Available online at www.sciencedirect.

com

ScienceDirect
Cognitive Systems Research 58 (2019) 342–350
www.elsevier.com/locate/cogsys

Narrative structure in the mind: Translating Genette’s

narrative discourse theory into a cognitive system

Taisuke Akimoto
Kyushu Institute of Technology, 680-4 Kawazu, Iizuka, Fukuoka 820-8502, Japan

Received 14 May 2019; accepted 17 August 2019

Available online 22 August 2019

Abstract

An episodic memory is considered to be mentally encoded or constructed information instead of a copy of the past events. However,
the mechanism that constructs episodic memory has not been systematically formulated in previous studies on cognitive architecture and
system. In this study, the term ‘‘story” is used, rather than episodic memory, to refer to a mental representation of temporally and lin-
guistically organized events and entities. The main difference between a story and an episodic memory is that a story involves a form of
mental representation, while the definition of an episodic memory is generally based on the function or content (i.e., it enables the rec-
ollection of past events). A story is considered as a uniform mental representation involving episodic memory, current situation, prospec-
tive memory, planned or imagined future, and fiction. To ensure systematic formulation of the story-form memory construction, this
study introduces Genette’s narrative discourse theory, which is a representative work in narratology. Genette provided a systematic clas-
sification of terms used to describe a narrative structure, with particular focus on how a narrative is structured on a text instead of what is
told. In this study, Genette’s narratological terms are analogically translated to methods of story-form memory construction in a cog-
nitive system. Furthermore, the narratological methods are rearranged using cognitive terms including temporal segmentation of the
world, generalization of stories, theory of mind, metacognition, self-formation, memory organization, and sociocultural aspect of cog-
nition. Thus, a story is characterized as integrative, multifaceted, and multifunctional information in the mind.
Ó 2019 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/
licenses/by-nc-nd/4.0/).

Keywords: Story; Memory; Narratology; Genette; Cognitive system

1. Introduction systematically formulated in previous studies on cognitive

In psychological studies on memory, an episodic mem-
ory is generally defined as a memory that enables the recol-
lection or remembrance of past events or experiences
(Tulving, 1983, 2002). It is not assumed as a copy of the
past events themselves and is instead assumed as corre-
sponding to mentally encoded or constructed information
including reinterpretation, sensemaking, abstraction, gen-
eralization, errors, and forgetting. However, the mecha-
nism that constructs episodic memory has not been
E-mail address: akimoto@ai.kyutech.ac.jp
architecture and system. For example, in the Soar cognitive
architecture, an episodic memory is considered as sequen-
tial snapshots of the agent’s working memory (Nuxoll &
Laird, 2004, 2012). Faltersack, Burns, Nuxoll, and
Crenshaw (2011) presented an episodic memory structure
that involves a hierarchy from lower primitive elements
to higher compound elements. In the ICARUS cognitive
architecture, an episodic memory is encoded via generaliza-
tion based on similarity with previous episodes (Ménager &
Choi, 2016). The aforementioned episodic memories
involve only few linguistic aspects. On the other hand,
Bölöni (2011) presented an episodic memory model includ-

https://doi.org/10.1016/j.cogsys.2019.08.007
1389-0417/Ó 2019 The Author(s). Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
 T. Akimoto / Cognitive Systems Research 58 (2019) 342–350
ing conceptual information. Recently, several studies
focused on narrative structure in an episodic memory.
Anderson (2015) discussed the mental processes of con-
structing an episodic memory including temporal segmen-
tation of event and creating a relationship between
events. León (2016) formalized the relational structure of
a narrative memory that connects narrative objects in
terms of kernels and satellites. Although the aforemen-
tioned studies captured a partial aspect of episodic mem-
ory, it is necessary to explore a unified computational
theory of episodic memory construction.
In this study, the term ‘‘story” is used, rather than epi-
sodic memory, to refer to a mental representation of tem-
porally and linguistically organized events and entities.
The main difference between a story and an episodic mem-
ory is that a story involves the form of the mental represen-
tation, while the definition of an episodic memory is
generally based on the function or content (i.e., it enables
the recollection of past events). As described in Akimoto
(2018), a story is considered as a uniform mental represen-
tation involving episodic memory, autobiographical mem-
ory, current situation, prospective memory, planned or
imagined future, and fictional or hypothetical story. They
include long-term, short-term, and working memories.
Hence, the generative cognition of story constitutes a com-
mon basis for a cognitive system or an integrative autono-
mous artificial intelligence. The importance of story
cognition in artificial intelligence was explored by Schank
(1982, 1990). Winston (2012)’s strong story hypothesis also
argues for the generality of story in an intelligence.
In order to ensure a systematic formulation of the story-
form memory construction, this study introduces Genette
(1980)’s narrative discourse theory. It is a representative
work in narratology (which is a discipline that examines
fundamental structures, principles, and properties of narra-
tives). The theoretical background of narratology includes
structuralism, semiology, Russian formalism, linguistics,
and literary studies. Narratological theories are applied in
artificial intelligence studies on narrative generation
(Gervás, Lönneker-Rodman, Meister, & Peinado, 2006;
Lönneker, 2005; Ogata, 2016), interactive narrative
(Montfort, 2007), narrative analysis (Mani, 2013), and cog-
nitive architecture (Anderson, 2015; León, 2016; Akimoto,
2018; Szilas, 2015; Samsonovich & Aha, 2015).
Genette’s narrative discourse theory is characterized by
its systematicity. He provided a hierarchical classification
of terms to describe a narrative structure, with particular
focus on how a narrative is structured on a text instead
of what is told. The aim of this study involves formulating
the methods of story-form memory construction via ana-
logically applying Genette’s systematic theory. Further-
more, the narratological methods are rearranged using
cognitive terms.
The rest of the paper is organized as follows: Section 2
provides an overview of Genette’s narrative discourse the-
ory including the analytical framework and structural cat-
egorization. In Section 3, Genette’s analytical framework is
343
translated into the structure of a cognitive system. In Sec-
tion 4, Genette’s categories are reinterpreted as methods
of story-form memory construction. Section 5 rearranges
narratological methods using cognitive terms. Finally, Sec-
tion 6 concludes the study.
2. Genette’s narrative discourse theory
This section provides a brief overview of Genette’s nar-
rative discourse theory (Genette, 1980). The theory was
developed based on analyzing a novel À la recherche du
temps perdu (In Search of Lost Time) written by Marcel
Proust. The basic analytical framework is explained in Sec-
tion 2.1. Subsequently, categorization of narrative dis-
course structure is described in Section 2.2.
2.1. Analytical framework
In order to deal with the structural analysis of a narra-
tive, Genette distinguished a narrative into three aspects,
namely discourse (narrative), story, and narrating. A dis-
course refers to the text of a novel or narrative itself. A
story refers to the content information recounted in the dis-
course. Thus, it contains chronologically organized events.
Narrating refers to the action of producing a narrative by a
narrator(s) and narratee(s). In this context, a narrator and
narratee correspond to persons inscribed in the narrative
text and not the author and reader of a narrative. In the
aforementioned three aspects, only a discourse exhibits a
materiality while a story and narrating are interpreted from
the discourse. Hence, the object of the analysis corresponds
to a discourse.
The categorization of structural properties of a narrative
discourse is constructed based on the relationship among
the discourse, story, and narrating. Specifically, it consists
of the following three broad categories (see also Fig. 1):
tense refers to the relationships between the temporal
aspects of a discourse and story; mood refers to the modal-
ities of expressing the story in the discourse; and voice
refers to the situation of narrating, in the relationships with
the discourse and story.
2.2. Categorization of narrative discourse structure
Genette established hierarchically organized terms to
describe a narrative structure. The terms are arranged as
Narrating
Voice
Story Discourse
Tense (narrative)
Mood
Fig. 1. Analytical framework of the narrative discourse theory.
344 T. Akimoto / Cognitive Systems Research 58 (2019) 342–350
subcategories of the aforementioned three broad cate-
gories. The hierarchical categorization of structural terms
and their brief explanations are provided below (see also
Prince (2003) for more strict definitions of the narratolog-
ical terminologies):

Tense: Relationships between the temporal aspects of a
discourse and story.
– Order (anachronies): Relationships between the
chronological order of events in a story and the order
in which the events are recounted in a discourse.
There are two major subcategories as follows:
* Analepsis: Going back to past events from a pre-
sent temporal position via a flashback, recollec-
tion, or other methods.
* Prolepsis: Going forward to future events from a
present temporal position via a flashforward, pre-
diction, or other methods.
– Duration (anisochronies): Relationships between the
length of a discourse (for e.g., words, lines, and
pages) and amount of time of the recounted events
(for e.g., seconds, minutes, hours, days, months,
and years).
* Summary: A sequence of events is briefly
recounted via relatively short text.
* Scene: A sequence of events is recounted in detail
via relatively long text.
* Pause: The story’s temporal progress is stopped
in a discourse via the narrator’s commentaries or
descriptions of objects, things, or situations.
* Ellipsis: A certain time period in a story is explic-
itly or implicitly omitted in the discourse (for e.g.,
‘‘some years passed”).
– Frequency: Relationships of frequency between the
occurrences of an event(s) in a story and recounted
number in a discourse.
* Singulative narrative: An event is recounted once,
i.e., a normal narrative.
* Repeating narrative: The same event is recounted
two or more times.
* Iterative narrative: Similar events are recounted
once (e.g., ‘‘every day of the week I went to bed
early”).

Mood: Modalities of expressing a story in the discourse.
– Distance: A mode of regulating narrated information
in terms of the degree of the narrator’s mediation.
The notion can be understood as a contrast between
mimesis or showing (smaller distance) and diegesis or
telling (larger distance).
– Focalization: A mode of regulating narrated informa-
tion based on a choice or not of a restrictive perspec-
tive. It should be noted that Genette distinguished a
perspective from a narrator because a narrator can
narrate a story via considering another character’s
perspective. The three basic types of focalization are
listed below.
* Zero focalization: A nonfocalized discourse with-
out information regulation based on a specific
perspective.
* Internal focalization: A discourse is composed
from a character’s perspective including the char-
acter’s mental actions.
* External focalization: A discourse presents only
the external behavior of characters (i.e., charac-
ters’ mental actions are not recounted).

Voice: Situations of narrating in the relationship with
the discourse and story.
– Time of the narrating: Relationships between the tem-
poral positions of narrating and the narrated story.
* Subsequent: A past-tense narrative.
* Prior: A predictive narrative (generally presented
in the future tense).
* Simultaneous: A present-tense narrative.
* Interpolated: Temporal position of the narrating
is interpolated between the moments of action.
– Narrative level: The notion can be understood as the
nesting structure of narrating and is explained as ‘‘a
narrative narrated within a narrative.” A first-level
narrative is produced in its external level (extradie-
getic position).
– Person: Relationships between a narrator and charac-
ters in the story.
* First-person: A narrator appears in the story as a
(central or secondary) character.
* Third-person: A narrator does not appear in the
story.
3. Translating the analytical framework into a cognitive
system
This section translates the relationship among discourse,
story, and narrating into a cognitive system. Fig. 2 illus-
trates an overview of the framework. First, the three
aspects of a narrative are replaced into representational
and procedural elements of a cognitive system: the story
corresponds to a mental representation containing infor-
mation of structured events; discourse corresponds to an
expressive structure of a narrative; and narrating corre-
sponds to the action of producing stories and discourses.
With respect to the aspect of narrating, the narrator essen-
tially corresponds to the agent itself. The narratee is
unspecified although it corresponds to an environment or
an objective that directs the production of a story and
discourse.
In the relationship with physical and social environ-
ments, a story that forms a temporal structure of a current
situation constitutes a foundation of a higher-level action–
perception system (Akimoto, 2018). Conversely, narrative-
communication with another person or agent is always
mediated via a discourse.
 T. Akimoto / Cognitive Systems Research 58 (2019) 342–350
Agent’s Cognitive System
Narrating
(production)
Story Discourse
(memory) (expressive
structure)
Action–Perception Narrative
communication
Physical and social environments
Fig. 2. Framework of story construction in a cognitive system.
Genette’s theory does not clarify whether a story refers
to events and things themselves (in an external world) or
a mental representation of events and things. In this study,
a story is clearly positioned as a mental representation.
Hence, it can be considered that a discourse as well as a
story is linguistically constructed. The aforementioned per-
spective blurs the boundary between a discourse and story.
However, the matter appears to constitute the essential nat-
ure of the mutual relationship between the memory (story)
and expression (discourse).
In the framework, a cyclic relationship exists between a
story and discourse, i.e., a story is reconstructed via struc-
turing a discourse based on that story. The cyclic relation-
ship involves the following three cognitive processes of
story construction: construction of a story via interaction
with the environment; reconstruction of a story via exter-
nally or internally narrating it; and construction of a story
from a narrative via another person or agent. Various
aspects of the discourse structure are reflected into a story
via the aforementioned cycle.
4. Translating the narratological categories into story
construction methods
In this section, the categories of narrative structure are
translated into methods of story construction in a cognitive
system. In dealing with the issue, we focus on processes
wherein a relatively raw (simple) story is reconstructed by
narrative methods. In order to provide a formal description
of the methods, we define a minimal representation formal-
ism of a memory containing stories.
A memory M denotes a tuple hS; T i, where S denotes a
set of stories fs1 ; . . . ; sn g and T denotes a set of times
ft1 ; . . . ; tm g. Time ti contains one or more stories:
½timefsi g. A story si is represented by a tuple hE; Ri, where
E denotes a sequence of events ðe1 ; . . . ; el Þ and R denotes a
set of relationships fr1 ; . . . ; rk g. An event ei represents a
345
character’s action or a stative information including
description and commentary.1 An event can contain a
sequence of two or more sub-events: ½ei ðe1 ; . . . ; ej Þ. A rela-
tionship ri relates two events: ½relationshipðep ; eq Þ. Table 1
shows an example of memory containing three simple sto-
ries. The example is subsequently used to explain story con-
struction methods.
In this study, the order is not applied to the construction
method for a story itself; instead, it exhibits a close rela-
tionship with a memory organization. Therefore, we start
with duration, and the relevance of temporal ordering to
a memory is discussed in the last subsection.
4.1. Duration
The four subcategories of duration (i.e., summary, scene,
pause, and ellipsis) are translated as follows.
Summary and scene correspond to a change in the level
of detail of a sequence of events. In summary, two or more
events are abstracted into a comprehensive event. For
example, s2 changes to ðe2:1 ; ½e2;x ðe2;2 ; e2;3 ÞÞ, where e2;x
(e.g., ‘‘We had coffee at a coffee shop.”) is newly generated
from e2;2 –e2;3 . If the sub-events (e2;2 –e2;3 ) are forgotten, then
s2 corresponds to ðe2;1 ; e2;x Þ.
In scene, an event is detailed into two or more sub-
events. For example, s2 changes to
ðe2;1 ; ½e2;2 ðe2;2a ; e2;2b ; e2;2c Þ; e2;3 Þ, where e2;2a –e2;2c (e.g. ‘‘We
walked to a coffee shop. We entered the coffee shop. We
ordered two cups of coffee and two shortcakes.”) corre-
spond to newly generated events from e2;2 .
Pause linguistically composes of a situational or scenery
information or commentary that is associated with an
event. For example, s1 changes to ðe1;1 ; e1;2 ; e1;2d ; e1;3 Þ where
e1;2d refers to a scenery description of the beach (e.g., ‘‘The
sea was sparkling in the morning sun.”).
Ellipsis can be interpreted as a choice of information that is
discarded from a story. For example, if e2;2 –e2;3 in s2 are dis-
carded (consciously or unconsciously), then the story short-
ens to ðe2;1 Þ, ‘‘Yesterday, I met an old friend at a park.”
4.2. Frequency
Only iterative narrating can be translated into a method
of memory construction from the subcategories of
frequency.2 In this method, two or more similar stories or
events are compressed into a single story. For example,
s1 and s3 can be transformed into a compressed story
s4 , ‘‘(Every morning recently) e4;1 : I wake up early. e4;2 : I
take a walk. e4;3 : I eat breakfast.” The story is contained
1
In narratology, a description of situational or scenery information and
a commentary by a narrator is distinguished from an event. However, in
this formalism, they are treated as the same type of representational
element for purposes of simplicity.
2
Singulative narrative is a normal form of story construction. The act of
repeating can be considered as a memory rehearsal although it does not
produce a copy of an event into a story.
346 T. Akimoto / Cognitive Systems Research 58 (2019) 342–350
Table 1
Example memory.
M ¼ fSfs1 ; s2 ; s3 g; T ft1 ; t2 gg
s1 ¼ fEðe1;1 ; e1;2 ; e1;3 Þ; Rfgg
s2 ¼ fEðe2;1 ; e2;2 ; e2;3 Þ; Rfgg
s3 ¼ fEðe3;1 ; e3;2 ; e3;3 Þ; Rfgg
t1 ¼ ½yesterday fs1 ; s2 g
t2 ¼ ½today fs3 g
Contents of stories (in English):
s1 : ‘‘(Yesterday) e1;1 : I woke up early in the morning. e1;2 : I took a walk at a beach. e1;3 : I ate breakfast.”
s2 : ‘‘(Yesterday) e2;1 : I met an old friend at a park. e2;2 : We went to a coffee shop. e2;3 : We chatted over our coffee.”
s3 : ‘‘(Today) e3;1 : I woke up early in the morning. e3;2 : I took a walk in a park. e3;3 : I ate breakfast.”
in a new time t3 : ½every-morning-recentlyfs4 g. The origi-
nal stories can be forgotten, and in this case
M ¼ fSfs2 ; s4 g; T ft1 ; t3 gg.
4.3. Distance
Example stories in Table 1 contains only a sequence of
simple events without relational information. Given that
the stories involve only a few mediations of the narrator
and that they appear to constitute showing as opposed to
telling, it is assumed that the stories are created from a rel-
atively small distance. A basic method to increase the dis-
tance involves adding a relationship between story
elements because a relationship is considered as an imma-
terial object created by a narrator’s mentality. Conversely,
decreases in the relational information from a story
decrease the distance.
We provide several examples of the idea. If a causality
between events is added into s1 , then it changes to
fEðe1;1 ; e1;2 ; e1;3 Þ; Rfr1 ½reasonðe1;1 ; e1;2 Þgg, which implies
‘‘e1;1 : Because I woke up early in the morning, e1;2 : I took
a walk at a beach. e1;3 : I ate breakfast.”. A creation
of a new event to provide a reason for an action
increases the distance to s1 , e.g., fEðe1;1 ; e1;2r ; e1;2 ; e1;3 Þ;
Rfr1 ½reason ðe1;2r ; e1;2 Þgg which implies for example,
‘‘. . .e1;2r : Because the morning sun was calling me, e1;2 : I took
a walk at a beach. . . .”.
Another method to increase the distance is to add a
commentary about a story element (e.g., ‘‘. . .e1;2 : I took a
walk at a beach. e1;2c : I believe that walking is good for
mental health. . . .”). In this example, the narrator’s value
judgment is added into the story.
4.4. Focalization
In a story construction process, focalization works as a
perspective-based regulation of information to be con-
tained in the story. In an agent’s cognitive system, a natural
focalization considers the internal focalization of the agent
itself. Hence, an important issue in focalization involves
taking a different perspective from the agent’s internal per-
spective. It generates a different version of a story. For
example, a different version of s2 (i.e., s20 ) is produced via
simulating the internal focalization of the old friend, e.g.,
‘‘e20 ;1 : An old friend met me at a park. e20 ;2 : He felt nostalgic
about me. . . .”. A perspective is distinguished from the nar-
rator, and thus the first-person character does not change
in this version.
4.5. Voice
In Genette’s theory, voice considers the aspect of narrat-
ing and not the structure of a discourse and the story itself.
As shown in Fig. 3, subcategories of voice (i.e., narrative
level, time of the narrating, and person) can be translated
into a set of elements to form a meta-story structure via
interpreting the notion as background conditions in the
cognitive action of constructing a story.
Narrative level corresponds to an essential aspect of the
meta-story structure to create a division between a story
and cognitive action to produce the story. It is represented
via a nesting structure and a first-level story is produced in
its external (extradiegetic) level by the narrator.
Time of the narrating corresponds to the temporal rela-
tionship between a story’s temporal position and present
time of producing or remembering the story. The informa-
tion is essentially reflected in the linguistic tense (i.e., past,
future, or present) when a story is expressed as a narrative.
Person corresponds to whether the narrator appears in a
story as a character. In an agent’s cognitive system, the
basic narrator corresponds to the agent itself. However,
another person or agent (as a character in a story) can
become the narrator of a second-level story. For example,
if s20 (the friend’s internal focalization presented in Sec-
tion 4.4) is nested as the friend’s story, then it changes to
‘‘I met K at a park. I felt nostalgic about K.   ” (K refers
to the original ‘‘me”).
In order to represent the meta-story structure, story for-
malism is expanded into si ¼ ½vðsi Þ. The term v denotes a
tuple hnarrator; narTime; leveli, where narrator indicates
the narrator by its identical symbol, narTime corresponds
to a temporal position of si relative to the time of the nar-
rating (e.g., ‘‘past,” ‘‘present,” or ‘‘future”), and level indi-
cates the extradiegetic level of si (e.g., it corresponds to
‘‘zero” when si denotes a first-level story, and ‘‘first: sx ”
when si denotes a second-level story contained in sx ). The
meta-story structure is dynamically generated when a story
is remembered or focused.
 T. Akimoto / Cognitive Systems Research 58 (2019) 342–350
Person
(appearance in the story or not)
Self Self
(as narrator) (as character)
Time
Time of the narrating
Fig. 3. Meta-story structure including narrative level, time of the narrating, and person.
4.6. Order
Events in a story are chronologically organized, and
thus the temporal order of events can be essentially manip-
ulated only in the discourse dimension. Hence, we do not
apply temporal ordering to a story construction method.
However, the mechanism of forming a connection between
events or stories over the chronological order and a tempo-
ral gap is relevant to the organization of stories in a mem-
ory. Therefore, we provide a short consideration on a
structure underlying temporal ordering in a discourse.
We use only one example that associates s2 with an event
in s3 via a flashback: ðe3;1 ; e3;2 ; ðe2;1 ; e2;2 ; e2;3 Þ; e3;3 Þ. The
structure is expressed as ‘‘Today, e3;1 : I woke up early in
the morning. e3;2 : I took a walk in a park. (Yesterday,
e2;1 : I met an old friend at the park. e2;2 : We went to a coffee
shop. e2;3 : We chatted over our coffee.) e3;3 : I ate break-
fast.”. The structure can be explained as an association
mediated by the ‘‘park.” It is considered that association
among stories in the human mind is mediated by various
types of similarities and relatedness including semantic,
structural, non-verbal, and instance-level. In order to gen-
erate these types of phenomena, it is necessary to organi-
cally organize stories in a memory. The issue is relevant
to memory retrieval including index- and similarity-based
approaches (Schank, 1982; Kolodner, 1983; Thagard,
Holyoak, Nelson, & Gochfeld, 1990; Forbus, Gentner, &
Law, 1994), and our initial idea about story association is
presented in Akimoto (2019).
4.7. Summary of story construction methods
Story construction methods including meta-story struc-
ture and temporal order as derived from Genette’s theory
are summarized in Table 2. Fig. 4 lists their graphical
explanations excluding the meta-story structure.
347
First-level story
Second-level story
(a mental story about
another person or
agent.)
Another
person or agent
Time
5. Rearrangement in cognitive terms
The aforementioned narrative methods and structures
are relevant to various cognitive issues and are categorized
as follows:
Temporal segmentation and abstraction. Stories are con-
structed in different degrees of duration between scene and
summary. The phenomenon arises from flexibility in
abstracting events with different densities of temporal seg-
ments of the world.
Generalization. Similar stories are compressed into a
story via iterative narrating. The phenomenon corresponds
to generalization that captures a shared information or
structure among stories.
Non-verbal memory. Processes of story construction
include verbalization of non-verbal memory such as detail-
ing events via remembering an experience by scene and
constructing a scenery description accompanying an event
by pause. In order to enable this type of cognition, it is nec-
essary for non-verbal (or unverbalized) memory to be asso-
ciated with linguistically structured stories.
Memory organization. Temporal ordering of events
including remembering a story over the chronological
order and temporal gap is underpinned via an organic
memory organization.
Sensemaking and relating. Stories are constructed in dif-
ferent degrees of distance. The process of increasing a
story’s distance via the creation of relationships among
story elements is similar to subjective sensemaking.
Theory of mind and point of view. In an agent’s cognitive
system, a story is essentially constructed from its own per-
spective. However, different versions of stories are gener-
ated via simulating another individual’s internal
focalization (perspective). The phenomenon is relevant to
the theory of mind, i.e., the ability to imagine another indi-
vidual’s mental state.
348 T. Akimoto / Cognitive Systems Research 58 (2019) 342–350

Table 2
Summary of story construction methods.
Narratological terms Operation or structure
Summary Generating a comprehensive event from two or more events
Scene Generating two or more detailed events from an event
Pause Adding a situational or scenery description or commentary associated with an event
Ellipsis Discarding part of a story
Iterative narrating Compressing two or more similar stories or events into a single story
Distance (increase) Adding a relational information to a story
Distance (decrease) Removing a relational information from a story
(Internal) Focalization Generating a different version of a story via simulating another individual’s perspective
Narrative level* Division between a story and its production via a nesting structure
Time of the narrating* Relationship between the time of a story and time of producing or remembering the story
Person* Whether the narrator appears in a story as a character
Order** Associating a part of a story with another part or story over the chronological order and a temporal gap
*
Meta-story structure.
**
Association among stories.

Summary Scene Pause Ellipsis

e e
e e e e e e
e e e e e e d

Iterative narrating Distance Focalization

r r
s
s
e e e
s s s c s’

Order (as association among stories)

s: Story
s s e: Event
d: Description of situational or scenery information
c: Commentary
s
s r: Relationship

Fig. 4. Graphical explanations of story construction methods derived from Genette’s narrative discourse theory.

Stories Discourses Stories

as as as
mental information social information mental information
communicated among
individuals

Mind Society Mind

(cognitive system) (social systems) (cognitive system)

Fig. 5. Narrative-based interpenetration between an agent’s cognitive system and a society.

 T. Akimoto / Cognitive Systems Research 58 (2019) 342–350
Metacognition and consciousness. The self as narrator
(narrator-self) and self as character (character-self) are dis-
tinguished based on meta-story structure derived from
voice. The distinction constitutes the basis for structural
modeling of metacognition, namely cognition about one’s
own cognition. This structure may also lead to the organi-
zational understanding of consciousness.
Self-formation. In psychological and philosophical stud-
ies, narrative and autobiographical memory are often
regarded as the foundation of self-formation. The pro-
posed narrative methods and structures, particularly voice-
based metacognition and sensemaking, will be the basis for
computational modeling of self-formation.
Sociocultural aspect of cognition. Narratives (discourses)
consist of social information communicated in various
social fields such as education, art, mass and social media,
and family. Styles of narrative composition are developed
by an individual mind and societies via co-creative commu-
nication among individuals. Therefore, the proposed the-
ory involves a sociocultural aspect of cognition. In the
proposed framework (see Fig. 2), the relationship between
stories and discourses is formulated as a cyclic relationship.
Based on this structure, the interaction between an individ-
ual mind (cognitive system) and societies (social systems)
can be understood as an interpenetration built on narra-
tives, as illustrated in Fig. 5. This system view on societies
and the relationship between societies and minds is rooted
in Luhmann’s social systems theory (Luhmann, 2012–
2013).3
Computational modeling of generative cognition of sto-
ries exhibits potential to lead to an integrative view on the
collaboration of the aforementioned different cognitive
aspects.
6. Conclusion
In search of a unified computational theory of story-
form memory construction, this study analogically trans-
lated Genette’s narrative discourse theory into a cognitive
system. First, the three basics of a narrative, i.e., discourse,
story, and narrating, are replaced by representational and
procedural elements in a cognitive system. Second, story
construction methods and meta-story structures are formu-
3 fall symposium (pp. 42–49).
Luhmann (2012–2013) incorporated the autopoiesis theory, which was
originally developed in the field of biology by Maturana and Varela
(1980), into a theory for describing modern societies. In Luhmann’s social
systems theory, a person and the society are divided into different
autonomous systems, i.e., mental and social systems. The modern society
is characterized as a complex interrelationship among functionally
differentiated subsystems including politics, economy, art, education,
science, religion, and family. The only element forming a social system is
communication, whereas the communication is basically conducted via
mental systems. A social system autonomously operates through recursive
chains of communication in which communication produces new com-
munication. Each of the social and mental systems is operationally closed,
however, it incorporates stimulations from other systems (i.e., environ-
ments). Hence, mental and social systems are closely coupled in the form
of interpenetration.
349
lated based on Genette’s systematic categorization of nar-
rative structure. Further, the proposed narratological
methods and structures are interrelated with cognitive
issues, including temporal segmentation of the world, gen-
eralization of stories, theory of mind, point of view,
metacognition, self-formation, memory organization, and
sociocultural aspect of cognition. Thus, a story can be con-
sidered as integrative, multifaceted, and multifunctional
information in one’s mind. Hence, a story-centered
approach to a cognitive system exhibits the advantage of
capturing integrative working of the mind. Our proposed
theory acts as the foundation to develop a story-centered
cognitive system or architecture.
The proposed theory describes the general aspects of a
story structure in the mind. However, it does not provide
the cognitive process and knowledge underlying the
production and manipulation of stories. In computational
narrative generation studies, various approaches such as
goal-oriented planning, case-based reasoning, analogical
reasoning, schema-based top–down cognition, and
combination or transformation of existing stories have
been suggested. We consider that generative story
cognition is a combination of various cognitive processes,
including the ones mentioned above. Hence, in a future
study, we will explore fundamental principles for the
unification of generative story cognition.
Funding
This work was supported by JSPS KAKENHI Grant
No. JP18K18344.
References
Akimoto, T. (2018). Stories as mental representations of an agent’s
subjective world: A structural overview. Biologically Inspired Cognitive
Architectures, 25, 107–112. https://doi.org/10.1016/j.bica.2018.07.003.
Akimoto, T. (2019). Key issues for generative narrative cognition in a
cognitive system: Association and blending of stories. In Story-enabled
intelligence, AAAI 2019 spring symposium.
Anderson, T. S. (2015). From episodic memory to narrative in a cognitive
architecture. In Proceedings of 6th workshop on computational models
of narrative (pp. 2–11).
Bölöni, L. (2011). An investigation into the utility of episodic memory for
cognitive architectures. In Advances in cognitive systems, AAAI 2011
Faltersack, Z., Burns, B., Nuxoll, A., & Crenshaw, T. L. (2011). Ziggurat:
Steps toward a general episodic memory. In Advances in cognitive
systems, AAAI 2011 fall symposium (pp. 106–111).
Forbus, K. D., Gentner, D., & Law, K. (1994). MAC/FAC: A model of
similarity-based retrieval. Cognitive Science, 19, 141–205. https://doi.
org/10.1016/0364-0213(95)90016-0.
Genette, G. (1980). Narrative discourse: An essay in method. NY: Cornell
University Press (Trans. Jane E. Lewin. Original work published
1972).
Gervás, P., Lönneker-Rodman, B., Meister, J. C., & Peinado, F. (2006).
Narrative models: Narratology meets artificial intelligence. In Pro-
ceedings of satellite workshop: Toward computational models of literary
analysis. 5th international conference on language resources and
evaluation (pp. 44–51).
350 T. Akimoto / Cognitive Systems Research 58 (2019) 342–350
Kolodner, J. L. (1983). Maintaining organization in a dynamic long-term
memory. Cognitive Science, 7, 243–280. https://doi.org/10.1016/S0364-
0213(83)80001-9.
León, C. (2016). An architecture of narrative memory. Biologically
Inspired Cognitive Architectures, 16, 19–33. https://doi.org/10.1016/j.
bica.2016.04.002.
Lönneker, B. (2005). Narratological knowledge for natural language
generation. In Proceedings of the 10th European workshop on natural
language generation (pp. 91–100).
Luhmann, N. (2012–2013). Theory of society (vols. 1–2). CA: Stanford
University Press (Trans. R. Barrett. Original work published 1997).
Mani, I. (2013). Computational modeling of narrative. CA: Morgan &
Claypool.
Maturana, H. R., & Varela, F. J. (1980). Autopoiesis and cognition: The
realization of the living. MA: D. Reidel Publishing Company.
Ménager, D. H., & Choi, D. (2016). A robust implementation of episodic
memory for a cognitive architecture. In Proceedings of the 38th annual
conference of the cognitive science society (pp. 620–625).
Montfort, N. (2007). Generating narrative variation in interactive fiction
(Ph.D. thesis). University of Pennsylvania.
Nuxoll, A. M., & Laird, J. E. (2004). A cognitive model of episodic
memory integrated with a general cognitive architecture. In Proceed-
ings of international conference on cognitive modeling (pp. 220–225).
Nuxoll, A. M., & Laird, J. E. (2012). Enhancing intelligent agents with
episodic memory. Cognitive Systems Research, 17–18, 34–48. https://
doi.org/10.1016/j.cogsys.2011.10.002.
Ogata, T. (2016). Computational and cognitive approaches to narratology
from the perspective of narrative generation. In T. Ogata & T.
Akimoto (Eds.), Computational and cognitive approaches to narratology
(pp. 1–74). PA: IGI Global. https://doi.org/10.4018/978-1-5225-0432-
0.ch001.
Prince, G. (2003). A dictionary of narratology (revised ed.). NE: University
of Nebraska Press.
Samsonovich, A. V., & Aha, D. W. (2015). Character-oriented narrative
goal reasoning in autonomous actors. In Goal reasoning: papers from
the ACS workshop, technical report GT-IRIM-CR-2015-001
(pp. 166–181).
Schank, R. C. (1982). Dynamic memory: A theory of reminding and
learning in computers and people. NY: Cambridge University Press.
Schank, R. C. (1990). Tell me a story: A new look at real and artificial
memory. NY: Charles Scribner’s Sons.
Szilas, N. (2015). Towards narrative-based knowledge representation in
cognitive systems. In Proceedings of 6th workshop on computational
models of narrative (pp. 133–141).
Thagard, P., Holyoak, K. J., Nelson, G., & Gochfeld, D. (1990). Analog
retrieval by constraint satisfaction. Artificial Intelligence, 46, 259–310.
https://doi.org/10.1016/0004-3702(90)90018-U.
Tulving, E. (1983). Elements of episodic memory. Oxford: Oxford
University Press.
Tulving, E. (2002). Episodic memory: From mind to brain. Annual Review
of Psychology, 53, 1–25. https://doi.org/10.1146/annurev.
psych.53.100901.135114.
Winston, P. H. (2012). The right way. Advances in Cognitive Systems, 1,
23–36.