Academic literature on the topic 'Sign language – Translating'

Thesis (MSc)--University of Stellenbosch, 2007.<br>ENGLISH ABSTRACT: Machine translation, especially machine translation for sign languages, remains an active research area. Sign language machine translation presents unique challenges to the whole machine translation process. In this thesis a prototype machine translation system is presented. This system is designed to translate English text into a gloss based representation of South African Sign Language (SASL). In order to perform the machine translation, a transfer based approach was taken. English text is parsed into an intermediate representation. Translation rules are then applied to this intermediate representation to transform it into an equivalent intermediate representation for the SASL glosses. For both these intermediate representations, a tree adjoining grammar (TAG) formalism is used. As part of the prototype machine translation system, a TAG parser was implemented. The translation rules used by the system were derived from a SASL phrase book. This phrase book was also used to create a small gloss based SASL TAG grammar. Lastly, some additional tools, for the editing of TAG trees, were also added to the prototype system.<br>AFRIKAANSE OPSOMMING: Masjienvertaling, veral masjienvertaling vir gebaretale, bly ’n aktiewe navorsingsgebied. Masjienvertaling vir gebaretale bied unieke uitdagings tot die hele masjienvertalingproses. In hierdie tesis bied ons ’n prototipe masjienvertalingstelsel aan. Hierdie stelsel is ontwerp om Engelse teks te vertaal na ’n glos gebaseerde voorstelling van Suid-Afrikaanse Gebaretaal (SAG). Ons vertalingstelsel maak gebruik van ’n oorplasingsbenadering tot masjienvertaling. Engelse teks word ontleed na ’n intermediˆere vorm. Vertalingre¨els word toegepas op hierdie intermediˆere vorm om dit te transformeer na ’n ekwivalente intermediˆere vorm vir die SAG glosse. Vir beide hierdie intermediˆere vorms word boomkoppelingsgrammatikas (BKGs) gebruik. As deel van die prototipe masjienvertalingstelsel, is ’n BKG sintaksontleder ge¨ımplementeer. Die vertalingre¨els wat gebruik word deur die stelsel, is afgelei vanaf ’n SAG fraseboek. Hierdie fraseboek was ook gebruik om ’n klein BKG vir SAG glosse te ontwikkel. Laastens was addisionele nutsfasiliteite, vir die redigering van BKG bome, ontwikkel.

Thesis (PhD)--Stellenbosch University, 2014.<br>ENGLISH ABSTRACT: A sign language tutoring system capable of generating detailed context-sensitive feedback to the user is presented in this dissertation. This stands in contrast with existing sign language tutor systems, which lack the capability of providing such feedback. A domain specific language is used to describe the constraints placed on the user’s movements during the course of a sign, allowing complex constraints to be built through the combination of simpler constraints. This same linguistic description is then used to evaluate the user’s movements, and to generate corrective natural language feedback. The feedback is dynamically tailored to the user’s attempt, and automatically targets that correction which would require the least effort on the part of the user. Furthermore, a procedure is introduced which allows feedback to take the form of a simple to-do list, despite the potential complexity of the logical constraints describing the sign. The system is demonstrated using real video sequences of South African Sign Language signs, exploring the different kinds of advice the system can produce, as well as the accuracy of the comments produced. To provide input for the tutor system, the user wears a pair of coloured gloves, and a video of their attempt is recorded. A vision-based hand pose estimation system is proposed which uses the Earth Mover’s Distance to obtain hand pose estimates from images of the user’s hands. A two-tier search strategy is employed, first obtaining nearest neighbours using a simple, but related, metric. It is demonstrated that the two-tier system’s accuracy approaches that of a global search using only the Earth Mover’s Distance, yet requires only a fraction of the time. The system is shown to outperform a closely related system on a set of 500 real images of gloved hands.<br>AFRIKAANSE OPSOMMING: ’n Gebaretaaltutorstelsel met die vermo¨e om konteks-sensitiewe terugvoer te lewer aan die gebruiker word uiteengesit in hierdie proefskrif. Hierdie staan in kontras met bestaande tutorstelsels, wat nie hierdie kan bied vir die gebruiker nie. ’n Domein-spesifieke taal word gebruik om beperkinge te definieer op die gebruiker se bewegings deur die loop van ’n gebaar. Komplekse beperkinge kan opgebou word uit eenvoudiger beperkinge. Dieselfde linguistieke beskrywing van die gebaar word gebruik om die gebruiker se bewegings te evalueer, en om korrektiewe terugvoer te genereer in teksvorm. Die terugvoer word dinamies aangepas met betrekking tot die gebruiker se probeerslag, en bepaal outomaties die maklikste manier wat die gebruiker sy/haar fout kan korrigeer. ’n Prosedure word uiteengesit om die terugvoer in ’n eenvoudige lysvorm aan te bied, ongeag die kompleksiteit van die linguistieke beskrywing van die gebaar. Die stelsel word gedemonstreer aan die hand van opnames van gebare uit Suid-Afrikaanse Gebaretaal. Die verskeie tipes terugvoer wat die stelsel kan lewer, asook die akkuraatheid van hierdie terugvoer, word ondersoek. Om vir die tutorstelsel intree te bied, dra die gebruiker ’n stel gekleurde handskoene. ’n Visie-gebaseerde handvormafskattingstelsel wat gebruik maak van die Aardverskuiwersafstand (Earth Mover’s Distance) word voorgestel. ’n Twee-vlak soekstrategie word gebruik. ’n Rowwe afstandsmate word gebruik om ’n stel voorlopige handpostuurkandidate te verkry, waarna die stel verfyn word deur gebruik van die Aardverskuiwersafstand. Dit word gewys dat hierdie benaderde strategie se akkuraatheid grens aan die van eksakte soektogte, maar neem slegs ’n fraksie van die tyd. Toetsing op ’n stel van 500 re¨ele beelde, wys dat hierdie stelsel beter presteer as ’n naverwante stelsel uit die literatuur.

Thesis (MSc (Computer Science))--University of Stellenbosch, 2005.<br>Sign Languages such as South African Sign Language, are proper natural languages; they have their own vocabularies, and they make use of their own grammar rules. However, machine translation from a spoken to a signed language creates interesting challenges. These problems are caused as a result of the differences in character between spoken and signed languages. Sign Languages are classified as visual-spatial languages: a signer makes use of the space around him, and gives visual clues from body language, facial expressions and sign movements to help him communicate. It is the absence of these elements in the written form of a spoken language that causes the contextual ambiguities during machine translation. The work described in this thesis is aimed at resolving the ambiguities caused by a translation from written English to South African Sign Language. We designed and implemented a preprocessor that uses areas of linguistics such as anaphora resolution and a data structure called a scene graph to help with the spatial aspect of the translation. The preprocessor also makes use of semantic and syntactic analysis, together with the help of a semantic relational database, to find emotional context from text. This analysis is then used to suggest body language, facial expressions and sign movement attributes, helping us to address the visual aspect of the translation. The results show that the system is flexible enough to be used with different types of text, and will overall improve the quality of a machine translation from English into a Sign Language.

The purpose of the thesis was to create a data glove that can translate ASL by reading the finger- and hand movements. Furthermore, the applicability of conductive fabric as stretch sensors was explored. To read the hand gestures stretch sensors constructed from conductive fabric were attached to each finger of the glove to distinguish how much they were bent. The hand movements were registered using a 3-axis accelerometer which was mounted on the glove. The sensor values were read by an Arduino Nano 33 IoT mounted to the wrist of the glove which processed the readings and translated them into the corresponding sign. The microcontroller would then wirelessly transmit the result to another device through Bluetooth Low Energy. The glove was able to correctly translate all the signs of the ASL alphabet with an average accuracy of 93%. It was found that signs with small differences in hand gestures such as S and T were harder to distinguish between which would result in an accuracy of 70% for these specific signs.<br>Syftet med uppsatsen var att skapa en datahandske som kan översätta ASL genom att läsa av finger- och handrörelser. Vidare undersöktes om ledande tyg kan användas som sträcksensorer. För att läsa av handgesterna fästes ledande tyg på varje finger på handsken för att urskilja hur mycket de böjdes. Handrörelserna registrerades med en 3-axlig accelerometer som var monterad på handsken. Sensorvärdena lästes av en Arduino Nano 33 IoT monterad på handleden som översatte till de motsvarande tecknen. Mikrokontrollern överförde sedan resultatet trådlöst till en annan enhet via Bluetooth Low Energy. Handsken kunde korrekt översätta alla tecken på ASL-alfabetet med en genomsnittlig exakthet på 93%. Det visade sig att tecken med små skillnader i handgester som S och T var svårare att skilja mellan vilket resulterade i en noggrannhet på 70% för dessa specifika tecken.

ABSTRACT Context: Communication is a primary human need and language is the medium for this. Most people have the ability to listen and speak and they use different languages like Swedish, Urdu and English etc. to communicate. Hearing impaired people use signs to communicate. Pakistan Sign Language (PSL) is the preferred language of the deaf in Pakistan. Currently, human PSL interpreters are required to facilitate communication between the deaf and hearing; they are not always available, which means that communication among the deaf and other people may be impaired or nonexistent. In this situation, a system with voice recognition as an input and PSL as an output will be highly helpful. Objectives: As part of this thesis, we explore challenges faced by deaf people in everyday life while interacting with unimpaired. We investigate state of art work done in this area. This study explores speech recognition and Machine translation techniques to devise a generic and automated system that converts English speech to PSL. A prototype of the proposed solution is developed and validated. Methods: Three step investigation is done as part of thesis work. First, to understand problem itself, interviews were conducted with the domain experts. Secondly, from literature review, it is investigated whether any similar or related work has already been done, state of the art technologies like Machine translation, speech recognition engines and Natural language processing etc. have been analyzed. Thirdly, prototype is developed whose validation data is obtained from domain experts and is validated by ourselves as well as from domain experts. Results: It is found that there is a big communication gap between deaf and unimpaired in Pakistan. This is mainly due to the lack of an automated system that can convert Audio speech to PSL and vice versa. After investigating state of the art work including solutions in other countries specific to their languages, it is found that no system exists that is generic and automated. We found that there is already work started for PSL to English Speech conversion but not the other way around. As part of this thesis, we discovered that a generic and automated system can be devised using speech recognition and Machine translation techniques. Conclusion: Deaf people in Pakistan lack a lot of opportunities mainly due to communication gap between deaf and unimpaired. We establish that there should be a generic and automated system that can convert English speech to PSL and vice versa. As part of this, we worked for such a system that can convert English speech to PSL. Moreover, Speech recognition, Machine translation and Natural language processing techniques can be core ingredients for such a generic and automated system. Using user centric approach, the prototype of the system is validated iteratively from domain experts.<br>This research has investigated a computer based solution to facilitate communication among deaf people and unimpaired. Investigation was performed using literature review and visits to institutes to gain a deeper knowledge about sign language and specifically how is it used in Pakistan context. Secondly, challenges faced by deaf people to interact with unimpaired are analyzed by interviews with domain experts (instructors of deaf institutes) and by directly observing deaf in everyday life situations. We conclude that deaf people rely on sign language for communication with unimpaired people. Deaf people in Pakistan use PSL for communication, English is taught as secondary language all over Pakistan in all educational institutes, deaf people are taught by instructors that not only need to know the domain expertise of the area that they are teaching like Math, History and Science etc. but they also need to know PSL very well in order to teach the deaf. It becomes very difficult for deaf institutes to get instructors that know both. Whenever deaf people need to communicate with unimpaired people in any situation, they either need to hire a translator or request the unimpaired people to write everything for them. Translators are very difficult to get all the time and they are very expensive as well. Moreover, using writing by unimpaired becomes very slow process and not all unimpaired people want to do this. We observed this phenomena ourselves as instructors of the institutes provided us the opportunity to work with deaf people to understand their feelings and challenges in everyday life. In this way, we used to go with deaf people in shopping malls, banks, post offices etc. and with their permission, we observed their interaction. We have concluded that sometimes their interaction with normal people becomes very slow and embarrassing. Based on above findings, we concluded that there is definitely a need for an automated system that can facilitate communication between deaf and unimpaired people. These factors lead to the subsequent objective of this research. The main objective of this thesis is to identify a generic and an automated system without any human intervention that converts English speech into PSL as a solution to bridge the communication gap between deaf and unimpaired. It is identified that existing work done related to this problem area doesn’t fulfill our objective. Current solutions are either very specific to a domain, e.g. post office or need human intervention i.e. not automatic. It is identified that none of the existing systems can be extended towards our desired solution. We explored state of the art techniques like Machine translation, Speech recognition and NLP. We have utilized these in our proposed solution. Prototype of the proposed solution is developed whose functional and non functional validation is performed. Since none of existing work exactly matches to our problem statement, therefore, we have not compared the validation of our prototype to any existing system. We have validated prototype with respect to our problem domain. Moreover, this is validated iteratively from the domain experts, i.e. experts of PSL and the English to PSL human translators. We found this user centric approach very useful to help better understand the problem at the ground level, keeping our work user focused and then realization of user satisfaction level throughout the process. This work has opened a new world of opportunities where deaf can communicate with others who do not have PSL knowledge. Having this system, if it is further developed from a prototype to a functioning system; deaf institutes will have wider scope of choosing best instructors for a given domain that may not have PSL expertise. Deaf people will have more opportunities to interact with other members of the society at every level as communication is the basic pillar for this. The automatic speech to sign language is an attractive prospect; the impending applications are exhilarating and worthwhile. In the field of Human Computer Interface (HCI) we hope that our thesis will be an important addition to the ongoing research.<br>Ahmed Abdul Haseeb & Asim ilyas, Contact no. 00923215126749 House No. 310, Street No. 4 Rawal town Islamabad, Pakistan Postal Code 44000

Sign language is a visual language. It is an effective way of communication for hearing and speech impaired community. In general, all visual languages are multi-modal which utilizes hand gestures, facial expressions and other non-manual features to effectively consider the linguistics of the language while communicating with others. In the recent few years, due to the advancement of computer vision and the NLP field, Sign Language Recognition(SLR) and Sign Language Translation(SLT) topic has attracted many researchers to find an effective way to translate sign language videos to spoken language sentence. Over a decade many ideations have been published but most of them focused on SLR as a mere gesture recognition problem without considering the linguistic structure. In the literature review, we dive deep in to the understanding of various Senor and vision based approaches which were used in the earlier days followed by Deep learning techniques which are offering state-of-the-art results in the recent days. Applying a mid-level Sign Gloss Representation is a key component of performing a successful SLT. Hence, an effective joint learning of mid-level sign Gloss into the Text translation is crucial to improve the performance. In this dissertation, we propose Deep Directive Attention Network (DDAN)-based sign translation framework that allows aligning key-tokens in sign Gloss with key-words in Text. Directive attention transformer is successfully used in this approach to have better inter-intra modal relationship between Gloss sequences and Text sentences which aids in higher translation accuracy of Sign videos to Text sentences. The proposed DDAN contains the Self-Attention (SA) of each sign Gloss and Text, as well as the Gloss Directive-Attention (DA) of Text. Those two attention units, SA and DA, can be placed and integrated in three different proposed DDAN variants, including DA, SDA and SSDA. We evaluate the translation performance of our Sign2(Gloss+Text) and Gloss2Text approaches on the two challenging benchmark datasets PHOENIX-Weather 2014T and ASLG-PC12. The data statistics were analyzed as the first step. Then, three different model variants were evaluated on the above mentioned datasets. The model variant SSDA outperformed the baseline models in both datasets with higher translation accuracy of Sign videos to Text sentences as well as Gloss sequences to Text sentences . Furthermore, we evaluated on various numbers of encoder and decoder to see the optimal count of layers in which the model outperformed the baselines. The hyper-parameter testing result shows the robustness of the proposed framework. In addition to quantitative analysis results, we also provide the qualitative results of the evaluations which shows the generated text sentences has translation precision close to gold standard text along with evident improvement in the morpho-syntax. Based on all the evaluations and analysis results, we demonstrate that out DDAN-based SLT framework outperforms all the state-of-the-art SLT models and achieve better and higher translation accuracy score.

CoordenaÃÃo de AperfeiÃoamento de Pessoal de NÃvel Superior<br>Este estudo, inserido no Ãmbito dos Estudos da TraduÃÃo, ratifica o reconhecimento da LÃngua Brasileira de Sinais (Libras), graÃas à sua inclusÃo no currÃculo das instituiÃÃes de Ensino Superior brasileiras. O objetivo principal deste trabalho à identificar e apontar a funÃÃo da traduÃÃo em Libras como estratÃgia de compreensÃo e interpretaÃÃo do gÃnero textual artigo cientÃfico em LP na disciplina âPsicologia e EducaÃÃo de Surdosâ, pertencente ao eixo Fundamentos da EducaÃÃo de Surdos, ministrada no Curso de Licenciatura em Letras Libras da Universidade Federal do Cearà (UFC). Com um olhar sobre os discentes surdos, esta pesquisa pretende averiguar se e de que forma a traduÃÃo de textos cientÃficos na direÃÃo LP/Libras serve como instrumento para se alcanÃarem os nÃveis de compreensÃo necessÃrios à realizaÃÃo das atividades que exigem leitura, traduÃÃo e produÃÃo textual/discursiva com o gÃnero textual artigo cientÃfico escrito originalmente em LP. Para chegar ao objetivo precÃpuo do trabalho aqui apresentado, esta investigaÃÃo baseou-se metodologicamente em uma pesquisa-aÃÃo (VIEIRA, 2009), partindo, por um lado, de anÃlises de fatores extratextuais e intratextuais de um texto-fonte (TF) em LP e do texto-alvo correspondente (TA) em Libras. Essas anÃlises foram realizadas por 6 (seis) discentes surdos regularmente matriculados na disciplina âPsicologia e EducaÃÃo de Surdosâ, que antes foram instruÃdos sobre os fundamentos teÃricos propostos por Nord (2016) no domÃnio da Abordagem Funcionalista da TraduÃÃo. Por outro lado, como parte da metodologia, foram aplicados questionÃrios aos mesmos 6 (seis) discentes surdos e tambÃm à docente ministradora da disciplina âPsicologia e EducaÃÃo de Surdos.â Como suporte teÃrico no campo tradutÃrio, fundamentamos esta pesquisa mediante a Teoria de Escopo, concebida por Reiss & Vermeer (1996), e a Abordagem Funcionalista da TraduÃÃo, com nossa atenÃÃo voltada especificamente para os trabalhos de Nord (2012; 2014; 2016). Nossa preocupaÃÃo parte de uma de nossas hipÃteses: a necessidade de preparaÃÃo dos professores que ministram a disciplina âPsicologia e EducaÃÃo de Surdosâ do Curso de Letras Libras da UFC em questÃes relativas ao uso da traduÃÃo como estratÃgia para que os alunos surdos possam alcanÃar melhores objetivos de aprendizagem, ao serem confrontados com textos cientÃficos escritos em LP. Dada a complexidade do gÃnero textual artigo cientÃfico (p. ex.: linguagem tÃcnica e/ou cientÃfica, registro de portuguÃs formal, termos especÃficos etc.), os docentes precisam apresentar, aos discentes, traduÃÃes dos textos em Libras, fornecidas no formato de DVD. Portanto, vislumbramos em nossas hipÃteses que os discentes envolvidos normalmente utilizam a LP como segunda lÃngua, carecendo, assim, de conhecimentos aprofundados que lhes permitam compreender com facilidade textos cientÃficos redigidos nessa lÃngua. TambÃm lhes faltam conhecimentos e experiÃncias com anÃlises textuais que levem em consideraÃÃo diferentes fatores intratextuais e extratextuais como aqueles propostos por Nord (2016). Entendemos essa necessidade como um desafio atual para o Curso de Licenciatura em Libras da UFC. AlÃm disso, compreendemos que, se forem adotadas medidas que conduzam a um entendimento prÃtico desses exercÃcios, auxiliadas por teorias oriundas do campo dos Estudos da TraduÃÃo, os futuros profissionais surdos licenciados em Libras na UFC terÃo uma formaÃÃo mais completa e um melhor domÃnio da compreensÃo leitora em LP. Isto tambÃm certamente se refletirà no 9 desempenho de sua futura funÃÃo de educadores, em que deverÃo estar empenhados na inclusÃo de crianÃas e jovens surdos na sociedade em geral, mas sem deixar de lado os fatores prÃprios da Cultura Surda. De maneira geral, os resultados por nÃs obtidos apontam que o grupo-alvo participante da pesquisa-aÃÃo nÃo està familiarizado com um modelo de anÃlise textual do TF e do TA nos moldes daquele fornecido pela Abordagem Funcionalista de Nord (2016), o que certamente dificulta sua compreensÃo dos conteÃdos apresentados no texto original em portuguÃs e na traduÃÃo em Libras. De maneira especÃfica, podemos concluir, dentre outras coisas, que os alunos surdos examinados/entrevistados: a) precisam se familiarizar mais com as estruturas sintÃticas e o vocabulÃrio em LP; b) precisam conscientizar-se das peculiaridades de sua lÃngua natural/materna perante o PortuguÃs como segunda lÃngua (PSL); c) necessitam fazer leituras mais atentas de textos cientÃficos; d) carecem de conhecimentos mais profundos sobre conceitos tÃpicos da LinguÃstica Textual; e) declaram aumentar seu conhecimento de terminologias especializadas em Libras atravÃs dos textos cientÃficos traduzidos da LP para Libras; f) afirmam a importÃncia da escola inclusiva/bilÃngue para surdos como preparaÃÃo para o Ensino Superior; g) lanÃam mÃo de diferentes mÃdias eletrÃnicas para obterem uma melhor compreensÃo leitora de textos cientÃficos; h) nÃo se consideram, em sua maioria, capazes de traduzir textos cientÃficos da LP para Libras; i) reconhecem que a traduÃÃo à uma ferramenta que traz vantagens para a compreensÃo leitora.<br>This study, which is embedded in the field of Translation Studies, corroborates the recognition of the Brazilian Sign Language, the so-called Libras, thanks to its inclusion in the Brazilian curricula of Higher Education institutions. The main objective of this research is to identify and point out the function of translation in Libras as a strategy for understanding and interpreting the text type scientific article in Portuguese in the subject âPsychology and Education of Deaf Peopleâ, within the framework of the subjects studied in area of the discipline Fundamentals of Education, part of the Libras Undergraduate Course curriculum (Teachers Training Program) at Federal University of Cearà (UFC). With special attention given to deaf students, this research intends to investigate if and how the translation of scientific texts in the Portuguese/Libras direction serves as an instrument to reach the levels of comprehension necessary to carry out the activities that require reading, translation and text/discourse production with the specific text type scientific article originally written in Portuguese. To reach the main objective of this work, this research was methodologically based on an action-research (VIEIRA, 2009), starting with, on the one hand, analyzes of extratextual and intratextual factors of a source text (ST) in Portuguese and the corresponding target text (TT) in Libras. These analyzes were carried out by six (6) deaf students regularly enrolled in the discipline âPsychology and Education of Deaf Peopleâ at UFC, who were previously instructed on the theoretical foundations proposed by Nord (2016) in the field of the Functionalist Approach to Translation. On the other hand, as part of the methodology, questionnaires were applied to the same 6 (six) deaf students, and also to the teacher of the subject âPsychology and Education of Deaf Peopleâ. As a theoretical support in the field of translation, in this research we make use of the fundamental principles of the Scope Theory, conceived by Reiss & Vermeer (1996), and of the Functionalist Approach to Translation, with our attention focused specifically on Nordâs works (2012; 2014; 2016). Our concern is based on one of our hypotheses: the need for preparation of teachers who teach the discipline âPsychology and Education of Deaf Peopleâ in the UFC Libras Undergraduate Course on questions related to the use of translation as a strategy for deaf students to be able to achieve better learning objectives by being confronted with scientific articles written in Portuguese. Because of the complexity of the text type scientific article (e. g. technical and/or scientific language, formal Portuguese language register, specific terms etc.), teachers need to present, to the students, translations of the texts in Libras, exhibited in DVD format. Thus, in our hypotheses we envisage that the students involved usually use Portuguese as a second language, thus lacking in-depth knowledge that could enable them to easily understand scientific articles written in that language. They also lack knowledge and experience with textual analyzes that take into account the different intratextual and extratextual factors such as those proposed by Nord (2016). We understand this need as a current challenge for the UFC Libras Undergraduate Course; we also think that, if measures are adopted leading to a practical understanding of these exercises, aided by theories from the field of Translation Studies, future deaf professionals that will graduate from the UFC Libras Undergraduate Course will have a more complete training and a better command of reading comprehension 11 in Portuguese. This will also certainly be reflected in the performance of their future role as educators, in which they should be committed to include deaf children and young people in society in general, nevertheless without neglecting the characteristics of the culture of deaf people. In general, our results indicate that the target group participating in the action-research is not familiar with a textual analysis model regarding both ST and TT, such as that provided by Nordâs Functionalist Approach to Translation (2016); this fact certainly makes it difficult for deaf students to understand the contents presented in the original Portuguese texts and the corresponding translation into Libras. Specifically, we can conclude, among other things, that the deaf students that we examined/interviewed during our action-research: a) need to become acquainted with the Brazilian Portuguese syntactic structures and vocabulary as well; b) need to be aware of the peculiarities of their natural/mother tongue as compared to Brazilian Portuguese as a second language; c) need to make more careful readings of scientific articles; d) lack deeper knowledge about typical concepts that are used in Text Linguistics; e) declare to increase their knowledge of specialized terminology in Libras through the scientific articles translated from LP to Libras; f) assume the importance of inclusive/and bilingual school for deaf people as a preparation for Higher Education; g) use different electronic media to obtain a better reading comprehension of scientific articles; h) majorly do not consider themselves capable of translating scientific articles from LP to Libras; i) recognize that translation is a tool that really brings advantages to reading comprehension.

Thesis (S.M.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, June 1999.<br>Includes bibliographical references (leaves 130-132).<br>by Robbin Nicole Chapman.<br>S.M.

This dissertation presents the first corpus-based system for translation from Arabic text into Arabic Sign Language (ArSL) for the deaf and hearing impaired, for whom it can facilitate access to conventional media and allow communication with hearing people. In addition to the familiar technical problems of text-to-text machine translation,building a system for sign language translation requires overcoming some additional challenges. First,the lack of a standard writing system requires the building of a parallel text-to-sign language corpus from scratch, as well as computational tools to prepare this parallel corpus. Further, the corpus must facilitate output in visual form, which is clearly far more difficult than producing textual output. The time and effort involved in building such a parallel corpus of text and visual signs from scratch mean that we will inevitably be working with quite small corpora. We have constructed two parallel Arabic text-to-ArSL corpora for our system. The first was built from school level language instruction material and contains 203 signed sentences and 710 signs. The second was constructed from a children's story and contains 813 signed sentences and 2,478 signs. Working with corpora of limited size means that coverage is a huge issue. A new technique was derived to exploit Arabic morphological information to increase coverage and hence, translation accuracy. Further, we employ two different example-based translation methods and combine them to produce more accurate translation output. We have chosen to use concatenated sign video clips as output rather than a signing avatar, both for simplicity and because this allows us to distinguish more easily between translation errors and sign synthesis errors. Using leave-one-out cross-validation on our first corpus, the system produced translated sign sentence outputs with an average word error rate of 36.2% and an average position-independent error rate of 26.9%. The corresponding figures for our second corpus were an average word error rate of 44.0% and 28.1%. The most frequent source of errors is missing signs in the corpus; this could be addressed in the future by collecting more corpus material. Finally, it is not possible to compare the performance of our system with any other competing Arabic text-to-ArSL machine translation system since no other such systems exist at present.