Proceedings of the 2nd Workshop on Building Educational Applications Using NLP,
pages 45–52, Ann Arbor, June 2005. c©Association for Computational Linguistics, 2005
Towards a Prototyping Tool for Behavior Oriented Authoring of  
Conversational Agents for Educational Applications 
 
 
Gahgene Gweon, Jaime Arguello, Carol Pai, Regan Carey, Zachary 
Zaiss, Carolyn Rosé 
Human-Computer Interaction Institute/ Language Technologies Institute 
Carnegie Mellon University 
5000 Forbes Avenue, Pittsburgh, PA 15213 USA 
Ggweon,jarguell,cpai,rcarey,zzaiss,cp3a@andrew.cmu.edu 
 
   
Abstract 
Our goal is to develop tools for facili-
tating the authoring of conversational 
agents for educational applications, and 
in particular to enable non-
computational linguists to accomplish 
this task efficiently.  Such a tool would 
benefit both learning researchers, al-
lowing them to study dialogue in new 
ways, and educational technology re-
searchers, allowing them to quickly 
build dialogue based help systems for 
tutoring systems. We argue in favor of 
a user-centered design methodology.  
We present our work-in-progress de-
sign for authoring, which is motivated 
by our previous tool development ex-
periences and preliminary contextual 
interviews and then refined through 
user testing and iterative design.   
1 Introduction 
This paper reports work in progress towards 
developing TuTalk, an authoring environment 
developed with the long term goal of enabling 
the authoring of effective tutorial dialogue 
agents.  It was designed for developers without 
expertise in knowledge representation, artificial 
intelligence, or computational linguistics.  In our 
previous work we have reported progress to-
wards the development of authoring tools spe-
cifically focusing on robust language 
understanding capabilities (Rosé et al., 2003; 
Rosé & Hall, 2004; Rosé, et al., 2005).  In this 
paper, we explore issues related to authoring 
both at the dialogue and sentence level, as well 
as the interaction between these two levels of 
authoring.  Some preliminary work on the un-
derlying architecture is reported in (Jordan, Ro-
sé, & VanLehn, 2001; Aleven & Rosé, 2004; 
Rosé & Torrey, 2004).  In this paper we focus 
on the problem of making this computational 
linguistics technology accessible to our target 
user population.   
We are developing the TuTalk authoring en-
vironment in connection with a number of exist-
ing local research projects related to educational 
technology in general and tutorial dialogue in 
particular.  It is being developed primarily for 
use within the Pittsburgh Sciences of Learning 
Center (PSLC) data shop, which includes devel-
opment efforts for a suite of authoring tools to 
be used for building the infrastructure for 7 dif-
ferent computer enhanced courses designated as 
LearnLab courses.  These LearnLab courses, 
which are conducted within local secondary 
schools as well as universities, and which in-
clude Chinese, French, English as a Second 
Language, Physics, Algebra, Geometry, and 
Chemistry, involve heavy use of technology 
both for the purpose of supporting learning as 
well as for the purpose of conducting learning 
research in a classroom setting.  Other local pro-
jects related to calculus and thermodynamics 
45
tutoring also have plans to use TuTalk.  In this 
paper we will discuss specifically how we have 
used corpora related to ESL, physics, thermody-
namics, and calculus in our development effort. 
To support this multi-domain effort, it is es-
sential that the technology we develop be do-
main independent and usable by a non-technical 
user population, or at least a user population not 
possessing expertise in knowledge representa-
tion, artificial intelligence, or computational lin-
guistics.  Thus, we are employing a corpus based 
methodology that bootstraps domain specific 
authoring using examples of desired conversa-
tional behavior for the domain. 
2 A Historical Perspective 
While a focus on design based on standards 
and practices from human-computer interaction 
community have not received a great deal of 
attention in previously published tool develop-
ment efforts known to the computational linguis-
tics community, our experience tells us that 
insufficient attention to these details leads to the 
development of tools that are unusable, particu-
larly to the user population that we target with 
our work. 
Some desiderata related to the design of our 
system are obvious based on our target user 
population.  Currently, many educational tech-
nology oriented research groups do not have 
computational linguists on their staff with the 
expertise required to author domain specific 
knowledge sources for use with sophisticated 
state-of-the-art understanding systems, such as 
CARMEL (Rosé, 2000) or TRIPS (Allen et al., 
2001). However, previous studies have shown 
that, while scaffolding and guidance is required 
to support the authoring process, non-
computational linguists possess many of the ba-
sic skills required to author conversational inter-
faces (Rosé, Pai, & Arguello, 2005). Because the 
main barrier of entry to such sophisticated tools 
are expertise in understanding the underlying 
data structures and linguistically motivated rep-
resentation, our tools should have an interface 
that masks the unnecessary details and provides 
intuitive widgets that manipulate the data in 
ways that are consistent with the mental models 
the users bring with them to the authoring proc-
ess.  In order to be maximally accessible to de-
velopers of educational technology, the system 
should involve minimal programming.   
The design of Carmel-Tools (Rosé et al., 
2003; Rosé & Hall, 2004), the first generation of 
our authoring tools, was based on these obvious 
desiderata and not on any in-depth analysis of 
data collected from our target user population.  
While an evaluation of the underlying computa-
tional linguistics technology showed promise 
(Rosé & Hall, 2004), the results from actual au-
thoring use were tremendously disappointing.  
A formal study reported in (Rosé, et al., 2005) 
demonstrates that even individuals with exper-
tise in computational linguistics have difficulty 
predicting the coverage of knowledge sources 
that would be generated automatically from ex-
ample texts annotated with desired representa-
tions. Informal user studies involving actual use 
of Carmel-Tools then showed that a conse-
quence of this lack of ability is that authors were 
left without a clear strategy for moving through 
their corpus.  As a result, time was lost from an-
notating examples that did not yield the maxi-
mum amount of new knowledge in the generated 
knowledge sources.  Furthermore, since authors 
tended not to test the generated knowledge 
sources as they were annotating examples, errors 
were difficult for them to track later, despite fa-
cilities designed to help them with that task.   
Another finding from our user studies was 
that although the interface prevented authors 
from violating the constraints they designed into 
their predicate language, it did not keep authors 
from annotating similar texts with very different 
representations, thus introducing a great deal of 
spurious ambiguity.  Thus, they did not naturally 
maintain consistency in their application of their 
own designed meaning representation languages 
across example texts.  An additional problem 
was that authors sometimes decomposed exam-
ples in ways that lead to overly general rules, 
which then lead to incorrect analyses when these 
rules matched inappropriate examples.   
These disappointing results convinced us of 
the importance of taking a user-centered design 
approach to our authoring interface redesign 
process. 
 
 
46
3 Preliminary Design Intents from 
Contextual Interviews 
The core essence of the user-centered design 
approach is designing from data rather than from 
preconceived notions of what will be useful and 
what will work well.  Expert blind spots often 
lead to designs based on intuitions that overlook 
needs or overly emphasize issues that are not 
centrally important (Koedinger & Nathan, 2004; 
Nathan & Koedinger, 2000).  Contextual inquiry 
is used at an early stage in the user-centered de-
sign process to collect the foundational data on 
which to build a design (Beyer and Holtzbatt, 
2000). Contextual Inquiry is a popular method 
developed within the Human Computer Interac-
tion community where the design team gathers 
data from end users while watching what the 
users do in context of their work. Contextual 
interviews are used to illuminate these observa-
tions by engaging end-users in interviews in 
which they show specific instances within their 
work life that are relevant for the design process.  
These methods help define requirements as well 
as plan and prioritize important aspects of func-
tionality.  At the same time, the system design-
ers get a chance to gain insights about the users’ 
environment, tasks, cultural influences and diffi-
culties in the current processes.  
Many aspects of the Tutalk tool were de-
signed based on contextual inquiry (CI) data. 
The design team conducted five CIs with users 
who have experience in using existing authoring 
tools such as Carmel-Tools (Rosé & Hall, 2004). 
The design team leader also spent one week ob-
serving novice tool users working with the cur-
rent set of tools at an Intelligent Tutoring 
Summer School.  Here we will discuss some 
findings from those CIs and observations and 
how they motivated some general design intents, 
which we flesh out later in the paper.  
A common pattern we observed in our CIs 
was that having different floating windows for 
different tasks fills up the computer screen rela-
tively quickly and confuses authors as to where 
they are in the process of authoring.  The TuTalk 
design addresses this observed problem by an-
choring the main window and switching only the 
components of the window as needed.  A stan-
dard logic for layout and view switching helps 
authors know what to expect in different con-
texts.  Placement of buttons in TuTalk is consis-
tently near the textboxes that they control, and a 
bounding box is drawn around related sets of 
controls so that the user does not get lost trying 
to figure out where the buttons are or what they 
are for.   
We observed that authors needed to refer to 
cheat sheets and user documentation to use their 
current tools effectively and that different users 
did not employ the same terminology to refer to 
similar functionality, which made communica-
tion difficult.  Furthermore, their current suites 
of tools were not designed as one integrated en-
vironment.  Thus, a lot of shuffling of files from 
one directory to another was required in order to 
complete the authoring process.  Users without 
Unix operating system experience found this 
especially confusing.  Our goal is to require only 
very minimal documentation that can be ob-
tained on-line in the context of use.   
TuTalk is a single, integrated environment 
that makes use of GUI widgets for actions rather 
then requiring any text-based commands or file 
system activity.  In this way we hope to avoid 
requiring the users to use a manual or a “cheat-
sheet” reference for the commands they forget. 
As is common practice, TuTalk also uses consis-
tent labels throughout the interface to promote 
understandability and communication with tool 
developers as well as other dialogue system de-
velopers. 
4 Exploring the User’s Mental Model 
through User Studies 
As an additional way of gaining insights into 
what sort of interface would make the process of 
authoring conversational interfaces accessible, 
we conducted a small, exploratory user study in 
which we examined how members of our target 
user population think about the structure of lan-
guage.   
Two groups of college-level participants with 
no deep linguistics training were asked to read 
three transcribed conversations about ordering 
from a menu at a restaurant from our English as 
a Second Language corpus.  The three specific 
restaurant dialogues were chosen because of 
their breadth of topic coverage and richness in 
linguistic expression.  Participants were asked to 
perform tasks with these dialogues to mimic 
47
three levels of conversational interface author-
ing: 
 
Macro Organization Tasks (dialogue level) 
Level 1. How authors understand, seg-
ment, and organize dialogue topics 
Level 2.  How authors generalize across 
dialogues as part of constructing a 
“model” script 
Micro Organization Task (sentence level) 
Level 3.  How authors categorize and 
decompose sentences within these dia-
logues 
 
The first group (Group A, five participants) 
was asked to perform Macro Organization Tasks 
before processing sentences for the Micro Or-
ganization Tasks.  The second group (Group B, 
four participants) was asked to perform these 
sets of tasks in the opposite order. 
Our findings for the Macro Organization 
Tasks showed that participants effectively broke 
down dialogues into segments that reflected in-
tuitive breaks in the conversation.  These topics 
were then organized into semantically related 
categories.  Although participants were not ex-
plicitly instructed on how to organize the topics, 
every participant used spatial proximity as a rep-
resentation for semantic relatedness. Another 
finding was the presence of primacy effects in 
the “model” restaurant scripts they were asked to 
construct. These scripts were heavily influenced 
by the first dialogue read. As a result, important 
topics that surfaced in the other two dialogues 
were omitted from the model scripts. 
Furthermore, we found that participants in 
Group B took much longer in completing the 
Micro Organization Task (35-40 minutes as op-
posed to 25-30 minutes) without performing the 
Macro Organization Tasks first. In general, we 
found that participants clustered sentences based 
on surface characteristics rather than creating 
ontologically similar classes that would be more 
useful from a system development perspective. 
In a follow-up study we are exploring ways of 
guiding users to cluster sentences in ways that 
are more useful from a system building perspec-
tive. 
Our preliminary findings show that getting an 
overall sense of the corpus facilitates micro-
level organization. This is hindered by two fac-
tors:  First, primacy effects interfere with macro-
level comprehension. Second, system developers 
struggle to strategically select portions of their 
corpus on which to focus their initial efforts.  
5 Stage One: Corpus Organization 
While existing tools from our previous work 
required authors to organize their corpus data 
prior to their interaction with the tools, both our 
contextual research and user studies indicated 
that support for organizing corpus data prior to 
authoring is important.   
In light of this concern, the TuTalk authoring 
process consists of three main stages.  Corpus 
collection, corpus data organization through 
what we call the InfoMagnet interface, and au-
thoring propper. First, a corpus is collected by 
asking users to engage in conversation using 
either a typed or spoken chat interface. In the 
case of spoken input, the speech is then tran-
scribed into textual form. Second, the raw cor-
pus data is automatically preprocessed for 
display and interactive organization using the 
InfoMagnet interface.  As part of the preprocess-
ing, dialogue protocols are segmented automati-
cally at topic boundaries, which can be adjusted 
by hand later during authoring propper.  The 
topic oriented segments are then clustered semi-
automatically into topic based classes. The out-
put from this stage is an XML file where dia-
logue segments are reassembled into their 
original dialogue contexts, with each utterance 
labeled by topic. This XML file is finally passed 
onto the authoring environment propper, which 
is then used for finer grained processing, such as 
shifting topic segment boundaries and labeling 
more detailed utterance functionality.   
Our design is for knowledge sources that are 
runable from our dialogue system engine to be 
generated directly from the knowledge base cre-
ated during the fine-grained authoring process as 
in Carmel-Tools (Rosé & Hall, 2004), however 
currently our focus is on iterative development 
of a prototype of the authoring interaction de-
sign.  Thus, more work is required to create the 
final end-to-end implementation.  In this section 
we focus on the design of the corpus collection 
and organization part of the authoring process. 
 
48
5.1 Corpus Collection  
An important part of our mission is developing 
technology that can use collected and automati-
cally pre-processed corpus data to guide and 
streamline the authoring process. Prior to the 
arduous process of organizing and extracting 
meaningful data, a corpus must be collected.  
As part of the PSLC and other local tutorial 
dialogue efforts we have collected corpus data 
from multiple domains that we have made use of 
in our development process. In particular, we 
have been working with data collected in con-
nection with the PSLC Physics and English as a 
Second Language LearnLab courses as well as 
local Calculus and Thermodynamics tutoring 
projects.  Currently we have physics tutoring 
data primarily from one physics tutor (interac-
tions with 40 students), thermodynamics data 
from four different tutors (interactions with 27 
students), Calculus data from four different tu-
tors (84 dialogues), and ESL dialogues collected 
from 15 pairs of students (30 dialogues alto-
gether).  
While we have drawn upon data from all of 
these domains for testing the underlying lan-
guage processing technology for our develop-
ment effort, for our user studies we have so far 
mainly drawn upon our ESL corpus, which in-
cludes conversations between students about 
every-day tasks such as ordering from a restau-
rant or about their pets.  We chose the language 
ESL data for our initial user tests because we 
expected it to be easy for a general population to 
relate to, but we plan to begin using calculus 
data as well.   
5.2 InfoMagnets Interface 
As mentioned previously, once the raw dia-
logue corpus is collected, the next step is to sift 
through this data and assign utterances (or 
groups of utterances) to classes conceptualized 
by the author. Clustering is a natural step in this 
kind of exploratory data analysis, as it promotes 
learning by grouping and generalizing from 
what we know about some of the objects in a 
cluster. For this purpose we have designed the 
InfoMagnets interface, which introduces a non-
technical metaphor to the task of iterative docu-
ment clustering. The InfoMagnets interface was 
designed to address the problems identified in 
the user study discussed above in Section 4.  
Specifically, we expected that those problems 
could be addressed with an interface that:  
1. Divides dialogues into topic based 
segments and automatically clusters 
them into conceptually similar classes 
2. Eliminates primacy effects of sequen-
tial dialogue consumption by creating an 
inclusive compilation of all dialogue 
topics 
3. Makes the topic similarity of docu-
ments easily accessible to the user  
 
The InfoMagnets interface is displayed in 
Figure 1.  The larger circles (InfoMagnets) cor-
respond to cluster centroids and the smaller ones 
(particles) correspond to actual spans of text. 
Lexical cohesion in the vector space translates 
into attraction in the InfoMagnet space. The at-
traction from each particle to each InfoMagnet is 
evident from the particle’s position with respect 
to all InfoMagnets and its reaction-time when an 
InfoMagnet is moved by the user, which causes 
the documents that have some attraction with it 
to redistribute themselves in the InfoMagnet 
space.  
 
 
Figure 1 InfoMagnets Interface 
 
Being an unsupervised learning method, clus-
tering often requires human-intervention for 
fine-tuning (e.g. removing semantically-weak 
discriminators, culling meaningless clusters, or 
deleting/splitting clusters too fine/coarse for the 
author’s purpose). The InfoMagnets interface 
provides all this functionality, while shielding 
the author from the computational details inher-
ent in these tasks 
49
Initially, the corpus is clustered using the Bi-
secting K-means Algorithm described in (Kumar 
et al., 1998).  Although this is a hard clustering 
algorithm, the InfoMagnet interface shows the 
particles association with all clusters, given by 
the position of the particle. Using a cross-hair 
lens, the author is able to view the contents of 
each cluster centroid and each particle. The au-
thor is able to select a group of particles and 
view the common features between these parti-
cles and any InfoMagnet in the space. The inter-
face allows the editing of InfoMagnets by 
adding and removing features, splitting In-
foMagnets, and removing InfoMagnets. When 
the user edits an InfoMagnets, the effect in the 
particle distribution is shown immediately and in 
an animated way.  
5.3 XML format 
The data collected from the conversations 
in .txt format are reformatted into XML format 
before being displayed with InfoMagnet tool.  
The basic XML file contains a transcription of 
the conversational data and has the following 
structure: Under the top root tag, there is <dia-
logue> tag which designates the conversion 
about a topic. It has an “id” attribute so that we 
can keep track of each separate conversation. 
Then each sentence has a <sentence> tag with 
two attributes “uid” and “agent”. “uid” is a uni-
versal id and “agent” tells who was speaking.  
Additionally, sentences are grouped into seg-
ments, marked off with a <subtopic> tag. 
The user’s interaction with the InfoMagnet in-
terface adds a “subtopic-name” attribute to the 
subtopic tag. Then, the authoring interface 
proper, described below, allows for further ad-
justments and additions to the xml tags.  The 
final knowledge sources will be generated from 
this XML based representation. 
6 Authoring 
The authoring environment proper consists of 
two main views, namely the authoring view and 
tutoring view. The authoring view is where the 
author designs the behavior of the conversa-
tional agent. The authoring view has two levels; 
the topic level and the subtopic level. The tutor-
ing view is what a student will be looking at 
when interacting with the conversational agent. 
Our focus here is on the Authoring view. 
Authoring View: Topic Level 
The Topic level of the authoring view allows for 
manipulating the relationship between subtopics 
as well as the definition of the subtopic. Figure 2 
shows the topic level authoring view, which 
consists of two panels. In the left, the author in-
puts the description of the task that the student 
will engage in with the agent. The author can 
specify whether the student will be typing or 
talking, the title of the topic, the task description, 
an optional picture that aids with the task (such 
as a menu or a map of a city), and a time limit.  
In the right panel of the topic level authoring 
view, the structure imposed on the data by inter-
action with the InfoMagnets interface is dis-
played in sequential form. The top section of the 
interface (figure 2, section A) has a textbox for 
specifying an xml file to read. The next section 
(figure 2, section B), “Move / Rename Subtopic” 
displays the subtopics. The order of the subtop-
ics displayed in this section acts as a guideline 
for the agent to follow during the conversation. 
Double-clicking on a subtopic will display a 
subtopic view on the right panel. This view acts 
as a reference for the agent’s conversation 
within the subtopic and is explained in the next 
section. The author can also rearrange the order 
of subtopics by selecting a subtopic and using 
the “>” and “<” buttons to move the subtopic 
right or left respectively. “x” is used to delete 
the subtopic. The author can also specify 
whether the discussion of a subtopic is required 
(displayed in red) or optional (in green) using 
the checkbox that is labeled “required”. Clicking 
on the “Hide Opt” button will only display the 
required subtopics. 
The last section of the right panel in topic 
level authoring view (figure 2, section C) is ti-
tled “move subtopic divider”. A blue line de-
notes the border of the subtopic. The author can 
move the line up or down to move the boundary 
of the subtopics automatically inserted by the 
InfoMagnets interface. The author can also click 
on any part of conversation and press the “split” 
button to split the subtopic in two sections. In 
addition, she can change the label of the sub-
topic segment using the drop down list. 
50
  
 
 
Figure 2: Topic Level Authoring View 
 
Authoring View: Subtopic Level 
While the Topic View portion of the authoring 
interface proper allows specification of which 
subtopics can occur as part of a dialogue, which 
are required and which are optional, and what 
the default ordering is, the Subtopic Level is for 
specification of the low level turn-by-turn details 
of what happens within a subtopic segment.  
This section reports early work on the design of 
this portion of the interface. 
The subtopic view displays a structure that the 
conversational agent refers to in deciding what 
its next contribution should be.  The building 
blocks from which knowledge sources for the 
dialogue engine will be generated are templates 
abstracted from example dialogue segments, 
similar to KCD specifications (Jordan, Rosé, & 
VanLehn, 2001; Rosé & Torry, 2004).  As part 
of the process of abstracting templates, each ut-
terance is tagged with its utterance type using a 
menu-based interface as in (Gweon et al., sub-
mitted).  The utterance type determines what 
would be an appropriate form for a response.  
Identifying this is meant to allow the dialogue 
manager to maintain coherence in the emerging 
dialogue.  Users may also trim out undesired 
portions of text from the actual example frag-
ments in abstracting out templates to be used for 
generating knowledge sources. 
Each utterance type has sets of template re-
sponse types associated with them. The full set 
of utterance types includes Open questions, 
Closed questions, Understanding check ques-
tions, Assertions, Commands/Requests, Ac-
knowledgements, Acceptances, and Rejections. 
The templates will not be used in their authored 
form.  Instead, they will be used to generate 
knowledge sources in the form required by the 
backend dialogue system as in (Rosé & Hall, 
2004), although this is still work in progress.  
Each template is composed of one or more ex-
changes during which the speaker who initiated 
the segment maintains conversational control. If 
control shifts to the other speakers, a new tem-
plate is used to guide the conversation.  After 
each of the controlling speaker’s turns within the 
segment are listed a number of prototypical re-
sponses.  One of these responses is a default re-
sponse that signals that the dialogue should 
proceed to the next turn in the template.  The 
other prototypical responses are associated with 
subgoals that are in turn associated with other 
templates.  Thus, the dialogue takes on a hierar-
chical structure.   
Mixed initiative interaction is meant to 
emerge from the underlying template-based 
structure by means of the multi-threaded dis-
course management approach discussed in (Rosé 
& Torrey, 2004).  To this end, templates are 
meant to be used in two ways.  The first way is 
51
when the dialogue system has conversational 
control.  In this case, conversations can be man-
aged as in (Rosé et al., 2001). The second way in 
which templates are used is for determining how 
to respond when user’s have conversational con-
trol.  Provided that the user’s utterances match 
what is expected of the conversational partici-
pant who is in control based on the current tem-
plate, then the system can simply pick one of the 
expected responses.  Otherwise if at some point 
the user’s response does not match, the system 
should check whether the user is initiating yet a 
different segment.  If not, then the system should 
take conversational control. 
7 Future Plans 
In this paper we have discussed our user re-
search and design process to date for the devel-
opment of TuTalk, an authoring environment for 
conversational agents for educational purposes.  
We are continuing our user research and design 
iteration with the plan of end-to-end system test-
ing in actual use starting this summer. 
 
Acknowledgements 
This work was supported in part by Office of Naval 
Research, Cognitive and Neural Sciences Division 
Grant N00014-05-1-0043 and NSF Grant 
SBE0354420.  

References  
Aleven , V. and Rosé, C. P. 2004.  Towards Easier 
Creation of Tutorial Dialogue Systems: Integration 
of Authoring Environments for Tutoring and Dia-
logue Systems, Proceedings of the ITS Workshop 
on Tutorial Dialogue Systems  
Allen, J., Byron, D., Dzikovska, M., Ferguson, G., 
Galescu, L., & Stent, A. 2000. An Architecture for 
a Generic Dialogue Shell. NLENG: Natural Lan-
guage Engineering, Cambridge University Press, 6 
(3), 1-16. 
Beyer, H. & Holtzblatt, K. (1998). Contextual De-
sign, Morgan Kaufmann Publishers. 
Gweon, G., Rosé, C., Wittwer, J., Nueckles, M. 
(submitted).  Supporting Efficient and Reliable 
Content Analysis with Automatic Text Processing 
Technology, Submitted to INTERACT ’05. 
Jordan, P., Rosé, C. P., & VanLehn, K. (2001). Tools 
for Authoring Tutorial Dialogue Knowledge. In J. 
D. Moore, C. L. Redfield, & W. L. Johnson (Eds.), 
Proceedings of AI-ED 2001 (pp. 222-233). Am-
sterdam, IOS Press. 
Koedinger, K. R. & Nathan, M. J. (2004).  The real 
story behind story problems: Effects of representa-
tions on quantitative reasoning.  The Journal of the 
Learning Sciences, 13(2). 
Nathan, M. J. & Koedinger, K. R. (2000).  Moving 
beyond teachers’ intuitive beliefs about algebra 
learning.  Mathematics Teacher, 93, 218-223. 
Porter, M. 1980.  An Algorithm for Suffix Stripping, 
Program 14 {3}:130 – 137. 
Robertson, S. and Walker, S., 1994.  Some simple 
effective approximations to the 2-poisson model 
for probabilistic weighted retrieval Proceedings of 
SIGIR-94. 
Rosé, C. P., and Torrey, C. (2004). ,DRESDEN: To-
wards a Trainable Tutorial Dialogue Manager to 
Support Negotiation Dialogues for Learning and 
Reflection, Proceedings of the Intelligent Tutoring 
Systems Conference. 
Rosé, C. P. and Hall, B. (2004). A Little Goes a Long 
Way: Quick Authoring of Semantic Knowledge 
Sources for Interpretation, Proceedings of SCa-
NaLu ‘04. 
Rosé, C. P. 2000. A framework for robust semantic 
interpretation. In Proceedings of the First Meeting 
of the North American Chapter of the Association 
for Computational Linguistics, pages 311–318. 
Rosé, C. P., Pai, C., Arguello, J. 2005. Enabling Non-
Linguists to Author Advanced Conversational In-
terfaces Easily. Proceedings of FLAIRS 2005.  
Steinbach, Kepis, and Kumar, A Comparison of 
Document Clustering Techniques, pg. 8. 
http://lucene.apache.org.
