- tiate is quite indigestible to some researchers. Instead, they take the view that text comprehension is an "outside-in" process, where prior knowledge is important but only to answer questions which cannot be answered from the text data. This view argues for the primacy of print in determining the meaning of a story. Text com- prehension involves a careful consideration of all of the text data, using complex in- ferencing rules to fill in the gaps which are not suggested by the text itself. When this is not done, then question-answering may well be plausible but it won't be accurate. Proponents of the "outside-in" point of view would agree with Thorndike (1917, p. 326), that—

"The mind is assailed, as it were, by every word in the paragraph. It must select, repress, soften,Downloadedemphasize,from jlr.sagepub.comcorrelateAirlangga Universityand organize,on March 10, 2013all under the in- fluence of the right mental set or purpose or demand."

Where Do TheirAnswers Come From?


Thorndike argued that in this complex interplay of ideas competing for the reader's attention, "errors" occur because certain words exert too much or too little influence on the child's construction of the meaning of the text. Support for the "outside-in" hypothesis has come from researchers whose findings are at odds with those who take the "inside-out" view. Warren, Nicholas and Trabasso (1979, p. 49) have criticized the findings of researchers such as Stein and Glenn (1976), arguing that "the understander would be a more flexible processor of text if he were allowed to induce a structure rather than apply a preconceived one."

Some researchers have also criticized the notion of "story grammars" as both speculative and unworkable (Black & Wilensky 1979, Cirillo & Foss 1980). Further support for the "outside-in" view has come from research by Rickards and Denner (1979) who found young children underlined the least important details in stories, even though their task was to underline only the important details. Olson et al. (in press) have argued that this happens because some propositions in the text seem im- portant at first, yet are not central to the final representation of the text meaning.

To sum up, the basic difference between the "inside-out" and "outside-in" no- tions of text comprehension is in the importance they assign to text data. According to the "inside-out" view, what matters is what the reader brings to the story. Those who advocate the "outside-in" view would argue the opposite—that what matters is the information the story brings to the reader.

One way of testing the importance of text data in story comprehension is to create "errors" or inconsistencies and to embed these within the stories themselves. Markham (1977) did this and found that young children seemed to overlook anomalous text data. Baker (Note 4) tried a similar procedure with college students but she found that they were aware of the inconsistencies. They had either decided to ignore them or else forced them to "fit" the story.

The problem with such "simulation" research is that it may not reflect the naturalistic situation. It is one thing for a child to read "a snail is an animal" and then later read "a snail is not an animal." It is another thing altogether to read "a snail is an animal" yet later read that "a snail is a plant." In the first instance, the anomaly is a negation; in the second instance, it is a contradiction. The difference in the type of anomaly may be important. In brief, we need to be cautious about results based on one kind of text information which may or may not simulate what children think are inconsistent.

Another way of testing the importance of text data without destroying the naturalistic situation, is to manipulate systematically the basic structure of the story. Text "connectedness" (Schank 1975) is one factor which seems important. We -already know that children are less likely to make their own inferences when logical connectives, such as "because, "are missing from the text (Bormuth et al. 1970, Pear- son & Nicholson 1976, Montague, Note 5). Yet we also know that cohesive texts can be misinterpreted. Children still seem to misunderstand or overlook explicit, relevant information in texts (Irwin 1980).

So there is more to text comprehension than explicitness and cohesion—but what? Perhaps the type of explicit information is an important variable. For example, Omanson et al. (1979) found that the effect of goal information in stories was par- ticularly striking when it was negative. It may be that some kinds of text information

Downloaded from at Airlangga University on March 10, 2013

are more salient or "potent" (to use Thorndike's 1917 term) than others.


Journal of Reading Behavior

When conducting preliminary studies for this experiment, we noticed that children were able to recall some explicitly stated causes with considerable ac- curacy, yet not others (Nicholson, Note 6). In one story, which was about a "pet lamb1' called "Midnight" (see Figure 1), the question we asked was "why did Mid- night begin to run wildly?" The causal statement which we had placed in the text itself was "a train whistled as it went past." We expected the children to write as an answer "because of the train" or something similar. Instead, we got a wide range of responses. Only two of the twelve children gave the expected response.

Jeremiah went into the store to buy some candy. He took his little black lamb, Mid- night, with him. "I want to buy some candy," Jeremiah said. "I'll keep Midnight right beside me."

"Well, buy your candy quickly and get that trouble-maker out of here," said Mr. Grundy. He smiled as he said it, so Jeremiah knew he wasn't angry. Midnight was so- quiet that Jeremiah kept on looking. It was hard to make a choice. Jeremiah kept on looking.

Then it happened. (Insert—A train whistled as it went past). Midnight began to run wildly around the store. He butted into pans and pails that were hanging on one side of the store. Down they came. He knocked jars and boxes off the candy counter. The floor was covered with candy and broken glass. He leapt wildly over the counter and knocked groceries all over the floor.

Mr. Grundy decided that something had to be done. He picked up a broom and tried to chase Midnight out of the store.

Just then, Uncle Hiram walked into the store. He saw at once what was happening. As the frightened lamb tried to leap over the candy counter, Uncle Hiram caught him. The wild chase was over.

Questions used for this story—

1.Why was it hard to make a choice?

2.Why did Midnight begin to run wildly around the store?

3.Why did Uncle Hiram come into the store?

This result was not unusual except in one way—children did more poorly on this types of causal statement than on many of the others we tried. Some of their answers were—

"he was frightened" "he did not like it" "he was naughty"

"he wanted to get Mr. Grundy" "hewasbord"

"he smalletthe candy"Downloaded from at Airlangga University on March 10, 2013

Where Do Their Answers Come From?


At first, we thought that the explanation was simple. These responses fitted perfectly the "inside-out" view. They reflected children's schema for "pet lambs"—that they were naughty, frisky, very excitable. Yet when we thought about it, the results were also explicable from the "outside-in" point of view. If "the mind is assailed by every word in the paragraph" then we could also argue that the alter- native cues in the text itself were just too "potent"—hence, the range of responses given by the children.

It seemed therefore, that the more stringent test of these opposing views of text comprehension would be to embed two kinds of causal statement in each story. We could create one version of, say, the "pet lamb" story, in which the embedded cause was "unpredictable" (such as "the train whistled...") and another version in which there was a more predictable reason (such as "he wanted to have some fun").

What would be achieved? It would, in fact, be a better test of the validity of the outside-in hypothesis. If the overall context of, say, the "pet lamb" story suggested a "naughty lamb" schema, then we would expect children to give the "predictable" cause more often. If, however, there was no difference in scores for both the "predictable" and "unpredictable" versions, then we could argue that the "outside-in"hypothesis was supported—that comprehension was more text based than schema based.

How would we determine what was "predictable" and "unpredictable"? This could be done by deleting the existing, causal statement—for example, "the train whistled." The story could then be given to children and their task would be to guess the most likely reason, based on overall story context.

In fact, when this "control" version of the "pet lamb" story was tried out, without the "train whistle" cue, we found that the children did, in fact, give answers which seemed to reflect a "naughty lamb" schema. In this version, the "train whis- tle" reason was not given by the children at all. In a sense, it was "unpredictable. "So, we decided to use this procedure as an operational way of selecting causal statements which were either "predictable" or "unpredictable."

Having decided that the type of causal statement may be an important factor in text comprehension, we then wondered whether it was perhaps more important than other variables. Chafe (1977), for instance, had suggested that "fore-grounding" was an important aspect of text cohesion. Yet the effect of this factor may be related to the type of causal statement used in the text. We decided to test this "foregrounding" variable by varying the distance between the causal statement and the target event probed by the question.

The position of the causal statement was also varied by placing it either before or after the text event to which it was related. Beaugrande and Colby (1979) have argued that the ideal narrative would put causes and effects in their corresponding temporal order—that is, "cause" followed by "effect."Yet narrators sometimes alter this order to create interest or suspense. We wanted to find out whether changes in position would make the causal statements less salient and thus more like- ly to be overlooked by the reader.

It also seemed important to include "story familiarity" as a factor in the study. After all, an unfamiliar story would be less likely to elicit relevant background

knowledge, simply because the concepts in the story were less likely to be known by

Downloaded from at Airlangga University on March 10, 2013

the reader. Research by Nezworski et al. (Note 7) suggests that story content (or


Journal of Reading Behavior

familiarity) may be a more important variable than the location of information in the text. Presumably, in "schema" terms, there would be more "slots" available than information to "instantiate"them. An "unpredictable" causal statement would be given more attention in an unfamiliar story, as the reader tried to fill in all the "slots" in the schemata. If this occurred, then we would expect children to give more "unpredictable" answers in the "unfamiliar" stories. In contrast, we would expect them to give more "predictable" answers for the "familiar" stories. In other words, the "inside-out" view could account for differences in children's answers, depen- ding on the familiarity of the story. The most powerful test of the "outside-in" view, however, would be with the familiar stories. If there was no difference between the "predictable" and "unpredictable" conditions, then the "outside-in" hypothesis would be supported.

Finally, we decided to include "text access" as a variable in the study. Some researchers have found that text access makes very little difference for short nar- rative stories (Tuinman &Farr, 1972), yet others had found that the effect of access depended on the type of question asked and the difficulty of the story (Nicholson et al. 1979). We felt that access to the text could result in the use of "search and destroy" tactics (Pearson 1978). The children might simply match the question transform to its text equivalent and copy out what was stated in the text. They would get the "answer" but it would not reflect what they did when they usually made sense of stories.

Our study thus focused on five variables. These included causal relation type, story familiarity and text access. In addition, two variables related to text cohesion were also examined—proximity to the target structure in the story, and position in relation to it. Our purpose was to control systematically the effects of these factors on children's ability to why-type questions about explicit causal structures in nar- rative stories.



Subjects. Eighty children (40 girls, 40 bojrs), all eight years of age, were selected from three schools, each representing a cross-section of the population of the city of Hamilton, New Zealand. All children had achieved scores which were average or above on the Progressive Achievement reading tests (NZCER, 1969). Subjects were assigned randomly to treatment groups, the only constraints being that sex ratios and the range of reading ability were the same for each group.

Materials. Six narrative stories, all about 200 words in length and, in our judge- ment, "well-formed" (Rummelhart, 1975), were used. Three were "familiar" to readers of this age; three were "unfamiliar"

Differences in story "familiarity" were established in two steps. First, they were equated for "readability," using two different readability formulae (Dale-Chall 1948, Fry 1975). Having done this, we assumed that any differences in difficulty

among them would be dueDownloadedto variationsfrom jlr.sagepub.comin contentat Airlangga Universityfamiliarityon March.10, 2013

In order to test for these differences, three why-type questions were prepared

WhereDo Their Answers ComeFrom?


for each story. The questions probed events for which no causes were explicitly mentioned in the texts, although the causes would be inferred if the texts had been understood (see Figure 1for sample questions).

A small group (N = 12) of eight year old children, all above average readers, were given the stories and the questions to answer as best they could. We scored their answers as either "sensible" or "not sensible." The results showed that out "familiar" stories elicited more sensible answers than did the "unfamiliar stories" (proportion correct = .74 for familiar; proportion correct = .29 for unfamiliar stories).

Three versions of each story were then created. In the first version, there was no causal statement explicitly stated in the story—this was the "control" version. In the second version, the "predictable" causes were embedded in the story. A predic- table causal statement was defined as one which readers would be likely to give just from the overall story context. For example, "he wanted some fun" was embedded in the "pet lamb" story as a reason for the lamb's strange behavior. In the third ver- sion, "unpredictable" reasons were embedded in the story. These were defined as unlikely to be given by the reader without having carefully read the text—that is, they were not predictable just from the overall story context or from prior knowledge. For example, ' 'the train whistle" was embedded in the '"pet lamb'' story as an alternative explanation, unlikely to be given unless it was explicitly stated in the story itself.

All of the embedded causal statements were text dependent (Tuinman 1973-1974). In other words, they were unlikely to be given unless the children had actually read the story. Even the "predictable" causes, such as "he wanted some fun" could not be easily elicited just be reading the question alone—"Why did Mid- night begin to run wildly?" We wanted to be sure that children had actually read the stories. Text dependency for questions and explicit causes was established by asking another group of children (N = 20) to answer the questions alone, without reading the stories.

The "text cohesion" factors were then incorporated. This meant that further versions of each story were needed. First, the "distance" of each causal statement from the probed event was varied so that it was either "near"—that is, the preceding or following sentence—or 'far1—that is, with an intervening sentence between it and the probed event.

Next, the "position" of each causal statement was varied so that it either came "before" or "after" the probed event.

The result of these text manipulations was the creation of nine story versions (2 causes x 2 distances x 2 positions, as well as the "control" version).

The final factor included was text accessibility. The stories were arranged in booklet form so that some children could easily look back at the story. Others, however, had to turn the page in order to get to the questions. These children were discouraged from looking back. Instead, they were to answer the questions from memory.

Design. In all, the design of the experiment was a 2* x 3 factorial, using

repeated measures. The between subjects factors were distance, position and text

Downloaded from at Airlangga University on March 10, 2013

access; the within subjects factors were story familiarity and type of causal state- ment (see Figure 2).


Journal of Reading Behavior

The children were randomly assigned to the between groups, the only constraint being that there was a balance of boys and girls in each group, and a similar range of reading achievement scores. Stories were randomly assigned to children with the constraint that each child received both types of embedded version as well as the non-embedded version for both familiar and unfamiliar stories. Each story was assigned in random order and was different in content from every other story. In all, the children each read six different stories, answering three questions about each story.

Procedure. Subjects were tested in groups of 20-30 in each school, with both ex- perimenters supervising the administration. The no access group was separated to facilitate close supervision. Children were told that their task was to read each story and then answer the three questions about each story as best they could. They were told they could take as long as they wanted because there was no time limit. Each child was then given a test booklet and an answer sheet. After the children had finished answering the questions for three of their stories, they were given a short break. They then completed the final three stories. Children were instructed not to talk about the stories during the break.

Analysis. Responses were analyzed according to what we called a "rewrite" criterion, where the meaning of the child's response was essentially the same as that of the embedded causal statement, even though there may have been slight lexical or syntactic changes. It was decided that a "rewrite" rather than an "exact" criterion be used, simply because very few responses were exact matches of the embedded causal statements (Xpredictable = .1083; unpredictable = .0854). This result in itself was interesting and suggested that these children did not use "search and destroy" techniques—that is, they did not seem to copy answers straight from the text.

Using the "rewrite" criterion for the "pet lamb" story, where the embedded reason was, say, "a train whistled as it went past,"a rewrite response would have been " a train wentpastto oting its horn" or "he heard a loud noise."

In the "control" versions, all answers were scored as "rewrites" if they had the same meaning as the predictable or unpredictable causal statements given in the other versions. In fact, however, hardly any qf the unpredictable reasons were guessed—less than one per cent. This supported our criteria for determining un- predictability. Some of the "predictable" reasons were guessed, however, and these were scored as rewrites.

Inter-scorer reliability coefficients (R = .95) indicated that scoring criteria could be reliably applied to responses, especially when they were scored in terms of "rewrite,""sensible" and "not sensible."

The data were analyzed using five-way ANOVA procedures. Since a con- siderable number of F tests were being made, only results with p < .01 were con- sidered significant.

Calculations of conditional probability were also carried out in order to assess the extent to which the "predictable" and "unpredictable" causal statements could be guessed, even if they were not explicitly stated in the text. In this way, we could gauge the extent to which the "predictable" causes were text dependent and

Downloaded from at Airlangga University on March 10, 2013

whether the "unpredictable" were in fact unpredictable, especially when story con- text was available to the reader.




o Q
























Unfamiliar stories

Familiar stories


Control Predictable Unpredictable Control Predictable Unpredictable







































Figure 2. A model of the design of experiment 1. Solid lines specify the between subjects conditions broken lines the within subjects condi- tions. Within each of the 48 cells, 10students each answered 3 why-type questions, resulting in a total of 1440data points.

Downloaded from at Airlangga University on March 10, 2013


Journal of ReadingBehavior

Results and Discussion.

Causal type. There were three types of causal statement—"control," predictable and unpredictable. The ANOVA showed an_effect for this factor, F(2, 144) = 39.57, p < .01 (X control = .2438, Xpred = .5250, XUnpred = .4250). Follow up contrasts, using Student-Newman-Keuls procedures (Winer 1972] showed that the predictable causes in the text were elicited significantly more often than the unpredictable causes, and that both those types of cause were elicited more often than in the "control" version, which contained no causal statement. In the "control" version, children's guesses were similar to the embedded causes in about 25 percent of the cases. These guesses, however, almost always were similar to the predictable causal relations, as would be expected, since these were intended to "fit in" with the meaning suggested by the overall story context.

Put briefly, these results indicated that children were more likely to give predic- table rather than unpredictable causal statements as answers to why-type questions. The extent to which they did this, however, seemed to depend on other factors, since causal type also interacted with both "distance" and "story familiarity." The nature of these interactions are discussed under the results for these other factors.

Story familiarity. There was no effect for story familiarity, F (i,72) = 5.60, p >

.01 (Xfam = .3694, Xunfam = .4264), although the result was almost significant. There was, however, _an interaction between story familiarity and causal type^ F (2, i44) =

4.91, p_< .01 (Xfam/cont = .2650, Xfam/pred = .5042, Xfam/unpred = .3417, Xunfam/cont =

.2250, Xunfam/pred = .5458, Xunfam/unpred = .5083). The nature of the interaction sug- gested that children were more likely to take notice of unpredictable causal statements when they were answering questions about an unfamiliar story.

Distance. There was a significant effect for the "distance" factor, F n, 72) = 14.84, p < .01 (Xnear = .4569, Xf.^ = .3389). Distance, however, interacted with

causal type, F(2, m)= 7.05, p < .01(X^/coyi = .2333, XBmlpit& = .6208,XMM/impred =

.5167, Xfu/Cont = .2542, Xfu/pred = .4292, Xfai/nnpred = .333). Inspection of the means revealed that the interaction was due mainly to the "control" version, which chang- ed only slightly for the "near" and "far" versions. This was to be expected, of course, since this version remained the same in terms of the distance factor. In other words, it seemed that children's ability to give an embedded text reason as an answer to a why-type question was in fact dependent on the distance of the answer from the probed event. The nearer the answer to the event, the more likely it was to be given as an answer. This result supported the notion of "foregrounding."

Position. There was no effect for position, Pp, 72) = 0.74, p < .01 (Xbefore =

.4111, Xaftet = .3847). Although it seemed logical to us that the cause should precede the event to which it was related (for example, "train whistle" —"lamb runs wildly"), this did not seem to matter to these young fluent readers. It may have been that temporal order was less important than the effect of the causal statement in establishing a "train" scenario which was then linked to the probed event.

Downloaded from at Airlangga University on March 10, 2013

Text Access. There was no effect for text access, F u, 72) = 3.80, p > .01(XacCess = .4278, Xn0 access = .3681). This was an unusual result, since it was expected that

Where Do Their Answers Come From?


children who had access to the text would have made more use of it. Text access, for some of the children, may have been a "hit and miss" affair. Our own observations during the administration sessions had been that some children did not look back, even when encouraged to do so. When we talked to the children later, they gave two reasons for not looking back—first, they usually "remembered" the answer anyway; second, it was "hard" to find the answer when looking back.

Further support for our "hit and miss" interpretation came from an analysis of the children's "non-sensible" responses. We found that children in the "access" groups were much more likely to give answers which were "text intrusions" (Pear- son & Nicholson, Note 8)—that is, answers which seemed to have been taken at ran- dom from the story and which did not make sense. Put briefly, we suspected that children in the "access" group did in fact look back at the text to find answers, but they did not do so in any systematic way. In fact, the process of looking back at the text may have confused rather than helped some children.

Follow-up analysis of causal type reponses. A further analysis was carried out, using the "control,'' predictable and unpredictable versions of each story, to find out the "conditional" probability of the children's answers being either "predictable" in the unpredictable version, "unpredictable" in the predictable version or both types in the "control" version. In the "control" versions, the conditional probability of a child giving the predictable cause was reasonably high (p = .48), although it was very low for the unpredictable cause (p = .02). It was expected that children were more likely to give the predictable causes since, by definition, they were intended to be plausible (but not obvious) causes of the events which were questioned. In the predictable versions of the stories, the conditional probability of a child giving the embedded predictable cause out of all relevant responses was very high (p = .92), and again very low (p = .008) for giving unpredictable causes. In the unpredictable versions, the conditional probability of a child giving the predictable reason, out of all the relevant responses was quite low (p = .15), yet the probability of a child giv- ing the embedded unpredictable reason was quite high (p = .69). In fact, when calculated only for responses to the unfamiliar stores, it was very high indeed (p =


What did we conclude from the results of this first experiment? In general, it seemed that these children's ability to give answers which had the same meaning as the embedded text causes, or even answers which were quite sensible in terms of story context, depended on the distance of the text cause from the text event, on the content familiarity of the story and on the type of causal statement embedded in the story. The effect of text access was less precise and may have been more confusing than helpful for some children.

These results did not support the "outside-in" hypothesis, although they did not destroy it either. As we predicted, story comprehension appeared to be more "schema-based" for the familiar stories and more "text-based" for the unfamiliar stories.

Yet the analyses of conditional probability also indicated that the children at- tended to the embedded text data, particularly when compared with the "control" versions. What seemed to happen was a kind of complex interplay between com- peting yet plausible Downloadedinferencesfrom at Airlanggaby theUniversitytextondataMarch.10,When2013 we re-examined the kinds of inferences generated by the "control" versions of some of the stories, we


Journal of ReadingBehavior

also noticed that some events predisposed children towards a certain kind of in- ference. In the "pet lamb" story, for example, where the question was "why did Midnight begin to run wildly," children's answers tended to favor "internal" causes—that is, the motives of the pet Iamb, such as "he wanted to get Mr. Grun- dy." The children were less likely to give "external" causes such as "Mr. Grundy was trying to hit him with a broom.''

We decided that some types of cause may have been more "potent" than others in different stories and may have been masking the effects of the causal type factor. For example, if children were predisposed toward an "internal" cause in the "pet lamb" story, then this may have interfered with the effect of the embedded "external" cause.

In order to examine this possibility, a "causal preference" factor would need to be included with "causal type." This could be done by embedding systematically "external" causes for some questions and "internal" causes for others. If the children were predisposed toward one type of cause, then the effects of this preference would be controlled.

We decided to test for the effects of "causal preference." If this factor turned out to be important, then we could argue that the results of the first experiment may have been due to the salience of certain preferred causes rather than the fact that the cause fitted the overall story context.

A second experiment also seemed necessary to check on a possible "story ef- fect" caused by the random assignment of stories to subjects in the first experiment. Although each child received each of the six stories in the between subjects condi- tions, the balance of stories in the within subjects conditions was uneven. We suspected that the results for the within factors—causal type and story familiarity—may have been affected by this imbalance of stories.



Subjects. Forty-four children (22 boys, 22 girls), all eight years of age, were selected from two schools. All children were average and above average readers. All subjects received all treatments.

Materials. Four narrative stories were selected from the first experiment—two were "familiar"; two were "unfamiliar." Two versions of each story were created. "Predictable" causes were embedded in the first version and "unpredictable" causes in the second. The "causal preference" factor was varied within each ver- sion. For example, in the "pet lamb" story, there was one question for which inter- nal causes were embedded and a second question for which "external" causes were embedded (see Figure 1for "pet lamb"story):

Question 1 —Why did Midnight begin to run wildly?

External causes—

Predictable = Mr. Grundy stood on Midnight's toe

UnpredictableDownloaded= at Airlanggaas it wentUniversitypaston March 10, 2013

Where Do Their Answers Come From?


Question 2—Why did Uncle Hiram come into the store?

Internal causes—

Predictable = Uncle Hiram wanted to buy something

Unpredictable = Uncle Hiram had a present for Jeremiah

We did not include the text cohesion factors, "distance" and "position" in this experiment. Instead, the causes were placed in the "near-before" position in relation to each target statement—for example, "Mr. Grundy stood on Midnight's toe. Mid- night began to run wildly..." The text access factor was also dropped. Instead, children were encouraged to answer the questions from memory. In other words, we wanted them to rely on their own representations of each story.

Design. In all, the design of the experiment was a 2> factorial, using repeated measures on all factors. The within subjects factors were story familiarity, causal preference and causal type.

The children randomly received all stories and all conditions. The only con- straint being that there was a balance of stories assigned to each within-subjects con- dition.

Procedure. Subjects were tested in groups of 20 in each school, with two resear- chers supervising the administration. The children were given a "practice" story to help them understand the task. The practice story required the children to think about the "best" answer to the question. The children were asked to read each story, then turn the page and answer the questions, without looking back.

Analysis. The data were analyzed using 3 way ANOVA procedures, using the "rewrite" scoring criterion, as in the first experiment.

Results andDiscussion

The results were seemingly very_uncomplicated.JThere was an effect for causal preference, Fti,43) = 11.94, p < .01(Xextenui = .7784,X^mi\ = .6193).There were no effects for story familiarity or causal type. There were no significant interactions, although the interaction of story familiarity, causal preference and causal type was almost significant, F(i,43j = 3.12, p < .08.

These results did indeed provide some support for the "outside-in" hypothesis. When we controlled for the effects of "causal preference," there was no significant difference between embedded causes which were "predictable" or "un- predictable."

When we inspected the results for the near-interaction effect, however, it was clear that we had not controlled all the factors which influenced the question- answering process. For example, in one of the unfamiliar stories, a Maori chief, Te Kanawa and the hunting party which he led, were visited at night by ghosts—

"his men were afraid. They wanted to run away, but it was too dark... Te Kanawa thought he could make the ghosts go away. He took off the tiki that he wore. He hung the tiki on a stick. In the light of the fire, the tiki make a

shadow on the ground. The ghosts stopped singing... Suddenly the ghosts

Downloaded from at Airlangga University on March 10, 2013

went. No one saw them again..."

124 Journal of Reading Behavior

The question was "Why did Te Kanawa hang the tiki on a stick?" We embedded two types of "internal" (motive) causes in the story which could be given as answers—

Predictable = He thought the tiki would scare the ghosts.

Unpredictable = The tiki was a present for the ghosts.

In this particular story, the children seemed to reject the "unpredictable" cause. Instead, they had apparently decided in favor of the more predictable reason. Their answers included—

To try to make the ghosts go away To scare them

To make a shadow on the ground

So the shadow would scare the ghosts away

To make a big shadow on the ground to make it look like a big monster

When we inspected the results for the familiar stories, we found a similar effect in the "possum" story, where the baby possum was saved from a steel trap by his mother—

"...Hop was crying in pain. Hop's mother did the only thing she could. She took him by the back of the neck and pulled. Hop didn't like that. Hop's mother pulled and pulled. At last,. Hop's foot came away from the trap. Bits of fur and foot were left in the trap. He jumped on his mother's back. He clung to her soft fur, crying..."

The question was—"Why did Hop jump on his mother's back?" The two types of "internal" cause which we had embedded were—

Predictable = Hop wanted his mother to help him

Unpredictable = Hop did not want to be punished

In this story, and with this question, we noticed that children seemed to reject both types of causal statement. Instead, they had apparently decided that the reason was basically related to Hop's injury—

He was friten

He felt safer there

His toes nont

He was hurt and scared

He couldent walk

In other words, there were strong competing explanations suggested by the text in both the "possum" and the "Te Kanawa" stories—both explanations were related to "fear." With the "possum" story, the children's answers were still acceptable and relevant. In the "Te Kanawa" story, however, the children's answers were plausible but not compatible with the embedded causal statement.

The impression we got was that children do consider all of the text data but that they often do not appear to do so. The reason for this is that some causal statements, although explicitly stated in the text, are just not as convincing as other competing

explanations suggested by the overall story context. If we want children to attend to

Downloaded from at Airlangga University on March 10, 2013

certain causal statements, then these statements will need to be "competitive" with other, more plausible, inferences.

WhereDo TheirAnswers Come From?



The notion that prior knowledge and text data compete for priority in question- answering is an important one. It explains why children often appear to overlook ex- plicit text reasons for events and give their own reasons instead. We find ourselves agreeing with Thorndike (1917, p. 326)—

"understanding a paragraph is like solving a problem in mathematics. It consists in selecting the right elements of the situation and putting them together in the right relations, and also with the right amount of weight or influence or force for each.''

The structure of the text may facilitate or interfere with this complex process of constructing meaning. Yet text comprehension is not solely determined by "cohesion" factors, even though the distance of relevant causal information from the probed event is clearly very important. What does matter a great deal is the ex- tent to which important causal data are "competitive" with other inferences that could be made.

Recent research on teaching reading comprehension has not given sufficient credit to the part which text data can play in helping children to make better sense of stories. The research emphasis has been on the build-up of prior knowledge. The results have shown that these techniques have been useful for specific stories, but do not generalize to new materials (Tierney &Cunningham 1980).

While not denying the importance of background knowledge in general, we would argue that chilren will learn more from texts by learning how to learn from them—by focusing on "the right elements." If we do not help children to do this, then we may be doing them a disservice. They may learn more about texts by study- ing these "competing" elements within stories—by trying to find out whether "this" matters more than "that."

Some texts are not structured to help children to ask these kinds of questions. As a result, they will appear to have overlooked relevant text information. Yet they may have simply asked their own kinds of questions—and answered them. When this happens, it is difficult not to agree with Smith (1977, p.67)—

"They (teachers) will ask—'But did you understand that the spy's failure to steal the secret plans was really a symbol of man's ineluctable helplessness in the face of manifest destiny?" and if you say, "No, I just thought it was a jolly good thriller," they will tell you that you did not really understand what the story was about. But basically what they are saying is that you were not asking the kind of question they think you should have asked while reading the book."

We think that better comprehension and better teaching may result if we take the "outside-in"view that the text matters. Children may benefit more from ac- tivities which encourage them to find out not only where their own answers come from but also whether better answers are possible (Nicholson 1979).


Downloaded from at Airlangga University on March 10, 2013

The results of this study provide some support for the "outside-in" view that


Journal of Reading Behavior

text data compete with prior knowledge for priority in question-answering. What seems to be the case is that children usually do consider all of the text information, but these data have to compete with "behind the eyeballs" knowledge.

The extent to which text data compete successfully, depends not only on the "cohesiveness" of the text but also on whether important text data conflict with cer- tain "preferred" inferences suggested by overall story content.

This notion of the reader being "assailed every word in the paragraph" is consistent with the findings of other researchers on the kinds of inferences we make when trying to make sense of discourse (Hayes-Roth & Hayes-Roth 1977, Thorndyke & Hayes-Roth 1979, Lehnert, Note 9).


Firstly, the study was limited to young readers, about eight years of age, all average or above in reading ability. The reading behaviors of less fluent or older readers may well differ from those of this sample. Secondly, although the stories were carefully piloted for relative difficulty, these differences were less apparent in the actual study. In other words, if the familiarity gap had been wider among stories, this factor may have been more influential. Thirdly, by manipulating the distance of the causal relations in Experiment 1, especially the "far" condition, we may have altered the natural flow of some of the stories. We tried to avoid this wherever possi- ble but we may not always have succeeded. Finally, these results are limited to nar- rative stories. They many not apply to expository texts.


Although we have reservations about some of the "simulation" research already conducted in this area of text comprehension, the technique itself still has potential to provide another way of analyzing the extent to which question- answering reflects "outside-in" or "inside-out" processing.

We have already begun to experiment with an activity called the "summary game" as a way of probing children's awareness of explicitly stated causal informa- tion in texts. What happens is that children first read a story. They are then given a "summary" of the story which contains inferences "errors" in the sense that they are plausible yet not those which were explicitly stated in the text. The children's task is to "spot the mistake" and then write what they think is the actual inference that should be made. This "simulation" technique seems to be potentially a more naturalistic way of probing the kinds of inferencing problems which confront children when they try to make sense of stories. It also provides a way of tapping in- to the inferences which children sometimes make, but discard in favor of more plausible explanations (Nicholson & Katterns, Note 10, Nicholson 1981).

Other techniques for probing the interaction of text data and prior knowledge also need to be explored. Attempts should be made to add a qualitative dimension to comprehension research. For example, the notion of "concurrent interviewing" in which children are encouraged to talk about their question-answering strategies as they use them, may provide us with considerable insights (Simons 1980).

Downloaded from at Airlangga University on March 10, 2013

Where Do Their Answers Come From?



In this study, the nature of the interaction between text information and prior knowledge in question-answering was investigated. In many ways, the results are a cautionary tale for those who conduct research in this complex area of text com- prehension.

In our first experiment, the findings seemed to support both the "inside-out" and the "outside-in" views of text comprehension. The eight year old readers who participated seemed to overlook important causal details in stories, particularly when those stories were on familiar topics. We thought that the text data had been ignored in favor of children's own presuppositions about why certain text events occurred.

In the second experiment, however, an attempt was made to take into account the fact that certain kinds of causal explanations may be preferred by children. When this is done, the results provided more support for the "outside-in" view of text comprehension. In this experiment, the children seemed to attend to the text data yet sometimes rejected these explicit details in favor of more plausible explanations.

The impression we have gained is that the reader is, indeed, "assailed by every word in the paragraph." When children do not give the answers teachers expect, it is probably because these explicit text data are not as "competitive" as other, more powerful inferences, suggested by the text.

We are left with the view not only that text data are important but that they com- pete for priority in question-answering. Clearly, background knowledge can be very helpful. Yet, in the final analysis, the ability to decide that "this," in the text, is more important than' 'that,'' may be what really matters.


1.PICHERT, J. W., &ANDERSON, R. C. Taking different perspectives on a story, (Technical Report No. 14). Urbana, Ill: University of Illinois, Center for the Study of Reading, November 1976.

2.MEYER, B. T., & FREEDLE, R. The effects of different discourse types on recall. Unpublished manuscript, Educational Testing Service, 1977. (Referenced in R. Freedle (Ed.) Discourse production and comprehension. Norwood, New Jersey: Albex, 1977, 333).

3.ANDERSON, R. C., REYNOLDS, R. E., SCHALLERT, O. L., &GOETZ, E. T. Frameworks for comprehending discourse. (Technical Report No. 12). Urbana, Ill: University of Il- linois, Center for the Study of Reading, July, 1977.

4.BAKER, L. Do I understand or do I not understand: that is the question. (Reading Education Report No. 10). Urbana, Ill: University of Illinois, Center for the Study of Reading, July, 1979.

5.MONTAGUE, A. P. The influence of concept explanations and logical connectives on children's reading comprehension. Unpublished MA thesis, University of Waikato, 1980.

6.NICHOLSON, T. Scripts, stories, and the young reader. Paper presented at ANZAAS Con-

gress, Melbourne, August 1977.

7.NEZWORSKI, T., STEIN, N. L., &TRABASSO, T. Story structure versus content effects on children's recall and evaluative inferences. Technical Report No. 129). Urbana, Ill: University of Illinois, Center for the Study of Reading, June, 1979.

8.PEARSON, P. D., &NICHOLSON, T. Scripts, texts and questions. Paper presented at the

Downloaded from at Airlangga University on March 10, 2013

National Reading Conference, Atlanta, December, 1978.


Journal of Reading Behavior

9. LEHNERT, W. The process of question-answering, (Research Report No. 88), Yale University: Department of Computer Science, 1978.

10.NICHOLSON, T., &KATTERNS, R. Spot the mistake—a way of helping children to make better sense of stories. Unpublished manuscript, University of Waikato, 1981.


BARTLETT, F. C. Remembering. Cambridge: Cambridge University Press, 1932. BEAUGRANDE, R. de, & COLBY, B. Narrative models of action and inaction. Cognitive

Science, 1979, 3, 43-66.

BLACK, J. B., &WILENSKY, R. An evaluation of story grammars. Cognitive Science 1979, 3, 213-230.

BORMUTH, J. R., MANNING, J. C., CARR, J. W., &PEARSON, P. D. Children's comprehen- sion of between- and within-sentence syntactic structures, Journal of Educational Psychology, 1970, 61, 349-357.

BOWER, G. H. Experiments on story understanding and recall. Quarterly Journal of Ex- perimental Psychology, 1976, 28, 511-534.

CHAFE, W. L. Discourse structure and human knowledge. In J. B. Carroll & R. O. Freedle (Eds)

Language, comprehension and the acquisition of knowledge. Washington D. C.: Winston, 1972.

CIRILO, R., & FOSS, D. Text structure and reading time for sentences. Journal of Verbal Learn- ing and Verbal Behavior, 1980, 19, 96-109.

DALE, E., & CHALL, J. A formula for predicting readability. Educational Research Bulletin, 1948, 37(1), 11-20, 28.

FRY, E. The readability principle. Language Arts, 1975, 52, 847-851. HALLIDAY, M. Cohesion in English. London: Longmans, 1976.

HAYES-ROTH, B., & HAYES-ROTH, F. The prominence of lexical information in memory representations of meaning. Journal of Verbal Learning and Verbal Behavior, 1977, 16, 119-136.

IRWIN, J. The effects of explicitness and clause order on the comprehension of reversible causal relationships. Reading Research Quarterly, 1980, 15, 477-488.

KINTSCH, W., & DIJK, T. A. Toward a model of text comprehension and production.

Psychological Review, 1978, 85, 363-394.

MANDLER, J. M., & JOHNSON, N. S. Remembrance of things parsed: story structure and recall. Cognitive Psychology, 1977, 9, 111-151.

MARKHAM, E. M. Realizing that you don't understand; a preliminary investigation. Child Development, 1977, 46, 986-992.

NICHOLSON, T., PEARSON, P. D., & DYKSTRA, R. The effects of embedded anomalies and oral reading on children's understanding of stories. Journal of Reading Behavior, 1979, 11, 339-354.

NICHOLSON, T. An anatomy of reading. Sydney: Horwitz-Graham Books, 1981.

NEW ZEALAND COUNCIL FOR EDUCATIONAL RESEARCH (NZCER). The progressive achievement tests (reading comprehension and reading vocabulary). Wellington: NZCER, 1969.

OLSON, G. M., DUFFY, S. A., & MACK, R. L. Knowledge of writing conventions in probe comprehension. In W. J. McKeachie and K. Eble (Eds.), New directions in learning and teaching. San Francisco: Jossey-Bass, 1981.

OMANSON, R. D., WARREN, W. H. &TRABASSO, T. Goals, inferential comprehension and recall of stories by children. Discourse Process, 1978, 1, 337-354.

PEARSON, P. D., & JOHNSON, D. Teaching reading comprehension. New York: Holt,

Rinehart and Winston,Downloaded1978. from at Airlangga University on March 10, 2013

Where Do Their Answers Come From?


RUMMELHART, D. E. Notes on a schema for stories. In D. G. Bobrow and A. Collins

Representation and understanding: studies in cognitive science. New York: Academic Press, 1975.

RICKARDS, J., & DENNER, P. Depressive effects of underlining and adjunct questions on children's recall of text. Instructional Science, 1979, 8, 81-90.

SCHANK, R. The role of memory in language processing. In C. N. Cofer (Ed.), The structureof human memory. San Francisco: W. H. Freeman, 1975.

SIMONS, H. Verbal reports as data. Psychological Review, 1980, 87.

SMITH, F. Understanding reading. Second edition. New York: Holt, Rinehart and Winston, 1977.

SMITH, F. Conflicting approaches to reading research and instruction. In L. B. Resnick and P. Weaver(Eds.), Theory andpractice of earlyreading. Vol. 2., Hillsdale, N. J.: LEA, 1979.

SPIRO, R. J. Remembering information from text: the "State of schema" approach. In R. C. Anderson and R. J. Spiro (Eds.), Schooling and the acquisition of knowledge. Hillsdale, N. J.: LEA, 1977.

STEIN, N. L., & GLENN, C. G. An analysis of story comprehension in elementary school children. In R. Freedle (Ed.), Discourse processing: multidisciplinaryperspectives. Nor- wood, N. J.: Ablex, 1978.

THORNDIKE, E. Reading as reasoning: a study of mistakes in paragraph reading. Journal of Educational Psychology, 1917, 8, 323-332.

THORNDYKE, P. Cognitive structures in comprehension and memory of narrative discourse.

Cognitive Psychology, 1977, 9, 77-110.

THORNDYKE, P., & HAYES-ROTH, B. The use of schemata in the acquisition and transfer of knowledge. CognitivePsychology, 1979, 11,82-106.

TIERNEY, R., & CUNNINGHAM, J. Research on teaching comprehension. In P. D. Pearson (Ed.) Handbook onresearchin reading. New York: Longman (in press).

TUINMAN, J. Determining the passage dependency of comprehension questions in five major tests. ReadingResearch Quarterly, 1973-1974, 9, 206-223.

TUINMAN, J., &FARR, R. Passage availability when answering passage based questions for long and short reading selections. Measurement and Evaluation in Guidance, 1972, 4, 234-237.

WARREN W., NICHOLAS, D., &TRABASSO, T. Event chains and inferences in understan- ding narratives. In R. Freedle (Ed.) New directions in discourse processing. Vol. 2., Hillsdale, N.J.: LEA, 1979.

WINER, B. J. Statisticalprinciples in experimental design. Second edition. New York: McGraw Hill, 1971.

Downloaded from at Airlangga University on March 10, 2013