words are inserted into syntactic trees under lexical categories such as N, V, and Det. Hence the grammar has to
generate a VP dominating a Vand a definite NP before insertingkick, the, andbucketindividually under V, Det, and N
respectively; then the semantic component has to interpret the concatenation of these three words idiomatically. This
idio matic interpretation has to co me directly fro mthe lexicon, not fro mthe usual co mbinatorial interpretive process;
but in the insertion of the words one by one, their unity as an idio mhas been lost. So how can the lexicon record the
idio mas a unit, and how can this unit be inserted into larger co mbinatorial structures?
One solution (Chomsky 1981: 146, n. 94) treatskick the bucketas a lexical verb rather than a verb phrase, giving it the
internal structure shown in (10).
Because (10) is dominated by a V node, it can be inserted as a whole into a V position in syntax. It thus can carry its
idiomatic information through to the level of semantic interpretation.
Such a solution has a number of fatal difficulties. First, as pointed out as early as Emonds (1970), there are many
discontinuous idioms such as those in (11), in which a direct object intervenes between two constituents of the idiom.
(Proe designates a pronoun that must be coreferential with the subject;proidesignates a pronoun that must be
coreferential with the NP subscriptedi.)
(11) take NP to task
take NP for granted
show NP the door
give NP the boot
give NP a piece ofproe'smind
put NPiin proi's placeA structure like (10), in which the entire idio mis do minated by V, predicts that these idio ms should instead have the
for m*[vtake to task] NP, with the direct object to the right of the entire idiom. One can save the analysis by stipulating
thatto taskmoves around the direct object in the course of the syntactic derivation.