The Generic Frame Protocol

IJCAI 1995

TL;DR

This paper introduces the Generic Frame Protocol (GFP), a Common Lisp API that provides a uniform interface to diverse Frame Knowledge Representation Systems (FRSs). GFP aims to reduce duplicated effort in developing AI tools by enabling them to operate across multiple FRSs, promoting knowledge reuse and portability.

What Problem Does This Paper Address?

The field of Artificial Intelligence has a long history of using Frame Knowledge Representation Systems (FRSs) for tasks like planning and expert systems. However, a significant challenge has been the fragmentation of the FRS landscape. With over 50 identified FRSs, each with its own idiosyncrasies, developing generic tools like graphical browsers or editors that work across different systems is incredibly difficult and leads to massive duplication of effort. Furthermore, reusing knowledge represented in one FRS within another application or system is often hindered by incompatible representations. This paper addresses the need for a standardized way to interact with these diverse FRSs, making them more accessible and interoperable.

Methods

The core of GFP is a set of Common Lisp functions that act as a generic API. This API is built upon a common conceptualization of frames, slots, facets, and inheritance, abstracting away the specific details of individual FRSs. GFP defines a reference model that captures commonalities across FRSs, using terms like "knowledge base," "frame," "class," "instance," "slot," and "facet."

A key aspect of GFP is its handling of FRS diversity through "behaviors." These behaviors allow an FRS implementation to declare which of its capabilities are supported, and an application can specify which behaviors it requires. This enables applications to adapt to the capabilities of the underlying FRS, or for GFP to emulate missing behaviors. GFP also employs a "runtime translation" paradigm, where knowledge is translated into a GFP-compatible form during application execution, contrasting with specification-time translation used in other knowledge-sharing efforts.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • The "Art of Compromise": The authors acknowledge that achieving all design goals (simplicity, generality, no legislation, performance, consistency, precision, language independence) simultaneously is impossible. GFP represents a deliberate compromise to balance these often conflicting objectives.
  • Behaviors as a Flexibility Mechanism: The concept of "behaviors" is crucial. It allows GFP to remain generic while accommodating the inherent differences between FRSs. This is a clever way to avoid "legislating" how FRSs should operate, instead focusing on what they can do.
  • Programmatic vs. Declarative Interface: GFP opts for a procedural (programmatic) interface, arguing it's more convenient for embedding within other software and can offer better performance than declarative (Tell/Ask) interfaces, especially for meta-level queries.
  • Runtime Translation for Interoperability: The runtime translation approach is a significant departure from specification-time translation. It allows for dynamic adaptation and reuse of knowledge and tools without requiring pre-conversion of entire knowledge bases.

Key Results

The paper details the design goals and reference model of GFP, outlining its programmatic interface and implementation strategy. It reports on four FRS-specific implementations of GFP: for LOOM, Ontolingua, THEO, and SIPE-2. Experimental evaluations using a LOOM implementation showed that GFP operations were 1%-50% slower than direct LOOM operations, with higher overhead for operations without direct counterparts in LOOM. However, the absolute overhead was noted as very small. GFP is already in use in two applications: the SIPE-2 planning system and the EcoCyc project at SRI.

Limitations and Future Work

The authors acknowledge that the GFP reference model is not exhaustive; it does not provide rules, methods, or contexts, nor does it offer a full concept-definition language. The model of knowledge bases is also simplistic, lacking notions of dependencies or imports. Future work includes extending GFP to cover more FRS features, allowing GFP calls to traverse a network, and potentially developing a declarative assertional facility.

Summary updated

BibTeX
@inproceedings{1995generic,
  title = {{The Generic Frame Protocol}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {1995},
}
Figure