Case-Based Reasoning Abstracts

This dissertation is about adversarial, case-based reasoning and the HYPO program that performs adversarial reasoning with cases and hypotheticals in te legal domain. The dissertation identifies and describes basic case-based operations, an adversarial, case-based reasoning process, a schematic structure for case-based arguments, the kinds of counter-examples that arise and the knowledge sources necessary to support adversarial, case-based reasoning.

Law is an excellent domain for studying case-based reasoning. Expert system designers use case-based reasoning to capture expertise in domains where rules are ill-defined, incomplete, or inconsistent. As an indispensable supplement to reasoning deductively with legal rules, attorneys and judges reason analogically with precedent cases; rule predicates are simply not sufficiently well-defined for them to infer correct decisions deductively. In fact, one "right answer" seldom exists to legal questions. Legal experts make competing arguments instead, pitting conflicting interpretations of cases and facts against each other.

This paper describes CABOT, a case-based system that is able to adjust its retrieval and adaptation metrics, in addition to storing cases. It has been applied to the game of OTHELLO. Experiments show that CABOT saves about half as many cases as similar systems that do not adjust their retrieval and adaptation mechanisms. It also consistently beats these systems. These results suggest that existing case-based systems could save fewer cases without reducing their current levels of performance. They also demonstrate that it is beneficial to distinguish failures due to missing information, faulty retrieval, and faulty adaptation.

We have built a hybrid Case-Based Reasoning (CBR) and Information Retrieval (IR) system that generates a query to the IR system by using information derived from CBR analysis of a problem situation. The query is automatically formed by submitting in text form a set of highly relevant cases, based on a CBR analysis, to a modified version of INQUERY's relevance feedback module. This approach extends the reach of CBR, for retrieval purposes, to much larger corpora and injects knowledge-based techniques into traditional IR.

In this paper we discuss a hybrid approach combining Case-Based Reasoning (CBR) and Information Retrieval (IR) for the retrieval of full-text documents. Our hybrid CBR-IR approach takes as input a standard symbolic representation of a problem case and retrieves texts of relevant cases from a document corpus dramatically larger than the case base available to the CBR system. Our system works by first performing a standard HYPO-style CBR analysis and then using texts associated with certain important classes of cases found in this analysis to "seed" a modified version of INQUERY's relevance feedback mechanism in order to generate a query composed of individual terms or pairs of terms. Our approach provides two benefits: it extends the reach of CBR (for retrieval purposes) to much larger corpora, and it enables the injection of knowledge-based techniques into traditional IR. We describe our CBR-IR approach and report on on-going experiments.

We discuss the use of Case-Based Reasoning (CBR) to drive an Information Retrieval (IR) system. Our hybrid CBR-IR approach takes as input a standard frame-based representation of a problem case, and outputs texts of relevant cases retrieved from a document corpus dramatically larger than the case base available to the CBR system. While the smaller case base is accessible by the usual case-based indexing, and is amenable to knowledge-intensive methods, the larger IR corpus is not. Our approach thus provides two benefits: it extends the reach of CBR (for retrieval purposes) to much larger corpora, and it enables the injection of knowledge-based techniques into traditional IR. Our system works by first performing a standard HYPO-style CBR analysis, and then using texts associated with certain important classes of cases found in this analysis to ``seed'' a modified version of INQUERY's relevance feedback mechanism in order to generate a query. We describe our general approach and report the results of experiments performed in two different legal domains.

In this paper we discuss a hybrid approach combining Case-Based Reasoning (CBR) and Information Retrieval (IR) for the retrieval of legal documents. Our hybrid CBR-IR approach takes as input a standard symbolic representation of a problem case and retrieves texts of relevant cases from a document corpus dramatically larger than the case base available to the CBR system. Our system works by first performing a standard HYPO-style CBR analysis and then using texts associated with certain important classes of cases found in this analysis to "seed" a modified version of INQUERY's relevance feedback mechanism in order to generate a query. Our approach provides two benefits: it extends the reach of CBR (for retrieval purposes) to much larger corpora, and it enables the injection of knowledge-based techniques into traditional IR. We describe our CBR-IR approach and report on on-going experiments performed in two different legal domains.

In this article we evaluate the BankXX program from several perspectives. BankXX is a case-based legal argument program that retrieves cases and other legal knowledge pertinent to a legal argument through a combination of heuristic search and knowledge-based indexing. The program is described in detail in a companion article in (this issue of) the Journal of Artificial Intelligence and Law. Three perspectives are used to evaluate BankXX: (1) classical information retrieval measures of precision and recall applied against a hand-coded baseline; (2) knowledge-representation and case-based reasoning perspectives, where the baseline is provided by the functionality of a well-known case-based argument program, HYPO [Ashley, 1990]; and (3) search perspective, in which the performance of BankXX run with various parameter settings, for instance, resource limits, is compared. In this article we report on an extensive series of experiments performed to evaluate the program. We also describe two brief experiments on ancillary questions regarding the program's search behavior and knowledge representation. Finally we offer some general conclusions that might be drawn from these particular experiments.

The BankXX system models the process of perusing and gathering information for argument as a heuristic best-first search for relevant cases, theories, and other domain-specific information. As BankXX searches its heterogeneous and highly interconnected network of domain knowledge, information is incrementally analyzed and amalgamated into a dozen desirable ingredients for argument (called argument pieces), such as citations to cases, applications of legal theories, and references to prototypical factual scenarios. At the conclusion of the search, BankXX outputs the set of argument pieces filled with harvested material relevant to the input problem situation. This research explores the appropriateness of the search paradigm as a framework for harvesting and mining information needed to make legal arguments. In this first of two articles, we describe how legal research fits the heuristic search framework and detail how this model is used in BankXX. We describe the BankXX program with emphasis on its representation of legal knowledge and legal argument. We describe the heuristic search mechanism and evaluation functions that drive the program. We give an extended example of the processing of BankXX on the facts of an actual legal case in BankXX's application domain-the good faith question of Chapter 13 personal bankruptcy law. We discuss closely related research on legal knowledge representation and retrieval and the use of search for case retrieval or tasks related to argument creation. Finally we review what we believe are the contributions of this research to the understanding of the diverse disciplines it addresses.

In this paper we describe a system, called BankXX, which generates arguments by performing a heuristic best-first search of a highly interconnected network of legal knowledge. The legal knowledge includes cases represented from a variety of points of view: cases as collections of facts, cases as dimensionally-analyzed fact situations, cases as bundles of citations, and cases as prototypical factual scripts, as well as legal theories represented in terms of domain factors. BankXX performs its search for useful information using one of three evaluation functions encoded at different levels of abstraction: the domain level, the argument-piece level, and the overall argument level. Evaluation at the domain level uses easily accessible information about the nodes, such as their type; evaluation at the argument-piece level uses information about generally useful components of case-based argument, such as best cases and supporting legal theories; evaluation at the overall-argument level uses factors, called argument dimensions, which address the overall substance and quality of an argument, such as the centrality of its supporting cases or the success record of its best theory. BankXX is instantiated in the area of personal bankruptcy governed by Chapter 13 of the U.S. Bankruptcy Code, which permits a debtor to be discharged from debts through completion of a court-approved payment plan. In particular, our system addresses the requirement that such Chapter 13 plans be 'proposed in good faith.'

We discuss the indexing of cases for use in precedent-based argument. Our focus is on how multiple, related indices into a case base of legal precedents are exploited by an argument-generation program called BankXX. This system's architecture and control scheme are rooted in a conceptualization of legal argument as heuristic search. Although our framing argument as search is not discussed in detail in this paper, we describe the main features of this view to provide context for a discussion of an indexing scheme that facilitates argument creation. We describe five inter-related index types-citation, prototypical story, factor, family resemblance, and legal theory indices-and show how they can be used to access, view, widen, or filter a set of cases. The application domain is a U.S. Federal statute that governs the approval of bankruptcy plans.

In this project we study the effect of a user's high-level expository goals upon the details of how case-based reasoning (CBR) is performed, and, vice versa, the effect of feedback from CBR on them. Our thesis is that case retrieval should reflect the user's ultimate goals in appealing to cases and that these goals can be affected by the cases actually available in a case base. To examine this thesis, we have designed and built FRANK (Flexible Report and Analysis System), which is a hybrid, blackboard system that integrates case-based, rule based, and components to generate a medical diagnostic report that reflects a user's viewpoint and specifications. FRANK's control module relies on a set of generic hierarchies that provide taxonomies of standard report types and problem-solving strategies in a mixed-paradigm environment. Our second focus in FRANK is on its response to a failure to retrieve an adequate set of supporting cases. We describe FRANK's planning mechanisms that dynamically re-specify the memory probe or the parameters for case retrieval when an inadequate set of cases is retrieved, and give an extended example of how the system responds to retrieval failures.

Rules often contain terms that are ambiguous, poorly defined or not defined at all. In order to interpret and apply rules containing such terms, appeal must be made to their previous constructions, as in the interpretation of legal statutes through relevant legal cases. We describe a system CABARET (CAse-BAsed REasoning Tool) that provides a domain-independent shell that integrates reasoning with rules and reasoning with previous cases in order to apply rules containing ill-defined terms. The integration of these two reasoning paradigms is performed via a collection of control heuristics, which suggest how to interleave case-based methods and rule-based methods to construct an argument to support a particular interpretation. CABARET is currently instantiated with cases and rules from an area of income tax law, the so-called "home office deduction". An example of CABARET's processing of an actual tax case is provided in some detail. The advantages of CABARET's hybrid approach to interpretation stem from the synergy derived from interleaving case-based and rule-based tasks.

We describe how a genetic algorithm can identify prototypical examples from a case base that can be used reliably as reference instances for nearest neighbor classification. A case-based retrieval and classification system called Off Broadway implements this approach. Using the Fisher Iris data set as a case base, we describe an experiment showing that nearest neighbor classification accuracy of over 95% can be achieved with a set of prototypes that constitute less than 5% of the case base.

We discuss several aspects of legal arguments, primarily arguments about the meaning of statutes. First, we discuss how the requirements of argument guide the specification and selection of supporting cases and how an existing case base influences argument formation. Second, we present our evolving taxonomy of patterns of actual legal argument. This taxonomy builds upon our much earlier work on 'argument moves' and also on our more recent analysis of how cases are used to support arguments for the interpretation of legal statutes, which provides the framework for the CABARET system. Third, we show how the theory of argument used by CABARET, a hybrid case-based/rule-based reasoner, can support many of the argument patterns in our taxonomy. Lastly, we discuss how some of these observations and models can be extended to the situation in which a conclusion is sanctioned by a general warrant and not just the application of a rule.

A model of case-based reasoning is presented that relies on a procedural representation for cases. In an implementation of this model, cases are represented as knowledge sources in a blackboard architecture. Case knowledge sources define local neighborhoods of similarity, and are triggered if a problem case falls within a neighborhood. This form of 'local indexing' is a viable alternative where global similarity metrics are unavailable. Other features of this approach include fine-grained scheduling of case retrieval, a uniform representation for cases and other knowledge sources in hybrid systems that incorporate case-based reasoning and other reasoning methods, and a straightforward way to represent the actions generated by cases. This model of case-based reasoning has been implemented in a prototype system ('Broadway') that selects from a case base automobiles that meet a car buyer's requirements most closely and explains its selections.

Precedent-based domains are areas where one appeals to previous cases to support a solution, decision, explanation, or an argument. In such domains, experts typically use care in choosing cases, and apply such criteria as case relevance, prototypicality and importance. In precedent-based domains where both cases and rules are used, experts use an additional selection criterion: the generalizations that a particular group of cases support. Domain experts use their knowledge of cases to forge the rules learned from those cases.

In this paper, we explore inductive learning in a ``mixed paradigm'' reasoning setting, one where both rule-based and case-based reasoning methods are used. In particular, we consider how the techniques of case-based reasoning in an adversarial, precedent-based domain can be used to aid a decision-tree based learning algorithm for (1) training set selection, (2) branching feature choice, (3) induction policy preference, and (4) deliberate exploitation of inductive bias. We focus on how precedent-based argumentation may inform the selection of training examples used to build classification trees. The resulting decision trees may then be re-expressed as rules and incorporated into the mixed paradigm system. We discuss the heuristic control problems involved in incorporating an inductive learner into CABARET, a mixed paradigm reasoner. Finally, we present an empirical study in a legal domain of the classification trees generated by various training sets constructed by a case-based argument module.