Avatars for Visual Communication
Submitting Institution
University of East AngliaUnit of Assessment
Computer Science and InformaticsSummary Impact Type
SocietalResearch Subject Area(s)
Information and Computing Sciences: Artificial Intelligence and Image Processing
Studies In Creative Arts and Writing: Film, Television and Digital Media
Language, Communication and Culture: Linguistics
Summary of the impact
The primary impact of our research has been through systems that provide
information for the deaf community via animated sign language. Tools
exploiting our research into automatic generation of human motion,
especially for sign language, are used in several ways:
- as a mechanism for face-to-face communication between deaf and hearing
people;
- for the provision of public information in sign language in a
semi-automatic way;
- to display the content of web sites in sign language;
- to provide sign language for TV programmes.
The same avatar (animated character) technology has also been applied in
two other arenas: as a foreign language learning resource for primary
school children and to add value to cultural heritage presentations.
Underpinning research
Interactive information in content using human gestures cannot be
effectively generated from video material; this is particularly true of
sign language. An attractive alternative approach uses 3D character
animation techniques.
The Virtual Humans Group at UEA undertakes distinctive research bringing
together computer scientists with a wide range of expertise: speech and
language recognition, computational linguistics, AI, and computer graphics
for 3D character animation. Over a period of 14 years, dating back to
1999, this Group has researched advanced techniques for general animation
with speaking characters, including scripted animation of 3D characters
supporting animation of deaf sign language. Such animations must not only
appear convincing, but must be fully comprehensible to those using only
the visual channel for communication.
Initially `Simon the Signer' (1999: Bangham, Glauert) explored
translating text subtitles to animated sign sequences. Linguistic analysis
drove the presentation of motion-captured signs in English language order,
which resulted in natural-looking animations. `Simon the Signer' won two
Royal Television Society awards in 1999. However, although the system was
an important step forwards, British Sign Language (BSL), like all sign
languages, has its own syntax and grammar which is only loosely related to
the syntax of spoken English.
The TESSA project (1999: Cox, Lincoln) developed a speech recognition
system to identify phrases spoken by a counter clerk for presentation to
deaf customers in true BSL, which was an important advance [1].
The key insight was that the discourse between deaf customer and counter-
clerk is highly constrained in topic and scope, so that the vocabulary and
syntax required can be largely pre-defined [2]. TESSA won the BCS Gold
Medal (2000) and was exhibited in the Science Museum, Kensington in July
2001.
This early work built sequences from motion-captured signs and achieved
realistic animation for fixed phrases. However, signs translated from
spoken language undergo considerable modification/inflection in context.
These issues were addressed by techniques developed in the ViSiCAST
project (2000: Glauert, Bangham, Cox, Kennaway, Marshall, Safar, Eliott),
which produced what is still the most mature system anywhere for
translating spoken language to sign language, using a Discourse
Representation Theory semantic representation, and Head-Driven Phrase
Structure Grammar rules for sign language generation, for three national
sign languages [3]. A sophisticated treatment is made of linguistic
features in sign language such as handling of plural verbs, management of
signing space for anaphoric references, and use of non-manual movements
(such as eyebrow position) to match sentence mode. The output was based on
Hamburg Notation System (HamNoSys), a recognised `phonetic' transcription
system for signs [4]. Two further EU-funded projects, eSIGN (2002 Glauert,
Cox, Lincoln, Elliott) and Dicta-Sign (2009 Glauert, Elliott, Kennaway,
Safar), developed novel algorithms to synthesise natural motion data for
3D characters from HamNoSys. Planning for hand shape, orientation, and
location for the initial sign posture, leads to generation of movement and
hand shape changes for subsequent transitions and postures making up a
sign. Inverse kinematics generates the necessary arm positions and
realistic dynamic trajectories create natural-looking motion [5].
These projects created a bespoke avatar development and animation toolkit
to support several cross-platform applications and browser plugins for
interactive real-time virtual signing [6]. Conventional 3D characters are
enhanced with information on key body sites linked to abstract locations
in HamNoSys transcriptions and morph targets for facial expressions used
in signing. Avatars and animations can be exported to industry-standard
tools (Maya and 3ds Max).
Key Research Personnel
Lead academics: Professors John Glauert; Stephen Cox; Andrew Bangham; Dr
Ian Marshall; Dr Ralph Elliott
Research Associates: Richard Kennaway; Mike Lincoln; Eva Safar
References to the research
(UEA authors in bold)
[1] S.J. Cox, M. Lincoln, M.J. Nakisa, M. Wells, M. Tutt and S.
Abbott The development and evaluation of a speech to sign translation
system to assist transactions Int. Journal of Human Computer Interaction 16
141(2003) URL: http://hdl.handle.net/1842/1103
[2] S.J. Cox Speech and Language Processing for a Constrained
Speech Translation System Proc. Int. Conf. on Spoken Language Processing
(ICSLP 2002), Denver, Colorado, USA (2002) Copy held on file at UEA
[4] R. Elliott, J.R.W. Glauert, J.R. Kennaway and I. Marshall The
Development of Language Processing Support for the ViSiCAST Project Proc.
ASSETS 2000, 4th ACM SIGCAPH Conference on Assistive Technologies,
Arlington, Virginia, USA (2000) URL: http://dl.acm.org/citation.cfm?id=354349
[5] R. Elliott, J.R.W. Glauert, J.R. Kennaway, I. Marshall and E.
Safar Linguistic Modelling and Language-Processing Technologies for
Avatar-based Sign Language Presentation UAIS (Universal Access in the
Information Society) 6 375 (2008) doi: 10.1007/s10209-007-0102-z
[6] R. Kennaway, J.R.W. Glauert and I. Zwitserlood Providing
Signed Content on the Internet by Synthesized Animation ACM Transactions
on Computer-Human Interaction 14 15 (2007) doi:
10.1145/1279700.1279705
External funding supporting the research
1. Simon the Signer
Bangham, Glauert; Independent Television Commission (1999-2000) £26K
2. TESSA
Cox; Post Office (1999-2002) £203K
3. ViSiCAST
Glauert, Bangham, Cox; EU FP5 (2000-2002) £563K with additional funding of
£519K from ITC, Post Office, BBC
4. eSIGN
Glauert, Cox; EU eContent (2002-2004) £168K
5. LinguaSign
Glauert; EU Socrates Lingua 2 (2006-2010) £93K
6. Dicta-Sign
Glauert; EU FP7 (2009-2012) £516K
Details of the impact
The primary beneficiary of our research into the automatic generation of
human motion has been the deaf community, in the UK and elsewhere. The
systems developed at UEA have been applied in a variety of situations to
improve both direct communication and information provision. In addition,
the avatar systems have been used to provide a new way of teaching
languages in primary schools and to add an additional dimension to
cultural heritage sites.
-
Avatar Signing for the Deaf Community
For many prelingually deaf people, their first language is natural sign
language, with approximately 50,000 people in Britain having British
Sign Language (BSL) as their first language. Our algorithms and 3D
animation technology have been developed to enable rich natural sign
language to be generated for communication with deaf people. Systems
arising from this research provide deaf people with the means to access
information and services in their chosen language, respecting the
cultural and social needs of a recognised minority community.
Our systems allowing face-to-face communication between deaf and
hearing people, with speech recognition and the subsequent automatic
generation of signing were first used by the Post Office in 2000. During
the current REF impact period, this technology has been taken forward by
IBM through Say It Sign It (SiSi), a real-time system
that allows sign language users to communicate with hearing people by
converting speech and text into natural sign language. A further
important development is the modification of this system to allow use on
mobile devices, which involves Orange working alongside IBM. IBM have
stated (corroborating source A):
`IBM chose the virtual signing technology developed at UEA as the
most advanced and flexible system available.'
Semi-automatic signing based on pre-defined phrases for information
provision is an important aspect of the impact of this technology for
the deaf community. This has been used in train announcement systems and
weather forecasting where the required phrases are from a limited set
and the key feature is the ordering that is required to produce the
appropriate information. The ViSiCAST project developed a tool to allow
non-signers to build weather forecast summaries using drop-down menus to
choose phrases covering a full range of general weather conditions.
Weather phrases have been mapped to natural British (BSL), Dutch (NGT),
and German (DGS) sign language, along with corresponding text.
Working with the Royal National Institute for the Deaf (now Action on
Hearing Loss), we have developed resources that enable authoring of sign
language content for websites. eSIGN has produced numerous websites
utilising our technology to provide sign language as a key method of
communication for non-hearing customers. Key examples include web sites
on employment in the Netherlands (IvD), deaf services in the UK (Deaf
Connections) and for the Federal Ministry of Labour and Social Affairs
(BMAS) in Germany (see corroborating source B).
A closely related application is the provision of information for
attendees at public festivals or events. A recent example is the
successful use of the JASigning software, which is based on UEA
research, at the Visual Festival organised by the Swiss Association of
the Deaf in May 2013. The Festival organiser has stated (corroborating
source C) that:
`the technology provided a novel and attractive way to communicate
information effectively for use by deaf participants.'
With signed-TV, the aim is to use 3D animation techniques to supplement
the use of on-screen human interpreters. For example, in conjunction
with Gamelab London we developed Performing Hands, a range of
interactive educational materials for the BBC which uses virtual signing
to support learning via interactive games. This resource was a finalist
for a BAFTA award in 2008. The same technology has been used for
programmes commissioned by the British Sign Language Broadcasting Trust.
Examples can be seen in on-line programmes such as Wicked Kids
and Little Wicked (see corroborating source D). Gamelab
(corroborating source E) say:
`these resources are unique animated materials for use by deaf
children and all depend on the avatar technology developed at UEA.'
and the British Sign Language Broadcasting Trust (corroborating source
F) state:
`production makes substantial use of the virtual signing avatar
technology created by researchers at UEA.'
The production of signed-TV is on-going and our technology is used by the
Netherlands Organization for Applied Scientific Research (TNO) to produce
accessible programmes for deaf children for Dutch TV, such as the popular
programme Het Zandkasteel. TNO have stated (corroborating source
G):
`We identified the UEA technology as the most suitable for our needs as
it enables us to produce cost-effective 3D animations that are
attractive to children.'
-
Foreign Language Learning
Our virtual signing technology has other applications where there is a
requirement to combine gestures and speech. Resources for primary
language education in the form of animated stories on DVD are available
under the brand LinguaSign (see http://www.linguasign.com/).
These stories combine standard animation for the main story line
alongside speech. The story characters speak with lip-sync animation
while performing gestures that reinforce the acquisition of vocabulary.
The stories have been reproduced using matching phrases in English,
Dutch, French, and Portuguese, voiced by native speakers. In 2013, more
than 50 primary schools undertook a trial of the resources for Key Stage
2 children. 62% of respondents to the survey at the end of the pilot
strongly agreed that the DVD was effective at improving pupil's speaking
and listening skills in a new language (corroborating source H) whilst
one teacher commented:
`My class and I really enjoyed using it. We found it engaging and
fun. A great way to prepare them for MFL [Modern Foreign
Language Learning]at high school.'
-
Cultural Heritage
In a further use of our technology, exploiting animation scripting
algorithms from the virtual signing system, animated characters have
been used to provide engaging stories to enhance the presentation of
cultural heritage sites managed by the Heritage Economic and
Regeneration Trust (HEART). The CEO of HEART has stated that
(corroborating source I):
`The technology developed by UEA, and utilised by us, has made a real
difference to the way in which we can present our cultural heritage to
a wide range of audiences and has enabled us to engage with
particularly hard to reach groups.'
Sources to corroborate the impact
[A] Letter from IBM confirming that IBM and UEA are developing the SiSi
system, held on file at UEA
[B] Letter from BMAS (German Federal Department for Social Affairs), held
on file at UEA
[C] Letter from Visual Festival 2013, held on file at UEA
[D] Wicked Kids and Little Wicked programmes available to view on-line
at:
http://www.bslzone.co.uk
[E] Letter from Gamelab London, held on file at UEA
[F] Letter from British Sign Language Broadcasting Trust, held on file at
UEA
[G] Letter from Netherlands Organisation for Applied Scientific Research,
TNO, held on file at UEA
[H] Analysis of SurveyMonkey results following completion of the primary
schools pilot 2013, held on file at UEA
[I] Letter from Heritage Economic and Regeneration Trust (HEART) Chief
Executive Officer, held on file at UEA