Miriam Butt and Tracy Holloway King (Editors)
CSLI Publications, 1996
Not all aspects of f-structure are equally language-independent, however. Although there was originally no stipulation that every f-structure must contain a PRED attribute, the extended coherence condition as stated in Bresnan and Mchombo 1987 ensures that every f-structure containing at least one syntactic function will also contain a PRED.
| (1) | |
In (2) the topic is related semantically to an overt constituent in
the comment. In this example, there is a part/whole relationship between
the door and the office.
| (2) |  |
In (3) the topic is not related to any specific constituent in the comment.
| (3) |  |
In each case, however, the topic specifies the semantic domain that the comment must be interpreted in relation to. I have argued that a uniform analysis for all these constructions may be achieved by using both a TOPIC and a COMMENT function in f-structure (Rosén 1988). The problem that arises with such an analysis is that this involves an f-structure nucleus with no PRED.
Several authors have proposed distinguishing between internal and external topics, see for example Aissen (1992) and King (1995). Internal topics are bound in the sense of Bresnan and Mchombo mentioned above, while external topics are not. For Vietnamese, this kind of analysis would mean that the topic in (1) would be an internal topic, whereas the topics in (2) and (3) would be external topics. Although such an analysis is certainly possible, it does not reflect the syntactic structure of the topic-comment construction.
| (4) |  |
The first NP in this sentence is a predicative complement, and the second is the subject. Semantically, the predicative complement predicates a property of the referent of the subject phrase, in this case the property of being a stallion.
According to Otto Jespersen, this type of sentence is "extremely frequent" in many languages. He mentions both Greek and Russian as examples of Indo-European languages "which have a copula, but do not use it as extensively as e.g. English." (Jespersen 1924:121) In Russian, for instance, the copula is not used in this type of sentences in the present tense, but it must be used in other tenses. In Maori, however, there is no copula, and thus the relation between the two nominal phrases is never expressed by a verb, but only by the grammatical construction as a whole.1
| (5) |  |
This f-structure is incoherent, since it contains two governable functions that are not governed by a local predicate. But since the sentence in (4) does not have a verb, there is nowhere to get a PRED with an argument list from.
In The Mental Representation of Grammatical Relations, predicative
complements are not extensively discussed, but several authors do treat
them. Jane Grimshaw suggests that the French verb
devenir 'become'
has the lexical form shown in (6) (Grimshaw 1982:137).
| (6) |  |
This lexical form would permit the f-structure in (8)2
for the sentence in (7).
| (7) |  |
| (8) |  |
This f-structure looks just like the f-structure for a transitive verb, except that there is an NCOMP rather than an OBJ.
Other authors have analyzed predicative complements in a way that departs more radically from the treatment of objects. In his discussion of the analysis of predicate adjectives in Icelandic, Avery Andrews says the following:
There are three ways in which this could be achieved. First, the adjective could be the functional head, with the copula serving merely as a mood- and tense-bearing element. Second, the copula could be a head taking the adjective as a complement under functional control. Third, the copula could be a head taking the adjective as a complement under anaphoric control. (Andrews 1982:446)
| (9) |  |
Andrews rules out his first suggestion on metatheoretical grounds, and
rejects the third alternative because he says such a structure would require
that the copula be a predicate relating an NP and a proposition, and that
there is no reason to suppose that the copula takes the SUBJ
as a predicate argument, since it does not itself predicate anything of
it. He opts therefore for the second possibility, which results in the
f-structure in (10) (Andrews 1982:447).
| (10) |  |
The control equation (^ ACOMP SUBJ) = (^ SUBJ) in the lexical entry for the copula ensures that the SUBJ of the ACOMP will be functionally controlled by the SUBJ of the copula.
As we can see, Grimshaw and Andrews agree that the predicative complement is semantically a predicate, but this has different consequences for their analyses. Grimshaw does not suggest that this semantic property of predicative complements should have an effect on how the f-structure looks. For Andrews, however, it is important that ACOMP and NCOMP are so-called open functions. The open functions are those which must receive a SUBJ through functional control.
When there is no verb that links the subject and the predicative complement,
it becomes more apparent that using an open function does not solve the
problem of having a nominal phrase that functions as a semantic predicate.
| (11) |  |
It doesn't help to have the PRED of the NCOMP subcategorized for a SUBJ, the f-structure is still incoherent since the NCOMP is not subcategorized for by any PRED. And in any case, this analysis would mean that all nouns would have to be subcategorized for subjects, which is certainly not desirable. In order to analyze nominal sentences, then, we must either introduce a PRED in some other way than by using a verb, or we must change the well-formedness requirements for f-structures.
LFG's lexical forms capture the intuition that verbs require certain
syntactic functions. The verb in a sentence helps us to interpret the nominal
phrases in a sentence as filling various syntactic functions. In a nominal
sentence, however, there is no verb that helps us to interpret the various
nominal phrases. It is the syntactic construction itself that provides
the information about how these phrases are to be interpreted. In Maori,
the first NP is the predicative complement and the second is the subject.
Since this information comes from the syntax and not from the lexicon,
it might seem natural to let the phrase structure rule for this sentence
type introduce a PRED that could subcategorize for
these functions.
| (12) |  |
This would result in a coherent f-structure.
| (13) |  |
But in his article 'Semantics for Lexical-Functional Grammar', Per-Kristian Halvorsen says that this type of semantic form is not permissible in LFG.
| (14) |  |
Since it is a characteristic of the imperative construction that the understood subject is in the second person, it does not seem unreasonable that this information should be introduced in a phrase structure rule. Of course, this rule could instead include the schema: (^ SUBJ PERS) = 2. But that would not solve the problem being addressed in (14), namely that the f-structure needs a PRED for its SUBJ. In this case, however, there is a verb that this information could be associated with instead. Halvorsen does not say what kind of relations he has in mind when he mentions "universally significant relations" that could be "treated specially in the semantics", but both the topic-comment construction and nominal sentences are so widespread among the languages of the world that they would seem to be candidates.
If weighty arguments are found against the introduction of PREDs
in phrase structure rules, the only other alternative would be to allow
f-structures without PREDs for these constructions.
In order to do this, we need to have some other way of doing the work otherwise
done by the PRED. One of the main functions of PREDs
is to specify the subcategorization requirements of lexical items. In addition,
they specify, albeit in a kind of shorthand notation, how the semantic
arguments of a predicate are assigned to syntactic functions. Instead of
using this shorthand, we can spell out more explicitly how this linking
is done by associating for instance situation schemata with the f-structures.
In the following, I will use situation schemata similar to those used in
Fenstad et al. (1987). First we will look at a simplified feature structure
for the English sentence in (15).
| (15) |  |
Since each f-structure will have a situation schema associated with
it, the equations used in lexical entries and phrase structure rules must
specify whether they refer to f-structures or situation schemata. The abbreviations
FS and SS in the following stand for f-structure and situation schema respectively.
The most important equations for this example are introduced under the
(partial) lexical entry for the verb ate, as shown in (16).
| (16) |  |
The abbreviated feature structure in (17) shows how these equations
establish the correct connections between semantic arguments and syntactic
functions.
| (17) |  |
Returning to Maori, let us examine the nominal sentence in (4) and its
f-structure (5), repeated here as (18) and (19).
| (18) | |
| (19) | |
Since we don't have any PRED that comes from a
verb, we need to find another way of expressing the fact that the relation
expressed by the predicative complement is to be interpreted as a predicate.
We can in fact let the f-structure be just as it is in (19) and still get
the relation expressed by the head of the predicative complement to be
the main relation in the outer situation schema. Since we don't have any
PRED
that can carry this information, we need to have the necessary equations
in the phrase structure rule. The rule we need for a nominal sentence in
Maori with an NCOMP is given in (20). (This analysis
is inspired by Helge Dyvik's treatment of predicative complements in the
machine translation system PONS (Dyvik 1990:47-48).)
| (20) |  |
This will give us the feature structure in (21).
| (21) |  |
As we can see, by following a path from the outer f-structure, we find the relation that we want for the outer situation schema. We don't need to do anything to the f-structure in order to express that the head of the NCOMP is interpreted as a predicate.
The same type of equations may be used in the analysis of objective
predicative complements. For the sentence in (22), we can derive the feature
structure in (23).
| (22) |  |
| (23) |  |
The equations necessary for this structure may this time be included
as part of the lexical entry for the verb call, as shown in (24).
| (24) |  |
On this analysis, a verb like call takes three arguments: an agent, a theme, and a proposition. We employ the same technique to link elements of the f-structure with elements of the situation schema. Specifically, the relation of the proposition is identified by following a path from the f-structure of the NCOMP, and the argument of this relation is identified by following a path from the f-structure of the OBJ. Just as in the nominal sentence in Maori, the relation expressed by the predicative complement is not associated with a verb.
Either one of these alternatives will work for the constructions I have discussed in this paper. In choosing one alternative rather than the other, one might of course want to consider other factors than those I have mentioned. In a recent exchange on the LFG-list, several researchers have questioned the role of PRED in current versions of LFG. In this discussion, Ron Kaplan said:
Andrews, Avery: 1982, 'The Representation of Case in Modern Icelandic', in Joan Bresnan (ed.), The Mental Representation of Grammatical Relations, The MIT Press, Cambridge, Massachusetts, pp. 427-503.
Biggs, Bruce: 1969, Let's Learn Maori, Reed Education, Wellington.
Bresnan, Joan and Sam A. Mchombo: 'Topic, Pronoun, and Agreement in Chichewa', Language 63:4, 741-78
Dyvik, Helge: 1990, The PONS Project: Features of a Translation System, Skriftserie fra Institutt for fonetikk og lingvistikk, nr. 39, serie B, Universitetet i Bergen.
Fenstad, Jens Erik, Per-Kristian Halvorsen, Tore Langholm and Johan van Benthem: 1987, Situations, Language and Logic, D. Reidel Publishing Company, Dordrecht.
Grimshaw, Jane: 1982, 'On the Lexical Representation of Romance Reflexive Clitics', in Joan Bresnan (ed.), The Mental Representation of Grammatical Relations, The MIT Press, Cambridge, Massachusetts, pp. 87-148.
Halvorsen, Per-Kristian: 1983, 'Semantics for Lexical-Functional Grammar', Linguistic Inquiry 14:4, 567-615.
Jespersen, Otto: 1924, The Philosophy of Grammar, George Allen & Unwin, Ltd., London.
Kaplan, Ronald M.: 1996, LFG-list, June 13.
Kaplan, Ronald M. and Joan Bresnan: 1982, 'Lexical-Functional Grammar: A Formal System for Grammatical Representation', in Joan Bresnan (ed.), The Mental Representation of Grammatical Relations, The MIT Press, Cambridge, Massachusetts, pp. 173-281.
King, Tracy Holloway: 1995, Configuring Topic and Focus in Russian, CSLI Publications, Stanford.
Li, Charles N. and Sandra A. Thompson: 1976, 'Subject and Topic: A New Typology of Language', in Charles N. Li (ed.), Subject and Topic, Academic Press, New York, pp. 457-490.
Rosén, Victoria: 1988, 'Unbounded Dependencies and Empty Pronouns in Vietnamese', in Victoria Rosén (ed.), Papers from the Tenth Scandinavian Conference of Linguistics, volume 2, Skriftserie fra Institutt for fonetikk og lingvistikk, nr. 31, serie C, Universitetet i Bergen.