In this paper, we describe a new approach to information extraction that neatly integrates top-down hypothesis driven information with bottom-up data driven information. The aim of the kelp project is to combine a variety of natural language processing techniques so that we can extract useful elements of information from a collection of documents and then re-present this information in a manner that is tailored to the needs of a specific user. Our focus here is on how we can build richly structured data objects by extracting information from web pages; as an example, we describe our methods in the context of extracting information from webp ages that describe laptop computers. Our approach, which we call path-merging, involves using relatively simple techniques for identifying what are normally referred to as named entities, then allowing more sophisticated and intelligent techniques to combine these elements of information: effectively, we view the text as providing a collection of jigsaw-piece-like elements of information which then have to be combined to produce a representation of the useful content of the document. A principle goal of this work is the separation of different components of the information extraction task so as to increase portability.