Terminal Time: An Ideologically-biased History Machine

Michael Mateas; Steffi Domike; Paul Vanouse

Carnegie Mellon University
Computer Science Department (Michael)
Art Department (Steffi)
Studio for Creative Inquiry (Paul)
5000 Forbes Avenue
Pittsburgh PA 15213
[email protected]; [email protected]; [email protected]
Abstract
Terminal Time is a machine that constructs ideologically-biased documentary histories in response to audience
feedback. The audience answers multiple-choice questions via an applause meter. The answers to these questions
influence which historical events are chosen from a knowledge base, how these events will be slanted to embody
the bias implied in the audience's answers, and how the events will be connected together to form a historical narrative. Once the narrative has been generated, video and sound clips are selected from a term-indexed multimedia
database. The resulting documentary, consisting of the newly generated narrative spoken by a speech synthesizer,
and the video and sound clips, is then presented to the audience.

1 Introduction
Terminal Time is a machine that constructs ideologically-biased documentary histories in response to audience feedback. It is a cinematic experience, designed
for projection on a large screen in a movie theater setting. At the beginning of the show, and at several
points during the show, the audience responds to multiple choice questions reminiscent of marketing polls.
The audience interaction in relationship to the viewing
experience is depicted in Figure 1. In the first question
period, an initial ideological theme (from the set of
gender, race, technology, class, religion) and a narrative arc (e.g. is this a progress or decline narrative) are
established.

ogy and religion). The third set of questions further

refines the theme(s) and introduces the possibility for a
reversal (e.g. a decline narrative becoming a progress
narrative). An example question (from the first question period) is shown in Figure 2.
Which of these phrases do you feel best represents
you:
A. Life was better in the time of my grandparents.
B. Life is good and keeps getting better every day.
Figure 2: Example question

The audience selects answers to these questions via an

applause meter the
first questions
second questions
answer generating the
third questions
posed to
posed to
posed to
most applause wins.
audience
audience
audience
The answers to these
questions allow the
2 min.
Discussion
system to create histori6 min.
6 min.
6 min.
Q2
Q1
Q3
cal narratives that atwith
Intro.
Section 2:
Section 3:
tempt to mirror and
Section 1:
Audience
roughly
roughly
Similar to
1000 to
often exaggerate the
1750-1950
1950-present
Masterpiece
1750 AD.
audiences biases and
Theater
desires. By exaggeratIntroduction
ing the ideological position implied in the audiences answers, Terminal Time
Figure 1: Audience interaction
produces an uncomfortable history that encourages the
audience to reflect on the influence of ideology on hisThe second set of questions refines the ideological
torical narratives.
theme chosen in the first set, and possibly introduces a
Terminal Time, a collaboration between a comsub-theme (e.g. combining race and class, or technolputer scientist specializing in AI-based art and enter-

tainment, an interactive media artist, and a documentary filmmaker, benefits from understandings of narrative drawn from AI, the arts, and documentary film.
Terminal Times architecture consists of the following major components: knowledge base, ideological goal tree (Carbonell 1979), template-based natural
language generator, a blackboard for event ordering
and story-arc maintenance (Englemore and Morgan,
1988), rhetorical devices, and a database of indexed
audio/visual elements primarily consisting of short
digital movies and sound files containing music. See
Figure 3 at the end of the paper for a diagram of the
architecture. The knowledge base contains representations of historical events. This is the raw material out
of which the ideologically-biased histories are constructed. Examples of historical events are the First
Crusades, the invention of Bakelite, and the rise of
enlightenment philosophy. Ideological-goal trees represent the current ideological-bias being pursued by the
computational narrator. The goal-trees consist of rhetorical goals ordered by subgoal and importance (to the
ideologue) relationships. These goals are used both to
select historical events to include in the story and to
spin the event in an ideologically-consistent manner.
The template-based generator generates the narrative
text once specific facts have been selected and connected to make a story. The blackboard serves as a
working memory for processes that impose a narrative
order on event spins created by the goal-tree. Constraints can also be passed back down to the goal tree
from this level. Rhetorical devices are connecting
pieces of text with accompanying constraints on story
structure. These devices are used to create narrative
connections between historical events. An example
rhetorical device is Yet progress doesnt always yield
satisfaction. Finally, the multimedia database contains
the audio/visual elements for the assembled documentary. First the system builds the narrative track using
the structures and process briefly described above.
Once a narrative track has been constructed, information retrieval techniques are used to match the best
indexed multimedia elements to the appropriate pieces
of text. Once the multimedia elements have been selected, the resulting documentary is displayed, layering
text-to-speech synthesis of the narrative track, and the
video and audio elements.
The Terminal Time project has been on-going
since the summer of 1997. During this time, several
prototypes have been built in order to experiment with
different representation schemes. The most recent prototype, finished in April of 1998, has been performed
in front of theater audiences at several venues. This
prototype has only one question period (at the beginning) and constructs a mini-history of the period
1900-1929. Since the construction of this last prototype, we have been working on the implementation and
knowledge encoding for the architecture outlined
above, performing historical research, and creating
media elements. This final version will take advantage

of the upcoming end of the millennium to tell the history of the last 1000 years. The project will be completed in June of 1999.
In the rest of this paper, we will describe the artistic aims of the project, discuss the relationship of this
project to computer creativity, and describe in more
detail the knowledge base and ideological goal trees.

2 Artistic goals
2.1 Documentary form
Ever since the first moving images were recorded,
filmmakers were aware of the power of this medium to
effect historical meaning. The historical documentary
became one of the first identifiable film genres. The
popular model of this form in America today, most
clearly exemplified by Ken Burns' "The Civil War,"
has the familiar structure of Western narrative: each
program has a distinct dramatic arc, a beginning, middle and an end. The rhetorical structure -- also familiar
and now almost universally expected -- invariably involves a crisis situation, a climax, and a clear resolution. Generally there is one prevailing narrative, one
interpretation of the historical facts presented. Most
usually, the narrative is delivered to the audience by an
unseen, yet obviously white, male narrator. So popular
is this model that networks and cable channels, including the public television networks, rarely show programs that diverge from it; thus the form has become
even more codified.
With Terminal Time we intend to imitate the
model of this cookie-cutter documentary with a machine that produces and reproduces it, until the model
itself is revealed for the tool of ideological replication
that it has become. Although dominant in popular media today, the cookie-cutter documentary is just one
form of historical documentary. Terminal Time derives
its impetus from the dominance of this archetype, as
well as from independent attempts to challenge the
authority implied in the historical documentary and to
posit alternative forms.

2.2 Utopian navigation

There is a great deal of industry hype surrounding interactive media and computing. Typically such experiences are promoted through a rhetoric of utopian navigation. According to such rhetoric, the computer provides unlimited access to information and experience, a
pure source of empowerment that imposes no interpretation on the data that is processed. Other familiar
tropes in this rhetoric include: Real-time, Immersion
and Virtuality -- promising the thrill of reality or hyperreality, without the effort, right from ones own PC.
Microsofts ads softly beguile us with the question
Where do you want to go today?
With Terminal Time, we play with these notions
by building a program that engages in active interpreta-

tion and construction of the interactive experience.

While the resulting constructed histories clearly respond to audience input, the system has a mind of its
own, pushing the story into extremes that the audience
did not intend. Thus value-free navigation gives way to
a value-laden interpretation. Terminal Time is a program that bites back.

2.3 Audience experience

Utilizing indirect questionnaires as a user interface, the
system essentially target markets each audience with an
appropriate history. Rather than asking audiences what
type of history they would like, or how they would like
to navigate through history, they are asked questions
about their own demographics and psychographics:
what type of home they live in, what cultural trends
they find most disturbing, how well they get along with
others, etc. The resulting history is like holding a funhouse mirror to the audience; it reflects an exaggerated
and distorted view of the audiences biases.
An applause meter is used to measure audience
response to the questions. The applause meter was chosen as the input device for two reasons: ease of setup in
different venues and the audience dynamic created by
public applause. The applause meter requires no special setup in a theater. All that is required is a good
quality directional microphone and a small mixing
board. Alternative input devices, such as buttons or
knobs placed at every seat, would be difficult to install.
Such devices would effectively prevent Terminal Time
from traveling to many venues. More importantly, applause metering enables interesting and entertaining
audience dynamics. With applause, the audience members can gauge how the audience as a whole is responding to questions. At performances of the prototype, the audience laughed when choices met with silence. Sometimes the applause would grow into a
groundswell of whistling and clapping as it became
clear that certain choices were nearly unanimous. In
other words, applause metering allows the audience to
watch and respond to its own behavior. In addition,
there is a nice incongruity when questions of "serious
import" (such as "What is the worst problem facing the
world today?") are answered by a popularity vote using
a social mechanism (applause) often associated with
entertainment and sporting events.
The audience should be kept just on the edge of
incredulity. As the history begins 1000 years ago, the
audience should experience a comfortable sense of
historical authority engendered by the familiar documentary form and the remoteness of the historical
events. As the history unfolds, the effect of the periodic
audience polls becomes more and more apparent. The
system will sometimes have to make wild causal leaps
in order to maintain the ideological story. In addition,
the system may foreground obscure historical episodes
while leaving out well known ones. These occasional
lapses, combined with the reflection of the audience

polls in the content, begin creating a tension with regard to the veridicality of the history (a sense of wait a
minute, this doesnt seem quite right...). Ideally, this
tension should reach a maximum as the piece moves
into modern history.
In order to fully appreciate the piece, an audience
should see it more than once. In a typical hour-long
performance, an audience will be able to see two performances. In the second viewing, even if the audience
answers the polls in exactly the same way, they will
experience a different history.1 Seeing two different
histories back-to-back should make fully apparent the
effect of ideological bias in historical construction.

3 Terminal Time and computer

creativity
Terminal Time is informed by a conception of AI as an
expressive medium (Mateas 1999; Sengers 1998). Expressive AI conceives of AI systems as cultural artifacts. The concern is not with building something that
is intelligent independent of any observer and cultural
context. Rather, the concern is with building an artifact
that seems intelligent, that participates in a specific
cultural context in a manner that is perceived as intelligent. Expressive AI views a system as a performance.
Within a performative space, the system expresses the
authors ideas. The system is both a messenger for and
a message from the author. Expressive AI thus changes
the focus from the system as a thing in itself (presumably demonstrating some essential feature of intelligence), to the system as a communication between
author and audience. At the technical level of building
the artifact, the technical practice becomes one of exploring which architectures and techniques best serve
as an inscription device within which the authors can
express their message.
As authors, we have specific artistic goals and
audience experiences we are pursing with Terminal
Time. The project would lose meaning if we could not
exert authorial control over the histories generated by
the system. Of course, maximum authorial control
would consist of writing a fixed set of canned histories;
audience interaction would select one of these canned
histories. But this extreme of control is inappropriate
for this project on several grounds. Conceptually, the
project depends on the machine really constructing
the histories. The critique of the computer as a passive
conduit of information requires that the computer actually take on an active role as a semi-cooperative genie,
obviously responding to the choices voted on by the
audience, but taking these choices to extremes. And on
practical grounds, the number of possible histories resulting from all possible answers to all the questions is
too large to build by hand. So, even if the conceptual
1

In the event that the polls are answered in the same way, the
differences will appear in the specific events chosen and the
text generated for these events, not in the ideological bias.

purity of the piece did not demand it, practical necessity would require that the computer play an active role
in story construction. As we reject the extreme of pure
hand-authoring, we also reject the extreme of strongly
emergent architectures, that is, architectures in which
as little high-level knowledge as possible is given to
the system, with all high-level behavior resulting from
large numbers of statistical combinations of low-level
elements. Such architectures by definition make authorship highly problematic. In a sense, they provide
no authorial hooks, no places within the architectural
in which an author can exert specific control. Much of
the architectural work that went into the iterative prototyping of Terminal Time was a search for an architecture providing authorial hooks on the right level of
abstraction: low-level enough to allow significant
combinatorial possibilities and the capability for surprise, yet high-level enough to allow the exertion of
authorial control over multiple levels of the story construction process.

4 Knowledge base
4.1 Upper Cyc ontology
The knowledge base consists of second order predicate
statements about historical events, definitions of ontological entities used in the historical event descriptions
(individuals and collections), and inference rules. Terminal Time's ontology is based on the Upper Cyc Ontology, the top 3000 most general terms in the Cyc
ontology (Lenat 1995). The Upper Cyc Ontology is
available free of charge from Cycorp2. It does not include any other components of Cyc (theorem prover,
natural language engine, database, etc.); it only provides definitions of the top 3000 most general terms.
However, the upper ontology provides a useful set of
distinctions in terms of which the more specific ontology needed by Terminal Time can be defined.

4.2 Example historical event

Figure 4 shows part of the representation of The First
Crusades. Those terms preceded by a "$" are defined in
the Upper Cyc Ontology. Those terms not preceded by
"$" are defined within the TT ontology in terms of the
Upper Cyc Ontology. The intent of the representational
syntax in Figure 4, translated into English, is:
The First Crusades, a historical event occurring in the 11th century, involved a war and a
transfer of possession of something. The First
Crusades can be divided into three sub-events:
the call for the First Crusades, the march to Jerusalem, and the taking of Jerusalem (in that
temporal order). The call for the First Crusades
is a communication act in which Emperor Al2

http://www.cyc.com/

; TheFirstCrusades
($isa TheFirstCrusades HistoricalEvent)
($isa TheFirstCrusades $WagingWar)
($isa TheFirstCrusades $TransferringPossession)
(circa TheFirstCrusades (CenturyFn 11))
($comment TheFirstCrusades "The First Crusades is
the first attempt of European Christians to
take back Jerusalem from the Muslims - 11th century")
($firstSubEvents TheFirstCrusades
CallForFirstCrusades)
($subEvents TheFirstCrusades
FirstCrusadesMarchToJersalem)
($lastSubEvents TheFirstCrusades
FirstCrusadesTakeJerusalem)
($isa CallForFirstCrusades
$Requesting-CommunicationAct)
($senderOfInfo CallForFirstCrusades
(CompositeInteligentAgentFn PopeUrbanI
EmperorAlexander))
($recipientOfInfo CallForFirstCrusades
(CompositeIACollectionFn EuropeanChristians))
($infoTransfered CallForFirstCrusades
(RequestFn
(takeByForce
(CompositeIACollectionFn
EuropeanChristians)
(CompositeIACollectionFn
MiddleEasternMuslims)
Jerusalem TheFirstCrusades)))
exander and Pope Urban I, acting in concert,
asked the community of European Christians
to take Jerusalem by force.
The syntax gains its representational intent both from
inference rules that allow new terms to be proved true
given the knowledge base, and from actions taken by
the rest of Terminal Time when terms are proved true.
Figure 4: Example knowledge base representation
Terminal Time's full representation of the First Crusades can be found in Figure 5 at the end of the paper.

4.3 Inference engine

The inference engine, implemented in Common Lisp,
is based on the interpreter implementing higher-order
hereditary Harrop logic described in Elliott and Pfenning (1991). Hereditary Harrop logic allows knowledge
base entries (the program, thinking in logic programming terms) to consist of Horn clauses, and queries
(goals) to consist of all the standard Prolog-like goals
(atomic goals, conjunctions, disjunctions, existentials),
plus embedded implications (assumptions). The interpreter also includes extra-logical support for operations
such as unifying logic variables against a function
evaluated by Lisp.

(solve '(and
($purposeInEvent ?Agent FirstCrusades
?Purpose)
($isa ?Purpose $WagingWar)
($subEvents FirstCrusades ?Sub)
($isa ?Sub $Requesting-CommunicationAct)
($senderOfInfo ?Sub ?Agent)
($infoTransfered ?Sub (RequestFn ?Request))
(match ?Purpose ?Request)
(not (and
($subEvents FirstCrusades ?Sub2)
($isa
?Sub2
$Requesting-CommunicationAct)
($startAfterEndingOf ?Sub ?Sub2)
($infoTransfered ?Sub2
(RequestFn ?Request2))
($isa ?Request2 $WagingWar)))
($hasBeliefSystems ?Agent ?Bsystem)
($isa ?Bsystem $Relgion)))
The inference engine is used to answer all queries
about historical events. For example, in the discussion
below of ideological goal trees, the historical event
tests that are mentioned are all made using the inference engine. For example, the query "Does the instigator of a war (e.g. The First Crusades) have a religious
belief?" could be represented as a query as depicted in
Figure 6. An English rendition of this query is:
Figure 6: An example query
Is it true that some agent in the First Crusades had
the purpose of waging war, that this same agent
requested some other agent to engage in this war,
that this request to wage a war occurred before
any other request by any other agent to engage in
a war, and that the agent who requested the war
(and whose purpose is to wage war) holds religious beliefs?
It may be the case that a query appears in several different places within Terminal Time (e.g. in several different rhetorical goals); it would be inconvenient to
have to repeat such complex queries in multiple places.
Much of this query can be pushed into inference rules.
For example, one could define a predicate (instigator
AGENT WAR) which means that AGENT is the instigator of the WAR. All of the query in Figure 6 down to
the $hasBeliefSystems formula could then become an
inference rule for proving that an agent is an instigator.
As additional ways of proving that someone is an instigator are needed, they can be added as additional inference rules. This collection of instigator rules becomes
part of the knowledge that Terminal Time has about the
script $WagingWar. The query in Figure 6 has been
partially unpacked (removing inference rule chaining)
in order to provide a clearer example of the kinds of
queries made of the knowledge base.

5 Ideological goal trees

Terminal Time organizes ideological bias with goal

trees, adapted from Politics (Carbonell 1979). In Politics, ideology is encoded as a set of goals held by the
ideologue. The goals are organized via subgoal links
(not corresponding exactly to either the conjunctive or
disjunctive notion of subgoal) and relative importance
links. The relative importance links place an importance partial order over the subgoals. For example, in
Politics, the US Conservative ideologues most important goal is Communist Containment. This goal has a
number of subgoals such as Have a Strong Military,
Aid Anti-Communist Countries, etc. Though Have a
Strong Military and Aid Anti-Communist Countries are
sibling subgoals, Have a Strong Military has a higher
relative importance. In addition to their own goal tree,
an ideologue also possesses beliefs about the goal trees
of others. In Carbonells system, the goal trees were
used to organize inferences made by a news story understanding system.
In Terminal Time, the goal tree has been modified
to represent the goals of an ideological story-teller.
Rather than having goals to modify the world, the
story-teller has rhetorical goals to show that something
is the case. For example, the Hard Core Anti-Religious
Rationalist might have the goals shown in Figure 7.
Figure 7: Example rhetorical goals
G2 and G3 are subgoals of G1. G4 has a lower relative
importance than G1.
The leaf goals in the goal tree are used to organize
two kinds of knowledge: a set of tests for recognizing
G1: Show that religious thought leads to evil
G2: Show that religious thought leads to war
G3: Show that religious thought leads to oppression
G4: Show that religious thought is the same around
the world
when a historical event is potential fodder for satisfying the rhetorical goal, and a set of plans for actually
constructing the description of the event to satisfy the
goal (the event spin).

5.1 Tests for event applicability

An ideologue needs a way of recognizing when a historical event could be used to satisfy a goal (make an
ideological point). For example, the Hard Core Rationalist must be able to recognize that the First Crusades
can be used to show that religious thought leads to war.
This involves recognizing that the First Crusades was a
war, and that the people who started it had a religious
belief. An example specification for recognizing that
an event can be used to show that religious thought
leads to war is shown in Figure 8.

Event contains a war (W1)

The aggressor (A1) of the war (W1) has a religious
belief (B1)
Figure 8: An example applicability test
The event tests are purposely simplistic. For example,
many wars may involve participants with religious
beliefs without the religious beliefs playing a significant causal role. But Terminal Time pursues its ideological goals monomaniacally - it takes whatever grain
of event knowledge it can find that is relevant to its
current ideological bias and treats this knowledge as
the whole truth.
In the process of satisfying the test, parts of the
event representation are bound to roles in the test.
These roles are used for focusing attention when generating the event spin. For example, a complex event
may involve several sub-events and multiple actors.
Yet if this event passes the test above, the actor who is
the aggressor of the war and that actors religious belief
will have been bound to the roles A1 and B1, singling
them out for special attention. Other knowledge known
about the event can then be ignored during construction
of the event spin.

5.2 Plans for event-level story generation

Once an event as been recognized as applicable to a
rhetorical goal of the ideologue, additional knowledge
is necessary to spin the event in such a way as to satisfy the rhetorical goal. This knowledge is represented
as rhetorical plans. Such plans coordinate the bindings
generated by the applicability test, available natural
language templates associated with the event, and rhetorical devices.
An example plan for Show that religious thought
leads to war is shown in Figure 9.
Describe the individual who called for the war,
mentioning their religious belief
Describe the religious goal of the war
Describe some event happening during the war
Describe the outcome

Figure 9: An example rhetorical plan

Associated with this plan is an ideological tone. Given
a choice a sentence templates representing the same
meaning, the ideological tone will be used to select the
sentence template most consistent with the ideologue.
For example, several sentence templates may describe
an individual calling for a war. The Hard Core Rationalist will prefer a template with a rationalist-tone, but,
if such a template cant be found, will accept a template with anti-religious tone, or, if one cant be found,
with generically negative tone. In the event that only a

neutral description is available, then the neutral description may be combined with boiler-plate text (rhetorical devices) to set the appropriate tone. For example, the Anti-Male Feminist may match on the First
Crusades as an example of men once again causing
pain and suffering (in this case, by starting a war). But
there may be no templates providing a gender tone. So
a rhetorical device, such as Once again, the male sex
revealed their fundamentally anti-life outlook would
be combined with a neutral description such as Pope
Urban I called for the First Crusades to set the appropriate ideological tone.
If sentence templates associated with the event can
not be found to satisfy a rhetorical plan, the system
backtracks, attempting other rhetorical plans if multiple
plans are available for a goal, or backtracking over the
bindings established by the event applicability tests.

5.3 Audience interaction influences goal

trees
The primary effect of audience interaction is to change
the goal tree. Audience interaction may add, delete, or
change goals in the goal tree. Two different ideological
positions can be mixed by combining goals from two
goal trees. For example, the audiences answers to the
first set of questions may select the Hard-core Rationalist goal tree. Answers to the second set of questions
may determine that racial equality (perhaps exaggerated as a homogenized Benneton commercial multiculturalism) is a sub-theme. The goal tree will be modified to include Corporate Multiculturalist goals in addition to Hard-core Rationalist goals, thus producing a
hybridized ideological narrative. Some responses to
questions (particularly questions in the third and last
set) will modify the tree more subtly, adding and removing individual goals in the tree.

6 Related work
Hovy's work investigating pragmatic constraints on
natural language generation (1987) has some similarities to Terminal Time. Hovy's system, Pauline, generates event descriptions that satisfy rhetorical goals. In
Pauline, rhetorical goals include goals of opinion (e.g.
show that our side has good goals or takes good actions) and goals of style (level of formality, level of
simplicity). This notion of rhetorical goal differs from
that used in Terminal Time. In Terminal Time, rhetorical goals are goals to argue for specific ideological
positions by providing historical examples. In Pauline,
rhetorical goals are goals to provide a spin of a single
event consistent with a specific style. The user tells
Pauline which event to describe, Pauline's orientation
towards the event, the hearer's orientation towards the
event, and which stylistic constraints to apply. Given
these inputs, Pauline produces a textual description.
Though Pauline only knows about three events, it can
produce 100 different texts for an event.

Pauline is concerned with rich natural language

generation parameterized by pragmatic constraints.
Terminal Time is concerned with selecting and ordering historical events so as to support an ideological
position. Pauline spends much more effort crafting
individual sentences; Terminal Time uses templatebased generation. But Terminal Time spends much
more effort selecting events consistent with an ideological goal tree and ordering these events so as to
form a story; Pauline is told which single event to describe and with which specific slant. The area where
the two systems most overlap is in topic selection and
ordering within an individual event, that is, in the processing that is handled by rhetorical plans in Terminal
Time.
Finally, a rather large difference between the two
systems is in the intentionality of the design. Pauline is
an AI research system intended to further the understanding of the knowledge structures and processes
required for language generation under pragmatic constraints. Terminal Time is a performative artwork intended to function as a provocative culture artifact, a
peculiar machine raising questions about the construction of history, the nature of ideological reasoning, and
the nature of user agency afforded by computer technology.

7 Conclusion
Terminal Time constructs ideologically biased documentary histories in front of theater audiences, utilizing
marketing-style polls to allow an audience to vote for
the history they want. But, like technology itself, Terminal Time is a fickle genie, using the audiences biases
and desires to display histories that become uncomfortably extreme. The conception of AI employed in
this project is expressive AI: AI systems viewed as a
communication between author and audience. Thus the
AI architecture was designed to afford combinatorial
possibilities while supporting authorial control. The
ideological goal tree is one representational mechanism
used to organize ideological bias in historical construction.

References
J. Carbonell. Subjective understanding: Computer
models of belief systems. Ph.D. Thesis, Computer
Science Department, Yale University, Research
Report #150, 1979.
C. Elliott and F. Pfenning. A semi-functional implementation of a higher-order logic programming
language. In Peter Lee, editor, Topics in Advanced
Language Implementation, pages 289-325. MIT
Press, 1991.
R. Englemore and T. Morgan, eds. Blackboard Systems. Addison-Wesley, 1988.

E. Hovy. Generating Natural Language Under Pragmatic Constraints. Ph.D. Thesis, Computer Science Department, Yale University, Research Report #521, 1987.
D. Lenat. Cyc: A Large-Scale Investment in Knowledge Infrastructure. Communications of the ACM,
38, no. 11, November 1995.
M. Mateas. Not your Grandmothers Game: AI-Based
Art and Entertainment. Working notes of the AI
and Computer Games Symposium, AAAI Spring
Symposium Series. Menlo Park: Calif.: AAAI
Press, 1999.
P. Sengers. Anti-Boxology: Agent Design in Cultural
Context. Ph.D. Thesis, School of Computer Science, Carnegie Mellon University, 1998.

Figure 3: Terminal Time architecture

;; The terms describing the First Crusades

($isa TheFirstCrusades HistoricalEvent)
($isa TheFirstCrusades $WagingWar)
($isa TheFirstCrusades $TransferringPossession)
(circa TheFirstCrusades (CenturyFn 11))
($comment TheFirstCrusades "The First Crusades was the first attempt of Eurpopean Christians to
take back Jerusalem from the Muslims in the 11th century")
($firstSubEvents TheFirstCrusades CallForFirstCrusades)
($subEvents TheFirstCrusades FirstCrusadesMarchToJersalem)
($lastSubEvents TheFirstCrusades FirstCrusadesTakeJerusalem)
($purposeInEvent
(CompositeIntelligentAgentFn PopeUrbanI EmperorAlexander)
TheFirstCrusades
(takeByForce
(CompositeIACollectionFn EuropeanChristians)
(CompositeIACollectionFn MiddleEasternMuslims)
Jerusalem
TheFirstCrusades))
($isa CallForFirstCrusades $Requesting-CommunicationAct)
($senderOfInfo CallForFirstCrusades
(CompositeInteligentAgentFn PopeUrbanI EmperorAlexander))
($recipientOfInfo CallForFirstCrusades
(CompositeIACollectionFn EuropeanChristians))
($infoTransfered CallForFirstCrusades
(RequestFn
(takeByForce
(CompositeIACollectionFn EuropeanChristians)
(CompositeIACollectionFn MiddleEasternMuslims)
Jerusalem
TheFirstCrusades)))

($toLocation FirstCrusadesMarchToJerusalem Jerusalem)

($mainTransportees FirstCrusadesMarchToJerusalem
FirstCrusadesArmy)
($subEvents FirstCrusadesMarchToJerusalem
FirstCrusadesMurderJews)
($isa FirstCrusadesMurderJews $Killing-Biological)
($victim FirstCrusadesMurderJews
(CompositeIACollectionFn EuropeanJews))
($performanceLevel
FirstCrusadesArmy
FirstCrusadesMarchToJerusalem
$performedBy
DegreeOfOrganization
($LowFn DegreeOfOrganization))
($isa FirstCrusadesTakeJerusalem $TransferringPossession)
($isa FirstCrusadesTakeJerusalem $WagingWar)
(takeByForce FirstCrusadesArmy
(CompositeIACollectionFn MiddleEasternMuslims)
Jerusalem
FirstCrusadesTakeJerusalem))
($successfulForAgents FirstCrusades
(CompositeIntelligentAgentFn PopeUrbanI EmperorAlexander))
($successfulForAgents FirstCrusades
(CompositeIACollectionFn EuropeanChristians))
($failureForAgents FirstCrusades
(CompositeIACollectionFn MiddleEasterMuslims))
;; Causal structure for FirstCrusades
($causes CallForFirstCrusades FirstCrusadesMarchToJerusalem)
($causes FirstCrusadesMarchToJerusalem
FirstCrusadesTakeJerusalem)

($assistingAgent CallForFirstCrusades PeterTheHermit)

($isa FirstCrusadesArmy $Army)
($genls (MemberCollectionFn FirstCrusadesArmy)
EuropeanChristians)
($circa FirstCrusadesArmy (CenturyFn 11))
($comment FirstCrusadesArmy "Represents the four Christian armies in the First Crusades (represent as a single entity.")
($isa FirstCrusadesMarchToJerusalem $TransportingPeople)
($fromLocation FirstCrusadesMarchToJerusalem
$ContinentOfEurope)

;; Soon Jerusalem was lost to the Christians

($isa JerusalemRetakenByMuslims1 $TransferringPossession)
($isa JerusalemRetakenByMuslims1 $WagingWar)
(circa JerusalemRetakenByMuslims1 (CenturyFn 12))
(takeByForce
(CompositeIACollectionFn MiddleEasternMuslims)
(CompositeIACollectionFn EurpeanChristians)
Jerusalem
JerusalemRetakenByMuslims1)
(reverses JerusalemRetakenByMuslims1 TheFirstCrusades)

Figure 5: Full representation of the First Crusades