|
| ||
|
Expression Composer
Yolanda Gil (gil@isi.edu, (310)448-8794)
Jihie Kim (jihie@isi.edu, (310)448-8769)
Jim Blythe (blythe@isi.edu, (310)448-8251)
Expression Composer During the summer evaluation, users frequently had difficulty finding an object in a CMAP that was not linked in a place they expected, or was referred to in a different way. We witnessed similar problems during the HPKB year 2 evaluation where, for example, a SME wanted to refer to "the units" to access the military units involved in a task, but the knowledge base required the more precise expression "the units of the blue forces assigned to the task". This spurred our development of the Expression Composer (EC), which takes an imprecise user expression and searches for a valid KB representation that matches the expression. We have already built the Expression Composer, which works with the EXPECT English editor, and which sparked quite a bit of interest when we demoed it in Tucson. Our current EC takes an expression that the user types, such as "max airport landing", and shows the user precise expressions based on the exact representation of those terms in the KB, such as "find the maximum of the available landing distance of the runways of the airport" (see http://www.isi.edu/expect/rkf/composer.ppt). In this way the EC allows users to make imprecise statements and still refer to relevant entities in the KB. While the Shaken project has shown in year one that the graphical approach to knowledge entry is an effective one, we need to develop ways to help users manage the clutter that can easily appear as CMAPs get larger. We'd like to integrate the EC with SHAKEN and KM, and extend it to support not only KB search but also search/expansions on a CMAP to support faster knowledge entry. For example, by typing "cell nucleus" the user could open relevant CMAPS in which the paths leading to the appropriate objects are already expanded. This provides a useful tool to help reduce CMAP clutter. Another extension would be to use lexical/linguistic resources such as thesauri (our SME dictionary and others). This would allow the composer to handle constructs such as the use of the possessive (e.g. "the cell's membrane") and synonyms, and would help us move towards a more natural interaction style for SMEs. We expect to draw from related work in conceptual query systems for information retrieval (e.g. [Sparck-Jones and Willett 97; Cunningham 99]). A very important new area that this would allow us to embark on is helping a user teach the system techno-speak, i.e., domain-specific vocabulary that will make the student use more technical terms to describe what was learned. This is something human teachers do once the student has conceptually understood the material, and we can see that SHAKEN this past year was built to do that part of understanding the material but had no facilities to let a user teach it to speak in domain terms proficiently. A brief description of the expression composer is given in this IJCAI paper, in section 4. More details can be found in these powerpoint slides. | ||