Avoiding syntactically incorrect individuals via parameterized operators applied on derivation trees

Evolutionary algorithms can be efficiently used to solve many different problems, without knowing much about the nature of the solution. One of the most appreciated property of these algorithms is simplicity. However, this simplicity causes an unguided nature; superfluous or even invalid individuals may be produced taking valuable time from the algorithm. In this paper a preliminary method is described for avoiding such individuals during the evolution of strings from a context-free language. Our method operates on the derivation trees of the underlying context-free grammar and keeps nevertheless the simplicity and randomness of the evolutionary algorithms. Though the system is designed for context-free languages, the method can be extended to higher level languages, too.