systems that generate natural language output as part of their interaction with a user have become a major area of research and development. typically, natural language generation is divided into several phases, namely text planning (determining output content and structure), sentence planning (determining abstract target language resources to express content, such as lexical items and syntactic constructions), and realization (producing the final text string) (reiter, 1994). while text and sentence planning may sometimes be combined, a realizer is almost always included as a distinct module. it is in the realizer that knowledge about the target language resides (syntax, morphology, idiosyncratic properties of lexical items). realization is fairly well understood both from a linguistic and from a computational point of view, and therefore most projects that use text generation do not include the realizer in the scope of their research. instead, such projects use an off-the-shelf realizer, among which penman (bateman, 1996) and surge/fuf (elhadad and robin, 1996) are probably the most popular. in this technical note and demo we present a new off-theshelf realizer, realpro. realpro is derived from previous systems (iordanskaja et al., 1988; iordanslcaja et al., 1992; rambow and korelsky, 1992), but represents a new design and a completely new implementation. realpro has the following characteristics, which we believe are unique in this combination: we reserve a more detailed comparison with penman and fuf, as well as with alethgen/gl (coch, 1996) (which is perhaps the system most similar to realpro, since they are based on the same linguistic theory and are both implemented with speed in mind), for a more extensive paper. this technical note presents realpro, concentrating on its structure, its coverage, its interfaces, and its performance.systems that generate natural language output as part of their interaction with a user have become a major area of research and development. this technical note presents realpro, concentrating on its structure, its coverage, its interfaces, and its performance. we are grateful to r. kittredge, t. korelsky, d. mccullough, a. nasr, e. reiter, and m. white as well as to three anonymous reviewers for helpful comments about earlier drafts of this technical note and/or about realpro. the input to realpro is a syntactic dependency structure. the development of realpro was partially supported by usaf rome laboratory under contracts f3060293-c-0015, f30602-94-c-0124, and f30602-92-c-0163, and by darpa under contracts f30602-95-2-0005 and f30602-96-c-0220. this means that realpro gives the developer control over the output, while taking care of the linguistic details. realpro is licensed free of charge to qualified academic institutions, and is licensed for a fee to commercial sites. the system is fully operational, runs on pc as well as on unix work stations, and is currently used in an application we have developed (lavoie et al., 1997) as well as in several on-going projects (weather report generation, machine translation, project report generation). the architecture of realpro is based on meaningtext theory, which posits a sequence of correspondences between different levels of representation. |