18.104.22.168 D-structure and Theta Theory
Let us consider the nature of D- and S-structure a little more closely. An obvious question is why it is that some elements start off in one position and then move to another. To answer this question we have to ask about why elements occupy the positions they do at any level of description. This is a matter of distribution: there are grammatical principles which determine the range of possible positions of categories of certain types. X-bar principles obviously have a large role to play in this, determining head, complement, specifier and adjunction positions. But as both D- and S-structures conform to X-bar principles, this clearly is not what differentiates the two. Obviously there must be other grammatical principles holding at D-structure which are not applicable at S-structure and vice versa.
A D-structure principle may then require a constituent X to occupy a certain position and an S-structure principle may require X to occupy a certain position, and if these two positions are not the same then X will have to move from its D-structure position to the required S-structure position. Thus, explaining movement is a matter of finding out the principles which determine the distribution of elements at D- and S-structure.
Turning to D-structure first, an important consideration which has been present in all developments of this concept, first proposed by Noam Chomsky in the 1950s, is that D-structure positions are somehow basic. For example, in a passive sentence, what sits in the subject position at S-structure is interpreted as the object of the verb and hence is assumed to occupy the object position at D-structure:
|(63)||S-structure:||Ken was confused|
|D-structure:||was confused Ken|
The idea is that the way an element is interpreted in terms of its thematic status indicates its D-structure position and thus if something is interpreted as an object it will be in an object position at D-structure. Moreover, an element that is interpreted as the subject or object of a predicate will be in the relevant subject or object position of that predicate at D-structure:
|(64)||S-structure:||Ken was considered to be confused|
|D-structure:||was considered to be confused Ken|
In this example, although Ken is sitting in the subject position of the verb consider, this element is interpreted as the object of confused and thus is in the object position of this predicate at D-structure.
D-structure then is a pure representation of thematic relations. Anything which is interpreted as the subject or object of a given predicate will be in the subject or object position of that predicate at D-structure no matter where it is found at S-structure.
The principles that determine D-structure positions must therefore have something to do with thematic relationships. We saw in chapter 1 how Θ-roles are encoded in the lexical entry of predicates. Yet in a sentence it is the arguments that are interpreted as bearing these Θ-roles. It must be the case therefore that these Θ-roles are given from the predicate to the argument. We can refer to this process as Θ-role assignment. For example:
The verb arrive is a one-place predicate, having one Θ-role to assign which it assigns to the argument an unexpected package in (65a). The verb mend is a two-place predicate. It assigns the agent role to its subject and the patient role to its object.
Where can a predicate assign its Θ-roles to? If there were no restrictions on this then arguments would not have distributions at D-structure as they could appear anywhere. We are assuming that this is not so and hence there must be conditions which determine where Θ-roles can be assigned. One fairly clear condition on Θ-role assignment that can be seen in (64) is that Θ-roles are not assigned over long distances. For an argument to receive a Θ-role from a predicate it must be close to it. We can see this from the fact that the following sentence has just one interpretation:
|(66)||Sophie suspects that Linda loves Dwain|
We can only interpret this sentence with Sophie doing the suspecting, Linda doing the loving and Dwain getting loved and there is no way to get Sophie associated with love or Linda and Dwain with suspect. This is simply because Sophie is structurally closer to suspect and Linda and Dwain are close to love. If love could assign its Θ-roles over long distances, Sophie might be able to be interpreted as one of its arguments.
We will adopt the following restrictive condition on Θ-role assignment:
|(67)||the Locality Restriction on Theta-role Assignment|
|a predicate assigns its Θ-roles to either its complement or its specifier|
It is a long standing assumption that there is a uniformity in Θ-role assignment which links certain Θ-roles to certain positions. The reason why the object is assumed to move in a passive sentence is precisely because of this assumption. In an active sentence the object occupies the object position, following the verb, and so it is assumed that in the passive sentence the argument that is interpreted identically to the object in the active originates from the same position that we see it in in the active:
|(69)||a||Monika munched the sandwich||=||active|
|b||was munched the sandwich||=||D-structure of passive|
|c||the sandwich was munched||=||S-structure of passive|
Thus it is assumed that there is a uniform position to which the patient Θ-role is assigned across different structures. We will actually adopt a very rigid form of this idea which was first proposed by Baker (1988), called the Uniform Theta-role Assignment Hypothesis (UTAH):
|(70)||the Uniform Theta-role Assignment Hypothesis|
|a Θ-role Þ is assigned in the same structural configuration in all structures in which it is present|
Then it follows from the UTAH that all themes in all structures will be assigned to the specifier of the verb that they are related to. We will see that this is a very restrictive theory of Θ-role assignment that will force certain analyses of structures which, while not at first obvious, turn out to have a number of positive features which go to support them and in turn this supports the assumption of the UTAH in the first place.
There are other aspects of the assignment of Θ-roles than those to do with where they are assigned. We saw in chapter 1 that for some predicates an argument that they select as a lexical property does not have to be realised as a syntactic entity but may be present only at a semantic level. Such an argument would be understood, but unable to play any role in a sentence such as licensing a reflexive pronoun:
|(72)||a||Paul ate the pie by itself|
|b||*Paul ate by itself|
|(73)||a||Fiona found the book|
|c||*found the book|
It is not well understood what determines when a Θ-role may be left understood, but it seems to be an idiosyncratic property of certain predicates. It is generally the case that Θ-roles must be assigned. The Θ-role assigned to the subject, for example, cannot be left as understood. Therefore we might propose that there is a general grammatical condition ensuring the assignment of Θ-roles, unless they are marked in the lexical entry of a predicate as being able to be understood. Moreover, a theta role can only be assigned to one argument and cannot be ‘shared out’ between more than one:
|(75)||a Θ-role must be assigned to one and only one argument|
|b||*Sam smiled the cat|
The verb smile is intransitive and therefore does not have a Θ-role to assign to an object. If we provide this verb with an object, we therefore have an argument that receives no Θ-role, which as we see from (76) is ungrammatical. Moreover, an argument cannot receive more than one Θ-role. So if a predicate must assign more than one Θ-role, it cannot assign them both to the same argument:
If it were possible for one argument to bear both Θ-roles of a predicate, (77b) would mean the same thing as (77a) which has a reflexive pronoun in one argument position taking its reference from the other argument. The unacceptability of (77b) can therefore not be a semantic fact.
It is also not possible for an argument to bear two Θ-roles assigned from different predicates. Consider the following:
|(78)||Knut knows Dennis danced|
This sentence is grammatical, but only with the interpretation that what Knut knows is that Dennis danced. In other words, the arguments of know are Knut, a DP, and Dennis danced, a sentence in which Dennis is the argument of danced:
|(81)||an argument must bear one and only one Θ-role|
Together the conditions in (75) and (81) are called the Theta Criterion:
|(82)||The Theta Criterion|
|a Θ-role must be assigned to one and only one argument|
|an argument must bear one and only one Θ-role|
We have now reviewed three simple and basic principles which regulate the assignment of Θ-roles within a structure: the Locality Condition on Theta-role Assignment, the UTAH and the Theta Criterion. All of these apply to D-structures, restricting the distribution of arguments at this level of representation. Collectively, the principles which govern Θ-role assignment are often referred to as Theta Theory and this can be considered as a part of the grammar, similar to the principles of X-bar theory which regulate the general formation of structures. A final important contributor to the well-formedness conditions of D-structure is the lexicon which provides structures with categorial information and Theta theory with the Θ-roles to be assigned. We might represent this in the following way: