A neural network integrated with hypertext for legal document assembly

Hypertext technology is increasingly finding application in law firms, since much of the work of a lawyer involves accessing, assessing and sculpting text. There are some severe limitations inherent in hypertext. In document assembly, users break documents into small units (clauses and subclauses) and link them into a complex web, but links between text units do not cater for all associations of ideas that users may wish to make. Much research aims to provide means of accessing information not restricted by the hypertext structure. Recent methods include free text retrieval (FTR), vector retrieval. FTR is unsuitable for complex legal applications-it is necessary to distinguish between clauses which may share a common vocabulary. Vector retrieval does not allow indirect associations of words, documents. The authors have implemented a novel, neural network-based approach to information retrieval in legal hypertext systems. Practical considerations in the design include handling an often changing collection of text units