The document discusses an evolutionary algorithm approach for evolving natural language grammars without supervision. It aims to extract the set of grammar rules underlying a language's constructions from unlabeled text. The algorithm works by incrementally constructing a grammar by considering sentences one by one and evolving parse trees for each through genetic operations. Experimental results on the Penn Treebank show the approach performs comparably to other unsupervised grammar induction systems, though there is room for improvement in the fitness function. Future work could refine the model and test it on other languages and corpora.