[Prev][Next][Index]

[XTAG_summary]


Summary of Meeting 11-12-98
Questions about Gerunds

There are several different kinds of structures that I'm putting under the
label "gerund". These all involve an underlying V with an -ing suffix acting
nominal at some level. 

All gerunds can appear in places where normal NPs appear (subject, object,
etc). What I'm calling the NP gerunds are the most nominal (take a determiner,
don't assign case to their object (so need preposition), and allow adjectives
to modify). 

NP (ing-of) gerunds
-------------------
1. [the smoking of cigars] 

Possessor (Poss-ing) gerunds
----------------------------
2.	a. John's smoking cigars	
	b. *the smoking cigars (*under relevant reading)

Accusative (Acc-ing) gerunds
----------------------------
3.	a. [John smoking cigars] 
	b. [him smoking cigars]

PRO gerunds
-----------
4. [smoking cigars]


I. First Issue
The NP gerunds are ambiguous in some cases between "gerunds" (that is, forms
that are verbs in the lexicon but become nouns in the syntax) and
nominalizations (that is, forms that become nouns in the lexicon). 

5. John's sighting of UFOs bothered us

There are suffixes that take a verb and make it a noun in the lexicon. Examples
of these are -tion (destroy -> destruction), -ance (annoy -> annoyance), and
*-ing* (sight -> sighting; build -> building). 

This nominalizing -ing is different from the general syntactic gerund formation
-ing. We know this because gerunds are completely productive (any non-stative
verb can be a gerund), but the nominalizer only occurs in certain cases. 
We can tell the difference between the two because the nominalized one
pluralizes whereas the gerund does not.

6. cuttings, buildings, helpings, teasings, killings, sightings...

7. *destroyings, *annoyings...

Many verbs nominalize using -ing, but exactly the ones that take a different
suffix to nominalize (like -ion, or -ance) are the ones that do not allow the
plural. "destroying" and "annoying" are unambiguously gerunds. "cutting",
"building" etc are ambiguous between nominalizations (nouns in the lexicon) and
gerunds (should not be listed as nouns in the lexicon). 

So the proposal was to make our lexical entries reflect this fact. Right now,
we have some -ing forms listed in the lexicon as nouns that should not be --
like "eating". We cannot say "eatings", so "eating" should not be listed as a
noun. 

Morph db: 
	eating
	------
		eating	N 3s
		eating	A
		eat	V PROG

	hitting
	-------
		hit	V PROG


The trick is how to correct our lexical entries without hand-correcting 40,000
entries (or whatever it is). One suggestion was to do a corpus search for
plural -ing forms. Then at least we would know that all of those forms were
nouns. We have actually done this since the meeting. We did a Wall Street
Journal search and a Brown search, extracted all the -ings forms. So we are
going to add these -ing forms to the lexicon as nouns. 

The other suggestion was to find all the verbs that have another form of
nominalizing. These should *not* then have an -ing form in the lexicon. This
however proved more difficult than expected for a number of reasons. So we were
not able to go through with this. 


II. Second Issue
In the NP gerunds, the genitive is a Det, whereas in the Poss-ing gerunds the
genitive is an NP. Why should there be this difference? We decided that this
was ok since in the Poss-ing gerunds the genitive is getting a theta role from
(i.e., is really the subject of) the verb, whereas in the ing-of gerunds this
is not the case. In the Poss-ing case in (8), Mary is the agent of the hitting,
necessarily. But in the ing-of case in (9), Mary is not necessarily the agent,
but could be the organizer of some hitting event. Thus the relationship between
the -ing form and the genitive is the same as with regular nouns.

8. Mary's hitting John

9. Mary's hitting of John

So the intuition is that -ing-of gerunds act more like nouns in that the
genitive is a determiner, whereas the Poss-ing gerunds act more like verbs in
having an NP as a subject.

III. Third Issue

Related to previous issue -- we get over-generation with the genitive NPs in
that they can go where any other NPs can go. In most cases we need this anyway
because of the elliptical genitives.

10. I read mine and Bill read his. 

However, we need different pronouns for the gerund genitive subjects than we do
for these elliptical cases. 

And also we are getting parses for "his smoking of cigars" where "his" is an NP
taking "smoking of cigars" as a relative clause. So this was tangential, but we
decided that we need to fix relative clauses so that they don't modify
pronouns. The problem was that when relative clauses are non-restrictive, then
they can modify pronouns, and in cases like: "he who arrived last". So we need
to investigate when it's actually possible to modify pronouns.