Narrative: Text Generation Model from Data

w w w . i j m r e t . o r g I S S N : 2 4 5 6 - 5 6 2 8 Page 20
International Journal of Modern Research in Engineering and Technology (IJMRET)
www.ijmret.org Volume 2 Issue 8 ǁDecember 2017.
Narrative: Text Generation Model from Data
Darío Arenas1
, Javier Pérez1
, David Martínez1
,David Llorente1
, Eugenio
Fernández1
, Antonio Moratilla1
,
1
(Computer Science/ University of Alcalá, Spain)
ABSTRACT : The generation of digital content has undergone a great increase in recent years due to the
development of new technologies that allow the creation of content quickly and easily. A further step in this
evolution is the generation of contents by automatic systems without human intervention. Thus, for decadesit has
been developing models for the Natural Language Generation (NLG) that allow the transformation of content to
the form of narratives. At present, there are several systems that enable the generation in text format. In this
paper we present the Narrative system, which allows the generation of text narratives from different sources,
and which are indistinguishable for user from those made by a human being.
KEYWORDS –Automatic Digital Content Generation, D2T, NLG
I. INTRODUCTION
Nowadays, and in a general way, we
understand digital content as every data,
information, knowledge or a set of those, capable
of being treated or stored by digital devices, and
transmitted through telecommunication networks.
The format of that digital content is usually
categorized basically in four types or formats: text,
image, audio and video; and if web pages, blogs
social media, or digital newspaper could be
considered in technical terms as sets of elements
from those four basics elements, eventually they
start to treat it as per se formats; for other digital
content as software in general or videogames where
the concept is not that evident define by those
basics formats. Regardless of this kind of
classification, the creation of digital content of
every type has suffered an exponential increase in
last decades.
This tremendous growth in digital content
generation has been affected positively by different
factors. On the one hand, we have the fast
evolution in hardware and software which has
allowed the creation of content faster, easier and
more efficient each time. On the other hand, digital
content has arrived to everyone at final user level,
this implies that each day more people are using
technology and it has been developing more
complex content to satisfy that public, such as web
generation, and treatment of images, audios or
videos. Besides, there are social factors which have
allowed this huge increase in digital content
generation, as we were saying, like the contents
demanding from users which have even created
many moments of saturation in certain content.
From now on we will be referring to text digital
content as information, because it is the approach
of this work. That kind of information which digital
media publish for users consuming in digital
newspaper, blogs or webs for electronic commerce.
For example, those pieces of information can be a
journalist article, a product description, a brief
event explanation, a set of rules or the explanation
of weather predictions. In this way, in this work,
the subset of digital content we are going to deal
with is digital information in text format. However,
the model of automatic generation we are going to
expose in this work is perfectly valid for other
different subsets of digital content, not only for
texts ones.
II. NATURAL LANGUAGE
PROCESSING
Automatic generation of information in
text format is usually approached from Natural
Language Processing (NLP in what follows), which
combines techniques from Artificial Intelligence,
like Logic or Machine Learning, with Statistics,
Applicated Linguistics or Procedural Programming,
in order to develop methods that allow us tasks
such as understanding and computer assisted
processing of information given in human
languages for certain jobs, like automatic
translation, narratives generation, knowledge
extraction from texts, etcetera. NLP is divided in

two large areas, Natural Language Generation (NLG hereafter) and Natural Language
Understanding (NLU hereafter), which
can be considered as the inverse task of NLG. On
the one hand, NLG aim is to generate text
messages in human language from structure and
complete data of the domain we want to describe.
On the other hand, NLU aims is obtain structure
and understanding data, given texts in human
language. In this work, we are keen on generate
narratives automatically, so we are interested in
NLG. It is not a new discipline, and it has support
from other disciplines which implies a lot of
techniques from those fields. NLP was born in the
1960’s, although it was not recognize until the
decade of 1980 [1],[2],[3],[4].
From the works of different researchers, principally
Reiter’s and Dale’s ones (1997), it has been
developing a model in recent years for solving
NLG problems, it is based on sequentially and
well-define stages ([5],[6],[7],[8],[9],
[10],[11],[12],[13],[14],[15],[16],[17],[18]), from
which it has developed specific models to apply to
a large number of fields, such as meteorology,
health or industrial processes [19].
This model is based on six phases or activities
grouped in three stages, so each one works in
different levels of linguistic representation
(semantic, lexicon, syntactic): Text Planning
(activities 1 and 2); Sentences Planning (activities
3, 4 and 5); Linguistic Realization (activity 6).
Those six basics activities which are represented on
Fig. 1 are [20]:
 Contents determination: in this first
activity, from given data, we generate a
set of simple messages summed up in an
intermediate language which distinguishes
labels, objects, concepts and relations of
interest in the domain of application, and
they stablish the information that will be
on the text.
 Speech Planning: we order and structure
the messages to transmit.
 SentencesAggregation: we group the
messages on sentences to improve text
fluency.
 Lexicalization: we choose the specific
words and expressions that we will use to
express the concepts and relations on the
domain.
 Generation of reference expressions: we
select word or expression which identify
objects from the domain, in that way, we
can ensure that the system provides
enough information to distinguish each
object from the others.
 Linguistic Realization: we apply
grammatic rules to provide a right final
text fromlexicon, syntactic and semantic
point of view.
III. TEXT GENERATION FROM
DATA
At present, the different developed systems based
on that model generate two kinds of texts. On the
one hand, those we can name “objective texts”, that
present information about a well-defined domain
(sports, finance, meteorology, products description,
etcetera) and those that usually define the domain
to deal with, in terms of data. On the other hand,
there are “subjective texts”, in which we express
stories with a high semantic weight and they have a
lot of emotionalconnotations, in contrast with
objective text which rarely get it because they
present objective information with no emotional
charge.
In this work we will focus on objective text,
inwhich we find a well-defined domain from a
finiteset of data. In the last years, that has been
named D2T models (data-to-text models). It has
increased the interest of NLG researchers [21] due
to the explosion in data generation, and the easy
access to it and their treatment. Therefore,
researchers have appreciated a practical application
for those techniques that have been developing for
the last years, in order to try to create systems
which permit to generate natural language
SISTEMA GENERACIÓN LENGUAJE NATURAL
Contents
determination
Speech
Planning
Sentence
Aggregation
Lexicalization Generation of
reference
expressions
Linguistic
Realization
Figure 1. Components of a NLG system.

successfully, that have been applied with more or
less success to task like automatic translation of
languages. Our goal is to achieve an approximation
to the model in
Fig. 1 in a context where data are the content
definition, several D2T models in the last years
have been developed, and now, there is a
consensual structure followed by the vast majority
of practical implementation [22].
This model (Fig. 2) is based on 4 activities:
Signals Analysis (it analyzes given data looking for
patterns and trends); Data Interpretation (it
identifies complex messages in the domain from
patterns and trends found in the first activity,
identifying relations among the messages);
Document Planning (it decides which messages
must appear in the final text); Microplanning (it
transforms messages into several linked
expressions) and Realization (those expressions are
transformed into text strings). Later, it has been
modified and used by different authors
generatingsimilar approximations, like in [18]
which propose those activities: Data Analysis,
Content Defining (select and structure the
information given from the data analysis that we
want to transmit), Aggregation and Microplanning
(it transforms information into linked expressions),
and Realization (those expressions are transformed
into text strings). With that model structure as
base,specific D2T models have been developed like
WeatherReporter, FoG[5], MultiMeteo[23],
SumTime-Mousam [24], British Met Office [25],
TEMSIS [26], and MARQUIS [27] in meteorology
field; SUREGEN-2 [28] and BabyTalk[21] in
health field, Patent Claim Expert [29] y Sum-Time-
Turbine [30] in industry field; IDAS [31] to
automatic generation of online documents;
ModelExplainer[32] to generate text description of
software models oriented to objects; PEBA [33] to
entity description in a knowledge base; STOP [34]
to automatic generation of personal letters to give
up smoking. In relation with generation systems for
subjective text there are several systems like
TALE-SPIN [35],GESTER [36], MINSTREL [37],
MEXICA [38], BRUTUS [39], Cast [40].
If we analyze those and other implemented
systems, beyond minimal difference with the model
in Fig. 2 to obtain a specific D2T model, we can
classify those system in two types, those based on
writing scripts, and those which do not use it. From
a formal perspective we can say that both system
should get similar solutions in quality terms, it is
truly that there are some practical aspects that
makeit impossible to obtain successful results
without using writing scripts, such as execution
time, aspects related with journalistic style,
semantic aspects like ambiguities, and those related
with emotional charge in the messages. Given
those aspects is rather difficult to get automatic
messages with D2T systems. Anyway, authors as
Van Deemter[41] have analyzed and compared
those two types of systems and have concluded that
there are no difference in quality results in system
based on writing scripts.
Another relevant aspect in development of that
kind of system, is the validation of obtaining results
and we do that by using evaluation methods to
check if they accomplish the requirements a priori,
which is not easy, because there are requirements
difficult to express and quantify, like those related
with the style in narratives. Others as users or
consumer acceptation can not be established until
checking if the obtained results satisfy their
necessities. Eventually, it is subject of study and
that is a consensual thought in the community of
researchers. In that way, we found that the vast
majority of known methods for evaluation are
quantitative ([42],[43]) and they try to obtain a
numeric measure about quality (making
questionnaires to human experts; similarity
measures among generated texts and representative
texts, etcetera). There are qualitative methods used
to identify and to correct some aspects in content
analysis stages and in speech stages ([44],[45]).
Signal
Analysis
Data
Interpretation
Document
Planning
Microplanning and
Realization
Patterns FINAL
TEXT
…………
Messages
Relations
Selected
messages
Data Events
Figure 2. Components of “data-to-text” NLG component system.

IV. NARRATIVE MODEL
From an applied perspective, the
development of D2T system is justified by the
constant increase on contents demanded by users
(information consumers), eventually this implies
situations in which it is impossible to generate that
amount of information with enough quality because
we do not have the human resources to make it
possible. Thus, it is necessary to ask ourselves in a
general way, if it is possible to automatize text
generation with developed techniques and with
enough quality, that is, showing complete and
accurate information, grammatically correct, with a
proper style, etcetera, which satisfy consumers
requirements about the content.
To achieve an illustration of our problem, we are
going to describe three real examples solved by
Narrative (www.narrativa.com) as an answer to
users demanding, in that case companies, which are
looking for solutions to their digital content
generation problems in text format:
 First one is about sport information
offered by written media. In all terms, a
digital written media can offer high
quality information to users about more
relevant matches in principal leagues of
popular sports. However, when it is a
minority sport or they are lower leagues,
offering high quality information to users
is non-viable in economics terms
(understanding high quality information as
complete an accurate information),
because that requires some task by the
professionals as covering all matches that
is economically non-viable.
 Second example comes from the necessity
of giving a quality description about a
huge quantity of products that offer some
webs for online shopping. In this case,
quality concept acquires more importance
because we do not achieve it only by
enumerating or simply describing
characteristics of the products, we need
appropriate descriptions for each product
in a specific way departing from their
characteristics and other data, so that this
generated information could identify
completely and accurately every product
among others similar, providing useful
information to users or consumers when
taking right decisions. This kind of
information is usually described by an
expert on the area, but for millions of
products, it is non-viable for human
experts to generate that much information
on time and shape.
 The third one deals with the problematic
of these situations that need a quick text
generation and it is impossible for humans
to generate on time. For instance, task
execution orders for people, which are in
base of data, and must be modified in
seconds, such as products delivery
planning. Simply, we have to imagine a
group of truck drivers loading product into
the truck. In headquarters someone is
planning the load and the delivery of the
products, from that moment, drivers have
to receive text orders in voice format,
which indicate them at real time the route
they must be keeping (the route will be
modified depending on traffic, business
interest, etcetera…) and some other useful
information.
Those examples, which can be extrapolated it to
similar situations, illustrate a current situation in
which users demand information with different
quality degrees, and that is impossible to generate
on time and shape, because it requires human
participation in the generation process.It has an
effect on business models, which make them non-
viable and generate the lack of certain useful
information demanded by users at presents. In that
situation, what matters, if possible, is automatize,
that is, to develop automatic systems that allow to
generate that information with enough quality to be
offered to users and consumers. Narrative’s system
answers that question showing the possibility to
generate information in text format like narratives,
from sets of representative data from a specific
domain, such as some of those commented before.
To design our model, we start from the issue that
generating text in natural language from given data,
we must develop systems which have to face three
phases (Dara Analysis, Content Determination and
Realization) with high complexity in scientific
terms, but eventually is possible to develop
approximate methods, principally from Big Data
techniques, statistics and Artificial Intelligence,
which permit to obtain useful solutions and
acceptable to practical endings.

First phase is related to data treatment, that is,
determine, obtain, analyze, classify and categorize
necessarily data needed to obtain the expected
narratives with the required quality, that is,
complete and accurate narratives [3]. Data must be
expressed in data bases and in knowledge bases
which will lead to proper treatments later.
As a general rule, the vast majority of development
system until now approach given data to the model
in plain text, that is, they work exclusively with
structure data, generally numeric. However, as we
can see it in the model displayed and in Fig. 3, this
concept to another and more complex given data
(documents with simple data, documents
elaborated in grammatical and emotional aspects,
images, audios, videos, social media, emails, web
pages, etcetera) from an analysis and extraction
data module, which obtain known and relevant
information with Big Data techniques, with the
goal of getting a final domain represented in a
specific knowledge base, undergoing through an
intermediate process of data bases generation that
will allow to aggregate proper data of the client, to
realize an unification process, and to keep a format
to extrapolate to any other model of knowledge
base later.
In second phase (Fig. 4), from the defined domain
in previous phase, and including client
requirements and other external agents which can
conditionate the final information, then it extracts
the relevant information that is subjected to appear
in the final text, using new analysis and exploring
techniques, principally of Artificial Intelligence,
about the knowledge base in that case.
Finally, once it has been obtained the relevant
information to express in the final text, it
realizesone and last phase to conform the final text
so that it fulfils all the editorial requirements from
the media or the client, such as style or proper
aspects of the language (Fig. 5).
In parallel with second and third phases, the model
is based on the use of written scripts at different
levels of granularity. It is known that humans when
generating text information are capable of
managing with a huge amount of information and
knowledge related with common sense, modelling
information and preparing messages adding
subjective aspects related with context or mood.
Thus, automatic text generation presents lack of
naturality an emotion, and poverty in semantic
content, when comparing with human texts.
Scientifics propose that, nowadays, humans are
capable of making complete and more complex
content than generating by automatic systems
based on NLG techniques, and it will last years,
probably decades, until obtaining automatic
approximations similar in quality terms.
Nevertheless, we hold that, from an applied
perspective, there is a way to deal with that
inconvenient. In those applied fields, there are
experts capable of generating narratives with those
complex aspects mentioned, and that the automatic
system is unable to show them in the final text. It is
possible to realize a process of atomic written
scripts, representatives of the domain, structures at
different levels of granularity, which add all the
emotional, subjective and style aspects, which an
automatic system is unable of. The module of
atomic written scripts is used as base to train the
system and to adapt to final texts, so that they show
information similar to a text generated by experts.
Figure 3. Components of Data Analysis phase.
Figure 4. Components of Content
Determination phase.
Figure 5. Components of Realization phase.

Narrative’s model is based on those three stages
shown in Fig. 1, 2 and 3 and on the module for
managing the written scripts. It is shown in Fig. 6 a
descriptive text of a football match summary
generated automatically by the system from
Premier League provided by the Company Opta
(www.optasports.es).
V. CONCLUSION.
The development of automatic systems for
narratives generation in text format has undergone
a constant development in the last decades, in
research of all the aspects related to natural
language. Nevertheless, we are far from showing
subjective aspects like emotions and complex
semantic of human texts in automatic generation
systems. However, eventually, it is possible by
using the advantages provided in that field and
tools like written scripts, to build automatic useful
systems which allow massive generation of
contents on real time, which is impossible only
with human intervention. In addition, they have
characteristics that make them indistinguishable
from human ones. In this work, we present the
Narrative’s system which correspondsto that
philosophy, and which is used at present to solve
practical problems in different real situations of
content production for objective public who
consume it through web platforms.
REFERENCES
[1] Goldman, N. (1975). Conceptual generation. En R.C.
Schank, Conceptual.
[2] Davey, A. C. (1978). Discourse production. Edinburgh:
Edinburgh UP.
[3] McKeown, K. R. (1985). Discourse strategies for
generating natural-language text.
[4] Appelt, D. E. (1985). Planning english sentences.
Cambridge University Press.
[5] Goldberg, E., Driedgar, N., & Kittredge, R. (1994).
Using natural-language processing to produce weather
forecasts. IEEE Expert, 9, 45-53.
[6] Dalianis, H. (1996). Aggregation as a subtask of text and
sentence planning. Proceedings of the 9th Florida
Artificial Intelligence Research Symposium (FLAIRS-96),
1-5. Florida (EEUU), 20-22.
[7] Not, E. (1996). A computational model for generating
referring expressions in a multilingual application
domain. Proceedings of the 16th International
Conference onComputational Linguistics (COLING'96).
2, 848-853. Copenague (Denmark).
[8] Reiter, E. & Dale, R. (1997). Building Applied Natural-
Language Generation Systems. Journal of Natural-
Language Engineering. 3, 57-87.
[9] Bateman, J.A. (1997). Enabling technology for
multilingual natural language generation: the KPML
development environment. Journal of Natural Language
Engineering, 3, 15-55.
[10] Cheng, H. & Mellish, C. (1997). Aggregation based on
text structure for descriptive text generation. Proceedings
of the PhD Workshop on Natural Language Generation,
9th European Summer School in Logic, Language and
Information (ESSLLI97). Aix-en-Provence (France), 18-
22.
[11] Shaw, J. (1998). Clause aggregation using linguistic
knowledge. Proceedings of the 9th
International Natural
Language Generation Workshop (INLG’98). Canada.
138-147.
[12] Cahill, L. &Reape, M. (1999). Component tasks in
applied NLG systems. Technical Report ITRI-99-05.
Information Technology Research Institute, University of
Brighton (United Kingdom).
[13] Teich, E. (1999). Systemic functional grammar in natural
language generation: Linguistic description and
computational representation. Cassell Academic
Publishers,Londres (United Kingdom).
[14] Theune, M. (2000). From data to speech: language
generation in context. Tesis doctoral, Eindhoven
University of Technology (PaísesBajos), 2000
[15] Reiter, E. & Dale, R. (2000). Building Natural Language
Generation Systems. Cambridge University Press,
Cambridge.
[16] Bangalore, S. &Rambow, O. (2000). Corpus-based
lexical choice in natural language generation.
Proceedings of the 38th Annual Meeting of the
Association for Computational Linguistics (ACL 2000),
(pp. 464- 471). Hong-Kong (China).
[17] Díaz-Hermida, F, Ramos-Soto, A. &Bugarín, A. (2011).
On the role of fuzzy quantified statements in linguistic
summarization. En Proceedings of 11th International
Conference on. Intelligent Systems Design and
Applications (ISDA), 166–171.
[18] Hunter, J., Freer, Y, Gatt, A., Reiter, E., Sripada, S. &
Sykes, C. (2012). Automatic Generation of Natural
Language Nursing Shift Summaries in Neonatal Intensive
Care: BT-Nurse. Artificial intelligence in medicine.
56(3), 157-172.
Figure 6. Narrative’s text example from given
data.

[19] Bateman, J.A. (2001). Natural language generation: an
introduction and open-ended review of the state of the
art, http://www.fb10.unibremen.de/.
[20] Bautista, S. (2008). Generación de textos adaptativa a
partir de una elección léxica basada en emociones.
Trabajo de Máster en Investigación en Informática.
Universidad Complutense de Madrid.
http//eprints.ucm.es/10061/1/ProyectoFinMaster.pdf.
[21] Portet, F., Reiter, E., Gatt, A., Hunter, J., Sripada, S.,
Freer, Y. & Sykes, C. (2009). Automatic Generation of
Textual Summaries from Neonatal Intensive Care Data.
Artificial Intelligence. 173(7). 789-816.
[22] Reiter, E. (2007). An architecture for data-to-text
systems. Proceedings of the 11th European Workshop on
Natural Language Generation (ENLG). Germany. 97-
104.
[23] Coch, J., Dycker, E.D., García-Moya, J.A., Gmoser, H.,
Stranart, J.F., &Tardieu, J. (1999). Multimeteo:
adaptable software for interactive production of
multilingual weather forecasts. En Proceedings of the 4th
European Conference on Applications ofMeteorology
(ECAM 99), Norrk¨oping, Sweden).
[24] Sripada, S., Reiter, E., & Davy, I. (2003):
Sumtimemousam: Configurable marine weather forecast
generator. Expert Update. 6(3), 4–10.
[25] Sripada, S.G., Burnett, N., Turner, R., Mastin, J., &
Evans, D. (2014): A case study: Nlgmeeting weather
industry demand for quality and quantity of textual
weather forecasts. En INLG-2014 Proceedings.
[26] Busemann, S. &Horacek, H. (1997). Generating air-
quality reports from environmental data. EnBusemann,
S., Becker, T., Finkler, W., eds.: DFKI Workshop on
Natural Language Generation, DFKI Document D-97-
06.
[27] Wanner, L., Bohnet, B., Bouayad-Agha, N., Lareau, F.,
&Nicklass, D. (2010). Marquis: Generation of user-
tailored multilingual air quality bulletins. Applied
Artificial Intelligence. 24(10), 914–952.
[28] Hüske-Kraus, D. (2003): Suregen 2: A shell system for
the generation of clinical documents. En Proceedings of
the 10th Conference of the European Chapter of the
Association for Computational Linguistics (EACL-2003).
215–218
[29] Sheremetyeva, S., Nirenburg, S., &Nirenburg, I. (1996):
Generating patent claims from interactive input. En
Proceedings of the 8th. International Workshop on
Natural Language Generation (INLG’96),
Herstmonceux, England. 61–70.
[30] Yu, J., Hunter, J., Reiter, E., &Sripada, S. (2001) An
approach to generating summaries of time series data in
the gas turbine domain. En Proceedings of IEEE
International Conference on Info-tech & Info-net
(ICII2001). Beijing. 44–51.
[31] Reiter, E., Mellish, C., & Levine, L. (1995). Automatic
generation of technical documentation. Applied Artificial
Intelligence. 9, 259-287.
[32] Lavoie, B. & Owen, R. (1997). A fast and portable
realizer for text generation. Proc. of the Fifth Conference
on Applied Natural-Language Processing. 265-268.
[33] Milosavljevic, M. & Dale, R. (1996). Strategies for
comparison in encyclopedia descriptions. Proceedings of
the Eighth International Workshop on Natural-Language
Generation. 161-170.
[34] Reiter, E., Robertson, R., & Osman, L. (1999).
Knowledge acquisition for natural language generation.
En Proceedings of the Joint European Conference on
Articial Intelligence in Medicine and Medical Decision
Making. 389-399.
[35] Schank, R.C. (1969). A conceptual dependency
representation for a computer-oriented semantics. Phd
thesis, University of Texas.
[36] Pemberton, L. (1989). A modular approach to story
generation. En 4th European Conference of the
Association for Computational Linguistics. Manchester.
[37] Turner, S.R. (1992). Minstrel: A computer model of
creativity and storytelling. Informe Técnico UCLA-AI-
92-04, Computer Science Department, University of
California.
[38] Pérez y Pérez, R. & Sharples, M. (2004). Three
computer-based models of storytelling: BRUTUS,
MINSTREL and MEXICA. Knowledge Based Systems
Journal, 17(1), 15-29.
[39] Bringsjord, S &Ferrucci, D. (1999). Artificial
Intelligence and Literary Creativity. En Inside the mind
of Brutus, a StoryTelling Machine. Lawrence Erlbaum
Associates, Hillsdale, NJ.
[40] León, C. &Gervás, P. (2008). Cast: Creative storytelling
based on transformation of generation rules. En P.
Gervás, R. Pérez y Pérez, y T. Veale, editores,
Proceedings of the 5th International Joint Workshop on
Computational Creativity.
[41] Van Deemter, K., Krahmer, E. &Theune, M. (2005). Real
Versus Template-based Natural Language Generation: a
False Opposition. Computational Linguistics. 31(1), 15-
24.
[42] Belz, A. & Reiter, E. (2006). Comparing automatic and
human evaluation of nlg systems. In Proc. European
Conference Association for Computational Linguistics
(EACL’06). 313–320.
[43] Papineni, K., Roukos, S., Ward, T., & Zhu, W.J. (2002).
Bleu: A method for automatic evaluation of machine
translation. En Proceedings of the 40th Annual Meeting
on Association for Computational Linguistics. ACL ’02.
Stroudsburg, PA, USA, 311–318.
[44] Sambaraju, R., Reiter, E., Logie, R., McKinlay, A.,
McVittie, C., Gatt, A., &Sykes,C. (2011). What is in a text

and what does it do: Qualitative evaluations of an nlg
system the bt-nurse using content analysis and discourse
analysis. En Proceedings ofthe 13th European Workshop
on Natural Language Generation. 22 – 31
[45] Reiter, E. (2011). Task-based evaluation of nlg systems:
Control vs real-world context. En Proceedings of the
UCNLG+Eval: Language Generation and Evaluation
Workshop. UCNLG+EVAL’11, Stroudsburg, PA, USA.
28–32

Narrative: Text Generation Model from Data

More Related Content

What's hot (7)

Similar to Narrative: Text Generation Model from Data (20)

More from International Journal of Modern Research in Engineering and Technology (20)

Recently uploaded (20)

Narrative: Text Generation Model from Data