The Brazilian Sign Language

Published Date: 02 Nov 2017

and is a language of signs, as used in different countries, presents rules

account for the formation of the signs and the organization of these structures and phrasal

speech.

Unlike spoken languages, the primary organizers of sign languages ​​are

hands that move in space in front of the body and articulate signs in certain

locations in space (SECRETARY OF EDUCATION, 2008).

Research on sign languages ​​have shown that these languages ​​are comparable

in complexity and expressiveness any oral languages​​. These languages ​​express

ideas subtle, complex and abstract. Your users can discuss philosophy,

literature or politics, and sports, work, fashion and use them with aesthetic function

to poetry, stories, drama and humor (FELIPE, 1997).

Thus we have, for the deaf, the LBS is their mother tongue and Portuguese

(writing) the secondary language is one that is perceived difficulty of much

deaf community in learning Portuguese. This difficulty is due, among others,

lack of sense of reasoning based on phonemes as we (listeners) do (associating

the phoneme to grapheme; association is completely arbitrary to the deaf flag).

This type of difficulty is the motivation for the creation of tools that assist

deaf in learning and understanding of texts written in Portuguese.

Moreover, there is also the need for listeners to learn sign language,

primarily educators who must ensure compliance with the Federal Decree 5626,

of December 22, 2005, which states that students must have a deaf education

bilingual, in which the Brazilian Sign Language is the first and Portuguese, in

written modality, the second.

Chapter 1. INTRODUCTION 2

1.1 Purpose

The primary objective of the project is to produce a translator able to analyze a

text in Portuguese and generate a graphical output in an avatar that performs the corresponding signals

in LBS. A translator that we named "Translator Poli-Pounds."

The translator must not only make a word for word translation, but do the considerations

syntax and context to the output approaches the maximum possible

natural expression in LBS.

Besides the basic function and more direct, which would support the deaf to understand

a text in Portuguese, this translator can be used in some contexts or other

Products such as:

? Learning tool to learn POUNDS listener;

? Learning tool for deaf learn Portuguese;

? Generator power for video POUNDS increase the accessibility of websites.

The translation result, it would be corresponding to "text in POUNDS" should also

can be stored in order to create a form of digital encoding of POUNDS

.

1.2 Motivation

As explained, the deaf in Brazil usually communicate by POUNDS,

this being the language of these people, while Portuguese is like a second

language.

Given the significant proportion of this population that has difficulty reading and understanding

texts in Portuguese, it would be desirable to obtain them access content encoded

in LBS.

Thus, this project aims to provide an alternative to this problem.

However, not only the instant translation is desired, but also expects

translation result, coded in pounds, is a product that can be manipulated

Chapter 1. INTRODUCTION 3

by other tools aimed at deaf audience, in order to enable the creation of a

legacy digital encoded POUNDS.

In Brazilian academy there are other projects with similar goals, however,

in general, these are not available to the general public, except for publications

academic articles. Therefore, another goal of the project is the publicity of the results,

including tools produced, so that they are public for use by

all.

The availability to the general public includes not only end users, but any

developers working on related projects, and to meet

this audience another project goal is to develop the system in a modularized,

so that certain modules may be reused independently.

For third parties to take advantage of these modules in the construction of new systems

is also fundamental to our system of release on a free license so that

permits. This is because the situation when something is available and anything that sets the

can be made, the rule of nothing is allowed until they receive specific permission.

In this context arise licenses used in open source software, also called

open source systems (open source), which ensures the user access to códigofonte

system, allowing them to make changes to adapt the software to their

needs, helping with the development or even generate a new product with

Based on the original.

Thus, to encourage and facilitate the production of new tools that meet

the needs of the deaf community, our systems will be published on licenses

free.

1.3 Related Work

In this section talks about other jobs that have similarities with the

proposed project, highlighting the differences, advantages and disadvantages of each.

Dictionary Accesses Brazil

A related work widely used in the deaf community is the dictionary POUNDS

Accesses of Brazil (LIRA; SOUZA, 2008), which is an online tool

query signals from Portuguese words. The dictionary is in fact quite complete,

But our proposal presents several differences. First, the fact translate

complete sentences and not just words. Furthermore, our vocabulary is espansível and

scalable for display output generated computationally, as opposed to video, and

Our dictionary module also presents a formal model based on signals

in XML, so that it can be reused for other applications.

Falibras

Falibras is a project done by the Institute of Computing at the Federal University of

Alagoas (CORADINE et al., 2007) whose purpose is very similar to ours: to make a

Portuguese translator for LBS, with animated output. The project went through Falibras

Various developments, starting with a purely lexical and translation of short phrases,

after implementing parsing and ambiguity reduction, then modularizing

components. There is also an alternative approach to translation

based memory. The problem with Falibras is that despite being very easy to find

several related articles, there is nothing available to the end user,

neither the developers. There are tools that actually help to community

deaf in general, and that is the difference with our project, provide tools

for the open-source community.

TLibras

Another translator Portuguese-POUNDS we find is the TLibras (LIRA, 2003), project

OSCIP coordinated by Brazil and accessibility that has 3 different teams:

one of LBS, the FENEIS (National Federation of the Deaf Education and Integration)

one of NILC-USP for Natural Languages ​​(Interinstitutional Center of Linguistics

Computational) and computer graphics, the very accessibility Brazil. This project

has more than 20 people, which demonstrates the complexity of the present proposal

work. They planned three stages, beginning with texts in Portuguese, adding recognition

voice in the second, and the third in implementing digital TV.

The process of translating the TLibras uses the UNL (Universal Networking Language)

(UCHIDA, ZHU, 2001) to bridge between languages. The UNL is a proposed

intermediate language for translations, an interlanguage (top of the pyramid of Vauquois cited

Conceptual Aspects of the section) to which all languages ​​and can be translated

be translated. That is, to use the UNL, the project team had to TLibras

basically do a mapping of various texts POUNDS to UNL, including

morphological information, syntactic and semantic, and then the bridge would be made because the

UNL-Portuguese translation already exists.

A module is quite interesting TLibras which proposes creating primarily

a 3D avatar that speaks POUNDS, which could be used in various applications through

an entry in the Notation-Pounds (which was not specified in the article read). However,

This project uses technologies not easily compatible with the Web, and as our aim

is to have a translator available via internet, using this avatar would be infeasible (always

would be generated videos, which consume a lot of resources).

However, one difficulty we had was the fact finding nothing available to

public of this project, though its release was scheduled for 2004.

All we found was a story of Monica's gang translated that into (LIRA, 2003)

that was referenced as a restriction of the initial stage, the only text that would be transcribed

UNL was for a small comic with 5 sentences.

Rybená

The player Rybená (FERNEDA; COSTA; ALMEIDA, 2003) is a web tool that

promises to translate texts of web pages for LBS through his animated avatar.

A highlight of this work is the possible integration with mobile phones to send

torpedo, although only 4 handset models are currently supported 1.

This translator was the only one who could actually test, as there is a version

functional on your site. However, we found that its use is not very intuitive and

translation was performed word for word, creating the so-called Portuguese signaled what

syntactically very different from LBS and therefore, the deaf tend not to accept.

Not to mention that the Rybená is a commercial product, disagreeing with our free philosophy.

1.4 Organization

This document is divided into the following sections:

1 According to the Rybená own site. Available at: http://www.rybena.com.br/produtos/

devices_list.jsp? ckRybena = marked. Access on 05/12/2010

Chapter 2 - Conceptual Aspects: the basics are set to be discussed

throughout the document, the characteristics of LIBRAS, concepts

of formal languages ​​and translation methods;

? Chapter 3 - Technologies Assessed: studies on APIs and existing tools

that have been evaluated and, among some of them, used in the project;

? Chapter 4 - Project Specification: In this section you define the requirements of

design, use cases and explanation of some decisions. In particular definise

the architecture of the system with its divisions into modules, specifying

the functions each including a template definition herein of data

encoding the signals;

? Chapter 5 - Implementation: This chapter describes the methodology and

technical details on the implementation of the modules specified in the architecture;

? Chapter 6 - Testing and Evaluation: This chapter describes the main tests performed

to validate the functionality of the translator;

? Chapter 7 - Conclusions: This chapter assesses the project's results and highlights

future projects that will be based on this project;

? Appendix A - Grammar free-of-context to model the Portuguese;

? Appendix B - Organization of enclosed CD.

Chapter 2

CONCEPTUAL ASPECTS

2.1 POUNDS

The Brazilian Sign Language is the language used by deaf people to communicate in

Brazil. Sign languages ​​are not much different from spoken languages, being as complex

and expressive as they are. The LBS is present at all levels of analysis of

other languages, but also have their own grammar, its own syntax and its own

vocabulary. The main difference compared to languages ​​spoken is the fact of being POUNDS

a language visual-spatial, which is done through gestures and expressions with hands, head

and body, perceived by sight. It is therefore different from Portuguese for example, which is a

auditory-oral language uses sounds as perceived by hearing.

2.1.1 The morphology of the signal

The formation of a signal is defined basically by five parameters, which is four

refer to the hands: configuration, location (or pivot point), motion and orientation

palms, and a fifth that refers to the use of non-manual features, which

include facial expressions, mouth movements, gaze direction, which allows the expression

a significantly greater number of linguistic information.

The configuration of the hand refers to the shapes of the hands, which may be the dactylology

(Digital alphabet) or other shapes made by the dominant hand (right hand for righties)

or by both hands. Since the location is the place in the body or space, in which the signal

is articulated, the hand can touch any part of the body or be in a neutral space.

The movement involves both the internal movements of the hand, the wrist movements

directional space, and the set of movements in the same sign. The orientation of

palms is the direction in which the palm of the hand points to produce the signal.

It may be upward, downward, into the body, forward, to the left or

right. Finally, traces involve non-manual facial expression, body movement and

look. Signs like CUTE, cute and handsome, for example, are represented

the same configuration, position, orientation and movement of the hand, however have

different facial expressions.

2.1.2 Conventions Written form

This section describes the conventions we adopt throughout the paper as

the representation of LBS in written form.

One. Signs in POUNDS are represented by uppercase letters in Portuguese. Exs:

CASA BALL.

2nd. Two or more words separated by hyphens represent a single signal. Exs.: CORTARCOM-

KNIFE, EAT APPLE

3rd. When a word is represented by Datilogia (manual alphabet), it appears

separate letter by letter. Exs.: M-A-R-C-E-L-O

4th. Number agreement / staff or place is made with elements subscribed to

signals. Ex: JOÃOa MARIAb aDARb THIS (the verb to agree with the points

b, which refer to John and Mary, respectively).

5th. Facial expressions / body and adverbs of intensity are given envelopes

the signal.

Exs: ANDARrapidamente, S IMbalanca head ô€€€

6th. Why do not usually have an ending to differentiate between male and female

and singular or plural signs, they will be represented with an @ in the end.

So AMIG @ means "friend", "friend", "friends" or "friends."

2.1.3 SignWriting

The notation for writing in POUNDS described in section 2.1.2 is the alternative usually

used by several articles and studies in this field, since there is no

official form of graphic representation. However, as is evident, this brings notation

little or no information about the morphology of the signals used, this

is from the lexeme CASA is impossible to imagine how it would be a corresponding signal, in

contrast, the Portuguese words are formed by a sequence of syllables

represent sound units well known, allowing the reader to "synthesize" the sound

the word, even without ever having seen it before.

But there is also a notation write enable signal language form

clarify the morphological features of the signal, being named SignWriting, and that can

be used to write any signs of sign language, the example of our alphabet

Roman also allows the writing of several different languages.

This system was invented in 1974 by American Valerie Sutton, a dancer who

two years earlier had developed DanceWriting.

With a symbol SignWriting you can specify characteristics of the signal as

position, shape of the hand, the finger joints, facial expressions and body etc.. an example

SignWriting of text can be seen in Figure 2.1.

Figure 2.1:

Figure 2.1: Example of text SignWriting

The system began to receive attention in Brazil since 1996, as one of

most relevant works on the subject Illustrated Encyclopedic Dictionary Trilingual (Capovilla;

RAPHAEL, 2001), whose author is the psychologist responsible for the adaptation of SignWriting

for POUNDS; another important work is the Manual SignWriting (SUTTON, 2003),

which is an American adaptation of the work originally written by the Valerie Sutton,

being focused on showing SignWriting in the context of LBS.

One of the main advantages of SignWriting is to enable deaf to write in

their own language, without having to resort to phonetic writing oral language to

symbolize the signs. However today the acceptance by the community of notation

deaf is still low, being considered one of the main reasons that its complexity

makes it difficult to learn.

Another fact noted by the group is that in practice several signs of LBS are difficult

to write in SignWriting the difficulty of determining the values ​​of attributes

forming the symbol SignWriting, and this, we believe, can also be a reason for

not receptive to SignWriting.

Also joins the fact that, unlike pounds, a creation SignWriting

artificial, made by listeners as a solution to be assimilated by the deaf (although

the assumption that there is some involvement to an enhancement of this SignWriting

sense).

2.1.4 Classes grammatical

Classes or categories correspond to grammatical paradigms on a body of

words. Through these new paradigms are obtained from other lexemes. There

Also paradigms connecting elements different parts of speech. The example

cites to number agreement between nouns and verbs.

The set of grammatical categories of LBS is well mapped in Portuguese,

there is only one in each category is not present in another. The following observations

on categories of LBS (FELIPE, 1997):

? Verbs

Basically, there are two kinds of verbs: those who do not agree and that

form (also called plain and non-plain, respectively). Those who do not agree

are always in the infinitive, being simpler.

Ex: U.S. OFFICE WORK.

As for those who agree, they can do so in three ways:

One. Concordance number-staff:

The orientation mark people's speech. The starting point agrees with the

subject and the final, with the object.

Eg = 1sPERGUNTAR2s "I ask you" is different from 2sPERGUNTAR1s

= "You ask me"

2nd. Gender agreement (person, animal, thing): hand configuration changes;

Eg for signs of the verb walk, "walking person", "car walking", "animal

walks "

3rd. Compliance with the location: occurs with verbs designating actions

begin or end at a certain place. The simplest explanation is

through instance, to indicate that something was placed on the table work must,

before signaling the verb 'put', represent the table at some point

articulation, the sign of the verb must then be carried out having as destination

the pivot point signal table.

? The classifiers are classifiers settings hand, replacing

name that precede them, can come along to the verb to classify the subject or object

which is linked to the action of the verb. So classifiers are markers in POUNDS

of gender agreement: PERSON, ANIMAL, THING. the classifiers

ANIMAL PERSON for and can have plural, which is marked up to represent two

people or animals simultaneously using both hands or making a motion

repeated for the number.

? Article in Portuguese If there is no direct counterpart of classifiers POUNDS

In this language there is no corresponding article.

? pronouns

One. personal pronouns

This subcategory of pronouns in POUNDS not only has the ratings

singular and plural, but to indicate more precisely the right amount

many people's speech. For example, there are signs for: EU, U.S. 2,

WE 3, 4 WE, WE-WE-ALL and GROUP. The same goes for the second

(YOU) and third (IT) people.

Full list:

first person (singular, dual, trial, and plural quatrial): U.S., WE-2, WE-3,

WE-4, GROUP WE-WE-TOD @;

(b) second person (singular, dual, trial, and plural quatrial): YOU, YOU-2,

YOU-3-4 YOU, YOU GROUP-YOU-TOD @;

(c) third person (singular, dual, trial, and plural quatrial): @ EL, EL @ -2, EL @ -

3, EL @ -4, @-EL GROUP, EL-TOD @ @

2nd. Demonstrative pronouns / adverbs of place

The demonstrative pronouns and adverbs of place have basically the

same signal on LBS. @ EST / HERE, @ ESS / and THEN THAT / THERE are made

pointing to the appropriate locations accompanied by a look at the

same.

3rd. interrogative pronouns

The pronouns WHICH, WHO, WHY, WHAT are generally used in the early

sentence, while WHAT, HOW, TO, WHAT, WHERE and WHO (meaning

"Who is?") Are used in the end.

Where both the pronouns question words or there is a

facial expression indicating that the sentence is in the interrogative form.

? adjectives

Adjectives usually come after the noun they qualify and represent the

characteristic iconic form.

? adverbs

Adverbs of intensity or so and have no signs themselves are indicated

modifying the speed or repeating several times the sign referred to.

Since there is no time stamp in verbs, adverbs TODAY PAST (or YESTERDAY

Yesterday) and FUTURE (TOMORROW) are generally used in the early

sentence to give this idea.

2.1.5 syntactic structure in POUNDS

According Tables (1999) order based on LBS is present SVO (Subject-Verb-

Object). Order base, for any reason that this structure could be understood

order and other orders that can be derived from this and not otherwise. this not

means that the most common order is SVO. There are many situations in which the

OSV order is better (in which the phenomenon occurs topicalization, which consists in

highlight the issue first, contextualization) or SOV.

Briefly, in POUNDS allowed to SVO, SOV and OSV (the latter two with

restrictions). These constraints are, in general, due to the requirement of concordance

between verb and subject / object and the presence of non-manual markings.

SOV is preferable figure in sentences where certain "iconic" as in PIE WOMAN

PUT-IN-OVEN. That is, there is shown first and then the cake is placed in the oven.

This order is useful to first locate themselves objects in space and then

signal action, linking them.

Eg JOAOa MARIAb aDARb BOOK.

OSV is also widely used (many point to as the most common order, as

BRITO (2006)), because it represents the topic-comment order, allowing first that

explain the context for the caller then explain the action that occurred. It is quite

Common since there are no restrictions on its use.

2.1.6 Degrees of Complexity POUNDS

Considering all these characteristics and difficulties, we set the

that would be a definition of LBS in different complexity levels, where each level

adds an extra layer of complexity to the language. This scale was divided into

2: one only analyzes the signal and its internal aspects (lexical level) and other analyzes

sentence structure as a whole (syntactic level). The purpose of this scale is also serving

guide to an iterative implementation of labor, increasing the complexity of each

level.

Figure 2.2 shows the grading level of the word and figure 2.3, the level of the sentence.

Figure 2.2: Graduation POUNDS - word level

Figure 2.3: Graduation POUNDS - sentence level

2.2 Linguistics

2.2.1 Grammars for Natural Languages

A formal language is a set of strings, where each string can be formed

combining the symbols of an alphabet?, then forming a subset of? * (all

possible combinations of symbols). A grammar is a set of rules that

define the formation of these chains, so a grammar defines a language. So

being, there is an equivalence relation between grammars and languages, as one

there is always at least one matching the other.

The linguist Noam Chomsky in 1959, defined a hierarchy with four types of grammars

(Languages) and the type 0 (recursively enumerable languages) that no

presents no restriction and each subsequent type 1 (context sensitive languages),

2 (context-free languages) and 3 (regular languages) more restrictive than the

above. That is, both for grammars for languages ​​such as: Type3? type2? type1?

tipo0. (Chomsky, 1959)

Besides the correspondence between languages ​​and grammars, there is correspondence

those with recognizers which accept sentences belonging to the corresponding language

and not reject the sentences corresponding to the language; matches

are regular languages ​​$ finite automaton, context-free languages ​​$ automaton

stack; $ recursively enumerable language Turing machine.

A grammar must be able to produce all sentences syntactically possible

the corresponding language and must be unable to generate sentences syntactically

invalid for the same language. However, because of the enormous complexity of languages

natural, you can not find a perfect grammar that represents a

language, and any grammar defined in the context of natural languages ​​approximation

of reality, ie, generate and recognize some sentences invalid.

Jobs linguists have sought to analyze the syntactic structure of sentences through

syntactic tree (FIORIN, 2005), which leads us to the use of free grammars

context because these correspond to stack automata, which are able to recognize

structures in trees. Chomsky himself has also developed the theory of grammars

transformative to perform analysis and natural language processing syntactic

based on context-free grammars (Friedman et al. 1971).

CHAPTER 2. CONCEPTUAL ASPECTS 16

Furthermore, the use of context-free grammar for describing a natural language,

What at first appears to correspond to an unrestricted grammar, is given by the fact that the

use of higher types imply a very high computational complexity for

sentence recognition, making its use impractical in applications where the

processing time is important, for most of them.

In (LUFT, 2002) we have a job of a linguist who seeks to generate a description

the Portuguese with the use of production rules of context-free grammars also

presenting the syntactic structures analyzed in the form of trees.

2.2.2 Grammars transformational

The theory of transformational grammar was created and described by Noam Chomsky in

his book Aspects of the Theory of Syntax (Chomsky, 1965) that seeks to create a

general theory of linguistics showing the inherent substrate common to all languages

Human (this set of standards is the very foundation "language", the innate ability

any human being to learn their mother tongue) (LYONS, 1970).

Chomsky makes a significant step towards formalizing the theoretical area

language, however this still falls short of formalization necessary to use

on computers. Thus, researchers sought to interpret the computing

the proposal of Chomsky, filling gaps and making changes when necessary

something to make it computationally tractable. The first proposal of formalizing

Transformational grammar was presented by the book (Friedman et al., 1971), written

in 1971.

Originally Chomsky describes three types of language representation:

? Generative grammar (linear generation of sentences)

? Phrase structure grammar (generating sentences with nesting)

? Grammar transformative

A transformative grammar consists of:

? Sentence structure

? Dictionary (lexicon)

CHAPTER 2. CONCEPTUAL ASPECTS 17

? Transformations

The phrase structure grammar is a context-free ordered in the same way that

context-free grammars, but with some additional restrictions.

The nodes of the trees, which describe the syntactic structures may be qualified

for complex symbols.

A complex symbol is formed by a list of "feature value +" where value =

*, +, -, * Being a sign of uncertainty, the obligatory presence of + and - of

mandatory absence (and presence of ausença featuure).

Features can be of the following types:

? Category: verb, noun, article etc.

? Inherence: subjective qualification, as HUMAN ABSTRACT, THING, ANIMAL,

ANIMATED etc..

? Context: describes a tree that is a subtree in the parse tree, can

obligation to determine the presence or absence of a particular element (qualified

by feature class or inherently).

The dictionary transformational grammar comprises a set of definitions

features of a redundant set of rules and a set of lexical entries,

each of these inputs being formed by a vocabulary and a complex symbol

(Vocabulary is formed by several words).

The lexical insertion is the process by which one enters the vocabulary tree nodes that

so permits. The insertion must be done through the analysis of complex symbols (a

word has a complex symbol, which must be compatible with the complex symbol

node that will be inserted).

Analysis for transformation to occur, a particular structural description

(Structural description) must be present in the tree; beyond said a test is carried out

inclusion complexes between symbols (from the description of the tree structure).

For a lexical insertion occurs, a contextual feature must be determinda

present in the tree, and besides if one performs compatibility testing complex symbols

the lexical entry and element (element node is where lexical insertion occurs)

Example of transformation:

PLADEL TRANS / / ID

SD% 2 INDEF N |-SG |%. / / Description of the structure

/ / Which applies the transformation

SC ERASE 2. / / Structural change applied

Changes can also be conditional change. Looks like this: IF restriction

THEN ELSE shift change.

Besides the transformations, the third component of transformational grammar can

also count on a control program, which is actually a program whose instructions

are as defined transformations, namely the control program is a form of

order transformations and determine on which points they are applied.

Examples of control programs:

1) CP PASSIVE. / / PASSIVE applies the transformation

/ / Change is made if any sub-tree

/ / Condiza with the structure defined in the transformation.

2) CP PASSIVE, FLIP, regdel. / / Apply various transformations

3) CP I. / / Apply all the transformations of the set I

/ / For each existing tree S

/ / This is done cyclically until

/ / Can not operate any more change

As a programming language, the driver also has

control elements, which are: IN, RPT (repeat), IF, FLAG, GOTO, TRACE, STOP.

It is based on these studies is that it is perceived that the transformative grammar

is a powerful tool in the analysis of syntactic structures and already considers

Predicted various aspects relevant to translation processes, an example is

in fact the verb POUNDS have an agreement as to be subject THING, ANIMAL OR

MAN. This is a case here to be treated with the features of inherence.

CHAPTER 2. CONCEPTUAL ASPECTS 19

It becomes clear here that a translator itself can actually be in a program

Control transformative grammar that would invoke the appropriate transformations,

appropriate times in the appropriate paragraphs.

2.3 Machine Translation

The problem of the translation of a text into another done by machines is a problem with

various solutions adopted by many pesquidores (Vauquois, 1976). One approach

was widely used for the translation rules, namely rules are defined morfogógicas, syntactical

or semantics to bridge the gap between the two languages. This method requires high

knowledge of languages, and large participation of specialists in them to create the

rules. Currently, it is said that we are in the "era of statistical translation," which follows a

different approach by the rules. This "new era" were a basis of statistical models

whose parameters are derived from the analysis of a language corpus, consisting of several

texts in the two languages ​​(Brown et al. 1990) and is widely used for solutions of

translation today (LOPEZ, 2008). However, as further explained in section 4.3

the statistical method is not the most suitable for the case of translation POUNDS.

Within the field of translation rules based on a scale of complexity level

and sophistication can be described by pyramid Vauquois (Vauquois, 1976)

(Also called a triangle Vauquois) which has a slightly modified version

shown in Figure 2.4. It has levels for the translation, and as

higher in the pyramid, the more "deep" is the analysis of language and therefore better

translation would be performed. In the case of this figure, the original pyramid was divided into

four levels. The first level, morphological, is only based on a translation

words, ie based on only one dictionary. This type of translation is poor because

disregards any sentence structure or relationship between words. The second level is

the syntactic analysis that performs the syntax of the language, then considering these aspects

mentioned. The third level corresponds to a study of semantic features

text, taking into account the meaning of words, enabling a reduction of ambiguity

below present levels and identifying linguistic expressions. The top

pyramid would be a perfect translation, using a single intermediate language in which

the other languages ​​can be described completely.

Figure 2.4: Pyramid Vauquois - a modified version

Chapter 3

TECHNOLOGIES EVALUATED

In various parts of the project we rely on existing technologies in order

for more agility in development and greater quality in the final work obtained,

striving to leverage the existing solutions to solve particular problems

more peripheral our system, i.e. those which are not directly related

translation algorithms.

In this section we describe the technologies that were examined in this context, being

some selected and used by us.

3.1 Morphological Analyzers

Were raised possible morphological analyzers for Portuguese to

which to base the implementation of some morphological analyzer that satisfied

raised to the interface, which handles relations between syntactic analyzers and

Morphological.

Both analyzers were evaluated and the MXPOST JSpell. We studied the complitude

the information provided by each facility and the degree of integration planned for

with the other modules of the project. The conclusions were:

MXPOST

The good of truth, unless a morphological analyzer, a program is MXPOST

labeler (tagger), ie, it applies labels to the words they are presented, so

based on a set of training data that it provided a learning

supervised. Using the work "Taggers NILC's" 1 in which researchers applied

MXPOST the training of morphological labels attached to texts in Portuguese, one may

use the program to discover the morphological classes of words in the language

Portuguese.

1 NILC's Taggers. Available at: <http://www.nilc.icmc.usp.br/nilc/tools/

nilctaggers.html>. Access on 05/12/2010

CHAPTER 3. TECHNOLOGIES EVALUATED 22

The results presented MXPOST high success rate in tests.

An example pair of input / output MXPOST trained:

Input: The red cars hit the post.

Output: Os_ART carros_N vermelhos_ADJ bateu_VERB no_PREP + ART poste_N

However, the MXPOST has some limitations that led us to preteri it.

Firstly, it would be easy to use it as a library: the existing classes had

little intuitive names and found no documentation beyond also not be

open-source and therefore not being able to see your source code in Java. Furthermore, the

information provided by the program are limited and may be insufficient because

it only returns the morphological class: no returns gender, number, or tense and

transistividade for verbs

JSpell

Another alternative was the JSpell2. This is a morphological analyzer code

open, which presents greater opportunities and information in its output.

For example, given a lexeme entry if it matches more than one interpretation,

all are presented. Even if there is no derivation itself to the lexeme

(Ie it is a word unknown to the JSpell), it returns a set of possible

solutions. In English, these solutions are approximate sequences of letters

calls near misses, and is just as JSpell the means; during the project described

herein, however, the nature of the information represented, treated

As the "approximate solutions".

The above example shows how the JSpell is complete varieties over the universe

around a lexeme, but says nothing about how the lexeme itself is described by the tool.

An explanation follows: for each possible interpretation, to be exact lexeme,

referring to approximate solutions, it informed the dictionary form of the word, its

categoriaEXEMPLOS p .. or link description and information relevant to the category (eg: for

verbs, one pertinent information would be the number, person, etc., while for nouns,

gender, etc.).

2 JSpell. Available at: <= http://natura.di.uminho.pt/wiki/doku.php?id

tools: jspell>. Access on 05/12/2010.

CHAPTER 3. TECHNOLOGIES EVALUATED 23

The only problem is that it was made into Portuguese of Portugal. A big advantage

JSpell of an outflow with more information:

? If there is doubt or several possibilities for a particular word, it shows all

options;

? Gender and number;

? Tense;

? Identification of suffixes and prefixes of compound words

3.2 Tools Graphics

To implement the synthesizer signal is essential to use computing technologies

graphically, these tools being divided mainly into two categories:

modeling tools that generate three-dimensional models and APIs that handle

these models at runtime.

Below is a listing of the tools researched and considered in this process,

including tools that fall into one of two categories or even

both.

Flash

At first we considered the possibility of using technology to be a

widespread, both among and between users and developers up by

observe translation solutions for LBS (CORADINE et al., 2007) using Flash.

However, with the use of Flash more suitable for 2D, it was observed that although

would be relatively easy to create an animation of a talking avatar POUNDS, would

create a very complex model with segmented morphological units

so they could be synthesized and integrated at runtime, for this type

Task 3D models show themselves much more appropriate.

CHAPTER 3. TECHNOLOGIES EVALUATED 24

Blender

Blender is a suite of 3D content creation open source, available under the GNU

General Public License (GPL) for all major operating systems.

Besides allowing the modeling of 3D objects, also contains a built-in game engine

allowing the manipulation of objects in real time, including using

scripts in Python (which is equivalent to the tools that define the type gráfiicas

API).

In this regard, one of the latest tool was made in the production of a game,

Yo Frankie!, available for download, including the project files of Blender.

The site of the tool has ample documentation, so we start

learning through their tutorial on the wiki site 3. Furthermore the group held

purchase book Mastering Blender (MULLEN, 2009), which discussed in depth

issues related to the game engine, since the user community in general can find

more focused on how to use Blender 3D modeller.

Following these sources we build an animation of a hand, being built

based on a hand model downloaded from the Internet, the model contained only the

"Skin" and we created the armor structure (bones) to the model and the manipulation

Armour was generated video with animation.

Figure 3.1: Sample screen of Blender

One disadvantage of using Blender is that there seems to be no way of

integrating the result into an environment like the web or even generate a stand

alone without the user himself has Blender installed on your machine.

Another supposed disadvantage is the high learning curve of Blender feature

for which it is famous. Shows is that the community recognizes that

Blender is not designed to be easy or intuitive, but to be a tool

high productivity as soon as you master it. However in our experience with

environment, this difficulty ended up being much lower than expected due to this

overall impression.

Processing

It consists of an open source Java API for generating interactive applications

3D simplest usage generally didactic, which allows direct manipulation of models

OBJ format text, supported by virtually all 3D modeling tools.

Its main advantage is the ease of learning, but has serious limitations,

the impossibility to handle designs which have bones that are used

to create linkages and movement restrictions between parts of the model to be animated;

in general it can be said that is not suitable for handling animated characters,

although sufficient for practical and other graphics applications, such as puzzles by

example.

Another advantage is that the resulting application is actually an applet, or one

Java application integrated with web browsers.

Panda3D

Like Processing, Panda3D is also an open source API for manipulating

runtime of graphical models.

Its advantage is that it is a professional tool, has been developed and

used by Disney and Carnegie Mellon University's Entertainment Technology Center.

One of the main advantages of Panda3D Processing is about the possibility of

dealing with the concept of bones at runtime.

CHAPTER 3. TECHNOLOGIES EVALUATED 26

Although it has a very small user base, its developer community

proved very open to newcomers, providing aid to those who really friendly

wish to dominate technology, contrary to what typically happens in the community

Blender, which prefers appreciate the extensive documentation.

It also presents an advantage over the game enngine Blender, which has no

documentation as wide as compared with the total application documentation as

a whole and also because of the Blender game engine is the most immature of

tool.

But as Processing is only API would still need a modeling tool,

so that a good combination of Panda3D as would be the API and Blender

as modeling tool.

3DStudio

As Blender is also a 3D modeling environment. The main difference

is that the 3DStudio be a proprietary tool and face, but at the same time,

have a much greater market penetration.

Both Blender when 3DStudio can generate models compatible with Processing

and Panda3D.

Comparison

In the table below we summarize the main characteristics of the computing tools

Graphical considered.

Note1: considering the language of the table, it would be more advantageous for us

Java would, by now possess a larger domain and the rest of the project is in Java;

Note2: not listed in the table, but remembering that Panda3D still has the advantage

Additional warmth of your community.

With this analysis, we conclude that the principle would be the best combination using

Blender for modeling and the Panda3D for the API, however, for reasons that will be

explained in the section that deals with the implementation of the system, already anticipate that the combination

chosen was the 3DStudio for modeling and for the Processing API

Table 3.1: Comparison of computer graphics technology

Flash Processing Blender Panda3D 3Dstudio

Modeling Yes (2D) Yes No No Yes

API Yes Yes Yes Yes (at best) not

User base Large Medium Small Small Large

Documentation Small Medium Large Wide ?

language Action Script Python Java C + + / Python -

Integrating web Yes No Yes Yes -

Learning curve Medium Large Small Medium ?

Open source No Yes Yes Yes No

Web 3.3 Frameworks

For the development of modules web technologies that could be studied

and used to provide a gain in productivity, preventing loss of time with the

writing code repetitive Chamdo the boilerplate code, whether the construction of interfaces and

manage navigation between pages, as in automating the mapping

object-relational to store information in databases.

Java EE - Servlets

(Java EE or J2EE or Java 2 Enterprise Edition, or Java Enterprise Edition in Portuguese)

is a platform for server programming in the programming language

Java.

The basic technology for the processing of requests are the Java EE platform

servlets, while the interface is mainly generated by JSP (Java Server Pages).

Servlets are basically user-defined classes that performs processing

the web application and generates the HTML page response, or more properly,

Arrow parameters and forwards the request to a JSP page that is more appropriate

for interface design. The part of the system that manages the lifecycle of servlets and

they forward the requests coming to the server is called container, and the

best known open source implementations Tomcat and Glassfish.

In general, the most advanced frameworks for Java / web are based on technology

servlets.

CHAPTER 3. TECHNOLOGIES EVALUATED 28

In our work first extensively studied these technologies through

a book (BATES; SIERRA; BATES, 2008) and initiated the implementation based

therein.

However, we find that your use codes demanded extensive and repeated for

treat navigation as well as making the mapping between request parameters and

objects.

So we decided to verify the possibility of using more advanced frameworks.

Struts

Struts was one of the first Java EE frameworks, implementing a layer

parent, according to the model MVC web application, simplifying the use of

servlets, requiring only the creation of classes Action

This framework was originally developed by Craig McClanahan and donated to

the Apache Software Foundation in 2002, which continues to be developed according to

this standard foundation.

JBoss Seam

It is a more advanced framework developed by Red Hat, part of lot

used JBoss application server, which consists of a platform that encompasses the

container and most other facilities for developers to create Java applications

server.

The idea of ​​the framework is basically an easy way to create together (hence the

name) technologies EJB components (business logic) and JSF interfaces (one

"Evolution" of the JSP). It also provides components that greatly facilitate the creation of

CRUD's (forms for creating, editing, updating and removing entities from the database

data).

However, being very powerful, also has a considerable learning curve,

In particular we must deal with several complex configuration files.

CHAPTER 3. TECHNOLOGIES EVALUATED 29

VRaptor

It is an MVC Framework for Java Web development focused on fast, promising

its users a high productivity with a low learning curve, saving

time in developing solutions.

VRaptor is developed by Caelum, school that offers courses in Java are

Paul.

As in other frameworks, the idea is to release the developer of repetitive code

tedious and connected to technology so you can focus on the development of logic

business.

Some aspects is utilized to automate the conversion of request parameters

HTTP objects expected by the controller and ease of navigation control

between pages.

An interesting feature is that the need of writing configuration files

and generally supplanted by the use of conventions in programming parts of

application that interacts with the framework.

By having a simpler structure, with a direct documentation and easy to understand,

and be sufficient for our needs (given the relatively small size

our web modules) that was our choice for web framework.

Hibernate

When dealing with web applications is very common the need to store

information in the database, but since the application we deal with objects in general

you must perform a complex process of converting data into objects that can

be registered in the database tables and vice versa.

Hopefully this work can be automated with the use of frameworks that implement

JPA (Java persitence API) that defines how Java EE applications should perform

this conversion.

The open source implementation of the most known and used Hibernate JPA is that

features support for multiple databases available, including MySQL,

database also open source used by us.

CHAPTER 3. TECHNOLOGIES EVALUATED 30

What's more, several frameworks, including Seam and VRaptor already have mechanisms

integration that facilitate the configuration and use of Hibernate.

Chapter 4

SPECIFICATION PROJECT

4.1 Requirements

functional Requirements

? The system must perform the translation of a text to a text in Portuguese

POUNDS, considering the syntactic aspects of languages ​​involved.

? The body of the dictionary should be expandable, allowing words to be cadastras

gradually.

? The basis of syntactic rules of the Portuguese language should be easily modifiable,

allowing iterative adjustments during project development and facilitating

linguists interactions with the system.

? The translation result must be viewed through an animation, since

few deaf know a system of written notation of sign language, as

occurs with SignWriting.

? The repository data dictionary should be available for other systems

can use it (via a service architecture)

Non-functional requirements

? Although the output occurs with computer graphics, it is expected that the animation

executable on typical computers without special plates for graphics acceleration.

? One feature desired for the system is the modularity, so that components

System can be used in other projects involving languages

signals.

? Waiting for the translator is also a system as transparent as possible without

there are difficulties user to operate it.

CHAPTER 4. SPECIFICATION PROJECT 32

? It is hoped that the system is implemented using platforms and technologies

opened.

4.2 Use Cases

This section will describe the expected use cases of the system. Figure 4.1 contains the

diagram of use cases, which are detailed below.

Figure 4.1: Use cases

UC1 - Title: Translate phrase

Actors: User

Preconditions: none

Primary flow of events:

One. User accesses the system by selecting option "translate"

2nd. User tells the system the desired text in Portuguese

3rd. User calls the translation

4th. System responds with the same text, but in Pounds.

Alternative Flows:

CHAPTER 4. SPECIFICATION PROJECT 33

2, if the user enters a text in another language or unfamiliar words,

the system responds with words by spelling the same manual,

ie, the word is spelled signals through the loan of Pounds.

UC2-Insert new signal

Actors: expert Pounds

Preconditions: none

Flow winds primer:

One. User accesses the system by selecting option "Insert new signal"

2nd. User provides the signal parameters

3rd. System responds as would be the signal described

4th. User validates the output

5th. User provides signal meaning in Portuguese

6th. Signal system adds to your base

Alternative flow of events:

In 4, you can reject the response and return to step 2.

Postconditions: new sign added to the system.

UC3 - Edit existing signal

Actors: expert Pounds

Preconditions: none

Flow winds primer:

One. User accesses the system by selecting option "Edit existing signal"

2nd. System displays list of all available signals

3rd. User chooses a signal to be edited

CHAPTER 4. SPECIFICATION PROJECT 34

4th. System displays all current parameters of the signal

5th. User modifies the parameters you want

6th. System responds as would be the signal described

7th. User validates the output

Eight. User provides signal meaning in Portuguese

9th. System saves signal modifications to their base

Alternative flow of events:

7, user can reject the response and return to step 5.

Postconditions: modified signal in the system.

UC4 - Insert / Change grammar rule

Actors: expert Pounds

Preconditions: none

Flow winds primer:

One. User accesses the system by selecting option "Change grammar rule"

2nd. System displays list of all the grammar rules

3rd. User edits or inserts new rules

4th. System saves changes to the base

Postconditions: modified rules in the system.

4.3 Design Decisions on Machine Translation

Although for Machine Translation in the statistical models are the most commonly

used, this work has a great restriction on the use of this method:

the language is the object pounds, which is constituted by signs and visual space is not

CHAPTER 4. SPECIFICATION PROJECT 35

having written form. Therefore, there are hardly any written texts in POUNDS

to serve as a corpus and that there are at notations difficult to be processed

or underused, as SignWriting. Given these facts, we chose to use

method of translation rules for this work.

Various levels of translation presented in Section 2.3, this translator will act in

second, the syntactic level, ie, the bridge between languages ​​is the transformation of the structure

syntactic Portuguese into the syntactic structure of LBS. This level was chosen for

represent a cost-effective, not showing the simplicity unwanted

a translation morphological but not going further merit of semantic analysis for

that present very high complexity that would flee the scope of this work.

The thesis defended by Quadros (1999) presents a proposal of what would be the structure

phrasal LIBRAS, more specifically the structure of a sentence. She divided it into two

categories, for verbs plain and non-plain, ie defined syntax trees for the two

POUNDS. Therefore these structures will be considered in this work to define

rules.

4.4 Architecture

The system architecture translation poly-pounds, i.e. the representation of the units

Software constituting the system (also called modules) and the relationship

among them can be described briefly in figure 4.2.

It is noteworthy since several of these blocks are independent units that can be

directly reused in other applications

The following describes each of these system modules by specifying function, input

and expected outputs.

4.4.1 Data Model

A key issue of this work is how can digitally encode

a sequence of signals (pounds or any other sign language) so that

they can be processed by various applications.

Figure 4.2: System Architecture

This question corresponds approximately to the problem of how to register for

texts written in sign language, which as already described in section aspects

conceptual, still not well resolved within the deaf community.

Initially, a chance for our translation system would provide the output now

this notation SignWriting, which also contribute to building more easily

a legacy written in LBS. However, as has also been exposed, now at

Brazil SignWriting acceptance of the deaf community is low, so we chose

By using this model only encoding system as "internal" systems that

use, ie so that the user need not know know POUNDS SignWriting

to operate our system.

Thus, our coding system has attributes corresponding directly

attributes expressed graphically in a symbol SignWriting.

We did this based mainly on correspondence (SUTTON, 2003), work

which describes extensively the attributes and possible values ​​in SignWriting;

CHAPTER 4. SPECIFICATION PROJECT 37

complementation of some aspects of the notation (possibilities of moving

hands) were removed from the work of Tables (1999).

Thus, we can define a data structure capable of encoding signals,

represented by figure 4.3

Figure 4.3: Data Model

The figure shows that a signal is modeled as a sequence of symbols, and

a symbol of what can be represented by a symbol SignWriting; created

this distinction between sign and symbol due to certain "composite signals" as AA-BB in

that "AA" has a form of writing in SignWriting and "BB" another, and thereby

"AA-BB" is being the signal, while "AA" and "BB" are symbols.

Besides the symbol sequence, the aggregate signal also stamp information

syntactic translator that can assist in your task.

Thus, we have the following hierarchy of attributes to describe the signal:

Signal: expresses an idea, which usually corresponds to a

Portuguese word;

name: name of the signal, like a word;

words of Portuguese words that can match the

sign;

literal: words that must occur in the context of the signal;

CHAPTER 4. SPECIFICATION PROJECT 38

inerências: classes of words that should appear in

context of the signal (eg, words that represent

animals);

Symbol: represents a combination of morphological

that can be expressed with a symbol, or

a figure of SignWrite;

Location: where in the body the hands are positioned;

Contact: form of interaction of the hand with the lease;

Hands on unit: indicates that the two go hand to

point of lease or if the non-dominant hand

remains neutral space;

Facial expression: defining characteristics of cheek

eyes, eyebrows, forehead, eyes, mouth,

nose, teeth, tongue, and others;

Hand (dominant and nondominant):

Setup: how the fingers are articulated;

Guidance: If the palm is facing the emitter

or against him, facing the floor or

to heaven;

Plane: the arm with the hand are vertically

or horizontally;

Fingers: how fingers move.

Based on this data structure defined primarily synthesized on a set

class, according to the paradigm of Object Orientation.

These classes are then used to serve as input to processes

tables to define structures in a database capable of storing data

and structure (or schema) XML documents that can also represent these

data.

Morphological Analyzer 4.4.2

Figure 4.4 illustrates the specification of this module.

For parsers have focused on parsing of sentences, and

This, as the name implies, your ultimate goal, it is certain independence necesserária

Figure 4.4: Specification of the morphological analyzer

concerning achievement of lexemes. Aiming to provide this decoupling the analyzer

syntactic, arises the idea of ​​using a morphological analyzer. It is said that this structure

to that underlies because a parser is used to obtain morphological analyzers

tokens as syntactic analysis, syntactic, continues.

It should be noted that the verification scores and, logically, diacritics,

this structure, it is the morphological analyzer. But why not treat function of these

signs, he should notify them to the overlying layer through syntactic tokens agreed.

Typically, a morphological analyzer uses morphological tokens until most

possible amount of consunta is, it returns to some automaton overlying

corresponding token.

In the case of natural languages ​​it is common that the same sequence of tokens morphological

join more than one syntactic sense. Is exemplified: The phrase "as a

wave in the sea, "the morphological analyzer would be dubious about deciding whether the sequence of letters

(And hence morphological tokens) "as" is an adverb, a conjunction, or even if

corresponds to a combination of first-person singular present indicative of voice

An active verb "eat".

In fact, this decision will not fit. He should just tell the parser

possible meanings of morphological token in question. That is, to the previous example,

the three possible meanings for the lexeme "how" should be returned.

4.4.2.1 Tokens morphological

Were planned for the tokens morphological characteristics of the classes listed

Then, being a token that does not necessarily have all these classes defined

(However, there classes of features that can not coexist grammatically).

These classes are very much inspired by the default dictionary provided by the project JSpell.

CHAPTER 4. SPECIFICATION PROJECT 40

It is worth noting also that only the characteristic "category" is Mandatory, it is

essential for analysis superjacent.

CAT, "category".

Possible values ​​and meanings:

adj: adjective

adv: adverb

Article: Article

a_nc: adejetivo / common noun

card: