A Flexible Interface for Linking Applications to Penman’s Sentence Generator

A Flexible Interface for Linking Applications to Penman's Sentence Generator Robert T. Kasper USC/Information Sciences Institute 4676 Admiralty Way, Suite 1001 Marina del Rey, CA 90292 U.S.A. Abstract The Penman text generation system has been used within several different experimental application domains, demonstrating that it provides the basis for an adaptable general purpose text generation capability. Linking with these applications also indicated several ways that Penman's interface with applications could be improved. Penman's interface with applications is described, focusing on SPL, a newly developed sentence plan language. SPL is a notation that can be used by text planning programs to specify plans for sentences at multiple levels of abstraction and varied amounts of detail. Sentence plans are interpreted with respect to a collection of predefined knowledge sources, thereby minimizing the size and complexity of inputs that must be dynamically constructed by the application to generate individual sentences. 1 Introduction Penman is designed as a domain-independent text generator that can be installed in an application system to generate text for the application on demand. Motivating this design is the hypothesis that a significant amount of knowledge about language and how it is produced can be re-used in different application domains. Assuming that this hypothesis is correct, the construction of an effective interface between a general-purpose text generator and application programs is a significant problem to be solved in text generation research. Penman's interface with applications is the subject of this report, focusing on how it has been designed to achieve a high degree of flexibility. In designing Penman's interface, we faced the following dilemma: how to make the system easy to use by an application programmer without compromising its expressive power. Penman's grammar can control several hundred different (semantic) features. If the application program were required to specify values for all of these features for every sentence to be generated, the system would be too complex for most practical purposes. If multiple texts are to be generated for the application, the interface can be made more efficient by factoring it into two components: 1. preparatory knowledge sources: making knowledge of the application domain available to the generator; 2. demands for text: inputs specifying information to be expressed in each particular sentence. By installing various preparatory knowledge sources for each application domain, the size and complexity of inputs that must be dynamically constructed to generate individual sentences can be minimized. 153 Sentence Application Generator Program INTERFACE--i m,. NOTATION TEXT '~IP'Q UTP UT Figure 1: Linking an Application to Per~man. 2 Basic Design of the Interface The principal components £nd sources of information used by Penman are shown in Figure 1. This diagram shows only those components that are necessary to generate one sentence at a time, although the same design has been augmented to generate coherent paragraphs. We describe first the components that are provided by Penman, and comprise its knowledge about the (English) language. Then, we describe those components whose contents depend on the application. 2.1 Resources Provided by Penman Grammar: …

Content truncated to 500 words