Single Indefinite Lexical Subjects in English Conversation

Hartwell S. Francis

September 20, 1999

Recently there has been a good deal of interest in linguistics in what Prince (1981a:247) has called the "conspiracy of syntactic constructions." According to this conspiracy, noun phrases that refer to entities that are new to the discourse tend to be kept from occurring in subject position. This conspiracy results from a conflation of the pragmatic status of referents in discourse and the function of subjects in clauses. Subjects are the grammaticization of clause topics (Mithun 1991), but subjects do not necessarily always function as clause topics (Lambrecht 1994). The end result, though, is a close relationship between pragmatic status of topic and syntactic status of subject.

Based on this low occurrence of new information in subject position, scholars propose that a variety of constructions exist in a given language that allow the speaker to structure discourse so that new information does not occur first in an intonation unit (Chafe 1987, 1994) or a clause (Birner & Ward 1998). The questions suggested by the theory that there is a conspiracy of syntactic constructions and by the data have led scholars to study non-canonical word order. For example, Birner & Ward (1998) look at pre-posing, post-posing, and argument reversal, and Prince (1997) and Gregory & Michaelis (1999) examine left-dislocation and topicalization.

The research agenda suggested by these theoretical concerns has come at the expense of canonical word order. The theoretical implications for canonical word order have been left relatively unquestioned. In particular, there has been no large scale corpus-based examination of the constraints purported to apply to subject position in English conversation. There have been studies that look at every entity and its syntactic and pragmatic status (Chafe 1987, 1994; Du Bois 1987; Prince 1992), studies that are the basis of some of the claims made about referring expressions in subject position, but they have been limited to small data sets.

One exception is Francis, Gregory, and Michaelis (to appear). In their study, the matrix subjects of over 30,000 sentences were examined with the goal of documenting the use of referring expressions in subject position and testing some of the constraints on subject position that have been proposed in the literature. Francis et al., however, were mostly interested in whether or not formally definite lexical subject NPs were violations of a specific constraint. They did not closely consider indefinite lexical subject NPs.

In this study I will examine the indefinite lexical subjects identified by Francis et al. (to appear). I will look at the following types of singular indefinite lexical NPs in subject position: a/an + N (see 1.a); indefinite-this + N (see1.b); the singular quantifier one + N (see 1.c); and the plural quantifier some + singular count N (see 1.d).

I will consider these referring expressions with respect to the constraints on new entities in subject position, and I will consider their function in discourse.

In the following section I will review the literature concerning constraints on subject position, pragmatic statuses of indefinite NPs, and discourse functions of indefinite NPs. In section 2, I describe the corpus and discuss five measures I will use to quantify the data. In section 3, I discuss the results of the measures I apply to the data. Section 4 is a more general, speculative discussion of the function of singular indefinite lexical subjects. Finally, in section 5 I consider the theoretical implications of the findings of this study and I suggest some areas for possible future research.

1. Review of the literature

Lexical subjects in and of themselves are uncommon (Chafe 1994, Du Bois 1987, Givón 1983, Lambrecht 1987). In a study of lexical noun phrases in subject position in English conversation, Francis et al. (to appear) found that only nine percent of the subjects of over thirty thousand sentences were lexical noun phrases. Example 2 is typical of the type of lexical subject they found.

Based on similar findings, several constraints on subject position have been proposed.

Chafe (1987) proposes one new piece of information per intonation unit coupled with a light starting point. Chafe (1994) restates the second component of this constraint as a light subject constraint. Du Bois (1987) proposes one new argument per clause and a given transitive subject. Lambrecht (1994) proposes the Principle of Separation of Reference and Role (PSRR) stated as a maxim: "Do not introduce a referent and talk about it in the same clause" (p.185).

All of these constraints seem to fail to prevent the data in 3.

Unless the referents of the four sentences are considered in some sense light, Chafe’s light subject constraint is violated. Consider 3a. The referent introduced in subject position is the topic of this sentence and the topic of the next clause. Example 3b and 3c appear to violate Du Bois’ proposed constraints. Both sentences contain two new referents in transitive clauses. Likewise, all of these examples appear to violate Lambrecht’s principle.

Chafe (1994) and Lambrecht (1994) have addressed the occurrence of indefinite lexical subjects. Example 4 (emphasis added, stress in original) is Chafe’s (1994:91) example 25b:

1. ...this van pulls in there,

Chafe interprets the use of the new, indefinite-this subject of 4 as an iconic representation of the speaker’s surprise: "In effect, he expressed his surprise through the surprising placement of a new, though trivial referent in the subject role" (1994:91). The subject is trivial in the sense that reference to the van does not persist in the ensuing discourse. The subject therefore does not violate the light subject constraint.

Example 5a (emphasis added) is Lambrecht’s (1994:82) example 3.7 and example 5b (emphasis added, stress marked in original with small caps) is Lambrecht’s (p169) example 4.38:

5a. A book is a useful thing to have in a doctor’s waiting room.

1. A boy was run over by a car.

A book in 5a is generic. Generic referring expressions "may be said to have identifiable referents" (Lambrecht 1994:82). Although the book referent is first mentioned in a clause in which something is said about it, its identifiability allows this (Francis et al. to appear). In example 5b, however, a boy refers to a specific boy. In this case the referent is not the topic of the sentence: "Instances of acceptable sentences with brand-new subject referents may also be found in thetic sentences of the event-reporting type" (Lambrecht 1994:169; his small caps). The Principle of Separation of Reference and Role makes predictions about entities in the pragmatic role of topic. It does not apply in this case in which the pragmatic role of topic encompasses more than the grammatical role of subject.

The constraints discussed above do not disallow indefinite subject NPs. Instead, they predict the functions of indefinite subject NPs and the types of indefinite subject NPs that can occur. Chafe’s (1994) constraint predicts that indefinite subject NPs will be available from the previous discourse or unimportant in subsequent discourse. Lambrecht’s (1994) constraint predicts that indefinite subjects will be generic or they will not be the topics of their clauses. Additionally, neither constraint would preclude a previously mentioned entity from occurring in subject position. The predictions can be verified by examining the use of indefinite subjects in conversation.

The pragmatic status of indefinite NPs has been widely discussed in the literature. For the purposes of this paper, the relationship between pragmatic status in terms of accessibility or identifiability and topicality will be discussed.

An indefinite subject NP will only violate the constraints discussed above if its occurrence is the first mention of the entity. Normally, this is the interpretation of an indefinite referring expression. This assumption, however, must be questioned. Gundel, Hedberg, and Zacharski (1993) propose the GHZ Givenness Hierarchy in which the form of the referring expression is determined by the speaker’s assumptions about the knowledge the hearer has of the referent. The GHZ Givenness Hierarchy ranges from pronouns coding information which is in focus to indefinite NPs which are type identifiable. Gundel et al. (1993:276) agree with other scholars that "the different forms serve as processing signals to the addressee."

Gundel et al.(1993), however, argue that the statuses ranked on their Givenness Hierarchy are implicationally related, from in focus down to type identifiable. In other words, an entity that is in focus can be referred to with any form, but a type identifiable entity must be referred to with an indefinite a/an form. This leaves open the possibility that an entity that is referred to with an indefinite form can have any status from in focus down to type identifiable.

The GHZ Givenness Hierarchy correlates with the ranking of familiarity statuses presented by Prince (1981a). Entities which are in focus are likely to have been evoked in the prior discourse. Entities which are only type identifiable are likely to be brand-new. This, in turn, leads to the relationship between form and pragmatic function proposed in the topic acceptability scale presented by Lambrecht (1994:p165). According to the topic acceptability hierarchy, active referents, referents which are in focus, are the most acceptable topics and brand-new unanchored referents, likely to be only type identifiable, are the least accessible topics.

This discussion leads to two salient possibilities for indefinite subject NPs. One, they are topics. Either their referents are referred to as indefinite NPs despite having a givenness status higher than this form generally encodes in order, perhaps, to convey a Gricean implication (Gundel et al. 1993), or they are unlikely topics, perhaps with generic reference. Two, they are not topics. Either the whole sentence is the focus, or the subject argument is the focus and the predicate is the topic. Determining if indefinite subject NPs are generic or anchored will provide evidence of these entities’ pragmatic status.

The generic possibility mentioned above is only a possiblility for a NP determined with the indefinite determiner a/an. The indefinite a/an is ambiguous between specific and non-specific or generic reference (Prince 1981b, Wald 1983). Hawkins (1991:416) finds that "sentences with a are logically neutral to the uniqueness entailments of corresponding sentences with the." Example 6a shows an indefinite subject with generic reference and example 6b shows an indefinite subject with specific reference.

Indefinite-this, on the other hand, is unambiguously specific (Prince 1981b). The use of indefinite-this (or new-this (Wald 1983)) entails the existence of the referent. It is not logically neutral in this sense. The generic possibility does not arise. Furthermore, by definition referents referred to with indefinite-this have not been previously mentioned in the discourse. Thus, indefinite-this subjects are either not topics or they are unconventional topics in terms of the topic acceptability hierarchy outlined above.

The indefinite use of some with a singular referent presents more of a problem. This use of some has not been well studied. It is indefinite, as are the other forms under consideration, based on "the classic test for indefiniteness, occurrence in existential there sentences"(Prince 1981b:233):

7. and there's some other city. (sw_0016)

Although sentence 7 occurs naturally, the referents in subject position referred to with some, as in example 8a, can be put in this construction felicitously, as in the constructed example 8b.

8. a. A143: and, and some kid goes out and gets the kid (sw_0221)

The examples cited here seem to indicate that indefinite-some is used to refer to specific referents. Accordingly, indefinite-some falls in the same group as indefinite-this with respect to pragmatic status as topic.

The form one to indicate indefinite status is fundamentally different from the other forms discussed here. The use of one invokes a larger set (of which the referent of the referring expression containing one is a part) that may have been introduced into the discourse.

Thus, the referent referred to with one and a noun, as an accessible referent, is a much more likely topic. Furthermore, there is no possibility of a generic reading.

Part of the reason this group of referring expressions with one is included is that there is a functional overlap between one and a/an. Example 10 shows this overlap..

In 10, a set of entities, turtles, is entered into the discourse, and one member of this set is singled out for some reason, biting in this case.

Gundel (1988a:210) provides a particularly clear definition of topic status:

This definition of topic makes clear that the topic role is in principle distinct from the discourse (givenness or familiarity) status of a referent. As Lambrecht & Michaelis (1998) argue, evoked status does not entail topic status (pronouns, both deictic and anaphoric, may be foci) and topic status does not entail evoked status (a referent may be established in the topic role in the very act of commenting about it). Lambrecht’s (1994) generic subject sentence 5a above and example 10 demonstrate this conception of topic status.

11. B70: and, uh, uh, a girl is very different.

In example 11, the referent is entered for the first time into a relation in which the

speaker is providing more information with respect to the referent. In this example, the speaker maintains the referent as the topic with another full indefinite subject NP.

As the "peg on which the message is hung" (Halliday 1970:161), though, a topic should be relatively stationary, i.e., predictable. This is the basis of the topic acceptability scale discussed above. By their very nature, and in particular indefinite-this and –some, singular indefinite NPs are not predictable. As seen in example 11, they may be acceptable as topics due to generic reference or, as discussed above, these entities may be part of a set that has been previously mentioned.

Whether or not the indefinite subject NPs are topics, however, is an open question. There is not a one-to-one mapping between grammatical function of subject and the role of topic (Givón 1983a, Gundel 1988b, Lambrecht 1994). Subjects may instead be focal. A lexical subject may be a narrow, or argument, focus or it may be the subject of a thetic or, equivalently, sentence focus sentence (Kuroda 1972, Lambrecht 1994).

Argument focus sentences express pragmatically presupposed open propositions (Jackendoff 1972), as in example 12.

In the series of clauses following the first sentence, the subject NPs clearly identify the variable in a presupposed open proposition ‘Someone caught fish’ (x=my daughter, his daughter, him).

Rather than identifying a variable in an open proposition, sentence focus sentences present entities and/or report states of affairs. Lambrecht & Michaelis (1998:496) define the function of sentence focus to be "introducing a new discourse referent or expressing an event involving such a referent." There is no topic presupposition and the entire proposition is asserted. Example 12 demonstrates sentence focus.

Prior to this utterance, the discussion concerned drugs and crime. Example 13 is typical of sentence focus constructions in that the verb is an intransitive verb and a verb encoding the appearance of an entity on the scene.

If sentence focus constructions are a significant source of lexical coding of subjects in the data, these lexical subjects will be correlated with intransitive predicates, since sentence focus sentences tend strongly to contain unaccusative verbs (Lambrecht 1994). Further, since unaccusative verbs select for undergoer-type subjects, sentence-focus sentences tend strongly to have patient subjects. Measures of these factors will indicate if indefinite subject NPs are topics.

Chafe (1994) contends that indefinite lexical subjects are trivial based on the finding that the entities they introduce do not persist in the discourse. His choice of example (example 4, repeated here) puts him directly at odds with other researchers.

4. ...this van pulls in there,

Indefinite-this is described in the literature as a mechanism for introducing important elements into the discourse. Prince (1981b) comes to the conclusion the function of indefinite-this is to introduce discourse topics based on the finding that 86 percent of the referents introduced by indefinite-this in a corpus of narratives were referred to again within a few clauses. Empirically, Gernsbacher and Shroyer (1989) report that entities introduced with indefinite-this are significantly more likely to be referred to again than entities introduced with a/an. Do indefinite subject NPs introduce important referents?

One way to measure this is by examining the type of shift that occurs after an entity occurs in subject position in the discourse. According to centering theory (Walker, Joshi, & Prince 1998), the basis of a model of which entities are at the center of attention of the participants in a discourse, an entity in subject position is the preferred forward looking entity. A subject can be described as a catatphoric device as discussed by Gernsbacher & Shroyer (1989:536): a device that marks "concepts that might play a pivotal role in the upcoming discourse."

In centering theory, transition states are ranked according to how cohesive the transition makes the discourse. The highest ranked transition is continue: "the speaker has been talking about a particular entity and indicates an intention to continue talking about that entity" (Walker & Prince 1996). This transition occurs when an anaphor with reference to a subject entity in the previous utterance is realized as the subject of the present utterance. If the referents of indefinite NP subjects continue to be referred to in subject position of the following clause, these referents are established as the center of attention. According to this measure of a referents importance, do indefinite subject NPs introduce important referents?

I used the data from Francis et al. (to appear) for this study. Francis et al. examined subjects from a subset of the Switchboard corpus of English telephone conversations (Godfrey et al. 1992). The Switchboard corpus is composed of approximately 2,400 telephone conversations between unacquainted adults. The participants in the conversations vary in age and represent all major dialect groups. From this corpus, the 400 conversations that were syntactically parsed (Marcus et al. 1993) were examined. A total of 31,021 subjects of declarative sentences were collected. Of these, 91 percent are pronouns and only 9 percent are lexical NPs.

The lexical subject NPs in the data are overwhelmingly definite. Only 55 of the lexical subject NPs contain the indefinite determiner a/an. There were 35 instances of lexical subject NPs containing one, 14 instances of lexical subject NPs containing indefinite-this, and only 5 instances of lexical subject NPs containing indefinite-some with a singular NP. In total, this represents 0.4 percent of the total number of subjects in the corpus or 4 percent of the total number of lexical subjects.

The NPs containing a/an and one were taken directly from the Francis et al. data. The NPs containing indefinite-this were collected based on the determination that there had been no previous mention of the referent. The NPs containing indefinite-some were collected on the basis of the occurrence of a singular count head noun.

Based on the preceding discussion I develop five measures to quantify the data. I code for previous mention, anchoring, generic reference, transitivity, and immediacy of reference.

Partial answers to the questions whether or not the indefinite subject NPs are topics and whether or not they are trivially new depends on the determination of their information status. Are the referents new to the discourse?

To answer this question, I looked at the previous discourse. Referents which have been mentioned in the previous discourse were coded as not new. The second mention of a small bowl in 13 is an example of an indefinite subject NP coded not new.

Referents which are part of a set of referents that has been mentioned in the previous discourse are coded as not new. Example 9, repeated here, is an example of an indefinite subject NP coded as not new.

All other referents are coded as new.

Partial answers to the same questions depend on the extent to which entities are new. Prince (1981a) distinguishes between brand-new and brand-new anchored. Francis et al. (to appear) argue that brand-new anchored referents make acceptable sentence topics. Furthermore, this type of referent is considered light in Chafe’s (1994) terms "because it is directly associated with information that is or was fully active" (p87). Indefinite NPs containing relative clauses (15a) and prepositional phrases (15b) were coded as anchored.

Fraurud (1990:430) raises just this question, "one might ask whether the interpretation of indefinite NPs also may involve anchoring," in the conclusion of a study of the nature of new definite NPs.

The nature of the indefinite referring expression must also be examined. As noted above, indefinites with a/an may be generic. This is a level of reference that is acceptable for new entities in topic status. Indefinite this, some, and one cannot code generic status as the application of a substitution test shows. In 16b the replacement of some with any results in a completely different meaning.

16. a. A17: Y-, you know, some health related thing is going to blow me away.(sw_0140)

b. Any health related thing is going to blow me away.

On the other hand, in 17b replacement of a with any doesn’t significantly change the meaning of the sentence.

Generic indefinite NPs were identified and coded as such through the use of the any substitution test in combination with an examination of the tense and aspect of the verb and subsequent reference to the referent.

Transitivity should give some indication as to whether or not an entity is in a topic relationship or a focus relationship. The correlation between the occurrence of unaccusative predicates and sentence focus suggests that indefinite NPs will be the subjects of intransitive sentences more frequently than they are the subjects of transitive sentences if they tend to be in sentence focus constructions. The predicates of the indefinite subject NPs will be compared with a reference group of predicates of a random sample of third person singular subjects from the same corpus.

Immediacy of reference will measure the importance of the indefinite NPs according to the hierarchy of shift types proposed in centering theory. As subjects, the indefinite NPs are the preferred forward looking entities in their clauses. If these entities are referred to again in subject position in the following clause the result will be a state in which the preferred center of the clause is the backward center of the clause. That is, the referent of the indefinite referring expression will be the subject of consecutive clauses. This continue transition is the highest ranked transition and it is an indication of cohesive discourse. The indefinite NPs will be coded according to whether they are referred to in subject position in the subsequent clause. The results will be compared to a random sample of third person singular subjects and to a random sample of indefinite a/an objects

All the indefinite subject NPs were examined to determine if they had been previously mentioned. The results are presented in Table 1.

Table 1. Previous mention of indefinite subject NPs.

The occurrence of zero previously mentioned indefinite-this tokens is not surprising given the procedure for identifying indefinite-this required no previous mention. In total, sixteen percent of the indefinite subject NPs have been previously mentioned.

The high number of previously mentioned entities referred to with a/an requires some discussion. Twenty-five percent of the referents in this category had been mentioned before the use of an indefinite referring expression. Of these referents, twelve are generic. Example 18 is typical of this set.

Of the remaining two previously mentioned referents referred to with a/an one seems to function as a thematic boundary. I will discuss this example below. The remaining token is similar to the previously mentioned referents referred to with one.

As discussed above, referents determined to be members of a set of similar referents previously mentioned in the discourse are counted as previously mentioned. In the explanation above, example 10, containing a turtle bit me, is an example of this type of previous mention. All 7of the previous mentions with one are of this type.

All indefinite subjects were examined to determine if the NP contains a relative clause or a prepositional phrase. The results are shown in Table 2.

Table 2. Indefinite subjects with relative clause and prepositional phrase anchors.

The high number of referents introduced with one is due to a construction for introducing propositions into the discourse. This construction one thing + relative clause + copula, shown in example 19, accounts for 21 of the 28 tokens.

Three of the remaining tokens are also previously mentioned. The relative clause serves to pick them out of their set as in example 20.

The prepositional phrase of mine occurred in 9 of the 13 tokens of indefinite a/an subjects containing a prepositional phrase. In 8 of the 9 occurrences the token was a friend of mine. In the remaining case it was a colleague of mine. The one case of indefinite-this with a prepositional phrase was this friend of mine.

One of the other three tokens was a variant of the one thing construction discussed above. The remaining two tokens contained a locative prepositional phrase. Example 21 is one of these two.

Except for the three indefinite one NPs with relative clauses, none of the referents of subjects with relative clauses or prepositional phrases has been previously mentioned. A scale of indefinite NPs begins to emerge: 21 are previously mentioned, 43 have not been previously mentioned but are anchored, and 45 appear to be brand new.

As discussed above, only the indefinite a/an subject NP category can refer to a generic. In this category, 29 of the 55 entities are generic. As mentioned above in the discussion of the previous mention result, 12 of these have been previously mentioned. One of the generics is modified by a locative prepositional phrase. There are 16 unanchored first mention generic referring expressions of the indefinite a/an type.

Of these 16, 2 are tautologies as in example 22a, 2 are passive, and 7 have predicates with will or would as in 22c. Example 22d, with a stative predicate, is typical of the remaining 5 tokens of the category.

Generic indefinite NPs comprise 16 of the 45 brand new indefinite subject NPs. In the corpus of over thirty thousand sentences there are 29 brand-new, unanchored indefinite subjects that refer to specific referents. Of these 29 brand-new unanchored referents, 10 are a/an, 13 are this, 3 are one, and 3 are some.

The indefinite subject NPs were examined to determine the distribution of transitive, copula, and intransitive predicates. The results of this examination are presented in Table 3.

Table 3. Transitivity for indefinite subject NPs.

The high occurrence of copula predicates in the one category has been discussed above.

A comparison of the distribution of the total number of predicate types for indefinite subject NPs with the distribution of predicate types for a sample of third person singular pronominal subjects indicates the distributions are not significantly different (Chi sq. (2, 136) = 0.1327, n.s.). From this comparison it appears that the predicates of indefinite subject NPs are not significantly more likely to be intransitive.

An examination of the verbs in question reveals that none of them are unaccusative constructions with an theme as the single argument of a transitive verb. There are several passive constructions (example 23a) and there appears to be a prevalence of verbs of appearance (example 23b).

There were no passive constructions and no verbs encoding appearance in the random sample of sentences with third person singular pronominal subjects.

The sentence following the sentence in which an indefinite subject NP occurred was examined to determine if the referent of the indefinite NP was maintained in subject position. The difficulties inherent in tracing reference to a proposition led to the exclusion of the 21 one thing constructions that introduced propositions. The results are shown in Table 4.

Table 4. Results of the measure of immediate reference.

A comparison of the total distribution of immediate and non-immediate mentions of referents introduced with indefinite subjects with the distribution of immediate and non-immediate mentions of a third person singular subject indicates that there is no significant difference in the distribution (chi sq. (1, 115) = 0.5352, n.s.). A comparison of the total distribution of immediate and non-immediate mentions of referents introduced with indefinite subjects with the distribution of immediate and non-immediate mentions of indefinite a/an objects indicates that there is no significant difference in the distribution (chi sq. (1,113) = 0.0340, n.s.). These indefinite subject NPs are no more important than similar types of referring expressions in object position or than other types of referring expressions in subject position.

Very few indefinite subject NPs refer to brand-new unanchored referents. In total, 45 of these subjects were found in the corpus. More than a third of these were generic referring expressions. Only 29 tokens in the corpus were identified as indefinite subject NPs referring to brand-new unanchored specific referents. Except for this small group, most of the indefinite subject NPs are topics based on the measures.

When the indefinite subject NPs are taken as a group and compared with third person singular subjects, there is no difference in the distribution of transitive, copula, and intransitive predicates. The same comparison indicates there is no difference in immediacy of reference. Furthermore, when the indefinite subject NPs are compared with indefinite a/an object NPs, there is no difference in immediacy of reference.

Through the measures applied to the data several distinct groups of indefinite subject NPs can be identified. These groups include the generic referring expressions, the NPs containing the prepositional phrase of mine, the one thing construction, and the brand-new unanchored specific referents. The heterogeneous nature of the set of indefinite subject NPs renders the measures of transitivity and immediacy of reference problematic. In this section, I discuss the groups that have been identified in general and with respect to the measures of transitivity and immediacy of reference.

All of the generic indefinite NPs are clause topics based on the definition of topic given above. The indefinite subject NP referent occurs in relation to a predicate that increases the addressee’s knowledge about that referent. Example 24 illustrates this relationship.

In this example, the speaker wishes to increase the addressee’s knowledge about the generic referent.

Example 24 also accounts for two tokens of the immediacy of reference measure. The referent is counted twice due to the possibility of repeating a generic referring expression. This presents a problem for the measure because it counts the same referent as a non-trivial referent twice. In addition the measure does not adequately define the nature of generic referring expressions. Some of the referents turn out to be important by this measure and some like example 25 do not.

25. A5: or if it 's at the beach, the beach is great. A pool is great. And, I,

B6: Where, where do you usually go?(sw_0227)

Immediacy of reference does not show that generic subject NPs introduce important entities into the discourse because by this measure some of the entities are important and some are not. The function of a generic subject NP does not seem to be to introduce important referents into the discourse.

The referents of the referring expression a friend of mine in subject position are generally embedded in a larger discourse topic. The discourses are not about these friends. Even in the two tokens in which the referent is immediately referred to again in subject position the entities are introduced as a demonstration that the point of the discourse is important. In example 26, the friend referent is introduced as part of an indication of how serious a golfer the speaker is.

This example suggests that immediacy of reference fails as a measure to distinguish important discourse referents. All 9 examples work this way. The friend entity introduced with an indefinite NP with a prepositional phrase serves as a participant in a part of a discourse segment that demonstrates the importance or significance of a proposition that is under consideration. The prepositional phrase seems to be unnecessary as example 27 indicates.

Here there is the discourse segment ending series of evaluations of a previous segment in B76 and the first clause of A77. The introduction of the new discourse topic chicken enchiladas is based on the friend entity which does not have any other importance in the discourse. It serves only to identify how the speaker has come to be in this situation. 4.3 The one thing construction

The one thing construction skews the relative clause and prepositional phrase distribution and the transitivity distribution. Tokens from the indefinite one NP category account for 28 of the 31 relative clauses and 29 of the 44 copula constructions. The description of indefinite one also indicates that it is different from the other categories of indefinites: indefinite one evokes a set of entities.

Of the 35 indefinite one subject NPs 20 are the one thing + relative clause + copula type of construction. Example 28 demonstrates this construction.

Example 28 gives some indication that the one thing construction is grammaticized. One thing seems to be semantically bleached. The speaker seems to like a number of things about fabric painting. The mismatch in agreement between the subject and the copula in 29 also indicates some degree of grammaticization.

In this type of construction the indefinite one thing is invariably new to the discourse. The referent of one thing is part of the set evoked in the relative clause. Things I thought about and things I like drawn from 28 and 29 are typical examples of these sets. The predicate is often a proposition as in 28 but it can be an entity as in 29.

The construction seems to function to introduce propositions and entities. The one thing subjects are clause topics. The construction is a conspirator in the conspiracy of constructions to keep new information out of the pragmatic role of topic. Immediacy of reference is a poor test of the importance of a proposition given the difficulties inherent in tracing reference to a proposition in discourse.

There are 29 brand-new unanchored specific referents. Several are related to the anchored referents discussed above. Of these, one is one thing is and one is a good example would be. Both of these exemplify strategies for keeping new information out of topic role. They seem to be referring expressions in which an anchor has been omitted due to its obviousness. Similarly, examples 30a and 30b seem to be anchored to the speaker although the anchor is not specifically mentioned.

Excluding these indefinite subject NPs, there are 25 brand-new unanchored specific referents.

Of these 25, 11 encode the appearance of the referent or the existence of the referent. Example 31a is typical of indefinite-this introducing a referent, and 31b is typical of the encoding of the existence of a referent.

Of the remaining 14 referents in this category, 4 are passive constructions. Example 32 is typical of these 4.

32. B108: How did this work? A mill rat had been bought.(sw_0338)

Of the remaining 10, 8 are transitive constructions like example 33.

Some of the brand-new unanchored indefinite subjects function to introduce new discourse segments. In example 32, the introduction of the mill rat coincides with the thematic boundary (Tomlin 1987) of a narrative of how the speaker came to own a rat. Evidence for this interpretation of this construction comes from example 34.

In this example, the referent is introduced via an existential there construction and referred to with a proper name. Nevertheless, at the start of the narrative concerning what happened to the girl, the speaker refers to her with an indefinite unanchored expression in subject position. The use of an indefinite referring expression when an expression higher on the GHZ Givenness Hierarchy could have been used suggests that the speaker is signaling a thematic boundary with the use of the indefinite expression.

In this paper, I have described the indefinite lexical subject NPs in a large corpus of spoken English. The results of this description are several categories of indefinite subject NPs. Indefinite subject NPs can be generic, anchored with relative clauses or prepositional phrases to the discourse, or brand-new. Generic and anchored indefinite NPs may refer to previously mentioned entities or sets of entities in the discourse.

I have tried to explain the function of indefinite lexical subject NPs by taking a measure of transitivity and a measure of immediacy of reference for these subjects and comparing them to third person singular subjects and to indefinite a/an objects. The comparisons resulted in no significant differences between the groups. Indefinite subject NPs, by these measures, do not appear to be topics more often nor do they appear to be more important in the discourse.

Nevertheless, the heterogeneity of the indefinite lexical subject NP group leads to the conclusion that the measures should have been applied to the subgroups of indefinite subjects. In particular, the group of brand-new indefinite subject NPs with specific reference seems to exhibit a high incidence of focus constructions. There were more constructions of appearance of entities and more passive constructions in this group.

Furthermore, based in part on the evidence of a previously mentioned entity referred to in subject position with an indefinite lexical referring expression, the group of brand-new specific entities seems to have a thematic boundary marking function. Despite the low number of this type of referent, this group promises to be an interesting area for further study.

In addition, the one thing construction that is identified in this study seems to be an additional means for keeping new referents from filling the pragmatic role of topic. This construction introduces entities and propositions into the discourse. It too may have a thematic boundary marking function, although this function has not been discussed in this paper. Also, the introduction of propositions and reference to proposition is a promising area for future research.