THE NEWSPAPER GAME
Automatic Identification of Text Genres to Enhance Language Learning
Debora Ramalho Barros, Carlo Emmanoel T. de Oliveira
Carla Verônica M. Marques and Cláudia Lage Motta
Institute of Information Technology, University Federal of Rio de Janeiro
Av. Brigadeiro Trompowsky, s/n, Rio de Janeiro, Brazil
Keywords: Text genres, Literacy, Naive bayes, Information retrieval, Genre features, Text linguistic, Genre classes,
Interdisciplinary.
Abstract: Genre has been receiving a lot of thought in Brazilian schools, due to its great importance on language
learning. According to the Brazilian Education Boarder the work with genre and text production is a
possible way to solve its issue on high rate functioning illiteracy. This paper aims to discuss ways of
implementing an intelligent-web based system that analyses characteristics of 8 textual genres found in the
newspaper support media. During the game, the player can build a newspaper and exercise writing different
genres, while a Bayesian agent recognizes genres.
1 INTRODUCTION
Genre has been receiving a lot of thought in
Brazilian schools (Marcuschi, 2008), due to its great
importance on language learning. According to the
Brazilian Education Boarder the work with genre
and text production is a possible way to solve its
issue on high rate functioning illiteracy (Soares,
1998). It’s been proved that the more you are in
contact with different types of texts either producing
or reading them, the more illiterate you become.
Hence, we came up with this project of creating a
text classifier sort of game with the purpose of
helping the production of different genre texts so as
to create context and opportunity for freer writing
practice as well as providing feedbacks on the
writing performance, which shall initially only
concern the genre compositional structure.
Firstly, we view text genre or the style of text as
characterizing the purpose for which the text has
been written. Examples for genre are: research
article, novel, poem, news article, editorial,
homepage, advertisement, manual, court decision
etc. Text-based applications have become more
increased, different aspects of text, such as genre,
can prove useful for various purposes. Not to
mention that, characterizing text differently than the
usual subject or prepositional content, has been the
focus of many information retrieval and
classification research. In this article we address the
issue on automatic detection of the genre class of
text.
Genre classes are clearly different from subject
classes that most classification research has dealt
with. Even though a set of documents may belong to
the same class because they share the common topic,
they often times serve different purposes, falling into
diverse genre classes. As such, classifying
documents based on genre would result in a totally
different outcome than that from ordinary subject-
based classification. From the traditional
information Retrieval point of view, a retrieval
query about a certain topic such as “Sports” would
retrieve many documents related to many different
sources of things when submitted to an Internet
search engine, but they may be of different genre,
such as a sport’s TV channel page, sports news,
product advertisement, or critical review of a certain
game. Genre provides a new dimension for text
retrieval and classification, in addition to topicality,
and help users become more familiar with the
intrinsic structure required in different texts formats.
Automatic genre classification has been studied
in the recent past by Bretan et al. (1997), Dewe et al.
(1998), and Dillon et al. (2000). Karlgren and
Cutting (1994) explored the use of structural cues
and rather simple cues such as counts of third person
pronouns in text with discriminating analysis. In
437
Ramalho Barros D., T. de Oliveira C., M. Marques C. and Lage Motta C..
THE NEWSPAPER GAME - Automatic Identification of Text Genres to Enhance Language Learning.
DOI: 10.5220/0003349004370444
In Proceedings of the 3rd International Conference on Computer Supported Education (CSEDU-2011), pages 437-444
ISBN: 978-989-8425-49-2
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
subsequent work (1996) she investigated the
relationship between the genre of retrieved versus
un-retrieved documents and relevant versus non-
relevant documents. Used features are simple
statistics, such as sentence length and word length,
and syntactic complexity such as average depth of a
parse tree. Identifying text genre would be beneficial
to many text-based applications. For instance, if the
genre of every document is known a priori,
information retrieval results could be better
presented to the user, depending on the preference
the user has. As pointed out by Kessler et al. (1997),
the performance of many natural language
processing tools, such as part of- speech tagging,
parsing, and word sense disambiguation, could be
enhanced since some language usages embedded in
grammatical constructions and word senses are
related to the genre of text. In Web applications,
genre detection would help wrappers that attempt to
extract specified information from semi structured. .
Kessler et al. (1997) identified cues in four
categories: structural cues (e.g. counts of POS tags),
lexical cues (e.g. words used in expressing dates),
character-level cues (e.g. punctuation marks), and
derivative cues (e.g. average sentence length as a
ration and standard deviation in sentence length as a
variation). They decided not to use the structural
cues because of the high computational cost. Their
computational methods were logistic regression and
neural networks (a simple perception and multi-layer
perception) that combine 55 cues.
More recently, Stamatatos et al (2000) reported
on their work for genre detection using word
frequencies and punctuation marks. Instead of using
sophisticated linguistic cues, they attempted to
develop a method that works for unrestricted text in
any domain and language with minimal
computational cost in extracting cues.
Lee and Myaeng (2004) took the stance more
related to traditional information retrieval and text
categorization approaches than to deep natural
language processing for genre identification. Our
text analysis is based on their work.
Features generated from the text analysis are
used for genre-based classification of documents.
Our approach is similar to the one of Lee and
Myaeng (2004) in that they developed their own
deviation-based statistical feature selection method
utilizing subject-based classification information of
the training documents and we are following their
steps.
Having developed a genre classifier, they began
to investigate on the issue of how text genres help
classifying documents based on the subject content
of documents. This is a corollary to our hypothesis
that subject classification would help identifying the
genre class of a document automatically. Thus
helping to enhance writing by providing a feedback
according to the genre they choose to write.
Some experimental results are provided, but the
work included in this paper is quite exploratory in
nature and is still on research.
2 THE GENRE CLASSIFIER
2.1 Overall Method
for Feature Extraction
The genre classifier we used is no different from the
traditional learning-based classifier. The learner
extracts features representing genre classes from
training documents whose genre classes are known
already, and a classification algorithm determines to
which class a new document should belong using the
learned representations of the genre classes. In
comparison with previous genre classification
approaches, furthermore, we agree with Lee and
Myaeng (2004) that the difference lies in the types
of features used as well as the extraction method
itself. The major difference lies in the method by
which features are extracted and their weights are
calculated, but it is still on research, and we are
unable to contribute to it in this paper.
The feature extraction method was derived from
text linguistics theory and observation that the
frequency of a feature (e.g. a noun, a connector, or
verb tense suffixes, etc …) may be high in a set of
document belonging to a genre class, because it
represents the particular genre class. This
phenomenon is likely to happen especially when the
training documents are collected randomly from the
entire document space.
As such our feature selection method shall use
the statistics from two different class sets, genre
classes and subject classes, in the training data.
The weight of a feature for a genre will be computed
based on three factors Lee and Myaeng (2004):
how many training documents belonging to the
genre contain the feature (test 1)
how evenly the feature is distributed among the
subject classes
that divide the genre class (test 2),
how discriminating the feature is among
different genre classes (test 3).
With the first two factors Lee and Myaeng (2004)
found features that are found in as many genre
documents as possible and distributed as evenly as
CSEDU 2011 - 3rd International Conference on Computer Supported Education
438
possible among all the subject classes that divide the
training documents in the genre class.Doing so Lee
and Myaeng (2004) have found out that a good
genre-revealing feature show up across different
subject classes even if a feature appears in many
documents belonging to a particular genre class.
Furthermore, according to Lee and Myaeng (2004)
the third factor ensures good features are as specific
to a genre class as possible by downgrading the
features that happen to occur in several genre
classes. Hence, our experiment counts with the
characteristics implied to the genres we decided to
work with.
As it is an interdisciplinary research project, we
counted on a text Linguistic approach towards the
classifier, taking into consideration a reliable source
of information taken from recognized text linguistics
theories and published data.
In an attempt to teach the classifier what’s
fundamentally determine each genre type, so that it
shall enables a more reliable return of deviating
characteristics when a text is tested. Our hypothesis
is that if trained documents based on genre class
help classifying documents we are able to help
students enhance their writing skills by giving them
a feedback on their work according to the genre they
choose to write.
The feature extraction method was derived from
a sieve for seven deliberately chosen pre-established
genres, which has been created to calibrate the
results given by the Bayes analyses.
2.2 Computation
So far is this research we have done nothing more
simply than store as many genre samples. From the
period of November to August 2010 we limit the
source by using online newspaper as a reference to
gather data for analysis. According to Heckerman
(1996) Bayesian classification needs a huge amount
of data in order to provide as accurate result as
possible, bearing in mind the fact that even the
largest amount of perfect data entry has only been
proven to be 80% solid, Hence what has been
measured by us is enough to show how relevant the
result is to the purpose of the game. In order to
gather different genre text data we used Crawling
and indexing content, that is the process by which
the system accesses and parses content and its
properties, sometimes called metadata, to build a
content index from which search queries can be
served. Meanwhile, we transformed the scrutiny of
the selected genres, which in fact, corresponds to a
contribution of our research. We organized a matrix
for the dimensions for classifying text genres into
taxonomy of genres according to their definitions
found in different sources of teaching materials and
dictionaries. Moreover, it aids in the construction of
attributes for the solidification of the students’ texts
Bayesian classification.
2.2.1 Features
We created a variable grammatical structure based
on one of the six criteria taken from Marcushi
(2008), often used to name genres in Brazilian
Portuguese language and complemented by our
research of genre definitions and concepts taken
from various sources. According to the definition of
those six criteria we organized the content from the
definition of those seven genres into grids, each one
corresponding to a dimension, which shall represent
a weight after the experiments are concluded.
Photo 1: Features of classification.
3 EXPERIMENT
This section reports on our first testing of the initial
Hypothesis that if the learner extracts features
representing genre classes from training documents
THE NEWSPAPER GAME - Automatic Identification of Text Genres to Enhance Language Learning
439
whose genre classes are known already, a
classification algorithm will determine to which
class a new document should belong using the
learned representations of the genre classes.
Unfortunately the feature extraction method is
not ready yet we are still running experiments to see
to what extent several different types of features
(photo1) such as nouns, pronouns, proper nouns,
verb endings, exclamations, and special characters
are useful for genre-based classification.
The experiment results you shall see below
represents the first phase of testing counting only on
a simple naïve Bayes approach, which helped us
visualize the areas in which the system should be
calibrated and extemporized.
3.1 Testing Ground
The documents we used for training and testing were
collected from the Web2 in seven genre classes:
article, chronicle, critical review, horoscope, news
report, recipe and headlines. The collected
documents were classified into subject categories
using the hierarchy to which they were assigned in
the Web sites and the results were examined
manually to correct possible errors. The total
numbers of Portuguese documents collected are
5,000. The documents collected by a linguist and a
computer scientist.
Due to lack of options available just 10 portal
sites were used in the collection building process to
eliminate possible bias toward document types
determined by the Web sites. Each document was
examined by two people for inclusion in the
collection as well as in the designated genre and
subject classes. A half of the collected documents in
Portuguese was used for training and the other
half for testing.
Grid 1 shows the numbers of documents in each
genre class which have been tested.
Genre classes number
articles
375
Reviews
141
chronicle
137
horoscope
1395
recipe
1286
headlines
1360
news report
1621
Grid 1: Stored text data.
Effectiveness has been measured by using real
text data texts taken form a research done with
students in a Brazilian Language school during the
first semester of 2010. As you see in the grid below,
not all of the genres were produced, due to varied
reasons such as the lack of knowledge of structure
and content of some genres, which they have not
seen in school yet.
Their texts were applied to the learner that
extracted the features and provided results, which
determined to which class a new document should
belong using the learned representations of the genre
classes.
Children’s texts number
articles
0
Reviews
2
chronicle
0
horoscope
3
recipe
3
headlines
10
News report
10
Grid 2: Number of text produced by students.
3.2 Overall Effectiveness
The first experiment was only made to see how
effective is the direct application of the Naïve
Bayesian approach to genre-based classification.
Moreover, it has proved it not at all very reliable,
even though it has provided some very accurate
results. However, we are aware of the fact that Lee
and Myaeng (2004) feature selection method is an
alternative of research in order to calibrate these
results, and enabling the game to help on the
production of different genre texts so as to create
context and opportunity for freer writing practice as
well as providing feedbacks on the writing
performance, which shall initially only concerns the
genre compositional structure.
3.3 Results
Surprisingly, just three genres had their results
around 50%, as you can see in the grip below.
Therefore, this lead us to mistrust the whole
experiment itself, due to the fact that having
satisfying results to some of the genres may have
been just fortune and we should not rely on fortune
on scientific experiments. Results, data and figures
must be accurate and by no means refuted.
The classification algorithm which determines to
which class a new document should belong it must
be rather known and calibrated using the learned
representations of the genre classes we provide it
CSEDU 2011 - 3rd International Conference on Computer Supported Education
440
with, so as to being able to automatic return a
feedback with the features the text either should or
shouldn’t have present.
Children’s texts number
articles
X
Reviews
0.5444
chronicle
X
horoscope
0.455
recipe
0.5222
headlines
0.999
news report
0.8766
Grid 3: Results.
An experiment which aims at helping text
production and complement classroom work, should
not only be able to understand the criteria which its
results are taken from, in other words, how the naïve
bayes algorithm was able to imply what genre class
a text belongs. But also explain why it belongs.
All in all this experiment has been the first step
of a long research project which shall next apply Lee
and Myaeng (2004) method in order to interfere and
guarantee a guided result to tests and results by
calibrating the results with the features shown in
photo 1, which has been the investigative work of a
linguist and shall grant this experiment to
more accurate results.
4 ROLES OF GENRE
IN LANGUAGE LEARNING
The linguistic fundaments in this research is also
based on the proposed National Curriculum
Parameters (PCNs) to support the teaching of
language, both oral and written, in the genres of
speech, which has triggered several studies aimed at
describing a considerable amount of genres from
heterogeneous texts as well as providing suggestions
for teaching using texts as examples and reference
sources for a particular genre. For example, the book
edited by Helen N. Brandão (2000), Genres of
discourse in school or the various theories about
gender and their learning, among which we mention:
Deborah C. P. Costa (2001), The construction of
secondary genres in kindergarten: the emergence of
genres and news entry. Campinas: UNICAMP,
Daniella L. Days (2001), Interview by computers: a
proposal to analyze the configuration of gender,
Belo Horizonte: PUC; Lusinete V. de Souza (2001),
The achievements of children: the ordinary lines of
the text view, São Paulo: PUC. (Kleiman, 2010).
Regarding the Portuguese language teaching the
PCNs propose that texts should be worked out
according to the axis USE = REFLECTION = USE,
aiming to "enable the student to expand the use of
language in private bodies in order to use it
effectively in public offices knowing how to take on
the word and produce texts, both oral and written,
coherent, cohesive, appropriate to their recipients,
the objectives will be proposed" (1997, p.41). Thus
the teaching unit to be considered is the text in its
various forms, called by the PCN, gender towards
empirical texts (1998, p.13).
The PCN emphasize the responsibility of all
disciplines to teach students to use texts that make
use a more systematic approach (1997, p.31)
However, when we refer to the work with genres,
including presenting a remarkable distribution cycles
(1997 p.111-112, 128-129, 1998 p.49, 52) The
PCNs are not self-explanatory: "A competent writer
is someone who, when producing a speech, knowing
the possibilities that are offered culturally , know
how to select the genre in which his speech will take
place by choosing whichever is appropriate to their
objectives and circumstances described in question.
For example, if what you want is to convince the
reader, the writer responsible selects a genre that
allows it to produce a predominantly argumentative
text, whether it is making a request to a particular
authority, probably draft a letter (...)" (1997 , p.65).
That is, the document contained only the statement
that the teacher shall choose the most appropriate
genre, since the Curriculum is not a textbook, it
works rather like a guideline, helping teachers
towards the understanding of the light concepts of
linguistic theories as to the didactic in the classroom
in order to develop activities using texts in the
classroom.
The Newspaper game was designed as an
attempt to allow the execution of a task in education
in general, and in the classroom in particular, that
can address the gender perspective in the discussion
here and lead students to generate and analyze the
most diverse linguistic events, whether written or
oral, and identify the characteristics of each genre.
Although always suggested by the literature and
rarely applied in the classroom, due to its high
demand of time, effort and cost, the game works as
an exercise which, besides instructive, also allows
the text production practice.
We believe that production will enable the
student to create a daily newspaper that implicitly do
reflect some of the genre, such as its main
THE NEWSPAPER GAME - Automatic Identification of Text Genres to Enhance Language Learning
441
characteristics in terms of content, composition,
style, language level and purposes. Clearly, this task
can be rephrased in many ways, according to the
intersets of each teaching situation. However, we
believe that it is more modest for the analysis; it will
always be very promising. And even more
promising, will enable the student to the practice of
producing these genres through the game's
newspaper.
Working with genre is an extraordinary
opportunity to deal with language in its various daily
uses. "Within a linguistic perspective everything we
do will be done at some genre. So everything we do
can be treated linguistically in either gender
"(Marcuschi, 1996). The game of the newspaper
brings seven genres that appear in various media
that, are produced systematically and with great
impact on daily life, without excluding the virtual
media to be worked, the internet and computer.
An analysis of textbooks of language teaching
materials shows that there are a variety of texts types
present in these works. However, when we look
more closely reveals that the variety does not match
an analytic reality.
However, we believe that students learn naturally
by producing the various genres written in everyday
use. As it is common to naturally learn the more
formal oral genres, as well observe Joaquim Dolz
and Bernard Schneuwly (1998). Thus, there is not an
ideal type of genre for teaching language. But it is
likely that one can identify progressive difficulties
with gender, level of less formal to more formal,
more private to more public and so on. We decided
to choose the genres at random, just observing the
universe of genres found in the printed newspaper in
our country today.
4.1 The Game
The game platform was developed according to the
web platform called Phidias (Carla et al, 2009a) and,
therefore, designed with three environments: two
interfaces one for the student and the other to the
applicator (teacher), to monitor and receive
evaluations of the process through Bayesian
algorithms, which contains measures of cognitive
requirements (screens) to help the diagnosis and
mapping of the process of textual development. At
the child’s interface, a newspaper similar to the
printed version, composed with empty text boxes
corresponding to the seven genres studied here,
along with the specification of the text required at
each box. The genres were carefully selected from a
variety of genres found in real newspapers here in
Brazil. Each text produced shall be subjected to a
Bayesian analysis which must return a percentage
result of how faithful to the genre is its text,
resulting in a diagnosis on the state of knowledge
you have each genre linguistic composition and
structure.
Photo 2: The game layout of the child’s interface.
The intervention of the applicant (teacher) occurs
after the diagnosed from the results the bayes
analysis, through a method known as Fio Condutor;
in which the teacher evaluates and rehabilitates
mentally the child through continue use of the
Newspaper game. During the sessions, the child can
go through seven stages as shown in photo 3 below.
Photo 3: Fio Condutor (Marques, 2009).
In the first phase, the teacher introduces the
game and the student can explore the possibilities in
part 1, the proposal is that student tells what they
see. In the second everything on screen can move
and they build their own newspaper, while the
applicant (teacher) observes and asks children to
report what they did. This is a first attempt to the
student reflection about their activity. In a third
instance, the child can move across the screen and
rebuilt it and fell free to write the texts. In version 4,
called ‘Elaboração dirigida’, created by psychologist
Seminério (1997), the user is subjected to a
discursive and reflective questionnaire based on the
results provided. In step 5, a new game with a new
CSEDU 2011 - 3rd International Conference on Computer Supported Education
442
set of genres. This project will serve until the fifth
stage, and the subsequent remain as future proposals.
4.2 Theoretical Framework
The conceptual contribution lies in two main topics
related to this study: assessment instruments and
cognitive development, and textual typology. At
first, we point out the theory by Franco Lo Presti
Metaprocessual Seminério (1997), developed based
on the following theories. A. Bandura (1980)
Modeling (Transmission Model or rules) as a means
of promoting learning, which is a result of stocking
about models that do not need to be strengthened at
the time of purchase, Bruner (1966) generative
power or value in order to generate new hypotheses
and combinations. Also called generative rules.
Noam Chomsky (1968) the rule of recursion is the
innate basis for the development of logic and
recursion. Flavell (1963) the metacognitive strategy
and comprehensive model of cognitive control,
Gestalt (1945) dynamism inherent in the cognitive
structuring the state of cognition to metacognition
through insight. Vygotsky (1930) participation in
teacher learning and social environment.
5 CONCLUSIONS
In this paper we show the failure of an Bayesian
analysis experiment which has only helped us to
invest in a better methodology for genre
classification using taxonomy and statistics rather
than just the naïve Bayes approach. The proposed
method uses concepts from selected genre-revealing
features from the textual Linguistic literature. The
deviation formula will make use of both genre-
classified and conceptual features to eliminate
features that can interfere in the classification and
return useful information, so that teachers will be
able to intervene in their students learning processes
in a more effective way.
All in all we plan to have this mechanism
running by the end of this year so that we undergo
school experiments to validate the game next year.
ACKNOWLEDGEMENTS
Our thanks go to Rackel Reis and Maíta Carvalho,
Allan Valente and Yago for either helping us with
the programming part of the experiment as well as
Lee and Myaeng for orienting us with information
retrieve and the Natural language process state of art
available to complement our research and methods
thus introducing us to a great method.
REFERENCES
Moore, R., Lopes, J., 1999. Paper templates. In
TEMPLATE’06, 1st International Conference on
Template Production. SciTePress.
Smith, J., 1998. The book, The publishing company.
London, 2
nd
edition.
Wolkert, Jussi Karlgren, “Web-Specific Genre.
Visualization”, Proc. of the 30th Hawaii International
Conference on System Science, Jan 1997.
Johan Dewe, Jussi Karlgren, Ivan Bretan, Assembling a
Balanced Corpus from the Internet”, 11th Nordic
Conference of Computational Linguistics, pages 100-
107, Copenhagen, 1998.
Andrew Dillon, Barbara A. Gushrowski, “Genre and the
Web: Is the Personal Home Page the First Uniquely
Digital Genre?”, JASIS, 51(2):202-205,2000.
Jussi Karlgren, “Stylistic Variation in an Information
Retrieval Experiment”, Proc. of the 2
nd
International
Conference on New Methods in Language Processing-
NeMLaP, 1996.
Jussi Karlgren, Ivan Brettan, Johan Dewe, Anders
Hallberg, Niklas Wolkert, “Iterative Information
Retrieval Using Fast Clustering and Usage-Specific
Genres”, 8th DELOS Workshop on User Interfaces in
Digital Libraries, pages 85-92, 1998.
Jussi Karlgren, Douglass Cutting, “Recognizing Text
Genres with Simple Metrics Using Discriminant
Analysis”, Proc. of COLING94, Kyoto, 1994.
Brett Kessler, Geoffrey Nunberg, Hinrich Schotze,
Automatic Detection of Text Genre”, ACL’97, pages
32-38, July 1997.
Yong-Bae Lee and Sung Hyon Myaeng, “Text Genre
Classification with Genre-Revealing and Subject-
Revealing Features, Proceedings of the 25
th
ACM
SIGIR Conference, pages 145-150, Tampere, Finland.
D. Lewis and M. Ringuette, “Comparison of two learning
algorithms for text categorization,” Proc. Of the 3rd
Annual Symposium on Document Analysis and
Information Retrieval, 1994.
H. J. Oh, S. H. Myaeng, and M. Lee,A practical
hypertext categorization method using links and
incrementally available class information”, Proc. of
the 23rd ACM SIGIR Conference, pages 264-271,
Athens, Greece, 2000.
E.Stamatatos, N.Fakotakis, G. Kokkinakis, “Text Genre
Detection Using Common Word Frequencies”, Proc.
of the 18th International Conference on
COLING2000, 2000.
Y. Yang and X. Liu, “A re-examination of text
categorization methods,” Proc. Of the 22nd ACM
SIGIR Conference, 1999.
Marcushi, Luiz Antônio, “Produção textual, análise de
gênero e compreensão.” Parábola, 2008.
THE NEWSPAPER GAME - Automatic Identification of Text Genres to Enhance Language Learning
443
Marques, C. V. M. (2009). Laboratório de
neuropsicologia cognitiva-projeto geral: avaliação de
crianças deficientes visuais. Rio de Janeiro:
NCE/UFRJ, 10 p. (Relatório Técnico, 02/09).
Marques, C. V. M; Motta, C. L. R; Oliveira, C. E. T.;
Vrabl, S. D. P.; Lapolli, F., Pereira, A. P. M; Daflon,
L. (2009a). Avaliação de Crianças Deficientes Visuais
através de jogos neuropedagógicos. SCA 2009 -
Simpósio de Computação Aplicada, Passo Fundo, Rio
Grande do Sul.
Marques, C. V. ; Oliveira, C. E. T. ; Motta, C. (Org.). A
revolução cognitiva; um estudo sobre a teoria de
Franco Lo Presti Seminério.Rio de Janeiro:
PPGI/IM/NCE, (2009b). (Relatório Técnico, 04/09).
Travaglia, L. C. (2007) A caracterização de categorias de
textos: tipos, gêneros e espécies. Alfa: Revista de
Lingüística, v. 51, p. 39-79, 2007. ISSN/ISBN:
19815794.
PCN - Parâmetros curriculares nacionais: língua
portuguesa (1998). Brasília: MEC/SEF.
Seminério, F. L. P. (1997). Novos rumos na Psicologia e
na Pedagogia: metacognição: uma nova opção.
Arquivos Brasileiros de Psicologia, Rio de Janeiro, v.
49, n. 3, p. 5-22, jul./set.
Soares, Magda. Letramento: um tema em três gêneros.
2ed. Belo Horizonte: Autêntica, 2000.
CSEDU 2011 - 3rd International Conference on Computer Supported Education
444