CERC Technical Report Series

Technical Memoranda

CERC-TR-TM-92-001

KNOWLEDGE-DIRECTED GRAPHICAL AND NATURAL LANGUAGE

INTERFACE WITH A?KNOWLEDGE-BASED

CONCURRENT?ENGINEERING?ENVIRONMENT

Zenon KULPA

Michael SOBOLEWSKI

August 1992

ACKNOWLEDGEMENT: This effort has been sponsored by Defense Advance Research Projects

Agency (DARPA), under contract No. MDA972-88-C-0047 for DARPA Initiative in Concurrent

Engineering(DICE).

Concurrent Engineering Research Center

West Virginia University

Drawer 2000, Morgantown WV 26506

2

KNOWLEDGE-DIRECTED GRAPHICAL AND NATURAL LANGUAGE INTERFACE WITH A?KNOWLEDGE-BASED
CONCURRENT?ENGINEERING?ENVIRONMENT(*)

Zenon KULPA(**) and Michael SOBOLEWSKI

Concurrent Engineering Research Center
West Virginia University, Morgantown, WV 26506

Abstract

Effective on-line communication between cooperating users with the mediation of machines (Virtual Team concept) is a necessity in new-generation concurrent engineering design and manufacture systems. The communication means for solving this problem are natural language and graphics. Managing the graphical and natural language user interfaces is a task full of ill-structured problems for which the knowledge-based approach offers a general solution methodology. Our knowledge-based expert system shell, implemented in Smalltalk, provides support mechanisms for flexible construction of such user interfaces. The natural language knowledge definition and editing is based on the novel percept-calculus knowledge representation scheme used in the system. The graphical module of the interface is capable of presenting diverse kinds of qualitative and quantitative data in various graphical forms which can be made interactive, serving as data input, data navigation, or design variation tools. An appropriate graphical knowledge base contains knowledge about general methods of graphical data presentation.

Keywords

Graphical interface; Knowledge base; Collaborative environment.

1. Introduction

Knowledge-based systems offer engineers and manufacturers various types of expertise on design and manufacturing requirements and constraints during various stages of design and manufacture. As a result, the knowledge-based programming is becoming the current methodology in various engineering applications, especially within the concurrent engineering framework aimed at uniting the whole design and manufacturing process. However, low-level knowledge description languages (frames, semantic networks) and conventional programming languages (LISP, C) as well as the traditional, command-oriented user interfaces used in most of these systems are too difficult and unnatural for designers and product developers to use. Consequently, long system and product development times result, with negative impact on product cost, quality, and supportability.

In the essence of the problem of integrating machines and human experts within new-generation concurrent engineering design and manufacture systems lies the necessity of effective on-line

* This work has been sponsored by Defense Advanced Research Projects Agency (DARPA), under contract

No.?MDA972-88-C-0047 for DARPA Initiative in Concurrent Engineering (DICE).
** Current address: Z. Kulpa, Artificial Intelligence Center, George Mason University, Fairfax, VA 22030.

3

communication between cooperating users with the mediation of machines (Virtual Team concept). The communication should be easy, fast, data-intensive, and error-free. To achieve this, it should be natural for users, despite their great diversity of types and needs; namely, it should be at the same time conceptually unified and diversified. The communication means that are able to satisfy these seemingly contradictory requirements are natural language and graphics. Natural language is general and unified, being common to all users, and can easily be specialized to their particular needs. Graphics, in addition, besides being a long-established means of communication within the engineering and manufacturing communities, is also a medium able to efficiently convey a vast amount of diverse data and knowledge.

Moreover, a lot of problems encountered in both engineering and graphic design are not amenable to purely algorithmic solutions. These problems are often ill-structured (the term ill-structured problems is used here to denote problems that do not have an explicit, clearly defined algorithmic or formal solution), and experienced designers deal with them using informal judgement and gathered experience. In the construction of graphical interfaces and the graphical presentation of complex data, such problems are abundant. For example, such a seemingly simple task as drawing binary trees nicely cannot be solved algorithmically in less than exponential time (Supowit and Reingold, 1983). Instead, to get suboptimal but practical solutions to problems of this kind, the knowledge-based approach has been proposed (Berztiss, 1990). Also, to make true integration of users and machines possible, the interface system should be able to supply by itself most of the obvious and repetitive data using context-dependent default values and intelligent guesses of user intent, based on the domain knowledge, the task at hand, and the user profile. To make this possible, modern intelligent interfaces incorporate knowledge-based techniques both at the interface design stage and as part of user interface managers (Reilly and Roach, 1984; Myers, 1988).

The implementation of natural language knowledge definition and editing critically depends on the intricacy of translation between natural language constructs and internal knowledge representation structures. This is a function of the chosen knowledge representation method. Three major paradigms compete in the representation of knowledge: frames, rules, and logic. All are based on some abstract concepts of knowledge. However, in humans, knowledge and thinking are believed to be based mostly on perceptual data processing (Arnheim, 1969). This fact is, in turn, reflected in the structure and semantics of natural languages. The novel percept calculus knowledge representation scheme (Sobolewski, 1989a; 1990; 1991b), based on this assumption, facilitates the natural-language knowledge definition.

In Section 2 of this paper we introduce the architecture and concepts underlying the DICEtalk knowledge-based engineering environment. In Section 3, we discuss in more detail the simplified natural language method of knowledge definition implemented in the system, and in Section 4, we address the graphical knowledge base approach to the graphical interface and data presentation capabilities of the system. In Section 5 we summarize briefly our experience with implementing a number of experimental applications within the concurrent engineering framework.

2. Knowledge-Based Environment

The DICEtalk is a knowledge-based engineering design and development environment (Sobolewski, 1990). It has been implemented in the Smalltalk-80 object-oriented language. The general architecture of the system was dictated by the need to efficiently integrate the knowledgebased programming with intelligent graphical data presentation and collection capability. It consists of an inference engine, graphical and application knowledge bases, and a graphical user interface. The inference engine implements rule-based deduction with a procedural attachment mechanism

4

Graphical
Knowledge
Base