Topics in the Syntax and Semantics of Infinitives and Gerunds

Gennaro Chierchia

• Publisher: Routledge
• Year: 2016
• ISBN:
  9781315459073
• Edition: 1
• Topic: Lingue e linguistica
• Subtopic: Grammatica e punteggiatura

This title, first published in 1988, is an inquiry into the nature of predication in natural language. The study is based on the hypothesis that infinitives and gerunds are not clausal or propositional constructions and attempts to provide support for such a hypothesis, whilst also drawing from analysis of various anaphoric phenomena. This title will be of interest to students of language and linguistics.

C H A P T E R IV
CONTROL AND SEMANTIC STRUCTURE

In the preceding chapters we have argued on the basis of semantic and syntactic (distributional) characteristics of infinitives and gerunds that these constructions should be regarded as being VPs (or, in the case of some gerunds NPs derived from VPs) associated with nominalized properties. Henceforth, we will refer to constructions of this sort with the label ‘verbal arguments’ (i.e. VPs that occur in argument position). Now, there are in English, as well as in many other languages, a number of constructions where verbal arguments are understood as if they had a subject. They are illustrated in what follows:

(1)	a.	John tried to get a job
	b.	Bill forced John to go to Alaska for a vacation
	c.	Bill believes John to have gone to Alaska
	d.	John thinks it is important to be at the meeting

In (1) the NP John bears various grammatical relations to the sentence in which it occurs (in (1a) and (1d) it is the subject, in (1b–c) the object) and acts simultaneously as the understood subject of the infinitival complement (though not necessarily so in (1d)). In the tradition of generative grammar, control theory deals with the principles that determine the ways in which a constituent may be selected as the understood subject of verbal arguments. In particular, (1a, b, d) are known in the literature as Equi structures while (1c) as a raising (to-object) structure. Raising structures are characterized by the fact that the NP controlling the relevant verbal argument does not seem to be a ‘logical’ argument of the matrix verb, as the following examples illustrates:

(2) a. John believes it to snow

b. *John forces it to snow

The dummy it (a typical example of ‘virtual’ or ‘non logical’ argument) can occur as object of believe but not of force. Control theory has represented and represents one of the major concerns of current linguistic debates also because it has important consequences for linguistic metatheory in general. It has been argued, for example, that a theory of control has to make crucial use of phrase structural configurations (Williams (1980), Chomsky (1982), Manzini (1982), Koster (1981)) or of grammatical relations (e.g. Bresnan, 1982). Control theory has also represented one of the traditional bases for the claim that it is necessary to posit the existence of phonetically unrealized subjects either in the syntax (GB and related approaches) or in the semantics (LFG, many versions of MG).

Within the framework we are adopting here, verbal arguments do not have subjects neither in the syntax nor in the semantics, and reference to phrase structural configurations or to grammatical relations in syntactic and semantic rules is, in general, disallowed. Thus we are forced to look in different directions to provide the basis of a theory of control. In fact, we will try to argue that the crucial empirical generalizations having to do with control phenomena can be captured in a satisfactory way in terms of a very simple set of conditions on what can constitute a proper model (in the sense of model theory) for natural languages. We will try to provide a truth conditionally explicit approach to control that does not have any of the characteristics mentioned above (i.e. crucial appeal to phrase structural geometry or to grammatical relations, postulation of null subjects), and try to compare it with other available alternatives. Our approach can be regarded as an attempt to put together the guiding principles of semantically based theories of control (especially, Bach 1979) with ideas developed within LFG (Bresnan 1982).

For our purposes it is convenient to distinguish two basic kinds of control structures: argument control and modifier control. By argument control we will refer to constructions where a predicative expression lacking an overt subject (i.e. the controllee) is semantically an argument. In this sense (1a–d) are all cases of argument control. Whatever the syntactic constituency of infinitives may be, there is in fact general agreement on the hypothesis that the semantic function-argument structure of, for example, (1) looks more or less as follows:

(3) try* ( get a job’)(j)

Where try’ is a function and get a job’ and j are its arguments. Hence, what we call argument control includes all the cases that within the standard theory were analyzed as Equi, Super-Equi and Raising constructions.

By contrast, modifier control concerns structures where the controlled item is (part of) a modifier, which semantically are analyzed usually as functions. Modifier control concerns, therefore, constructions such as purpose and rationale clauses, adjuncts, infinitival relatives and the like. A consideration of modifier control phenomena goes beyond the scope of the present work, and we will consider only argument control. Throughout, in speaking about control (with no further qualifications) we will mean argument control (unless otherwise specified).

1. Relevant Data.

In what follows we will review some of the facts that any adequate theory of control must deal with. In doing so we will draw from various sources (especially Williams, 1980 and Bresnan, 1982) although the empirical typology of control phenomena we will present differs somewhat from most of these sources. In particular we will suggest that control phenomena form three natural classes that we call obligatory, semi-obligatory and prominence control.

1.1. Obligatory control.

What we call obligatory control phenomena correspond more or less to what Williams (1980) calls by the same label (though extended here as to include raising structures) or to what Bresnan (1982) calls ‘functional control’. Relevant examples are the following:

(4) a. John tries to leave

b. Bill persuades John to leave

c. John promises Bill to leave

d. Bill believes John to leave

e. John seems to leave

(5) a. Bill considers John a friend

b. Bill thinks of John as a friend

c. John strikes Bill as pompous

d. John painted the room red

(6) a. John likes drinking beer

b. Mary accuses John of being a communist

There are at least six properties that I think make the constructions exemplified in (4)–(6) a natural class. They can be identified in the following terms.¹

First, the relation between the controlling NP and the controlled verbal argument is strictly local. In particular, it can never cross an S boundary:

(7)	a.	*Mary believes that John tried to kill himself
	b.	*Mary believes that Bill strikes everybody as proud of herself.
	c.	*Mary believes that John likes making a fool of herself

Second, the understood subject of the constructions in question can never receive an arbitrary (i.e. generic or contextually specified) interpretation.

Third, the controlling NP is uniquely determined, even when there is more than one potential controller. For instance objects can obviously be obligatory controllers (see e.g. (4b), (5a) or (6b)). However there are verbs that have objects but do not allow them to control a verbal argument in their complement structure:

(8) a. *John promised Mary to wash herself

b. *John strikes Mary as proud of herself

c. *Mary accused John of killing herself

This means that the controller must bear a specific θ-role with respect to the matrix predicate (i.e. must be the agent or the goal, etc.) We will call this property ‘thematic uniqueness’.

Fourth, the controlled verbal argument can’t have more than one controlling NP. This phenom...