Generating natural language descriptions with examples: differences between introductory and advanced texts

Examples form an integral and very important part of many descriptions, especially in contexts such as tutoring and documentation generation. The ability to tailor a description for a particular situation is particularly important when different situations can result in widely varying descriptions. This paper considers the generation of descriptions with examples for two different situations: introductory texts and advanced, reference manual style texts. Previous studies have focused on any the examples or the language component of the explanation in isolation. However, there is a strong interaction between the examples and the accompanying description and it is therefore important to study how both these components are affected by changes in the situation. In this paper, we characterize examples in the context of their description along three orthogonal axes: the information content, the knowledge type of the example and the text-type in which the explanation is being generated. While variations along either of the three axes can result in different descriptions, this paper addresses variation along the text-type axis. We illustrate our discussion with a description of a list from our domain of LISP documentation, and present a trace of the system as it generates these descriptions.