1 Introduction

The English verb change is an example of a verb undergoing the causative alternation, illustrated in (1).

A standard view on the causative alternation is that the transitive, causative verb has a more complex event structure than the intransitive, anticausative verb. According to this view, the lexical-causative variant of the alternation is roughly analyzed as ‘cause to V-intransitive’ (Dowty 1979; Levin & Rappaport Hovav 1995), projecting two event variables, one for a cause event, and the other for a become (inchoative) event leading to some state that satisfies the stative property lexicalized by the verb. For instance, Parsons (1990: chap.6) can be attributed the following analyses of break in its anticausative and its (agentive) causative variant.¹

Other authors such as Rapp & von Stechow (1999), Kratzer (2005) or Alexiadou et al. (2015) argue that change-of-state verbs have the same event semantics in their transitive and intransitive uses: both denote a set of events that yield some state satisfying the predicate of state lexicalized by the verb.² For example, Alexiadou et al. (2015) propose that causatives and anticausatives involve the same verbalizing v-head taking a RootP introducing the internal argument DP as its complement; see (3). In both structures in (3), this v-head introduces a causative event leading to the state property expressed by the root and predicated over the internal argument. Lexical causatives and anticausatives therefore do not differ in their event structure. Rather, the causative alternation is, for these authors, a Voice alternation: Voice, the syntactic head responsible for introducing external arguments (Kratzer 1996), is projected in causative (transitive) construals, and absent from anticausative (intransitive/unaccusative) construals. The Voice-head does not, itself, introduce any event, but relates a DP in its specifier to the event denoted by the vP in its complement position, in (3b) as the agent of this vP-event.

In this paper, we provide a fresh look at the event structure of verbs of change-of-state undergoing the causative alternation by addressing two related questions:

Whereas we adopt Alexiadou et al.’s (2015) view that alternating change-of-state verbs have one event argument only in both their intransitive and their transitive use,³ we will answer to the second question in the affirmative.

Our proposal will be that in sentences built with a change-of-state verb, the verb and the vP leave the relation between the verbal event and the ensuing state underspecified. It is the syntactic context that determines the nature of this relation.⁴ The hypothesis we argue for is that in all syntactic contexts except when an agentive external argument is present, this relation is interpreted as become, which we define as a relation between a change that an entity x undergoes and some new state of x this internal change leads to. In the presence of an agentive external argument, it is interpreted as cause. A key difference between become and cause events is that while the former only have the theme x of the ensuing state as a participant, the latter necessarily involve a second participant (an agent or a causer).

We will refer to this hypothesis as the ‘inchoative hypothesis’, as the idea is that in their default meaning, change-of-state verbs have inchoative semantics, i.e., express just a change. As the distinction between agentive and non-agentive external arguments is therefore crucial to determine the type of relation between the verbal event and the state, we need a clear difference between the two types of external arguments. An operative distinction cannot rely on the criterion of animacy only, since agents can also be inanimate (Cruse 1973; Piñón 2001; Fauconnier 2012; Folli & Harley 2008 a.m.o.).

The paper is structured as follows. Section 2 critically discusses different views on the event structure of change-of-state verbs. Section 3 provides a brief background on the Voice framework adopted here. It focuses on the semantic differences between agent vs. causer subjects (that is, between agentive and non-agentive external arguments), and brings together tests that make it possible to distinguish between agents and causers without resorting to the animacy feature. One of our conclusions is that while inanimate DPs can be associated with the agent role, animate DPs can only be causers in restricted contexts (such as the subject position of causative psych-verbs). Section 3.1 focuses on inanimate subjects, and Section 3.2 on animate subjects. Section 4 summarizes data supporting the view that causative predicates express inchoative events when they are combined with a causer subject, and full causation events when combined with an agent subject. Section 5 presents an approach which accounts for this generalization in the semantics in a compositional way. This approach treats Voice heads as syncategorematic expressions, as opposed to categorematic expressions. Section 6 turns to transitive anticausatives (as in The room changed its temperature), as a further instance of cases where change-of-state verbs are used in transitive contexts but nevertheless describe inchoative events.

2 The event structure of change-of-state verbs

One of the main arguments for not assuming that there is an additional event argument for the agentive transitive variant compared to the intransitive anticausative variant of alternating verbs is that this assumption predicts that separate modification is possible for the agentive subevent (the action) and the inchoative subevent (the change) (Fodor 1970; Rapp & von Stechow 1999; Higginbotham et al. 2000; Pylkkänen 2008; Alexiadou et al. 2015 a.o). But separate modification is not possible, as illustrated by the contrast in (5) (see Higginbotham et al. 2000; Pylkkänen 2008; Martin & Schäfer 2014 for related examples with manner modifiers that point to the same conclusion). The sentence in (5b) is not felicitous in the relevant context established in (5a) where Charlotte did nothing more to Jean-Paul other than pushing him out of the window on Sunday. In this scenario, the agentive kill-subevent (the pushing) takes place on Sunday, and the inchoative kill-subevent (the dying) takes place on Tuesday night. If the two subevents are modified separately, this scenario cannot be lexicalized with the verb kill. Sentence (5c) illustrates the same problem. For us, this is because the verb in its agentive use describes the mereological sum of the action performed by the agent and the resulting change-of-state of the theme.⁵

The problem of (5b/c) results from trying to modify two subevents separately. This requires the verb kill to have two different event arguments, which it does not have according to the authors mentioned above. Contrary to what is often assumed (including by Fodor 1970),⁶ the problem of (5b/c) is not due to the inability of lexical causatives to lexicalize a scenario as above, or more generally to express indirect causation, that is, causation between two spatiotemporally disjoint events. As noted for instance by Danlos (2000), as long as there is no attempt to modify the action and change subparts of the vP-event separately, a lexical causative like kill is perfectly fine as a description of a causation event such that the agentive component and the inchoative component are spatiotemporally disjoint. This is shown in example (6), which is inspired by a similar example in Danlos (2000) (see also Neeleman & Van de Koot 2012; Martin 2018 for arguments in favour of the idea that lexical causative verbs can convey causation between temporally disjoint events).

Research that shares the assumption that change-of-state verbs have a single event argument in both intransitive and transitive uses typically assumes that the nature of the relation between the event and the state is specified at the level of the verb or vP (but see Wood & Marantz 2017 for a different view; cf. fn. 5). In this perspective, change-of-state verbs encode the same relation between the event and the state across transitive and intransitive uses. A first view is that events described by a change-of-state verb are become events. So for instance, for Rapp & von Stechow (1999), change-of-state verbs are decomposed into v-become and a state predicate across transitive and intransitive uses, and the agentive transitive use involves, on top, Voice introducing the agent of this become-event, as illustrated in (7)–(8).

A problem for this view is that, in agentive transitive uses, the verbal event sometimes starts when the action of the subject’s referent starts, not when the change begins. This is unexpected if the verbal event is only a become-event (at least under the common view that a become-event is the coming about of some state s, and only has the theme of this state as a participant). In English, this is especially obvious in progressive sentences. Take for instance (9):

Let us imagine that Ali Baba knows a long magic formula which, once recited, causes the opening of the cave door. In this context, (9) is true as soon as Ali Baba starts to recite this formula. In the context of an agentive subject, the opening event can therefore start before the door has started to open. The same observation holds for languages like French that do not have a futurate reading for the progressive (Martin 2015), and therefore cannot be explained away by the futurate use of the progressive in English (on such readings of the English progressive, see Dowty 1977; Copley 2014 among others).⁷

A second view is that change-of-state verbs always denote a set of cause events. For instance, Alexiadou et al. (2015) decompose change-of-state verbs into v-cause and a state predicate across intransitive and transitive readings (see also Wood & Marantz 2017). Adding Voice to the structure of the anticausative yields the transitive use, as shown in (10) and (11).

This solves the previous problem since the causing event can plausibly have the agent’s action as a part. More specifically, Alexiadou et al. (2015) can say that in (9), the causing event is already in progress when Ali Baba recites the magic formula.⁸ But it is the analysis of the variants without an agent (anticausatives or non-agentive causatives) that turn out to be problematic. For Alexiadou et al. (2015), anticausatives and non-agentive causatives are all descriptions of causing events (cf. Kratzer 2005). The way in which they analyse causer subjects or causer-PPs such as the English from-PPs like in (12b) makes this particularly clear. Building on Solstad (2009), they assume that the from-PP introduces an event that gets identified with the verbal event of the anticausative. They assume the same for causer subjects in non-agentive causatives, such as (12a) (following Pylkkänen 2008). Thus for them, the event introduced by the causer DP or PP gets identified with the verbal event in (12a, b).

A problem with this view is that separate modification shows that the event introduced by the causer is different from the verbal event. This is observed in Martin (2018; 2020) for causer subjects, see e.g. (13). Note that in (13a/c), the adjectives constant or continued ensure that the subject DP denotes an event and is not coerced so as to get a result or a referential reading (see Iordǎchioaia 2020 and references therein on the distributional restrictions of these adjectives):⁹

The same observation extends to anticausatives modified by PPs introducing causers:

Martin (2020) concludes from examples like in (13) that the event denoted by an external argument causer DP causes the verbal event rather than being identified with it (i.e., in (12a), the wind causes the door-opening event). Data as in (14) point to the same conclusion for anticausatives combining with causer PPs.

Martin further argues that in non-agentive causative statements (as in (12)–(13)), the verbal event is simply the change, as in anticausative statements. An argument for this view is that in such statements, the verbal event necessarily starts when the change starts, and not before as in agentive causative statements, as we just observed in the Ali Baba example above. For instance, for (15) to be true, the door must start to open, just as it is the case in the anticausative (16). In (15), the cause of this change is not part of the (ongoing) verbal event; rather, this cause is referred to by the subject the wind (we return to this point in Section 4).

Martin’s (2020) idea that in a non-agentive causative statement like (15), the verbal event is a become-event is, in fact, exactly the proposal of Rapp & von Stechow (1999) discussed above. Rapp & von Stechow (1999), however, assume that this applies to causative predicates in general, whether they are agentivized or not. We are then back to the problem raised by progressive agentivized causative predicates like in the Ali Baba example (9): events in the denotation of such predicates cannot be just become-events.

Obviously, the verbal event is not identified in the same way in agentive and non-agentive statements. What the data reviewed so far shows us, and what we will argue for further below is summarized as follows:

Apart from the separate modification facts discussed above, we will present further evidence for the hypothesis in (17) based on two kinds of contrasts: (a) whether the occurrence of a change-of-state is entailed by the transitive progressive sentence, and (b) whether the occurrence of a change is entailed by a transitive change-of-state verb in the complement of begin/start. These empirical arguments will be developed in Section 4.

To investigate the semantics of change-of-state verbs, we break with the habit in much literature of only contrasting an alternating verb’s anticausative use (cf. (18e)) with its causative use involving an in-control animate agent subject (cf. (18a)). Rather, we take into account a wider range of uses of alternating verbs. We will also discuss inanimate agents (for example the meteorite in (cf. (18b) and non-agentive causatives involving causer subjects (cf. (18c)). Finally, we will discuss transitive anticausative statements (cf. (18d)), which express the same meaning as canonical anticausatives such as (18e), and, thus, must be distinguished from examples like (18a–c).

Our proposal will be that sentences like (18c/d) exhibit a syntax/semantics mismatch: they are syntactically transitive, but the verbal predicate receives the same interpretation as in intransitive anticausative construals such as in (18e). That is, they form transitive predicates with inchoative semantics.

To develop this proposal, we turn next to the semantic and pragmatic differences between change-of-state sentences built with an agent and a causer subject.

3 Agent vs. causer subjects

We couch our analysis in the Voice framework originating in Kratzer (1996). Following Alexiadou et al. (2006), Schäfer (2008), Alexiadou et al. (2015), and Schäfer (2024), we differentiate between three different Voice heads: Agent Voice, Causer Voice and Expletive Voice (we focus on active versions of these Voice heads; passivized or non-active versions of these heads are possible, too; cf. Bruening 2013; Alexiadou et al. 2015; Schäfer 2017). We give a semantics for the composition of the vP with these different voice heads in Section 5, after the presentation of the facts supporting this analysis in Section 4. Expletive Voice is addressed in more detail in Section 6, devoted to transitive anticausatives.

Agent Voice (henceforth Voice_ag) associates the thematic role ‘agent’ with the DP merged in its specifier. All kinds of agents (exerting agent control on their behaviour or not, acting intentionally or not, animate or inanimate) are introduced by the same Voice_ag. We furthermore assume that in order for a DP to be associated with the role of agent, it must denote an individual (as opposed to an eventuality), and therefore be of type e.

Causer subjects are introduced by Causer Voice (henceforth Voice_c). We take Causer Voice to introduce an eventuality or a fact and to state that this eventuality/fact causes the verbal event. That is, Voice_c introduces cause-semantics (Martin 2020), and in the specifier of Voice_c, we find either eventuality-denoting DPs, as in (18c), or fact-denoting DPs, as in (19).

Fact-denoting DPs are virtually never addressed in the Voice framework, but are regularly found in the subject position of some change-of-state verbs such as cause (Vendler 1967; see also Rose et al. 2021 on subject DPs denoting omissions such as in The lack of sunscream caused Jane’s skin to burn), kill, or change (Martin et al. 2024, see Section 3.1.1). However, treating facts as causer subjects requires some revision of the semantics of Causer Voice (i.e., Voice_c). For Pylkkänen (2008) and Alexiadou et al. (2015), Voice_c has an event-denoting DP in its specifier. This event e is identified for them with the vP-event e′, see (20) (Voice is combined with a vP via Event Identification, as in Kratzer 1996). So for instance for these authors, in The tornado destroyed the city, the tornado e ends up being identified with a causing event e′ in the denotation of the vP.

Our first modification then concerns the ontological type of causer subjects. We take eventualities and facts to be of type s, and use the variable i as ranging in the domain of situations, which is the union of the set of eventualities, facts and other states of affairs. As our second modification, and following Martin (2020), we depart from the view proposed in Pylkkänen (2008) and Alexiadou et al. (2015) according to which the eventuality e (or, we add, the fact) introduced in the specifier of Voice_c is identified with the vP-event.

A first problem with the latter view is that if the causer subject is a verbal gerund with a genitive (a poss-ing gerund) and thereby denotes a fact (Vendler 1967; Asher 1993) or another sort of abstract object (Huang 2024), this fact or abstract entity cannot be identified with the verbal event.¹⁰ For instance in (22a), the killing event cannot be the fact denoted by the subject. A fact cannot be identified with an event; e.g., the fact that my neighbor did not water my plants is distinct from the killing-my-plants event (Martin et al. 2024).

A second problem with such a view is that as already mentioned in the introduction, it predicts separate temporal modification for the event denoted by the causer subject and the vP-event not to be possible, which is not the case, as shown in (13a–c) repeated below:¹¹

Instead, we take Voice_c head to introduce an eventuality or fact i and state that this eventuality/fact i causes the verbal event. When the causally responsible entity is an event, this event can temporally precede the vP-event, which explains why separate modification as in (13) is possible.¹² On this view, Voice_c does not carry thematic information, because it does not associate a thematic role to a participant of the vP-event. Rather, it introduces the eventuality or fact that is causally responsible for the vP-event (or what Neeleman & Van de Koot 2012 call the crucial contributing factor).

In line with much of the literature (e.g. Levin & Rappaport Hovav 1995; Reinhart 2000; Davis & Demirdache 2000; Doron 2003; Levin & Rappaport Hovav 2005; Alexiadou et al. 2006; Kallulli 2006b; Folli & Harley 2008; Schäfer 2008; Alexiadou et al. 2015), we assume that external arguments of transitive causative predicates are necessarily either agents or causers.¹³ External arguments cannot be associated with the role Theme (see Kratzer 1996 on Holder Voice with stative transitives). Some nominative instrumental subjects can be subsumed under agents, others under causers (Schlesinger 1989; Alexiadou & Schäfer 2006).¹⁴

In (23a/b), we recapitulate the assumptions made in this section about the external argument of lexical causative change-of-state verbs.

It follows from (23b/c) that for us, in the context of change-of-state verbs, the distinction between agents and causers maps onto the distinction between individual-denoting DPs (of type e), and situation-denoting DPs (of type s): in the subject position of a transitive (and eventive) change-of-state verb, an individual-denoting DP is necessarily an agent (whether animate or not), and a fact- or eventuality-denoting DP is necessarily a causer.¹⁵

The distinction between agents and causers is not just a terminological issue; it is key, for instance, to understanding the difference between manner (or hit-) verbs and result (or destroy-) verbs (Levin & Rappaport Hovav 2005; Alexiadou et al. 2015). Whereas result verbs are compatible with event-denoting (causer) subjects, manner verbs are not. Thus for instance, the manner verb hit cannot host an event description in subject position, cf. (24a) vs. (24b): its subject must be of the semantic type e.

So for us, the stone is an agent in (24a), not a causer. We address agentive inanimate subjects in more detail in the next section.

4 How subjects affect the meaning of the verbal predicate

The next subsections gather data in favor of the view that the causative predicate is interpreted as denoting causing events when it is agentivized, and as denoting inchoative (become)-events otherwise. In the latter case, the inchoative event described by the verb is understood as caused by the event or fact denoted by the causer subject. Examples (41) illustrate the two cases; in the paraphrase, the description of the event contributed by the verbal predicate is put in italics, and the contribution of the Voice head is underlined.²¹

The causal relation in the paraphrases of (41a) and (41b) is not exactly the same: while in (41a), it is a relation between an event and a state, in (41b), it relates a situation (an event or a fact) and an event. We use the operator cause for the latter, and we introduce cause⁺ for the former. In the next subsection, we present data in favor of the idea that the “causing⁺ event” denoted by an agentivized causative predicate consist of the full causation event (a sum of a causing event and some ensuing change), rather than the causing event only (the action), or the ensuing change only.

5 A new semantics for Voice

The proposal developed below takes at face value the fact that at the level of the verb and the vP, the relation between the event and the state awaits further specification. Neither the verb nor the vP tells us whether the events in their denotation are causing⁺ or inchoative events, and this needs to be specified in order to obtain a full event description. To account for this ‘incompleteness’ of change-of-state verbs/vPs, we introduce a relation R as their argument (and thereby also of the change-of-state vP). As we adopt a Distributed Morphology approach (Halle & Marantz 1993), we take this argument to be inherited from the verbalizing head v responsible for forming change-of-state verbs and vPs. Since this argument will only be introduced for change-of-state verbs/vPs, and since we will need Voice heads to ‘see’ whether the vP is of this type, we introduce a way of syntactically marking change-of-state vPs. The easiest way to do this is to introduce a syntactic feature [±cos], where change-of-state verbs and vPs are [+cos], and non-change-of-state verbs and vPs are [–cos], so that we would have (e.g.) the following for open the door:

More generally, a verb or vP would be marked as [+cos] if it results from the combination of the verbal head v with a stative predicate (provided by a root or an adjective as in hammer flat), and it is marked as [–cos] otherwise.²⁴

Taking open the door as an example, we implement this suggestion that the corresponding vP bears the feature [+cos], and we attribute to it the following semantic representation with an additional relational argument R:

More generally, this proposal assumes that change-of-state verbs and vPs come with a relational argument R between events and states that needs to be applied to a particular relation between events and states.

The semantics given in (54) is the result of composing the verbalizing head v to the state predicate open in (55a). We assume that next to its verbalizing role, v also comes with an additional argument R, as shown in (55c). A vP composed with the verbalizing head v gets the feature [+cos].

Our second suggestion is that we have a rule that translates the combination of the head Agent Voice with a vP that is [+cos] as follows:

The rule in (56) says that the combination of the head Agent Voice with a vP that is [+cos] is translated as a relation between events e and individuals x such that x is the agent of e and the relation α′ is a relation between e and the relation cause⁺, where α′ is the semantic representation of α, the vP, and cause⁺ is a relation between events and states.

In the case where the vP is open the door (see (54)), the result of applying the rule in (56) is shown in (57).

Next, we adopt a rule that translates the combination of the head Causer Voice with a vP that is [+cos] as follows:

The rule in (58) says that the combination of the head Causer Voice with a vP that is [+cos] is translated as a relation between events e and situations i such that i causes e and the relation α′ is a relation between e and the relation become, where α′ is the semantic representation of α, the vP, and become is a relation between events and states.

In the case where the vP is open the door (see (54)), the result of applying the rule in (58) is shown in (59).

The rules in (56) and (58) need to be paired with a rule that says that in the absence of a Voice head introducing an external argument variable (an agent or a causer), the relational argument R is also applied to the relation become, as shown in (60).

Syntactically, the application of this rule is triggered if a semantically empty (expletive) Voice head is merged on top of the vP (thereby ensuring that no further argument DP can be added to the vP). This is covered by the first line of (60). The same rule applies if a non-verbal head (i.e., a head like Aspect, Tense, an adjectivizer or a nominalizer) is merged with a vP lacking any Voice head. This, once again, ensures that no further argument DP could be added to the vP. This second case is covered by the second line of (60).²⁵ That is, this rule applies in all cases not covered by the rules treating the composition of vP with Agent Voice or Causer Voice. It applies (for instance) in the formation of anticausative predicates; these are sometimes analyzed as involving an Expletive Voice projection introducing non-active/anticausative morphology but no external argument variable, or as involving no Voice projection at all (see e.g. Schäfer 2008; Alexiadou et al. 2015; Wood & Marantz 2017; Expletive Voice will be taken up in Section 6).

In the case where the vP is open the door (see (54)), the result of applying the rule in (60) is given in (61).

These rules treating the combination of change-of-state verbs with Voice heads are paired with an additional rule that treats the combination of Agent Voice with a vP that is [–cos] as not including the relation cause⁺, as seen in (62).

In this much simpler case, α (= vP) is represented by a one-place event predicate (= α′), and so there is no relational argument R from the vP to deal with.

Additional rules for Causer Voice or Expletive Voice are not necessary, since Causer Voice and Expletive Voice combine with change-of-state vPs only (Schäfer 2008; Alexiadou et al. 2015; Schäfer 2024).

We do not have the space to clarify in detail how modification is implemented in this approach. A crucial point, though, is that our proposal has the characteristic that the relations cause⁺ or become get specified only above vP. This means that inside of vP, we do not have a full event description yet (the relation between the verbal event e and the state s is still unspecified and therefore not assessable yet). As a result, we do not expect an ambiguity to arise depending on whether the modifier applies to the event argument within or above vP: the modifier needs to apply to the event argument above vP for the result to be interpretable.

The final point to settle concerns the behavior of the relation cause⁺. The principle in (63) says that for any event e, state s, property of state P and individual y such that e causes⁺ s, P holds of s, and y is the theme of s, there is an event e′ and an event e″ such that e is the mereological sum of e′ and e″, become holds between e″ and s, and there is an individual x such that x is the agent of e′ (see Piñón 2000 for an early application of mereological sums to the semantics of change-of-state verbs).

This ensures that the causing event denoted by agentivized predicate is the sum of an action performed by the agent and the ensuing change-of-state of the theme (see Section 4.1), and also that an agentivized causative predicate entails the corresponding anticausative predicate (Parsons 1990).

A further rule is not necessary to ensure that the non-agentive causative (with a causer subject) entails the anticausative, since the relation between e and s resolves as become in this case (recall (58)).

We now turn to a last type of transitive construal for change-of-state verbs with inchoative semantics, namely transitive anticausatives (The gas changed its temperature), and provide the first compositional semantic analysis for this type of statements, thereby updating Schäfer (2024). Whereas transitive causative statements report a whole causation event (denoted either entirely or only partially by the verbal predicate, depending on whether the external argument is an agent or a causer), transitive anticausative statements describe just a change, like intransitive anticausative statements.

6 Transitive anticausative vPs

6.2 An analysis of transitive anticausatives

Our analysis of transitive anticausatives will be presented in three main steps, beginning with a discussion of functional nouns (e.g., temperature, size, price, volume, …, see Löbner 2020 and literature cited therein on these nouns).

6.2.1 Functional nouns

Taking temperature as a canonical example of a functional noun, a straightforward idea is that temperature denotes a function that applies to an ordinary object x (its possessor) and then to a time t, yielding the temperature of x at t, as formalized in (80), where ι is the description operator and d is an individual variable for temperature degrees.²⁹

1. (80)

1. temperature ⇝ λxλt.ιd(temperature(t, x)=d) (the temperature d of x at t)

If (for simplicity) the gas is represented as the individual constant g, as in (81a), then the temperature d of g at an implicit time t′ is represented in (81b).

1. (81)
1. a.

1. the gas ⇝g

1.  
1. b.

1. the gas’s temperature (at t′) ⇝
2. [[λxλt.ιd(temperature(t, x)=d)](g)](t′) = (by two applications)
3. ιd(temperature(t′, g)=d) (for a value of t′)

If the function represented in (80) is applied only to g, then the result is the following function on times:

1. (82)

1. the gas’s temperature ⇝ λt.ιd(temperature(t, g))=d

By the way, note that the temperature of the gas would receive the same treatment as the gas’s temperature:

1. (83)

1. the temperature of the gas (at t′) ⇝ιd(temperature(t′, g)=d) (for a value of t′)

Returning to the gas’s temperature, we adopt a DP approach and treat it syntactically as a DP with the gas as a DP in its specifier position and with a syntactic feature [+poss(essive)] contributed by the possessive clitic ’s, which is its head, as shown in (84).

1. (84)

1. 

For convenience, we treat the possessive clitic ’s as semantically inert: normally, it would contribute a two-place possessive relation (e.g., in Hongyuan’s book), but since the property concept noun temperature already denotes a relation (indeed, a function), we can (again, for convenience) treat ’s as semantically inert here. Even so, ’s does contribute the syntactic feature [+poss].³⁰

This concludes our basic treatment of functional nouns, which was one of the three necessary conditions for transitive anticausatives (recall (76a)).

6.2.2 Underspecified-scale verbs of change

Taking change as a – if not the – canonical example of an underspecified-scale verb of change, another straightforward idea is that change denotes a change-of-state relation that applies to a function δ on times (of logical type ⟨e, e⟩), a relation R and an event e such that R holds between e and s and the value of δ at the beginning of e is not identical to the value of δ during s (which follows e), as shown in (85), where τ stands for a function that gives the time of an event or state.

1. (85)

1. [_v[+cos] change] ⇝ λδλRλe.∃s(R(e, s) ∧ δ(beg(τ(e))) ≠ δ(τ(s)))

On the anticausative use, change is unaccusative, and the syntactic argument corresponding to δ in (85) is an internal DP-argument (which moves to SpecTP to check nominative case and/or the English EPP):

1. (86)

1. 

Bringing together the analysis of the gas’s temperature as a function on times in (82) and the analysis of change in (85) is given in (87).

1. (87)
1. a.

1. [_DP the gas’s temperature] ⇝ λt.ιd(temperature(t, g) = d) (type ⟨e, e⟩)

1.  
1. b.

1. [_vP[+cos] [_v[+cos] change] [_DP the gas’s temperature]] ⇝
2. [λδλRλe.∃s(R(e, s) ∧ δ(beg(τ(e))) ≠ δ(τ(s)))]
3.     (λt.ιd(temperature(t, g) = d)) = (by application)
4. λRλe.∃s(R(e, s) ∧
5. ιd(temperature(beg(τ(e)),g) = d) ≠ ιd′(temperature(τ(s),g) = d′))

The last step consists in applying the rule (60) through which the relation R receives its default interpretation in any syntactic context different from Agent or Causer Voice. The derivation of the event predicate corresponding to The gas’s temperature changed (= (64b)), ignoring tense, is given in (88).

1. (88)

1. [_TP [_vP[+cos] [_v[+cos] change] [_DP the gas’s temperature]]] ⇝
2. [λRλe.∃s(R(e, s) ∧
3. ιd(temperature(beg(τ(e)),g) = d) ≠ ιd′(temperature(τ(s),g) = d′))]
4.     (λs′λe′.become(e′, s′)) = (by application)
5. λe.∃s(become(e, s) ∧
6. ιd(temperature(beg(τ(e)),g) = d) ≠ ιd′(temperature(τ(s),g) = d′))

This concludes our basic treatment of underspecified-scale verbs of change, which was another one of the three necessary conditions for transitive anticausatives (recall (76b)).

6.2.3 Transitive anticausatives

In this section, we show how transitive anticausatives are semantically equivalent to intransitive anticausatives despite being syntactically transitive, which is the challenge set out in Section 6.1.

Recall (77), repeated in (89).

1. (89)

1. [The gas]_i changed its_i/∗j temperature.(transitive anticausative sentence)

The LF representation of transitive anticausatives is given in Figure 1. Since transitive anticausatives are formally transitive (for example, they trigger have-selection, while their canonical anticausative counterparts select be in many languages; see Schäfer 2024 for data and additional arguments that these structures are formally transitive across languages), their structure must involve VoiceP hosting the external argument DP in its specifier. Furthermore, the internal argument DP involves a functional noun phrase with a possessive pronoun in its specifier.

Figure 1: LF representation of transitive anticausatives.

We need to explain 1) why the DP located in the canonical external argument position Spec,VoiceP is not interpreted as an agent or causer of the verbal event (as it would be in canonical lexical causative uses), and 2) why the external argument DP obligatorily binds the possessive pronoun inside the internal argument DP.

Schäfer (2024) argues that these two aspects are related and depend on Voice in transitive anticausatives being neither Agent Voice nor Causer Voice, but rather Expletive Voice (as proposed for various other purposes in Schäfer 2008; Wood 2015; Alexiadou et al. 2015; Myler 2016; Wood & Marantz 2017 a.o.). An expletive Voice head provides a specifier position for an external argument DP to be merged, but does not enter any semantic relation with this DP. In particular, it does not assign any Theta-role to it. This answers 1). As further explained below, 2) follows, since, in the absence of an external argument Theta-role for the subject DP, this binding relation is the only way to semantically integrate the subject DP into the overall structure.

We now proceed step by step to the semantic derivation of transitive anticausative statements, starting with meaning the verb change. The semantic representation of change as a transitive anticausative is the same as the semantic representation of change in (85), repeated below.

1. (90)

1. [_v[+cos] change] ⇝ λδλRλe.∃s(R(e, s) ∧ δ(beg(τ(e))) ≠ δ(τ(s)))

Turning to the internal argument DP, the functional noun temperature comes with a possessive pronoun its, which later must be co-indexed with the external argument DP the gas (i.e., bound by it) in order for the transitive anticausative to be available. The meaning of its₁ temperature is composed as follows (cf. (87a)):

1. (91)
1. a.

1. [_DP it(s)₁] ⇝x₁

1.  
1. b.

1. [_DP [_DP it(s)₁] [_NP temperature]] ⇝
2. [λxλt.ιd(temperature(t, x) = d)](x₁) = (by application)
3. λt.ιd(temperature(t, x₁) = d)

Note that the index of the free variable representing the pronoun in (91a) is the same as the index of the pronoun (here: 1).

Applying the meaning of change to the meaning in (91b), we obtain the following meaning for the vP, ignoring tense (cf. (87b)):

1. (92)

1. [_vP[+cos] [_v[+cos] change] [_DP its₁ temperature]] ⇝
2. [λδλRλe.∃s(R(e, s) ∧ δ(beg(τ(e))) ≠ δ(τ(s)))]
3.     (λt.ιd(temperature(t, x₁) = d)) = (by application)
4. λRλe.∃s(R(e, s) ∧
5.     ιd(temperature(beg(τ(e)),x₁) = d) ≠ ιd′ (temperature(τ(s),x₁) = d′))

Recall that since Expletive Voice is semantically inert, both Expletive Voice’ and Expletive VoiceP have the interpretation of the vP (except that R receives its default interpretation become as a result of applying the rule (60)). But since Voice is expletive and does not assign a Theta-role to the subject DP, this raises the question of how the DP in the specifier of Voice can pass the Theta Criterion (Chomsky 1981). Schäfer’s (2024) proposal is that this DP passes the Theta Criterion because it is interpreted as the possessor of the internal argument DP (recall that the transitive anticausative structure in (64c) and its canonical anticausative structures in (64b) entail each other). The only semantic effect at this Voice-level is then that the DP in the specifier of Voice binds the possessive pronoun inside the object DP, thereby acquiring the possessor role from the pronoun.

1. (93)
1. a.

1. Voice′[+expletive] ⇝
2. [λRλe.∃s(R(e, s) ∧
3. ιd(temperature(beg(τ(e)),x₁) = d) ≠ ιd′(temperature(τ(s),x₁) = d′))]
4.     (e, λs′λe′.become(e′, s′)) = (by application)
5. λe.∃s(become(e, s) ∧
6. ιd(temperature(beg(τ(e)),x₁) = d) ≠ ιd′(temperature(τ(s),x₁) = d′))

1.  
1. b.

1. 1 ⇝ λx₁

1.  
1. c.

1. • ⇝
2. λx₁λe.∃s(become(e, s) ∧
3.     ιd(temperature(beg(τ(e)),x₁) = d) ≠ ιd′(temperature(τ(s),x₁) = d′))

1.  
1. d.

1. the gas ⇝ g

1.  
1. e.

1. VoiceP ⇝
2. λe.∃s(become(e, s) ∧
3.     ιd(temperature(beg(τ(e)),g) = d) ≠ ιd′(temperature(τ(s),g) = d′))

Despite the significant syntactic difference between transitive anticausatives and intransitive anticausatives, they are semantically equivalent, which is illustrated by the fact that the event predicate in (93e) and the event predicate in (88) (representing The gas’s temperature changed) are identical.

If the subject DP does not bind the pronoun in the derivation in Figure 1, the string The gas changed its temperature turns out ungrammatical, as the subject DP falls victim to the Theta Criterion. The only way to rescue the string in the absence of binding is to resort to Agent or Causer Voice, via which the subject becomes an external argument. This leads to a (pragmatically dispreferred) causative interpretation of the sentence.

7 Conclusions

In this paper, we have addressed a specific and pervasive type of syntax/semantics mismatch: except in the presence of Agent Voice, change-of-state verbs used in transitive sentences have ‘intransitive’ (inchoative) semantics: they are used to describe mere changes (become events), just like when they are used in intransitive, anticausative sentences (what we called the inchoative hypothesis). In contrast, in the context of an agent, change-of-state verbs describe full causation events (sums of an action and the resulting change). In our proposal (like in the one developed by Wood & Marantz 2017), whether the relation between the verbal event and the state is interpreted as become or cause is determined by the syntactic context. While for Wood & Marantz (2017), the relation is causal in the context of any external argument, for us, it is so only if the external argument is an agent.

A interesting point concerns the link between the inchoative hypothesis explored in this paper and the Agent Control Hypothesis of Demirdache & Martin (2015). The latter states that across languages, zero-change readings of change-of-state verbs tend to require the subject’s referent to be associated with agentive properties (recall the agentive vs. non-agentive ‘wake up’ examples above in Section 4). The inchoative hypothesis captures the generalization that change-of-state verbs tend to describe only changes, except in the presence of an agent subject, in the context of which they describe full causation events (actions followed by their result). The agent control hypothesis derives directly from the inchoative hypothesis, since the latter says that when no agent is present, the verbal event is just a change. A ‘zero-change’ reading is impossible, since the change is then the only event whose existence can be asserted, in whole or in part.

Funding information

F. Martin’s research in the first phase of this project was supported by the European Research Council (erc) under the European Union’s Horizon 2020 research and innovation programme (LeibnizDream, grant agreement No 856421).

Acknowledgements

We are very grateful to Malka Rappaport-Hovav and our anonymous reviewers for valuable feedback and constructive criticisms. We also thank the audience of the workshop Change of State Verbs – Empirical and Theoretical Perspectives for their feedback, as well as Paola Fritz-Huechante and Antonio Machicao y Priemer for their editorial work and infinite patience. Thanks too to Claire Augusta Bergey, Bridget Copley, Julian Grove, Malka Rappaport-Hovav, Ben Sluckin and Todd Snider for sharing with us their insights on various data points of English. Unless otherwise indicated, the data presented in this paper reflect the judgment Christopher Piñón, a native speaker of English, and Fabienne Martin and Florian Schäfer, native speakers of French and German respectively, on the French/German counterparts of the English examples.

Competing interests

The authors declare that they have no competing interests.