Information Technology Reference
In-Depth Information
from the constraints of the clinical domain at hand (i.e., the constraints for meanings
defined by the static and dynamic descriptions of the clinical environment and the
clinical goal of operation) [9]. The key observation is that, in general, the understand-
ing of natural language cannot be separated from the clinical domain, (i.e., clinical
medicine in general and in specific operational environments). Thus, a proposed solu-
tion is domain-specific symbolic language modeling as initially developed and stud-
ied in [8-11] for emergency department chief complaints.
In the context of open clinical agents, a workable mechanism for natural language
understanding has to be capable of learning on the human abstraction level with the
implication that semantics can in principle be automatically and natively constructed
through a design-while-use paradigm to support gradual evolution of system func-
tionality and experience. In practice, this means the ability to learn, in a principally
unbounded case space, through user interactions having the expressivity of natural
language. Supporting the understanding and integrated learning process is a con-
stantly augmented and unrestricted linkage of explanations expressed as computable
explanations (here natural language labeled ontological relations). Initial work on how
this augmentation (i.e., learning) should be done is presented in the chief complaint
case study [8-11].
As part of an abstract architecture for reflective clinical agents (Figure 5), the link-
age (' Explanations ') provides the associative power to provide for the semantic dis-
ambiguation of both natural language user interactions complementing and working
towards the goal of operation (though the physical ' Action interfacer ' component) and
natural language labeled ontological relations of the linkage itself.
Fig. 5. Abstract architecture of an open clinical agent
Apart from general descriptions of the clinical domain, the linkage embeds a di-
verse, comprehensive set of direct observations characterizing the clinical domain as
well as the clinical goal(s) of operation itself. The descriptive annotations LT and ST
denote ' long-term memory ' and ' short-term memory ', concepts widely used in cogni-
tive computing and expert systems research. The observations include anything from
ordered laboratory result data to equipment and personnel location data and from
genetic information to medical imaging data. A key topic in this context is
Search WWH ::




Custom Search