An Agent That Helps Children to Author Rhetorically-Structured Digital Puppet Presentations.
- SourceAvailable from: psu.edu[show abstract] [hide abstract]
ABSTRACT: Recent years have witnessed the birth of a new paradigm for learning environments: animated pedagogical agents. These lifelike autonomous characters cohabit learning environments with students to create rich, face-to-face learning interactions. This opens up exciting new possibilities; for example, agents can demonstrate complex tasks, employ locomotion and gesture to focus students' attention on the most salient aspect of the task at hand, and convey emotional responses to the tutorial situation. Animated pedagogical agents offer great promise for broadening the bandwidth of tutorial communication and increasing learning environments' ability to engage and motivate students. This article sets forth the motivations behind animated pedagogical agents, describes the key capabilities they offer, and discusses the technical issues they raise. The discussion is illustrated with descriptions of a number of animated agents that represent the current state of the art.International Journal of Artificial Intelligence in Education. 01/2000; 11:47--78.
- [show abstract] [hide abstract]
ABSTRACT: The authors tested the hypothesis that personalized messages in a multimedia science lesson can promote deep learning by actively engaging students in the elaboration of the materials and reducing processing load. Students received a multimedia explanation of lightning formation (Experiments 1 and 2) or played an agent-based computer game about environmental science (Experiments 3, 4, and 5). Instructional messages were presented in either a personalized style, where students received spoken or written explanations in the 1st- and 2nd-person points of view, or a neutral style, where students received spoken or written explanations in the 3rd-person point of view. Personalized rather than neutral messages produced better problem-solving transfer performance across all experiments and better retention performance on the computer game. The theoretical and educational implications of the findings are discussed. (PsycINFO Database Record (c) 2012 APA, all rights reserved)Journal of Educational Psychology 11/2000; 92(4):724-733. · 3.08 Impact Factor
Conference Proceeding: The Persona Effect: Affective Impact of Animated Pedagogical Agents.[show abstract] [hide abstract]
ABSTRACT: Animated pedagogical agents that inhabit interactive learn- ing environments can exhibit strikingly lifelike behaviors. In addition to providing problem-solving advice in response to students' activities in the learning environment, these agents may also be able to play a powerful motivational role. To design the most effective agent-based learning environment software, it is essential to understand how students perceive an animated pedagogical agent with regard to affective dimen- sions such as encouragement, utility, credibility, and clarity. This paper describes a study of the affective impact of ani- mated pedagogical agents on students' learning experiences. One hundred middle school students interacted with animated pedagogical agents to assess their perception of agents' af- fective characteristics. The study revealed the persona effect, which is that the presence of a lifelike character in an interac- tive learning environment—even one that is not expressive— can have a strong positive effect on student's perception of their learning experience. The study also demonstrates the interesting effect of multiple types of explanatory behaviors on both affective perception and learning performance.Human Factors in Computing Systems, CHI '97 Conference Proceedings, Atlanta, Georgia, USA, March 22-27, 1997.; 01/1997
An Agent that Helps Children to Author Rhetorically-
Structured Digital Puppet Presentations
Paola Rizzo1, Erin Shaw2and W. Lewis Johnson2
1Department of Computer Science
University of Rome “La Sapienza”
Via Salaria 113, 00189 Rome, Italy
2Center for Advanced Research in Technology for Education
Information Sciences Institute, University of Southern California
4676 Admiralty Way, Marina del Rey, CA 90292-6695 USA
Abstract. This paper describes a pedagogical agent that helps children to learn
to author structured presentations about explanations of concepts.
Rhetorical Structure Theory analysis of a source Web page, the agent performs
pedagogical tasks to support the user's understanding of rhetorical relations,
stimulates reflection about the relations between the structure of the original
text and the structure of the presentations, and suggests ways to improve the
user's performance. Upon completion of the authoring, the presentations are
organized into coherent structures that can be performed by animated
characters, or Digital Puppets, in a learning-by-teaching classroom context.
When properly designed, multimedia presentations result in deeper learning,
compared to equivalent textual presentations . Likewise, authoring multimedia
presentations forces students to organize their thoughts and clarifies their
understanding of the subject material, encouraging the development of important
procedural and metacognitive skills while achieving mastery of the subject matter.
To support both multimedia presentation and multimedia authoring, we are
developing a system aimed at young children for authoring and generating Digital
Puppet presentations.Digital Puppets (DPs), like animated pedagogical agents
(APAs) , are animated characters that help learners understand a subject. Like
APAs, DPs use text-to-speech software and a variety of nonverbal gestures to provide
voiceover narration and personalized commentary, which have been shown to be
particularly effective multimedia presentation methods . Like APAs, we expect
DPs to evoke a positive affective response in the viewer, referred to as the persona
effect, which is produced even by lifelike characters that do not perform autonomous
pedagogical behaviors . Unlike APAs, however, DPs are not necessarily intelligent
or autonomous; hence, we call them puppets.
How can we assist students in structuring their knowledge and explanations about a
subject, so as to create coherent and effective presentations? Agents such as Adele,
STEVE, and Herman [5, 6, 7] are designed to generate explanations based upon
knowledge structures that support explanation. But young children have relatively
poor understanding of what counts as causation, evidence, etc. As a consequence,
agent authoring tools such as Diligent  and VIVIDS  are too complex to be used
by children.In contrast, off-the-shelf Web page and presentation building tools
provide little structure, and no guidance as to how to think about the knowledge
students must present.
As a basis for structuring knowledge and explanations we have adopted Rhetorical
Structure Theory (RST) by Mann & Thompson . According to the RST, a coherent
natural language text is structured in terms of functional relations that hold between
parts of it. The relations, such as elaboration, motivation, and evidence, represent
different pragmatic goals within the authoring tool. The goals provide a high-level
structure for presentation authoring.
We focus in this paper on a pedagogical agent that assists with presentation
authoring by monitoring the user’s performance, and by intervening to assist him.
The tutor is embedded in the Digital Puppet System’s authoring tool. Using an RST
analysis of the source text, the agent performs pedagogical tasks to support the user's
understanding of rhetorical relations, stimulates reflection about the relationships
between the structure of the original text and the structure of the presentations, and
suggests ways to improve the user's performance. Upon completion of the authoring,
the authored paragraphs are organized into coherent structures that can be presented
by DPs in a learning-by-teaching classroom context.
2. Related work
The Digital Puppet system is, at its core, a tool for building knowledge
representations, and is similar to systems like Belvedere  and ConvinceMe .
These hypothesis-oriented tools address high school and undergraduate level
scientific inquiry into broad-based problems in a collaborative setting. In contrast, we
present an RST-based tool for young children whose goal is to explain the
relationships of local artifacts (paragraphs of text) in a learning-by-teaching context.
Tools that address rhetorical issues, such as argument construction, e.g., SenseMaker
, and explanationconstruction,e.g.
fundamentally similar, though are not RST- or agent-based.
As far as we know, no computer-based learning environment is aimed at teaching
students how to organize presentations of a target text according to rhetorical
principles. Most Intelligent Language Tutoring Systems (ILTS) focus on vocabulary
and grammar; some systems (e.g. ) help learners plan the essay by dividing it into
functional units like introduction, body, and conclusion, but the text to be written is
self-contained and does not constitute a presentation of some other text.
The Rhetorical Structure Theory is used in the area of natural language processing
for both analyzing and generating texts in terms of rhetorical relations [15, 16].
Burstein et al.  use the RST to automatically summarize GMAT essays produced
by students. Their work differs from ours in that the GMAT essay is a self-contained
text, and the summary provided to the student does not outline the rhetorical structure
of the essay nor of the summary. André et. al  use an RST-based planning system
for controlling an agent that displays animated presentations of Web pages or
multimedia material. Our system is focused on helping the user to author the text of a
presentation, rather than on automatically building multimedia presentations, and the
text authored by the children is organized into a coherent structure using templates,
rather than by means of a planning algorithm. Other works on ITSs use natural
language methods, such as performatives designed from speech acts, for producing
coherent and effective dialogues between system and student . Here, we use the
RST to guide students through the process of building coherent presentations.
3. The Digital Puppet System
Digital Puppet presentations are authored in the Digital Puppet System (DPS), which
was created for authoring a simplified version of Adele (Agent for Distance
Education), a Web-based animated pedagogical agent technology . Whereas Adele
is designed to support simulation-based learning, DPS serves a complementary
functions, to augment Web-based presentations. There are three types of users of the
DPS: designer, author, and viewer. The designer annotates the text of a Web page
using the RST relations; the author, a fourth grader, creates a puppet-enhanced
presentation about a Web page; and the viewer, a classmate, teacher, or even the
author herself, plays the resulting presentation. The DPS, which includes the
pedagogical agent and RST annotation tool [20, 21], is illustrated in Figure 1.
Fig. 1. The DP system, including the pedagogical agent and RST annotation tool
The DPS consists of three main parts: (1) The designer’s tools for annotating the
text, (2) the authoring tool for creating the presentation, which this paper is focused
on, and (3) the browser environment for playing the presentations. The pedagogical
agent is embedded in the authoring tool, where it monitors and assists the author. The
operation of the system can be summarized in the following steps:
1.A Web page on the topic of study is either selected or created by a teacher or
instructional designer, and is input to an RST-based manual annotation tool.
2. The designer, by means of the tool, generates a paragraph-level RST analysis of
the page and produces hierarchical markup tags that represent the tree of
rhetorical relations, that are inserted into the Web page using XHTML.
The RST-annotated Web page is read in and displayed by the DP authoring tool.
The RST agent interacts with the author, explaining the meaning of the RST
relations, highlighting the structure of the page, and suggesting ways to structure
The DP authoring tool is a Java application that enables the learner to further
annotate the Web page for the purpose of adding introductory and explanatory
presentation text. RST relation boxes are presented as a means of outlining, or
structuring, the presentation. The authored text and associated interactive buttons
are inserted into the page with Java Script.
6. At anytime, the presentation-enhanced Web page can be displayed and tested in a
Web browser. The authoring tool creates a display Web page with two frames. In
which controls the Digital Puppet’s animated behavior and communicates with
the client-based text-to-speech engine. In the other frame, it displays the
annotated Web page that includes a synthesis of the authored text for each
explanation. The tool then calls a browser to display the results.
The viewer activates the Digital Puppet presentation by clicking on the
interactive buttons that have been placed before relevant paragraphs, i.e.
annotated paragraphs of the Web page by the DPA tool. The puppet then presents
the material to the viewer. In our initial version of the system, the interaction
between the puppet and viewer is minimal; the puppet may ask questions but
The author’s interface to the DPA tool is shown on the left in Figure 2. The Web
page is displayed in the main window, to the left. The user writes an introduction to
the page in the small window at the top-right. She then selects a paragraph from the
original page and writes an explanation about it in a second small window (bottom-
right). An annotation button is inserted before the main paragraph so that the authored
text can be retrieved for editing, and ultimately, presentation. The authored text is
linked to the original text by means of a cause relation that has been selected by the
author from a list of possible rhetorical relations. The agent monitors the author’s
activities and makes suggestions as appropriate, or provides help on demand via the
Help button. To activate the presentation, the author clicks on a button on the toolbar.
The viewer’s interface to the Web browser and the Digital Puppet is shown on the
right in Figure 2. The Web page is augmented with small buttons that, when pressed,
activate the animated puppet to present explanations that were authored for a
particular paragraph. The puppet uses a client-based text-to-speech synthesizer to
narrate the presentation and an XML-based animation engine to perform the
presentation. The teen persona of the Digital Puppet is designed as a character with
whom younger students can easily identify, and as a role model.
Fig. 2. DP authoring tool interface (left) and the puppet-enhanced Web page (right)
4. Organizing presentations using Rhetorical Structure Theory
According to the RST , a text can be analyzed in terms of a tree of relations, each
of them holding between two non-overlapping text spans named nucleus and satellite
respectively. The nucleus is more important for expressing the writer’s intention, and
is independent of the satellite. A relation specifies a set of constraints on nucleus and
satellite, and a pragmatic effect that the writer intends to produce in the reader. For
example, the “evidence” relation has as intended effect that the reader’s belief in the
nucleus is increased. The following text, taken from , p. 10, and regarding a federal
income tax program, shows an instance of this relation; the first span is the nucleus,
while the second span is a satellite providing evidence for the nucleus: “(1) The
program as published for calendar year 1980 really works. (2) In only a few minutes, I
entered all the figures from my 1980 tax return and got a result which agreed with my
hand calculations to the penny.”
Table 1. Rhetorical relations for the Introductory and the Explanatory Tutorials
Introductory TutorialExplanatory Tutorial
We have selected and partially modified a subset of the relations originally
proposed in , and we have classified them according to their purpose: 1) structuring
an Introductory tutorial, i.e. a text that introduces the Web page as a whole, and 2)
structuring Explanatory tutorials, i.e. texts that are aimed at explaining specific
portions of the Web page (see Table 1). For Introductory relations, the whole text of
the Web page is considered a nucleus, for which three types of satellite information
are to be provided by the author: the introduction, the background, that should
increase the reader’s ability to comprehend the nucleus, and the motivation, that
should foster the reader’s desire to read the original text. As for the Explanatory, the
author is requested to identify some important spans within the original text, and to
provide one or more satellites for each of them, choosing suitable relations. For
example, the author might provide some evidence or restatement for several concepts
mentioned in the original text.
The connection between the original text (i.e., the Web page) and the authored text
is illustrated in Figure 3. Each piece of text written by the author corresponds to a
relation. There is only one introductory presentation, comprising three pieces of text,
one per relation. There may be several explanatory presentations, each comprising at
least one piece of text about a unique paragraph.
Fig. 3. Relations between the original text and the authored texts
5. Pedagogical tasks and knowledge of the tutor
In order to realize RST-based presentations, especially the explanatory ones, the
author should perform three tasks while interacting with the DPA tool: (1) understand
the basic structure of the original text; (2) build sensible relations between the original
text and the texts to be authored; (3) edit the texts of the presentations. The tutor helps
the author perform these tasks by executing several types of behaviors, that can be
classified according their “focus”, i.e. the issue they concern, and to four pedagogical
functions: stimulating reflection, supporting comprehension, encouraging action, and
improving performance (see Table 2 for examples).
Tutor’s behaviors focused on the original text.
Thanks to an a priori RST analysis of the original text, the pedagogical agent is able
to highlight, comment, and explain the following items in the original text:
1. The paragraph containing the basic nucleus, and the most significant satellite
paragraphs. This information helps the author to choose which paragraphs to
work on first, or can be used by the agent for giving feedback to the author about
his choice of paragraphs to annotate.
2. The rhetorical relations holding between paragraphs. This helps the author
understand the structure of the original text, provides him with examples of the
relations he wishes to edit, and gives him guidelines about how to structure his
own explanatory text.
Table 2. Examples of pedagogical behaviors of the agent
Structure of original text Preparation of
Can you think of
other relations you
may write wrt this
When explaining a
result, it is good to
evidence for it
Why don’t you try to
elaborate on this
It would be good to
elaborate on this
What do you think is the most
Do you think what
you have is long
Look, this paragraph is an
example of an evidence
For writing some
evidence, look at the
Try to identify some evidence
in the text. If you can’t, I can
What you have highlighted is
not a generic elaboration, it is
Maybe you could
I suggest that you try
to rewrite the text in
your own words
Tutor’s behaviors focused on relations between original and authored texts.
The RST is a descriptive rather than a prescriptive theory; hence, the tutor can only
give some general suggestions about the relations to choose for linking the authored
paragraphs to paragraphs in the original text, based on preferences shown in Table 3.
The preferences are based on the idea that a relation instance in the original text
should be linked to the most relevant and specific relation instances in the authored
Table 3. Preference-based rules for suggesting relations to author
If relation instance in original text is:
Then agent suggests user author:
CauseResult, then Evidence, then Elaboration
ResultCause, then Evidence, then Elaboration
EvidenceCompare-contrast, then Elaboration
Restatement[no specific suggestion]
Tutor’s behaviors focused on the authored text.
Two kinds of performance variables about the author can be continuously measured
by the agent: global and local. Global variables refer to the relations and relation
instances the author creates, regardless of the specific paragraphs they are attached to.
Local variables refer to the relations and relation instances concerning specific
paragraphs in the original Web page. The variables are listed in Table 4. In the names
of the variables, the prefixes“o-” and “a-” stand for “original” and “authored”
respectively. Some of the variables are computed on the basis of two constants:
total_rels, the number of all the relations the author can choose from (see Table 1);
and total_o-pars, the number of paragraphs in the original text.
The agent can also keep track of other variables that refer to the actual actions
performed by the user: tool_actions, concerning the author’s use of the tool in general
(for example, how often and about what topic help is requested), and task_actions,
regarding the user’s progress during the authoring task (for example, how often he
edits, clears and deletes relations).
By means of these variables the agent can notice, for example, whether the author
is (a) over/under-using some relations; (b) adequately explaining the original text or
ignoring some parts of it; (c) plagiarizing the original text; (d) authoring text that is
too long or short with respect to the original paragraph. The values of the variables
are used for inferring the author’s needs and for triggering agent’s behaviors aimed at,
for example, encouraging him to experiment with all the available relations, or better
explaining to him the overlooked relations.
Table 4. Global and local variables for measuring the authoring task outcomes
of all authored
of each authored
piece of text
of used relations
number of annotated original
relative_a-rels: number of
a-relations authored for
difference between length
of o-paragraph and length
of the a-text
distance between o-
paragraph and a-paragraph
of instances authored
for a given relation
6. Evaluating the RST-based approach
The work presented in this paper is new and ongoing. During the 2001-2002 school
year we are scheduled to perform field trials of the authoring tool at an Elementary
School in Los Angeles Unified school district with English Language Learners
(students whose first language is not English).
preliminary study to evaluate the usability of the tool and understanding of the task,
and a pilot study to assess its effectiveness. To evaluate how well an arbitrary ten-
year-old might create an RST-structured presentation, we asked a fourth grader to
prepare a sample presentation. His answers give us an indication of the challenges we
will face in our future work, including how to parse awkward sentences, how to assist
at both grammatical and semantic levels, and how to synthesize authored components:
Our evaluation plan calls for a
Motivate: “Say if your in the USA and you wanted to go some where cold you would go down to the
southern hemisfear because it would be winter there”
Evidence: “The cause of seasons is that it is on one imaginary line called and axis. The earth rotates around
the axis and form an angle of 23.4. the the ear start to tilt and a cause a certin amout of area.”
Backgrnd: “The axis cause a tilt that mak the north or south hemisfear to receive more sun then the other.”
“Hi my name is Dale Lin. I am Doing a report of seasons.”
The challenge is then to turn the authored results into a coherent presentation. We
may be able to catch some of the grammatical errors before the components are
synthesized, and the authoring tool enables the student test out a presentation before it
is performed for the peer group. However, we expect for younger children, especially
children whose first language is not English, that there will be awkwardness and that
the awkwardness will be a catalyst for learning. The following is an example of one
synthesis of the components above. Italicized text is the student’s verbatim text:
Hi! My name is Oliver, and I will be speaking for Dale today. We are doing a report of seasons. It’s good
to know what causes the seasons. Why? Say if your in the USA and you wanted to go some where cold you
would go down to the southern hemisfear because it would be winter there. What causes the seasons? The
cause of the seasons is that it is on one imaginary line called and axis. The earth rotates around the axis
and form an angle of 23.4. The ear start to tilt and a cause a certin amout of area. The axis cause a tilt that
mak the north or south hemisfear to receive more sun then the other.
For the pilot study , as designed by Mayer, we will work with a class of fourth
grade students. We will evaluate both learning performance and outcomes for both a
puppet and control group, analyzing the performance based on learning process
dimensions such as number of self-explanations, and inquiry episodes generated by
students, to determine whether the Digital Puppet learners scored significantly higher
on process measures of cognitive activity during learning. To study the learning
outcomes, we will analyze the data from a battery of instruments in order to determine
whether the Digital Puppet task improved student learning.
We would like to thank Kate LaBore for her support and insightful comments, Nancy
Chang, for performing the student evaluation, Andrew Marshall, who created the first
animation engine for Adele, Karyn Cordova, who is creating the new animation
engine, Daniel Marcu and Edward Hovy for insightful discussions, and Stefano
Levialdi and Marilena De Marsico for useful comments on a previous version of this
paper. This work was supported by an internal R&D grant from the USC Information
Sciences Institute, and by a scholarship from NATO and the Italian National Research
Council awarded to Paola Rizzo during her visit to the USC.
1. Mayer R. E.: Multimedia Learning. Cambridge University Press, Cambridge, UK (2001)
2. Johnson, W.L., Rickel, J.W., Lester, J.C.: Animated Pedagogical Agents: Face-to-Face
Interaction in Interactive Learning Environments.
Intelligence in Education 11 (2000) 47-78
3. Moreno, R. and Mayer, R. E.: Engaging students in active learning: The case for
personalized multimedia messages. Journal of Educational Psychology, 92 (2000) 724-733
4. Lester, J. C., Converse, S. A., Kahler, S. E., Barlow, S. T., Stone, B. A., Bhogal, R. S.: The
persona effect: Affective impact of animated pedagogical agents. In: Proceedings of CHI '97
International Journal of Artificial
5. Shaw E., Johnson W. L., Ganeshan R.: Pedagogical agents on the Web. In: Proceedings of
the Third International Conference on Autonomous Agents (1999) 283-290
6. Johnson, W.L., Rickel, J., Stiles, R., Munro, A.: Integrating Pedagogical Agents into Virtual
Environments, Presence: Teleoperators and Virtual Environments 7(6) (1998) 523-546
7. Lester, J. C., Converse, S. A., Stone, B. A., Kahler, S. E., Barlow, S. T.: Animated
pedagogical agents and problem-solving effectiveness: A large-scale empirical evaluation.
In: Proceedings of the Eighth World Conference on Artificial Intelligence in Education. IOS
Press, Amsterdam (1997) 23-30.
8. Angros, R., Johnson, W.L., Rickel, J.: Agents that Learn to Instruct. In: AAAI 1997 Fall
Symposium Series: Intelligent Tutoring Systems Authoring Tools, Technical Report FS-97-
01, AAAI Press, Menlo Park (CA) (1997)
9. Mann W. C. and Thompson S. A.: Rhetorical structure theory: A theory of text organization.
Technical Report ISI/RS-87-190, Information Sciences Institute, University of Southern
California, Marina del Rey, CA. (1987)
10.Suthers, D. D.: Representational Support for Collaborative Inquiry. In: Proceedings of the
32nd Hawai'i International Conference on the System Sciences (HICSS-32), Maui, Hawai'i
(CD-ROM), Institute of Electrical and Electronics Engineers, Inc. (1999)
11.Schank, P, and Ranney, M.: Improved reasoning with Convince Me. In: Human Factors in
Computing Systems CHI '95 Conference Companion, Association for Computing
Machinery, New York, NY (1995) 276-277
12.Bell, P. Using argument representations to make thinking visible for individuals and groups.
In: Proc. Computer Supported Collaborative Learning ‘97, University of Toronto (1997)
13.Sandoval, W. A., and Reiser, B. J.: Evolving explanations in high school biology, Paper
presented at the Annual Meeting of the American Educational Research Assn., Chicago, IL
14.Rowley, K. and Crevoisier, M.: MAESTRO: Guiding students to skillful performance of the
writing process. In: Proceedings of the Educational Multimedia and Hypermedia conference,
Calgary, CA. (1997)
15.Marcu D.: The Rhetorical Parsing of Unrestricted Texts: A Surface-Based Approach.
Computational Linguistics, 26 (3) (2000) 395-448
16.Hovy, E.H.: Automated Discourse Generation Using Discourse Structure Relations.
Artificial Intelligence 63(1-2), Special Issue on Natural Language Processing (1993) 341-
17.Burstein, J. and Marcu D.: Towards Using Text Summarization for Essay-Based Feedback.
In: La 7e Conference Annuelle sur Le Traitement Automatique des Langues Naturelles
TALN'2000, Lausanne, Switzerland (2000)
18.André, E., Rist, T., Müller, J.: Integrating Reactive and Scripted Behaviors in a Life-Like
Presentation Agent. In: Proc. of the Second Int’l. Conf. on Autonomous Agents (Agents
'98), Association for Computing Machinery (1998) 261-268.
19.Penstein Rosé C. and Freedman R.: Building Dialog Systems for Tutorial Applications. In:
AAAI Fall Symposium 2000, AAAI Press Technical Report FS-00-01, Menlo Park (CA)
20.Marcu D., Amorrortu B., Romera M.: Experiments in Constructing a Corpus of Discourse
Trees. In: The ACL'99 Workshop on Standards and Tools for Discourse Tagging, Maryland
(1999) (RST tool: http://www.isi.edu/licensed-sw/RSTTool/index.html)
21.O'Donnell, M.: RSTTool 2.4 -- A Markup Tool for Rhetorical Structure Theory. In:
Proceedings of the Int’l Natural Language Generation Conference (INLG'2000), Mitzpe
Ramon, Israel (2000) 253-256
22.Johnson, W.L., Mayer, R., and Shaw, E.: K12 Agents: USC/ISI grant 12-1540-0002 (2001)