those adjectives
LFG List
dalrympl at parc.xerox.com
Tue Sep 2 17:32:32 UTC 1997
Hi,
I've been designated by ParGram to compose an answer to the vexed
question of adjectives that Veronika Knueppel posted about recently.
As part of a parallel grammar (ParGram) development project, (see
http://www.parc.xerox.com/istl/groups/nltt/pargram/ for more info) a
group of us has been working on a French, German and English grammar
for the last year or two. In writing the grammars, the question
of how to treat adjectives arose as well, of course.
And I guess I've been given the job to briefly summarize our "internal"
debates here in this forum because I turned out to be one of the ones
with the loudest opinions.
One feature of the ParGram project is to try to use as many
"universal", i.e., common analyses for each of the languages as
possible. As part of that effort, we tried to converge on a treatment
of adjectives.
One of the first issues we tackled was whether adjectives should
subcategorize for subjects or not.
(NOTE: that adjectives sometimes need to subcategorize for things
like OBJ or OBL is fairly clear from "proud of" or similar examples.
The question addressed here is not whether adjectives *sometimes*
need subjects, but whether they should always be assumed to
subcategorize for one.)
The standard/traditional approach had been to have adjectives subcategorize
for subjects:
`red<SUBJ>' etc.
We, the gang of fresh-faced young grammar writers, thought briefly
(about 10 minutes) and decided that there was no real reason to really
have this analysis any longer, given the larger reliance on argument
structure concepts in recent years. So, we nixed subjects in
adjectives and lived happily with that for a long time.
With time, problems arose from several sources, some of the
ideological, some of them practical. One of the practical ones was,
just to take an example, that the German grammar began to have
problems when coordinating "simple" adjectives and deverbal adjectives
as in
Die gelbe und rennende Katze -- `The yellow and running cat'
This was because the `yellow' had no subject, but the `running'
did from its verbal subcat list, and there was a clash in the
way the coordination is implemented.
I fixed this by simply suppressing the verbal subject via a
lexical rule in the process of the deverbalization.
I still think this is a reasonable thing to do, but it didn't
seem to make very many other people happy.
And so we did spend a rather long time discussing the issue, and it
doesn't seem to be resolved as yet.
Here are the pros and cons we came up with:
A) Why have subjects in adjectives at all???
1) Agreement comes "for free".
I think this argument came mainly out of languages like French,
where you get agreement between adjectives and nouns, as follows:
Predicative Use Attributive Use
--------------- ------------------
Ils sont rouges `they are red' Les chiens rouges `the red dogs'
Il est rouge `he is red' Le chien rouge `the red dog'
(I hope this French is right....)
That is, you get the adj-N agreement in both attributive
and predicative uses.
So, then the idea is that if you allow adjectives to have
subjects universally, you get to capture both of these
patterns with the same mechanism, and to boot a mechanism
that already exists in the grammar: subject-verb agreement
extended to a more universal subject-predicate agreement.
Time saving, economical, neat.
However, other languages don't in fact confirm this pattern
at all. German, for example, would instead appear to speak
very strongly against such a treatment.
German has quite a complex adj-N agreement system, including
gender, number, and a distinction between "weak" and "strong"
declensions (one of the most ridiculous things I've seen a
language do, I must say).
Predicative Use Attributive Use
---------------- ----------------
Sie sind rot `they are red' Die roten Hunde `the red dogs'
Er ist rot. `he is red' Der rote Hund `the red dog'
German, in fact, shows adj-N agreement only in the attributive
use. In the predicative (and appositive) use, an invariant
form of the adjective is used.
Thus, German makes a very strong argument *against* conflating
the predicative and attributive uses of adjectives.
2) Anaphora Data
There are a series of data that would appear to indicate that
adjectives, PPs and NPs must subcategorize for subjects.
English:
He is proud of himself. (AP)
She was beside herself with joy. (PP)
He is a force unto himself (NP)
John_j considers Bill_i too proud of himself_i,*j.
German:
Der sich rasierende Professor
he self shaving professor
`The professor shaving himself.'
Norwegian:
Vi gjorde Jon_i glad i seg self_i
we made John fond of him self
`We made John fond of himself.'
*Jon_i gjorde meg glad i seg self_i
John made me glad of him self
`John made me fond of himself.'
That is, subject oriented reflexives appear to be sensitive
to the subcat frame of the adjective. This is particularly
evident in the Norwegian contrast. So, the argument goes, the
subcat frame of the adjective must contain a subject.
However, I think it is also possible to construct examples
in which there is no subject (impersonal constructions) and where
the reflexive can still find an antecedent. Consider the German:
Hier wird sich nur gemalt, nicht fotografiert.
here is self only painted, not photographed
`Here one only paints, not photographs oneself.'
Hier wird gelernt, auf sich stolz zu sein.
here is learned, on self proud to be
`Here it is learned to be proud of oneself.'
These sentences are examples of what the Germans call
"subjectless" constructions: impersonals.
Here, the reflexive is good, even when there is no overt
subject of the sentence.
If one admits such sentences into the debate, then they
serve to cast doubt over the assumption whether one should
indeed resolve anaphora only via a recourse to grammatical
functions, or whether one should do anaphora resolution
at the level of argument structure (in addition, or perhaps
even instead).
(In fact, argument structure effects on anaphora resolution
have been argued for).
3) Copular Constructions: Who does the Predicating?
Another strong argument in favor of the adjective-have-subjects
view has been a consideration of predicative constructions.
In many languages, a copular form of the verb "be" appears
in these constructions.
The dog is red.
However, there are other languages which have a "null"-copula.
Like Bengali:
oTa lal
that red
`that is red'
In addition, languages with agreement also display more
complex constructions. There is the French example that
Veronika cited, where the adjective agrees with the object.
Le chevalier trouve la femme belle.
`The chevalier found the woman pretty.'
And the English-type "consider"-constructions.
So, the question with all of these predicative constructions
is: where does the predication come from??
In languages like English, which have an overt copula,
a likely analysis might be that the "be" subcategorizes
for a subject and a thing that is predicated over (ACOMP,
PCOMP, VCOMP in old LFG).
But then --- what does one do with languages like Bengali??
So, an alternative analysis that would cover both types
would posit simply that the adjective (or PP or NP) in
predicative constructions in fact does it's own predicating ---
it is responsible for providing all the arguments via it's
own subcat frame.
This would mean for English that the "be" is just a tense
carrier for the clause, but in fact doesn't have to do any
of the predicative work.
And then in the "consider"-type constructions
She considered him idiotic.
you could apply the same analysis, which is basically a small clause
analysis.
This is a fairly reasonable way to proceed, but I personally
would rather adopt an approach in which the overt copulas do
get to have predicative force. In those constructions without
an overt copula, I would want to adopt something akin to
a constructional grammar approach --- we know it has to be
a copular construction by it's structure, so it gets a
subcat frame supplied *by virtue of the construction*
(in practical terms through an empty category).
B) Advantages of not having adjectives subcategorize for subjects
1) F-structures are less cluttered.
This is not only an aesthetic argument (your f-strs become
much easier to read without all those extraneous subjs floating
around --- try it!), but also an efficiency argument from
the point of view of grammar writing: if you have that many
less subjects to keep track of and control into and agree with,
and such, you have that many less rules to write in your grammar.
2) If you decide to believe the above arguments from anaphora
and copula constructions (arguments A2 and A3), then you also
commit yourself to having all nouns (and PPs) have subjects.
In practical terms this means that you must have a disjunction
in all your noun entries:
Hans<Subj>
and Hans
That is because you will not always want your noun to come
with a subject ---- you actually only want it in predicative
contexts, but not in: Hans sees a dog.
3) Having overt subjects in APs, PPs and NPs tends to lead you
to the question of what the semantics of things like
Hans saw the Hamster in the cage
are. That is, "in the cage" now has a SUBJ. Is this SUBJ
controlled by Hamster? Or is it controlled by Hans?? Who
is in the cage?
That is, you are now doing some of the PP-attachment ambiguities in terms
of subject control, which (to me at least), seems a bit odd.
On the other hand, those who believe that s(emantic)-structure
should feed entirely off of f-structure information are much
in favor of this approach, as the semantics can then be
constructed by taking the SUBJ of adjectives into account
(but, in my view, even the adj<SUBJ> can't provide all of the
necessary semantic and pragmatic information needed to do
a good semantic analysis, so why not simply do it differently
in the first place).
C) Summary
Okay, so that's about the extent of the central pros and cons as
we could determine them. A number of other concerns did enter
the debate, such as how to do the French agreement facts if
adjectives did not have subjects --- however, these were concerns
that could readily be provided with an alternative solution
in terms of grammar implementation and so were not deemed
"central".
The debate as it stands within the Pargram project has so
far arrived at the following conclusions:
1) Some things are more practical to implement when adjectives
do have subjects.
2) Some things are more practical to implement when adjectives
don't have subjects
(Here the pros and cons are about 50-50, so practical
implementation issues don't seem to be able to play a decisive role)
3) Much of the analyses that has been attributed to the
presence of a SUBJ in the adjective subcat frame could be
reformulated (in light of the ever growing work on
argument structure) in terms of argument structure properties.
That is --- the predicative power of the adjective could
be expressed at argument structure, rather than positing
the realization of grammatical functions (and then worrying
about how to control them). However, some people (reasonably)
feel a bit iffy about the nature and properties of argument structures
and would thus prefer to stick to the "known" quantity
of f-structure.
4) It seems to boil down to a matter of belief of what
grammatical functions are for, versus argument structures
(if you even admit those in your grammar).
D) Finally --- a direct answer to Veronika's question.
Having said all that, I'll here try to answer how Veronika
could solve her problem.
To repeat and summarize their approach:
Sample Adjective
/belle/ V, (^PRED) = 'Beau <(^SUBJ)>'
(^SUBJ NUM) =SG
(^SUBJ GEN) =FEM
Copular "be"
/est/ V, (^PRED) = 'Etre<(^SUBJ) (^ACOMP)>'
(^ACOMP SUBJ) = (^SUBJ)
...
No problems with the copular.
>The problems arise when we try to treat adjectives in attributive use in
>the same way. Here, I cannot see any possibility of indicating where the
>adjective's SUBJect can be found. The adjective in this case is not
>governed, so there is no way to have anything like functional control.
1) with adjectives subcategorizing for SUBJ
This actually didn't prove to be a problem in our grammars. You
simply decide that the head noun will always be your SUBJ. And
then you write a bunch of rules within an NP that identify the
head noun as the SUBJ of the ADJ. Since the head noun is always
in a position where it can be uniquely identified, this can be done
quite straightforwardly.
It actually does require some intricate bookkeeping and lots of
disjunctive rule-writing in that that you end up continually passing
around the subject, but it works. So, as far as we could tell, it was
a pain, but it could be done.
None of the ParGram grammars has adjectives with subjects at the moment
(and actually hasn't had any for quite some time) so I can't post any
sample rules, unfortunately --- but we did have a working grammar
with the SUBJ-plus approach about 1 1/2 years ago.
2) without adjectives subcategorizing for SUBJ
>whereas the other
>has no subcategorization frame, being the one that is attributively used.
>The agreement of noun and adjective would then simply be treated by means
>of unification
Yes, this would then be the alternative method. And, in fact, as
I've said above, one needn't even assume the SUBJ for the predicative
adjectives in French --- one could do without SUBJ in the subcat
frames for adjectives altogether.
I believe some tricky problems do crop up with adjective agreement,
especially with regard to French, which make a subject-less treatment
seem unappealing, but I believe they can be done even so
(Anette Frank at Grenoble would be able to provide more information
here).
Miriam
More information about the LFG
mailing list