🚀 go-pugleaf

   PART II of Glenberg & Mathew: WHEN MINIMALISM IS NOT ENOUGH:
                                 MENTAL MODELS IN READING COMPREHENSION
                                 Paragraphs 31 - 50, Lines 493 - 911
                                 Tables, References

-------------------------------------------------------------------------
psycoloquy.92.3.64.reading-inference-2.1.glenberg-mathew  Sunday Dec 13 1992
ISSN 1055-0143           (50 paragraphs, 12 references, 4 Tables, 911 lines)
PSYCOLOQUY is sponsored by the American Psychological Association (APA)
                Copyright 1992 Arthur M. Glenberg & Shashi Mathew

                WHEN MINIMALISM IS NOT ENOUGH:
                MENTAL MODELS IN READING COMPREHENSION

                Arthur M. Glenberg
                Department of Psychology
                University of Wisconsin - Madison
                Madison, WI 53706
                glenberg@macc.wisc.edu

                        and

                Shashi Mathew
                Department of Psychology
                Harvard University
                Cambridge, MA 02138
                mathew@isr.Harvard.edu

    0.0 ABSTRACT: McKoon and Ratcliff (1992) contrast a minimalist
    theory of reading with a constructionist or mental-model theory
    "that readers automatically construct a full representation of the
    real-life situation described by a text." We take issue with three
    aspects of McKoon and Ratcliff's arguments. First, the contrast
    between the minimalist and the mental-model account is unwarranted:
    The minimalist account applies to automatic inferences, whereas
    mental models are not often constructed automatically. Second,
    mental models are not full representations of real-life situations.
    Last, we report the results of two experiments demonstrating
    problems with their "salience" account of phenomena previously
    attributed to mental models.

    0.1 KEYWORDS: reading comprehension, mental models, inferences,
    minimalism, salience

----------------------------------------------------------------------
2.0  PART II

2.6.0  EXPERIMENT 1

2.6.1  How does one measure salience?  McKoon and Ratcliff suggest that
salience is related to topicality, and that an associated GM&L target may
be "treated during comprehension as more relevant to the topic of its
discourse" (p. 460) than a dissociated target. To measure salience, we
asked subjects to make paired-comparison salience judgments. Each
subject was presented with a pair of texts labeled Version A or Version
B (see Table 1 for an example). One text in the pair consisted of a
setting sentence and the associated sentence. The other consisted of
the setting sentence and the dissociated sentence. In addition, a
critical word (either the GM&L target or the M&R target) was
identified. The subject's task was to determine the text in which the
critical word was more salient. The subjects were instructed as
follows. "You should decide if the critical word is more important in
Version A or in Version B. That is, if the text continued, would the
critical word play a more important role in Version A or Version B?
Once you have decided, write the letter in the space provided."

_________________________________________________________________________

                        Table 1
                Example of Text Used in Experiment 1

Laundry

Version A: John was preparing for a marathon in August. After doing a
           few warm-up exercises, he put on a sweatshirt from the
           laundry and went jogging.

Version B: John was preparing for a marathon in August. After doing a
           few warm-up exercises, he put his sweatshirt in the laundry
           and went jogging.

_________________________________________________________________________


2.6.2  We will demonstrate that subjects can make these sorts of
salience judgments reliably. The question of interest is then whether
salience is affected by the associated/dissociated variable. If it is,
we would expect the preponderance of choices to be of the associated
version of the text. That is not the result we obtained, however.

2.7.0  METHOD

2.7.1  SUBJECTS. A total of 48 students enrolled in introductory
psychology classes at the University of Wisconsin-Madison participated
in the research. They received extra credit points in payment for their
participation.

2.7.2  MATERIALS AND DESIGN. Each subject received a booklet consisting
of a consent form, two pairs of practice texts (on one page), and 24
pairs of experimental texts (on eight pages). Each pair of texts was
laid out as in Table 1. For half of the experimental pairs in a booklet
the associated text appeared on top and was labeled Version A; in the
other half the associated text appeared on the bottom and was labeled
Version B. There were four versions of the booklet. In two of them the
24 experimental pairs had the M&R target highlighted and in two the
GM&L target was highlighted. Considering the two versions in which the
M&R target was highlighted, in one version a given pair of texts was
presented so that the associated text was on top and in the other
version that pair was presented so that the associated text was on the
bottom. This was also true for the two versions in which the GM&L
target was highlighted.

2.7.3  Each subject received one of the four versions of the booklet.
Thus, there were two between-subjects variables, whether the salience
of the M&R target or the GM&L target was judged, and counterbalancing
of the location (top or bottom) of the associated text in each pair.

2.7.4  PROCEDURE. Subjects were run in groups of up to 12. Within each
group, the four versions of the booklets were randomly distributed
among the subjects under the constraint that by the end of the
experiment a total of 12 subjects received each version. After signing
the consent form, the subjects worked on two practice texts and then
completed the remaining texts at their own speed. Subjects also rated
the difficulty of deciding on the text (in a pair) for which the
critical target was most salient. These data were not analyzed, however.

2.8.0  RESULTS AND DISCUSSION

2.8.1  The basic data consist of the frequencies of choosing the
associated text as the one in which the critical word is most salient
(important). After collecting the data we discovered that one text in
one of the versions was mistyped. We did not score this text; responses
to the other texts were converted to percentages before statistical
analyses were performed.

2.8.2  Before answering the questions of interest, we need to
demonstrate that our measure of salience is sensible. We believe it has
face validity. Furthermore, we can demonstrate that the measure is
reliable. Suppose the degree to which a particular term is salient
varies with the text. That is, in some texts, associating the critical
target with the actor may make it highly salient, whereas in other
texts the salience may be enhanced little by association or it may even
be enhanced by dissociating the target and the actor. If texts differ
in fact in salience of the target items, and if subjects can reliably
judge that salience, then we should observe significant correlations
(computed across texts) between judgments of separate groups of
subjects. To compute these correlations, we divided the subjects into
four groups corresponding to the versions of the booklets they
received. Thus, two of the groups had rated the GM&L target (one in
each order of presentation of the associated and dissociated version)
and two had rated the M&R target. For each group and for each of the 24
texts, we determined the proportion of times the subjects choose the
associated version as most salient. We then computed the correlation
between the two sets of proportions for the counterbalanced versions of
the (24) M&R targets and the correlation between the two sets of
proportions for the counterbalanced versions of the (24) GM&L targets.
These correlations were .59 for the M&R target and .71 for the GM&L
target, p<.01 for both. Clearly, subjects can reliably judge salience.

2.8.3  Now, is salience controlled by the associated/dissociated
variable? Subjects judged the GM&L target to be more salient in the
associated condition 57% of the time. This effect is reliably greater
than 50% (the forced-choice neutral point) by subjects, t(23) = 2.26,
but not by texts, t(22) = 1.63. Furthermore, subjects judged the M&R
target to be more salient in the associated condition only 40% of the
time. This effect is reliably less than 50% by both subjects, t(23) =
-4.41, and by texts, t(23) = -3.34.

2.8.4  These data indicate that salience cannot explain faster
responding (as found in McKoon and Ratcliff's Experiment 5) to the M&R
target in the associated condition compared to the dissociated
condition. The problem is that subjects find the M&R target to be more
salient in the dissociated condition than in the associated condition.
There is some evidence, however, that salience is positively correlated
with the associated/dissociated variable for the GM&L target. One
interpretation of this finding is that salience is indeed the
controlling factor for the GM&L targets. Given the very strong evidence
against salience playing a controlling role in responding to the M&R
target, however, this seems unlikely. An alternative interpretation is
that salience and association/dissociation are correlated, but that the
causal direction is opposite to that supposed by McKoon and Ratcliff
(1992). That is, when objects are represented in a mental model, they
are more likely to be judged as salient than when the objects are
absent from the mental model.

2.9.0  EXPERIMENT 2

2.9.1  This experiment is a conceptual replication of McKoon and
Ratcliff's (1992) Experiment 5, but using rewritten texts. The rational
for the experiment was provided in PART I, paragraphs 4.8 - 4.13.

2.9.0  METHOD

2.10.1  MATERIALS AND DESIGN. We wrote 84 texts. Four of these were
used for practice, 40 were filler texts, and 40 were experimental
texts. The 40 experimental texts had the following characteristics (see
Table 2 for an example). The first sentence introduced a main actor and
a situation. The second, critical, sentence could appear in one of two
forms. In the associated condition, a target object (the GM&L target)
was introduced and spatially associated with the main actor. In
addition, an initial location for the GM&L target, the M&R target, was
described in a prepositional phrase immediately after the GM&L target.
In the dissociated condition, the GM&L target was spatially dissociated
from the main actor. Otherwise, the associated and the dissociated
sentences were very similar because we equated (almost exactly) the
number of words, grammatical structures, and propositional structures of
the two forms of the critical sentence. A third sentence reinforced
implied movement of the main actor from the initial location (the M&R
target). Thus, the third sentence forced a dissociation between the
main actor and the M&R target and the third sentence reinforced any
separation between the main actor and the GM&L target in the
dissociated condition. The recognition probe (the M&R target or the
GM&L target) always appeared after this third sentence. Finally, for
each text we wrote a comprehension question, which could be answered
with a yes or no.

________________________________________________________________________

                        Table 2
        Example of a Text Used in Experiment 2

Setting sentence:       Sonia spent the afternoon reading in her
                        apartment.

Critical (associated):  Realizing that it was such a nice day, she took
                        her book from the table and went outside.

Critical (dissociated): Realizing that it was such a nice day, she
                        left her book on the table and went outside.

Filler:                 She walked down to the lake to enjoy the
                        sunshine.

Comp. Question:         Was Sonia going to walk to the park?

Note: The GM&L target is "book" and the M&R target is "table."
Neither was highlighted in the texts read by subjects.
________________________________________________________________________

2.10.2  The 40 filler texts were designed to hide the nature of the
critical texts. Although they had the same general format as the
experimental texts (an introduction and additional sentences) they
differed in several important respects. First, the correct answer to
the recognition probe was "no" for 30 of the filler texts, whereas it
was "yes" for all of the experimental texts. Second, the probe came
after the first sentence (an average of) 16 times, after the second
sentence (an average of) 16 times, and after the third sentence (an
average of) 8 times, whereas the probe always followed the third
sentence in the experimental texts. Finally, the probe word was a non-
noun for half of the filler texts, whereas it was always a noun for the
experimental texts.

2.10.3  The two variables of interest were manipulated
within-subjects. These were whether the associated or the dissociated
versions of the second sentence were presented and whether the
recognition probe was the GM&L target or the M&R target. Ten texts were
presented in each format. The assignment of the 40 experimental texts
to the four conditions was randomized for each subject. Also, the order
in which the 80 (filler and experimental) texts were presented was
randomized for each subject within the following constraint: Each set
of eight successive texts contained one example of each of the four
experimental conditions and four filler texts.

2.10.4  PROCEDURE. Subjects participated individually. They were
instructed on the use of the computer and given four practice texts.
Each text was preceded by the word "Ready," which was dismissed by the
simultaneous pressing of a microswitch labeled "yes" and another
labeled "no." Each of the first three sentences followed in turn and
each was dismissed in the same way. Upon the dismissal of the third
sentence (in the experimental texts), a probe word appeared in a
Macintosh "dialog" box and was underlined by asterisks. Subjects
pressed the "yes" switch to indicate that the probe word had appeared
in the text and the "no" switch to indicate otherwise. The "yes" label
was always assigned to the switch pressed by subject's dominant hand.
The response to the subject's probe word was followed by feedback, the
comprehension question, the subject's response to the comprehension
question, and feedback on that response.

2.10.5  SUBJECTS. The subjects were run in three waves. The first set
consisted of 24 subjects who participated midway through the academic
semester. Because the random assignment of texts to conditions resulted
in some texts with very few observations in some conditions, we were
unable to perform an analysis by texts at this point. We then collected
data from another 23 subjects who participated near the end of the
academic semester. To our chagrin, adding these new subjects greatly
increased the variability and reduced the statistical significance of
the results. Thus, we added a third set of 12 subjects during the
summer session. The initial 47 subjects participated in exchange for
extra credit in their introductory psychology courses. The remaining
subjects were members of the University's summer community and were
paid for their participation.

2.11.0  RESULTS AND DISCUSSION

2.11.1  We analyzed the probe recognition times when the responses to
both the probe and the comprehension question were correct, calculating
a median recognition time for each subject for each of the four
conditions. The recognition time data that are presented are the means
of the subjects' medians. As noted in the method section, adding the
subjects who participated near the end of the academic semester
increased the variability and reduced the significance of all of our
effects. That these subjects were from a population different from that
of the initial subjects is supported by a subsidiary analysis. In
particular, the second wave of subjects was slower than the first wave
by 126 msec, t(45) = 1.87, p=.07. Thus, we are presenting the data
grouped by wave of subjects.

2.11.2  The means of the median correct recognition times for the first
wave of subjects are presented in the top row of Table 3. Note that
there is quite a large effect of the associated/dissociated variable
(62 msec) for the GM&L target but a slight reversal for the M&R target
(-6 msec). The only effect that was significant in an analysis of
variance was the interaction between the associated/dissociated
variable and type of target, F(1,23) = 6.54, MSE = 4387. The data from
the second wave of 23 subjects are presented in the second line of
Table 3. The only significant effect was for the type of target,
F(1,22) = 7.83, MSE = 18108. Thus, the critical interaction did not
replicate for these subjects. The data from the third wave of 12
subjects are presented in the third line of Table 3. As with the first
wave of subjects, the only significant effect was for the interaction
of type of target and the associated/dissociated variable, F (1,11) =
6.13, MSE = 4326. We also performed an analysis including all of the
subjects, using wave as a between subjects variable. The type of target
had a significant effect, F(1,56)=6.08, MSE=13072, indicating faster
responding to the GM&L target than to the M&R target. The critical
interaction between type of target and the associated/dissociated
variable almost reached the standard level of significance, F(1, 56) =
3.52, MSE = 9231, p = .07. The last line of Table 3 reports the
results from using the data from all 59 subjects to compute a median
recognition time for each condition for each text. The means (over the
40 texts) of these medians are reported in Table 3. In the analysis of
these data, the effect for type of target was significant, F(1, 39) =
8.119, MSE = 30158. The interaction was not significant; however, the
effect of the associated/dissociated variable was significant for the
GM&L target, F(1,39) = 4.36, MSE = 9520, but insignificant for the
M&R target, F=0.

________________________________________________________________________

                                Table 3
                        Mean of Median Recognition Times for Experiment 2

                              M&R Target              GM&L Target

                              Assoc   Dissc   D-A     Assoc   Dissc   D-A

First 24 Subjects              938     932     -6      862    924     62

Second 23 Subjects            1080    1078     -2     1010    992    -18

Third 12 Subjects              982     913    -69      940    965     25

By Text (59 Subjects)         1003    1003      0      902    948     46

________________________________________________________________________

2.11.3  Two types of error rates are presented in Table 4. For the
first type, a text was scored as an error if the subject responded
incorrectly to either the recognition probe or the comprehension
question. Errors of this sort were used to remove responses from the
recognition time analysis. The second type of error rate (in
parentheses) is the proportion of texts for which the subjects erred on
the recognition probe regardless of the response to the comprehension
question. This second type of error rate corresponds to those reported
by McKoon and Ratcliff. For both types of error rate, for all three
waves, and for the analysis by text, the only significant effect was
for type of target: subjects made significantly fewer errors responding
to the GM&L target than in responding to the M&R target. Significantly,
these error data remove any concern for speed-accuracy tradeoffs in
interpreting the interactions found for the recognition data.

________________________________________________________________________

                                Table 4
                      Mean Percentage Errors for Experiment 2

                           M&R Target           GM&L Target
                      Assoc   Dissc   D-A     Assoc   Dissc   D-A

First 24 Ss           22(15)  26(14)  4(-1)   17(7)   21(8)   4(-1)

Second 23 Ss          26(18)  21(13) -5(-5)   16(7)   16(6)   0(-1)

Third 12 Ss           25(16)  24(11) -1(-5)   17(7)   14(5)  -3(-2)

By Text (59 Ss)       24(15)  23(13) -1(-2)   17(7)   18(8)   1(1)


Note: The numbers in parentheses are the percentage errors in responding
to the probe. The other numbers are the percentage of times the probe
or the comprehension question was answered incorrectly.

________________________________________________________________________

2.11.4  It is of course a little disconcerting that the critical
interaction (in the recognition times) was found for the first and
third waves of subjects but not the second. Although we cannot be
certain, it seems reasonable to attribute the failure to replicate to
the oft-discussed problem of end-of-semester subjects. We can take some
comfort in the following. First, the interaction of the
associated/dissociated variable and type of target did replicate for
two different populations (those populations sampled by the first and
third waves). Second, for the M&R target, the associated/dissociated
variable never had a positive effect. Third, the critical interaction
was very close to conventional levels of significance in the analysis
of all of the subjects. Fourth, in the analysis by text, the
associated/dissociated variable was significant for the GM&L target but
not for the M&R target. Overall, the case is fairly secure that the
associated/dissociated variable has a positive effect for the GM&L
target (to which the variable refers), but not for the M&R target. Thus
the prediction derived from the salience account has been disconfirmed.
Further discussion of these results may be found in PART I, Section 1.5.

REFERENCES

Bryant, D. J., Tversky, B., & Franklin, N. (1992). Internal and
external spatial frameworks for representing described schemes.
Journal of Memory and Language, 31, 74-98.

Gentner, D., & Stevens, A. L. (Eds.). (1983). Mental models.
Hillsdale, NJ:  Erlbaum.

Glenberg, A. M., Kruley, P., & Langston, W. E. (in press). Analogical
processes in comprehension:  Simulation of a mental model. In M. A.
Gernsbacher (Ed.), Handbook of Psycholinguistics.

Glenberg, A. M., & Langston, W. E. (1992). Comprehension of
illustrated text:  Pictures help to build mental models. Journal of
Memory and Language, 31, 129-151.

Glenberg, A. M., Meyer, M., & Lindem, K. (1987). Mental models
contribute to foregrounding during text comprehension. Journal of
Memory and Language, 26, 69-83.

McKoon, G., & Ratcliff, R. (1992). Inference during reading.
Psychological Review, 99, 440-466.

McKoon, G, Ratcliff, R., & Seifert, C. (1989). Making the connection:
Generalized structures in story understanding. Journal of Memory and
Language, 28, 711-734.

Morrow, D. G., Greenspan, S. L., & Bower, G. H. (1987). Accessibility
and situation models in narrative comprehension. Journal of Memory and
Language, 26, 165-187.

Palmer, S. E. (1978). Fundamental aspects of cognitive representation.
In E. Rosch & B. Lloyd (Eds.), Cognition and categorization. Hillsdale,
NJ:  Erlbaum.

Radvansky, G. A., & Zacks, R. T. (1991). Mental models and the fan
effect. Journal of Experimental Psychology:  Learning, Memory, and
Cognition, 17, 940-953.

Sanford, A. J., & Garrod, S. C. (1981). Understanding written
language. Chichester:  Wiley.

Sidner, C. L. (1983). Focusing and discourse. Discourse Processes,
6, 107-130.

AUTHOR NOTE

This research was funded in part by an Air Force Office of Scientific
Research grant (89-0367) to Arthur Glenberg. We thank Will Langston for
help in programming, and we thank Heather Rippl for help with
preparation of the stimulus materials and data collection.

----------------------------------------------------------------------
[Note: PART I and PART II, posted in 2 parts, will be archived as a
single article]