The pivot function .1 The concept

In the previous chapter, we examined the concept of subject from the perspec-tive of argumenthood, and concluded that the subject is the most prominent core argument of the verb,g f. We saw that some subject properties, specifi-cally those that are shared by uniform-subject languages and mixed-subject lan-guages (Type 1 properties), are explained by this view of subjecthood. These are properties which are based in one way or another on the relational hierarchy of argument functions: the alignment of the relational hierarchy with other hierar-chies (agenthood, topichood), the specification of properties of arguments by the head verb (null pronominals, imperative addressee), and anaphora (anaphoric prominence, switch-reference).

However, we still need to account for the Type 2 properties, the ones that differ in uniform-subject and mixed-subject languages. These properties are the following.

(1) a. Shared argument in coordinated clauses Controlled argument (PRO) (in some languages) Raising

Extraction properties b. Obligatory element

Definiteness or wide scope

“External” structural position

These “subject properties” differ from the ones discussed in the previous chap-ter. The properties ofg f are the result of the status ofg f as an argument in hierarchical relation with other arguments; they are relative properties which are, in some languages, shared with other arguments. The properties in (1) are related neither to argumenthood nor to hierarchies. They have nothing to do with hierarchies because they are unique properties of a single distinguished element in the clause. They have nothing to do with argumenthood because 73

74 Subjects and their properties

they are not properties that relate the “subject” to a head that selects it. We therefore would not expect theg f function to result in these properties; they must be the consequence of a different grammatical function. The fact that these properties characterize a different element from the argument-related properties in ergative and Philippine-type languages reinforces the conclusion that these properties do not follow from the nature of the functiong f.

We propose that the Type 2 properties are associated with a grammatical function which we call piv (pivot), loosely following Foley and Van Valin (1984) and Dixon (1979, 1994).The familiar concept of subject in uniform-subject languages is thus an amalgam of two distinct grammatical functions:

g f and piv. The realization that there is more to subjecthood than argumenthood has led some researchers in LFG (such as Bresnan 2001) to cross-classify the s u b jfunction as a grammaticized discourse function, but the Type 2 properties are no more discourse-related than they are argument-related. We therefore do not consider piv to be a grammaticized discourse function. We need to take a closer look at the piv-related properties to determine the nature of the piv function.

We begin our discussion of the piv function by considering the properties in (1a), which we take to be the core properties of piv. These properties relate to the sharing of a single element by more than one clause. In the coordination construction in question, an argument is shared by the coordinate clauses. In control and raising constructions, the main clause and subordinate clause share an argument. Since extraction is often long-distance, cross-clausal sharing of an element is often a factor in extraction constructions as well. These properties are inherently non-local, and lead to the conclusion that the piv function is the function of cross-clausal connections, or cross-clausal continuity.

(2) The piv is the element with the function of connecting its clause to other clauses in the sentence.

This function is unrelated to questions of argument realization. It thus contrasts sharply with the g f function discussed in the previous chapter, and is not inherently related to it.

We will have less to say about the properties in (1b), which we take to be secondary properties. Unlike the (1a) properties, these properties do not relate elements of different clauses. However, they are similar to those other piv properties in that they are not related to argument hierarchies either. Instead, they seem to be based on the notion that the piv is a distinguished element of the clause, with properties beyond being in a particular position on the relational hierarchy. There is also a topic-like quality to some of these properties – in

Pivot 75 particular, definiteness and wide scope. We will discuss these properties briefly later.

In order to understand the piv function better, we begin by noting that the grammatical functions generally assumed in theories like LFG (as in, for exam-ple, Bresnan 2001) can be divided into three groups:

(3) a. Argument functions: local, selected by predicate



b. Adjunct functions: local, not selected by predicate a d j , x a d j, etc.

c. Grammaticized discourse functions: not local, related to discourse t o p i c

f o c u s etc.

Of these, the argument and adjunct functions are local in their scope – they function to express relations within their clause, and they are locally licensed.

Argument functions are licensed by being selected, and adjuncts by modi-fying meaningful elements. The grammaticized discourse functions (focus, t o p i c, etc.), on the other hand, relate otherwise-licensed elements to the larger discourse within which they are embedded. That is to say, all elements are locally licensed,¹ but an argument (or adjunct) can be assigned an additional, not locally relevant, function. This is reflected in LFG’s Extended Coherence Condition (and in transformational notions such as D(eep) structure and Merge atθposition, which give argument “positions” a special status). This property of the grammaticized discourse functions is captured particularly well termi-nologically in RG, where these functions are referred to as overlay functions (or relations). We will follow the RG terminology here.

Something is missing from this set of relations expressed by grammatical functions. We have grammatical functions that are local to the clause in which they are located and grammatical functions that relate a clause to the larger discourse. What we seem not to have is a function expressing the relation between elements of a clause and the sentence (i.e. larger syntactic structure) of which it is a part. It is this gap that we propose to close with the function piv. The

1 A possible exception to this can be found in a subset of what are sometimes known as topic-oriented languages. We will discuss this briefly in Chapter 6.

76 Subjects and their properties

p i vis a kind of sentence-internal topic.²Just as a discourse topic (represented syntactically in many languages as the grammatical function topic) identifies a single participant as the common thread running through a discourse, the piv is the common thread running through clauses that make up a sentence. Every clause in a syntactic structure (sentence) will have a piv.

As we conceive of it here, piv is an overlay function, but crucially not a discourse function. There is nothing inherently discourse-related about the p i v. It relates exclusively to syntactic properties. In this sense it is sui generis, although (as an overlay function) it is related to the discourse functions.

3.1.2 Formalization: the Pivot Condition

In a formal theory like LFG, the idea that piv is the function of syntactic cross-clausal continuity needs to be expressed in terms of the technical concepts of the framework. It is the role of the formalism to provide a precise expression of intuitions of linguistic analysis. This formal instantiation will play a major role in our understanding of the properties of piv.

As we saw in Chapter 1, the major formal tool for expressing relations between elements in LFG is the functional constraint, annotated to phrase struc-ture rules or encoded in the lexicon. It was noted in passing in Chapter 1 that these functional constraints designate paths through the f-structure. To take an example from the previous chapter, if a verb includes the information that its object is a (covert) pronoun (that is to say, the obj has the attribute pred with the value ‘pro’), this is expressed formally through the following constraint in the verb’s lexical entry.

(4) (↑ obj pred)= ‘pro’

The parenthesized expression on the left side of this equation defines a path through the f-structure, where ‘↑’ represents the local f-structure where the path begins:

2 In class lectures on this material, I have anthropomorphized the concept and referred to the piv as the clause’s ambassador to the rest of the sentence. I think that this metaphor actually goes a long way towards explaining the concept and some of its consequences.

Pivot 77 In early LFG (Kaplan and Bresnan 1982) it was proposed that such paths be limited to a length of 2, by what was called the Functional Locality Condition.

This idea was subsequently abandoned with the advent of the formalism of functional uncertainty (Kaplan and Zaenen 1989) for licensing long-distance dependency constructions.³The abandonment of the Functional Locality Con-dition, justified though it was, has left LFG with no formal expression of the intuitive idea that arguments are beholden exclusively to the predicates of which they are arguments. The piv function allows us to return to the intuition that the theory needs to express this.

The core of Kaplan and Bresnan’s Functional Locality Condition is the idea that a functional expression should not be allowed to directly specify properties of an argument function in a lower or coordinate clause. As suggested in the previous paragraph, this follows from the nature of argumenthood. Arguments are selected by their local predicates. As we saw in Chapter 2, the properties of arguments can be determined by their local predicates. Arguments are strictly local in their scope. A formal theory based on grammatical functions should express this.

The piv function is not an argument function, and therefore is not local in its scope. It is an overlay function, a second function assigned to a locally licensed element. Assigning the piv function to an element which bears an argument function provides a formal escape hatch to the locality of arguments: it allows higher clauses to specify information about it. We propose to formally restrict functional designations in such a way that the only way to refer to a function in a lower or coordinate clause is through the function piv. We refer to this as the Pivot Condition.

The Pivot Condition needs to constrain two types of paths: the path inward from a superordinate argument domain to a subordinate one (argument or adjunct), and the path from one conjunct of a coordinate structure to the other.

The former case can be shown schematically as follows:

(5)

A functional constraint associated with f cannot refer to a non-piv function inside g: it cannot assert its identity to an element in f or specify any features for it. It is crucial that g be a distinct predicate-argument domain; we do not want

3 We will discuss long-distance dependencies in Chapter 4.

78 Subjects and their properties

to rule out reference to, say, the object of a non-predicative “Case-marking”

preposition by a designator such as (↑ oblθo b j). The formal statement will therefore have to distinguish argument-taking preds. The second kind of path is illustrated by the following f-structure and corresponding c-structure:

(6)

Here, the restriction will be against a constraint associated with h referencing a non-piv element of i, and against a constraint associated with i referencing a non-piv element of h. We can think of a path from h into i or from i into h informally as a sideways path.⁴ Formally, we want to restrict the form of a path stated in terms of “f-structure element corresponding to the right sister” –

(∗>) – or “f-structure element corresponding to the left sister”– (< ∗). We include both an informal version of the Pivot Condition and a formal version.

(7) The Pivot Condition Informal statement

A path inward through f-structure into another predicate-argument domain or sideways into a coordinate f-structure must terminate in the function piv.

Formal statement⁵

In a functional designation of the form (↑ . . . α . . .  γ) where α or ((<*. . . γ)) or ((*> . . .  γ)), if  is a (→ pred arg1)’

grammatical function and eitherγ= /0 or γ is a feature,  = piv.

The Pivot Condition is the formal statement of the functional role of piv. It plays a major role in pivot properties, because it restricts reference from one clause to a lower (or coordinate) clause to the piv of the lower clause.