Constraint structures are presented as a new knowledge structure to encode both engineering part designs and constraints on those designs. Using a common representational form for both designs and constraints results in a number of practical advantages for a design evaluation system grounded on constraint structures. First, a graph unification algorithm generalized to operate on constraint structures is used as a mathematically elegant means of applying constraints to designs. As a side-effect of constraint structure unification, constraints are attached to design representations. Once the constraints become a part of the design representation, a part is able automatically to reevaluate relevant constraints following interactive changes to design parameters without re-invoking the unification algorithm.>

Unification-based NL parsers that copy argument graphs to prevent their destruction suffer from inefficiency. Copying is the most expensive operation in such parsers, and several methods to reduce copying have been devised with varying degrees of success. Lazy Unification is presented here as a new, conceptually elegant solution that reduces copying by nearly an order of magnitude. Lazy Unification requires no new slots in the structure of nodes, and only nominal revisions to the unification algorithm.

This paper presents a discussion of the techniques developed and problems encountered during the design, implementation, and experimental use of a portable natural language processor. Datalog (for “database dialogue") is an experimental natural language query system, which was designed to achieve a maximum degree of portability and extendibility. Datalog uses a three-level architecture to provide both portability of syntax to new and extended tasks and portability of semantics to new database applications. The implementation of each of the three levels, the structures and conventions that control the interactions among them, and the way in which different aspects of the design contribute to portability are described. Finally, two specific, implemented examples are presented, showing how it was possible to transport or extend Datalog by changing only one “layer” of the system's knowledge and achieve correct processing of the extended input by the entire system.