Filler-gap mismatches

[Robert Levine]

I have never found the treatment of clausal subjects as structurally
nonsubjects especially plausible; the point of the passive example
vis-a-vis the topicalization case is just that in both instances,
there is a problem relating a clausal constituent to a valence element
to which it should correspond, but which in each case it cannot be
literally realized as. For example, in the case of passives, the
passive lexical rule will normally map a COMPS element of a finite
verbal head to SUBJ element of a passive; but here the element showing
up as subject cannot actually show up as a complement of the finite
form. By the same token, on a BMS version of extraction, one of the
possible forms of a verb which may select a certain complement has
that complement missing from its COMPS list, but a corresponding
SLASHed element on its ARG-ST list which then shows up on the verb's
SLASH list. In both cases---very roughtly speaking---there is an
apparent relationship between lexical entries, but one in which the
combinatorial possibilities of one of the entries does not bear the
correspondence to the other that the relationship (finite/passive,
valence n/valence n-1) normally involves.

The brute force way to do this this would just be to say that the
passive form here simply does not correspond to the yield of the
passive lexical rule. We know of other such cases (`Robin was
said/rumored to have been a spy for the Ostrogoths'/*We said/rumoured
Robin to have been a spy for the Ostrogoths') so one more may not be
that big a deal. By the same token, one could imagine that there is a
lexical entry of the form V[SLASH Z, ARG-ST <..., SYNSEM|LOC Z, ...>
where there just doesn't happen to be a corresponding form V[COMPS
<...,Z...>]. We have, instead, an idiosyncratic lexical entry. Such
cases seem to me to be predicted by a hypothesis in which V[SLASH {}]
and V[SLASH {X}] are different lexical entries. In other words, if
heads licensing gaps (with gaps as missing rather than empty
categories) are separate lexical entries from those in which a valent
actually appears, then whatever you do with the passive cases seems
applicable to the extraction cases, no? (I don't think you necessarily
have to accept the valence reduction hypothesis for extraction to get
such results, NB).

Bob


To: hpsg-l at lists.Stanford.EDU, mike_maxwell at sil.org
Subject: Re: Filler-gap mismatches