So far we have talked about the kind of grammatical knowledge that can be expressed in terms of a constituent structure tree --- information about the constituent units, and the parts of speech. But there are other kinds of information implicit in these representations which it is useful to make explicit. In particular, information about which phrases fulfil which grammatical relations or grammatical functions such as SUBJECT, OBJECT and SENTENTIAL COMPLEMENT. English SUBJECTs are normally the NPs which come before the verb, and OBJECTs normally occur immediately after the verb. In other languages these relations may be realised differently with respect to the verb. For example, in Japanese the normal word order is SUBJECT OBJECT VERB, and in Irish and Welsh it is VERB SUBJECT OBJECT. In many languages, such as Russian , the VERB, SUBJECT and OBJECT can appear in essentially any order. (In such languages the different grammatical relations can often be recognized by different forms of the noun -- usually called cases. In English, this only occurs with pronouns --- he, she, etc., are only possible as SUBJECTs). What this suggests, of course, is that while the constituent structures of languages differ greatly, they may appear more similar when described in terms of grammatical relations .
Phrases which serve as SUBJECT, OBJECT, etc., should also be
distinguished from those which serve as MODIFIERs, or ADJUNCTs, of
various sorts. For example, in the sentence (
)
You is the SUBJECT of the verb clean, the printer casing
is its OBJECT, whilst the prepositional phrases with a
non-abrasive compound and at any time are ADJUNCTs.
ADJUNCTs are prototypically optional --- unlike SUBJECTs. For example,
a sentence which omits them is still perfectly well formed: there is
nothing wrong with (
a), but omitting the SUBJECT, as illustrated in (
b)
produces an ungrammatical result.
There are various ways of representing sentences in terms of
grammatical relations , but it is essentially not very different
from that of constituent structure tree representation, which we have
seen earlier in this chapter. The basic idea is
to represent sentences in terms of their constituent parts (so a tree
representation is convenient), but since one wants to represent the
grammatical relation which the parts have to the whole, it
is common to mark either the branches or the nodes with the
appropriate relation. Figure 
gives a
representation of (
). This can be compared with a constituent
structure representation for the same sentence in
Figure .
Figure: A Representation of Grammatical Relations
Figure: A Constituent Structure Representation
In Figure , the relations are marked on the
nodes, and a new relation HEAD has been introduced. The HEAD element
is, intuitively, the most important element from the point of view
of the grammar of the whole phrase --- the element which makes the
phrase what it is. This is the noun in an NP, the verb in a VP or
sentence, the preposition in a PP.
There are three important differences between this tree representing
grammatical relations , and those representing constituent structure .
First, instead of consisting of an NP, and a VP (containing a V and an
NP), the representation of grammatical relations consists of a V and
two NPs -- the VP node has disappeared. Second, in this grammatical
relations representation, the order of the branches is
unimportant. This is possible, of course,
because the grammatical relations have been indicated and this gives
information about word order implicitly. Figure
could be redrawn with the branches in any order, and it would still be
a representation of The temperature affects the printer,
since this is the only sentence that has these items with these
relations. By contrast, reordering the branches in a constituent
structure tree might produce a representation of a quite different
sentence, or no sentence at all.
The third difference is that some of the words have been missed out
from Figure , and have been replaced by
features , that is, pairs that consist of an attribute, such
as def, tense, and aspect, and a value, such
as +, pres, and perfective. The features
aspect=perfective and tense=pres indicate that
the sentence as a whole is in the present perfect tense. It is called
perfect because it is used to describe events or actions that
have been `perfected' or completed, unlike, for example, a sentence such as
The temperature was affecting the printer, where the
`affecting' is still going on at the time the writer is referring to.
It is called present perfect because the auxiliary verb is in a
present tense form ( has not had). The feature def=+
on the NPs means these NPs are definite. This definiteness
indicates that the writer and reader have some particular object of
the appropriate kind in mind. Compare, for example, The printer has
stopped where one particular printer which is in some sense known
to both writer and reader is being discussed, with A printer has
stopped, where this is not the case.
These three differences are all intended to represent what is expressed by the sentence, abstracting away from the way it is expressed: we abstract away from the division into NP and VP, from the particular word order, and from the way in which the definiteness of the NPs and the tense and aspect of the sentence are realized (in English it is by the determiners, and the auxiliary verb respectively; in other languages it might be realized differently).
When it comes to describing the relationship between constituent
structure , and what we might call relational structures, such as
Figure , there are basically two approaches. One is
simply to add information about grammatical relations to the grammar
rules.
SNP{SUBJECT} AUX VP{HEAD} VP
V{HEAD} NP{OBJECT} PP{ADJUNCT}* AUX
has{aspect=perfective, tense=pres}
The idea is that these annotations can be interpreted in such a way
that a representation like Figure can be
constructed at the same time as the constituent structure tree . To do
this requires a convention to `flatten' the constituent structure tree
`merging' a structure (e.g. the structure of S) that is associated
with the LHS of a rule with that of the HEAD daughter on the
RHS, and
a convention which simply merges in information that comes from items
which do not have a grammatical relation , such as the AUX.
A second approach is to have special rules which relate the constituent structure representation to the representation of grammatical relations. One such rule might look like this:
NP:$1, AUX:$2,
V:$3, NP:$4
HEAD:$3, SUBJ:$1, OBJ:$4
In this rule, $1, $2, etc. are variables, or temporary names
for pieces of structure. The idea is that such a rule matches a
constituent structure such as that in Figure , and
assigns (or `binds') the variables to various pieces of structure.
For example the NP containing temperature becomes bound to the
variable $1. The rule can then be interpreted as an instruction to
transform the constituent structure tree into a tree like
Figure . This involves making this NP into the
SUBJECT, making the V into the HEAD, and missing out the AUX entirely,
among other things. The rule is rather simplified, of course, since it
does not mention putting the information about perfective aspect into
the grammatical relation representation, and ignores the problem of
dealing with PPs, but it should give some idea.
The reader may also notice that the arrow used in this rule is
bidirectional . This is intended to suggest that the rule simply
states a correspondence between constituent structure , and grammatical
relation representations, without suggesting that one is
prior to the other. Thus, the idea is that one could
equally well use the rule to transform Figure into
Figure and vice versa. Similarly, the annotation
approach is not supposed to be directional (though this may be
somewhat harder to appreciate).
Many verbs have what are called active and passive forms, as in the following.
. ( Active)
are affected by temperature. ( Passive)
Notice that the object in the active sentence corresponds to the
subject in the passive. This raises the question of what the
grammatical relations SUBJECT and OBJECT should mean. One possibility
is to use the the terms in the sense of the `surface' grammatical
relations. The SUBJECTs of actives and the corresponding passives
would be different, then. In particular, temperature would be
the SUBJECT of (
a), and printers would be the SUBJECT of
(
b). The alternative is to adopt a notion of a deep
relation which picks out the same elements in both active and passive
sentence. We would then say that (in English) the D-OBJECT (`deep'
OBJECT) corresponds to the noun phrase after the verb in active
sentences and to the noun phrase that precedes the verb in the
corresponding passive. In active sentences, the surface and deep
relations are the same, but they are different in passives, as can be
seen from the following (in the passive sentence there is no surface
OBJECT, and the D-SUBJECT has become a sort of ADJUNCT, in a PP with
the preposition by).
Interpreting SUBJECT as deep subject is clearly consistent with the general idea of abstracting away from surface characteristics in the grammatical relational representation. But it is not obviously the right move to make. For example, English verbs often vary their form depending on the nature of their subject (this is called agreement -- as the following makes clear, there is also agreement of demonstratives like this/ these with their head noun).
However, the point to notice is that the agreement is with the
surface subject, not the deep subject. Thus, if one wants to use a
representation of grammatical relations to describe the phenomenon of
agreement , the notion of SUBJECT had better be surface subject.
This is not, in itself, a critical point here. The point we are making
is simply that there is a range of options, and that the option
chosen can make a difference for the overall description.
