The MikroKARAT Distributed

Knowledge Acquisition Environment

MikroKARAT (Mikrokosmos Knowledge Acquisition and Review Assistance Tools) is a multi-user, client-server-based knowledge acquisition environment written in C++ using the Motif user interface under X Windows.

Goals

The goal of MikroKARAT is to develop a fast and easy-to-use acquisition environment which can be used by multiple geographically dispersed knowledge analysts to simultaneously update multiple large knowledge bases of various kinds.

Users should be prevented from accidentally obliterating each other's updates, yet be given the tools with which to cooperate as if they were in the same room rather than thousands of miles apart.

Updates should be checked immediately to prevent invalid data from being entered.

Possible Alternatives

The decision to develop an entire new acquisition environment stemmed from the deficiencies of the existing environment which was already in use, ONTOS. ONTOS is an ontology acquisition environment developed at Carnegie Mellon University in 1988, written in Lisp with an X Windows interface.

ONTOS has the following drawbacks:

  1. very slow
  2. single-user -- no arbitration on database access
  3. few consistency checks
  4. buggy display code
of which the first two are not only the most serious, but would have required major rewrites of the ONTOS code. An examination of the code indicated that a major effort had already been made to speed up execution, and retrofitting distributed access would have required an effort on a scale similar to reimplementing the underlying FrameKit frame representation package without resulting in any significant speedups.

Because of the perceived effort required to attempt an update to ONTOS, it was decided to implement an entire new system based around a new frame representation package implemented in C++ (FramepaC).

Client-Server Architecture

To support multiple simultaneous users, MikroKARAT uses a client-server architecture with a central database server to arbitrate access among the concurrent users. Each user runs an acquisition environment which acts as a client of the database server.

Knowledge Representation

!!! Frames are an associative memory: 
named frames contain
    named slots which contain
	named facets which contain
	    actual filler values
Values are retrieved by giving the name of the frame, slot, and facet containing the value(s) of interest.

Frames and Knowledge Bases

In our system, a knowledge base consists of a set of linked frames. Ontologies are a hierarchy of concepts connected by IS-A and PART-OF relations. Text Meaning Representations (TMRs) are linked by relations such as AGENT, THEME, ASPECT, and ATTITUDE. Lexicons are have sense frames linked by N, V, ADJ, etc. to superentry frames.

Permanent Storage of Frames

!!! FramepaC provides transparent demand-loading of frames from either a disk file or a server. Changes necessary to use persistent frames instead of in-memory frames:
  1. initialization call: initialize_VFrames_disk or initialize_VFrames_server
  2. cleanup call: shutdown_VFrames
  3. optional update calls: store_VFrame or synchronize_VFrames

Multiple databases of frames can be active simultaneously, each in a separate name space.

Frame Editor

A highly-configurable structured editor for frames is the main knowledge-entering tool.

The editor configuration can specify:

  1. required slots or facets
  2. permissible slots or facets
  3. types of fillers
  4. default values
  5. aliases for slot names

The editor thus supports both free-form frames such as ontological concepts and very rigid frames such as lexical entries.

Graphical Browser

The browser provides an overview of a knowledge base, and the ability to !!!vertical browser!!!
!!!TMR browser!!!

Configuration

Browser Configuration Editor Configuration

Glosser

A very shallow, heuristically-driven English generator for TMRs.

Glosser Display
Glosser Output

Linguistic Utilities

MikroKARAT provides access to all the usual utilities:
  1. dictionary lookup
  2. corpus lookup
  3. KWIC searches

It also provides a special frame-search capability. The names of all frames in a knowledge base which match the specified criteria are displayed in a pick list, from which the user may select one or more to edit.

KWIC Searches

KWIC Search Dialog

Frame Searches

Query Screen
Result of Query

Inter-User Communication

Chat Utility

To support cooperative updates to a knowledge base, MikroKARAT provides interactive communication between users. (not yet implemented, but one-line messages can be sent to other users) Sending a One-Line Message

Work Remaining

Planned Enhancements

  1. Concept Cluster Manager
  2. To-Do lists
  3. Annotations
Abstract
Images
Tables

Ralf Brown (ralf+@cmu.edu)