It's true that most natural languages cannot be parsed with a CFG, but he gives the wrong example. In fact, the language of the grammar that he describes can pe

It's true that most natural languages cannot be parsed with a CFG, but he gives the wrong example. In fact, the language of the grammar that he describes can perfectly be generated by a CFG. E.g. for a rule such as: sentence --> noun_phrase(Num), verb_phrase(Num). You could create the CFG equivalent: sentence --> noun_phrase_sg, verb_phrase_sg. sentence --> noun_phrase_pl, verb_phrase_pl. It sure is a drag, but this is how many CFG grammars/parsers work. Of course, it's generally agreed upon that DCGs generalize better even when used with grammars that can be described with a CFG. Most natural languages are not context-free, but he gives the wrong example. To provide a more convincing example - in Dutch you can construct phrases like: dat ik Marie Jan de dozen zie helpen inpakken (lit: that I Marie Jan the boxes see helping packing) The NPs and the verbs correspond in the following manner: ik - zie, Marie - helpen, Jan - Inpakken. Or, more formally, the structure is: np_a np_b np_c np_x v_a v_b v_c. Of course, this highly resembles a COPY language, which cannot be expressed using a CFG, but (obviously!) can be expressed using a DCG: s(L) --> w(L), w(L) w([a|Rst]) --> [a], w(Rst). w([b|Rst]) --> [b], w(Rst). w([]) --> [] (This grammar parses/generates a string of zero or more a's and b's, followed by exactly the same string.)