Computational Learning in Adaptive Systems for Spoken Conversation
NEW: annotated data archive released : archive link
Details: Project Number 216594 funded under EC FP7, Call 1. Start date: 1 March 2008, Duration : 36 months
IEEE Signal Processing Society Web article on The CLASSiC Project (february 2011)
News: All France Telecom 1013+ customer service calls are now handled by a CLASSiC system for automatic appointment scheduling (March 2010)
Public deliverables available (updated august 2011)
YouTube Video demonstrations, updated february 2009
CLASSiC researchers ran a Special Session "Machine Learning for Adaptivity in Spoken Dialogue Systems" at Interspeech 2009
CLASSiC researchers ran a Tutorial on "Reinforcement Learning for Adaptive Dialogue Systems" at EACL 2009
The overall goal of the CLASSiC project is to facilitate the rapid deployment of accurate and robust spoken dialogue systems that can learn from experience. The approach is based on statistical learning methods with a unified treatment of uncertainty across the entire system (speech recognition, spoken language understanding, dialogue management, natural language generation, and speech synthesis). This will result in a modular processing framework with an explicit representation of uncertainty connecting the various sources of uncertainty (understanding errors, ambiguity, etc) to the constraints to be exploited (task, dialogue, and user contexts). The architecture supports a layered hierarchy of supervised learning and reinforcement learning in order to facilitate mathematically principled optimisation and adaptation techniques. It is being developed in close cooperation with our industrial partner in order to ensure a practical deployment platform as well as a flexible research test-bed.
For the People involved at each site, please see PeopleThe CLASSIC project Wiki (members only)
The CLASSIC project aims at a qualitative leap in the robustness, flexibility, efficiency and naturalness of spoken dialogue systems, through new technologies based on the paradigm of computational learning and statistical modelling.
The project is expected to produce new mathematical models and computational techniques for spoken language understanding, dialogue management, and natural language generation. It will produce 4 different “showcase” spoken dialogue systems illustrating these advances, and it will provide new software and corpora for the future development of statistical spoken dialogue systems. Computational learning techniques will also be integrated with industry standard dialogue system development tools used by our industry partner, France Telecom/Orange Labs.
Potential Impact and Use
For end users (i.e. members of the general public and professionals who will use spoken dialogue systems in the future) the impact of CLASSiC will ultimately be more useable, robust, and efficient human-computer spoken dialogue interfaces, which are context-aware and adaptive.
CLASSiC therefore develops key technology and tools which will help meet some of the general goals of the ICT programme in FP7 – tools for the delivery of information technology to different individuals in a natural, user-tailored, adaptive, and intuitive manner. Speech interfaces are inherently inclusive in their support for non-expert (and even illiterate) users, who are only required to have basic spoken conversational skills in order to interact with IT services and devices.
One potential impact of CLASSIC is the development of a new paradigm for generic technology which will enable human-computer interaction in a conversational, user- adaptive manner, based on the user and their situation. This type of technology contributes to the EU’s objectives of providing European citizens with more efficient, robust access to IT services, and removing educational and linguistic barriers.
CLASSiC System demonstrations:
- "Lifting the Lid": Scottish Informatics Research for SMEs, 2008.
- Interactive Technology Festival, Cambridge, Sept 2009.
- Disney, Technical Team, December 2009 .
- Cambridge Science Festival, March 2010.
- ICT, Brussels, September 2010.
CLASSiC researchers took part in Google's 2010 Faculty Summit , Zurich on 8-10 February 2010.
Nhumi Technologies (Electronic Healthcare records: http://nhumi.com/): CLASSiC methods for semantic annotation transfer, feb 2010.
The PARLANCE project (European Commission, FP7, starting November 2011)