A Computer-Based Approach For Deriving And Measuring Individual And Team Knowledge Structure From Essay Questions

J. EDUCATIONAL COMPUTING RESEARCH, Vol. 37(3) 211-227, 2007
A COMPUTER-BASED APPROACH FOR DERIVING AND
MEASURING INDIVIDUAL AND TEAM KNOWLEDGE
STRUCTURE FROM ESSAY QUESTIONS
ROY B. CLARIANA
The Pennsylvania State University
PATRICIA WALLACE
The College of New Jersey
ABSTRACT
This proof-of-concept investigation describes a computer-based approach
for deriving the knowledge structure of individuals and of groups from
their written essays, and considers the convergent criterion-related validity
of the computer-based scores relative to human rater essay scores and
multiple-choice test scores. After completing a classroom-based course,
undergraduate participants in a sophomore-level management course com-
pleted a 100-item multiple-choice final examination and then answered an
extended-response essay question comparing four management theories. The
essays were quantified with ALA-Reader software applying both sentence-
wise and linear lexical aggregate approaches, and then analyzed with
Pathfinder KNOT software. The linear aggregate approach was a better
measure of essay content structure relative to the sentence-wise approach,
with significant Spearman correlations of 0.60 and 0.45 with the human rater
essay scores. The group network representations of low and high performing
students were reasonable and straightforward to interpret, with the high group
being more similar to the expert, and the low and high groups more similar to
each other than to the expert. Suggestions for further research are provided.
Following Anderson (1984), we view structure as the essence of knowledge (p. 5),
and human neural networks are the biologically plausible source of knowledge
structure (Elman, 1993; Rumelhart & McClelland, 1986). In our connectionist
211
Ó 2007, Baywood Publishing Co., Inc.
doi: 10.2190/EC.37.3.a
http://baywood.com

view, knowledge structure is the antecedent underpinning of knowledge and
the precursor of meaning (Goldsmith & Johnson, 1990; Goldsmith, Johnson, &
Acton, 1991). Specifically, an individual’s knowledge structure is an expression
of their language association neural network. Unique in each individual, this
network is a sort of lexicon (i.e., albeit a multi-dimensional mental representation)
that is at the most fundamental level of memory (pre-meaningful). We hold that
measurers of knowledge structure are primarily measures of this association
network (Deese, 1965) but are not necessarily measures of semantic meaning
(e.g., word association approaches), though some do also measure meaning
(e.g., concept maps).
The notion of structure of knowledge has become so main stream that a report
by the National Research Council (2001) stressed the importance of classroom
assessment that measures it. Knowledge structure has been measured in a variety
of ways including card-sorting tasks, network diagrams, concept maps, and
Pathfinder networks derived from word association data. Jonassen, Beissner,
and Yacci (1993) provide details for 22 different ways to measure knowledge
structure.
Compare-contrast type essay questions have also been used to assess relational
understanding that is part of knowledge structure (Gonzalvo, Canas, & Bajo,
1994). Goldsmith, Johnson, and Acton (1991) state “Essay questions, which
ask students to discuss the relationships between concepts, are perhaps the most
conventional way of assessing the configural aspect of knowledge” (p. 88). It
is rather critical to keep in mind that an essay contains different kinds of infor-
mation, and that the scoring approach determines what is actually measured
(Nitko, 1996), and most if not all essay scoring approaches do not intentionally
measure knowledge structure.
But whether it is intentionally measured or not, essays contain a reflection of an
individual’s knowledge structure. For example, essay questions require students to
recall information about a topic, select the most relevant aspects in the light of
the particular question prompt, and then to organize the material into a logical
and coherent response (Taber, 2003). Recalling, organizing, and integrating ideas
while writing requires far more structuring of response than that required by
objective test items (Jonassen & Wang, 1992; Valenti, Neri, & Cucchiarelli, 2003)
and this structure ought to reflect the individual’s internal knowledge structure.
Thus, essays contain many aspects of knowledge structure, but as Goldsmith,
Johnson, and Acton (1991) point out, “there is no simple and objective method to
derive the structural relations from the written answers” (p. 88).
This current investigation purports to provide a simple and objective method
for deriving and measuring individuals (and teams) knowledge structure from
essays. In addition, two alternate scoring methods are considered, a within-
sentence aggregation approach that focuses on key concepts within sentences
(Koul, Clariana, & Salehi, 2005) and a new linear aggregation approach that
considers key concepts both within and across sentences.
212 / CLARIANA AND WALLACE

CONVERTING ESSAYS INTO CONCEPT MAPS
Lomask, Baron, Greig, and Harrison (1992) devised a method for scoring
student essays based on essay content structure. As part of the Connecticut
statewide assessment of science content knowledge, trained teacher raters
converted student essays into concept maps and then hand scored the concept
maps as a measure of science content knowledge and comprehension (Shavelson,
Lang, & Lewin, 1994). For example, students were asked to “describe the possible
forms of energies and types of materials involved in growing a plant and explain
fully how they are related.” One student responded in part:
Plants need solar energy, water, and minerals. The chlorophyll is used to
give it color. And I also think that the plant grows toward the suns direct
(Lomask et al., p. 22).
This text sample is shown as a concept map in Figure 1. Plants are the central
and most connected concept node in the map representation of this text. Note
that the terms in the map differ from the student’s written text response because
several more or less equivalent terms (synonyms and metonyms) can stand for
the same concept (Shavelson, Lang, & Lewin, 1994, p. 7).
The resulting concept maps were scored by Lomask’s team relative to an expert
map based on (a) the number of concept terms used by the student, (b) the number
of correct links between concepts, and (c) the expected number of links among all
of the concepts mentioned in an essay. These concept map scoring features in
Lomask’s et al. rubric considered the extent of knowledge, declarative knowledge,
and one aspect of knowledge structure.
DERIVING AND MEASURING KNOWLEDGE STRUCTURE / 213
Figure 1. Concept map derived from a portion of a student’s
essay (in Lomask et al., 1992).

SOFTWARE FOR CONVERTING ESSAYS
INTO NETWORK GRAPHS
Along these lines, Clariana (2004) developed a software tool called
ALA-Reader (ALA, analysis of lexical aggregates) for converting short essays into
concept map-like Pathfinder network representations (PFNet) by aggregating
pre-selected key concepts from the essays at the sentence level (Shavelson, 1972).
For example, these two sentences taken from a participant’s essay in the present
investigation (with the key words underlined) were analyzed by ALA-Reader and
then represented as a PFNet (see Figure 2).
Humanists believed that job satisfaction was related to productivity. They found
that if employees were given more freedom and power, then they produced more.
ALA-Reader aggregated the individual sentence networks into a Pathfinder
proximity data file. Then Knowledge Network Organizing Tool software (KNOT,
Schvaneveldt, 1990) was used to convert the proximity data into a single network
PFNet representation of the essay (see Figure 2). Productivity is the most con-
nected concept term with five links (degrees) and so is the central term.
PFNet representations of essays from ALA-Reader can be compared to the
PFNet representations of a referent PFNet (i.e., an expert essay) to produce a
measure of structural similarity between the student and expert essays. Using this
approach, a student’s essay can be reduced to a number that represents the amount
of similarity between the student’s essay and the expert referent. Similarity is
defined as the intersection of the links in common between the student and expert
PFNets divided by the union of all links in those two PFNets.
Using this ALA-Reader sentence aggregation approach (e.g., the co-occurrence
of key concepts within each sentence of the essay), Koul, Clariana, and Salehi
(2005) considered the criterion-related validity of ALA-Reader essay scores. In
pairs, teachers (N = 22) enrolled in a graduate-level course researched a science
topic online and created a concept map of the topic. Later, individually they
wrote a short essay from their concept map. The concept maps and essays were
scored by the computer-based tools and by human raters using rubrics. The
ALA-Reader computer-based essay scores, aggregated at the sentence level, were
an adequate measure of essay content compared to human-rater scores (r = 0.71).
SENTENCE LEVEL AGGREGATION VERSUS
AGGREGATION ACROSS SENTENCES
The present investigation uses ALA-Reader to aggregate key text terms at
the sentence level (within sentences), but adds another method that aggregates
terms both within and across sentences (linearly). This new linear approach was
developed because (a) the sentence aggregation approach obtains disconnected
graphs, (b) writing is a serial artifact of a multi-dimensional mental representation,

DERIVING
AND
MEASURING
KNOWLEDGE
STRUCTURE
/
215
Figure 2. ALA-Reader sentence-wise PFNet representations of the first sentence (left panel), the second sentence
(middle panel), and the combined aggregate of the first and second sentences (right panel).

and (c) because of the body of research that shows the influence of serially
contiguous associations on cognitive processing, such as free association (Deese,
1965), associative models of memory (Anderson, 1983; Shavelson, 1974), the
effects of text signals (Brooks, Dansereau, Spurlin, & Holley, 1983), and Bayesian
essay scoring (Rudner & Liang, 2002). On the other hand, Landauer, Foltz, and
Laham (1998) argue that word order is not very important for indexing an essay
(Landauer, Laham, Rehder, & Schreiner, 1997), although their approach, Latent
Semantic Analysis (LSA), must depend on the serial contiguity of words at least
partially since the linear order of words in the training texts inevitably and
profoundly influences the structure of the final LSA semantic space.
Therefore, a linear aggregation approach was developed and added to the
ALA-Reader software for this investigation. During linear analysis, initially,
the developing network is fairly linear (see left panel of Figure 3), but as key
terms are repeated in the essay, such as productivity, the representation begins
to fold back on itself creating a network structure. This network structure can
be represented as a force-directed graph (see right panel of Figure 3). As with
the sentence-wise approach (compare to the right panel of Figure 2), productivity
is the central and most connected concept term in this text sample; however
the linear approach reduces the text to only six links while the sentence-wise
approach has nine, the same six links plus three other links. In addition, the linear
approach will always produce a network structure that is a connected graph, while
the sentence-wise approach is likely to produce multiple unconnected clusters (a
disconnected graph), especially if the text is not coherent or uses pronouns or other
substitute words extensively (the problem of linguistic anaphoric reference).
Figure 3. Using the same text that was used to derive the PFNet in Figure 2,
the ALA-Reader linear representation of the first and second
sentences (left panel) and the force-directed graph of this
same representation (right panel).

Besides ALA-Reader, there are a number of computer-based tools for scoring
essays (Valenti et al., 2003). Of these, ALA-Reader is most similar to Conceptual
Rater software (called C-Rater) in that both approaches: (1) evaluate whether a
response contains information related to specific domain concepts (writing style is not
assessed), (2) do not require a large collection of graded answers for training, (3) can
score short or long essays, and (4) use a correct-answer referent for grading purposes.
This is important because most of the other computer-based tools are costly to setup,
involving the development of a large dataset of example text and marked responses (at
least 500 essays) either for training (neural network, Bayesian, and natural language
processing approaches) or calibration (setting weight values through correlation
analyses). Further, the training or calibration for these other computer-based essay
scoring tools only apply to a specific essay prompt and content sub-domain, and so
substantial cost is incurred for each new writing prompt. ALA-Reader is neither
prompt nor domain specific, and so does not have a high setup cost. It can score essays
given only a list of important terms and an expert referent essay.
USING ALA-READER TO REPRESENT
TEAM KNOWLEDGE
Perhaps more radically, KNOT software has a data averaging feature that can be
used to combine the proximity files from several students’ essays into a group or
team representation of that content. A group representation can provide a clearer
view of the team’s knowledge structure by averaging out the idiosyncrasies in
individual student responses. Using this feature, it is possible to compare the
knowledge structure of a group during pre- and post-instruction to determine
growth and change of the group; and also to compare subgroups within a group,
for example, low versus high performing students.
SUMMARY
This current investigation reports the results of the ongoing development
of ALA-Reader software by adding a new linear aggregation approach that knits
terms together both within and across sentences. The linear scores will be com-
pared to within sentence aggregation scores as well as to human rater scores and
multiple-choice test scores. In addition, the use of this tool for measuring team
(group) knowledge structure will be considered. This formative information
will contribute to the next round of development of the software tools.
METHOD
Participants
One section of a sophomore-level course that must be taken by all Business
majors, Management 201, consisting of 29 students was selected as a sample of

convenience for this investigation. There were 12 females and 17 males. The
smaller number of female students participating in the study is directly related
to the disproportionate number of female students versus male students that are
enrolled in the Management major.
Instructional Setting
Management 201, Managing in the 21st Century, considers the historical roots
of management theory and the basic functions of management including the
essential management skills for planning, organizing, leading, and controlling a
department or an entire organization The course, which is required of all School
of Business students, utilizes experiential learning activities such as cases, team
work, and exercises in order to realize the role of the manager in a changing
workplace. The course met twice a week for a semester.
During the regularly scheduled examination week at the end of the semester,
all students completed the customary final examination for the course (worth
25% of their final course grade) and also answered an extended-response essay
question for extra credit. The final examination, which was designed to evaluate
the student’s ability to apply concepts learned in the course, consisted of a
100-item multiple-choice test that was comprehensive in scope and was delivered
by computer using ExamView testing software. The final examination score
mean was 78.0 (sd = 8.3).
The essay question was typed by the participants using Microsoft Word
software and the finished essay was printed in the computer lab and submitted
to the course instructor. Participants were given a maximum time of 30 minutes
for the essay. The extended-response essay question stated, “Describe and contrast
in an essay of 300 words or less the following four Management theories:
Classical/scientific management, Humanistic/Human Resources, Contingency,
and Total Quality Management.”
Essays Scored by ALA-Reader and by Raters
The 29 essays contained 7,917 total words, of which 1,289 were unique. The
longest essay was 490 total words, the shortest was 97 words, and the average
essay length was 273 words. The 20 most frequent stop words (i.e., a, and, the;
Luhn, 1959) accounted for about 33% of all words, and the top five stop words in
order are the (5.2% of all words), and (3.3%), of (3.3%), a (3.1%), and to (3.2%).
ALA-Reader software must be provided with a list of key terms (maximum
limit of 30) with synonyms and metonyms. In order to optimize the value of the
information across the entire set of 29 essays, a frequency list containing all of the
content terms used in the participants’ essays was prepared. A common important
term, such as employee, had a high frequency count, occurring 90 times in
the 29 essays. ALA-Reader analysis requires that functionally equivalent terms,

such as employee, worker, and subordinate, be grouped together as the same
concept list word. When the synonyms and metonyms for employee are combined,
the number of occurrences of this concept in the 29 essays increases to 150
times, or about 5 occurrences per essay. The frequency list of terms was used
to create a list of the 30 most important conceptual terms. The 30 important
concept terms arranged in order of frequency of occurrence are: Management
(manager, boss, employer), Employee (workers, subordinates), Company (corpor-
ation, organization), Work (job, task), Classical (scientific), Environment (situa-
tion, place), Productivity (performance, output), Contingency, Efficiency (well,
best), Empowerment (freedom, satisfied), Needs (encourage, appreciate),
Humanistic, Concerns (stress, problem), Leadership, Plan (process, implemen-
tation), Human, Feelings (happy, positive, intrinsic), Customer (consumer,
everyone), Relationship, Motivation, Individual (unique), Goal, Success, Machine
(mass production), TQM, Quality (reputation), Pay (salary), Serve (service),
Measured, and Focus (emphasis).
Next the list of 30 terms was stemmed. Stemming refers to the process of
carefully removing suffixes to obtain word roots that have similar meanings.
For example, “educ” is the common stem for educate, education, educates,
educating, educational, and educated (Rudner & Liang, 2002). Stemming may
be a crucial step in essay analysis, though the potential value of stemming
is not clear. Probably, stemming is better in some cases and not in others,
depending on both the specific terms involved and the essay domain content,
for example whether it is important to distinguish between educated and
educational or not. In previous ALA-Reader studies, stemming was easier and
direct for the biology content used, where specialized technical vocabulary
comprised most of the important terms in an essay. For example, Yager
(1983) estimated that introductory college biology courses have more new
vocabulary, about 3,000 unique terms, than even foreign language courses.
The management essays in this investigation contained relatively fewer
specialized words, and substantially more synonyms than the previous study
involving biology content.
All of the essays were also scored by a pair of human raters using a generic
content rubric. The essay scoring rubric considered whether the content was
clear, relevant, accurate, and concise. The four essay sub-sections (classical,
humanistic, contingency, and TQM) were scored separately and each section
was worth 2.5 points, thus the maximum essay score was 10 points. To simplify
analysis comparisons, the ALA-Score PFNet common subtest raw scores were
linearly transformed to a 2.5 point scale.
The Cronbach alpha reliability for the essay scores based on four sources,
the two raters and the two ALA-Score scores (sentence and linear) are: classical
subtest alpha = 0.70, humanistic subtest alpha = 0.61, contingency subtest alpha =
0.65, TQM subtest alpha = 0.69, and total test alpha = 0.81.

RESULTS
Data analysis consists of the subtest means and standard deviations, evaluation
of the criterion-related validity of the computer-based essay scores relative to the
rater scores and the multiple-choice tests, and finally description of the group
network structure for low and high performing students. Means and standard
deviations for raters’ and for the ALA-Reader essay scores are shown in Table 1.
Rater 2 tended to award relatively higher essay scores, while Rater 1 was
stricter. The ALA-Reader mean scores tended to fall between the two human rater
mean scores, with the ALA-Reader sentence-wise scores greater than the linear-
aggregate scores.
Spearman rank correlation for the combined essay test score (i.e., sum) for
Rater 1 and Rater 2 was rs = 0.71 (see Table 2). Rater 1 total essay scores were
most related to the 100 item multiple-choice test scores, rs = 0.53. Rater 2’s and
ALA-Reader linear-aggregate essay scores were also significantly related to the
multiple-choice test scores, although ALA-Reader sentence-wise essay scores
were not. Shavelson, Ruiz-Primo, and Wiley (2000) reported correlations between
multiple-choice tests and measures of knowledge structure (i.e., concept maps)
in the range of 0.45 to 0.64, and proposed that both measurement methods tap
aspects of declarative knowledge that are interrelated; concept maps reflect the
structure of knowledge while multiple-choice tests reflect the extent of knowl-
edge. Thus the ALA-Reader measures (i.e., knowledge structure) should not
necessarily correlate highly with the multiple-choice test scores (extent of knowl-
edge). The ALA-Reader linear total essay scores correlated significantly with
Table 1. Essay Score Means and Standard Deviations
Essay subtest
Classical* Humanist* Contiguity* TQM* Sum**
Rater 1
Rater 2
ALA-Reader
(sentence)
ALA-Reader
(linear)
1.14
(0.75)
1.84
(0.71)
1.77
(0.50)
1.47
(0.44)
0.98
(0.54)
1.91
(0.52)
1.50
(0.53)
1.21
(0.51)
0.66
(0.57)
1.47
(0.79)
1.09
(0.48)
1.14
(0.63)
0.64
(0.63)
1.53
(0.80)
1.58
(0.84)
0.90
(0.41)
3.41
(2.01)
6.76
(2.34)
5.94
(1.47)
4.73
(1.39)
*2.5 points maximum; **10 points maximum.

Rater 1 and Rater 2 total essay scores (r = 0.60 and 0.45). The ALA-Reader
sentence-wise total essay scores correlated less well with Rater 1 and Rater 2
total essay scores.
Group Comparisons
The students were grouped into high (n = 14) and low (n = 15) groups based on
median split of their human rater essay scores, and then the individual essay
proximity files within each group were averaged to obtain a PFNet for each group.
Then the similarity of the low and high groups’ PFNets and the expert’s PFNet
were computed (see Table 3).
Similarity is defined as the intersection of the links in common between two
PFNets divided by the union of all unique links in the two. Similar to a percent
test score that ranges from 0 to 100%, similarity ranges from zero, no similarity,
to one, perfect similarity. The high group’s essay network structure was more
similar to the expert than was the low group’s (i.e., 0.31 compared to 0.19), and
the high and low groups were most like each other (i.e., 0.41).
Both groups’ network structure (see Figure 4) showed employee and manage-
ment as the central and most connected concepts. The terms company, environ-
ment, individual, productivity, and work were also important concepts for both
low and high performing groups. The four super-ordinate management theory
categories from the essay prompt (classical, humanistic, contingency, and TQM)
were all associated with management. Action terms such as leadership and success
were also associated with management, while emotion-related terms such as
feelings, concerns, and motivation were associated with employee. Both groups
associated productivity and pay, although the high group had a richer repre-
sentation of productivity that included customers and efficiency. The main dif-
ferences between the PFNets displayed in Figure 4 are that the high-performers
had relatively more terms that were central (e.g., with more than 1 link) and the
association structure was more extensive and complex.
Table 2. Spearman Correlations of Essay Scores and
the Multiple-Choice Test Score
MC test R1 R2 ALA (S) ALA (L)
MC test
Rater 1
Rater 2
ALA-Reader (sentence)
ALA-Reader (linear)
1
0.53**
0.43*
0.17
0.39*
1
0.71**
0.47*
0.60**
1
0.29
0.45*
1
0.73** 1
*Sig. at .05. **Sig. at .01.

In general, an expert’s extensive knowledge base is organized into elaborate,
richly connected and integrated structures; whereas novices tend to posses
less domain knowledge and a less coherent organization of it (personal com-
munication from Reviewer B; Chi, Glaser, & Farr, 1988; Chi, Glaser, & Rees,
1982), and this qualitative difference has been shown even for “stronger novices”
compared to “weaker novices” (Zajchowski & Martin, 1993). The PFNets in
Figure 4 clearly demonstrate this idea; the high-performers knowledge structure
as represented by the group PFNet (see the right panel of Figure 4) is visually
more integrated and elaborated than that of the low-performers (see the left
panel of Figure 4).
DISCUSSION
This investigation described a computer-based approach using ALA-Reader
software to aggregate key terms at either the sentence level or else linearly across
sentences. The data from ALA-Reader were analyzed by Pathfinder KNOT
software to derive the knowledge structure of individuals and of groups from their
written essays. Based on the linear aggregation approach compared to human rater
scores (r = 0.60 and 0.45) and to performance on a multiple-choice test (r = 0.39),
the convergent criterion-related validity of the computer-generated essay scores
was supported to some degree. Further, the representations of group knowledge
structure seemed reasonable and consistent with expectations (e.g., the high
group is more similar to the expert than the low group). This supports the premise
that this approach using PFNet representations of ALA-Reader text aggregate
data provides a method for comparing the knowledge structure of sub-groups in
a classroom to each other and to some standard. Thus the approach may also be
useful for measuring team knowledge structure.
There are limitations and caution notes regarding the findings of this
investigation. Writing prompts are a critical consideration in essay writing
and scoring. Following Gonzalvo, Canas, and Bajo’s (1994) use of compare-
contrast type essay questions to assess relational understanding that is part of
knowledge structure, in this investigation, students were asked to “Describe and
Table 3. Similarity of the Expert’s, and the
Low and High Groups’ Essay PFNets
Exp. High Low
Expert
High group (n = 14)
Low group (n = 15)
1
0.31
0.19
1
0.41 1

DERIVING
AND
MEASURING
KNOWLEDGE
STRUCTURE
/
223
Figure 4. PFNets of the low (left panel; n = 15) and high (right panel; n = 14) groups.

contrast . . . the four management theories.” However, all students “described”
but very few of the students followed the “contrast” aspect of the prompt and
so the essays are less likely to show integration or elaboration. Disregarding
the “compare-and-contrast” part of a writing prompt may be quite common
(Biggs, 1992). Such “knowledge telling” results in essays that are little more
than un-integrated lists of terms, and so is less likely to reveal the students’
knowledge structure.
Another limitation of this investigation is that human raters’ essay scores
are not necessarily an appropriate measure of knowledge structure (Shavelson,
Ruiz-Primo, & Wiley, 2000). What do raters actually evaluate when they judge
student essays holistically (Barritt, Stock, & Clark, 1986)? In this case, the
raters were not intentionally looking for the student’s knowledge structure; they
were looking for a meaningful and complete response to the prompt. However,
we hold that among the many other things present in an essay, essays reflect the
structure of the student’s association network (knowledge structure). Note that
the ALA-Reader measure of knowledge structure does significantly correlate
with two raters’ holistic essay scores (e.g., r = 0.60 and 0.45).
Another limitation of this investigation involves the centrality of the expert
essay referent for scoring purposes. The students’ scores depend entirely on
the quality of the expert essay referent, since the score is the number of links
in common between the student essay PFNet and the expert essay referent
PFNet. Though the essay prompt is well constrained, it seems likely that dif-
ferent experts could write different essays, and so the resulting student scores
would vary depending on which expert referent essay was used for the analysis.
Further research is needed in order to specify how to derive the expert
essay referent.
For the express purpose of essay marking, this approach is probably less
powerful than other approaches, such as Latent Semantic Analysis. Although
further research and development are needed, since ALA- Reader is free
shareware, then this approach has promise right now as a low cost measure of
knowledge structure.
Considering the next steps, instruction intends to increase not only the
number of what Novak (2003) refers to as “valid notions” but also to decrease
the number of “invalid notions.” The approach described in this investigation
can be used for instructional purposes such as highlighting correct, incorrect,
and missing propositions in students’ essays. For example, the participants’
network representations contain correct links that agree with the referent
network representations, and these “correct” propositions could be pointed
out, such as with check-plus marks beside sentences, green highlighting, or
underlying. Sentences without such highlights could be reconsidered by the
student. Also, the links contained in the expert essay that were not included
in the student’s essay could be pointed out to the student in various ways, such
as with hints and clues, so that the student could add this missing content to

their essay. Future development of the free ALA-Reader software will include
these features.
REFERENCES
Anderson, J. R. (1983). The architecture of cognition. Cambridge, MA: Harvard
University.
Anderson, R. C. (1984). Some reflections on the acquisition of knowledge. Educational
Researcher, 13(10), 5-10.
Barritt, L., Stock, P. L., & Clark, F. (1986). Researching practice: Evaluating assessment
essays. College Composition and Communication, 37, 315-327.
Biggs, J. (1992). A qualitative approach to grading students. Higher Education Research
and Development Society of Australasia News, 14(3), 3-6
Brooks, L. W., Dansereau, D. F., Spurlin, J. E., & Holley, C. D. (1983). Effects of headings
on text processing. Journal of Educational Psychology, 75, 292-302.
Chi, M. T. H., Glaser, R., & Farr, M. J. (1988). The nature of expertise. Hillsdale, NJ:
Erlbaum.
Chi, M. T. H., Glaser, R., & Rees, E. (1982). Expertise in problem solving. In
R. S. Sternberg (Ed.), Advances in the psychology of human intelligence (pp. 1-75).
Hillsdale, NJ: Erlbaum.
Clariana, R. B. (2004). ALA-Reader. Downloaded May 3, 2006, from
http://www.pesronal.psu.edu/rbc
Deese, J. (1965). The structure of associations in language and thought. Baltimore, MD:
The Johns Hopkins Press.
Elman, J. L. (1993). Learning and development in neural networks: The importance of
starting small. Cognition, 48, 71-99.
Goldsmith, T. E., & Johnson, P. J. (1990). A structural assessment of classroom learning.
In R. W. Schvaneveldt (Ed.), Pathfinder associative networks: Studies in knowledge
organization (pp. 241-253). Norwood, NJ: Ablex Publishing Corporation.
Goldsmith, T. E., Johnson, P. J., & Acton, W. H. (1991). Assessing knowledge structure.
Journal of Educational Psychology, 83, 88-96.
Gonzalvo, P., Canas, J. J., & Bajo, M. (1994). Structural representations in knowledge
acquisition. Journal of Educational Psychology, 86, 601-616.
Jonassen, D. H., Beissner, K., & Yacci, M. (1993). Knowledge structure: Techniques for
representing, conveying, and acquiring knowledge structure. Hillsdale, NJ: Lawrence
Erlbaum and Associates.
Jonassen, D. H., & Wang, S. (1992). Acquiring structural knowledge from seman-
tically structured hypertext. Proceedings of Selected Research and Development
Presentations at the 14th Annual Convention of the Association for Educational
Communications and Technology. (ERIC Document Reproduction Service No.
ED 348 000).
Koul, R., Clariana, R. B., & Salehi, R. (2005). Comparing several human and computer
based methods for scoring concept maps and essays. Journal of Educational
Computing Research, 32(3), 261-273.
Landauer, T. K., Foltz, P. W., & Laham, D. (1998). Introduction to latent semantic analysis.
Discourse Processes, 25, 259-284.

Landauer, T. K., Laham, D., Rehder, R., & Schreiner, M. E. (1997). How well can passage
meaning be derived without using word order? A comparison of latent semantic
analysis and humans. In M. G. Shafto & P. Langley (Eds.), Proceedings of the 19th
annual meeting of the Cognitive Science Society (pp. 412-417). Mahwah, NJ: Erlbaum.
Lomask, M., Baron, J. B., Greig, J., & Harrison, C. (1992, March) ConnMap: Connecticut’s
use of concept mapping to assess the structure of students’ knowledge of science.
Paper presented at the annual meeting of the National Association of Research in
Science Teaching, in Cambridge, MA.
Luhn, H. P. (1959). Keyword-in-context index for technical literature (KWIC index).
(Technical report RC-127, IBM Corporation, Advanced Systems Development
Division). Reprinted in C. K. Schultz (Ed.). (1968). H. P. Luhn: Pioneer of information
science: Selected works (pp. 227-235). New York: Spartan Books.
National Research Council. (2001). Knowing what students know: The science and
design of educational assessment. Washington, DC: National Academies Press.
Nitko, A. J. (1996). Educational assessment of students. Englewood Cliffs, NJ: Merrill.
Novak, J. D. (2003). The promise of new ideas and new technology for improving teaching
and learning. Cell Biology Education, 2, 122-132.
Rudner, L. M., & Liang, T. (2002). Automated essay scoring using Bayes’ theorem.
Journal of Technology, Learning, and Assessment, 1 (2). Retrieved May 5, 2005,
from http://www.jtla.org
Rumelhart, D., & McClelland, J. (1986). Parallel distributed processing (Vols. I and II).
Cambridge, MA: MIT Press.
Schvaneveldt, R. W. (1990). Pathfinder associative networks: Studies in knowledge organ-
ization. Norwood, NJ: Ablex Publishing Corporation.
Shavelson, R. J. (1972). Some aspects of the correspondence between content structure
and cognitive structure in physics instruction. Journal of Educational Psychology, 63,
225-234.
Shavelson, R. J. (1974). Some methods for examining content structure and cognitive
structure in instruction. Educational Psychologist, 11, 110-122.
Shavelson, R. J., Lang, H., & Lewin, B. (1994). On concept maps as potential “authentic”
assessments in science (CSE Tech. Rep. No. 388). Los Angeles, CA: University
of California, Center for Research on Evaluation, Standards, and Student Testing
(CREST).
Shavelson, R. J., Ruiz-Primo, M. A., & Wiley, E. W. (2000). Windows into the mind. Based
on an invited address, Facolta’ di Ingegneria dell’Universita’ degli Studi di Ancona,
June 27, 2000. Retrieved May 5, 2005, from
http://www.stanford.edu/dept/SUSE/SEAL/Reports_Papers/Windows%20into%20
the%20Mind_8_4_03_Final.doc
Taber, K. S. (2003). Examining structure and context—Questioning the nature and
purpose of summative assessment. Seminar presentation to Cambridge International
Examinations, University of Cambridge Local Examinations Syndicate, July 2003.
Retrieved May 04, 2005 from
http://www.leeds.ac.uk/educol/documents/00003134.htm
Valenti, S., Neri, F., & Cucchiarelli, A. (2003). An overview of current research on
automated essay grading. Journal of Information Technology Education, 2, 319-330.
Retrieved May 14, 2005 from
http://jite.org/documents/Vol2/v2p319-330-30.pdf

Yager, R. E. (1983). The importance of terminology in teaching K-12 science. Journal
of Research in Science Teaching, 20(6), 577-588.
Zajchowski, R., & Martin, J. (1993). Differences in the problem solving of stronger and
weaker novices in physics: Knowledge, strategies, or knowledge structure? Journal
of Research in Science Teaching, 30(5), 459-470.
Direct reprint requests to:
Dr. Roy Clariana
30 E. Swedesford Road
Malvern, PA 19355
e-mail: RClariana@psu.edu

A Computer-Based Approach For Deriving And Measuring Individual And Team Knowledge Structure From Essay Questions

A Computer-Based Approach For Deriving And Measuring Individual And Team Knowledge Structure From Essay Questions

More Related Content

Similar to A Computer-Based Approach For Deriving And Measuring Individual And Team Knowledge Structure From Essay Questions (20)

More from Angie Miller (20)

Recently uploaded (20)

A Computer-Based Approach For Deriving And Measuring Individual And Team Knowledge Structure From Essay Questions