Specification by Reformulation: A Paradigm for Building Integrated User Support Environments

John Yen, Robert Neches, Michael DeBellisAAAI 1988

Abstract

This paper presents "Specification by Reformulation," a generalized human-computer interaction paradigm designed to unify diverse tasks—such as database querying, electronic mail composition, and note-taking—into a single, consistent environment. By implementing the BACKBORD system, the authors demonstrate how iterative refinement of partial specifications, supported by a formal knowledge representation, allows users to navigate complex data structures without mastering specialized query languages. The system acts as an "interface shell," reducing development overhead and cognitive load for the user.

Introduction

In the late 1980s, the proliferation of large databases and complex software created a significant usability gap. Users often lacked the knowledge of internal data structures or the specific query languages required to retrieve information. While early systems like RABBIT introduced "retrieval by reformulation" (allowing users to find data by tweaking examples), these tools were often limited to specific applications and lacked a "top-down" navigation component. This paper addresses these limitations by proposing a generalized paradigm for building integrated user support environments that bridge the gap between user intent and system requirements.

The Specification by Reformulation Paradigm

The core of the proposed approach is the Reformulation Loop. Interaction is treated as a cycle where the user provides a partial specification of the desired object or action. The system then provides feedback by displaying matching instances and guidance on how to refine the description. This process is built on several technical pillars:

  • Knowledge Representation (NIKL): The system utilizes NIKL, a "KL-ONE" style language, to represent data as a hierarchy of "concepts" and "roles" (similar to frames and slots).
  • Search in Description Space: Users navigate a "description space" rather than just raw data. Operations such as Generalize, Specialize, and Siblings allow for movement through the knowledge taxonomy.
  • Automatic Classification: An underlying classifier determines where the user's current partial description fits within the existing taxonomy, enabling the system to suggest relevant constraints and identify missing information.
The BACKBORD interface illustrating the reformulation menu and matching instances for electronic parts.

The BACKBORD System

BACKBORD serves as a domain-independent "interface shell," similar to an expert system shell. By providing a domain-specific taxonomy and a set of actions, developers can quickly create new applications that leverage the same interaction logic. This architecture ensures Consistency as a Feature: by using the same interaction logic for email, notes, and database queries, the cognitive load on the user is significantly reduced.

Furthermore, the system enables Contextual Integration. Because different tools share a common paradigm and underlying knowledge base, they can pass context seamlessly. For example, if a user finds a specific "Part" in the database browser and then opens the mail tool, the system can automatically populate the email fields with that part's details.

Integration of the FAST mail interface and TINT note-taking facility within the unified BACKBORD shell.

Key Results and Applications

The authors validated BACKBORD by integrating three distinct services:

  1. Database Retrieval: A retrieval aid for standard electronic components used in procurement.
  2. FAST (Electronic Mail): A structured interface for sending procurement messages.
  3. TINT (The Intelligent Note Taker): A facility for creating and organizing structured notes.

The results demonstrated that building new applications became a process of customization rather than ground-up development, leading to reduced code size and easier maintenance. The system also proved capable of providing active user guidance, such as identifying which fields were missing to make a procurement email "sendable."

Limitations and Future Work

The effectiveness of Specification by Reformulation is heavily dependent on the quality and depth of the underlying NIKL knowledge base; a poorly defined taxonomy renders the system's guidance useless. The authors also noted limitations in the version of NIKL used, such as an inability to handle disjunctive value restrictions (e.g., "Part A OR Part B"), which required custom extensions. Future work involves exploring how well this guidance logic scales to much more complex or less structured domains.

Summary updated

BibTeX
@inproceedings{yen1988specification,
  author = {John Yen and Robert Neches and Michael DeBellis},
  title = {{Specification by Reformulation: A Paradigm for Building Integrated User Support Environments}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {1988},
}
Figure