Automatically Generating Natural Language Reports in an Office Environment

AUTOMATICALLY GENERATING NATURAL LANGUAGE REPORTS IN AN OFFICE ENVIRONMENT Jugal Kalita and Sunil Shende Department of Computer and Information Sciences University of Pennsylvania, Philadelphia, PA 19104 ABSTRACT In this paper, we describe a system which models a set of concurrent processes that are encountered in a typical office environment, using a body of explic- itly sequenced production rules. The system employs an interval-based temporal network for storing histor- ical information. A text planning module traverses this network to search for events which need to be mentioned in a coherent report describing the current status of the system. In addition, the planner also combines similar information for succinct presentation whenever applicable. Finally, we elaborate on how we adapt an available generation module to produce well- structured textual report for our chosen domain. 1. INTRODUCTION This paper describes the implementation of a text gen- erating system which produces natural language re- ports on the status of a system of inter-related pro- cesses at various stages of progress. The motivations behind this project were to model a system of concur- rent processes and to successfully generate well-formed text about their temporal behavior. We view a process as an ordered sequence of events over time where an event refers to an atomic action by one of the partic- ipating agents. In many AI applications, monitoring the state of a system of processes is deemed essential. The ability of a system in such an environment to de- scribe its actions in natural language will be certainly very useful. As our sample domain, we chose a scenario where the system is used to assist the secretary for an aca- demic journal by keeping track of a paper submitted for publication. The process being modeled is that of paper-submission with the usual participating agents being the author of the paper, the journal editor, the reviewers assigned to evaluate the paper and the jour- nal secretary. In the rest of the paper we present methods to rep- resent the knowledge of the chosen domain, to model historical knowledge base of events and processes, to appropriately order the contents of the historical in- formation store for presentation based on temporal re- lationships and other relevant factors, and to produce coherent English text describing the events in the do- main. The system as described herein has been fully implemented, and currently avenues for further im- provement are actively being pursued. Section 2 of this paper gives a brief description of the domain being modeled along with an overview of the system components. Section 3 describes the rep- resentation of the processes in terms of a network of rules. Section 4 elaborates the nature of the historical information store. Section 5 deals with text planning which is crucial for selecting information to be pre- sented in a coherent manner. Section 6 details the actual process of text generation. Finally, we con- clude with a discussion about enhancements to the existing system which can contribute towards a more general implementation with the …

Content truncated to 500 words