Automatic Geographical Hypertext “Multi-scaled Links” Generation

Abstract

Because text can express what can’t be drawn, and graphic can draw what can’t be written [Jol93], a great number of documents contain both expression modes. Importance in an understanding task of text/picture association has been highlighted in psycho-linguistic studies, especially in understanding spatial concepts and relations [Den89]. In this kind of composite document there are two different usual ways to link text and graphics: explicit links, authors use some specific linguistic expressions like : cf., see fig XX, ...to refer from the text directly to a picture (in this case it’s quite a simple problem to automatically find such links). Implicit links, in this case authors don’t use any specific linguistic expressions they just “talk” about a same subject in the text and in the graphic. Our research has focused on developing a computational model for automatic extraction and “light” interpretation of some appropriate classes of these implicit links. To develop such a theory it’s useful to select a working corpus. In this work geographic information based collection have been selected to be the working corpus [RR94] and the class of implicit links is of course geographical one.