Antecedent-Contained Deletion

George Lakoff lakoff at COGSCI.BERKELEY.EDU
Sun Jun 17 19:48:04 UTC 2001

This was one of the old arguments for generative semantics, that is,
for syntactic distribution to be based on semantic considerations.
The kinds of examples I used back in 1966 to argue the case were
distinctions like:

(1) [The man who deserved [it]j ]i got [the prize [he]i wanted]j.

(2) *That Harry [believed that Bill had []j ]i indicated that Sam
would [claim that Max had []i ]j.

(1) is the old Bill Woods sentence (written about by Bach and Peters).

(2) shows that predicates don't work the same way as arguments, for
semantic reasons.

The fact so-called VP-deletion is actually the omission of an
identical predication also explains the ill-formedness of (5) in
Dan's letter.

 From a theoretical perspective, it would not be surprising if
functional considerations fit semantic considerations. After all,
semantically ill-formed sentences tend not to occur and therefore
would be outside of positive functional principles.

It should be said that such considerations also hold in all
contemporary theories where semantic considerations determine
syntactic distributions: e.g.,cognitive grammar and embodied
construction grammar (aka neural grammar).

It's nice to see argument forms from 35 years ago surfacing again.


>I am wondering if anyone reading this list knows of or has worked on
>functional approaches to Antecedent Contained Deletion. Consider the
>following (inspiration for this posting comes from David Pesetsky's new
>book, Phrasal Movement and Its Kin, MIT Press):
>In thinking of functionalist accounts of ACD, some of the simple cases seem
>to work out. So, consider
>crucial pairs like the following:
>(1)  Mary suspected everyone that I did.
>(2) *Mary suspected that I did. (under verbal ellipsis reading, i.e. where
>  'did' is 'suspected')
>(1) is supposed to be good in a Minimalism account because the entire
>quantified d.obj.
>  'everyone that I did' raises at Logical Form and then the quantifier raises
>  i.e. 'everyone' to the far left of the phrase, the CP position. And this
>  movement is supposed to eliminate the infinite regress difficulty. Since
>  there is no quantifier in (2), such LF movement is
>impossible, hence we are led to an infinite regress and the sentence is out
>  people don't usually mention that an infinite regress ought only to be a
>  performance problem.)
>If I were to take a simple-minded approach to this, I would say that, in
>RRG terms, you cannot delete/omit verbal material (or any other) from a
>'core' argument position (the d.obj. especially) because new information
>is presented here. Notice that the grammatical 'quantificational
>structure' doesn't involve deletion of a core argument but, rather, a
>position of exactly the kind that old information often turns up in (and
>old information is often realized as a clitic or zero). The other condition
>on ACD is
>that the antecedent VP is supposed to c-command the elided VP. So no
>passives: *That everyone did was suspected by John. But c-command, in my
>experience, can be paraphrased in RRG-functional terms as CORE argument or
>argument of CORE argument. In any case, the sentence just given is bad
>presumably because subjects never take objects as antecedents. C-command
>is unnecessary. Now consider
>(3) Everyone suspected. That everyone did was suspected by John.
>(4) John met everyone that Mary did.
>(5) *Everyone that Mary did was met by John.
>(6) ??Mary met people. Everyone that Mary did was met by John (too).
>  (perhaps special stress is needed on 'did' for this to work.)
>So, it doesn't look utterly implausible to suggest an
>information-structure approach to ACD, rather than a structural
>In fact, an information-structure approach is rendered even more
>plausible by the well-known, but often ignored, fact that intonation affects
>cases of deletion and displaced constituents. Is anyone on this list aware
>of any
>attempts to reanalyze ACD facts in terms of information structure?
>Dan Everett

More information about the Funknet mailing list