ABSTRACT
Symbolic and connectionist approaches are obviously complementary for problem solving, leading to a growing interest in hybrid systems: those involving the co-operation of both approaches in a single integrated system. However, some general problems may be observed in current literature on the subject. These problems, usual in new research arenas, may render the achievement of advances in the field difficult:
First, most of the research effort until now has addressed specific applications, generating specific hybrid solutions. Such results rarely evolve toward general models of hybridization.
Second, applications are usually developed in a purely ad hoc manner. Portability, re-usability, scalability and, finally, applicability to real world problems are seriously hampered.
In this situation, comparing the plausibility, generality and performance of different approaches, models or implementations (in general, any kind of experimental research) becomes an impossible task.
