The Knowledge Engineering Review, Vol. 00:0, 1–00. c 2005, Cambridge University Press DOI: 10.1017/S000000000000000 Printed in the United Kingdom Multi-modal reasoning with CBR CINDY MARLING , EDWINA RISSLAND and AGNAR AAMODT School of Electrical Engineering and Computer Science, Ohio University, Athens, Ohio 45701, USA E-mail: marling@ohio.edu Department of Computer Science, Box 34610, University of Massachusetts, Amherst, Amherst, MA 01003-4610, USA E-mail: rissland@cs.umass.edu Department of Information and Computer Science, Norwegian University of Science and Technology,Sem Saelands v. 7-9, N-7491, Trondheim, Norway E-mail: agnar@idi.ntnu.no 1 Introduction Case-based reasoning (CBR) has been successfully integrated with other reasoning modalities and computing paradigms in multi-modal reasoning (MMR) systems. The approaches most frequently integrated with CBR are rule-based reasoning (RBR), model-based reasoning (MBR), constraint sat- isfaction problem (CSP) solving, information retrieval (IR), and planning algorithms. Tasks that seem especially amenable to MMR include interpretation and argumentation, design and synthesis, planning, and the management of long-term medical conditions. Among the benefits of using a mixed mode approach that includes CBR are: more accurate modeling of the knowledge available in a problem domain; compensation for the lack of a complete model in a problem domain; simplification of the knowledge acquisition process; improved solution quality; improved explanatory power; improved run- time efficiency; leveraging of past problem-solving experiences; and the ability to compensate for the shortcomings of one approach by capitalizing on the strengths of another (Aha & Daniels, 1998; Rissland & Skalak, 1991). Early forums for discussing and disseminating work on MMR were the 1998 AAAI Workshop on Case-Based Reasoning Integrations (Aha & Daniels, 1998) and the 1998 AAAI Spring Symposium on Multimodal Reasoning (Freuder, 1998). A comprehensive overview and survey of CBR integrations is available in Marling et al., 2002. In this commentary, we briefly summarize the state of the art in CBR integrations and provide pointers to the literature for the interested reader. 2 CBR Integrations 2.1 Integration with rule-based reasoning The first reasoning modality to be successfully integrated with CBR was rule-based reasoning. The earliest CBR/RBR systems were built in statutory legal domains, where statutes naturally correspond to rules and legal precedents naturally correspond to cases. CABARET used a rule-based agenda mechanism to integrate past cases with legal regulations in the domain of U.S. tax law (Rissland & Skalak, 1991). CABARET pioneered a domain independent architecture in which there are independent CBR and RBR co-reasoners, each of which monitors and communicates its own processing and results, and an agenda- based controller that proposes and prioritizes tasks for the two co-reasoners. Another early legal system, GREBE, integrated CBR and RBR to determine and justify legal conclusions for cases in the area of Texas employment law (Branting, 1991). IKBALS operated in the domains of Australian worker disability law and lending by financial institutions (Zeleznikow et al., 1994). This system also integrated information retrieval techniques to give users access to legal treatises. CBR/RBR hybrids have since proliferated, in diverse domains and applications, ranging from planning nutritional menus (Marling et al., 1999) to harmonizing melodies (Sabater et al., 1998). ANAPRON inte- grated CBR and RBR for speech synthesis in pronouncing American surnames (Golding & Rosenbloom, 1991). This system used CBR to improve upon the accuracy of a primarily RBR system by handling exceptions to pronunciation rules. SaxEx integrates background musical knowledge into a primarily CBR system for generating expressive musical performances (L´ opez de M´ antaras & Arcos, 2002). In SaxEx,