CHAPTER FOUR: TYPES OF PRONOUNS
1.Introduction
In a nutshell, the focus of Philosophical interest
in Anaphora can be said to have been the question of the different sorts
of ways that Pronouns appear to behave Semantically and Pragmatically --
or (to put it another way) the question of how many types of Pronouns there
are. When I talk of the "behaviour" of Pronouns, I am essentially referring
to the various kinds of relationships that can obtain between pronouns and
their antecedents. As a point of entry into this topic, let us look at some
examples from Solan (1983,38):
(1) John1 knows that he1 has to pay taxes.
(2) Everyone knows that he has to pay taxes.
(3) Few people know that they have to pay taxes.
(4) That John1 has to pay taxes annoys him1.
(5) That everyone has to pay taxes annoys him.
(6) That few people have to pay taxes annoys them.
The way Solan presents the problem, it consists of a perceived need to explain why (4) tends to be considered as Grammatical, whereas (5) and (6) tend not to be. Solan (op.cit., 39) makes the following suggestion:
The difference simply involves the scope of the constraint, anaphora being more difficult to establish when bound variables are involved. Generally, a quantifier must be less deeply embedded than a pronoun it binds, while definite noun phrases must only be no more embedded than the pronouns associated with them.
Not only would I, as a Cognitivist, propose a different solution -- I would also present the problem differently. I see the problem as originating with the selection of the data: the structures in (4)-(6) are so rare, I would suggest, as to be almost nonexistent in actual speech. And, even for those people who might actually utter (4) (as opposed to:
(7) That John1 has to pay taxes is what annoys him1)
utterances such as (5) and (6) would still tend to be avoided, because they would probably cause confusion in the hearer's mind as to the intended scope of the quantifiers. Compare:
(5') Having to pay taxes is annoying to everyone.
(6') Few people are annoyed by having to pay taxes.
Examples (5') and (6') are much more easily processable alternatives to (5) and (6), respectively, as there is no definite pronoun present to cause scope ambiguities. By contrast, there is no more easily comprehensible alternative to (2) or (3).
Thus depth of embedding does not need to be invoked here as an explanation. The data problem is caused by the fact that Autonomist Linguistics looks at Language as strings generated in some quasi-mathematical way, and so it must be prepared to account, on an indiscriminate basis, for any examples whatsoever that may be generated. However, Autonomous and Functionalist Linguistics look at Language as (primarily) a means to an end. It is crucial to consider what means to a given end a given speaker has available to him on a given occasion. Not just actual, but also potential utterances must be considered as being part of the relevant data.
2.Some historical remarks
As regards Definite NP's, the three relationships between Pronouns and Antecedents that have been commonly argued for in the literature are Coreference, Coindexing, and Referential Dependency. Here are some examples of the ways that the first two have been used in the literature:
(8) 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.D.Z.), the one in the domain is a pronoun (Reinhart 1976,14).
(9) 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 (Solan 1983).
In the present chapter, I would like to discuss these two types of relationship, together with that of Referential Dependency (Evans 1980). The latter work, together with Reinhart (1983), represent a conscious break with the tradition of making do with the one term Coreference for all cases of Definite NP Anaphora. Evans (1980) and Reinhart (1983) have in common a desire to assimilate quantifier-bound pronouns to (at least some cases of) Definite NP's (or vice-versa). Here are the NP Anaphora Restrictions which they, respectively, formulate:
(10) A term can be referentially dependent upon an NP iff it does not precede and c-command that NP (Evans, op.cit.,358).
(11) Coindex a pronoun P with a c-commanding NP , ( not immediately dominated by COMP or S).
conditions:
a. If P is an R-pronoun, must be in its minimal governing category.
b. If P is a non R-pronoun, must be outside its minimal governing categories (Reinhart 1983).
I could point out here in passing that Lasnik (1976,1981) and Chomsky (1981,1982) have made (mostly vague and terse) comments to the effect that a term such as Intended Coreference might be more adequate than Coreference on its own. I will return to this suggestion later.
3.Referential dependency
Evans (1980) identifies four uses of pronouns, of which we shall be concerned with only the following two in any detail:
(a) having a definite NP as antecedent
(b) being bound by a quantifier expression.
Evans is interested in pointing out the similarities between these two uses. One property he wants to say relates them is the relation of Referential Dependency, which he thinks holds between Antecedent and Pronoun.
(12) Every man loves his donkey.
(13) John loves his donkey.
Just as in (12) a formal Semantic representation would link the Pronoun to the Quantifier Phrase by the device of Quantifier Binding, so (according to Evans) in (13) the Pronoun "picks up its reference" from the Proper Noun. Notice that this is saying more than that John and his happen to be coreferential: they are necessarily coreferential, because the Pronoun picks up its reference from the Noun. Evans still allows for the possibility of "accidental" Coreference, as well as Referential Dependency, as we will see below.
I have three criticisms to make of the notion Referential Dependency: a) Evans fails to define it adequately (as Reinhart (1983) point out); b) Evans' constraint employing this notion is empirically inferior to (otherwise similar) constraints which do not; and (c) the notion itself is Psychologically implausible (from a mentalist standpoint).
Apart from a passing use of the phrase "pick up its reference from" (Evans 1980,358), the nearest that Evans comes to providing a definition for Referential Dependency is the following:
(14) let us use the expression t is referentially dependent on t' to mean that t is to be understood as being taken to have the same reference as t' (ibid,358).
Unfortunately, (14) in no way distinguishes Referential Dependency from Coreference. Consider:
(15) Remember John ? ... If he comes, tell John to sit here.
(16) Remember John ? ... If he comes, tell Jane to sit here.
(17) Remember Jane ? ... If he comes, tell John to sit here.
(18) Remember John ? ... If he comes, tell Paul to sit here.
In such structures as (15)-(18), the pronoun can be coreferential with either a preceding or a following NP. Assuming that Evans was clear in his own mind as to what he intended, by what criteria could one determine whether he was referentially dependent on the first, or the second instance of John in (15) ?
This lack of definition for Referential Dependency shows up crucially when Evans attempts to use it to express a syntactic generalisation, as in (10) above, which has the logical structure:
(19) Rxy = - Cxy
(where C = precede & c-command; and
R = can be referentially dependent on)
What (19) makes clear is that, in such Anaphora Constraints, we have (at least) two crucial terms which must be independently definable for the empirical generalisation as a whole to have any significance. Since, as we have seen, Evans fails to provide a definition for Referential Dependency (and I will argue below that the notion is fundamentally incoherent from a Cognitive point of view), it is clear that (10) has no value as an empirical generalisation -- irrespective of the status of the notion precede and c-command, to which I also have objections on quite different grounds.
In fact, Evans does claim that (10) is empirically superior to a hypothetical constraint (20):
(20) A term can be coreferential with an NP iff it does
not precede and c-command that NP,
which is identical to (10), except for the substitution of coreferential with for referentially dependent on.
Evans bases this claim on examples such as:
(21) Everyone has finally realised that Oscar is incompetent. Even he has finally realised that Oscar is incompetent.
Evans writes:
... a discourse like (21) shows that a pronoun can both precede and c-command an NP with which it is intended to be coreferential -- so long as it does not pick up its reference from that NP (ibid, 358).
Example (21), then, is intended to show the empirical superiority of (10) over (20), as the latter does not allow for (21). Evans obviously judges that he is referentially dependent on the first, but not the second instance of Oscar, whereas it is coreferential with both. Of course, he gives us no hint as to how he determined which of them the pronoun is referentially dependent on, but, for present purposes, we can assume that his guiding principle is that a pronoun can be taken to be Referentially Dependent on any semantically appropriate NP, provided the syntactic configuration is not such that this would involve a contravention of Evans' principle (10). This is rather circular, but Evans leaves us no option.
However, as discussed in Chapter Five (Anaphoric Configurations), examples such as (21), which involve contrastive stress on the pronoun, constitute a principled exception to the kind of Anaphoric Constraint I myself would argue for.
In addition, we can note the peculiar features of example (21):
(a) the predicate involves not new information, as is usually the case, but old (given) information; and
(b) the example involves a large amount of verbatim repetition, so that it might come under the loose Pragmatic heading of "word-play", which may or may not make it ipso facto a principled exception, as well.
If one accepts the above argument, then one can conclude that, even apart from Evans' lack of definition for Referential Dependency, he has failed to show that his Constraint (10) is empirically superior to otherwise similar Generative constraints.
My final major criticism of Referential Dependency takes a frankly Mentalistic/Cognitivistic standpoint. It does not seem to me to make any sort of psychological sense to say that the reference of a given pronoun depends on the reference of a given noun -- in just one of however many occurrences that noun has within the relevant space/time frame.
The antecedent of a pronoun may in principle be indefinitely distant (for speech: in time -- for writing: in space) from it. Although his more "linguistic" paper (Evans 1980) contains examples where antecedent and pronoun are in different sentences, in his more philosophical treatment of anaphora (Evans 1977) he appears to fall into the common trap of making the tacit assumption that the sentence is the largest linguistic unit that needs be taken account of. In the latter paper, his reliance on short printed sentences on paper perhaps made the idea of one word being Referentially Dependent on another word seem very natural -- an artefact of his methodology, perhaps ?
In psychological terms, however, this implies that we would have to consider at least every NP (and if every NP, then perhaps every phrase, or every word) as being present in a "surfacy" form in memory for an indefinitely long time -- so that any pronoun indefinitely distant from it in hearing/reading time will have a noun to "pick up its reference from", as Evans puts it. Even the interesting claims of Bates et al. (1980) go nowhere near as far as that.
Moreover (as we have already noted, in fact), there may, in discourse, be more than one occurrence of an NP near a pronoun that we might want to say is Referentially Dependent on one of them. Which one do we choose ? And what Psychological sense does it make to make such a choice ? And what is the referential relationship between the two Nouns in Evans' system ? Some mental representation in non-surfacy form is surely involved and it must be this to which all coreferential NP's relate or "refer". It may, of course, be the case that one particular NP triggers the creation or activation of a mental address, but subsequent coreferential NP's relate back to the address, rather than to the "vanished" NP itself.
As a philosopher, Evans might not have paid much attention to the notion that Linguistics is part of Cognitive Psychology, or, at least, has been considered to be such by many scholars (following Chomsky). As I stated earlier, Evans identifies four uses of pronouns, and attempts to show that two of them (bound variables and pronouns with Definite NP's as Antecedents) are very similar to each other. He contrasts this approach with that of Lasnik (1976), where, by contrast, the latter class of pronouns is related more closely to pronouns that have "pragmatic" antecedents (e.g. antecedents in other sentences), and where no particular similarities between either of these two classes and bound variables are alluded to. Evans is aware that it might seem that what he gains by generalisations linking his particular pairing of pronoun-types he loses (by comparison with Lasnik) by way of losing any link to Pragmatic Pronouns. He counters this possible objection as follows:
... we are not obliged to postulate a different mechanism of understanding in the two cases, as though in one case a book labeled Rules of the English Language is consulted, while in the other case, it is the book labeled How to Make Sense of One's Fellow Men. It is just that we describe a propensity to interpret what speakers say as being in accordance with a rule of the language only under certain conditions: one of these conditions is when the interpretation of other utterances obliges us to ascribe semantic properties to sentence types considered independently of context (Evans 1980,352).
Most Cognitivists would, I think, have to reject this recourse to the Competence-Performance Distinction (specifically: C/P 1-4 -- see Chapter One). I think that even most Autonomists would have difficulty accepting it, as well.
4.Coindexing
Reinhart (1983) claims that Definite NP's can be divided into two groups: one group can be treated as bound pronouns and is subject to the Coindexing rule (11), whereas the other group is subject to Pragmatic conditions governing Coreference. A Sloppy Identity Test is the proposed criterion for dividing definite NP's into these two Semantic groups. This Semantic taxonomy would be of limited interest, were it not for the fact that Reinhart makes (tentative) claims that this grouping correlates with a Syntactic grouping on the basis of the C-Command relation (Rule (11)).
Reinhart (1983) states her main point in the following terms:
The crucial point in my analysis is that what is needed in the grammar to account for anaphora is a mechanism determining when a pronoun can be translated as a bound variable (ibid,73).
Other authors, of course, have had views on the translation of some kinds of pronouns into bound variables. Reinhart's originality lies in gratly extending the class of pronouns which are to be so translated. On the one hand, she follows the well-established tradition of assuming that reciprocals, reflexives, and pronouns bound by quantifiers fall into this class. On the other, she goes on to divide definite NP's into two sets: those that are to be translated into bound variables; and those that are not. Thus, the class of Pronouns that are to be translated as bound variables comprises reciprocals, reflexives,pronouns bound by quantifiers, and one set of the class of definite NP's.
As she herself says (ibid,80):
... the distinction between bound anaphora and coreference in the case of definite NP's is not arbitrary ..., and it is directly testable in each case by the sloppy-identity test.
It is clear that she intended this test to prove positive also for reciprocals and reflexives, though I can find explicit reference only to applicability to quantifier-bound pronouns:
The crucial point for this paper is that sloppy identity interpretation is possible precisely in the same contexts where bound anaphora with quantified NP antecedent is possible (ibid,67).
This quotation falls short of asserting a perfect correlation between sloppy identity and Q-binding contexts: all that it actually claims is that such a context (presumably including reciprocals and reflexives) is necessary for sloppy identity readings -- not that it is sufficient, as indeed we can see from:
(22) Every farmer on the outskirts of Los Angeles enjoys eating the fruits of his labours -- and so, in many cases, do the local suburban children as well !
(23) He taught himself. I didn't -- that's for sure !
(Marjorie James, 19 August 1982, Brighton, U.K.,
where the intent was understood to be: "I didn't
teach him").
Example (22) is a made-up example. Example (23) is an actual utterance. In both cases, I am, of course, claiming only that they are comprehensible in the required readings. Examples (22) and (23) show that strict identity readings are possible with quantifier-bound and reflexive pronouns. However, in the following examples (the first of which is acknowledged by Reinhart herself), we can see that sloppy
identity readings are possible even when (with definite NP's) there is no coindexing possible (according to Rule (11)), because there is no C-Command relationship between the relevant NP-Pronoun pair. This means that Q-binding-type contexts are neither necessary nor sufficient for sloppy identity.
(24) Those who know her respect Zelda; the same goes for Sue.
(25) When he comes, Paul will sit near Jane, and Sam likewise.
(26) When Paul comes, he will sit near Jane, and Sam likewise.
(27) John's au pair is great. She takes great care of his children. On the other hand, so does Paul, though he's a single parent too, and hasn't got an au pair or nanny.
Reinhart's primary distinction is between those pronouns whose interpretation is decided on the basis of considerations within the syntax, and those whose interpretation follows from semantic and pragmatic considerations outside the syntax. The former enter into relationships which she characterises using the notion Coindexing, and only the latter enter into relationships that she is willing to label using the term Coreference. As Reinhart operates within the traditional Generative framework of Sentence Grammar (assuming the C/P 10 Distinction -- see Chapter One), only the latter relationship can obtain between pairs of NP's which are not in the same sentence. This, of course, is true of many Autonomists apart from Reinhart, but it is worth citing the following counterexamples, in view of the crucial nature (for Reinhart's theory) of the notion that quantifier anaphora, reciprocals and reflexives are cases of Coindexing, par excellence:
(28) Any member who has been squealing that he hasn't had
had a say has already had a say. He's been sent
pages and pages of information.
(U.K. National Union of Railwaymen General Secretary,
Sidney Weighell, at 1318 hours on 27 June 1982,
during an interview on B.B.C. Radio 4).
(29) Paul tells a lot of jokes, particularly about Sam.
The ones I like best, however, are the ones about
himself (modified example from Dick Hudson).
5.Alternative treatment
Szwedek (1981) points out three ways that the term reference (and thus also coreference) has been used in the literature in the following ways:
a) as a relation between a name and the thing named....
b) as an association between two 'occurrences of phrases in the text'....
c) as an association 'between noun phrases of natural language and mental entities present in the language-user's mind'
Szwedek (op.cit) prefers version c), which he associates principally with Sampson (1969). Other authors, such as Perry (1979), have taken essentially similar positions. In this section, I will also be basing myself on this version, in order to argue for a solution to the problems we have encountered in connection with the terms coindexing and referential dependency.
I will argue that the traditional notion Coreference is perfectly adequate on its own -- provided an adequate Cognitivist model of Reference is employed as a basis. Let us look at the type of examples which have been used in the past to argue against the use of this term in Anaphoric Constraints:
(30) Consider Quintus the war-hero who is suffering from amnesia. He is discussing the war-hero with a friend, We report "He (= Quintus) admired him".... Or consider the sentence: Who is that guy over there ? He's Joe Blow (Bach and Partee 1980,6).
Bach and Partee (1980) centre their discussion on an Anaphoric Constraint from Reinhart (1979):
(31) Two NP's must be interpreted as non-coreferential if one is in the domain of the other and is not a pronoun.
They then offer (32) as an explicit and literal version of (31):
(32) In any sentence of L (English) in which an NP1 (not a pronoun) is in the syntactic domain of another NP2, the individual denoted by NP1 is distinct from the individual denoted by NP2.
Bach and Partee rightly point out that (32), and thus (31), would erroneously rule out the examples in (30).
The notion Intended Coreference has (e.g. Lasnik 1976, Footnote 5) been sometimes mentioned in the apparent belief that it would be immune from the problems that examples such as (30) pose for the term Coreference. However, this belief is unsustainable, as we can see if we rephrase (32), substituting the suggested term where appropriate:
(33) In any sentence of L (English) in which an NP1 (not a Pronoun) is in the syntactic domain of another NP2, the individual denoted by NP1 is intended to be distinct from the individual denoted by NP2.
We can see here that the problem remains, as in the counterexamples in (30) the individuals are not only in fact pairwise non-distinct from each other -- they are also intended to be nondistinct. Thus they are counterexamples to (33), as well as to (32) and (31).
However, Chomsky (1981,314, Footnote 2 & 1982b,83) hints at the necessity for mental representations of entities as intermediaries between NP's and real world entities, though he does not say anything very explicit on this matter. This is the approach I myself would favour (where necessary), though my own Pragmatic Anaphora Constraint (see Chapter Five), being based on Functional notions of explicitness in context, does not actually mention such mental representations.
Jackendoff (1983) is a relevant work written from a Cognitivist point of view (whether he would himself agree with that characterisation, I do not know, however). His thesis, in essence, is that it is inappropriate for Philosophers or Linguists to talk as if reference were a relation between a linguistic item and some state of affairs or object in the real world, because the mind (as is well known) plays an active part in imposing its own structure and interpretation on its environment -- to the extent of sometimes creating believable states of affairs and objects which, from the "objective" standpoint of a later time or another person, do not exist or have not existed at all !
So much is well-known. The point is to integrate this fact fruitfully into a theory of anaphora:
We can now say that the information language conveys, the sense of linguistic expressions, consists of expressions in the inner code. What the information is about, that is, the reference of linguistic expressions, is not the real world as in most semantic theories, but the projected world. The referring expressions of natural language will be just those expressions that map into projectable expressions of the inner code (Jackendoff 1983,68).
Anyone interested in exactly how Jackendoff defines the crucial terms in the above quotation is advised to read his article direct. I think his general intent will already be clear, however. As far as the examples in (30) are concerned, we can see that (in Jackendoff's terms) what is involved is a case of Projected Worlds within Projected Worlds. We are all, of course, aware that other people have the unfortunate, and more or less predictable tendency to be "wrong", or even "deluded" from time to time, so we have the routine capacity to cope with this by incorporating, in our own Projected Worlds, models of the Projected Worlds that we assume are in other people's minds.
When we (as in (30)) say "He (= Quintus) admired him", we are taking into account the fact that, in the amnesiac Projected World of "the current Quintus", there are (at least) two distinct entities, viz. Quintus himself and the war hero. Quintus may have said something like, "I admire that war-hero", and we, tongue-in-cheek, are reporting his admiration (however expressed), presumably with some amusement to ourselves. Since we have here a (not uncommon) Projected-World-within-Projected-World situation, we have the option of basing our linguistic utterances either on the "embedded" Projected World, or on the "real" one (i.e. our own matrix Projected World). pragmatically speaking, there is nothing wrong with choosing either option, though it is far more common to base oneself on the matrix Projected World. This tends to lessen the risk of confusion in the Hearer/Reader's mind as to what it is that we ourselves believe.
The second example in (30): "Who is that guy over there? He's Joe Blow," is straightforward enough. The speaker of the second sentence has a Projected World in which "that guy over there" and "Joe Blow" refer to the same entity, and he is just making use of the appropriate linguistic structure to inform the speaker of the first sentence (i.e. the questioner) of that fact, as he obviously does not have a Projected World with that particular feature already in it.
I would conclude that, provided the notion Reference is understood as involving Projected Worlds, rather than real objects or states of affairs directly, the term Coreference is perfectly adequate on its own -- i.e. without being supplemented or replaced by Referential Dependency, Coindexing, and the like -- although it does not seem to me that any such term is in fact necessary for Anaphoric Constraints.
As far as quantifier-bound pronouns are concerned, I adopt Johnson-Laird (1983,390ff.)'s treatment of them, which specifically addresses the points that Evans (1980) raises about the similarities between such pronouns and those with definite NP antecedents:
Since quantified assertions can be interpreted as mental models, it follows that when a pronoun functions like a bound variable it takes its referent in such a model. The assertion:
Every man loves a woman
can be represented in a mental model of the form:
man ---- woman
man ---- woman
(woman)
where the arrow stands for the appropriate relation. The interpretation of the assertion:
Every man loves a woman who loves him
builds on the previous model to produce:
man ---- woman
man ---- woman
(woman)
This interpretation, as Evans desired, closely corresponds to the interpretation of a sentence containing an anaphoric pronoun:
John loves a woman who loves him.
The model that would be constructed for this assertion:
John ---- woman
is created for each individual in the set representing every man (Johnson-Laird, op.cit., 390-391).
Thus I am proposing a Cognitivist theory of Reference which incorporates Jackendoff's theory of Projected Worlds, which is to incorporate, in turn, aspects of Johnson-Laird's theory of Mental Models.
This approach, applied to pronouns, has the best of both worlds: Like Lasnik's, it unifies deictic and definite anaphoric pronouns, in that, once there is an entity in the Mental Model that pronouns can refer to, it makes no difference whether that entity entered the Mental Model by linguistic or non-linguistic means; and, like Evans' approach, Johnson-Laird's account is able naturally to represent the similarities between quantifier-bound and other pronouns.
Let us now go back to examples (22)-(29), which were cited as counterexamples to claims associated with weak or strong versions of the theory put forward in Reinhart (1983). Let us see how the present Cognitivist approach would handle them.
(22) Every farmer on the outskirts of Los Angeles enjoys eating the fruits of his labours -- and so, in many cases, do the local suburban children as well !
This case of VP Anaphora would come under the Phonological Identity Criterion proposed in Chapter Three.
(23) He taught himself. I didn't -- that's for sure ! (Marjorie James, 19 August 1982, Brighton, U.K., where the intent was understood to be "I didn't teach him").
In the Mental Models of Speaker and Hearer, the entity of the male person referred to both by "he" and by "himself" would occur twice -- once as the teacher, and once as the taught. There would be no reason for this entity not to be subsequently available to serve as the mental representation of the pupil whom Marjorie failed to teach. The only qualification that needs to be made here is that the potential for ambiguity (in certain contexts) between a sloppy identity reading and a strict identity reading of such combinations of Reflexives and VP Anaphora has the consequence that the armchair intuitionist has a greater than usual tendency to consider such sequences to be"ungrammatical".
(24) Those who know her respect Zelda; the same goes for Sue.
(25) When he comes, Paul will sit near Jane, and Sam likewise.
(26) When Paul comes, he will sit near Jane, and Sam likewise.
Assuming that the Mental Model makes available, as potential Antecedents, not just entities, but also relations between entities, I see no problem with examples (24)-(26).
(27) John's au pair is great. She takes great care of his children. On the other hand, so does Paul, though he's a single parent too, and hasn't got an au pair or nanny.
Like example (22), example (27) would come under the Phonological Identity Criterion.
(28) Any member who has been squealing that he hasn't had a say has already had a say. He's been sent pages and pages of information (U.K. National Union of Railwaymen General Secretary, Sidney Weighell, at 1318 Hours on 27 June 1982, during an interview on B.B.C. Radio 4).
Example (28) parallels example (23). Once an entity is established in the Mental Model, the latter is access-flexible enough for this entity to be used freely as an Antecedent -- irrespective of whether (in Formal Semantics) a quantifier bound the term which first introduced that entity. Quantifier-binding has a limited "range" in natural language, but this example shows that, whatever it is that limits the range/scope of this binding, it is not the sentence-boundary as such (as has been claimed by the proponents of the notion "C-Command").
(29) Paul tells a lot of jokes, particularly about Sam. The ones I like best, however, are the ones about himself (modified example from Dick Hudson).
This example has nothing to do with Mental Models or the Phonological Identity Criterion. It is just a lexical fact that "himself" is the most convenient option available to communicate the intended meaning, as another "John" might be taken to refer to another person, also named "John", and "him" would probably be taken as referring to Paul, though this would not necessarily appear certain to the hearer/reader. It is worth pointing out that this particular NP (himself/him/John) is bound to have contrastive stress anyway, in context, and so the option of using contrastive stress to direct the Hearer's antecedent-hunting is not a possible one.
Of Evans' four types of pronouns, there is one which we have so far not discussed: what he calls "E-type pronouns", e.g.:
(34) Few MP's came to the party but they had a good time.
The present approach would say much the same about (34) as it would about the examples (23) and (28) discussed above.
Hintikka and Kulas (1985) cite the following examples(35-41) , which they consider to be problematic for the notion of Coreference:
(35) If Stewart buys a car or a motorcycle, he will take good care of it.
(36) *If Stewart buys a car and a motorcycle, he will take good care of it.
The above pair of examples is not particularly problematic for a Mental Models-type of approach. In the case of (35) both Speaker/Writer and Hearer/Reader have in mind two parallel hypothetical situations -- one in which Stewart buys a car, and the other in which he buys a motorcycle. In the former case, he is thought of as taking good care of the car, and in the second of the motorcycle. The "referent" for "it", in each case, is whatever Stewart bought, in the Mental Model.
In (32), the difference is that the word "and" does not indicate the existence of two situations, and thus the singular pronoun "it" is inappropriate to the non-singular antecedent.
(37) Yes, there certainly is night life in Jerusalem. Unfortunately, this weekend she is in Tel-Aviv.
Hintikka and Kulas (ibid) state that "she" is "obviously" not coreferential with "night-life". However, that is the very point of the joke. Jokes often distort language, and it makes no sense to recognise this fact, and at the same time to label as "ungrammatical" or "unacceptable" that very distortion which makes them into jokes. Moreover, the mere fact that "she" is feminine and "night life" is neuter is not sufficient to render them noncoreferential. Coreferentiality is a semantic notion which is independent of such considerations as grammatical gender.
(38) Bill is not the only man who loves his wife.
This example, which the authors cite in relation to the ambiguity between the "bound" and "referential" readings, is also not a problem. The ambiguity is quite simply caused by the fact that two NP's ("Bill" and "man") are available to act as antecedents for the pronoun "his". If "Bill" is the Antecedent, then you have the "referential" reading. If "man" is the antecedent (irrespective of the bracketing that Formal systems set up in such circumstances), then you have the "bound" reading.
(39) Both Jill and Jane told Jim about herself.
Here the authors argue that, if "herself" is coreferential with both "Jill" and "Jane", then "Jill" and "Jane" must be coreferential with each other. However, the Mental Model for (39) again involves two separate scenarios, with "herself" referring to a different lady in each case.
(40) If you extend an invitation to every friend of yours, someone will disregard it.
Here, because of the quantifier, "it" is not straightforwardly coreferential with "an invitation". However, the reader can doubtless see for himself how easy it would be to adapt Johnson-Laird's way of dealing with quantifiers to this example.
(41) Did you hear that Greta Garbo has been seen in public ?
Who is she ?
Here the authors complain that "she" does not even refer, let alone refer to "Greta Garbo". However, we have discussed similar examples above, and the solution is that "she" refers to "Greta Garbo" characterised as being some person referred to by the first speaker. The second speaker is requesting information from the first speaker which will allow the former to expand his concept of, or "file on" Greta Garbo.
A final type of pronoun (see (42) below), what I call a "Phonaphoric Pronoun", is discussed in Chapter Three:
(42) The man who hid his paycheck from his wife was wiser than the man who gave it to her unopened.
6.Conclusion
In this chapter, I have argued for a return to the notion of Coreference, though this is to be understood in conjunction with a slightly more sophisticated model of reference than is usually implied. As far as the various types of pronouns are concerned, I have relied on a notion of "Mental Model" that is powerful enough to accomodate the hitherto problematic types I reviewed above, apart from those that come under the Phonological Identity Criterion. Though the Mental Model is conceptually "powerful", this power is subject to the limits imposed, not only by linguistic data, but also by psycholinguistic theory and experimentation.