U.S. patent application number 11/930338 was filed with the patent office on 2008-09-25 for method and system for automatic computation creativity and specifically for story generation.
This patent application is currently assigned to International Business Machines Corporation. Invention is credited to Selmer Conrad Bringsjord, David Angelo Ferrucci.
Application Number | 20080235576 11/930338 |
Document ID | / |
Family ID | 39059558 |
Filed Date | 2008-09-25 |
United States Patent
Application |
20080235576 |
Kind Code |
A1 |
Bringsjord; Selmer Conrad ;
et al. |
September 25, 2008 |
METHOD AND SYSTEM FOR AUTOMATIC COMPUTATION CREATIVITY AND
SPECIFICALLY FOR STORY GENERATION
Abstract
A computer-implemented method (and system) of automatically
generating a story, includes selecting a theme of the story,
examining elements of the theme and instantiating the theme, and
using the theme to select and control other aspects of the
story.
Inventors: |
Bringsjord; Selmer Conrad;
(Brunswick Hills, NY) ; Ferrucci; David Angelo;
(Yorktown Heights, NY) |
Correspondence
Address: |
MCGINN INTELLECTUAL PROPERTY LAW GROUP, PLLC
8321 OLD COURTHOUSE ROAD, SUITE 200
VIENNA
VA
22182-3817
US
|
Assignee: |
International Business Machines
Corporation
Armonk
NY
|
Family ID: |
39059558 |
Appl. No.: |
11/930338 |
Filed: |
October 31, 2007 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
09471689 |
Dec 23, 1999 |
7333967 |
|
|
11930338 |
|
|
|
|
Current U.S.
Class: |
715/256 |
Current CPC
Class: |
G06N 5/00 20130101 |
Class at
Publication: |
715/256 |
International
Class: |
G06F 17/21 20060101
G06F017/21 |
Claims
1. A computer-implemented method of automatically generating a
story, comprising: selecting a theme of said story; examining
elements of said theme and instantiating said theme; using said
theme to select and control other aspects of the story generation,
including a plot of said story which employs knowledge-generated
characteristics, relationships, and events.
2. The computer-implemented method according to claim 1, further
comprising inputting said elements of said theme into a stage, said
stage being a collection of elements of said story and their
interrelationships.
3. The computer-implemented method according to claim 1, further
comprising inputting said elements into a simulation engine to play
out a series of events over time, thereby to generate a plot, said
plot including characters, their characteristics, and their
respective interactions, and a history of events and their temporal
relationships, said simulation engine including a predetermined
randomness such that random elements from said databases are
selected.
4. The computer-implemented method according to claim 1, further
comprising inputting said elements into a simulation engine to play
out a series of events over time.
5. The computer-implemented method according to claim 1, further
comprising generating a plot.
6. The computer-implemented method according to claim 5, wherein
said plot comprises characters, their characteristics, and their
respective interactions, and a history of events and their temporal
relationships.
7. The computer-implemented method according to claim 5, wherein
said simulation engine comprises a predetermined randomness such
that random elements from said databases are selected.
8. The computer-implemented method according to claim 5, wherein
said plot comprises at least one of characters, their
characteristics, and their respective interactions, and a history
of events and their temporal relationships.
9. The method according to claim 1 further comprising: developing a
scenario, including a plurality of activities, based on said plot
generated, such that details of said activities are developed and
bridged.
10. The method according to claim 9, wherein said scenarios are in
a formal, logic based language independent of a spoken language,
said method further comprising: inputting said scenarios into a
natural language generator such that said scenarios are converted
from said logic-based language to a natural language, said natural
language being a spoken language understandable by a human reader,
said conversion from said logic-based language to said natural
language influencing at least one of story grammars, literary
constraints, words, phrases, and sentence structure used in said
scenario.
11. A signal-bearing medium tangibly embodying a program of
machine-readable instructions executable by a digital processing
apparatus to perform a method of story generation, said method
comprising: selecting a theme of said story; examining elements of
said theme and instantiating said theme; using said theme to select
and control other aspects of the story generation, including a plot
of said story which employs knowledge-generated characteristics,
relationships, and events.
12. A computer-implemented method of automatically generating a
story, comprising: inputting elements of a story into a simulation
engine to play out a series of events over time, thereby to
generate a plot, said plot including at least one of characters,
their characteristics, and their respective interactions, and a
history of events and their temporal relationships.
Description
[0001] The present application is a Continuation Application of
U.S. patent application Ser. No. 09/471,689, which was filed on
Dec. 23, 1999.
BACKGROUND OF THE INVENTION
[0002] 1. Field of the Invention
[0003] The present invention relates to a computer system, and more
particularly to a computer system and method for generating a
plurality of artifacts in a specified language which people (e.g.,
humans) are likely to find interesting.
[0004] 2. Description of the Related Art
[0005] The notion of general creativity, like intelligence, is
ultimately a humanistic and subjective concept. Moreover, the
notion of computational creativity is not bound to the same
procedural attributes one might associate with human
creativity.
[0006] Indeed, the procedural attributes of human creativity may
remain largely a mystery, while the procedural attributes of
computational creativity may be well-defined. This comparison is
akin to the comparison between the human intelligence at work
playing a game of chess and the computational mechanism at work in
a computer that rivals the human player. Both mechanisms manifest
an effective performance in, what humans consider, an
intellectually demanding task. However, the procedural attributes
of the human remain largely ill-defined, while the procedural
mechanisms used by the computer may be rigorously explicated in a
formal computational language and ultimately vary significantly
from the human mechanism.
[0007] Therefore, the association of creativity with an agent is
focused on manifest behavior, not on implementation. One ultimately
judges an agent as "creative" if it can produce an interesting
artifact in some language of expression (e.g., natural language,
music, sculpture, etc.), starting from a point sufficiently distant
from the end result.
[0008] Two key concepts in realizing a creative agent are the
notions of interestingness and creative distance. "Interestingness"
asks the question of: "will one find an agent's creation
interesting or will one find it an incomprehensible blob with no
communicative value?"
[0009] An agent may generate voluminous works that are never
assigned any value by the agent's audience. The agent, rather than
being judged creative, is considered an arbitrary generator of
random artifacts. There are several ways to consider
interestingness.
[0010] First, the agent may understand the cognitive cultural
context of its audience well enough to invent the interesting. That
is, the agent may extend and redefine the audience's cognitive
cultural awareness with a novel creation and demonstrate its value.
One typically associates genius with such capability. Computational
creativity does not address genius.
[0011] A second way to create interesting artifacts is to begin
with a known seed of interestingness (e.g., what is referred to as
a "theme"). For example, it is known that particular stories about
the fruits or snarls of romantic love have human interest.
Variations based on skillful use of language, plot twists. rich
characterizations etc. that appeal to audiences are considered
creative and interesting, albeit perhaps short of genius.
[0012] While the output of a generative agent may be an interesting
well-crafted story, a harmonious piece of music or a delicious
recipe, the creativity of the agent is ultimately predicated on the
input's creative distance from the output.
[0013] For example, if an agent for story generation requires a
complete story as an input and outputs different stories differing
from the input only by the names of the characters, then the agent,
while legitimately generating a unique artifact, would not be
considered creative. Indeed, the "distance" between the input and
the output would be considered negligible and insufficient. Nor
would the artifact produced by the agent effect significant human
interest in light of the input.
[0014] However, if a literary theme such as the "evil of betrayal"
or the "destructive force of ambition" were the input to the agent,
and the agent output complete stories about the indicated theme
varying according to characters, plot, story-structure and
language, then the agent would be considered a "creative literary
agent". The artifact, while anchored to the theme, would exhibit
sufficient expansion (e.g., creative distance) relative to the
input.
[0015] Any creative agent must begin with a seed of interestingness
and maintain that theme in the generation of skillful variations
that are sufficiently distinct from the input. Hitherto the
invention, there has been no system which performed the above
operations and in which interestingness and creative distance were
even considered.
[0016] Thus, the conventional story generation system has been
deficient in a number of areas as briefly discussed above.
[0017] Regarding knowledge sources, the conventional story
generation system and method have used such sources only
individually, in an unintegrated fashion. There has been no
integrated, composite approach to story generation.
[0018] That is, the conventional methods have demonstrated analogs
of lexical knowledge, in the form of natural language lexicons and
grammars. Further, compositional knowledge, in the form of story
grammars have been demonstrated. Additionally, domain knowledge,
represented in many different ways from the logically formal (e.g.,
such as ("(X) isa(X, man).fwdarw.isa(X, mortal)) isa(socrates,
man).fwdarw.isa(socrates,mortal) to the ad hoc (e.g., such as
"socrates was mortal becomes he was a man.").
[0019] However, these knowledge sources have neither been
integrated and related to cooperative roles of an overall system
architecture for story generation, nor of a more general
architecture for computational creativity.
[0020] Regarding processes, the conventional story generation
system and method has demonstrated process analogs for stage
evolution, typically using planning and simulation techniques to
generate story plots. Additionally, process analogs for structural
expansion, typically using generative story grammars have been
developed. Lastly, natural language generation has been
developed.
[0021] Regarding system architectures, two basic architectures have
emerged in story generation. The first architecture is based
principally on plot development, whereas the second architecture is
based on structural expansion. Both include some form of natural
language generation.
[0022] An example of plot development through planning and
simulation would be to provide a planning engine in which would
implicitly ensure that the plot involved a character trying to
achieve some goal. However, a problem is that sometimes a
character's striving for a goal is tedious.
[0023] Thus, this approach to story generation is dominated by the
process of plot expansion to produce variability in stories.
However, the results have lacked a thematic anchor, and therefore
struggled with the notion of "interestingness". They also had no
explicit knowledge component or mechanism for achieving
impressionism (or in the special case of story generation,
rhetoric). Story structure is part of an explicit architecture, and
thus cannot be easily expressed or varied.
[0024] Further drawbacks include no representation or process for
producing interestingness, no representation or process for
structure expansion, and no representation or process of
identifying and exploiting impressionistic knowledge.
Impressionistic knowledge captures how an artifact might impact or
impress human sensibilities through style and form, rather than
explicit content. Knowledge about how the sounds of certain words
affect the reader is considered "impressionistic knowledge".
Another example is knowledge about how different words, even though
they may refer to the same thing, would produce different emotional
states in the reader. Describing weapons of destruction as
"ordnance" produces a different impression in the reader than
describing them as "harbingers of death". This class of knowledge
may be acquired through the practice and study of literature and
rhetoric and/or may be acquired and inferred statistically.
[0025] Another conventional implementation approach is structural
expansion through story grammars in which through iterative
structural expansion of a story grammar provides and builds
increasingly detailed outlines. However, in this approach it
becomes extremely difficult to represent the complex plot and
literary variations in the declarative forms required by story
grammars.
[0026] Thus, this approach to story generation focused on
generative grammars and ignored the potential to achieve creativity
and variability through plot expansion.
[0027] Thus, hitherto the present invention, there has been no
system and method adequate for story generation in which a creative
agent could begin with a seed of "interestingness" and maintain
that theme in the generation of skillful variations that are
sufficiently distinct from the input.
SUMMARY OF THE INVENTION
[0028] In view of the foregoing and other problems, disadvantages,
and drawbacks of the conventional methods and structures, an object
of the present invention is to provide a method and system for
story generation.
[0029] A further object of the invention is to provide a system and
method for story generation in which a plurality of artifacts are
generated in a specified language which humans are likely to find
interesting.
[0030] Yet another object is to provide a system and method for
story generation in which the theme of the story is selected
initially and the theme is anchored for constraining subsequent
choices made in generating the story.
[0031] Another object of the invention is to provide a system and
method for story generation in which a creative agent begins with a
seed of interestingness and maintain that theme in the generation
of skillful variations that are sufficiently distinct from the
input.
[0032] In a first aspect of the present invention, a
computer-implemented method (and system) of automatically
generating a story, includes selecting a theme of the story,
examining elements of the theme and instantiating the theme, and
using the theme to select and control other aspects of the
story.
[0033] In a second aspect of the invention, a programmable storage
medium is provided for storing a program implementing the steps of
the inventive method.
[0034] With the unique and unobvious features of the present
invention, story generation can take place automatically in which a
plurality of artifacts are generated in a specified language which
humans are likely to find interesting. Further, the story
generation is provided by a creative agent which begins with a seed
of interestingness and maintains that theme (e.g., anchors the
theme) in the generation of skillful variations that are
sufficiently distinct from the input.
[0035] Additionally, the present invention integrates a plurality
of knowledge sources (e.g., compositional, domain, literary, etc.)
relates them to cooperative roles of an overall system architecture
for story generation.
[0036] Thus, the invention is a composite, theme-based story
generation system and method. To provide the requisite
interestingness, the invention provides a mathematization (e.g.,
formal representation) of the theme independent of plot, language,
and story structure. The invention focuses on the theme (e.g.,
betrayal, the power of ambition to corrupt the soul, romantic love,
etc.) as a critical aspect for providing "interestingness", and
anchors the theme regardless of plot, characters, language, story
structure, etc.
[0037] By doing so, plot, setting, characters, story structure,
language, etc. can be varied while keeping intact/constant the
essential element (e.g., theme) which keeps the story interesting
and while making the theme permeate the entire story. As such, the
theme influences all of the other aspects and processes of the
story generation such as setting, characters and their properties,
language, story structure, selection of the words, sentences, and
paragraphs used in the generation of the language, plot, etc. and
these other aspects can be varied while maintaining the theme which
keeps the story interesting.
[0038] Additionally, besides orchestrating a composite architecture
and making the inventive process theme-based, thereby creating wide
variability in the other aspects of the story generation around the
same theme, the invention also allows capturing/representing
literary devices and techniques (e.g., using certain phrases,
words, etc.) in order to evoke certain emotions from the
reader.
[0039] Thus, for example, when generating a natural language, the
system and method of the invention will use (and indeed will be
constrained to use) certain words based on the theme selected and
appropriately classified in a database which lead the reader to
believe that the character (antagonist) is, for example,
"psychologically conscious" or have some other property. Hence,
only certain words might be available for usage based on certain
themes. That is, the selection of the theme will constrain the
words, phrases, etc. available for use in generating the story.
Thus, the invention will be constrained to use certain words or
phrases which would lead the reader to think the charter is
psychologically conscious. Such words are preferably
pre-classified, for example, in a database for use in generating
the story grammar to evoke a predetermined literary response from
the reader.
[0040] Unlike the conventional systems and methods which anchor the
story to characters or plot, etc., the invention uses the theme and
is captured independently to maintain a creative distance between
the input and the output and to provide interestingness to the
story.
BRIEF DESCRIPTION OF THE DRAWINGS
[0041] The foregoing and other purposes, aspects and advantages
will be better understood from the following detailed description
of a preferred embodiment of the invention with reference to the
drawings, in which:
[0042] FIG. 1 illustrates a block diagram of a composite
architecture 100 for story generation according to the present
invention;
[0043] FIG. 2A illustrates the process level of the composite
architecture 100 and the knowledge level(s) used by the composite
architecture 100 according to the present invention;
[0044] FIG. 2B is another view showing the generalized form of the
invention's different process steps and knowledge bases used by
each step;
[0045] FIG. 2C shows the inventive process (and the accessing of
respective knowledge bases) specialized for story generation;
[0046] FIG. 3A illustrates a flowchart of the method for story
generation according to the present invention;
[0047] FIG. 3B illustrates a sample frame hierarchy, including some
instances, for an exemplary story;
[0048] FIG. 3C illustrates the inventive system capturing the
knowledge used by a story structure expansion step 308 and language
generation step 307 (of the process shown in FIG. 3A) in a
three-level grammar hierarchy;
[0049] FIG. 4 illustrates an exemplary information
handling/processor system 400 for implementing the system and
method of the present invention; and
[0050] FIG. 5 illustrates a programmable storage medium 500 for
storing a program implementing the method of story generation
according to the present invention.
DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS OF THE INVENTION
[0051] Referring now to the drawings, and more particularly to
FIGS. 1-5, there are shown preferred embodiments of the method and
structures according to the present invention. It is noted that,
for consistency and clarity, the same reference numerals are used
in different figures to represent identical steps and/or
elements.
[0052] Referring to FIG. 1, a first embodiment of an upper level
system block diagram according to the present invention will be
described below.
[0053] FIG. 1 shows a composite architecture 100 which explicitly
reflects a story generation process that starts with a symbolic
encoding of a theme to anchor interestingness. A theme may be
selected by the system or by a user as a first process of the
invention.
[0054] The instantiation of the theme performed by a thematic
instantiation module 101 from a knowledge-base of characters and
events is used to input to a stage module 102, guide plot
generation in a plot generation module 103 to input to a scenario
module 104, perform story structure expansion via a story structure
expansion module 105 to input to a story outline in story outline
module 106, and natural language (e.g., English prose) generation
via language generation module 107.
[0055] The result is a quicker path to scenarios and stories 108
that conform (necessarily) to the explicitly represented theme,
which is likely to be interesting from the beginning. The idea of a
composite architecture that "covers all the bases" with
differentiated and independently varying components seems to have
helped breathe new life into the vision of a storytelling AI.
[0056] FIG. 1 illustrates the orchestration of the inventive
architecture's different components, including thematic
instantiation, story expansion, and plot and language generation,
all working together with different classes of knowledge.
[0057] FIG. 2A illustrates that the architecture of the inventive
system can be decomposed into two levels including a knowledge
level 100A and a process level 100B. Such levels interact with and
are controlled by a controller 100C, as shown in FIG. 2A.
Controller 100C controls the knowledge level, to perform the
process level 100B. The controller 100C is programmed to perform
the processes of the process level while interfacing with the
knowledge level.
[0058] The knowledge level 100A comprises the different types of
knowledge required to generate a written story about a particular
theme. The process level 100A contains the processes (e.g.,
described briefly above with regard to FIG. 1) that together use
the knowledge level 100 to generate a story. The process level
represents the system's computational approach to story
generation.
[0059] Prior to further describing the knowledge and process
levels, it is noted that an exemplary implementation of the system
and method of the invention has been developed using FLEX, a
multi-paradigm artificial intelligence (AI) programming system
based in the programming language Prolog. It is commercially
available from Logic Programming Associates, Ltd., incorporated in
the United Kingdom. Obviously, other systems could be used
including other programming languages.
[0060] Turning now in detail to the knowledge level 100A, the
knowledge level 100A contains representations of different types of
knowledge required to generate the written story:
[0061] For example, the "knowledges", typically contained in
databases or the like, include domain knowledge 100A1 (e.g.,
people, places, things, events, goals, behaviors, etc.), linguistic
knowledge (e.g., sentences, phrases, words, parts of speech etc.)
100A2 ("paragraph and sentence grammars" 100A5 and "lexicon"
100A6), literary knowledge (e.g. thematic descriptions, literary
structures, story grammars, etc.) included in "story grammars"
100A2, "thematic knowledge" 100A4 and "literary constraints"
100A3.
[0062] Generally, a story is a natural language description of
objects, their attributes, relationships, behaviors and
interactions. It may or may not be centered around any particular
theme or have any particular message. A story may or may not be
interesting (and, uninterestingness quickly follows from an absence
of themes and structure). However, all stories include a
description of some set of objects and their interactions.
[0063] Domain knowledge (e.g., 100A1) encodes a formal
representation of objects, attributes, relationships, goals,
behaviors, and events (e.g., a formal description of a domain).
Domain knowledge is not the story itself, but is a description of a
collection of concepts about which some story may be written.
[0064] Domain concepts may include the fundamental elements that
might compose a story's settings (e.g., places, things, and their
attributes and interrelationships), its characters (e.g., people,
their physical and psychological makeup, their goals and behaviors)
and the events that occur (e.g., steal, kill, murder, give, take,
run, sign, buy, borrow, loan, kiss, etc.). Such domain concepts are
formalized and stored into the domain knowledge storage 100A1.
[0065] Hence, a story is a description of a set of objects and
events. The description itself is of course encoded in a natural
language. A story therefore takes the form of a sequence of words
of some language strung together to conform to the language's
grammar. While domain knowledge may be considered a description of
domain concepts, it does not qualify as a story because, among
other reasons, the description is not encoded in a natural
language.
[0066] Linguistic knowledge is knowledge required to produce a
natural language description of domain concepts from a formal,
logical representation of domain knowledge. Linguistic knowledge
formally describes the linguistic structure of paragraphs,
sentences, phrases, and words. It categorizes words of a lexicon as
verbs, nouns, adjectives, adverbs, etc.
[0067] Literary Knowledge is also used by the invention. With but a
little domain knowledge and some linguistic knowledge, a story
generation system can produce a story. However, the story likely
will not be particularly interesting. A weak story may look more
like a laundry list (e.g., of descriptions of places, things,
people, and events) than engaging narrative. Such a story does not
have a theme or a message, nor will it be told in a way that holds
readers or steers them toward some intended meaning.
[0068] Literary knowledge is independent of a story's domain
content or grammatical integrity. It is knowledge of the high art
of storytelling. Literary knowledge enables the compelling
communication of interesting interpretations of domain knowledge
through the medium of natural language.
[0069] Generally, the "Story Grammars" 100A2 and "Literary
Constraints" 100A3 of FIG. 2A make-up Compositional Impressionistic
Knowledge (CIK) (discussed below with regard to FIG. 2B) and
"Paragraph and Sentence Grammars" 100A5, "Lexicon" 100A6, and
"Literary Constraints" 100A3 make-up Lexical Impressionistic
Knowledge (LIK) (discussed below with regard to FIG. 2B). More
generally, all together they represent the components of
Impressionistic Knowledge.
[0070] Specifically, the inventive system encodes literary
knowledge to generate stories that can achieve key literary
objectives, including 1) triggering readerly imaging, 2) project
psychological consciousness, 3) engage readers in classic themes,
and 4) instantiate classic story structures.
[0071] 1) and 2) are achieved through the characterization and
selection of words, phrases, sentence, and paragraph structures in
terms of their ability to trigger images and
psychological-consciousness in the reader. 3) is achieved through
the formalization of thematic knowledge. This formalization is
encoded as part of the literary knowledge-base in a structure
referred to as a thematic relation. This relation is an
implementation structure used to gather essential ingredients for a
story about betrayal from a domain knowledge-base.
[0072] 4) refers to the familiar flow of a story such as how it is
structured at a high-level in terms of setting, character
introduction, etc. Story grammars are what we use to represent the
high-level "look-and-feel" of story structure. These grammars are
part of the literary knowledge-base (KB).
[0073] One can view the domain knowledge-base as a pool of story
elements (characters, places, things, events, etc.) that can be
configured to produce any number of stories about any number of
themes.
[0074] The literary knowledge-base's thematic relation defines a
particular theme independently of any particular set of domain
elements. It is used to identify a set of elements from the domain
and their interrelated roles required to tell a story about that
theme. The literary knowledge base 100A3's story grammars are used
to orchestrate the high-level structure of the story that
determines which paragraphs to write first and what sentence forms
they contain. Literary knowledge is also used to select the key
words and phrases that best communicate the theme by triggering
readerly imaging and projecting P-consciousness. Finally,
linguistic knowledge is used to configure grammatically correct
English sentences.
[0075] Thus, as shown in FIG. 2A, the present invention utilizes
many knowledge sources. For example, the invention uses Lexical
Knowledge (LK). That is, the Language of Expression, L, is the
medium in which the artifact is manifest.
[0076] For story generation, the medium is typically some natural
language like English or German. Obviously, any language
understandable by a human is possible. For music generation, it
would be musical notes for example.
[0077] A grammar for L (e.g., G(L)), describes syntactically valid
ways of combining elements of L. (This is stored in "paragraph and
sentence grammars" 100A5 of FIG. 2A.)
[0078] The invention also utilizes Compositional Knowledge (CK).
That is, the language of Artifact Composition, C, is a language
naming structural components used to assemble an artifact. For
example, in story generation, the language of composition would
include names of things such as introduction, climax, conflict,
setting, protagonist, antagonist, resolution, etc. In cooking, the
language of composition might include things such as side dish,
sauce, glaze, dessert, etc.
[0079] C is L-neutral (e.g., language expression neutral). For
example, the same language of composition for stories would apply
regardless of whether L was German, English, Chinese or some other
natural language (e.g., spoken language understandable by human
beings).
[0080] A grammar for C (e.g., G(C)), describes valid ways of
structurally combining elements of C. For story generation, G(C) is
typically called the "story grammar" 100A2. The story grammar is a
type of outline of the story in which the sequence of activities
are presented.
[0081] As mentioned above, the present invention also utilizes
Domain Knowledge (DK) 100A1, which is a representation of objects,
agents, attributes, activities, events, goals, behaviors, etc. that
exist in some world (e.g., real or artificial) that the elements of
L are intended to represent. In story generation, for example,
domain knowledge would include a representation of characters and
their behaviors, actions, objects etc. In music generation, domain
knowledge may include a representation of instruments and their
characteristics and different ways to play them (e.g., potential
behaviors).
[0082] Referring to FIG. 2B, a generalized form of the invention is
shown (e.g., for producing an artifact 208 as opposed to
necessarily a story 108 showing in FIG. 2A). Generally, the "Story
Grammars" 100A2 and "Literary Constraints" 100A3 of FIG. 2A make-up
Compositional Impressionistic Knowledge (CIK) (discussed with
regard to FIG. 2B) and "Paragraph and Sentence Grammars" 100A5,
"Lexicon" 100A6, and "Literary Constraints" 100A3 make-up Lexical
Impressionistic Knowledge (LIK) (discussed with regard to FIG. 2B).
More generally, all together they represent the components of
Impressionistic Knowledge. The invention also utilizes
Impressionistic Knowledge (IK). This knowledge is captured in
application of "literary constraints" 100A3 to "story grammars"
100A2 and "paragraph and sentence grammars" 100A5. Impressionistic
knowledge represents the effects that different language or
compositional elements of the lexical and compositional knowledge
sources 201, 202 may have on the audience. In literature, this type
of knowledge is considered "rhetoric". The usage of certain words
or sets of words in the construction of sentences will produce
different types of impressions on the reader, ultimately affecting
communicative results.
[0083] Further, the invention uses Lexical Impressionistic
Knowledge (LIK) 203. LIK is a specific type of impressionistic
knowledge. In terms of FIG. 2, it is stored in "paragraph and
sentence grammars" and in "literary constraints" databases. It is
realized in the artifact by the application of literary constraints
to the expansion of the paragraph and sentence grammars. Lexical
impressionistic knowledge 203 relates lexical elements and/or
phrases, thereof, to their impressionistic effects. This knowledge
is used to shape the expression in L to achieve different
impressionistic effects on the audience.
[0084] There are a variety of devices used by authors to
emotionally impress their audience. For example, in story
generation, different classes of words or sentence structures may
be used to produce tenor or suspense in the mind of the audience.
Other devices involve selecting certain classes of verbs to known
to communicate psychological consciousness of their subject.
[0085] The system and method of the present invention also uses
Compositional Impressionistic Knowledge (CIK) 204. CIK is a
specific type of impressionistic knowledge represented by the
application of "literary constraints" 100A3 to "story grammars"
100A2. Compositional impressionistic knowledge relates
compositional elements and/or phrases thereof to their
impressionistic effects. This knowledge is used to shape the
structure of an artifact to achieve different impressionistic
effects.
[0086] For example, in story generation, a story may open with a
tragic event that chronologically belongs at the end of the story.
Such a story order may be for the purposes of "shocking" the
audience. The shocking hook is an impressionistic effect. Its
association with an alternative story structure is part of the CIK
204.
[0087] Furthermore, as mentioned above, thematic knowledge (TK)
100A4 is used by the invention. That is, the inventive system and
method introduce the explicit notion of a theme T and a language
for expressing themes, T. The introduction of the theme assists in:
1) maintaining a certain "creative distance" between the input and
output of the creative agent; and 2) focusing the agent on
producing interesting artifacts. Distance between the input and the
output is maintained by the subsequent elements of the process of
story generation being anchored to (e.g., constrained by) the theme
selected. In this regard, along with the random generator
encompassed in the controller, the system can produce interesting
artifacts.
[0088] It is noted that, while each of the "knowledges" are shown
as separate storages/databases in FIGS. 2A and 2B, a single,
partitioned database may be provided, as would be known by one of
ordinary skill in the art taking the present specification.
[0089] FIG. 2C illustrates the generalized form of the invention
(e.g., of FIG. 2B) having been specialized for story 108
generation. In FIG. 2B, the steps of structural template 205, stage
evolution 102A, etc. are described further below with regard to
FIG. 3.
Formalizing a Theme
[0090] Turning now to how to formalize a theme, thematic knowledge
is a collection of themes. Composing particular themes in T (e.g.,
in natural language) is the job of the human. The creation of
interesting themes is outside the scope of the present
invention.
[0091] A thematic expression in T may be expressed in a syntactic
structure of a formal language, like first-order logic or FLEX. A
theme is the primary input to a creative agent of the inventive
system. The theme may be input by the controller 100C accessing the
thematic knowledge 100A4 and selecting a theme (e.g., betrayal,
etc.), or by the user selecting a theme. The following is an
example representation of "Betrayal" in the language FLEX. [0092]
betray(Betrayor, Betrayed) [0093] Evil is some goal whose plan is
an EvilPlan [0094] and whose agent is a Betrayor [0095] and Saying
is included in the Evilplan [0096] and Saying is some say [0097]
and Thwarting is included in the EvilPlan [0098] and Thwarting is
some thwart [0099] and Betrayeds_Goal is the prevented goal of
Thwarting [0100] and Betrayors_Lie is the theme of the Saying
[0101] and Betrayors_lie is some support of the Betrayeds_Goal
[0102] and Betrayed is some person [0103] whose goal is the
Betrayeds_Goal [0104] and whose beliefs include the Betrayers_Lie.
The symbols beginning with a capital letter are variables. The
system of this invention will search for and produce appropriate
values for those variables as part of the story generation
process.
[0105] Other elements of this FLEX representation refer to
primitive structures supplied in the thematic knowledge-base (e.g.,
"lie", "thwart", "plan"). These elements are thematic building
blocks includes as part of this invention. The input command is a
simple selection. Choosing from a multiple choices of pre-defined
themes stored in the knowledge-base. It is simplistic. The casual
user, however, does not define the theme. Someone capable of
representing knowledge in first-order logic or in FLEX for example
(both well-defined formal systems accessible to any logician or
engineer) would be capable of creating the requisite representation
for a theme using elements from the thematic knowledge base only.
These elements would include, in the example of "betrayal"
predicates like "betrayer", "betrayed", "evil goal", "betrayer's
objective", etc. These elements would be combined in a formal
expression to create the theme's "input" representation (i.e., the
representation stored in the knowledge base). Once stored, the user
might select it, among others, by some name (arbitrarily selected
by the person who created the theme's formal or "input"
representation".
[0106] It is noted that it is possible to design and implement a
system that would allow a casual user to compose a theme's formal
representation from constituent elements, stored in the thematic
knowledge-base, without prior knowledge of the formal language.
This, however, is not part of the current invention.
[0107] A theme is independent of the language of expression, the
language of composition, domain knowledge and impressionistic
knowledge. The elements composing the representation of the theme
are drawn from thematic knowledge only. They are independent of the
other classes of knowledge and they are independent of the
mechanisms that result in the story's plot, language, style and
structure.
[0108] Such independence is an important aspect of the present
invention. It ensures the "creative distance" between the input and
the output by constraining the process choices made subsequently.
The selection or even construction of the input theme may be
accomplished without access or prior knowledge of the classes of
story generation knowledge or process. Based on the theme alone,
the story will make selections to produce a variety of stories
about the selected theme.
[0109] While large and varied knowledge sources will lead to
varied, well-structured artifacts, the theme anchors the creative
agent to producing "interesting" artifacts where interesting is
ultimately defined subjectively by the human composing the theme.
The theme performs such anchoring by constraining subsequent
choices made by the controller 100C/random generator in the plot,
characters, dialogue, setting, literary structure, etc.
Process of the Invention
[0110] Turning now to FIG. 3A, the story generation method/process
300 of the invention will be described hereinbelow. It is noted
that while different reference numerals are shown in FIG. 3A than
in FIG. 1 and FIG. 2, the method of the invention shown in FIG. 3A
is substantially the same as that shown in the composite
architecture 100 shown in FIG. 1 and the process level of FIG.
2.
[0111] In the inventive system, the story generation process 300 is
decomposed into four high-level processes including thematic
concept instantiation, plot generation, story structure expansion,
and language generation. Each process uses knowledge represented as
part of the knowledge level and stored in the inventive system's
knowledge-base.
[0112] First, in step 301, a theme is selected by either the system
(controller) or by a user from the thematic knowledge.
[0113] Then, based on the theme selected, the theme is instantiated
in the thematic instantiation step 302. Thematic concept
instantiation begins with a description of a theme. This
description is independent of specific objects or events that might
appear in a particular domain knowledge-base. Given a specific
domain KB (e.g., reference numeral 100A1), the theme is
instantiated, which serves to identify particular objects, events,
characters, etc. from that KB to play the general roles required to
realize the theme. The result of this process (e.g., step 302) is
called a "stage" 303.
[0114] The stage 303 is input to a plot generation step 304. Plot
generation 304, through planning and simulation, weaves the details
of a specific plot for the cast of characters identified in the
stage 303. Plot generation 304 completes what is called a scenario
305. The scenario 305 is the stage plus a completed set of events
and the effects these events had on the state of world.
[0115] Story structure expansion 308 is the root of a separate
process thread in the inventive system. A high-level story
structure, represented in the inventive system as a story grammar,
may be input or randomly selected. As already noted, story grammars
describe how a story is organized with respect to introduction,
character descriptions, conflict, resolution, conclusions, and so
on. Story structures in the inventive system, are preferably
independent of plot and theme.
[0116] The story expansion step 308 recursively expands structural
components of a story represented in the story grammar until a
series of paragraph types are reached. (It is noted that expanding
a formal grammar to produce "grammatical" expressions of that
grammar is well within the grasp of one ordinarily skilled in the
art). Paragraph types are further expanded into a series of
sentence types. The story expansion step 308 is performed in
parallel to the steps 301-305 described above.
[0117] The result is a detailed story template (e.g., reference
numeral 205 in FIG. 2C), or story outline 306, that may be entirely
independent of story content. Literary constraints (e.g., from
literary constraints knowledge 100A3 shown in FIG. 2) may be used
to influence otherwise random choices in the generation of a story
outline. It is through these constraints that decisions regarding
theme or plot may carry over to achieve congruent influence over
story structure.
[0118] The scenario 305 developed in plot generation step 304,
along with the outline 306 produced in the story structure
expansion 308, are input to the final process (e.g., language
generation step 307). In this process, linguistic and literary
knowledge are used to produce the written story 310.
Implementation of the Invention
[0119] In an exemplary implementation and as mentioned above, the
inventive system was implemented using a variety of knowledge
representation and programming techniques available in the
logic-programming system called FLEX, originally developed by Logic
Programming Associates (LPA), Ltd. FLEX is based in Prolog, a
logic-programming language based on a general computational method
that starts with a goal statement and searches a database of facts
to find a proof for the goal. This method models a form of
reasoning often referred to as "goal-directed reasoning." An
in-depth treatment of Prolog can be found in Sterling et al., The
Art of Prolog (Cambridge, Mass., MIT Press (1986)) and R. O'Keefe,
The Craft of Prolog, (Cambridge, Mass.; MIT Press (1990)). The
genealogy of logic programming is well-documented in J. A.
Robinson, "Logic and Logic Programming", Communications of the ACM,
Vol. 35:3, pages 40-65 (1992). The present inventors have also
written on Prolog's genealogy, and its power in Bringsjord, S., and
Ferrucci, D. "Logic and Artificial Intelligence: Divorce,
Separated, Still Married . . . ?" Minds and Machines, Vol. 8; pages
273-308 (1998).
[0120] FLEX provides the developer with complete access to Prolog
and enhances the paradigm with frame-based structures, relations,
production rules, and an English-like syntax. The syntax and
semantics of FLEX are well-documented in Vasey, P., LPA-flex
Technical Reference (London, England; Logic Programming Associates
Ltd. (1989).
[0121] FLEX's frames allow the developer to group sets of facts
around a specific domain entity, enabling a more intuitive
structuring of the knowledge. A frame is used to represent a domain
entity, and its attributes are used to represent facts or
properties about that entity. From a programmer's perspective,
frames are data structures that have a name and a set of attributes
with default values. Frames can be organized in a generalization
hierarchy via the relationship
is-a. Attributes and default values are inherited from a parent in
the frame hierarchy to its children. Instances are special kinds of
frames that have actual values rather than default values.
Instances are leaves in the frame generalization hierarchy. A
sample frame hierarchy, including some instances, for an exemplary
story, is illustrated in FIG. 3B.
[0122] A portion of a sample frame intended to represent the
prototypical person, is shown below. Specific people would be
represented as instances with specific values that might override
the default values inherited from the person frame. The value for
an attribute in a frame can be a list of things. Lists are
represented by { }s. [0123] frame person is an agent [0124] default
height is 5.sub.--5 and [0125] default weight is 130 and [0126]
default gender is female and [0127] default eye_color is brown and
[0128] default goals are {eat, sleep} and [0129] default beliefs
are {"computers write interesting stories"}
[0130] Relations in FLEX are used to define abstract properties
about a frame or a sets of interrelated frames. Relations are
processed by Prolog's goal-directed reasoning mechanism to find the
specific frames and their elements that satisfy the relation.
[0131] For example, a relation may be developed to find all frames
in a knowledge-base that contain the necessary features to play the
role of the betrayed in a story about betrayal. The following FLEX
relation (e.g., noted above and repeated here for ease of
understanding) captures the abstract relationships between various
persons, goals, plans and actions sufficient for the inventive
system to identify the components of a simple story about betrayal.
[0132] relation betrayal_p [0133] if Evil is some goal whose plan
is an EvilPlan [0134] and whose agent is a Betrayor [0135] and
Saying is included in the Evilplan [0136] and Saying is some say
[0137] and Thwarting is included in the EvilPlan [0138] and
Thwarting is some thwart [0139] and Betrayeds_Goal is the
prevented_goal of Thwarting [0140] and Betrayors_Lie is the theme
of the Saying [0141] and Betrayors_lie is some support of the
Betrayeds_Goal [0142] and Betrayed is some person [0143] whose goal
is the Betrayeds_Goal [0144] and whose beliefs include the
Betrayers_Lie.
[0145] The above relation is actual (exemplary) code. Strings
starting with an uppercase letter are logic variables. The strings
say and thwart refer to types of actions defined in the domain
knowledge-base. Goals agents, and plans are also concepts defined
in the domain RB. The English-like syntax of this relation is
translated by the system into standard Prolog programs. These
programs drive a goal-directed search through the knowledge-base
and assign the logic variables to specific domain concepts.
[0146] Production rules or condition-action rules have a condition
and an action. If the condition is met, the actions are executed.
These rules are processed in FLEX by a forward reasoning engine.
They are used in the inventive system to represent the reactive
behavior of agents. The forward reasoning engine is exploited in
the inventive system to provide a computational facility for
simulation that runs in the course of plot generation. A set of
production rules is input to the forward reasoning engine. All
conditions are checked, and rules whose conditions are satisfied
are "fired" (e.g., their actions are executed). The result of the
actions may change the state of the KB, potentially causing other
rules to fire. The process continues until all rules have been
processed (some may fire and some may not). If no rules fire, then
no rules were intended to fire. The wording "fail to fire" is
intended to mean the rule did not fire because it was not supposed
to given the current state of the knowledge-bases. There is not
real "failure" of any kind. FLEX provides a variety of control
features for specifying the order in which rules should fire and
for resolving conflicts if more than one rule can fire at a
time.
The Knowledge Level
[0147] Turning now in detail to the knowledge level, as mentioned
above, domain knowledge 100A1 (in FIGS. 2A and 2B, with specialized
domain knowledge 100A1' being shown in FIG. 2C) includes a static
description of the world, or domain, about which a story is
written. Domain knowledge includes descriptions of people, places,
things, events, actions, goals, and behaviors. These descriptions
are represented in the inventive system as FLEX frames.
[0148] Thus, for example, agents perform actions and participate in
events. In stories about people, people are obviously an important
class of agents. People are represented as agents with physical and
psychological attributes. In addition to the typical physical
attributes like height, weight, and eye color, a person's
attributes include a set of goals and a set of beliefs. The latter
set is an important element in plot generation, as would be evident
to one of ordinary skill in the art.
[0149] Consider frame excerpts describing the characters Striver, a
doctoral student/candidate at the University of Rome, and Professor
Hart, Striver's mentor and one of three reviewing members of
Striver's doctoral thesis, and who has intimated to Hart that
Hart's thesis should pass any opposition. It is in part through the
actions associated with characters' goals that a plot is developed
from a stage. Hart's evilGoal refers to a goal that includes
actions intended to prevent Striver from achieving his goal. The
interplay of these various intentions is integral to the betrayal
theme. [0150] instance hart is a person [0151] name is `Hart` and
[0152] gender is male and [0153] goals is {evilGoal}. [0154]
instance striver is a person and [0155] name is `Dave Striver` and
[0156] gender is male and [0157] goals is {to_graduate} and [0158]
beliefs is {harts_promise}.
[0159] Events are concepts that relate people, places, and things.
In the inventive system, the focus is on the specific event
attributes required to facilitate representation of key story
elements.
[0160] Hart and Striver are described as people with physical and
psychological attributes. They are interrelated in the domain KB
100A1 through their roles in common events. For example, Striver's
thesis defense is a thesis defense, which is a type of event, where
Striver is the examined and Hart is a member of the examining
committee. Among other attributes, the examined and the committee
are the key elements of this event, which link Striver and Hart.
[0161] instance strivers_defense is a thesis_defense [0162] thesis
is strivers_thesis and [0163] where is university_of_rome and
[0164] examined is striver and [0165] committee is
{hart,meter,rodgers,walken} and [0166] status is scheduled.
[0167] The set of beliefs associated with a person describes in
part what that person believes is true about the world. However,
that which a person believes need not be true In the inventive
system, a person, by default, is assumed to believe all that is
true about the physical state of the world as represented in the
domain KB (e.g., hereinafter simply "the world"). This excludes
knowledge about other people's psychological attributes or beliefs.
A person is not assumed to have knowledge about other people's
beliefs, goals, or behavior patterns, unless otherwise explicitly
stated.
[0168] For example, if Hart states publicly that he supports
Striver's success, then Striver, by default, believes this
statement was made. However, the system does not conclude by
default that Striver believes that Hart is telling the truth (or is
lying). Statements about another person's beliefs are explicitly
inferred in the inventive system. Explicit belief statements are
included by the designer of the knowledge-base or may be inferred
as the result of a rule.
[0169] Consider Striver's belief about Hart's promise to help
Striver and support his thesis defense. A promise is a type of
statement. An exemplary frame representing Hart's promise follows.
[0170] instance harts_promiae is a promise [0171] agent is `Prof.
Hart` and [0172] utterance is `I will sign your thesis at your
defense.` and [0173] promised is `Dave Striver` and [0174]
intended_state is signatures of striver_thesis include `Prof.
Hart`.
[0175] The intended_state associated with a promise describes some
aspect of the state of the world that would be true if the promise
were kept. The following exemplary frame demonstrates the
representation of Striver's belief about the truth of Hart's
promise: [0176] instance striver is a person [0177] name is `Dave
Striver` and [0178] gender is male and [0179] goals is
{to_graduate} and [0180] beliefs is {harts_promise}.
[0181] In the exemplary implementation, persons may also believe
statements that contradict the current state of the world. For the
person with the false belief, these contradictory statements
override that actual state of the world. Contradictory or false
beliefs must be stated explicitly in the list of beliefs associated
with a specific person. For example, if the saying of harts
promise, represented by the following frame say101, [0182] instance
say101 is a say [0183] agent is `Prof. Hart` and [0184] statement
is harts_promise. existed in the domain knowledge-base and Striver
did not believe that Professor Hart made the promise (i.e., that
Hart uttered the statement), then Striver's beliefs must contain an
explicit negative belief. The following frame represents that
Striver does not believe that say101 occurred. [0185] instance
striver is a person [0186] name is `Dave Striver` and [0187] gender
is male and [0188] goals is to_graduate and [0189] beliefs is
{not(say101)}.
[0190] Goals, plans, and actions are used to represent a
character's proactive behavior. Once a character is set in motion,
the character attempts to achieve its goal by executing the actions
associated with the goal's plan. Each action has preconditions and
state-operations. The pre-conditions ensure that the setting is
right to execute the action; the state-operations attempt to change
the state of the domain. If, for some reason, a state-operation
fails to execute, then the action fails.
[0191] If all the elements of a plan succeed, then the goal is
successfully completed by the corresponding agent. Goals, plans,
and actions are considered part of the proactive behavioral model
because they are initiated by the character as part of his or her
explicit intentions.
[0192] A goal is a concept that has three main attributes of: 1)
Agent: someone or something that can achieve the goal, typically a
person; 2) Plan: a list of actions; and 3) Success state: a
condition that if true would indicate that the goal has been
satisfied.
[0193] For example, Hart may have the goal to destroy Striver's
hope to graduate. To achieve this goal, there would be an exemplary
sequence of actions that the agent, Hart, must execute. This
sequence is the "plan". It may include, for example, conveying to
the examining committee that Striver's ideas are worthless by
refusing to sign his thesis. Hart may execute all elements of the
plan, but there is no guarantee that his goal will succeed. A
goal's success state is a condition which if true of some elements
of the domain, indicates that the goal has been achieved (e.g., the
agent has succeeded). In this case, a success state may be the
failure of Striver's thesis defense.
[0194] The following frame represents Hart's goal to thwart
Striver's goal to graduate. [0195] instance evilGoal is a goal
[0196] agent is hart and [0197] plan is {lie101, refuse_to_sign101}
and [0198] success is status of strivers_defense is failed.
[0199] lie101 and refuse_to_sign101 are specific actions that will
execute during the plot generation step 304 on Hart's behalf when
their preconditions are met.
[0200] Reactive behavior describes how a character reacts to
changes or events in the world. Reactive behavior is represented by
a set of condition-action rules (or production rules) associated
with agents. If the condition is met, then the actions are
executed.
[0201] With reactive rules, the inventive system allows character
behavior to be described in terms of how a character may respond to
different events independently of the character's specific goals.
Reactive behavioral knowledge is ideal for describing basic
elements of a character's default behavior independently of
specific roles in a theme. It provides a representation mechanism
for capturing particular behavioral dimensions that "trademark" a
character.
[0202] Rules associated with a character's behavior may be as
specific or as general as the knowledge-base designer feels is
appropriate For example, Professor Hart might have specific
behavior rules that describe exactly how he may react to a given
state. Alternatively, without a specific rule, Professor Hart might
react according to a default behavioral rule associated with all
persons.
[0203] Reactive behavior in the inventive system functions both to
increase the variability of a plot's outcome during simulation and
to produce dialogue.
[0204] The inventive system combines proactive and reactive
behavior in simulation in order to vary plot generation while still
maintaining a theme-driven focus. The thematic frame sets key
characters in motion by triggering their proactive behavior.
Effects may, in turn, trigger the reactive behavior of other
characters who can twist the plot by helping or hurting a
character's proactive agenda.
[0205] The following production rule describes the typical reactive
behavior of, for example, a member of a thesis committee. In
summary, this rule states that if a member of a thesis committee is
asked by the chair of the committee to sign the candidate's thesis,
the member reacts by signing the thesis. [0206] rule
committee_members_behavior [0207] IF [0208] Candidate is some
person and [0209] Thesis is the thesis of Candidate and [0210] the
committee of the Candidate includes Member and [0211]
Request_To_Sign is some request and [0212] Member is the requestee
of Request_To_Sign and [0213] the requester of Request_To_Sign is
the chairman of the committee and [0214] Thesis is the document of
subject of Request_To_Sign and [0215] status of Request_To_Sign is
pending [0216] Then [0217] do answer(Member, Request_To_Sign) and
[0218] do sign(Member, Thesis).
[0219] Proactive behavior, by default, overrides reactive behavior.
In the above-mentioned exemplary story of betrayal, for example,
Professor Hart intends to thwart Striver's goal to graduate, by
refusing to sign his thesis. Professor Hart does not react like a
typical committee member. His proactive refusal to sign Striver's
thesis must override (e.g., constrain) the reactive behavior
imposed by the production rule previously described.
[0220] Such is accomplished in the inventive system by ensuring
that all actions associated with plans are given a chance to
execute in response to changes in the knowledge-base before
production rules are fired in response to the same changes. In the
exemplary case, the fact that Professor Hart refused to sign the
thesis would occur first, ensuring that the reactive rule would not
execute for him, since the request directed to Professor Hart would
no longer be pending when the reactive rule fired.
[0221] By changing behavioral rules and/or actions, and activating
different sets of rules, the simulation process will unfold
different plots about the same theme and the same stage. For
example, varying Professor Hart's reaction to pressure from the
chairman to sign the thesis can result in a happy or sad ending for
the candidate, Dave Striver. While the story is still about
betrayal, the difference revolves around whether or not the
betrayer succeeds in the quest to thwart the betrayed's goals.
[0222] While domain knowledge, the thematic relation, and the
results of simulation determine what the inventive system will
write about, how the inventive system will express the story is
determined by linguistic and literary knowledge.
[0223] As described above, linguistic knowledge is knowledge about
natural language. The written story is formed from "legal"
expressions in a natural language. For example, in the inventive
system, the natural language may be English, German, etc.
[0224] Linguistic knowledge is explicated through observations of
how humans express internal knowledge in distinct language
structures. Linguistic theory describes language structures at a
number of different levels. These levels include: 1) Discourse; 2)
Paragraph; 3) Sentence; 4) Phrase; and 5) Word.
[0225] The discourse level may be viewed as the most aggregate
level, and is composed by configuring elements from the paragraph
level. The paragraph level, in turn, is composed by configuring
elements from the sentence level, and so on.
[0226] Words are defined as the smallest independent units that can
be assigned descriptive meaning. However, words can be further
decomposed into sets of letters that act as prefixes, roots, and
suffixes. These can be combined to form different words. The study
of how these combinations occur is termed "word formation
morphology". The morphology of words may be influenced by semantic
properties relating to number, gender, case, tense, etc. The study
of the forms of words as they relate to these properties is termed
"derivational morphology". Finally, words may take on different
forms based on agreement with the forms of other words which occur
simultaneously in a phrase. For example, the number of a noun
influences the form of the verb in a noun phrase. The study of the
types of agreement and how they influence the form of words is
termed "inflectional morphology".
[0227] Encoded in the linguistic knowledge-base are a lexicon
containing a set of words and the morphological rules necessary to
demonstrate story generation examples. R. C. Dougherty, Natural
Language Computing: An English Generative Grammar in Prolog
(Lawrence Erlbaum Associates; Mahwah, N.J. (1994) provides details
on encoding morphological rules.
[0228] In general, a sentence in a natural language, L, is a string
of words taken from a fixed list of words that constitute L's
lexicon. A natural language grammar for L is a device that
distinguishes the grammatical from the ungrammatical sequences of
words. This defines the valid (i.e., grammatically correct)
sentences of L. For a treatment of formal grammars for a subset of
English, and their use in mechanized communication, the present
inventors refer to S. Russell et al. Artificial Intelligence: A
Modern Approach (Prentice-Hall, Englewood Cliffs, N.J. (1995).
[0229] The inventive system encodes (e.g., via the controller and
its interfaces with the "Story Grammars" knowledge base 100A2 and
its interfaces with the story expansion and other process steps of
the invention), as Prolog goals, a variety of sentence-level
grammars, which include phrase-level grammars, for generating
grammatically correct sentences. These generative grammars
presuppose lower level linguistic procedures to ensure, among other
things, verb/subject agreement, punctuation, etc.
[0230] Sentence grammars are categorized in the inventive system to
represent classes of sentences that have a particular structure.
These sentences vary around pivotal elements. Inventive system uses
numerous sentence grammars. Some examples include sentence types
designed to describe a setting, to describe a character, to reveal
a character's goals, to narrate different classes of events, etc.
Variations can result in the generation of a negative or positive
tone, or the selection of words to produce different classes of
imagery.
[0231] As an example, a sentence grammar for producing a sentence
type called an "independent parallel setting description" is
illustrated below: [0232] INDPSD.fwdarw.SETTING verb FP [0233]
FP.fwdarw."its" FEATURE FP `its` FEATURE [0234]
SETTING.fwdarw.noun_phrase [0235] FEATURE.fwdarw.noun_phrase
[0236] In sentence grammars, uppercase words in the grammar are
non-terminals. Words in quotes are literals used exactly as is in a
generated sentence. Lowercase words are terminals in the sentence
grammar. The terminals are phrase- or word-types that are selected
and/or constructed from the available lexicon when a sentence is
generated. Sentence generation procedures use sentence grammars to
produce individual sentence that conform to the grammar. Typically,
certain elements are fixed in the application of a sentence
generation procedure. Inventive system supports different
generation procedures for the same grammar. A generation procedure
pivots around the elements which it allows to be fixed. For
example, a procedure that pivots on SETTING will generate all
possible sentences that can be built from the lexicon where SETTING
is fixed to a particular noun. Other sentence generators for a
given grammar may pivot on more than one element.
[0237] An example of an independent parallel setting description
sentence where SETTING is fixed to the noun university is as
follows: [0238] The university has its ancient and sturdy brick,
its sun-splashed greens, and its eager youth.
[0239] Thus, a noun that is a setting is chosen, and some
possessive verb is used to link the setting to a list of feature
phrases that describe a set of objects considered characteristic of
the setting.
[0240] For sentences of the above form to have literary effect,
elements of the sentence preferably are somehow associated with
literary knowledge. As described below with regard to linking
literary and linguistic knowledge, augmented grammars that enhance
linguistic structures with literary objectives are described.
[0241] A description at the discourse level in the inventive system
is realized by a structure called the "story grammar" described
above. As mentioned above, such a grammar describes, at a high
level, how the story will be laid out in terms of paragraphs that
describe the setting, characters, and events that make up the
story. The stories generated can be viewed as literary concepts as
opposed to linguistic concepts, since they are ultimately
language-independent. Indeed, the same story grammar can be used to
both lay out the frames in a silent movie and generate a written
story.
[0242] While the linguistic knowledge-base identifies types of
words, phrases, and sentences as they function in the English
grammar, as mentioned above, literary knowledge describes different
ways to use words and phrases together to achieve a variety of
literary objectives.
[0243] Literary objectives may include generating imagery in the
reader's mind, suggesting a character's landscape of consciousness,
and producing a positive, secure mood, or a negative, anxious one
for the reader. Literary objectives like these may be achieved by
selecting appropriate words and sentence structure combinations.
Literary concepts, found in the inventive system's literary
knowledge-base, define attributes of linguistic concepts as they
relate to domain concepts so that text generation procedures can
produce expressions that achieve literary objectives.
[0244] Objects described in the domain knowledge-base are
interrelated and linked to linguistic concepts (e.g., words,
phrases, etc.) as one way of capturing literary knowledge. The
resulting associations between concepts and language elements are
called "literary associations". They are used to generate sentences
that satisfy specific literary objectives.
[0245] The inventive system includes a plurality of types of
literary associations including: 1) Iconic features; 2) Literary
modifiers; and 3) Literary analogs.
[0246] For example, objects are related to other objects in the
inventive system via an attribute called "iconic features". An
object's iconic features are represented as a list of other objects
that are typically associated with the first object in literary
settings. For example, ivy and clocktowers might be iconic features
of a university (as in the exemplary story mentioned above).
Wheels, engines, and speed might be iconic features of a
motorcycle. The list of iconic features associated with an object
is further specialized with respect to general notions of positive
or negative imagery. This classification is highly subjective and
may vary with respect to context. However, it can be very effective
for imaging.
[0247] Consider the following frame fragment describing a
university.
[0248] frame university is a object [0249] defalt
positive_iconic_features is [0250] {clocktowers. brick. ivy, youth,
architecture, books, knowledge. scholar, sports) and [0251] default
negative_iconic_features is [0252] {tests, competition,
`intellectual snobbery`}.
[0253] In this frame, ivy is listed as a positive iconic feature of
a university, while tests is listed as a negative one. Though the
subjectivity of these classifications is apparent, the
representational framework allows for capturing and configuring a
variety of literary descriptions.
[0254] Additionally, there are positive and negative literary
modifiers. That is, the concept of literary modifiers is an
association between modifiers and the objects that they typically
modify. In the inventive system, these associations are grouped
into positive and negative classes. The associations are linked to
the frames representing the modified objects with the
attributes
[0255] positive_literary_modifiers and
[0256] negative_literary_modifiers.
[0257] As is the case with iconic features, the negative and
positive classification of literary modifiers is entirely
subjective, but nonetheless this classification is profitably
configurable in the literary KB. For example, given the frame for
university and the following frame fragment for ivy [0258] frame
ivy is an object [0259] default positive_iconic_features is
{leaves. vines) and [0260] default negative_iconic_featurs is
{poison} and [0261] default negative_literary-modifiers is
{poisonous tangled} and [0262] default positive_literary_modifiers
is {spreading. green, lush}. a text generation procedure for
independent parallel setting description sentences can produce
positive imagistic sentences such as: [0263] The university has its
lush, green ivy. If the knowledge-base were reconfigured and ivy
was also listed as a negative iconic feature of university, then
the following negative imagistic sentence can be exemplarily
produced: [0264] The university is full of tangled poison ivy.
[0265] Sentence generation in the inventive system are
parameterized and can be called to produce negative or positive
phrases.
[0266] Literary Analogs Metaphors are used in order to support
sentence types in the inventive system. These literary devices are
supported by an underlying representation structure called
"literary analogs". A literary analog identifies a single main
object, another object called the "analog" that is used to
represent the main object in a literary fashion, and the intended
imagery suggested by the analog.
[0267] The following frame is an instance of an analog for the main
object eye and one of its analogs, the sun. [0268] instance analog1
is a literary_analog [0269] object is eye [0270] analogs are sun
[0271] images are {warmth. power trust}.
[0272] To the extent that sentence grammars capture the grammatical
rules of a natural language, they are considered linguistic
concepts. However, in the inventive system, these structures are
augmented to contain elements that reflect the linkages between
literary objectives and grammatical structure.
[0273] Literary augmented grammars (LAGs) are used in the inventive
system to represent and use literary knowledge to produce
compelling and interesting prose. Consider the following LAG for
independent parallel description sentences:
[0274] INDPSD.fwdarw.SETTING verb (isa possessive_verb) FP
(n=3)
[0275] FP.fwdarw.`its` FEATURE FP `its` FEATURE
[0276] SETTING noun_phrase (has role setting)
[0277] FEATURE noun_phrase (isa iconic_feature_of SETTING)
[0278] Elements on the rightside of the LAG rules are annotated
with expressions in parentheses. These expressions are called
"literary constraints". It is through literary constraints 100A3
that linguistic and literary knowledge interact. While sentence
grammars drive the construction of classes of grammatically correct
sentences in a natural language, literary constraints in LAGs are
designed to shape the contents of the sentence by using literary
and domain knowledge to achieve various literary objectives.
[0279] Sentence generation procedures limit the instantiation of
terminals to words or phrases that satisfy the literary constraint.
The constraints describe the literary roles that the elements of
the sentence must assume independently and/or with respect to one
another. Constraint processing relies on how words (and their
associated objects) are classified and linked in the literary
knowledge-base. The constraints are used to search the literary and
linguistic Knowledge Bases, where nouns, modifiers, verbs, etc. are
categorized and associated with one another by a variety of
classifications and associations. For example, brick, greens, and
youth are nouns classified as iconic features of a university.
[0280] The constraint
[0281] (isa possessive_verb)
in the preceding grammar constrains the preceding terminal, verb,
to be instantiated from a certain class of verbs, namely,
possessive verbs. Membership in this class is captured in the
knowledge-base.
[0282] Similarly, (has role setting) constrains the selection for
the terminal noun_phrase in the SETTING rule of this grammar to be
instantiated by a noun phase whose subject can function as a
setting according to the classifications contained in the literary
knowledge-base. The constraint isa feature_of SETTING is used to
narrow the selections for a noun phrase based on its relationship
with a choice for SETTING. In this case, the constraint ensures
that the noun in
the associated noun phrase is represented as an iconic feature of
whatever object is selected in the noun_phrase for SETTING.
[0283] The constraint (n 3) in the first rule of this grammar
instructs the sentence generation procedures to produce exactly
three descriptive features in any generated sentence.
[0284] Regarding imagistic expertise, there have been four
techniques identified for triggering images in the minds of the
reader, including 1) Exotic or bizarre material; 2) Visual
perception and P-consciousness-related verbs; 3) Familiar
reference; and 4) Voyeurism.
[0285] The inventive system includes a framework in the literary KB
for classifying linguistic elements to enable the automation of
these techniques.
[0286] For example, to capture the bizarre, modifiers are linked
with objects in frames named bizzaro_modifiers. Consider the
following instance describing the bizzaro modifier bleeding.
[0287] instance bleeding is a bizzaro_modifier
[0288] objects are (sun, plants, clothes, tombs, eyes}.
An action analogy LAG may be augmented with constraints to
stimulate bizarre images in the mind of the reader. The following
LAG for action analogies,
[0289] BizarreActioniaalogy.fwdarw.NP VP like ANP
[0290] NP.fwdarw.noun_phrase
[0291] ANP.fwdarw.modifier (isa bizzaro_modifier) noun (isa analog
of NP)
in conjunction with bizzaro_modifiers can be used to generate the
following sentence.
[0292] Hart's eyes were like big bleeding suns.
[0293] The inventive system also triggers images with the technique
of visual perception by constraining the principal verb in the
appropriate LAG to be a member of a class of verbs called vp_verbs.
These include verbs for seeing, looking, glancing, and so on.
[0294] Similarly, the inventive system narrates stories from a
particular character's point of view using verbs selected from a
special class of verbs represented in the literary KB called
pc_verbs. These include verbs for feeling, thinking, understanding,
wanting, etc. PC verbs give the reader the sense that the subject
of these verbs has a psychological life.
[0295] A variation of the parallel setting description sentence is
a sentence that describes a setting from a particular character's
point of view. The following is the LAG for that sentence type,
which includes a literary constraint that forces the vise of a PC
verb to convey the character's consciousness:
[0296] POV.fwdarw.Agent (is a person) Verb (is a PC Verb)
FirstFP
[0297] FirstFP.fwdarw.Setting FEATURE FP
[0298] FP.fwdarw.its FEATURE FP I "."
[0299] FEATURE.fwdarw.noun_phrase (is a feature of SETTING)
[0300] SETTING.fwdarw.noun_phrase (is a setting)
[0301] Stories that appeal to a reader's familiar experiences are
known to quickly engage their audience. Triggering images of
familiar things in the reader's mind keeps him or her interested
and gives the author opportunity to manipulate the reader's
expectation, better enabling the effect of plot twists and
emotional turns. The inventive system can trigger images with
familiar reference by classifying settings, objects and features as
familiar to a class of readers and ensuring that they are used in
stage and language generation through the elaboration of the
thematic relation and the application of literary constraints.
[0302] In the exemplary story of betrayal, the university setting
is a familiar reference to a wide class of readers and is expected
to quickly form familiar and comforting images in the minds of
these readers. These images are reinforced with the choice of
particular modifiers and features. Of course, in the above
examples, the imminent betrayal is quickly suggested, shifting the
focus from the positive reference of the university to the
universal image of classic betrayal.
[0303] Another literary "trick" known to trigger images and spark
the reader's interest is voyeurism. Indeed, Franz Kafka, for
example, often included in stories scenarios in which one person
sneaks a peak at the behavior of others, or even places the reader
in the position of outright voyeur.
[0304] For example, consider an elaboration of betrayal,
voyeuristic betrayal. The inventive system may include a thematic
relation for voyeuristic betrayal that requires a few new stage
elements, including a new character, the voyeur, and a new action
secretly watch. The relation may compose a stage where the voyeur
is envious of the betrayer and engages in regular acts of secretly
watching the betrayer's activities. The stage will initialize the
story with the potential for an additional subplot involving the
voyeur secretly witnessing the betrayer reveal his plans to betray
the betrayed to some confidant. The Plot generation step will run
the stage through a simulation, as described above, working out the
details regarding the voyeur's experience and decision to tell what
he witnessed to the betrayed. Regardless of how the plot develops,
well-crafted sentences and word choices that describe scenes
involving the secret witnessing of another's behavior will serve to
stimulate the emotions and imaginations of human readers.
[0305] Returning now to the process level 100B of the invention (as
shown in FIG. 2A and FIGS. 2B-2C for the generalized and specific
embodiment of the invention), the process of thematic instantiation
selects components from the domain knowledge-base sufficient to
generate a story about a particular theme. The process of thematic
instantiation uses a FLEX relation, based on a formal
mathematization of a thematic concept like betrayal, and the
goal-directed reasoning mechanism underlying Prolog, to search the
domain knowledge-base 100A1 and assemble the story elements
(characters, goals, events, settings, etc.) required to construct a
story about the theme.
[0306] The following FLEX relation represents the components and
their interrelationships sufficient to generate a story about the
theme of betrayal. [0307] relation betrayal_p (A_Betrayal) [0308]
if Evil is some goal whose plan is an EvilPlan [0309] and whose
agent is a Betrayor [0310] and Saying is included in the EvilPlan
[0311] and Saying is some say [0312] and Thwarting is included in
the EvilPlan [0313] and Thwarting is some thwart [0314] and
Betrayeds_Goal is the prevented_goal of Thwarting [0315] and
Betrayers_Lie is the theme of the Saying [0316] and Betrayers_Lie
is some support of the Betrayeds_Goal [0317] and Betrayed is some
person [0318] whose goal is the Betrayeds_Goal [0319] and whose
beliefs include the Betrayors_Lie.
[0320] A rough intuitive translation of how this particular
relation is interpreted from a procedural perspective runs as
follows.
[0321] First, look in the domain knowledge-base and find some goal
with an associated plan and agent, where the plan includes a saying
action and a thwarting action. This agent might be engaged in an
act of betrayal; the agent might be the betrayer.
[0322] Then, look to find that the alleged betrayer's thwarting
action is intended to thwart someone else's goal. The person who
has this goal might be vulnerable to a betrayal. That person might
be the betrayed.
[0323] Finally, look to find that the alleged betrayer's statement
is in support of the betrayed's goal and that the alleged betrayed
believes the statement.
[0324] The search of the domain KB 100A1 described here is
performed automatically by the Prolog inference engine. If the
search is successful, then the thematic instantiation process
builds the stage. The stage, implemented by a FLEX frame,
represents the key roles in a story about the designated theme.
[0325] In the inventive system, thematic instantiation 101 captures
the essential characteristics of betrayal by building the following
frame: [0326] frame betrayal is a concept [0327] default betrayer
is a person and [0328] default betrayed is a person and [0329]
default betrayers_goal is a goal and [0330] default betrayeds_goal
is a goal and [0331] default betrayers_lie is a statement and
[0332] default betrayal_location is a place and [0333] default
betrayers_evil_action is a action.
[0334] While the domain elements of the story are selected as a
result of thematic concept instantiation, the story generation
process has yet to develop a specific plot. The domain concepts are
static. They represent a snapshot in time, one capturing different
characters and events that are the principal ingredients for a
story. In plot generation, the ingredients are cooked; the details
of plot are produced based on the behaviors of the characters.
[0335] In the inventive system, thematic instantiation 101 requires
that the domain knowledge-base include many of the specific objects
required to instantiate the theme.
[0336] Thereafter, the plot is developed through simulation. The
process of plot generation 103 enables the inventive system to
produce a variety of plots around a particular theme and a
particular stage.
[0337] Through a forward reasoning process provided by FLEX, the
inventive system can simulate character behavior. The principal
knowledge used in plot generation is behavioral knowledge. As
discussed earlier, characters have proactive behavior, represented
by goals and plans, and reactive behavior represented by
condition-action rules. These rules represent how characters react
to particular states. Once set in motion, they result in actions
that affect new states, and more rules fire; this process simulates
a chain of reactive behavior. Eventually the process of
forward-reasoning halts, indicating that a final state has been
achieved and the simulation is over.
[0338] Changes made to behavioral knowledge lead the inventive
system to produce different plots given the same theme and stage.
In this way, the inventive system may be used to write different
stories about the same theme with the same cast of characters,
events, and initial states.
[0339] Plot generation 103 is very much a computational device for
dynamically extending the domain knowledge-base through time.
Initially, the domain KB contains basic story elements that may
interact in a variety of ways. The use of rules and simulation
enables the process of plot generation to play out a scenario by
simulating the behaviors of the selected set of characters and
recording the results of the simulation in terms of new events,
states, and dialogue (i.e., the results of speaking actions) in the
domain KB. At the end of the simulation, the plot is developed and
a particular outcome is realized. The new domain KB contains the
completed plot or the scenario.
[0340] For example, suppose the process of thematic instantiation
101 has produced, from the domain KB, a stage 102 for the theme of
betrayal, including the following elements (expressed informally
for readability): [0341] betrayer: Professor Hart [0342] betrayed:
Dave Striver [0343] location: University of Rome [0344] betrayed's
goal: to get all members of thesis committee to approve and sign
[0345] thesis [0346] betrayer's promise: to approve betrayed's
thesis.
[0347] Plot generation will extend the story elements though time
by executing actions associated with character goals, and will
process behavioral rules, the final result being a detailed
scenario. Specific actions that execute in the development of this
stage include: [0348] sign [0349] refuse-to-sign [0350] say [0351]
answer [0352] request_signatures [0353] demand
[0354] Behavioral rules governing the typical behavior of committee
members and the committee chairperson would execute during plot
generation. For example, at the end of the thesis defense, the
following rule would trigger the committee chairperson to request
the members to sign the thesis. [0355] rule
committee_chairs_behavior 1 [0356] if Committee is a thesis defense
committee [0357] and chair is the chair of Committee [0358] and
Defense is the subject of the Committee [0359] and the status of
Defense includes completed [0360] and unjudged [0361] then do
(request_signatures(Chair,Committee)).
[0362] The following rule would result in the typical committee
member agreeing to sign the thesis when asked by the committee
chair. [0363] rule committee_members_behevior [0364] IF [0365]
Candidate is some person and [0366] Thesis is the thesis of
Candidate and [0367] the committee of the Candidate includes Member
and [0368] Request_To_Sign is some request and [0369] Member is the
requestee of Request_To_Sign and [0370] the requester of
Request_To_Sign is the chairman of the committee and [0371] Thesis
is the document at subject of Request_To_Sign and [0372] status of
Request_To_Sign is pending [0373] THEN [0374] do answer(Member,
Request_To_Sign) and [0375] do sign(Member, Thesis)
[0376] The result of plot generation 103 produces a scenario 104
that would, for example, include the following series of actions
and states (expressed informally for readability):
[0377] Prof. Hart tells Dave Striver "I will support your defense
and sign your thesis" at T0.
[0378] Dave Striver completed his thesis defense at time T1.
[0379] Prof. Rodgers requests approval signatures of members of
Dave Striver's committee at
[0380] time T2.
[0381] All committee members except Prof. Hart sign Dave Striver's
thesis at time T3.
[0382] Prof. Hart refuses to sign Dave Striver's thesis at time
T4.
[0383] Prof. Hart says "Dave does not deserve to graduate." at
T5.
[0384] Prof. Rodgers insists that Prof. Hart signs at T6.
[0385] Prof. Hart refuses to sign at T7.
[0386] Dave Striver's thesis is rejected.
[0387] All the actions that take place in this series are encoded
in the domain KB and are related through the stage to the roles
that participate in the theme of betrayal. This relationship allows
the inventive system to infer, among other things, that Professor
Hart's refusal to sign Striver's thesis is the specific act through
which the betrayer successfully thwarted the goal of the betrayed
(i.e., Striver's goal to graduate).
[0388] The inventive system must now compose the language to write
the story based on the scenario and its relationship to the theme.
Thus, writing the story includes proceeding to the outline and
language generation.
[0389] As mentioned above, thematic instantiation and plot
generation produces a stage and then a scenario respectively. The
scenario will be input into the process of language generation. In
this process, sentences are constructed based on the characters,
goals, events, etc. present in the scenario. The sentence-types
chosen and the sequence in which they appear depend on the story
outline.
[0390] The process of story structure expansion builds a story
outline 306, as described above and as shown in FIG. 3A. Story
grammars, discussed above, are used to represent a variety of
possible story structures. The process pursues a path through a
grammar hierarchy, making either random or constrained choices as
non-terminals in the grammars are expanded until a string of
terminals, representing sentence types, is produced. The sequence
of sentence types is the story outline 306 (e.g., see FIG. 3A).
[0391] The inventive system captures the knowledge used by story
structure expansion and language generation in a three-level
grammar hierarchy, as shown in FIG. 3C. This hierarchy is termed
"the literary-to-linguistic grammar hierarchy" because of its
successive levels takes story expansion from a high-level story
structure (literary knowledge) down to an-English-sentence grammar
and word selection (e.g., linguistic knowledge).
[0392] The top level of the hierarchy is composed of story
grammars. Story grammars may be organized in a taxonomy, in which
top-level story grammars are very generic and may apply to a wide
variety of story types. Lower levels in the story grammar taxonomy
may contain story grammars specialized for particular kinds of
stories, like stories about betrayal, for example.
[0393] The terminals, or leaves of story grammars, are names of
paragraph types. These are associated with structures in the second
level of the grammar hierarchy called "paragraph grammars". The
terminals of these grammars are in turn sentence-types.
[0394] The final level in the grammar hierarchy is composed of
literary-augmented sentence grammars (LAGs), which are formal
language grammars that represent components of English syntax
augmented with literary constraints. The augmentation enables the
generation of grammatically correct sentences that achieve
particular literary objectives. The leaves of LAGs are variables
instantiated to words that represent domain concepts, as described
above.
[0395] Starting with a story grammar, the structures in the grammar
hierarchy are expanded until a sequence of sentence-types is
generated. This entire process may be performed independently of
the scenario produced by thematic instantiation and plot
generation. However, augmenting the story and paragraph grammars
similarly to the way LAGs do for sentences allows for shaping the
story outline by constraining choice points in its expansion based
on literary objectives. These objectives may be determined by
different elements of the theme or plot. More generally, the
inventive system architecture suggests that any parameter (e.g.,
author styles or user preferences, etc.) may be used to constrain
choices in the expansion of a story outline. Alternatively, the
notion suggested by LAGs only to direct sentence generation.
[0396] Language generation takes the scenario and the story outline
and uses literary and linguistic knowledge to generate a story in
English. The language generation process is directed to "choosing
words". The story outline is a map identifying the sequence of
sentence types that will be used to tell the story. Each of these
sentence types must be instantiated (e.g., subjects, objects,
verbs, adjectives, etc. must be chosen from a lexicon). These
choices are directed by the story elements represented in the
scenario and the constraints imposed by literary objectives
captured in the LAGs.
[0397] Consider the following fragment (simplified to enhance
readability) from the inventive system's story grammar specialized
for short stories about betrayal:
1.
Story.fwdarw.Setting+Goals_and_plans+Betrayers_evil_action+betrayed`
s_state 2.
Goals_and_plans.fwdarw.Betrayed's_goal+Betrayers_promise+Betrayers_goal
3. Setting.fwdarw.setting_description(betrayal_locat ion,pov
betrayed) 4. Betrayed's_goal.fwdarw.personal_goal_sentence
(betrayed) 5. Betrayed's_goal.fwdarw.personal_goal_sentence
(betrayer) 6.
Betrayers_promise.fwdarw.promise_description(betrayer_betrayed) 7.
Betrayers_evil_action.fwdarw.narrate_action(betrayers_evil_action)
[0398] This grammar specifies a structure that first exposes the
reader to a description of the setting where the betrayal occurs
from the point of view of the betrayed.
[0399] It then describes the betrayed's goal and the betrayer's
promise to assist the betrayed in achieving that goal.
[0400] Finally, the story ends with a description of the betrayer's
evil action that ultimately thwarts the goal of the betrayed, and a
statement about the betrayed's state. Non-terminals in this grammar
begin with capital letter. Terminals begin with lowercase letters
and indicate the names of paragraph grammars or LAGs. Arguments to
grammars are roles contained in the stage-produced thematic
instantiation and are assigned domain entities from the stage
completed in plot generation.
[0401] The third rule in this story grammar leads to a paragraph
grammar containing two sentences for describing a location from
some character's point of view. Following from our sample stage,
the paragraph grammar called in this third rule would be
[0402] setting_description(university_of_rome,pov, `Dave
Striver`).
[0403] The terminals in the following paragraph grammar lead to
LAGs, with one for the parallel description sentence described
above directed to the inventive system's knowledge level.
1. Setting_description(Loc,pov,Person)/pc_of
(Person,Loc)+parallel_dscrp(Loc,pov,Person)
[0404] A brief sample story generated from the simple grammars just
described and the theme and stage produced earlier in our example
follows.
[0405] Dave loves the University of Rome. He loves its studious
youth, ivy-covered clocktowers and its sturdy brick. Dave wanted to
graduate. Prof. Hart told Dave, "I will sign your thesis at your
defense." Prof. Hart actually intends to thwart Dave's plans to
graduate. After Dave completed his defense and the chairman of
Dave's committee asked Prof. Hart to sign Dave's thesis, Prof. Hart
refused to sign. Dave was crushed.
[0406] The brief examples here taken from the inventive system's
grammar hierarchy illustrate how story grammars can drive a variety
of story structures that lead to paragraphs and sentence
structures, which are in turn tied directly to domain entities and
literary objectives through LAGs. Significant storytelling and
literary variability can be achieved by altering, adding, or
selecting different story, paragraph, or LAGs. Content variability
can be achieved by creating or modifying thematic relations,
behavioral rules, and domain knowledge.
[0407] As described above, the betrayal (theme) relation and frame,
and the process by which the inventive system generates stories,
starting with an interesting theme, are described. Variations can
be realized by adjusting the thematic relation to describe
different outcomes.
[0408] For example, while the betrayal relation previously
described allows for success or failure of the betrayer's goal, a
betrayal frame for generating specialized stories about betrayal
where the betrayer necessarily succeeds can be built, as shown
below. The thematic instantiation process would ensure that the
stories generated will include only events such that the betrayer
will succeed in thwarting the betrayed. The instantiation of this
type of betrayal might entail associated variations in plot, story
structure, and language.
[0409] A similar approach can be taken to produce a variety of
specializations. For example, cases where the betrayer's goal is
considered altruistic by a general audience but taken as an offense
by the betrayed. This is a case that may apply to a parent
betraying a child's trust to ultimately save the child from a
tragic end (or similarly Brutus' murder of Julius Caesar could be
viewed as an altruistic/patriotic act to save the Roman Empire).
Many variations may be achieved by adding or relaxing constraints
in the thematic relation. [0410] relation successful(Betrayal)
[0411] if Betrayer is the betrayer of the Betrayal [0412] and
Betrayers_Evil_Goal is the goal of the Betrayer [0413] and
Condition1 is the success of Betrayor_Evil_Goal [0414] and
Condition1 [0415] and Betrayed is the betrayed or Betrayal [0416]
and G is the goal of the Betrayed [0417] and Condition2 is the
success of C [0418] and not (Condition2) [0419] and RName is the
name of the Betrayed [0420] and DName is the name of the Betrayor
[0421] and !.
[0422] Thus, as described above, the system of the present
invention employs a plurality of processes in story generation. For
example, the invention uses thematic instantiation, which is the
process by which a theme produces a stage. A stage links a theme to
a specific set of domain entities that are sufficient to realize
the theme in some artifact in some language.
[0423] In the exemplary system developed for story generation, a
theme is expressed as a logical formula (e.g., in first order logic
translated into the processing language of choice such as Prolog in
the case at hand). That is, "interestingness" is formalized.
Variables in the above formula are instantiated with elements from
the domain knowledge base. The result is a collection of key
ingredients and their initial states called a "stage".
[0424] Additionally, the invention uses the process of Stage
Evolution 102A (e.g., see FIG. 2B for the generalized view of the
invention for generating an artifact). The process of stage
evolution is used to achieve width variability and creative
distance from an initial input. Stage evolution takes the static
entities and their behaviors composing a stage and extends the
entities' interaction through time producing new entities and
events. Stage evolution may be implemented using simulation and
planning techniques. In story generation, stage evolution may be
specialized as plot development. The result of the process is a
detailed plot or scenario.
[0425] Additionally, the invention uses structural expansion, which
uses compositional knowledge and a sequence of compositional
sentences. These sentences are independent of scenario and
language. This enables the easy use and reuse of different story
structures in combination with different themes, plots, literary
devices or languages. They form a structural template that
describes the surface structure of an artifact.
[0426] In story generation, for example, structural expansion
produces a detailed story structure that describes which parts of a
story will be told when. For example, the protagonist will be
introduced, then the setting will be described, then the
antagonist's physical appearance will be described, then . . .
etc.
[0427] Additionally, the invention incorporates language
generation, which is the process that uses the lexical knowledge to
produce grammatically correct sentences in the language L.
[0428] Further, the invention employs impressionistic shaping 107
(e.g., see FIGS. 2B and 2C), which is the process of generating
expressions in L or C to satisfy some set of impressionistic
objectives described in impressionistic knowledge sources.
[0429] The present invention employs a man-machine interface such
as a Graphical User Interface (GUI). The GUI can be a low-level
which can monitor and display, for example, the next result of the
system and which prompts the user by telling the user what the
system is going to do and please confirm (e.g., "I am going to
select John Doe for the betrayer. Does the user want to change this
selection?"). Alternatively, the GUI could allow higher level
control by asking: "Please select a theme", "please select a
character", "please select a character's property (ies)", "please
select how a level of psychological consciousness of the following
characters", "please select a story grabber", etc.
[0430] The man-machine interface allows for control of the creative
process. While changing the contents of the different knowledge
sources will effect different artifacts, the system 100 of the
present invention identifies specific interface points that give
the human user an opportunity to interactively control and guide
the creative agent's process and influence the resultant artifact.
Thus, for example, such interface points may be in the process
steps, in the databases, etc. such that the plot elements or stage
elements etc. can be anchored. By the same token, if such elements
are anchored or changed which diverge from the requirements of the
system or the theme, then a different result may occur different
from what the system would have allowed.
[0431] Basically, any choice made by the system (e.g., choosing an
event, choosing a plot, choosing a sentence, choosing a story
grammar, choosing a word, choosing a character, choosing a
character's property, etc.) may be made by the user. The interface
points are realized by variables (places where certain elements
(examples enumerated above) are initially left unspecified and
whose range of possible values become more and more limited as the
story generation process progress and choices for these variables
are made incrementally--either by the system's random processes or
by user's choices).
[0432] Specifically, if the FLEX representation of the theme is
taken as an example, the capital letters are variables. As the
system makes choices based on random processes or user input these
variables are assigned values from the knowlege-base. Once an
assignment is made, the variable's value is fixed (for a single
pass through the system) and the value will represent a fixed
element of the story. Any time a variable is encountered the system
can prompt the user to make a choices from a list of possible
values derived from the appropirate database.
[0433] Thus, the invention maintains a creative distance between
input and output. In contrast, in the conventional systems for
example, a machine turning an English story into one in German is
not creative. Likewise, if the machines was given a list of
characters, settings, plot, etc., which had to be used or merely
changed the characters etc. for one another, then the machine would
not be creative if such lists had to be used and there was a direct
mapping from input to output. Instead, "creativeness" is provided
by the invention since a disjoint, randomness occurs in selecting
and containing various characters, etc. from various databases,
while keeping the theme anchored thereto.
[0434] Thus, to provide creativeness, the invention merely needs to
be provided a theme (e.g., betrayal, romantic love, ambition,
jealousy, hatred, defined therefor and then the system can provide
the story including the characters etc. after the theme is input.
The them is the core to the story generation system, and the system
makes random choices as appropriate and constraint-based
choices.
[0435] Thus, if a betrayal theme is selected, then the fact that
betrayal was selected may in turn limit the system in terms of
which specific characters (or plots, or settings, etc. having
different characteristics) in, for example, a character database
could be selected by a random generator to achieve variability and
randomness. Thus, the system knows which characters are "good" for
being a betrayer in a betrayal theme story.
[0436] Likewise, the system can select certain words, phrases, etc.
based on the theme, as literary devices to evoke certain emotions
from the reader. Again, with a betrayer theme, a betrayer must be
selected having certain characteristics selfish, jealous, ambitious
etc. to narrow down the choices of characters. However, if a
character is randomly selected, then one may end up with a
psychologically conscious" character (as opposed to one that is
not) will constrain the words, phrases, adjectives, and adverbs
used.
[0437] Hence, in the story of Julius Caesar's assassination, Brutus
might be described with words, phrase, adjectives, adverbs, etc.
such that the reader believes Brutus was "psychologically
conscious" of what he was doing in killing Caesar. For example,
Brutus might be described as "The inventive system thought . . . "
as opposed to "Brutus mindlessly knifed Caesar . . . ". By the same
token, the system also knows that other characters which are not to
change or be constrained.
[0438] As shown in FIG. 4, a typical hardware configuration of an
information handling/computer system 400 in accordance with the
invention preferably has at least one processor or central
processing unit (CPU) 411. The CPUs 411 are interconnected via a
system bus 412 to a random access memory (RAM) 414, read-only
memory (ROM) 416, input/output (I/O) adapter 418 (for connecting
peripheral devices such as disk units 421 and tape drives 440 to
the bus 412), user interface adapter 422 (for connecting a keyboard
424, mouse 426, speaker 428, microphone 432, and/or other user
interface device to the bus 412), communication adapter 441 (for
connecting an information handling system to a data processing
network), and display adapter 436 (for connecting the bus 412 to a
display device 438).
[0439] As shown in FIG. 5, in addition to the hardware and process
environment described above, a different aspect of the invention
includes a computer-implemented method for performing a document
component importation and reconciliation, as described above. As an
example, this method may be implemented in the particular hardware
environment discussed above.
[0440] Such a method may be implemented, for example, by operating
the CPU 411 (FIG. 4), to execute a sequence of machine-readable
instructions. These instructions may reside in various types of
signal-bearing media.
[0441] Thus, this aspect of the present invention is directed to a
programmed product, comprising signal-bearing media tangibly
embodying a program of machine-readable instructions executable by
a digital data processor incorporating the CPU 411 and hardware
above, to perform a document component importation and
reconciliation.
[0442] This signal-bearing media may include, for example, a RAM
(not shown) contained within the CPU 411, as represented by the
fast-access storage for example. Alternatively, the instructions
may be contained in another signal-bearing media, such as a
magnetic data storage diskette 500 (FIG. 5), directly or indirectly
accessible by the CPU 411.
[0443] Whether contained in the diskette 500, the computer/CPU 411,
or elsewhere, the instructions may be stored on a variety of
machine-readable data storage media, such as DASD storage (e.g., a
conventional "hard drive" or a RAID array), magnetic tape,
electronic read-only memory (e.g., ROM, EPROM, or EEPROM), an
optical storage device (e.g. CD-ROM, WORM, DVD, digital optical
tape, etc.), paper "punch" cards, or other suitable signal-bearing
media including transmission media such as digital and analog and
communication links and wireless. In an illustrative embodiment of
the invention, the machine-readable instructions may comprise
software object code, compiled from a language such as "C",
etc.
[0444] While the invention has been described in terms of a
preferred embodiment, those skilled in the art will recognize that
the invention can be practiced with modification within the spirit
and scope of the appended claims.
[0445] With the unique and unobvious features of the invention, the
method and system of the present invention provide an architecture
for computational creativity captures an explicit model of the
creative process in terms of computational processes and formally
encoded knowledge. It demonstrates how a formally represented theme
can be used to seed and maintain interestingness as an artifact
develops through different phases of the creative process.
[0446] It is noted that the present invention is not limited to
story generation. For example, music generation would find great
benefit from the present invention. Moreover, there are many other
possible applications (including other artistic forms given the
appropriate effectors and "lexicon" sculpture, painting, cooking
etc.). The system and method of the present invention, specialized
for story generation as described in above, produces a unique and
more robust architecture than is present in the conventional system
and methods. It is believed that this is also true for music as
well as other languages of human expression.
[0447] The system architecture for computational creativity
described above can revolutionize how computers and computation are
applied to human endeavors. It is the foundation for engineering
creative agents that generate interesting stories, scripts,
adventure games, musical compositions, recipes, paintings,
sculptures etc.
[0448] While the application of the inventive system and method to
each of these areas requires engineering knowledge bases of
considerable expertise, the present invention provides a
demonstrable approach to building effective computational engines.
These engines, termed the "creative agent", can pursue various
humanistic themes, explore wide varieties of alternative scenarios
and build expressions in humanistic languages while applying
impressionistic knowledge to craft that language to best impact
human audiences. Creative agents, as the functional core of many
products in entertainment, education and business analysis, may
have great benefit as creative tools.
* * * * *