PURPOSIVE CONSTRUCTIONS IN ENGLISH         

Thorold May

ACKNOWLEDGMENTS

Thanks are due to the Linguistics staff at the University of Newcastle, N.S.W., who gave me an intellectual home for so long. In particular, Professor Ray Cattell, Dr. Peter Peterson and Dr. George Horn acted at various times as supervisors during the preparation of this research. Peter Peterson and anonymous referees offered comments on an earlier draft prior to publication in the Australian Journal of Linguistics, (1990:Vol. 10,1 ). Special proxy thanks should also be given to Charles Jones and Emmon Bach whose respective papers, "Agent, Patient & Control in Purpose Clauses" and "Purpose Clauses & Control", I used as stalking horses throughout the analysis, sometimes in ways that might have surprised them.

ABSTRACT

This thesis explores some of the syntactic & semantic properties of Purposive Constructions in English. The term "purposive" is recognized as a semantic concept which finds regular expression in a small range of syntactic configurations. Purpose Clauses (PCs) and Rationale Clauses (Rat.Cs) are examined in some detail. Briefer reference is made to several other configurations, notably Because Clauses, So-That Clauses and Infinitival Relatives. In general Purposive Constructions comprise rather fuzzy semantic categories. Nevertheless, the main syntactic features are fairly clear. Interpretation of the constructions requires a systematic account of the control of empty slots (ellipted NPs) by thematic elements in the matrix clause. General conditions of Government and Binding appear adequate to predict the distribution of gaps in most Purposive Clauses. However, the relationship between propositions predicated of a common argument in these constructions is found to sometimes require matching conditions too subtle for syntax alone to predict. A concept of Thematic Coextensiveness is introduced to account for such matching.

CHAPTER I : STANDARD PURPOSE CLAUSES

The aim of this thesis is to re-examine existing notions of

Purposive Constructions, with particular attention to Purpose

and Rationale Clauses. It will be shown that they can be

accounted for as possible structures of English at two levels

of the grammar. Syntactic principles preclude certain structures and

control paradigms. However these alone are not sufficient to

constrain the range of acceptable sentences. In addition a

semantic level comes into play in ways which are sometimes only

marginally linguistic. It will emerge that conventions of

pragmatic construal, particularly those relating to an Agentive

Source for action, must be applied in order to predict linguistic

outcomes. Such conventions depend effectively upon knowledge of

the world and control many of the interpretive possibilities of

what I generically call Purposive Constructions.

A Purpose Clause (as identified by Bach, Chomsky, Jones and others) has the following forms:

a) Object Purpose Clause

^1 They brought Johni along [e to talk to ei].

b) Subject Purpose Clause

^2 They brought Etheli along [ei to talk to the children]

A Rationale Clause has the following forms :

^3 Maryi bought Bill a book [(in order) ei to help him].

^4 Mary bought Bill a book [in order for him to succeed].

2. Attempts to Define Purpose Clauses

"Purpose" is a semantic concept which has been attached to

some generally recognised syntactic configurations. This is a

familiar procedure (Causation is another example) but it can lead

to considerable analytic ambivalence. Nowhere is this more

evident than in the papers of Bach and Jones which I refer to

extensively below.

Bach in particular seeks a wider semantic

perspective, without, however, exploiting it very successfully.

He feels that "...the meaning of Purpose Clauses is to be

explicated within a general account of purposeful activity and

our language about it..."(Bach 1982:53).

Williams (1980), who briefly examines Purpose Clauses in his

study, Predication, and Chomsky (1980), who mentions them in On

Binding, have an essentially syntactic perspective. That is,

grammatical explanation, even Logical Form (semantics), is seen

as a matter of mutually relating lexical configurations rather

than of relating form to psychological meaning.

This thesis on the other hand presumes a skepticism about the

autonomy of syntax in the unmapped territory of mind, and the

slant of explanation accordingly embraces a wider cognitive base.

The more focussed work on Purpose Clauses starts from

defined configurations, enquires about the nature of their

syntax, and may then posit various kinds of semantic facts as a

rationale.

The following Statement I, a description by Jones

(1985:105), makes a useful starting point, although it turns out

to be neither sufficient nor quite accurate; (n.b. nine

Statements are distributed throughout this thesis. They are loose

working hypotheses. Eventually, some are not sustained by the

investigation). Purpose Clauses are of two main types :

Purposive Constructions : Statement I

a) "Subject Purpose Clauses ... have an obligatorily
      controlled gap in subject position".

b) "Object Purpose Clauses ...have an obligatorily
      controlled gap in object position and an empty subject
      position".

3. The Syntactic Status of Purpose Clauses

It has been argued that Subject Purpose Clauses have a

fundamentally different syntactic structure from Object Purpose

Clauses. Williams proposed that some PCs (OPCs, though he didn't

call them that) are dominated by S~, while others (SPCs) are

dominated by S only. The Object gap in OPC was to be a trace

governed by "WH Movement" while the Subject gap, and the gap in

SPCs, would be governed by Control (i.e. co-indexed with a matrix NP). Thus :

^5 [OPC]

Williams' theory of predication claims that S in VP is a

predicate controlled by the matrix Theme. This has some relevance

to our later discussion of PC Environments (Section 11).He links

the control of S to the control of the gap in PCs. Although his

model of predicates is open to challenge, the final link between

the matrix Theme and PC gaps is beyond question.

In terms of Chomsky's Binding Theory, PRO2 in ^5 is assigned

control by it in the matrix (Minimal Distance Principle), while

PRO1 is allowed to be indexed to matrix NP1 once control by V1 on

PRO2 has been exercised (Chomsky 1980:42;1981:77). The innovation

of PRO2 indexed to a trace is necessary to overcome an Opacity

Condition in this model.

The Williams/Chomsky analyses may have some problems.

Chomsky's very brief treatment of Purposive Constructions

overlooks the differences between Purpose and Rationale Clauses.

The WH slot of 1980 vintage is later conceived of as a general

complementiser slot fronting all clauses, and (pruning apart)

would seem to be as applicable to ^6 as to ^5 (refer to Williams'

distinction between S~ and S Purpose Clauses). Another of

Williams' examples, ^7, seems to illustrate the more general

paradigm, where the lexical insertion or ellipsis of for governs

the possibility of a lexical Subject in the lower sentence. It

will be interesting to observe in the chapter on Rationale

Clauses that where the COMP slot is governed by the complex

conjunction, in order then there is no possibility of a trace

(i.e. of ellipsis) in the Object position of the lower sentence.

Note incidentally that ^7 is an OPC whose Subject theta

position is not empty, contrary to Jones' definition in Statement Ib:

^7 S

The main task of any syntactic model applied to PCs is to

account for control of the "understood", non-lexical elements in

their structures in a regular way. The PRO insertions of ^5, ^6

and ^7 are model-specific. We might dispute them. Thematic

elements such as Agent, Patient or Theme (theta roles) are less

controversially assigned by verbs, and if we accept something

like Chomsky's Projection Principle, must find representation in

the grammar. The possibility, first proposed by Jackendoff

(1972), that thematic relations govern the application of

control rules, will strongly influence the present analysis.

A point on notation : in this thesis, where a thematic role

has no lexical correlate, I normally mark a slot "e" (empty) at

an appropriate point in the lexical string. In most cases I avoid

any direct judgement about the model-theoretic status of "e" as

PRO, trace or whatever. Where PRO or trace are labelled as part

of the argument, the general sense intended is as in LGB Theory.

However, generative models are shaky in so many aspects of

convention and detail that I have used them as an aid rather than

an orthodoxy, and deviated where necessary.

The explanation of "e" slot control will partly turn on the

constituents in which thematic roles are seen to be embedded.

While Williams and Chomsky assume that all PCs have the

constituent structure of sentences within a matrix verb, Jones

considers SPCs to be sentences and OPCs to be VPs. Jones'

approach is to suggest that the NP slots in PCs are empty because

they are assigned no Case. In SPC there is no [+TENSE] element to

assign Case to Subject position, and "...Object-Case absorption

by V (in OPC) goes hand in hand with VP's inability to assign an

external theta role;" (Jones 1985:111).I find these linkages a

bit obscure.

"Object Case Absorption" seems to be an unsupported notion.

The Object theta position in PCs is certainly governed (by the

verb, or the particle to). The argument for TENSE to assign Case

to the Subject theta position has a longer history. For an

independent discussion of grammatical relations pertaining to

Subject, see Williams (1984). If SPCs cannot assign Case to

Subject position, then neither can OPCs since neither are

tensed. Most analysts assume an ungoverned PRO in the empty

Subject slot of such infinitival phrases. It is therefore not

obvious why SPC should be treated as a sentence and OPC as a verb

phrase.

My own intuitive feeling about both ^5 and ^6 is that give

and hold do assign an external, Subject position theta role whose

coreference is controlled by thematic rather than strictly

syntactic considerations. Above I mentioned "understood" elements

in PCs, rather than using the more precise term ellipsis.

Although the arguments of a verb are not lexically present, and

may even be precluded by the surface syntax, a speaker may still

interpret their effect. Consider ^8:

^8. We brought himj along [ei to talk to ej ].

The Theme of ^8. is indexed to ej while ei is "free" but must have an

implicit or explicit (co)referent : probably we in this case. Ellipsis

normally implies that a reduced structure has a full lexical

analogue which is entirely synonymous. A difficulty with this

argument for treating PCs as conventionally ellipted sentences is

that the insertion of lexical material sometimes changes the PC

into a Rationale Clause, which has different properties :

^9. We brought him along (in order) for you to talk to him.

Therefore if PCs are "sentences" (clauses), then sentence

here means something more than the surface behaviour of lexical

strings in clauses. The word sentence is applied rather to the

subcategorized pattern of thematic relationships which define a

verb. A sentence would minimally be said to exist in a string

when the thematic relationships defining some verb could be

uniformly interpreted for (co)reference in the string by

competent users of that language. Grammars may have many

applications, and not all of them may need to engage thematic

concepts. However thematic roles are important explanatory tools

for the job at hand.

4. The argument for both SPCs and OPCs as sentential adjuncts

Both types of Purpose Clauses can be characterized as

sentential adjuncts, rather than strict complements to a matrix

verb phrase. The terminology is rather fluid here. A

prepositional phrase or a restrictive relative clause in a NP is

clearly an adjunct modifying a head. It sharpens the specificity

of the head, or put in another way, it is presumed to be non-

controversial information (Grice's sense; 1981) defining a topic.

On the other hand, adjuncts such as PCs in a VP are frequently

much more complement-like, adding information to the comment-

element of the discourse. The truth value of such adjuncts in a

VP may well be challenged. Thus although PCs are syntactically

optional (a defining property of adjuncts), their analysis lends

itself to the kind of machinery usually reserved for regular

(syntactically obligatory) complements.

5. The Predication of PC Adjuncts

Williams in his study, Predication,(1980) recognized that

any category can be a predicate. He observed that predicates

could be either thematically or grammatically governed, although

the former were said to all involve predicates in the VP, and the

predication was always of the Theme of that VP. However Williams

does restrict the notion of predication to what he calls

Obligatory Control by a lexically defined co-referent. I

ultimately find it more productive to think in terms of semantic

predicates whose substance and antecedents are propositions

rather than clauses.

Propositions may be explicit or implicit. It

is not our task here to challenge Williams' original

argumentation in detail, but it will emerge that the government

of predicates in adjuncts like PCs can be a very slippery matter.

See especially Section 8 below, where it turns out that the

antecedent of a semantic predicate can be a non-lexical exophoric

referent (which nevertheless determines the grammaticality of

sentences).

Jones (1985:107) develops an argument that Purpose Clauses

are predicates to the whole of their matrix verb phrase, and not

merely the matrix verb itself. His evidence for this proposal is

based here on a semantic, not a syntactic, observation : that

successful attachment of an adjunct depends upon the total

meaning of the main verb phrase rather than any single lexical

collocation with the matrix verb. For example,

^10. I sent John out of the room to call the children.

^11.* I sent John out of the room to talk to later.

Further instances would be ^12 and ^13 :

^12 I poured the moet to complement the cheese .

^13 *I poured the Moet down the drain to complement the cheese.

The verb pour, it is argued, cannot determine subsequent

agentivity by itself since it is primarily the interpretation of

the whole matrix complement's meaning (e.g. the meaning of pour +

the moet + down the drain ) which determines the matrix verb's

power to link that complement with any adjunct (e.g. a Purpose

Clause) expressing the intent of the matrix Agent.

However the difficulty with ^13 seems to me to arise from a

violation of conventional inferences between connotations in the

matrix complement and the adjunct, rather than any

"incompleteness" in the volitional scope of pour. That is, pour

down the drain suggests waste, a negative purpose, while to

complement the cheese suggests pleasure, a positive purpose. If

anything, the infelicity of ^13 proves that pour (and hence the

matrix Agent) must control and reconcile both phrases

simultaneously.

Perhaps we can hypothesize from ^13 that any class of

structures which retains a constant semantic label (such as

"purpose" or "cause") is going to be susceptible to clashes of

semantic felicity in complex environments. Pragmatic, normative

proposals can be made about such felicity constraints, but they

probably relate to "communicative competence" rather than

"linguistic competence" in the narrower sense used by Chomsky.

More formally, for a sentence to be felicitous and coherent, the

arguments of its main verb must be semantically compatible.

6. The Control of Purpose Clauses by a Matrix Agent or Patient

6.1 Thematic Hierarchy Condition

The initial description of PCs as constructions with an

obligatorily controlled gap must now be expanded by adding a

thematic explanation. Where there are two gaps (OPC), what I will

call a Hierarchy of Thematic Control Condition generally applies:

Purposive Constructions : Statement II

The THEME of the matrix clause controls that gap in a PC
adjunct which is the optimum available selection from a
thematic hierarchy of INSTRUMENT > PATIENT > AGENT.
In a typical PC adjunct containing a transitive verb, the Object
of that verb (V2) will be the Patient of the clause, and if
lexically empty will be the optimum match for the matrix Theme.

Hence Jones' description of OPC behaviour. Where the Object

position of V2 is not empty and control defaults to an empty

Subject position (SPC), other thematic properties such as Agency

may have an influence; (e.g. see the Condition of Manifested

Intent later in this thesis). The thematic behaviour of an

Instrumental phrase creates special conditions, also dealt with later.

I assume a definition of Theme roughly comparable to that

of Fillmore (1968) : that it is the element which moves with

respect to a verb of motion, or which in general is acted upon by

the behaviour of the verb. Thematic properties are discussed in

more detail in the Appendix.

Thematic hierarchies have been proposed in many studies of

"deep case" or "thematic relations" (Fillmore 1968, Jackendoff

1972, et al). It has always been possible to find exceptions to

such rules. Sentences ^18-^20 pose difficulties for the present

Hierarchy of Thematic Control Condition. With this in mind I

think it best to treat all such formulations as statements of

pragmatic tendency. Why would such a tendency exist ? I suspect

that discourse coherence has something to do with it. In a

sentence such as

^14 I brought himi along e to talk to ei.

... there is probably a natural assumption that the Theme of the

matrix will persist as a Theme in the discourse, all other things

being equal. That is, talk to in the PC adjunct of ^14 also

subcategorizes its Object position as Theme. Where the theta

position is empty, it will tend to take the matrix Theme as

coreferent. The thematic hierarchy itself no doubt reflects the

pragmatic probability of a matrix Theme assuming various semantic

roles in the extended discourse.

6.2 Unmarked thematic assignment to Subject and Object

It has been argued that in English, and evidently in most

other languages, the unmarked thematic assignment on verbs (i.e.

the assignment to theta position) is Agent in Subject position

and Patient in Direct Object position; (e.g. Jones 1985:111). It

is certainly common to a large number of verbs, and especially

those active matrix verbs most associated with Purpose Clauses.

Later I will question the precision of these thematic notions,

and the value of such a presumed association with verbs as a

class. Jones tried to establish an interpretive framework for

defining and accounting for Purpose Clauses. The assumed

canonical distribution of Agent and Patient was said to supply

this as follows.

Purposive Constructions : Statement III

In PCs.. a)"The Subject gap is controlled by a subsequently possible Agent".
              b)"The Object gap is controlled by a subsequently possible Patient".

"Subsequently possible" turns out to be a fairly pragmatic

idea in which possibility is somewhat extended by metaphor or the

linguist's imagination. More on this later. The core of the

notion is that subordinate verbs, like all verbs involved in PCs

(the assumption goes) will have Agent and Patient type argument

places. Where one of these argument places corresponds to a gap

it will be coreferent with an NP in the matrix clause. The

appropriate matrix NP will have the semantic potential to be a

proxy Agent or Patient for the empty theta position, subsequent

to the action of the matrix verb.

6.3 Variations on the Subject-Agent / Object-Patient paradigm

We can begin qualifying the proposal for an unmarked

association of Subject/Agent by noting that the subject gap of an

adjunct may have an argument which is ambiguously coreferent with

either the Subject or the Object of the matrix, both being

possible subsequent Agents.

^15 I brought JBi along [ei to impress the board].

^16 Ii brought JB along (in order)[ei to impress the board].

This ambiguity is resolved by the insertion of in order in

^16, demonstrating that the sentence contains a Rationale rather

than a Purpose Clause. Note that although JB can be the primary

Agent controlling ei in ^15, the Intention of I is also entailed.

^15 is actually reminiscent of Saksena's (1980) description

of the Affected Agent, a condition which is marked

morphologically in Hindi, and thus more susceptible to a

convincing syntactic explanation. Affected Agents in Hindi are

those which upon causativization of a complement (rather than an

adjunct as in our problem) in the environment of certain verbs

(only) are marked with the suffix -kao, all other Agents taking -see :

^17 a) mai-nee-ram-koo/*see khaanaa khil-aa-yaa
                 I-AGT  Ram-OBJ/AGT        food         eat-DC-PAST
                 I-CAUSE [Ram to eat food]
             I fed Ram

^17 b) mai-nee-ram-see/*koo peer kat-aa-yaa
                I-AGT Ram                           tree cut-DC-PAST
            I made Ram cut the tree

The point of the Hindi analogy here is that a particular

language may formalize a semantic distinction through morphology

or configuration. It is then amenable to syntactic "explanation".

In Rationale Clause sentences like ^16 there is a formal control

relationship between the Subject/Agent of the matrix verb and the

putative Subject of the adjunct's verb. However the syntactic

signals become vague when in order is excluded as in ^15, and we

seem to have no reliable syntactic marking in English for

semantic concepts such as Affected Agents. To explain what is

going on in the grammar in such environments we may have to go

beyond purely syntactic explanation.

6.4 Ambiguity of thematic co-reference for Object gaps

It is also possible (rarely) for the Object gap to be

ambiguous about the thematic coreferent. The situation may arise

when the verb and its NPs are exceptionally neutral about

relationships, as in

^18. Theyi had a planej [ei to catch ej].

^19 They had friends [ei to advise ej].

^20 They employed counsellorsi [ei to advise ej].

The unmarked version of ^18 seems to require that both of its

gaps be controlled by matrix arguments. The unmarked reading of

^19 probably parallels that in ^18, neither sentence being

purposive. That is, ^19 has ej coreferent with friends, although

the other, purposive, interpretation is certainly possible.

Have seems to permit alternative value assignments to its theta

positions, which must throw serious doubt on the generality of

Jones' rule for associating Agent with the Subject gap and

Patient with the Object gap. Sentence ^20 on the other hand

definitely marks counsellors as the advisers (ei), while leaving

the reference of ej as arbitrary, and perhaps typically

substituted by a lexical noun. In other words, ^20 is an unusual

SPC, not an OPC, and it is ei rather than ej which is subject to

obligatory control.

It is interpretive convention and pragmatics which link ei

in sentence ^18 to the matrix Subject. Imagine (hard!) where ei

could have exophoric reference, as in the case when they (the

company) had planes available for us (ei) to catch. This sense of

have available (rather than have = obligation) would render ^18

and OPC with ei under arbitrary control.

By a similar vigorous exercise of imagination, note how they in

^19 could be the Mafia who have friends (ei) to advise defaulting

debtors (ej). The shift in control of ei from the matrix Agent to the

Theme would render ^19 an SPC. The difference between these exotic

versions of ^18 and ^19 is influenced by the verb in the adjunct and

discourse context.

However, the default to SPC or OPC is

determined by the possibility of the matrix Theme acting as an

Agent in the subordinate clause. Clearly it cannot do this in ^18.

Thus even configurational contrast in the syntax of SPCs and OPCs

is not invariant; ultimately it is tied to semantic interpretation.

The exceptions just discussed do not weaken an argument that

the interpretation of PC gaps is linked in a principled way to

the distribution of thematic arguments in the matrix clause, but

they do show that interpretation is not restricted to a

particular thematic type predetermined by structural dominance

amongst the constituents (as Case is determined in English).

The examples reaffirm that obligatory control affects one gap; (the

literature says only one gap, but there are always bothersome

sentences like ^19). Furthermore, the matrix argument which

exercises obligatory control in Purpose Clauses is invariably the Theme.

7. Resultant States

The notion of Resultant State is developed by Bach to

describe the kind of semantic evidence which shows that an Agent

in matrix Subject position may exercise Intent and retain control

over subsequent events in the conjunct of a sentence.

^21 Johni went to New York for three days (but ei only stayed for two).
         Resultant State" of control
         over events in the conjunct

^22 Dinosaursi appeared on earth for three million years
          (*but only stayed for two million).

From Bach's examples we see that what are broadly called

Action verbs (as opposed to Stative verbs) may lead to a

"Resultant State" of the Agent. To me it looks more generally,

and more simply, like a pragmatic matching condition between

verbs in the respective clauses of the conjunct sentence.

A critical element in this case seems to be that the verbs must

imply an Effect through volitional action. Appear in ^22 implies

no volition or purpose. ^21 itself does not contain a Purpose

Clause, although it is purposive in meaning. The term "resultant

state" is a bit unfortunate since both John and the dinosaurs

enter into a state which differs from their inceptive condition.

Perhaps what needs to be isolated is an Induced Result. The

significance of Verbs of Induced Result for PCs will be developed

a little later.

Jones, diverging from Bach, adopts Resultant State to

label the status of a Theme in the context where it exercises

control through its matrix verb over the theta position of the

lexically null argument in a PC. Thus the Resultant State of a

matrix Theme is said to meet the condition for PCs when the lower

theta role that it controls represents a "subsequently possible"

Agent\Patient (depending upon the PC gap).

Note that the matrix Theme itself is normally found in

Object position, but with a verb like BE, as in ^23, may occur in

Subject position (assuming that we accept such sentences as PC

constructions). The Resultant State of the books in ^23 seems to

derive from the entire matrix predicate. Sentence ^23 is borrowed

from Jones, who does accept it as a PC.

           THEME     PREDICATE        BENEFICIARY          THEME2(?)
^23 The booksi are ready [for the children to receive ei]
                                                                       Resultant state

Although ^23 has many of the structural characteristics of a

PC, it seems to me that there are good grounds for believing that

it is semantically quite different, at least from the kind of

"purpose" that we have been considering. The matrix is a stative

copula sentence, and although there might be a pragmatic

inference of Agency somewhere, that is fortuitous to the context

of situation. The copula does not signal any induced condition.

Note the analogous sentence ^24 below in which there is no

necessary inference of Purposive Agency related to the matrix at all.

^24 The corn is ready to harvest.

Sentence ^23 again demonstrates the perils of trying to assign

theta roles like Patient and Agent within an unexceptional

framework of constituent structure. Detailed subcategorization in

the lexical component may be more promising.

If there is anything like a Patient in the traditional sense in the adjunct of ^23,

then it must be the children, for this is more or less how

receive subcategorizes its argument structure here. Beneficiary

is actually a better description. Therefore the books is not a

"subsequently possible Patient", but rather whatever it is

that we want to call the argument in the Object gap of the OPC,

if it is an OPC : Theme again? Transfer Element?.

Our starting analysis, Purposive Clause Statement III

divided the Sentient particles in PC environments into Agents and

Patients. These roles were assigned mechanically, according to

constituent structure. The procedure has proved unsatisfactory.

The purely syntactic assignment of thematic elements might render

them semantically empty (hence redundant). The preceding examples

from Bach and Jones fail, perhaps, to exemplify purposive

properties very well, but they do show that the semantic

interpretation of syntactically empty slots turns upon the

matching semantic properties in associated verbs.

Existing formulations of "resultant state" are not adequate

then to capture the essential nature of Purpose Clauses. It is

true nevertheless that some kind of semantic/interpretive

condition is needed to filter these constructions.

8. Constructions related to Purpose Clauses

We have already seen that one necessary criterion for identifying

Purpose Clauses is the matrix Theme's control of an argument (theta

position) in the lower construction. This is not a sufficient condition of

course since many other kinds of sentences conform to such a pattern.

8a. NP + BE + PARTICIPLE as a Matrix to TVP Clauses

BE as a matrix verb to TVP clauses offers some firm

evidence that PCs are merely a special variant of a wider class

of syntactic phenomena. What all such TVP clauses seem to have in

common interpretively is a sensitivity to the control of one or

more of their theta positions by the matrix Theme.

Although I use the term "control" here, it often describes

no Volitional or Causative relationship. It is more like a

coherence factor, defining the likelihood of an actor, object or

event participating in a certain way in two propositions. It is

sufficient in many models to simply define control as co-

indexing. The kind of graduated semantic constraints on co-

indexing found in Peterson's sentences (Section 9) mean that our

view of control in this analysis has to be more delicate.

It is worth taking a little time to explore some examples of

the wider set of TVP sentences. Only in this way is it possible

to see what is unique to Purpose Clauses. Consider the following

sets of sentences, which could not really be called purposive

without diluting the meaning of that term to the point of insignificance.

^25 Pengi is exciting a) e to talk to ei

                                b) for us to talk to ei

                                c) * ei to talk to us.

^26 Pengi is excited a) * e to talk to ei

                               b) * for us to talk to ei

                               c) ei to talk to usi

^27 Pengi is too excited a) e to talk to ei

                                     b) for us to talk to ei

                                     c) ei to talk to us

Sentence sets ^25 and ^26 seem to be in some kind of

complementary distribution, so what is going on ?

Firstly, for the purposes of this exercise, BE is a

syntactic predicator, but probably not part of the semantic

predicate; (eg. refer to Huddlestone 1984:182). My real concern

here is with the structure of propositions, so "predicate" shall

refer to semantic predicates.

Next, it is necessary to consider what the matrix predicate

is predicated of. Williams (1980:208) assumed the antecedent of

all predicates to be a lexical category. We will see that this is

not necessarily so.

In the case of ^25, exciting is a state predicated of some

other actor (possibly the speaker) relative to Peng. In the case

of ^26, excited is an involuntary state predicated of Peng.

Finally in ^27, too excited is a state predicated of Peng, but

the attributing Agent is vague, possibly the speaker or possibly

Peng himself.

Thus since the ^25 sentence set has a matrix predicated of

an exophoric referent, its adjuncts must do likewise. ^25c, which

tries to assert a predication of Peng, is uninterpretable.

Conversely, the ^26 set of sentences is predicated of Peng, so

its adjuncts cannot be predicated of any exophoric referent, as

^26a) & b) would have to be. The vague predication of the ^27 set

of sentences permits all interpretations.

The consequences of these differences in predication are

reflected in the acceptability of the various subordinate

clauses. A common thread seems to be that the matrix predicate of

TVPs must be attributed to the same referent as the predicate of

the second clause. There is a further familiar pattern : it is

always the Theme which controls an empty argument place in the

lower clause. The properties discussed here we have also been

attributing to Purpose Clauses.

The preceding section considered some examples of Resultant

State taken from Bach, and using the verb ready. Such structures

are subject to the same controls on predication as excite, which

does not of course make them PCs :

^28 Unaisii says shei is ready a) * e to talk to ei

                                             b) * for us to talk to ei

                                             c) ei to talk to us.

As with ^26, the adjuncts in ^28 a) & b) are predicated of an

exophoric argument while the matrix predicate is predicated of

the matrix lexical Theme in Subject position : an unacceptable

disjunction. Sentence ^28 c) is no problem of course since the

matrix and adjunctive predicates are predicated of a common

argument (Unaisi) .

8b. Purpose Clauses as a sub-set of TVP Clauses

The adjectival constructions just surveyed had

matrices rather different from PCs, although the control

properties turned out to be rather similar. Even without straying

too far from the notion of "purpose", it seems that some matrix

verbs are rather idiosyncratic in subcategorizing for particular

kinds of "Purpose Clauses" :

^29 We brought himi along [e to talk to ei].

^30 * We asked himi along [e to talk to ei].

^31 We asked himi along [ei to talk to us].

Sentence ^31 above illustrates what appears to be a Purpose

Clause but which can only occur in SPC, not OPC form. This is not

characteristic of other instances of Purpose Clauses : a matrix

verb accepting one kind of PC will normally accept the other.

With the sentences above, some might argue that the TVP in ^29 is

an adjunct whereas the TVP in ^31 is a complement of the matrix

verb. This looks more plausible with ^32 ,^33, ^34 although the

judgment is murky.

Below are sentences containing three more matrix verbs

which seem unable to accept OPC-type adjuncts, although SPC-type

adjuncts are fine. However two of these verbs can scarcely be

conceived to generate "purposive" semantic environments.

Sentences ^32 to ^34 are more examples of syntactic

configurations with which PCs have much in common. The extent to

which grammaticality is constrained by interpretive criteria

within this syntactic commonality is therefore instructive.

^32 We invited him [e to talk to us\*e].

^33 We expected him [e to talk to us\*e].

^34 We wanted him [e to talk to us\*e].

The behaviour of the verbs just mentioned can be clarified

somewhat by exploring their factive implications.

^35 *We brought him along, but he didn't come.

^36 We asked him along, but he didn't come.

It seems that for an adjunct or complement ( such as those found

in ^29 to ^34) to accept the matrix Theme as controlling

Subject/Agent for itself, the effect of the matrix verb on its

own direct Object/Patient must be induced. In other words the

matrix THEME must be Affected so that its "resultant state", as

projected into the theta role of the PC, matches the Intent of

the matrix AGENT. A verb such as invite is not subcategorized to

induce an EFFECT on the Theme. It merely projects a possibility.

Consider ^29. Him is the matrix Theme. The condition of him

is induced by brought along. While this interpretation stands,

him can be projected into an empty theta role (Object position)

of the adjunct verb talk.


The Condition below tries to formulate the semantics succinctly.

Purposive Constructions : Statement IV

CONDITION OF MANIFESTED INTENT

In Purpose Clauses a lexically empty theta role may be
controlled by the matrix Theme if that Theme is induced to
manifest the Intent of the matrix Agent.

coda : Thematic control implicitly imposes a temporal
condition on PCs. The projected action of V2 in a PC must
always be subsequent to the presupposed action of the matrix
verb V1.

The Condition of Manifested Intent successfully excludes

sentences ^30 to ^34 from the typology of standard PCs. It

captures the useful elements of Jones' and Bach's "resultant

states" without the complications. It also assigns a very clear

function to the notion of Agent in the grammar. The status of

being "manifested" entails a factive presupposition for the

matrix (except where it is in the future aspect).

9. Caveats on The Condition of Manifested Intent

Peter Peterson (private communication) has observed that PC-

type constructions such as

^37 *? The housei was painted [e to selli]

are more or less unacceptable despite "a pragmatic inference of

Agency", and despite having Themes manifesting an Induced Effect.

However the difficulty here has less to do with the deleted

passive Agent than with what we might call "the domain of

control" exercised from the matrix clause into the purposive

adjunct. Compare these three sentences :

^38 a) Wei painted the house (in order) [ei to sell it].

^38 b) *? Wei painted the housej [ei to sell ej].

^38 c) Wei built the housej [ei to sell ej]

Sentence ^38a) is a Rationale Clause; ^38c) is an OPC; ^38b) can be neither.

What seems to be going on here is that the matrix Agent in

^38a) and ^38b) exercise only selective Effect on the

Theme/Patient, house, through the matrix verb, paint, but a more

holistic Effect into the adjunct theta position ei through sell.

It is a bit hard to express this notion coherently, but it seems

clear that sell and build Effect the totality of an object in a

way that paint does not. The grammatical consequences of the

difference are intriguing.

Purposive Constructions : Statement V

Where the influence of the Agent exercised through both
verbs is Thematically Coextensive, then the Object of the
second verb may be deleted provided that it (the second
Object) is coreferential with the Object of the first verb.
However, where the Agent's influence is not Coextensive,
then the Object of the second verb has to be represented by
at least a lexical pronoun.

The preceding discussion certainly adds a semantic slant to

so-called Equi-deletion. Thematic coextensiveness is open to

subtle, and perhaps idiosyncratic interpretation, but it is an

effective constraint. Note that it is directly applicable to the

earlier discussion on Resultant States.

A couple of important conclusions seem to derive from these

examples. Firstly, the thematic terms such as Agent and Theme

which we have been dealing with are meta-labels for internally

complex phenomena whose compositional nature may differ

significantly in different environments.

Secondly, the kind of analysis of Purpose Clauses which is attempted

here is not comprehensive. Working at a finer level of delicacy, semantic

constraints of the kind just noted are likely to keep cropping

up. This is also good evidence of course that an entirely

"syntactic" analysis (in the traditional sense) is not likely to

capture these finer distinctions either. Similar phenomena are

endemic in the grammar. Elsewhere (May 1987) I have explored

subtle changes in the value of controlling Themes in other

environments with similar results.

10. Summary of the properties of Purpose Clauses


It may be useful at this stage to summarize some properties of standard Purpose Clauses.

Purposive Constructions : Statement VI

1. Purpose Clauses are a subset of the class of TVPs
    (infinitival verb phrases);eg. compare with sentences ^31- ^36.

2. Purpose Clauses are sentential adjuncts to the matrix sentence.
    See Section 4.

3. Every PC has at least one lexically null NP position.
    OPCs normally have two lexically null NP positions; (that
    is, if they are interpreted as sentential adjuncts in TVP
   constructions).

4. A matrix verb which can accept SPCs can usually also
    accept OPCs, and vice versa, according to most analyses, but
    see sentence ^31.

5. One theta-role of a PC is controlled by the matrix Theme.
     Refer to Section 6, and the discussion on ^18-^20.

6. A Hierarchy of Thematic Control Condition normally
    applies in PCs such that the Theme of the matrix clause
    controls that gap in a PC which is the most nearly optimum
    thematic selection from a hierarchy of Instrument> Patient > Agent;
    (refer Section 6.1; see also Section 17 below).

7. The Theme-controlled theta-role in a regular PC is
     subject to an interpretive Condition of Manifested Intent;
     (refer to Statement IV).

8. The projected action of V2 in a PC is always subsequent
     to the presupposed action on the verb in the matrix; (refer
     Statement IV).

9. In a purposive construction, the ellipsis of the Object
    of the verb in the adjunct is only possible when the matrix
    verb and the adjunctive verb subcategorize for semantically
    co-extensive Themes;(refer to Statement V).

We might add a tenth point, more pragmatically determined :

10. The second lexically null position in an OPC takes the
      matrix Agent as controller by default;(see sub-section 6.1
      for a pragmatic explanation of this). The pragmatic context
      of situation can override this default when appropriate;
      for example, see the discussion of sentences ^18-^20. See
      also the discussion on "purposeful possession", Section 14
      below.

The first part of this thesis has identified the

typical formulations of PCs. As adjuncts they always occur in

association with matrix constructions, but not with all matrix

constructions. It will therefore help to orientate the analysis

if some attempt is made to identify relevant matrix environments.

11. The Classification of Matrix Verb Environments

Because verbs play such pivotal roles in the construction of

strings of natural language, and have as a consequence a

multiplicity of functions, there are many ways in which they can

be classified. For example, Bach, in speaking of "verbs of choice

and use" (Section 16 below) attempts some sort of classification

on the basis of participation in purposive-type constructions.

If  the overall analysis in this thesis shows anything, it must be

that such classification can never extract more than fuzzy

categories. It is with this caveat in mind that I propose a

series of verb environments which vary in their hospitality to

purposive expression. These environments centre on the

subcategorization of matrix verbs. Note that it is likely (if

other lexical behaviour is any guide) that certain lexical verbs

will participate in more than one environment.

11a. ENVIRONMENT I

Environment I matrix verbs subcategorize a matrix Theme

which has extensible reference: the Theme may be supplemented by

a subordinate adjunct. "Extensible reference" means that an

argument of the verb (Theme in this case) has its own, optional,

extended argument set (in this case, an adjunct). This is a way

of referring to the notion of control from another perspective.

The matrix Theme in such constructions will exercise control into

the adjoined construction. There is an implication, in standard

purposive examples of the environment, of Volition-at-Source

(which might not be lexically explicit). Examples of verbs

eligible for Environment I are : bring, buy, choose, use, be,

have, propose, invite, marry, build, want ...

Sample Environment 1 sentences might be :

^39 a) We used a hatchet. (..unextended THEME)

^39 b) We used a hatcheti [ei to strip the saplings] (..extended THEME)


Semantically, the extension of the Theme has the effect of

modifying its condition or scope relative to the matrix verb.

11b. ENVIRONMENT II

Environment II matrix verbs subcategorize a matrix Theme

which has terminal reference. "Terminal reference" means that the

arguments of the matrix verb fully satisfy the unmarked semantic

implications of the verb in a relevant context. This is vague,

but so are semantic parameters.

By way of illustration, we typically ask of an instrumental verb

like use, which has extensible reference : "What for?", ...while no such

query is normally implicit in a verb like eat. Standard purposive examples

in the environment imply Volition at Source (which might not be

lexically explicit).

The matrix of verbs with Terminal Reference may be

supplemented by a conjoined string with an independent structure.

The matrix Theme cannot exercise control into such a conjoined

structure (the Agent might) and any empty category may be of a

PRO form.

An example of such a structure would be a Rationale

Clause. Thematic coreference for these verbs may also be

supplemented by devices such as relativization and prepositional

location.

Examples of verbs eligible to participate in Environment II

are: read, eat, see, come, leave, kill, enjoy (where volition can

be inferred).

Verbs eligible for Environment I are generally eligible for

Environment II also, but the reverse is not true. Sample

Environment 2 sentences might be :

^40 a) Jezebel eats ginger.
        (..THEME with terminal reference)

^40 b) *Jezebel eats gingeri [ei to flavour everything]
         (..THEME not extensible)

^40 c) Jezebeli eats ginger [in order ei to keep the doctor away].
        (..Conjoined supplement to S1)

Environment I really defines (although not sufficiently)

where Purpose Clauses may be found, while Environment II is

congenial to Rationale Clauses.

11c. ENVIRONMENT III

Environment III is a collective description for the bulk of

matrix types which obligatorily take so-called Sentential

Complementizers : THAT, FOR-TO & POSS'-ING. In Environment III

the matrix clause cannot exist independently. The matrix verb

makes obligatory transclausal reference to arguments in the

complement construction.

Where particular verbs are able to operate in simple sentences

(i.e. NP+V+NP) as well as Environment III, then they may have

different meanings for each application. For example:

^41 a) Harry likes that woman.

^41 b) Harry likes that woman to work for him.

Examples of verbs eligible to participate in Environment 3 would

be: like, want, believe, know, order, force, tough...

CHAPTER II : POSSIBLE VARIATIONS ON STANDARD PURPOSE                               CLAUSES

12. Some Specific Purpose Clause Environments

Bach (1982: 38) lists a number of environments to which he

claims PC matrices are restricted, although by the criteria

established in the last section he is not sufficiently

restrictive. It will be productive to examine some of these

environments. Note that I turn them in ways which the author

never intended.

13. HAVE Expressing Obligation

Bach describes typical purposive environments for have and

be as being "... in place, on hand, available, at one's disposal,

in existence." The following examples are Bach's:

^42 Mary has her motheri [to consider ei].

^43 War and Peacei is available [(for X) to read ei] to the students.

If ^43 is purposive (which is debatable) it is certainly no

ordinary PC. Sentence ^43 is comparable in most ways to the

adjectival construction, ^27: the predicator of the matrix and

subordinate clauses is a vague attributing Agent. It is true that

things are usually "available" for a purpose while too excited is

not a volitional condition. In this instance however, the

semantic distinction seems to have no material bearing on the

well-formedness or interpretive felicity of the sentences.

There are good reasons for believing that ^42 is not a

Purpose Clause at all. Recall that a necessary condition for PCs

was that the matrix and subordinate verbs be sequenced in terms

of action. Consider is not subsequent to have, but part of the

same concept. That is, have (X) could almost be called a modal

property of consider.

The have in ^42 thus expresses Mary's obligation, a

traditional modal function. An intriguing facet of this

quasi-modal role for have is that it does affect the syntax of

the subordinate clause by making it more complement-like than

adjunct-like. In this it differs from possessive HAVE.

14. HAVE Expressing Possession

There is a sense of have which seems amenable to an

interpretation of "latent Agency", and that is its possessive

use. A possessor in some way has "command" of an advantage which

may be exploited, and a Purpose Clause can express the nature of

that potential exploitation.

^44 Maryi has her motherj [ ei to talk to ej].

^45 The troops have enough suppliesi [ei to last them through winter].

The objection might be raised that in these sentences also, have

and V2 exhibit no sequence. However it seems to me, on

reflection, that possession does have an ontological priority to

V2 in both ^44 and ^45. Furthermore, the subordinate

constructions are adjuncts, not complements, a syntactic property

evidently stemming from the semantics of the situation. Some

short texts may make this more obvious :

^46 + Can Lisa come ?

        - * No, she has her mother. [OBLIGATION]

^47 + Will the child be cared for ?

        - Yes, he has parents. [POSSESSION]

The one property which is missing here is any surface

expression of active purpose itself. It is normally the Intent

and active Volition, the Agency, of NP1 relative to the adjunct

which leads us to talk of Purpose (and Rationale) Clauses. This

is not to say that a purposive phrase (with quite arbitrary

reference) can't be read into the sentences :

^48 Mary has her mother [serving the purpose of someone] to talk to.

^49 The troops have enough food [for the purpose of] lasting them through winter.

The periphrastic insertions in ^48 and ^49 suggest perhaps that

the Theme of have in these sentences is presupposed to manifest

the Intent of the matrix Subject, which thereby assumes Agentive

overtones. That is, we reinterpret have in this context to imply

something more than possession.

15. "Transitive Verbs Involving Continuance or Change of State"

Bach's description of on type of PC environment (1982: 38)

referred to : "... transitive verbs which involve continuance or

change in the states of affairs indicated in (the matrix) and are

of a "positive" sort...". "Positive" seems to mean that the

action expressed in the adjunct is not contrary to the purpose

implied by the matrix.

As examples he gives sentences such as the following :

^50 We always keep a fire extinguisher in the kitchen [to use e in case of fire].

^51 I got it [to prop up the porch with e ].

Precluded sentences lacking the necessary "positive"

qualities were said to be :

^52 ? I sent him out [for us [to talk to e ]].

^53 ? I keep it out of my office [to manage my students with e].

&n