Task support in an office system

Reviewer: Ian Hugh Witten

This paper describes a system which acts as an interface to a set of existing office tools that accomplish such tasks as mail, forms processing, calendar maintenance, etc. It combines a semantic database, a library of procedures, and a natural language query and help facility. The semantic database records properties of office objects, such as forms and mail messages, as well as world objects such as vendors and supplies. The procedure library, which is built by a systems analyst, contains hierarchically-organized routines for tasks such as “purchase items” and “hire graduate students.” The natural language component allows a user to make simple requests about the status of the current tasks, and generates explanations of user errors. The center of the system is the method of procedure specification. Predefined procedures help the user to structure his activity. They are specified in an intermediate language which allows concurrency. Each procedure definition includes an English description of what the procedure is supposed to accomplish, a sequence of actions, constraints which must hold during execution, and precondition and satisfaction clauses. The actions constitute other procedures and, at the lowest level, invocations of system tools. Precondition and satisfaction clauses cause demons to be attached to uninstantiated database objects which will fire on instantiation. The system is open-ended in that a primitive action might simply ask the user to do something (obtain information, check a constraint) which cannot be automated. This might involve him making a phone call, composing mail, etc. The user is free to employ any method for carrying out such steps. Procedure satisfaction clauses allow the system to monitor the overall activity and check that the specified goal is met. In this way, both structured and unstructured tasks can be addressed. The work is interesting and novel. In particular, the use of a single mechanism to initiate and monitor both internal and external activities required to complete a task may be a major step forward, especially if the system can meaningfully accommodate external actions, which only partially satisfy their goals, and readjust its plan accordingly. However, the description of the actual mechanisms involved is rather sketchy. From the paper alone, it is difficult to judge exactly what has been accomplished and what remains at the idea stage.