SEQUENCE OF TENSES IN ENGLISH

RENAAT DECLERCK

1. INTRODUCTION

Comrie (1986) raises the question of why English speakers use the preterite in the complement clauses of examples like (1) Arthur said that he was sick. The fact that we use was in the subclause, and not is (as in the Russian equivalent of (1)), is traditionally traced back to the so-called 'Sequence of Tenses' (SoT) rule, which is claimed to be operative in indirect speech in English. It is difficult, however, to determine the exact nature of this rule. Is it a formal (syntactic) rule, which is applied mechanically, or is it a rule that is semantically motivated? And if it is semantically motivated, which is the relevant semantic factor? As far as I know, the linguistic literature shows three different hypotheses to answer these questions. The first is that the subclause of (1) is in the past tense because it refers to past time. Comrie (1986) calls this the 'absolute deixis hypothesis', because it treats was as an absolute tense form, i.e. as a tense which relates a situation (i.e. event, state, etc.) directly to the moment of speaking. The second hypothesis is that was is in the past tense because it expresses that the subclause situation is simultaneous with the head clause situation, which lies in the past. Let us call this hypothesis the "relative time hypothesis'. The third hypothesis is that indirect speech in English is subject to a formal SoT rule which automatically backshifts the tense forms from direct speech when the introductory verb is in the past tense. Comrie (1986) aims to prove that only the third hypothesis is consistent with the facts of English.

Folia Linguistica XXIV/3-4 0165^004/90/24-513 $ 2. (C) Mouton/de Gruyter, Berlin - Societas Linguistica Europaea

514 My aim in this paper is to show that Comrie's attempt fails. I will argue that the 'formal SoT hypothesis' does not always make the correct predictions, and that the tense of a complement clause in indirect speech may be either a relative tense (as in (1)) or an absolute tense.
2. A THEORY OF TENSE

It is not possible to argue for or against one of the three hypotheses without incorporating them into a coherent theory of tense. Such a theory is offered in Declerck (forthcoming). Since expounding a full theory of tense is obviously a major undertaking, only the bare essentials can be pointed out here. However, I think that the following brief sketch should be sufficient to render it possible to evaluate the alternative answers to the question raised in section 1. English speakers view a situation as either past or non-past with respect to the moment of speech. That is, they locate a situation either in the 'past time-sphere' or in the 'present timesphere' The past time-sphere lies wholly before the moment of speech (i.e. it does not include the time of utterance). To locate a situation in it the speaker uses the preterite. The present timesphere is divided into three 'sectors' by the moment of speech: the portion of the present time-sphere that precedes the moment of speech is the 'pre-present sector'; the portion that is centered around the moment of speech is the 'present sector'; and the portion that follows the moment of speech is the 'post-present sector' The tenses used to locate situations in these three sectors are the present perfect, the present tense and the future tense, respectively. For ease of reference we can use the term 'absolute sectors' to refer to the above three sectors plus the pa.st time-sphere. (I call these timespans 'absolute' because they are defined in direct relation to the the moment of speech.) The four tenses that are used to locate situations in one of these four sectors can therefore be called 'absolute tenses' When two situations are located within the same time-sphere, there are two possibilities: either both of them are represented as related to the time of speech, or one situation is related to the time

515 of speech while the second is related to the first. To capture these possibilities we can introduce the notion of 'temporal domain' A (temporal) domain is a time interval taken up either by one situation or by a number of situations that are temporally related to each other by means of special tense forms. Consider, for example: (2) John said that he had worked hard all day, that he was tired and that he would go to bed early. The first clause contains an absolute tense form (said) which locates a situation in the past time-sphere and by doing so creates a past time-sphere domain (or 'past domain' for short). The second clause contains a relative tense form (had worked), which relates the situation of working to the situation of the first clause, and not to the moment of speech. That is, the second clause does not establish a domain (as the first clause does) but incorporates its situation into the already existing domain. The third clause does exactly the same thing, only the relation is now different: whereas had worked represents its situation as anterior to the 'central situation' of the domain (said), was tired represents its situation as simultaneous with it. The fourth clause, finally, again incorporates its situation into the domain, but the situation in question is now represented as posterior to the central situation. When a situation is introduced into a domain, it need not always be related to the central situation. The 'binding situation' may also be another situation (which is itself directly or indirectly related to (i.e. bound by) the central situation). For example, in (3), where the three situations referred to are again located within a single domain, the third clause represents its situation as simultaneous with the situation of the second clause, while the latter is respresented as anterior to the central situation referred to in the first clause: (3) John said he had felt very tired when he was working. It should be noticed that in (2) and (3) the relative tense form used to express siniultaneity is each time the past tense, irrespective of whether the binding situation is or is not the central situation of the domain. This means that there is a single system of relative tenses to represent domain-internal relations

516 in the past time-sphere: we always use the past tense for simiiltaneity, the past perfect for anteriority, and the conditional tense (trouW+present infinitive) for posteriority. It appears, then, that the past tense can be used in two ways: either as an absolute tense (establishing a past domain) or as a relative tense (expressing the domain-internal relation of simultaneity). The past perfect and conditional, on the other hand, can only be used as relative tenses. As noted before, when two situations are located within the past time-sphere, the second may or may not be incorporated into the domain established by the first situation. If there is no incorporation, the second situation establishes a new domain. In that case we can speak of a "shift of domain' For example, in John went to the door, opened it and left the reference is to three consecutive past domains. A shift of domain is also to be noticed in (4), which can be compared with (5): (4) The boy ran away from home. He never came back. (5) The boy ran away from home. He would never come back. There is a shift of domain in (4): both clauses establish a domain of their own, using the preterite as absolute tense. In (5), in contrast, the first clause (using an absolute past tense) establishes the domain and the second clause (using a relative tense) incorporates a new situation into it. It should be noted that when there is a shift of domain in the past time-sphere, no temporal relation between the two domains is expressed by the tenses. Absolute past tense forms relate their situations to the moment of speech, not to the time of some other situation. The hearer interpreting sentences like (4) will therefore have to deduce the temporal order of the situations from the time adverbials (if any), from contextual information, or from pragmatic considerations. In (4), where there is no definite time adverbial or context, the hearer will deduce the order of the situations from his pragmatic knowledge: he knows that one can only come back after one has left. If the pragmatics of the situation cannot help him either, the hearer can still rely on the fact that, by convention, we preferably report situations in the order in which they have happened (cf. Dinsmore 1982:218). However, this is far from being an absolute principle.

517 In John stood by the window and Mary sat on the sofa, the most plausible interpretation is probably that the two situations were simultaneous. Why this should be so is a question we need not go into here. For our present purpose the only important thing is that it should be clear that situations which establish their own domains are not temporally related to each other by the tense system (since tenses that establish domains are absolute tenses, i.e. tenses that relate situations directly to the moment of speech). Examples like (4) illustrate the possibility of shifting the domain within the same time-sphere. It is clear, however, that a shift of domain may also be a shift from one time-sphere to the other. Thus in / am sad because my dog died the head clause establishes a present domain and the subclause a past one. In He left because his mother-in-law is in town today it is just the other way round. Apart from this possibility of switching from one absolute sector to another, there is also the possibility of 'shifting the temporal perspective', A shift of temporal perspective is a type of shift which may occur, not when a new domain is established, but when a domain-internal relation is expressed. It means that the time to which the situation is related is treated as if it belonged to an absolute sector that is different from the one to which it actually belongs, so that the relative tense used to relate another situation to it is a tense which is actually characteristic of another sector. The following sentences illustrate the phenomenon: (6) Has the woman ever told you that she loved you? (7) Has the woman ever told you that she had been beaten by her husband? (8) I have never promised that I would help you. In each of these examples the head clause establishes a prepresent domain. In each case the subclause is temporally bound by the head clause. However, the relative tenses used for this are those typical of past domains. This means that the pre-present domain behaves as if it were a past one. In other words, once the domain is established there is a shift of perspective from the pre-present to the past.

518 There is a similar shift of perspective in post-present domains. The central situation of a post-present domain behaves as if it were a present situation (i.e. as if it coincided with the moment of speech) when other situations are related to it. This means that in order to relate situations to this post-present time we use the same tenses as we use to relate situations to the moment of speech. That is, the set of relative tenses used in a postpresent domain coincides with the four absolute tenses. This is clear from examples like the following: (9) (said when planning someone's murder:) The police will think that he was killed by accident. (10) His excuse next time will be that he has been ill all week. (11) He will tell you that he does not know anything. (12) Bill will not believe that you will help him. In each of these examples the head clause establishes a postpresent domain and the subclause situation is incorporated into it. Since the speaker treats the time of the central situation as if it were the moment of speech, he uses a 'pseudo-absolute' tense form in the subclause. That is, the tenses used to express relations in a post-present domain are the preterite or present perfect for anteriority, the present tense for simultaneity and the future tense for posteriority.
3. TENSE IN INDIRECT SPEECH

The principles outlined above constitute only a portion of what is needed for a full theory of tense in English,^ but they suffice to describe and explain the use of tenses in indirect speech in English. As far as I can see, the principles governing this use can be summarized as follows:
For example, nothing has been said about the status of such tenses as the future perfect and the conditional perfect. In Declerck {forthcoming) it is argued that the future perfect combines the functions of absolute tense (since it establishes a domain) and relative tense (because it also expresses the domaininternal relation of anteriority), whereas the conditional perfect expresses two domain-internal relations at once. However, for our present purpose it is not necessary to take these tenses into account.

519 Principle A: In indirect speech the complement clause can in principle use either absolute or relative tense. That is, the complement clause situation can either shift the domain (i.e. establish a new domain) or be incorporated into the domain referred to in the head clause. Principle B: If both clauses refer to the same time-sphere, the use of relative tense in the complement clause is the unmarked choice. This means that, in such sentences, relative tense is always possible, whereas Grice's (1975) maxims of conversation allow the use of absolute tense only if the temporal order of the situations (which is not expressed by the absolute tense forms) is clear from a temporal adverb, the context or from the hearer's pragmatic knowledge of the world. These two principles account for the following array of possibilities: a. They predict the grammaticality of canonical examples of the application of SoT, such as John said that he was ill. Bill believed that he had seen a ghost, Betty threatened that she would leave him. In each of these sentences the subclause uses relative tense to incorporate the situation into the domain established by the head clause. Since the domain is a past domain, the relative tense forms in question belong to the system of relative tenses typical of the past time-sphere. b. The principles predict the grammaticality of sentences like John said that Bill was in London the day before. In such sentences the subclause shifts the domain instead of incorporating its situation into the head clause domain. The tense used in the complement clause is therefore the preterite (which is the only absolute tense that can establish a past time-sphere domain). Because there is no tense form expressing a domain-internal relation, the temporal order of the situations is expressed only by the time adverbial (the day before). If we omit this adverbial (John said that Bill was in London), the idea that one situation is anterior to the other is no longer expressed.

520

c. Since the expression of temporal relations in a pre-present domain may involve a shift of temporal perspective to the past (cf. above),^ principle A correctly predicts that a head clause in the present perfect can be followed by a past time-sphere tense in the complement clause (as in He has always admitted that he had made a mistake). d. Principles A and B predict the grammaticality of examples like (13)-(14), because they provide for the possibility of using an absolute tense in a complement clause. (13) John said that he will come. (14) John said that New York is an interesting city. Moreover, principle B states that these sentences realize a marked possibility (the unmarked possibility being the use of past time-sphere tenses in the that-clauses, as in (15)-(16)). This accords with the fact that there are restrictions on the use of sentences like (13)-(14), whereas there are no restrictions on the use of (15)-(16). (15) John said that he would come. (16) John said that New York was an interesting city. The restriction on the use of (13)-(14) is that these sentences can only be used if, in the opinion of the (reporting) speaker, the subclause situations are still valid at the moment of speech. There is no such condition for the use of (15)-(16). e. Our theory correctly predicts that 'backshifting' of tenses will not always take place after a reporting clause in the past tense. 'Backshifted' tense forms are by definition relative tense forms. There is therefore no 'backshifting' if the t/ia<-clause shifts the domain. Moreover, even if the that-clai\ise is temporally subordinated, there is no 'backshifting' of tense forms that express a relation in a past domain. Consider:

The shift of perspective from the pre-present to the past takes place only if the situation referred to by the present perfect is over at the moment of speech. If the situation continues into the present, the subclause situation must be directly related to the moment of speech, as in / have known for a long time that BUI i unreliable.

521 (17) I realized that I was late and that I would have to hurry. (18) Bill said that he had realized that he was late and that he would have to hurry. The verb of the head clause of (17) is 'backshifted' in (18) because the situation is now represented as anterior to Bill said instead of as anterior to the moment of speech. The other tense forms, however, are not 'backshifted' because the temporal relations which they express are not affected by this change. Both in (17) and in (18) was late expresses simultaneity and would have to hurry expresses posteriority. (Remember that, in a past domain, we use the preterit for simultaneity and the conditional for posteriority irrespective of whether the binding situation is or is not the central situation of the domain.) The conclusion from this section is that principles A and B make correct predictions in connection with the different cases of indirect speech that present themselves. According to these principles, the tenses in indirect speech complement clauses are usually relative tenses, but absolute tenses can also appear, subject to certain conditions. This means that what I propose is basically a 'relative time hypothesis' supplemented with some elements from the 'absolute deixis hypothesis',
4. COMRIE'S ARGUMENTS

The above conclusions run counter to Comrie (1986), where it is claimed that the formal SoT hypothesis is the only of the three hypotheses that is in keeping with the data of English. It is therefore necessary that we should go into Comrie s arguments and show in what way they are deficient. Before doing so, however, I would like to point out that there is one point on which I agree with Comrie (1986), viz. where he claims that the absolute deixis hypothesis cannot (by itself) account for the use of tenses in English indirect speech. Comrie argues this point effectively by showing that the absolute deixis hypothesis (which holds that indirect speech complement clauses make use of absolute tense forms) makes the wrong predictions in each of the following cases:

522 a. The absolute deixis hypothesis predicts that the indirect speech version of (19) is (20), and not (21): (19) Diana will say, 'I am dancing.' (20) Diana will say that she will be dancing. (21) Diana will say that she is dancing. Comrie's formal SoT hypothesis (according to which the tense of the original utterance is retained in the that-c\a,xise if the tense of the verb of reporting is non-past) accounts for the fact that the correct reported version of (19) is not (20) but (21). The hypothesis I have been arguing also accounts for this fact, because it states that the that-c\a,\ise in (21) represents the situation of dancing as simultaneous with the head clause situation. As we have seen, the relative tense used to express simultaneity in a post-present domain is the present tense. b. The second argument concerns sentences like the following: (22) In 2010, Ebenezer will say, 'I got tenure in 2000.' (23) In 2010, Ebenezer will say that he got tenure in 2000. (24) *In 2010, Ebenezer will say that he will get tenure in 2000. Since the year 2000 is future with respect to the moment of speech, the absolute deixis hypothesis predicts that Ebenezer's statement should be reported in the form of (24). However, the correct indirect version of (22) is not (24) but (23). Comrie's formal SoT hypothesis is in keeping with this, as it states that a past tense should be retained after a verb of reporting in a non-past tense. Our own hypothesis also accounts for the data, because, as we have seen, it provides for the possibility that the expression of temporal relations in a domain established by a future tense involves a shift of temporal perspective to the past time-sphere. That is, the past tense is used to represent a situation as being past relative to the central situation of the future domain. c. Consider: (25) Tomorrow, FYances will say, 'I was absent yesterday.' (26) Tomorrow, Frances will say that she was absent today. (27) *Tomorrow, Frances will say that she is absent today.

523 The absolute deixis hypothesis wrongly predicts that (26) should be ungrammatical (since an absolute past tense cannot refer to present time) and that (27) should be the correct indirect speech version of (25). The formal SoT hypothesis, on the other hand, correctly predicts the grammaticality of (26) and the ungrammaticality of (27). Our own hypothesis does so too, because it provides for the use of the past tense to represent a situation as past with respect to the central situation of a future domain. This is the case in (26), where today indicates a time that is past with respect to the time indicated by tomorrow. It should be noted, however, that sentences like (27) are not always ungrammatical. The use of the present tense in the subclause is occasionally possible, witness the following examples: (28) One day John will regret that he is treating me like this. (Dowty 1982:50) (29) They will report tomorrow that Harry is transmitting. (Ejerhed Braroe 1974:47) In these examples, the </ia<-clauses can be interpreted as referring to present (rather than future) time. In that case they are the indirect representations of a thought or utterance which, in direct speech, would not be phrased in the present tense, but rather in the preterite. That is, (29) (on the relevant interpretation) is the indirect version of (30), not of (31): (30) They will report tomorrow: 'Henry was transmitting (yesterday).' (31) They will report tomorrow: 'Henry is transmitting. Similarly, the thought which is reported in (28) and which is the cause of the subject's regret is 'I treated (or: was treating) her badly', and not 'I am treating her badly Sentences like (28)-(29) are subject to very severe restrictions: the substitution of the present tense for the preterite is only exceptionally possible. Even so, there are examples (such as (28)-(29)) that are grammatical, and this grammaticality should be accounted for by an adequate theory of indirect speech. And now we see that Comrie's formal SoT hypothesis is no longer up to the mark: since it holds that "the tense of the original utterance is retained" "if the tense of the verb of reporting is non-past"

524 (Comrie 1986:279), it cannot account for the fact that a thought or utterance in the past tense becomes a present tense statement in indirect speech. The theory I have been arguing, however, does provide for the grammaticality of (28)-(29). It holds that the tense in an indirect speech complement clause may be either a relative tense (expressing a domain-internal relation) or an absolute tense (shifting the domain), and that the latter possibility is marked, and therefore subject to restrictions. This theory predicts that the normal (unmarked) reported version of (30) is They will report tomorrow that Henry was transmitting today (in which was transmitting is a relative tense),^ but it also provides for the possibility that the that-c\a.\ise shifts the domain from the future to the present by using the present tense (as absolute tense). Moreover, it correctly predicts that this shift of domain is subject to restrictions, since it is the marked possibility. In sum, this is a clear example of a ca.se in which our hypothesis proves to be superior to Comrie's formal SoT analysis. d. The fourth argument advanced by Comrie (1986) against the absolute deixis hypothesis concerns sentences like the following: (32) In 1970, Graham said, 'I will get tenure in 1980.' (33) In 1970, Graham said that he would get tenure in 1980. (34) *In 1970, Graham said that he got tenure in 1980. The absolute deixis hypothesis incorrectly predicts that (34) should be grammatical (since the subclause refers to a time that is past with respect to the moment of speech). The formal SoT analysis rules out (34), as it states that will must be backshifted to would after a verb of reporting in the past tense. Our own hypothesis also predicts the grammaticality of (33), as it states that the tense in a complement clause is normally a relative tense, and the relative tense needed here (to express posteriority in a past domain) is, indeed, the conditional. (Note that, if (34) had been grammatical, we would have accounted for that in terms
As noted in section 2, the tenses used to relate situations to the central situation of a post-present domain are the same as the (absolute) tenses we use to relate situations directly to the moment of speech. This explains that we can use the preterite (was transmitting) to express a situation as anterior to the central situation (will report) of a post-present domain.

525 of a shift of domain. However, we have seen that shifting the domain is a marked possibility, which is subject to (often severe) restrictions, and there is such a restriction forbidding the shift of domain in sentences like (34). The restriction in question is that what is uttered as a prediction cannot be reported as a past fact. Such a report would not do justice to the contents of the original utterance.) e. The final piece of evidence adduced by Comrie against the absolute deixis hypothesis concerns examples like the following: (35) Yesterday, Henrietta said, 'I will be absent tomorrow' (36) Yesterday, Henrietta said that she would be absent today. (37) *Yesterday, Henrietta said that she is absent today. The absolute deixis hypothesis incorrectly predicts that (37) should be grammatical ("since a time point that is subsequent to a time point prior to the present moment could happen to coincide with the present moment" (p.282)). Comrie's hypothesis (which requires backshifting after said) accounts for the grammaticality of (36) and the ungrammaticality of (37). Our hypothesis does so too. It accounts for the use of the conditional tense in (36) (since the conditional is the tense to be used to represent a situation as posterior to the central situation of the past time-sphere domain). And it accounts for the ungrammaticality of (37) because it holds that the use of an absolute tense (shifting the domain) in an indirect speech complement clause is a marked possibility which is often blocked. (Sentence (37) is unacceptable as a report of (35) for the same reason as (34) is unacceptable as a report of (32): the contents of a prediction (which are by definition nonfactual) should not be reported as a (past or present) fact.) The above discussion can be summarized as follows. There are five sets of data which the absolute deixis hypothesis cannot account for. All these data are consistent with both Comrie's hypothesis and ours. However, the discussion has revealed that there are other data (viz. sentences like (28)-(29)) which our hypothesis can account for, whereas Comrie's cannot.

526 Let us have a look now at the evidence which Comrie adduces against the 'relative time hypothesis'. (This evidence should also refute our hypothesis, since this is basically a relative time hypothesis, supplemented with the claim that the use of absolute tense is sometimes possible too.) Since a relative time hypothesis is more plausible than the absolute deixis hypothesis I know of only three linguists that have advanced the latter, viz. Brecht (1974), Riddle (1978) and Heny (1982:124), whereas the former analysis is fairly widespread we might expect Comrie to devote even more space to refuting the former than he has devoted to refuting the latter. However, this is by no means the case. Comrie treats the relative time hypothesis as if it were hardly worth considering, and adduces only two arguments against it. The first of them is really a nonargument: "Russian does use relative time reference; English differs from Russian, therefore English cannot use relative time reference" (p.278). Let us skip this argument and move on to the second. It hinges on the example Andrew said that he was sick (although he now claims to be better) and runs as follows:
One possible interpretation of the indirect speech, and the one that we shall be concerned with here, is that Andrew's actual words were / am sick, i.e. that the sickness in question is simultaneous with Andrew's utterance. If tense were conditioned by relative time reference, however, the past tense would be inappropriate in these circumstances, or at least the present (simultaneity with a contextually given reference (Kiint) should also be possible, but in fact the present is not possible on this interpretation (in particular, where Andrew's real or putative sickness does not extend to the present). Thus tense in indirect speech in English is not conditioned by relative time reference, (p.278)

There are two (explicit or implicit) claims underlying this argument: (a) there is no tense in English that expresses relative time; (b) if the expression of relative time were possible, simultaneity would always have to be expressed by the present tense. Claim (a) is made explicitly elsewhere in the article. After pointing out (on p.272) that the preterite, the present tense, and the future tense express absolute time in independent clauses, Comrie goes on to claim that the same thing is true in embedded clauses:

527
It is important to note that these characterizations of the meanings of these three tenses apply just as much in subordinate clauses as in main clauses, as in (19): (19) While Ian was singing, Jennifer was dancing. In (19), the past tenses in both clauses are determined by past time reference relative to the here-and-now, i.e. are absolute deictic expressions; the fact that the actions of singing and dancing were simultaneous (relative deixis) is not expressed grammatically in English.

It is clear that we cannot subscribe to this view, which is asserted dogmatically, without a single piece of evidence supporting it. How can one accept (as Comrie does) that there are special tenses (viz. the pluperfect and the conditional tense) to represent a situation as anterior or posterior to a past situation, while at the same time denying that it is possible to represent a situation as simultaneous with a past situation? I certainly know of no language (particularly in the Indo-European group) whose tense system allows the expression of anteriority and posteriority, but has no form(s) to express simultaneity. Moreover, the claim that the past, present and future tenses can only be used as absolute tenses is flatly contradicted by examples like Jennifer will be dancing when Ian is singing or Tomorrow you will say that you are sick, where the present tense forms (is singing, are) cannot be interpreted as referring to the moment of speech and can only be taken to be relative tense forms expressing simultaneity. The second claim underlying Comrie's argument against a relative time hypothesis is equally vacuous. If English allows the expression of simultaneity, why should it be the case that only the present tense could be used for this? This is again a claim that is presented without evidence, and which is obviously in contradiction with the facts of English. I maintain that the preterite is the tense to be used to express simultaneity in a past time-sphere domain, and all the examples involving past time reference that have been discussed so far support this view. In sum, Comrie's arguments against a relative time hypothesis turn out to be no arguments at all because they are based on fallacies. This takes us back to the conclusion of the previous section, viz. that so far we have found one set of examples that are inconsistent with Comrie's analysis, but none that are inconsistent with the hypothesis I have been arguing.

528
5. FURTHER FACTS FROM COMRIE ( 1 9 8 6 )

Comrie (1986) also considers some "apparent problems" (p.283) for the formal SoT hypothesis. Let us see if these data are problems for our hypothesis too. a. A formal SoT rule stipulating that backshifting of tenses is obligatory after a verb of reporting in the past tense wrongly predicts that sentences like the following should be ungrammatical: (38) Yesterday, John said that he will arrive tomorrow. (39) Bill said that he is ill. Comrie solves this problem by adding a restriction to his rule of SoT. The rule now stipulates that backshifting after a verb of reporting in the past must take place, except "if the content of the indirect speech has continuing applicability" In that case "the backshifting is optional" (p.285). Comrie also points out that some of these cases of continuing applicability cannot be handled by the absolute deixis hypothesis. For example: (40) * Yesterday, Linda said she will arrive tomorrow, but she immediately changed her mind. Comrie's argument runs as follows:*
Absolute deixis is incapable of distinguishing between grammatical examples like (39) and ungrammatical examples like (40), because in both sentences the verb in the future has future time reference relative to the here-and-now of the reporter, and should therefore be grammatical. But whereas in (39) the content of the original utterance still has continuing applicability, in (40) the final clause makes it clear that the content of the original utterance does not have continuing applicability. What is crucial in the examples discussed in this section is thus not the time reference of the content of the original utterance, but rather its continuing applicability, and examples like (40) show that these are not necessarily the same, (p.286)

It is clear that this criticism applies to the absolute deixis hypothesis but not to the one we have been arguing. Our hypothesis explains the use of present time-sphere tenses in the <Aa<-clauses
I have substituted my own reference numbers for Comrie's in this quotation, and in another that follows.

529 of (38)-(39) as resulting from a shift of domain. It also says that a shift of domain is a marked possibility (the unmarked alternative being the use of relative tenses), and is therefore subject to restrictions. In the case of (38)-(40) this restriction is that the statement must have continuing applicability. (By relating the subclause situation to his own here-and-now the speaker makes clear that he believes in the truth of the subclause. Of course he cannot do this at least not without violating Grice's Maxim of Quality if he has (and reports) information to the contrary.) Our hypothesis is thus perfectly in keeping with Comrie's remarks. Moreover, it is superior to Comrie's analysis in that it accounts for the use of present time-sphere tenses in the that-c\a,\ises of (38)-(39) without having to introduce an additional (and rather ad hoc) stipulation. Furthermore, it should be pointed out that Comrie's solution (according to which backshifting is optional if the original utterance has continuing applicability) fails to account for the fact that it is sometimes impossible to use the present tense, in spite of the fact that there is continuing applicability. The following is a case in point: (41) This is John's wife. Yes, I THOUGHT he was married. As pointed out by Oakeshott-Taylor (1984:4), it is not possible to substitute is married for was married in the reply sentence (which is to be read with the nuclear accent on thought). Yet it is obvious that the speaker takes the truth of the that-claaise for granted. Examples like this make clear that there are restrictions on the use of the present tense in indirect speech clauses depending on a head clause in the past tense. This is in accordance with our theory, which treats the use of the present tense as a shift of domain, i.e. as a marked possibility, which is naturally subject to restrictions. Comrie's theory, however, does not provide for such restrictions.*
The restriction that is at work in (41) has to do with the fact that the verb of the head clause is a cognition verb (thought). In general, cognition verbs very seldom allow a shift of domain. The reason is that the report of a thought is not felt to be a faithful report if it involves a shift of domain, because the shift changes the structure of the thought. This follows from the fact that the speaker relates the contents of the t/iat-clause to his own here-and-now rather than to that of the original speaker.

530 b. The second of Comrie's "apparent problems for the SoT hypothesis" arises from examples like the following, which should be interpreted as uttered on Wednesday: (42) On Friday, Oswald will say, 'I arrived on Thursday.' (43) On Friday, Oswald will say that he arrived on Thursday. (44) ?0n Friday, Oswald will say that he arrived tomorrow. The absolute deixis hypothesis wrongly predicts that (43) and (44) are ungrammatical. The formal SoT analysis predicts that they should both be grammatical, and therefore also meets with a problem, for it turns out to be the case that many speakers find (44) ungrammaticaJ. Comrie's solution to this problem is that, for those speakers who reject (44), a collocation restriction should be formulated which says that a sentence is ungrammatical if it combines a tense that has past time reference with a "time adverbial whose meaning incorporates future time reference" (p.288). (Note that on Thursday in (43) "does not have future time reference as part of its meaning" (p.288). It is only because of the context that it is here interpreted as referring to the future.) Comrie concludes: "If this collocation restriction account is correct, then the ungrammaticality of (44) (for some speakers) is accounted for by a principle independent of the sequence of tenses rule, and the validity of this rule is unaffected." (p.288) Since this statement is also applicable to our hypothesis we will also have to accept the existence of a restriction of this kind the conclusion is that cases like (44) do not provide evidence for or against either of the competing hypotheses. c. Comrie stresses that his SoT rule "is completely independent of the meaning of the tense forms involved, it is a purely formal operation" (p.29O). It follows that sentences like (46), in which the predicted backshifting of the past tense to the pluperfect does not take place, form a challenge for the formal SoT theory:

531 (45) Yesterday, Wendy said, 'I arrived yesterday.' (46) Yesterday, Wendy said that she arrived the day before yesterday. (47) Yesterday, Wendy said that she had arrived the day before yesterday. The fact that both (46) and (47) are grammatical presents no problem for our hypothesis: (47) illustrates the normal rule that the past perfect is used when a situation that is incorporated into a past time-sphere domain is represented as anterior to another situation in that domain; (46) is an instance of a shift of domain, i.e. the that-cldMse situation is not incorporated into the existing domain but is represented as establishing a domain of its own (by the use of the preterite as absolute tense). However, the grammaticality of (46) does present a problem for Comrie's theory. In order to solve it, Comrie proposes that we should accept the following two principles: (a) "Where the verb in the original utterance is in a past tense already, it remains unchanged after a main clause verb in the past tense in indirect speech" (pp.290-291) (b) "In English, any event in the past can be referred to by the past. If that event is located prior to some contextuaJly established reference point in the past, then the past may be replaced by the pluperfect" (p.292) According to Comrie, principle (a) explains the grammaticality of (46), while principle (b) accounts for that of (47). Several things should be noted here: 1) These principles again constitute ad hoc supplements to the SoT rule. Our hypothesis, in contrast, accounts for them without any additional apparatus. 2) Principle (b) involves reference to a temporal relation. This means that Comrie is moving away here from a pure formal SoT hypothesis (which hinges on the claim that SoT happens independently of temporal relations) and is trespassing on the ground of the relative time hypothesis. (As a matter of fact, this is the second time that Comrie has recurred to semantics, for the notion of continuing applicability is also a semantic notion.)

532 3) Principle (a) is flatly contradicted by the observation that 1 loved her cannot be reported as He said he loved her, unless there is a context that excludes interpreting loved as a relative tense form expressing simultaneity. (Since Comrie's principles are of a purely formal nature, they operate independently of the context. Principle (a) therefore wrongly predicts that / loved her can always be reported as He said he loved her.) (Note that our theory naturally accounts for the fact that this kind of report is not always possible, as it holds that a shift of domain is only admissible if it does not obscure the temporal relations.) 4) The two principles do not account for the restrictions that are to be observed in the following examples: (48) (a) (b) (c) (49) (a) (b) (c) (50) (a) (b) (c) (51) (a) (b) (52) (a) (b) (53) (a) (b) I spoke to her when she came home. He said he had spoken to her when she came/ had come home. *He said he spoke to her when she had come home. (ungrammatical as a report of (48,a)) I spent some time with John, who felt lonely. She said that she had spent some time with John, who felt/had felt lonely. *She said she spent some time with John, who had felt lonely.(ungrammatical as a report of (49,a)) I was sad while I was alone. She said she had been sad while she was/ had been alone. *She said she was sad while she had been alone. I decided not to buy the house because it weis on a main road. He explained he had decided not to buy the house because it was/*had been on a main road. John said: 'I told Betty that I was feeling ill. John said that he had told Betty that he was feeling/*had been feeling ill. I left the money where it was. He said he had left the money where it was/ *had been.

533 In our theory the use of the past tense in the most deeply embedded subdauses of the (b) sentences (i.e. in the subclauses that appear in final position) follows naturally from the fact that these subclauses represent their situations as simultaneous with a past binding time. The fact that the past perfect is sometimes also possible in these clauses follows from the fact that in some types of subclause 'indirect binding' is allowed. This means that the subclause in question is temporally bound, not by its own matrix (superordinate clause), but by the clause that also binds its matrix (cf. chapter 2 of Declerck, forthcoming). In Comrie's theory, principle (b) will have to account for the past perfects in the sentence-final clauses of (48,b), (49,b) and (50,b): we will have to say that the past perfect can each time be used in the third clause because the situation referred to is anterior to the situation of the first clause (whose verb is said). However, this explanation fails to account for the fact that the past perfect is not possible in the third clause of (51,b), (52,b) and (53,b), where the same anteriority relation exists between the third clause and the first. (Our analysis can account for this, since indirect binding is a restricted option, which is only available if it does not obscure the temporal relations.) Another problem for Comrie's theory is that it wrongly predicts that the (c) sentences of (48)-(50) are grammatical reports of the (a) sentences. According to Comrie the past tense of the original report can always be retained (which is what happens in the first subclause of each (c) sentence) or can under certain conditions be backshifted to a past perfect (which is what happens in each of the final clauses of the (c) sentences). The condition under which the past perfect can be used is that there is some contextually given past reference point to which the relevant situation is anterior. This condition is satisfied in the (c) sentences, since the situation of the final clause is each time anterior to that of the first clause. Still, the (c) sentences are no grammatical reports of the (a) sentences, which means that Comrie's theory makes incorrect predictions. (In our theory the observation that the final clause of the (c) sentences cannot be in the past perfect is explained from the fact that this clause must be bound

534 by its matrix and that the relation to be expressed is that of simultaneity. (Using a past perfect would be an instance of indirect binding, but this is not possible here because the second clause is not bound; it shifts the domain. The second and third clauses thus belong to a domain that is different from the one established by the first clause. Self-evidently this means that the first clause cannot bind the third.)) 5) Last but not least, Comrie's principles (a) and (b) imply that, in pairs like (46)-(47), where the t/iat-clause refers to a situation that is anterior to the head clause situation, the preterite is the unmarked choice in the that-cla.vLse, whereas the use of the past perfect is a marked alternative. This runs counter to our theory, which holds that the use of the past tense (which shifts the domain) is the marked possibility. The unmarked alternative is the use of the past perfect (which incorporates the that-cl&use situation into the existing domain). Now, it is not difficult to prove that it is indeed the past tense that is the marked choice, i.e. the choice that is subject to restrictions. When a direct speech statement in the past tense is embedded under a verb of reporting in the past tense, the 'backshift' from the absolute preterite to the pluperfect is always possible. It is not subject to any restrictions whatsoever. The use of an absolute past tense, however, requires that the anteriority relation between the that-c\a,}ise situation and the head clause situation should be clear from an adverbial, the context, or the hearer's knowledge of the world (cf. above). If that is not the case, the preterite will maJce for a sentence that is unacceptable because it violates the generally accepted principles of conversation. Thus a sentence like John said he had been ill, if presented out of context, cannot be replaced by John said he was ill without violation of Grice's (1975) Maxim of Quantity (which requires that the speaker should make his contribution sufficiently informative to guide the hearer unequivocally to the intended interpretation). Failing evidence to the contrary, John said he was ill will be interpreted as a report of / am ill, not as a report of / was ill.

535 Examples of domain shifting in temporal clauses confirm this conclusion. (I feel entitled to include time clauses in the discussion because Comrie (1986:292) does so too: He refers to the tense vacillation in After he ate/had eaten breakfast, Xenophon put on his armor as evidence for his principle (b).) Like complement clauses, time clauses allow a shift of domain (entailing that the situation is no longer related to any other situation) only if the temporal relation between the situations of head clause and time clause are recoverable in some other way. In the above example, the relation is recoverable because it is explicitly expressed by after. But the relation is no longer recoverable if we replace after by a temporally vague conjunction such as when. In consequence, the Maxim of Quantity forbids replacing When he had eaten breakfast, Xenophon put on his armour by When he ate breakfast, Xenophon put on his armour. We must conclude that Comrie's treatment of sentences like (46) (in which the preterite from the original utterance is not 'backshifted') is not satisfactory. This is another case, then, where the formal SoT hypothesis is deficient, whereas the analysis argued here neatly accounts for the data.
6. FURTHER ARGUMENTS

Let us now consider some data which are not mentioned in Comrie (1986) and which provide further evidence against his SoT hypothesis. a. Examples can be found of indirect speech sentences whose complement clauses are in the present perfect although the tense of the head clauses is past: (54) The Secretary of Labor stated the other day that in the past couple of months there have only been 200 lost man days. (Ota 1963:115) (55) Not too long ago (...) a gentleman who'd been on foreign service mentioned that Latin America has not had its New Deal, (ibid) (56) The Prime Minister stated in Parliament yesterday that Britain's economy has looked up considerably during the past two months.

536 In our theory such sentences are explained in terms of a shift of domain: the head clause establishes a domain in the past timesphere, whereas the subclause establishes one in the present timesphere. As usual, the shift is semantically motivated. In this case the motivation is that the embedded statement concerns a period which started in the past and reaches up to the present. As is well-known, the present perfect is the normal tense to refer to such a period. Sentences like (54)-(56) run counter to the formal SoT hypothesis because they exhibit no backshifting in spite of the fact that the verb of reporting is in the past tense. To salvage the SoT rule one would have to formulate still another ad hoc restriction providing for this kind of exception. Needless to say, this would erode the rule even further. b. We now come to a piece of evidence which is really detrimental to the formal SoT hypothesis. As noted briefly in section 2, domain-internal relations within a domain established by a socalled 'indefinite' perfect are normally expressed by the system of tenses typical of the past time-sphere.^ That is, once the present perfect has established the domain, there is a shift of temporal perspective to the past time-sphere. The following sentences exemplify this: (57) He's been here once or twice, while his wife was on holiday. (58) Have you ever spoken to her after she had had one of her flts? The same shift of temporal perspective is to be observed in indirect speech sentences:^
An 'indefinite perfect' refers to a pre-present situation that is completed before the moment of speech. We can also find indirect speech sentences that use a present perfect in the complement clause after a present perfect in the head clause: (i) I have always maintained that I have never said that. In this sentence the speaker does not use a relative tense incorporating the subclause situation into the head clause domain. Instead, he uses an absolute tense which shifts the domain, i.e. which creates a new pre-present domun. The semantic motivation for this shift is that the speaker wants the complement clause to refer to a time interval reaching up to the moment of speech.

537 (59) I have never said that you were stupid. (60) (He has met her several times but) he has never told his wife that he had met her. (61) (He has met her many times and) he has always promised her that he would not say anything to her husband. When we reconstruct the direct speech statements that are reported in these sentences, we see that they are in the present, present perfect, and future tense, respectively. This means that, in the terminology of the formal SoT hypothesis, (59)-(61) are instances of backshifting after a verb in the present perfect. Such sentences constitute an insurmountable problem for Comrie's theory, as they are incompatible with the most basic claim of the hypothesis, viz. the claim that backshifting is "a purely formal operation" (p.290) which is triggered by the past tense of the verb of reporting in the head clause and which consequently cannot occur if the tense of the verb of reporting is non-past. Since sentences like (59)-(61) have a reporting verb in the present perfect, they present a case of backshifting which simply falls outside the scope of the formal SoT rule. (The present perfect is a non-past tense. Comrie himself treats it this way on p.290 of his article.) This thrusts to the heart of the SoT hypothesis. This time the hypothesis can no longer be salvaged by the formulation of an ad hoc restriction on its applicability. c. Up to now I have only adduced arguments that are based on observations. Obviously, these arguments in themselves are amply sufficient to demonstrate that the formal SoT hypothesis is untenable. However, it is worth pointing out that this analysis also raises all kinds of theoretical problems. 1) As has often been pointed out, a quotative analysis of indirect speech (i.e. an analysis which derives indirect speech sentences from direct speech ones) meets with the problem that some verbs (e.g. realize, regret) can be followed by an indirect speech (or indirect thought) clause but not by a direct quote. 2) Comrie himself argues that the "adaptation' of deictics other than tense (i.e. pronouns and adverbs of time or place) in indirect speech is not the result of a formal rule. The deictics in question

538 are related either to the here-and-now of the reporting speaker (as in He told me that he would do it tomorrow) or to the time of the head clause situation (as is the case for the next day in He said he would do it the next day). Given this fact, it is hard to imagine that the verbal deictics should be analysed in a totally different way. Our hypothesis, which holds that verb forms in indirect speech are either absolute deictics (relating the situation to the here-and-now of the reporting speaker) or relative ones (relating the situation to the time of the head clause situation), treats verbal deictics in exactly the same way as the pronominal and adverbial ones. 3) Comrie's analysis of indirect speech is not applicable to socalled 'free indirect speech' (erlebte Rede), unless it is assumed that a free indirect speech sentence is the result of ellipting the reporting clause of a direct speech sentence. (The latter solution is not without problems. To mention only one: how is the hearer to know whether the verb of the deleted clause is thought or believed, said to himself, realized, noticed, etc.? That is, this kind of deletion would not satisfy the well-known recoverability requirement.) Our theory, however, does not need to posit a deleted hyperdause in order to account for the 'backshifted' tenses in free indirect speech. As is well-known, the speaker may assume the temporal standpoint of a participant in a past situation in order to suggest that the situation is expressed from his point of view. This means that he relates the situations to a particular past time (the time when the relevant character said, thought or observed something) rather than to his own here-and-now. The past time-sphere tenses in free indirect speech are thus explained as tenses that are bound by the past time that the speaker is focussing on. Our analysis therefore does not need to consider free indirect speech sentences as derived (through deletion of the reporting clause) from indirect speech sentences (which, on this hypothesis, are themselves derived from sequences involving a direct quote). 4) Comrie is at pains to argue that he can account for the use of tenses in indirect speech without referring to the semantics of the temporal relations. He repeatedly makes statements of the following kind:
One feature of tense backshifting that takes place in indirect

539
speech after a main verb in the past tense is that (apart from the stipulation concerning continuing applicability) it is completely independent of the meaning of the tense forms involved, it is a purely formal operation. (Comrie 1986:289-290)

However, statements like this raise intriguing theoretical questions. How can we reconcile this view (that the use of the tenses in indirect speech bears no relation to the meaning of the tense forms in question) with Comrie's definition of tense as "the grammaticalisation of location in time" (Comrie 1985:1)? The claim that backshifting is a "purely formal" operation, which is "independent of the meaning of the tense form involved", suggests that was in John said he was ill has the same meaning as am in / am ill. If this is not the case, Comrie's SoT rule is not a purely formal rule, as it has semantic import. But of course was is semantically different from am. Comrie himself admits as much when he formulates the 'collocation restriction' which leads to the reduced acceptability of' On Friday, Oswald will say that he arrived tomorrow as follows:
what is wrong with (86) is the apparent direct conflict between the past tense, part of whose meaning is paist time reference, and a time adverbial whose meaning incorporates future time reference. (Comrie 1986:288)

If past time reference is part of the meaning of arrived in the above example, then that element of meaning must have been introduced by the backshifting operation. But if so, we cannot maintain the fiction that the operation is a purely formal one. Moreover, once this is clear we must ask the question whether the element of past meaning results from the fact that the form of the verb is changed from present tense to preterit or whether it is the other way round (i.e. the choice of verb form is determined by the meaning to be expressed). In my opinion, there is little doubt that the latter hypothesis is the correct one. If this were not the case, the 'backshifting* rule would be without any motivation whatsoever. If one reads the article carefully, one easily sees that Comrie does not succeed in eliminating semantics from his allegedly "purely formal" rule. Each of the three additions to the original formulation of the rule crucially refers to the meaning of the

540

tense forms. Moreover, even the flrst formulation encroaches on semantics, and Comrie himself points this out when he writes:
...the sequence of tenses rule (...) says either (for non-past main clause verb) that the tense of the original utterance is retained, or (for past tense main clause verb) that a transposition is to be applied to the tense of the original utterance. In either case, the tense of the original utterance is crucial (...) Thus, there is a very real sense in which relative deixis, relative to the speech situation of the original utterance, is part of tense in indirect speech in English. (Comrie 1986:283)

However, if one accepts this, there is absolutely no reason left to assume the existence of a formal backshifting rule: given that tenses in indirect speech (at least, the 'backshifted' ones) express relative deixis, and given the fact that the relative tenses to be used for this if the binding time belongs to a past domain are past time-sphere tenses, the use of past time-sphere tenses in indirect speech after a head clause in a past tense is automatically accounted for. 5) One of the questions raised by the formal SoT hypothesis is why it is completely impossible to backshift the past perfect. Comrie's (1986:291) answer to this is that there is no verb form '''had had seen. (In fact *had have seen would have been a more likely candidate.) However, if backshifting were a rule that was operative in English, it is rather unlikely that the form *had have seen would not have existed. After all, this is not a very complex form. It is no more complex than would have seen or will have seen and even less complex than has been going to see, was going to have seen, will have been going to see, etc. This suggests that the nonexistence of *had have seen fails to be explained by the suggestion that this is too complex a form to be acceptable. Moreover, we also observe that would see is never backshifted in indirect speech. Yet, we note the existence of the forms had been going to see and would have seen. One of these would surely do if English needed a backshifted version of would see. The conclusion is clear: if the past perfect and the conditional tense are not backshifted, it is not for lack of suitable verb forms. It is simply because there is no such thing as a formal backshifting rule in English. The truth is that the past perfect and the conditional express a temporal relation in a past domain and that they do so irrespective of the location of the binding time in the domain. This means that the

541 reason why the past perfect is not backshifted when we report I had already left when Mary arrived as Bill said he had already left when Mary arrived is that in both sentences Bill's leaving is expressed as anterior to Mary's arrival. And a similar remark can be made in connection with the conditional tense: I thought I would meet her here is reported as He said he had thought he would meet her there without there being any need for 'backshifting' would meet because the temporal relation remains the same. 6) According to Comrie, one of the advantages of the formal SoT hypothesis is that it offers a unified account of the data and is therefore more economical than an analysis which is based on two or more principles:
Sequence of tenses (...) provides a homogeneous solution to the general problem of tense in indirect speech in English, covering all examples with a single set of principles. Of course, it would surely be possible to take absolute deixis in those instances where it happens to make the same predictions as sequence of tenses, then take some other principle (perhaps relative deixis) in some of the cases where absolute deixis makes incorrect predictions, plus perhaps some other principle where even this combination fails to make the correct prediction, and come up with a composite account which would have different principles or sets of principles operating in different sets of cases. This simply illustrates a commonplace of science, namely that given a particular hypothesis which covers a certain range of data, it is always possible to come up with some more complex theory that covers the same range of data. The beauty of the sequence of tenses rule is that it covers the whole range of data with just a single simple set of principles. (Comrie 1986:282-283)

At first sight Comrie's formal SoT analysis might seem to be simpler than our analysis (which holds that indirect speech complement clauses use either relative or absolute tense forms). However, it soon becomes clear that this is not true. The above quotation refers to the preliminary version of the SoT rule, i.e. to the version which covers only the 'backshifted' tense forms. Our analysis also needs only one principle to account for these, since the cases in which there is 'backshifting' coincide with the cases which we analyse in terms of temporal subordination. However, immediately after the above quotation Comrie has to add a couple of ad hoc stipulations to his rule in order to accommodate the various cases in which there is no 'backshifting', In our theory all these cases are analysed as shifts of domain. This analysis.

542 which works with two principles only, is therefore more economical than Comrie's, which works with a basic rule and a number of exceptions. Moreover, our analysis is applicable not only to indirect speech subclauses but to subdauses generally. Comrie, by contrast, treats the 'backshift' of tenses in indirect speech complement clauses as an isolated phenomenon, unrelated to the behaviour of tenses in other environments. This not only makes his analysis less general and hence less economical than ours but also appears unjustified, since we notice a striking parallelism between what happens in indirect speech and what happens in other clauses (both embedded and unembedded). Compare: (62) (a) The man said he had heard a shot in the street. (b) The man phoned the police. He had heard a shot in the street. (c) The man, who had heard a shot in the street, phoned the police. (d) The man phoned the police because he had heard a shot in the street. (e) The man reported the fact that he had heard a shot in the street to the police. In the theory I have argued, the use of the past perfect receives the same explanation in each of these examples: the past perfect is used because the situation referred to is incorporated into the past time-sphere domain established by the head clause or preceding context, and because it is represented as anterior to the central situation of that domain. (The first element of the explanation accounts for the use of a past tense (the past perfect), the second for the use of a tense signalling anteriority (the past perfect).) Our hypothesis therefore makes a much larger generalization than the formal SoT hypothesis could ever achieve: it accounts for the fact that 'tense harmony' takes place not only in complement clauses after verbs of saying or thinking (where it is called SoT) but also in other types of noun clause (e.g. It was clear that he was lying, There was a chance that she would die. The idea that she was perhaps dead had not occurred to me), and in fact in any kind of clause that allows temporal subordination (e.g. Last week he made a promise which he would not keep, I got the job because I knew the manager well. They needed someone

543 who had studied languages).^ It also accounts for the fact that tense harmony may be 'violated' both in complement clauses and in other types of clause (e.g. Last week he made a promise which he will never keep): tense harmony disappears whenever the speaker prefers to shift the domain.
7. CONCLUSION

We have investigated three a priori possible explanations of the behaviour of tenses in indirect speech, viz. the absolute deixis hypothesis, the formal SoT hypothesis, and a hypothesis which holds that the tenses in indirect speech complement clauses are mostly relative, but sometimes absolute tenses. We have subscribed to Comrie's (1986) claim that there are data which cannot be accounted for by the absolute deixis hypothesis. We also agree that a 'relative time hypothesis' which holds that the tenses in indirect speech complement clauses must invariably be relative tenses is untenable. However, contrary to Comrie (1986), we have shown that the formal SoT hypothesis is not adequate either. Several arguments have been adduced to this effect, the most important of which are the following: a. There are too many data that are incompatible with the basic claims of the formal SoT hypothesis, and which can only be treated as exceptions. b. The formal SoT hypothesis does not cover cases of backshifting after a reporting verb in a non-past tense. c. The formal SoT hypothesis does not cover cases in which there is no backshifting in spite of the fact that the reporting verb is in a past tense. (I am referring here to the fact that tense forms
g

The following show how tense harmony works in larger discourses: (i) John has a very good memory, which he has always thought of as a great advantage. He stores up the things his friends say to him and if he does not understand them, he goes back to them for clarification after he has turned them over in his mind. (ii) John had a very good memory, which he had always thought of as a great advantage. He stored up the things his friends said to him and if he did not understand them, he went back to them for clarification after he had turned them over in his mind.

544 expressing a temporal relation in a past domain are not backshifted in indirect speech.) d. We have mentioned no fewer than six objections of a theoretical nature that can be made against the formal SoT hypothesis. One of the most serious is that it misses an important generalization because it treats tense in indirect speech complement clauses as totally unrelated to tense in other environments. The theory we have argued, according to which complement clauses may use either relative tenses or absolute tenses (the former possibility being the unmarked one), is subject to none of these objections. It accounts naturally for all the possibilities that manifest themselves, both in indirect speech and in other contexts. Address of the author: Renaat Declerck K.U. Leuven, Campus Kortrijk Universitaire Campus B-8500 Kortrijk (Belgium)
REFERENCES Brecht, R.D. 1974. "Deixis in Embedded Structures." Foundations of Language 11, 489-518. Comrie, B. 1985. Tense. Cambridge: Cambridge University Press. Comrie, B. 1986. "Tense in Indirect Speech." Folia Linguistica 20, 265-296. Declerck, R. (forthcoming). Tense in English: Its Structure and Use in Discourse. London: Routledge. Dinsmore, J. 1982. "The Semantic Nature of Reichenbach's Tense System." Glossa 16, 216-239. Dowty, D.R. 1982. "Tense, Time Adverbs, and Compositional Semantic Theory." Linguistics and Philosophy 5, 23-55. Ejerhed Braroe, E. 1974. The Syntax and Semantics of English Tense Markers. Stockholm: University of Stockholm Institute of Linguistics. Grice, P. 1975. "Logic and Conversation." In: P. Cole & J.L. Morgan (eds) Syntax and Semantics 3: Speech Acts. New York: Academic Press, 41-58. Heny, F. 1982. "Tense, Aspect and Time Adverbials: Part 2." Linguistics & Philosophy 5: 109-154. Oakeshott-Taylor, J. 1984. "Factuality and Intonation." Journal of Linguistics 20: 1-21. Ota, A. 1963. Tense and Aspect of Present-day American English. Tbkyo: Kenyusha.