Automated knowledge acquisition for instructional text generation

Language generation systems have often been advocated for use in the generation of user documentation, but in practice, it is hard to use them because of the amount of knowledge they require as input. Without a readily available source of input, the knowledge acquisition bottleneck will prevent generation technology from ever being used. Unfortunately, knowledge acquisition for any domain-varying language generation application is difficult because there is seldom any single knowledge source or acquisition method that is adequate for acquiring the appropriate knowledge. This paper argues that this required knowledge can, in practice, be acquired from a heterogeneous set of complementary sources. It then presents Isolde, a support environment for technical writers that can mine these sources using an extensible set of specially tailored acquisition tools. These tools acquire instructional knowledge on-the-fly and consolidate it into a knowledge base that can drive instructional text generation.