Storytelling with Adjustable Narrator Styles and
Boyang Li, Mohini Thakkar, Yijie Wang, and Mark O. Riedl
School of Interactive Computing, Georgia Institute of Technology
{boyangli, mthakkar, yijiewang, riedl}
Abstract. Most storytelling systems to date rely on manually coded
knowledge, the cost of which usually restricts such systems to oper-
ate within a few domains where knowledge has been engineered. Open
Story Generation systems are capable of learning knowledge necessary
for telling stories in a given domain. In this paper, we describe a technique
that generates and communicates stories in language with diverse styles
and sentiments based on automatically learned narrative knowledge. Di-
versity in storytelling style may facilitate different communicative goals
and focalization in narratives. Our approach learns from large-scale data
sets such as the Google N-Gram Corpus and Project Gutenberg books in
addition to crowdsourced stories to instill storytelling agents with linguis-
tic and social behavioral knowledge. A user study shows our algorithm
strongly agrees with human judgment on the interestingness, concise-
ness, and sentiments of the generated stories and outperforms existing
1 Introduction
Narrative Intelligence (NI), or the ability to craft, tell, understand, and respond
to stories, is considered a hallmark of human intelligence and an effective commu-
nication method. It follows that Narrative Intelligence is important for Artificial
Intelligence that aims to simulate human intelligence or communicate effectively
with humans. In this paper, we focus on computational NI systems that can
generate and tell stories.
A significant challenge in building NI systems is the knowledge intensive na-
ture of NI. To date, most computational systems purported to demonstrate NI
are reliant on substantial amount of manually coded knowledge, whose avail-
ability is limited by the time and financial cost associated with knowledge en-
gineering. Consequently, most systems are designed to operate in only a few
micro-worlds where knowledge is available. For example, an automated story
generator may be told about the characters and environment of Little Red Rid-
ing Hood; that system can tell a large variety of stories about the given set of
characters and topic, but no stories about other characters or topics.
Open Story Generation systems (e.g. [5, 16]) have been proposed in order
to tackle the challenge of generating and telling stories in any domain. Such
systems can learn the needed knowledge for story generation and storytelling
without a priori knowledge engineering about a particular domain. We previ-
ously described an open story generation system, Scheherazade [5], which uses
crowdsourcing to construct a commonsense understanding about how to perform
everyday activities such as going to a restaurant or going to a movie theater.
Given a topic, the system learns what it needs to generate a story about the
topic. However, the system does not reason about how to tell a story, or how to
translate a sequence of abstract events into natural language.
A story may be told to achieve communicative goals, such as to entertain, to
motivate, or to simply report facts. Different goals may require narrators to adopt
different storytelling styles. Additionally, the narrative technique of focalization
involves describing the same events from different characters’ perspectives, pos-
sibly with opposing sentiments (cf. [2,18]). As a first step, we tackle the problem
of creating different storytelling styles for Open Story Generation. Style param-
eters are learned from large data sets including the Google N-Gram corpus [8]
and books from Project Gutenburg ( We offer methods
to tune the storytelling with different levels of details, fictional language, and
sentiments. Our user study indicates our algorithm strongly agrees with human
readers’ intuition of linguistic styles and sentiments, and outperforms existing
1.1 Background and Related Work
Story generation and interactive narrative have a long history. See Gerv´as [3]
and Riedl & Bulitko [12] for overviews. Several Open Story Generation systems
have been proposed before. The SayAnything system [16] generates stories from
snippets of natural language mined from weblogs. McIntyre & Lapata [7] learn
temporally ordered schema from fairy tales, merge schema into plot graphs,
and use a genetic algorithm to maximize the coherence of generated stories.
Crowdsourcing has been proposed as a means for overcoming the knowledge
bottleneck. Sina et al. [14] use case-based reasoning to modify crowdsourced
semi-structured stories to create alibi for virtual suspects in training. None of
these approaches explicitly model discourse or generate different narration styles.
The work in this paper builds off our previous work on the Scheherazade
system [4, 5], which learns the structure of events in a given situation from
crowdsourced exemplar stories describing that situation. As opposed to other
story generation systems, Scheherazade is a just-in-time learner; if the system
does not know the structure of a situation when it is called for, it attempts to
learn what it needs to know from a crowd of people on the Web. This results in
a script-like knowledge structure, called a plot graph. The graph contains events
that can be expected to occur, temporal ordering relations between events, and
mutual exclusions between events that create branching alternatives.
The learning of the plot graph proceeds in four steps [4, 5]. After exemplar
stories about a social situation are crowdsourced from Amazon Mechanical Turk
(AMT), the learning starts by creating clusters of sentences of similar semantic
meaning from different exemplar stories. Each cluster becomes an event in the
plot graph. In order to reduce the difficulty in natural language processing,
crowd workers from AMT have been asked to use simple language, i.e., using
one sentence with a single verb to describe one event, avoiding pronouns, etc. The
second and third steps identify the temporal precedences and mutual exclusions
between events. The final step identifies optional events. Story generation in
Scheherazade is the process of selecting a linear sequence of events that do
not violate any temporal or mutual exclusion relations in the script [5]. However,
telling the generated story in natural language with different storytelling styles
has not been previously realized.
Focalization in narration refers to telling stories from different viewpoints
(e.g. of an omniscient entity or any story character; cf. [2]), potentially requiring
multiple narration styles. Most computational implementations focus on plot
events [11, 18] instead of linguistic variations. Curveship [10] generates focal-
ized text based on manually coded knowledge. Our work directly addresses the
problem of diverse language use by implied or explicit narrators.
Automatic generation of distinct linguistic pragmatics for narration has also
been studied. The Personage system [6] maps the Big Five psychological model
to a large number of linguistic parameters. Rishes et al. [13] used Personage
to create different tellings of stories generated from a semantic representation
consisting of events and character intentions. The generated linguistic styles
differ mostly in aspects independent of content, such as in the use of swear
words, exclamation marks and shuttering. Instead of generating from symbolic
representations with precise semantic meaning, we select from existing sentences
that are similar but not strictly synonymous to describe an event (i.e. sentences
may differ in content). We consider parameters directly related to word choices:
degree of details, fictionality, and sentiments.
2 Storytelling with Different Styles and Sentiments
This section describes the process of telling stories in natural language with a
variety of personal styles. The architecture for story generation and communi-
cation is shown in Figure 1. Plot graph learning is typically an offline process
that incrementally constructs a knowledge base of models of social situations,
from which stories can be generated [4, 5]. A story is generated as one possible
total-ordered sequence of events that respect all constraints in the plot graph.
The discourse planning stage selects some interesting events from the complete
sequence to be told, which is beyond the scope of this paper and not used in the
evaluation. This paper focuses on the last stage of the architecture: describing
the selected events with personal styles and affects, which we explain below.
Recall that each event in the learned plot graph is a cluster of natural lan-
guage descriptions of similar meaning. Given a generated story (a complete,
linear sequence of events), natural language text is generated by selecting the
sentence from each cluster that best matches the intended narration style. We
describe two criteria for selecting sentences: (1) the interestingness of the text
and (2) the sentiment of the text. We aim to create a diverse set of storytelling
styles that may be suitable for different occasions. For example, some narrators
Plot Graph
A plot
Fig. 1. The system pipeline.
or story characters may speak very succinctly, whereas others can recall vivid de-
tails. A positive tone may be used if the narrator wants to cheer up the audience;
a negative tone may be suitable for horror stories and so on. We also present
a Viterbi-style algorithm that considers preferences on individual sentences and
inter-sentence connections to produce coherent textual realizations.
We performed a second round of crowdsourcing to obtain a variety of event
descriptions that can reflect different narration styles and sentiments. The orig-
inally crowdsourced exemplar stories were written in simple sentences that help
to simplify natural language processing [4]. Thus, only simple event descriptions
were available for selection. The second round of crowdsourcing asked AMT
workers to provide “interesting” event descriptions. For $1, workers wrote de-
tailed descriptions for each event in a given story; each description may contain
more than one sentences. We allowed workers to interpret “interesting” however
they wanted, though we suggested that they describe characters’ intentions, fa-
cial expressions, and actions. Each worker saw a complete sequence of events
to make sure they understand the story context. We accepted all stories that
describe the events we provide, and did not perform a manual check of interest-
2.1 Textual Interestingness
We investigate two aspects of language that affect the interestingness of stories.
The first is the amount of details provided, and the second is the degree that the
story language resembles the language used in fictions. We model the amount of
details with the probability of a sentence in English, since Information Theory
suggests a less likely sentence contains more information (i.e. more details). We
compute the probability of an English word as its frequency in the Google N-
Gram corpus. Due to the large size of the corpus, these frequencies approximate
word probabilities in general English. We compute the probability of a sentence
using the bag-of-word model, where the probability of sentence S containing
words w
, w
, . . . , w
, each appearing x
, x
, . . . , x
times is
P (S) =
P (w
where P (w
) is the probability of word w
. For our purpose, the average frequency
over the 10-year period of 1991 to 2000 in the “English 2012” corpus is used.
Stop words are removed before computation.
We further consider the style of language as how much it resembles fictional
novels. The language used in fictions has distinctive word choices as fictions tend
to accurately describe actions (e.g. “snatch” instead of “take”) and emotions,
and make less use of formal words (e.g. “facility”, “presentation”). If a word
appears more frequently in fiction books than in all books, we can presume that
its use creates a sense that the story is being told in a literary manner. Therefore,
the fictionality of a word w is the ratio
= P
(w)/P(w) (2)
where P (w) is the probability of a word computed previously and P
(w) is
the probability of a word appearing in the “English Fiction 2012” corpus from
the Google N-Gram corpus. The fictionality of a sentence is aggregated from
fictionality values of individual words as an exponentiated average:
fic(S) =
card(W )
where W is the multiset of words in sentence S, and card(W ) is its cardinality.
α is a scaling parameter. The exponential function puts more weights on words
with higher fictionality, so that a few highly fictional words are not canceled off
by many words with low fictionality.
Table 1 shows some example sentences. We observe that the most proba-
ble sentence (MostProb) usually provides a good summary for the event. The
most fictional (MostFic) sentence usually contains more subjective emotions and
character intentions, whereas the least probable (LeastProb) sentence is usually
longer and contains more objective details. We balance and combine the MostFic
and the LeastProb criteria by using the harmonic mean in order to create the
sentence with most interesting details (MID). Let us denote the ranks of each
sentence under the LeastProb and the MostFic criteria as r
and r
tively. For example, the least probable sentence has r
= 1, and the second
most fictional has r
= 2. The harmonic mean rank r
is computed as
2 r
+ r
). The sentence with the lowest r
is picked as the one
with the most interesting details.
2.2 Textual Sentiments
Stories may be told with positive or negative sentiment. To detect sentiments
of sentences in each event cluster, we construct a sentiment dictionary called
Table 1. Example Sentences Selected with the Probability, Fictionality, and Sentiment
Example event 1: Sally puts money in bag
MostProb: Sally put $1,000,000 in a bag.
LeastProb: Sally put the money in the bag, and collected the money from the 2 tellers
next to her.
MostFic: Sally quickly and nervously stuffed the money into the bag.
MID: Sally quickly and nervously stuffed the money into the bag.
Positive: Sally continued to cooperate, putting the money into the bag as ordered.
Negative: Sally’s hands were trembling as she put the money in the bag.
Example event 2: John drives away
MostProb: John drove away.
LeastProb: John pulled out of the parking lot and accelerated, thinking over which
route would make it easier to evade any police cars that might come along.
MostFic: John sped away, hoping to get distance between him and the cops.
MID: John sped away, hoping to get distance between him and the cops.
Positive: As the stoplight turned green and the daily traffic began to move, John
drove away.
Negative: John slammed the truck door and, with tires screaming, he pulled out of
the parking space and drove away.
Smooth SentiWordNet (SSWN). SSWN builds off SentiWordNet [1], which tags
each synset (word sense) in WordNet [9] with three values: positivity, negativity,
and objectiveness, the three summing to 1. SentiWordNet was produced by prop-
agating known sentiments of a few seed words along connections between words
in WordNet to provide good coverage, but this automatic approach can produce
many erroneous values, resulting in unreliable sentiment judgments. Smooth
SentiWordNet uses an unsupervised, corpus-based technique to correct errors
found in the original library and expand its coverage beyond words appearing
in WordNet. The intuition behind SSWN is that words that are nearby should
share similar sentiments, and words closer should have a stronger influence than
words farther away. We take sentiment values from SWN and “smooth” the val-
ues based on word location using Gaussian kernel functions, in order to alleviate
errors and further expand the coverage.
We perform smoothing with a corpus of 9108 English books from Project
Gutenberg that are labeled as fiction. These books are tagged with parts of
speech (POS) with the Stanford POS Tagger [17]. Each pair of word and POS
is considered a unique word. For every word we want to compute sentiment
value for, we consider a neighborhood of 100 words, 50 to its left and 50 to
its right. The target word is at position 0 and denoted as w
. The words to
its immediate left and right are at position -1 and 1, and so forth. The posi-
tions of these words are included in the index set N. For word w
at position
i N, its influence at position j is modeled with a Gaussian kernel function
: g
(j) = exp
(i j)
, where parameter d determines how fast the func-
Table 2. An example partial story with most interesting details. The first 7 sentences
in the story are omitted for space reasons.
(. . . the first 7 sentences omitted)
When it was his turn, John, wearing his Obama mask, approached the counter.
Sally saw Obama standing in front of her and she felt her whole body tense up as her
worst nightmare seemed to be coming true.
Once Sally began to run, John pulled out the gun and directed it at the bank guard.
John wore a stern stare as he pointed the gun at Sally.
Sally saw the gun and instantly screamed before she could stop herself.
John told her she had one minute to get the money and shook the gun at her.
John gave Sally a bag to put the banks money in.
John struggled to stuff the money in his satchel.
Sally was quietly sobbing as John grabbed the bag full of money.
John strode quickly from the bank and got into his car tossing the money bag on the
seat beside him.
John pulled out of the parking lot and accelerated, thinking over which route would
make it easier to evade any police cars that might come along.
tion diminishes with distance, and is empirically set to 32. Only nouns, verbs,
adjectives and adverbs in complete sentences can influence the target word.
In the each neighborhood that the target word w
appears, its sentiment s
is computed as a weighted average of all kernel functions at position 0:
where s
is the sentiment value from SentiWordNet, i.e. the difference between
the positive and negative polarity. The SentiWordNet value for the target word
has no influence on itself, i.e. 0 / N. As a word can appear multiple times
in different neighborhoods, the final sentiment value for w
is the average over
all neighborhoods it appears in. We aggregate sentiments of individual words in
sentence S, again using the exponential average:
sentiment(S) =
sign (s
) exp (β|s
card(V )
where card(V ) is the cardinality of the multiset V , containing only nouns, verbs,
adjectives or adverbs in sentence S. β is a scaling parameter. The exponential
function ensures that words expressing strong sentiments are weighted more
heavily than words with weak sentiments.
We selected a subset of English words that are of interest to our task. The
exemplar stories in two previously crowdsourced social situations—dating at the
movie theater and bank robbery—contain 1001 unique nouns, verbs, adverbs
and adjectives. We selected highly influential adjectives and adverbs from their
direct neighbors, producing a total of 7559 words. We normalize the raw values
produced by smoothing, so that 1 percentile and 99 percentile of the values fall
in the range of [1, 1], to account for outliers.
Table 1 shows some of the most
positive and most negative sentences. We find the results to reflect the valences
of individual words. Although this approach works most of the time, there are
cases such as sarcasm where the sentiment of a sentence could be the opposite
of that of individual words. SSWN is evaluated in Section 3.
2.3 Connecting Sentences
For each event, we can find individual sentences ranked highest for any criterion
or combinations of criteria using the harmonic mean. However, this selection
does not consider the coherence between sentences and may results in incoherent
texts due to two major problems: (1) previously mentioned objects can suddenly
disappear and previously unmentioned objects can appear, and (2) a sentence
can repeat actions in the previous sentence. To address this problem, we propose
a Viterbi-style algorithm, which considers both selection criteria for individual
sentences and the connection between sentences.
In a hidden Markov model (HMM), the Viterbi algorithm finds a sequence of
hidden variables that best explains a sequence of observed random variables. The
algorithm relies on two things in an HMM: One, the probabilities of a hidden
variable generating any observation. That is, the observation indicates preference
over values of the hidden variable. Two, the probabilities of a hidden variable
transiting to the next hidden variable. That is, we have preferences over pairs of
values for adjacent variables.
Our problem is similar as we want to find the highest scored sentence sequence
based on preferences over sentences in each event cluster, and preferences on how
adjacent sentences connect. In this paper, we do not consider connection between
non-adjacent sentences. Specifically, we score the connection between any two
sentences s
, s
as log ((sn(i, j) + 1)/(sv(i, j) + 1)), where sn(i, j) is the number
of nouns shared by the two sentences, and sv(i, j) is the number of verbs shared
by the two sentences. Similarly, we score individual sentences as the reciprocal
of their ranks according to any selection criterion c : score(s
) = 1/rank
Our algorithm is shown as Algorithm 1. The BestSeqEndingIn function is
recursive, because in order to find the best sequence ending in a given sentence
from the j
event cluster c
, we need to consider the scores of best sequences
ending in every sentence from the previous cluster c
, in addition to the con-
nection between every sentence from cluster c
and s
. Due to the Markov
property, we do not need to consider clusters c
, . . . , c
. We can then iterate
over every sentence from cluster c
to find the best sequence ending in cluster c
A dynamic programming approach can be used to store every sequence ending
in every sentence from every cluster and their scores. For a sequence of n clusters
and m sentences in each cluster, the time and space complexity are O(m
n) and
O(mn). An example partial story is shown in Table 2.
The list of books can be downloaded at
list.txt. The resulted dictionary is at:
Algorithm 1 Generation of Story Text
function GenerateText(event sequence hc
, c
, . . . , c
for each sentence s
, s
, . . . , s
} in event cluster c
, score(seq
)) BestSeqEndingIn(s
, c
end for
return the highest scored sequence from seq
, seq
, . . . , seq
end function
function BestSeqEndingIn(s
, c
for each sentence s
, s
, . . . , s
} in event cluster c
, score(seq
)) BestSeqEndingIn(s
, c
) stored previously
new seq
+ s
score(new seq
) score(seq
) + score(s
, s
) + score(s
end for
seq the highest scored sequence from new seq
, . . . , new seq
return (best seq, score(best seq))
end function
Table 3. Statistics of crowdsourced interesting stories
Movie Date Bank Robbery
# Stories 20 10
# Sentences 470 210
# Words per sentence 14.53 13.7
# Verbs per sentence 2.36 2.6
3 Evaluation
We performed a user study to test if the results of our algorithm agree with
human intuition. We investigated two social situations: dating at a movie theater
and bank robbery. In addition to the originally crowdsourced exemplar stories,
we crowdsourced interesting stories using procedures described in Section 2.
Some statistics of these stories are shown in Table 3. Some of these sentences can
be seen in prior examples showing least probable and most fictional sentences
for particular clusters (the most probably sentence typically comes from the
original, simplified language exemplars).
With the newly crowdsourced sentences added, for each situation we generate
two groups of stories with the Viterbi-style algorithm with different sentence
selection criteria. We do not perform discourse planning to avoid confounding
factors. The first group includes stories generated from the most interesting
details (MID) criterion, the most probable (MostProb) criterion, and a story
where we use the MID criterion but penalize long sentences. After reading the
stories, participants are asked to select the most interesting story, the most
detailed story and the most concise story. Our hypothesis is that human readers
will select the MID story as containing the most details and the most interesting,
and the MostProb story as the most concise. We set α to 12. The second group of
Table 4. Participant agreement with our algorithm. § denotes p < 0.0001. * denotes
p < 0.0005.
Participant Agreement %
Test Movie Date Bank Robbery
Most Concise Story 90.38
Most Detailed Story 97.92
Most Interesting Story 88.46
Positive/Negative Stories 86.54
stories include a story containing the sentences with the most positive sentiment
from each event, and a story containing the sentences with the most negative
sentiment. We set β to 16 and 2 for the movie data and bank robbery situation
respectively. After reading the second group, participants are asked to select a
positive and a negative story. We hypothesize human readers will agree with the
algorithm’s sentiment judgments.
A total of 52 undergraduate, master’s, and doctoral students participated in
our study. Table 4 shows the percentage of human participants that agree with
the algorithm. All results are predominantly positive and consistent with our
hypothesis, strongly indicating our algorithm can capture the human intuition
of interestingness, conciseness, and sentiments. We use one-tailed hypothesis
testing based on the multinomial/binomial distribution and find the results to
be extremely statistically significantly above a random baseline.
However, it is arguably easier to detect the sentiment of an entire story than
to detect the sentiment of individual sentences, because in a story, a few sentences
labeled with wrong sentiments mixed with many correctly labeled sentences can
be overlooked by human readers. To further evaluate our sentiment detection
algorithm, we perform a sentence-level comparison. We first take out the top 3
most positive sentence and top 3 most negative sentences from 45 event clusters
in both situations. One positive and one negative sentences are randomly selected
from the top 3, and shown to participants, who labeled one sentence as positive
and the other as negative. In total, 52 participants performed 4678 evaluations of
265 unique pairs of sentences. The results are shown in Table 5 . Overall, 70.76%
of participants’ decisions agree with our algorithm. The majority opinion on each
pairs of sentences agrees with our algorithm for 80.75% of the time.
We further compare our algorithm with SentiWordNet. We replaced word
sentiments in Equation 5 with values directly taken from SentiWordNet, and
label a sentence as positive if its sentiment is higher than the median sentence
in a cluster, and negative if lower. Results are matched against the participants’
decisions. We tuned β to maximize performance. We also compare SSNW with
the technique by Socher et al. [15] from Stanford University, which directly labels
a sentence as positive, negative or neutral. The results are summarized in Table
5. SSWN outperform SWN by a margin of 11.16% to 16.22%, and outperform
Socher et al. by 34.85% to 41.5%, although Socher et al.’s algorithm targets
Table 5. Comparing word sentiment values from SentiWordNet and the values
computed by our smoothing technique. § denotes p < 0.0001.
Participant Agreement %
Test Smooth SWN SentiWordNet Socher et al.
Sentence Sentiments 70.76 59.60
Sentence Sentiments by Majority Vote 80.75 64.53
movie reviews and has not been tuned on our data set. A Chi-Square test shows
the difference between conditions are extremely statistically significant.
4 Discussion and Conclusions
Open Story Generation systems can learn necessary knowledge to generate sto-
ries about unknown situations. However, these systems have not considered how
to tell the generated story in natural language with different styles. Such a ca-
pability is useful for achieving different communicative goals and for projecting
a story to perspectives of story characters. For example, a story with mostly ob-
jective details is suitable for conveying information, whereas interesting stories
tend to describe characters’ subjective feelings. A positive tone may be used to
cheer up the audience, or to describe things from a cheerful character’s perspec-
tive. As a first step toward solving these problems, we discuss Open Storytelling
with different styles, such as attention to detail, fictionality of language, and sen-
timents. Our technique employ the same knowledge structure learned by Open
Story Generation systems and large data sets including the Google N-Gram
Corpus and Project Gutenberg. We develop a method for selecting interesting
event descriptions and build a sentiment dictionary called Smooth SentiWordNet
by smoothing out errors in sentiment values obtained from SentiWordNet. Our
user study with 52 participants reveals that corpus-based techniques can achieve
recognizably different natural language styles for storytelling. Future work will
investigate newer fiction corpora, such as weblogs labeled as stories, than Project
Gutenberg, which may not fully reflect the language use of this day.
Our storytelling techniques help to overcome the authoring bottleneck for
storytelling systems by learning from data sets consisting of crowdsourced ex-
emplar stories, the Google N-Gram Corpus, and books from Project Gutenberg,
Building off existing work [4, 5], the effort presented in this paper moves the
state of the art towards the vision of computational systems capable of telling
an unlimited number of stories about an unlimited number of social situations
with minimum human intervention.
5 Acknowledgments
We gratefully acknowledge DARPA for supporting this research under Grant
D11AP00270, and Stephen Lee-Urban and Rania Hodhod for valuable inputs.
1. Baccianella, S., Esuli, A., Sebastani, F.: SentiWordNet 3.0: An enhanced lexical
resource for sentiment analysis and opinion mining. In: The 7th Conference on
International Language Resources and Evaluation (2010)
2. Bae, B.C., Cheong, Y.G., Young, R.M.: Automated story generation with multiple
internal focalization. In: 2011 IEEE Conference on Computational Intelligence and
Games. pp. 211–218 (2011)
3. Gerv´as, P.: Computational approaches to storytelling and creativity. AI Magazine
30, 49–62 (2009)
4. Li, B., Lee-Urban, S., Appling, D., Riedl, M.: Crowdsourcing narrative intelligence.
Advances in Cognitive Systems 2 (2012)
5. Li, B., Lee-Urban, S., Johnston, G., Riedl, M.: Story generation with crowdsourced
plot graphs. In: The 27th AAAI Conference on Artificial Intelligence (2013)
6. Mairesse, F., Walker, M.: Towards personality-based user adaptation: Psycholog-
ically informed stylistic language generation. User Modeling and User-Adapted
Interaction 20, 227–278 (2010)
7. McIntyre, N., Lapata, M.: Plot induction and evolutionary search for story gener-
ation. In: The 48th Annual Meeting of the Association for Computational Linguis-
tics. pp. 1562–1572 (2010)
8. Michel, J.B., Shen, Y., Aiden, A., Veres, A., Gray, M., Brockman, W., The Google
Books Team, Pickett, J., Hoiberg, D., Clancy, D., Norvig, P., Orwant, J., Pinker, S.,
Nowak, M., Aiden, E.: Quantitative analysis of culture using millions of digitized
books. Science 331, 176–182 (2011)
9. Miller, G.: WordNet: A lexical database for English. Communications of the ACM
38, 39–41 (1995)
10. Montfort., N.: Generating narrative variation in interactive fiction. Ph.D. thesis,
University of Pennsylvania (2007)
11. Porteous, J., Cavazza, M., Charles, F.: Narrative generation through characters
point of view. In: The SIGCHI Conference on Human Factors in Computing Sys-
tems (2010)
12. Riedl, M.O., Bulitko, V.: Interactive narrative: An intelligent systems approach.
AI Magazine 34, 67–77 (2013)
13. Rishes, E., Lukin, S., Elson, D., Walker, M.: Generating different story tellings
from semantic representations of narrative. In: The 6th International Conference
on Interactive Storytelling (2013)
14. Sina, S., Rosenfeld, A., Kraus, S.: Generating content for scenario-based serious-
games using crowdsourcing. In: The 28th AAAI Conference on Artificial Intelli-
gence (2014)
15. Socher, R., Perelygin, A., Wu, J., Chuang, J., Manning, C., Ng, A., Potts, C.:
Recursive deep models for semantic compositionality over a sentiment treebank.
In: The Conference on Empirical Methods in Natural Language Processing (2013)
16. Swanson, R., Gordon, A.: Say anything: Using textual case-based reasoning to
enable open-domain interactive storytelling. ACM Transactions on Interactive In-
telligent Systems 2, 1–35 (2012)
17. Toutanova, K., Klein, D., Manning, C., Singer, Y.: Feature-rich part-of-speech
tagging with a cyclic dependency network. In: The NAACL-HLT Conference (2003)
18. Zhu, J., Onta˜on, S., Lewter, B.: Representing game characters’ inner worlds
through narrative perspectives. In: The 6th International Conference on Foun-
dations of Digital Games. pp. 204–210 (2011)