JOURNAL OF
SEMANTICS VOLUME 4, No.2. 1985
SwETS 11A
I>S 1"$
SEil�ICI!
Repnntcd voith pcnmSSJon of the orig�nal publishers by
S\VETS & ZEJTLINGER BV LISSE
-
n1E ffiTHERLANDS
-
1997
JOURNAL OF
SEMANTICS VOLUME 4, No. 2, 1985
Reprinted with permission of the original publishers by
SWETS & ZEITLINGER BV LISSE THE NETHERLANDS 1997 -
-
JOURNAL OF SEMANTICS
Volume 4, Number 2
June 1985
CONTENTS
FRITS BEUKEMA: Chronological Relations and Free Adjuncts in English. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 GUNTHER ROHDENBURG: Dogs, Bitches and other Creatures . . . 117 JAN DE VUYST: The Present Perfect in Dutch and English . . . . . . 137
Review article
ARNIM VON STECHOW: Max J. Cresswell, Structured Meanings. The Semantics of Propositional Attitudes. . . . . . . . . . . . . . . . .
165
Bookreviews
PETER ROLF LUTZEIER: Fred Landman and Frans Veltman (eds.) , Varieties of Formal Semantics
. . . . . . . . . . . . . . . . . . . . . . . . 193
THOMAS PECHMANN: J. Engelkamp and H.D. Zimmer, Dynamic Aspects of Language Processing
. . . . . . . . . . . . . . . . . . . . . . . 200
WALTER J. PERRIG and HUBERT D. ZIMMER: A.J. Sanford and S.C. Garrod, Understanding Written Language: Compre hension Beyond the Sentence . . . . . . . . . . . . . . . . . . . . . . . .
204
Journal ofSemantics 4: I 01-115 CHRONOLOGICAL RELATIONS AND FREE ADJUNCTS IN ENGLISH*
FRITS BEUKEMA
ABSTRACT
1.
Free adjuncts and Rigler 's theory of tense
Traditionally, free adjuncts are considered to be tenseless sentence adverbi als, which can occupy sentence-initial, sentence-final and, less frequently , sentence-internal positions. Examples of free adjuncts in these positions are given in (I )-(3). (1)
Having no money, I couldn't buy a computer
(2)
John felt at peace again, after having seen Mary
(3)
John, not feeling very well, had to lie down
Free adjuncts are related when they share a semantic argument with the clause that they modify . 1 In example ( 1 ) , the free adjunct having no money is interpreted in such a way that the subject of the matrix clause is coreferential with the empty subject of the free adjunct. The empty subject in related free adjuncts will be labelled PRO. The two other types * The research for this article was part of the Lciden University Research Project 'Word Order and Syntagmatic and Paradigmatic Structure', financed by the Dutch Ministry of Education. I am grateful to Bob Rigter for his comments on an earlier version.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In this article I provide a discussion of chronological relations in the con text of English free adjuncts, which is based on the theory of tense pro posed in Rigter ( 1 982). A fter a brief review of the main points of this theory, I address the problem of selecting the relevant domain in which the time perspective of the free adjunct and the modified clause are pro cessed. I then discuss the available means to express chronological rela tions in the free adjunct. The final section deals with the influence of se quence indicators like BEFORE and A FTER .
102
(4)
RELEVANT CHRONO LOGICAL RELATION
> Fn 3. -(Pn > F n) 4. (F n > T n) V (F n � T n)
2. p n
LEXICAL SELECTION If a < +tense> node If a perfect node is available : is available: +- domain +PAST shift tense If not: HAVE +PAST F-tense -PAST HAVE +- T-tense
1+-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
of free adjuncts - absolute and unrelated - will not be discussed in this article. Temporal clause-modifying adverbials establish a focal episode which serves as the reference time for the focus of the matrix clause (Vet 1 984, Rigter 1 985). As free adjuncts have no tense node, the formal means of marking chronological relations by using ±PAST is not available in verbal free adjuncts. The only remaining oppositional mechanism is therefore the selection or non-selection of the auxiliary of the perfect HAVE. Note that the choice of ±HAVE is also available in other non-finite domains, e .g. in infinitival complements, nominalizations and in what u sed to be called reduced relative clauses. In order to be able to account for the chronological rel ations holding between the free adjunct and the matrix clause, we need a tense theory that is capable of relating chrono logical conceptualisations to the formal oppositions ±PAST and ±HAVE. Such a tense theory is found in Rigter ( 1 982, 1 985). The central idea in Rigter's theory is that the referent of a tenseless proposition can only be assigned a truth value if it can be located in the chronology of the relevant intensional domain. The primary intensional domain (or PID) represents the world of the discourse. If the world of the discourse repre sents the speaker's conceptualisation of reality at the time when he speaks , the present (P) o f the primary domain coincides with the speaker's present. The PID may contain one or more intensional subdomains (or ISDs), which can be introduced by world-creating verbs such as say, think, seem, etc. Every intensional domain has its own chronology . The present P of an lSD coincides with the time when the proposition of the relevant world-cre ating verb in the embedding domain holds true. The variables in terms of which the chronology of any intensional domain can be described are P, the present of the relevant domain , F, the episode on which the speakers focuses his attention, and T, the time at which the relevant proposition holds true in the chronology of the domain. The relevant chronological relations and the lexical choices dependent on these chronological relations as described in Rigter ( 1 982) can be sum m arized as in (4).
103 NB
a> a= a< a�
b b b b
b a a b
lies in the past of a and b are simultaneous lies in the past of b begins in the past of a , and continues into a
(5)
a. b.
1 ohn thinks that Pete is ill
p l= F l= T l =P2 = F2=T2
In the primary intensional domain (=PID), which is the speaker's con ceptualisation of reality , it is true that John t'1inks something, but it is not necessarily true in the PID that Pete is ill. The truth of 'that Pete is ill' holds in an intensional subdomain (=l S D) which represents the speaker' s conceptualisation of the world of John's thoughts. The P of the subdomain in (5) is located at the T of the higher domain, i.e. at the moment at which John thinks something. The lexical selection of ±PAST and ±HAVE for a sentence such as (5) follows the schema in (4). This se lection is determined by the relation between the P of a higher and the P of a lower domain, as indicated in rule 1 of ( 4), by the relation between F and P in the same domain , as indicated by rules 2 and 3 , and by the relation between F and T in the same domain, as indicated in rule 4 of schema (4). In (6), I provide the three possible chronological relations between p n , which is the present of the domain unde r consideration, and pn-1 , which is the present of the next higher domain. Note that rule 1 in (4) can only apply in finite embedded clauses, and then only in the case of possibility (6a). The result of the application of rule 1 is the selection of a +PAST domain-shift tense in the embedded clause . This is illustrated in (7). Note 2 2 that the choice of the +PAST tense in was cannot result from the P = F relation in the in tensional sub domain. Hence , this +PAST cannot be an Ftense. (6)
a. b. c.
p n-1 > p n p n- 1 = p n p n- 1 < p n
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Domains have numbers. The highest domain is l. The first embedded domain is 2. A domain embedded in the first embedded domain is 3 . I n (4) n i s the number o f the relevant domain, and a i s any domain num ber which is n or which is lower than n (so : a � n). Consider now the application of (4) to the sentence in (Sa), which involves two intensional domains. The appropriate chronology is given in (Sb).
104 (7)
John saw that Mary was ill p i > F I= TI=p i=F2=T2
2.
The selection of the domain of computation
As stated earlier, the < +tense> option is not available in free adjuncts, which entails a drastic reduction of the formal possibilities for the ex pression of chronological relations. If we could also formulate restrictions that hold in the area of the conceptual relations between free adjunct and matrix clause , the relation between chronological orderings and their formal expression in free adjuncts becomes amenable to a rigorous and economical description. We may pose the question as to whether the free adjunct and the matrix clause are evaluated for their truth values in two diffe rent intensi onal domains, as was the case with the clauses in (5), or in one and the same intensional domain, namely, the speaker's primary intensional domain. To discover this, let us compare the sentences in (8) and (9). (8) (9)
John thought that Mary was ill, but that was not true ?Playing with our cat, I hurt my finger, but that was not true
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The choice of ±HAVE can be co-determined by domain-shift : when rule I in (4) has selected +PAST for the tense node , application of rule 2 in ( 4) cannot fill the (already occupied) tense node and thus leads to the selection of HAVE if in the embedded domain p n > F n . In (5) and (7) both the headclause and the subclause contain finite verb forms , which implies that a <+tense> node is available in the syntactic representation of these clauses. In non-finite constructions such as free adjuncts a < +tense> node is not available. The lexical choice is thus restricted to HAVE or no HAVE. HAVE is inserted if the chronological relations in rules (2 ) and/or (4) in schema ( 4) are con ceptualised in non-finite constructions. In all other cases there is no special lexical item to accommodate these chronological relations. The ordering of chronological relations within intensional domains which I have briefly summarized above is the core of Rigter's m9st recent exposition (cf. Rigter 1 985). However, the accompanying syntax is now cast in the mould of Government-Binding theory , while the semantics of chronological relations as they are expressed in a sentence or discourse are now formulated in terms of discourse domain representation struc tures. For further details I must refer to Rigter ( 1 985) and , in a broader context, to Seuren ( 1 985).
1 05
( 1 0)
From now on, the superscript a will be used to refer to the free adjunct , while the superscript b identifies the matrix clause . The numeral 1 in dicates the domain of computation ; thus, the relevant chronologies of both the free adjunct and the matrix clause are computed in the speaker's PID, of which P 1 is the presen t. In this domain , both propositions have their own chronology, i.e. r l a =rib or -(T l a = ri b ). Similarly , F l a F ib or -(F l a F 10). However, for both propositions, pi is the point of orientation. =
=
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As argued earlier, the sentence in (8) involves two intensional domains: in the PID it is true that John thinks something, but it is not necessarily true in the PID that Mary is ill. The addition of but that was not true may serve as a test to bring out the respective truth values of the propo sitions in the intensional domains in (8). The use of the anaphoric pronoun that in (8) brings about a backward reference that does not cause a contra diction. In (8) the speaker can apparently freely deny the truth of Mary was ill in his own PID without violating the truth of Mary was ill in the subdomain. Since two propositions that have contradictory truth values cannot be true in one and the same intensional domain, it follows that the utterance in (8) does not construct one intensional domain in which such a clash of increments could arise, but at least two, i.e. the PID and an lSD. Now consider the sentence in (9). Whether the reference of the pronoun that in (9) is determined by the free adjunct or the matrix clause or the constitute of free adjunct and matrix clause is immaterial : the sen tence remains contradictory. If (9) involved two intensional domains, such that, for example, the free adjunct could be true in a subdomain, and the matrix clause in the PID, or the other way about , the sentence would have to show a non-contradictory reading. In the absence of any other domain in which the truth of the three propositions in (9) could be evaluated, we are forced to conclude that the three propositions are evaluated in the same domain, i .e . in the speaker's primary domain. 2 I f this line of reasoning is correct, there cannot be any domain shift between matrix clause and free adjunct, i .e. the conceptual relation in (6a) is not in evidence. This means that the matrix clause and free adjunct are both computed in the PID. Rigter ( 1 982) represents the general format of a PID computation of two propositions as in ( 1 0), in which the chro nological relations have not yet been specified .
1 06 3.
Chronological relations and their realization in free adjuncts
,
( II )
Feeling ill, he went to bed
Application of the rules in (4) would be expected to select HAVE in the f ree adjunct in ( II ), which we do not discern in ( 1 1 ), however. The implication would seem to be that rule 2 in schema (4) does not apply in free adjuncts. If this observation is correct, the occurrence of HAVE in free adjuncts is attributable solely to rule 4 in schema (4) . Notice , how ever, that, although the F-P relation does not play any discernible formal role, the F of the free adjunct is relevant to rule 4 in schema (4) and also to possible specification by adverbials. Thus, in ( 1 2) there is both an F specifying adverbial at that moment and a T-specifying adverbial only half a minute before in the free adjunct . ·
( 1 2)
Having heard the news only half a minute before at that moment , he was not in a position to give any comment
C onsider now the chronology represented in ( 1 3). ( 1 3)
Signalling his wife to raise he r glass, he said a few words of thanks PRO i SIGNAL . . . HE i SAY . . .
A s indicated by the two different chronologies in ( 1 3), the event referred to in the free adjunct may e ither be simultaneous with or anterior to the i event in the matrix clause. Let us first examine the possibility that r a is 1 b simultaneous with T , as is indicated in ( 1 4a).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Considering the application of the system in (4) to free adjuncts - i.e. to tenseless structures evaluated in the same domain as the headclause we see that the first rule is not applicable because there is no finite tense , and that the two remaining conceptual relations- the F-P relation and the T-F relation - must be associated with the lexical selection of HAVE , where F-tense representation has precedence over T- tense representation . Having delimited the set of possible conceptual relations by the elimina tion of domain shift, we can narrow down still further the system in (4) in its application to free adj uncts. To see this, consider the sentence in ( 1 1 ) in which the F of the free adjunct is in the past of P, the speaker's present.
1 07 ( 1 4)
a.
/
( 1 4)
> pl a
b.
=
Now let us examine the second chronology indicated in ( 1 3) , in which a I T precedes T I b . This is indicated in ( 1 5a). ( 15)
a.
> pl b
=
As already indicated in ( 1 5 a) , the chronology of the matrix clause in ( 1 3) is identical to that of (I 4b ). Since the free adjunct does not contain HAVE, T l a=pia . The two remaining relations can only be filled in as is done in ( 1 5b ), because any other solution would result in a logical inconsistency . (15)
> pl a
b. pl
Tl a J\
J\ > pl b
=
Tl b
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
i n ( 1 4a) T l a is simultaneous with T 1 b . The other chronological relations follow automatically, as we shall see. In the free adjunct in ( 1 3), there is no HAVE. This entails that pi a = T 1 a . The matrix clause in ( 1 3) does not contain HAVE either. As we are dealing with a finite construction l l in the PID, this means that pib = T ib . If T l a= T 1 b and p a = T b la lb 1 b l b and p = T , it follows that p is simultaneous with p . The occur rence of said in the m atrix clause of ( 1 3 signifies that, as we can read off from schema (4), pi b is in the past of P . Entering all these relations into ( 14a) , we conclude that ( 1 4b) represents the chronology of ( 1 3) in the event that the signalling and the speaking are co-temporaneous .
1 08 A comparison of ( 1 4b) with ( I Sh) shows that it is possible either to keep the focus constant between free adjunct and matrix clause, or to shift it. 5.
Sequence markers in free adjuncts
Let us now consider the influence that sequence markers such as WHILE, AFTER, BEFORE may exert on chronological relations expressed in free adjuncts. Consider ( I 6)-( 1 8). While signalling . . .
,
he said
( 1 7)
After signalling . . .
,
he said
Tia < Tib
( 1 8)
Before signalling . . .
,
he said
Tia >T1 b
We have argued just now that it is only rule 4 in schema (4) that applies to free adjuncts. If a free adjunct does not contain HAVE, as is the case in (1 6)-( 1 8), then rule 4 states that T does not ( artially) precede F. f It may thus be assumed that in ( 1 6)-( 1 8) Tia = F a, as is indicated in ( 1 6')-(I 8'). ( 1 6')
>Fib ( I 7' )
a
(?) F i A
( 1 8')
Ti b =
y1a A
b >F i
yib
ia
y1 a
().) F
@
>Fib
v
y1b
What does the addition of WHILE, AFTER and BEFORE contribute? The chronological semantics of these sequence markers is presented after the examples in ( I 6)-( I 8). These relations are also accommodated in ( 1 6 ')-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( 1 6)
109 (18'). The relation holding between P and F 1b and between F 1 b and rl b is identical to (14b) and (15b). The circled chronological relations in (1 6')-(18') can be derived by logical deduction, with the exception of the relation between P1 and F 1a in (18), which is enclosed by a dotted circle. But even in this case, it may be argued that (F 1a T1a) lies in the past of P1 although, formally speaking, F 1a is at some unspecified moment with respect to P1. However, on pragmatic grounds, it may be assumed that the signalling has already taken place at the moment P when the speaker reports about the signalling. Let us now apply what we have done in (1 3)-(18) to the same free adjuncts, this time containing HAVE. Parallel to (13), we now get (19). ==
Having signalled his wife . . . , he said . . .
The possibility that ri a== ri b is excluded in (1 9), because the event in the free adjunct is anterior to the event in the matrix clause. We can therefore plot out the following chronological relations in (20a). (20)
a.
> pl a
A
Rule 4 of schema (4 has inserted HAVE in the free adjunct on the grounds that F 1a >T a. Two relations are still open in (20a): F 1a - F 1b and P1 - F 1a. If we assume that the speaker does not shift his focus, (20b) gives a complete picture of the chronological relations in (19).
1
(20) b.
>
>F ib
::::
Let us review what we have done so far. We have argued that rule 4 in the rule schema in (4) operates in the selection of ± HAVE in free ad· juncts. If we accept (4), then signalling in (13) expresses the chronology F la ==ria, while having signalled in (19) expresses that F l a >ria_ Thus,
the different linguistic signals in (13) and (19), i.e. the absence or presence of HAVE, are correlated with a difference in chronological perspective, which is ascribed to rule 4 in schema (4). This rule is crucially concerned with F-T relations, thus involving the focal episode in the lexical selec-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(19)
110 tion of HAVE or no HAVE. F thus appears to be a relevant variable in the chronological computation of free adjuncts on theory-internal grounds. As external corroboration I have adduced evidence based on F specification by adverbials (cf. (12)). Rigter (1985) addresses the issue of the relevance of F in a principled fashion. Using the chronology of examples (21)-(24) as evidence, he fol lows the tradition based on Reichenbach (1 947) in assuming that in simple headclauses a focal episode (i.e. F) must be postulated, because (22) (24) are formally different. This formal evidence cannot be explained if F is not involved. Jane is angry
p
=
F
T
p = T
(22)
Jane has been angry
p
=
F >T
p >T
(23)
Jane was angry
p >F = T
p >T
(24)
Jane had been angry
p >F >T
p >T
=
It is well-known that, when (22)-(24) are embedded under a raising verb such as APPEAR, the formal differences are neutralized, as appears from (25), in which Jane has been moved to the subject position of APPEAR. (25)
Jane appears to have been angry
Rigter notes that the chronology expressed by the non-finite embedded clause is left vague by the speaker who uses (25). If the speaker wants to be explicit, he can embed (22)-(24) under APPEAR as finite clauses. However, when interpreting (25), the hearer is free to infer on the basis of contextual support whether the embedded non-finite clause has a focal episode or not, and if so, whether this is on P (as in (22)), or on T (as in (23)) or between P and T, as in (24). Thus, the F of the embedded non finite clause in (25) can only be 'inference-driven'? I shall now explore what happens if in free adjuncts with HAVE we use the same sequence markers in (16)-(18). Relevant examples are provid ed in (26)-(28). (26)
While having signalled his wife
he said
(27)
Before having signalled his wife . . .. ,
he said
(28)
After having signalled his wife . . . . ,
he said
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Disregarding F:
(2 1 )
Ill
(29)
a.
>
The speaker makes a statement about the past, both in the matrix clause and in the free adjunct. On the assumption that the focus is kept constant, i.e. F1 a F1b , the picture in (29b) emerges. =
(29)
b.
>
Keeping the focus invariant would lead to problems in the case of example (27), in which the free adjunct is introduced by BEFORE. A logical in consistency would arise, because the use of BEFORE implies that the chronological relation between T1a and T1b in (29b) would have to be inverted when we replace AFTER by BEFORE. Let us for the moment keep all the other relations as they are in (29b). The result would be (30), which is an unacceptable diagram. (30)
*
> v =
A
Tlb
8
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Of these three examples, (28) is clearly the best. Example (26) is uniform ly rejected by informants, the reason being that WHILE implies simul taneity. The sentence in (27) is accepted by informants with some reluc tance; a much improved version would have had said in the matrix clause. Let us first consider the best example, i.e. (28). In (16)-(18) F I a is non distinct from Tl a and Fib non-distinct from T1b, so that it has not become clear from the analysis of (16)-(18) whether WHILE, BEFORE and AFTER affect the location in time of T or of F. This is what we have to decide first. Let us therefore consider examples in which these operators occur and in which the T of the free adjunct precedes the F of the free adjunct. Assuming now that AFTER applies to T, we must accept (29a) as a first approximation of the chronological relations holding in (28).
112
If we follow the path of arrow A, F1a and T I b are simultaneous. If we follow the path of arrow B, T 1b is in the past of F I a. This logical incon sistency makes (30) unacce able. If now we accept instead that BEFORE applies to F, i.e. F l a >F b, then our assumption of constant focus has to be given up. We then arrive at diagram (3I ).
w
(3I)
> v
v
ceptability of (30) and the acceptability of (31 ) that BEFORE is to be regarded as affecting F, and not T. Returning to AFTER, we see that AFTER, like BEFORE, can be regarded as affecting F. Consider the well formed (29c) as an alternative to (29b). (29)
c.
> A
@
Notice that two relations are implicit in (29c), as indicated by the circles: pia and Tla - rib_ We can arrive at the conclusion that the event referred to in the free adjunct is in the past of the event referred to in the matrix clause, i.e. rib >T13 -which is, of course, the natural interpre tation of example (28) - if we accept the following:
p
-
- if AFTER, just like BEFORE, applies to F, then Fib >F l a in (29c). - the occurrence of HAVE in the free adjunct in (28) allows us to deduce that rla is in the past of F1a, in keeping with rule 4 of schema (4). - the non-occurrence of HAVE in the matrix clause allows us to deduce that T I b is not in the past of F I b. This concludes my discussion of the chronology of free adjuncts in rela tion to the chronology expressed in the modified clause. What has emerged from this discussion is that free adjuncts and their modified clauses are evaluated for their chronologies in the same intensional domain. Since free adjuncts are non-finite, they do not contain a <+tense>node.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This picture is internally consistent as far as the conceptualisation of the relevant chronologies is concerned; the two possible paths from Fia to T1b both allow F1a to precede T1b. We may conclude from the unac
113 Therefore, the domain-shift tense - the first rule in (4) - will never manifest itself in free adjuncts. It has appeared that P-F relations do not affect the selection or non-selection of HAVE in free adjuncts, so that the occurrence of HAVE in free adjuncts is the formal expression of a F >T relation. Assuming that the focal episode (i.e. F ) is crucially involved in the selection of HAVE, I have argued that sequence markers like BE F ORE, AFTER, WHILE, affect the position of F and not ofT.
NOTES l. Though it is sometimes suggested (see e.g. Mohanan 1 9 83) that the sentence be comes ungrammatical or stylistically dubious when other semantic arguments are shared between related free adjuncts and the modified clause than the subject argu ment, this is a misconception. The following examples appear to be fully gram matical, and show that the antecedent may have different syntactic functions.
(i) PRO having no money, the idea hit William to rob the bank (ii)PRO having no money, it became clear to Mary that she had to rob a bank There is, however, a strong tendency for the subject argument to be shared between the related free adjunct and the modified clause. See Beukema ( 1 985) for further details. Backlund ( 1 984) also provides numerous examples of control from object position. See e.g. pp. 1 7, 25, 33, 5 2 , 69, 1 08 , 1 1 4 etc. of her book. 2. The test for intensionality that I have used here is a simple adaptation of Frege's ( 1 892) discovery that, especially under world-creating predicates, intensional domains are created where co-extensive terms cannot freely replace each other sal· va veritate. The co-extensive terms Frege referred to are definite NPs. In my analysis they are the referents of propositions and the anaphoric definite pronoun that. The test of existential generalization, which is often looked upon as a standard test for intensionality, is less suited to demonstrate that free adjunct and modified clause contribute to the construction of one and the same domain. It is less suited because it does not apply directly to the truth of propositions in that domain, but to the presence in the domain of discourse addresses (see Seuren 1985) referred to by def inite NPs in the free adjunct and the modified clause. It must be assumed that, if free adjunct and modified clause contribute to the construction of one and the same domain, it is impossible for the free adjunct to refer to a discourse address which can not be present in the domain of the modified clause. Conversely, it is impossible for the modified clause to refer to a discourse address which cannot be present in the domain of the free adjunct. Falsification of this assumption would constitute an argument for entering free adjunct and modified clause in different intensional domains. It is my view that this assumption will resist falsification, but, of course, this conviction will not serve as proof. However, consider the following example.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Department of English University of Leiden POBox 9515 . 2300 RA Leiden The Netherlands.
1 14 (i) Looking for the unicorn, John wore out his shoes
�
(=( 1 1 ))
(i) PRO feeling ill, he went to bed INFL
(ii)
D
--------'
y2
PRO FEEL ILL (t=T ]
INFL
1
1
� l 6
INF
2
+P ST
HE GO TO BED
[t =P> F J
[ t =T]
2 The v node of the free adjunct is syntactically adjoined to the clause that it mod ifies. Suppose that the internal structure of the free adjunct is such that there is no 1 c-commanding chronology indicator. Then the T of FEEL is in P , i.e. it would be located in the speaker's present, which is contradicted by the meaning of (i). The T 1 of FEEL is clearly in the past of P , as noted before. This implies that the free ad j unct should contain a chronology indicator c-commanding the verb FEEL. The log ical candidate is the F, because T is involved with F in chronological relations. Syntactically. there must therefore be a node in the free adjunct that c-commands the verb FEEL and bears the relevant chronological specification. This node could then be held responsible for the induction of the verbal -ing morphology. Such a node could be a manifestation of the [-N -V] INFL node described in Reuland ( 1 983).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
It might be suggested that whereas John and his shoes really exist, the unicorn is likely to be an intensional entity. Yet, this observation does not constitute a falsif ication of the assumption referred to above, since the predicate look for is itself intensional with respect to its object term : it allows for the selection of a discourse address for its object term in an intensional subdomain (with respect to the domain of look for). Since only discourse addresses in the PID are required to correspond to really existing entities for truth of the discourse, it does not follow from (i) that there is a really existing unicorn that John was looking for when he wore out his shoes. See Seuren ( 1 985, chapters 5, 6) for further details. 3. Rigter ( 1 985) also contains a discussion of focus in free adjuncts. His discussion provides independent evidence for the presence of F in the chronology of free ad j uncts. In Rigler's proposal verbs are provided with the chronological specification (t=T], in which the arrow is an instruction to search for the nearest, syntactically c-commanding. chronology indicator. A node A c·commands a node B iff A does not dominate B and the first branching node dominating A also dominates B. When there is no c-commanding chronology indicator available, Ri er assumes that the arrow deictically points to the moment of speech, i.e. to the P shared by the free adjunct and the modified clause. Now consider the example in (i) and the syntactic represen tation in (ii).
1 15 REFERENCES
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Backlu nd , I., 1984 : Conjunction-headed Abbreviated Clauses in English. U ppsala: Studia Anglistica Upsaliensia, vol. 50. Beukema, F.H., 1 98 5 : Lexical structures in English free adjuncts. In : G .A.J . Hoppen brouwers, P.A.M. Seuren & A.J.M.M. Weijters (eds), Meaning and the Le'Cicon . Foris, Dordrecht : 190-199. Frege, G., 1 89 2 : Ueber Sinn und Bedeutung. Zeitschrift fur Philosophie und Philo sophische Kritik, 1 00 : 25-50. Mohanan, K.P., 1 983: Functional and anaphoric control. Linguistic Inquiry, 1 4 : 64 1 -6 7 5 . Reichenbach, H . , 1 94 7 : Elements of Symbolic Logic. Macmillan, New York. Reuland, E.J ., 1 983: Government and the search for auxes: a case study in cross linguistic category identification . In F. Heny & B. Richards (eds), Linguistic Categories: Auxiliaries and Related Puzzles. Reidel, Dordrecht. Vol. 1 : 99- 1 6 8 . Rigter, G . H . , 1982 : Intensional domains a n d the use o f tense, perfect and modals in English. Journal of Semantics 1 : 94- 1 4 5 . Rigter, G .H., 1 985 : Focus matters. In V. Lo Cascio & C. Vet (eds), Temporal Representation in Sentence and Discourse. Foris, Dordrecht. Seuren, P .A.M., 1 985 : Discourse Semantics. Blackwell, Oxford. Vet, C., 1 984 : Is there any hope for the 'futur'? In H. Bennis & W.U.S. van Lessen Kloekc (eds), Linguistics in the Netherlands 1 984. Foris, Dordrecht; 1 89-196.
Journal of Semantics 4:117-135 DOGS, BITCHES AND OTHER CREATURES
GUNTHERROHDENBURG
ABSTRACT
1.
In a closely argued and thought-provoking article Kempson ( 1 980) re considers the traditional distinction between the concepts of homonymy and polysemy. Disagreeing with Lyons ( 1 97 7) she begins by suggesting that it is homonymy that should be maximalized with polysemy being restricted to certain special phenomena. The next point she makes in volves a constraint on sentential ambiquity: It was pointed out in Zwicky and Sa dock 1 97 5 that any proposed cases of sentence ambiguity where the sentence in question is claimed to have two interpretations, one entailing the other, defy any of the standard
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In an attempt to uphold a specific constraint on sentential ambiguity Kempson ( 1 980) has proposed an elaborate framework for dealing with the semantic duality of items such as dog. Her analysis culminates in a duality principle relating the specific and the general interpretation of the terms in question. These proposals are shown to have a number of serious shortcomings. First, they do not allow for the fact that the two relevant interpretations may vary in strength, and that independently of each other. Second, the duality principle is inadequate in several respects. In a number of sets containing just two members like dog and bitch, the prin ciple is incapable of establishing precisely the specific interpretation of the general term. Nor does it correctly predict the absence of such specific uses in certain cases. Worse still, the principle is incompatible with the fact that the kind of duality under study is also found in a great variety of multiple sets. In addition, there are analogous phenomena associated with various types of part-whole relations which are not captured by the principle. Finally, as was shown in Horn ( 1 983), case in which a given term functions as its own hyponym may originate in other ways than those predicted by Kempson's thesis.
1 18 test procedures for ambiguity. In Kempson 1 97 9 , however, it is sug gested that wherever such an ambiguity is claimed, the sentence is invariably unambiguous with respect to the proposed distinction but has only the more general of the two interpretations as its semantic representation . . . . The proposal is that lexical ambiguity is excluded in those cases where the lexical ambiguity is reflected in a corresponding sentential ambiguity of which one interpretation entails the other. In accordance with the ambiguity constraint outlined above , any such cases are predicted to have only the more general of the two inter pretations in question. (Kempson 1 980 : 10-1 1)
( I)
That's not a dog: it's a bitch.
( 2)
They've got four dogs including a bitch and two puppies.
According to the ambiguity constraint "dog must be entered in the lexicon as unambiguous, denoting merely the entire class of canines" (Kempson 1 980: 1 2). However, this does not account for the sex-restricted inter pretation of dog nor does it correctly predict the "non-contradictory" nature of (I). In an attempt to reconcile the behaviour of dog and similar items with the ambiguity constraint Kempson considers three different solutions to the problem. Under the analysis adopted dog is treated as a single polysemous item. In addition, the notion of polysemy is redefined in terms of the following two conditions (I.e. 1 4) : In certain contexts the two relevant interpretations can be simultaneously present. II All cases of polysemy arising in natural language can be pre dicted by general rule. Both conditions are satisfied by concrete nouns such as book which are also associated with an abstract interpretation. Dog and similar cases certainly meet the first condition . But is their semantic duality really predictable by general rule? Kempson believes it is. Considering the additional contrasts in (3) to (7) she proposes a principle relating the general and specific interpretations of items such as dog.
(3)
That's not a duck : it's a drake .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Briefly, the constraint on sentential ambiguity stipulates that no sentence should be assigned two (or more) distinct representations if one inter pretation entails the other(s). This raises the question as to how the semantic duality of, for instance , dog as in ( I ) and (2) should be characterized .
1 19 (4)
That's not a cow: it's a b ul l .
(5)
That's not a cow : it's a calf.
(6)
That's not a line : it's a curve .
(7)
That's not a rectangle : it's a square.
The duality principle is provisionally stated in the following terms:
Kempson notes, however, that the principle in its preliminary version fails to capture cases like (8) and (9) . (8)
That's not a book: it's a novel.
(9)
That's not a room: it's a lounge.
Unlike ( 1 ) and (3) - (7) these examples "seem contradictory" (I .e. IS). This is because there are not any readily available specific uses of book and room which are incompatible with novel and lounge, respectively . Com paring examples like (8) and (9) on the one hand and examples like (3) (7) on the other, Kempson concludes that the occurrence of both a general and a more specific interpretation in items such as dog is restricted to cases involving only one lexically designated subset along any one dimension of contrast . Accordingly the duality principle is revised as follows : If a lexical item L1 has as its extension a set s1 which includes the set S2 which a second lexical item L2 has as its extension, and s2 is the only lexically designated subset of the extension of L1 along any one dimension of contrast , then the lexical item L1 may be used to denote that subset of s1 which ex cludes S2. (Kempson 1980: 1 5)
The preceding discussion makes it clear that the initial if is to be read as 'if and only if. In a previous article (Rohdenburg 1985), I have discussed a greater range of data bearing on the problem under investigation. Drawing on these and some additional data I now want to show that the discussion leading up to the duality principle contains two erroneous assumptions and - above all - that the revised principle still falls short of its goal . In
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
If a lexical item L1 has as its extension a set s1 which includes the set s2 which a second lexical item L2 has as its extension, then the lexical item L1 may be used to denote that subset of s1 which excludes S2. ( K empson 1 980: 1 5)
1 20 Horn ( 19 83), a wide-ranging attack is launched on Kempson ( 19 79 , 19 80) and a number of other papers which also seek to support the constraint on sentential ambiguity . Unlike Horn's work the present paper is exclusively concerned with matters of lexical semantics. In this area, Horn's evidence is largely complementary to mine , and some of his findings will be men tioned below.
2. 2.1.
( 1 0)
?That isn't a lion : it's a lioness.
In addition , Horn ( 19 83: 1 1 3) finds (7) "somewhat odd". My British in formant, however, agrees with Kempson on this point . In either case , it is assumed that there is no specific interpretation of the superordinates con cerned. Yet this is not necessarily a valid argument. Constructions like ( 1 0) may remain unacceptable despite the fact that the general term has a more specific use alongside its general interpretation . In that case, the specific interpretation simply is not readily available in contexts of type ( 1 0), though it may occur in other contexts . Consider in this respect the following examples : 1•2 (I I )
a . Leather and suede need special cleaning, . . . (Gundrey) b . ? That's not leather : it's suede .
( 1 2)
Haunt Fields, pastures , marshes, tidal mud-flats and moors (Benson) b .?? That's not a field : it's a (cow) pasture.
( 1 3)
a.
Fresh Meats, Purveyors of Quality Meat , Poultry and Game (London advertisement) b . ? That's not meat : it's game.
( 1 4)
a . Nest . . . in b rambles, creepers or a bush (Benson) b . ?? Those aren't bushes : they are b rambles.
a.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Apparently Kempson believes that contrasts like that in ( 1 ) are necessarily diagnostic of the absence of any specific interpretation associated with the general (or superordinate) term in question . This is an assumption which is shared by other linguists including Zwicky & Sadock ( 19 75: 7 -8) and Horn ( 19 83: 1 12 - 1 1 3). Zwicky & Sadock ( 19 75: 8) point out that ( 1 0) is unacceptable as opposed to ( 1), which is fine .
121 a. They also stock pianos and grands b . ? That's not a piano : it's a gran d .
( 1 6)
a. Haunt Lakes, meres and river flats (Benson) b . ? That isn't a lake : it's a mere.
( 1 7)
a. Beers, Ales and Lagers (London advertisement) b . ? That isn't beer: it's lager (ale).
( 1 8)
a. Those are all my trousers and b reeches. b. ? Those aren't trousers: they are b reeches.
( 19)
a. The recipe calls for a pound of fat or suet. b. ? That isn't fat : it's suet.
(20)
a. He will . . . point to toys and objects that he wants. (Boots) b .?? That's not an object: it's a toy.
(2 1 )
a.
locust . . . (kinds of) migratory African and Asian winged insect which . . . destroys crops and vegetables (Oxford Advanced Learner's Dictionary of Current English) b .?? Those aren't crops : they're vegetables.
(22)
a.
Ideally begonias for pots and containers should be started in a heated greenhouse d uring February or M arch . . . (Spalding) b .?? That's not a container: it's a pot.
(23)
. .. the heavily-built Pyrotherium, which was very similar to the early elephant Phiomia in its trunk, t usks and teeth (Cox) b . ?That's not a tooth: it's a tusk .
(24)
a.
Home can be a dangerous place for an infant or child with in satiable curiosity , . . . (Boots) b . ? That's not a child : it's an infant.
(25)
a.
An understanding o f these fundamental cycles is essential in caring for b ab ies and children . . . (Gay Gaer Luce) b . ? That's not a child : it's a b aby.
(26)
a.
a.
All these factors apply also to toddlers, and to a lesser extent , to children too. (Parish) b . ? That's not a chil d : it's a toddler.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( 1 5)
1 22 a.
This will help you decide that the use of s uch drug preparations in b ab ies and infants is . . . unnecessary . . . (Parish) b . ?That's not an infant : it's a bab y .
(28)
a.
Two 3-pint (2 litre) saucepans for vegetables and potatoes (Burb ridge) b . ?That's not a vegetable: it's a potato.
(29)
a. He left his coat at home, b ut he did b ring a mac. b . ? That's not a coat: it's a mac.
(30)
a. Eyries and nests were protected , . . . (Falkus) b . ? That's not a nest: it's an eyrie.
(3 1 )
a . The analysis is restricted to vowels and diphthongs. b . ? That's not a vowel : it's a diphthong.
(32)
a.
A great variety of pictures and photographs will also be on display. b . ? That's not a picture : it's a photograph .
(33)
a. She has taken all the curtains and nets down . (Townsend) b . ? That's not a curtain: it's a net.
As is suggested in Rohdenb urg (I 985 : 69-70) we can distinguish at least three b road categories of contextual contrast : weak , strong and very strong. The strength of the specific interpretation may accordingly be rated as weak, strong and very strong. The examples in ( I 1 a) - (33a) and ( 1 1b) - (33b) exemplify the weak and strong categories, respectively . The weak contexts include coordination by means of and as in (1 5), various forms of (mere) juxtaposition in the same sentence or in neighbouring sentences as in (29) and coo rdination by means of or as in ( 1 9). In the latter case, the sentence may not be construed as presupposing the truth of one , and only one, of the relevant disjunctive propositions. Alternative questions (cf. Quirk et a/. 1 9 72 : 395), which presuppose the truth of one, and only one, of the propositions put forward by the speaker (Lyons 1 977 : 762), defmitely belong to the category of strong contrasts. This category is represented in ( I Ib) - (33b) by statements in which one term is predicated and the other negated. Apparently, the strong contexts are among those in which the occurrence of two items standing exclusively in a hyponymous relationship leads to a contradiction while the weak con texts would only produce tautology in such a case . Unlike ( I I) (33) both categories involve fully acceptable examples in (34a-b) and (35 a-b). -
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(27)
1 23 (34)
a. Cats and Kittens (An Observer Book) b. Is it a cat or a kitten?
(35)
a . The dog has four puppies. b . That's not a dog : it's a puppy.
Exchanges such as (34c) and (35 c) illustrate the very strong contexts. (34)
c. ? A: Is it a cat? B : No, it's a kitten .
(35)
c. ? A: Is it a dog? B: No, it's a puppy.
2.2.
In several places Kempson ( 1 980 : 1 2 , 1 4 , 1 6) refers to the specific inter pretation of items like dog as an extended use or an extension of meaning. This is in keeping with the way the duality principle is formulated . Im plicit in this characterization is the assumption that it is invariably the general interpretation which is b asic or central to the lexical item. It is true that all the examples provided by Kempson and perhaps most examples of the relevant kind are adequately characterized in this way . There is , how ever, a large number of items that cannot be described in these terms. Consider such contrasts as the following: (36)
shoe - clog, monkey - ape , bed -cot , house - cottage , bicycle moped , soldier - officer, town - city , feathers - down , soup - stew , worm - caterpillar, glove - mitten , iron - steel , bicycle - tricycle, bus - coach, sauce - gravy, (fly - dragon-fly), garden - allotment , cake - biscuit , orange - t angerine , grasshopper -locust, mushroom - toadstool, raisin - currant, (swallow - (house) - martin), shoe sandal , (turtle - tortoise), (semantics - pragmatics), (table - desk),
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
What this shows is that certain specific interpretations acceptable in the weak and strong contexts are no longer fully appropriate in the very strong contexts. The contexts distinguished can be represented in terms of an im plicational scale . Any specific interpretation arising in the very strong contexts is also available in the strong contexts, and any specific inter pretation b rought out by the strong contexts is also available in the weak contexts. Although native speakers with different b ackgrounds may differ somewhat with respect to the exact cut-off point they are nevertheless committed to the implicational scale itself. Incidentally , my informant agrees with Horn and Sadock & Zwicky in rejecting ( 1 0). However, the contrast involved is judged to be acceptable in the weaker contexts.
1 24 wood - copse , man - woman, cow - bull, trousers - shorts, hat cap, plant - tree, cat - lion, (heron - bittern), meat - chicken , plum - prune, animal - bird , stocking - sock , candle - taper It is immediately obvious that the specific interpretation of the first term each case is comparatively speaking more prominent than that of the relevant term in ( 1 1 ) - (33). In quite a few cases the specific interpretation is even much stronger than the general interpretation itself. It seems possible to determine the strength of the general interpretation of a particular item by classifying the contexts in which it can function as a superordinate of its hyponym. A first step in this direction is taken in Rohdenburg ( 1985 : 68-69). It is suggested there that the use of the general term to refer anaphorically to the specific term may represent the strongest type of context. Consider, for instance , the German pair Uhr ('clock'/'watch') and Wecker ('alarm clock'), which may be contrasted in the weak contexts. Uhr is readily available as a superordinate of Wecker in contexts such as (37). in
Wecker und andere Uhren ('alarm clocks and other clocks')
However, noun phrases like die Uhr ('the clock'/'the watch') cannot normally be used to refer to an object that has been previously identified as a Wecker. Similar restrictions hold for many of the superordinates in (36). Obviously, superordinates which are only available in metalinguistic glosses (such as male cow for bull (Lyons 1 9 77 : 3 09)) must be regarded as very weak. In the area under study there is a considerable amount of variation to be found amongst native speakers. Even in the weakest con texts not all of the first-mentioned terms in each pair are accepted by everybody as superordinates of the second te rms. This is true, in particular, of the b racketed items. It might be supposed that the strength of a specific or general inter pretation correlates with the extent to which it is established in the language community. This is far from being generally true. For instance , the specific interpretation of vegetable associated with contexts such as (28a) must be regarded as generally established. Nonetheless the specific use of vegetable is restricted to weak contextual contrasts such as (28a) . T o a large extent the relative strengths o f the general and specific inter pretations of a given term may vary independently of each other. Though only one of the two uses may be weak, both uses may be very strong, as in the case of dog paired with bitch. Apparently it is only when both interpretations are fairly strong that examples like (38) and (39) begin to be accepted by native speakers.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(37)
1 25 (38) (39)
That dog isn't a dog: it's a bitch. (Zwicky & Sadock 1 975 :7) ?That lion isn't a lion : it's a lioness. (ibid.)
In cases like (I), (3) - (7), or ( 1 0) the only requirement seems to be that the specific interpretation of the potential superordinate should be rel atively strong. Here, the strength of the general interpretation is simply irrelevant . Consider the following examples involving strong specific inter pretations and weak general interpretations of the superordinate in question : a. Those are not trousers: they're shorts b .?? Those trousers are not trousers : they're shorts
(4 1 )
a. That isn't a plant : it's a tree b .?? That plant isn't a plant : it 's a tree
(42)
a. That isn't a hat : it's a cap b.?? That hat isn't a hat: it's a cap
(43)
a. That isn't a bus: it's a coach b .?? That bus isn't a bus : it's a coach
In Zwicky & Sadock ( 1 97 5 : 7) and Horn ( 1 98 3 : 1 1 2) examples like (39) are used to argue for the non-existence of a restricted interpretation of a given superordinate . We have already seen that contexts like ( 10) or ( 11 b) -(33b) cannot be relied on to make such a prediction . Comparing the two types of context in (40) - (43) we find that the diagnostic value of the second type is distinctly lower in this respect .
3.
3.1.
This bring us to the duality principle itself. Throughout Kempson glosses the specific interpretation of dog as 'male canine' or [Canine] + [MALE] . However, this is not entirely correct, since the contrast between dog and bitch is only relevant to grown-up or mature dogs (Pilch 1 976: 95). 3 But the general interpretation of dog, as is pointed out by Kempson herself (1980 : 12 , 1 4), denotes the entire class of canines. The duality principle requires that the extensions of bitch and dog in its specific interpretation should exhaust the extension of dog in its general interpretation . We have just seen that this is not the case. Therefore , the principle cannot be
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(40)
1 26 invoked to derive the specific use of dog. Nor can it be applied in most similar cases involving the dimension of sex. This is a point that is also overlooked by Horn ( 1 983 : 1 1 2- 1 1 3). In this respect , t he relationship holding between dog and bitch is paralleled , for instance , by the pairs of examples listed in (44).4
(44)
duck - drake , goose - gander, lion - lioness, tiger - tigress, man woman, bull - cow
(45)
Duuw ('mature female pigeon/dove') - Duffer ('mature male pigeon /dove'), Katt ( 'mature female cat') - Bolls ('tomcat'), Aant ( 'mature female duck') - Woort ('drake')
Naturally, the first term of each pair has also a general interpretation denoting the relevant species as such. We must conclude , then , that the presence or absence of special terms denoting the young is irrelevant to the issue under consideration . This conclusion is supported by the pair mankind - womankind, since there is no corresponding term childkind. Another important area where the duality principle fails to apply more often than not involves the dimension of (im)maturity. This includes potential contrasts such as the following: (46)
cat - kitten, sheep - lamb , dog - puppy, pig - piglet , horse - foal, lion - cub, cow - calf, goose - gosling, duck - duckling
According to the principle, the extension of the general term in its general interpretation must be neatly divisible into two mutually exclusive subsets associated with the specific term denoting the young and the general term in its specific interpretation . However, this is generally not the case in non technical colloquial English. Here , there tends to be some middle ground not covered by the two terms in question. In more technical varieties the two terms are often redefmed in such a way as to cover completely the extension of the general term in its general interpretation. It is only these technical uses that are captured by the duality principle. Much the same situation obtains in German and other European languages.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
It might be thought that the failure of the duality principle to apply in this area was due to the presence of special terms denoting the young or immature members of the relevant species. It is true t hat there are such terms alongside all of the English examples given so far. But similar pairs to those listed in (44) do exist without there being any special terms denoting the young of the species. The clearest evidence I know of comes from Northern Low German as spoken , for instance , in the area lying between Hamburg and Bremen . Consider the contrastive items in (45).
1 27 Consider now the English and German contrasts in (4 7) and (48). (47)
beef - veal, mutton - lamb
( 48)
Rindfleisch ('cattle meat') - Kalbfleisch ('calf meat'), Schaffleisch ('sheep meat') - Lammfleisch ('lamb meat')
(49)
cat - tomcat , Katze ('cat') - Kater ('tomcat')
Cat and Katze may serve as superordinates of tomcat and Kater, res pectively. Since the English terms may not be contrasted there is no specific interpretation of cat. The situation in German is less clear than in English. Some speakers, including myself, would regard contrasts such as (50) as inappropriate . (50)
?Wir haben drei Katzen und einen Kater. (*We have three cats and one tomcat)
But apparently, there are others who readily accept Katze in a sex-restricted specific interpretation ( cf. Steinitz 1 974:258). I suspect that in at least some cases this may be attributable to Low German interference ( cf. (45)). It is hard to see how the lack of a specific interpretation in the case o f Katze can b e accounted for. On the other hand , the behaviour o f cat and tomcat could be treated as only a particular manifestation of a more general tendency. As is also pointed out in Rohdenburg ( 1 985 :67-68),
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In German, the relationship holding between the two terms is largely predictable on the basis of their constituents. This means that the first term in each case could be analysed as a superordinate of the second term. Furthermore , both the specific term and the general term in its specific interpretation may have d ifferent extensions complying with or failing to comply with the duality principle.5 By contrast, the relationship between cow (or cattle) and calf or sheep and lamb is not paralleled by that holding between the relevant items in (4 7). Beef and mutton are not normally available as superordinates of veal and lamb, respectively. We have seen, then, that the d uality principle is incapable of determining precisely the specific interpretation of a large number of lexical items. In most examples considered so far the only correct prediction it makes is that there is a specific interpretation associated with the general term in question. But even this prediction t urns out to be incorrect in a number of cases. Among the counter-examples in English and German· we find the following pairs of items :
128
(5 1 )
lecturer - senior lecturer
(52)
executive officer - senior executive officer
(53)
professor - assistant professor
Unlike the pattern using senior as modifier, combinations involving as sistant tend to be treated as exocentric. Consequently, they are not available as hyponyms of the corresponding heads. This situation is il lustrated in (54). (54 )
registrar - assistant registrar
3.2.
The duality principle fares no better in the area of multiple sets. Ac cording to the principle the more specific uses of the general term are excluded in all those cases where there is more than one lexical item denoting a subset of the extension of the general term along the same parameter. This is supposed to account for the Jack of any specific uses of the relevant kind in the case of book and room and accordingly con trasts such as ( 8) and (9) are predicted by Kempson to be contradictory .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
superordinates and their hyponyms cannot i n general be contrasted if the hyponym is a compound (or phrasal lexeme in the sense of Lyons 1 977: 23 , 3 1 0) whose head is represented by the superordinate term. In this respect , compounds typically behave like syntactic modifier-(nominal) head constructions but unlike derivatives.6 This explains why 'beheaded' compounds tend to be easier to contrast with their superordinates than their corresponding full forms. While, for instance , the phrase oranges and mandarins is fully acceptable, the supposedly equivalent oranges and mandarin oranges is generally felt to be deviant . Compare also ( 1 5 a) and (33a) with pianos and grand pianos and czutains and net curtains, which are odd. The phrase plums and Victorias, which I have seen in print, is considered to be marginally possible by some speakers. On the other hand, plums and Victoria plums is downright unacceptable. Incidentally, cats and toms is not acceptable either. Most exceptions to this tendency I know of concern the dimension of professional seniority. Here, the d isinclination to contrast a lexical item with a compound using it as head is outweighed by the overriding interest in and awareness of differentials. Thus, we find that the two terms in (5 1 ), (52) and (53) may be treated as superordinates and hyponyms as well as incompatibles.
1 29 A closer analysis, however, reveals that this assumption is simply incorrect . It just is not t rue t hat cases analogous to (8) and (9) do not exist . A num ber of counter-examples are presented in Rohdenburg ( 1985 :66), and many such examples may be constructed from those listed in { 1 1 ) - (33) and (36). For the purposes of this article it may be sufficient to take a brief look at three multiple sets. child - bab y , infant , toddler
(56)
pig - piglet , weaner 7
(57)
plum - damson, (green)gage
In each case , the first and general term rna y be contrasted with all of the other terms listed . While in (5 5) this results in several different specific uses of child (cf. (24) - (26)), the specific interpretations of pig and plum remain the same throughout. There are other distinctions that could be drawn in this area. For instance, unlike (5 5) - (5 7) there are some cases where the general term can only be contrasted with one particular member of a multiple set . For some further details see Rohdenburg ( 1 985 : 66-67). 4.
The examples discussed so far are all concerned with general and specific terms standing in a hyponymous relationship. Hyponymy bears a good deal of resemblance to the part-whole relation. Both are hierarchical relations. In the case of part-whole relations the item denoting the whole is of course hierarchically superordinate to the one denoting the part . Hyponymy is a transitive relation and the part-whole relation is at least in part transitive (Lyons 1977 :3 1 1 ff; Cruse 1 9 79). In addition, there is a certain amount of overlap between the two sense-relations (Lyons 1 977 : 3 1 6). A further parallel - and one that is only hinted at in Pilch ( 1 976 : 95) and Horn { 1 983 : 1 1 4 , 1 26) - can b e se e n in the phenomenon under investigation . The general term, i.e. the term denoting the whole, may have a restricted interpretation which is in contrast with one or more than one subordinate term denoting a part of the whole. Some relevant examples are listed in (58) - (65). (58)
flower - leaf, stem, etc.
(59)
body - arm, leg, head, tail, wing, etc.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(55)
1 30 (60)
shirt - collar, cuffs , etc.
(6 1 )
day - night , (dusk, dawn)
(62)
week - week-end
(63)
Europe - England, Ireland, Iceland, etc .
(64)
London - Streatham, Tooting etc.
(65)
New York - Brooklyn , Queens, etc. (Horn 1 98 3 : 1 26)
5.
So far , we have restricted our attention to the synchronic point o f view. In addition, it seems perfectly legitimate to interpret the duality principle as a diachronic claim (cf. § 2 .2 .). This is the general approach adopted by Horn ( 1 983 : 1 10- 1 1 7). He criticizes the principle mainly on the grounds that cases in which a given term functions as its own hyponym may originate in many other ways than those envisaged by Kempson. The phenomena covered by Horn may be reduced to two general classes, which will be discussed in turn. First of all, there are those cases where the semantic restriction of the "autohyponym" cannot be due to a contrast provided by other specific
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Obviously cases such as these fall outside the scope of the duality principle in its present version . Nevertheless they may pose the same problem for the constraint on sentential ambiguity as do items like dog. Such terms as body may also occur in sentences whose two relevant in terpretations entail each other. In order to rescue the ambiguity constraint Kempson might wish to treat the semantic duality of body etc. as a case of polysemy. Within Kempson's framework this means that the two inter pretations should be predictable by general rule. One might consider extending the duality principle to cover part-whole-relations. As far as I can see , this proposal is doomed to failure , and largely for reasons similar to those that were put forward concerning items such as dog, cat, plum, and child. What is more , the strength of the two relevant interpretations of the general term may also vary considerably. For instance, in (58) both interpretations are very strong indeed. In (63) the specific interpretation of the item denoting the whole is, for a large number of Britisch speakers, much more prominent than the corresponding general interpretation. In (60) the situation is reversed.
131 terms. I n some pairs like mankind - womankind the specific term happens to be a later development . In other examples the restricted use of the superordinate simply duplicates the range of a previously existing specific term rather than complementing it . Relevant examples include those in (66). (66)
colour - hue, smell - stink (Horn 1 983 : 1 1 5 - 1 1 6)
(67)
a. I'm a bit short. b. I'am a bit short of money.
Horn (I 983 : 1 1 6) suggests that the "restricted use of drink . . . evolved from the general use . . . via Gricean conversational implicatures". A large number of autohyponyms have presumably originated in this way. Some additional examples may include the nouns in (68), whose specific uses are illustrated in (69) - (78). (68)
result , manners, taste, voice , speed, price, style, head, chest , back, heart
(69)
They knew had to get a result (sc. in a soccer match). (BFBS)
(70)
He has no manners/taste.
(7 1 )
. . . a n aged prima donna without a voice (W. Somerset Maugham)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
So although a specific term is available in these pairs which could have restricted the superordinate it has failed to do so. In yet other instances of autohyponymy t here do not seem to be any relevant specific terms at all. Horn mentions the verb drink, whose restricted use is glossed as 'partake of alcoholic beverages'. Here , the semantic specialization is associated with "indefinite complement deletion" (Allerton 1 975 :2 1 7 -2 1 8). This phenomenon, which is not uncommon in English and other European languages, has been extensively discussed in recent work on syntax (cf. also Matthews 1 9 8 1 : 1 28- 1 29 ; Allerton 1 982 : 7 1 ,1 35). With many examples of this kind the general interpretation of the verb is also available, if only marginally so , in the case of indefinite complement deletion. Thus t his type of semantic specialization poses a real threat to the constraint on sentential ambiguity. The same kind o f semantic specialization is character istic of a number of predicative adjectives occurring with an appropriate adverbial modifier (Rohdenburg 1 986). For example, on the relevant interpretation of short, (67a) is normally equated with (67b ).
1 32 I cycled there ar speed. (Boyd)
(73)
. . . t hey were offered sex at a price . . . (Evening Standard)
(74)
The celebration was carried out in style . . . (Sunday Times)
(75)
I've got a head. I think I'll go and lie down. (Christie)
(76)
Grandma said, "There's a nip in the air and Maxwell has got a chest".(Townsend)
(77)
Your first patient's Ted Winchel, . . . He 's got a back. (Duncan)
(78)
A : It's his heart , Jack . B : H e hasn't got a heart. A: But he . . . crawled upstairs. (Duncan)
The second class of examples which is incompatible with the diachronic claim associated with the duality principle involves autohyponyms whose general use is derived from a previously existing specific use. The phenom enon is exemplified in Horn ( 1 983 : 1 1 4) by trade name labels such as Xerox or Scotch tape. A large number of autohyponyms involving geo graphic labels have also arisen as a result of semantic generalization. Typically, these have developed strong general interpretations . Conversely, most of the superordinates in (36) whose general interpretations are relatively speaking weaker than their specific interpretations appear to have undergone various processes of semantic restriction . There is, then , no direct correlation between the present-day strength of the general and specific interpretations of a given autohyponym and the processes of semantic generalization or specialization t hrough which it originated.
6.
I n conclusion, my critique of Kempson's proposals concerning the se mantic duality of items such as dog has uncovered a number of serious shortcomings. Kempson's analysis does not allow for the fact that both the specific and the general interpretation of the terms in question may vary in strength, and that independently of each other ( § 2 . 1 :• § 2 .2 .) . The duality principle relating the relevant interpretations was shown to be in adequate in several respects. First , the principle is incapable of establishing precisely the extension of the specific interpretation of the general term in sets containing just two members ( § 3 . 1 ., § 5 .). Nor does it correctly
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(72)
1 33
Westfalische Wilhelms- Universitiit Englisches Seminar Johanisstr. 12·20 D-4400 Munster, F R Germany NOTES I . Here and elsewhere, any attested examples have been marked by the addition of the author's name or a similar indication.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
·
predict the absence of such specific uses in certain cases ( § 3 . I .). Second, and more important, the principle is incompatible with the fact that the kind of duality u nder study is also found in a great variety of multiple sets ( § 3.2 .). Moreover, there are analogous phenomena associated with various types of part-whole relations which are not captured by the principle either ( §4.). Finally, as is shown by Horn ( 1 983), autohyponymy may originate in other ways than those predicted by Kempson's thesis ( § 5 .). There is no overall principle in sight capable of covering the range of data presented in this paper. This suggests that Kempson's restricted notion of polysemy should not be invoked to account for (all of) the relevant phenomena . Furthermore , cases such as I bought a dog continue to stand as counter-examples to the constraint on sentential ambiguity. It does not follow that the duality principle should be abandoned alto gether. However, its relevance seems to be more a matter of pragmatics than (synchronic) lexical semantics . When confronted with contexts contrasting a general term and its (only) hyponym the co-operative hearer attempting to resolve the apparent contradiction or t autology will search for a more specific interpretation of the general term . If none is readily available the hearer may feel obliged to construe an appropriate interpretation. Presumably, the most obvious way to go about it corre sponds to the b asic insight embodied in the duality principle . This involves restricting the interpretation of the general term as much as is required for the purpose of avoiding a contradiction or tautology but no further . Given the appropriate modifications the principle could easily b e extended to cover those part-whole relations where there is only one lexical item denoting a part of the whole . Although some such principle certainly plays an important part in comprehension processes its scope is limited by other factors, which are only poorly understood . For instance, in cases where there is felt to be a prototypical subset or part of what is denoted by the general term its interpretation may be restricted even further to apply just to this core area. With multiple sets of both types of lexical relation the tendency to look for such core interpretations seems to be much more pronounced .
1 34 2 . The acceptability j udgements of the invented examples reflect the views of one educated speaker of British English. I am, however, aware of the fact that the area under study involves a certain amount of inter-speaker variation. Incidentally, some of the attested examples in ( I I ) ( 3 3 ) are considered to be of marginal acceptability by my informant. 3 . The mistake of interpreting bitch and dog in its sex-restricted interpretation simply as 'female dog' or 'male dog' has of course a respectable tradition. It seems to be characteristic of virtually all monolingual English dictionaries. 4. It should not be assumed that contrasts such as these can never be used to refer to any babies or immature members of the species involved. An example of the relevant kind is found in (i). -
(i)
But such cases are perhaps explicable by reference to a general principle : any evolving entity may under appropriate circumstances be referred to by a noun denoting its earlier or later manifestations. 5. The term Schaff/eisch appears to be relatively recent and may not be generally established . The more usual contrast involves the incompatibles Hamme/fleisch ('meat from castrated rams') and Lammfleisch. 6 . Interestingly enough, superlatives in German may show the same behaviour as bona fide derivations. Some of t he relevant contrasts I have come across include the fo llowing : (i)
K lein- und Kleinstwagen ( 'small and smallest cars'), Klein- und Kleinst familien ( 'small and smallest families'), Klein- und Kleinstarbeit ('detailed and most detailed work '), Tief- und Tiefstfltige ( 'low and lowest flights'), Niedrig- und Niedrigstpreise ( 'low and lowest prices'), Hoch- und Hochst leistungen ( 'great and greatest achievements')
Though typically associated with compounds the items contrasted may also occur as attributive modifiers of noun phrases . (ii)
hohe und hochste Anforderungen ( 'high ar.d highest demands'), die hohen und hochsten Gerichte ( 'the high and highest courts')
In cases like (i) and ( ii) the interpretation of the relevant roots is predicted by the duality principle. 7 . Note that the three-member set including weaner belongs to a relatively technical register. The same is true of the corresponding set in German : (i)
Schwein ('pig') - Ferkel ('piglet'), Uiufer ('weaner')
R EFERENCES Allerton, D.J ., 1 9 75 : Deletion and proform reduction. Journal of Linguistics 1 1 : 2 1 3-2 3 7 . Allerton, D.J. , 1 98 2 : Valency and the English Verb. Academic Press, London. Cruse, D.A., 1 97 9 : On the transitivity of the part-whole-relation. Journal of Lin guistics 1 5 : 29-38.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I lifted the pups out one by one and examined them . . . "Three dogs and t hree bitches", I said. "Nice even litter". (Herriot)
1 35
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Grice, H.P. , 1 97 5 : Logic and conversation. In: Cole, P. & Morgan, J . L. (eds.), Sylllax and Semantics 3. Academic Press, New York. 4 1 -58. Horn, L.R., 1 9 8 3 : Ambiguity, negation and the London school of parsimony. In Jones, C. & Sells, P. (eds.), Proceedings of NELS 14. CLSA, UMass, Amherst. 1 08- 1 3 1 . Kempson, R., 1 97 9 : Presupposition, opacity and ambiguity. I n : Dineen, D.A. & Oh, C.-K. (eds.), Syn tax and Semantics 1 1 : Presupposition . Academic Press, New York. 283-29 7 . Kempson, R., 1 980 : Ambiguity and word meaning. In Greenbaum , S . , Leech, G. & Svartvik, J . (eds.), Studies in English Linguistics: For Randolph Quirk. Longman, London. 7 - 1 6 . Lyons, J ., 1 9 7 7 : Semantics I. Cambridge University Press, Cambridge. Matthews, P.H., 1 98 1 : Syn tax. Cambridge University Press, Cambridge. Pilch, H., 1 976: Empirical Linguistics. Francke, Munchen. Quirk, R., Greenbaum , S., Leech, G. & Svartvik, J . , 1 97 2 : A Grammar of Contem porary English. Longman, London. Rohdenburg, G., 1 985 : Unmarked and marked terms in English. In: Hoppenbrouwers, G., Seuren, P. & Weijters, A. (eds.), Meaning and the Lexicon . Foris, Dordrecht . 63-7 1 . Rohdenburg, G., 1 986 : Eingeschriinkt auslassbare Ergiinzungen bei englischen Ad jektiven. Vernachliissigte Erscheinungen in der Valenzbeschreibung. In : B urk hardt, A. & Korner, K.-H. (eds.), Pragmantax: Akten des 20. Linguistischen Kol loquiums. Niemeyer, Ttibingcn. Steinitz, R., 1 974 : Nominate Pro-Formen. In : Kallmeyer, W., Klein , W., Meyer Hermann, R., Netzer, K. & Siebert, H . J . (eds.), Lektiirekolleg zur Textlinguistik ll · Reader. Fischer/Atheniium, Frankfurt. 246-265. Zwicky, A.M. & Sadock, J . , 1 97 5 : Ambiguity tests and how to fail them. I n : Kim ball, J . (ed.), Syn tax and Semantics 4 . Academic Press, New York. 1 -3 6 .
Journal ofSemanrics 4: 13 7-163 THE PRESENT PERFECT IN DUTCH AND ENGLISH*
JAN D E VUYST
ABSTRACT
0. INTRODUCTION
There are some well-known differences between the use of the present perfect in English and the use of that form in other languages such as, for example, Dutch. Thus Dutch may use a present perfect where English uses a simple past : (I)
heeft me bedrogen ik sprak De man met wie The man to whom I talked has me cheated
(2)
The man to whom I talked cheated me :f The man to whom I talked has cheated me
(3)
Harry is gisteren gearriveerd 1 Harry has yesterday arrived
* This paper is based on material from a chapter of De Vuyst ( 1 9 8 3b). Financial support by ZWO ("Netherlands Organisation for the Advancement of Pure Research"), grants 30-1 28 and S30-200 , is gratefully acknowledged. Comments b y Ewan Klein , Sjaak de Mey , Marc Moens and Mark Steedman on an earlier draft of the paper were very helpful.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Certain differences between Dutch and English use of the present perfect are considered in relation to a more general difference between the two languages that involves temporal inclusion . It is shown that Dutch and English exploit different means of expressing a temporal inclusion relation between two events , notably where Vendler's activities and accomplish ments are concerned. Precisely in those cases Dutch and English use the present perfect in different ways. But when there are no differences in the expression of temporal inclusion, notably where Vendler's states are concerned , the two languages use the present perfect in the same way.
1 38 (4)
Harry arrived yesterday cf. : *Harry has arrived yesterday
(5)
J an is naar huis gelopen toen de born explodeerde Jan has home walked when the bomb exploded
(6)
Jan liep naar huis toen de born explodeerde Jan walked home when the bomb exploded
Sentence (5) expresses the same (7)
as
the English sentence
Jan walked home when the bomb exploded
namely , that first the bomb exploded , and then (probably as a result) Jan walked home , as represented in (8) : Jan walk- home
(8)
the bomb explodewhere t is the time-axis and i the time of evaluation . 3 But the Dutch sentence (6), with a past-tense main clause , is ambiguous between a reading of the type represented in (8), and a reading which in English would be expressed as Jan was walking home when the bomb exploded, in which the bomb explodes during Jan's walk : (9)
Jan loop- naar huis Jan walk- home I
de born explodeerthe bomb explode-
::>.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
On the b asis of these and similar sentences it has been argued more than once that in terms of earlier/later relations on the time-axis, the Dutch present perfect is temporally equivalent to a past tense . In the words of McCoard ( 1978 : 1 56) : "In spoken Dutch, the only past tense normally used is the perfect..., so there really is no regular opposition with the preterite ... .' 2 But this assumption is too simple. There is an important opposition between the Dutch simple past and the Dutch present perfect and I will argue that it is a regular one. Consider:
I 39
( I 0)
gewerkt Hij heeft nu a! een hele poos in de tuin He has now quite�-time in the garden worked
{I I)
He has worked in the garden for quite some time now
{ I 2)
Hij is n u al een hele poos ziek geweest He has now quite�-time ill been
{ 1 3)
He has been ill for quite some time now
Neither language , however, accepts the following kind of sentence : { I 4)
*Jan heeft nu al een hele poos een b rief Jan has now quite�-time a letter
geschreven written
( I S)
*Jan has written a letter for quite some time now
Without the addition of an appropriate time-adverbial, and without any further context, there is no obvious temporal distinction between the simple past and the present perfect in either Dutch or English. Thus in terms of some earlier/later relation the following sentences cannot be distinguished : { 1 6)
Harry was ziek Harry was ill Harry was ill
{ 1 6a) Harry is ziek geweest Harry has ill been Harry has been ill
will argue that from a purely temporal point of view the present perfect requires essentially the same analysis for Dutch and English and
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Since (5), the Dutch present-perfect sentence , allows other orderings on the time -axis than (6), the Dutch simple past sentence , one has to take into account at least some opposition between the Dutch simple past and the Dutch present perfect. So before one can tackle the question why e .g . {5), the Dutch present perfect, expresses the same a s {7), the English simple past, one must consider j ust what the difference is between the Dutch sentences {5) and {6). But there are also important similarities between the Dutch and English use of the present perfect. In each of the two languages the present perfect can express a temporal link between past and present {I will have very little to say on pragmatic links which have been claimed in terms of "relevance", "existence", etc.). Both languages usually add time-adverbials in such a link reading :
1 40 that the differences between the languages as regards the use of present perfect versus simple past result , either directly or indirectly , from dif ferent ways in which Dutch and English make use of what Vendler ( 1 967) classified as activities and accomplishments.
1. ACCESSIBI LITY
( 1 7)
*Sue swims in the river (at the moment)
( 1 8)
*Harry writes a Jetter (at the moment)
In Dutch, on the other hand , such sentences are fully acceptable : ( 19)
Sue zwemt in de rivier (op het moment) Sue swims in the river at the moment
(20)
Harry schrijft een brief Harry writes a letter
(op het moment) at the moment
For the truth of a present-tensed sentence it is required that the relevant event-time includes (at least) the time of evaluation i : (2 1)
event-time
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Consider again the representations (8) and (9). The Dutch simple past allows the "bomb explode"-time to be included in the "Jan walk home" tim e , but the Dutch present perfect and the English simple past do not allow such temporal inclusion . In fact, this is an instance of a far more general temporal/aspectual opposition between Dutch and English, which in de Vuyst ( 1 983b) was dubbed "accessibility". If for the time being we take accessibility as an intuitive label for what seems to be the difference between (8) and (9), then the "Jan walk home "-time in (9) can be ac cessed by the "the bomb explode"-time , but in (8) no such access is allowed. Consider the following further examples of this accessibility opposition between Dutch and English. In English, an activity or accomplishment does not , normally , occur in the simple present tense : 4
141 Dutch activities and accomplishments appear t o allow the required in clusion in the simple present , but the English activities and accomplish ments do not . In other words, in Dutch the activities and accomplish ments appear accessible to the time of evaluation whereas in English they appear inaccessible. Note that under this assumption the states in both languages are accessible : (22)
Richard is very happy at the moment
(23)
Richard is erg gelukkig op het moment Richard is very happy at the moment
(24)
John ate at 6 o'clock
(25)
John at om 6 uur John ate at 6 o 'clock
In both Dutch and English, the combination of an activity and a point time adverbial can mean the activity started at the indicated time : (26)
John eet John eat
6 uur 6 o'clock The Dutch (25), however, allows an additional reading in which the time as specified by the time adverbial occurs sometime during the activity , a reading which in English would require a progressive like John was eating at 6 o 'clock :
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
It might well be argued of course that it is the present tense , rather than the activities and accomplishments, which is different for the two languages . There are other cases, however, other than the simple present tense , quite independent of the position of the time of evaluation actually, which nevertheless show this accessibility opposition between Dutch and English activities and accomplishments. Consider the sentence:
1 42 (27 )
John eet John eat-
6 uur 6 o 'clock
I
(28)
Richard was very happy at 6 o'clock
(29)
Richard was erg gelukkig om 6 uur Richard was very happy at 6 o 'clock
(30)
Richard zij n- gelukkig Richard be- happy t --�-------.--�---r--�� 6 uur 6 o'clock
A third area for which the accessibility opposition between Dutch and English can be demonstrated concerns temporal connection . The following when-sentences, for instance, (3 1)
J an liep naar huis toen de born explodeerde Jan walked home when the bomb exploded
(32)
Jan walked home when the bomb exploded
were already shown to be different in the introduction : The Dutch (3 1 ) allows, as one o f its readings, that the "Jan walk home"-time is accessed by the "the bomb explode" -time, but the English (32) has no such reading
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The same holds again for the accomplishments : The Dutch lk schreef een brief om 6 uur (I wrote a letter at 6 o 'clock) can mean both that at 6 o'clock I commenced the writing or that 6 o'clock was j ust one of the times of letter-writing, parallel to what the English I was writing a letter at 6 o 'clock expressed .5 So in terms of accessibility again, the Dutch activities and accomplishments allow , in the above sentences, access by the adverbial time, b ut the English activities and accomplishments do not. Note that the states are accessible in both languages and that the following sentences allow a reading in which "6 o 'clock" was j ust one of the times ofhappiness :
1 43 available .6 The main clauses in (3 1 ) and (32) are accomplishments, b ut the same difference Dutch/English holds for activities, as in : (33)
John worked in the garden when I called him
(34)
John werkte in de tuin toen ik hem belde John worked in the garden when I him called
The reading for the English (33) is that John started his work in the garden j ust after (and probably because of) my call :
t --.-�L-------�--��� I call The Dutch (34), however, allows that the "John work in the garden"-time is accessed by the "I call"-time , a reading for which English would use a sentence like John was working in the garden when I called him : John werk John work-
(36)
t ------�-.----L---��� ik bel
/ call-
As in the previous cases , states allow access in both languages: (37)
John was at home when the bomb exploded
(38)
John was thuis toen de born explodeerde John was at home when the bomb exploded John zijn- thuis John be- at home
(39)
t
-L
__ __
-,
--
--
------------
de born explodeer the bomb explode-
L-
--
--.-� �
--
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
John work-
(35)
1 44 There are still further examples o f this specific opposition between Dutch and English, b ut the above examples may suffice as a demonstration of "accessibility": Dutch activities and accomplishments are , still in tuitively, accessible, whereas English activities and accomplishments are inaccessible. The sentences (5) , (6) and (7), repeated here , J an is n aar huis gelopen toen de born explodeerde Jan has home walked when the bomb exploded
(4 1)
Jan liep naar huis toen de born explodeerde Jan walked home when the bomb exploded
(42)
Jan walked home when the bomb exploded
show that not only the English simple past (42) and the Dutch simple past (4 1 ) , b ut also the Dutch simple past (4 1 ) and the Dutch simple perfect (40) differ in terms of accessibility. In accordance with the re presentations (8) and (9), the Dutch simple past of an accomplishment allows access, b u t the Dutch present perfect of an accomplishment does not. The same holds for a Dutch activity , which is accessib le in the simple past, b ut inaccessible in the present perfect. A sentence like (43)
Jan heeft viool gespeeld toen de born ontplofte Jan has violin played when the bomb exploded
forces a "just after" reading of toenfwhen : first the bomb explodes, and only then Jan starts playing the violin : 7 (44)
Jan heb - viool gespeeld Jan have- violin played
de born explodeer the bomb explodeIn contrast , as demonstrated , the simple past of an activity is accessible : (45)
J an speelde viool toen de born explodeerde Jan played violin when the bomb exploded
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(40)
1 45 (46)
Jan speel- viool Jan play- violin
de born explodeer the bomb explode-
2. THE STATUS OF ACCESSIBILITY
Since accessibility involves temporal inclusion, it presupposes the existence of time-intervals. But something like "accessibility" cannot , by definition , be a property of those intervals. An interval can be defmed as a (convex) set of moments, and qualifications such as "accessibility" simply do not apply there : one cannot, in other words, distinguish sections of the time axis in terms of accessibility.9 Neither is there any need to do so. In accordance with other recent proposals I will assume that the temporal/ aspectual interpretation of natural language sentences is not directly tied to the time-axis, is not to be achieved in terms of moments and intervals , in other words!° Consider the following very b rief discussion of the main ingredients that are needed for an interpretation. The fact of anything existing, moving, occurring, etc., in the real world I will call an event. Language can represent those events in many different ways and I will call those linguistic representations situations. I will have nothing more to say on the real world events, except that there must be a "compatibility" relation R that holds for some event e and
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
What this boils down to is that the Dutch present perfect of an activity/ accomplishment expresses the same as the English simple past of an activity/accomplishment in terms of accessibility: both express an in accessible activity/accomplishment in the past . Since , however, activities/ accomplishments are different in the two languages as regards accessibility , Dutch and English must use different means to express the same. Where as English uses a simple past. Dutch uses a present perfect to express an "earlier than i" relation in a manner to be considered below. Whatever else the semantics of the perfect in Dutch might involve , it apparently functions as an inaccessibility-operator. 8 But this still leaves us without an analysis of the perfect and the notion of "accessibility". That analysis must account for "possible", or perhaps "allowed", inclusion on the time-axis, for the link-reading of the perfect in the relevant cases , and for the fact that the present perfect can be used to relate defmitely past events.
1 46 some situation s that is to represent e ; sRe , that is. In what follows, only situations will be discussed and depicted. The relevant sections of the time-axis are the results, or the value, of a map from the situations to the time-axis, and will be depicted as follows : (47) t
A -..-....-.-
......
J
(48)
�
Tense , a simple present in this case , triggers a map to t : (49)
J Truth conditions for the simple present will require that for the truth
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
where the section J of the time-axis t will be referred to as the situation time. Thus if the situation in (4 7) is some activity, say a "Sue swim" situation , then J will be the corresponding "Sue swim"-time , or , more generally, an activity-time. I will assume that the utterance/writing/processing of a sentence , in short its mere existence, is also a situation, therefore not a predetermined section of the time-axis . This situation E, for Evaluation , will result in the time of evaluation i, when mapped to t . S o suppose there is som e event o f John's being ill , linguistically re presented as a "John be ill"-situation by a sentence like John is ill. The utterance of the sentence itself is represented in the following picture by E, and the "John·be ill " -situation is pictured again as a cloud/oval:
1 47 of the sentence John is ill relative to a time i , there is some "appropriate" (see below) situation s, and an event e such that sRe , and that J, the value of the map from s to t, includes the time of evaluation i. 1 1 One of the many ways in which language may represent real-world events is to represent them as states, activities , accomplishments and achievements. For the interpretation of a tenseless sentence such as John walk-, we need a set of "John walk" -situations, representing a possibly endless number of "John walk" -events. So if [ ] indicates the interpretation function, then
Any activity-situation involves movement and will therefore have subsituations : movement always allows further subdivision in terms o f e .g. beginning, middle and end . I n other words, any activity-situation m ust be non-minimal, since minimal situations preclude such further subdivision . Moreover, any non-minimal subsituations of an activity situation can be qualified in the same way as the entire activity-situation , that is , any non-minimal subsituation of e .g. a "Sue swim in the river" situation is itself a "Sue swim in the river"-situation . So any "Sue swim in the river"-situation in the interpretation-set for the sentence Sue swims in the river may happen to be the subsituation of some other situation in the same set . Therefore we can give the following characterization of the interpretation-set of a tenseless activity-sentence : (5 1 )
[tenseless activity-sentence] = {s 1 . . . sn l where I) each s is non-minimal 2) the set is closed under non-minimal subsituations
The interpretation-set for a tenseless accomplishment-sentence will not contain any minimal situations either, b ut in addition , no subsituation of an accomplishment-situation will be in the same interpretation-set : no
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
says that the interpretation of the tenseless sentence John walk- is the set of situations s 1 to sn , where all situations in that set , to be referred to as the interpretation-set, are "John walk"-situations. Two kinds of situations are to be distinguished, namely minimal and non-minimal, 12 and , at least according to a classification such as Vendler's , four different interpretation-sets. These wil! be distinguished both in terms of minimal versus non-minimal and in terms of subsituation relations, quite parallel in fact to the way in which Interval-Semantics, following Taylor's ( 1977) predicates , uses moments and (sub)intervals for its defmitions of aspectual classes . 13
1 48 subsituation of e .g. a "Vincent paint a van Gogh"-situation is also itself a "Vincent paint a van Gogh"-situation: (52)
[tense less accomplishment-sentence]
=
{s 1 . . . sn }
where 1 ) each s is non-minimal 2) no two s's stand in a subsituation-relation
(53)
[tenseless state-sentence] = {s 1 . . . sn } where the set is closed under arbitrary subsituations
The achievement-set , finally , will only contain minimal situations, about which I will have nothing more to say in this paper. The phrase "appropriate situation s", as used above in the informal truth-conditions for the present tense can now be given substance : an appropriate situation is a situation from the interpretation-set of the sentence under consideration . Note that there must be restrictions on the map from the situations to t . So for instance no "cross-overs" are to occur : (54)
*
s2 -time
s 1 -time
A more specific condition, in fact a language-dependent one in the sense that it may differ for each language under consideration , concerns access ibility . In accordance with the example sentences so far, Dutch activity and accomplishment-situations may be mapped to t without any restric tions on some resulting temporal inclusion on t ; but the English activity and accomplishment-situations are only to be mapped to t if the corres ponding activity- and accomplishment-times do not include any other situation-time(s).14 The following configuration therefore represents an admissible map for Dutch , resulting in an activity- or accomplishment-time which temporal ly includes the S-time , where "S-time" is just any situation-time:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The interpretation-set of a tenseless state-sentence will typically contain both minimal and non-minimal situations. Every subsituation of e .g. a "shirt be blue"-situation will also be a "shirt be blue"-situation, hence for the state-set we need not restrict the kind of situations it contains:
1 49 (55)
S-time act./acc.-time
(56)
-__..
S -time The "accessibility-condition" is merely demonstrated here in order to show that there are other differences between Dutch and English than j ust the use of the present perfect , differences in fact which have consequences for the use of tense forms, where the activities and accomplishments are concerned. So far, however, nothing has been said on the interpretation of the (present) perfect , which appeared to involve inaccessibility in contrast to the simple past ; cf. the Dutch sentences (5) and (6).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The S in (55) can stand for an E-situation (for the present-tense cases), or for a 6 o 'clock-situation (for the point-time adverbial cases), or for a situation from the set that interprets some temporal subclause (for the when-sentences) or for any other situation. Regardless of the particular S involved, and that includes the evaluation-situation E , the Dutch activities and accomplishments can be accessed by it, as represented in (55). For English activities and accomplishments , however, there must be a condition which prevents a map such as the one in (55): regardless again of the particular S involved, English activities and accomplishments cannot be accessed by it :
1 50 3 . THE INTERPRETATION OF THE PERFECT
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As shown in the introduction, the present perfect can, in both Dutch and English, express a temporal link between past en present (examples 1 0 - 1 3) for some though not all cases (cf. 1 4 and 15), and it can be used for a situation that is clearly earlier than the time of evaluation ( cf. examples 1 6 and 1 6a). An important issue in theories on the (English) present perfect has always been to account for that link on the one hand and the similarity with the past tense on the other hand . How is the perfect to be analysed and how does it differ from the past? In Jespersen ( 1 924) , a proponent of what McCoard ( 1 978) has called the "Current Relevance" theory (CR), the present perfect is claimed to "represent the present state as the outcome of past events, and may therefore be called a retrospective variety of the present tense" (Jespersen 1924: 269) . Thus for a sentence l ike He has become mad the present state is He is mad whereas no such present state can be inferred from He became mad. But a general problem with the CR theory is that "relevance" is very hard to define . McCoard gives several terms that have been associated with "relevance", such as "recency", "present existence", "present validity", e tc . , b ut confronts all of them with counter-examples (McCoard 1 978 : 44-64), and it does not seem that CR-proposals can be made sufficiently precise or concrete to provide a b asis for a semantic account of sentences ( 10-1 6a). Other attempts to pin down the meaning of the present perfect are proposed in the "Indefinite Past" theory (ID), which, in the words of McCoard ( 19 7 8 : 1 8) , can be characterized as claiming that the present perfect "expresses a past event which is unidentified as to time". This claim is b ased on the distribution of time-adverbials such as last June, at 6 o 'clock, etc . , adverbials which refer to a specific or definite span of time. These adverbials can be combined with the simple past (John arrived last Junefat 6 o 'clock) b ut they cannot be combined with the present perfect (*John has arrived last Junefat 6 o 'clock). McCoard argues, against this theory, that the present perfect can be combined with e .g. today, a definite time adverbial (John has arrived today) , and that , moreover, it is not the case that the simple past cha racteristically refers to definite time-spans (Man descended from the apes) , hence that "defmiteness", whatever its further definition might be , does not distinguish simple past and present perfect . Even apart from McCoard's arguments against ID-theories, such theories could never be used in an account of both Dutch and English present perfect , since despite the similarities as demonstrated by ( l 0-1 6a), Dutch present per-
151
(57)
[PRES(cf>)] i
=
1 iff 3 s, 3 e , sRe,
S€
[cf>] & i c J
where cf> is a tenseless sentence , i the value of the map from E to the time-axis t and J the value ·of the map from s to t . So for the truth of a present tensed sentence there must be some situation from the inter pretation-set of the corresponding tenseless sentence , some e with which s is compatible and the situation-time must temporally include the evaluation time . Under the assumption that the present perfect is a present tense , (5 7) requires for the truth of PRES (John have-worked) relative to i that a "John have worked"-time temporally includes i. That in turn raises the question of what the interpretation-set for John have worked looks like , in other words how the situations that interpret HAVE (John work-) will differ from the situations that interpret John work-. In order to meet the requirement of inclusion one might argue , parallel to Dowty's definition of the XN-predicate but adapted to the present framework, that a "John
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
feet can easily combine with defmite time-adverbials (Jan heeft om 6 uur gebeldfJan has at 6 o 'clock phoned = Jan phoned at 6 o 'clock). Yet another theory on the present perfect , the one preferred by Me Coarct , is the "Extended Now" theory (XN), which has attracted a lot of attention in recent semantic proposals (e.g. Dowty 1 979 , Saurer 1 98 1 , Richards 1 982). The XN-theory claims, in McCoard's words again ( I 978 : 1 8) , that the present perfect expresses a "past event within a time span that is continuous with the present , not differentiated in "then" versus "now"." So in this view, the analysis of the perfect requires two time units, one which is essentially present , and another which is located within that present "time span". In Dowty's ( 1979 : 3 4 1 ) words, the XN-theory "is the view that the perfect serves to locate an event within a period of time that began in the past and extends up to the present moment . . . " The b asic claim of the XN-theory that there are two time units involved in the analysis of the present perfect offers a significant starting-point for an analysis of both Dutch and English present perfect . In Dowty ( 1979 : 342 and 3 53) a predicate XN is formulated for the perfect on a par with predicates for past and present tense . It is defined in such a way that , informally , a perfect sentence will be "true" with respect to an interval i if the corresponding non-perfect sentence is "true" with respect to an interval j and i is a fmal subinterval of j . That Dowty uses predicates rather than e.g. operators is of no importance here , but it does not seem appropriate to consider the perfect a tense j ust like PAST and PRES , giving it independent truth conditions. 1 5 I will assume that the present perfect is a specific instance of the present tense . Truth conditions for the present perfect might be formulated as follows:
1 52 have worked" -sit uation must be a "John work"-situation, the final subsi tuation of which is co-extensive with E :
� � John have worked
{58)
work
t
.__..
:;;...
� On t , the time of evaluation is a final subinterval for the "John work" time , quite in accordance with Dowty's proposal. But note that in this approach the "John work"-time will last till and including i: E is co-extensive with the final subsituation of the "John have worked"-situation, not with its final bound , which means that the only reading of the present perfect that can be accounted for in this way is a link reading in which the relevant event goes on while the sentence is being uttered (recall that E represents the existence of the sentence itself). This may be fine for sentences such as Some time has elapsed now or Harry has been ill for three days now (if uttered sometime during the third day) , b ut it will not account for e .g. Three minutes have elapsed now, if uttered exactly after the three minutes, or for John has been ill for three days now, if uttered just after the third day . In Heny { 1982) it is proposed , in a different framework , that the "John work"-situation is a non-final subsituation of the "John have worked" -situation . 16 Together with the requirement that the final subsituation of a "John have worked" situation is co-extensive with E, this leads to something like the follow ing: {59)
John have worked
But this approach also has a problem : it suggests that the "John work"-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
J
1 53
{60)
[HAVE{>)] = {s l . . . s 0 } where each s has an initial subsituation which is also a member of [ >] , and a final subsituation which is co-extensive with E .
Note that E i s explicitly mentioned in this condition since the present perfect is being dealt with here . 1 7 So the interpretation of John have- worked will be a set of situations , the initial subsituations of which are "John work"-situations and the final subsituations of which are co-extensive with the corresponding E 's {that is, each time the E of the sentence under consideration). In {6 1 ) one such "John have worked"-situation is depicted : {6 1)
John have worked
For a present-tensed sentence such as John has worked, the "John have worked" -situation is mapped to t , and the resulting situation-time will include i, as required by the conditions for the present in (57). Thus for the sentence John has worked the situation-time of the non-perfect situation "John work" is in fact earlier than i, b ut the "John have worked" time includes i. For the past reading of the perfect, however, we need one
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
time can always last till i, though not including it . That of course leaves e .g. Harry has been ill [or three days now, as uttered during the third day , u naccounted for. More seriously, however, neither the approach in {58) nor the approach in {59) can account fo r the perfect sentences which locate a situation as defmitely and unambiguously earlier than the time of evaluatio n : they cannot account for the past reading of the perfect as in sentences { I ) and {3) or { 1 6) and { 1 6a). I will argue that the "John work"-situation must be an initial sub situation of the "John have worked"-situation , rather than just a non-final one , and that in addition E must be co-extensive with the final sub situation of a "John have worked"-situation . Thus if > is a non-perfect tenseless sentence :
154 more condition, namely that there is some non-empty interval between i and the relevant situation-time : the situation-time must not adjoin to i but be earlier than i. At the representational level of the situations this boils down to the requirement that the perfect situation must have more subsituations than just the initial and final one : the perfect situation must be larger than j ust the union of final and initial subinterval . Hence : (62)
[HAV E(t/>)]
=
{s 1 . . . s n }
In this way , the third condition will guarantee that in e .g. (6 1) the initial "John work"-situation does not immediately adjoin the final "John have worked" -situation, hence that the "John work" -time is definitely earlier than i. 18 At this point we may have taken care of the past reading of the perfect , but it seems that (62) will not account for the link-reading of the perfect, as illustrated in sentences ( 10-13), precisely because of the third con dition that was needed for the past reading. Thus for a sentence like
(63)
John has wo rked in the garden for some time now
the analysis of the perfect as proposed in (62) requires that the "John work (in the garden)"-time is earlier than i, whereas the sentence ex presses that the "John work"-time lasts till (at least) i. But in fact, as I will argue below, all the link-readings as the ones in ( 1 0-1 3) follow from the definitions o f the various interpretation sets, given in (5 1 -53). The perfect in (63) demands that the "John work"-situation is earlier than E : (64)
t
�
__ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __
whereas the time-adverbial for some time 1ww requires that the "John work"-situation "reaches E" (which is short for "can last till (at least) the final subsituation of the perfect situation which is co-extensive with E") :
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
where 1 ) each s has an initial subsituation si , si € [1/>] 2) each s has a final subsituation sf, such that sf is co-extensive with E 3) each s properly contains the union of its si and s f
1 55 (65)
® t
�
__ __ __ __ __ __ __ __ __ __ __ __ __ __ __ __
(66)
Jan heeft nu al een hele poos in de tuin gewerkt Jan has now quite-a-time in the garden worked
can be accounted for. The definition of the interpretation-set of an activity is the same for Dutch and English, and so is the analysis of the perfect in (62). Therefore the account for the Dutch (66) and the English (63) can be exactly the same . The same , mutatis mutandis , holds for states in the link-reading of the present perfect in both languages : (67)
John has been ill for quite some time now
(68)
John is nu al een hele poos ziek geweest 20 John has now quite-a-time ill been
The adverb ial demands that the "John be ill " -situation reaches E, (69)
�
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Given the acceptability of (63), and under the assumption of course that (62) is an appropriate analysis of the perfect , these two commands do not seem to contradict each other. In fact, there is no such contradiction b ecause John work- (in the garden) is a tenseless activity-sentence . The defmition of the interpretation-set for activities (5 1 ) states that sub situations of an activity-situation can be characterized in the same way as the entire situation: all (non-minimal) subsituations of an activity-situation are in the same interpretation-set as that activity situation . Therefore , the two demands as depicted in (64) and (65) are not incompatible : the initial subsituation in (65) is a "John work"-situation , as required by the analysis of the perfect (62), although it is a 5ubsituation of a larger "John work" -situation. Hence for the interpretation of the link-reading of the present perfect as in (63) we need not change the analysis of the perfect in (62) . In exactly the same way the link-reading of the Dutch present perfect in a sentence like
156 while a t the same time (62), the perfect that is, demands that the initial subsituation of the "John have been ill"-situation is a "John be ill" situation. Given the interpretation-set for states, (53), these two com mands are not incompatible : the initial subsituation in (69) is a sub situation of a "John be ill"-situation, therefore also itself a "John be ill " situation. Hence both demands can be met at the same time . . But that is different for the accomplishments. Both Dutch and English reject sentences like : *Harry has written a letter for some time now
(7 1 )
*Harry heeft n u a! een hele poos een brief Harry has now quite-a-time a letter
geschreven written
The adverbials for some time nowjnu a/ een hele poos demand that the non-perfect "Harry write a letter"-situation reaches E : (72 )
® At the same time the analysis of the perfect, (62), demands that the in itial subsituation of the perfect situation is a "Harry write a letter" situation. But given the interpretation-set for the accomplishments, (52) , that is not possible : the initial subsituation as in (72) is the subsituation of an accomplishment-situation, hence by defmition not also itself an accomplislunent-situation. The first condition in (62) and the demand that E be reached contradict in the case of the accomplishments, hence the unacceptability of (70) and (7 1 ) So (62) starts out, as it were , other than proposals such as Dowty's , with an account o f the past reading o f the (present) perfect . The dis t ribution of acceptable link-readings of the present perfect then simply follow from (62) and the definitions of the various Vendler-classes, in terms of the interpretation-sets (5 1 -53). .
4 . CONCLUSIONS
We have now two means for distinguishing the simple past and the present perfect in Dutch and English , namely accessibility and the analysis (62). Consider again the sentences we started out with, here renumbered :
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(70)
1 57 (73)
Jan is naar huis gelopen toen de born explodeerde Jan has home walked when the bomb exploded
(74)
naar huis toen de born explodeerde Jan liep Jan walked home when the bomb exploded
(75)
Jan walked home when the bomb exploded
(76)
J an is naar huis gelopen toen de born explodeerde Jan has home walked when the bomb exploded
I
==t< I
w
Jan loop naar huis Jan walk home
/ born explodeer ""'\ '- bomb explode /
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
We can now characterize these sentences as follows : (73) , Dutch present perfect , is inaccessible and perfect , in the past reading of the perfect , as drawn in e .g. (6 1 ) ; (74) is accessible and past , (75) is inaccessible and past . Hence the similarity between Dutch present perfect and English simple past , claimed to be of a temporal nature, is in fact a similarity of inaccessibility : both (73) and (75) express an inaccessible situation which is earlier than i. This is corroborated by the fact that the similarity of present perfect/simple p ast is limited to the activities and accomplishments, which appeared different in Dutch and English as regards accessibility . Thus were the states are concerned, which in both languages appeared accessible, Dutch and English use present perfect and simple past in the same way . For instance, the Dutch equivalent of a sentence like That man knew Chinese is Die man kende Chinees (simple past: that man knew Chinese) , and not the present perfect Die man heeft Chinees gekend (that man has Chinese known). Finally note that whenjtoen as used in the sentences (73-75) allows several o rderings over situations, b ut that the accessibility-condition limits the possibilities , not allowing the following maps:
1 58 (77)
Jan walked home when the bomb exploded
II� �
I
I
I \It
bomb expl . 1" � >K 1
\jl
(78)
Jan is naar huis gelopen toen de born explodeerde Jan has home walked when the bomb exploded
(79)
Jan walked home when the bomb exploded
So the reading that remains available , as it were , for both (73) and (75) is that first the bomb explodes and then Jan walks home - which is pre cisely the reading that is wanted of course . In the same way , the Dutch simple past of activities and accomplishments , as in (74), will allow both an inclusion-reading and an earlier-than reading, quite in accordance with
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In neither (76), representing the Dutch present perfect , nor (77), the English simple past , can the "Jan walk home"-situation be mapped to t in such a way that the corresponding situation-times have to temporally include another situation-time . In both cases, however, another order over situations leads to no problems , since the map will not result in temporal inclusion:
1 59 the dat a , because non-perfect activities and accomplishments in Dutch are accessible : (80)
Jan liep naar huis toen de bom explodeerde Jan walked home when the bomb exploded
Grote Rozcnstraat
97 1 2
31
TG Groningcn
The Netherlands
NOTES
I.
Dutch uses both
hebben (have)
and
zijn (be)
for the construction of the perfect .
Since nothing in what I will have to say on the perfect h inges on any possible differ en ces between
2.
hebben and zijn, I will for both of them gloss have. ( 1 96 1 : 60- 6 1 ) , Ebeling ( 1 96 2 : 9 7 ) , M cCoard ( 1 9 78 : 1 55 - 1 56),
See e .g . Bosker
who agree that t h e Dutch present perfect can e xpress an event t i m e which i s definitely earlier than the time of evaluation , despite claimed "pragmatic links" with the presen t. Cf. Comrie
3.
(1976: 8
and
5 3)
for si milar remarks o n German.
I will use tenseless sentences like
John walk- home
rather than an "underlying
present tense" - a linguistically unwelcome residue , I think, from Prior's
(1 96 7 )
logically mot ivated om ission o f a present operator fro m his temporal logic.
4.
E'\cept of course when used generically o r reportively.
(1 967) classification in (desire. be at home, know Chinese) , ACTI VITIES (walk, swim, run ) , ACCOM P L I S H M ENTS (walk home, write a letter, build a villa) , and, tho ugh little will be said about the m , ACH I E VEM ENTS (spot, flash, snap). 5 . Some speakers of Dutch prefer a locative here, like lk was een brief aan het scl!rijt•en om 6 uur (I was a letter at the writing at 6 o 'clock). I will ret urn to this Th roughout the rest of the paper I will use Vendler's
terms of STATES
issue in connection with the English progressive in note
14.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Institute for General Linguistics Gro nmgen University
160
1 2 . Minimal situations are not to be associated with moments, but with instantaneit y , a s expressed i n achievement-sentences such as The cable snap-, The light flash, and the like. It may very well be, of course , that if one would put a "cable snap"-event under the microscope , as it were, various further subdivisions of that snap emerge. Thus someone interested in such processes may want to say e .g . When a cable of this size is snapping it has to go through three phases or whatever else could be said on snapping cables (also note the progressive on snap in this e xample). The snap-event is
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
6. The possible access in a case such as ( 3 1 ) is undisputed, but there are other cases where intuitions of native Dutch speakers differ . Sometimes no access is allowed at all, particularly in cases where causality forces a "j ust after" ,eading of when/toen as in Hij vertelde het hele verhaal toen ik hem er naar vroeg (He told the entire story when I asked him for it) . I have disregarded such sentences in this paper . 7. Only in a "deductive" modal reading (which for Dutch requires a change in intonation and usually the addition of e.g. natuurlijk (of course)), corresponding to the English John m ust have played the violin when . . . , may one infer that the bomb exploded sometime during the violin-playing. Even then , not all speakers of Dutch accept such a reading. The availability of a "deductive" modal reading in Dutch is not limited to the present perfect , however, since sentences in the simple past and the simple present can also be used deductively if the intonation is appropriately adapted. I will not be concerned with the fact that Dutch may use a change in intonation where English uses "deductive" must, and will only concentrate on the temporal relations involved. 8. This is an important difference with a language like German, where the perfect does allow access. Thus the perfect in e .g . Sie hat gearbeitet als ich anrief (She has worked when I called) is accessible : the call can be included in the "she work" tim e , which in English would be rendered as She was working when I called her. 9. Because of the "temporal inclusion" relation, presupposing the existence of time-intervals, an obvious theory within which to try and account for the Dutch/ English data would be so- called Interval-Semantics (Bennett and Partee 1 972/78, Cresswell 1 97 7 , Dowty 1979, among others). In de Vuyst ( l 983b : 2 7 -50) it is argued, however, that I nterval-Semantics cannot account for the Dutch data, nor for all the English data, essentially because the various Vendler-classes are inter preted in terms of sections of the time-axis, in terms of moments and intervals , that is. Since moments and intervals are o b viously the same for any language under consideration, Interval-Semantics turns o ut to be too inflexible to deal with other languages than English. 1 0 . See, among others, ter Meulen ( l 9 8 3a). 1 1 . This inclusion is not necessarily a proper one. For a reportive present, such as Bach's example I utter tl!is sentence, E and the relevant situation will temporally coincide:
1 61
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
represented in this case not as an achievement but as an activity. This is not, how ever , how the a verage language user will perceive , and subsequently report on, the snap of a cable. Note that also the other way round is possible, i.e. the representation of an activity as an achievement in e.g. an enumeration like At three I pruned the roses, at four I washed the car, at six I mowed the lawn, . . . Minimality, therefore , must not b e associated with real-world events or moments , but with situations - the representations of t hose events. 1 3 . See e.g. Dowty ( 1 9 7 9 : 1 6 6 ff). 1 4. Non-progressive activities/accomplishments, that is. In the same vein that the perfect was called an "inaccessibility-operator", as at least one facet of its semantics, the progressive may be called an "accessibility-operator", also as at least one facet of its semantics. Thus in all cases where Dutch activities and accomplishments were found to be accessible , English progressive sentences could be used to clarify the meaning of the Dutch sentences : in English, one of the t hings that the progressive does is to turn otherwise inaccessible situations into accessible ones. As pointed out in e .g . Konig ( 1 980), many analyses of the progressive are in terms of part versus whole, which, in the framework proposed here, means that progressive forms are concerned only with parts of the situations with which the non-progressives are concerned ( cf. ter Meulen 1 98 3 b , who takes a similar line). An analysis of t he pro gressive would then be that the interpretation of a progressive tenseless sentence is a set of situations each of which is the (nonterminal) part of the corresponding sit uations that interpret the corresponding non-progressive tenseless sentence . In that way, a progressive activity is still an activity (since non-minimal parts of activity""it uations are also themselves activity-situations) so that an inference like John was walking, so John walked holds (an activity is inferred from an activity). But a progressive accomplishment is not itself also an accomplishment (a situation which is part of a n accomplishment""ituation i s not i n t h e same interpretation-set), s o that the in ference John was writing the letter, so John wrote the letter does not hold (one cannot infer an accomplishment from a non-accomplishment), quite in a ccordance with Dowty's ( 1 9 7 9 : 1 3 3-1 34) "Imperfective Paradox". So it need not be claimed , in order to account for these things , that progressives are states (as in e.g. Vlach 1 98 1 ; cf. Bach 1 9 8 1 ), which they obviously are not. Progressives, unlike states, ca n very naturally express movement. Moreover, a non-minimal subsituation o f a pro gressive situation is in the same interpretation""et. Which is precisely the defmition of an activity: progressives are accessible activities (cf. de Vuyst 1 98 3 a : 17 2-1 74 ). Note that the same holds, mutatis m utandis, for the Dutch locative: it can turn an accomplishment into an activity. He ncr. I do not think that the preference for a locative in some cases in Dutch (also cf. note 5 ) , for instance for sentences such as H1j was een brief aan het schrijven toen . . . (He was a letter at the writing when . . . = He was writing a letter when ... ) , has anything t o do with accessibility - accomplish ments are accessible in Dutch - but with a choice/preference for an activity rather than an accomplishment. That for the when-sentence above an activity is preferred (for other cases such as the simple present that preference is clearly much less or non-existent) seems only natural given the inferences associated with activities/ accomplishments: in the above when-sentence one can, as far as a ccessibility is concerned, use a simple accomplishment, b ut at the same time one does not want to claim that the letter was actually finished . Hence an activity rather than an ac complishment. 1 5 . The perfect is not , unlike tenses , deictic. The occurrence of have is insensitive to the occurrence of such clearly deictic time adverbials as yesterday and tomo"o w
1 62
20. Dutch has an a lternative here which English has not , namely the use of a simple present tense as in Jan is nu a/ een hele poos ziek (Jan is now quite-a-time ill = Jan has been ill for quite some time now). A further compariso n , however, of the use of the present perfect versus the simple present in Dutch and English, is quite beyond the purpose of this paper.
REFE RENC ES Bach. E . , 1 981 : On time , tense, and aspect : an essay in English metaphysics. I n : P. Cole (ed.), Radical Pragmatics, Academic Press . New York. P p . 63-8 1 . Bennett, M . and B . Partee, 197 2 : Toward the logic o f tense and aspect in English . Distributed by I U LC 1 978. Bosker, A., 1 96 1 : Het gebruik 1·an her lmperfectum en her Perfectum in her Neder /ands, het Duits, het Frans en het Engels. Wolters. Groningen . Comrie, B . 1 97 6 : Aspect. Cambridge University Press . Cambridge. .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(ci. Palmer 1 974 : 49-50). There are morphological reasons for not considering the perfect a tense in its own right (cf. McCoard 1 9 78 : 6-7). The interaction of tenses and subject-quantifiers appears to point to the same (Tedeschi 1 98 1 : 240-243). All things considered , it does not seem reasonable to consider have a realisation of tense. 1 6 . Heny ( 1 982) is a proposed revision/adaptation o f Richards' ( 1 982 : 9 1 ) truth conditions for HA VE(A ) , which are , informally, that HA VE(A) will be true relative to an interva l i if there is a subinterval j of i such that A is true relative to j. Since in the framework proposed here the i is the va lue of the map from E to t , Richards' approach would boil down to a literal extension of E, resulting in t he incorrect interpretation that a perfect sentence is true if the corresponding non-perfect situation occurs sometime during the utterance of the sentence. Heny ( 1 982 : 1 42-14 3 ) does speculate about , but does not decide on, an analysis involving two subintervals of the perfect interval. l 7 . In a more general account that also includes the past perfect and the com bination of future modals and the perfect , this condition must be formulated in such a way that not only E, but also other "available" situations can be co-extensive with the final subsituation of the p erfect situation, can function, in effect ,as reference times for the perfect. E is j ust one of those reference-times, quite in accordance actually with Prior's ( 1 9 6 7 : 1 2-1 5 ) discussion of Reichenbach's famous R. So for a sentence like He had waited for half an hour when I came into the room , the "I come"�ituation functions as a reference-time for the past perfect in the main clause. 1 8 . Note that such a requirement in Interval-Semantics would automatically involve a separation of two events on the time-axis; in fact , Dowty uses this requirement for the truth-conditions of the simp le past. In the approach proposed here, however , there is still one perfect situation , and the requirement of separation applies to the internal make-up of that one perfect situation only. Incidentally, such internal make-up may also be affected when, in a more elaborate analysis , time-adverbials are considered . 1 9 . Compare a simple-past representation for e.g. John worked :
1 63
,
.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Cresswell, M .J ., 1 977 : Interval semantics and logical words. In : Chr. Rohrer (ed .), On the Logical A nalysis of Tense and Aspect, Narr, Ti.ibingen. Pp. 7-30. Dowty. D.R., 1 97 9 : Word Meaning and Mon tague Grammar. Reidel , Dordrecht. Ebeling, C.L., 1 96 2 : A semantic analysis of the Dutch tenses. Lingua 1 1 : 86-99. Heny, F.W. , 1 98 2 : Tense , aspect, and time adverbials II. Linguistics and Philosophy 5 . 1 : 5 9- 1 07 . Jespersen, 0., 1 924 : The Philosophy of Grammar. Allen and Unwin , London. Konig, E., 1 98 0 : On the context- dependency of the progressive in English . In : Chr. Rohrer (ed.) , Time, Tense. and Quantifiers, Niemeyer, Ti.ibingen. Pp. 26929 1 . McCoard , R .W., 1 978 : The English Perfect: Tense-Choice and Pragmatic Inferences. North-Holland, Amsterdam . ter Meulen, A.G . B . , 1 983 : The representation of time in natural language. In: A . ter Meulen (ed.) , Studies in Mode/theoretic Semantics, Foris, Dordrecht . P p . 1 77 191. Palmer, F.R., 1 9 74 : The'English Verb. Longmans, London. Prior, A.N., 1 96 7 : Past, Present, and Future. Clarendon Press, Oxford. Richards, B., 1 982 : Tense , aspect , and time adverbials I. Linguistics and Philosophy 5 . 1 : 5 9-1 07 . Saurer, W., 1 98 1 : A Formal Semantics of Tense, A spect, and A ktionsarten. PhD-diss. Pittsburgh University. Taylor, B., 1 97 7 : Tense and continuity . Linguistics and Philosophy 1 .2 :_ 1 99-220 . Tedeschi, P . J . , 1 98 1 : Some evidence for a Branching-Futures semantic · model. In : P. Tedeschi and Z. Zaenen (eds.), Tense and A spect, Academic Press, New York. Pp. 239-269 . Vendler, Z . , 1 967 : Linguistics in Philosophy. Cornell University Press, Ithaca. Vlach, F., 1 98 1 : The semantics of the progressive. In: P. Tedeschi and A. Zaenen (eds.), Tense and Aspect, Ao.:ademic Press, New York. Pp. 27 1 -292. de Vuyst , J . 1 98 3 a : Situation-descriptions: temporal and aspectual semantics. In: A. ter Meulen (ed.), Studies in Mode/theoretic Semantics, Foris, Dordrecht . P p . 1 6 1 - 1 76. de Vuyst, J , 1 98 3 b : A Semantic A nalysis of Temporal Elements in Dutch and English. PhD-diss. Groningen University.
Journal of Semantics 4: 1 65-191 REVIEW ARTICLE
Max J . Cresswel l , Structured Meanings. The Semantics of Propositional A ttitudes. MIT Press, Cambridge, Mass., 1985. Pp. x+202 , $26.9 5 (cloth). ARNIM VON STECHOW
1 . THE LAST GREAT PROBLEM FOR TRUTH-CONDITIONAL SEMANTICS*
(I )
a. Veronica believes that 5+7= 1 2 b . Veronica believes that 1 2= 1 2
How can this be? If the two embedded sentences have the same truth conditions, they express the same proposition in truth- conditional se mantics because the truth-conditions of a sentence and the proposition it expresses are one and the same in this approach, regardless of whether you reconstruct truth-conditions as sets of possible worlds (as Cresswell does), as sets of possible situations or whatever you like . Thus, Veronica believes and does not believe the same proposition under the same cir-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Semantics is a discipline that comes close to our ideal of a cumulative science , although certainly not everyone will agree with this statement . Among the great achievements of semantics within the last hundred years are these : Frege's invention of the semantics for bound variables and quantifiers, Russell's theory of descriptions, Tarski's model theory for the predicate calculus, Carnap's reconstruction of the notion of intension and its application to the semantics of natural language by Montague , K ripke's semantics for modal logic, the Stalnaker-Lewis semantics for counter factuals, the semantics for indexicals and context- dependency in general by Scott, Montague, Kaplan and others. The success of the field has to my mind been possible because of a common methodological ground shared by these semanticists . They believe that the meaning of a sentence consists in its truth-conditions. One of the last unsolved problems of the discipline has been the pro blem of propositional attitudes, !1encefo rth PAs. In a way this problem challenges the entire issue . It is precisely PAs where truth-conditional semantics seems to fail . It is convenient to distinguish at least two facets of the problem, namely problematic de dicta and de re attitudes. Let us start with attitudes de dicta. The problem arises, because logical ly equivalent sentences like "5+7= 1 2" and " 1 2= 1 2" seem to have the same truth-conditions, but ( I a) may nevertheless be false whereas ( 1 b) is true.
1 66 cumstances, a logical impossibility. This is the de dicta aspect of the problem of PAs. The problem is not confined to attitudes towards mathe matical statements. On p. 4, Cresswell gives the following two truth conditionally equivalent sentences . (2)
a. Robin will win. b . Everyone who does not compete , or loses, will have done something Robin will not have done.
(3)
a. Ralph believes that the man at the beach is a spy. b . Ralph believes that the man with the brown hat is not a spy.
The case posited makes it clear that both the man at the beach and the man with the brown hat is Ortcutt , whom Ralph has seen on two different occasions. Therefore Ralph holds the following two beliefs about Ortcutt . (4)
a. Ralph believes that Ortcutt is a spy. b . Ralph believes that Ortcutt is not a spy.
(4a) and (4b) certainly are correct reports of the two beliefs held by Ralph , though Ralph would not express them in this way but rather by means of the two descriptions used in (3a) and (3b ). It is clear from the case posited that the two beliefs of Ralph are mutually compatible, but the usual methods of semantics don't allow to express this . If we let the two des criptions have narrow scope , then the object of Ralphs beliefs is a different proposition in (3a) and (3b). But then the information is lost that Ralph's beliefs are about Ortcutt . So these are not formalizations of the de re readings. I f, on the other hand, we give wide scope to the two descriptions, we obtain the wrong result that Ralph holds contradictory beliefs about Ortcutt : (5)
a. There is exactly one man at the beach such that Ralph believes that he is a spy. b. There is exactly one man with a brown hat such that Ralph believes that he is not a spy.
The embedded propositions are contradictory, because the m an at the beach and the man with a b rown hat are, by supposition, the same . There is at least one further aspect of the problem of PAs , namely the so called de expressione aspect. I 'll come to this later.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Obviously, Miriam may believe (2a) without believing (2b ). The de re aspect of the problem is best illustrated by Quine's famous Ortcutt example .
1 67 2 . ORGANISATION OF THE BOOK AND OF THE REVIEW
3 . TWO NON-TRUTH-CONDITIONAL SOLUTIONS
People have taken quite diffe rent stands toward the problem . To mention two extremes: there are the followers of Quine , who claim that PAs are (somehow refined) relations to expressions, whereas others say that they are relations toward somehow refined propositions, i .e . , toward propo sitions that can't be identified with the truth-conditions of a sentence but have to be something more refined . Let me illustrate the two solutions, together with some criticism. The first solution , viz. that PAs are directed to expressions, is above all boring, because the problem of propositional attitudes doesn't arise there , since it has its origin in the meaning of expressions , not in the nature of expressions qua expressions. It is hard to see what it could mean to have an attitude toward an expression without having a relation to its meaning. And if the latter is conceded , then the problem of PAs will arise again in its full generality, a point made by Cresswell, when he discusses alleged solutions by artificial intelligence researchers (vide below). Let me mention two particular problems this kind of solution has to face . On p. 6, Cresswell discusses the following sentence :
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
After this general setting, let me give some more detailed comments on the contents of the b ook . The book is organized in three parts, Part 1 : Sense and Reference, Part I I : What Meanings Are, Part I I I : Formal Se mantics. Each part has five chapters. The final part is an extensive Biblio graphical Commentary. The latter reviews some of the most important contributions of the enormous amount of literature about the problem. It may be said then that the book has a nice architecture. The first part is the most important one because it develops the problem of PAs and its solution. The second part is a justification of the truth-conditional ap proach to semantics. It is mainly addressed to those who believe that meanings are representations of some sort , i .e. to the majority of psy chologists, AI-specialists, but also to a vast number of linguists, I guess. The third part gives an exemplary formal semantic analysis of a PA-verb , viz. the verb say. In this review , I want to concentrat� on some issues which I consider central to the enterprise. Given the long history of unsuccessful attempts to solve the problem of PAs, the technical elaboration of the theory is very important, in order to see how possible objections are met . There fore , I want to devote some attention to a detailed discussion of Cress well's exemplary analysis of the PA-verb say.
I 68 (1)
Yvonne knows what 5 +7 is.
It is very hard to reconstruct the knowing relation here as a relation to ward an expression , because the embedded sentence is a nominalisation having roughly the logical structure "that x such that x is 5+7". The other problem is connected with Quine's solution of the Ortcutt example , which is something like this : (2)
a . Ralph believes of Ortcutt 'is a spy'. b . Ralph believes of Ortcutt 'is not a spy'.
(3)
[=( 1 ), p. 54] Ambrose says that it will rain.
is true , then Amb rose will have utte red a particular sentence . This sentence is called the object of the PA. The content of the PA reported by (3) is the meaning of "that it will rain". It is obvious that for our understanding of (3), the object doesn't play any role, because we understand (3) without knowing the sentence Ambrose actually produced . We understand (3) even if Amb rose produced no sentence at all . So it is the content that matters for the meaning of a PA-sentence . Cresswell points out that the semantic p roblem of PAs arises when PAs are relations toward a content . On the other h and . the psychologist might be interested in investigating relations
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Thus, Quine assumes that a relational belief goes toward an individual and an expression . This kind of analysis assumes an ambiguity for PA-predicates . If we t ake the sentence discussed by Russell in The problems of philo sophy "Othello believes that Desdemona betrays him with Cassio" , then the belief relation will have more argument places , namely one for Othello, one for Cassio and one for 'betrays'. Even if we admit that this kind relational belief is correct (and , as we will see , it is correct in some sense), then we have to say that the analysis can't stop here , because we have to spell out what the logical connection between the different belief-relations is, because , intuitively , the belief-relation is always the same . Even if one does not like Quine's solutions , which are dictated by ontological parsimony, one certainly has to admit that Quine himself is always aware of the real difficulties of the problem of PAs. Unfortunately , this cannot always be said about the work done by psychologists , artificial intelligence researchers , and linguists. In our days, it is fashionable to say that the objects of attitude are representations in some internal language . The purpose of Chapter 6 Why Meanings Are Not Internal Representations is to show that this position is, to say the least , insufficient. A well known supporter pf this view is Fodor ( 1 98 1 ). In discussing Fodor's work on PAs, Cresswell introduces an object - content distinction . I f
1 69 between an organism and objects , where the latter may among other� be representations in a language of thought. Cresswell points out that what ever such investigations might find out, they won't contribute anything to the semantic problem of PA. For instance , (4) shows , that the object of diffe rent PAs are often quite different things , viz. a representation in an internal language as opposed to a representation in a public language , whereas the content is the same . (4)
[=(2), p . 5 5 ] Beatrice believed what Callum said.
" . . . their arti cle gives the strong impression that workers in the com putational sciences feel able to provide answers to questions that logicians had been fumbling about with for many years. It is important to realize , therefore , that the question these workers have answered are not the questions that logicians have been asking."
I take it that the last sentence is again pure politeness.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Cresswell suggests that the semanticist and the psychologist are apparent ly engaged in different enterprises. I am inclined to believe that this statement mainly betrays a cautious politeness. On page 54, Cresswell gives an argument that makes it doubtful that objects , i.e. representations in some internal language , can play a crucial role in explaining how beliefs influence our behaviour. If two people have the same belief, then accord ing to Fodor each one has a particular sentence in his mind (more ac curately, a token). Even if it is the same sentence (which is very unlikely for an Englishman and an Eskimo), this sentence can only play a similar causal role , because it represents the same proposition. It follows that not objects but contents play the essential part in a theory of causal ex planation for beliefs . It seems to me that a consideration of this kind defeats Fodor's enterprise , though Cresswell does not claim this . What he claims is that Fodor doesn't have to say anything about the semantic problem of PA. Cresswell uses plainer language when he discusses a work of the com puter scientists Moore and Hendrix (I 982) who maintain that 'A believes that S' is true if A has in his belief set a formula that has the same meaning as S. This analysis is supposed to solve the problem of PAs. Cresswell notices that the analysis lacks any credit , because , if 'has the same meaning' means 'expresses the same proposition as', then the relation is in reality not to a sentence only , but also to a proposition . But then the problem of PA will arise again. For a similar argument , vide p. 42 ff. Let us illustrate Cresswell's attitude toward this kind of solution with his own words :
170
"The reason I find this unsatisfactory is t hat it seems to me that the meaning of - is completely exhausted b y its truth-conditional des cription. I find it very odd to imagine that - is a propositional operator whose meaning I only partially know." ( p . 7 4 )
On the othe r hand, to make propositions finer won't help anyway , for nothing prevents us to define a super-strict-negation, which has the proper ty that --p is the same as p. Then we have our problem again, because it must be possible to believe --a without believing o:. The moral of this is always the same : the problem of PA arises with compositonality, no matter what kind of entities we assume for semantics. This certainly is a challenge for all those who try to solve the problem of PAs by looking for fmer-grained properties and propositions. If Cresswell is right , then their search will be as successful as a search for unicorns . What both kinds of solutions have in common is that they take it for granted that the truth-conditions of the embedded sentence can't be the right object for the PA. If one of these positions were correct, then truth conditional semantics would be seriously deficient . So the problem of PAs has presumably been the last nail in the coffin for anyone who believes in t ruth-conditional semantics but who cannot solve the problem. Cresswell w rites on this :
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
So far, w e have been concerned with positions claiming ·that PAs are relations to expressions. Let us come now to the second kind of solution, which consists in the assumption that the obj ects of attitudes are proposi tions, but that propositions are something finer than truth-conditions. {I feel free to use the term "object" in a different way than Cresswell in his discussion of Fodor's work.) In particular, a proposition is more than just a set of worlds. There might be two propositions p and q which have the same intension but are nevertheless different. Thus, one might have an attitude toward p without having that attitude toward q. Cresswell calls this kind of solution a "propositional approach". Cresswell held this ' position in his dissertation more than twenty years ago , but he gave it up since for the following reason [vide Ch. 7 ] . His discussion refers to Thomason ( 1 980). Suppose we have propositions that have the same intensions but are nevertheless different, i.e. p=Fq but I(p)=I(q), where I{ ) stands for "the intension of'. What could be the meaning of the negation - ? We certainly want 1{-"1') to be I(p). But, of course , -"1' can't be the same as p , because one can believe a without believing --a. So there must b e some thing in the meaning of - that makes --p different from p .
171 "The problem of propositional attitudes has intrigued me for over twenty years. At one time or another I have probably espoused virtual ly every solution available." [ p . 6 ]
4 . THE STRUCTURED MEANING APPROACH
(I)
a. b.
where (a) consists of the relation which holds of any two numbers if they add up to 1 2 and the numbers 7 and 5 . Similarly, but with a few further complications for (b). The idea is that a that-clause ambiguously ex presses sequences , where P is an n-place property and a 1 , . . . ,an are arguments of the right type . Any such sequence is called a structured meaning. The structured meanings of a that-clause have in common that they all express the same proposition when the predicate is applied to its arguments. Thus for both (a) and (b) we obtain the ne cessary proposition, when we apply the first member of the sequence to the following members. But , of course , (a) and (b) are different . So there is at least no logical problem for Veronica to stand in the believing relation to (a) without bearing that relation toward (b). Given that that- clauses are ambiguous as to the structured meaning they express, even one sen tence may report different beliefs. Thus, different sentences will a fortiori report different beliefs , at least in general . So the object of the belief relation in "Veronica believes that 1 2= 1 2 " is perhaps the structured meaning (2)
<=, 1 2 , 1 2>
There is no reason to assume a priori that Veronica should always stand in the belief relation to (2) when she bears that relation to (I ar or ( I b). This example concerned the de dicto part of the problem. The de re aspect is solved in exactly the same way . The Ortcutt examples are analyzed in the following manner.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The solution advocated by Cresswell is indirectly indicated by the title of the book. A PA is not a relation to a meaning simpliciter but to a structured meaning. For instance , "Veronica believes that 7+5= 1 2 " will presumably not report a relation between Veronica and the set of all possible worlds, i.e . the necessary proposition , but a relation between Veronica and the sequence (a) or (b) (or even something more complicated),
1 72
(3)
a. Ralph believes . b. Ralph believes .
(4)
a . Veronica believes that 1 2 is 1 2. b . Veronica believes that the square root of 1 44 is 1 2 .
But (4b) can't mean the same a s (4a), because (4a) may b e true under circumstances where (4b) is false . At first sight, this objection seems to defeat Cresswell's approach completely. We will see, however, that this is not so. This shows, at any rate, that the structured meaning approach has to be carefully spelled out. 5. THE A�1BIGUITY OF PA-COMPLEMENTS
One of the main points of C resswell is that the key for a solution of the problem of PAs is to recognize that the complements of PA-predicates
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In other words, in (3a), the object of belief is the structured meaning consisting of the property of being a spy and Ortcutt, whereas in (3b) it is the property of not being a spy, and Ortcutt. One has to say, of course , what it means for an attitude to be toward a structured meaning. Further more , one should indicate how a structured meaning is obtained from the sentence expressing it . I will come b ack to the first question below. As to the second question, the answer is that the formal language used by Cresswell , viz a X- categorial l anguage , is so close to natural language that a translation into it certainly poses no more difficulties than a translation say into the language of intensional logic. It should be noted that Cresswell's solution is entirely within a t ruth conditional framework viz a possible world approach. It would seem then that the approach is open to the same objections that have been raised against "classical propositional accounts" of the problem, i.e . accounts where the that- clause expresses a set of possible worlds or something equivalent. The problem arose because we can't substitute that- clauses expressing the same proposition sa/va veritate. Now, the members of Cresswells structured meanings are intensions which are denoted by constituents of the that- clause . It seems then that we can replace these constituents by any constituent denoting the same intensions. Consider, fo r instance (4a). Since the term "the square root of 1 44" has the same intension as "1 2 " , we would expect (4a), when taken as a report of an attitude toward the structured meaning (2), to be synonymous with (4b ), where (4b) is obtained from (4a) by replacing the first occurrence of " 1 2 " by "the square root of 1 44".
1 73 are ambiguous. Cresswell makes this point with the arithmetical sen tence "7+5 = 1 2 " . If we try to fmd out whether a statement of this kind is true, "We begin with certain n umbers, perform an operation on them, and end up with a certain result. And in all cases the same moral can be drawn . /t must be logically possible to get it wrong. " [ p . 1 1 )
This becomes clearer, perhaps, when we think of how the meaning of [=(2) , p. 1 2 ) Xavier has calculated what 5 +7 is.
can be described. Cresswell offers the paraphrase (2)
[= (3) , p. 1 2 ] Xavier has taken the two numbers 5 and 7 and per formed on them the operation of addition.
This statement is not about the number 5+7 , i.e . not about 1 2 , but about the numbers 5 and 7 . " I shall argue that the key to solving the problem o f propositional attitudes lies in recognizing this ambiguity in the expression '5+7 ' : it sometimes refers t o the two 'input' numbers 5 and 7 together with the operation to be performed on them, and it sometimes refers to the result of performing that operation." [ p. I I f. 1
At this point a historical remark may be in order. When Church in troduced functional abstraction by means of the A.-operator, one of the motivations was precisely to remove the ambiguity of the notation (3)
x+y
for this notation could mean "the result of applying addition to the num bers x and y" or "the function of addition simpliciter". The introduction of the A.-notation removes the ambiguity. In Church's formalism, which has since become the Coca-Cola of the semanticist , (3) can only mean the former, whereas the function has to be represented as A.xy [x+y] or +. What this shows is that the ambiguity is really there, or otherwise , Church couldn't have disambiguated the notation. Similarly, "5+7" is ambiguous between 5+7 , i.e . 1 2 , A.x [x+7 ](5), A.y [5+y](7) and A.xy [x+y](5 ,7). As we will see , the same kind of ambiguity can be found in that-clauses quite generally. It is presumably helpful to say a bit more about the truth-conditions
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(I)
1 74 underlying (2) . So let me b riefly speculate about them. Perhaps, it is something like this : Xavier had in mind a particular method of addition (say Adam Riese's algorithm or the abacus). Therefore , he stands in a de re relation to addition . Then he applied that method to the appropriate rep resentations of 7 and 5, thus bearing a de re relation toward the two numbers. Then he applies the algorithm to the numerals. Perhaps, he applies it correctly, perhaps he makes a mistake at some point . Thus , the outcome of the "calculation" is a contingent matter. (Notice, by the way, that the formalization of ( 1 ) is not a t rival matter. The logical form of ( I ) is presumably something like (4) : Xavier has calculated the z .
The problem with this representation is that we have to give sense to the description at the object position. For some discussion of a similar problem , see von Stechow ( 1 984).) Similar considerations apply to the analysis of knowing. An appropriate paraphrase of (5 a) is something like (5b) : (5)
a. b.
[= (6), p. 1 3 ] Yvonne knows what 5+7 is. [= (7), p . 14] Yvonne knows what the result would be o f taking the numbers 5 and 7 and (correctly) applying to them the operation of addition.
A third example for the structural ambiguity of the PA-complement is this : (6)
a. b.
[= (3), p. 34] The map indicates that it is 5+7 kilometers from A to C . [= (4), p. 34] The map indicates that it i s 7+5 kilometers from A to C .
The former may be t rue and the latter false . This i s only possible , if we structure the meaning of the embedded clause , because the two that clauses e xpress the same proposition. Obviously, the order in which the numbers 5 and 7 are added play a role for the correct interpretation of the two sentences. Notice that these examples show that Cresswell uses the term PA in a rather wide sense . Any verb or operator that embeds a clause falls into this category. This generalization is justified by the fact that the problem of PA arises with any such predicate . When we were discussing Quine's analysis of relational belief, I criticized the assumption that PA-predicates are ambiguous. One may ask then why it is preferable to situate the ambiguity in the PA-comple-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(4)
1 75 ments. In Ch. 4, Cresswell gives an argument in favour of this position. Suppose we want to analyse a PA-sentence of the form (7)
x
(8)
x believes iff there are descriptions d 1 , . . . , d n that pick out a1 , . . . , an respectively such that x believes the proposition P(d 1 , . . . , d n).
Applied to the Ortcutt examples (3a) and (3b) of Section 4, this rule yields the truth-conditions (9a) and (9b) respectively [cf. p. 22]. (9)
a. There is a way of picking out Ortcutt (viz being the man at the beach) such that Ralph believes the proposition that whoever is so picked out is a spy. b . There is a way of picking out Ortcutt (viz being the man with the b rown hat) such that Ralph believes the proposition that whoever is so picked out is not a spy.
The analysis (8) leaves believe unambiguous. As a special case of this analysis we get ordinary propositional belief. This happens if we don't structure the that-clause. To be sure, the belief relation takes differ ent structured meanings as objects, but this doesn't make it more am biguous than any other function that is defined for more than one ar gument . For instance, we would not say that the operation of addition is ambiguous because it is defined for different numbers. This approach certainly makes the content of the PA a contingent proposition . Recall that Lewis ( 1 979) has shown this analysis still to be too crude, because
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
and we find out that it expresses several different propositions. This will in general be the case, as the discussion of the previous chapter has shown . Then there must be an ambiguity somewhere in (7). Where is the ambiguity to be located? If we assume that
1 76
( 1 0)
a. Max believes that Miriam steals the spoons. b . Max believes that Miriam steals the spoons and eithe r eats the cheese or does not eat the cheese. "But the use of such a sentence would almost certainly indicate t hat a different logical form is being assumed. Although the theory allows for a logical form in which the more complex sentence can replace the simpler one , it is often the very use of the more complex sentence that indicates t hat the more complex structure is involved." ( p . 79 ]
The conclusion is that a PA-verb should at least respect the macro structure of the that- clause , i.e . at least the embedded connectives, if there are any. For { l Ob), it follows that the structured meaning of the that-clause has at least the components "x steals the spoons", "x eats the cheese" and "x doesn't e at the cheese". That a PA-verb should respect the macrostructure of the that- clause is c alled the Macrostructure Constraint. It follows that ( l Oa) and ( l Ob) can't have the same meaning, if the macrostructure constraint is obeyed . A constraint of this kind can meet the objection against the structured
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the content of a PA should rather be a property. But Cresswell is not worried about refinements of this kind. The point of the chapter is to show that a de re analysis of PAs requires structured meanings or some thing equivalent. It should be noted that, given the analysis (8), the content o f Ralph's belief still is a proposition. But it isn't the proposition expressed by the unstruct u red that-clause . The trick is that parts of a structured meaning are replaced by descriptions which are determined by the context . When this replacement is done, we obtain a quite different proposition, which is the actual content of the belief. So one origin of the enormous vagueness of sentences reporting attitudes is caused by the fact that there are literal ly infinite possibilities to pick out particular components of a structure within a particular context. There is more to be said about this, of course . I refer the reader to Lewis ( 1 979) and to Cresswell and Von Stechow ( 1 982). The other origin of the vagueness rests on the structural ambiguity. The question is : How much structure should be given to the meaning of a that-clause? Cresswell doesn't try to answer this question exhaustively , but, in Chapter 9 , gives a partial answer, which is motivated by the follow ing example . Suppose we have the property 'steals the spoons' as a con stituent of some structured meaning. This property is intensionally alike to 'steals the spoons and either eats the cheese or does not eat the cheese ' . It seems to follow that the following two sentences have the same truth conditions.
1 77 meaning approach which was mentioned at the end of Section 4. There we said that we could replace "12" by "square root of 1 44" and obtain the same structured meaning. Presumably , the latter term induces further structure , if it is used. Otherwise, its use would be pointless.
6 . ATTITUDES DE EXPRESS/ONE
(1)
[= (25), p . 48) Prunella has j ust learned that Venus is uninhabitable. "Now let us imagine that Prunella lives in a community of scientists who use the name 'Phosporus' to refer to Venus and that Prunella has never heard the name 'Venus'. These scientists, however, have sent their own rockets to Venus and have obtained information about its atmos phere. They have just learned that it is uninhabitable. So ( I ) is true . But the following sentence is false : "
(2)
[= (26), p. 49] Prunella has j ust learned that the thing of which 'Venus' is a name is uninhabitable.
This analysis assumes an ambiguity of names. Names are mostly directly referential but sometimes they denote descriptions. Such a move puts de expressione attitudes closely to what is usually called a de dicto analysis ; Notice that the Ortcutt example obviously can't be analyzed as a de expressione case . Neither can mathematical attitudes generally be analyzed in this way as shown by (3).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
There is a long tradition in the lite rature which speaks of a de re reading when a variable in an opaque context is bound from outside ("quantifying in"). It should have become clear from the preceding discussion that this is not the way the term de re is used here. This term is reserved for e xamples of the Ortcutt kind. The essential feature of a de re attitude is that the subject of the attitude stands in a relation of acquaintance to one or more entities of the structur ed meaning. The discussion has furthermore shown that there is no real distinction between de re and de dicta attitudes, because the latter are special cases of the former. Could it be said then that every attitude is de re? It seems to me that this could indeed be maintained . Cresswell , however, introduces a third kind of attitudes, namely the so called attitudes de expressione. The idea is that in a PA-context, names sometimes denote descriptions. For instance, "Ede" sometimes means 'the person called "Ede".' Whenover this occurs, we have a de expressione attitude . A typical example is
178 (3)
[= (27), p. 50) Trevor is now able to prove that 13 is a prime.
This can be true even if the de expressione interpretation of 3, viz (4), is false . (4)
Trevor knows that the number whose name is " 1 3" is a prime.
t.
THE FORMAL SEMANTICS OF INDI RECT DISCOURSE
So far the discussion has been informal. Let me finally come to the formal p art of the book. This part is not only written to please those who only believe in formal theories. The reason is rather to test wheter a semantic t heory incorporating the ideas mentioned is possible at all. Sometimes you see the untenability of an idea only if you want to write it down properly. Cresswell gives the formal semantics for just one PA-verb , viz "saying". I n several respects, this verb is less problematic than "believing" [vide p . 5 7 , for some discussion], but the problem of PA arises here with all its ramifications. So it seems reasonable to assume that a formal semantics for all PA-verbs is possible , if a formal semantics for "saying" is possible. The formal language used by Cresswell is a categorial language with A-abstraction. The language is introduced explicitly in Cresswell { 1 973). For the following I will presuppose familiarity with the notational con ventions (which are easy enough). The ontology is more o r less like the one found in Montague's "English as a Formal Language" or Lewis' "General Semantics". There is one feature that should perhaps be pointed out to people who are used to thinking in the PTQ framework. There are no extensions in this system. Predicates are functions from individuals to propositions, i.e. to sets of worlds. For simple predicates it is however possible to define the extension . For instance, the extension of the one place predicate P in a world w is the set of those individuals x, such that w E P(x). For the following discussion, "V" will denote the inter pretation function.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This is so because Trevor may speak a language where 1 3 is not called " 1 3". Notice , by the way, that the de expressione reading gives us the same proposition we would obtain by means of Stalnaker's method of diagonalization, because , if I am right, diagonalization turns a name into a description , namely "the object having the name". Presumably, Cresswell's view that names are ambiguous between a referential and a descriptive reading is more revealing. I remember that it took me quite a while to realize what is behind diagonalization (and , perhaps, I have a wrong understanding of this method even today).
1 79 The most important meaning rule is the one for that, because it is this rule which generates structured meanings. Cresswell assumes a family of symbols that ((O/a 1 , ... , an)a 1 , ... ,an), each of type (1 /(0/ a 1 , ... , an ), a , ... , a ). For any such that, a , ... , a in D 0 1 , ... D 0n respectively and 1 n n 1 w E D(O/ al , . . . ,an) we have it that ( 1 ) V (that) (w,a1 , . . . ,a ) is <w,al , . . . ,a >. n n
(2)
V(says) (a,b) is true in a world w iff a produces a sentence a in a A.-categorial l anguage whose interpretation V* says that V*(a) is b
This is tantamount to stating that to say b means to utter a sentence which has the meaning b. This rule is only appropriate for 'propositional saying'. For instance, (3a) could be analysed as (3b ), where it is raining is an unanalyzed expression of category 0 . (3)
a. b.
Max says that it is raining. <Max, says, >
Chapter 1 2 considers cases where says operates on a structured meaning. The first example concerns a de re attitude which is analogous to the Ortcutt case. Suppose Cynthia utters the sentence (4)
The man in the blue coat is singing.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
It may be worth noting that Cresswell's ontology presupposes that D 1 is very large . I t contains any other domain as a subset and it must contain sequences like the one mentioned in the meaning rule , though Cresswell does not explicitly stipulate the latter when he builds up his ontology. (Presumably, a sharp formalist will discover a number of small negligences of this kind.) So his type theory is in some sense cumulative . It would be rather cumbersome to express a meaning rule like ( 1 ) in a strictly typed framework like e .g. Montague's Intensional Logic. Presumably, we would have to defme an entirely new system of types for that- clauses together with the appropriate semantic domains. This would complicate things considerably, and not yield any new insight but almost certainly obscure the idea which lies behind the rule. Cresswell makes a remark as to the policy motivating his framework : "Montague Grammar c an be very terrifying" [p. 9 5 ] . I think that's right. Chapter 1 2 , Indirect Discourse, gives the first tentative meaning for the OA-verb says [p. 1 06]. Roughly, it is this:
1 80 If this is taken as a de re statement of Cynthia about Bruce, then the logical form of (S a) must be something like (Sb), provided Bruce is the man in the b lue coat whom Cynthia has in mind . (5)
a. b.
Cynthia says that Bruce sings .
>
The meaning rule for says has to be extended to cover the de re case. This proceeds exactly according to the idea which underlies the analysis of relational belief (vide the discussion in Section 5). For the special case underlying (Sb), we obtain something like : V(says)(a,< w , b>) is true in w iff a utters a A.-categorial sentence of the form in w where a is a nominal that picks out b for a in w and V*(/3) is w , where V* is the interpretation of the language a is speaking in w.
According to this semantics, (Sb) may be true if Cynthia utters a A.-cate gorial version of (4) and the nominal "the man in the blue coat" pies out Bruce for her. (We have to say what "picking out" means, of course. See for this p. 1 1 3 .) Notice, that we could have represented the same de re statement as (7), provided, Bruce is the man with the blue coat, (7)
�The-man-in-the-blue-coat,>
The example shows that we can quantify in a structured meaning. This is not astonishing, because any that is an ordinary intensional operator. 7 An analysis of the same kind applies to mathematical cases like (8a), whose logical form is perhaps (8b). (8)
a. Cynthia says that 59 is prime. b.
Notice that none of these cases involves quotation . But there are , of course , such cases , which can be seen from the obvious difference in meaning between (9a) and 9b ). (9)
a. (= (8), p. l i S ) Elinor said that I whistled. b. (= (9), p. l i S) Elinor said 'I whistled'.
These are analyzed by means of the quotation symbol qu, which is in troduced on p. I 03, and which gives us for any expression the very
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(6)
181 expression as the value , i .e . , for any ex we have V( is ex. By stipulation , we make sure that is a name , i .e. an expression o f type 1 . Neg lecting tense , the logical form of (9b) is
( I O) <Elinor, says,>>
( 1 1) [= ( 16), p. 1 1 7] Bill says that Arabella perspires. Since it is impolite to use the word "perspire" for ladies, who only "glow", the word itself matters. On the other hand , Bill hasn't said "Ara bella perspires" (he doesn't even know her by name). He has said rather { 1 2) [= ( 1 7) , p. 1 1 7 ] Doesn't that dame in the corner perspire? Therefore, a de re case is involved. And, of course, the content of "pers pires" matters, too, because Bill has attributed a particular property to Arabella. The solution, Cresswell indicates, is something like this ( 1 3). ( 1 3)
< BiU, says,>
qu* is a content-quotation symbol that gives us for any ex the content of ex and ex itself as value , i.e . V() is the ordr.red pair . We have to extend the meaning rule of says to account for such a case . This could be something like this : ( 1 4)
V(says)(a,>) is true in a world w iff at w a produces a sentence <{3,ex> in which (3 is a nominal that picks out b for a at w and where V* is the interpretation for the language a is speaking at w and V*(ex) = c
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Notice that in this case the object of says is a name which denotes the A.-categorial sentence , in other words, the content is not a structured meaning. The 'simple' meaning rule (2) for says guarantees that ( 1 0) is t rue iff Elinor utters a token of . Given that we can obviously formulate a general meaning rule for says where the unstructured case is a special case of the structured one , we have the nice result that the quotational reading is obtained without stipulating any ambiguity for the verb says. This is in accordance with the constraint mentioned in Section 5 which says that the PA-verbs are univocal, the ambiguity lying in the sentential complement. The most complicated cases are those which are a mixture of a de re analysis and a quotational analysis . Cresswell discusses the case where a shocked lady says
1 82
8. DISCOURSE DE SE
So far my comments have been entirely positive . Chapte r 14, Discourse De Se, calls for objections . It seems to me that something has gone wrong here. Discourse de se is a special case of what Lewis (I 979) calls an attitude de se. David Lewis was the first person to realize that in many cases the object of an attitude can't be a proposition but rather has to be a property. Consider the sentence (I)
[= (2), p. 1 2 1 ] Zeus says that he is Zeus .
If what Zeus has said were a proposition , then it would have to be the proposition that Zeus is Zeus, i.e. the set of all possible worlds . But this can't be right, because this proposition is trivial, noninformative , whereas Zeus has said something informative. He might , for instance have said "I am Zeus", when Leda asked him who he was. Another example is (2)
[= ( 1 5), p. 124] It is now four o'clock .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(Cresswell's book has V*(a)=b, which is obviously a misprint .) For the evalu ation of ( 1 3) , we have to apply V(says) to Bill and the sequence >. This will give as the proposition that Bill produces a sentence <(3,perspires> suchs that t3 picks out Arabella for him V*(perspires)=V(perspires), where V* is Bills interpretation of the language (whereas V is ours). The latter requirement makes sure that Bill means the same with the predicate "perspires" as we do. A purely quotational ap proach couldn't incorporate this feature into the meaning. For further discussion, vide p. 1 1 7 . So this is a de re analysis as before, the only difference being that the qu*-symbol guarantees that a particular constituent of the structure has to be expressed by a particular expression. It should be obvious from this remark that this is a more general account than the plain de re account , which we have been discussing most of the time , which doesn't specify the way a res is expressed . So the plain de re attitude can be subsumed unde r the mixed attitude. Cresswell doesn't work this out . It is clear, however, that a general meaning rule for says has to be shaped after ( 14), according to Montague's strategy of 'generalizing to the worst case'. Cresswell doesn't offer a general analysis covering all cases. Such an analysis might not be entirely trivial, but it is reasonable to assume that it is possible. All this is certainly still very crude and needs some refinements. I must admit , however, that I find this analysis of indirect discourse highly ingenious.
1 83 Suppose I say this at four o'clock . Since "now" refers to the time of utterance, it would seem that I expressed the proposition 'At four o 'clock it is four o'clock', a triviality. The solution Lewis suggests for these apparently paradoxical findings is that the object of an attitude is a property. In the case of ( l ) it is the property of being Zeus, in the case of (2) it is the property of being temporally located at four o'clock . The descriptive problem is how to get this property from the that- clause that has to express it. This is the problem of this chapter. Unfortunately, Cresswell doesn't discuss possible solutions by considering sentences like ( 1 ) or (2), where the problem becomes particularly salient. He switches to the sentence : [p. 1 2 1 ] Zeus says that he throws thunderbolts.
This sentence is not so illuminating , because the complement is informative even under a classical propositional account. I will therefore discuss Cresswell's solution by considering ( l ) . It will become clear very soon that his solution is untenable . Among the possible solutions which Cresswell considers is a de re analysis , viz (4)
< Zeu s, says , < that((O
/ 1 ), 1 ) , Zeus, is , Zeus>> .
Cresswell comments on this : "Perhaps this might be sufficient to close the matter" [p. 1 22]. In Stechow ( 1 984) I defended the view that this is indeed the case. For, as Lewis has pointed out , a de re attitude becomes a de se attitude , if the suitable relation that connects the subject of the attitude to the res is the relation 'the person I am' , i.e. identity. To remind the reader how this works, I repeat Lewis' de re analysis. (5)
x ..ps P of y iff (i) there is a suitable relation R connecting x with y and (ii) x ..ps the property of bearing R uniquely to some object which is P
Suppose now, R is identity. For (4) this means that it is true iff (i) Zeus is Zeus, which is trivially so, and (ii) Zeus says the property of bearing identity uniquely to something that is Zeus. But the said property is the property of being Zeus, as can be seen if we spell it out by means of logical notation : (6)
A.x( 3 y [ V'z [x=z <=> z=y] & y = Zeus])
This is the same as A.x [x=Zeus] i .e . 'being Zeus'. So (4) yields the right
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(3)
! 84 reading. We must , of course , accommodate the de re analysis along the lines indicated by Lewis. Notice that the me aning of says makes it possible for Zeus to make (4) true by uttering: (7)
Zeus is Zeus.
Let us see now what Cresswell has to say to these cases. Following Castaneda, he assumes that pronouns are ambiguous between their ordinary reading and a de se reading. He formalizes the latter by means of the star: he* ,it* denote de se pronouns . What is the semantics of these? First , Cresswell correctly points out that the sentence : (8)
[= ( 14), p. 1 24] It is four o'clock .
denotes the set of world-time pairs < w ,t> such that t = four o'clock. (Never mind that the time has to be relativized to a particular day and place.) This is what we would expect , having in mind Lewis' analysis. For, according to this analysis, this sen tence should express the property of times of being four o'clock, and this property can be identified with the world-time pairs satisfying it . Cresswell assumes accordingly that pro positions are sets of world-time pairs. He furthermore assumes that his X-categorial language contains the expression : (9)
[= ( 1 6) , p. 1 24] it* is four o'clock
which denotes the aforementioned set . The info rmative reading of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
or, more accurately, a A·Categorial version thereof. This might seem strange , because we would rather expect him to utter a sentence like "I am Zeus" in order to get the de se inte rpretation . Perhaps this is the reason why Cresswell didn't stop with the analysis of de se at this point instead of trying a different analysis. Personally, I believe that it is a matter o f convention whethe r (7) can have a de se interpretation . When I was young, I read many books about Red I ndians , where Big Bear used to speak like this: "Sig Bear has spoken. Hough , hough! " Clearly this is to be taken in the de se sense. The only thing we have to make sure of in order to ge t the de se sense is that the subject term must be able to refer to the subject itself, when the subjec t utters the sentence. Thus, speaking to himself, Zeus might attribute to himself the property of being Zeus even by "Pull yourself together, old chap. You are Zeus". And , of course, de se was involved, when Queen Victoria commen ted on Ibsen's Nora "We are not amused".
1 85 ( 1 0) [= ( 12), p. 124] Zeus says that it is four o'clock. can then be formalized as ( 1 1 ) [ = ( 1 7), p. 1 24] >
( 1 2) ,>>> he* is an operator that makes a nominal out of a name , i .e . he* is a sym bol of type ((0/(0/ 1 ))/ 1 ). The meaning is described by the following rule [cf. p. 1 25 ) : ( 1 3) V(he*) i s a function r such that for any a € 0 1 i n its domain and w € 0(0/ 1 ) in the domain of r (a) and any <w,t,p>, <w,t ,p> Er(a)(w) iff <w ,t,a>Ew(a) According to this rule , V(<, is, Zeus>) is true at <w,t ,p> iff <w ,t ,Zeus> EV(>)(Zeus). And this is the case iff Zeus is Zeus. This means, however, that this sentence expresses a tautology, and this can't be right, because the object of the attitude reported in ( 1 ) shouldn't be a tautology but should rather be the property of being Zeus. In other words, it should denote those <w,t ,p> where p=Zeus. It seems to me that this argument defeats Cresswell's analysis. But let me try to repair it. In view of what I have said the role of de se pronouns seems to be to 'bind' the implicit person parameter, thus yielding a property of persons. The revised rule for he* would be then this: ( 1 4)
he* is a nominal. For any property w , V(he*)(w) is true at <w ,t,p> iff <w,t ,p> € w(p) .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I have no objections to this. Notice that Cresswell is able to obtain the correct result within a classical propositional account , i .e . by working with that0 , because he has adopted a 'fatter' notion of proposition in taking a proposition to be a set of world-time pairs instead of a set of worlds simpliciter. A fatter proposition of this kind will automatically be a property of times . It seems straightforward then to extend the strate gy of fattening to the de se cases, since, as we have seen, the content of a that-clause will here typically be a property of persons. So it seems natural to consider propositions as sets of world-time-person triples, for which the notation <w ,t ,p> is henceforth used. We would expect then that the de se reading of the sentence "I am Zeus" denotes the set of those < w ,t ,p> where P is Zeus. Unfortunately, our expectations are disappointed. Let us see how Cresswell would analyse our critical sentence ( 1 ). The formaliza tion he offers is this:
1 86 Cresswell's formalization ( l l) should be replaced b y : (1 5)
>>>
This time we obtain the correct result , because >> is true at <w,t,p> if p is Zeus . Notice that this analysis carries over to the unanalyzed (9). Assuming that the it* in (9) abstracts over the implicit time parameter t, we can replace (9) by ( 1 6) : >>
My repair is compatible with everything Cresswell says in the rest of the chapter. He claims, for instance , that the pronoun "I" has to be analyzed in the same way as he* with the exception that a he* translating "I" must have wide scope with respect to any PA-operator. According to this analysis, the logical form of ( l 7a) is ( l 7b): ( 1 7)
a. I am Cresswell . b . >>
This certainly is a correct formalization . Of course, this is a rather trivial case , because there isn't any PA-operator in the sentence. A more genuine example is ( 1 8)
[= (20) , p . 1 26] Zeus says that I throw thunderbolts.
which has to be formalized as ( 1 9), where the formalization presupposes the revised semantics for he* : ( 1 9)
>>>> .
Roughly speaking this is the set of those <w ,t ,p> where Zeus utters in <w ,t ,p> a sentence that expresses the proposition that p throws thunder bolts. Notice that the p is the speaker of the sentence ( 19). <w ,t,p> may be regarded as an egocentric situation . It is of course not forbidden that Zeus says something in the very same situation . Cresswell comments on his analysis of "I" : ''Thus, we have given a semantic characterization of the pronoun "I" [p. 1 24]. This might seem strange to u s, used as we are to describing the semantics of "I" by double indexing. But I think that Cresswell's position is consistent, because the uttermost index <w,t,p> may be regarded as the context of utterance . The general convention that all genuine indexicals have wide scope certain-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( 1 6)
1 87 ly makes it possible to describe the meaning of the deictic words correctly, because these can be defined with respect to the utterance context <w ,t ,p>. For instance, "you" denotes the person addressed by p at <w ,t,p>. [vide p . 1 27 ] . So it seems that this account can replace double indexing by the wide scope convention. A problem not noticed by Cresswell is that this account doesn't carry over to sentences like the following: (20)
a. You say that you are Cresswell. b . You people say that you are the Hallelujah Ramblers.
(2 1 )
a . >>>> b . >>>>
Examples like these are a bit troublesome for the theory, for this time we cannot say that "you" is simply a surface device indicating wide scope for a he* , because that would obviously give us unwanted readings. Further more , even if (2 1 a) and (4 1 b) are correct representations of the intended readings, these examples certainly don't speak very much in favour of Cresswell's analysis. The formalizations look rather artificial. Notice, by the way , that these sentences don't represent difficulties for a de re analysis, as I have argued in Stechow ( 1 984). To summarize , this chapter is a bit confusing and partially wrong. It can , however, be repaired along the lines indicated.
9. SEMANTIC PARADOXES
In logic, and still more in mathematics, the most severe objection against a theory is to show that it is inconsistent . In Cresswell's account , virtually every interesting meaning rule is inconsistent. And no attempt is made to remove the inconsistencies. This certainly is a rather unexpected attitude for a logician. So let me illustrate where the inconsistencies arise and let me show Cresswell's reaction. A first paradox is generated by the harmless looking meaning rule (2) of Section 7 according to which <x,says ,y> is true if x utters a sentence whose content is y. Unfortunately, this semantics is not innocent at all . Suppose that ( I ) is the only sentence Felicity utters, then ( I ) will be true at the occasion of utterance if it is false at the very same occasion :
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Obviously, the nontrivial interpretation of these involve an attitude de se and should therefore be analyzed as (2 1 a) and (2 1 b) respectively:
188 (I)
[== (5), p. 108 ] What Felicity says is false.
We are assuming that ( 1) is analyzed as (2)
[== (6), p . 109] <<what, >>)s false>
''This book is not concerned with the semantic paradoxes, and I have no more to say about what to do with cases like (6). I shall assume that V(says) is a function that gets as close as is consistently possible to what is desired ."[p . I 1 0 ]
Chapter 1 0, Iterated A ttitudes , points a t another difficulty of the structured meaning approach. The sentence : (3)
[==( I ), p. 8 5 ] Mortimer believes that Natasha believes that the earth is flat .
allows for one analysis, which has this logical form : (4)
> .
The problem is that the belief-operator contains a structure containing the belief-operator itself. A closer analysis of the theory will show that this operator has to be a function where one argument is the function itself, a set-theoretical impossibility. So (4) generates a paradox. There are several possible reactions to this . First , V(betieves) is a partial function which is not defined for a case like this. Furthermore , a bit of reflection shows that the reading corresponding to (4) is not a very likely one . Another reaction would be to introduce a hierarchy of 'believe' predicates, which might be necessary on independent grounds in view of examples like (5), which are semantically paradoxical. (5 )
[(I 9), p. 9 2 ] Syllabub believes that all of her beliefs are false .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
what has to be analysed as if it were the definite article, i .e., as making a description out of the A.-abstract, and is false assigns to any proposition its complement , i .e . this predicate simply is negation. Let us see why the meaning rule for says leads to a paradox. Suppose w is the occasion where Felicity says (2) but nothing else . Then V((2)) is true at w if the proposition p that Felicity says p at w is false at w . Now the case posited makes it clear that this p is in fact V((2)). Therefo re V((2)) is false at w . What this shows is that we can't have a meaning rule like (2) o f Section 7. Cresswell, however, doesn't try to repair the rule.
1 89 A third solution, viz to develop a theory where functions may be applied to themselves is not discussed by Cresswell, since he believes that paradoxes of this kind belong to the nature of semantics. It follows that a theory where such paradoxes simply don't arise misses something of the essence of semantics. And there are, of course , the well-known paradoxes generated by the truth-predicate. If we allow the truth-predicate to apply to expressions, then we can deduce a paradox along the well-known lines indicated by Tarski . But Cresswell argues that we don't have different truth-predicates, one applying to expressions and the other to meanings . Consider the following two examples: a. '5+7= 1 2' is true. b . That 5 +7 = 1 2 is true.
If the predicate "is true" we re different in (6a) and (6b), then the con straint mentioned in Section 5 that predicates of PAs are unambiguous would be violated . Cresswell gives some arguments to the effect that in both cases true looks at the parts of the embedded sense and gives us the proposition which is obtained if we apply the parts onto each other in the appropriate way . So in each case true will yield the necessary proposition for these examples, because we obtain it , if we apply e .g. = to (5+7), i .e . 1 2 , and 1 2, or if we first apply + to 5 and 7 , and then = to the result of that ap plication and 1 2 . So this semantics looks rather innocent , like what is known as the redundancy of truth. But the analysis is not harmless, be cause according to an idea of Thomason we can construct a sort of liar paradox by means of this special truth predicate , Godelization and so on. I refer the reader to p. 39 f. for some details. So this analysis leads to an inconsistency. This is how Cresswell comments on this result : "What are we to make of this? The q uestion is whether these problems lie in the nature of semantics itself or are the consequences of a n erroneous theory . M y opinion is that they lie in t h e nature of semantics itself." ( p . 40 I
So these paradoxes don't worry Cresswell too much. To be sure , he would not claim that a meaning rule leading to a contradiction is correct . In fact , there can't be such a meaning rule. The alleged meaning rule is an ideal (like e .g. the unrestricted principle of comprehension in set-theory) which the analyst has to approach as closely as possible . There are several ways of achieving this and Cresswell expresses no preference for any particular t reatment of semantic paradoxes.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(6)
1 90 One might think that this is a rather sloppy way of dealing with problems that have worried great thinkers for centuries. This would be a valuable objection if one could show that any solution of the problem of PA depends on a solution of the problem of semantic paradoxes. Cresswell thinks that the two problems can be treated independently from each o ther. If one admits this, then Cresswell's attitude is rational and , as I think, the only fruitful attitude . One would achieve nothing if one tried to solve all problems of semantics at once or, to use a slogan of Chomsky's : if a theory doesn't explain everything, then this does not imply that it explains nothing.
Has the problem of propositional attitudes been solved? An answer depends on what one considers to be a solution. It seems to me that the structured meaning account is basically correct. But even if this is so , there is more to be said about the problem of PA. Cresswell's book is in large parts very sketchy. I would , for instance , have liked to read a bit more about the formal semantics of verbs like calculate. The discussion offerend in Section 5 is extremely vague, and this is perhaps no accident. Despite what I said in that section, it is not at all clear to me what would be a satisfactory analysis of this verb. I certainly believe that the object of the attitude must be something like a structured meaning. But I feel that more needs to be said about the semantics of calculate and similar PA-predicates in order to generate the decisive 'aha'-effect. One of the merits of the book is that it offers a principled dis cussion of different kinds of solutions in a theory-neutral way. This alone makes the book worth reading. In this review , I haven't been able to discuss the Bibliographical Commentary, which contains a competent discussion of most of the relevant literature on the problem o f PAs. I dare say that nothing important has been left out. It is obvious from everything I have said that I believe this to be an important book. It treats the last great open question in truth conditional semantics and it suggests an answer to it. I believe it is a correct answer. On the other hand, many people certainly won't believe that the riddle has been solved now. Thus, the book represents a challenge to all those who believe in a different kind of solution , for instance those who look for finer-grained entities, those who believe only in internal representations and so on. In other words , the debate will go on. No doubt, other shortcomings like that in Chapter 14 will be found, and the book, therefore, represents a challenge also to those who basically agree with Cresswell.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
1 0 . HAS THE R I DDLE BEEN SOLVED?
191 The book is written in very simple language, but is not easy to understand . The reason presumably is the intricacy of the problem of PAs . So, if you are interested in the deepest problem of truth conditional semantics and in the study of meaning in general, I advise you to buy this book. Universitiit Konstanz Fachgruppe Sprach wissenschaft Postfach 5560 D· 7750 Konstanz
I would like to thank Ulli Haas-Spohn for critical comments on an earlier version of this review and Frans Plank for checking my English.
REFERENCES Cresswell, M.J , 197 3 : Logics and Languages. Methue n , London. Cresswell M.J. & A. von Stechow, 1 9 8 2 : De re belief generalized. Linguistics and Philosophy 5 :503-5 3 5 . Fodor, J.A. 1 9 8 1 : Representations. MIT Press, Cambridge, Mass. Lewis , D.K., 1 979 : Attitudes de dicta and de se. The Philosophical Review 8 8 : 5 1 3-43. Moore, R.C. and G.G. Hendrix, 1 98 2 : Computational models of belief and the se mantics of belief sentences. I n : Peters and Saarinen (eds), Processes, Beliefs and Questions. Reidel, Dordrecht. Thomason , R.H., 1 980 : A model theory for propositional attitudes. L inguistics and Philosophy 4: 47-70. Stechow, A . von , 1984: Structured Propositions and Essential Indexicals. I n Land man and Veltman (eds), Varieties of Formal Semantics. Foris, Dordrecht. .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
*ACKNOWLEDGEMENTS
1 93 BOOK REVIEWS Fred Landman and Frank Veltman (eds.) ,
Varieties of Formal Semantics. 1 982 Semantics 3). Faris Publication s ,
Proceedings of the Fourth Amsterdam Colloquium , September (
=
Groningen-Amsterdam Studies in
Dordrecht-Holland. Peter Rolf Lutzeier Gone are the days when one religiously consulted the works of Richard Montague, if one was concerned about semantic problems of natural is confirmed by this collection of
19
for a colloquium in Amsterdam in
papers which were mostly written
1 982.
However , just as Montague's
work is unthinkable without Alfred Tarski's previous work on model theory for formal languages all the papers here are still indebted in one way or another to Montague. This holds in the most explicit way for the papers by Emmon Bach , Ken-ichiro Shirai, Annabel Cormack, Theo M .V . Janssen, Godehard Link and Barbara H . Partee . Bach's syntactic paper 'Some Generalizations o f Categorial Grammars' i s concerned with the liberation from the idea of categorial grammar that all that matters is the semantic perspective of functor- and argument expressions. First , he suggests adding something like traditional mo rphosyntactic features which can be seen as giving partial specifications for interpretations of expressions
(5-6);
a fairly
uncontroversial addition, one may say. Secondly, for the formulation of syntactic rules he b asically makes use of sequences . This is shown to be useful for a number of problems in natural languages , especially for Dutch complement structures. Sequences, by the way , have played a major part in the syntax of integrational linguistics for a long time al ready (cp. Lieb
( 1 983)).
Bach's article is patchy in parts and could have
been improved by some rigorous editing. Shirai in 'Where locative expressions are located in Montague Gram mar' gives a syntactic and semantic treatment of Japanese phrases marked with the locative particles
ni
and
de
(their translations are mainly
in, on
for both of them). He accepts a Montague type framework, but on the semantic level he combines it freely with a decompositional approach. The decomposition of predicates into other (more basic?) predicates comes in handy , for instance, for explaining the difference of the
ni
and de-phrases with regard to their causative entailments : Ni-phrases in contrast to de-phrases seem to require predicates with a CAUSE-element
(368).
Shirai also motivates his claim that the two phrases ought to be
assigned to di fferent syntactic categories , especially by showing the un-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
language and wanted to apply some formal methods to them. Such a view
1 94
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
grammaticality of certain sentences (380). De-phrases are seen best as t raditional modifiers of intransitive verb-phrases whereas ni-phrases are seen best as extra arguments of certain transitive verb-phrases. Although the issue is not mentioned at all, the examples in connection with exis t ential entailments seem to suggest that one can do without an assignment to the sentence modifier category . VP anaphora is still a puzzling case , as one can gather from Cormack's paper 'VP anaphora : variables and scope'. We find a discussion of scope variation and deletion into relative clauses. The latter provides examples for the fact that we cannot simply assume a connection with the whole antecedent VP . In o rder to formulate her views on a number of interesting examples she rather cautiously t ries to do without an essential use of variables and lambda-expressions within intensional logic. I am sym pathetic to such an approach , but wonder whether one should not try once to discuss these cases without the assumption of a VP category. Semantic problems are the main targets of the following articles. In dividual concepts as functions from the set of contexts into the set of in dividuals are one of the easiest examples of intensions. Despite that they have not found widespread application . Janssen starts his defence of them in 'Individual concepts are useful' by looking at some of the criticisms made t owards Montague's use of them for the temperature paradox: if The temperature is ninety and The temperature rises, then Ninety rises. He cleverly elaborates on his theme by providing a good review of different examples ('normal' common nouns, concealed questions , pronouns), which can all be treated quite naturally by means of individual concepts. Unfortunately, readability is hampered at times by a large number of misprints. Term phrases like the boy and the girl who met yesterday found the interest of Link in 'Hydras. On the logic of relative constructions with multiple heads'. In his semantics of plurality he avoids working with sets of individuals. Instead of that he establishes an ontology consisting of ' individual sums', which are meant to be on the same level as individuals, and 'groups of individuals' (247). The phrasal conj unction and turns out to be at least two-ways ambiguous b etween the reference to individual sums and the reference to sums of groups (249). Although one has to accept the analyses given of the sentences discussed , I fmd the ambiguity claim for and a bit hard to swallow and would like to have first a more detailed look into the question of verb semantics for the verbs involved . The question of compositionality lies at the heart of any semantic theory which does not deny a relation between the syntactic build-up of a sentence and its meaning. Therefore it is not surprising that this issue is mentioned in quite a few of the articles in this collection. However, Partee dedicated her whole paper 'Compositionality' to it . She rightly
1 95
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
stresses the theory-dependence of any clear formulation of the principle and then more or less settles for a discussion of its versions within Mon tague's framework. There is a valuable d iscussion of a wide range of possible challenges ; the only obvious exclusions I can think of are the challenges caused by metaphors and quotation. The principle still comes out as a good working hypothesis (304), especially if we take seriously two messages conveyed by Partee's discussion : first , syntactic analysis has to include information about the relational (functional) structure of the construction . Again a point which has been stressed by integrational linguistics for quite a while (cp. Lieb ( 1983)). Secondly , more weight has to be put on the lexical level. What I cannot understand is the fact that we do not fmd anywhere in the paper a clear division between sentence meaning and utterance meaning. If such a division had been made , some of the challenges could have been dealt with in a clearer way (cp. my discussion of the principle in Lutzeier { 1985)). Applying model theory to natural language has shown convincingly that traditional propositions are too coarse a means for several problems there . So either some semantic refmement , some syntactization or even some pragmatization is called for. Papers with roughly such an aim are the ones by Fred Landman, John Perry , Arnim von Stechow, Henk Zeevat , Alice G .B. ter Meulen, Jeroen Groenendijk/Martin Stokhof, and David Lewis. Propositional attitudes are certainly one of the most obstinate problems in this area . Landman in 'Data semantics for attitude reports' wants to give an account for the differences between a sentence like John sees it rain and John sees that it rains. Landman mainly argues for a difference in awareness between the 'Naked infmitive' -sentence and the 'That-complement' sentence { 1 99). In his assumed framework of data semantics he does not start with propositions b ut with facts, which represent our direct in formation about the world (20 1). With regard to the examples above , the 'Naked infinitive'-sentence reports a visual relation between John and a set of facts, whereas the 'That-complement' -sentence reports a visual relation between John and a set of one isolated fact. This is how the awareness-difference can be accounted for. For more complex examples propositions which are constructions out of facts are involved . Such a basically constructive approach is undoubtedly interesting and ought to be encouraged, although one may have some doubts whether we can realistically ascribe to ourselves all these detailed construction capacities . In addition , Landman points out one of the weaknesses of situation semantics (cp. Barwise/Perry ( 1983)) : it cannot differentiate between contradictions. However, as Perry in 'Contradictory Situations' demon strates, this can easily be remedied : the interpretation of an utterance does not have to contain only coherent events (32 1 ). The paper also gives a short introduction into the basic ideas of situation semantics.
1 96
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Stechow in 'Structured propositions and essential indexicals' success fully replaces, in his relational analysis of attitudes, traditional propositions by structured propositions, which are sequences of n-place properties and n individuals. Indexicals as well as names are j ust referential for Stechow . This sh1;>Uld help to differentiate , for example , the two sentences This is Peter and Peter is Peter; a paradox which, of course , in the case of names, b rought Frege to the assumption of senses. Here the means of structured propositions provides the extra information needed : an indexical or name turns out to be essential if it denotes the individuals of the stmctured proposition involved (392). Zeevat in 'Belief advocates a theory of belief which stresses the strong connection between belief and action (405). This in itself makes interesting reading. He also carefully d istinguishes de re-terms from de dicta-terms (408), b ut I do not find his familiar claim that we need an intersubjective meaning representation in order to explain communication a convincing starting point (407, 4 1 5-4 1 6). For instance , reference can already ef fectively play the role of a common anchor despite more subjective meaning representations (cp. Lutzeier (1 985)). Natural language expressions force ter Meulen in 'Events, quantities and individuals' to have a closer look at the internal structure of se mantic entities. Divisibility plays a central part because she is concerned with differences between homogeneously referring expressions and hetero geneously referring expressions. I have my doubts about her criterion for heterogeneously described entities: she maintains that they remain the same entity when parts of them change (26 1 ). There must be limits to such a changing process. At most, this describes a way of gradual differ ence for homogeneously described entities. The traditional and notorious ly difficult linguistic issue of aktionsarten and aspect comes into the discussion too . A process structure is the basic framework for description . Closure conditions for subsets of processes are then defined , some of them requiring something on parts of processes . All this is based on good linguistic insights, especially because an inference test is used. Pragmatic issues are touched on by Groenendijk/Stokhofs paper 'On the semantics of questions and the pragmatics of answers'. It is a prag matic notion of information which they want to employ for answers, b ut it depends also on the semantics of the question posed whether something will count as a suitable answer. Such an analysis of the question answer relation is given within the possible worlds framework. The problem of indirect answers demands an understanding of the relations 'dependence' and 'conversational equivalence' between questions . The authors rightly point out that all this can only be achieved on the b asis of a richer theory of communication ( 1 65). This is a clearly written article from which a non-specialist reader can learn quite something.
1 97 Lewis' 'Individuation by acquaintance and by stipulation' makes a stimulating piece of philosophical reading. He again puts his counter part theory to the test of cross-identification over different worlds. For the counterparts by acquaintance the subject of attitudes obviously plays a central role ; so they really presuppose the cross-identification of the subject
(226).
For this he recommends stipulation , which in my eyes un
fortunately forces us into a godlike position. He then advocates a dis tinction between 'possibilities' and 'possible worlds' in order to be able to talk about less global possible things . Each possibility is related to exactly one world , but a world may provide several different possibilities
(234-235).
Unquestionably, we need such a distinction. However, I think
thing like this. Why , for instance, sho uld we not have a notion of possible worlds for which parts o f a world (in the naive sense) can count as possible worlds (in the theoretical sense)? (Compare Lutzeier
( 198 1 ) where possible
worlds are also populated ; so Lewis' examples of duplicate worlds can count as examples of different worlds.) Refinements of traditional approaches can also be found in the papers by Manfred Pinkal and Pieter A .M . Seuren. Pinkal's topic in 'Consistency and context change : the Sorites paradox' is semantic vagueness , exempli fied by the Sorites paradox which in one version deals with grains of sand which gradually make a heap . The paradoxical character arises out of the transition from a non-transitive similarity between individuals with respect to a property to transitive equivalences of sentences assigning the property to these individuals (326) . Pinkal chooses carefully an extension of classical logic whereby the unconditional application of the transition gets blocked . This is guaranteed by using some rather locally operating rules and a context changing mechanism which can create local areas of equivalences . Seuren in 'Logic and truth-values in language' picks up the classical issue of presupposition and treats it as a semantic phenomenon
(35 3) . He
argues for a system with truth-value gaps for atomic sentences and three truth·values . Parallel to that he distinguishes two negation s : a minimal one which protects presupposition and a radical one which does not protect presupposition . The diffe rence is especially nicely reflected by the behaviour of negative polarity items compared to the behaviour o f positive polarity items
(3 54-355).
I can only unde rline Seuren's general
intention of estab lishing a logic of language instead of using, for instance , a logic of mathematics. Finally, the papers by Jan van Eijck, Kit Fine , Renate Bartsch and Johan van Benthem have to be looked at . The non -specialist reader may fmd it a bit armoying that Van Eijck's paper 'Discourse representation , anaphora and scope' cannot be unde rstood without familiarity with Kamp
( 1 98 1 ).
Van Eijck's aim is to formulate some anaphora and scope
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the notion 'possible world' itself is flexible enough to incorporate some
1 98 rules as constraints on the important discourse-representation level estab lished by Kamp. In doing that he supports the sensible view that anaphoric linking goes beyond pure sentence syntax. Fine in 'A defense of arbitrary objects' takes us back to the practicalities of 'day-to-day'-mathematical reasoning, which despite all logical qualms makes heavy use of arbitrary objects, especially as 'variable' numbers. In a step-by-step approach he steers his readers cleverly around several objections against arbitrary objects. I can follow Fine in most of his arguments. Nevertheless I find
his defence against the objection o f logical incoherence the weakest one . We eventually move into lexical semantics with Bartsch's paper 'The meanings are seen there as polysemic complexes, i .e ., as something like sets of readings interconnecte d with each other . Yet there is an assumption of a kernel meaning
(26) , which one may fmd easier to accept diachronically
than synchronically. Realization of metaphorical use is fairly uncontro versially related to a feeling of discrepancy. Also, metaphor is based on associations of similarity and metonymy is based on associations of factual relatedness
(36).
Bartsch advocates a model of context-dependent inter
pretation for words, which is also supposed to explain the possibility of metaphorical and metonymical transfer. There is a fme and very helpful
illustration for all this by means of the verb run.
Van Benthem's paper 'The logic of semantics' has to be seen on a more metatheoretical level. By looking at issues like the ties of semantic analysis to syntactic structure , explaining inferences , quantifiers and dynamic interpretations he sees room for foundational studies on semantics. At the same time he reminds everyone involved in such endeavours not to turn a blind eye to empirical facts
(56 , 77).
A reminder
I
can wholeheartedly
endorse , especially in view of the demands for a linguistic semantics. The editors arranged the papers in alphabetical order . My grouping could only weave one possible set of threads between the papers and no doub t , the reader will think of other possible ones. He will also fmd a lot more interesting points in the individual papers which I did not have the space or patience to point out . Perhaps not surprising with regard to the topics chosen there is not much room for the non-specialist reader who j ust wants to inform himself about new developments in formal semantics. Obviously , the collection cannot cover .all trends in semantics. What I have missed most are refer ences to attempts at finding better mental representations or more holistic notions in cognitive psychology , computer science and linguistics. Also, lexical semantics is certainly unde rrepresented . Nevertheless , this is a collection of papers worth looking at.
Fachbereich 1 6 - Germanistik Freie Universitiit Berlin
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Structure of word meanings : polysemy , metaphor, metonymy' . Word
1 99 REFERENCES
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Barwise, Jon, and John Perry, 1 9 8 3 : Situations and A ttitudes. M IT-Press , Cambridge, Mass. Kamp, Hans, 1 98 1 : A theory of truth and semantic representation. In: J .A.G. Groe nendijk, T.M.V. Janssen, and M.B.J . Stokhof (eds.), Formal Methods in the Study of Language. Mathematisch Centrum, Amsterdam. Pp. 277-3 22 . Lieb, Hans-Heinrich, 1 98 3 : /ntegrational Linguistics. Vol. 1: General Outline. John Benjamins. Amsterdam. Lutzeier, Peter Rolf, 1 9 8 1 : Words and worlds. In: H .-J . Eikmeyer and H. Rieser (eds .) , Words, Worlds, and Contexts. New A pproaches in Word Semantics. De Gruyter, Berlin. Pp. 75-106. Lutzeier, Peter Rolf, 1 985 : Linguistische Semantik. Metzlersche Verlagsbuchhand lung, Stuttgart.
200 BOOK REVIEWS I. Engelkamp and H .D . Zimmer, Dynamic Aspects of Language Processing. Sp ringer Verlag, Berlin , 1983 . Pp. x + 102 . Price : $ 2 5 .00. Thomas Pechmann
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Psycholinguistic research concerned with syntax typically studies the syntactic complexity of particular sentences as predicted by particular linguistic theories on the one hand compared to the ease with which people process these sentences on the other. The more complex a sentence is in terms of syntax, the more time a person should need, for example, to understand it or to verify it against a picture. Much less attention has been paid to another question regarding the syntactic structure of sentences: what is the communicative meaning conveyed by particular syntactic forms as opposed to others? The book written by Engelkamp and Zimmer contributes to the fllling of this gap. Its central concern is to show how a small number of syntactic devices are used by speakers to transmit specific communicative meanings and what effect the use of these devices has on listeners. The syntactic devices under consideration are subjectivization , clefting, and article use . The authors discuss whether these devices are used to mark the focus of attention in sentences and to differentiate between given and new sentential info rmation. The book consists of two parts. The introduction is followed by five chapters in which the authors give an outline o f their general conception o f how knowledge i s represented and communicated . I n the remaining four chapters Engelkamp and Zimmer present 1 6 experiments investigating the above mentioned relationships between syntactic devices and communi· cative meanings. It is common to conceive of the representation of knowledge in terms of propositional structure consisting_ of a predicate and one or more arguments. On the one hand , Engelkamp and Zimmer fully subscribe to this model, which they characterize in a very concise , but perspicuous w ay. Also, the terminology of the whole book is b ased on this propositional model. On the other hand , shortcomings of the model are their main starting point . One problem with the model is that the same propositional structure is assigned to different sentence forms. Active and passive sentences, for example, to which the same propositional structure is assigned , are treated as equivalents by the model . By contrast, Engelkamp and Zimmer claim that different sentence forms to which the same proposi tional structure is assigned do differ with regard to their communicative meanings, and their main concern is to show in what respects this is the case .
201 I n Chapter
2,
some more details of the propositional model of knowl
edge representation are discussed . The authors particularly differentiate between factual knowledge which " . . . refers to entities of which it is assumed that they actually exist in the outside world as individual people , objects, or events"
(p. 1 1 )
and general knowledge which " . . . is assumed
to refer to classes of potential referents" (ibid.). In the next chapter, the distinction between given and new knowledge is taken up. By further diffe rentiating between knowledge and
relationally
conceptually
new/given
new/given knowledge , the authors link the
new/given distinction with the propositional model. Conceptual knowl edge refers to knowledge of individual entities which are called argu lational knowledge refers to knowledge as to whether particular argu ments participate in particular predicates. In the same chapter Engelkamp and Zimmer introduce the notions o f
activated, deactivated,
and
focussed
knowledge . Activated knowledge is
knowledge which is presently being processed by a person . If it is activated for the first time , it is new information . If it has already been part of long-term memory , it is given information which is now being reactivated . Thus, " . . . given knowledge can be present in an activated or a deactivated state , whereas new knowledge is activated for the first time and can only be present in activated form"
(p . 20).
In linguistic communication , the
predicates and argume nts which establish the propositional structure of an utterance are activated knowledge . This activation of knowledge is conceived of as the directing of attention . The authors further assume that an activated proposition consists of a nucleus which represents a maximum of activation . This nucleus is also called the focus of attention . The information which is in the focus of attention is called focussed info rmation . Finally Engelkamp and Zimmer introduce the notion o f thematized knowledge , which " . . . i s unde rstood t o b e the part of our knowledge structure to which we add new knowledge for storage" (p .
25).
Thematized knowledge must be activated , since something i s predicated t o i t . I t i s , furthermore, assumed t o represent a focus of attention . However , Engelkamp and Zimmer claim that thematized knowledge is not necessarily given information although it is rather likely to be . They stress that new information can also be the theme of an utte rance to which something is predicated . If linguistic information affects the cognitive knowledge structure and if the processing of visual information depends on this knowledge structure then linguistic info rmation should a ffect the processing of visual in formation. This is the issue which is taken up in the fourth chapter. Engelkamp and Zimmer argue more specifically that linguistic information can be used to focus the attention of an interlocutor on particular visual
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
ments when they are part of a propositional structure . By contrast , re
202 information. They p redict that focussed visual information should be subject to longer eye fixation times. This prediction is later tested in a number of experiments. In Chapter 5 the authors outline a general model of communication based on their assumption that verbal knowledge is represented by pro positional structures. They claim that " . . . the hearer must locate in his knowledge structure those individual concepts to which something is predicated, and integrate the predicates into his conceptual knowledge"
(p.
33) . In addition , the hearer must check which concepts are new and
which are given and also which are thematized and in the focus of atten tion . To accomplish this task, the listener is helped by the speaker, who which is given . Now Engelkamp and Zimmer have set the stage for presenting a series of experiments investigating whether speakers use particular syntactic stylistic devices to transmit particular communicative meanings to the listener and whether the listener makes use of these devices. I will restrict myself to a brief summary of some of the most important results. The authors argue that " . . . the relative clause o f a cleft construction contains the relationally given information" (p . 4 1 ). Since relationally given information must also be conceptually given , " . . . the concept used in the relative clause of a cle ft sentence should as a rule be marked by the definite article" (p . 4 1 ) . The first two experiments show this to be true . Experiment 3 studies the influence of perceptual features on thema tization and subjectivization . It shows that small objects are perceived as figures against a b ackground . Therefore , they are likely to be chosen as the theme of an utterance and , consequently, to be subjectivized . The fourth experiment demonstrates that in active and passive sen tences which include two concept s , it is the first concept which is usually thematized. This concept is usually given information and therefore marked by the defmite article. Experiment 5 provides evidence for the general claim that passivization is primarily used to thematize a known concept , but not to communicate a particular given-new structure. ·
Experiment
6
investigates why speakers use cleft constructions. It is
argued that the main reason for using this syntactic device is to fill and focus an already existing info rmation gap on the listener's part . Experiments
7
through
9
show that cleft sentences have an attention
focussing effect . Concerning the listener's scanning of visual representa tio ns it can be shown that a concept is observed longer when it is placed in the focus of attention by clefting than when it is not . Experiments 10 through 1 5 study whether speakers avoid constructing sentences with two foci of attention. In sentences including two concepts
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
uses particular cues to indicate, for example , which information is new and
203
The book is less than 1 00 pages long. It is very concisely written , omitting any lengthy consideration of details of minor importance. The argu mentation throughout the volume is straightforward. The points are made clearly. The experiments reported in the second part are theoretically well prepared for in the first part of the book. Of course , the brevity of the book, while enhancing its readability , has its costs. The authors present their points of view on various topics without considering any alternatives. One point I would particularly like to raise relates to the syntactic devices which are studied in detail by Engelkamp and Zimmer. They provide evidence that article use , sub jectivization and clefting are used by speakers to indicate which information the speaker wants the listener to focus his attention upon . Their ex periments demonstrate that the devices under consideration do fulfil these functions. However, one might wonder why the authors do not take into account the most versatile linguistic device which speakers use both to differentiate between given and new information and to focus the listener's attention on some part of the verbal message : prosodic accent. As has now been shown in a number of experiments, accentuation is effective with respect to both functions. It is not unlikely that accentuation could also moderate or overrun the effectiveness of clefting and sub jectivization in the functions discussed by Engelkamp and Zimmer. There fore , although the authors are more concerned with written than with spoken language, it is surprising that they do not even mention that prosodic devices are highly relevant for the communicative functions they consider in their b ook. But again , given that Engelkamp and Zimmer make their points straight forwardly and clearly, this is a very interesting book for everyone who is interested in the relationship between syntactic structure and its com municative meaning. Max-Plonck .fnstitut fur Psycho/inguistik Nijmegen
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
one concept can be focussed by means of subjectivization, the other by clefting. The experiments show that speakers do indeed avoid two foci of attention in sentences consisting of two concepts. Furthermore, they demonstrate that the focus of attention in the listener is stronger when subjectivization and clefting work together than when they work against each other. The experiments also show that both clefting and subjectiviza tion h ave an attention-focussing effect, but that subjectivization clearly has less effect on the hearer's attention-focussing than clefting. Finally, the last experiment demonstrates that listeners use a clefted concept of a sentence they hear as the subject of their own next sentence. This can be taken as further evidence that clefting focusses the listener's attention.
204 BOOK REVIEWS
A .J . Sanford and S .C. Garrod , Understanding Written Language: Compre hension Beyond the Sentence. J . Wiley & Sons , Chichester, New York , Brisbane , Toront o , 198 1 . 32 figures , 224 pages , ISBN 047 1 278424. Walter
J . Pe rrig and
Hubert
D. Zimmer
The book deals with the problem of how readers come to understand written tex t . The book is divided into two sections. In Part I the authors provide a b ackground review and an introduction to a number of issues
meaning, semantic
components and primitives, procedural semantics ,
representation of verbs, events and full discourse are discussed , as well as the implications of structures and processes of semantic networks , pro positional hierarchies, story grammars and schema-based theories of com prehension and memory. The language and presentation of this first part of the book is clear and straightforward , and the well thought-out dis cussions are restricted to general problems. In summary , Part I is a fme introduction for students or everyone interested in text or language comprehension and memory . Part I I o f the book is in this respect completely different from Part I . I t requires , and , as we will point out , also deserves t o b e read thoroughly by scientists working in psycholinguistics or language processing. The b ook j acket promises an analysis and synthesis of ideas from linguistics , p sychology and artificial intelligence on the question o f how readers understand written text. By the second part of the book at the lates t , it i s obvious that the scientific contribution this book makes i s based o n a classical psychological approach. I n fact, almost a l l of the analyses p resented are b ased on explicitly described experimental studies. Sanford
&
Garrod concentrate on reference resolution . They believe that
if one can understand how refe rence is resolved , then an understanding of other parts of the general comprehension process will follow auto matically because
reference resolution constitutes a cornerstone for
building an appropriate mental model of what is being said . The investi gations presented include anaphoric reference, reference to things previously implied and reference to things explicitly mentioned in the prior context . In a series of ingenious experiments, the authors demonstrate pro cedures which monitor on-line the moment-to-moment processing decisions made by human
readers. Monitoring methods are response time and
reading tim e . Sanfo rd
&
Garrod advocate a theoretical position in which
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
which are of central importance to the question of language compre hension , and at the same time, they lay the foundation for Part I I . Word
20 5 a linguistic input addresses knowledge directly. Their "scenario account" describes a constructive concept- driven approach . Processing o f every new input and construction of cohesion is influenced by a previously retrieved and/or constructed domain o f re ference or situational knowledge. This approach is opposed to an interpretative data-driven approach where reference resolution is mainly d riven by cues from the surface structure of the text (e .g. Clark
& Haviland , 1 977).
A processing framework is proposed which identifies operationally distinct components of memory. A dynamic part of memory is distin guished from a static part . The dynamic part contains an explicit focus and an implicit focus representing different partitions of working memory. in the text , and is of limited capacity. In the implicit focus there is a currently selected subset of general knowledge which corresponds to the current scenario. The dynamic memory is characterized by continuously changing content as the text unfolds . By contrast, the static memory contains a long-term memo ry partition for the actual discourse described as a specific subset of episodic memory and a long-term semantic memory partition or knowledge base . Reference resolution is achieved by retrieval and construction pro cesses . As long as these processes act upon the limited reference or search domain in "focus", resolving reference is a "primary process". If this primary process fails, the "secondary process' continues its search in the static domain of memory . Predictions of reading and response time are based on this processing framework. The proposed theoretical framework includes a number of ad ho c assumptions and is far from achieving the status of a complete mode l . Nevertheless, i t i s a very attractive hypothetical construct , capable o f providing plausible explanations for a series of phenomena i n crucial comprehension problems. The e ffects of topicalization and emphasi s , foregrounding, t h e amount of intervening material and t h e centrality of scenario characters are predicted by the framework on the b asis o f work space allocation in explicit focus. The final chapter of the book is dis cursive and relates a number of different psychological considerations to each other, such as language production in conversation and discourse , or the relation o f comprehension and memory to the theoretical framework presented . Of course, the reader of this book will find some problems, pitfalls or limitations in the account presented . Some of them can be attributed to somewhat vaguely specified entities in terminology and theory , leading sometimes to inconsistent inte rpretations of different b ut comparable results. For instance , it is not quite clear how one can distinguish between
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The explicit focus contains representations of entities and events mentioned
206
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
specificity or complexity of verbs (die/murder) and nouns (vehicle/tank) and their effects on processing time . For verbs , Sanford & Garrod (p. 1 07) argue : " . . . if verbs are thought of as simply calling up their schemata when they are enco untered , there is no reason to suppose that the com plexity of the schemata will influence retrieval time ." On page 1 20 , fmding noun-specificity-related time differences, they argue : " . . . a general term cannot serve to retrieve episode structure of any detail from longterm memory and therefore is in a sense less well understood." The rationale b ehind this different interpretation is not clear. Moreover, the relationship between the terminology of the partitions of memory in the processing framework and existing distinctions made in p sychology is often not explicit enough. For example , the partition of ''implicit" and "explicit focus" within working memory remains obscure , particularly if it is considered that only the explicit focus should be of limited capacity . Another terminological problem remains with respect to the definition of given and new information . It may not be enough to introduce given information as "a certain amount of action knowledge (that) will be represented" (p . 1 1 1 ). For an adequate modelling of reference resolution and language processing, a more precise distinction into con ceptual and relational given and new information is necessary (cf. Engel k amp & Zimmer 1983). Also , Sanford & Garrod's use of tokens in "explicit focus" and scenario , where n o tokens seem to exist for explicitly given verbs that are already implicitly mentioned , is somehow different from the general use of the type-token distinctions, where a token (verb or noun) simply specifies an entity in actual use (e .g. Norman & Rumelhart 1975). Other problems are of a more general kind . One of them is the question : What are the units of representation or of the cognitive processes? For example , a sentence like "three Englishmen , an Irishman, and a Scotsman were having a discussion" should lead to "three tokens for Englishman" (p. 1 65 ) . It is not obvious why there could not be only one token repre senting 'the group' of three Englishmen . Under what conditions are the different units constructed? Furthermore , the primary process can only use "simple noun phrases" but the secondary process can use "information from an entire sentence" (p. 1 69). Does it follow from this that refe rence resolution for "the old man" or "the man who won the prize" is not possible within the fast primary process, but has to wait for the slow secondary process? Such questions and related ones arise if the proposed framework is used to analyse text processing. These problems might be identified as arising from the gap between theory and empirical data but they might b e partially based in the framework itself. To sum up, at different points the reader will miss further specifications of processes and assumptions or supporting empirical results for some
207 speculations. Here he might either wait for Sanford & Garrod to put forward their work or he will try to do it himself. In any case , there is no lack of explicit or implicit theoretically interesting problems. In this sense , Sanford & Garrod should not be criticized for these problems and limitatio.ns of their account ; these are, rather , evidence of the value and scope of their contribution. Such problems also motivate the reader to conduct further research into language processing, and they sharpen the eye for the problems of frameworks for handling these processes. There fore we highly recommend the book to anybody interested in the psycho logical processes of text and language comprehension .
R E F E RE NC E S
1 9 7 7 : ComprehensiOn a n d t h e given-new contract . Dtscourse Processes: A dvances in Research and Theory. 1 -40.
Clark, H . H . a n d S . E . Haviland, In :
R.O.
Freedle (Ed.),
Ablex , Norwood . Pp. Engelkam p ,
J.
and H . D . Z1mmer,
1 983 : Dynamic A spects of Language Processing.
Springer, Berhn. Norman, D.A . and D . E . R umelhar t , San Francisco.
1 975 : Explorations in Cogn,ition.
Freema n ,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
University of the Saarland Saarbnicken, West Germany