Generating text in context

While the World Wide Web sets the scene for the future of information sciences, a great opportunity and challenge is created for researchers in natural language generation. The present WWW paradigm in which the user is looking for information on the net, trying to interpret it, to determine what is important and useful becomes problematic as the amount of information explodes. A new paradigm needs to be worked out, in which information entities proactively try to capture the attention of the user, in which the information adapts to the user and her particular extra-linguistic context. We are investigating in how far the current state of the art in natural language generation can contribute to that paradigm and how further developments can be targeted towards the generation of context-sensitive meta-text for information sources. This paper presents a research programme including work on different fronts for capturing the user’s context, eliciting appropriate linguistic expressions and enabling on the fly generation of relevant meta-text adapted to the ‘hic-et-nunc’ situation of the user. 1. Objective The age of the information society has been initiated by the WWW enterprise: an unlimited network connecting everyone, everywhere, everything. The result today is an ever-growing network of on-line information available to an ever-growing number of users at the most diverse locations of the world. However, putting information on-line is not the most revolutionary fact, indeed it immediately creates also the huge problem of a mass of users facing a mass of on-line documents. The real challenge is to exploit the computational potential of the net and to make services available rather than (static) information. In this spirit a new challenge for natural language generation (NLG) research appears. The combination of computational power, extensive networks and NLG technology can drastically change our conception of texts. In this new technological context, information sources need not be the same for everyone everywhere at anytime. Several researchers in natural language generation have started early investigating the perspective of adaptive text, argumenting that in order to improve communication, different texts should be created for different readers [Moore and Swartout 1991], [Paris, 1987], [Sparck Jones, 1991]. We aim to work further on this direction within the new technological context just described. The research challenge at the basis of our project consist in investigating in how far the natural language generation process can be determined by the extra-linguistic context. Whereas current and past research projects often focus on adaptation to the most stable characteristic of the context the user’s profile-, we aim at taking into account the more dynamic facets related to the extra-linguistic situation and even to the minute-to-minute interests of the user. We propose to approach the problem from related disciplines, each one focusing on a number of generally acknowledged subtasks of NLG. How can the context of the user be sensed on a minute-to-minute basis: content determination and discourse planning is viewed as an AIproblem. What are the linguistic means that humans use to relate their utterances to the context of their hearer/reader ? A linguistic investigation contributes to the lexicalisation process. What is a good computational framework to perform the text generation task in real time? Realisation, referring expressions generation and sentence aggregation are to be considered from the computational linguistics point of view.