A statistical procedure for evaluating similarities among very divergent languages