Most of the literature on information gathering deals with locating, gathering and selecting the best response to a query from among a multitude of responses from different data repositories[Oates, Nagendra Prasad, &Lesser1994][Bowman et al.1994][Arens et al.1993]. Nagendra Prasad, Lesser and Lander[Nagendra Prasad, Lesser, &Lander1996][Nagendra Prasad, Lesser, &Lander1995] introduced a different model of response to a query where no single source of information may contain the complete response to a query; necessitating piecing together mutually related partial responses from disparate and possibly heterogeneous sources. A complex query is presented to a set of agents, each of which is responsible for retrieving information relevant to a part of the query. The agents negotiate to piece together a mutually acceptable response to the query. This type of retrieval, defined as Negotiated Retrieval[Nagendra Prasad, Lesser, &Lander1996][Nagendra Prasad, Lesser, &Lander1995], adopts the above view of a query to a set of distributed case bases in the corporate memory context. More specifically, a response to a query involves assembling related pieces of information from different case bases to form a composite case. The agents have to cooperatively retrieve mutually acceptable responses while negotiating compromises to resolve conflicts. Each agent retrieves subcases from its local case base and all agents together assemble a mutually acceptable overall case from these subcases to produce a response to the user's information needs.
Information requirements of many real life applications rely on such distributed case bases. Let us illustrate this with an example: a management consultancy firm is faced with the need to quickly build a cross-functional team by drawing from an organization-wide talent pool for the purpose of helping an inventor research the market for a product, analyze pros and cons of the competition, and locate interested venture capitalists. We can imagine an automated assistant for querying a multi-agent system that assembles the team by letting each agent access its own resume database of various experts for a particular aspect of the project: technical, management, sales, etc. In assembling the team, the agents need to consider interactions between the requirements of experts for different aspects like for example, all experts willing to work on that type of project or all technical experts on the team being familiar with a particular computing environment. The distributed case bases in this example are the resume databases for different expertise, with descriptor generators that extract features like ``project_types_willing_to_workon'' and ``familiar_computing_environments''.