<< Prev | - Up - | Next >> |
Of course, we could express the same linguistic information by using standard trees. In order to do so we need some convention. We can assume for instance that the number feature does always reside at the first term position:
s(pl
np(pl
det(pl the)
n(pl men))
vp(pl
v(pl run))
Now, we could even encode the above rules within a context free grammar. As terminals we chose sg,pl
and for the non-terminals we need s_sg,s_pl,np_sg,np_pl,vp_sg,vp_pl
.
s -> s_pl
s -> s_sg
s_pl -> pl np_pl vp_pl
s_sg -> pl np_sg vp_sg
This kind of an encoding does not look very promising. Now, categories carry additional information which is structured in itself. Therefore, it is inconveniant to encode categories as atoms; it is much nicer to encode all information of a category by a feature tree.
<< Prev | - Up - | Next >> |