Evolutionary Parsing for a Probabilistic Context Free Grammar

Abstract

Classic parsing methods are based on complete search techniques to find the different interpretations of a sentence. However, the size of the search space increases exponentially with the length of the sentence or text to be parsed, so that exhaustive search methods can fail to reach a solution in a reasonable time. Nevertheless, large problems can be solved approximately by some kind of stochastic techniques, which do not guarantee the optimum value, but allow adjusting the probability of error by increasing the number of points explored. Genetic Algorithms are among such techniques. This paper describes a probabilistic natural language parser based on a genetic algorithm. The algorithm works with a population of possible parsings for a given sentence and grammar, which represent the chromosomes. The algorithm produces successive generations of individuals, computing their “fitness” at each step and selecting the best of them when the termination condition is reached. The paper deals with the main issues arising in the algorithm: chromosome representation and evaluation, selection and replacement strategies, and design of genetic operators for crossover and mutation. The model has been implemented, and the results obtained for a number of sentences are presented.