Gene ación de his o ias a pa i de un análisis
emá ico de simulaciones
S o y gene a ion using hema ic analysis o e a
simula ion
T abajo de Fin de G ado
Cu so 2023–2024
Au o
Pablo Cageao Hondu illa
Di ec o
Pablo Ge ás Gómez-Na a o
Gonzalo Rubén Méndez Pozo
G ado en Ingenie ía In o má ica
Facul ad de In o má ica
Uni e sidad Complu ense de Mad id
Gene ación de his o ias a pa i de un
análisis emá ico de simulaciones
S o y gene a ion using hema ic analysis
o e a simula ion
T abajo de Fin de G ado en Ingenie ía In o má ica
Au o
Pablo Cageao Hondu illa
Di ec o
Pablo Ge ás Gómez-Na a o
Gonzalo Rubén Méndez Pozo
Con oca o ia: Junio 2024
G ado en Ingenie ía In o má ica
Facul ad de In o má ica
Uni e sidad Complu ense de Mad id
27 de Mayo de 2024
Dedica o ia
A mi amilia y compañe os, po apoya me cuando en o en pánico po que no me
da iempo a en ega el g.

Ag adecimien os
G acias a mi amilia, po el apoyo y la comp ensión ecibidos du an e los úl imos
años de mi iempo en la Uni e sidad.
A Ped o Pablo y Ma co An onio, po c ea TeXiS e ilumina nues o camino.
G acias a los di ec o es de es e p oyec o, Gonzalo, po no en a en pánico cuando
se leyó la memo ia 2 días an es, y Pablo, po hace las p egun as co ec as que han
pe mi ido la ealización de es e p oyec o.
ii
Resumen
Gene ación de his o ias a pa i de un análisis emá-
ico de simulaciones
El es udio di ec o de los emas como una mé ica pa a alo a his o ias y como
una he amien a pa a cons ui las no ha enido mucho uso en el campo de la ge-
ne ación au omá ica de his o ias. El obje i o de es e abajo es diseña un modelo
emá ico an o como mé ica como sis ema pa a ex ae his o ias de un g an co pus
de e en os. Pa a log a es a a ea, se u iliza á una simulación pa a gene a los e en-
os, una implemen ación del modelo emá ico ex ae á his o ias in e esan es de él
y los algo i mos de combinación de sub amas jun o con una ex ensión del modelo
se u iliza án pa a cons ui his o ias más complejas y cub i las debilidades que el
modelo emá ico p esen a al ex ae his o ias.
Palab as cla e
Gene ación de his o ias, Combinación de sub amas, Análisis emá ico, Mé icas en
gene ación de his o ias, Simulación.
ix

Lis o ables
3.1. Example o idea acquisi ion om he ea u es o e en s . . . . . . . . 20
3.2. Example o idea acquisi ion om he ea u es o e en s a e he
change in oduced o sol e he p oblem in he ep esen a ion o hemes.
he hemes ep esen ed by *idea a e he hemes in oduced o sol e
hep oblem................................. 22
3.3. Possible desc ip ion o he e en s in able 3.2 om Bob’s poin o
iewins eado Alice’s........................... 23
3.4. 20 Mas e Plo s as p esen ed by Tobias (2012), able ex a ed om
(Ge áse al.,2015)............................ 27
x ii
Chap e 1
In oducción
Toda his o ia iene un ema, o una
p oposición subyacen e que indica su ipo. El
ema pod ía se e iden e y des acable, u
ocul o y ago; siemp e es á ep esen ado, y
di e encia un ipo de his o ía de odos los
demas. Sob e un ema se pueden con a
in inidad de his o ias
Cook (2011)
En la úl ima década, muchos a ances en el campo de la gene ación de his o-
ias, es deci , el campo que es udia como gene a his o ias au omá icamen e, han
a anzado eno memen e. A pa e de se un in e esan e eje cicio in elec ual es una
o ma de cons ui mane as de p esen a in o mación de o ma in e esan e lo que
pod ía ali ia el p oblema de la sob eca ga de in o mación o ayuda a pe sonas con
p oblemas de memo ia.
El es udio de los emas an o como componen es indispensables a la ho a de
con a his o ias como elemen os esenciales a la ho a de descompone , es udia y, a
pa i de ello, c ea nue as his o ias, es usado po muchas ob as muy in luyen es
como guias pa a diseña his o ias o iginales, ease Cook (2011) o Lemon and Reis
(1965), los han usado como pun o de clasi icación esencial, sin emba go, en mi
in es igación sob e el ema, he encon ado pocas aplicaciones di ec as de los emas
a la gene ación au oma ica de his o ias con lo que he decidido dedica es e abajo a
es udia la iabilidad de un sis ema basado en emas pa a la gene ación de his o ias.
1.1. Mo i ación
Ci ando a Lemon and Reis (1965) “La emoción enlazada a un ema juega un
papel impo an e man eniendo el in e es”. Como Lemon and Reis (1965) muchos
o os abajos li e a ios han de endido la impo ancia an o de ansmi i un ema
en las ob as a is icas (no necesa iamen e li e a ias, ease la in e p e ación musical
1
2Chap e 1. In oducción
de Lea i (2010)) como la de cen a se en el es udio de emas pa a cons ui nue as
his o ias, ease Cook (2011). Es os es udios sugie en la impo ancia de in es iga
es a u a de gene ación de his o ias p ocedu almen e, an o como heu is a a la ho a
de juzga las como seleccionado a pa a cons ui amas emá icamen e consis en es
a pa i de e en os dados. Ademas, dada la na u aleza sub ex ual de los emas,
pe mi e ansmi i in o mación de o ma más e icien e cuando se usan his o ias pa a
es e p oposi o.
1.2. Obje i o
El obje i o de es e abajo es diseña un modelo que ep esen e la ep esen ación
de emas en una his o ia y u iliza lo pa a cons ui un p o o ipo de una mé ica que
se á capaz de medi qué an bien una his o ia ansmi e una idea de e minada y
u iliza la pa a gene a un conjun o de his o ias.
Pa a la ealización del p oyec o, p ime o se elegi án algunos e en os y se u i-
liza án pa a gene a una población conside ablemen e g ande de e en os. Se desa -
olla á el sis ema pa a es udia , e alua y ex ae e en os basados en emas.
Es os emas se án de inibles po el usua io del sis ema sin necesidad de cambia
el sis ema, ya que los emas de una na a i a son subje i os, el usua io debe pode
de ini los componen es basados en su p opio en endimien o subje i o. Es o signi ica
que el sis ema debe pe mi i modi icaciones en odos los aspec os elacionados con
los emas. De es a mane a, se espe a hace el modelo emá ico al amen e adap able
sin la necesidad de eesc ibi el código.
Después de es o, se u iliza án algo i mos y heu ís icas pa a combina di e en es
amas, in en ando man ene los emas con es a ecombinación. Finalmen e, las a-
mas comple as se pasa án a algo i mos de gene ación de lenguaje na u al u ilizando
ecnologías basadas en IA pa a hace las más dige ibles.
Aunque es e p oyec o se cen a en aplica la heu ís ica a un g an co pus gene ado
median e simulación, se espe a que la heu ís ica pueda aplica se pa a medi cualquie
his o ia.
1.3. Es uc u a del documen o
Es e abajo se di ide en 4 capi ulos, en es os se expone los aspec os que se han
conside ado ele an es de es e abajo. La siguien e lis a mues a un esumen de
cada capi ulo:
En el capi ulo 1, es e capí ulo, se in oducen nociones básicas de lo que en en-
de emos po emas y su impo ancia así como los obje i os de es e abajo.
1.3. Es uc u a del documen o 3
En el capi ulo 2 se mues an abajos p e ios en los campos de gene acion
de his o ias que se usa an en es e p oyec o así como en es udios na a i os
necesa ios pa a el diseño de los algo i mos desc i os.
El capi ulo 3 desc ibe el abajo ealizado con las ideas y azonamien os que
lo acompañan.
El capi ulo 4 e lexiona sob e los obje i os conseguidos, compa a el abajo
ealizado con abajos an e io es y p opone posibles ias de in es igacion pa a
expandi el abajo.
Los apendices A, B y C con ienen in o mación sob e los da os conc e os u i-
lizados pa a las p uebas del p oyec o y sus conclusiones.
El Apéndice D con iene his o ias gene adas po el sis ema que se u iliza án
pa a e alua lo y encon a sus debilidades y o alezas en el capí ulo 4.

Chap e 1
In oduc ion
E e y s o y has a Theme, o an unde lying
p oposi ion ha indica es i s ype. The
Theme may be clea -cu and dis inc , o
shadowy and ague; i is always in e idence,
and di e en ia es one ype o s o y om all
he o he ypes. A ound each Theme any
numbe o dis inc ly di e en s o ies may be
w i en.
Cook (2011)
In he las decade many ad ances in he ask o s o y gene a ion, ha is, he
au oma ic gene a ion o s o ies, has ad ance g ea ly. Apa om being an in e es -
ing in ellec ual exe cise, i is a way o cons uc me hods o p esen in o ma ion in
an engaging manne , which could alle ia e he p oblem o in o ma ion o e load o
help people wi h memo y issues.
The s udy o he cen al ideas o a s o y, o hemes, bo h as indispensable compo-
nen s when elling s o ies and as essen ial elemen s when b eaking down, s udying,
and he eby c ea ing new s o ies is used by many highly in luen ial wo ks as guides
o design o iginal s o ies. (Cook, 2011), (Lemon and Reis, 1965) ha e used hemes
as an essen ial classi ica ion poin . Howe e , ew di ec applica ions o hemes in he
ield o au oma ic gene a ion o s o ies ha e been ound. To pallia e his de ici his
wo k will be s udying he easibili y o a heme-based sys em o s o y gene a ion.
1.1. Mo i a ion
Ci ing (Lemon and Reis, 1965) “The emo ion a ached o a heme plays a majo
ole in main aining in e es”. Like Lemon and Reis (1965), many o he au ho s ha e
de ended he impo ance o con eying a heme in a is ic wo ks, no necessa ily li -
e a y (Lea i , 2010), as well as ocusing on he s udy o hemes o cons uc new
s o ies (Cook, 2011). These s udies sugges he impo ance o in es iga ing his
5
6Chap e 1. In oduc ion
ou e o p ocedu al s o y gene a ion, bo h as a heu is ic o judging hem and as a
selec o o cons uc ing hema ically consis en plo s om gi en e en s. Addi ion-
ally, gi en he sub ex ual na u e o hemes, i allows o mo e e icien in o ma ion
ansmission when s o ies a e used o his pu pose.
1.2. Objec i es
The objec i e o his wo k is o design a model ep esen ing he ep esen a ion
o hemes in a s o y and use i o build a p o o ype o a me ic which will be able o
measu e how well a s o y con eys a ce ain idea and use i o gene a e a se o s o ies.
Fo he ealiza ion o he p ojec , i s , some e en s will be chosen and hey will
be used o gene a e a conside ably la ge popula ion o e en s. The sys em will be
de eloped o s udy, e alua e, and ex ac e en s based on hemes.
These hemes will will be de inable by he use o he sys em wi hou he need o
change he sys em, as he hemes o a na a i e is subjec i e he use mus be able
o de ine he componen s based on hei own subjec i e unde s anding, his means
he sys em should allow o modi ica ions o all he aspec s ega ding he hemes.
In his way, i is expec ed o make he hema ic model highly adap able wi hou he
need o ew i e code.
A e his, algo i hms and heu is ics will be used o combine di e en plo s, y-
ing o main ain hemes wi h his ecombina ion. Finally, he comple e plo s will
be passed o na u al language gene a ion algo i hms using AI-based echnologies o
make i mo e diges ible.
Al hough his p ojec is conce ned wi h applying he heu is ic o a la ge co pus
gene a ed ia simula ion, i is expec ed ha he heu is ic can be applied o measu e
any s o y.
1.3. S uc u e o he Documen
This wo k is di ided in o 4 chap e s, in which he aspec s conside ed ele an o
his wo k a e p esen ed. The ollowing lis shows a summa y o each chap e :
In chap e 1, his chap e , basic no ions o wha we will unde s and by hemes
and hei impo ance, as well as he objec i es o his wo k, a e in oduced.
In chap e 2, p e ious wo ks in he ields o s o y gene a ion used in his
p ojec , as well as na a i e s udies necessa y o he design o he desc ibed
algo i hms, a e p esen ed.
Chap e 3 desc ibes he wo k done wi h he accompanying ideas and easoning.
1.3. S uc u e o he Documen 7
Chap e 4 e lec s on he achie ed objec i es, compa es he wo k done wi h
p e ious wo ks, and p oposes possible esea ch a enues o expand he wo k.
Appendices A, B, and C con ain in o ma ion abou he speci ic da a used o
he p ojec ’s es s and hei conclusions.
Appendix D con ains s o ies gene a ed by he sys em which will be used o
e alua e i and ind i s laws and s eng hs in chap e 4.

Chap e 3
Desc ip ion o he sys em
The objec i e o his wo k is o de ine a me ic o measu e he ele ance o hemes
in a s o y and o use ha me ic o ex ac a s o y om a la ge co pus o e en s
gene a ed using a simula ion.
This wo k is di ided in o ou dis inc asks, each wi h i s own sec ion, a schema ic
ep esen a ion o his di ision can be ound in igu e 3.1:
Fi s sec ion: Explains he au ho ing p ocess ollowed o decide he e en s
and he way hey a e combined o o m he e en lis om whe e o ex ac
s o ies. This s ep e u ns a conside ably long lis o e en s.
Second sec ion: This sec ion is whe e he la ges con ibu ions o his wo k
eside, i desc ibes he hema ic model and i s applica ion o he ex ac ion
and e alua ion o s o ies om he e en lis . I e u ns a lis o possible plo s
wi h hema ic e alua ions om he e en s gene a ed in sec ion 1.
Thi d sec ion: Desc ibes me hods and me ics o combine di e en plo s and
c ea e mo e complex s o ies om he lis o plo s ecei ed om sec ion 2 and
an applica ion o he hema ic model o mul i plo s o ies. I e u ns a mo e
complex lis o plo s wi h a simila s uc u e as he one ecei ed.
Fou h sec ion: Desc ibes he gene a ion o na u al language om he na -
a i e gene a ed in sec ion 3.
3.1. E en s and hei gene a ion
The i s pa o his p ojec consis in he gene a ion o a la ge co pus o e en s.
In o de o achie e bo h empo al and causal consis ency be ween a la ge amoun
o e en s, a compu a ionally cheap simula ion has been equi ed. The simula ion
gua an ees he consis ency while i s low compu a ional cos allows he gene a ion o
a la ge numbe o e en s in a ela i ely low ime. To build he simula ion a g oup
o e en s o be simula ed and an engine o un he simula ion ha e been chosen.
15
16 Chap e 3. Desc ip ion o he sys em
Figu e 3.1: Schema ic ep esen a ion o he pipeline desc ibing he s o y gene a ion
sys em designed in his wo k
3.1.1. E en selec ion
E en hough a hema ic analysis does no es ic us o a conc e e g anula i y,
he e en s ha e been es ic ed o simple bu impo an e en s in he li es o cha -
ac e s which could be desc ibed wi h a single e b such as “Alice and Bob ha e a
child named Cha ly”.
On one hand, a g anula i y oo high, ha is, di iding he s o y in oo sim-
ple componen s, wo ds o example, as he minimum uni o he simula ion, would
educe he amoun o in o ma ion con ained in he co pus o e en s and, mo e im-
po an ly, his in o ma ion would be oo simple o exp ess any signi ican meaning
on hei own. A hema ic analysis on in o ma ion o his kind, e en i possible,
would equi e an holis ic analysis in he meaning o mul iple e en s as a uni , which
is ou o he scope o his p ojec .
On he o he hand, e en s such as “Alice igh s a ha d ba le wi h he conscience;
she inds i a losing ba le and makes an impo an e ela ion in o de ha she may
achie e peace o mind”1a e usually designed wi h a ce ain pu pose in a s o y and
as such ha e been designed o con ene a ce ain heme. An analysis made on a
co pus buil wi h he desi ed objec i e in mind would be poin less.
1E en ex ac ed om (Cook, 2011)
3.1. E en s and hei gene a ion 17
The simula ions p oposed in (Ma in e al., 2018) and (Chen e al., 2021) a e
in e es ing as hey sugges s using neu al ne wo ks and a huge da ase o s o ies o
gene a e e en s and p obabili ies o he simula ion , his app oach comes close
o he idea o compa a i e mo i as elemen epea ed in many s o ies desc ibed
in (Lemon and Reis, 1965) bu uses many s o ies wi h well de ined hemes which
would in e e e wi h he hema ic analysis his p ojec is conduc ing. Fo his ea-
son a simple au ho ed app oach has been used, ew i ing and expanding he lis o
e en s de ined in (Johnson-Bey e al., 2022) o allow o a g ea e exp essi i y in
he s o ies gene a ed. A desc ip ion o he e en s and he ac o s equi ed in hose
e en s, implemen ed as a plugin o he sys em desc ibed in (Johnson-Bey e al.,
2022), can be ound in he Appendix A.
3.1.2. In e ence engine
The sys em desc ibed in (Johnson-Bey e al., 2022) has been used o gene a e he
e en s o be p ocessed. I compu es e en s as as i mani es hem andomly whe e
he p obabili y o an e en appea ing is gi en by he mean o a base p obabili y, he
p obabili ies desc ibed in p econdi ions. Fo example an e en desc ibing a cha ac e
A s ealing money om a cha ac e B may be desc ibed as:
A s a e a l s money om B:
base p obabili y =0.2
p econdi ion name : A doesn l i k e B
p o b a b i l i y : 1−how much A l i k e s B no malized
p econdi ion name : A needs he money
p o ba bi li y : i A needs money igno e e l s e 0.01
In his example i he lo e A p o essed B was e y low, ep esen ed as a 0.1
ou o 1 o example hen i A needed money he p obabili y o A o s eal om B
would be (0.9+0.2)/2=0.55 bu i A did no need money he p obabili y would be
(0.9 + 0.2 + 0.01)/3=0.37. The simplici y o he sys em makes adding o ew i ing
e en s in i simple, making his simula ion con enien o expe imen a ion.
The e a e simula ions based on gene a i e AI, such as he one p esen ed in
(Pa k e al., 2023), which a e mo e complex and exp essi e han he one used in
his p ojec , bu he complexi y adds a di icul y in adding and emo ing e en s,
a signi ican inc ease in compu a ional ime and hey p esen an undesi ably high
g anula i y.
The simula ion models de ined as comp omises be ween he simula ion and he
au ho -cen ic app oaches a e immedia ely disca ded because hey would comp o-
mise he esul s o he e alua ion unde he hema ic model as many au ho -cen ic
echniques a e designed using p io na a i e wo ks de ined o con ey a conc e e
heme.
18 Chap e 3. Desc ip ion o he sys em
3.2. Thema ic engine
When building a hema ic engine he desi ed sys em should be able o measu e
and sa is y he equi emen s lis ed in sec ion 2.2.1, ha is, he sys em mus ex-
ac ex ual and sub- ex ual mo i s om he a omic elemen s o he s o y and in e
hemes om hei combina ion in ch onological o de . The epe i ion o mo i s mus
be an impo an ac o in he calcula ion o he ele ance o a heme. Mo i s ha e
been de ined by he e en s o he s o y as sugges ed by Lemon and Reis (1965) and
Lea i (2010).
Gi en he subjec i e na u e o he hemes o a s o y, an AI-based algo i hm
would no be an e icien op ion o ou sys em as each in e p e a ion o hemes
would equi e he cons uc ion o a pe sonalized da ase which would make he
au ho ing p ocess se e ely ine icien . Fo his eason a mo e adi ional app oach
inspi ed by (Ha good e al., 2010) has been used.
3.2.1. Rep esen a ion o hemes
When ep esen ing hemes, he ep esen a ion ound in (Ha good e al., 2008),
(Ha good, 2009), (Ha good e al., 2010) and (Ha good e al., 2018) is an in e es ing
s a ing poin which d aws hea y inspi a ion om (Lemon and Reis, 1965). This
model assumes he s o y is compiled om small segmen s o na a i e, also known
as Na a i e-a oms o na oms, o example a pa ag aph o a wo d, in his p ojec
he e en s de ined in he simula ion could be conside ed hese small segmen s o
na a i e which compose he s o y. These na oms a e ich wi h in o ma ion bu he
model would only ha e access o he meaning p o ided by he au ho , his meaning
is wha will be called ea u es. A single na om could con ain a my iad o ea u es
as he e may be mo e han a single in e p e a ion o he meanings p esen ed in an
ac ion, e en unde a single eade ’s pe cep ion, o example he ac ion “Alice kills
Bob” could e oke he co up ion o Alice i she was a mo ally igh eous cha ac e
un il his poin o a sense o loss i we knew Bob had a lo e wai ing o him.
The meanings con ained in he na oms con ey simple ideas, mo i s, impo an
o he en i e s o y, he mo i ep esen ed by “Alice kills Bob” could be “ agic lo e”
o example i we knew bob had a lo e wai ing o him, was in lo e o , mo e gene -
ically, he heme o “lo e” was an impo an pa in Bob’s s o y.
The epe i ion o basic ideas, o mo i s, along a s o y would, in u n, con ey a
b oade idea encompassing he whole s o y, in o he wo ds, he epe i ion o mo i s
con ey a heme. Fo example he epe i ion o mo i s such as “co up ion”, “b oken
u h” o “injus ice” would c ea e a s o y whose main heme may be desc ibed as
“The da k na u e o li e”. The main heme o a s o y can be in e ed om o he
hemes and mo i s, o example i he idea o “hope” can be ound in a s o y abou
“The da k na u e o li e” he main heme could be ein e p e ed as “Hope in he
da kes o places”. An schema ic ep esen a ion o his model can be ound in he
igu e 3.2.
3.2. Thema ic engine 19
Figu e 3.2: Model desc ibed in (Ha good e al., 2010) o he ep esen a ion o
hemes
The dis inc ion be ween mo i s and hemes bo h in (Ha good e al., 2008) and
(Lemon and Reis, 1965) is mos ly educed o he numbe o na oms needed o
ep esen an idea, i a single na om could ep esen an idea, ha idea is a mo i ,
meanwhile, i a se ies o na oms is necessa y o con ey an idea, ha idea is a heme.
In his con ex he idea o “ agic lo e” desc ibed in a p e ious example would be
close o a heme han o a mo i as i needs bo h a ea u e om whe e o in e “lo e”
and a ea u e om whe e o in e “ agic lo e” i “lo e” has al eady been in e ed.
This model allows a con enien sepa a ion be ween e en s and ideas h ough he
concep o ea u es, making he ea u es om whe e o in e mo i s and hemes he
cen e o he au ho ing p ocess while he ideas, mo i s and hemes, a e he building
blocks o he cen al idea o a s o y and, as such, he main componen s in i s analysis.
The s o y p esen ed in 3.12can be used as an example o he model in ac ion.
The ea u es and ideas used in he example a e no necessa ily he bes a aluing
he s o y bu a possible de ini ion in ended o make he example easy o ollow.
We can see he mo i “lo e” has been gene a ed by he ea u es o he na oms “Alice
de elops a c ush o Bob” and “Alice and Bob dissol e hei enmi y” while he heme
“powe _o _lo e” has been o med by he combina ion o he mo i "lo e" and he
2This is a simpli ica ion o a s o y gene a ed wi h an ea ly model o he one p esen ed in his
wo k.

20 Chap e 3. Desc ip ion o he sys em
seen ideas e en (na om) ea u es
1 Alice and Bob became enemies enmi y
i wa : e ils_o _wa
2 enmi y Alice de elops a c ush o Bob lo e
3 enmi y
lo e
Bob goes o wa wa
4 enmi y
lo e
wa
Alice and Bob dissol e hei
enmi y
i lo e: powe _o _lo e
lo e
enmi y
lo e
wa
lo e
powe _o _lo e
powe _o _lo e
Table 3.1: Example o idea acquisi ion om he ea u es o e en s
ea u e o he na om “Alice and Bob dissol e hei enmi y”. The ideas p esen ed by
he end o he na a i e would be “enmi y”, “lo e”, “wa ”, “lo e” and “powe _o _lo e”,
o which he hemes and mo i s “enmi y”, “lo e” and “lo e” would o m he heme
“Complex ela ionships”. The epe i ion in he ideas ound is impo an as epe i ion
o an idea is wha gene a es a heme bu his will be u he explo e in he nex
chap e s.
3.2.2. Implemen a ion o he in e ence o mo i s
The model p oposed by Ha good e al. (2008) and Lemon and Reis (1965) is
highly heo e ical and as such an implemen a ion is needed. Ha good e al. (2018)
De ines an implemen a ion in ended o he analysis o hemes in images, as de ined
in (Lemon and Reis, 1965), images a e desc ip i e wo ks, wo ks wi h no empo al
no causal ela ionship be ween hei ideas, as opposed o s o ies like he ones s ud-
ied in his wo k, whe e he s udy o hemes in hei con ex is essen ial.
Fo he implemen a ion, he minimum elemen s o he s o y, o na oms, has been
de ined as he e en s gene a ed by he simula ion, e e y na om has ea u es: ex ual
o sub- ex ual in o ma ion abou he mo i s i ep esen s, which can be de ined by
he use o he sys em. The ea u es de ine bo h he mo i s, as ideas always ex ac ed
om an e en , and he hemes, in he o m o condi ional ideas only mani es ing
i he equi ed ideas ha e been ound be o e. This dis inc ion allows he au ho o
de ine ideas ins ead o ha ing o deal wi h he abs ac dis inc ion be ween hemes
and mo i s.
Fo his app oach o wo k, he model needs o p ocess he s o y sequen ially o
main ain he o de . I ca ies a mul ise o ound mo i s a e e y s ep. The hemes
3.2. Thema ic engine 21
Figu e 3.3: Schema ic ep esen a ion o he implemen a ion o he hema ic model
p oposed in his wo k.
would no be able o mani es i he equi ed mo i s a e no ound in he mul ise ,
he mo i “ agic lo e” could no be ound wi hou he mo i "lo e" in he mul ise
o example. A main heme, he main idea o e alua e he s o y on, is de ined as a
se o ideas, o example, one main heme o be e alua ed in a s o y could be “The
da k na u e o li e” which would be ep esen ed by he mo i s and hemes “co up-
ion”, “b oken u h” and “injus ice” while he heme “ agic lo e” could be de ined
by he single idea “ agic lo e”. An schema ic ep esen a ion o he implemen a ion
can be ound in he igu e 3.3.
Repe i ion mus be he basis o e alua e he ele ance an idea has in a s o y
as s a ed in (Lea i , 2010). Fo an idea o s and ou , i mus appea signi ican ly
be ween he ideas o a s o y, ha is, epe i ion should be measu ed as a p opo ion
be ween desi ed ideas and ound ideas in he s o y. I he idea o wa would appea
only once in a s o y wi h housands o ideas, i would be di icul o in e “wa ” is
he heme o he s o y, meanwhile, i he s o y only had wo ideas and one o hem
was “wa ”, i would be na u al o hink he main heme o he ex has some hing
o do wi h wa .
Unde his assump ion, hemes, which equi e o o he mo i s and hemes o
mani es , a e e alua ed wi h a conside able disad an age, as mos o he ideas com-
p ising he hemes a e spen on p epa a ion and a e no coun ed when calcula ing
he p opo ion. Conside he example gi en in able 3.1 whe e “powe _o _lo e” is
clea ly a main heme because he s o y has been building o i s appa i ion by in o-
ducing lo e and enmi y, bu only coun s wi h a single appea ance and as such would
no be conside ed an impo an idea. To sol e he p oblem o he ep esen a ion o
hemes, he ideas gene a ing a heme which we e al eady ins an ia ed when he las
ins ance o he heme is being ins an ia ed a e duplica ed as ins ances o he heme.
22 Chap e 3. Desc ip ion o he sys em
seen ideas e en (na om) ea u es
1 Alice s eals Bob’s ca sel ishness
i sel lessness: co up ion
2 sel ishness Alice gi s money o Bob sel lessness
i sel ishness: edemp ion
3 sel ishness
sel lessness
edemp ion
* edemp ion
Alice s eals Bob’s easu e sel ishness
i sel lessness: co up ion
4 2 sel ishness
sel lessness
edemp ion
* edemp ion
co up ion
*co up ion
Alice gi s a easu e o Bob sel lessness
i sel ishness: edemp ion
5 2 sel ishness
2 sel lessness
2 edemp ion
co up ion
*co up ion
2 * edemp ion
Alice gi s a ca o Bob sel lessness
i sel ishness: edemp ion
2 sel ishness
3 sel lessness
3 edemp ion
co up ion
*co up ion
2 * edemp ion
Table 3.2: Example o idea acquisi ion om he ea u es o e en s a e he change
in oduced o sol e he p oblem in he ep esen a ion o hemes. he hemes ep e-
sen ed by *idea a e he hemes in oduced o sol e he p oblem.
An example o he solu ion can be seen in he able 3.2. In s ep 2 one “* e-
demp ion” is added because “sel ishness” p oduces “ edemp ion” and he e has been
ound one ins ance o “sel ishness” when “ edemp ion” was ound. in he s ep 3 one
ins ance o “sel ishness” has been added bu he ex a ins ances o “* edemp ion”
a e no upda ed as hey a e only upda ed when a new ins ance o “ edemp ion” is
ound which happens in s ep 4. In s ep 5 a new ins ance o “ edemp ion” is ound
and as such he ex a ins ances “* edemp ion” a e upda ed, bu because he num-
be o ins ances o ‘sel ishness” emain he same so do he numbe o ins ances o
“* edemp ion”.
The heme associa ed wi h an ac ion is dependan on he ac o we a e ollowing.
The ideas p oduced by Alice making he obbe y in “Alice s eals Bob’s ca ” may be
some hing simila o ”sel ishness” while he ideas associa ed wi h Bob being obbed
could be some hing close o “ agedy” o “loss”. This in o ma ion mus be aken
3.2. Thema ic engine 23
seen ideas e en (na om) ea u es
1 Alice s eals Bob’s ca loss
i iendship: b oken_ us
2 loss Alice gi s money o Bob o une
3 loss
o une
Alice s eals Bob’s easu e loss
i iendship: b oken_ us
4 2 loss
o une
Alice gi s a easu e o Bob o une
5 2 loss
2 o une
Alice gi s a ca o Bob o une
2 loss
3 o une
Table 3.3: Possible desc ip ion o he e en s in able 3.2 om Bob’s poin o iew
ins ead o Alice’s.
in o accoun when de ining he ea u es and e alua ing he ideas. Fo his eason a
mul ise o hemes simila o he one used o he en i e s o y be o e will be gi en
o each cha ac e in he s o y, and he e alua ion o a heme would be ca ied ou
wi h s o ies whe e all he ac ions in ol e a p o agonis , he mos p ominen cha -
ac e in a s o y. This in e p e a ion is simila o he one adop ed by (Lemon and
Reis, 1965) because “ he p o agonis , a he , is he esul o he o ma ion o he
s o y ma e ial in o a plo ”. This educes he complexi y a s o y can ake o a se o
e en s ollowing a single cha ac e , bu he algo i hms in sec ion 3.3 will es o e he
complexi y emo ed wi h his concession. While he example p esen ed in able 3.2
would be he one gene a ed by Alice, he example ound in able 3.3 would be he
ideas gene a ed om Bob’s poin o iew.
The ea u es a e ep esen ed as a map wi h he keys being an iden i ica ion o
he ype o e en and he alues a se o uncondi ional and condi ional mo i s, he
uncondi ional mo i s a e ep esen ed as a map wi h key ole_o _ac o being he ole
o he ac o hese ideas a e going o a ec and alue [idea_gene a ed] being a lis o
ideas added o he mul ise o ideas o he ac o , and he condi ionals a e ep esen ed
as a map wi h key ole_o _ac o being he ole o he ac o hese ideas a e going o
a ec and alue a lis o (condi ions_ideas, idea_gene a ed) whe e condi ions_ideas
is a lis o mo i s, his las uple ep esen s he ollowing ins uc ion:
i condi ion_idea p esen in he idea mul i se o he ac o
whose o l e i s ole_o _ac o in h i s ac ion
o e e y condi ion_idea in condi ion_ideas :
add idea_gene a ed in o he ac o ’ s mul ise
These ea u es a e gene a ed om a JSON ile whe e hey a e codi ied wi h he
ollowing s uc u e:
{
"e en _name " : {
" co n d i io n a l ": {
30 Chap e 3. Desc ip ion o he sys em
3.3.3. Thema ic model applica ion o he alua ion o s o ies
wi h mul iple p o agonis
Using he in e p e a ion o suspense ul e en s as e en s wi h ideas opposing he
main heme o he s o y a me ic o alue s o ies wi h mul iple p o agonis s can be
designed. When e alua ing his s o ies h ee hings mus be aken in o accoun :
1. I is impo an o he heme o e e y p o agonis o emain consis en , his
means he hema ic alue o each subplo de ined as all he ac ions o a p o-
agonis s mus be aken in o accoun .
2. The hemes o he p o agonis s should no clash wi h each o he , in o he
wo ds, he hemes o he p o agonis s mus be compa ible in pai s. Compa -
ibili y could be de ined using he in e sec ion o ideas be ween hemes, bu
wo simila hemes could be de ined wi h di e en ideas, “lo e” comp ised o
he idea“lo e” and “cha ac e g ow h” comp ised o ” edemp ion“ and “ iend-
ship” o example, o wo opposing hemes could ha e ideas in common, “lo e”
and “ agic lo e” o example. Fo his eason he ela ionship be ween main
hemes has been le as pa o he au ho ing p ocess.
3. How he hemes o he di e en subplo s in e ac wi h each o he mus be
aken in o accoun . In his p oblem he in e p e a ion o suspense ul e en s as
e en s wi h ideas opposing he main heme o he s o y comes in o play. Le
OP, named om opposing e en s, be he se o e en s in which a p o agonis
is playing a ole whe e an idea opposing he main heme o ha p o agonis
is gene a ed. a me ic could be calcula ed as
#|OP ∩e en s p eceding a change in p o agonis |
#|OP|
Conside ing he poin s s a ed abo e a me ic o s o ies wi h mul iple p o agonis
could be de ined as he mean be ween he alues o he hemes o each p o agonis
and he me ic de ined as a hema ic in e p e a ion o suspense.
3.3.4. Gene ic algo i hm o subplo combina ion
The easies model o di ec ly apply o a simula ion based s o y is he subplo
combina ion using gene ic algo i hms desc ibed in (Ge ás e al., 2022a), (Ge ás
e al., 2022b) and (Ge ás e al., 2023). In hei gene ic ep esen a ion hey o de
he di e en subplo s hey wan o me ge and cycle be ween hem when changing
subplo . The genes a e de ined as 2 ec o s o leng h he o al numbe o e en s
in he plo , he i s ec o , de ined as “ ec o o genes o subplo 4change” wi h
a 1 in he posi ion o an e en when he nex e en is om ano he subplo and 0
o he wise. The second ec o , de ined as “Vec o o genes on numbe o subplo s
o skip on change” which has an elemen o each one on he “ ec o o genes o
4in hei wo k hey e e o he subplo s as AoI(Axes o In e es ) because he subplo combina-
ion is an in e media e s ep in hei s o y gene a ion app oach and as such a e no ye comple ely
o med subplo s.

3.3. Subplo planning and combina ion 31
Figu e 3.4: Image ex ac ed om (Ge ás e al., 2023) ep esen ing he way hey
codi y he subplo combina ion p oblem in o genes o a gene ic algo i hm.
subplo change”, his elemen n ep esen s a change o he nex nsubplo in he
lis , i i eaches he end o he lis i s a s again. An example o he model can be
ound in igu e 3.4.
By de ining he plo s his way hey can make su e when a 1 is ound in " ec o
o genes o subplo change" he subplo changes as long as no elemen o "Vec o o
genes on numbe o subplo s o skip on change" is bigge o equal o he numbe o
subplo s. They use an in ege o ep esen he s a ing subplo , which we will no
need as i will be he main subplo , he subplo o he main cha ac e , and a hi d
ec o o de ine he cha ac e s who a e no ye ins an ia ed in hei model, as his
is no he case in his model his hi d ec o will be omi ed.
The unc ion o gene a e andom genes has been simpli ied om he o iginal
pape o be e sui a se o al eady cons uc ed plo s as some ex a s eps we e
equi ed o he s a ing subplo gene and he hi d ec o which a e no equi ed
in his wo k:
Fo he ec o o decisions on whe he o swi ch, andom choice be ween 0 and
1 is sui able.
Fo he ec o o decisions on skip size a each swi ch, andom choice be ween
1 and N-1 (wi h N he o al numbe o plo s being combined) is sui able.
The mu a ion o his algo i hm is equally simpli ied om he o iginal pape and
de ined by he ollowing ope a ions:
Fo he swi ch poin ec o , alues a a single poin chosen a andom a e
mu a ed.
Fo he skip size ec o , alues a a single poin chosen a andom a e mu a ed
o a alue chosen a andom wi hin he equi ed ange.
The new combina ion is de ined in he ollowing way:
32 Chap e 3. Desc ip ion o he sys em
Fo he swi ch poin ec o , a poin in he ec o is chosen a andom and he
co esponding hal es o he ec o s o he wo indi iduals a e swapped o e .
Fo he skip size ec o , a poin in he ec o is chosen a andom and he
co esponding hal es o he ec o s o he wo indi iduals a e swapped o e .
This me hods does no ensu e a he numbe o e en s in a subplo is main ained,
ha means he gene may equi e he e en numbe n+10 o a subplo wi h ne en s,
o a oid his a pos p ocess a e each o he p e ious ope a ions will be an:
Whene e a subplo is in i s las e en a 1 will be w i en in he “Vec o o
genes on numbe o subplo s o skip on change” and a one will be w i en on
he " ec o o genes o subplo change".
Whene e a subplo is going o change o an emp y subplo 1 will be added
o he “Vec o o genes on numbe o subplo s o skip on change”.
3.3.5. Implemen a ion o subplo combina ion
Wi h he subplo s c ea ed in sec ion 3.2.5 he combina ion p ocess can be im-
plemen ed. Fi s ly, possible subplo s o be combined mus be selec ed, when ex-
pe imen ing wi h he model only wo subplo s o less han hi y e en s each ha e
been combined a once o ensu e as eadabili y o he s o ies, bu any numbe o
subplo s may be combined wi h he me hods desc ibed in his sec ion. To conside
subplo s ap o combining hey mus sa is y:
The in e sec ion o cha ac e s be ween all he subplo s mus be abo e a ce ain
h eshold.
The pe cen age o appea ance, ha is, he pe cen age o e en s in which i
appea s, o one o he p o agonis s ( he one we a e calling main cha ac e )
mus be abo e he pe cen age o appea ance o any o he cha ac e by a ce ain
h eshold.
The p o agonis s who a e no he main cha ac e mus appea in he main
cha ac e ’s subplo a e hei subplo s ha e ended.
The main hemes o all he subplo s mus be ela ed, o ha a se o pai s
a e de ined in appendix C whe e a pai o hemes will be ela ed i a pai
con aining bo h is in he se .
When c ea ing he uples i mus be assu ed he e en s o each subplo a e in-
dependen , in he case o in e sec ion he longe subplo will be he only one o
main ain he e en as he longe a subplo is he mo e p incipal i is, being he
longes subplo he one called he main subplo .
Once he uples ha e been o med, he gene ic algo i hm desc ibed in he p e-
ious sec ion will be applied o combine hem. The i ness unc ion o e alua e he
possible combina ions will be de ined in he ollowing way:
3.4. Gene a ing na u al language o exp ess plo s 33
1. The hemes o he subplo s and hei e alua ions will be e ie ed.
2. Fo each cha ac e , he ension will be measu ed as he numbe o e en s
gene a ing ideas opposing he main heme o a cha ac e in ol ed in he e en
whe e a jump is aken di ided by he numbe o e en s gene a ing an idea
opposing opposing he main heme o a cha ac e in ol ed in he e en , i no
mo i opposing he heme exis s in he s o y he alue o his me ic will no
be aken in o accoun .
3. The ime consis ency will be measu ed simila ly o how (Ge ás e al., 2023)
measu es i :
Assigning 1 o any p ecedence cons ain ha is me ( o A + B, A
appea s be o e B in he discou se sequence).
I a equi ed p ecedence cons ain is no me , a pa ial sco e be ween 0
and 1 is assigned co esponding o he numbe o posi ions ha one o
he elemen s would need o shi o he cons ain o hold (no malised
o e he leng h o he sequence).
The a e age o all sequencing cons ain s is aken as he o al sequencing
sco e.
4. Finally, he mean o he sco es will be aken o decide a inal sco e.
3.4. Gene a ing na u al language o exp ess plo s
The e en s gene a ed a e capable o ep esen ing he na a i e bu o make he
na a i e mo e diges ible a p esen a ion is necessa y. Fo his he AI model Cha -
GPT3 will be used.
The model will be gi en a desc ip ion o e e y e en and he oles each ac o
plays in hem, simila o he desc ip ions shown in appendix A and a simple ex ual
ep esen a ion o he e en s which o m he na a i e. The ep esen a ion has been
asked wi h he ollowing que y:
Na a e h i s s e o e en s , no ex a in o ma ion mus be
added , he da es a e i e l e a n , add conec o s be ween he
sen ences
Chap e 4
Conclusiones y abajo u u o
A lo la go de es e abajo se ha p esen ado an o una mé ica pa a medi emas
en his o ias que siguen a un único pe sonaje y una ampliación pa a his o ias con
a ios pe sonajes, como una aplicación de es a a la ex acción de his o ias in e e-
san es a pa i de e en os.
Ya solo queda explica la si uación ac ual del abajo, discu i su iabilidad y
p opone caminos sob e los que expandi es a in es igación pa a, con sue e, gene a
ideas uc í e as pa a el campo de la gene ación de his o ias.
El abajo, en su es ado ac ual, Es capaz de e alua his o ias con uno o a -
ios p o agonis as. Es e modelo pe mi e al au o in oduci sus concep os de ideas
impo an es pa a una his o ia así como sus emas p incipales y elaciones de comple-
men a iedad y oposición en e es os pa a es udia na a i as. Algo menos uc í e o
ha sido el uso del modelo pa a la ex acción de his o ias ya que, aunque uncional, la
calidad de las his o ias gene adas no es compa able a la de sis emas mas comple os.
Es o se discu i á con mas de alle en la p óxima sección.
4.1. Conclusiones sob e el es ado ac ual del modelo
Las his o ias gene adas po es e sis ema, aunque a eces puedan llega a se in-
e esan es, no pueden compa a se a abajos an e io es como (Swa jes and Theune,
2008). En el apéndice D se pueden encon a es ejemplos de his o ias gene adas
con es e sis ema. Como se puede e en el ejemplo D.1 las modi icaciones en la
mé ica pa a e i a múl iples acciones epe idas no ue on su icien es pa a elimina
el p oblema comple amen e. Es bas an e p obable que es o se deba a una com-
binación de una p obabilidad muy al a pa a cie os e en os en la simulación, y
una de inición de ca ac e ís ica e ideas que a o ecen la apa ición de es os e en os.
Siendo es e el caso, es p obablemen e un e o ácilmen e subsanable sin cambios
signi ica i os en el modelo g acias a la acilidad de edición de ca ac e ís icas y emas.
Un e o más esencial se puede e en el ejemplo D.3. Se a a de una his o ia
35

36 Chap e 4. Conclusiones y abajo u u o
pe ec amen e uncional que ha ep esen ado dos emas a la pe ección, sin emba go,
un ema bien de inido no hace necesa iamen e a la his o ia in e esan e. Un buen
ema es necesa io pa a gene a una his o ia in e esan e pe o no es su icien e. Es e
p oblema es una consecuencia de depende casi po comple o del modelo emá ico
pa a gene a la his o ia, decisión omada especí icamen e pa a pode hace es as al-
o aciones. Desde un pun o de is a más posi i o, es un p oblema muy in e esan e
pa a una posible con inuación de es e p oyec o.
Un pun o en el que el sis ema ha uncionado mejo de lo espe ado es el ilus ado
en el ejemplo D.2. El cambio de sub ama en la boda del pe sonaje ágico es muy
e ec i o, la boda de Dako a gene a ideas de elicidad que con as an con su ine i able
agedia. Es o sugie e que las suposiciones hechas pa a el análisis emá ico de his o-
ias con a ios p o agonis as, en especial aquella que elaciona suspense y oposición
de emas, no an desencaminadas.
El modelo emá ico como componen e independien e, po o o lado, ha esul-
ado bas an e e ec i o. Es más lexible que modelos an e io es y pe mi e una mayo
exp esi idad de his o ias. Cumple con c eces los obje i os p opues o, un modelo
capaz de alo a el en oque emá ico de una his o ia, no es ingido po el núme o
de p o agonis as, y ácilmen e edi able po un au o g acias a la can idad de in o -
mación exp esable en a chi os auxilia es como los desc i os en los apéndices B y C.
A pesa de que el modelo cuen a con mucha exp esi idad y es ácilmen e pe son-
alizable, ca ece de oda la exp esi idad que se pod ía ene pues, e en os, emas y
ol de los ac o es, no son siemp e in o mación su icien e pa a conside a el alo co -
ec o de un ema. Véase el ejemplo 3.1 donde el hecho de que apa ezca la enemis ad
en e Bob y Alice le da mucha ele ancia al momen o en que es a acaba, pe o es o
no puede se cap u ado con emas gene ales. Las ideas gene adas po un e en o no
solo dependen del e en o, o as ideas encon adas an es y el ol del pe sonaje en el
e en o. In o mación de con ex o, como quienes ienen o os oles en el e en o y que
elación hay en e los pe sonajes, son piezas cla e pa a en ende las ideas gene adas
que es e modelo es incapaz de es udia .
4.2. Compa ación con el es ado del a e
Conside ando la ele ancia que han enido abajos pasados pa a el desa ollo
de es e es in e esan e discu i que apo aciones se ha hecho en es e abajo y el
esul ado de es as en compa ación con abajos pasados.
4.2.1. Compa ación del modelo emá ico p opues o con o os
modelos
En el campo de la gene ación de his o ias y la clasi icación de es as po el es-
udio de emas li e a ios el abajo ealizado es poco y dispe so. Como se comen ó
en el capi ulo 2 es o se puede debe a a ias azones, p incipales p obablemen e
4.3. T abajo u u o 37
siendo la al a de un consenso sob e una de inición de ema de una his o ia, an o
a ni el na a i o como compu acional, la al a de in e és debida a la p ominencia
de modelos basados en ex apolación de o as his o ias con emas in eg ados po
un au o humano p e iamen e, y la subje i idad de emas, en endidos como ideas
impo an es pa a una na a i a.
T abajos como Ma ia e al. (2000) usan una in e p e ación mas supe icial del
ema p escindiendo del sub ex o y e aluando las ideas ex uales. Es as suposiciones
hacen que puedan de ini un modelo ela i amen e obje i o usando edes neu onales,
como al es más ápido de con igu a , pues apenas equie e con igu ación y más sim-
ple pues es cons uido sob e modelos ya conocidos de edes neu onales. A cambio
es un modelo muy poco lexible pues pa a in oduci nue os emas se equie e de
gene a y en ena un nue o conjun o de da os y iene di icul ades pa a analiza el
sub ex o de una ob a a ís ica.
De en e los abajos sob e el es udio compu acional de los emas con los que
se pod ía hace una compa ación, los es udios de Ha good e al. (2018) son los más
ce canos a lo que se ha hecho en es e p oyec o. Su modelo, aunque inicialmen e
pensado pa a el análisis de his o ias ((Ha good e al., 2008),(Ha good, 2009)), es a
diseñado e implemen ado pa a el análisis de imágenes (Ha good e al., 2010). Su
modelo es igual de exp esi o que el comen ado en es e abajo sal o po los com-
ponen es causales y empo ales. Al se un modelo diseñado pa a imágenes, que no
con ienen in o mación empo al ni causal, esa in o mación no puede se ep esen-
ada po su modelo. Su implemen ación es una in e p e ación más li e al de las
ideas de Lemon and Reis (1965) y po ello hacen el modelo concep ual lige amen e
más simple, pe o con ello hace el abajo de au o ía mas complicado pues, como
se ha is o en es e abajo, la dis inción en e ema y mo i o no es an cla a en la
p ac ica y o za la sob e el modelo complica el abajo del au o .
Un ul imo g ado de exp esi idad que añade es e abajo sob e el an e io es la
gene ada po la hipó esis de que el ema no es independien e de la p esen ación de
la his o ia, en el modelo in oducido en es e abajo se ha hecho encape en o za
una dis inción del pun o de is a de una acción a la ho a de asigna le un ema, una
dis inción inexis en e en abajos an e io es.
En cuen o a la in e p e ación de suspense u ilizada en es e abajo, no es an
e ec i a como la p opues a en (O’Neill and Riedl, 2014) en la que se basa, pe o
es o a heu ís ica que puede se complemen able y pod ía llega a se más ácil de
implemen a . Pa ece un campo de es udio p ome edo .
4.3. T abajo u u o
Como se ha is o en las secciones an e io es, aunque se han hecho a ances sob e
abajos an e io es es as ideas y ecnologías es án lejos de alcanza su máximo po-
encial. En es a sección se in oduci án posibles aplicaciones pa a el sis ema ac ual
38 Chap e 4. Conclusiones y abajo u u o
y posibles ías de in es igación que segui pa a mejo a lo.
4.3.1. Posibles aplicaciones pa a el modelo emá ico
La aplicación mas e iden e es con inua con el abajo desc i o en es e docu-
men o y usa el modelo emá ico pa a mejo a y es udia his o ias gene adas po
o os modelos. Pues el mayo e o en la gene ación de his o ias de es e abajo es
depende casi exclusi amen e del modelo emá ico pa a la ex acción de his o ias.
El modelo emá ico se puede usa , y pod ía llega a se una he amien a pode osa,
como mé ica pa a la e aluación de o as his o ias. Aunque pa a ello unos e en os
más exp esi os y un es udio li e a io p o undo de emas se án necesa ios, po sue e
el modelo hace ambas cosas muy sencillas de cambia lo que pe mi i á bas an e ex-
pe imen ación.
Po ul imo la aplicación del modelo como mé ica de suspense pod ía se una
apo ación in e esan e al campo de combinación de sub amas.
4.3.2. Mejo as p opues as pa a el sis ema
En lo que espec a a la ex acción de e en os usando el modelo emá ico, es
a eas equie en un impo an e es udio.
Pa a la ob ención de e en os usa más a iedad de e en os y sis emas mas ex-
p esi os como el desc i o en (Pa k e al., 2023) o los desc i os en (Ma in e al.,
2018) y (Chen e al., 2021) pod ían mejo a conside ablemen e el sis ema.
Un es udio de ideas e icaces pa a exp esa in o mación impo an e sob e una
na a i a se á necesa io.
Sis emas auxilia es y mé icas p obadas deben se aplicadas pa a hace la
ex acción iable pues man ene un ema, aunque necesa io, no es su icien e
pa a ga an iza una buena his o ia, éase el ejemplo D.3.
Cuando se combinen his o ias hay muchas o mas que no se han podido al-
o a en es e abajo pa a añadi p o undidad, como la in oducción de una
sub ama cuya única a ea sea añadi con ex o como se sugie e en (Po eous
e al., 2016).
Se ha de dedica mas in es igación a la p esen ación de los e en os.
Pa a la mejo a del modelo emá ico es necesa io da mas exp esi idad a la de ini-
ción de ideas, muchas ideas son incomple as si no pueden ene en cuen a las si ua-
ciones que lle an a esas acciones, y o as ideas an e io es no son siemp e in o mación
su icien e. El con ex o, como se conocen los ac o es, que elación ienen, po que es a
ocu iendo dicha acción,... Es una pa e esencial del ema que debe se es udiada.
4.3. T abajo u u o 39
El es udio li e a io sob e como se compo an los emas es muy amplio y es
bas an e p obable que ideas impo an es no se hayan encon ado y usado en es e
abajo con lo que más in es igación es necesa ia. También es impo an e conside a
la o ma en la que los lec o es abso ben in o mación pa a juzga co ec amen e la
e ec i idad de las ideas gene adas po un ex o, pa a es o la aplicación de es udios
psicológicos como el p esen ado en (Zhang, 2005) puede esul a muy bene icioso
pa a un a ance en el modelo.
Finalmen e un modelo emá ico comple o debe ía ene la capacidad de es udia
como se asocian emas en una his o ia in e namen e, po ejemplo, el ema de la na i-
dad queda asociado al pe sonaje Ruby Sunday po epe ición en la se ie b i ánica
de ele isión Doc o Who, siemp e que apa ecen mo i os na ideños como la nie e
se sabe que en ese momen o algo elacionado Ruby es a eniendo luga . El modelo,
en su es ado ac ual, no es capaz de alo a e e encias emá icas in e nas en una
his o ia di ec amen e.

Bibliog aphy
Be ns ein, M. (2001). Ca d sha k and hespis: exo ic ools o hype ex na a i e.
In P oceedings o he 12 h ACM con e ence on Hype ex and Hype media, pages
41–50.
Booke , C. (2004). The se en basic plo s: Why we ell s o ies. A&C Black.
Chang, H.-M. and Soo, V.-W. (2008). Simula ion-based s o y gene a ion wi h a
heo y o mind.
Chen, H., Shu, R., Takamu a, H., and Nakayama, H. (2021). G aphplan: S o y
gene a ion by planning wi h e en g aph. a Xi p ep in a Xi :2102.02977.
Cook, W. (2011). PLOTTO: he mas e book o all plo s. Tin House Books.
Figgis, M. (2017). The Thi y-Six D ama ic Si ua ions. Fabe & Fabe .
Ga be, J., K eminski, M., Samuel, B., Wa d ip-F uin, N., and Ma eas, M. (2019).
S o yassemble : an engine o gene a ing dynamic choice-d i en na a i es. In
P oceedings o he 14 h In e na ional Con e ence on he Founda ions o Digi al
Games, pages 1–10.
Ge ig, R. J. and Be na do, A. B. (1994). Reade s as p oblem-sol e s in he expe-
ience o suspense. Poe ics, 22(6):459–472.
Ge ás, P. (2019). Gene a ing a sea ch space o accep able na a i e plo s. In
10 h In e na ional Con e ence on Compu a ional C ea i i y (ICCC 2019). UNC
Cha lo e, No h Ca olina, USA.
Ge ás, P., Concepción, E., and Méndez, G. (2022a). E olu iona y cons uc ion o
s o ies ha combine se e al plo lines. In In e na ional Con e ence on Compu-
a ional In elligence in Music, Sound, A and Design (Pa o E oS a ), pages
68–83. Sp inge .
Ge ás, P., León, C., and Méndez, G. (2015). Schemas o na a i e gene a ion
mined om exis ing desc ip ions o plo . In 6 h Wo kshop on Compu a ional
Models o Na a i e (CMN 2015). Schloss-Dags uhl-Leibniz Zen um ü In o -
ma ik.
47
48 BIBLIOGRAPHY
Ge ás, P., Méndez, G., and Concepción, E. (2022b). E olu iona y combina ion o
subplo pa e ns in o meaning ul plo s.
Ge ás, P., Méndez, G., and Concepción, E. (2023). E olu iona y combina ion
o connec ed e en schemas in o meaning ul plo s. Gene ic P og amming and
E ol able Machines, 24(1):7.
Ha good, C. (2009). Explo ing he Impo ance o Themes in Na a i e Sys ems.
PhD hesis, Uni e si y o Sou hamp on.
Ha good, C., Milla d, D., and Weal, M. (2010). Cap u ing he semio ic ela ionship
be ween e ms. The New Re iew o Hype media and Mul imedia, 16:71–84.
Ha good, C., Milla d, D. E., and Weal, M. J. (2008). A hema ic app oach o
eme ging na a i e s uc u e. In P oceedings o he hype ex 2008 wo kshop on
Collabo a ion and collec i e in elligence, pages 41–45.
Ha good, C., Milla d, D. E., and Weal, M. J. (2018). The hema ic modelling o
sub ex . Mul imedia Tools and Applica ions, 77:28281–28308.
Johnson-Bey, S., Nelson, M. J., and Ma eas, M. (2022). Neighbo ly: A sandbox o
simula ion-based eme gen na a i e. In 2022 IEEE Con e ence on Games (CoG),
pages 425–432. IEEE.
Klein, S. (1973). Au oma ic in e ence o seman ic deep s uc u e ules in gene a i e
seman ic g amma s. Technical epo , Uni e si y o Wisconsin-Madison Depa -
men o Compu e Sciences.
Lea i , J. (2010). My heme and mo i : Lé i-s auss and wagne . In e sec ions,
30(1):95–116.
Lemon, L. T. and Reis, M. J. (1965). Russian o malis c i icism: Fou essays,
olume 405. U o Neb aska P ess.
Ma ia, N., Sil a, M., Cincias, F., and Lisboa, C. (2000). Theme-based e ie al o
web news.
Ma in, L., Ammanab olu, P., Wang, X., Hancock, W., Singh, S., Ha ison, B.,
and Riedl, M. (2018). E en ep esen a ions o au oma ed s o y gene a ion wi h
deep neu al ne s. In P oceedings o he AAAI Con e ence on A i icial In elligence,
olume 32.
Mason, S., S agg, C., and Wa d ip-F uin, N. (2019). Lume: a sys em o p ocedu-
al s o y gene a ion. In P oceedings o he 14 h In e na ional Con e ence on he
Founda ions o Digi al Games, pages 1–9.
O’Neill, B. and Riedl, M. (2014). D ama is: A compu a ional model o suspense.
In P oceedings o he AAAI Con e ence on A i icial In elligence, olume 28.
Pa k, J. S., O’B ien, J. C., Cai, C. J., Mo is, M. R., Liang, P., and Be ns ein, M. S.
(2023). Gene a i e agen s: In e ac i e simulac a o human beha io .
BIBLIOGRAPHY 49
Po eous, J., Cha les, F., and Ca azza, M. (2016). Plan-based na a i e gene a ion
wi h coo dina ed subplo s. In ECAI 2016, pages 846–854. IOS P ess.
P opp, V. (1968). Mo phology o he Folk ale. Uni e si y o Texas p ess.
Swa jes, I. and Theune, M. (2008). The i ual s o y elle : S o y gene a ion by
simula ion. In BNAIC, pages 257–264. Ci esee .
Thompson, S. (1955). Mo i -Index o Folk-Li e a u e, Volume 4: A Classi ica ion o
Na a i e Elemen s in Folk Tales, Ballads, My hs, Fables, Mediae al Romances,
Exempla, Fabliaux, Jes -Books, and Local Legends, olume 4. Indiana Uni e si y
P ess.
Tobias, R. B. (2012). 20 mas e plo s: And how o build hem. Penguin.
Zhang, H. (2005). Ac i a ion o hemes du ing na a i e eading. Discou se P o-
cesses, 40(1):57–82.
Appendix A
De ini ion o e en s in he Neighbo ly
simula ion
In his appendix I will name he ull lis o e en s conside ed wi h a b ie de-
sc ip ion o wha hey ep esen and hei ac o s as his a e necessa y o de ine he
ea u es o he e en s using he model desc ibed in 2.2.1:
BecomeAdolescen E en : T ansi ion e en when a cha ac e becomes an
adolescen .
Ac o s:
•subjec : he ac o eaching adolescence.
BecomeAdul E en : T ansi ion e en when a cha ac e becomes an adul .
Ac o s:
•subjec : he ac o eaching adul hood.
BecomeEnemies: E en ep esen ing wo cha ac e s becoming enemies.
Ac o s:
•subjec ,o he : he enemies.
BecomeF iends: E en ep esen ing wo cha ac e s becoming iends.
Ac o s:
•subjec ,o he : he iends.
BecomeSenio E en : T ansi ion e en when a cha ac e becomes a senio .
Ac o s:
•subjec : he ac o eaching senio i y.
BecomeYoungAdul E en : T ansi ion e en when a cha ac e becomes a
young adul .
Ac o s:
51

52 Appendix A. De ini ion o e en s in he Neighbo ly simula ion
•subjec : he ac o eaching young adul hood.
Bi hE en : E en igge ed when a new cha ac e is bo n.
Ac o s:
•subjec : he ac o being bo n.
B eakUp: E en ep esen ing he end o a oman ic ela ionship.
Ac o s:
•subjec ,O he : ac o s b eaking he ela ionship.
BuildT easu e: E en ep esen ing he c ea ion o a easu e. The build
e en s ha e been designed o pu in o ci cula ion in o he wo ld objec s ha
can be s olen o gi ed.
Ac o s:
•subjec : he builde .
BuildVehicle: E en ep esen ing he cons uc ion o a ehicle.
Ac o s:
•subjec : he builde .
BuildWeapon: E en ep esen ing he cons uc ion o a weapon.
Ac o s:
•subjec : he builde .
BusinessClosedE en : E en igge ed when a business is closed down.
Ac o s:
•subjec : he ac o who closes he business.
•business: The business being closed.
BuyT easu e: E en ep esen ing a cha ac e pu chasing a easu e om
ano he cha ac e .
Ac o s:
•buye : Ac o who buys.
•selle : Ac o who sells.
BuyVehicle: E en ep esen ing a cha ac e pu chasing a ehicle om an-
o he cha ac e .
Ac o s:
•buye : Ac o who buys.
•selle : Ac o who sells.
BuyWeapon: E en ep esen ing a cha ac e pu chasing a weapon om an-
o he cha ac e .
Ac o s:
53
•buye : Ac o who buys.
•selle : Ac o who sells.
ChangeResidenceE en : E en igge ed when a cha ac e changes hei
place o esidence.
Ac o s:
•subjec : Ac o who changes esidence.
•new esidence: Whe e he ac o is mo ing o.
Dea h: E en igge ed when a cha ac e dies.
Ac o s:
•subjec : he ac o who has died.
Depa DueToUnemploymen : E en igge ed when a cha ac e lea es due
o unemploymen .
Ac o s:
•subjec : he ac o lea ing.
Depa Se lemen : E en igge ed when a cha ac e lea es a se lemen .
Ac o s:
•subjec : he ac o lea ing.
Dissol eEnmi y: E en ep esen s when enmi y be ween wo cha ac e s
ends.
Ac o s:
•subjec ,O he : he ex-enemies.
Dissol eF iendship: E en ep esen s when iendship be ween wo cha ac-
e s ends.
Ac o s:
•subjec ,O he : he ex- iends.
Fi edF omJob: E en ep esen s a cha ac e being i ed om a job.
Ac o s:
•subjec : The ac o being i ed.
•business: The job being i ed o .
•job_ ole: he job ole he subjec no longe has.
Fo mC ush: E en ep esen s a cha ac e o ming a c ush on ano he cha -
ac e .
Ac o s:
•subjec : The ac o o ming a c ush.
54 Appendix A. De ini ion o e en s in he Neighbo ly simula ion
•o he : The ac o who subjec has a c ush on.
Ge Di o ced: E en ep esen ing a di o ce.
Ac o s:
•subjec ,ex_spouse: The ac o s ge ing di o ced.
Ge Ma ied: E en ep esen ing a ma iage.
Ac o s:
•subjec ,subjec : The ac o s ge ing ma ied.
Ge P egnan : E en igge ed when a cha ac e becomes p egnan . Ac o s:
•subjec : The ac o ge ing p egnan .
•pa ne : The o he pa en .
Gi T easu e: E en ep esen ing a cha ac e gi ing a easu e o ano he
cha ac e .
Ac o s:
•gi e : Ac o who makes a gi .
•gi ed: Ac o who ecei es a gi .
Gi Vehicle: E en ep esen ing a cha ac e gi ing a ehicle o ano he cha -
ac e .
Ac o s:
•gi e : Ac o who makes a gi .
•gi ed: Ac o who ecei es a gi .
Gi Weapon: E en ep esen ing a cha ac e gi ing a weapon o ano he
cha ac e .
Ac o s:
•gi e : Ac o who makes a gi .
•gi ed: Ac o who ecei es a gi .
Ha eChildE en : E en igge ed when a cha ac e has a child.
Ac o s:
•subjec ,subjec : The pa en s.
•child: The child.
JoinSe lemen E en : E en igge ed when a cha ac e joins a se lemen .
Ac o s:
•subjec : he ac o en e ing he se lemen .
•se lemen : Se lemen being joined.
55
Kill: E en ep esen ing a cha ac e killing ano he cha ac e .
Ac o s:
•subjec : The ac o who killed.
•o he : The ac o who was killed.
LaidO F omJob: E en ep esen ing a cha ac e being laid o om a job.
Ac o s:
•subjec : Ac o lea ing he job.
•business: The business he subjec was wo king on.
•job: The job he subjec had.
Lea eJob: E en ep esen ing a cha ac e lea ing a job.
Ac o s:
•subjec : Ac o lea ing he job.
•business: The business he subjec was wo king on.
•job: The job he subjec had.
P omo edToBusinessOwne : E en ep esen ing a cha ac e being p o-
mo ed o business owne .
Ac o s:
•subjec : Ac o being p omo ed.
•business: The business he subjec is wo king on.
• o me _owne : The ac o who was he owne be o e.
Re i e: E en igge ed when a cha ac e e i es om wo k.
Ac o s:
•subjec : Ac o lea ing he job.
•business: The business he subjec was wo king on.
•job: The job he subjec had.
S a ANewJob: E en ep esen ing a cha ac e s a ing a new job.
Ac o s:
•subjec : Ac o en e ing he job.
•business: The business he subjec is going o wo k on.
•job: The job he subjec will ha e.
S a Business: E en ep esen ing a cha ac e s a ing a new business.
Ac o s:
•subjec : Ac o c ea ing he business.
•business: The business he subjec is c ea ing.
62 Appendix B. Desc ip ion o ea u es assigned o he e en s
170 ],
171 " se lemen ":[]
172 }
173 },
174 "ChangeResidenceE en ":{
175 "condi ional":{
176 "subjec ":[],
177 " new_ esidence ":[]
178 },
179 " uncondi ional ":{
180 "subjec ":[
181 " begining "
182 ],
183 " new_ esidence ":[
184 " begining "
185 ]
186 }
187 },
188 "BuildVehicle":{
189 "condi ional":{
190 "subjec ":[]
191 },
192 " uncondi ional ":{
193 "subjec ":[
194 " ambi ion ",
195 " c ea ion "
196 ]
197 }
198 },
199 " BuyVehicle ":{
200 "condi ional":{
201 "buye ":[],
202 "selle ":[]
203 },
204 " uncondi ional ":{
205 "buye ":[
206 ],
207 "selle ":[
208 "weal h"
209 ]
210 }
211 },
212 " Depa DueToUnemploymen ":{
213 "condi ional":{
214 "subjec ":[
215 [["sel ishness"]," e ibu ion"]

63
216 ]
217 },
218 " uncondi ional ":{
219 "subjec ":[
220 " ailu e",
221 "ending"
222 ]
223 }
224 },
225 "Depa Se lemen ":{
226 "condi ional":{
227 "subjec ":[]
228 },
229 " uncondi ional ":{
230 "subjec ":[]
231 }
232 },
233 "BuildWeapon":{
234 "condi ional":{
235 "subjec ":[]
236 },
237 " uncondi ional ":{
238 "subjec ":[
239 "dea h",
240 " c ea ion "
241 ]
242 }
243 },
244 " BuyWeapon ":{
245 "condi ional":{
246 "buye ":[],
247 "selle ":[]
248 },
249 " uncondi ional ":{
250 "buye ":[
251 "dea h"
252 ],
253 "selle ":[
254 "weal h"
255 ]
256 }
257 },
258 " BuildT easu e ":{
259 "condi ional":{
260 "subjec ":[]
261 },
64 Appendix B. Desc ip ion o ea u es assigned o he e en s
262 " uncondi ional ":{
263 "subjec ":[
264 "weal h",
265 " c ea ion "
266 ]
267 }
268 },
269 "Gi Vehicle":{
270 "condi ional":{
271 "gi e ":[
272 [
273 [
274 "sel ishness"
275 ],
276 " edemp ion "
277 ]
278 ],
279 "gi ed":[
280 [
281 [
282 "sel lessness"
283 ],
284 " edemp ion "
285 ],
286 [
287 [
288 "sel ishness"
289 ],
290 " injus ice "
291 ]
292 ]
293 },
294 " uncondi ional ":{
295 "gi e ":[
296 "sel lessness"
297 ],
298 "gi ed":[
299 " iendship "
300 ]
301 }
302 },
303 " BecomeF iends ":{
304 "condi ional":{
305 "subjec ":[],
306 "o he ":[]
307 },
65
308 " uncondi ional ":{
309 "subjec ":[
310 " iendship "
311 ],
312 "o he ":[
313 " iendship "
314 ]
315 }
316 },
317 "Gi T easu e":{
318 "condi ional":{
319 "gi e ":[
320 [
321 [
322 "sel ishness"
323 ],
324 " edemp ion "
325 ]
326 ],
327 "gi ed":[
328 [
329 [
330 "sel lessness"
331 ],
332 " edemp ion "
333 ],
334 [
335 [
336 "sel ishness"
337 ],
338 " injus ice "
339 ]
340 ]
341 },
342 " uncondi ional ":{
343 "gi e ":[
344 "sel lessness",
345 " iendship ",
346 " us "
347 ],
348 "gi ed":[
349 " us ",
350 "gi ed",
351 " iendship "
352 ]
353 }
66 Appendix B. Desc ip ion o ea u es assigned o he e en s
354 },
355 " Gi Weapon ":{
356 "condi ional":{
357 "gi e ":[
358 [
359 [
360 "sel ishness"," dea h "
361 ],
362 " edemp ion "
363 ]
364 ],
365 "gi ed":[
366 [
367 [
368 "sel lessness"
369 ],
370 " edemp ion "
371 ],
372 [
373 [
374 "sel ishness"
375 ],
376 " injus ice "
377 ]
378 ]
379 },
380 " uncondi ional ":{
381 "gi e ":[
382 "sel lessness",
383 " iendship "
384 ],
385 "gi ed":[
386 "dea h",
387 "gi ed",
388 " iendship "
389 ]
390 }
391 },
392 "BuyT easu e":{
393 "condi ional":{
394 "buye ":[],
395 "selle ":[]
396 },
397 " uncondi ional ":{
398 "buye ":[
399 "weal h"
67
400 ],
401 "selle ":[
402 "weal h"
403 ]
404 }
405 },
406 "S ealVehicle":{
407 "condi ional":{
408 "s eale ":[
409 [
410 [
411 "sel lessness"
412 ],
413 " co up ion "
414 ]
415 ],
416 "s olen":[
417 [
418 [
419 "sel ishness"
420 ],
421 " e ibu ion"
422 ]
423 ]
424 },
425 " uncondi ional ":{
426 "s eale ":[
427 "sel ishness",
428 " a el"
429 ],
430 "s olen":[
431 "b oken_ us "
432 ]
433 }
434 },
435 "S ealWeapon":{
436 "condi ional":{
437 "s eale ":[],
438 "s olen":[
439 [
440 [
441 "sel ishness"
442 ],
443 " e ibu ion"
444 ]
445 ]

68 Appendix B. Desc ip ion o ea u es assigned o he e en s
446 },
447 " uncondi ional ":{
448 "s eale ":[
449 "sel ishness",
450 "dea h"
451 ],
452 "s olen":[
453 "b oken_ us "
454 ]
455 }
456 },
457 " S ealMoney ":{
458 "condi ional":{
459 "s eale ":[],
460 "s olen":[
461 [
462 [
463 "sel ishness"
464 ],
465 " e ibu ion"
466 ]
467 ]
468 },
469 " uncondi ional ":{
470 "s eale ":[
471 "sel ishness",
472 "weal h"
473 ],
474 "s olen":[
475 "b oken_ us "
476 ]
477 }
478 },
479 " S ealT easu e ":{
480 "condi ional":{
481 "s eale ":[
482 [
483 [
484 "sel lessness"
485 ],
486 " co up ion "
487 ]
488 ],
489 "s olen":[
490 [
491 [
69
492 "sel ishness"
493 ],
494 " e ibu ion"
495 ],
496 [
497 [
498 " us "
499 ],
500 "b oken_ us "
501 ]
502 ]
503 },
504 " uncondi ional ":{
505 "s eale ":[
506 "sel ishness",
507 "weal h"
508 ],
509 "s olen":[
510 "b oken_ us "
511 ]
512 }
513 },
514 "BecomeSenio E en ":{
515 "condi ional":{
516 "subjec ":[]
517 },
518 " uncondi ional ":{
519 "subjec ":[
520 "weal h",
521 "wo k",
522 " ambi ion "
523 ]
524 }
525 },
526 "Dea h":{
527 "condi ional":{
528 "subjec ":[]
529 },
530 " uncondi ional ":{
531 "subjec ":[" dea h "]
532 }
533 },
534 "BecomeAdul E en ":{
535 "condi ional":{
536 "subjec ":[]
537 },
70 Appendix B. Desc ip ion o ea u es assigned o he e en s
538 " uncondi ional ":{
539 "subjec ":[]
540 }
541 },
542 " BecomeEnemies ":{
543 "condi ional":{
544 "subjec ":[
545 [
546 [
547 "lo e"
548 ],
549 " agic_lo e"
550 ],
551 [
552 [
553 " iendship "
554 ],
555 " agic_ iendship"
556 ]
557 ],
558 "o he ":[
559 [
560 [
561 "lo e"
562 ],
563 " agic_lo e"
564 ],
565 [
566 [
567 " iendship "
568 ],
569 " agic_ iendship"
570 ]
571 ]
572 },
573 " uncondi ional ":{
574 "subjec ":[
575 "enmi y"
576 ],
577 "o he ":[
578 "enmi y"
579 ]
580 }
581 },
582 " Fo mC ush ":{
583 "condi ional":{
71
584 "subjec ":[
585 [
586 [
587 "lo e"
588 ],
589 " agic_lo e"
590 ]
591 ],
592 "o he ":[
593 [
594 [
595 "lo e"
596 ],
597 " agic_lo e"
598 ]
599 ]
600 },
601 " uncondi ional ":{
602 "subjec ":[
603 "lo e"
604 ],
605 "o he ":[
606 "lo e"
607 ]
608 }
609 },
610 "S a Da ing":{
611 "condi ional":{
612 "subjec ":[
613 [
614 [
615 "lo e"
616 ],
617 " equi ed_lo e "
618 ]
619 ],
620 "pa ne ":[
621 [
622 [
623 "lo e"
624 ],
625 " equi ed_lo e "
626 ]
627 ]
628 },
629 " uncondi ional ":{
78 Appendix C. Desc ip ion o hemes used and hei ela ionship
And hei ela ionships a e de ined as a JSON below:
1{
2" ela ed":[
3[" C ime Pu sued by Vengeance ","Ri al y"],
4[" C ime Pu sued by Vengeance ","Lo e And
Ha ed"],
5[" C ime Pu sued by Vengeance ","Remo se"],
6[" C ime Pu sued by Vengeance "," Injus ice o
Li e "],
7["Ri al y"," Lo e And Ha ed "],
8["Ri al y"," P o esional Ambi ion "],
9[" Lo e And Ha ed ","Remo se"],
10 [" Lo e And Ha ed "," Injus ice o Li e "],
11 [" Lo e And Ha ed "," P o esional Ambi ion "],
12 ["Remo se"," Injus ice o Li e "],
13 ["Remo se"," Libe a ion F om Wo k "],
14 [" P o esional Ambi ion "," Libe a ion F om
Wo k "],
15 [" Injus ice o Li e "," Libe a ion F om Wo k "]
16 ],
17 " opposi e ":{
18 " C ime Pu sued by Vengeance ":[" iendship ",
"lo e"," peace "," ailu e"],
19 "Ri al y":[" iendship "," equi ed_lo e ","
peace"," ai h "],
20 " Lo e And Ha ed ":[" agic_lo e","
agic_ iendship","sel ishness","dea h "
,"enemy"," ailu e"],
21 "Remo se":[" c ea ion "," begining ","dea h",
"sel ishness"," ailu e"],
22 " Libe a ion F om Wo k ":[" weal h "," ambi ion "
,"wo k"," begining "," ailu e"],
23 " Injus ice o Li e ":[" iendship "," lo e ","
peace","weal h"],
24 " P o esional Ambi ion ":[" iendship ","lo e "
,"peace"," ailu e"]
25 }
26 }

Appendix D
Examples o s o ies gene a ed using he
sys em desc ibed in his documen
Lis ing D.1: The p oblem wi h lack o a ia y is consis en e en a e he adjus men s
in he me ics o a oid i .
( ’ I n j u s i c e o Li e ’ , ’ Lo e And Ha ed ’ ) : 0.3755606522743312
2632−03:
BecomeEnemies (
subjec = Kayleigh Gold (23521) ,
o he = Callen Ame son (23362)
)
2642−08:
Fo mC ush (
subjec = Dominick Hamble (23406) ,
o he = Kayleigh Gold (23521)
)
2642−08:
BecomeEnemies (
subjec = Kayleigh Gold (23521) ,
o he = Kayleigh Gold (23521)
)
2643−08:
S a Da ing (
subjec = Dominick Hamble (23406) ,
pa ne = Kayleigh Gold (23521)
)
2643−10:
Ge Ma ied (
subjec = Dominick Hamble (23406) ,
subjec = Kayleigh Gold (23521)
)
2648−09:
BecomeEnemies (
subjec = Kayleigh Gold (23521) ,
79
80
Appendix D. Examples o s o ies gene a ed using he sys em desc ibed in his
documen
o he = En ique Gold (23630)
)
2651−05:
BuildWeapon (
sub je c = E s e l l a Jinke son (23646)
)
2651−06:
BuildWeapon (
sub je c = E s e l l a Jinke son (23646)
)
2651−12:
Gi Weapon (
g i e = E s e l l a Jinke son (23646) ,
g i e d = An onella Icke (23574)
)
2652−06:
Gi Weapon (
g i e = An onella Icke (23574) ,
g i e d = E s e l l a Jinke son (23646)
)
2652−07:
Gi T easu e (
g i e = E s e l l a Jinke son (23646) ,
g i e d = An onella Icke (23574)
)
2652−09:
Gi Weapon (
g i e = E s e l l a Jinke son (23646) ,
g i e d = Alden Blackaby (23635)
)
2652−10:
Gi Weapon (
g i e = E s e l l a Jinke son (23646) ,
g i e d = Kayleigh Gold (23521)
)
2652−11:
Gi T easu e (
g i e = E s e l l a Jinke son (23646) ,
g i e d = Alden Blackaby (23635)
)
2652−12:
Gi T easu e (
g i e = E s e l l a Jinke son (23646) ,
g i e d = Bl ai Hoa (23606)
)
2653−01:
Gi Weapon (
g i e = E s e l l a Jinke son (23646) ,
g i e d = An onella Icke (23574)
81
)
2653−06:
Gi V ehic le (
g i e = Alden Blackaby (23635) ,
g i e d = E s e l l a Jinke son (23646)
)
2653−08:
Gi T easu e (
g i e = Alden Blackaby (23635) ,
g i e d = E s e l l a Jinke son (23646)
)
2653−09:
BecomeF iends (
sub je c = E s e l l a Jinke son (23646) ,
o he = Kayleigh Gold (23521)
)
2654−05:
Gi Weapon (
g i e = E s e l l a Jinke son (23646) ,
g i e d = Kayleigh Gold (23521)
)
2654−05:
Dea h (
sub je c = E s e l l a Jinke son (23646)
)
2658−12:
Kill (
subjec = Kayleigh Gold (23521) ,
o he = Kayleigh Gold (23521)
)
2658−12:
BecomeEnemies (
sub je c = E s e l l a Jinke son (23646) ,
o he = Kayleigh Gold (23521)
)
Tex ual ep esen a ion :
Kayleigh Gold and Callen Ame son became enemies . La e , Dominick Hamble de eloped a c ush on Kayleigh Gold . Howe e , Kayleigh Gold a l so became he own enemy . Subsequen ly , Dominick Hamble and Kayleigh Gold s a e d da ing . Sho ly a e , Dominick Hamble and Kayleigh Gold go ma ied . Meanwhile , Kayleigh Gold and En ique Gold became enemies .
Du ing h is ime , E s e l l a Jinke son b u i l a weapon and hen b u i l ano he weapon . Subsequen ly , E s e l l a Jinke son g i e d a weapon o An onella Icke , and in e u n , An onella Icke g i e d a weapon o E s e l l a Jinke son . Addi ionally , E s e l l a Jinke son g i e d a e as u e o An onella Icke .
Fu he mo e , E s e l l a Jinke son g i e d a weapon o Alden Blackaby and al so g i e d a weapon o Kayleigh Gold . She hen g i e d a ea su e o Alden Blackaby and ano he ea su e o Bla i Hoa . Mo eo e , E s e l l a Jinke son g i e d ano he weapon o An onella Icke .
Following his , Alden Blackaby g i e d a eh i c le o E s e l l a Jinke son and al so g i e d a eas u e o he . La e , E s e l l a Jinke son and Kayleigh Gold became i e n d s . Subsequen ly , E s e l l a Jinke son g i e d ano he weapon o Kayleigh Gold .
Un o una ely , E s e l l a Jinke son died . E en ually , Kayleigh Gold k i l l e d h e s e l . A e wa d , E s e l l a Jinke son and Kayleigh Gold became enemies .
82
Appendix D. Examples o s o ies gene a ed using he sys em desc ibed in his
documen
Lis ing D.2: Example showing he e ec i ness o unde s anding suspense as clashing
hemes.
( ’ Ri al y ’ , ’ Lo e And Ha ed ’ ) : 0.4853982721133222
2266−09:
S a Da ing (
subjec = Dako a Scowden (20730) ,
pa ne = Sam Calco e (20880)
)
2270−04:
Fo mC ush (
subjec = Sam Calco e (20880) ,
o he = E in Buley (20630)
)
2282−08:
Ge Ma ied (
subjec = Dako a Scowden (20730) ,
subjec = Sam Calco e (20880)
)
2283−11:
BuildVehicle (
subjec = Megan Men(21059)
)
2283−12:
BuildVehicle (
subjec = Megan Men(21059)
)
2284−01:
BuildT easu e (
subjec = Megan Men(21059)
)
2284−07:
S ealT easu e (
s e a l e = Megan Men(21059) ,
s o l e n = Dallas S ingham (20986)
)
2284−08:
BuildT easu e (
subjec = Megan Men(21059)
)
2285−06:
S ealT easu e (
s e a l e = Megan Men(21059) ,
s o l e n = Dallas S ingham (20986)
)
2286−10:
BecomeEnemies (
subjec = Megan Men(21059) ,
o he = Dallas S ingham (20986)
)
83
2287−07:
BecomeEnemies (
subjec = Megan Men(21059) ,
o he = Sam Calco e (20880)
)
2288−02:
Kill (
subjec = Megan Men(21059) ,
o he = Sam Calco e (20880)
)
2289−01:
BecomeEnemies (
subjec = Ced ic Scowden (20929) ,
o he = Megan Men(21059)
)
Tex ual ep esen a ion :
Dako a Scowden and Sam Calco e s a ed da ing . La e , Sam Calco e de eloped a c ush on E in Buley . E en ually , Dako a Scowden and Sam Calco e go ma ied .
In he mean ime , Megan Men b u i l a e h ic l e and hen b u i l ano he e h i c l e . Subsequen ly , Megan Men b u i l a e as u e . La e , Megan Men s o l e a e as u e om Dallas S ingham . Addi ionally , Megan Men b u i l ano he ea su e . Megan Men hen s o l e ano he e as u e om Dallas S ingham .
E en ually , Megan Men and Dallas S ingham became enemies . La e , Megan Men and Sam Calco e a lso became enemies . Subsequen ly , Megan Men k i l l e d Sam Calco e . Following his , Ced ic Scowden and Megan Men became enemies .

84
Appendix D. Examples o s o ies gene a ed using he sys em desc ibed in his
documen
Lis ing D.3: This example shows a s ong heme is no su icien o gua an ee an
in e es ing s o y
( ’ Lib e a io n F om Wo k’ , ’ P o e s s i o n a l Ambi ion ’ ) : 0.48278886249443476
0003−09:
S a ANewJob (
subjec = F e l i c i y Ma s on (306) ,
business = a m (95)
)
0003−09:
Re i e (
sub je c = N e l l i e Lashley (114 ) ,
business = a m (95)
)
0015−06:
Re i e (
subjec = Louis Fel on (356) ,
business = a m (95)
)
0042−07:
P omo edToBusinessOwne (
subjec = F e l i c i y Ma s on (306) ,
business = a m (95) ,
o me _owne = B yce Jewel (117)
)
0042−07:
Re i e (
subjec = B yce Jewel (117) ,
business = a m (95)
)
0042−07:
Lea eJob (
subjec = F e l i c i y Ma s on (306) ,
business = a m (95)
)
0043−06:
BecomeSenio E en (
subjec = F e l i c i y Ma s on (306)
)
0043−10:
Re i e (
subjec = F e l i c i y Ma s on (306) ,
business = a m (95)
)
0043−10:
Lea eJob (
subjec = F e l i c i y Ma s on (306) ,
business = a m (95)
)
85
0043−10:
BusinessClosedE en (
subjec = F e l i c i y Ma s on (306) ,
business = a m (95)
)
Tex ual ep esen a ion :
F e l i c i y Ma s on s a e d a new job a a a m . A he same ime , N e l l i e Lashley e i e d om he same a m . La e , Louis Fel on a ls o e i e d om he a m .
Yea s l a e , F e l i c i y Ma s on was p omo ed o business owne o he a m , aking o e om B yce Jewel . Consequen ly , B yce Jewel e i ed . Sho ly a e , F e l i c i y Ma s on l e he job a he a m .
The ol low in g yea , F e l i c i y Ma s on became a sen io . Subsequen ly , F e l i c i y Ma s on e i e d om he a m again and l e he job . This s e i e s o e en s led o he c lo su e o he a m .