For all the four components listed in 1.1.1
At the Information Theoretical observation level, digital bandwidth and
digital communication concepts are relevant, describing the pure
informational characteristics of the channels (e.g., entropy). At this
level of observation, issues of data compression, but also of optimal
search processes play a role. An example is the number of bits/s needed to
faithfully transmit the recorded images of a human talking face over a
COMHIC channel.
At the Cognitive level, representational and procedural
aspects have to be made explicit, in syntax and semantics.
Furthermore involved are all those components of pattern recognition
processes which exceed the level of the basic signal features, and require
information of a more abstract nature. An example is the use of linguistic
``top-down'' information in the disambiguation of speech, handwriting, or
the use of high-level information in the interpretation of facial
movements. Other typical topics at the cognitive level are the notion of
short-term memory with its limited capacity in the human:
conceptual items [223,43]. And, last but not least, an essential
property of the Cognitive level is the process of learning.
Because of the fact that the Cognitive level is the mediating level between
low-level processing and the high level of intention, to be clarified
in the next paragraph, it is at the core of the project.
At the Intentional level, the goals and beliefs of the involved agents have to be made explicit. It is still one of the problems of current computer science and AI that this level is often ignored, whereas it is essential to the human user. When asked what he/she is currently doing on the computer, the answer of the user will most likely not be on the level of the tools (e.g., ``I'm pressing this button''), but on the level of goals and intentions (``I'm looking for this nice picture in the Louvre homepage''). Typically, much of the research in human-computer interaction involved experimental conditions of goal-less behavior. At best, goals artificially imposed on the user subjects are used. However, there is a fundamental difference between ``playful browsing activity'' and "goal-oriented search in a financial database under time pressure". Such a difference at the intentional level will have a large influence on low-level measures such as movement times, reaction times, and most probably even on perceptual acuity. Another area where the disregard of goal-oriented behavior of the human users leads to serious problems is in the automatic recognition of HOC data. In most pattern recognition methods, training of the algorithms is needed, using a database of human-produced data. However, using isolated words which are collected outside a ``natural'' application context for the training of speech or handwriting recognizers, will yield data which is not generalizable to the type of human output (HOC) data produced in an application context involving realistic goals. The problem of intention becomes also clearly apparent in the case of teleoperating applications, where the goals of the user must be continuously inferred from action by the computer.