This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
or e] -
]t u [ e] t, will be added to Opt. The •t• operation, 'intersection with W;', distributes over the union operation representing or when inf(W;) - inf8(W;). i.e. (46) inf ]t c [e] t [ ] t 11 [ e] t # ] t
u
[ e])t.
The effect of permission is therefore to add ([] u [ e])t to Opt. Note well that P4/K2 prescribes that the worlds added to Opt by permissions of <1>, e, or ore are not the whole ofinf
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Or denotes boolean set-union, which operates on its arguments before feeding the result to the state-changing operation. The problem now is that [ or e] , computed directly, is satisfied if either one of its disjuncts is. This looks fine, at first blush, if and e are equally reprehensible. And bad if they are not.
A. Merin
1 17
according to P4/K2, physico-nomologically independent; i.e., the intersection and symmetric difference of[ ] and [0] are both non-empty in Pos: ([]\[0]) 11 Pos # <jJ # ([ 0 ]\[]) 11 Pos By definition we also have and [ e] t �dr [0] 11 inf8(Wi); (48) a. [
(49) a. b. c. d.
[] t � [ e] t [ ] t 11 [ e]t �
and
i.e. [] 11 inf
Each condition represents an equivalence class of elements ofPOS X ORD (cf section 2.5). It is vital to remain aware that these are initial conditions which, if P4/K2 is to do its work, must obtain prior to, and independently of, permission being granted, not the other way round. Of course, this does not exclude that psychologically or by conventional fiction the empirical action of giving permission represents one of these initial conditions, after the fact, as having obtained prior to the act, i.e. as an implicature or, better, presupposition. The label is immaterial and I shall use the linguistically less loaded term presump tion. Formally it might be explicated by familiar means (Gazdar 1 979) as a property of the context, ostensibly held true of it by the speaker. This means, of course, as (another) transformation of the context of utterance from a state where this property possibly had not been assumed to hold, to a state in which it is an ostensible assumption of the speaker.18 The effects of these four cases of prior conditions on atomic permissions [PER()] and [PER(0)] and on disjunctive permissions [PER( V 0)] differ significantly: 49a. Permission of one amounts to permission of the other, and indeed ofboth. Taking up one means taking up the package deal. Disjunctive permission and the atomic permissions coincide. 49b. Permission of one does not give permission of the other. Permission of the disjunction is exclusive: either one may be freely taken up, but not both.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Consider now the four relevant types of cases of initial conditions for equi reprehensible and e (49a-d). I shall characterize each of them in three ways: (i) set-theoretically (see also Figure 1), (ii) in terms of what K2 predicts for the effects of atomic permissions and (iii) of the corresponding disjunctive permission sentences:
I I8
Permission Sentences
[49b ] i
[49c]
[49d]
Figure
1
49C. Permission of e also permits ct> but does not require that ct> also be taken up if e is taken up. Permission of also permits e, but only as a package deal: if one is taken up, the other must be taken up, too. Disjunctive permission is asymmetrically inclusive: it offers choice between e and the package deal. 49d. Permission of one also permits the other, does not require joint uptake, but does not rule out joint uptake either. Note that here and e are also deontically independent, i.e. independent even within inf
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
[49a]
A. Merin
1 19
What can we conclude? 1 . (49d) is the only set-up that permits disjunctive K2 permission of equi reprehensibles to be interpreted conclusively. But at the cost of a wholly counterintuitive condition. For even if smoking and drinking are equi reprehensible, and even if their reprehensibility is not additive, an utterance intuitively permitting just one of them should not, intuitively, permit the other as well. 2. The only K2 set-up for equi-reprehensible and 0 which does not license seriously counterintuitive possibilities for atomic permissions is (49b). The exclusive reading it predicts firs paraphrastic intuitions given by McCawley ( I 98 I : 23 I ) and by my own experimental subjects, and the intuitions predicted by K I . I believe it is the one common for most permissions; especially for those with unmarked rise-fall (rather than marked rise-rise) intonation. Note also on what formal assumptions it predicts what e.g. Hilpinen ( I 98 I : 162) demands for disjunctive permissions of equi-reprehensibles in saying that restriction of choice to just one option would be 'arbitrary' while license of joint uptake would be 'gratuitous'. 3· (49b) makes reprehensibility of joint realization of and e a function of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Case (49b): Permission to smoke does not entail permission to drink, nor vice versa. Permission to smoke or drink permits just one or the other, choice being free: disjunctive permission is exclusive. Since smoking and drinking are not, however, physically incompatible habits (i.e. ([
1 20
Permission Sentences
J.J.I .3.2 Non-equi-reprehensible disjuncts As Kamp notes, whenever one of the options is more reprehensible with respect to >w"• scheme P4/K2 predicts that only the less reprehensible option is being (or has been) permitted, since a disjunction is satisfied as soon as one of its disjuncts is. Thus (so) You may take a $ 1 0 apple or (take) a $20 pear will permit at most the $ 1 0 apple to be taken.19 Let (s • ) [] := S takes a $ 1 0 apple;
[8] :- S takes a $20 pear
(52) a. [] t = [ <1>] II inf(Wi ) := [] II Wk b. [E>]t - [ 8] II inf8(Wi ) := [] II Wm c. k < m By (s2c) (see also Figure 2, disregarding case distinctions) we get (s 3) [] t �� [ E>] t = � t.e. 1.e. [<1>] 11 Wk 11 [8] 11 Wm = � [] II inf(W;) II [ 8] II inf8(W;) = � We need not distinguish cases for the time being, since we now have (s4) inf<�>ve(W;) = inf(Wi),
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
their individual reprehensibilities, strictly increasing in both arguments. This suggests that, in addition to ordering structure, some form of concatenative, algebraic (say, additive) structure may be defined with respect to reprehensibility. Using it could take us well beyond the LSK paradigm into alternative ways of inducing recursion (see note 30 below). 4· The intuition of exclusive disjunction requires precondition (49b) to be predicted by K2. fu a presumption it would, therefore, be an integral part of the MEANING of a typical or-permission. That is, it must count among the cost of K2, since one cannot assume that permission sentences can be used with their typical reading only in situations in which the requisite presumptions (49b, or sometimes 49d) are actually common ground (undisputed mutual knowledge) prior to the utterance. Now, as a tacit representational assumption these might inform a computational mechanism reflected in robust intuitions, REGARDLESS of whether the requisite conditions actually obtained in the real world. We should therefore have a potential conflict between adequate reasoning about the world and the rationale underlying the linguistic data. This must be unwelcome news to the analyst who feels that linguistics and philosophy of language are a joint enterprise. Similar considerations might also serve to diminish the linguistic import of the next problem.
A Merin
121
Hence, the net increment to Opt effected by [PER( or E>)] , where [ or] - v, is (55) ([] v [E>])t - ([] v [E>]) 11 inf<�><-e(Wi ) - ([] v [E>]) 11 inf(Wi) [] 11 w k [ ] t =
=
which equals that effected by [PER()] alone. The t operation of K2 does not generally distribute over asymmetric example: (56) [ <1>]t v [e] t - ([ <1>] 11 wk) v ([E>] 11 w .J � ([] v [E>]) 11 inf<�>(Wi) - [] 1.e.
11
v.
In our
w k - [
([] v [E>])t � [<1>]t v [e]t
reprehensibility gradient with implicature from brevity The 'theory of conversation', familiar from Grice ( 1 967, 1 975), and specifically its sub-maxim of rational conversational conduct-'Be brief'-are called to the rescue of K2. The argument (1 979: 259), translated to these examples, has two steps. The first is firmly taken. Had the speaker merely wished to grant the lesser alternative, he should have simply said J.J . I .4 K21:
(57) You may take a $ 1 0 apple. The second is more tentative. One might, says Kamp, hope to find additional conversational principles which justify the further inference that a $20 pear is also among the options intended to be permitted. Note that a less scrupulous analysis would simply take both steps in a single stride and maintain that flouting of Breviry generates that inference in the most obvious (and if you probe, inexplicable) of ways as a CONVERSATIONAL IMPLICATURE. Now, what is felt to be missing from the bare K2 permission is (formally) the set [E>]t. Intuitively, this is the previously more reprehensible option of taking the $20 pear. And this is what the implicature from 'Breviry' would have to
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In fact, t will distribute over v only for the equi-reprehensible special case Per J . J . I .J . I , as shown above. By contrast, the operation ofKr always distributes over v. This is no greater a feat than the correct 'prediction' ofK r even for the case of unequal reprehensibiliry. The first properry is stipulated, and so is the prediction that goes with it, so that the problem simply does not arise. But we can now recognize a formal correlate of adequate prediction: the unary pragmatic operation that delivers permission sets distributes over the binary semantic operation denoted by or. Put as a desideratum: the pragmatic operation should be a boolean join (union) homomorphism from truth sets to permission sets. The counterintuitive prediction ofK2 naturally challenges Kamp to propose tmprovements.
1 22 Permission Sentences
restore. Let K21 designate the implicature-added version ofK2. Kamp himself does not go into further detail here. But an obvious way to restore the undersized permission of K2 to its intended liberality is by a further context changoing operation. It should expand Opt u ([
'Brevity'
(s s b) Opt ..... Opt u ([
We distinguish again four special cases of initial conditions (see Figure 2). By (5 3), relevant distinctions can reside at most in relations between those parts of [
[
n n n n n n n
(inf9(W;)\inf
59a. Permission of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
By associativity of u we can rebracket and represent the combined effect of utterance of (so), scheme K2, and the putative implicature from Brevity as
A. Merin 1 23
@] [59b]
[59a]
Wm
@] [59d]
[59c]
Figure 2
(59b) corresponds to typical intuitions, i.e. to an exclusive permission. Thus, K21 with presumption (59b) would amount to a reconstruction of K r as far as the result goes. Cases (a) and (c) should again be disregarded as being too perverse. And, as before, case (d) serves to license the inclusive reading. And at similarly high cost. Smoking is not only more reprehensible than drinking, but so reprehensible that ceteris paribus it makes no deontic difference if Slave drinks once he smokes. 'You might as well be hanged for a sheep as for a lamb,' says the proverb. In case (59), however, permission to take a sheep implies permission to take a lamb as well-without a word having been said about the lamb. Kr by contrast should always predict the exclusive reading, unless one of the disjuncts entails the other (that is, throughout Pos). For only in that case would permission of a single disjunct have to amount to permission of the other, as a 'package' entailed by the nature of things. Disregarding for now the problematic, if not to say dubious motivation ·of the implicature, K2 1 plus condition (59b) thus predict better than expected or stated by Kamp. Prima facie, appeal to Grice and UBR would suggest that Brevity first of all restores the effects of a boolean or, i.e. inclusive permission. Then another implicature from Grice's Maxim of Quantity (or whatever neo Gricean substitute there is for it) would have to be invoked to get the usual exclusive reading. But 'Quantity' would force consideration of the expression alternative with and, with fatal consequences for the K-family of theories (see section J.J.r.6). And one may feel that Brevity is enough of a liability already (see below).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Wk
124 Permission Sentences
That said, it would be premature to identify, as in K 1 , the union operation of the set-object [ «P] t v [E>]t with the boolean denotation of the or occurring in (so). The two occurrences of union in the final output representation (Opt v [ c�»] t) v [E>]t of(s s) are very differently conditioned. The first represents the expanding act-pragmatic effect of (so), in which the or and its right-hand sequel get lost altogether. The second represents the implicature-pragmatic effect of or, called in to do the repair job. Gricean doctrine embodied in Kamp's
Something along these lines would provide part of the explanation for R Lakoff's (1972: 239) observation regarding English modal verbs and their respective periphrastics: 'when the speaker agrees with or takes upon himself the atomic meaning of the model, he can use the simple modal form. Otherwise he must use the periphrastic version.'20
But it remains unclear what, apart from the bare instruction 'look for a distinct (and thus presumably unusual) meaning', Brevity itself can counsel. The implicature would have to be known in advance and triggered; it is in no way 'calculable', as Grice should have to maintain, from truth-conditions, expression length and general rules of cooperative discourse. Thus, already for good cases the implicature is likely to be more 'conventional' (i.e. ad hoc) than 'conversational'. Grice himself ( 1 989: 372) came to take a pessimistic view of 'Manner's' explanatory potential. I see little reason, then, to consider it as anything but a fudge, at least in cases of putative 'generalized conversational implicature', where the data to be explained are always highly robust, fairly sharp and often non-parochial. Moreover, if we consider the proposed application in K2 1 , it would seem that meaningfulness of or, rather than local meaninglessness, is a less likely candidate for unusual meaning than for usual meaning. Hence, the putative effects of Brevity would go against the very interpretive rule that lends it a semblance of generality. There are also more abstract methodological objections. Like K 1 , K21 relies on identifying each disjunct with a distinct performance of a state-changing act. In K1 the acts are virtual (recall the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
proposal and our reconstruction of recursion steps will make it entirely an implicatural effect of the utterance of (so). Hence, NEITHER occuRRENCE OF BOOLEAN UNION in ( S S) , nor therefore its OCCUrrence in [ $r V [ $ ] t IS A BOOLEAN DENOTATUM OF oR . The meaning of or that enters into the recursion scheme is entirely a matter of implicature from Brevity. UBR is at best non-predictive and at worst an obstacle to Brevity's workings. 'Brevity', part of Grice's set of 'Manner' maxims (and, as brevitas, of classical rhetoric) has not led to serious formal predictions. Gazdar ( 1 979: 44) in passing mentions 'quantifying over the length of expressions at some level of representation', conventionalizing the shorter of two 'potentially synonymous' expressions as the speaker's default choice and instructing hearers to assign the longer one a 'distinct' reading:
A. Merin 1 2 5
(6o) A X[X u ([ct>] u [6])t]
=
AX[X u ([ct>] u [6]) n infue(W;)],
into another, (6 r ) A X[X u ( [ ct>] u [ 6])t 1 ] : - AX[X u ([ ct>] u inf(W;) ) u ([ 6 ] n inf8(W ;))] [t' ] i.e. AX[X u ( [ct>] t u [ erJ. This formulation, too, indicates powers unlicensed by the received doctrine of implicature. In (6 1 ) the operation t1 is not a union-homomorphism: it does not distribute, being distinct from plain t. And we saw that t itself will not distribute over u in the general case (s8b). The effect of the putative implicature, embodied in the move from (6o) to (6 1 ), thus amounts to turning t into a union homomorphism, like its precursor Per of K 1 , or as it already is for the special case of equi-reprehensible disjuncts.
B
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
counterfactual specifications) and both o f the same kind. I n K21, they are each differently motivated and the incremental, procedural reconstruction of Kamp's proposal makes them relatively actual acts. One may rightly comment that even by fairly conventional Gricean accounts (Gazdar 1 979: ch. 3), computation of a conversational implicature is predicated on performance of a speech act. But if, in fact, one has a sentence of form (22) and reads (s 8a) in line with Kamp's sequence of procedural steps, one finds an implicature undoing the effects of another, relatively more semantic computation step: the one reducing the effect of (so) to that of {57). The computation undone lies prima facie further inside the mechanism of semantic recursion than conversational pragmatics could be for a committed Gricean. Quite apart from the fact that it is unclear, except to common sense, why and how the particular implicature that is taken to do the undoing should arise, the theory of implicature is not, in fact, equipped to undo such computations. Unless, perhaps, they themselves have the status of implicature. I say 'perhaps', since a Gricean with pretensions to the attribute 'principled' will in such cases at the very least require a clash of maxims, and criteria for deciding which maxim is to have right of way. No clash or criterion is apparent here. Nor do we have readily apparent substantive Gricean conditions, apart from the bare fact of an utterance, for the implicature undone to arise in the first place. What, then, would remain of the alleged foundations that gave 'implicature' its sheen ofphilosophical respectability? Nothing. Ofcourse, we can always label the phenomena into the irreducibly residual category of'convenrional implica ture'. But in the absence of a serious, formal, recursive non-truthcondirional semantics this is just a euphemism for 'unexplained'. The conceptual move from (s 8a) to (s 8b) also suggests a non-procedural formulation of the effect of the putative implicature from 'Brevity'. It amounts to a transformation of the K2 state-changing function whose argument is Opt,
I 26
Permission Sentences
(62) f.. X (X u ([
(t2]
- J. X(X u (([
·
Factoring out f, we therefore have results identical to those for the equi reprehensible cases (49a-d) for each respective condition (s9a-d).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
J.3. I .S Kz2: reprehensibility isoclines with implicature of indifference Kamp suggests another implicature as a rationale for the desired interpretation: that utterance of the or-coordinated sentence testifies to the speaker's indifference as to which option is taken up. Call this variant of the theory K22. The basic idea appears more attractive than mere Brevity, being more specific. Note, then, its parallels to the implicature of speaker's ignorance, which Schroder ( I 890: I, I 34 £) and elementary logic texts by Tarski ( I 94 I : 22) and Quine ( I 950: I 6) drew attention to before Grice ( I 96 I , I 967) caught the linguistic imagination. Note also that Gazdar ( I 979) classes it with a similar one for if(likewise: Quine 1 950) as a 'clausal quantity implicature'. But its putative Gricean motivation relies again on, yes, 'Brevity' (Gazdar I 979: 6 I ), as, indeed, do Quine, Tarski and Schroder. Without faith in the predictive import of boolean interpretation one may suspect that indifference or, in episternic cases, ignorance are direct con comitants of a basic meaning of or indexically allocating choice (c£ Dohmann 1 966; Dik I 978: 275; Merin I 99 1 : section 7) and that the so-called Gricean reconstruction is spurious; at best a normative rationalization. But suppose that, whatever its origins, the implicature supplements boolean union. Then it still raises problems deeper than apparent from Kamp's ac.count. Taken literally, the testimonial of indifference means that the distinction berween Wk (i.e. inf(W;)) and Wm (i.e. inf8(W;)) which was presumed to have obtained prior to utterance of PER(IP or E>) has now, IN VIRTUE OF THE UTIERANCE, collapsed. In the present framework, where isoclines are static (P4/ K2 proposes nothing else), this effect may be repr�sented by yet another variant of the t operation, call it tz, which yields a state-changing function
A. Merin
1 27
How do predictions of tt and t2 compare? Unaided intuition would not discern a difference. And it ought not, if the same result is desired. Rather than proceed laboriously case by case we can see that the effects are, indeed, equivalent by considering inclusion relations for the sets K2 1 and K22 each propose to add to Opt. Under K2 1 the set added is, by (6 1 ), (65) ([] u [ 8])t 1 = [ ]t u [ 8]t = ([] n infct>(Wi)) u ([ 8] n inf9(Wi) ) = ([] n Wk) u ([ 8] n Wm) Under K22, the set added is, by (62), (66) ( [] u [8])t2 = ([] u [8])
n
infct>(Wi) - ([] u [8])inf9(Wi)
But we have, by definition of the general case for (59a-d), the inclusions (67) ([] n [ 8]) n infct>(Wi) s;-;; ([] n infct>(Wi)) u ([ 8] n inf9(Wi)) s;-;;
([] u [ 8]) n inf9(Wi)
Le. (68) ([] u [8]) n Wk s;-;; ([] n Wk) u ([ 8] n Wm) s;-;; ([] u [8]) n Wm If the condition of (66) is indeed obtained by collapse of the distinction between Wk and Wm, then the three sets of (67/68) are identical. Hence, K2 1 and K22 make identical predictions about what is being permitted. These two conceptually distinct ways of arriving at the same primary intuitions might testify either to the richness of pragmatic life, or to the arbitrariness of implicatural arguments. We have already discussed methodological problems of tl. But t2 has its own peculiar difficulties. The rationale of t2 demands that utterance of an or-permission sentence changes not only the extent of what is granted but also the underlying preference structure of the model. Assuming that there are prior conditions (i.e. intrinsic relations among sets of worlds) of the kind illustrated in (59), it is the spheres or isoclines that change, not just the option set. Thus, t2 appears to adopt at least in part Lewis's critique ofP 4, now for empirical linguistic reasons. Having operations on preference structures is interesting not least because it raises the ancillary costs of a putatively boolean approach to a level at which the costs of an alternative might no longer seem very much higher. But even ifwe disregard these difficulties of foundations there is the problem shared by both K2 1 and K22: that no union operation in the body of the state changing function is the semantic denotation of or in the sentence. So, perhaps there is another boolean alternative?
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
= ([] u [8]) (') wk = ([] u [8]) (') wm·
1 2!! Permission Sentences
3.3 . I .6
K1/K2.:
implicature-free operations on pragmatic denotata
·
(69) [PER(ct> or E>)] - [ PER(ct>)] · [PE R(E>)] which also equals [ PER(E>)] [ PER(ct>)] by commutaOVIty. What could formally distinguish the intended effect of (22) or (so) from successive permissions of their respective disj uncts? (See section 3·3·S for Kamp's analogous proposal for conjoined assertions.) By all accounts, those should have the cumulative effect of permitting both options being taken up. Or coordinated permission is, however, usually exclusive. One can again think of a plausible implicatural rationale. Suppose the context were such that condition (59b) obtains: i.e. that the nearest honempty intersection of [ct>] and [E>] lies in some w n > w m of the reprehensibility ordering. Once the first permission is issued outright, its uptake could not be ruled out. Then issuing the second permission under t - A Y [Y n inf9(Wi)] ). Y[Y n Wm] , where Y ranges over 'propositional content' sets, would be vacuous. Its uptake would be precluded by condition (s9b). Opt would remain unchanged by union with the empty set. In this case an implicature generated by flouting of a prohibition against vacuousness ('Relevance', 'Manner' . . . ?) should presumably modify t to intersect [ E>] with the nearest isocline that would permit joint satisfiability, i.e., it should modify t to an operation t' = A Y[Y n Wn] · This amounts to a change of the function A Y[infv(Wi) ] that selects the least reprehensible subset of[ e ] to be added to Opt. Forget about the costs; there is still a problem. Assuming that sequential permission is intuitively equivalent to and-coordinated permission, i.e. ·
(7o) [PER(ct> and E>)]
=
[ PER(ct>)] [ PER(E>)] , ·
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Suppose we return to K 1 and adopt a reconstruction, suggested by a passing look at (s6) or (6 1 ), which might engender the belief that the second u is denoted by or. Then the denotatum of or operates on values of the operation •t•, not just on raw propositional contents. (Both, of course, are sets of possible worlds.) The weakness of implicatural motivations might suggest that, after all, radical pragmatism of this forthright kind is no less intelligible than white liberal pragmatism of a Gricean ilk. And we would have the correct prediction corresponding to the special precondition (49b/ s9b) without such investment. But even this bona fide boolean solution is faced with two further problems. One is discussed by Kamp himself (see section 3·3·4)· The other is more disquieting yet and raises a problem also for K21 and K22• Let ' ' designate functional composition. By associativity of union in (6 1 ) we have
A. Merin 1 29
then, at least in permissions, and and or WILL DIFFER SEMANTICALLY ONLY BY WAY OF THE IMPLICATURES THEY ENGENDER. But this undermines the core ofUBR and with it all broadly truth-conditional theories on composition of meaning. Criticism of Gricean or neo-Gricean implicature may have little force with those who, like Kamp at the time, admit to being resigned to a two-tier theory of robust, non-parochial prima facie semantic regularities. The theory: a rigorous recursive boolean or Intuitionist semantics drawing upon inputs from, and delivering outputs to, quite informal pragmatics. The criticism should gain force if implicture turns out to be doing all the explanatory work. The next section turns to and, which has boolean interpretation of permission sentences turn from a zero-valued asset into a net liability.
Extension of Lewis's proposal to pure, monotonic and-commands poses no problems for UBR. Let the composite label COM( and E>) stand for (7 1 ) You must take an apple and (take) a pear. Informally, its recursive interpretation could be glossed in an intermediate nonce language by must ((you take an apple) and (you take a pear)). With and denoting boolean intersection, we have formally
(72) [ COM( and E>)] : - f
n
and this agrees with intuitions. Opt shrinks so as to make impermissible all worlds in which Slave does not (at some not necessarily coincident times in the future) take an apple and take a pear. And, disregarding whatever difficulties attach to timing, this is good enough for present purposes. But now consider the and -conjoined permission, (73) You may take an apple and (take) a pear. KI&: UBR extension ofKI to and K 1 , extended to a version KI& such that Per would be a meet (intersection) homomorphism, predicts J . J .2. I
under whatever constraint scheme computes the individual conjuncts. But K1 requires that, for non-cointensional conjuncts, permitting one conjunct alone does not automatically permit the other. Hence, (74) always denotes the empty set and the permission, which denotes a function (75) [ PER( and E>)]K : Opt ,_. Opt u �. 1
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
3-3.2 Conjunctive permission p erformatives
1 30 Permission Sentences
has no effect at all. K r & is completely wrong. And interestingly so, because K 1 is a scheme of boolean (or Heytingean) recursion on context changes that might seem attractive as a deontic extension of,or adjunct to, epistemic state-change semantices (c£ again Gardenfors 1 984, Veltman, 1 99 1 , Groenendijk&Stokhof 1 99 1 ). K2&: UBR extension ofK2 to and Computing an interpretation for (7 3) by a scheme K2&, in line with K2 and UBR (where [ and 0] = [ <1>] II [0]) yields J . J .2.2
(76) [PER( and 0)] : = g
....
Wj ([] [0]) - tj>] II
II
The set [ II 0]t& := [([ ] II [0]) II inf
(78) You may take an apple provided you also take a pear; and vice versa would K2 yield the requisite reading. And it is not obvious that one should call (78) a permission, rather than a conditional offer.22 And since and 0 are sensibly chosen so as not to be cointensional there is no escape hatch. 3-3.2.3 An itnplicatural escape hatch None of the LSK theories examined so far can deal with and -permissions while retaining UBR. Once and is interpreted as intersection, no worlds can be added to Opt in which not both options are realized. If an implicature were to be
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
where
A Merin 1 3 1
called in to preserve UBR, it would have to add to the package-deal-worlds those worlds which K 1 and K2 1 /K22 under condition (49b/s9b) 'predict' to be added by the or-permission. I see nothing in Grice's theory which could allow one to do this with a straight face. We might, however, press on regardless and argue that permitting both options in cases where we are not dealing with a more or less obvious conditional permission should naturally also permit the likely-to-be less reprehensible individual option. There are immediate up-front costs: apparatus yielding a procedure for distinguishing such cases. Take that for granted. Then an implicature, appealing perhaps to a permission-sensitive version of Horn's (I 972) pragmatic scales containing and and or, would add the missing worlds after the locus communis of ancient rhetoric, omne maius continet in se minus. AGAIN, AN OBSTACLE TO THE SMOOTH WORKINGS OF WHOLLY PRAGMATIC PROCEDURES.
you disagree, try to write down exactly why-)23 3-3·3
(If
And as union
What, then, is the apparent interpretation of and in permission sentences? At first sight, intuitions appear to fit satisfaction conditions for plain inclusive, boolean or, i.e. such that the denotation of and in permission contexts cper is
(79) [ and] cper :- v.
Since union is, after all, an instance ofan upper semi-lattice join, this idea might (and, in verbal response to a seminar presentation, in fact did) not seem wholly unattractive to a proponent of lattice-algebraic accounts of plural and event structures (c£ Link 1 987). But what predicts? Presumably still the general framework ofKI or K2. And we have seen that interpreting or as set-union will, in the general case, mispredict for plain K2. J. 3 · 3 · I
K2&1: and as union with precondition presumption K21 and K22 offer a possibility that both and E> arejointly admissible without one entailing the other everywhere; but with the unpalatable consequences of conditions (49d/s9d). Now, if we are willing to bear with the asymmetric deontic entailments required and are ready to give up UBR to save a lattice operation interpretation for and, then we have escaped. But again, the difference between and and or would be purely implicatural: specifically in terms of presumptions 49b/ 59b and 49d/ 59d. Call this scheme K2&1•
3-3.3.2 K2&2: and as union with intersection-based constraint set Another variant is conceivable, too. Suppose for argument's sake that and again designated boolean union plus some mysterious additional factor, MY, able to
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
THIS LEAVES THE BOOLEAN INTERPRETATION OF CONNECTIVES AS LITTLE MORE THAN
1 32
Permission Sentences
work implicatural changes requisite for the computation of a permission set [ u e] t&. Then what it should deliver always is a permission set of the kind delivered by K21 /K2 2 for sub-case (49d/ 59d). This will be ensured if [ <�> u 9] is intersected with inf
([]\[9]) n inf
n
iff
[ 9]
In words: the least reprehensible worlds in which either or e is realized are, without exception, worlds in which both are realized. Again, this is a presumption in the above sense. The condition on Pos and the reprehensibility ranking is not, of course, very revealing. Hypothesizing MY, a relatively pragmatic procedure, tailor-made to ensure that inf
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(8 I ) ([ ] n [ e]) n inf
A. Merin
133
3 · 3 ·4 K3: The performarive-constarive theory
·
Suppose, now, that a prior permission is being reported by an utterance of (22) and that this is clear to speaker and hearer. Suppose further that no tacit or explicit rider 'but I don't know which' is understood or added. Then (22) will in general report that both taking an apple and taking a pear are permissible acts. (Add: but not taking both.) The entailments, which follow schema PD (- 25), then conflict with the standard account of disjunctive declaratives. Kamp ( 1 979: 264) proposes, in essence: A permission sentence f uttered assertorically is true iff all those worlds already belong to Opt which a performative use off would have added to Opt had they not already belonged
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
ad hoc, satisfaction conditions o fa permission sentence are specified recursively and thereby intelligibly. For . the therapeutic philosopher or the knowledge engineer this might do at least in the short run. But for the empirical linguist it is, I think, just a starting point for further analysis. Declarations of non intuitable, non-parochial homonymy (bank is intuitable and differently parochial to English and German) need not worry the purveyor of reliable tools of inference. But they are the linguist's declaration of bankruptcy. 'One word, one meaning' is a principle not to be lightly abandoned, at least as a methodological constraint. Suppose we were happy with K2&2• Then we might momentarily feel that a weakened version ofUBR could survive in K2&2 by way of duality principle, that is, by appeal to inverse inferential monotonicity of permission, which removes rather than adds constraints in respect of Opt. Indeed, a duality principle of this kind is suggested by Kamp's p-entailment, originally developed against the background of K 1 . If we pursue this line, a context governed by such inferential duality would map and's unmarked command- or assertion-relative boolean denotation (intersection) to its boolean dual (union). But then, by duality, or in permissions should behave like boolean intersection. Since (22) will not grant permission restricted to a package deal, or does not appear to be representable by intersection. This puts paid to hopes for retaining a meet/join interpretation of connectives by way of simple pragmatically conditioned lattice-theoretic duality. (But see note 30 for suggestions of a lattice-theoretic duality principle not predicated on treating connectives semantically as meet or join operations.) There is one drastic way ofpreserving UBR for and in permission indicatives analogous to those examined by Kamp. But it involves giving up SPR, makes essential appeal to the assertoric use of such sentences, and will therefore not extend to imperatives. That last failing should disqualify it outright (see section 3-3-7)- Before considering Kamp's corresponding perusal of computational conventions for or (section 3-3-S), let us however examine an intermediate proposal.
1 34
Permission Sentences
to it. He considers one possible explanation for PO in assertions: performative use as the primary use of permission sentences. This apparent reversal of the priorities embodied in Lewis's account is not gratuitous. Lewis does not address or-coordinated permissions and their 'strong' inference pattern [PD]. Formally, Kamp's proposal means augmenting the recursion steps A. B and C of K 1 , which are associated with clause (39) by another recursion clause, D , which determines the truth conditions of a constative (reporting) permission sentence, i.e., Jr], in terms of J f]Per: D. c[f] - {w: 3 c' [w - We· &c·[ r]Per � c·Opt] }
DETERMINED,
'
' SPEECH ACT NEUTRAL CONTENT CANNOT BE MAINTAINED FOR
ENGLISH (1 979: 270). This could be an objection to the theory, but one which might just as well crumble as a piece of dogma ill-applied to natural languages. If AVAILABILITY TO SPEAKERS (and to hearers) is what characterizes operations and operands of the recursive component of language, then, says Kamp, pragmatic elements available to speaker-hearers could just as easily be operated on recursively. But note that this argument takes us firmly into semantics as a branch of psychological anthropology or anthropological psychology. (And the theory that gives rise to it is predicated on a mechanism of counterfactual reasoning.) There are also more empirical objections to K3. First, constative use of (22) may well result in the 'weak' reading satisfying DI ( 26). This much has been clear from the outset. Secondly, so Kamp argues ( 1 973: 67; 1 979: 271), even performative uses of disjunctive permission sentences are sometimes inter pretable in the 'weak' sense. Suppose it is mutually known that both P and Q are, so far, prohibited. Then a person intuitively in authority utters one of the patr ·
=
(8 3) You may do P or Q, a. but first take counsel with my secretary b. but don't do the dangerous one. These are, says Kamp, permissions which appear to lift the prohibition against the disjunction considered as a whole ('weak' reading) while the individual prohibitions on P and on Q stay in force, e.g.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
where r expands, as before, to You may a, time coordinates are omitted and c·Opt represents the option set associated with the possibly counterfactual context c'. In other words, Jr] is finally determined as the value of a function, A.x(c[ x]) for an argument that is a function of c·[r]Per, which, in turn, was computed on the basis of another value of the function A.x(J x]).24 Similarly for the analogous steps ofK21 etc. with t1 etc. taking the place ofPer. Kamp concludes that, on this theory-call it performative-constative, K3the fundamental Fregean assumption that there is such a thing as a RECURSIVELY
A. Merin 1 3 5
(84) You go to Copacabana or (go) to Ipanema would be permitted, while each of (8 s) You go to Copacabana (86) You go to Ipanema
(87) Do P or do Q, a. but first take counsel with my secretary. b. but don't do the dangerous one. c. ??but I don't know which. d. but I won't be responsible for the consequences Since (87a,b,d), unlike (87c), are acceptable, it appears warranted to call (87) deontically performative. (87d shows that unacceptabiliry of 87c is not simply due to its indicative mood.) How, then, is one to capture Kamp's intuition that permission to do one or the other (say, to go to the beach, where (87) instantiates You may go to Copacabana or to Ipanema (beach) ) is granted, while neither of its possible realizations (8 s) or (86) is? At this point SPR enters a consideration on choice of theories. Kamp proposes to represent 'weak' performative use of (22) by recursion in which the union operation (interpreting or) is applied before the transitions from j Q] to j Q]Per take place, where Q is a verb phrase. That is, it operates on J(38a)] and J(38b)] rather than on j (23)]Per and J (24)]Per. Analysis trees of the kind familiar from Montague ( 1 97 3 ), here transliterated into bracketed notation, show the difference between putative recursion schemes for strong (RS] and weak (RW] readings (Kamp 1 979: 272). You may uike an apple or take a pear [ 2 3] analyses respectively into (88) [or [may [you [take an apple]]], [may [you [take a pear]]]] (89) [may [you [or [take an apple], [take a pear]]]]
(RS] (RW]
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
would remain prohibited. That interpretation seems more doubtful than the previous one. It makes a theoretical claim whose model-theoretic import is not spelt out. Failing such specifications it is not clear in what sense, other than in terms of syntactic operations on a redundantly specified database, DB, a prohibition against P V Q can be abrogated without abrogating prohibitions of its disjuncts (c£ below). One may also doubt whether (8 3) is, indeed, a grant rather than some form of report or even something altogether different. This should be decided on more secure linguistic grounds. Let us therefore check in how far the riders, (83a) or (8 3b), constrain introspectible act-imputation. Note that they are both in the imperative mood; unlike the rider, but I don't know which , in ex. (27). To verify the workings of the diagnostic, consider versions of (8 3) in the imperative mood, which excludes any constative reading:
1 36 Permission Sentences
Kamp notes three problems: I . On purely syntactic considerations [RS] is much less plausible than [RW]. Not only must or be inserted, but the second occurrence of may must be deleted. This violates SPR, which is associated with a more explicit constraint on the form of syntactic theories:
[SCP] Syntactic recursion neither deletes nor moves lexical items. SCP is a desideratum widely shared among grammarians who attempt to endow their syntactic structures with intelligible semantic interpretations. To argue in defence of [RS] that (22) be derived by verb phrase deletion from which can likewise be used for both constative and performative purposes, would not only violate SCP. It is also implausible in view of the primary data. For (9o) will generally force the 'unsure', report reading (c£ (27) ). Even without a rider. 2. This problem besets both analysis trees. RS yields (23) by combining a representation of may with a representation of you take an apple i.e. treating may as a sentential operator much like Lewis's i• which I notated PER. RW does the same for 'You take an apple or take a pear'. Adopting the SCP-consistent analysis -
(91 ) [You [may [take an apple]]] for (23) leaves two options: (i) retain the first semantic account, but give up SPR; or (ii) revise the theory to give J take an apple] a value appropriate for combination with may, from which then to compute c[ (23)]Per. Kamp argues that this is quite feasible but makes the theory rather more awkward. (Its exposition. I do not think that an explicit compositional account can, in fact get round the banal (Kamp: 'uninteresting') details of such a step.) 3· The permission sentences considered are in the indicative mood, primary use of which is for asserting; a view that is plausible enough not to be thrown overboard without serious attempts at accommodating it. Let me add that, quite apart from Kamp's second and third objections, it remains unclear how RW should act on the above K-proposals to give a 'weak' permission. And the specification that comes to mind makes it implausible that it should so act. Kamp's wording suggests a modified version of the syntactic database scheme, Ko, with a distinct prohibition sentence, -.(P V Q), stored in the database IN ADDITION TO p and -.Q. Its removal by RW-permission would, as desired, require further removal of the individual prohibitions to have any effect. Taken at face value, such redundancy is implausible. First, to anticipate -.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(90) You may take an apple or you may take a pear.
A. Merin
1 37
(A) One might treat (83) as a hybrid of 'sttong' and 'weak' readings thus: Knowledge or presumption of Master's authority to grant permissions forces a 'strong' reading-up to but, which heads a partial countermand that delegates responsibility. At this point Master has partially transferred authority and assumes the role of the 'weak' assertor; and this is how the sentence before the rider must now be reinterpreted. The speaker (Master) is now imputable with genuine ignorance, since choice does no longer reside with him nor, without further constraints, with the addressee. The only difference to (27) is now that the assertion is not a report, but a prediction. But this jars with the uniformly imperative case (87), which supports Kamp's intuitive description of (8 3) being used performatively. A solution accom modating this intuition must, I believe, lead to a more seriously jurisprudential analysis of permissions. As Kelsen ( 1 949: §77) pointed out, permission is not simply a lifting of restrictions. It is a conferral of rights on some party, and by that token a restriction of others' freedom of action. Conditional permission could thus be seen as a command issued to a lower instance ofauthority to grant permission once certain conditions are satisfied. Decision-making is again partially delegated to a third party: actual, or .virtual in the shape of an inter subjective criterion. (B) What distinguishes the worrisome 'weak' performative case from the 'strong' case is the allocation of choice. Default allocation to the addressee characterizes 'free choice permission'. The weak reading, by contrast, allocates choice to a third party, or to a person identical with the addressee in all respects save for a constraint on possible choices imposed by the rider (say, (8 3b) ). We might even, looking ahead, consider that allocation of choice to Nature or to chance takes us close enough to the 'weak' constative case to retain univocality for or. But then THE BOOLEAN INTERPRETATION WOULD ONCE AGAIN BECOME
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
any disjunctive eventuality for a set of n atomic prohibitions one should have to store in advance a set of 2" - ( n + 1 ) disjunctive prohibitions. Such cardinali ties bode ill for implementation. Secondly, permission of either ofP or Q singly would also be intuitively invalid as long as -.(P V Q)-which, after all, is proof theoretically equivalent to -.p&-.Q-remained there.25 The K2 family of model-theoretic proposals seems even more recalcitrant to such a move. Expanding Opt by a non-empty subset of [ ] v [ 0] must admit worlds in which Slave will actually perform one of or e. One might counter that these worlds are such as have passed the secretary's judgement or Slave's risk assessment. But how is this to be represented compositionally? Attempts at procedural explanation of data take us well beyond the capabilities of the K-family of proposals. And, as far as I can see, they all hinge on two plainly related notions: (A) delegation of responsibility and (B) allocation of choice.
1 3 8 Permission Sentences
(And rightly SO. If we proceed, the semantics of Or, one of the last refuges of apparent insensitivity to context, would become highly indexical.) LINGill STICALLY IRRELEVANT.
BUT THEN THE ONE RESPECTABLE REASON FOR RETAINING BOOLEAN OR INTUITIONIS TIC SEMANTICS AS THE MAINSTAY OF RECURSION IN THE FACE OF PREDICTIVE INADEQUACY IS LOST.
3 · 3 ·5 K4: The constative theory
The 'constative' scheme starts with the assertoric use of permission declaratives. SPR counsels the natural analysis bracketing for (23), namely (9 1 ). The crucial step now is the formation of (92) from (93) [= 3 8]: (92) may take an apple (93) take an apple. Kamp argues that both are intransitive verb phrases that ought to be assigned the same type of model-theoretic entity.26 This means: a function from worlds (i.e. world/time pairs) to sets of individuals. Specifically, those individuals j whose option sets license taking an apple in w. To compute c[ may Q] from c[Q] requires the recursion clause (94) J may Q] (w) = {j: 3w ' [w ' E Opt(wJ)&j E c[Q] (w)] } where Opt(wJ) is individual j's option set in world w. If the denotation of you is taken to be a function fad(c) of context c which delivers the individual being addressed,27 then we can explicate the last step of sentence formation in this scheme, call it K4, by treating sentence denotata as characteristic functions of sets of worlds: (95) J You may Q] (w)
-
((fad(c) )J may Q] (w).
Why, then, is an utterance of(23) ( You may take an apple) so often understood as performative, a grant of permission? Well, if speaker Mandy is known to be in authority and may be assumed to know her own mind and to observe
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Finally, consider a psychologistic qualification. Suppose we do not take the inferential structure of the database in Ko at face value. Then we might say that utterance of the qualified permission merely implicates (as a quasi legalistic fiction) the prior existence of appropriate disjuncts in the database. Much like K1 might account for some of the data if a fiction of logical independence governed intuitions. This would imply that actual interpretation (perhaps not unlike much of legal and political decision-making) largely proceeds by mechanisms distinct from those that inform the conventions legitimizing them. If so, it might be the former that underlie, at least in discernible part, our naive inferential intuitions.
A
Merin 1 39
I.
2.
It has not been established that there is an actual institution of giving permission by uttering such permission sentences. Kamp thinks that a diachronic account, however informal, of how performative use might have developed out of assertoric use would fill the explanatory gap. The effects of such performative use must still be specified. Kamp considers an informal version of the reasoning explicated in Lewis/ Stalnaker P4 ( I 979: 276). He is not satisfied. Truth-conditions and conversational principles general enough to be other than ad hoc to the task will not suffice to derive conventions of performative use, even for atomic permission sentences.28
Speculative diachrony, confined as it is to indicatives, might also be taken to imply that report of permission evolutionarily precedes expression of permis sion. A facetious implication, of course; the hoped-for diachrony does not address that larger question. All the same, it manages to stay aloof only by not considering expressions such as (96) Take an apple, which may acceptably be used to give permission, though not to report that permission has been given. The linguist will be quick to note that such bare imperative sentences are morphologically identical to the VP-complement of may and must. It suggests that there could be more to a performative account, and a more plausible beginning to evolutionary thought-experiment, than self imposed restriction to indicatives would allow for. Back to empirical and technical questions. Kamp notes that opting for a semantic analysis of (22) which gives or wide scope over may, as in (88), would not yield the strong reading with the means of the constative theory. (And it
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
conversational propriety by adhering to truthfulness (this is essentially Lewis's strategy), the effect is to make addressee Sandy realize that permission has been given. As Kamp notes, usually the addressee need not even decide which of the two acts is being performed. The practical consequences for him will be identical as long as he can take the utterance to be appropriate ( I 979: 264). One would do well to replace the protean aesthetico-teleological predicate 'appropriate'-synonym of Austin's (I962) 'felicitous' and pervasive in speech act theorizing-by something more specific, which Kamp himself regards as the warrant for performative use: backing by authority. Deontic authority (or power potentially backed by real or imagined sanctions) in the first case would then correspond to epistemic authority (or persuasiveness backed by real or counterfeit evidence) in the second. This, I believe, is the useful way to approach the constative/performative distinction and its much featured special cases of incontrovertibility and automatic effect. Kamp notes two different objections:
1 40
Permission Sentences
must violate SPR viz. SCP.) Computing the truth-conditions of (22) by the natural route satisfying SPR and SCP (97 ) (you (may (or ((take an apple], (take a pear]]]]]
To incorporate it into our theory requires that we articulate general principles which, together with specific assumptions about the utterance in question which can reasonably be imputed to the addressee, yield the desired conclusions by pure logic. (1979: 278)
This is a task he declines to undertake; and to date nobody seems to have been drawn to it. Kamp also observes that the principles warranting the inference of indifference used in pragmatic inference of the strong reading must contain at least one that applies only to a small subclass of all indicatives containing disjunctive constructions. An important point. Recall, therefore, my suggestion that speaker's apparent indifference correlates with performative interpreta tions, whereas properly constative interpretations correlate with speaker's apparent ignorance. (Disregarding the putative intermediate case of ex. 8 3.) A decision-theoretic, univocal interpretation of or might proceed from here.30 But Kamp shows that, even if accomplished, an addition of pragmatic principles on top of the recursive definition of truth-conditions would come up against strong readings of (22) when it is embedded in a complex sentence: (98) Usually you may only take an apple. So if you may take an apple or take a pear, you should bloody well be pleased. The truth conditions Kamp is committed to for conditionals with disjunctive antecedents are equivalent to those of (99) You should be pleased if you may take an apple, and you should be pleased if you may take a pear. How we get there-and whether to get there we have to make ad hoc assumptions in a truth-conditional theory-is immaterial now. We assume, as Kamp appears to, that intuitively
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
using clause (94), the world/time index w lies in J (22)] (equivalently, in terms of a characteristic function, J(22)] (w) - true) when there is at least one future world in Opt in which Sandy takes either an apple or a pear. It is not required that both apple-worlds and pear-worlds should be among Sandy's options. The 'theory of conversation'-presumably one or more of Grice's Maxims enjoining informativeness, relevance or brevity-is again invoked to produce the strong reading: if Mandy is known to be in authority, then use of(22) rather than of a single one of (23) or (24) must indicate either that she is not yet in a position to make a final decision or that she is indifferent on which disjunct Sandy realizes. Where the first possibility can be ruled out, we get the strong reading.29 Kamp himself finds his argument unsatisfactory as it stands.
A. Merin
141
( wo) If A or B, then C will be equivalent to ( 1 0 1 ) If A, then C; and ifB, then C. The intuition is unobjectionable; how to predict it is an open chapter in the theory of conditionals. By all accounts (c£ also Kamp 1 979: 2 80, n. 1 4) they cannot simply be identified with the material conditional for which we have, indeed,
3 . 3 .6 Ks: the demonstrative theory- bringing to attention
Kamp's final candidate for an explanation is avowedly a mere suggestion. Its formal pursuit might have the advantage of linking permission sentences explicitly with other constructions that exhibit similar properties of or (sometimes known as 'affective or'; c£ Ehrenkranz 1 973). The strong reading for disjunction, as Kamp ( 1 97 3) already noted, is available also in sentences exemplified by:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Kamp observes that the readings which the pragmatic component must produce on the basis of these truth conditions fails to entail them. Pragmatic procedures thus would not merely add an implicature to the truth-conditions, but would actually have to undo what the recursive component has established. This would go against the very presupposition of the theory of implicature. For conjoined declarative sentences in which one of the conjuncts is or coordinated and has the strong reading, Kamp is prepared to give each conjunct the status of an individual assertion, which the pragmatic component could then operate on individually. The motivation: connected discourse of individual sentences, each carrying implicatures, may often be equivalent to the conjunction of these sentences. (But here the objection of indistinguishability of and from or, outlined for the case of pragmatic recursion in section 3-3. I .6, must surely bite.) SSP would also be violated if one argued that certain implicatures become conventionally attached to sentences and then become somehow incorporated into embedded occurrences. And to preserve SSP by introducing a special semantic clause to govern the joint effects of may and or would, as Kamp also notes, go against the grain of compositionality in ad hoc fashion besides missing a generalization suggested by examples given in the next section (1 979: 280). Consider also that if we tried to emend the coordinate VP-complement take an apple or take a pear to the empirically smoother take an apple or a pear, the separation, by take, of may would foil even such a move.
I 42
Permission Sentences
( 1 03) We may go to France or stay put next Summer [epistet?ic/alethetic possibility 'may') ( 1 04) I can drop you at the corner or drive you home ( 105) If Ron had come to the party with Fanny or with Jane we might have had some fun. Sentences (22), ( 1 03), ( 1 04) and ( 105) are typically used in speech acts whose function it is to bring a certain number of possibilities to the attention of the audience. What distinguishes these speech acts from each other lies only in the different purposes for which they bring these possibilities into focus. (p. 28 I)
In all these cases the set of possibilities to which the hearer's attention is drawn consists of those in which the truth conditions of the relevant subsentence are fulfilled and which are among the less outlandish possible situations in which this is so. (pp. 28 I £)
K�mp attempts to explain the generalized strong reading in analogy to the special case. Use of disjunction testifies to speaker's indifference as to whether the possibilities he brings to the hearer's attention satisfy the first or the second disjunct. So one may infer that he wants to bring both possibilities to his attention. Again, we depend on purely implicatural extra-boolean properties of or, as in 3·3·1.5. And 'bringing to attention' is not, perhaps, the most useful generaliza tion. As a residual category it marks the outer perimeter of currently explored formal structure. It also leads to the embarrassing riches of common-sense speech-act classification. What (if any) recursion scheme of a specificity equal to that of K1-K4 would be wanted? If no explicit choice of theory is made, Kamp's proposal will step dangerously close to the quicksands of 'mention', historic point of departure for some of the least constrained proposals in soft core pragmatics. But reference to conditionals suggests recourse to P4, which is essentially a deontic version of the Stalnaker/Lewis semantics for conditional antecedents. If so, the new proposal should encounter old problems. But ask now: are connectives in conditional antecedents to be interpreted in essentially the same way as in permissions? 3·3.6.1 Conditional antecedents vs. pennission clauses In fact, not only the ecumenical basis of Ks but the whole P4/K2 family of theories is undermined by robust DIFFERENCES OF PRIMARY NATIVE SPEAKER INTUITION BETWEEN PERMISSIONS AND CONDITIONAL ANTECEDENTS. Consider
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Permission conveys deontic acceptability of the possibilities, ( 103) and ( 1 04) that they are to be reckoned with for planning, while ( 1 05) 'offers a set of possible situations to claim of each of these that a certain condition-expressed by the consequent of [( 105))-holds in it'.
A. Merin
143
(106) IfJohn had walked or talked, Mary would have danced (107) IfJohn had walked and talked, Mary would have danced.
,
3 · 3 ·7 And-permissions in a demonstrative or constative scheme
To the extent that K4 and Ks are to be explicated in a way analogous to conditional antecedents, we face either the objections against the K2 family of theories or the task of explaining the differing intuitions, or a mixture of both. If truth-conditions for (9o) are computed so as to satisfy SPR with and as intersection, the prediction of the constative scheme K4 would again only license the package deal. A false prediction. However, a recursion scheme violating SPR (and, in particular SCP) could derive (90) in analogy to (99) by deletion transformations from (108) You may take an apple and you may take a pear. This is primafacie a report that taking an apple is permissible and that taking a pear is permissible. As a report-and more specifically as an assertion-(Io8) will simply constrain the range of epistemic possibilities and relieve Sandy of any doubts about possibly being in a world where either or both were prohibited.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As regards or, ( 105) or (106) are typical in yielding an inclusive interpretation. As regards and, (107) intuitively requires the conjunction, formally repre sentable by adding to the set of worlds consistent with what we know to be true the least outlandish intersection of the sets of worlds in which John, respectively, walked and talked. In other words, the intuitive satisfaction conditions of conditional antecedents are directly predicted by an epistemic/ nomological interpretation ofP4/K2.31 This is not surprising given their origin. But it makes all theories of permission sentences that utilize P4-regardless of whether they are performa tive, constative or demonstrative-appear somewhat roundabout. A more general suspicion may now arise: that boolean interpretation of English connectives fits reasonably well in constructions used in CONTEXTS OF usE INVOLVING EXPLICIT DEDUCTION, i.e. when necessary or sufficient conditions for certain consequences to follow are being thematized. If so, 'logical form' as logicians understand it would more likely be a property of arguments than of English sentences (see section 4). Thus, when the syntactic implausibility of the performative account and the relative plausibility of informal reconstructions lead Kamp to conclude that the performative theory 'should probably be rejected' ( 1 979: 282) one may agree with the first reason while having doubts about the warrant for, and utility of, the second.
1 44
Permission Sentences
(1 09) Take an apple and (take) a pear has permissive uses (Don't be shy, Go right ahead, Be myguest will usually force it). But, to repeat, neither it nor its atomic kin can be used to make statements in any direct way. Hence, the assertoric escape hatch is closed and we should have to come up with a separate theory for imperatives facing the same basic problem. Hardly an auspicious start for cognitive science. Similar objections apply if we replace 'assertion' (K4) by 'indication' (Ks), unless, of course, the notion of demonstration is so vague that it could not possibly engender them. In which case it is unclear that it will serve to explain anything. I do not, then, object to treating the alleged source, (1 08), as an assertion which, intuitively, reports that both options are permitted, and which it licenses the intuitive inference that each one singly also is. Nor would I dissent from an account of permission indicatives which explains their performative use as based on conventional inference about the speaker's truthfulness and authority. It is the specifics of the Lewis/Kamp family of theories that seem unable to provide an economical explication of this semi-theoretical intuition.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As long as we do not worry too much about the formal pragmatics of assertion sets (as embodying constraints on admissible belief), this looks like a way out. But if we do, then we must decide whether ( 1 08) should be classed as an assertion, an admission, or a neutral statement; and on whether the pragmatics of such acts does not throw up problems similar to those of permission. Supposing they do not, or that we are dealing with an assertion which fits the description of 'assertion-entailment' for monotone constraint imposition, the solution is available. What if, however, a situation obtains which, in fact, underlies the Lewis/ Kamp scheme-prior prohibitions to be lifted? Then Sandy's current belief set will be inconsistent with the constraint imposed by ( 1 08). The remedy would have to be an account of non-monotonic belief revision that accomplishes more or less what permission is supposed to do as regards Opt. One may feel sanguine about undertaking the task it poses. (Wouldn't one have to get round to it anyway, sooner or later?) But whatever relief it might offer is at best tepid comfort. First, gross violations of SPR are hardly tenable within the more formal and semantically minded of current syntactic theories. Secondly, the intuitive difference of meaning, noted by Kamp, between reduced and unreduced versions for or-coordinated permissions (i.e. the empirical invalidity of equation [39]) would imply that and and or are syntactically of radically different kinds. Finally, the ploy with assertion will not work for imperatives:
A. Merin
145
Some quantifying determiners This must be very brief As Kamp ( I 973: 68 ff) pointed out, the close parallels 3 ·4
( I 1 0) You may take { every flower/all flowers} growing in the garden inherits all the problems of and, on either its individual or plural coordinating readings, and whether it is interpreted as a meet or as a join operation. Kamp suggests that both any and some represent set-theoretic union over classes of possible realizations, differing in any being associated with pragmatic operations analogous to those later outlined in Ks for or. In formal representa- tion this will reflect in a difference in scope of pragmatic operators; again as for or. If so, the same problems must arise. Yet it would seem that any distinctions there might be between ( 1 1 o) a. You may take some flowers growing in the garden b. You may take any flowers growing in the garden are perhaps less a matter of difference of existential quantifier scope representing (as Kamp suggested at the time) differences of pragmatic operations on arguments to set-union, than a question of allocation of, and concomitant constraints on, choice. This would again suggest an alternative line of analysis, similar to that proposed for or and hinted at above (see note 30). I shall not dwell on Kamp's I 97 3 analysis of determiners. It was offered a long rime ago and not taken up again in 1 979. Kamp's subsequent work on 'donkey sentences' ( I 98 I ) would suggest that he would be offering rather more developed and different analyses of such data today. But for now it suffices to point out that meet and join will face problems with permission whatever their domains of operation. On the final page of his I 97 3 paper, Kamp voices the strong suspicion that the semantical analysis of natural language is of fundamentally greater complexity than is imagined by those who believe that we can describe the semantics of at least the declarative
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
between coordinating connectives and certain quantifying determiners persist for their apparent behaviour in permission sentences. Without going into a critique of the detail of his analyses one can still safely say that something of the sort is not unexpected. Are not all and every reasonably interpretable as generalizations of and Gust as the universal quantifier generalizes conjunction for non-empty finite domains)? And do not similar parallels inform disjunction and the existential quantifier? Here parallels between or and some come to mind. But Kamp maintains that any, with its much discussed universal! existential ambiguity of readings, generalizes or in free-choice constructions; as, indeed, it generalizes or in conditional antecedents and questions. The detail ofKamp's analysis leaves room for disagreement on the data. But surely
1 46
Permission Sentences
parts of narural languages by theories which are based on no other concepts than those of satisfaction and truth.
But if boolean or lattice operations face problems of descriptive adequacy that turn supporting constructs into full substitutes, then the role of truth (or, for that matter, of information defined by entailment relations) in the recursive specification of large fragments of natural languages will be even smaller than Kamp appeared to be assuming. So the question of what should take their place will have to be posed. 4
C O N C L U S I O N S A N D O UT L O O K
I . Can the thesis of uniform boolean recursion (UBR) be maintained, on any of the possible analyses of permission sentences considered? That is, will it be tenable if intrasentenrial recursion of meaning operates (i) on speech-act independent denotata in the Fregean sense, or (ii) on speech-act-dependent denotata, or even (iii) on essentially speech-act-denoting objects? 2. Is the currently favoured symbiosis of boolean or intuirionisric formal semantics and informal (Neo)-Gricean pragmatics the optimal instrument of explanation within conceivable reach? I hope the above-average length of the analysis wiJJ give some weight to the largely negative answers I submit for consideration. To I : The Lewis/Stalnaker/Kamp (LSK) family of theories and the assumption of uniform boolean recursion are at odds with the data. In particular, UBR with and as intersection (meet, conjunction) and or as union Goin, disjunction) cannot be maintained. The problem is particularly acute for and, where even heavy support by implicatural auxiliary constructs (and, for the empirical reasons suggested by Kamp himself, syntactic deletion transforma tions) will not prove remedial. Even if we disregard the fatal case of imperatives: taking the K-farnily of theories down to fine detail shows that the whole burden of distinguishing and from or would have to be borne by powerful implicatural machinery of doubtful and informal provenance, working to undo boolean or more generally lattice-theoretic predictions. The problems remain, in one form or another, both for static (constarive) and kinematic {performarive) semantic analyses. With the former, we are on well trodden ground, unlikely to turn up surprise escape routes within a boolean or slightly weaker lattice-theoretical framework. Kinematic analyses, by contrast, are a very recent product. However, there is A Groves's reported proof of the equivalence of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Two questions, one relatively specific and one more general, tending to methodological, arise from the foregoing investigation:
A. Merin
147
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Segerberg's epistemic interpretation of the LSK P4/K2 isocline scheme to recent logic-based axiomatic theories of non-monotone constraint revision (Gardenfors 1 988; see note r in the present paper). This suggests that P4/K2, which in consequence embodies the rationality postulates of recent work on theory revision, will not easily be improved into conformity with the data by replacing possible worlds with other constructs. (At least as long as the lattice theoretic interpretation of connectives is retained in a model-theoretic semantics.) For if deductive closure is required, differences will, on current logic-based proposals (as discussed in Gardenfors 1 988: chs. 3-4) arise only in the specification of selection functions that contract a set of constraints or, dually put, expand the set of options admitted by constraints. Granted that prediction of spontaneous intuitions is a criterion of adequacy, then the above-mentioned schemes for belief-change should be inadequate if Gardenfors ( I 98 8: I o I ) is correct in assuming that derogations and amendments of norms throw up the same logical problem as contractions and revisions of belief sets. If he were right on this point, the permission data would falsify an attractive conjecture: that boolean or intuitionistic state-change semantices (e.g. Gardenfors 1 984) in which propositions are conceived of as operations on epistemic states (as, in our case, commands and permissions are operations on . deontic states) could be extended to viable empirical theories of spontaneous semantic native speaker intuitions even for the purely epistemic domain when belief revision is involved. However, in the light of section 3·3·6. r the assumption of problem equivalence will be empirically wrong to Gust) the extent that the semantics of conditionals treats antecedents as instances of ordinary belief revision.32 This would give epistemic applications, at least in conditionals, a breathing space. But it still gets in the way of extending such accounts beyond the narrow confines of certain kinds of assertoric argumentative discourse. Which casts doubt on the value of those accounts as linguistic semantices ready for the kind of surface-interpretation one would, if only on grounds of simplicity, like to attain, and without giving up univocality of and and or. In other words, simply changing the basic type of sentence-denotata from type WORLDS --- TRUTH-VALUES to type CONTEXTS --- CONTEXTS will not in itself circumvent empirical problems. Again, imperatives, which are prime candidates for the latter analysis (and doubtful ones for the first), should suffice to drive home that last point. Finally, to the extent that quantifying determiners such as every, all, each , any, and some are associated with intersection or union operations, they will inherit similar problems. To 2: Kamp's methodological hypothesis that empirical linguistics will have to deviate from the standard ideal of explanation in the natural sciences in that
148 Permission Sentences a combination of(i) a strict formal theory, and (ii) a conglomerate ofless formal explanations of various aspects of the formal theory might be the best we can do if we want to give an account of language which is precise and yet optimally illuminating (pp. 282 £)
should be open to serious doubt, even before a possible alternative theory is being considered. The hypothesis does not simply refer to the metasemantic considerations which motivated Kamp's final conjecture that, some time in the future, we will find there to be not just one compucation procedure for each (unambiguous) expression, but in general a multiplicity of such procedures-different computations being used by different speakers of the same language and perhaps even by the same speaker on different occasions. ( 1 979: 283} Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Rather, its component (ii) consists largely of a conglomerate of so-called 'pragmatic' implicatural remedies intended to bring the strict formal theory into line with the data. For Kamp continues: 'If this is indeed so, then it might turn out that each of the theories considered in this paper correctly mimicks a procedure that is actually in use.' 'Grice Saves' is the tide of a central chapter in McCawley ( I 98 I ). Recalling the author's publication history and the original slogan, one will do well to suspect a polite but firm recognition that an appeal to faith, hope and charity rather than to the leading graces is being made. And if my analysis is correct, then the explanatory burden within the symbiotic combination apparently referred to by Kamp is being shouldered by the Gricean conglomerate alone. This means that the distribution of explanatory weight berween formal main theory and informal auxiliary support is completely lopsided. Indeed, for the problematic cases examined, the elegant and prestigious theory of lattice theoretic coordination serves mainly to lend to the combination an aura of credibility which its intellectual junior partner could never have commanded on its own. The facts to be explained, of which I have addressed some instances, are as follows: Boolean, intuitionist or weaker lattice-theoretic interpretations of and and or appear to predict well in some linguistic contexts; mainly in those which can readily be construed to satisfy direct monotonicity of epistemic or deontic constraint accretion.33 They predict poorly if at all in others. Nevertheless, native speakers appear to have no more trouble communicat ing by means of sentences for which lattice-theoretic predictions are wrong.34 What is to be done? We might opt for increasingly baroque if not altogether desperate 'implicatural' remedies to cover those theoretically 'marked' contexts. In doing so we might also feel that we still accommodate the semi-theoretical, reflective intuition that and and or are univocal across all such contexts. But my argument has been that the pragmatic remedies required to sustain the faith
A. Merin 1 49
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
will, in the critical cases, turn the lattice-theoretic semantics into something like Lichtenberg's bladeless knife with the handle missing. We would not, after all, be doing justice to the reflective intuition of univocality. What if these intuitions are deemed to be of little import? One could then try to argue for more or less systematic homophony of and and or. In some contexts they would denote boolean or weaker lattice-theoretic operations. In other contexts their denotations would have to be specified in different ways. I believe there is a grain of truth in this; but no more than that. Even if we had solved the specification problem in the critical cases-and it has not, I think, been solved by the (neo-)Gricean or stronger means examined in the main text-an objection remains. How would an empirical theory of meaning or understanding look if it had to give these basic closed-class items multiple meanings, systematically across languages? Not very elegant. Nor would it have much phylogenetic or ontogenetic plausibility, if that is felt to matter. How then to pick up the grain of truth? And how to conclude an oppressively long recital of objections on something like an upbeat semiquaver note? A bare suggestion will have to do. The means of conveyance is a rough analogy. Its familiar part is Putnam's proposal of a theory of language understanding, based on the notion of a stereotype ( 1 975, 1 978b). Stereo types are offered as a solution to a problem posed by the work of Frege, Camap and Montague: how people can communicate by means of language in spite of their inability, in practice, to determine individually the reference of the expressions they use. What is 'grasped' by Putnam's speakers are stereotypes associated with 'standard minimum amounts of information' or, perhaps better put, sketchy theories about referents. By contrast, reference, specified in terms of necessary and sufficient conditions for an intuitively concrete or perhaps even more abstract object to be in the extension of a concept is, if not determined, so at least better specified by the relevant community of experts. Stereotypes are intersubjective within the speech community, at least sufficiently so for conversation to take off Knowledge of extensions proper may be required to deal more rather than less successfully with the physical world. Pumam speaks of tiger, gold and other natural kind terms. Open-class stereotypes and the associated two-tier view of meaning have become one of the less contentious items in the common travel kit of philosophers, linguists, and psychologists. Putnam still assumes that the recursive component poses no problems.35 But now we find it does. Why not, then, entertain the possibility of a two-tier theory of closed-class linguistic meaning in which one tier is not necessarily a conservative extension of the other? Note that stereotypes' mini-theories need not be subtheories (or else conservative extensions) of more realistic expert theories either.36 How could the analogy proceed? For one, individual determination of
1 so Permission Sentences
(i) preserve, as far as possible, the desideratum that semantics be a homo morphism from a syntactic algebra to an algebra of denotata with enough structure to account for linguistic data one would not wish to be predicted by one's syntactic formation rules alone; (ii) do not rely on a skeleton of essentially boolean, intuitionistic or locally weaker lattice-theoretic interpretation of coordination and related quantification strapped into place by pragmatic bandages; (iii) nevertheless, in certain contexts of use, do generate intuitions that are in line with those required by formalisms learned in the logic class and which are usually appealed to when rapid spontaneous intuitions are to be checked for argumentative validity and, perhaps, correspondence to Reality? If the question is posed in this way, Kamp's final conjecture is likely to be true, though in a way more specific, rather different, and perhaps more spectacular than apparently envisaged at the time. It is one aspect of the same speaker on different occasions or under different sociological perspectives that will concern us: how the speaker should come to compute meanings by procedures that predict the data elegantly and how such predictions might either coincide with, or be open to considered correction by, procedures that embody criteria of soundness and, in places, completeness. The explicit recognition on Kamp's part that one is, after all, aiming to investigate psychological phenomena which result in successful communica tion ( I 979: 282), already amounts to a shift of descriptive burden from formalisms designed or Platonically suited to satisfy inferential soundness
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(unobviously conditioned) reference should have for counterpart the instanta neous determination of (unobvious) consequences. Where would disanalogies of'the same level of generality be likely to arise? For one, Pumam's original stereotypes are mini-theories for all seasons ofconversational use.37 By contrast, systematic differences of meaning intuitions suggest relativization of any closed-class counterparts of mini-theories to distinct types of language game. The contrary conviction that these are not altogether 'disconnected' uses (c£ Pumam 1 978b: 97) is what an empirical theory of understanding should attempt to reconcile with apparent diversity. This would tend to make it rather more complicated than the context-invariant theory of inference. As a small sample of how one might begin to proceed on this issue, recall the discussion of indifference and ignorance in section 3·3·LS and further remarks in notes 3037· The question for an empirical theory that does not abandon the methodo logical standards set by Frege and byMontague ( I 970) altogether would be: how can speakers and hearers compute meanings by way of computational conventions that
A. Merin
151
Acknowledgements Preparation of this paper was supported by a research fellowship at the Centrum fiir Informations- und Sprachverarbeitung, Universitiit Miinchen. My thanks are due to its director, Franz Guenthner. ARTHUR MERIN
Seminarfur Philosophie, Logik und Wissenschaftstheorie, Universitiit Miinchen Ludwigstr. 3 1 8ooo Miinchen 22 Germany
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
criteria.38 I f so, the task need not be to demonstrate a 'logical form' o f sentences by unwrapping the pragmatic clothing they receive from conventions of utterance. On the contrary, one might just as well entertain the idea, fundamental but not exclusive to the cited papers of Gardenfors, Landman, and Veltman, that the algebra of logic (considered non-Platonically) derives from rationally motivated conventions of epistemic commitment. We might, with Lorenzen ( 1 962: ch. 2), wish to extend 'commitment' to 'argument'. Add now a further assumption consistent with, or even implicit in this view: that these con ventions may very well be extralinguistic, i.e. highly invariant with respect to the fine structure of conceivable natural languages, human or otherwise, and to some of the evolutionary contingencies their speakers have been subject to. Finally, consider the possibility that the algebra of meaning reflected in spontaneous native speaker intuitions, answering to some of those con tingencies and to somewhat different aspects of action and cognition, could differ quite sharply from those of our favourite logics. The task then will be to explain how conventions of argument license what is in effect the assignment of logical forms to utterances. This would give a concrete interpretation to E. J. Lemmon's conjecture that only forms of argument, not sentences of ordinary English, have logical forms ( 1 965: 1 67). Concrete because it need not leave a jawning gap between rules of deliberation and pure, autonomous formation-rule syntax.39 Nor treat as ineffable or non compositional those kinds of complex sentence, clause or utterance to which logical forms cannot be assigned readily, if at all. That is to say, at least not as readily, plausibly or uncontroversially as to sentences of natural languages used in arguments whose ideally formed exemplars are most frequently found in mathematical discourse.
I 5 2 Permission Sentences
N O TES (P4, extended by Kamp to K2) and axiomatically specified syntactic schemes of belief revision by 'partial meet contrac tion functions' jointly and singly worked out by Alchourron, Gardenfors and Makinson. I shall not address this widely discussed work in any detail, but readers should be aware of the relationship. Section J.J.6. I suggests empirical differ ences between the strucrure of intuitions for the deonric domain and for a promi nent application of theory change in the epistemic domain. They emerge for coordinate constructions and have not, to my knowledge, been addressed in the formal literature on non-monotonic theory change. Hilpinen ( I 98 I ) relates a tantalizing example of Cornides' ( I 969: I 24 I ) on conditional injunctions (also found on p. I O I of Gardenfors I988 and in Lewis I 979) and adds another tantalizing insight: that logically equivalent databases of prohibitions may be essentially dif ferent with respect to permission. Exam ple: work from I o h to I 8 h and rest from I 8 h to 10 h is commanded in systems I and 2. In system I also not to drink during working hours, in system 2 not to drink from I o h ro I 8 h. A permission to drink Saturdays from 1 0 h to I 8 h should modify system I ro one where drinking is prohibited all week except Saturday, and system 2 to one where work on Saturday I o h to I 8 h is prohibited. This example is a salutary reminder of the relevance of decision-theoretic, preference-based lin guistic analyses to the database engineer. 2 c£ Stalnaker ( I 972, I 978); Kamp (1979: 265-70); Gazdar (I 979: ch. I ). The formal notation and some example sentences are adapted for present pur poses. Minor points implicit in Lewis are made explicit. Some points discussed by Lewis are ignored. I omit reference to rime of compliance or uptake, or the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I Historically, Hamblin (I 97 I ) and Davies & Isard (I 972) pioneered the state-change approach to discourse. The former makes few, if any, empirical predictions; the latter, which takes its cue from Scott Strachey denotarional semantics for com puting languages, engages detail of data directly only in areas beyond those of bread-and-butter satisfaction conditions ofcoordination. Cornides ( I 969), cited by Lewis, was probably the first ro probe the logic of deontic state-changes abrogating commands and prohibitions. He usefully distinguishes 'formal derogations', which refer to a previous edict by citation and/ or date and which invalidate edicts as pro non dicta unproblematically, from 'rna. terial derogations', which do raise prob lems (pp. I 22o£). The distinction is akin to that between updates in 'location addressable' and 'content-addressable' databases. Lewis and Kamp deal exclusiv ely with the latter case. Cornides' approach lives in a jurisprudential setting and is syntactically phrased in terms of a propositional formal language. Rightly taking partial derogation to be the general case, Cornides shows conclusively the points where pure logic must give way to pragmatic considerations. The pragmatic revision process is procedurally repre sented by flow-diagrams after the man ner of Rescher ( I 966). A purely declarative reflex of deonric ranking considerations informing Cornides' pro cedural approach is his I 974 book (see note 1 0). Gardenfors ( I 988) outlines correspon dences between derogation and belief revision. At I 988: 83 f£ he sketches semi published work by A. Grove ('Two modellings for theory change', Auckland Philosophical Papers I 3, I 986) which estab lishes correspondences between the epis temic analogue of the most worked-our basic model for permission used by Lewis
------
--
A. Merin I 5 3 I 979 exposition o f permission, > would represent comparative distance from the actual world (c£ also Gardenfors I 988: 84 for a remark on similar differences in belief revision schemes.) In the present scheme I retain an index w• (making the relation ternary) with the different inter pretation of ( I 4.iii) so as to provide for the indexicality of the ordering relation that takes on empirical significance in section 2.3. 1.5 where orderings themselves are being transformed. This indexicality is also in line with Lewis's own critical discussion in the permission paper. 2. If �- were antisymmetric, then any two worlds equireprehensible with respect to w• would be identical. Lewis's objections to scheme PI tell against it. (The assump tion of antisymmetry is what most con spicuously distinguishes Stalnaker's I 968 conditional semantics from Lewis's.) Completeness may be hard to attain in real life, but we can plausibly restrict our paradigmatic attention to those compara tively uncomplicated cases where people (especially Masters and Slaves) are at most indifferent, rather than in two minds. (Terminological recall: an antisymmetric but not necessarily complete pre-ordering is a partial ordering.) 7 Cornides ( 1 974) bases the deontic logic of commands on preference relations to avoid familiar paradoxa; so does Hansson (1 990). See note 1 0. 8 Still rarely quoted; it is now reprinted, as an 'epilogue', in Kulas et a/. ( 1 988). 9 This fact of grammar is often brushed aside in the more self-consciously philo sophical literature-with dire consequen ces when ir extends its brief to argument about natural languages (c£ Hamblin 1987: ch. 2, and, for remedial discussion, Merin 1991). Kamp's use of the label 'permission sentences' for sentences of the form You may {3, where may has its deontic, 'root' sense and f3 is an intransi tive VP, does nor imply rhat other sentence forms cannot be used to report or grant permission. w·
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
problem of standing orders and prohibi tions, which Kamp notes in passing. Informally, Kamp ( I 979: 2 56) presents Lewis's scheme in terms of imperatives, such as Polish my shoes! He does not, however, extend his own subsequent analysis to them, nor consider the problems this would raise. 4 Not all contexts of utterance or proposi tions are, or provide, arguments for functions effecting certain social acts. Nothing hinges on this restriction for present purposes and henceforth the qualification 'partial' is to be implicitly understood. The example of computing languages will rightly suggest that a state change semantics-involving, after all, instances of store-to-store mappings will come up against cardinality problems with function spaces. I believe Theo Janssen has addressed such problems by way of partiality conditions (my source is a DYANA Newsletter of I 990). The problem is deep and important; I omit discussion entirely, on the understanding that a constructive account will have to deal with it at some stage. Its generality makes, I hope, local disregard acceptable. What about further nesting within Opt itself? In a paradigm in which, subject to the presupposed physical constraints defining Pos, every wish of Master's should be Slave's command, there is no point to it. Without such omnipotence, there would be. Similarly, one might consider representing differences in qual ity and prompmess of performance by further nesting. Which may not be of much help, but at least suggests that the two considerations are related. The rela tion becomes intuitive as soon as you start thinking of industrial action by 'work to rule'. So, perhaps, does the inadequacy of the Master-Slave model. 6 1. Lewis's {I973) theory of conditionals differs from the theory employed here in having nests centered around a single world, call it w•, the actual world. In that 197 3 scheme, and unlike in his informal
1 54 Permission Sentences
13
14
15 16
I
7
I
8
3· 3· 1 .2 for an analogous recursion schema. For ease of exposition I continue to use a propositional format up to the point where an issue hinges on phrasal decom position. We could, of course, define as 'semantic' everything that is recursively specified given an arbitrary supply of input para meters ('indices') representing contextual conditions. The Stalnaker ranking P4 would be just such an index. Bur so might the constraint scheme PI whenever conditions are met for it to work without handwaving. Kamp's metatheorerical discussion ( 1 979: 265-70) sets the stan dard of discourse on such matters. If 'the' recursive component is weakened to 'a' recursive component, then Gazdar's ( 1 979) and Soames's ( 1 982) accounts of implicatureand presupposmon projecrion exemplify recursive specifica tion of non-truthcondirional meaning components. See Merin ( 1 99 1 ); and please look out for a more detailed exposition of the positive argument, to appear soon. The mistake is a good example of the unfortunate influence Gricean teachings have had on standards of argument and investigation 111 linguistics. I like it particularly because its correction makes no appeal to conceptions of semantics that differ from the received view sym biotic with Grice. The intuitive desired paraphrase would be of form You must (/) or 8; bur since and e are formally sentences, recursion would be mirrored in a heuristic nonce language as must ((/) or e) where and e are of form { You/Slave) VPintranr Again, when thinking of imperatives, one may imagine and e to be of type VPintrans• relying on a context-based proposal such as Hausser's ( 1 980) to supply a missing subject referent, or of type S by simply admitting the addressee as parr of the syntactic string (Merin, 1 99 I : section 4). Other ways of explicating presumptions are conceivable, notably one informally
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
1 0 A valuable discussion of 'staric' and kine marie notions of permission is Cornides ( 1974: ch. p). Comides' own static logic rests centrally on a (transitive, irreflexive) preference relation R between proposi tions, intuitively interpreted in terms of normative desirability. He defines a rela tion of conditional obligation p 0 q : (p & q) R (p & �q) and unconditional obligationO'p :- (q V �q)Op. Inferences O'(p & q) I- O'p or O'p I- O'(p V q) are barred. The first of the barriers assimilates obligations to Menger's explication of wishes; somewhat dubiously, smce Menger ( 1 934. 1 939) actually took them to be distinct m this very point from commands. Comides is forthright about the difficulties of defining permission. A similar development is Hansson ( 1 990). He may be forgiven for being unaware of Cornides, on the grounds that German ceased being a lingua franca of science some rime around 1 93 3 · 1 1 A routine generalization o f this hypo thesis (Gazdar 1 980; in line with Lambek 1 9 5 8) will extend uniformity to subsen tential constituents and corresponding boolean operations on their functional denotata. In a Montagovian framework predicted on hierarchies of functional application this is formally reconstruct ible 111 terms of a family of similar operations constructed by lifting from codomain to function space. Recent truth-conditional accounts (e.g. Link 1 98 3) are agreed that UBR fails for many coordinate NPs bur do not question its viability for clausal and VP coordination. Extension of mass-term properties to algebras of event structures (Link 1 987) retains a lattice-theoretic framework of descriprion. 1 2 Kamp argues throughout in terms of the embedded verb phrases such as take an apple and take a pear. Their semantic interpretation is a property of individuals, i.e. a function from individuals to pro positions which the addressee is initially prohibited from instantiating. See section
A. Merin
where You can go now grants permission, You are able to go now will not (readily do so). Likewise for questions Can you open the door?, whose usual interpretations as requests are commonly analysed in terms of 'indirect speech acts'. Here performa tivity consists in an attempt to put the addressee under a direct moral obligation to extra-epistemic action. A referee pointed out, for can and be able to, that what, in the context of Gazdar's argument, I called the usual meaning (better: reading) of can will be the non-literal one; and the non-usual reading the literal one. One can take this further. Applying the association of performativity and non-literalness to may should support the 'constative' analysis of permission indicatives to be discussed in section 3·3 · 5· if literal meaning were assumed to have semantic (or computa tional) priority. The substantive issues raised by Lakoff's claim militate against Brevity as an explanans, both for permissive or and more generally. First, the correlation of the shorter expression with speaker's performative intent speaks against the desired explanation for or where the longer expression would so correlate. Secondly, length of expression, pace Gazdar, may be just a banal concomitant of a more specific explanans. For Lakoff's examples this might very well be the difference between bare (morphologi cally imperative-like) and to-infinitival complements. To pursue this incipient hypothesis is beyond my present brie£ But if there is something to it, Brevity is again a red herring, affording as it does a probability of correct prediction some where around !. 2 1 McCawley ( 1 98 1 : 23 1 ) notes this for On the $1.50 dinner you get a soup and a dessert, but not the problem it raises for a UBR semantics. 22 The same goes for a more ecological example suggested by a referee: You may close the door and open the window, spoken to
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
suggested by Ducrot ( 1 972: ch. 3 ) for presuppositions. He sees them less as avowals of the speaker, than as ostensibly undisputable commitments imposed on the addressee which cannot be rejected without rejection of the whole basis for the ongoing discourse. Note also that the term 'context of utterance' does not differ from 'context' or 'context of use' by mere accident I take it that various types of contexts are required to take care oflevels of ostensibility and to stave off the evident problems of self-reference raised by propositional attirudes about the context. This complication shall not be pursued here. 19 I have chosen priced goods to give a fairly obvious criterion of reprehensibility. Feel free to substitute your unworthiest and dearest of priceless possessions for the respective options. 20 'Atomic meaning' seems to be a place holder. One might crudely paraphrase it as 'usual reading'. But it really needs to be replaced with a theory of the pheno menon. Lakoff's own discussion is not only evidence against the predictive value of Brevity, bur also germane to the substantive discussion in section 3 · 3 · 5 of the present paper. Her examples include deonticJohn {[a] may![b] is allowed to) have a cookie. Citing contemporaneous work by D. Larkin and R. Binnick, she argues that in [a] 'the speaker is himself giving permission and therefore covertly indica ting his approval. In [b], he may be opposed to the idea of giving John permission . . . but is merely reporting that permission was given.' This is a bit strong; better speak of tendencies or relative likelihoods of reading. Similarly for will and be to, or must and have to. Ask next: is simple 'identification' criteria!? Rather than (deontic or, as the case may be, boulomaic) performativity? Lakoff reports a present inability to find a parallel distinction for can and be able to. Yet, is not 'peformative' vs. 'constative' the generalization here, too? In dialects
155
1 56
Permission Sentences arise. That is, the putative boolean inter pretation of and is met only in a context that requires very much more complex, extralogical reasoning based on world knowledge than do contexts that do not meet it. Moreover, in the corresponding imperative case the package reading goes with but, and not (readily) with and. (ii) The semi-theoretic analysis I have offered involves a give-and-take, or trade-off, which suggests an economic rather than a logical framework of analysis. On neither count can the boolean analysis be said to predict. 2 3 The pre-theoretical idea informing the locus communis and the putative escape route underlies my proposal for an alternative to the received view (c£ note 30 below and Merin 1 99 1 : section 7). One point to consider is that a connective semantics for and is conceivable which, as it were, plugs into or grows out of the utility-based considerations required to motivate the putative implicature. Set intersection as such does not. And, to repeat: contrary to the defining properties of implicature, the received view would, in the above case, also have to equip the locus communis with powers to undo the K-theory predictions of the 'semantic' component. Surely there is a point in looking for an alternative that need not invite intratheoretical incoherence, nor introduce a deep computational asym metry berween commands (or claims) and permissions (or concessions). 24 This holds always if r is (treated as) atomic. If r is complex, as (22) indeed is, then it will depend on choice berween K 1 and K2 1 whether the initial application of A.X(JX]) was to (22) or to its atomic constituents. 25 In the analogous case that prohibition of a conjunct is to be represented i.e. �P V -Q (viz. �(P & Q )), lifting it would amount to permission of P & Q. The first objection still holds; an analogue of the second would only bite on subsequent permission of a second disjunct.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
a child who complains about the heat in the speaker's office. So far, the door is open, the window closed, and Speaker doesn't want both door and window open: papers would fly and the room would soon get too chilly. Clearly, Child could not just open the window and argue, in reply to a scolding, that permission had been granted for this action too. What indicates a package deal reading? The example suggests complex delibera tions of physical/causal and boulomaic/ preferential relations. A naive causal theory of heat and fluid mechanics is needed to envisage the effects of open door, open window and their combina tion. A theory of local preferences is involved in which (for the simplest case) Child prefers cooling action (hence, increased air inflow) and Speaker prefers retention of hear and atmospheric tran quility (hence restrained airflow). In socio-economic terms, the permission is not a homogeneous concession (for which all acts envisaged would be con ducive to cooling, bur a combination of concession and counterclaim (it grants an action with cooling effect and further side-effects dispreferred by Speaker, and requires an action with contrary effects). This has a linguistic reflex in the corresponding imperative which would presumably go (All right, then. ) Open the window; (but!??and) (first) close the door. Choice of but indicates, as I should maintain it does everywhere, addition of a counterclaim. The infelicity of and is hardly explicable unless it is seen to bespeak inhomogeneity. The indica tive version would presumably be You may open the window (but!??and) you must open the door, which makes the speaker's meaning plain. Consider then, rwo facts: (i) Package deal readings are relatively rare and, as shown by the elaborate contextual stage setting that accompanied rhe example to motivate the reading, require relatively complex conditions and reasoning to
.------ - -
- --
A.
26 But a linguistic note of caution rings out: (93 ) is bare infinitive, while (92) is finite. 27 Kamp. like Montague, utilizes the ele gant logician's device of assignment to free variables. 28 The problems of 1 and 2, whose avatars prompted Lyons ( 1 9n: 1 7.6) to distin guish 'subjective' and 'objective' utter ance modalities, are already reflected in Husserl's careful avoidance of reference to 'reports' or 'statements' alongside the syntactic or pragmatically unspecific notion 'expression' (Ausdruck ): is, in the first instance, used not in actual face to-face interaction of commander and com mandee, but on any occasion when a more or less objective expression of one's own or another's
volitional
attitude/opinion
is
required; e.g., by third parties transmitting the command, or as an expression of legislatorial will in the law.
( 1 90 1 : VI, §69; my trans.)
Husser! cites Sigwart ( 1 898: I, 1 7 n.), who notes a directive vs. reporting ambiguity of modal indicatives. The background to ambiguity is this: Assertions [with Kasher (1 974), we might now say 'mood implicarures'] about the subjective state of the speaker, which lie in the fact of utterance and which are valid on condition of his truthfulness, accompany uniformly all utterances
and
therefore cannot generate
differences of different sentence-types. The imperative 'Be quiet!' will, of course, express 'I want that you are quiet'. However, it does not intend directly to communicate this fact, but rather to determine the addressee's will. It does not claim belief in its truth, but obedi ence. I call this property non-transferrable, since the addressee does not repeat in the same sense the act of will of one issuing a command as he would take up the thought of the speaker in believing an assertion. (Sigwart
18
1 898:
I,
n.)
How can we explicate the inruition that the peculiar force of performatively used indicatives derives from the very fact of their being ostensible declaratives? I submit that the impersonal backing typical of statements imposes consrraints on negotiability and imputed social
c
relations that are subtly different from those attending the imperative mood. Where the backing of imperatives typi cally rests on the utterer's will, that of indicatives typically rests on evidence which exists independently of the utterer. One may dispute and thus ultimately negotiate the evidence, but one can no longer negotiate quite as directly. 29 Three incidental comm�nts are perhaps in order: 1 . The constative account lends itself straightforwardly to the application of the 'lifting' scheme for recursion to produce the linguistically felicitous VP-complement a or f3 of ( 39). The performative account would seem to have difficulties additional to those attending to the case examined. On the other hand, once it has an explicit set theoretic interpretation in terms of state-changing functions (as proposed above), there need not, in principle, be an obstacle to applying analogous lifting procedures to boolean recursion on subsentential constiruents. The decisive question is whether it predicts correctly; and a crucial fact to consider when trying to avoid the undesirable equality ( 39) is that the embedded VP is infini rival. 2. The more idiomatic You may take an apple or a pear, nor considered by Kamp, might be SCP-derivable by routine semantic interpretation of or as NP-coordinating (c£ Lambek 1958 ) on any semantic analysis respecting compositionality. J. There is still a syntactic question about che place of or: some current descrip tions would see it as part of a constitu ent: [or [cake a pe� r]), rather chan as a middle sister. Routine Schonfinkel reduction should cake care of chis on just about any compositional seman tics. 30 Briefly (see also Merin 1 99 1 : section 7), the idea is co represent or by an operation
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The expression form with ought to (and must)
Merin 1 57
I 5 8 Permission Sentences ness (rather than reprehensibility, as in P4/K2). If we disregard rhe ceteris paribus conditions usual in causally interpreted conditionals, the rruth of the antecedent is a sufficient condition for the con sequent co be(come) true. This is in line with material implication. The condition of minimal implausibility tends to impose constraints which come close enough for practical purposes to making it a necessary condition, too. (Nothing less outlandish will do; though other, no less outlandish conditions might also suffice.) For or ( 1 06), equi-plausibiliry of dis juncts will cause no problems; otherwise truth of the less implausible one will intuitively suffice. Continued causal sufficiency of joint realization is not guaranteed by the formal scheme, but will hold on the assumption that disjuncts conducive singly to the consequent's realization do not get in each other's way. If joint performance is deemed more outlandish or implausible, the economy principle will favour the exclusive read ing. And, indeed, it would seem that assent to an 'inclusive' interpretation will be accompanied by a tacit rider 'though it isn't actually necessary', i.e. by a recogni tion of potential diseconomy. These assumptions seem to guide intuition. For and {I o7), only the conjunction is guaranteed to be a sufficient condition. {One should perhaps reserve the label 'package' for the special case of a quid pro quo package deal or its causal antecedent analogue.) If if . . . then was material implication, then to assume that it is a necessary condition would be to commit the fal lacy ofdenying the antecedent. But here the minimal departure constraint requires that, in general, nothing less outlandish will do: the antecedent con junction is {among) the least departure(s) from the status quo which suffice to bring about or ensure the consequent. Formal parallelism with the permission case would obtain only if we admitted a weakening of the relative necessity con-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
lxy( ax + ( I - a )y) where x and y range over elements of a linear space, '+' is vector addition and a e (o, I) is the value of a variable under the addressee's control for imperatives, and under (imperfectly known) Nature's for indicatives, i.e. in all cases ostensibly not under the speaker's present control. Assume there exist, for imperatives and indicatives respectively, orderings by utility and argumentative relevance of linear denotation spaces {this is where lattices do come in). Assume further, in line with any reasonable model of bargaining, that claims (and com mands) are qualified 'at least', and conces sions (and permissions) 'at most'. The former thus yield principal filters of propositions generated by rhe speech-act invariant linear combination, the latter principal ideals. Then rhe right predic tions on speaker's commitments are seen to follow. And expected upperbounding of claims by the other parry generates what is usually called 'scalar implicature'. (There is a wrinkle occasioned by occa sional inclusive permissions; but note that they typically carry a marked 'rise-rise' speech melody; and there is a whole set of intuitions about purely ordinal, rather than resource-limited criteria of evalua tion going with them.) The connective and simply denotes .l. xy(x + y), though when phrasal only for distributing read ings; which, given rhe principal filter/ ideal convention, again predicts correctly. To motivate the idea and defend it against rhe more obvious objections would, as rhe sympathetic reader may suspect, require a good-sized book. Bur I hope the brief hint has given some idea of how the observed parallel can be more rhan an idle curios ity: indifference and, crudely ar least, ignorance, are representable by way of equiprobability. 3 I Recall rhar in theories of conditionals developed by Ramsey, Rescher, Stalnaker and Lewis rhe intuitive interpretation of the ranking on worlds is in terms of increasing implausibility or outlandish-
A. Merin 1 59 srraint, or, alternatively, if we raised to the status of that constraint the expectation that a rational permission will always take up a homogeneous permission fully. Causal analogues of permissive package deal readings (facilitated slightly by inserting also as in Ifjohn had walked and also talked, Mary would have danced) or, for that matter, 'even if' readings and expli citly concessive conditionals, will again raise issues of positive and negative conduciveness analogous to those of note 22.
33
34
35
37
38
39
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
32
In both cases the semi-theoretical reconstruction appeals to notions of prima (though not always secunda) facie additive utility and disutility or positive and negative causal relevance. In neither case can we rely on the predictive powers of lattice-theoretically interpreted con nectives to differentiate, let alone explain. Gardenfors (1 988: 84) remarks, however, that Lewis's conditional semantics, in having isoclines centred around a single world rather than a whole set ofworlds, is thereby unsuitable as a theory of belief revision. Note that the permission model has centring around a set of worlds. At least ifwe disregard problems with the intuitive acceptability of English should be correlates of characteristic lattice theoretic axiom or theorem schemata such as the idempotent, absorption and distributive laws. These are less well known than problems with associativity, commutativity, and with schemata involving complementation. The reader will no doubt be relieved to hear that I intend to drag home their import on another occasion. And, in return, do have such trouble with certain sentences alluded to in note 3 3 that should be well-formed by syntactic and lattice-semantic criteria and which are, qua strings, no longer than readily acceptable ones: try Sandy is clever or lucky, and clever or tenacious. His own brief suggestion of a logic
36
interpreted in terms of provability within an empirical theory ( 1 978a: 25 f£) rather than in terms of truth leaves untouched the usual axioms and/or rules of infer ence such as Gentzen (1 934) style intro duction and elimination rules for conjunction and disjunction. The standard example is gold, which, so we are told by honest-looking authority, is white when pure and becomes the 'yellow metal' only in virtue of being alloyed with copper. But note that reasonably ecological com prehension experiments (Roth & Shahen 198 3) suggest systematic contextual varia tions of their psychological counterparts. I take it that Kamp's Discourse Represen tation Theory ( 1 98 1 ) pursues such a line for broadly anaphoric phenomena: psy chologistic procedures for which rules of correspondence with truth-conditional model theoretic semantics can be estab lished. Whatever role such a notion might play in natural languages; c£ Morrill & Carpenter (1 990) for one that minimizes it. From the early 1 970s on, ]. Hintikka, developing proposals by P. Lorenzen, has proposed a game-theoretical semantics (GTS) which 'encourages looking at the logical form of a sentence as being determined by the entire [add: reason ably narrow satisfaction-conditional language-) game connected with it and not just by its surface form' ( 1 989: 5 3). The idea, though not Hintikka's specific interpretation in terms of games against Nature, is consistent, I think, with Lemmon's view. But GTS has not found much favour with linguists; one plausible reason may be precisely the descriptive gap left between such games and surface form Filling it should be the task of a psychologistic yet algebraic viz. model theoretic empirical semantics or prag matics, or whatever one might wish to call it.
1 60 Permission Sentences
REFERENCES Austin,]. L. (1 962), How to do things with words, versity (mimeograph 1 967: published in in J. 0. Urmson & M. Sbisa (eds), 2nd edn, Grice 1 989). Oxford University Press, Oxford, I 975· - (1 975), 'Logic and conversation', in P. Barwise, J. & Perry, J. ( 1 98 3), 'Situations and Cole & J. L. Morgan (eds), Syntax and attitudes', Bradford, Cambridge, MA. Semantics J: Speech Acts, pp. 4 1 -58, Aca Chellas, B. ( 1 980), Modal Logic: An Introduc demic Press, New York (part of Grice tion , Cambridge University Press, Cam 1 967)bridge. - (1 989), Studies in the Way of Words, Harvard University Press, Cambridge, MA Comides, T. (1 969), 'Der Widerruf von Groenendijk, J., Janssen, T. & Stokhof, M. Befehlen', Studium Genera/e. 22, 1 2 I 5-6 3 · - (1974), Ordinale Deontik: Zusammenhiinge (eds) ( 1 98 1 ). Formal Methods in the Study of zwischen Priiferenztheorie, Normlogik und Language, Mathematical Centre Tracts, Rechtstheorie, Springer, Vienna. Amsterdam, part repr. as Truth information and interpretation, Selected Papers from the Davies,]. & Isard, S. D. ( I972), 'Utterances as programs'. In B. Meltzer & D. Michie (eds), Third Amsterdam Colloquium March 1 980, Machine Intelligence, 7 325-39. Edinburgh Foris, Dordrecht, 1 984. University Press, Edinburgh. Groenendijk, J. & Stokhof, M. ( 1 99 1), 'Dynamic predicate logic', Linguistics and Dohmann, K. ( 1 966), 'Zur Semamik und Philosophy, 14, 39- 1 00. Etymologie der sprachlichen Darstellung der dyadischen Funktoren', Studium Gene Hamblin, C. L. ( 1 97 1 ), 'Mathematical models rale, 19, 402- 1 5. of dialogue', Theoria , 37, 1 3 3-5 5· Dik, S. C. ( I 968), Coordination: Its Implications - (1 987), Imperatives, Blackwell, Oxford. for the Theory ofGeneral Linguistics, North Hansson, S. 0. ( 1 990), 'Preference-based Holland, Amsterdam. deontic logic (PDL)',journal ofPhilosophical Ducrot, 0. (I 972), Dire et ne pas dire, Her Lagic, 19, 75-93. mann, Paris. Harper, W. L., Stalnaker, R. & Pearce, G. (eds) Ehrenkranz, J. (1 973). 'Sometimes "or" ( 1 98 I ), Ifs, Reidel, Dordrecht. "and" ', Linguistic Inquiry, 4, 24I-2. Hausser, R. (1 980), 'Surface compositionality Gardenfors, P. (I 984). 'The dynamics of belief and the semantics of mood', in J. R. Searle, as a basis for logic', British Journal for the F. Kiefer & M. Bierwisch (eds), Speech Act Philosophy ofScience, 35, 1-Io. Theory and Pragmatics, Reidel, Dordrecht, - (1988), Knowledge in flux, MIT Press, 1 74-93· Cambridge, MA Hilpinen, R ( 1 98 1 ), 'On normative change', Gazdar, G. ( I 979), Pragmatics, Academic Press, in E. Morscher & R Stranzinger (eds), London. Ethics: Proceedings of the Fifth International - 'A cross-categorial semantics for co Wittgenstein Symposium 1980, Holder, ordination', Linguistics and Philosophy, 3, Pichler, Tempski, Vienna, 198 1 , 1 5 5...Q4. 407--9· Hintikka,J. (1 989), 'Logical form and linguis Gentzen, G. ( 1 934), 'Untersuchungen iiber tic theory', in A. George (ed.), Reflections on das logische SchlieBen', Mathematische Chomsky, Blackwell, Oxford, 4 1-57. Zeitschrifi, 39, 1 76-2 1 0, 405-3 1 . Hom, L. R. ( 1 972), 'On the semantic proper Grice, H. P. ( 1 96 1 ), 'The causal theory of ties of logical operators in English', Ph.D. perception', Proceedings of the Aristotelian thesis, UCLA, distr. IULC, Bloomington, Society, Supplementary Volume, 35, I 2I -S2. Indiana, 1 976. - (1967!I 989), 'Logic and conversation', Husserl, E. ( 1 901), Logische Untersuchungen , William James Lectures, Harvard Uni2nd edn, I 9 I 3-2 I , Niemeyer, Tiibingen. .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
.
A. Merin 1 6 1 Lorenzen, P. ( 1 962), Metamathematik, Biblio graphisches Instirut, Mannheim. Lyons,]. ( 1 977), Semantics, vol. II, Cambridge University Press, Cambridge. McCarthy, ]. & Hayes, P. ( 1 969), 'Some philosophical problems from the stand point of artificial intelligence', in B. Meltzer & D. Michie (eds), Mac/tine Intelli gence 4, Edinburgh University Press, Edin burgh, 463-502. McCawley, ]. D. (198 1 ), Everything Linguists Have Always Wanted to Know About Logic {But Were Afra id to Ask), Chicago Uni versity Press, Chicago. Menger, K. (1 934), Moral, Wille und Weltgestal tung: Grundlegung zur Logik der Sitten , Springer, Vienna, trans. Morality, Decisions, and Social Organization: Toward a Logic of Ethics, Reidel, Dordrecht, 1 974. - (1 939), 'A logic of the doubtful: on optative and imperative logic', Reports ofa Mathematical Colloquium, 2nd Series, Issue I (ed. K. Menger), 5 3-64, Notre Dame University Press, Notre Dame, Ind. Merin, A. ( 1 99 1 ), 'Imperatives: linguistics vs. philosophy', Linguistics, 29, 669-702. Montague, R. (1 970), 'Universal grammar', Theoria , 36, 373-98. - ( 1 973), 'The proper treatment of quanti fication in ordinary English', in]. Hintikka, J. Moravcsik & P. Suppes (eds), Approaches to Natural Language: Proceedings of the 1970 Stanford Workshop on Grammar and Seman tics, Reidel, Dordrecht, 22 1 -42. Morrill, G. & Carpenter, B. (1 990), 'Com positionality, implicational logics, and theories of grammar', Linguistics and Philo sophy, 1 3 , 38 3-92. Muskens, R. (1 989), 'Meaning and partiality', Ph.D. thesis, University of Amsterdam. Partee, B. H. (1 984), 'Compositionality', in f. Landman & f. Veltman (eds), Varieties of Formal Semantics, foris, Dordecht, 28 1 3 1 I. Putnam, H . (1 975), 'The meaning o f mean ing', in Mind, Language, and Reality: Philo sophical Papers Vol. 2 , Cambridge University Press, Cambridge, 2 1 5-7 1 . - ( 1 978a), 'Meaning and knowledge', in
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Kamp, H. (1 973). 'free choice permission', Proceedings oftheAristotelian Society, N.S. 74, 57-74- (1 979), 'Semantics versus pragmatics', in f. Guenthner & S. J. Schmidt (eds), Formal Semantics and Pragmaticsfor Natural Langua ges, Reidel, Dordrecht, 25 5-78 (repr. in Kulas et a/. 1 988, 349-8 1 ). - ( 1 9 8 1 ), 'A theory of truth and semantic representation', in Groenendijk eta/., 25578 ( 1 984: 1 -41). Kasher, A. ( 1 974). 'Mood implicarures: a logical way of doing generative prag matics', Theoretical Linguistics, I, 6-38. Kelsen, H. (1 949), General Theory of Law and State, Harvard University Press, Cambridge, MA. Kempson, R. M. ( 1 986), 'Ambiguity and the semantics-pragmatics distinction', in C. Travis (ed.), Meaning and Interpretation, Blackwell, Oxford, 77- I OJ. Kulas, ]., Fetzer, ]. H. & Rankin, T. (eds) ( 198 8), Philosophy, Language, and Artificial Intelligence, Kluwer, Dordrecht. Lakoff, R. (1 972), 'The pragmatics of mod ality', Papers from the 8th Regional Meeting, Chicago Linguistic Society, 229-46. Lambek, ]. ( 1 958), 'The mathematics of sentence structure', American Mathematical Monthly, 65, 1 54-70. Lemmon, E.J. ( 1 965), Beginning logic, Nelson, London. Lewis, D. K. ( 1 973), Counterjactuals, Harvard University Press, Cambridge, MA. - (1 979), 'A problem about permission', in E. Saarinen, R. Hilpinen, I. Niiniluoto & M. Provence-Hintikka (eds), Essays in Honour ofJaakko Hintikka , Reidel, Dor drecht, 1979, 163-79 (MS 1970). Link, G. ( 1 98 3), 'The logical analysis of plu rals and mass-terms: a lattice-theoretical approach', in Bauerle R., Schwarze, C. & von Stechow, A. (eds), Meaning, Use, and Interpretation of Language, De Gruyter, Berlin, 302-22. - (1 987), 'Algebraic semantics of event structures' in]. Groenendik, M. Stokhof & f. Landman (eds), Papers from the Sixth Amsterdam Colloquium, April t987.
I62 Permission Sentences Sigwart, C. (I 898), Logik, Vol. I, 2nd edn, J. C. B. Mohr, Tiibingen. Soames, S. (I982), 'How presuppositions are inherited: a solution co the projection problem', Linguistic Inquiry, 13, 483-545. Stalnaker, R. C. ( 1 968), 'A theory of con ditionals', in N. Rescher (ed), Studies in Logical Theory, Blackwell, Oxford. - (I 972), 'Pragmatics', in D. Davidson & G. Harman (eds), Semantics of Natural Lon guage, Reidel, Dordrecht, 3 80-97. - (I978), 'Assertion, in P. Cole (ed.), Syntax and Semantics g: Pragmatics, Academic Press, New York, 3 I 5-32. - (n.d), 'Comments on Lewis's problem about permission', (unpubl. MS; cited in Lewis 1 979 and Kamp 1979). Tarski, A. ( I 94 I ), Introduction to Logic, Oxford University Press, New York. Veltman, F. ( 1 99I), 'Defaults in update semantics', ITLI Prepublications for Logic, Semantics and Philosophy ofLonguage LP-9 1oz, Instituut voor Taal, Logica en Infor matie, University of Amsterdam. von Wright, G. H. ( I 9 5 I ), 'Deontic logic', Mind, 6 o, I - I 5·
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Meaning and the Moral Sciences, Routledge, London, 5-80. - (1 978b), 'Reference and understanding', in Meaning and the Moral Sciences, Rout ledge, London, 97-I I 9. Quine, W. V. 0. ( 1 950), Methods of Logic, Routledge, London, 1 952. Ramsey, F. P. ( I 922), 'General propositions and causality', in Foundations ofMathematics and Other Logical Essays (ed. R B. Braith waite), Routledge, London, I 9 3 I . Rescher, N . ( 1 96 I ), 'Belief-contravening suppositions and the problem of contrary to-fact conditionals', Philosophical Review, 6o, I 76-96. - (1 966), TheLogic ifCo m mands , Routledge, London. Ross, A. ( I 94 1 ), 'Imperatives and Logic', The oria , 7, 5 3-7 1 . Roth, E. & Shoben, E. J. ( I 98 3 ), 'The effect of context on the structure of categories', Cognitive Psychology, 1 5, 346-78. Schroder, E. (I 89o), Vorlesungen iiber die A�ebra der Logik, Vol. I, B. G. Teubner, Leipzig, repr. Chelsea Pub!. Comp., Bronx, NY, 1 966.
Journal oJSemanrics 9: 163-178
© N.I.S. Foundation (1992)
So Be It: The Discourse-Semantic Roles of So and It FRAN C I S C ORN I S H
University ofKent at Canterbury
Abstract
1
I N T RO D U C T I O N
The basic aim of this article is to examine one type o f predicate and proposi tional anaphora expressed by two closely related anaphors, so and it , in terms of its discourse-building role. I want to show, first, that the anaphora created via each of these expressions (in conjunction with their host verb, and the mood, tense and aspect values associated with it) has several important functions beyond that of simply retrieving a predicate or proposition already used as such in prior discourse. And second, I will try to pinpoint the precise semantic and pragmatic contribution made by so and it to the discourse content they help to structure, as well as the contextual factors that favour or constrain the choice of one or other of these two anaphors. I shall start by briefly defining the view of anaphora as a semantic- pragmatic phenomenon to which I subscribe, and will then spend the rest of the article examining the examples to be presented later under various different headings. These examples are mostly attested, naturally occurring utterances embedded in some specific discourse context.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The aim of the article is to determine whether so and it fulfil distinct discourse-building roles, in their predicate- and proposition-anaphoric guise. Is the choice between them d etermined by the syntactic or semantic nature of their 'antecedent', or of the context in which the choice between them is to be made? Or do they in fact themselves determine their 'antecedent' (or at least superimpose upon it a characteristically distinctive interpretation)? I shall argue for the latter analysis, indicating a range of factors (of a mainly semantic nature) which co-determine both the choice between these two anaphors, and their full utterance-level interpretation. A wide range of attested and constructed examples will be presented as the basis for the discussion.
I 64
So Be It : The Discourse-Semantic Roles of So and It
2.
P RE L I M I N A R I E S
2. I
Discourse anaphora
( I ) '. . . In countries like Israel and Sweden, academics and authors write their work in English for world publication, and only then have it translated into their own language . . .' (The Guardian , 6 June I 989)
In the process of reaching an interpretation of then , a deictic temporal anaphor, it is clear that an inference is triggered such that the Israeli and Swedish academics and authors in question have infact published their work in English, something which is not entailed by what I am calling the 'antecedent-trigger' (c( Cornish I 986 : 8 and passim ) of the anaphor. This inference is derived from
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Discourse anaphora is intimately connected with deixis. Both are concerned with the management of discourse in terms of the coordination ofspeaker's and addressee's attention. Deixis centrally has to do with the switching of that atten tion focus (itself a deictic motion) from a previously existing focus to a new one-but which is derived from the current discourse context. Anaphora, on the other hand, is a set of procedures for maintaining both speaker's and addressee's attention focus on objects, substances, situations, processes, events, concepts etc. which are already in (relatively) high focus (cf Ehlich I 982; Bosch I 98 3). There is thus a kind of inverse relation between the use of deictic and of anaphoric expressions. Deixis presupposes that the referent in question has a relatively low level of focus ('intermediate' in Ariel's I 990 scheme) at the point where it operates, so that a high degree of concentration and cognitive effort is needed by the addressee to bring it into high focus; whereas anaphora presupposes that the referent already enjoys a high level of focus at the point where it functions, so that only a relatively low level of concentration is needed by the addressee to retrieve (or even construct) the referent in question. The formal properties of deictic and anaphoric expressions reflect these different discourse functions though not perfectly so (c( Ariel I 990, who conceives of the distinction in terms of three main levels of referent accessibility). One major cognitive goal of the speaker or writer is to get his or her addressee to synthesize a mental model of the situation being evoked via the text (and via other non-linguistic sources of input) that matches his/her own model to some degree (cf Johnson-Laird I 9 8 3). Deixis and anaphora constitute specific procedural instructions for organizing and constructing these models manipulating, not expressions in the co-text ('antecedents', as traditionally conceived), but conceptual representations of entities, properties, simations, etc. within the relevant discourse models. To set the scene for what is to come, let us examine example ( I ).
F. Cornish 1 6 5
2.2
The distribution ofso and it
Before we look in detail at the examples of so and it to be presented below, we need to establish, first, which type of so and it we are interested in, and, second, whether in fact each type of expression we are concerned with is in comple mentary distribution with the other. If the latter is the case, it should be possible to specifY the particular formal environments which determine the selection of one or other of these anaphors in their predicate- or proposition-anaphoric gmse. 2.2 . r Non-anaphoric occurrences
Let us first examine certain types of non-anaphoric occurrences of so . (2a) Conjunction: . . . so , as I say, we do have a case after all . . . (2b) Intensifier: Your book was so absorbing that I completely forgot to cook the dinner! (2c) Variant of also : . . . and so can/does 0 Pete. All three uses of so illustrated in (2a-c), namely the conjunction in (2a),2 the intensifier in (2b), and the variant form of also in (2c), need to be set aside in our investigation. Note that does in (2c) is not a form ofthe full verb do in do so , but the auxiliary that provides 'do -support' in polar questions, negations, etc.; the actual anaphor in instances of this kind is the zero form 0, rather than either so or the auxiliary. Two types of non-anaphoric uses of it are illustrated in (3a, b).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the result of the processing of the first clause of the discourse fragment: then (which would receive a high degree of stress and pitch if spoken, due primarily to the focusing presence of only immediately preceding it )1 thus refers to this particular time stage in the process, which evidently follows the writing of the works in question. The contextually derived referent of then here would thus be something like 'Once the Israeli and Swedish authors have had their work published in English in publications reaching a world audience'. This example illustrates an important feature of anaphora generally. Anaphors serve to integrate the segment of text in which they occur with the result of the processing of a preceding segment. They point to entities etc. in the relevant discourse model and may even (as we shall see) serve partially to con struct such entities. Anaphora is not, then, a purely textual intra-linguistic relation between two or more expressions in a co-text (the 'antecedent' and one or more anaphors). The importance of the predicative context in which an anaphor occurs (in particular, the valency properties of its governing verb or other predicate) is paramount in the way this phenomenon operates.
166
So Be It: The Discourse-Semantic Roles of So and It
(3a) Expletive: It 's terribly hot in here. It seems that a June election is unlikely. (3b) Neither the 'ambient it ' in (3a) nor the dummy, non-argumental it in (3b) is anaphoric; hence, these uses of it do not fall within the type of use I am con cerned with here. Furthermore, I am also not concerned here with the 'entity referring' use of it as illustrated in instances such as: john bought a new umbrella, but it didn't keep the rain out! 2.2.2 Predicate- or p roposition-anaphoric occurrences
(4) So occurs as adjunct of certain complement-taking verbs (e.g. hope , appear, remain , suppose); (5) So may follow certain WH -words: how/why so ? (6) So may follow the conjunction if and even : . . . ifso, then that will be the end of the matter; even so, you shouldn't do it . (7) So can occur in pro-clitic position: '. . . and elements that cannot so occur are inside the verb phrase ' (G. Lakoff&J. R. Ross r 976, 'Why you can't do so into the sink', p. ws); (8) So can occur in pre-adjective position: so-called , so-defined . (9) It occurs in all argument positions (subject, direct object, indirect object, oblique object). It (� so ) can thus occur as subject of passivized verbs: ifyou can apologize, it!•so should be done immediately.
It can occur in none of the environments in which so may occur (namely those specified under (4)-(8) above); and, conversely, so may occur in none of those indicated under (9). Though superficially, it might appear that so and it share an environment in common, namely where so occurs as the adjunct of certain verbs (e.g. do , as in do so), and where it occurs as direct object of certain of these same verbs (e.g. again do , as in do it ), this is only a 'surface' impression, as the impossibility of passivizing do so in the example given under (9) demonstrates. This example already enables us to see how different is the grammatico-seman tic relationship between it (an argumental pronoun, unlike so ) and its host verb, on the one hand, and so and its host, on the other.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Having narrowed down the range of types of uses of so and it to those in which the expressions are interpreted in terms of a discursively salient predicate or proposition, we need now to determine whether in fact the choice between them is determined by the environment in which each may occur: if it is the case that each type of expression may be interpreted in terms of the same kinds of referents (predicates or propositions), and that each expression-type occurs in a di}forent set of environments (as the data presented below suggest may well be the case), then the conclusion that they are in complementary distribution would appear to be inescapable. Let us examine environments (4)-(9) below.
F. Cornish 1 67
The information presented so far, then, strongly suggests that so and it , in their predicate- and proposition-anaphoric role, are in complementary distri bution. Further support for this conclusion is provided by a pair of occurrences found in a children's book:
( w) . . . "Could you be king of the Clowns one day?" "I suppose so but I don't '
think it's really likely" '
The conclusion must be, then, that whereas so is syntactically adverbial (and see Bouton 1 970 for still further arguments to this effect), it is nominal.
3 .1
Insensitivity ofso and i t as predicate orpropositional anaphors to the syntactic category oftheir antecedent trigger
Let us look now at how these anaphors receive their full, utterance-level inter pretation. First of all, the examples presented under the above heading indicate that the 'antecedent triggers' (as I am calling them) of both so and it are syntactically diverse. Consider the following: ( 1 1 a) I'm extremely busy at the moment, and expect to be so for the next two hours at least. ( 1 1 b) John isn't usually worried, but today he certainly looks it ( 1 2a) The entire factory is on strike, and is forecast to stay so for some considerable rime ( 1 2b) A:-You look on top of the world today! ( 1 2b) B:-Kind of you to say so; I really feel it too! ( 1 3) '. . . It looks very much as if arts graduates are becoming a drug on the market and are likely to remain so ' ( Tize Sunday Times , 6 June 1 97 1 ) ( 1 4a) I know that he's coming, because he told me so himself ( 1 4b) Max is a superb linguist, and he knows it In ( 1 1 ), the antecedent triggers are adjective phrases (extremely busy in ( 1 1 a), worried -not # usually worried -in ( 1 1 b)); in ( 1 2), they are prepositional phrases (on strike in ( 1 2a), on top oftlze world in ( 1 2b) ); in ( 1 3), it is a nominal (a drug on the market), and in ( 1 4) they are clauses (that he's coming in ( 1 4a), and the entire initial conjunct in ( 1 4b) ). Thus so and it (whose type of anaphoric value is constant throughout this series of examples) are clearly not dependent on the syntactic nature of their antecedent-trigger for their operation. The antecedent triggers in ( 1 1 ) to ( 1 4) are predicates . Now it is well known that predicates are to a large extent independent of syntactic category. Indeed, Napoli ( 1 9H9: 3 so)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
3· S Y N TA C T I C A N D S E M A NT I C C H ARACTE R I S T I C S O F S O A N D IT
168 So Be It : The Discourse-Semantic Roles of So and It
suggests that predicates need not have any particular syntactic characteristics at all. They clearly do not have to correspond to single lexemes (as they do in ( 1 1 a), ( 1 1 b) and ( 1 2b), where on top ojthe world is an idiom, that is, a single lexical unit). This is shown by the antecedent-triggers in ( 1 2a), ( 1 3) and ( 1 4a, b). Predicates can be headed by nouns (often abstract nominals like destruction , or relational nouns like father (of), by adjectives, by verbs, or even by some types of preposi tion (those that express a relation, usually a spatial relation, between two or more entities). What is important is that as predicates, they must have at least one external argument ('role player', in Napoli's terminology), and must predicate some property of it (or, where there are two or more such arguments, some relation holding between them).
Semantic independence cfmood and modality cfantecedent trigger
The examples presented under this heading suggest that the interpretation of the occurrences ofso and it here may be independent of the particular choice of mood or modality in whose scope the antecedent trigger predication occurs. Observe the following:
( I sa) '. . . Does the great presentation the following day necessitate the dis ( I sb) ( I6a) (16b) (1 7a) ( 1 7b)
astrous evening the night before? At Rank Xerox we don't think so' (advertisement, The Guardian , 1 8 March I 99 I , p. s). 'Was Johnson badly advised? He would have detected it first' (title of article in Times Higher Educational Supplement , 1 0 November 1 978, p. 9) 'Don't get lost-it wastes petrol' (notice in garage, Canterbury) 'Stay low, or you'll live to regret it ' (government anti-drink-and-driving advertisment) Willard may have cooked the dinner, but I doubt it (from Peterson 1 982, ex (44)) 'WARNING'-Do not leave purses, wallets etc. unattended. If you see anyone acting suspiciously, report it to a member of staff' (notice placed at entrance to library, University of Kent at Canterbury)
In ( 1 sa) and (b), the antecedent-trigger predication occurs within the scope of an interrogative operator; in ( 16a) within that of a negative imperative; in ( 16b) within that of an affirmative imperative; in ( I 7a) within that of an epistemic modal; and in (1 7b) within that of a counterfactual or conditional. In none of these instances, then, is the antecedent-trigger predication asserted. And in none of them is the particular modality involved part of the interpretation of the respective anaphors. Semantically, then, these examples suggest an analysis whereby the inter rogative, negative, imperative, modal or conditional components of the clauses at issue would be treated as operators, located outside the predication which
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
3 .2
F. Cornish 1 69
occurs within their logical scope (this is exactly the way in which these sentence or utterance modalities are handled in Dik's (1 989) Functional Grammar model). Note that the actual antecedent of it in (16b) is not 'stay low' but the ellipsed predication (required to be instated for the purposes of coherence) 'in the event that you do not stay low' (that is, 'your alcohol blood content is high', or 'not low'). (In passing, it is worth noting that this is good evidence in favour of my distinction between the notions 'antecedent' and 'antecedent-trigger'.) This is effected via the instruction to disjoin the two propositions conveyed by the connective or in terms of truth value, and by the fact that it is the object of a factive verb (regret), presupposing the truth of the proposition to which it is anaphoric. A similar analysis holds for (16a) and (1 7b).
4. 1
The 'discourse-operatorJunction ofso and it
Having seen in section 3 that the anaphors so and it are not sensitive to the particular syntactic character of their potential antecedent-triggers (c£ 3 . I ) and that their interpretation is independent of certain aspects of the semantic properties of those triggers as well (c£ 3-2), we will now see further evidence of the syntactic-semantic independence of these anaphors, namely evidence of the discoursally 'creative' abiliry of anaphors, in conjunction with their host or governing verb together with its mood, tense and aspect values, to extract or 'construct' predicates from within elements performing non-predicative functions in their textual context. ,
( 1 9a) ' . . . some of the soldiers [on the Dunkirk beaches] themselves told me they were starving, and exhausted, and unorganised, and very frightened. I would have been so too' (letter to Radio Times , 2 ]une-4 July 1 980, p. 6o) ( I 9b) Predicate constructed via the interpretation ofso in ( 1 9a): 'J. (x) (Starving&Exhausted &Unorganized &Very-Frightened, x)' (I 9c) ' . . . Only a virgin can join [the new religious order launched recently by the bishop of Verona, 'in defiance of a sociery infatuated by the idols of consumerism and sex'], and she must remain so for the rest ofher days "as a sign of martyrdom' " (The Guardian , March 1 99 I ) ( I 9d) '. . . H e went on to claim that the allegedly high-spending Labour authorities had, by so doing , damaged industry and lost jobs' (lliES, 30 September 1 98 3) (2oa) 'I'm not a farmer any more-because there isn't enough time to do it ' (Christopher Mayhew, BBC Radio 4, March I 982) (2ob) 'Rumania-the odd man out in Eastern Europe, and proud of it ' (BBC Radio 4, 22 January I 979)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
4· D I S C O U R SE - LE VE L P R O PE R T I E S
1 70
So Be It : The Discourse-Semantic Roles of So and It
4.2
Further distinguishingfeatures ofso and it
So far, we have seen a number of different semantic and pragmatic values associated with so and it as predicate or proposition anaphors. Below arc illustrations of divergences between so and it along two further dimensions.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In (r 9a), the use of so by the writer (and its interpretation by the reader) in this context serves to 'abstract' the individual predicators (applied severally in the complement clause of the antecedent-trigger predication to the specific group of soldiers who spoke to the writer) into a single, complex, predicate. This latter is represented in logical terms in (r 9b). In (r9c), the predicate 'virgin' in the subject expression of the initial clause is syntactically nominal and semantically denotes a set (the class of human females defined by the property ofbeing virgins). Note that the NP this noun-predicate heads is already generic, genericity corresponding to a less highly referential (and hence more fully predicative) contextual value. Note also the variation between she (referring to the prototypical virgin eligible to join the religious order in question), on the one hand, and so on the other, sharing the same antecedent trigger-though with very different discourse model referents. High spending in (9d) is an attributive· adjective, a modifier.3 The example provides further evidence for the view that so and do so are not dependent for their interpretation on a syntactically parallel antecedent trigger (contra Hankamer & Sag r 976; Sag & Hankamer r 984, who claim that (do ) so is a surface anaphor, requiring surface syntactic parallelism of a potential ante cedent trigger). The interpretation of so doing here isolates the modifier and gives it full predicative status-as an independent cause of the state of affairs predicated by the main assertion of this utterance. It can also behave in this way-though I have collected fewer such attested examples than in the case of so . (zoa) is similar to (r 9c), where, however, the antecedent-trigger predicate is not being used predicatively (i.e. as a predicator) in its immediate context; yet do it isolates from it, not the defining property, but the activity (due to the fact that it is governed by the activity pro-verb do). That is, it is the activity predicate 'do farming' rather than the property predicate 'be a farmer' which is constructed here. In (zob), the process of interpreting it , as the direct object of the factive adjective proud (of), converts the predicate corresponding to the antecedent trigger predicator the odd man out in Eastern Europe into a proposition having the status of a mutually validated fact; that is, its true value is presupposed by the speaker.
F. Cornish 1 7 1
4.2.1 Possibility of 'sloppy identity' readings
(2 r a) John kissed his·wife, and then Bill did so (2 r b) John kissed his wife, and then Bill did it
So allows so-called 'sloppy identity' readings (i.e. where a nominal contained within the antecedent trigger is consrrued as a bound variable rather than as a constant) more readily than it : see (2 r a) in contrast to (2 r b). The most natural interpretation of did so in (2 r a) is given as (2 1 a'): (2 ra') '(Bill,) A (x) (kiss x's wife, x)' (2 I b') 'kiss j's wife' (where J' = John) This no doubt ties in with the greater ability of so and it to extract or construct predicates from within expressions not used predicatively (that is, as pre dicators) in the antecedent trigger. (2 r a') reads '(Bill) has the property ofkissing his own wife' (tense not being represented here). Under this interpretation, what has happened is that in the process of interpreting do so , the hearer has constructed a new predicate, 'kissing one's own wife', on the basis of the occurrence in its immediate context of a related one, 'kissing John's wife'. 4.2.2 Possibility of occurrence under pragmatic control
Conversely, it is much more readily interpretable under pragmatic, rather than purely linguistic control (c£ Hankamer & Sag's 1 976 and Sag & Hankamer's 1 984 'deep' vs. 'surface' anaphor dichotomy). Observe (22a) in contrast to (22b): (22)
[1 1 p.m. A couple is preparing to retire for the night. Before entering the bedroom, the wife gives a 'knowing' look at the front door] (a) Husband:-# I'll do so , don't worry! (b) I'll do it , don't worry! 4·3
Interim conclusions
At this point, 1t IS necessary to take stock of the situation regarding the differences between so and it . It is clear from the examples and analyses presented up to this point in both section 3 and the present section that so is semantically intensional and indefinite: it is interpreted in terms of a predicate (which may be quite complex). It , on the other hand, is semantically extensional and definite: it is normally interpreted in terms of a presupposed or asserted proposition having the status of a fact (or treated as such by the speaker, who assumes the hearer will do likewise).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
And the most natural interpretation of did it in (2 rb) is represented as (2 r b'):
1 72
So Be It : The Discourse-Semantic Roles of So and It
So seems to mean 'true' or 'the case' (Bolinger I 970: 1 4 1 defines its meaning as 'it as represented', and suggests that 'imposing a condition is a proper setting for do so '). It , on the other hand (in this type of anaphofi:c usage), seems to be equivalent semantically to 'the fact' (at least, when it occurs as the subject or object of a factive predicate: observe the occurrences of it in ( I 4b), (I 5 b), ( I6b) and (I 7b) ). Otherwise, its semantic value is 'the salient state of affairs established in the current context'. 5 THE SE MAN T I C S O F THE H O S T VE RB
5.1
Verbs hosting so only
(23) below provides a (non-exhaustive) list of host verbs restricted to taking so : (23) appear, be, become, hope, presume, remain , suppose, think . . . Of these, appear, be, become and remain are copulative, or quasi-copulative verbs. And apart from this sub-set (with the single exception of hope), these verbs are what Hopper (I975) calls 'weak assertive predicates', and Lysvag (I97 5) calls 'verbs of hedging'. Given these properties, it is predictable that they will occur with so as the governed predicate anaphor, rather than it -so carrying no presupposition as to the truth value of the predication in terms of which it is interpreted (c£ also Lysvag I 975: I 27). 5.2
Verbs hosting it only
(24) below contains a list (also non-exhaustive) of host verbs restricted to taking it as the governed propositional anaphor. (24) acknowledge, be surprised (at), claim , deny, doubt, ignore, know, prove, realize, regret, understand . . . Unlike the verbs listed under (23), those appearing under (24) (with the exception of doubt and possibly also claim ) are factive: that is, they carry a presupposition on the part of the speaker that their complement proposition is true (see Kiparsky & Kiparsky I 97 I }.4 As such, they require a direct object syntactically and a single internal argument semantically. As to their semantics, they are 'strongly assertive predicates', in Hopper's { I 975) terminology. Doubt is also, in a sense, a strongly assertive predicate, since it is used to assert that the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Let us examine now the issue of the influence of the semantics of the particular host verb on the selection of so or it as the predicate or propositional anaphor.
F. Cornish
17 3
5·3
Verbs capable ofhosting either so or it
The most interesting cases, however, are the type of verbs listed under (zs) below. These are verbs which may take either so or it as their anaphoric complement, but with a systematic change in sense.
(zs) assume, be afraid (of), believe, do, expect, Jear, guess, imagine, say, seem , tell, wish . . . The interesting thing is that, when it is selected, the host verb has its full, relic, strongly assertive sense, as is the case with the verbs in Mary assumed it, they are afraid ofit, I believed it,Jack did it, we expected it,joe andAnnefeared it, you'veguessed it, she imagined it, you said it, it seemed it, they told me it, I wished it . On the other hand, when constructed with so , the sub-set of these verbs which involves the expression of a psychological or propositional attitude (that is, all the verbs in this category apart from do , say and tell) assume a 'weakly assertive' or 'parenthetical' sense, something like 'be of the opinion that . . . but without wishing to be committed to it'. Thus we have I assume so, I'm afraid so, I believe so , we expect so, Ifear so, Iguess so, I imagine so, , it seems so , I wish so .5 As far as the exceptions to this generalization are concerned, with say and tell , the sense in combination with so seems to be such that it is the content , not the form , of what is said or told that is in focus. I will examine do so , the remain ing exception, below. The examples of'ambi-valent' verbs, verbs which may govern either so or it, which I present below, mainly involve do , the most frequent of all verbs (apart from be). This is simply due to the fact that my corpus of examples has yielded what I consider to be particularly good examples of the phenomenon I am interested in here, involVing do as the governing verb in the anaphoric clause.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
speaker cannot accept the validity (that is, positive truth value) of the comple ment proposition. In other words, the speaker does take a definite stance (c£ Cushing 1 972) with respect to the truth of the complement proposition, namely that it is false. Hence, all these verbs assign a particular theta role (that is, a generalized semantic role in the state of affairs described by the predicator)-namely 'Theme'-to their object argument, as well as a semantic role as a function of their sense (here, that the proposition denoted by the complement has the status of a fact). Because of this, and because of the definiteness of it (in contrast with so), such verb + pronoun combinations are telic (that is, they denote 'accomplishments' in Vendler's ( 1 967) classification of semantic verb types, accomplishments being events that terminate at a definite end point). The verb + so combinations, on the other hand, are atelic , denoting activities or states.
1 74
So Be It : The Discourse-Semantic Roles of So and It
First, example (26) involves do it , and this is clearly motivated by the fact that the predication at issue is a specific, goal-directed action (viz. 'playing his first villain'), the accomplishment of which is of considerable importance to the referent of the subject-topic, Michael Keaton. Hence this predication is relic, according to the test for telicity (c£ Comrie 1 976: 44-8): viz. 'Michael Keaton was playing his first villain' does not pragmatically entail 'Michael Keaton had played his first villain'; and it is therefore highly appropriate. The use of so here would, as predicted, have led to relative incoherence. In (27), 'making a save [of a shot at goal in football]' is again a goal-directed (pun not intended!) action: the test for telicity is again positive (viz. 'Shilton is making a save' does not pragmatically entail 'Shilton has made a save'), and it , once again, is the most appropriate choice of anaphor. So is, however, marginally acceptable here, since the sense of 'activity in progress' is to some extent salient in this discourse fragment-particularly via the use of the progressive aspect in the anaphoric verb phrase. By contrast, do so is appropriate in (28), since the framing predicational context of the anaphoric clause is that of a description of the usual role of a nurse in this particular institution: the mood is generic, or 'gnomic', the nurse in question being construed generically, not specifically. The test for telicity is in fact negative here (that is, 'the nurse is (continually) evaluating the care given' does pragmatically entail 'the nurse has evaluated the care given'. It , as predicted, would clearly be inappropriate in this context, as it would suggest 'accomplishment' rather than mere 'activity'. In (29), the contextually determined sense of believe has to be the ti.t l l
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(26) '. . . [Michael] Keaton was nervous of playing 'his first villain. People were telling him he was a big star so he shouldn't do it'/# so (Radio Times , 9- 1 s February 1 99 1 , p. 37) (27) 'Shilton has made more saves this season than any other goalkeeper, and he's doing it/? so again now' (lTV, 28 May 1 980) (28) '. . . So the care for each patient is very much tailor-made. And the nurse continually evaluates the care given, updating the system as she does so'/?# it (advertisement for 'Nursing', Observer Magazine , r o March 1 99 1 , p. 39) (29) '. . . "He went on and on, absolutely deadpan. Nobody interrupted him because nobody could believe it/# so . It was so obviously sincere" . . .' (Observer, ro March 1 99 1 , p. 2 r ) (3o) '. . . We tell you about these things not just to outrage you, but because there is something simple and effective you can do to help stop them. Join Amnesty [International]. Please. Do it/? so now' (Observer, ro March 1 99 1 ) (3 r ) 'The Americans have developed a new method o f shooting down inter continental ballistic missiles in space: they did it/# so on Sunday' (BBC Radio 4 , I 2 June 1 984)
F. Cornish
I7S
6 CONCLUSION The conclusion to be drawn from this discussion is that so and it are not, after all, in a relation of straightforward complementary distribution with one another. If this were the case, it would suggest that their occurrence is linguistically determined in terms of formal environments. Yet the thrust of the evidence and analyses presented so far has led to the conclusion that the choice between them is subject to the speaker's control in terms of his or her intensions. Table 1 below summarizes the respective properties of so and it as established in the above discussion. The choice of it or of so in particular positions is the outcome of an interaction between purely linguistic constraints (both syntactic and semantic) and the pragmatic value of the anaphor containing utterance as determined by the speaker. As we have seen, that choice may affect both the sense of the governing or host predicate and the discourse status of the situations, properties, actions, processes, events or propositions
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
'strongly assertive' sense, involving a commitment by the referent of the subject expression to a particular belief that such-and-such is the case. Hence, so would be quite unacceptable, as it would force believe to be construed 'parenthetically', that is 'non-committedly', as it were. The urgency conveyed via the predicative context in (30) (where the illocutionary stance adopted by the writer is one of exhortation to the reader to perform some specific action) seem to require it rather than so , which, if used here, would markedly weaken that sense of urgency. Finally, in (3 I), the anaphoric predication introduces a specific event which is situated at a definite time in the then recent past (viz. 'on Sunday'). If so had been used in place of it, the 'accomplishment' sense, with its concomitant assumption of agency on the part of the referent of the subject of do it (the Americans) would be lost, and the discourse fragment as a whole would become relatively incoherent. Note that the full interpretation of do it in this context docs not require a plurality of intercontinental ballistic missiles, but only one, unlike the antecedent-trigger predication. 'Shooting down inter continental ballistic missiles in space' is in fact atelic (viz. 'X is shooting down intercontinental ballistic missiles in space' pragmatically entails 'X has shot down intercontinental ballistic missiles in space'). Yct the 'accomplishment' sense conveyed by the choice of do it as anaphor is coherent here, since it co occurs with the definite past tense and the deictic time adverbial on Sunday , creating a new (though related) sub-topic from the theme introduced in the initial statement. This is yet another piece of evidence for the relative semantic independence of overt (as opposted to elliptical) anaphors in relation to their antecedent-triggers (c£ also Garcia 1 990: n. 4, p. 3 1 7).
1 76 So Be It : The Discourse-Semantic Roles of So and It
Table
I
Summary of the properties of anaphoric so and
it
Parameter
so
it
Syntactic category: Type of reference: Semantic status: Nature of antecedent (� 'antecedent-trigger'): Aspcctual status of host verb + anaphor combinations:
Adverbial Intensional Indefinite {Complex) Predicate Atclic
Nominal Extensional Definite Mutually validated Proposition Tclic
2.2.2.
Acknowledgements This article is based on a paper entitled 'So and it : complementary distribution, or comple mentary anaphoric function?' which I presented at the sth Annual Symposium on the Comparison and/or Description ofEnglish and Greek, held by the School of English, Aristotle University, Thessaloniki, Greece on 28 March 1 99 1 . I would like to record my thanks to Alan Garnham for his helpful comments on a preliminary version of the article, as well as to the rwo anonymous referees who reported on the version submitted to this Journal. FRANCIS CORNISH School ofEuropean and Modern Language Studies Cornwallis Building University ofKent Canterbury England
N O TES I am indebted to one of the anonymous Journal ofSemantics referees for this latter point. 2 There is indeed a diffuse anaphoric pro perry encoded in the conjunction so , as one of the anonymousJS referees pointed out its pragmatic value is something like 'To 1
return to the topic we were dealing with earlier . . .' However, it is equally clear that it cannot have the anaphoric values or roles assumed by the predicate anaphor so . Napoli ( 1 989: 1 9-2 1 ) distinguishes modi fication from predication, by saying that a modifier serves to construct the reference
·
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
which are already salient in the mental discourse model at the point where the pronouns are used; and it may even actually serve to create new such referents on the basis of already existing ones. At all events, I think it is clear that the situation regarding anaphoric it and so is quite a bit more complex than it might at first sight appear, on the basis of the distributional data given in section
F. Cornish I 77 involving anaphoric so where the subject o f its governing verb is third person, and where the verb is in the definite past tense (both of these features suggesting reference to some 'objective' reality, rather than to a subjectively undergone experience): (i) ?Jane believed/was afraid so . This restriction does not apply to it, how ever, as the examples illustrating it immediately below example (2 5) demon strate. In fact, as the reader will already have noted, with two exceptions, all these examples of propositional it are in the definite past tense. Changing this to the present would in certain cases result in unacceptability or extreme oddity (c.f., e.g., ? # you guess it). I am indebted to Alan Garnham (p.c.) for pointing this out.
R E FE R E N CE S Ariel, M. ( I 990), Accessing Noun Phrase Ante cedents , Routledge, London and New York. Bolinger, D. ( I 970), 'The meaning of do so ', Linguistic lnquiry , 1: I, I 40-4. Bosch, P. ( I 98 3), Agreement and Anaphora , Academic Press, London. Bouton, L. F. (I 970), 'Do so : do + adverb', in J. M. Sadock (ed.), Studies Presented to R. B. Lees by his Students . Linguistic Research, Edmonton, I 7-38. Comrie, B. (I 976), Aspect , Cambridge Uni versity Press, Cambridge. Cornish, F. (I 986), Anaphoric Relations in English . and French: A Discourse Perspective , Croom Helm, London & Canberra. Cornish, F. (I 99 I), 'Non-discrete reference, discourse construction, and the French neuter clitic pronouns', journal 4 French LAnguage Studies , 1:2, 1 2 3-38. Cushing, S. (I 972), The semantics ofsentence pronominalization', Foundations 4 Langu age , 9:2, I 86-208. Dik, S. C. ( I 989), The Theory of Functional
Grammar, Part I: The Structure 4 the Clause, Foris, Dordrecht. Ehlich, K. ( I 982), 'Anaphora and deixis: same, similar or different?', in R J. Jarvella and W. Klein (eds), Speech, Place and Action , Wiley, Chichester & New York, 3 I 5-38. Fraurud, K. (I 990), 'Definiteness and the pro cessing of NPs in natural discourse', journal ofSemantics , 7:4, 395-434· Garcia, E. C. (I 990), 'A psycho-linguistic crossroads: frequency of use', journal of Semantics , 7:3, 30I - I 9. Hankamer, J. & Sag, I. A. ( I 976), 'Deep and surface anaphora', Linguistic Inquiry , 7=3, 39I-428. Hopper,]. B. ( I 97 5), 'On assertive predicates', in J. P. Kimball (ed.), Syntax and Semantics, Vol. 4, Academic Press, New York, 9 I- I 24. Johnson-Laird, P. N. (I98 3), Mental Models , Cambridge University Press, Cambridge. Kiparsky, P. & Kiparsky, C. ( I 97 I ), 'Fact', in D. D. Steinberg & L. A. Jacobovits (eds), Semantics: An Interdisciplinary Reader in Philosophy, Linguistics and Psychology ,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
of the NP whose head noun it modifies, and is therefore not independent of that reference; predication, however, involves the application of a predicate (in the shape of a predicator) to an argument (an NP, syntactically) whose reference is inde pendently determined. 4 In the case of claim , there is a strong assumption attached to the use of this verb that the complement proposition is true; and in that of deny , there is an assertion by the speaker that it is false (i.e. not true). Thus deny certainly is a 'strongly assertive' predicate. Note that all of these examples involve first-person subjects, with the governing verb in the present tense. These properties underscore the 'subjectivity' of the sense of so which I am trying to explicate. It is in fact very difficult to construct an example
I 78 So Be It : The Discourse-Semantic Roles of So and It Cambridge Universiry Press, Cambridge, 345-69. Lysv:ig, P. {I 975). 'Verbs of hedging', in J. P. Kimball (ed.), Syntax and Semantics , Vol. 4, Academic Press, New York, I 2 5-54· Napoli, D.-J. (I 989), Predication Theory, Cambridge Universiry Press, Cambridge. Oim, H. { I 973). 'On che semantic creacmenc of predicacive expressions', in F. Kiefer & N. Ruwec (eds), Generative Grammar in Europe , Reidel, Dordrechc, 360-86.
Peterson, P. L. ( I 982), 'Anaphoric reference co faces, propositions, and events', Linguistics and Philosophy , s. 23 5-76. Sag, I. A. & Hankamer, J. ( I 984), 'Toward a theory of anaphoric processing', Linguistics & Philosophy , 1. 325-45 · Vendler, Z. (I 967), Linguistics in Philosophy , Cornell Universiry Press, Ichaca, NY.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
