CHAPTER THREE: TYPES OF ANAPHORA.
1.Introduction
Excluding Ellipsis (which is dealt with in Chapter
Six), the present chapter is an attempt to decide what types of Anaphora
proper exist. The point of entry is Verb Phrase Anaphora (VP Anaphora), building,
in part, upon work that has already appeared which classifies this type of
Anaphora (see below). As puns figure prominently among the data I use to
discuss VP Anaphora, I start here with a short statement of what I understand
puns to involve.
2.What are puns ?
As I understand puns, they involve misleading the hearer/reader by at first making him assume that phonological identity is indicative of semantic identity (which is usually a safe assumption for the hearer/reader to make in the cooperative activity that communication is usually assumed to be). When the hearer/reader realises that he has been "tricked", he experiences surprise, and then (hopefully) amusement.
3.Identity criteria for VP anaphora
My starting-point is Cormack (1981), which is essentially an attempt (since repudiated by the author herself -- personal communication) to provide an explanation for the following example from Dahl (1973) in terms compatible with the traditional Logical/Generative approach:
(1) a. John realises that he is a fool
b. but Bill does not,
c. even though his wife does.
The problem, for Cormack, is that we have here
two successive clauses containing VP Anaphora, both of which (in the traditional
analysis in terms of deletion under semantic identity of predicates) must
have been derived by deletion of one and the same logical predicate present
as the antecedent in (1a). The cause for concern is that (1b) must have been
obtained by the deletion of:
(2) !x (x realises x is a fool),
whereas (1c) could only have been derived by deleting:
(3) !x (x realises Bill is a fool),
(under the reading:
(1') a. John realises that John is a fool
b. but Bill does not realise that Bill is a fool,
c. even though Bill's wife does realise that
Bill is a fool).
However, if we assume that VP Anaphora is not always a matter of semantic, but sometimes of phonological identity (i.e. that phonological and semantic identity are both sufficient, but not necessary conditions for VP Anaphora), then (1) poses no difficulty, other than that of accounting for the low-level, morphological change that the verb undergoes between (1a) and (1b). In this view, the antecedent which is carried over from (1a) to (1b) is the phonologically-stored equivalent of:
(4) realises that he is a fool,
so that the expanded version of (1b) would be:
(5) but Bill does not realise that he is a fool;
and the sequence:
(6) realise that he is a fool
is then carried over from (1b) to (1c), so that the "de-anaphorised" version of (1c) is:
(7) even though his wife realises that he is a fool.
Thus a Phonological Identity Criterion (PIC) approach would have the advantage of providing a relatively simple and natural account of the otherwise problematic example (1). There is other evidence for the PIC, however, and some of it comes in the form of puns:
(8) John ran into a bank, and so did Bill.
Cormack (1981) cites this example, but overlooks the possibility of punning, which obviously exists, because of the ambiguity of both ran into and bank. With implicit reliance on the C/P 17 distinction (see Chapter One) between literal and figurative language, Autonomist linguists tend to shy away from examples which incorporate figures of speech, and the like. However, as was explained in the previous section, a pun differs from "normal language" only in that it uses Phonological identity to mislead the hearer into expecting (temporarily) a case of semantic identity. Puns do not necessarily involve VP Anaphora, nor do they usually occur in sentences which are syntactically odd in any way. Thus there is no good reason to discount the validity of puns as evidence where the syntactic nature of VP Anaphora is at issue. Indeed, because of the peculiar interplay between ( Noun-Phrase) Semantic and Phonological features that puns involve, they are particularly crucial evidence in the matter at issue.
Cormack's literal-minded approach here is in line with Generative tradition. McCawley (1981:7), for example, also seems to ignore the possibility of punning. He develops the argument that VP Deletion and One-Anaphora can be used to determine whether a given word is ambiguous or vague:
(9) Maxine married a bastard, and then Frieda married a bastard.
(10) Maxine married a bastard, and then Frieda married one.
(11) Maxine married a bastard, and then Frieda did.
(12) Susan caught a carp, and then George caught a carp.
(13) Susan caught a carp, and then George caught one.
(14) Susan caught a carp, and then George did.
McCawley's point is that (9) and (12) can both be understood four ways, because of the combinations of the two ways (at least) that "bastard" and "carp" can each be
understood: "bastard" can be understood as equivalent to "nasty person", or to "illegitimate person"; "carp" can be understood as meaning "male Cyprinus carpio" or "female Cyprinus carpio". He then goes on to claim (wrongly) that (10) and (11) have only two readings each, whereas (13) and (14) each have four readings, and that this demonstrates that "bastard" is truly ambiguous as between two distinct lexical items, whereas carp is merely vague as to sex, rather than being properly ambiguous.
In fact, we are now in a position to see that, although McCawley's conclusion may well be correct, his reasoning is not. Because of the possibility of punning, which he overlooks, all six of the above sentences can have four meanings each. However, it would still be possible to argue that there does exist the important difference between "carp" and "bastard" that only in the case of the former can these four readings be arrived at without recourse to punning. So the merits or otherwise of McCawley's overall argument as to the difference between vagueness and ambiguity (which I do not propose to go into here) remain substantially unaffected.
It cannot be entirely coincidental that McCawley (1981) both subscribes to the forma-logic approach to VP Anaphora and overlooks the possibility of punning in (10) and (11). Sag, the originator of this formal semantic approach, was himself by no means breaking new ground when he claimed (1976:533) that the following were two-ways, but not four-ways ambiguous:
(15) a. John likes flying planes, and Bill does too.
b. Betsy divulged when Bill promised to call me, and
Sandy did too.
c. The chickens are ready to eat, and the children
are, too.
In fact, punning can render all these examples four-ways ambiguous: Example (15a) can be uttered as a pun if John likes flying planes in his capacity as a professional pilot, and Bill likes watching planes passing overhead, as well as vice-versa, plus two readings whereby John and Bill each have the identical one of these two predilections in each reading in turn. Likewise, (15b) can be uttered as a pun if Betsy divulged the time that Bill made his promise, and Sandy let slip what time he was allegedly going to call me, as well as vice-versa, plus two readings whereby the readings of each conjunct are identical as to the predicate. And example (15c) can be said as a pun if in a black comedy infants had been chopped up for chicken-feed, and chickens chopped up for infant-food (plus three other readings as per the other two examples).
So far, the evidence is compatible, not only with a Phonological Identity Criterion, but also with a Lexical Identity Criterion. The following examples enable us to choose between these two possible criteria:
(16) The farmer is sowing, and my mother is sewing, too. (17) The farmer is sowing, and my mother is, too.
If we imagine these as spoken utterances, or actually say them out loud, we can see that both could be uttered as puns. Note that the presence of "too" is necessary, in order to link the two clauses together in a way whose inappropriateness is the essence of the pun -- i.e. the verbs concerned are Phonologically, but not Semantically similar. These examples demonstrate that both puns and VP Anaphora depend on Phonological, rather than Lexical identity, as "sew" and "sow" are Phonologically, but not, of course, Lexically identical.
Two related issues crop up here, which I would like to mention in passing, but not discuss in detail: one is the question of whether intonation and stress are relevant to the notion of Phonological Identity, as per the PIC. My inclination (inspired by an example from Max Cresswell) would be to assume that they are indeed relevant. The other matter is that of written puns and written VP Anaphora: is the latter subject to a Graphological Identity Criterion, or is the visual signal converted mentally into an aural one that is subject to the PIC ?
Another example comes from Grinder and Postal (1971) (who try to account for it by means of an ad hoc device):
(18) Pete painted his house, and so did his mother.
Grinder and Postal (op.cit.) see this example as having only the reading whereby Pete's mother painted Pete's house, and no reading whereby she painted her own house. By contrast:
(19) Pete painted his house, and so did his father
has both the reading whereby Pete's father painted his own house, as well as the reading whereby Pete's father painted Pete's house. This contrast between (18) and (19) tends to support the PIC over the traditional account of VP Anaphora as involving only a Semantic Identity Criterion.
Sag (1976) provides a counterexample to his own approach, which serves to strengthen the case for the PIC:
(20) They caned a student severely when I was a child, but not like Miss Grundy did yesterday.
As Sag points out, the first clause in (20) can be interpreted generically, at the same time as the deleted "a student" in the second clause can be interpreted specifically.
Note that this structure is also possible when the non-generic clause precedes the generic one:
(20') Miss Grundy caned a student severely yesterday, but not like they did when I was a child.
Finally, though all our examples up to now have involved VP Anaphora, it seems that some pronouns may also be subject to the PIC:
(21) The man who hid his paycheck from his wife was wiser than the man who gave it to her unopened.
It seems clear that the it refers back, not to the semantic content of of his paycheck, but to the phonological sequence it represents, which then derives its Semantic representation from the context of the second clause. The same phenomenon occurs between her and his wife. Lyons (1977,Vol.II) calls attention to this similarity between "pro-verbals" and "pronouns of laziness", though he does not explain it in Phonological terms at all. I shall call such cases Phonaphoric Pronouns, as the term "Pronouns of Laziness" seems to have both this narrow meaning and a broader one. Please note that puns can be overlaid on such sentences as (21), e.g.:
(21') The man who fed his ass oats was wiser than the man who wiped it with a $10 bill.
This strengthens the parallelism between Phonaphoric Pronouns and VP Anaphora.
Thus we have grounds for suspecting that it might be worth looking for further generalisations across the barrier that might otherwise seem to separate NP Anaphora from VP Anaphora.
4.Types of anaphora
I am not claiming that the PIC replaces the Semantic Identity Criterion entirely. The two criteria must coexist in some way, as there are many examples along the lines of the following, where it would not be plausible to claim that the PIC was involved in any way whatever:
(22) He gave me a quarterly report last week from a certified public accountant, showing that the estate has increased to date 22% above its original value, so I guess my son will get his million. And I will, too.
Under the most plausible interpretation of the last sentence, the antecedent is get x's million (where the referrent of x is the same as the referent of the subject of the relevant clause); the antecedent derivable by using the PIC, i.e. get his million, is obviously not appropriate here, where it seems that a Semantic Identity Criterion must be involved.
There are already two strands within the Generative tradition which argue for a bifurcation of Anaphora, and so I shall discuss these two below, assess them, and propose an analysis that is compatible with both the arguments adduced above and the "data" (though my attitude to Autonomist data remains as stated in Chapter One) which these two Generative strands discuss. Specifically, I will refer to Grinder and Postal (1971), who distinguish between identity of sense and identity of reference Anaphora, and Hankamer and Sag (1976), who distinguish between deep and surface Anaphora.
Grinder and Postal (op.cit.) illustrate their distinction with examples such as the following:
(23) Harry insulted his wife, and Bill did too.
(24) Max is looking for immortal zebras that can fly but there are no such zebras.
(25) Larry married a nurse who owned an iguana but Pete did not marry one.
(26) The caterpillar will be able to fly when it is a butterfly.
(27) The caterpillar will be able to fly when the caterpillar is a butterfly.
Grinder and Postal (ibid) consider that (23)-(25) involve Identity-of-Sense (IS) Anaphora, whereas (26) involves Identityof-Reference (IR) Anaphora. The distinction between IR and IS Anaphora is highlighted by (27), as the latter contains a selfcontradictory second clause, whereas (26) does not. The IR-Anaphoric "it" in (26) carries over the identity of the caterpillar/butterfly, which is treated as remaining constant (ontologically) despite its metamorphosis. The use of the explicit noun "caterpillar" in the second clause of (27), by contrast, focuses attention on the physical characteristics, rather than the ontological "essence", of the insect in question, and so creates a self-contradiction in that clause, according to Grinder and Postal, at least. Some may disagree that this actually involves a self-contradiction. Certainly, it is much easier to see (27) as a self-contradiction, than (26).
Despite that apparent self-contradiction, however, I find (27) comprehensible, though stylistically odd (because of the existence of the "normal" alternative (26)). Perhaps one can draw the conclusion from such examples that what I will call Resumptive Phrases (RP's), i.e. explicit or anaphoric phrases which refer back to an entity, description or action already mentioned in the discourse, are not just simple cues to the recall of some pre-existing entry/file in the Mental Model of the Speaker/Hearer -- they are active in shaping the perspective in which that entity, description or action is presented at that particular point.
Those examples which Grinder and Postal (ibid) see as involving IS Anaphora arguably involve the PIC, as the reader can see for himself by substituting the Phonological form of the Antecedent for the Anaphor in (23)-(25). It is worth noting at this point that, while structures such as (24) and (25) always involve the PIC, structures such as are crucially involved in (21) and (22) can be interpreted by the hearer as involving either the PIC or the SIC, according to context.
In a footnote, Grinder and Postal (ibid) mention a problem which is always present in the background in Autonomist accounts of Anaphora: because of the C/P 10 distinction between Sentence-Grammar and Pragmatics, Autonomists generally discuss only cases where the Antecedent is in the same sentence as the Anaphor, on the tacit assumption that if they are in different sentences, completely different explanations must be given for the relationship between them. This rather unnatural duplication of rule-types is avoided in Cognitive Linguistics, where the sentence-boundary is not required to assume the same degree of theoretical importance as in the writings of most Generative Grammarians.
The examples of this problem which Grinder and Postal (ibid) cite involve VP Anaphora, and Spencer-Smith (1987) mentions the same problem in connection with NP Anaphora:
If propositions come in sentence-sized chunks, we would have to regard the use of multi-sentential discourse as the expression of a sequence of distinct and discrete propositions. But it is implausible to suppose that a two-sentence discourse like
(28) Pandora has a jar. She opens it,
is essentially different from a single sentence
(29) Pandora has a jar, and she opens it,
simply on account of the full-stop.
The following alternative suggests itself: sentences in declarative discourse successively contribute to a complex representation of the world. The possibility exists for words to contribute to units of discourse larger than single sentences. Inter-sentential anaphora provides a good example, as in (28) we would normally interpret She, in the second sentence, as referring to Pandora, the subject of the first. So correct understanding of a discourse requires not only grasping the "vertical" connections between words and the world, but also the way sentences link together to express a complex propositional structure. This structure not only embodies the content of the discourse, it also forms part of the context for interpreting subsequent utterances. (ibid, 1-2)
There have, of course, also been developments out of Generative Grammar, as well as in other linguistics schools, that have taken the same approach from a linguistic, rather than a quasi-philosophic perspective. There seems to be a convergence of approach, here, between philosophers (whose subject-matter has, of course, always been hard to define or delimit), psychologists (who are inherently involved with what Spencer-Smith refers to as "representations of the world" in living creatures), and linguists, who cannot deny that mental representations play a role in language -- though some try to avoid referring to such representations as much as possible in their linguistic work.
It seems to me that, for these three groups (under the aegis of the Competence-Performance Distinction) to carry out what is essentially very similar research in parallel is a tremendous waste of resources. I do not think it even helps very much if they carry out these separate activities in the context of university departments that happen to be named "Cognitive Science" departments. Cognitive Linguistics, on the other hand, has the potential to become a truly integrative approach which can help to solve this problem. Once some Metatheoretical foundations have been agreed upon (and it is my hope that the present work is a modest step in that direction), then it is possible that Computer Scientists, Cognitive Psychologists, Linguists, and Philosophers can all work together to make "Cognitive Science" a truly cooperative and concerted venture.
Cornish (1986) equates Identity of Sense Anaphora with his own notion of Noun anaphora, by which he means,
...those anaphoric relations which may hold between a non-lexical "pro-form" and the head noun or nominal group ... of a Noun Phrase (ibid,9).
He lists a number of examples -- some of which do not seem to me to support his identification of IS Anaphora with Noun anaphora, e.g..
(30) "...Before then, the generally accepted "external
view" of Africa's history had been that there was none..." (B.Davidson 1984, Africa in History, Paladin, p.13).
(31) "....Some industries were doing well - but there were not enough of them..." (BBC Radio 4, 19.8.80).
Though both (30) and (31) may rightly (it seems to me) be considered to be instances of IS Anaphora, it is clear that none (in (30)) is Anaphorically related to Africa's history, rather than just to history, and that (31) does not mean that there were not enough industries, as such -- just that there were not enough industries as characterised by the whole Antecedent NP (those that were doing well). Thus (30) and (31) are instances of IS Anaphora, without being instances of Cornish's "Noun anaphora".
Hankamer and Sag (1976) and Sag and Hankamer (1980) essentially start where Grinder and Postal (1971)'s Footnote 4:
...the identity conditions on these processes are of at least two distinct sorts. One class of identity requirements must refer to the Semantic Representations. The other involves superficial syntactic properties
leaves off.
The following are the examples that Sag and Hankamer (1980) use to illustrate their distinction between Deep and Surface Anaphora:
(32) SURFACE ANAPHORA
(a) Robin will like Sandy. Chris will [ ], too. (VP Ellipsis).
(b) Robin will like Sandy. We're not sure why [ ]. (Sluicing).
(c) Robin will like Sandy, and Chris [ ] Pat. (Gapping).
(d) Robin will like Sandy. [ ] not [ ] this week. (Stripping).
(33) DEEP ANAPHORA
(a) Robin entered the room. SHE was angry. (Definite Pronominalisation).
(b) Terry left the room. I resented IT. ("Sentential" IT Anaphora).
(c) Lee smashed the Ford with a sledgehammer. Pat DID IT too. (DO IT Anaphora).
(d) Kim wanted to leave. I let her [ ]. (Null Complement Anaphora).
N.B. The capital letters highlight Anaphoric elements, rather than indicating location of stress.
Sag and Hankamer (1980) reiterates Hankamer and Sag (1976)'s position that Deep and Surface Anaphora (renamed "Model-Interpretive Anaphora" and "Ellipsis", respectively) can be distinguished by the facts that: (a) only Deep Anaphora can be used deictically; (b) only Surface Anaphora exhibits the "Missing Antecedent" phenomenon as per Grinder and Postal (1971); and (c) only Surface Anaphora requires parallelism in syntactic form between Anaphor and Antecedent.
Unlike Grinder and Postal (ibid), Sag and Hankamer (1980) gives clear criteria for distinguishing the two proposed types of Anaphora. I propose to evaluate these two taxonomies of Anaphora, and to compare them with my own PIC/SIC taxonomy, as well as Williams (1977)'s rival distinction:
Some pronouns are dominated by NP (HE, ONE, IT, and the missing N in such NP's as JOHN's [ ]) and some are not (such as missing VP's, SO, SUCH). It seems that if an anaphoric element is immediately dominated by NP, then it can be pragmatically controlled, and that it otherwise cannot be. (Williams, op.cit.: 695-6).
Williams' generalisation misses the point at issue by treating it as a basically Syntactic matter. The root of the matter is that entities are easier to refer to nonlinguistically than are actions, or other states of affairs typically represented by VP's. This means that entities are more likely to be available to "pragmatically control" Anaphors than are actions. Nevertheless, counterexamples to Williams' proposal do occur, e.g.:
(34) (Sam and Bill are watching a movie which is set in Roman times. In the Colisseum, one gladiator has another on the ground with his sword at his throat. The spectators are waiting to see whether the Emperor will give the thumbs up -- allowing the loser to live -- or the thumbs down -- meaning that the loser should be killed. Sam theatrically turns both of his thumbs down.
Bill says: I think so too !!
Let us now turn to the three criteria listed by Sag and Hankamer (1980) (we will retain the outdated Deep/Surface terminology for the sake of brevity):
As far as the deictic/pragmatic control criterion is concerned, counterexamples are not hard to find:
(35) (Context as per (32))
Bill: "I think the Emperor will, as well !"
This is a case of VP Ellipsis as per (32a), which is supposed to involve Surface Anaphora and thus to be incapable of having deictic antecedents.
As to the syntactic parallelism criterion, I think most native-speakers of English would find the following example hard to comprehend in the relevant (or indeed any) sense:
(36) Paul wanted to kiss Kim. I let her [ ] (= be kissed
by Paul).
This sentence involves "Null Complement Anaphora" as per (33d), which is supposed to involve Deep Anaphora, and therefore not to require syntactic parallelism between Antecedent and Anaphor. It is an example which appears to show that at least some cases of Deep Anaphora do require such a parallelism, in that the actual Antecedent is active, whereas the structure of the Anaphoric clause requires a passive Antecedent -- except that I myself would be inclined to call it Phonological, rather than Syntactic parallelism.
As far as the remaining criterion (relating to "Missing Antecedents") is concerned, let us look at an example of what Grinder and Postal (op.cit.) consider to be an instance of such a phenomenon:
(37) Harry doesn't have a wife but Bill does and she is a nag.
The Antecedent which is "missing" is that for "she".
Consider the following two counterexamples to this third criterion:
(38) Leroy gave his wife a skunk for Christmas, and the following year Paul did it too. She did not take it as badly as Lee's wife had done.
(39) Chad wanted to go for a drive. I let him, but he couldn't get it out of first gear.
From a Comprehensibility point of view I see no problem with either of these examples, although (38) involves DO IT Anaphora and (39) involves Null Complement Anaphora, and so neither is supposed to permit the "Missing Antecedent" phenomenon to occur. I am not particularly interested, here, in going further into a discussion of "Missing Antecedents". All that I want to demonstrate here is that none of the three criteria appears to stand up to close scrutiny.
It is interesting to note that Sag and Hankamer (op.cit.), though writing from a Generative perspective, see their proposals as a Performance, rather than a Competence model of Anaphora. Unfortunately, they neither discuss the Competence-Performance Distinction per se, nor do they give clear reasons for their preference for the term Performance over Competence in this context. Since they appear to be referring to what my Chapter One calls C/P 1-4, (a group of distinctions I find unnecessary, and even deleterious),however, I find myself in apparent sympathy with their overall approach to Anaphora.
The authors borrow Johnson-Laird's notion of a "Mental Model" (see Johnson-Laird (1983)) in their explanation of the way that discourse Anaphora is (in Psychological terms) represented, produced, and comprehended. What they previously termed "Deep Anaphors" are now thought by them to refer back to items in the Mental Model of the ongoing discourse which each party to the discourse has constructed in his mind -- rather than to any specific part of the preceding discourse itself. What they had previously called "Surface Anaphors" are still thought by them to have actual linguistic antecedents in some part of a propositional representation.
They base themselves partly on experimental results produced by Johnson-Laird, purporting to show that the level of propositional representation is psychologically real in language processing. However, an examination of Johnson-Laird's argument reveals that the evidence is equally compatible with a level of phonological representation being the one that is psychologically real.
The notion of a proposition has been (it seems to me) somewhat taken for granted in Autonomist Linguistics since its annexation from Philosophy, where it originated. I do not see a use for this notion as far as language comprehension is concerned. I would propose an alternative model of processing, whereby the basic units (e.g. arguments and functions) of semantic representations are integrated as far and as fast as possible into the Mental Model, without going through any intermediate "propositional" stage. This would account more naturally than the propositional model for the large proportion of utterances which are elliptical, anaphoric, or incomplete is some way.
We saw (above) that Spencer-Smith considers that the sentence is not a large enough unit, as far as some aspects of language-comprehension are concerned -- similarly, one could argue (in the wake of Psycholinguistic experiments carried out by Marslen-Wilson and associates) that the sentence is too big a unit as far as other aspects of language-comprehension are concerned. Thus Cognitive Linguistics, I predict with some confidence, will eventually see fit to demote the notion sentence from the preeminence it currently enjoys in Autonomist Linguistics -- or even banish it entirely to the realms of Normative Grammar. It would not be entirely coincidental for both the sentence and the proposition to be so demoted, as in Generative Grammar the proposition is the semantic representation which corresponds to a given (syntactic) sentence, and the two notions are linked historically, as well.
Sag and Hankamer (1980) cite experimental evidence that shows that there is a critical time-span separating antecedent and anaphor, which, if exceeded, results in anything other than model-based Anaphora becoming impossible. In other words, oral PIC-based anaphors must be relatively close in time to their antecedents. We could hypothesise that the PIC is the antecedent-hunting strategy ,of first resort (demonstratives apart, as they are a special case), but that the memory-type that the PIC involves has a limited capacity and a high turnover-rate.
Thus, rather than proposing that there is a clear Linguistic demarcation (which seems implausible to me on a priori Psychological grounds) between structures that involve the PIC and others which involve the SIC, I prefer to hypothesise that the short-term verbatim (phonological) memory is first searched for a plausible antecedent, and recourse is had to the Mental Model if and only if the first search is unsuccessful. In fact, the experimental recency effects apply to verbatim recall tasks, and this fits in much better with my proposal for a phonological representation than it does with the traditional propositional representation.
Sag and Hankamer (ibid) mention also the case of indexicals, which fit neatly into neither their proposals, nor into my own. They cite the following examples:
(40) A: Do you think they'll like me ?
B: Of course they will [ ]. ([ ] = like you).
(41) A: Are you coming over here ?
B: Yes, I am [ ] . ([ ] = coming over there).
Both propositionally and phonologically, me would be represented differently from you (in (40)), and here differently from there (in (41)). The solution which Sag and Hankamer propose is essentially similar to the one I will myself suggest:
Indexicals... will be assumed to be interpreted directly by immediate reference to the discourse model; we will further assume that this interpretation takes place simultaneously with the construction of propositional representations.... Under this interpretation, indexicals are just those elements for which there are special interpretation rules linking a constant term in the propositional representation with an entity in the discourse model according to discourse-situational conditions. (ibid, 12 (MS)).
From a functional point of view (and Cognitive Linguistics is fundamentally compatible with Functionalism), it makes sense to assume that the human organism should give high temporal priority to translating indexicals into a form relevant to itself. Thus, I venture the proposal that the hearer does indeed automatically translate you into me, and here into there, etc. (as appropriate to the physical setting of the particular discourse) -- but at the phonological level, rather than in some propositional form. My proposal at least has the advantage of involving fewer hypothetical levels (there being no propositional level).
Like theirs, my explanation involves intervention by the Mental Model in the composition of the "lower" (propositional/phonological) representation, as indexicals do not have deterministic or fixed translations -- e.g., the Mental Model has to provide the information that the hearer in question is the addressee, or one of the addressees, before the indexical you can be safely translated into me. If not, then him, or some other pronoun, might be more appropriate.
The output of this phonological adjustment of indexicals then becomes the input to the Mental Model and to the Phonological Buffer Memory (PBM), as I will call it. Whereas this input is interpreted semantically before being incorporated (probably word-by-word, rather than in clausal chunks) into the Mental Model, it passes without further change into the PBM.
We might recall at this point that we also provided for "tampering with the record" with respect to verbal Morphology in the case of (1) (above). Though these two kinds of "tampering" are dissimilar, accepting the principle that tampering is theoretically acceptable in one type of case makes it easier to contemplate the possibility that it might be involved in the other type of case, as well. Though these two cases make the overall theory of Anaphora less "simple", it is important to note that Cognitivists do not need to place so much (rhetorical) emphasis on aesthetic simplicity as Autonomists tend to do. This is because a Cognitivist (in my view) does not rely on the CPD (specifically, C/P 1-4) to isolate his theories from interplay with the experimental findings of Psycholinguists. Since any proposals I make are open to confirmation or otherwise by experimental research, it would be inappropriate to worry about how simple they are. Ultimately, it matters much more whether they are shown to be correct or incorrect.
5.Discussion
Though there is a tendency for definite pronouns to be subject to the SIC, rather than the PIC, there are exceptions to this, such as Phonaphoric Pronouns (as we have seen). The reason for this is that definite pronouns can have up to three of the following semantic characteristics : definiteness, anaphora, reference. Of these, definiteness is indispensible (by definition). Anaphora is indispensible if it is defined in such a way as to include "Exophora", i.e. deixis, as well. But reference is not indispensible, so that there are cases (eg. Phonaphoric Pronouns, and when Propositional Attitudes are involved) where the definiteness is coupled with Identity of Sense Anaphora, rather than Identity of Reference Anaphora. (Pronouns bound by quantifiers are also generally regarded as being non-referring, as "referring" is usually understood). We can note here in passing the similarity with Definite Descriptions, which (since Donnellan (1966)) have been recognised as also having uses where definiteness occurs without specific reference. There is the additional similarity that Definite Descriptions can be used Anaphorically. However, I do not mean to carry this analogy so far as to suggest that the so-called "Attributive" use of Definite Descriptions is subject to the PIC.
Leaving deictics aside as being a special case, it would appear to be the case that the core of the distinctions that Grinder and Postal (1971), Hankamer and Sag (1976) and Williams (1977) set up in fact has to do with this definiteness of definite pronouns. All the examples of Surface Anaphora in Hankamer and Sag (1976)'s examples involve what may broadly be termed "Ellipsis" -- and, of course, Ellipsis cannot be definite. And the pronouns ("such" and "one") that Grinder and Postal (1971) use as examples of IS Anaphora are both indefinite. By contrast, all but one of the examples provided by Hankamer and Sag (1976) involve Definite Pronouns.
The one exception is a case of so-called "Null-Complement Anaphora". I would argue that this actually involves two distinct phenomena. Only one of these is actually Anaphoric in nature, and it actually involves Surface, not Deep Anaphora. Compare:
(42) The oats had to be taken down to the bin, so
a. Sandy volunteered.
b. I let Sandy.
Structures similar both to (42a) and to (42b) are cited by Hankamer and Sag (1976) and Sag and Hankamer (1980) as examples of "Null Complement Anaphora". However, (42a) seems more comprehensible than (42b), as there exists a more likely alternative to (42b), namely:
(43) I let Sandy do it.
Thus, using Sag and Hankamer (1980)'s criteria, it seems preferable to assume that the structures represented by (42a) and (42b) are in fact distinct, and that the structure in (42b) is a case of Surface Anaphora, as it seems to require syntactic parallelism between Antecedent and Anaphor. Structures like (43) are also more comprehensible than structures like (42b) in contexts where deixis is involved:
(44) (Hankamer fires a gun at stage right. A blood-curdling female scream is heard).
Sag: a. I should have volunteered.
b. I shouldn't have let him. c. I shouldn't have let him do it.
As with (42), the intransitive (44a) type of "Null Complement Anaphora" seems to admit of "pragmatic control", but the transitive type (44b) does not. The latter seems more appropriate to someone talking to himself in self-reproach, whereas (44c) is much more appropriate and comprehensible when talking to someone else.
In the case of intransitive "Null Complement Anaphora" (e.g. (44a)), I see no reason to assume that any Anaphora or Ellipsis is involved at all. One could argue that (44a) is very succinct, but I doubt that a convincing case could be made that Ellipsis, rather than mere succinctness is involved in such cases (assuming such a distinction can be drawn with any lucidity).
With this reanalysis of so-called Null-Complement Anaphora, we have completed our examination of the three anaphoric dichotomies we have been discussing, and reached the conclusion that the only valid distinction is a fourth one -- that between Definite and Indefinite Anaphora. We found Williams' distinction to be both limited and incorrect. Postal and Grinder's distinction was too imprecise to be useful. Of Hankamer and Sag (1976)'s three criteria for distinguishing "Deep" from "Surface" Anaphora, we found the Missing Antecedent one to be irrelevant, and the Pragmatic Control one to be incorrect as stated. The remaining one -- Syntactic Parallelism -- is now irrelevant or meaningless, given that all remaining instances of "Deep Anaphora" involve Definite Pronouns. A Definite Pronoun has no "structure" which is capable of either being, or not being parallel with anything else, in a syntactic sense.
This Definite/Indefinite distinction does not parallel the PIC/SIC distinction. This is because Phonaphoric Pronouns, which, I assume, are subject to the PIC, are definite, whereas VP Anaphora is indefinite, and can be subject to either the PIC or the SIC.
6.Further discussion
The most important conclusion I would like to draw from the above discussion has to do with C/P 14: the distinction between "linguistic" and "nonlinguistic" mental faculties. Adopting a PIC explanation for even some kinds of Anaphora makes quite specific claims about memory, in a way that more traditional accounts of Anaphora do not.
If we were still to assume, despite the arguments discussed above, that the only relevant identity criterion for VP Anaphora was a semantic one, then this would amount to no specific claim about the type of memory involved in the processing of VP Anaphora. That is to say, the Antecedent might conceivably be stored in Phonological, Syntactic, Logical, or some mixed or intermediate form of one or more of these levels of representation, to be subsequently retrieved during the processing of the Anaphor.
Since, in processing terms, the semantic/logical form is the goal, and is recoverable/derivable directly or indirectly from any of the other, more surfacy levels, it would not be unreasonable to argue that the way that such anaphors and antecedents are processed is in principle irrelevant, since only a description of Competence was intended, and, in fact, that this was not something that purely linguistic evidence could be expected to shed any light upon.
However, as the PIC, which implies the processing of phonological data just as surely as conventional theories implied the processing of semantic data (in whatever form it was actually stored), does make a specific claim about the type of memory storage involved in the processing of certain types of Anaphora. This is because we cannot uniquely reconstruct the phonological form of an utterance from its semantic representation, whereas the converse is certainly sufficiently possible for human communication to be useful and adaptive to our species.
Thus we are in a position to hypothesise that memory for the phonological form of language is involved in at least some language-processing. This entails that phonological memorisation is constantly taking place during language-processing (it being obvious that a listener normally has no way of knowing that a pun is about to be made, so that he can make special provision for it).
There is evidence in the Psycholinguistic literature to the effect that a working memory exists which stores (possibly small) chunks of language for processing purposes, and I am aware of no evidence that this could not carry out the storage in a phonological form. Bard et al. (1988), for example, argue that hearers routinely recognize a substantial proportion of the words they hear only after the acoustic offsets of the words involved.
Within the framework adopted here, ... late recognition simply emerges from the regular functioning of a mechanism in which alternative hypotheses are entertained simultaneously until such a time as there is sufficient evidence to select one hypothesis above the others.... For a substantial proportion of the time, (the) ... processor must be pursuing hypotheses across a stretch of input corresponding to more than one word. (ibid, 407)
In all likelihood, the above model involves the existence of a special-purpose memory which contains the readily-accessible words against which the alternative hypotheses can be matched until such a time as an interpretive decision is reached by the hearer. I would hypothesise that the words are stored in phonological form in this memory, and that the PIC operates on the basis of the words thus stored.
What is perhaps most interesting here is perhaps not so much the claim itself, that Phonological memory does exist and play a role in language-processing -- references to sensory memory are not rare in the literature -- but, rather, the fact that purely Linguistic evidence, such as had hitherto been brought to bear on hard-core "Competence" matters only, has been found to have consequences and implications outside the "purely linguistic" sphere: memory, after all, is not a mental faculty restricted to the processing of language.
Cognitivist Linguistics is most radically opposed to Autonomist Linguistics as regards the relationship between Linguistics and Cognitive Psychology, which relationship (for Generativists) is encapsulated in the C/P 14 distinction. While Cognitive Linguists would probably admit the obvious fact that Language does have some characteristics that are unique to it, they might tend to prefer the research-strategy of looking for the connections between Language and other mental processes, as opposed to that of concentrating on the supposedly unique aspects of Language.
It seems to me that, if it is accepted that the PIC plays a role in Anaphora, then this must count as a contribution towards the case that Cognitive Linguists in general, and this book in particular, seem to be gradually building up -- to the effect that the Autonomist approach is unproductive, counterintuitive, unlucid, and metaphysical.
7.Conclusion
Any thoroughgoingly Cognitive approach to Anaphora will have to account for the results of experimenters such as Bates et al. (1980), who conclude:
This suggests that disambiguation of reference is not the only factor governing choice of surface forms.... It may be that the decision to use an explicit form of reference involves the need to highlight, dramatize, mark
topic shifts and important points, and generally "stage"
utterances in a way that assures the listener's attention.... Certainly the notion of "drama" will be more difficult to formalize than the notion of referential ambiguity. But a full linguistic theory of pronominalization and anaphora may need to include such concepts if we are to describe and explain the rules of reference in natural discourse. (ibid,47).
Nothing in the present work should be seen as detracting from the sentiments expressed in the above passage.