Temporal adverbials are another important source of temporal reference in Chinese.
Chinese temporal adverbials can be syntactically simple or compound. Examples of the former type are xianzai ‘now’, zuotian ‘yesterday’, gang ‘just now’, jiu ‘then’
mashang ‘immediately’, etc. Compound temporal adverbials normally consist of a temporal operator and some governed element— an NP or an S. They can be further divided into three classes, depending upon the position of the temporal operator. Let us abbreviate temporal operators as TO and the governed element as X. The three types are the following:
(a) TO X: qian san nian ‘the first three years’, hou san nian ‘the later three years’, cong san dian dao wu dian ‘from three o’clock to five o’clock’
(b) X TO: san nian qian ‘three years ago’, san nian hou ‘three years later’, san dian (de)-shi-(hou) ‘at three o’clock’, san tian (zhi)-nei ‘within/in three days’, liu shi niandai yilai ‘1960’s on onward’, xinnian yiqian ‘before the new year’
(c) TO X TO: zi xiatian yilai ‘since from the summer’, cong yi yue yi hao kaishi ‘from January 1st’
Many of the first two types can be further preceded by the preposition zai ‘at/in’
without changing the meaning. For example, san tian qian ‘three days ago’ is equivalent to zai san tian qian ‘three days ago’, and qian san nian ‘the first three years’ is equivalent to zai qian san nian ‘in the first three years’.
Temporal adverbials can also be categorized according to their function. For example, Klein (1994) has listed 6 types of temporal adverbials:
(a) positional temporal adverbials such as yesterday, at five o’clock, in the night (b) temporal adverbials of frequency such as often, always, once in a while (c) temporal adverbials of duration such as for one hour, during the autopsy
(d) temporal adverbials describing inherent temporal properties of a situation such as quickly, gradually, slowly
(e) temporal adverbials indicating the position of a situation such as firstly, at last, eventually
(f) temporal adverbials fitting none of the above class such as already and still
In this study I will focus on positional temporal adverbials (abbreviated as PTAs).
Following many previous researches on PTAs, I assume that they single out some interval or a set of intervals determined by the lexical content. Moreover, the interval may specify the event time or the reference time of the event, depending upon the syntactic position of the PTA (Stump 1985, Klein 1994, Lin 2000b). The idea that PTAs may function as a reference time is suggested in Reichenbach (1947), Bennett and Partee (1978), Bäuerle (1979), Dowty (1982), Kamp and Reyle (1993), Klein (1994), among many others. The interval that a PTA denotes may be directly determined by the lexical specification as in the calendaric time zai 1996 nian ‘in the year o f 1996’, or the lexical content of the PTA offers an open slot that allows
‘deictic rooting’ as in jintian ‘today’— the day which contains the speech time, or
‘anaphoric rooting’ as in wu fenzhong hou ‘five minutes later’— five minutes after some time given in the context. I refer the reader to Klein (1994) for a more detailed discussion of these three possibilities. Although the study of the ways by which PTAs determine their time span is very interesting of itself, this will not be the focus of this paper. The focus will be on how PTAs are semantically linked to the sentence that they modify.
The simplest hypothesis for the semantics of PTAs is that they denote nothing but intervals, i.e., entities of type e, as I assumed earlier. These intervals are identified with the reference time provided by Aspect when the PTAs are combined with the sentence they modify. On this hypothesis, a perfective sentence such as (77) is semantically computed as (78).
(77) Zuotian ta dapuo wan yesterday he break bowl
‘He broke a bowl.’
(78) a. [[ta dapuo wan]] =: λt2λt1∃y [t1 ⊆ t2∧ cup’(y) ∧ break’(y)(he’)(t1)]
b. [[zuotian ta dapuo wan]]
=: λt1∃y[t1 ⊆ yesterday’ ∧ cup’(y) ∧ break’(y)(he’)(t1)]
c. Existential closure = ∃t1∃y[t1 ⊆ yesterday’ ∧ cup’(y) ∧ break’(y)(he’)(t1)]
In plain English, the truth conditions in (78c) require that there be an interval t1 at which he breaks a bowl and t1 is included within the reference time yesterday’. Since the event time t1 is included within a past interval, (77) is correctly understood as a past tense sentence.
Next, let us consider an imperfective sentence such as (79). Its truth conditions are given in (80).
(79) San dian de-shihou ta zai shuijiao three o’clock when he be sleep
‘He was sleeping at three o’clock.’
(80) a. [[Ta zai shuijiao]] =: λt2λt1[t2 ⊆ t1∧sleep(he)(t1)]
b. [[San dian de-shihou ta zai shuijiao]] =: λt1[3-o’clock’ ⊆ t1∧ sleep’(he’)(t1)]
c. Existential Closure =: ∃t1[3-o’clock’ ⊆ t1∧sleep’(he’)(t1)]
The truth conditions in (80c) assert that the interval denoted by san dian de-shihou ‘at three o’clock’ is included within the interval t1 at which he sleeps. From this, of course, we cannot conclude that the sleeping event described in (79) has ended, because the truth conditions do not say when the event time is located at the time axis.
Indeed, (79) can be felicitously uttered in a situation in which the event is already completed before the speech time or a situation in which the event is still on going at the speech time. But generally (79) is preferably construed as denoting a past event.
Before explaining why this is the case, let me first show that (79) is compatible with a situation where the event has not ended at the speech time. The evidence can be easily adduced by adding another sentence to (79) as in (81).
(81) San dian de-shihou ta zai shuijiao, keshi xianzai shi-bu-shi hai zai shui, three o’clock when he be sleep but now be-not-be still be sleep wo jiu bu zhidao le
I then not know ASP
‘He was sleeping at three o’clock. But I don’t know whether he is still sleeping now.’
In fact, all stative sentences display the same property. For example, sentence (82), (82) Ta xiao shihou hen tiaopi
he small when very naughty
‘He was naughty when he was a kid.’
when read alone, implies that he is no longer naughty at the speech time. But given a proper context, it does not exclude the possibility that he is still naughty at the speech time. This is illustrated by (83).
(83) Ta xiao shihou hen tiaopi, xianzai zhangda-le haishi zheme tiaopi he small when very naughty now grow-up-ASP still this naughy
‘He was naughty when he was a kid. Now he has grown up, but he is still as naughty as before.’
Do the truth conditions that my analysis assigns to imperfective sentences with a PTA denoting a past interval reflect the fact that such sentences usually imply that they have a past tense interpretation? On my analysis, the sentence (82) has the following truth conditions.
(84) ∃t[the-period-when-he-is-a-kid’ ⊆ t ∧ naughty’(he’)(t)]
(84) says that there is an interval t at which he is naughty and the period when he was a kid is included within t. From this, of course, one cannot conclude that the property of being naughty is restricted to the interval when he was a kid. But then why does (82) strongly imply that the state of being naughty is a past state? This question can be answered by the Maxim of Quantity, a conversation principle formulated by Grice (1975), which recommends speakers to say as much as he can. According to this maxim, if a speaker knows that someone is naughty at the utterance time, he should say it as such. The use of a time adverbial denoting a past interval then generates the implicature that the subject does not have the property of being naughty at the speech time. I conclude that the implicature as we saw in (79) and (82) is a suggested inference instead of part of the truth conditions. (See also Kamp and Reyle (1993) for relevant discussion.)
Although analyzing PTAs as entities of type e may successfully explain most
sentences with a PTA, this approach is in fact inadequate. The illustrating examples above all contain only one PTA, but it is well known that a sentence may contain more than one PTAs--i.e., time adverbials are recursive, as is illustrated in (85) and (86).
(85) [Zuotian [sandian de-shihou [ta zai shuijiao]]]
yesterday three-o’clock when he be sleep ‘He was sleeping at three o’clock yesterday.’
(86) [Ta (shi) [zai 1994 nian [wuyue [diyige xingqi [qu meiguo (de)]]]]]
he be in year May the-first week go America PAR ‘He went to America in the first week of May in 1994’
If (85) has the structure as indicated and each time adverbial denotes an interval of type e, then it should translate as (87), where the interval denoted by san-dian de-shihou ‘at three o’clock’ fills the reference time t2 and the interval denoted by zuotian ‘yesterday’ fills the event time t1.
(87) a. [[ta zai shuijiao]] =: λt2λt1[t2 ⊆ t1∧sleep(he)(t1)]
b. [[san dian de-shihou ta zai shuijiao]] = :λt1[3-o’clock’ ⊆ t1∧sleep’(he’)(t1)]
c. [[zuotian san dian de-shihou ta zai shuijiao]]
=: [3-o’clock’⊆yesterday’ ∧sleep’(he’)(yesterday’)]
Unfortunately, the truth conditions thus yielded falsely predict the meaning of the sentence. For (85) to be true, the subject only needs to sleep at some subinterval t containing the interval of 3 o’clock and contained within yesterday. But this is not what (87c) says. (87c) says that (85) is true if and only if the subject was sleeping all day long yesterday and 3 o’clock is contained within the interval of yesterday. This is certainly false, because for (85) to be true, the subject does not need to sleep for the 24 hours yesterday. Also, the condition “3-o’clock’⊆yesterday’” does not provide any meaningful communicative information, because any interval within the 24 hours of a day is contained within that day.
A first solution to the above problem that comes to mind is to analyze zuotian sandian de shihou ‘at three o’clock yesterday’ as a single constituent with zuotian modifying san dian de shihou ‘three o’clock’. That is, (85) has the structure (88).
(88) [IP[DP zuotian sandian de shihou] [IP wo zai shuijiao]]
If the structure of (85) is as in (88), then the reference time of the sentence is the
3-o’clock of yesterday and the event time will be existentially closed as is usual. The problem thus seems to be avoided. This is not true, however, on closer inspection.
Even if the bracketing in (88) is possible, evidence indicates that the constituents zuotian ‘yesterday’ and sandian de shihou ‘three o’clock’ need not always form a constituent, because they can be separated by the subject of the sentence, as illustrated by (89).
(89) zuotian wo sandian-de-shihou zai shuijiao yesterday I three-o’clock at sleep ‘I was sleeping at three o’clock yesterday.’
Thus, the original problem remains in (89) regardless of whether the constituent structure in (88) is correct or not.18
From the above discussion, it is clear that to solve the problem of multiple time adverbials, we have to make them semantically recursive. The question is how this is possible. Before answering the question, let us first consider how the same problem is tackled in the literature as in Dowty (1979) and Ogihara (1996). Let us assume that propositions are temporal abstracts of <i,t> type, i.e., sets of event times, and that temporal adverbials serve as a restriction upon (event) time variables, denoting objects of type <<i,t>,<i,t>> , i.e., functions from properties of times to sets of time intervals. Now consider the example in (90), ignoring the problem of morphological tense.
(90) He broke a bowl at night.
The proposition ‘he break a bowl’ denotes the set of times at which he breaks a bowel, i.e., λt∃y[bowl’(y) ∧ break’(y)(he’)(t)]. On the other hand, at night denotes λP<i,t>λt[t
⊆ at-night’ ∧ P(t)]. The combination of these two is thus another object of type <i,t>
as (91) shows.
(91) λP<i,t>λt[t ⊆ at-night’ ∧ P(t)]( λt∃y[bowl’(y) ∧ break’(y)(he’)(t)]) = λt∃y[t ⊆ at-night’ ∧ bowl’(y) ∧ break’(y)(he’)(t)]]
The newly created object of type <i,t> can then serve as the argument of another temporal adverbial of type <<i,t>,<i,t>>. Thus, the denotation of the temporal
18 One might suggest that the NP zuotian ‘yesterday’ is moved to the topic position. Thus, at
D-structure, zuotian ‘yesterday’ still forms a consitituent with sandian de shihou ‘three o’clock’. This analysis, however, should be excluded by an independently motivated syntactic principle which disallows extraction out of an adjunct NP.
adverbial today can be combined with the denotation of the proposition ‘he eat an hamburger at night’ without any problem, as is shown in (92) below.
(92) λP<i,t>λt[t ⊆ today’ ∧ P(t)]( λt∃y[t⊆night’ ∧ bowl’(y) ∧ break’(y)(he’)(t)])) = λt∃y[t ⊆ at-night’ ∧ t ⊆ today’ ∧ bowl’(y) ∧ break’(y)(he’)(t)]]
After existential closure, the truth conditions then require that there be a time t which is included within the night time and which is included within today and he breaks a bowl at t. The recursive problem of English temporal adverbials thus can be solved by analyzing them as objects of type <<i,t>,<i,t>>.
The above approach, however, cannot be directly applied to Chinese under the assumptions of this current study. As I noted earlier, to correctly understand temporal reference in Chinese, the notion of aspect is crucial. Following Klein (1994), I have assumed that aspect is a relation between event times and reference times. Thus, what combines with a temporal adverbial is not simple properties of event times of <i,t>
type but objects of <i,<i,t>> type. Clearly, objects of type <i,<i,t>> cannot serve as an argument of a function of type <<i,t>,<i,t>>.
There is another reason not to adopt the above approach. Note that the illustrating example (90) is a perfective sentence. If a PTA is combined with an imperfective sentence, the above approach will not work. Consider (93a).
(93) a. Zhangsan xiawu shenti bu shufu Zhangsan afternoon body not feel-well ‘Zhangsan did not feel well this afternoon.’
b. ∃t[t⊆this-afternoon’ ∧ ¬fell-well’(Zhangsan’)(t)]
On the Dowty-Ogihara’s approach of PTAs, the situation time is always included within the reference time, no matter whether the sentence is perfective or imperfective.
Thus, (93a) has the translation in (93b). However, this leads to an incorrect interpretation of the sentence, because as we have already mentioned, the relation between the situation time and the reference time for imperfective sentences is reverse to that of perfective sentences (Kamp and Reyle 1993 and Klein 1994, among others).
Although Dowty-Ogihara’s approach of PTAs cannot be adopted, its spirit, together with Dowty’s (1982: 31) comment in another article that “Time adverbials… are… sentence operators which semantically have the effect of asserting that the reference time i is located at a particular time, say within yesterday” suggests to us a new analysis. Instead of type <<i,t>,<i,t>>, let us assume that PTAs are of type
<<i,<i,t>>,<i,<i,t>>> with the following semantics.
(94) [[PTA]] =: λP<i,<i,t>>λt2λt1[t2 ⊆ PTA’ ∧ P(t2)(t1)]
In essence, what the semantics of (94) does is to locate the reference time t2 within the time span singled out by the PTA. Since the semantic type of the argument of the function denoted by a PTA is the same as that of its resulting expression, the output expression can serve as the argument of another PTA. This explains why PTAs are recursive. Take (91) as an illustrating example. Its semantic computation is as follows:
(95) a. [[ta zai shuijiao]] =: λt2λt1[t2 ⊆ t1∧sleep’(he’)(t1)]
b. [[san dian de-shihou ta zai shuijiao]]
=: λP<i,<i,t>>λt2λt1[t2 ⊆ 3-o’clock’ ∧ P(t2)(t1)]( λt2λt1[t2 ⊆ t1∧sleep’(he’)(t1)])
=: λt2λt1[t2 ⊆ 3-o’clock’ ∧ t2 ⊆ t1∧sleep’(he’)(t1)]) c. [[zuotian san dian de-shihou ta zai shuijiao]]
=: λP<i,<i,t>>λt2λt1[t2 ⊆ yesterday’ ∧ P(t2)(t1)]( λt2λt1[t2 ⊆ 3-o’clock’ ∧ t2⊆t1∧ sleep’(he’)(t1)]))
=: λt2λt1[t2 ⊆ yesterday’ ∧ t2 ⊆ 3-o’clock’ ∧ t2 ⊆ t1∧ sleep’(he’)(t1)]
After existential closure, (95c) is equivalent to saying that there is an event time t1 at which he sleeps and a reference time t2, which is located within 3-o’clock of yesterday and t2 is included within the event time t1. This interpretation of the sentence is not only correct but avoids the problem of the perfective-imperfective distinction because the semantics of PTAs as given in (95) only identifies the reference time with the time span singled out by the PTA but says nothing about the relation between situation time and reference time. The relation is expressed via aspect, not via PTAs.
In order to solve the recursive problem of PTAs, I have argued that they do not denote intervals of type i but are expressions of type <<i,<i,t>>,<i,<i,t>>>. However, there is evidence showing that under certain circumstances temporal adverbials may function as type i expressions. Many authors have pointed out that –le cannot occur with a future time adverbial as is shown by (96).
(96) Wo mingtian (zhege shihou) likai-(*le) nanjing I tomorrow this moment leave-Asp Nanjing
‘I will have left Nanjing tomorrow.’
The ungrammaticality of (96) is quite straightforward under my analysis of –le and PTAs. The LF and translation of (96) are as follows.
(97) LF: [[IP1 wo [IP2 mingtian [AspP le [VP likai Nanjing]]]]
a. [[VP]] =: λt[leave’(Nanjing’)(x)(t)]
b. [[AspP]] =: λP<i,t>λt2λt1∃t3∃t4[t3< tpro & t3 = finitial(t4) & ftarget(P)(t4) & P(t1) &
t1⊆ t2](λt[leave’(Nanjing’)(x)(t)]) =: λt2λt1∃t3∃t4[t3< tpro & t3 = finitial(t4) &
ftarget(λt[leave’(Nanjing’)(x)(t)])(t4) & leave’(Nanjing’)(x)(t1) &
t1⊆ t2]
c. [[IP2]] =: λP<i,<i,t>>λt2λt1[t2 ⊆ tomorrow’ ∧ P(t2)(t1)]( λt2λt1∃t3∃t4[t3< tpro &
t3 = finitial(t4) & ftarget(λt[leave’(Nanjing’)(x)(t)])(t4) &
leave’(Nanjing’)(x)(t1) & t1⊆ t2])
=: λt2λt1[t2 ⊆ tomorrow’∧ ∃t3∃t4[t3< tpro & t3 = finitial(t4) &
ftarget(λt[leave’(Nanjing’)(I’)(t)])(t4) & leave’(Nanjing’)(I’)(t1) &
t1⊆ t2]]
=: ∃t2∃t1[t2 ⊆ tomorrow’ ∧ ∃t3∃t4[t3< Now & t3 = finitial(t4) &
ftarget(λt[leave’(Nanjing’)(I’)(t)])(t4) & leave’(Nanjing’)(I’)(t1) &
t1⊆ t2]]
From the last translation in (97), it is clear that the logical form is a contradiction, because on the one hand it requires that the initial subinterval of the target state, i.e., t3
must precede the speech time— i.e., I must have left Nanjing already by the speech time, but on the other hand it says that the event of leaving Nanjing is included within tomorrow. Similarly, if tpro refers to tomorrow, another time that may be made salient by the context, a contradiction arises. (96) is thus correctly predicted to have no meaning at all under my analysis of -le.
Despite examples like (96), Dai (1994) has observed that the verbal -le may actually occur with a future time adverbial as illustrated in (98).
(98) Mingtian (zhege shihou) wo yinggai yijing likai-le nanjing tomorrow this moment I should already leave-Asp Nanjing ‘I should have already left Nanjing by (this moment) tomorrow.’
The question is why (98) contrasts with (96) and how this contrast should be accounted for. Before explaining the contrast, I want to call the reader’s attention to an interpretational difference between the time adverbial in (96) and the one in (98).
While mingtian ‘tomorrow’ in (96) is interpreted as an interval within which the event takes place, i.e., it modifies the event time19, the same adverbial in (98) is understood
19 This reading is confirmed when the time adverbial mingtian ‘tomorrow’ in (96) is replaced by a time adverbial indicating a past time such as zuotian ‘yesterday’.
as a reference point before which the event takes place, i.e., it fills the value of tpro
introduced by -le. Another point about (98) is concerned with the relative position between the time adverbial and the modal verb. When the time adverb in (98) is placed after the modal verb, the sentence becomes ill-formed. This is shown by (99).
(99) *Wo yinggai mingtian (zhege shihou) yijing likai-le nanjing I should tomorrow this moment already leave-Asp Nanjing
‘I should have already left Nanjing by (this moment) tomorrow.’
Notice that time adverbials can actually appear after modal verbs but they must be construed as modifying the event time rather than serving as a reference point.
(100) Ni yinggai mingtian (zhege shihou) likai nanjing Ni should tomorrow this moment leave Nanjing
‘You should leave Nanjing tomorrow at this time.’
Finally, to obtain a reference point reading for (98), the adverb yijing ‘already’ seems obligatory. Thus, if yijing ‘already’ in (98) is deleted, the sentence becomes ill-formed.
This is shown by (101).
(101) *Mingtian (zhege shihou) wo yinggai likai-le nanjing tomorrow this moment I should leave-Asp Nanjing
‘I should have left Nanjing by (this moment) tomorrow.’
But if the verbal –le in (101) is replaced by the sentence-final le, the sentence becomes acceptable and the time adverbial can be construed as the value of tpro. In such cases, either a modal auxiliary or yijing ‘already’ is obligatory.
(102) Mingtian (zhege shihou) wo yinggai likai Nanjing le tomorrow this moment I should leave Nanjing ASP ‘I should have left Nanjing by (this moment) tomorrow.’
Summarizing the above discussion, in Chinese, when a time adverbial is interpreted as a reference time before which an event occurs, it must be attached to a position higher than a modal verb or the sentence-final le. From this I conclude that Chinese temporal adverbials are ambiguous. They may have a meaning of type i as when they appear in examples like (98) and (102) or a meaning of type <<i<i,t>,<i,<i,t>>> as when they are construed as modifying the event time (cf. Stump 1985, Thompson
1994). Which interpretation they have seems to depend upon two factors: (i) their syntactic position in tree structure and (ii) the presence/absence of a modal auxiliary, yijing and/or sentence-final le. A type i interpretation of a temporal adverbial is possible only when there is overt evidence such as a modal auxiliary or yijing
1994). Which interpretation they have seems to depend upon two factors: (i) their syntactic position in tree structure and (ii) the presence/absence of a modal auxiliary, yijing and/or sentence-final le. A type i interpretation of a temporal adverbial is possible only when there is overt evidence such as a modal auxiliary or yijing