GroupCoding and Communication

Coding and Communication

Coding and Communication

External homepage: http://www.bk.isy.liu.se/

The Coding and Communication research group is closely connected with the Image coding Group in the Electrical Engineering Department at Linköping University, which started in 1972 when Robert Forchheimer began his Ph.D. thesis work on "Vectorial Delta Modulation for Picture Coding" under the supervision of Prof. Ingemar Ingemarsson. An early contribution was the suggestion of doing motion compensation in the spatial domain in a hybrid coding loop, as presented by Robert Forchheimer in 1981. In the beginning, the scheme was critisized for being too computationally expensive, since it requires an inverse transform in the coder, but has by now become an established technique. All modern video coding standars, such as H.261, H.263, H.26L, MPEG-1, MPEG-2, and MPEG-4, are based on this principle. A later major contribution was the introduction of 3D-model-based coding in 1983, now becoming an established technique through standardization (MPEG4-SNHC).

People

Robert Forchheimer
Professor, Santa Anna Research Faculty
+46 70 522 33 10
robert.forchheimer [at] ri.se

Martin Mileros
Researcher
+46 70 835 12 36
martin.mileros [at] ri.se

Harald Nautsch
Researcher
+46 70 171 87 15
harald.nautsch [at] ri.se

Dialogue Management and Automatic Contact Center.

The group is involved in research on language technology and interaction from a variety of perspectives, ranging from empirical  investigations of human-computer natural language interaction to  methods and tools for development of dialogue systems.  At present  focus is on two projects, BCORN and ACC.
BCORN is a generic dialog strategy with domain and task-independent  conventional, information-providing and recommendation capabilities.  BCORN is an ambitious project aiming at developing a model for  dialogue management able to handle the dialogue in any natural  language interaction for any application.
The first version of BCORN is limited to conversational recommender  systems and an instance of such a system, CoreSong, has been  developed in the application domain of preference-based music  recommendation.
BCORN is inspired by the work on a  subsumption architecture for  building intelligent creatures , as advocated by Rodney Brooks 1991.  The BCORN model is constructed using dialog behaviors that each corresponds to a natural chunk of an agent¡Çs dialog strategy.  Some dialog behaviors are  general (e.g. a conventional dialog  behavior of greeting and farewell), and some are specific (e.g. a  recommendation or preference interview behaviour). A generic task and  domain-independent dialog agent thus needs a dialog model that  includes dialog behaviors that can co-exist but at the same time have  a clear order of priority. It is also imperative that the model  adjusts to the needs of different back-end resources at hand in a particular application.
Similar to the  model proposed by Brooks, BCORN is constructed using  state automata---called dialog behavior diagrams (DBD).  The DBDs  express dialog behaviors of the dialog agent that are both natural  conceptually and efficient computational mechanisms. The complete  dialog strategy of the agent is the result of running several DBDs in  parallel in a DBD strata machine, leading to an emergent coherent and  flexible agent behavior.
ACC is a project with the goal of building an Automatic Contact  Center. This research project is conducted in cooperation with  Icepeak. Icepeak develop and sell products for call centers to a  number of Swedish customers. The ACC project aims at a system that  can respond automatically to FAQs. A first step is to build a  database of FAQs automatically from humans calling the contact  center. The database can also be used for statistical analyses of  call canter traffic. In the next phase there will be a hypothesis  generator helping the agent respond to questions from customers based  on previous question-answers. Finally, an automatic FAQ will be  developed.
In this project we develop the non-speech parts, i.e. taking textual  input. The work is centered around ACC-core, a workbench for  experimenting with a variety of techniques, such as LSA and decision- tree learning, for information extraction. ACC-core also includes a  variety of filters, stemmers and other linguistic means for  transforming data.