Companions Research
The overall scientific and technological objective of Companions is the fusion of human-computer communication technologies at a higher level than anything achieved till now, so as to produce a human-machine interface that simulates a persistent human-like presence in a machine, one with emotional aspects that forms a long-term bond with its user.
Speech input to a Companion will require speech technologies to handle automatic speech recognition. Natural language processing includes natural language understanding, a dialogue manager (which draws on knowledge representation) and natural language generation, leading to an output, for example text to speech.
In order to achieve this we will be building on existing technology provided by the project partners, and researching and developing technology in original directions.
Research objectives
- Developing a new approach to dialogue management (i.e. understanding and deciding what to say next) based on machine learning.
- Advance research in Embodied Conversational Agents (ECAs) so as to combine them effectively with speech and language technology. This includes speech synthesis, speech recognition, dialogue management, and appropriate knowledge representation so as to allow the ECA to display personality, and use politeness and humour in different types of dialogues.
- Use and develop advanced machine learning techniques so as to improve performance, user adaptation, scalability and portability (to new languages and domains).
- Develop and extend speech technology by developing ASR and TTS for a new language (Czech), and extending the capability of existing systems to handle emotion and affective aspects of human-user interaction.
- Investigate and develop appropriate ways of having a persistent interaction with the user, with the Companion building up a representation of the user (including a narrative of the user's life events). Initially this research will focus on building a life narrative based on discussion of digital images.
Expected results
- An overall development methodology for multimodal ECAs based on a range of advanced language technologies, allowing for rapid adaptation to new contexts.
- A set of Companions components which can easily be adapted and integrated for different domains and languages. This will be shown by the demonstrators, which will in fact be just instances of what Companions can be about.
- A new paradigm for multimodal architectures, which will depart from overly complex software engineering approaches and will rely on adaptive architectures based on machine learning and specifically Bayesian methods.
- A set of demonstrators across domains (e.g. Senior Companion, and Health and Fitness Companion), across at least two languages (English and Czech), and across technologies (touch screens, PDAs and mobiles).
Updated: 16 May 2008 14:51 PM