Non-canonical word order and temporal reference in Vietnamese

2.1 Duffield (2007): Wh-dislocating for scope evasion

Duffield (2007) makes many theoretical claims about Vietnamese, one of which is the assumption that Vietnamese separates Tense from Assertion (according to Klein 1998, 2006, finiteness is composed of tense and assertion, and assertion can be realized independently of tense). This claim is the core of his account of the temporal effect. For illustration, consider the examples in (3).^[3]

Similar to (1), the questions in (3) are composed of identical material, with minimal contrast: the ‘when’-phrase occurs sentence finally in (3a), and sentence initially in (3b). These questions display distinct temporal locations: the former is temporally located in the past, while the latter is in the future. What brings about this contrast on Duffield’s analysis? In short, Vietnamese projects tense and assertion separately, having the Assertion Phrase take the vP as its complement, the effect of which is that the material within the scope of the Assertion head, namely the vP, is construed as denoting past events. It is important to note that according to Duffield a ‘when’-phrase, being temporally underspecified, obtains its temporal specification from its syntactic position. As a consequence, if it stays inside the scope the assertion, it is specified with the positive past [+PAST]; if it moves out of the Assertion Phrase (to evade the assertion scope in Duffield’s terms), it is marked with the negative past [−PAST]. It serves in the former as an enquiry about a past event, and in the latter as an enquiry about a non-past or future event. Temporal expressions, being inherently temporally specified, may appear freely either inside or outside the scope of the assertion operator. Diagram (4) represents Duffield’s scope evasion analysis.

2.2 Empirical issue

Duffield (2007) accounts nicely for the correlation between temporal reference and location of the ‘when’-phrase as displayed by examples such as (3), and (5) given below, but it fails to explain the lack of this correlation in a wide array of cases that involve atelic (unbounded) events.

It is known in the literature that argument and complement expressions contribute to the aspectual composition of the clause (Dowty 1979; Jackendoff 1991, 1996; Krifka 1989, 1992, 1998; Verkuyl 1972, 1993; among others). In Duffield’s example, the verb phrase involving the consumption verb ‘watch’ and the definite NP ‘that film’ conveys a telic reading. If we replace the full-fledged NP object ‘that film’ with a bare NP ‘film’ as in (6), an atelic reading obtains, and precisely due to this change the temporal effect disappears.

Before discussing further a few words are in order about the interpretative variation of bare NPs in Vietnamese. A bare NP like chó ‘dog’ can be interepreted either as a definite singular/plural or as an indefinite singular/plural NP. For instance, it can be used by an owner to refer to the one dog or the dogs he/she has, or by an observer informing other observers that he/she spots a dog or many dogs in the forest. For simplicity, just keep in mind that the bare NP in (6) can be interpreted either as a plural indefinite ‘films’ or as a plural definite NP ‘the films’. It is on the indefinite reading of the bare NP that the events denoted in (6) are atelic. Consequently, the temporal effect disappears, as indicated by the translations where the first reading of (6b) means the same as (6a). The questions are then about the hearer’s habit, meaning more or less like ‘when do you go to the movies/watch movies?’ The absence of the temporal effect is attested by the fact that it is natural to use an unspecified temporal expression like ‘Whenever I am free’ or ‘In my free time’ to answer the questions in (6).^[4]

By the same token, replace the subject pronoun ‘you’ in Duffield’s examples with a bare NP sinh viên ‘student’, and the temporal effect fails to emerge.

Analogous to the examples in (6), (7a) obtains a non-past temporal reference regardless of the (in-) definiteness reading of the bare NP subject, but (7b) is ambiguous, depending on the (in-) definiteness of the bare NP. The temporal effect arises if the bare NP subject is construed as definite, as indicated by the contrast between the second reading of (7b) and (7a). In the former the ‘when’-phrase occurs in final position giving rise to a past tense construal, in the latter it is in the initial position, hence non-past. Nevertheless, the temporal effect is absent if the bare NP is taken as being indefinite, as shown by the parallel between the first reading of (7b) and (7a): The ‘when’-phrase in final position does not bring about the past tense interpretation. This reading is salient in a context where the film under discussion is so highly restricted that students are recommended to watch it at some time after they finish their study. If the speaker wants to know when they will watch the film, he/she can use either (7a) or (7b).

The data above shows that it is not the location of the ‘when’-phrase, but the NP arguments, that play a critical role in bringing about the temporal effect. To be more precise, the temporal effect is sensitive to the telicity of the clause.

For instance, a question about a generic event, formed by an atelic clause, as shown in the exchange in (8), taken from Cao (1998) [gloss and translation mine], does not exhibit any temporal effect: The question in (8) obtains same temporal reference with the ‘when’-phrase in the initial position. Recall the absence of the temporal effect in questions with atelic clauses given in Section 1. Note that the example in (8), a clause with the change of state verb ‘bloom’, and the when-phrase ‘which season’, does not express a telic event, but if we replace the bare NP hoa cúc ‘chrysanthemum’ with a full-fledged NP cây hoa cúc đó ‘that chrysanthemum plant’, and another when-phrase like lúc nào ‘which time’, the clause obtains a telic reading, thus confirming that telicity is a clausal property, not a verbal property.

Next, Duffield’s assumption that ‘when’-phrases, being underspecified with respect to temporal values, hinge on their syntactic position for temporal specification, while temporal adverbials, being temporally specified, can occur freely is not attested. First, if this temporal effect was due to the temporal under-specification of the ‘when’-phrase, then what brings about the same temporal effect of argument wh-questions? But before we proceed, a few words on word order in Vietnamese are necessary. Vietnamese is a wh-in-situ, SVO language. Durative and referential or clock-calendar adverbials, and locational adverbials occur in final position. However, as discussed later on, deictic adverbials are syntactically constrained depending on their temporal value. It follows naturally that canonical argument wh-questions, ‘where’-questions, and ‘when’-questions with ‘when’ ranging over referential time, and ‘how long’-questions require wh-phrases to stay in the base position of their counterparts in declaratives.^[5] However, other adjunct wh-questions behave differently. For instance, a ‘why’-phrase is base generated initially, but a ‘how’-phrase finally (Bruening and Tran 2006). Now consider the examples in (9), where (9a) is a canonical question (a wh-in-situ question), and (9b) is a non-canonical question (a wh-ex-situ question). To see the contrast, imagine a context where a person is informed that Nam just put something into the cupboard, but is not certain what it is. If that person wants to know what Nam put into the cupboard, he/she will utter (9a), not (9b). By contrast, in a context where someone, upon being informed that Nam does/will not put certain things in the cupboard, wonders what Nam will put/puts in the cupboard, then the question he/she will ask is (9b), not (9a).^[6]

The first reading of (9a) and the licensed reading of (9b) indicate that the temporal effect is operative. Interestingly, the possibility of the second reading of (9a) suggests that the temporal effect can be neutralized or absent, a pattern displayed by when-questions with atelic predicates as in (8).

The argument wh-questions in (10) with cái gì ‘what’ pose a comparable pattern: While the wh-in-situ question is ambiguous, the wh-ex-situ question is not. It is less likely, but possible for (10a) to obtain a non-past reading, in addition to the salient past reading; yet, (10b) does not allow past reading. We witness again the absence of the temporal effect: the second reading of the wh-in-situ (10a), and the legitimate reading of the wh-ex-situ (10b) are equivalent.

We can explain the absence of the temporal effect of argument wh-questions by assuming that just like a bare NP, a wh-phrase like ‘what’ or ‘who’ is ambiguous between indefinite and definite reading. That is why it is natural to answer (10a) either with a specified expression like ‘The apple you bought’ or with an unspecified expression like ‘Apples’: the former giving rise to a telic reading, the former an atelic reading. On the atelic reading, the temporal reference remains the same, no matter whether the wh-phrase is in situ or ex situ. It is therefore plausible to hold that the temporal effect of argument wh-questions is sensitive to telicity.

Finally, contrary to Duffield’s prediction that temporal adverbials, being temporally specified, can occur either sentential finally or sentential initially, it is known in the literature (see Cao 1991; Tran et al. 1940; Truong and Nguyen 1963; among others) that temporal adverbials in Vietnamese surface in different locations, depending on whether they denote past or future times: While past time adverbials can appear sentence initially or sentence finally, future time adverbials must occur sentence initially. Examples (11), taken from Cao (1991) [gloss and translation mine], are illustrative.

The ill-formedness of (11b) due to the occurence of the future time adverbial ‘tomorrow’ in the final position is a mystery on Duffield’s analysis, so is the ungrammaticality of (12c) with the future time adverbial ‘next year’. Note, however, not all temporal adverbials obey this constraint. Referential adverbials like lúc ba giờ ‘at 3 o’clock’ can occur either initially or finally. We come back to this issue in Section 4.2.

In summary, Duffield’s scope evasion analysis is empirically unattested and inadequate, but its bright side lies in the idea of associating temporal interpretation with the vP. Since the temporal effect is also found in non-interrogatives, as demonstrated the next section, it is appealing to have a unified analysis for interrogatives and non-interrogatives, yet the phenomenon turns out to be less straightforward than it seems. However, what is consistent is the fact that the temporal effect is sensitive to telicity, which in turn suggests that an aspect-based investigation into Vietnamese temporal reference appears to be a step in the right direction.

3.1 Temporal reference of bare sentences

Vietnamese employs a few aspect/tense particles, but in natural discourse their occurrence is quite low in frequency. It is far more frequent to come across texts consisting mainly of bare sentences, sentences with lexical material only. The question is how do bare sentences obtain their temporal interpretation? Chinese, an isolating language, gives rise to the same question. Consider the data below taken from Lin (2003:262–263), where CL stands for classifier.

Lin (2003) argues that bare sentences as in (13) provide no evidence for the projection of TP because their temporal locations can be accounted for with or without covert tenses. On the covert tense analysis, a head tense is covertly realized, and so is the viewpoint aspect. Following Bohnemeyer and Swift (2001), Lin proposes that the default viewpoint aspect of a telic description is perfective, while that of an atelic description is imperfective. The value of the covert tense is determined by a selectional restriction, according to which the covert present tense selects the imperfective AspP, and the covert past tense selects the perfective AspP. Consequently, the telic sentence in (13a) is construed as denoting past tense because it involves a covert past tense that selects a covert perfective aspect, whereas the atelic (13b) is interpreted as present tense since it includes a covert present tense that selects a covert imperfective aspect. The alternative analysis without covert tenses can account for these facts by assuming that the topic time of bare sentences is the speech time. By pragmatic reasoning, the topic time of a non-future perfective durative sentence must be a past interval because a durative event cannot be included in the speech time. Comparatively, the topic time of a non-future perfective achievement event must be in the past because such an event must be completed before one can talk about it. The present tense construal of an imperfective sentence is obvious since the topic time/the speech time is included in the event time.

Lin’s non-covert tense analysis is in line with Smith and Erbaugh (2005) pragmatic analysis, according to which Chinese follows a default pattern of temporal form and interpretation, characterized in (14).

In terms of Vendler’s (1957) verb classification, (14a) and (14b) amount to saying that sentences denoting states and processes/activities are by default construed as present tense, while (c) indicates that achievements and accomplishments denote past events.

It is obvious that either the default temporal interpretation by Smith and Erbaugh (2005) or the covert tense analysis as in Lin (2003) can account for the data in (13), and it is equally plausible to use either of them to account for the Vietnamese data in this paper. However, given the limited scope of the paper we opt for an extension of Smith and Erbaugh’s (2005) temporal reference pattern to Vietnamese and leave it open the issues raised by a commitment to the existence of covert tense. First, a covert tense analysis will need to account for the absence of a phenomenon referred to as the lifetime effects (Arche 2006; Mittwoch 2008; among others): the correlation between the temporal interpretation and individual-level predicates. An individual-level predicate in the present or past tense triggers an inference (defensible) about the life or death of an individual. This is illustrated by the examples below, where (15a) licenses the inference that John is alive, while (15b), in out-of-the-blue contexts, licenses the inference that John is dead. Consequently, contradictory inferences arise when the subject of an individual predicate denotes an existing entity and a dead one, as in (16).

If a covert tense existed in Vietnamese, either present or past tense, the Vietnamese equivalent of (16) ‘Michael Jackson và Bob Dylan là hai nhạc sĩ người Mỹ’ should be problematic, contrary to fact. Second, a covert tense with a non-future value in the spirit of Mathewson (2006) would also find it difficult to explain the well-known fact reported in reference books, and formal study: bare sentences obtain future reading when modified by future-time temporal adverbials as illustrated in the preceding data and in (17). If a non-future covert tense was available in Vietnamese as it is in St’at’imcets, then on Mathewson (2006: 677) these sentences would be ill-formed.

Now consider how bare sentences in Vietnamese obtain temporal interpretations. Uttered out of the blue, it is impossible to interpret (18a) as expressing that the state of Nam liking Nga no longer holds at the speech time. Also impossible is the use of (18b), with the non-bare object NP quả táo ‘CL apple’, to express Nam’s habit of eating apples after school. To convey this habit, (18c) must be uttered, where the bare NP táo ‘apple’ is used. So, it is (18c), not (18b), that can be used as a continuation of ‘It is usual that after coming home from school…’

While (18a) confirms that state denoting sentences are temporally located at speech time, the contrast between (18b) and (18c), points to a critical issue, mentioned in Section 2, that is, that the structure of the argument NPs can affect the predicate telicity, and that event type is a clausal property, not a verbal property. This is well known in the literature. Krifka (1992) proposes that telicity is composed of the verb and its arguments. For instance, ‘run’ is an activity denoting verb, without a set terminal point (STP), and is an atelic predicate: run ⊆ ε∧¬STP (run). But when it is modified by a measure phrase ‘a mile’ it has a set terminal point, and is telic: run.a.mile ⊆ ε∧ STP (run.a.mile). Briefly, creation/consumption verbs give rise to telic reading (19b) with quantized reference arguments, and yield atelic reading with cumulative reference arguments (19a).

Along the same line, Verkyul (1972, 1993 provides a theory of aspectual composition to compute the telicity of the predicate and its arguments. In principle, aspectual composition, according to Verkyul, processes bottom-up, based on the value of the feature [ADD TO] of the verb, and of the feature [SQA] of the argument. Verbs that express dynamic, progress, change, non-stativity are assigned with the positive value of feature [ADD TO], hence [+ADD TO]; stative verbs are assigned with the negative value, namely [−ADD TO]. The feature [SQA] of the argument expresses whether the NP conveys a specified quantity of things or mass denoted by the head noun. If affirmative, it is assigned with the positive value [+SQA], otherwise it is [−SQA]. Formal definitions of how the values of [SQA] are determined are given in (20).

Finally, for the machinery to work, Verkuyl advances the Plus-Principle, which says that one minus-value below is sufficient to yield a [−T] at the top (T=Telicity). The Plus-Principle is thus to guarantee that a clause with a [+ADD TO] verb is specified with [+T], or telic, only if all of its arguments are specified with [+SQA].

Illustrative examples are given in (21).

An example of how telicity is composed in Vietnamese on Verkuyl’s theory is given in (22).

Given the [+ADD TO] achievement verb ‘arrive’, the telicity of (22) is dependent on the value of the bare NP. If it is taken as a definite NP, hence [+SQA], the sentence will be marked [+T], or telic, and is interpreted as past tense (22i). The sentence is felicitous as an answer to a question about the arrival time of a train known by both speaker and hearer. If it is construed as an indefinite NP, hence [−SQA], referring to an unknown number of trains, then sentence (22) is marked [−T], or atelic, and is non-past in temporal reference (22ii). The sentence is appropriate as a general statement about the arrival time of trains at a station.

In brief, Vietnamese demonstrates that telicity is a clausal, not verbal, property in that it is composed of the verb and its arguments/complements. It is plausible to extend either Krifka’s or Verkuyl’s to Vietnamese. For instance, a clause involving a dynamic verb and cumulative/unspecified arguments gives rise to an atelic event, while a clause involving a dynamic verb with quantized/specified arguments yields a telic event. Temporal reference of bare sentences is derived from Smith and Erbaugh’s (2005) default deictic pattern of temporal form and interpretation: achievement and accomplishment (telic) events are construed as past events; by contrast, states and activities (atelic) are non-past.

3.2 Temporal reference in non-canonical sentences

We observed in Section 2 that the temporal effect of ‘when’-questions and argument wh-questions is sensitive to telicity. Interestingly, non-interrogatives display similar behavior as illustrated by the examples below. The past interpretation of the first reading of the canonical (23a) and the non-past reading, the legitimate reading of the non-canonical (23b) strongly suggests that the temporal effect is at work. However, the fact that it is possible, though less likely, to grant the canonical sentence (23a) a non-past reading indicates that the temporal effect is absent when the predicate is atelic. This fact is parallel to what we observed in the argument wh-questions in (9), Section 2. In natural discourse, it is more appropriate to take (23b) as expressing a generic event, or a habit, but is less likely to do so with (23a). (23a) is instead prominently construed as denoting an episodic event in the past. Note again that the availability of these readings is dependent on the (in-) definiteness of the bare NP ‘money’ as rendered by the translation. Obviously, (23b), not (23a), is a felicitous continuation of a preceding sentence like ‘It is not Nam’s habit to put jewelry or gold into the cupboard, but…’ Meanwhile, a sentence like ‘Nam received the money, and then…’ should be followed by (23a), not (23b).^[7]

Comparable to (23), the examples in (24) indicate further that the temporal effect is conditional on the telicity of the clause. When the canonical sentence denotes an atelic event, there is no temporal effect, as illustrated by the second reading of (24a) and the legitimate reading of (24b), where the bare NP is construed as indefinite, giving rise to an atelic reading. By contrast, the first reading of (24a) and the legitimate reading of (24b) suggest that the temporal effect is operative. On this reading, the bare NP in (24a) is construed as definite, hence a telic reading obtains.

In summary, interrogatives and non-interrogatives display a common pattern regarding the temporal effect in that they are sensitive to telicity. The temporal effect is not operative in a declarative or a wh-question with atelic description: A change in word order (NP dislocation or wh-ex-situ) does not lead to a change in their temporal reference.^[8]

4.1 Accomplishment realization and telicity

We have witnessed the neutralization (or the absence) of the temporal effect in a sentence with an atelic/unbounded event description: a non-past interpretation obtains regardless of whether the wh-phrase, be it an argument or an adjunct when-phrase, or the argument NP stays in situ or is dislocated. Observe again the sentences in (23), disregarding the case where the bare NP in (23a) is construed as indefinite, neutralizing the temporal effect. The legitimate reading of (23b) expresses an activity of an unspecified length: Nam puts money in the cupboard in the past, and continues doing so at present and in the future, with the amount of money being unspecified. Consequently, (23b) does not state the temporal extent of the activity. By contrast, the first reading of (23a) conveys an activity of Nam putting the money into the cupboard, with the amount of money being identified due to the definite reading of the bare NP tiền ‘money’, plus a logical culmination such that the specified money ends up being in the cupboard. This reading expresses an accomplishment event, a complex event the structure of which is formalized as below by Rothstein (numbered 38 in Rothstein 2004), where x and y stand for the content of the activity and BECOME event.

An accomplishment event involves a sum of two small eventualities, e₁ and e₂, with e₁ being an activity, and e₂ a ‘becoming’ event, an incremental event that accompanies the activity. The theme of the activity e₁ is the incremental argument of the BECOME event e₂, and e₁ is incrementally linked to e₂, with C (e₂) indicating the culmination of e₂, the set of upper bounds of e₂.

In brief, the activity event denoted by (23b) differs from the accomplishment (23a) in that the latter denotes a complex dynamic event with a culmination point or set terminal point: the event comes to an end with all the amount of money being in the cupoard, whereas the former denotes a dynamic event without a culmination point: we do not know when the event is completed. Thus (23b) expresses a single event consisting of a series of Nam’s putting money into the cupboard. It is assumed that the lack of the culmination point of (23b) results from the blocking of the construction of the complex event of accomplishment.

To see how the blocking occurs, consider more examples that involve activity denoting sentences ‘John pushed the ball’, ‘Mary wiped the table’, and accomplishment denoting sentences that comprise verb-particle constructions and adjectival resultative constructions like ‘John pushed the ball out’, ‘Mary wiped the table clean’, respectively. Snyder (2012) analyzes these English verb-particle constructions and adjectival resultatives as involving a specific rule of semantic composition, and proposes a parameter, the Compounding Parameter. Languages differ as to whether the Compounding Parameter setting is positive or not. The positive setting of this parameter is responsible for endocentric compounding as a creative process through an operation called Generalized Modification (GM). For instance, speakers of English can easily grasp the meaning of a newly created word like ‘frog chair’, [_N frog [_N N chair]], as a kind of chair that stands in some contextually relevant relation to the denotation of the word ‘frog’. Simply put, [+TCP] languages make available endocentric compounding and constructions that give rise to accomplishments, similar to English verb-particle constructions and adjectival constructions.

If α and β are syntactic sisters under the node γ, where α is the head of γ, and α denotes a kind, then interpret γ semantically as a subtype of α’s kind that stands in a pragmatically suitable relation to the denotation of β (Snyder 2012:285).

For instance, the GM configuration that results in an adjectival resultative construction (26a) is given in (26b), where the head α is the verb ‘wipe’ and β is the resultative adjective ‘clean’. The meaning γ of the construction is given in (27).

It is highly plausible to extend this analysis to cases that involve verbs of path- -of-motion (28) and path expressions.

The activity event of (28a) is a simplex event of cart pushing; the accomplishment event of (28b) involves two eventualities, the cart pushing activity and the state of the cart being out of the yard. The GM configuration of (28b) is given in (29a), with α being the head ‘push’, β the small clause, the meaning γ resulted from GM is given in (29b). The accomplishment event is described as in (29c).

Back to Vietnamese, the data in (30a) and (30b) indicates that Vietnamese makes use of compounding as a productive means of word formation, and makes available adjective resultative constructions. Note that the connection between compounding and resultatives in Vietnamese is morphological: the availability of the compound verb in (30c) indicates that the verb and the adjective form a complex word. This connection in English, according to Snyder (2012:289) is semantic. Therefore, it is conclusive that Vietnamese is a [+TCP] language.

Accordingly, I propose that Vietnamese employs Generalized Modification (GM) to drive accomplishment predicates from activity verbs and predicates of location or path, or resultative adjectives. Differing from English, Vietnamese requires the compounding constituents, namely the head α and its sister β, to be fully realized and in sisterhood for GM application. Otherwise, GM does not apply and the accomplishment predicate does not obtain. With this constraint in place, the contrast in telicity between (23a) and (23b) can be explained as follows. The application of GM that results in the construal of the node γ as an achievement predicate is operative in (23a), as represented in (23’a), where the head verb ‘put’ and its sister, the small clause, are fully realized. By contrast, GM does not apply in (23b), as represented in (23b’), where the small clause is not fully realized given the NP ‘money’ being dislocated. Hence, the failure of accomplishment realization follows.^[9]

This assumption finds its indirect support from Zubizarreta and Oh (2007). The authors present and discuss the following example (numbered 440 in their work).

The sentence in (440) allows two possible readings due to the ambiguity of the PP ‘in the bag’. On the locative reading of the PP, it conveys an activity event interpretation, according to which John is inside the bag and is squeezing the clothes. On the directional meaning of the PP, it expresses an accomplishment event, meaning ‘John made the clothes go into the bag by squeezing them.’ What is relevant to our discussion is that when the PP is topicalized as in (441a), the sentence is disambiguated and allows only the activity reading. The unavailability of the accomplishment can be accounted for by the assumption that the dislocation the PP ‘in the bag’ blocks GM.

Parallel examples in Vietnamese are given in (31).

It is obvious that (31a) expresses a complex dynamic action, whereas (31b) is preferably taken as describing an activity. This is due to their difference in telicity. (31a) licenses a directional meaning, and as such it gives rise to an accomplishment reading that includes the activity event of Nam hanging the painting(s), and the becoming event of the painting(s) being on the wall. By contrast, (31b) allows for the locational reading only, hence the atelic. This contrast is borne out by the fact that (31a), not (31b), can be used as a continuation of a sentence describing a dynamic event as in (31c).

The infelicity of (31b) in forming a mini-discourse with (31c) is explained away, based on discourse research on aspect and narrative time. Dry (1981), and Smith (2003) point out that sentences that express bounded events, or achievement/accomplishment events, propel the narrative time, whereas stative and activity or unbounded sentences do not, in general. It is obvious that the dynamicity of the first sentence in (31c), plus the requirement of the adverb ‘immediately’ that the sentence it modifies must denote a dynamic event that instantly follows the first event, triggers the expectation that the following sentence must advance the narrative time. A sentence conveying an unbounded event as in (31b) does not propel narrative times. Hence, it is odd to continue (31c) with (31b).

The same explanation applies for the temporal effect of the dislocation of the argument NP as in (31d): like (31b), (31d) cannot be used as a continuation of (31c).^[10]

Back to example (24), it is helpful to look at the analogous examples, given in (32), that involve a telic particle, hết, glossed as T-PART in Fukuda (2007), and a non-bare NP. The contrast in the event structure of these sentences is interesting: (32a) conveys an accomplishment event, while (32b) expresses what seems like an activity event. The fact that the verb ‘eat’ even with the support of the telic particle in (32b) does not express a telic (accomplishment-like) event suggests that the predicate ‘eat’ in (24) is basically an activity denoting verb. On Krifka (1992), the telicity of (24a) is dependent on the direct object argument: if the bare NP is construed as a quantized argument, it is telic; if it is construed as a cummulative, it is atelic. The telic reading of (32a) is obviously derived from the quantized arguments ‘the three apples’. The atelic reading of (32b) can be explained if we assume that aspectual composition is a operational in surface syntax: Fronting the object NP blocks aspectual composition, as a result of which the activity reading remains unchanged.

Note that replacing the non-bare NP in (32) with the bare NP ‘apple’ does not change the telic-atetic flip-flop, but adds extra meaning to (32b′), as shown by the translation. Crucially, (32’a) states that Nam ate all the apples, whereas (32b′) expresses that whenever Nam eats apples, not other fruits, he eats them all.

In brief, canonical declaratives and wh-argument question can be aspectually telic or atelic because wh-phrases are analogous to bare NPs in that they are ambiguous with respect to definiteness or [SQA], (in Krifka’s terms, it can be either referentially quantized or cumulative). The telicity of a canonical argument wh-question or a declarative obtains with the wh-phrase or the bare NP being construed as definite or positively valued [+SQA] (quantized). By contrast, dislocating NPs and wh-phrases blocks GM (Generic Modification) or aspectual composition, as a result of which the remnant material is construed as denoting activity, or atelic/unbounded event. The dislocated wh-phrases and NPs are construed as indefinite or negatively valued [-SQA]. Table (33) summarizes what we have discussed.

(33) Word order, Generic Modification/Aspectual Composition, and Telicity

The following subsection shows that the temporal effect of ‘when’-questions is just superficial, but tied to the constraints on temporal adverbials, and that the location of the ‘when’-phrase does not contribute to the event structure or telicity. The temporal specification of a ‘when’-phrase is determined by its role in the temporal modification configuration.

4.2 The syntax of temporal adverbials

Temporal adverbials, in general, are known to be sensitive to the lexical aspect of the sentence they combine with (Dowty 1979; Kamp and Reyle 1993). For instance, an atelic predicate sentence must be modified by a temporal adverbial of the form ‘for temporal NP’, not ‘in temporal NP’: ‘He walked for two hours/*in two hours’, whereas a telic predicate sentence must be modified by an ‘in temporal NP’, not ‘for temporal NP’: ‘He walked to the store in two hours/*for two hours’. Given the telicity sensitivity of the temporal effect, it is expected that answers to the ‘when’-questions under investigation involve only deictic adverbials. Therefore, it is worth considering again the restrictive distribution of deictic adverbials in Vietnamese: future time adverbials must occur sentence initially, whereas past time adverbials are not subject to this restriction (see Cao 1991, 1998; Tran et al. 1940; Truong and Nguyen 1963). It is highly likely that an account for this constraint can be extended to the temporal effect of ‘when’-questions. But first let us investigate the syntax of deictic adverbials in Vietnamese.

Demirdache and Uribe-Extebarria (2004) analyze the semantics and syntax of temporal adverbials as being parallel to that of nominal modification; hence the term temporal modification. In nominal modification the modifier is base-generated as an adjunct to the NP it modifies, and thus restricts the reference of the NP, as illustrated in (34a). Modification is established via predication: the modifier is predicated of its sister NP. Analogously, in temporal modification, the temporal adverbial is base-generated as an adjunct to the sentence, and restricts its temporal reference. But how is the temporal modification realized? Basically, according to Demirdache and Uribe-Extebarria, a sentence containing a temporal adverbial as in (35) involves a dyadic predicate, represented by the preposition ‘in/after/before’ or a null ‘in’. The predicate establishes a topological relation (inclusion, subsequence, or precedence) between the two time denoting arguments, two Zeit Phrases. The external argument is the AST-T (assertion time) or EV-T (event time) of the sentence, and the internal argument is the temporal adverbial. Demirdache and Uribe-Extebarria assume that when either the head T or head Asp is without morphological content, its external temporal argument binds its internal temporal argument. Hence, in the example (35), where the head Asp is null, the assertion time is co-temporal with the event time.^[11]

I propose, following Demirdache and Uribe-Extebarria (2004)’s temporal modification, that a sentence like (36a) involves a null predicate ‘in’, establishing an inclusion relation between two time denoting arguments such that external argument is included in the internal argument, where the former is the bare sentence Nam cưới cô ấy ‘Nam marry she’ and the latter is the temporal adverbial năm ngoái ‘last year’. The diagram in (36b) is illustrative.^[12]

The next question is how does the temporal reference in (36) obtain? Before we try to answer this question, let us first look at the syntax and semantics of nominal modification in Vietnamese. Vietnamese allows a [NP NP ] nominal modification configuration, in which the meaning of the compound is dependent on that of the initial head (37a). Switching their order alters the meaning of the compound, as illustrated in (37b).

Extending this analysis to the interpretation of the temporal modification, the temporal interpretation of (36) is derived from the temporal interpretation of the bare sentence ‘Nam marry she’, the head, and the deictic adverbial ‘last year’, the modifier. Since the bare sentence conveys a bounded event, it is by default located in the past. The temporal modification, as illustrated in the diagram in (38), restricts the event time by establishing that it is included in the time denoted by the deictic adverbial.

The distributive constraint on future denoting adverbials, as shown in (39a), can be explained, based on the analysis above, as follows.

As shown in (39b) and (39c), the time denoting argument, the head, represented by the bare sentence ‘Nam marry she’, is construed by default as being located in the past on the temporal axis, marked as [+P], while the deictic adverbial is located in the future [−P]. The temporal modification fails because they are not in the same side on the axis, resulting in ungrammaticality. This explanation leads to the question of why (40a) is grammatical given that it lacks the compatibility between the temporal adverbial and the default temporal interpretation, supposedly giving rise to the same conflict observed in (39a)? In other words, the question is why does the temporal information given by the deictic adverbial ‘next year’ override the default temporal reference in (40a), not in (39a)?

This question implicates that the pragmatic rule must be revised. For the pragmatic rule (14d), that is, that explicit temporal information may override default temporal reference, to be operative, the deictic adverbial must be in a syntactic configuration that projects its temporal value. In the case of Vietnamese, it must be the head of the temporal modification. As illustrated in (40c), the head [−P] ‘next year’ projects its value [−P], and hence overriding the default past interpretation denoted by the bare sentence.

Unlike future denoting adverbials, past denoting adverbials are not subject to the sentence initial constraint as illustrated in (41a, b). This is because the past denoting adverbials and the bare sentence bear the same temporal value [+P]. The surface difference between (41a) and (41b) is presumably triggered by information structural consideration. That is why, (41b), not (41a), is felicitous as a continuation in a contrastive topic discourse: A asks B: ‘Tell me what Nam did recently.’ B replies: Tôi không biết Nam làm gì trong thời gian gần đây, nhưng … ‘I don’t know what Nam did recently, but….’. Also different is their temporal modification configuration: in (41b) as shown in (41b′), ‘last year’ is the head, and the bare sentence is the modifier; by contrast, in (41a) as displayed in (41a′), ‘last year’ is a modifier, and the bare sentence is the head.^[13]

Further evidence in support of this analysis is provided in what follows. It is known that since individual-level predicates express stable properties, they are incompatible with temporal modifiers. Thus, it is odd to utter ‘Mary knew German last night’, but is perfectly fine to say ‘Mary spoke German last night’ because ‘know’ is an individual predicate, expressing a stative eventuality. However, the verb ‘know’ can be coerced into expressing an achievement event.^[14] The examples below are illustrative, taken from Smith (1997: 18).

The coercion of the stative predicate ‘know the truth’ in (42a) into an achievement predicate in (42b) is thanks to the occurrence of the event modifying adverb ‘suddenly’. It is argued that Vietnamese coerces of a stative predicate into an achievement through temporal modification. However, the grammaticality and acceptability of the coerced sentences varies, depending on the temporal modification configuration they are in. Consider the achievement-stative flip-flop in the following.^[15]

Both (43a) and (43b) are grammatical, yet their acceptability is different. This is because the former retains the stative reading, whereas the latter does not. In other words, while (43a) allows for some degree of ambiguity in event type, (43b) completes the coercion, as indicated by the absence of the stative reading of (43b), and its availability in (43a). Structurally, these sentences are in the same configuration, with the temporal adverbial as the head and the bare clause as the modifier. The configuration of (43a) is (44a), and (43b) is (44b).

Recall that the head overrides the temporal value of the modifier. In this case, the modifier, the bare clause, is ambiguous between the achievement and stative reading, hence its temporal value is either [−P] or [+P]. The overriding mechanism requires the modifier to have a temporal value opposite of that of the head. Consequently, the modifier must take the [−P] value, the default temporal value of a stative sentence. That is why it is not quite acceptable: An individual-level predicate modified by a temporal adverbial.

By contrast, configuration (44b) of (43b) allows only the [+P] value of the modifier (so that it can be overridden by the [−P] head). This leads to the perfect coercion, as indicated by the absence of the stative reading.

The sentences in (45) are in a different configuration, giving a different pattern of acceptability and grammaticality. First, the cancellation of the default temporal reference does not happen because the head is not the temporal adverbial, but the bare sentence. It is arguable that this change leads to an adjustment of the null predicate: in a configuration as in (44a) and (44b) the predicate establishes a relation in which the modifier is included in the head, but in a configuration as in (46a) and (46b), the head is included in the modifier. Therefore, in order to execute the temporal modification, the bare clause in (45a) must take the [+P] value. This is the reason why (45a) is grammatical and obtains an achievement reading only. The configuration of (45a) is (46a).

The ungrammaticality and unacceptability of (45b) is predictable. The sentence licenses neither the stative nor the achievement reading. On the stative reading (ii), the bare clause must choose the [−P] value, the default temporal reference of a stative predicate, to be modified by the temporal adverbial [−P]. A stative predicate modified by a temporal adverbial is ill formed. The configuration of (45b) on this reading is (46b).

On the achievement reading (i), the bare clause takes on the [+P] value, while the modifier is [−P] valued, blocking temporal modification. Briefly, on this reading (45b) is ungrammatical for the same reason that makes (39a) ungrammatical: The event time and the deictic temporal adverbial are not located in the same side of the temporal axis, as indicated in configuration (39c).

To conclude, deictic adverbials locate bounded events by placing the events along the temporal axis, directly determining on which side of the temporal axis they are situated. In Vietnamese, temporal modification is endocentric in which the temporal value of the initial head determines the temporal value of the compound. It is therefore expected that future denoting adverbials are base generated in the initial position, as the head of the compound, to override the past time value (the default temporal interpretation) of the modifier, the bare sentence. Past time denoting adverbial must be base generated in the modifier position, the final position since the head, which is the bare sentence, is by default construed as denoting past time.

Now back to the ‘when’-questions, I assume, that the ‘when’-phrase, following Duffield (2007), is temporally unspecified. Then how do the ‘when’-questions obtain the temporal interpretations that we have observed? Extending the temporal modification analysis above, I propose that the ‘when’-phrase inherits the temporal value from the position it is base-generated. Recall that the default temporal interpretation is overridden only when the temporal adverbial is head of the temporal modification. It follows naturally that a ‘when’-phrase as the head obtains a temporal value opposite of the modifier, the bounded event sentence. By calculation this temporal value is non-past, [−P]. By contrast, a ‘when’-phrase in the modifier position must take on the temporal value of the bare sentence, the head, which is past [+P]

The diagrams in (47) summarize the discussion.