JOurn1al or in.lnerr.3du.3- Re--e-arch
..,Oluin e 5, i..ue- - [..3, 2i:,1:14
The Effects of Incorrectly Accessing Domain-Specific Knowledge on
Predictive Inference Generation
The effect of incorrect use of domain-specific knowledge on inference generation was assessed by manipulating
the participants' view of the study's purpose. Half the participants believed the study was testing their
general knowledge of psychology, while the other participants were only told to complete the task. All of
the participants were asked to predict what they thought would come next in a series of stories. There were
also different versions of each story, a low and high constraint version. Results showed that the participants who
had incorrectly accessed their domain-specific knowledge were not aware of the difference between the low and
high constraint versions of the stories, while the control participants read the low constraint stories significantly faster.
Reading comprehension is an essential aspect of college level learning. Nearly every facet of a student's study
life puts some demand on the student's reading comprehension ability, whether they are reading a course
textbook or taking notes from lecture slides during class. Due to the importance of reading comprehension for
college level learning, there have been many studies produced regarding reading comprehension and ways
to improve the reader's skills (e.g. Ainsworth & Loizou, 2003; Magliano, Graesser & Trabasso, 1999; Palinscar
& Brown, 1984). Reading strategies such as making predictive inferences during reading have been found
to particularly helpful for improving learning from texts (McGee & Johnson, 2003). Constraint levels also make
some inferences more difficult because low constraint stories are designed to be more ambiguous. The purpose of
the current study is to explore the factors that enhance or detract from a useful reading comprehension
strategy, specifically making predictive inferences during reading.
In addition to the use of good strategies, domain-specific knowledge has been shown to improve performance
in domains such as chess (Ericsson, 1996; Ericson & Kintsch, 1995; Simon & Chase, 1972) and the domain of
reading comprehension performance (Nusca, 2000). Domain-specific knowledge can be understood as
a concentration of a person's knowledge in a certain subject area. Research also suggests that individuals with
an expertise, or domain-specific knowledge in a certain area, process related information differently than
individuals without domain-specific knowledge. As it pertains to reading comprehension, individuals comprehend
texts better that pertain to their domain-specific knowledge than texts that do not (Recht & Leslie, 1988). People
with domain-specific knowledge also process words and text more quickly in their domain than other people
(Nusca, 2000). As it pertains to the particular reading process strategy under investigation in this study, research
has shown that predictive inferences are also made more quickly and easily in the individual's domain (Fincher-
Kiefer, 1992). Much research actually points to the advantages of making predictive inferences during
reading; therefore it is probable that making predictive inferences in the person's domain would only compound
these advantages. It seems that a combination of these two reading processes, using predictive inferences
and accessing domain-specific knowledge, is extremely beneficial to the reader.
Despite the abundance of published research, the focus of the studies produced to date seems to be skewed. That
is, there have been many studies produced regarding the positive affects of accessing domain-specific knowledge
on performance but there is no research examining if there are any drawbacks to using domain-specific
knowledge. In the current study, it is proposed that if domain-specific knowledge is incorrectly accessed, it
may interfere or inhibit cognitive processes and ultimately have a negative affect on performance. Specifically, it
is hypothesized that accessing incorrect domain-specific knowledge will inhibit predictive inference generation.
The aforementioned research that shows that people with domain-specific knowledge read certain texts differently
(e.g., Recht & Leslie, 1988) supports this hypothesis because it shows that different mental processes are
used during reading. In turn, this research also suggests that accessing inappropriate domain-specific knowledge
may influence other cognitive processes, such as making predictive inferences. Specifically, working-memory may
be significantly affected by the incorrect use of domain-specific knowledge. Research has shown that the
working-memory has a limited capacity and that one's capacity is highly related to reading comprehension
success (Just & Carpenter, 1992; Seigneuric, Erhlic, Oakhill & Yuill, 2000) and is related to making
predictive inferences successfully (Linderholm, 2002; St. George, Mannes, & Hoffman, 1997). If domain-
specific knowledge is being used at the wrong time this could potentially prevent valid information from
being examined, due to lack of resources in working memory. For this reason, it is crucial to examine the effects
of incorrectly accessing domain-specific knowledge on the making of predictive inferences.
Participants were 35 undergraduates (28 females and 7 males), enrolled in cognitive psychology courses. The
first language of all participants was English. It was assumed that since the participants were enrolled in a
cognitive psychology class and therefore had taken at least one other psychology course, they would have
domain-specific knowledge on the subject of psychology concepts and theories. The participants were
compensated for participation in the study with extra course credit.
Participants were randomly assigned to one of four groups. There were two experimental groups and two
control groups. The difference between the experimental and control groups was that the experimental group
was told that the study was regarding their general knowledge of psychology. This experimental manipulation
was meant to force participants to access their domain-specific knowledge, which in this case was inappropriate
for the reading task. The duplicity of the experimental and control groups was to counterbalance the presentation
of two separate versions of the stories. These two versions had a mixture of low and high constraint versions of
the same stories so that all participants would be tested in the low and high constraint condition. It has been
found that constraint levels affect predictive inferences (Linderholm, 2002), so it was likely that constraint
would have an effect on the conditions.
During the study, the participants read a series of stories sentence by sentence on a computer screen. There were
28 experimental stories and 12 filler stories. The stories used had been previously tested and had known
reliability (see Linderholm, 2002). Each story had a low and high constraint version that was designed to elicit
a particular predictive inference. The difference between low and high constraint versions were the degree to
which the story elicited the particular inference. That is, low constraint stories were designed to elicit several
possible inferences, one of which was the target inference. High constraint stories were designed to strongly point
to one particular inference.
At the end of each story, participants were prompted to think about what would come next in the story. That is,
they were to make a predictive inference. They were instructed that once they had their idea to press a key on
the computer and then write down their answer on paper provided. The computer recorded the time it took (in
ms) each participant to arrive at his or her predictive inference. At the completion of the predictive inference
making section, participants were presented with a brief quiz of general psychology questions. This quiz was
meant to both distract those in the control condition and reinforce the experimental group's belief that the purpose
of the study was to test their general psychology knowledge.
In addition to timing the participants' generation of predictive inferences, the predictive inferences were
also analyzed to assess accuracy. That is, participants' predictive inferences were compared to target
prediction words that the texts were originally designed to elicit. There was an inter-rater reliability of 85%
after jointly coding 15% of the data. Participants' predictive inferences were coded as accurate if their response was
a direct match of the anticipated inference or a synonym of the anticipated inference.
All response times that were plus or minus 2 standard deviations from the mean were considered outliers and
were replaced by the mean in the condition. This resulted in the replacement of 4% of the data. An analysis
of variance (ANOVA) was performed using Knowledge Condition and Causal Constraint as independent
variables; response times and accuracy served as dependent variables.
Response times. The analysis showed a main effect of Causal constraint, F (1,31) = 4.71, MSE = 1944535,
p< .05. Participants were faster at generating responses for high (M = 3,921 ms) versus low constraint stories
(M= 4,666 ms). However, this was qualified by a significant interaction between Knowledge Condition and
Causal Constraint, F (1,31 = 4.36, MSE = 1944535, p< .05. A closer inspection of the interaction showed that
those in the control condition were significantly faster at generating responses to high (M = 3,550 ms) compared
to low constraint texts (M = 5,013), t (15) = 2.48, SEM = 590, p< .05. Participants in the experimental
condition, however, were equally fast in the low (M = 4,320 ms) and high constrain conditions ( M= 4,292 ms), t
Accuracy. The ANOVA showed a main effect of Causal Constraint, F (1,31) = 109.43, MSE = .005, p< .001. All of
the participants produced inferences with greater accuracy for high constraint stories (M = 2.9) compared to
low constraint stories (M = .11).
Participants in the experimental condition did not detect the difference in difficulty between the high and
low constraint stories, as shown by response times, whereas participants in the control condition did. The
main finding of this study is that incorrect use of domain-specific knowledge may interfere with predictive
inference generation when an individual encounters a more difficult (or ambiguously structured) narrative text.
Thus, results indicate that there is interference occurring during the reader's initial predictive inference
making processes, which is caused by accessing incorrect knowledge.
Surprisingly, the incorrect use of domain-specific knowledge did not affect the accuracy of the content of
predictive inferences generated by participants. One explanation is that the stories were not difficult enough to
be affected by the accessing of incorrect knowledge. Readers have an extensive knowledge of narrative events,
based on general experience interacting in the world, and this may have superceded the interfering
domain knowledge. Another explanation is that perhaps interference would have occurred had the stories been
more closely related to the subject's domain-specific knowledge. For instance, if stories were not based on day-
to-day occurrences, but rather another scholarly based matter or issue. In this case perhaps the environment of
the story would be set in a psychologist's office as the events that occur un this setting are psychologically related
in nature alone.
There are many practical applications for research related to domain-specific knowledge and the potential
negative effects of accessing it incorrectly. Teachers' instructions likely affect when students access their
domain-specific knowledge. It is quite possible that tangents (e.g. talking about a recent vacation in the middle of
a lecture) from the instruction could cause interference for the students in the sense that students may be
accessing the wrong information. Lesson plans might also relate to this research as it could affect organization
(or transition) between subjects. Further research should be produced to detect if certain subject transitions
decrease or increase correct use of domain-specific knowledge.
These results may apply to study habits outside of the classroom as well. Study habits and how students
organize there study time could be changed to optimize correct use of domain-specific knowledge and
minimize instances of incorrect activation of domain-specific knowledge. For instance, switching back and
forth between textbooks for an English literature and a chemistry course may affect comprehension. That is to say
if the student's attention is divided between the subjects, it is possible that information from the previous
subject would remain in the working memory and, in turn, impair comprehension.
This research serves as a starting point for further analysis of the notion that incorrect use of domain-
specific knowledge may interfere with inference making during reading. The results of this study also suggest
that this incorrect use of domain-specific knowledge may inhibit the reader's ability to distinguish between
various types of information presented in text (e.g., low from high constraint texts) that are differentially useful
for making inferences.
Back to the Journal of Undergraduate Research
College of Liberal Arts and Sciences I University Scholars Program I University of Florida |
ï¿½ University of Florida, Gainesville, FL 32611; (352) 846-2032.
W UNIVERSITY of