CHAPTER FIVE: ANAPHORIC CONFIGURATIONS

1.Introduction

Anaphoric configurations have been one of the most-studied areas in Generative Grammar. Despite this, there seems to be little consensus that conclusive results have been attained here. The present chapter approaches the problems from a Cognitive/Functional perspective, in the hope that this perspective will be more fruitful than the Autonomist one has been hitherto -- as well as being better justified on metatheoretical grounds.

The usual Generative approach has been to treat Anaphoric Configurations as essentially a syntactic, i.e. intra­sentential, matter. However, many authors also make the point that:

anaphora is basically a discourse phenomenon and can occur between sentences, and in some cases, with no linguistic context present at all (Solan: 1983,36).

Though the present approach can be labelled Cognitive/Functional, it differs from most other Functionalist approaches, in that it does not accept as given a primary distinction between Sentence and Discourse (i.e. C/P 10). Instead, it takes a notion from Bever (1970) -- but conceived independently, as it happens -- and argues that Anaphora is essentially a unitary phenomenon -- whether operating across sentence-boundaries, or not. Indeed, the very focus of my attention is on the interface between Sentence and Discourse. Thus I will not primarily be concerned with Functionalist theories of Discourse Anaphora, though a preliminary discussion of some of them seems to me to be appropriate.

Functional approaches to Anaphora usually do not aim to account for what they assume to be intrasentential, Syntactic processes. They can be divided, roughly, into (a) those that seek to find features in the text/discourse to which Anaphoric Configurations can be shown to relate; and (b) those that take a more directly Cognitive approach, in that they seek to explain textual/discourse configurations (including Anaphoric ones) as being transparently related to mental factors such as "attention".

Fox (1987) and (1987a) are examples of the first kind of Functional approach. As far as conversational English is concerned, she assumes the existence of a kind of dialectic between the context which determines lexical choice (e.g. choice of Anaphoric form) and the influence that such a choice has on the current/subsequent conversational context:


1 Anaphoric form X is the "unmarked" form for a context like the one the participant is in now.

2. By using anaphoric form X, then, the participant displays a belief that the context is of a particular sort. 3. If the participant displays a belief that the context is of a particular sort, than the other parties may change their beliefs about the nature of the context to be in accord with the belief displayed (Fox 1987, 16).


Rejecting the idea (as in Givon 1983, for example) that Anaphoric Configurations can be usefully explained by referring to the factor of distance from the last mention of the relevant referent, Fox (1987) proposes that the fundamental pattern is as follows:


1 The first mention of a referent in a sequence is done with a full NP.

2 After the first mention of a referent, a pronoun is used to display an understanding of the sequence as not yet closed.

3 A full NP is used to display an understanding of the preceding sequence containing other mentions of the same referent as closed (ibid, 18-19).


Tomlin (1987a) is an example of the more Cognitive of the Functionalist approaches to Discourse Anaphora. He argues that "during the on-line task of discourse production subjects ... use a full noun to reinstate reference across an episode boundary, and they will use a pronoun to maintain reference within a particular episode" (p.457). He specifically excludes ambiguity resolution from the ambit of his study.


The basic claim ... is that the alternative use of a noun or pronoun in discourse production is a function of attention allocation by the speaker. During the on-line process of discourse production, the speaker uses a pronoun to maintain reference as long as attention is sustained on that referent. Whenever attention focus is disrupted, the speaker reinstates reference with a full noun, no matter how few clauses intervene between subsequent references (ibid, 458).


Tomlin (1987a)'s experimental results seem persuasive. It is particularly noteworthy that he established episode­boundaries independently and non-linguistically -- thus avoiding any hint of circularity in his reasoning. He even raises the possibility that "discourse units, like the paragraph, are more likely to be the artifacts of linguistic analysis than they are cognitive units utilized by speakers in discourse production" (ibid, 475).



2. Some Generative approaches


Since Langacker (1969), the Generative literature has concentrated its attention on why coreference is (said to be) impossible between nouns and pronouns in configurations like:


(1) She thinks that Jean should get lost.

The general principle that successive Generative writers have been concerned to revise and refine is broadly along the lines that:


If an NP serves as the antecedent of a definite pronoun to its left, the pronoun must be at least as deeply embedded as the NP (Wasow: 1979,56).


I will not here review the lengthy history of the debate within the Generative tradition as to the correct formulation of this principle. I will confine myself to examining some relatively recent versions, which are largely motivated by the desire to avoid deficiencies perceived in earlier formulations. As I differ from Generativists both as to data (Comprehensibility Intuitions, as opposed to Descriptive Grammaticality Intuitions), and as to Metatheory (Cognitive vs Autonomist Linguistics), it is not particularly relevant, from my point of view, to speculate as to whether Generativists will ever devise a formulation which will satisfy the majority of their School. However, as discussed in Chapter One, the quasi-metaphysical nature of Autonomist Linguistics (particularly in relation to the vagaries of intuitionising) is such that Autonomists are unlikely ever to agree on the data, let alone on how to account for it.

For me, the question of "depth of embedding" will not be an issue, as I do not regard the phenomenon as "syntactic", in any traditional sense of that term. As will become clear, I regard the matter as one of communicative cooperation between speaker/writer and hearer/reader.

A good starting-point is the notion of C-Command, as in Reinhart (1976):


(2) Node A c(onstituent)-commands Node B if the first branching node 1 dominating A either dominates B or is immediately dominated by a node 2 which dominates B, and is of the same category type as 1.


(3) Two NPs in a non strict reflexive environment can be coreferential just in case if either is in the domain of the other (i.e. is c-commanded by the other -- P.Z.), the one in the domain is a pronoun (ibid,14).


Taking (2) and (3) together as a single rule, in effect, Solan (1983) proposes some counterexamples:


(4) I told him1 that Bill1 will be here.

(5) We put him1 in John's1 bed.

(6) I told him1 about the party when Bill1 was here.

(7) I interviewed her1 after Mary's1 speech.


Solan's argument is that (2) and (3) together correctly predict that (4) and (5) are ungrammatical, and that (6) is grammatical. However, they do not predict the fact that (for Solan himself, as well as for "many speakers") coreference is much harder to obtain in the case of (7) than in the case of (6). Solan's explanation is that in (7) the pronoun and intended antecedent are in the same clause. He argues that (2) and (3) should be revised so as to take this into account.

Solan also produces the following counterexamples:


(8) In John's1 bed he1 put Sam.

(9) In John's1 bed Sam put him1.

(10) That John1 likes liver he1 won't admit.

(11) That Mary1 should start eating liver we told her1.


The problem for the C-Command analysis, according to Solan, is that, although (9) and (11) should be perfectly grammatical, in fact they are just as ungrammatical as (8) and (10), respectively. Solan also cites the following:


(12) Which book that John1 read did he1 like ?

(13) How angry that John1 came is he1 ?

(14) How angry that John1 came did that make him1 ?


Solan takes (12) to be grammatical, and (13) to be ungrammatical, while assuming that they have identical syntactic configurations as far as the C-Command constraint is concerned. He makes the point that a C-Command analysis would predict opposite polarities as between the grammaticality of (13) and that of (14), as the former involves a subject pronoun and the latter an object pronoun, whereas in fact he finds both to be ungrammatical.

Solan himself proposes the following improved version of the constraint:

(15) Backwards Anaphora Restriction (BAR):


Pro1 ... NP1 is impossible if:

Pro c-commands NP: or

Pro and NP are within the same minimal S, and Pro d-commands NP.


It must be understood that (15) does not stand alone: it comes as part of a package which includes the definition of c-command (i.e. (2) above), the definition of d-command (i.e. kommand from Lasnik (1976)):


(16) Node A d-commands another node B if (1) neither A nor B dominates the other; and (2) the cyclic node that most immediately dominates A also dominates B (where NP,S are cyclic nodes),


together with the understanding that (15) applies at both the level of Logical Form and at the level of S-Structure, as these terms are represented in the following diagram:

(17)


DEEP STRUCTURE

|

|

|

|

|

S-STRUCTURE

+ +

+ +

+ +

+ +

+ +

PHONETIC FORM LOGICAL FORM


(after Chomsky and Lasnik (1977));


and the understanding that "What the restriction does at logical form is to act as a filter, which eliminates pairs of indices that violate it." (Solan, op.cit.,70).

The BAR, then, is a restriction which blocks only backwards Anaphora if either of two configurations occurs at either of two levels of grammatical representation. It seems to me that this solution falls well short of the ideal of simplicity (let alone naturalness) which occurs so frequently in Chomsky's writings, and in some works on the Philosophy of Science. I hope that the Cognitive/Functional alternative that I advance in the next section will strike the reader as somewhat less contrived, at any rate, than the above.



3. A Cognitive/Functional perspective



3.1.Pragmatics or Syntax ?


Let us go back to the passage quoted earlier from Solan (1983:36), which I repeat here:


Anaphora is basically a discourse phenomenon and can occur between sentences, and in some cases, with no linguistic context present at all.


Here Solan expresses an insight which I propose to generalise to cover even those cases of Anaphora which (as we

saw above) have been considered by Generativists to be syntactic in nature. In other words, I wish to draw attention to the unnaturalness of an explanation which treats Anaphora as being basically pragmatic in nature, while making an (apparently ad hoc) exception of a small sub-group of Anaphoric phenomena, i.e. those that are deemed to be subject to the BAR.

In his chapter on Strategies and Contrastive Stress, Solan (op.cit.) reviews the literature relating to the way that contrastive stress, the parallel function strategy, context, knowledge of the world, and "processing principles" (including relative proximity of antecedents) affect how a hearer/reader chooses between those potential antecedents of a pronoun which are not precluded/excluded by the BAR. As he makes clear, these strategies are pragmatic in nature, rather than linguistic, and thus outside what Generativists generally consider to be the "Grammar".

Since the Generative tradition takes it as axiomatic that there should be a distinction between the BAR (in some formulation or other) and pragmatic strategies such as those alluded to above, Solan feels compelled to adduce only one brief argument in support of his strongly-worded contention that,


none of the (Pragmatic) factors described here, in spite of their central role in understanding language, can undermine the grammatical restrictions on anaphora (ibid, 148).


What he does is to demonstrate that contrastive stress on a pronoun leads the hearer to look for an antecedent other than that which would normally be the obvious choice according to considerations of relative proximity, the parallel function strategy, context, and knowledge of the world. He then goes on to claim that contrastive stress cannot achieve the same effect where the BAR is involved, in the sense of being able to link a pronoun with contrastive stress to an antecedent which is not a possible antecedent according to the BAR. He maintains that contrastive stress can override the effects of mere Pragmatic processing principles, but it cannot override the effects of a rule of Core Grammar, such as the BAR. He cites the following examples:


(18) He thought that John acted unethically.

(19) He thought that John acted unethically.

(20) *Mary thought that John should represent

herself.

(21) *Mary thought that John should represent

herself.

(22) *John1 saw him1.


Solan's point, in connection with (18) and (19), is that putting contrastive stress on he does not make it any more possible to consider he and John to be coreferential. Likewise, in (20), putting contrastive stress on herself (as in (21)) does not make it any more possible to consider the sentence as grammatical. And, in (22), the contrastive stress on him still does not succeed in achieving an interpretation whereby John and him are coreferential.

Even if one were to accept the rationale for Descriptive Grammaticality Intuitions, Generativists must still be criticised, in addition, for only seldom providing contexts for their example sentences. Evans (1980) provides examples (including relevant linguistic contexts) which tend to disprove Solan's remarks about (18)-(22). Consider (19), for example, in the following context:


(19') Absolutely everyone thought that John acted unethically. Even he1 thought that John1 acted unethically.


The syntactic relationship of the relevant pronoun and noun is identical in (19) and (19'), as far as the BAR is concerned, yet it is clear that (19') has the reading whereby they are coreferential, and that the example is perfectly comprehensible with that reading. This coreferentiality obviously has something to do with the contrastive stress, as well as other factors which come together (one might say) as a communicative "package".

Thus it would be misleading to say that the contrastive stress is (somehow) the cause of the BAR being "transgressed against" here. Rather, since there is an intention, on the speaker's part, to repeat verbatim the words "that John acted unethically" (for what one might call stylistic reasons), the contrastive stress is used to direct the hearer not to use his usual, first-resort antecedent-hunting strategy as far the contrastively-stressed pronoun is concerned.

Likewise, consider (21) in the following context:


(21') Mary's husband, John, wanted to help Sarah, and to represent her in an upcoming mock-trial. But Mary1 thought that John should represent herself1.


The only example out of (18)-(22) where I agree with Solan (op.cit.) is (22):


(22) *John1 saw him1.


Here the contrast is lexical in a very straightforward way: if coreference is intended in such clause-internal cases, then reflexive forms are available precisely for that purpose. Thus noncoreference in sentences such as (22) cannot be overridden by pragmatic/stress factors for the simple reason that the noncoreference in not a pragmatic, but a lexical matter.

On the other hand, the contrast between reflexive and nonreflexive pronouns across clause-boundaries is not quite so clearly delineated, so that, in (21'), her (with contrastive stress) could be substituted for the sentence-final reflexive without affecting comprehensibility or meaning. Compare:


(21'') Mary's husband, John, wanted to help Sarah, and to represent her in an upcoming mock-trial. Then there was Pauline, who was very keen on getting John to represent her. But Mary thought that John should represent herself.


In (21''), "herself" provides a clearer contrast with the "her" of the penultimate sentence than would another instance of "her". A speaker would be expected to have regard for how comprehensible his utterance would be to the listener, and to bear such "stylistic" factors in mind. This is one example of why the linguistic alternatives available to a speaker to express a given sentiment are relevant to linguistic description, and also an example of the lack of a clear boundary between "stylistic" considerations (which may, but need not be tied in with the notion of Presumed Comprehensibility) and what Autonomist Linguistics would view as matters involving "Grammaticality".

Thus we can see that pragmatic factors can indeed override the BAR. This is strong evidence that the BAR (in any of the various versions) is the wrong approach to take to Anaphoric restrictions, as not just most, but ALL the factors involved are in fact pragmatic.



3.2.The Proposal


In its first, preliminary formulation, the Pragmatic Anaphora Constraint (PAC) which I am putting forward as a replacement for Syntactic constraints such as the BAR reads as follows:



(23) A communicator who is cooperatively communicating is never any less explicit than he assumes to be necessary for communicative or other specifiable reasons -- and never more so.


There are some important examples where the PAC must be read in conjunction with Bever(1974)'s General Semiotic Principle:


(24) A symbol "S1" can stand for "S2" if (a) the prior connection is known or (b) there is an indication that a connection is about to be established.


Let us see how the PAC works in practice: In (18), for example,


(18) He thought that John acted unethically,


"he" would be used cooperatively only if something (e.g. prior context) had sufficed to tell the addressee who it was that "he" referred to. But, by the same token, if there was some background information which rendered "he" sufficiently explicit, then the proper name "John" would be over-explicit if it referred to the same person.

The argument has been adduced (Max Cresswell, personal communication) that the PAC could be used to show the opposite of what I interpret it as showing, with respect to (18):


In (18) "he" would be used cooperatively only if something in the prior context has sufficed to tell the communicatee who "he" refers to, or if something in the immediately following linguistic context will. If nothing in the prior context supplies a referent, then it must be supplied by the immediately following linguistic context. The only item which could supply it is "John", and so "he" and "John" are coreferential.


Where I disagree with the above analysis is in the second sentence. According to Bever's Principle, the hearer cannot automatically assume that a referent will be supplied in the subsequent context: there must be some reason given to him by the speaker for assuming that one will be. The issue is best dealt with, perhaps, by incorporating the relevant part of Bever's Principle into the PAC:


(25) A communicator who is cooperatively communicating is never any less explicit than he assumes to be necessary for communicative or other reasons -- and never more so. He makes assumptions as to the addressee's background knowledge and assimilation of prior context, and postpones explicit references only if some guidance is given to the addressee as to what the referent is likely to be, or some indication present that an explicit reference is upcoming. Usually, such an indication takes the grammatical shape of a clause or phrase that is obviously non-sentence-final, so that the addressee can expect to hear/read the Antecedent in the main clause, or main phrase of the relevant clause (in the case of Preposed Prepositional Phrases).


This expansion may appear lengthy and cumbersome, but that appearance is only an artificial consequence of the fact that Autonomists tend to limit Anaphoric Constraints to (what they regard as) syntactic factors, and "postpone" discussion of pragmatic factors into the indefinite future. It would seem to me that (25) should be compared, not just to Autonomist Anaphoric Constraints alone, but to these plus any relevant pragmatic explanations which are explicit or implicit in the Autonomist literature, and which cover the range of linguistic behaviour covered (in outline) by (25).

The essential contrast between the PAC (25), on the one hand, and Autonomist constraints such as the BAR, on the other, centres on their respective treatments of such examples as:


(26) If you mention his1 name, John1 will suddenly appear.

(27) Near him1, Dan1 saw a snake.

(28) In front of him1, Ben1 held a candle.

(29) In her1 kitchen, Zelda1 spent her sweetest hours.

(30) For his1 wife, Ben1 would give his life.

(31) How obnoxious to his1 friends Ben1 is.

(32) Fond of his1 wife though Ben1 is, I like her even more.


Such examples have been shown by Carden (1982), rebutting the claims of e.g. Kuno (1972a), to occur naturally in texts without necessarily needing preceding antecedents for the relevant pronouns. I will return to these two works in a later section.

Unlike the example which Cresswell discussed (above, (18)), all the examples (26-32) contain some early indication that the sentence contains at least two phrases/clauses, with the implicit promise that the second phrase/clause will contain an "antecedent" ("postcedent"?) for the earlier pronoun. In (26) that early indication consists of the word if; in (27) it is near; in (28) it is in front of; in (29) it is in; in (30) it is for; in (31) it is how obnoxious ( + exclamation intonation); and in (32) it is (sentence-initial) fond of.



3.3.Discussion


It might be helpful to discuss some more examples (some of which we have seen before) against this background:


(33) I told him1 about the party when Bill1 was here.

(34) I interviewed her1 after Mary's1 speech.


Solan(op.cit.) considered that (33) was grammatical, whereas he found it harder to consider (34) as grammatical. However, it seems to me that this contrast between the two sentences can be made to disappear, even for those using Descriptive Grammaticality Intuitions (as well as those thinking in terms of Comprehensibility), provided the intervals between Antecedent and Anaphor are equalized, and an appropriate context is provided for them e.g.:


(35) A: What about Sam ? And when did you tell Bill ?

B: I told him1 about the party when Bill1 was here.


(36) A: What about Sally ? And when did you interview Mary ?

B: I interviewed her1 long after the date of

Mary's1 speech.


The way I interpret these examples and the reasons speakers might have for using such anaphoric configurations, (35B) uses the pronoun him because the reference is made clear by the carrying-over of the verb tell from (35A). On the other hand the speaker of (35B) finds it advisable, if not actually essential, to use the explicit Bill later in the clause, to avoid confusion as to whether he was referring to Sam.

The same considerations apply to (36). The difference between (34) and (36B) is that in (34) the two NP's are too close together in time/space for the explicit Noun Mary to be so likely to be interpreted as coreferential with her, as the "reminder" function of an explicit NP is related to its distance from any previous coreferential NP, although other, more structural or stylistic factors are involved as well (as we saw above).



3.4.Preposed prepositional phrases


Let us also have another look at the following examples from Solan (op.cit.):


(37) In John's1 bed he1 put Sam.

(38) In John's1 bed Sam put him1.

(39) That John1 likes liver he1 won't admit.

(40) That Mary1 should start eating liver we told

her1.


The BAR includes a reference to the level of Logical Form in order to account for sentences such as (37)-(40), as it is at that level that (37)-(40) are identical to their non-preposed versions:-


(41) He put Sam in John's bed.

(42) Sam put him in John's bed.

(43) He won't admit that John likes liver.

(44) We told her that Mary should start eating liver.


The BAR already outlaws coreference between the relevant noun-pronoun pairs in (41)-(44), so making it apply at the level of Logical Form automatically outlaws it in (37)-(40) as well.

If the range of data is restricted to the above eight sentences, devoid of plausible discourse context, Solan's solution may well seem to "work". However, in my view this is not an insightful or revealing explanation. Leaving aside examples like (39) and (40) for the time being, I feel that the answer could well lie in the Pragmatic/Discourse characteristics of Preposed Prepositional Phrases (PPP's).

As far as (37), for example, is concerned, we have a choice, essentially, between searching corpuses for PPP's and investigating their characteristics in situ, on the one hand, and imagining a plausible context in which they might occur, on the other. The former method, though more laborious and time­consuming, is perhaps ultimately more convincing. However, the latter method can also be convincing enough, I believe, provided it is Comprehensibility Intuitions that one employs. Let us, then, imagine what sort of context sentences such as (37) could plausibly occur in:

(45) The two famous families of quintuplets had swopped houses for the vacation. Mr.Smith had to decide which Smith quintuplet to put in which of the Jones quintuplets' beds. In Paul Jones' bed he put Adam Smith, in Samantha's bed he put Jennifer, in Sarah's bed he put Caroline, in John's bed he put Sam, and in Rebecca's bed he put Josephine.


In the Generative approach to data, there seems to be the latent assumption that the examples cited can occur in practically any discourse context, so that the latter need not be treated as relevant. However, it is clear (to me, at any rate) that some contrast, as in (45), is necessary for PPP's to occur at all. This implies that PPP's have quite specific Pragmatic/Discourse functions. As in Chapter Six, I will have recourse to the Given/New distinction, which occurs (in various guises) in the writings of various Functionalists, even though various writers have cast doubt on the validity of such a simple binary distinction.

In (45), we can see that what is Given (in the last sentence) is the fact of Mr.Smith putting one of his children in one of the Jones children's beds. What is New (in each clause of that sentence) is the identity of each child involved, plus the identity of the child whose bed it is being put into.

PPP's need not necessarily involve two sets of new information per clause, as in (45). In the first sentence of the previous paragraph (unlike the current sentence itself), only one contrast is involved -- and even that is implied, rather than explicit (i.e. the contrast between (45) and any other example that might come to mind). In the sentence immediately preceding the current sentence, however, there is both a (New) contrast between two different sentences that are involved, and also another contrast between a case of only one contrast, and a case of two contrasts.

The fact that PPP's appear to require certain kinds of contexts, however, does not imply that only PPP's can occur in such contexts. Such contexts are necessary, but not sufficient conditions for the occurrence of PPP's. Example (45) can occur in a version devoid of PPP's. However, it is interesting to note that contrastive stress can be highlighted in writing for those phrases which (in (45)) were PPP's, whereas it is redundant to indicate contrastive stress in PPP's themselves, as they (in effect) always do have such stress in speech: In other words, PPP's are in themselves a Syntactic indication that a contrast is involved.


(45') .... (Context as per (45)): He put Adam Smith in Paul Jones' bed, he put Jennifer in Samantha's bed, he put Caroline in Sarah's bed, he put Sam in John's bed, and he put Josephine in Rebecca's bed.


What is interesting here, too, is the fact that the repetition of "he put" seems stylistically less felicitous than omitting all but the first occurrence of these words in (45'). Such omissions would result in what I elsewhere (see Chapter Six) call Conjunction Ellipsis. Likewise, such omissions in (45) would result in Gapping proper.


(45'') .... (Context as per (45)): He put Adam Smith in Paul Jones' bed, Jennifer in Samantha's bed, Caroline in Sarah's bed, Sam in John's bed, and Josephine in Rebecca's bed.


(45''') .... (Context as per (45)): In Paul Jones' bed he put Adam Smith, in Samantha's Jennifer, in Sarah's bed Caroline, in John's bed Sam, and in Rebecca's bed Josephine.


PPP's always involve contrast, and this contrast is always intersentential (Discourse/Pragmatic), rather than intrasentential. We can portray the relevant features of PPP's (in context) as follows:

(46) DTx + (P+NP1)(=PP).... (P+NP2)(=PPP) + DTx;


where DTx = Discourse Topic (entity x),

P = Preposition,

PP = Prepositional Phrase,

PPP = Preposed Prepositional Phrase,

and NP1 is not identical to NP2.


The normal, or unmarked case is for the Discourse Topic to be carried on from sentence to sentence in the Discourse -- always outside any PPP's that may occur. The only time the Discourse Topic itself would occur in a PPP is when there is actually a change of Discourse Topic -- i.e., the new Discourse Topic is in a PPP for the purpose of contrasting it with the old Discourse Topic. This is a simplification, to the extent that the only role in a PPP which a Discourse Topic can absolutely never play (except, as stated above, when there is a change of Discourse Topic) is that of HEAD (most highly accented word) of the PPP. The further one gets from this HEAD in the PPP, the more likely is the occurrence of the Discourse topic -- particularly in Pronominal form. Let us look again at some relevant examples, coupled with variations along the lines just discussed:


(47) In John's bed he put Sam.

(48) In John's ornate, 19th century, luxurious four-poster bed, he put Sam.

(49) That John likes liver he won't admit.

(50) That John has even the very slightest smidgeon of a liking for liver he would not for one moment admit.


I take it that the reader will agree with me that (48) and (50) are more likely to occur than are (47) and (49), respectively. That is because the explicit NP "John" is further from the most highly accented word in the PPP's of (48) and (50) than it is in (47) and (49).

I have not much, at present, to say about preposed constituents other than PPP's, except that I assume that, in context, there is always some contrast involved. I do not feel that I myself use, or use often, preposed structures such as those in (49) and (50), so I do not have substantial proposals to offer about their use. In fact, I doubt that their occurrence in modern English is frequent. Nevertheless, I would like to suggest plausible contexts where the two actual examples from Solan (op.cit.) that I have cited above ((49) and (40)) could plausibly and comprehensibly occur:


(51) When John was young, he used to hate eating sheep's brains and liver -- in fact, offal of any kind. I am sure he has well outgrown that stage now; but, though you occasionally see him eating brains with apparent pleasure, that John likes liver he won't admit.


(52) Mary's private nurse is somewhat hard-of-hearing. She thinks we told Mary that she should go for a swim in the river. But it was just a misunderstanding: That Mary should start eating liver we told her.


The Given/New distinction explains the referential differences between pronouns and nouns in PPP's: When a new entity is introduced into a discourse, it is never introduced by a (non-deictic) pronoun, but always by a noun. This means that, in a PPP, a pronoun can never be interpreted (unless it is deictic/exophoric) as being central to the contrastive, new information that the PPP is contrastively introducing.


(47) In John's bed he put Sam.

(53) In his bed John put Sam.


In (47), the identity of the possessor of the bed into which Sam was put might conceivably be central to the contrast embodied by the PPP as a whole, but not in (53), unless there is contrastive stress on "his", for some reason -- and in that case, the contrastive stress would make it impossible for "his" to be coreferential with "John". This is because the best way of expressing coreference, with emphasis on the pronoun, would be to use an emphatic pronoun:


(54) In his own bed John put Sam.


Thus the Anaphoric contrast between sentences such as (47) and (53), whereby (47) appears to rule out coreference between the relevant noun and pronoun, and (53) to allow it, has nothing to do with any hypothetical syntactic analysis (as in the BAR), but is a result of the Given/New Distinction and the functions of PPP's in discourse.

It should be noted that the role of the given/new distinction may not be limited to the noun/pronoun contrast. Szwedek (1981) suggests that, in Polish,


Nouns in sentence-final position are interpreted as noncoreferential ('new'), in sentence-initial position as coreferential ('given') to some earlier occurrence of the same noun.


Indirect empirical support for this sort of Functional analysis of PPP's is provided by Ramsay (1987), which is an empirical study of preposed and postposed When-clauses and If­clauses. She finds that the preposed versions are thematically associated to the preceding discourse as well as to the main clause, thus having braod scope. The postposed versions are only related to their main clause, thus having a very localised scope. Preposed When-clauses form an integral part of the narrative, in that they help to sequence the temporal contour of the main line of the narrative, whereas postposed When-clauses only function as added comments to the main clause.



3.5.Further examples

Other examples discussed by Solan and/or Reinhart can, I think, be disposed of rather quickly:


(55) Which book that John read did he like ?

(56) How angry that John came is he ?

(57) How angry that John came did that make him ?

(58) In Ford's home town he is considered a genius.

(59) In Ford's home town he built a house.


Solan considers (55) grammatical, and (56) and (57) ungrammatical under the coreferential readings. However, this is because of Lexical, not Syntactic factors. Compare:


(60) How angry that John lost is he really ?

(61) How angry that John lost did that make him ?


The armchair intuitioniser finds it easier to envisage someone's losing making them angry than someone's coming making them angry, and this makes (60) and (61) seem more Comprehensible/Grammatical than (56) and (57) with coreference between the relevant noun/pronoun pairs.

Solan and Reinhart consider (58) to be grammatical and (59) to be ungrammatical (if "Ford" and "he" are coreferential). They ascribe this to Syntactic factors, saying that the PPP in (58) is a sentential complement, whereas the equivalent in (59) is a verbal complement. They then go on to construct their tree­diagrams in such a way that their respective Anaphoric Restrictions permit coreference in (58), but outlaw it in (59). From my perspective, this seems a very arbitrary and unconstrained approach.

Examples (58) and (59) are not truly parallel in lexis, prosody, context, or length (as is often the case with examples that Generativists pair together to prove a point). With (58), the reader is likely to take "Ford" to refer to a famous U.S. public figure (which makes the sentence seem even more natural). In addition, a natural contrast springs to the intuitioner's mind immediately (it seems to me), whereby the world outside the home town is assumed to consider Ford to be mediocre or worse. The contrastive function of PPP's helps to set up this imagined contrast, which makes the sentence seem comprehensible and grammatical/acceptable.

Example (59), on the other hand, is lexically very mundane. No natural context springs to mind, and so the intuitioniser can see no good reason for it to have a PPP in it. Compare:


(62) In Mario Pei's home town he built his second pyramid.


Provided the intuitioniser knows something about Mario Pei and his relationship to pyramids, example (62), which is Syntactically identical to (59), is likely to seem just as grammatical/comprehensible as (58).


3.6.Counterexamples


I would like to conclude this section by discussing three apparent counterexamples to my PAC.


(63) The waiter brought them tea when the guests had finished the dessert (from Johnson-Laird 1983:390).


(64) I don't know if he's serious, but my roommate wants to walk a tightrope over Niagara Falls. (example from a student's essay cited in de Beaugrande and Dressler (1981)).


(65) As for the Opposition parties, whatever they say in public about the sooner the better, privately they are apprehensive. Their vastly greater financial resources puts (sic) the Conservatives in a much stronger position to fight an early election. ("The Observer", London, 17 April 1983, page 4).


These three sentences all have features which give strong guidance (though this does not take the usual form of a subordinating conjunction or preposition) to the listener/reader as to the location of the antecedent of the relevant pronoun.

In (63), the word "waiter" immediately conjures up a scenario in the listener/reader's mind, whereby a mental referent for "them" is already in place before this word is actually perceived. This makes the sentence seem grammatical/comprehensible in a way that the following version is not (assuming no prior antecedent for "them"):


(66) Paul brought them1 atlases when the guests1 had finished the dessert.


In (64), the introductory "I don't know if" is a standard formula which indicates to the hearer that this is to be a two­clause (at least) sentence, with the probability of an upcoming "Postcedent" for any preceding Anaphor.

In (65), although the Anaphor "their" might initially be taken as referring to either the Opposition parties or to the Conservatives, at least the field of possible referents (in the context provided) is limited to just those two possibilities. Example (65) involves a stylistic "stretching" of comprehensibility, but not beyond all limits.



4.The Non-inevitability of Autonomist analyses.


Because Autonomist Linguistics has been around for so long, and has inspired so much research by (in many cases) such clever people, there is a tendency in some quarters to be unimpressed when any one specific Autonomist tenet is refuted. "After all," such scholars seem to think,"that tenet has wider, proven application than merely in this particular case. So it cannot be disproved so easily." This defensive reaction may also be a factor as far as the notion C-Command is concerned.

It may be useful, therefore, to look at the kind of data that Chomsky (1981a) considers that his (then current) type of theory is particularly good at explaining -- and then seeing whether there are not other, at least equally insightful ways of looking at that data. Please note that Generative Grammar is in a constant state of flux, and it is not my aim to refute a specific (possibly dated) Generative analysis. My aim is, rather, to illustrate the fact that even data specifically chosen by Generativists as being favourable to their case is amenable to a contrary interpretation.


Chomsky (1981a) cites the following :


(67) everyone wants John to like him

(68) everyone wants him to be liked by John

(69) for every person x, x wants John to like x

(70 for every person x, x wants x to be liked by John


He points out that (67) can have the interpretation expressed in (69), whereas (68) cannot have the corresponding interpretation expressed in (70), despite the fact that John likes him and he is liked by John are "virtual synonyms".

The specifics of the Generative account of such data is not particularly relevant to us here. Suffice it to say that there seems to me to be a clear and obvious Functional explanation. That is, the English language possesses another (slighly simpler) way of expressing (70), viz.


(71) everyone wants to be liked by John,


so that it makes functional sense for the language to reserve the structure underlying (68) for the meaning that it does in fact express. Indeed, there is even a second, more contrastive way of expressing (70), namely:


(72) everyone wants himself to be liked by John.


It is to be expected that some readers will be tempted to consider (72) "ungrammatical", because of its rarity, but there is no doubt about its comprehensibility, which is what concerns me.

The crucial difference between (67) and (68) is that English does not permit clause-final NP Ellipsis, so that


(73) everyone wants John to like [ ]


is not an option available to the speaker/writer who wishes to express (69). This is why (67) is intrinsically ambiguous as to the reference of him, unlike (68).

Chomsky (1981a, 225) goes on to discuss:


(74) everyone wants that John like him

(75) for every person x, x wants that John like x

(76) everyone wants that he be liked by John

(77) for every person x, x wants that x be liked by John


Though in Chomsky's own "dialect", (74) and (76) are not "well-formed" (which seems to be an undefined term used in an attempt to avoid the serious problems surrounding the terms grammatical and acceptable -- see Chapter One), he nevertheless claims to know that (74) could have the meaning (75) and (76) the meaning (77).

I am sure he is right about that, but not about the reason for the contrast between (74)-(77), on the one hand, and (67)-(70), on the other. In fact, the reason is that English­speakers do not find NP Ellipsis comprehensible in the case of the subject of a that-clause, as in:


(78) everyone wants that [ ] be liked by John,


perhaps because of the ambiguity caused by the fact that that is a pronoun, as well as a conjunction.

This means that there is no natural alternative to (76) as a means of expressing (77), and this, in turn, results in the ambiguity of he in (76), paralleling the ambiguity of him in (74).



5.Kuno's Functionalist proposals


Because of the seemingly close compatibility of Cognitivist and "Functionalist" approaches, I think it appropriate to pay some attention to the interesting work done by Susumu Kuno. Kuno (1972) works within the Generative framework, while attempting to incorporate into it a Functionalist perspective. Thus, he accepts the parameters of the (then current) Generative statement of the problem of Anaphoric Configurations, including the notion that the Anaphoric Constraint "must apply to the stage in the derivation of sentences where Passivization has already applied" (ibid, 162), because of examples such as:


(67) The girl whose life he1 had saved before betrayed John1.

(68) The girl whose life John1 had saved before betrayed him1.

(69) John1 was betrayed by the girl whose life he1 had saved before.

(70) *He1 was betrayed by the girl whose life John1 had saved before.


However, Kuno puts forward some further examples which, he suggests, necessitate what he calls a "Direct Discourse" analysis:


(71) John1 expects that he1 will be elected.

(72) That he1 will be elected is expected by John1.

(73) *That John1 will be elected is expected by him1.

(74) John1 claimed that he1 was the best boxer in the world.

(75) That he1 was the best boxer in the world was claimed by John1.

(76) *That John1 was the best boxer in the world was claimed by him1.


Kuno (ibid, 162) explains:


The grammaticality of (72) and (75) is automatically accounted for because he in these sentences is in a subordinate clause that does not dominate John. However, the ungrammaticality of (73) and (76) is inexplicable because forward pronominalization ... is supposed to be unconditional.


Kuno (ibid) sets out to explain these (supposed) facts, together with the apparent dissimilarities between "expect" and "claim" (as above), on the one hand, and "deny" and "forget" (see below), on the other:


(77) John1 denied that he1 was sick.

(78) That he1 was sick was denied by John1.

(79) ?That John1 was sick was denied by him1.

(80) John1 forgot that he1 had an appointment at two.

(81) That he1 had an appointment at two was forgotten by John1.

(82) ?That John1 had an appointment at two was

forgotten by him1.


Kuno comments:


Although (79) and (82) are very awkward, they are not ungrammatical. Note that these sentences are at least considerably better than (73) and (76). What is it that makes (79) and (82) better than (73) and (76) ? This paper is an attempt to give an answer to this question.


Kuno develops this topic at some length, employing data from both English and Japanese. His explanation, in his own words, runs as follows:


One of the differences between verbs such as expect, claim, know, think, request, on the one hand, and verbs such as deny, forget, be unaware (of), on the other, is that the content of the complement clause of the former represents "more or less" the direct discourse of the matrix subject, while this is not the case for the latter.


I believe that it can be shown that the familiar Generative problem with data is at the core of the matter. All the allegedly problematic English examples ( 70, 73, 76, 79 & 82 ) can in fact be so contextualised as to no longer appear to pose Comprehensibility problems :


(70') Talking about betrayals, I've heard of some absolutely shocking cases ! For example, did you hear what happened to John1 about five years ago ? Well, he1 was completely and utterly betrayed by the very same girl whose life John1 had saved before -- in fact, only a few days before !


(73') John's1 chances in the upcoming elections are not rated very highly by most people around here. In fact, that John1 will be elected is expected by him1 and a few of his1 cronies, but by no one else, to speak of.

(76') John's1 abilities as a boxer were not rated very highly by most people in boxing circles. In fact, that John1 was the best boxer in the world was claimed by him1 and a few of his1 cronies, but by no one else, to speak of.


(79') John1 was always terrified of doctors, and he1 was most unwilling to ever admit that he1 was ill. Time and time again, that John1 was sick was denied by him1 most vehemently -- no matter how serious the symptoms.


(82') John's1 entire family is extremely forgetful. Take last Monday, for example: That John1 had an important appointment at two was forgotten by him1, his1 father, his1 mother, his1 two sisters, and his1 three brothers.


Kuno's analysis also fails when it is used to "predict" intuitions as to other sorts of examples: again, the problem arises from a failure to contextualise examples, together with the use of Descriptive Grammaticality Intuitions, instead of Comprehensibility Intuitions, or corpuses.

Kuno (1972a) lists a number of Hypotheses, of which nos. II, III, and IV involve Anaphora:


Hypothesis II. Backward pronominalization is possible in English only when the rightmost of the two coreferential noun phrases represents old predictable information.


This Hypothesis is based on the following examples, which are contextualised:


(83) a. What do you suppose I did when I saw Harry1

getting mad ?

b. ? I calmed him1 before Harry1 did something

rash.


(84) a. Who do you suppose I calmed when I saw him getting

mad ?

b. * I calmed him1 before Harry1 did something rash.

(85) a. Which of the two, Harry or Bill, do you suppose I calmed before I saw them getting mad ?

b. * I calmed him1 before Harry1 did something rash.


In fact, it can be argued that the strangeness (whether it be called "ungrammaticality" or "unacceptability" seems fairly arbitrary in terms of Generative methodology) of (84b) and (85b) is not due so much to the fact that Harry does not involve "old predictable information" -- but, rather, to the fact that (84b) and (85b) do not constitute relevant answers to the questions (84a) and (85a), respectively. Carden (1982) provides corpus/text examples which refute Hypothesis II. Indeed the present Chapter also contains counterexamples to this Hypothesis, viz. (63)-(65). In (63), the rightmost NP represents predictable, but not old information. In (64), the relevant information is neither old nor predictable. And in (65), the information is predictable, but need not necessarily be old.


Hypothesis III. A noun phrase that represents what the sentence is about, namely, the theme of the sentence, cannot be pronominalized intrasententially.

Kuno bases his case on the following example:


(86) What did John do for Mary ?

a. When he went to Boston, John took her out to dinner.

b. * When John went to Boston, he took her out to dinner.


(86b) is ungrammatical for some speakers because the subject of the matrix sentence, which is the theme of the whole sentence in this context, has been pronominalized in violation of Hypothesis III (Kuno 1972a, 305).


In fact, what is actually "odd" about (86b) is not that the subject of the matrix sentence has been pronominalised -- but that the first reference to John in the sentence, coming right after a question which contained the explicit NP John, is over-explicit, in that it is not pronominalised.

Compare:


(86) What did John do for Mary ?

c. When he went to Boston, he took her out to dinner.


Example (86c), unlike (86b), does not violate the PAC by being over-explicit. Of course, it does not violate Hypothesis III either, in that there is no possibility of "intrasentential pronominalisation". However, a somewhat longer, and therefore more realistic context (than Kuno's two-line question-answer pair) may clarify the issue:


(87) Q: What did John do for Mary ?

A: Well, he took her to see his cousin Jeff, who owns a sawmill in British Columbia. He also took her on a whirlwind visit to Las Vegas, and his relatives there. And when John went to Boston, he took her out to dinner in the swankiest restaurant in town.


Example (87) violates Hypothesis III (but not the PAC), and is perfectly comprehensible and natural-sounding.


Hypothesis IV. Noun phrases of exhaustive-listing interpretation cannot be pronominalised intrasententially.


Kuno gives the following example:


(88) Who killed Mary ?

a. ? With an accomplice that he1 hired, John1 killed Mary.

b. *With an accomplice that John1 hired, he1 killed Mary.


According to Kuno, Hypothesis II makes (88a) "awkward", and (88b) is "ungrammatical" because of Hypothesis IV. However, it seems to me that what is odd about (88a) is the fact that the question is not answered in a straightforward manner with "John", or "John killed Mary", or even "John killed Mary with an accomplice that he hired". This last version would put the important information first, and leave the incidental details (appropriately) till last.

And what is odd about (88b), it seems to me, is that the question demands an explicit NP as the core of the answer -- either alone, or as the subject of the predicate killed Mary in the main clause. Knowing whether this amounts to a paraphrase of Hypothesis IV or not would depend on a more precise characterisation of exhaustive-listing interpretation than Kuno (1972a) provides. What is also not clear is whether Hypothesis IV bans both forwards and backwards pronominalisation in the relevant cases (i.e. I read it as banning (88a) as well as (88b), as it stands now).



6.VP Anaphora configurations

Up till now, we have been dealing only with NP Anaphora configurations. But much the same sort of configurations occur with VP Anaphora. Consider:


(89) If Chris will, Robin will go to the movies

tonight too (VP Ellipsis).

(90) If Chris does it, Robin will go to the movies wearing a pink carnation too (DO IT Anaphora).

(91) If I let her, Kim wouldn't even want to leave (Null Complement Anaphora).

(92) Though we're not sure why, Robin will definitely be wearing a pink carnation (Sluicing).

(93) Although not this week, Robin will definitely be dating Sandy sometime soon (Stripping).

(94) If Chris Pat, then Robin will like Sandy (Gapping).


It seems from these examples (though contextual factors have not been investigated) that all these (supposed) types of Anaphora/Ellipsis (labelled in accordance with Generative terminology) allow Cataphora (i.e Backwards Anaphora) -- apart from Gapping.

This exception may be due to the fact that Gapping is the only indisputable case of total Ellipsis in (89)-(94). There is some doubt (in my mind at least) that it is necessary to assume that a verb is "understood" or "missing" in cases of so-called Null Complement Anaphora (91), Sluicing (92), or Stripping (93).

If I am right, then the correct generalisation is that Backwards Anaphora is not comprehensible in the case of Ellipsis (Zero Anaphora). See Chapter Six for further discussion of Ellipsis.



7.Conclusion


The reader will of course make up his own mind as to how convincing or otherwise he finds the proposals put forward in this chapter. Personally, I view the arguments in section 3.1 as to the pragmatic nature of the phenomenon of anaphoric configurations as being fairly strong ones. However, my proposals are not the only ones possible, even within a Functional/Cognitive framework. A lot depends on the analysis of preposed prepositional phrases, and on methodological approaches to data. I would be happy if I were merely pointing the reader in the general direction of an eventual solution to these matters.

Back to Contents Page