It is at this point the most developed approach and is implemented into actual use in France (Figure 2) for announcements in railway stations [BRAFFORT 2011]. The system can generate announcements like “Train No. xx from xx will arrive at platform xx and will be xx minutes late”, the ‘xx’ representing the variable elements of the announcement.
Annelies Braffort & Patrice Dalle Figure 2 : Transmission of Messages in French SL (LSF) at Gare de l’Est in Paris.
• Generating utterance from scripts. This approach is based either on video editing tools, or linguistic mechanisms [ELLIOTT 2008], [FILHOL 2010]. The latter approach requires a database of lexical descriptions and a set of grammar rules describing the sl’s functioning. These elements are still being investigated in foundational, long-term research, and the sl are not yet fully described.
• Generating utterance from text. These methods of machine translation, often inspired by those used for spl, are statistical mechanisms necessitating very large corpora, which are not yet available for sl.
Their other drawback is that they don’t model the sl characteristics of iconicity, spatiality, and multilinearity. Another, more recent approach [VENDRAME 2010] is to go through a semantic representation that better resembles sl discourse organization [GUITTENY 2007].
For animation generation, three approaches can also be identified :
• Generating animations via rotoscopy. This approach uses video to create a model for a specialised graphic designer, who then uses specialized software to transpose the signer’s filmed postures and movements onto an avatar. This method, which is widely-used in 3D animated film production, enables the high quality of animations that is indispensable for a satisfactory content comprehensibility and sl user acceptance. This approach’s disadvantages are that the quality of the result is heavily dependent on the graphic designer’s experience and talent, and that it entails a certain period of production. However, this approach makes it possible to construct an animation database that can then be concatenated and adapted. This is also the method that is currently used in sncf software [BRAFFORT 2011].
Annelies Braffort & Patrice Dalle • Generating animations via motion capture. This approach, like the previous one, consists of concatenating and adapting predefined animations, but in this case using motion capture databases [LU 2011].
It similarly requires prior construction of an sl corpus, but in this methodology the corpus is captured by infrared cameras. It has the advantage of providing data directly in 3D, but nevertheless requires a certain production period for a data “cleaning” process that is required prior to use. Moreover, it requires the use of a motion capture system.
• Automatic animation generation. This refers to generating animation from symbolic description. It stems from the research problematic of the fields of computer graphics and biology (physiology, movement modelling, etc.). Some websites, especially those dedicated to bilingual dictionaries, incorporate software built using this approach 9.
Automatically generated animations are still rather robotic and lacking in realism, particularly when it comes to the naturalness of movement and the animation of facial expressions. The latter difficulty arises from the current lack of knowledge about non-manual elements function in sl. Some studies are now beginning to address this issue [CHTELAT 2011].
The advances in sl generation are indisputable, and internationally increasing numbers of research teams are gaining interest in this field. To produce significant advances, upstream studies on sl function analysis must be developed, by creating large corpora and supporting linguistic and multidisciplinary studies that incorporate the results of this research in their models and software.
CYBERSPACE : TOWARDS NEW APPLICATIONS Until now, devices for the deaf enabling them to live bilingually, that is, to practice their own sign language in a predominantly hearing world using the country’s spoken language, consisted mainly of human assistance, essentially, the model of face-to-face interpreters, a model more or less unrelated to cyberspace.
New technologies and the internet have enabled these devices to be expanded using video conferencing systems : remote communication 9 www2.cmp.uea.ac.uk/~jrwg/DictionaryAnnelies Braffort & Patrice Dalle between signing and non-signing individuals via interpreters at a relay centre, or directly in sl between signing individuals 10, is now possible.
However, these mechanisms may only be used for communication between people.
New systems [DALLE 2011], focusing on content access or on the sl itself [LEFEBVRE 2010B], are now beginning to appear in several fields.
In terms of content access, the first applications targeted accessibility for existing websites. They aimed to integrate online assistance with sl video format, either to translate parts of a document or to add explanations in sl (to alleviate the difficulty many deaf people experience when reading).
However, this complementary form of sl can only be used for pages that rarely change ; and so, despite initial enthusiasm, these devices haven’t continued to develop. The WebSourd company website (Figure 3), which provides daily translations of afp (Agence France Presse) bulletins in sl video, remains an exception.
Advances in sl generation and analysis have not yet translated into new products in this area, but we can see in experimental products some hints of future developments :
• Website enrichment : eca (Embodied Conversational Agents) are used to support navigation ; virtual signers, viewable on demand, can be used to provide complementary sl 11 ; for example, by translating a site’s faq page. Compared to a video of a real signer, they offer the advantage of anonymity, they conform to the look of the site, and they can be modified, especially if the generation module is powered by a translation module from text to sl ;
• Wiki in sl : An even more ambitious prototype application, which has been explored during the European project DictaSign, is the realisation of a wiki in sl. The wiki consists of partial recognition of information in SL that the user provides via webcam, along with messages generated and performed by a virtual signer ;
• Bilingual sites : sl is now well established on the web, through high quality bilingual information websites 12. Beyond the useful and cultural value for those involved, they also serve to exhibit sl and make 10 http://www.afils.fr/index.php conseils “Du bon usage des centres relais”.
11 http://www.limsi.fr/Individu/jps/online/diva/geste/geste.main.htm 12 http://www.websourd.org Annelies Braffort & Patrice Dalle it accessible to all. Furthermore, these sites have an impact on the language itself : as they employ a more academic level of language than that used in everyday signing, these sites have a diffusive effect on the local lexicon and the spread of neologisms ;
Figure 3 : Intensive video use on the website WebSourd.
• sl dictionaries : Cyberspace is the perfect venue for this type of document. Although they remain rare, several encyclopaedia projects are underway (Elix 13, Ocelles [MOREAU 2010]), together with some thematic glossaries (uved 14) ;
The use of this sl content consists of several components [DALLE 2011] :
– Content realisation : This refers to sl and bilingual documents. Easyto-use tools [LEFEBVRE 2010B] for including a comment in sl using video (sl video, video of a text or a presentation) now exist ;
– Acceptance of signs used : In a rapidly evolving but little standardised language, this issue is particularly acute, and ongoing projects are providing mechanisms to identify the geographical origin of signs and their level of acceptance within the deaf community. This is especially helpful with neologisms ;
– Presentations in sl : Before working with a document in a sl, you need to know of its existence and be able to assess its relevance. As sl do not have a writing system, apart from text usage, special techniques 15 for presentation in image form, photosigns (Figure 4), and mini-video enable tables of contents and indexes to be provided in sl.
13 http://www.signesdesens.org/-e-learning-.html 14 http://www.irit.fr/GlossaireDD-LSF 15 http://www.usherbrooke.ca/liaison_vol41/n08/a_avaglyphe.html Annelies Braffort & Patrice Dalle – Content navigation in sl : Finally, the routes through an sl document should be able to be consumed non-linearly, and experiments are underway 16 to replicate in video the equivalent of a clickable link in text. For this, it is necessary to know how to display the link, and subsequently how to use it. On the other hand, research in pattern recognition should help in the short term to perform queries directly in sl, at least in the form of isolated signs.
• Distance education in lsf : Beyond creating online support for lsf, distance education in lsf is beginning to develop (In France : daeu Nancy 18, du lsf Rouen 19, du iels Toulouse 20). While the courses are most often offered in the form of downloaded sl videos, student progress monitoring can occur directly in lsf using online resources (videoconferencing, lsf forums, etc.) [TANAKA 2010] ;
• Learning lsf : The first websites in lsf were language presentation sites, essentially French lexicons providing each entry in French with its signed equivalent ; terms were listed either alphabetically or by theme.
In lsf, an order of presentation is defined, usually parametrically, with signs being classified at a first level according to hand shape (choosing from among around fifty configurations). It will soon become possible to make a request in lsf via webcam [LEFEBVRE 2010A]. However, these sites are of limited value for learning lsf. lsf teaching websites 21 are often online registered courses repositories where students can send their exercises in video form. Again, we can expect some interesting developments for self-learning, thanks to tools comparing signs and generation by virtual signers [ARAN 2009] ;
• Teacher resource platform of/in lsf : The recognition of sl and its introduction in education (bilingual education, an lsf option for the French Baccalaureate) has led to the creation of teacher resource sites (cndp, e-lsf ) and extensive use of sl video, but less so the most advanced lsf techniques.
18 http://erudi.free.fr/index.php page=daeulsf 19 http://formations.univ-rouen.fr/LSA31_864/0/fiche_formation 20 http://www.irit.fr/iels 21 http://www.signingsavvy.com Annelies Braffort & Patrice Dalle Figure 4 : Examples of photosigns DEVELOPMENT and SUSTAINABLE in LSF CYBERSPACE, AN OPPORTUNITY FOR SL Because deafness is not seen, it isn’t taken into account except when the deaf express themselves in sl. Cyberspace presents an important opportunity for sl :
– For minority languages that are not geographically localised : Their practice is doubly penalised ; it is therefore crucial to increase sites of sl expression, that is, to recreate social spaces ;
– Multimedia languages, for which video is the “written” form : For these languages, the web and its ability to disseminate image, animation and video, as well as its potential for interaction, constitutes a privileged space ;
– Languages for which the grammatical component is fairly universal :
These languages are facilitating international exchanges.
Cyberspace should enable mother-tongue education for deaf children.
Communication in sign language beginning at birth can restore a normal environment in which deaf children can built a real language, provided the sl used is correct. For these children, only sl can play the role of native language. To do this, it must be or must also become the parents’ language. Most children who are born deaf have hearing parents who have not mastered (or don’t know) sign language, and they must be trained early and at home. These families are dispersed throughout the country (there is no community of families of deaf children), so it is difficult to establish regular group training.
Annelies Braffort & Patrice Dalle Distance learning in sl, and the dissemination of language resources and sites via internet provide one answer to this problem. Self-teaching systems for sl, including corrective mechanisms via recognition or generation, would facilitate sl learning for families and its early acquisition by deaf children, who would then enter a life with language.
Deaf people often lack quality training and mastery of the written form of their country’s spoken language(s). sl supports, the web resources to distribute them, sl encyclopaedias, and online training can help fill these gaps. While production techniques for these supports exist, there are currently few operational solutions for sl content access (queries in sl, indexing, navigation in sl) and generation (content production by virtual signers). However, these systems would be very useful for the deaf, for their cultural development, and thus their social integration.
The development of these systems entails more complete and realistic computer modelling of sl, incorporating models of variations, either to recognise spontaneous utterances or to generate realistic productions by a virtual signer. Campaigns for the constitution of large high-quality corpora should be undertaken to accelerate the development of these models, support research into sl recognition and generation, and thus enable slp to approach the performance level of nlp.
Материалы этого сайта размещены для ознакомления, все права принадлежат их авторам.
Если Вы не согласны с тем, что Ваш материал размещён на этом сайте, пожалуйста, напишите нам, мы в течении 1-2 рабочих дней удалим его.