2.1. lǝ and telicity

In Yixing, as mentioned in the introduction, lǝ obligatorily imposes a telic reading on an event. In a lǝ -marked sentence, the most natural internal argument is the one that takes the role of ‘incremental theme’ (Dowty Reference Dowty1991, Rothstein Reference Rothstein2004). An incremental theme is usually an argument that measures up the event, representing a homomorphic mapping between the argument and the event. For example, a glass of wine is an incremental theme in the phrase drink a glass of wine. There is a one-to-one homomorphic mapping between the glass of wine and the drinking event. The consumption of the last drop of wine is also the endpoint of the drinking event. It is in this sense that predicates like drink are termed homomorphic predicates (Filip Reference Filip and Toman1997), which are mainly accomplishment predicates in terms of Vendler’s (Reference Vendler1957) classification. Achievement predicates in Vendler’s (Reference Vendler1957) classification also denote telic events, which often express a change of state that takes place instantly.

In Yixing, achievement and accomplishment predicates with an incremental theme can occur in a lǝ-marked sentence, provided that the quantity and object fronting requirements are met and a telic interpretation always arises, as is evidenced by the compatibility with the ‘in x time’ phrase shown in the introduction (see the examples in (4)). Also, as shown in (5), repeated below, even when the predicate is achievement or accomplishment and the object is a quantity NP, still telicity will not be attested if lǝ is not there.

It should be noted here that the ungrammaticality arises only because ǝ is not compatible with telic reading. In the above examples, the ‘in x time’ adverbial requires the telicity of the event, and lǝ is obligatory, which indicates that lǝ is responsible for telic reading. If the reading of telicity is not required, the sentence with ǝ becomes acceptable:

The above sentence without lǝ is grammatical, and it is obvious that the only difference between this sentence and the ungrammatical one in (10b) is that the adverbial ‘in x time’ that imposes telic reading requirement is not present. Although this sentence has both a quantity incremental theme and an accomplishment predicate, telicity is not attested, evidenced by the fact that the truth condition of this sentence does not require the three apples to be eaten up.

This presents a sharp contrast with English, which will express a telic event as long as the predicate is achievement or accomplishment and the object is a quantity NP:

What further shows the correlation between lǝ and telicity is that even if the sentence has an activity predicate, still telicity will be obligatorily derived:

It is well known, as shown in Vendler (Reference Vendler1957), that activity predicates in English do not denote telicity. For example, (13b) will be very unnatural if the adverbial in 30 minutes is added, showing the difficulty of deriving telic reading from this sentence. However, in the Yixing example in (13a), even without the adverbial sazǝ fǝŋoŋ lidou (in 30 minutes), telic reading is obligatory. The reading is that there is an endpoint of pushing these three carts of goods. For example, Zhangsan’s work is to push the goods away to some place, and in this context, this sentence means that three carts of goods have been pushed to that place. It should be emphasised here that lǝ obligatorily requires contextual information of this type to be compatible with the telic reading, which again shows that lǝ imposes telic reading.

2.2. The effects of lǝ on the object

There are three special issues regarding the effects of lǝ on the object: the quantity requirement, the object fronting requirement (only to definite and bare NPs), and the definiteness requirement (to the bare NP).

Firstly, if lǝ is attached to the verb, the nominal object must have a quantity reading. Whenever the object takes a non-quantity reading (like the readings of mass nouns and bare plurals in English), a sentence will be ungrammatical as long as lǝ is attached. A numeral NP therefore is a legitimate object in the lǝ-marked sentence:

In the above examples, when the object takes the bare plural reading in (14b) or the mass reading in (14c), the lǝ -marked sentence becomes unacceptable.

The quantity condition can also be met by kind-denoting nominals, which further shows that lǝ is only sensitive to quantity reading, regardless of the semantic denoting of the NP:

Another interesting issue regarding the effect of lǝ on the object is that bare NP and definite NP objects have to be fronted to the topic position if lǝ is attached to the verb. Let’s start with the bare NP. (14b) and (14c) seem to show that bare NPs are always rejected in lǝ -marked sentences. However, this is not the case. These examples only show that they are rejected if they stay in the post-verbal position, taking a non-quantity (bare plural or mass) reading. For such bare NP objects to be legitimate, they have to be fronted to the topic position. There are three topic positions in Mandarin Chinese: the sentence initial position, the position after the subject and before the verb, and the complement position of ba. The topic positions in Yixing are exactly the same, except that the counterpart of ba is nɔ. When a bare nominal is fronted to one of these positions, it is compatible with lǝ. Another phenomenon related to this case is that the bare nominal must take a definite reading:

In all the above examples, the bare nominal u (alcohol) not only has to be fronted to a topic position, but also must take a definite and quantity reading. That is, all the above sentences have to mean that a certain amount of alcohol known to the speaker and the hearer has been drunk up. The same constraint also applies in (14b). For this sentence to be grammatical, the bare nominal biŋgo (apple) must move to one of the topic positions, and this noun must mean a certain amount of apples, which is the old information in the context.

In addition to bare nominals, a definite NP object also has to be fronted to the topic position if lǝ is attached:

As we can see, the definite NP g̥ǝ sa ǝ biŋgo (these three apples) cannot stay in the post-verbal position if lǝ is attached to the verb; instead, it must move to the topic position. Note that if lǝ is not there, for example, when it is replaced by the particle ǝ, this movement is not obligatory. The definite NP object can either stay in the base post-verbal position or move to the topic position:

Below we see that other types of definite NPs, such as proper names, pronouns, and NPs modified by a relative clause, are also restricted by this constraint. In all the following examples, if the definite object NP is placed in the topic position, the sentence will be grammatical:

Since the definite and bare NP objects have to be situated in the topic position, it might be argued that there is a correlation between topic construction and this object fronting. However, it shows below that even in a topic construction, as long as there is an indefinite quantity NP in the object position, the sentence is still grammatical, indicating that it is the incompatibility between lǝ and definite/bare NP objects in the base post-verbal position that plays a role hereFootnote ⁵:

lǝ’s effect of object fronting only applies to bare and definite NPs, while numeral NPs are not constrained by this effect. In (14a), for example, the numeral NP sa ǝ biŋgo (three apples) stays in the post-verbal position and lǝ is attached to the verb. Of course, as Chinese grammar generally allows, this numeral NP can move to the topic position if required by specific information packaging requirements. The crucial point here is that only in lǝ -marked sentences, bare and definite NPs – but not numeral NPs – have to move to the topic position.

We summarise the effects of lǝ on nominal objects as follows:

2.3. lǝ and past event reading

In addition to telicity, another piece of temporal information related to lǝ is past event reading. The following examples illustrate this point:

As shown in (24a), when lǝ is attached, even without any past tense adverbial, the sentence obligatorily takes a past tense reading. In these examples, there is no possibility to override the past tense interpretation when lǝ is involved. (24b) shows that even with a habitual temporal adverbial like maetie (everyday), the present tense reading still cannot be yielded; instead, the sentence becomes ungrammatical, showing that the past tense reading imposed by lǝ cannot be overridden. The situation is the same in (24c), where a future temporal adverbial is not compatible with lǝ.

3.1. The syntax of telicity: Borer (Reference Borer2005b)

While telicity is often taken as a lexical property (hence the term ‘lexical aspect’) (Vendler Reference Vendler1957, Filip & Rothstein Reference Filip, Rothstein and Lavine2000, Rothstein Reference Rothstein2004), there are recent studies arguing that telicity is encoded in syntax (cf. Borer Reference Borer2005b, Thompson Reference Thompson2006, MaDonald Reference MacDonald2008, Travis Reference Travis2010), according to which a functional head is responsible for the derivation of telicity. In this paper, we will adopt Borer’s (Reference Borer2005b) theoretical framework, the Exo-Skeletal (XS) Model, on the syntax of events (also see Hu (Reference Hu2018)), which provides a specific account on the nature of telicity.

Like other researchers such as Bach (Reference Bach1986), Rothstein (Reference Rothstein2004), and many others, the XS Model captures the semantic parallelism between the domain of events (vP domain) and that of objects (DP domain). The XS Model takes a step further by specifying two parallel functional structures for events and nominals that explicate the semantic parallelism between these two domains. The functional structures encoding events and objects are EP (event phrase) and DP (determiner phrase). In an extended projection (i.e. functional structure), each functional head specifies an open value, which has to be assigned a range, hence giving rise to the semantic interpretation of the structure.

Range assignment can be either direct or indirect. The direct range assignment involves inserting a functional item in the corresponding functional head. A functional item can be an independent morpheme termed ‘f-morph’. Will in English is such an f-morph which assigns range to the open value specified on the T head. A functional item can also take the form of a bound morpheme termed ‘head feature’, such as the English past tense affix -ed. The indirect range assignment can be instantiated by an adverb of quantification, a discourse operator, and specifier-head agreement. Range assignment via specifier-head agreement means that the open value specified on a functional head can be assigned a range if the phrase in the specifier position contains this range. Borer (Reference Borer2005a, Reference Borerb, Reference Borer2013) postulates that the underlying reason for linguistic variation is often tied to how an open value is assigned the range. For example, variation might arise from whether the range is assigned in the shape of a bound morpheme or a morphologically independent functional item, or whether range assignment is achieved directly or indirectly. This is in line with the account of variation in minimalism, which attributes variation to features in the lexicon, which is often termed the Borer-Chomsky Conjecture, as is discussed in Baker (Reference Baker2008), Roberts & Holmberg (Reference Roberts, Holmberg, Biberauer, Holmberg, Roberts and Sheehan2010), and Borer (Reference Borer2013: 630). While there are various definitions of interpretable and uninterpretable features (Pesetsky & Torrego Reference Pesetsky, Torrego, Karimi, Samiian and Wilkins2007), in general the pair of open value and range is the equivalent to the pair of uninterpretable and interpretable features. Therefore, for ease of exposition, in the rest of this paper, we will use the terms of uninterpretable and interpretable features.

The extended projection, EP, starts from a lexical item, often a verb, which is dominated by several functional heads in a fixed and universal hierarchical structure, represented as follows:

The Asp_Q head in EP is the counterpart of the quantity head in DP and is responsible for the quantification of the event division, which is the source of telic reading of an event. Thus, in the XS Model, telicity comes from the valuation of the quantity feature specified on the Asp_Q head. In languages like English, the valuation of the quantity feature is often achieved via specifier-head agreement, which can copy the quantity value of a quantity DP in the specifier position of the Asp_Q onto the Asp_Q head, thereby giving rise to the interpretation of telicity.

We can take the following examples to illustrate the feature valuation of quantity in EP:

Following the XS Model, in (26a), it is the DP three apples in the specifier of the Asp_QP that provides the interpretable quantity feature to value the uninterpretable quantity feature on the Asp_Q head. The valuation of the quantity feature then gives rise to the semantic interpretation of the telicity of the eating event. On the other hand, in (26b), the bare plural apples does not bear an interpretable quantity feature, which means that, in this sentence, if an Asp_Q head projects, the valuation of the quantity feature cannot be achieved and therefore telic interpretation cannot be derived.

Just like the DP structure, in EP, the functional head specifying the quantity feature is optional. When the Asp_Q head does not project, which is exactly the case of atelic events, a layer of F^sP will appear in the otherwise Asp_Q position, and the [Spec F^sP] position will host a DP that is the theme of the event. Since this paper focuses on telicity, F^sP will not be discussed.

3.2. Capturing telic variation: Indirect vs. direct valuation of quantity feature

As we have seen from the outline of the XS Model, languages have the potential to vary due to the different mechanisms of valuing the features specified on the functional heads in an extended projection. Our concern is telicity. In English, the quantity feature on the Asp_Q head is valued via the indirect strategy, which is copying the quantity feature of a DP in the specifier position of Asp_QP (i.e. [Spec Asp_QP]) via agreement. In theory, it is possible that in some languages the direct valuation strategy might also be available; that is, there is a functional item in the lexicon that bears an interpretable quantity feature that can value the feature on Asp_Q.

In Borer (Reference Borer2005b), it is shown that this situation does exist in some Slavic languages. In languages like Czech, a perfective prefix serves as an event delimiter, which imposes a telic interpretation on the one hand, and also restricts the interpretation of bare nominal arguments by providing them with quantificational force:

In (27b), the prefixed perfective verb gives rise to a telic interpretation. In addition, the prefix also forces a definite and quantity reading on the bare noun object. Without the perfective prefix, no telic reading is attested. Additionally, the bare noun in (27a) does not need to take a definite reading or quantity reading.

Borer (Reference Borer2005b) takes such data as evidence of the paradigm of direct range assignment (feature valuation). In particular, the perfective prefix in Slavic languages is the functional item that bears the interpretable quantity feature, which is directly inserted in the Asp_Q head to value the uninterpretable quantity feature ([uQuan] for short). In addition, when a bare nominal theme argument is involved, the perfective prefix copies the quantity feature to the quantity head in the DP structure and provides a strong D feature (with a definite force) to value the uninterpretable D feature ([uD]) on the D head of the nominal theme, as shown in (27b). In this paper, we adopt Borer’s (Reference Borer2005b) general framework, but deviate in one aspect: the explanation of the definiteness of bare nominals in Slavic languages. For this aspect, we will draw on Filip’s (Reference Filip and Toman1997) account to be summarised in the following sub-section, and we will show how this can be assimilated into Borer’s framework.

3.3. The quantificational force of the telic marker: Filip (Reference Filip and Toman1997)

In the above examples, we already see that when a bare nominal is involved in a perfective prefix marked sentence in Slavic languages like Czech, it will not only take a quantity reading, but also a definite reading. In Borer’s account, the source of definite interpretation is straightforward: the telic marker (i.e. the perfective prefix) provides a strong [D] feature to the bare NP in the object position. However, this assumption is weakened by the fact that indefinite numeral NPs can also stay in Slavic telic sentences where a perfective marker is prefixed to the verb. If the perfective prefix itself provides the strong [D] feature (i.e. the definite force), it will follow that an indefinite NP object in the [Spec QuanP] position will be impossible because this NP is in the scope of quantificational force of the [Quan] head (i.e., the perfective prefix that imposes definiteness force to it). The definiteness imposed by the perfective prefix and the indefiniteness taken by the numeral NP will thus give rise to semantic incompatibility. However, in Slavic, an indefinite NP object is actually possible:

In the above examples, the objects are all numeral NPs, taking an indefinite reading and staying in the post-verbal object position. Borer’s (Reference Borer2005b) solution is as follows: when a numeral NP object appears, this means that the quantity feature of this NP is already valued by the numeral, and the telic marker (i.e. the perfective prefix) is not responsible for the quantity feature of this DP. On the other hand, this numeral NP still needs its D feature for referentiality, and it is the perfective prefix that provides the weak D feature (existential closure), on the assumption that if the quantity and D features are not both valued by the prefix, the latter will provide a weak D feature. We find this solution is more or less ad hoc, as it is, in a sense, inconsistent to claim that the same prefix can sometimes assign a strong (definite) D feature and sometimes assign a weak D feature. It is better to assume that the perfective prefix does not bear any definiteness relevant feature (strong D feature in Borer’s framework), and this is exactly the point taken in Filip (Reference Filip and Toman1997).

Filip’s (Reference Filip and Toman1997) account draws on studies about the division of two types of quantification: the D-quantification and the A-quantification (Partee et al. Reference Partee, Bach and Kratzer1987, Partee Reference Partee and Ingemann1990). While the D-quantification is expressed in the NP by determiner quantifiers, the A-quantification is achieved by elements such as sentence adverbs, auxiliaries, and affixes, among others. Filip (Reference Filip and Toman1997) points out that while a perfective prefix makes a verb telic or bounded, it also functions as an A-quantifier, which binds the variable introduced by the NP object, and hence extends a semantic effect to the NP with its quantificational force. Following Comrie (Reference Comrie1976), Filip (Reference Filip and Toman1997) makes the following proposal:

The above assumption can be assimilated into Borer’s (Reference Borer2005b) framework: the holistic function is roughly equivalent to the effect of the quantity ([Quan]) feature. The [Quan] feature provided by the perfective prefix imposes the interpretation that the event is a quantity one (i.e. telic, bounded, or holistic). This obligatory interpretation requires the incremental theme to be a quantity/holistic object. This is because, with a telic interpretation and an incremental theme, a homomorphic relationship is established, which means that the subevents of the holistic event are mapped onto the subparts of the incremental theme, hence the OTE (object-to-event) mapping (MacDonald Reference MacDonald2008). For Filip (Reference Filip and Toman1997), the perfective prefix, serving as an operator/quantifier, imposes the holistic force on the object, requiring the bare NP to take the quantity reading. Again, this assumption can be assimilated into Borer’s (Reference Borer2005b) framework if we assume that the [Quan] feature on the Asp_Q head is copied onto the bare nominal. However, Filip (Reference Filip and Toman1997) then points out that the perfective prefix does not require that the incremental theme be definite. That the bare object nominal in Slavic (mass nouns and bare plurals) has to take the definite reading is the presupposition of the quantity reading taken by the bare nominals. In particular, when a bare nominal is forced to take a quantity reading by the perfective operator, this quantity reading has to presuppose the existence of a ‘whole bounded entity’. We can take the examples repeated below to explain this point:

In (31b), the perfective marker provides the holistic function (or [Quan] feature), imposing the interpretation that the drinking event is bounded, hence a telic event. This interpretation, together with the fact that ‘wine’ is the incremental theme of this drinking event, entails the homomorphic relationship between the bounded drinking event and the ‘wine’. The final point of the drinking event is when the final part of the wine is consumed, which means the mass interpretation of the wine must be rejected. Instead, there must be a certain quantity of wine. That is, in order to get the quantity reading, we should hold the presupposition that there is a certain quantity of wine, which is known to both the speaker and the hearer (or the hearer has to resort to presupposition accommodation to take this reading), hence the definite reading of the bare nominal wine. Note that while the quantity reading of wine is derived from the perfective marker, either the quantificational force or the [Quan] feature in Borer’s (Reference Borer2005b) system, definiteness is required to accommodate the quantity reading. The perfective marker itself does not provide any definiteness force. This then explains the situation in (28) and (29) where the telic event has a numeral NP object that takes an indefinite reading. This line of explanation predicts that if the perfective prefix is not present, the whole entity or quantity reading of the bare nominal object is not required, which is supported by the fact presented by (31a).

There remains a question to be addressed: if the definite reading is not from the quantification of the perfective/telic marker, then from where is it derived? Firstly, we have to know that it is a linguistic fact that Slavic (like Chinese) languages allow a bare noun, whether it is count or mass, to occur in argument positions, taking different interpretations like definite, indefinite, and generic reading. This is different from languages like English where a bare count noun cannot stay in the argument position in most cases. The syntactic structure of nominals in Slavic is naturally an important domain in syntactic studies. Two approaches are taken, divided by whether a D head is present. Borer (Reference Borer2005a) argues that, for all languages, there is a universal functional structure DP that involves a D head responsible for (in)definiteness and existential closure, hence making the type shifting from $ \left\langle e,t\right\rangle $ to $ \left\langle e\right\rangle $ possible. If this approach is taken, a bare nominal in languages like Slavic takes either a definite or indefinite reading depending on the feature of the null D head. Note that for the null D to take a definite or indefinite effect (corresponding to the and a in English) can be a pure pragmatic issue. The syntax only provides a null D, while the specific value of D can be pragmatically determined. The other approach, represented by the series studies of Bošković (Bošković Reference Bošković2008, Reference Bošković2009a, Reference Boškovićb; Bošković & Hsieh Reference Bošković and Hsieh2015), argues that there is a parameter of DP and NP languages. Slavic and Chinese belong to the NP language, which does not have a D head in the nominal structure. Abstracting away technical details, in Bošković’s approach, instead of relying on a null D head, the (in)definiteness reading of a bare nominal is achieved via pure type shifting that does not have to resort to the valuation of the D feature on a certain D head. In this paper, we keep neutral to the two approaches to the nominal structure. The point crucial to our discussion is that the (in)definiteness interpretation of a bare NP in Slavic (and in Chinese) is not directly assigned by the telic feature. The linguistic system of these languages has a specific mechanism responsible for this interpretation, which is activated when interacting with the requirement of telic interpretation as discussed above.

To sum up, integrating studies in Borer (Reference Borer2005b) and Filip (Reference Filip and Toman1997), we assume the following theoretical points to be applied in the rest of this paper:

4.1. lǝ as a telicity functional item

Below is our first hypothesis on the nature of lǝ:

The above hypothesis directly addresses an issue summarised in Section 2: lǝ-marked sentences always express a telic event. In the XS Model, telicity is the result of the valuation of the quantity feature on the Asp_Q head; if lǝ bears an [iQuan] feature, telic reading is obligatory in lǝ-marked sentences.

With this hypothesis, we can also explain another issue: the shift of event types denoted by activity predicates into telic events:

In (34), although tae (to push) is not a telic predicate (as it is an activity verb), this sentence is forced to take a telic reading. This is because lǝ provides an [iQuan] feature to the Asp_Q head, hence requiring the event to take a quantity/holistic reading. Semantically, this is achieved because lǝ specifies the endpoint of the event, meaning that the pushing event is by no means homogeneous. Therefore, when this sentence is uttered out of the blue, ‘endpoint accommodation’ in the context is required. This does not mean that pragmatics determines syntax, but the other way around. Syntactic valuation of the [Quan] feature on Asp_Q requires this contextual accommodation. For example, we can imagine this context: Zhangsan’s work is to push carts away to some place. With this context, the interpretation is that Zhangsan has pushed three carts away, and the endpoint of this event is when the third cart is pushed to the destination.

We can also take accomplishment verbs to further support the hypothesis:

Both sentences have the same accomplishment verb ʧɛ (eat), and both involve a quantity incremental theme, sa ǝ biŋgo (three apples). However, only in (35a), the endpoint is invariably expressed, which is the consumption of the last bit of the three apples. This is evidenced by the fact that the second half in this example is unacceptable.

(35b), on the other hand, does not specify whether the three apples were eaten up. It only means that Zhangsan bit all three apples. He might not finish any of them. That’s why the second half in this example is natural. To sum up, when lǝ is involved, the endpoint is part of the truth condition, while its absence leaves the endpoint under-specified. This is clear evidence that lǝ is syntactically/semantically responsible for telic interpretation. In our framework, this is because it is a functional item carrying the interpretable [Quan] feature, valuing the [uQuan] on the Asp_Q head. Without this marker, the [uQuan] feature will not get valued, and hence the endpoint is not semantically specified.

4.2. Accounting for the effects of lǝ on objects

In (23), we have summarised the effects of lǝ on nominal objects: first, it forces a bare NP to take a definite and quantity reading; second, it seems to force bare NPs and definite NPs to be fronted to a topic position, while a numeral NP can still stay in the post-verbal position and take indefinite reading. We will address these effects one by one below.

4.2.1. Definite reading of bare objects

We account for the definite effect on the bare object in a lǝ-marked sentence first. Based on the theoretical toolkit in Section 3, we postulate the following hypothesis on lǝ:

The hypothesis in (36) is a straightforward application of Borer’s (Reference Borer2005b) framework. Like Slavic languages, in Yixing there is a functional item lǝ that takes the [Quan] feature, responsible for telicity, as we have argued above. This feature can then be copied onto the NP in the [Spec Asp_QP] position via the [Spec-head] agreement, as is proposed in Borer (Reference Borer2005b). This patterns with Slavic languages, imposing a contrast with English, where there is no functional item to value the [Quan] feature on the Asp_Q head, and the feature copy is from Spec to Head, instead of Head to Spec. This explains why in Yixing, just like Slavic languages, the bare NP in the telic sentence must take a quantity reading. This is also in line with Filip’s (Reference Filip and Toman1997) point on the holistic function of the perfective item: the perfective item derives telicity, hence holistic/bounded interpretation, which entails the holistic and boundedness interpretation of the incremental theme. A bounded entity in Borer’s (Reference Borer2005a, Reference Borer2005b) sense is a DP with a quantity feature.

Above we have accounted for the quantity reading of bare nominals in the lǝ -marked sentence in Yixing. Note that like Slavic languages, the bare nominal in telic sentences in Yixing also has to take a definite reading. We have pointed out in Section 3 that such definiteness is not directly derived from the telic functional item, hence deviating from Borer’s account. We presented evidence why definiteness is not part of the feature copied from the telic marker. In Slavic languages, a numeral NP with indefinite reading can be an object of a telic sentence, which is otherwise incompatible with the telic marker if it involves a definite feature. This type of evidence is also present in Yixing, as we have seen throughout this paper. We repeat one example below:

Following Filip’s (Reference Filip and Toman1997) assumption, we argue that a bare noun in a telic sentence in Yixing, just like the situation in Slavic languages, takes a definite reading because we have to presuppose the existence of this whole entity. It is also shown in Section 3 that, to fulfill this definiteness presupposition, that is, to assign a definite reading to the bare NP, we can either follow the spirit of Bošković (Reference Bošković2008, Reference Bošković2009a, Reference Boškovićb), Bošković & Hsieh (Reference Bošković and Hsieh2015) to claim that Yixing Chinese and Slavic languages are NP languages that lack a D feature, and definiteness is derived from a pure semantic type shifting from $ \left\langle e,t\right\rangle $ to $ \left\langle e\right\rangle $ , or adopt Borer’s (Reference Borer2005a) account, assuming that there is a null D that is responsible for definiteness. Both are compatible with the present analysis.

4.2.2. ‘Fronting’ of definite objects

So far, we can see that the telic marker lǝ in Yixing and the perfective prefixes in Slavic languages share almost identical syntactic and semantic properties. They both denote telicity, impose a boundary on the incremental theme, and bear a relationship with the quantity reading of the bare nominal object. But there is a prominent difference in the relationship between telicity and the syntactic positions of the object NP. As we have shown in Section 2, in a lǝ -marked sentence in Yixing, the definite object and the bare object have to be fronted to the topic position, while the numeral NP object can stay in the post-verbal position. We repeat relevant examples below:

The major hypothesis to be detailed in the rest of this section is summarised below:

We will first explain why it is a reasonable hypothesis that lǝ is a universal quantifier, after which it will be shown how this aspect of lǝ provides the underlying reason for the fronting of definite and bare NPs.

In the traditional sense, quantifiers will remind people of those elements that form part of the DP or the extended projection of nominals, such as all, every, a, etc., in English. However, Partee (Reference Partee and Ingemann1990), based on the cross-linguistic studies of languages like Salish and Warlpiri, shows that quantifiers in different languages might take different categories, and the D category is just one possibility. A quantifier might be a predicate, an adverb, or a verbal affix, among others, which are called A-quantifiers, whose semantic effects are comparable to those of the D-quantifiers like all and every in English. In fact, there is already evidence that lǝ is a quantifier. In Yixing, as we have shown, it is the telic marker lǝ that provides the quantity feature, which is unlike the situation of English telic sentences, where it is the DP that provides quantity feature to the inner Asp head.

What further supports the hypothesis in (42) is that lǝ always provides the interpretation that all the members of the set denoted by the object NP are ‘consumed’ in the event. In the (b) sentence of (38), for example, all the three apples were eaten. For the bare NP objects in (39) and (40), the bare NP takes a definite reading that involves a certain quantity for the reason presented in the last subsection, and the interpretation is that the whole quantity is consumed. We will later show that this also patterns like another non-nominal universal quantifier (hence also an A-quantifier) in Mandarin Chinese (i.e. dou), which is an adverbial universal quantifier.

Following the hypothesis in (42), the object fronting restriction is reduced to the following condition:

We argue that the crucial difference between the nominal quantifier and the A-quantifier is that the former but not the latter can always find a variable in the same nominal domain. Take the universal quantifier every in English, for example. Since it is part of a DP, it can always find a variable in the DP domain (for example, a variable provided by the nominal predicate). As a verbal quantifier, lǝ is not in a DP/NP domain, and in order to be a legitimate universal quantifier, it needs a variable from another source other than the DP domain. As the trace of NP movement provides a variable, it can be predicted that as long as an NP trace is within the binding domain of lǝ, the quantification requirement of lǝ would be met, making the appearance of lǝ legitimate (other things being equal). This then provides a straightforward account for the NP fronting constructions listed previously in this section. In all these examples, a trace is left due to NP movement. For example, in (38), the definite NP object ‘those three apples’, if in the original post-verbal object position, does not provide a variable, and hence lǝ as a universal quantifier does not have any variable to quantify over, resulting in the ungrammaticality of the sentence in (38a). After the movement of the definite object in (38b), a trace is left, serving as a variable for lǝ to quantify over. The resulting semantics is then roughly as follows:

Within this account, lǝ itself does not have any grammatical function to front a definite DP object (and this is why we put ‘fronting’ in quotation marks in the heading of this sub-section). The NP movement is enabled by other independent mechanisms available in Chinese, such as topic raising and ba construction, among others. What is required by lǝ is that it needs a variable within its binding domain. A prediction, then, is that a lǝ -marked sentence might still be legitimate if there is a variable within its binding domain which, however, is not created by NP movement. A typical situation of this scenario is the so-called tough-construction, where the trace is created not by DP/NP movement but by a null operator (cf. Chomsky Reference Chomsky, Wasow, Culicover and Akmajian1977, Keine & Poole Reference Keine and Poole2017, among others). The example below shows that such a scenario does legitimise lǝ:

The above example is a typical tough-construction, where a trace is left in the object position of the verb ʧɛ (to eat), and this trace is well known to be created by the movement of a null operator, which moves to the edge of the CP that is later merged with the subject. What is crucial is that, in the above example, the subject of the sentence is not moved from the object position but is base-generated in the subject position, a well-received conclusion in the literature on tough-constructions (see Chomsky (Reference Chomsky, Wasow, Culicover and Akmajian1977) for the original argument, and Keine & Poole (Reference Keine and Poole2017) for a detailed review and analysis of new arguments and debates). Following our hypothesis, this example is predicted to be grammatical because it meets the condition in (43): there is a variable in the binding domain of lǝ. Note that in this situation, the variable is created by the null operator movement, and the null operator movement is motivated independently, not related to any property of lǝ.

Now we need to further explain an issue: why can an indefinite quantity NP (i.e. the phrase in the shape of [numeral+classifier+N]) stay in the object position? According to our hypothesis, there must be a mechanism that provides a variable in the binding domain of lǝ. And this is indeed the case. Note that it is well known in the literature (Chierchia Reference Chierchia1998, Borer Reference Borer2005b) that at least in Chinese, the [numeral+classifier+N] chunk creates a nominal predicate ( $ \left\langle e,t\right\rangle $ type), and a nominal predicate provides a variable, which is often bound by a D element, like a nominal quantifier in the DP domain. However, without such a D element, this phrase is just a nominal predicate, which provides a variable to be quantified over. The verbal quantifier lǝ, therefore, can still find a legitimate variable to quantify over, and this explains why it is possible for an indefinite quantity NP to stay in the post-verbal object position. According to Diesing (Reference Diesing1992), an indefinite NP can get its existential closure within the VP domain (roughly equivalent to vP in the recent tradition). In our case, if the quantity NP gets its existential closure, again we would not have a variable for lǝ. However, in the previous section we have shown that lǝ is selected and inserted in the derivation within the vP domain, which means that the indefinite quantity NP will be quantified over by lǝ before the derivation of the vP phrase is completed. That is, the quantification provided by lǝ goes before the existential closure by the vP can take effect, which supports the current account.

But why do Slavic telic markers not trigger such NP fronting if they are also A-quantifiers? We assume, tentatively, that Slavic telic markers are not, by nature, universal quantifiers. Filip (Reference Filip and Toman1997), while showing the universal quantification semantics in some perfective sentences, does admit that not all perfective prefixes will give rise to the semantics of universal quantification. As is stated in Filip (Reference Filip and Toman1997), the perfective prefixes in Slavic languages actually take idiosyncratic lexical meanings not limited to universal quantificational meaning, but include distributivity and vague quantificational meaning like ‘many’, ‘much’, and ‘a lot (of)’, among others. Therefore, the quantificational meaning is from the idiosyncratic lexical meaning taken by the prefix instead of from an operator, which does not impose the condition requiring a variable in the binding domain. There is, in fact, evidence also in Yixing to support this assumption. What we have argued so far is that lǝ, in addition to being a telic marker, is a universal quantifier, an operator in nature. Apart from this operator, it is also possible to provide lexical meaning similar to the universal quantificational meaning. If this is the case, then in such a situation where no operator is involved, the restriction on NP fronting will not be observed. This is indeed the case. In Yixing, there are lexical particles like wə and gwaŋ (counterparts of wan and guang in Mandarin), which provide the meaning that the whole quantity is ‘consumed’. On the other hand, they are not telic markers and hence do not provide a quantity feature or serve as quantifiers. The examples below show that when such particles are involved, if they do not co-occur with lǝ, the sentence does not have a telic effect, and they do not have an object fronting effect:

The above examples are evidence that apart from universal quantifiers, there are also other non-operator elements that can provide lexical meaning of finishing or completion. Not being a universal quantifier, items like wǝ and gwa ŋ do not submit to the condition in (43). Additionally, such items do not share the other effects imposed by lǝ discussed in this paper. The Slavic perfective prefixes, while sharing the telic nature with lǝ in Yixing, are not, by nature, universal quantifiers. Instead, such prefixes provide idiosyncratic meanings with only some patterning like universal quantificational meanings. lǝ in Yixing, on the other hand, consistently provides the universal quantificational semantics, and we argue that it is indeed a universal quantifier.

Before proceeding, we address an issue regarding the assumption on variable binding proposed in this section. The current approach, as pointed out by a reviewer, seems to have the potential to violate the Bijection Principle (BP) by Koopman & Dominique (Reference Koopman and Dominique1982), as a variable in this account is bound by both the NP moved to the topic position and the universal quantifier. We would like to show that this situation presented in our paper does not pose a threat regarding the violation of BP. BP is proposed to account for the issue of crossover, which states that there is bijective correspondence between an A-bar operator and a variable. The gist of BP is that an A-bar operator should only bind one variable, and a variable should only be bound by one A-bar operator. BP, regardless of whether it is rejected or not in the later research (cf. Safir (Reference Safir1983)), does not impose a restriction on the non-A-bar operator and the variable. For example, it does not apply to A-operators, which can bind more than one variable. Returning to our situation, the variable is only bound by one A-bar operator (the NP in the topic position), while the other operator, the universal quantifier, is not an A-bar operator.

In fact, a variable bound by both an A-bar operator and a quantifier is not uncommon, as the following examples in English show:

In (47a), the variable created by the wh-movement is bound both by the wh-phrase and the universal quantifier all. In (47b), which is a topic construction, the pronoun them is bound by the topic DP as well as the universal quantifier all. In fact, McCloskey (Reference McCloskey2000) provides a syntactic analysis of such data that involve a wh-movement and a quantifier, further showing that this type of phenomenon is a linguistic fact that is attested cross-linguistically.

While there is no BP violation problem, it is a general issue as to how a variable is bound by both an antecedent and a quantifier. As far as we know, there is no formal account for this phenomenon (McCloskey (Reference McCloskey2000) mainly focuses on the syntactic structure, without explaining how the variable-binding relationship is realised). Providing a detailed account for this general issue is beyond the scope of this paper. What is relevant to the current purpose is the fact that a variable can be bound by an antecedent and a quantifier.

Below we present an initial analysis of the analogy between lǝ quantification and the quantification of dou in Mandarin Chinese, which further supports the hypothesis that lǝ is a universal quantifier. dou in Mandarin is often taken as a universal quantifier (cf. Lee Reference Lee1986, Lin Reference Lin1998, Pan Reference Pan2006, among others), and since it is not within the DP domain (often taken as an adverbial item), it is also an A-quantifier. It would therefore be predicted that it will impose the same requirement: requiring a variable in its binding domain. This is indeed the case. The NP quantified over by dou is always fronted to the left side of dou, and following the current analysis, this is because a variable is required:

Following our hypothesis, this universal quantifier, dou, which is not in the DP domain, also submits to the condition in (43); that is, it needs a variable to be within its binding domain. DP/NP movement, motivated by whatever mechanism in the grammatical system, can provide a variable, as indicated by the above example in (48a). Also, (48b) shows that, without such a variable, the sentence will be ungrammatical, reminiscent of our case in lǝ-marked sentences.

If our hypothesis is correct, then we would also predict that the NP movement is not triggered by dou itself, and other scenarios where a variable is created not by NP movement can also be legitimate in a dou sentence. Here again, the tough-construction kicks in:

Here, just like the case in (45), a variable is created in the post-verbal object position, not because of NP movement, but due to null operator movement, while the subject is base-generated. This variable is quantified over by the adverbial universal quantifier dou, making the latter legitimate. There is, however, one difference between the restriction of lǝ and dou. Recall that for lǝ, it is possible for an indefinite numeral NP to stay in the post-verbal object position. This is impossible for dou:

The ungrammaticality of the above sentence supports our account. It was argued above that indefinite quantity NPs are possible in the post-verbal position in lǝ -marked sentences because such phrases are, by nature, nominal predicates, which provide a variable to be quantified over by lǝ. The variable provided by the indefinite NP can be quantified over by lǝ because the quantification of lǝ goes before the existential closure can be provided by vP. In the case of dou, since this universal quantifier, as an adverbial item, is outside the vP domain (or merged with vP as an adjunct in the stage of late-insertion), the vP existential closure will quantify over the variable before dou can take effect. Therefore, when the indefinite NP is in the post-verbal object position, dou does not have a variable to quantify over, and movement, among others, is one way out to create a variable for dou.

A reviewer raises an issue regarding the involvement of shenme in dou-marked sentences presented below:

The interesting issue is that when shenme is in the object position, a wh-question reading is derived, but when the same object shenme is placed in the pre-verbal position, the question reading is impossible. Following Tsai (Reference Tsai1994), shenme in Chinese is not a wh-word like what in English, but a variable that can be unselectively bound by accessible operators. What is crucial in Tsai’s analysis is that the question operator (Q-operator) is null, which is attached in the C position, and is hence an A-bar operator. Since dou is in the vP domain, which then is not in an A-bar position, it is an A operator. As shown above, there is empirical evidence that a variable can be bound both by an A-bar operator and an A operator, and it is not against the BP. In (51a), the variable shenme is bound by the Q-operator to give rise to the question reading, and it is bound by the universal quantifier dou which gives the ‘all’ reading.

In (51b), the wh-question reading is not accessible, and the reason is that shenme is moved to a topic position (Chinese can have multiple topics), which then serves as an A-bar operator to bind its trace. This means that this trace can no longer be bound by an even higher Q-operator because the Q-operator is also an A-bar operator, and having two A-bar operators for a single variable is against BP. Of course we can also take Chomsky’s (Reference Chomsky, Wasow, Culicover and Akmajian1977) analysis of tough-constructions to assume that shenme is base-generated in this topic position; a null operator moves from the object position to the edge of the embedded CP, and the trace left by the movement creates a variable. But the result is the same: the variable is bound by an A-bar operator (the null operator in the edge of the embedded CP, which co-refers with the antecedent shenme), which blocks the accessibility of a higher Q-operator due to BP restriction. While providing this sketchy analysis, we have to emphasise that both dou and shenme are famously complicated topics in Chinese linguistics. However, we show that at least for the typical use of dou as a universal quantifier, it patterns like that of lǝ in Yixing. This analogy, in a sense, supports our hypothesis proposed in this section and further supports the assumption regarding A-quantifiers in Partee (Reference Partee and Ingemann1990) and Filip (Reference Filip and Toman1997). While there might be alternatives to the account proposed in this subsection, it seems to be reasonable to assume that the object fronting of lǝ -marked sentences and dou sentences share the same underlying mechanism, which is reduced to the nature of the A-quantifier style of the universal quantifiers, verbal for lǝ and adverbial for dou.

Before ending this section, a caveat is presentedFootnote ⁹. At this point, it is clear that lǝ, in addition to its grammatical functions like telic marking, has some idiosyncratic semantic content, denoting the meaning that the object is completely consumed, indicating that the object will disappear at the endpoint. This property of lǝ is not surprising in the context of Chinese grammar wherein most functional items also take idiosyncratic content. That is why functional items in Chinese are called semi-functional items in Huang’s (Reference Huang, Li, Simpson and Tsai2014) recent work on the analyticity of Chinese syntax. We should then ask a question like this: what if a telic event is to be expressed, but there is no reading such that the object (theme) disappears at the endpoint? Obviously, other ways to derive telicity have to be employed. In Borer (Reference Borer2005b: Chpater 12), it is argued that goal PPs and particles (like over in take over and up in stand up) can also assign a range to the quantity head (Asp_Q head). It would then be predicted that such ways in Yixing can be taken to provide telic (quantity) feature as a last resort when there is no complete consumption and disappearance involved in the telic event. This prediction does hold. The following examples are both telic, but neither involves lǝ. The first one takes a goal PP, and the second one takes a particle (roughly equivalent to up in English):

In (52a), it is PP that provides the telic feature (following Borer’s account), and lǝ is not required. A question can be asked: without PP, can lǝ be attached to the verb in this example to derive telic reading? The answer is no, and the reason is related to the semantic content of lǝ: lǝ expresses the reading that the endpoint is the ‘disappearance’ of the theme argument (such as via consumption or other forms of disappearance), while reaching a location like the case in the above example is not compatible with this reading. This also applies to the freeze example. Since the water only changes its form but does not disappear, lǝ is not compatible with this verb. To express the change of state, the particle ʧi-lae, which is the counterpart of qi-lai (roughly meaning rise and up) in Mandarin, has to be used. This also explains in general why unergative verbs are not compatible with lǝ. Since lǝ denotes the semantic content that the entity is completely ‘consumed’ and hence disappears, events involving an unergative verb will not be compatible with lǝ because there is no theme argument involved, not to mention an entity that is completely ‘consumed’. This is shown by the following examples:

In both examples that involve unergative verbs (ʃao (smile) and bao (run)), the sentence becomes ungrammatical as long as lǝ is involved.

By arguing that lǝ takes some idiosyncratic content that denotes the meaning of complete consumption, we might face this question, as is raised by both reviewers: is lǝ actually a secondary predicate like the particle wan (finish) in Mandarin, which, together with the matrix verb, forms a complex predicate (i.e. the resultative verb compound (RVC)) as is put in Sybesma (Reference Sybesma1997, Reference Sybesma, Sybesma, Behr, Yueguo, Handel, Huang and Myers2017)? There are reasons to argue against this possibility. First, in Yixing, there are items corresponding to particles like wan (finish). Even when such particles are attached to the verb, as shown by the example below, still lǝ is obligatorily required for the expression of telicity, showing that it is lǝ that is responsible for the syntactic encoding of telicity. Also, if lǝ is really the resultative predicate that is mainly responsible for denoting semantics of ‘fishing’ or ‘completion’ which is also expressed by wǝ (counterpart of wan in Mandarin), then we would not expect it to co-occur with wǝ. But the following example shows that they can co-occur, and actually to express telicity, even when wǝ is attached to the verb, the presence of lǝ is obligatory. This clearly shows that lǝ and particles like wǝ have different grammatical functions and hence are inserted in different positions in the structure.

Secondly, as shown throughout this paper, apart from the function of marking telicity, lǝ presents some special grammatical functions like the special effect on the object NP (like NP fronting) and temporal denoting in the sense of perfective aspect. Neither is exhibited by resultative secondary predicates like the Mandarin wan or its Yixing counterpart wǝ in the RVC. Therefore, we conclude that despite the possible historical relationship between lǝ and the resultative predicate, at the synchronic level, lǝ has developed into a functional item, although it still retains some idiosyncratic content which seems to be common to functional items in Chinese in general.

4.3. lǝ and perfective aspect

As described in Section 2, when lǝ occurs in a single clause, a past event reading is expressed, seeming to indicate that lǝ serves as a past tense marker. Moreover, lǝ seems to be impossible to occur in a clause that expresses either a habitual event or a future event. We repeat the relevant examples below:

Such data might easily lead us to assume that lǝ also takes a past tense feature that imposes an uncancellable past tense reading. However, a closer scrutiny at more data shows that the above phenomenon is not related to a tense marker, but concerns reference time:

The above examples show that as long as there is explicit reference point and this point is preceded by the event time, then the event time of the lǝ -marked sentence does not have to be past. Since the relationship between the reference time and event time is exactly related to the outer aspect, it is natural, therefore, to assume that lǝ, in addition to the telicity/quantity feature, also bears an outer aspectual feature. And this is exactly the hypothesis to be proposed in this section. We assume that in addition to the quantity/telic feature, lǝ also bears a perfective feature, which then serves as a relative tense marker as proposed by Lin (Reference Lin2000, Reference Lin2003, Reference Lin2007) in the seminal study on the verbal le in Mandarin Chinese. We will return to the difference between Yixing lǝ and the Mandarin le later in this section. The crucial property of the perfective marker is that the event expressed by the sentence is within the reference time, and therefore the event is taken as a complete whole (cf. Smith Reference Smith1997; Lin Reference Lin2000, Reference Lin2003, Reference Lin2007; Demirdache & Uribe-Etxebarria Reference Demirdache and Uribe-Etxebarria2014). This alone does not necessarily entail that a lǝ -marked sentence is forced to denote a past tense reading. What also plays a role in the derivation of the past tense reading is that the speech time is taken as reference time by default. Hence we have the following summary:

It then follows that, in the absence of an explicitly expressed reference time, a lǝ -marked sentence always expresses an event that is within the speech time taken as default reference time. Here we see that the dual-feature property of lǝ gives it two functions: it is responsible for the encoding of a telic event, and it is responsible for the viewpoint over the event. Because the event in a lǝ -marked sentence is telic, it is an atomic entity like a count noun. A special property of the atomic entity is that no sub-part of this entity is also the same entity, just as a part of an apple in itself is not an apple. This is crucial in the past tense reading of an event in the lǝ -marked sentence. The endpoint of this event must be within the reference time because it is situated by the perfective aspect, which encodes the viewpoint reading such that the event is within the reference time. If the default reference time is speech time, it means that the endpoint of the event is within the speech time, which then gives rise to the past tense readingFootnote ¹⁰. This can be summarised as follows:

We can take examples in (55) to illustrate the above points. In (55a), without a specific reference time, the speech time is taken as reference time by default. This means that the final point of the drinking of a cup of coffee must take place within the speech time, hence a past tense reading. We may then ask why can we not just use the same default reference time to make (55b) acceptable? The problem lies in ‘every day’. A habitual sentence obviously cannot take a specific speech time as reference time, and without a legitimate reference time, this sentence is unacceptable. The same logic applies to (55c). Again, in this sentence there is no explicitly expressed reference time, and the default reference time (i.e. the speech time) is the only choice. By taking speech time as the reference time, a semantic conflict will arise: the perfective reading plus the speech time as the reference time will blindly return a past tense reading, but the adverbial miŋzao (tomorrow) explicitly shows that the event will take place in the future. This obvious semantic conflict cannot be resolved, leading to the unacceptability of this sentence.

The syntactic structure involving lǝ is therefore as follows:

In the above structure, Asp_OUTER is the functional head responsible for the perfective aspect, which is above vP. The [iQuan] feature on lǝ will value the corresponding feature to yield telicity. In addition, the [uPerf] on the Asp_OUTER needs to be valued, which is achieved via the valuation still provided by lǝ due to its additional [iPerf] feature.

If this analysis is on the right track, it can be predicted that, as long as a legitimate reference time is provided, the sentences in (55) will be saved. This is indeed the case as shown in (56). In both (56a) and (56b), the specific time 9 o’clock indicates that the reference time is a temporal span with 9 o’clock as the endpoint. In (56a), the reading is that the endpoint of drinking a cup of coffee occurs before or exactly at 9 o’clock everyday, a reading that is not past tense. In (56b), the reading is that the endpoint of his drinking a cup of coffee will occur before or exactly at 9 o’clock tomorrow, again not a past tense reading.

So far, the perfective reading of lǝ is exactly like what Lin (Reference Lin2000, Reference Lin2003, Reference Lin2007, Reference Lin2017) has proposed for the interpretation of verbal le in Mandarin Chinese. However, there is a difference. As Lin (Reference Lin2017) shows, the verbal le-marked sentence in Mandarin can also express a present tense:

This is because, as Lin (Reference Lin2017) argues, le is a perfective marker which only requires one of the events expressed by the vP to be within the reference time. Since the event of raising rabbits is not a telic event, a subpart of it is also of the same event. Therefore we can have the reading that one subevent of rabbit raising is within the reference time (i.e. the speech time by default), while other subparts of this event, which are also rabbit raising events, can continue, hence the expression of the present tense reading. This analysis in fact indicates that lǝ in Yixing cannot occur in such sentences. This is because the predicate jaŋ (‘to raise’) is not compatible with a telic reading, but lǝ forces a telic reading due to its telic (quantity) feature, which le in Mandarin does not bear. This is indeed the case:

The above example indicates an important fact regarding the comparison between Mandarin le and Yixing lǝ: the two particles have the same perfective feature but differ in that the latter bears a telic feature. This explains why the two look quite similar when we focus on the interpretation of perfective reading. The Mandarin le, however, does not obligatorily express telic reading, nor does it exhibit those restrictions of lǝ like object fronting. And this is due to the lack of telic feature.