This article is in sharp contrast to the Furnas et al. article in that it focuses on a particular system, for indexing or analyzing patent documents. By picking a limited domain, one can benefit from the characteristics of the special sublanguage involved, such as stereotypical reference forms that allow resolution of the sticky problem of anaphora.
Also as a result of using a sublanguage, a semantic approach to parsing can be adopted, and a relatively small set of relations can be used to indicate how the elements of noun phrases fit together. There is enough detail, however, to illustrate how parsing and natural language processing might be applied to other similar document collections.
The representation for memory is like a semantic network, but specialized into a hierarchy for fast lookup / classification. Problems of disambiguation are at the heart of the discussion. An open question is how well this approach would scale up to larger domains, and to larger numbers of documents.