---
title: "STAMMER2: A Rule-Based Application"
authors:
  - "Robert J. Bechtel"
doi: ""
url: "https://ijcai.org/proceedings/1981/200"
pdf: "https://ijcai.org/proceedings/1981/0200.pdf"
venue: "Proceedings of the Seventh International Joint Conference on Artificial Intelligence"
year: 1981
pages: "1053--1"
track: "Program Demonstrations"
keywords:
---

# STAMMER2: A Rule-Based Application

Robert J. Bechtel

*Proceedings of the Seventh International Joint Conference on Artificial Intelligence*, pages 1053--1.

## Abstract



## Full text

## ST~~R2: A RULE-BASED APPLICATION*

## Robert J. Bechtel

Naval Ocean Systems Center

San Diego, CA 92152

STAMMER2 is a working rule-based system for identifying objects detected by sensors on board Navy vessels and for interactively explaining the identification process. It is currently in use as a demonstration system at the 1\aval Postgraduate School in Monterey, California, and serves as the basis of continuing artificial intelligence development efforts at the Naval Ocean Systems Center.

Many rule-based systems utilize a data base which is fixed and regarded as complete, except for inrerenees added by the system itself. In contrast, STAMMER2 operates in an environment which is never complete and where later information may supersede or even negate earlier data. Others also deal with the issues of incomplete and inconsistent data, but propose solutions differing from those adopted in STAMMER2. Besides handling unstable inl"ormation, STAMMER2 is required to process large amounts of data as it is received. This combination of constraints introduces problems for

which novel solutions have been adopted.

STAMMER2 operates on a data base of assertions relating to ocean-going craft. Periodically, packets of information in the form of messages are added to the data base. The unsolicited nature of data entry suggests a data-driven approach, and the rule interpreter operates primarily in a bottom-up, ,forward-chaining mode. To facilitate the bottom-up processing, inference . proceeds in an incremental fashion. That is, a rule may be partially evaluated on the basis of currently available inl"ormation. The evaluation is then suspended and may be revived when further relevant information is received. This technique eliminates duplication of effort by the rule interpreter. An implication of this approach is that if later information sa tist"ies the remaining conditions of a rule in a prior context, then that context is recalled and

the rule finally fires.

The confidence computation mechanism, while using ·formulas similar to those of MYCIN, has three novel aspects. The first two, cycle breaking and invalidation of earlier inferences by later information, are a result of a dynamic confidence calculation which is undertaken only on request. Confidences are not stored. The third innovation is the introduction of an "UNLESS" logical operator to permit default inferences based on an absence of inrormation.

This work performed in part while the author was employed by SOC Integrated Services under NOSC contract N00123~76-C-0172

A facility is provided for the system to explain the reasoning underlying its inferences. For this purpose, stored with each system inferred assertion is a "derivation record" which contains all the rules and particular condition satisfiers which participate in concluding the assertion.

The explanation system has four interesting features:

1. The specific assertions involved in a rule application are provided
2. 2 . Graphics are used to illustrate relevant aspects of the ocean situation
3. General question-answering are provided with respect base capabilities to the data
4. User interaction is in a constrained "natural language-like" format.

## ACKNOWLEDGEMENTS

Paul Horris and Dennis Kibler made major contributions to the development of the system.

## References

1. DC McCall, PH Morris, OF Kibler, and RJ Bechtel, "STAMMER2 Production System for Tactical Situation Assessment", Technical Document 298, Naval Ocean Systems Center, October, 1979.
2. R Bechtel, P Morris, and "Incremental Deduction in a Environment", In ~ Victoria, B.C. May, 1980. i:l Kibler, Real-Time CSCSI-80.
3. EH Shortliffe, Compyter-Based Medi.,al Consultations: MXCIN. American Elsevier, 1976.