2.1 The ‘well-known’ intransitivity alternation

Intransitive alternations (Levin 1993: 33) differ from other alternations in the type of constituent constantly realized. In this case, the mandatory presence of the external argument vs. the optional status of the theme is what sets Intransitive Alternations apart from the set of commonly studied alternations (transitivity alternations such as the causative, middle and conative alternations). Clear patterns have been established as to what kind of verb may enter (in)transitivity alternations and of which type.

Levin & Rappaport-Hovav (1998; 2008) note that, first, verbs allow for distinct argument structure and event representation options and, second, that major asymmetries in this respect seem to be linked to specific meaning components in the denotation. Following previous studies from neighboring disciplines like cognitive linguistics (Gentner 1978), a distribution is proposed whereby verbs from different semantic classes fall into two major types: manner vs. result verbs. Whereas manner-type verbs essentially express the way in which an action is carried out, result-type verbs refer to the establishment of a state as a result of a certain happening involving scalar change. The saliency of these meaning components is conceived in complementary distribution.

Further studies (Levin & Rappaport-Hovav 1995; Levin 2006) conclude that it is the saliency of manner/result denotation what defines major constraints on internal argument realization, defining frame alternations accordingly. For instance, manner verbs typically allow possessor raising (The dog bit me on the ear), unselected objects (The dog bit his way through town), and conative alternations (The dog bit at me).⁶ Crucially, for this class, unexpressed objects (This dog bites) are possible or even frequent. This is important for many reasons, but especially because in the opposite verb class (result-type), which produces typical examples of the probably most popular frame alternation – the Causative Alternation (8) – unselected, unspecified, or unexpressed objects (Levin 1993) are, by rule, disallowed. This condition, illustrated in (9)–(10), places a severe limitation: essentially, it reduces frame alternations for result verbs to transitive/unaccusative variants; that is, structures where the internal argument is invariably realized.

The MRC imposes differences in subevent representation as well. The question of which verbs allow intransitive alternates is explained also in relation to asymmetrically complex semantic structures. For instance, prototypical examples of manner verbs include surface contact and removal verbs like wipe and sweep (Rappaport-Hovav & Levin 1991: 129; Levin 1993; Erteschik-Shir & Rapoport 2010). As action verbs, these verbs express simple event structures consisting of a single subevent. Assuming that subcategorization options for any verb transparently reflect event structure (cf. Levin 1999), the general pattern explains why it is precisely verbs whose core denotation involves only one subevent, that require only one argument. This offers in consequence a less constrained context for object realization. Agentivity is in general central to this pattern: provided the core denotation is a manner of action, only one structural argument – the actor – is required, hence satisfying (11).

As a result, unexpressed or null object constructions become naturally available for certain verbs insofar as two conditions are met: (i) there is an explicit actor participant involved; (ii) the predication is eventive and structurally simple. These circumstances produce the variants in (12) (see also Levin 1993: 99).

What (12) further shows that intransitive alternations also come in many guises. Along with way-constructions (12)b, conative alternation (12)c and out-of (12)d variants, unexpressed, null, unspecified, and arbitrary (PRO-arb) object realizations (Levin 1993), together with objectless (Mittwoch 2005; Lemmens 2006) and deprofiled object (Goldberg 2001; 2005; Lemmens 2005) constructions like (12)a have been widely discussed.⁷ The fact that for all of them, the grammatical subject bears the same semantic relation to the verb (Levin 1999; Levin & Rappaport 2008) poses interesting parallelisms with other widely-studied transitivity alternations like the causative alternation, where it is also the constant argument (in this case, the internal argument) that preserves the semantic role (theme) in relation to the verb’s predication (Levin 1993: Levin & Rappaport 1995; 2008). An overarching notion thus emerges, as transitivity alternations commonly discussed – centrally, transitive-to-objectless and transitive-to-unaccusative variants – all preserve the semantic role across alternates.

Now, from all the types mentioned above, we will focus on a particular kind of intransitive alternation, the Characteristic Property of Agent Alternation [CPAA], taking (13) as a toy example.

Several properties of this alternation are central to the discussion. First, verbs allowing CPAA are almost exclusively instances of ‘manner’ verbs. As argued above, it is generally agreed that, among transitive verbs, those showing more flexibility for (non)specification of the object involve a prominent manner component in the denotation (Mittwoch 2005). A verb-lexicalized manner entailment and its characteristic global interpretation would explain the distribution drawn, namely, by adverbials like those in (14)a–b. Moreover, it would further capture the computation of frequency adverbials like always in (14)c in relation to a characteristic property of the subject constructionally created by pluractional implication (i.e. repeated action > manner of behavior).

Various studies stress that in all these variants the emphasis is placed on the action and the agent (Lemmens 2006). Goldberg (2001) notes that it is the coincidence of a de-emphasis of the patient and emphasis on the action that is essential to ‘objectless’ – strictly, omitted object – alternatives for otherwise transitive verbs, illustrated in (14). This captures two relevant conditions on representation – simple event, actor participant – conforming to the Structure Participant Condition ((11) above). Concomitantly, Rice (1998: 206) notes that because these constructions evoke general semantic scenarios, the object is “fairly unimportant as the pragmatic focus is on the activity itself”. Essentially, the intransitive variant is used to express that the subject typically shows a disposition for the action named by the verb.⁸

What (15) additionally illustrates is Rice’s, Levin’s (1993: 38), and Goldberg’s (2001: 506) intuition about an understood/implied easily retrievable object (e.g. much [alcohol]) (e.g. (14)a above) essentially based on cognitive grounds. The conditions summarized in (16) reflect these points.

From here – pragmatic reasons aside – the relative irrelevance of the theme/patient to the core predication and the weaker conditions on its realization seem to always follow from the kind of event structure involved, which is assumed to decompose into something like (17). This layout would be coherent with the vP configuration usually assumed for the same verbs in constructional accounts ([Init, Proc] verbs in Ramchand 2007: 214, V_DO-headed verbs in Folli & Harley 2005, to name a few) where the only logical (default) argument is the external one –i.e., one defined not by the verb, but rather by construction.

Properties and patterns characterizing this alternation will be discussed in detail below. Yet, before we proceed, a different type of (in)transitivity alternation must be introduced in order to make clear the contrast at issue here.

2.2 The ‘other’ transitivity alternation

The constructions addressed next offer examples of a relatively underexplored alternative. These variants feature monadic frames for result verbs, which are, as anticipated, structurally — semantically and syntactically — distinct from the monadic (unaccusative) frame commonly discussed: in these constructions it is the external argument that is kept constant, while the unique argument is not interpreted as a patient or resultee but as a possible cause. Basically, what is expressed in these constructions is that an entity (subject) has the potential to motivate a change of state associated with the lexical root.

The observation that verbs appearing in these constructions are precisely those entering the causative alternation is problematic for a simple reason: such verbs are generally classed as result-type verbs (cf. Levin 2006; 2017), and, under the transitivity pattern defined by the MRC, unexpressed, null, or unspecified internal argument occurrences So often seen in traditionally considered intransitive variants (e.g. cpaa), are disallowed (Levin & Rappaport-Hovav 2008 i.a.). However, occurrences like (18) are not rare. Similar examples involving canonical result verbs like kill figure prominently in the literature on optional and omitted objects (cf. Lemmens 2006).

Two things are especially noteworthy about constructions like (18). First, they contrast with the ‘intransitive’ null/unspecified/omitted object variants discussed in Section 2.1 above, which characteristically involve an active participant with active volitional attitude or disposition to act in a punctual (iterated) event (cf. (19)).

Second, the systematic productivity of constructions like (18) challenges the view that, for verbs allowing scalar change-of-state predication – a key property for result-type classification –, transitivity alternations seem naturally restricted to unaccusative and transitive frames. Nevertheless, Romance languages like Spanish, Catalan, Romanian, Portuguese and Italian, or even Greek (cf. Alexiadou & Iordachioaia 2014; M. Rasia 2021; Kallulli 2021) allow this alternative almost unrestrictedly, as (20) also shows.

Although more restrictedly (cf. (20)), English also allows for a fair set of what seem objectless alternatives of prototypical result verbs including causatives Stop;¹⁰ literal glosses to many Romance productive examples are in fact possible, as in (21).

There are three points to consider here.

First, the DP subject in these occurrences is necessarily inanimate. This sets a clear contrast with the usually-discussed intransitive alternates like (1), (13)-(14) and (19) above – e.g., Mary drinks or This dog bites.

Second, intransitive variants like (20)–(21) consistently follow stative patterns.¹¹ Both points are important because constructions like (2) deviate from the usual profile of ‘objectless’ predications as to the two relevant conditions: active participant and agent-controlled event with (volitional) denotation. Instead, they seem to lack them both (cf. (22) vs. (16) above).

Third, these constructions fail all relevant tests for null/_arbpro constructions.¹² Namely, Constructions like (23)–(24) disallow object-oriented secondary predication (Rizzi 1986) and null-object quantification. In Italian, ne-clitization and quantifiers used as null-object diagnostic (bare molti, Cattaneo 2008) are equally resisted.

Additional tests unanimously indicate that these are not null or missing object instances. Namely, Spanish shows deviant results with null-object quantifiers like muchos ‘many’, in parallel with (23).

(25) is interesting as it reveals a significant difference in the computation of the intensifier, which is not interpreted on a quantized (null/unrealized/unspecified/missing) object, but it rather is computed in reference to the disinfecting capacity of alcohol – that is, in the stative sense, maximizing the degree to which the gradable property holds for the subject as individual-level characteristic. The English gloss in (25)b, expressing the difficulty to interpret the partitive quantifier some has been used as a test for null and objectless constructions (cf. Mittwoch 1982: 120, Levin 1993); this contrasts with the occurrence of something, which is typically allowed with null and arbitrary/generic objects.

Similarly, (26) shows that further null/arb object-oriented quantifiers are fine in prototypical cpaa examples like (14)a above – repeated in (26)a – but not in (26)b. Also here, the quantifier, if allowed, gets once more a stative reading where the dehydrating capacity of caffeine as a possible cause is measured, just as in (25)a above.

In addition, and what is more important, there is the fact that subject-as-possible-cause monadics are systematically productive with both unpassivizable verbs and verbs by default incompatible with null objects (M. Rasia 2019).

Taken together, empirical observations of this sort contribute to our proposal that the constructions at issue constitute genuine monadic atransitive variants [AV]. This involves the assumption that we are not dealing here with an underspecified, null, or generic object, but rather with a structure where no object is structurally licensed at all. If correct, a configurational contrast between atransitive vs. arbitrary/unexpressed/unspecified¹³ object constructions would support the proposed existence of asymmetrically different (in)transitive alternations. Each type would Accordingly define have a distinctive behavior and distributional semantic pattern coherent with their relevant denotational and configurational properties, coherent with the MRC.

2.3 Two characteristic property alternations

There are three points objectless¹⁴ and av constructions share. First, cpaa and AV constructions fit the notion of characterizing sentences in the sense that they both express some distinctive trait in an entity, such that a change in the character of the entity arises if the description is altered (Deo et al. 2013). Second, they respond to the central condition in (27) (see also Park 2009). Third, cpaa and av constructions both apply for the Structure Participant Condition provided there is an only required argument (single participant) involved in the expression of a simple event (single eventuality).

Importantly, however, they apply in different ways. Essentially, CPAAs and AVs build on distinct sorts of simple atelic types – actions vs. states. This yields distributional differences suggestive of a configurational contrast imposing disparate selectional and interpretive patterns. There are three basic examples:

One. In standard generalizing sentences taking a missing/unrealized/unspecified object, as CPAAs do adverbs like usually (also, when, whenever) yield the entailment that there may be exceptions to the rule established by the predicate – essentially, a series of events that define the subject by habit or disposition to a certain behavior (Krifka et al. 1995), as in (28)a.¹⁵ Instead, if the object is realized, episodic event reports are entailed, as in (28)b.

By contrast, AVs fail to license habitual entailments in which exceptions or omissions are derived. This follows naturally from their structural semantic properties – specifically, three: (i) a clear lack of subject animacy/volitionality, shown in (29), (ii) distinctive generic dispositionality entailments, and (iii) a clear non-episodic nature.

Moreover, AVs yield non-deontic predications where real event instantiations are not required, as shown by (30) and epistemic effects arise with modals like must. Consider the contrast in (30).¹⁶

Two. Important distributional differences also touch on tense compatibility. Whereas prototypical examples of unexpressed/null object alternatives and objectless constructions usually involve perfective tenses (cf. Levin 1993: Mittwoch 2005; Lemmens 2006), for AVs, by contrast, the perfect tense is odd, showing instead the affinity with generic tenses typically expected from stative constructional variants.¹⁷

Consequently, if allowed, the past tense yields the LifeTime effect expected in temporally unbounded (Individual-Level) stative predications, entailing that the holder of this stable property no longer exists (Carlson 1977), as in (33). This contrasts with the discontinued habit reading in prototypical objectless constructions like those in (34), including normal contexts like concatenation ((34)a) and division of labor ((34)b).

Relatedly, in standard objectless constructions punctual adverbials define action iteration (event multiplicity) as a source of the habit(ual) entailment, as in (35), while frequency adverbs stress a cyclic reading, as in (36).

In fact, it has been often noted that adverbials facilitating habitual and iterative readings improve grammaticality in those cases where the object is selected but left unspecified (Rice 1988; Bender 1999; Lemmens 2006). Such observations are consistent with a subject characterized by repeated actions dispositionally performed by an actor.

The requirements on occurrence frequency in objectless constructions (cf. Deo et al. 2013) explain not only their usual combination with adverbs like typically, usually, generally, and phrases marking temporal/situational frames like (35), Moreover, the absence of similar adverbs is problematic with subject DPs with not-so-clear volitional reading:

Three. The proposed analysis effortlessly captures Rizzi’s intuition on prototypical null/arbitrary object examples like (37).

Four. Minimal pairs, like those in (40) below, show that with DP subjects eligible for inanimate/stative/involuntary causer readings, an av predication becomes available. This variant is incompatible with perfectivity, deonticity, and frequency, thus stressing the need of animacy/agency/volitionality in the subject to license a missing object frame——exactly as our analysis predicts. Instead, when the DP subject guarantees an agentive/active reading, and the verb allows entailments related to habitual events characteristically performed by a volitional actor, as in (40)b, the objectless (cpaa-like) construction becomes prominent again.

For all these facts, a natural explanation building on a constructional (configurational) contrast, based on the transparent event-to-syntax correlation pursued, is available and discussed next.

2.4 Analysis

From the previous (sub)sections, a major contrast between distinct intransitive alternation types emerges. According to a series of conditions evaluated above, it can be noted that intransitive alternatives commonly discussed in the literature (cpaa/objectless type) involve no major differences with respect to the dyadic variant: across frames, dynamicity and agentivity are generally preserved.

By contrast, when a monadic (atransitive) variant of a result-type verb is produced, all hallmark properties of the result class, like scalar change, resultativity, eventivity, and episodicity are disabled, leaving a mere (Individual-Level) predication instead. We propose that this major aspectual difference between result-type-based avs and internal-argument-licensing (causative/anticausative) forms of the same verbs – i.e., intransitive variants commonly discussed for this type –, like (42)b, can be explained grammar-internally, and so can the fundamental differences between AVs and standard objectless constructions with manner-type verbs like (42)a.

Our hypothesis is that this asymmetry – along with the specific implications analyzed in Section 3.3 – immediately follows from the absence of specific heads playing a critical role in argument structure and event representation. Consequently, the stativity of AVs would be a predictable configurational result of an optional non-merger of such heads. To propose this, we build on aspectual tests presented above offering evidence that the absence of the relevant (internal-argument-licensing) head correctly correlates with the absence of its (sub)evental (aspectual) structure counterpart.

The component in question is the one generally associated with the eventive denotation – notably, the BECOME prime in Levin & Rappaport-Hovav (1995), Dowty (1979) i.a. This component is structurally linked to the (sub)structure introducing the internal argument – the undergoer of the change in the thematic/argument structure. Following the same combinatorial logic, stative occurrences of result verbs, as a pure configurational result, would be by default possible. Importantly, this holds under the condition that the matrix verb is not combined with the inner verbal head furnishing the verb with the ability to license internal arguments and introducing the scalar change denotation (V₂ in the V₁→V₂ subevent schema in Hale & Keyser 2002; 2005; [vcause [vbecome]] embedding in Levin & Rappaport-Hovav 1995). The caveat emerging here is that transitivity in these verbs is a constructional result, automatically obtained, via Merge, by virtue of the complement (v-V) relation (Hale & Keyer 2005: 12), as in.

The aspectual properties noted in e.g. (42)a would therefore follow naturally, provided that the originally monadic verbal head is independently realized, licensing a sole subject participant who is a mere holder of a property, as in (44).¹⁹ This structural difference should account for their behavior in relation to the standard causative-inchoative alternation but also to unexpressed/unrealized object alternatives discussed.²⁰

Importantly, while still conforming to Hale & Keyser’s (2002; 2005) compositional process – obtained by direct root(sister-to-v)-incorporation also giving place to unergative verbs (cf. also Harley 2005) – the proposed structure also conforms to more recent constructional proposals specifically addressing causative stativity, like Rothmayr (2009). A direct combination of the matrix verb with rhematic information (that is, skipping merge with internal-argument licensing heads) would effortlessly accommodate the facts discussed above independent of properties that come with v-V composition. In doing this, they still meet the requirement that this kind of projection (Rheme) fill the lowest slot in the subject-holder configuration (Ramchand 2007, Berro 2015), as in (45), producing a stative verb in consequence (2007:115).²¹

In (45) the rheme projection is interpreted as a stative relation linking a property to the DP in specifier position. Since there is no ResP, and the rhematic material is not selected by a process-introducing head (proc) but rather by an aspectually underspecified projection (Init) headed by a defective v head (Ramchand 2007: 44),²² the argument is automatically interpreted as the holder of a property – an initiational state – and not as the holder of a ‘result’ (cf. Ramchand 2007: 53). Whether this property is associated with a change of state is irrelevant here, in the sense that the scalar change component is structurally not represented and is not entailed in the core denotation (ultimately, it is merely implied).²³

Now, reconsider the contrast in (46).

It is important to note here that the specific interpretation of the subject – whether it is a volitional agent, causer, force, experiencer (e.g., of a psych status), or holder – depends on lexical, conceptual and syntactic (configurational) factors. These include animacy (Arad 1998: Doron 2020), additional semantic underpinnings of the predicate (Alexiadou et al. 2017), and the presence of proc and res projections (V₂ and P_TerminalC on Hale & Keyser’s terms), respectively. In addition, provided that proc is not mediating between InitP (i.e. vP) and the rhematic information at the bottom of the configuration, the subject in (46)b is interpreted as a stative holder. This contrasts with (46)a, where the external argument gets instead an actor reading.

The configurational restriction to a stative holder computation explains that no relevant differences are found with DPs lexically limited to inanimate readings (cf. Zyklon B kills).²⁴ Animated (volitional) readings of the DP – think of Covid kills to survive –, require that the eventive core (proc) be merged.

Under the proposed structural homomorphism, eventivity means, in principle, an internal-argument-licensing structure (VP). This agrees with the premise that manner-patterning events characteristically involve a ProcP argument guaranteeing the entailment of an initiator (agent) licensed by the external-argument-introducing head (InitP in Ramchand 2013; 2018), which is, recall, aspectually unspecified. The Characteristic Property sense in Levin’s intransitive alternates would thus involve a manner-of-behavior property whose holder is originally an actor; specifically, the argument of the eventive proc associated with the V head. A structure like (47) would account for such (cpaa-like) cases.

If correct, two important conclusions arise: first, in cases like the cpaa, no real argument frame alternation seems involved; at least not to the extent seen in AVs (which characteristically lack internal-argument-licensing structure).²⁵ Concomitantly, atransitive (AV) structures, with the underlying configuration in (45), contrast in all the relevant points: there is no emphasis on an action and/or agent (Lemmens 2006), the subject is not described by manner-of-behavior or tendency to certain action, and, most importantly, it cannot be presented as a volitional actor performing an agent-controlled event, even if this parsing is lexically possible (though semantically and conceptually) odd. This disparity crucially explains the pattern illustrated at the beginning of this subsection ((42)a-b above). In turn, minimal pair contrasts like (48) empirically summarize the points just raised.

These selectional restrictions are important under the transparent mapping between event structure and participant selection rule. In both alternatives, a simple eventuality with a single mandatory participant appears – an activity in one case, a state in the other. In both cases, the subject is the holder, but in radically different characterizing predications. Distinct vP configurations, with predictable semantic properties, account for this.²⁶

Whether and how these findings can be extended to other verbs like psych verbs is the topic of the next section.

3.1 A productive asymmetry

There are two properties that Object Experiencer psych verbs and canonical causative verbs like those analyzed above share. First, they both alternate between transitive and unaccusative frames, taking special morphology for the anticausative (unaccusative) variant in languages with nonequipollent derivation like Romance (se-Cl) or Greek (NAct inflection). Second – and this is somehow new to the discussion – they both allow for monadic variants (avs) featuring a possible cause as only argument.²⁷ This option, along with the asymmetries defined by the (un)availability of intransitive (av) variants across languages (e.g., Greek, Romance vs. English), leads to four important, strictly empirical, observations.

One. Whereas AVs are, as noted above, possible in standard result verbs (cf. Section 3.2) in English, this alternative is apparently disallowed for psych verbs. Things are different in languages like Romance or Greek,²⁸ where atransitive structures are systematically productive with psych verbs, as (49) shows.

Importantly, this option is available even with psych subtypes not licensing themes or anticausative alternates, like (50) (inverted IO verbs, Nishida 2016).

Two. When AV production is disallowed, English recruits an analytic (be+-ing) construction to convey a similar meaning. In turn, the monadic variant gets an unaccusative (anticausative or middle-type) defective reading in which the single visible argument is generally interpreted as experiencer or undergoer (cf. Pesetsky 1995: 56; see also data in Lekakou 2005) (cf. (51)). This happens even when the DP is lexically eligible for agentive and/or cause readings, as (49)b above and (52) equally show.

In languages like Romance or Greek, by contrast, an unmarked monadic construction has defective AV (subject-as-possible-cause) reading, whereas special morphology (se-cl/NAct) aligns with anticausative or middle computation.

The relevant observation here is that the asymmetry in AV productivity seen above aligns with the asymmetry created by (non)equipollent derivation.²⁹

Three. Some psych verbs do allow intransitive alternates in English, as the gloss in (54) indicates. The difference is that, for these verbs, languages like Romance offer an option between AV and cpaa readings unparalleled by English, where possible readings instead alternate between anticausative and cpaa constructions. Thus, whereas Cause/Originator (Borer 2005) and Actor readings are equally available for the single participant of the eventuality in Romance and Greek, English opts between Actor and Experiencer defaults. The option, note, involves a major structural shift regarding the configurational slot assigned (external vs. internal argument). This disparity is avoided in Romance or Greek, provided the argument is consistently computed as external in both (cpaa/av) constructions.

Four. In Romance, animate nouns allow distinct readings coherent with the crosscut in intransitive alternatives. One reading describes a non-volitional property that characterizes the subject as a potential cause, subject matter or trigger (Levin & Grafmiller 2013) of a mental state. The other describes the subject as a volitional participant in a controlled, intentional action with an implied psych result defining the subject based on a dispositional behavior (action). The disparity is not trivial, as the glosses (boldened in (54)) show.

Tests like adverbial modification support this contrast: whereas quantifiers and intensifiers scope on the active/agentive event, yielding event frequency or intensity in the agentive reading, a purely stative sense (property degree) is yielded, coherent with its ILP nature, in the AV reading.

In English, the intransitive is possible as long as the subject can be interpreted agentively, and the verb allows for manner-of-behavior denotation. Since the only syntactically licensed alternative takes an external argument in an agent-controlled event, the reading in (55)b is effortlessly computed, whereas (55)a is not.³⁰ Consequently, if the subject cannot be interpreted agentively, as happens with concern (Levin 1993: 191), the English gloss fails.

Further English examples equally suggest that if the subject cannot be interpreted as an actor in an agent-controlled event, a single-specified participant construction fails. As a result, deviant acceptability is dependent on the choice of subject, following specific semantic properties like ±animacy/±volitionality. This is a predictable outcome when alternations are limited to one (cpaa) type. This crucial effect captures the distributional contrast in (57), showing that English psych verbs systematically refuse the inanimate subject-as-possible-cause seen in AVs.

In Romance, in principle, no such asymmetries emerge. ±animate subjects are effortlessly accommodated in intransitive frames under a stative cause reading (AV-type computation) remaining systematically available (cf. (58)). In turn, when the subject DP allows for animate/active readings, two alternative (AV/cpaa) computations coexist.

Summing up, various patterns indicate that in the set of verbs traditionally classed as psych, only those allowing an event-based agentive reading with a prominent manner component generally allow for intransitive alternates (manner-of-behavior cpaa), imposing interpretive and selectional restrictions on the subject accordingly. The output of the consequent productive asymmetry – vs. psych verbs freely deriving av variants outside English – and its crosslanguage expression, is illustrated in (59). The resulting distribution in complementary patterns offers new evidence of a major (psych-internal) verb class difference.

Relatedly, the contrast in auxiliary distribution for verbal passives in Italian – a Romance language in which auxiliary distribution is a prominent indicator of event/syntactic structure – is quite suggestive in this regard. Note the possibility of an eventive (venire da) passive construction depending on the verb type, as predicted by (58).

Importantly, eventivity goes hand in hand with manner-of-behavior denotation. In English, only active/agentive psych variants are systematically productive. Accordingly, adverbials expressing volitionality, event runtime, and specific situations for eventive instantiation are naturally predicted to fit well with the kind of intransitive variant available, as (61)a, as opposed to (61)b, indicates.

In the analysis, eventivity is crucial.³¹ Structurally, dynamicity is responsible for triggering the manner-of-behavior reading. It thus rests, under the standard analysis, on further composition with the inner V head introducing the eventive (process) component (given a standard v[V°] configuration) and providing the verb with transitive capacity. The structure proposed is repeated below.

This means that ‘intransitive’ alternates allowed in English would be configurationally closer to Levin’s intransitive – strictly, speaking unexpressed/unrealized/proArb object – alternates than to atransitive configurations assuming that an internal-argument-licensing head ultimately figures in the representation (with (47) above matching (62)).

If this turns out as the correct generalization, the conclusion that follows is that intransitive alternates in languages like English reduce, for both psych and non-psych verbs, to internal-argument-licensing (unaccusative or null/unspecified object) frames.³²

The general idea that the crosslanguage asymmetry seen in psych (oe) verbs correlates with the (un)availability of a construction amenable to the atransitive variant in standard causative verbs finds support in patterns. Importantly, these patterns draw similar contrasts with cpaa/objectless constructions. Namely, intransitive variants allowed in English for (some of the) verbs traditionally treated as psych are deontic, carry pluractional or iterative implications, can be situated in time, and, moreover, allow perception reports rejected by pure psych predicates (Arad 1998 i.a.) The general picture is, hence, too similar to the one just as seen in Section 2.

Many examples, like (63) and (64), reflect the conditions on objectless constructions listed by Goldberg (2001) and Lemmens (2006), including habitual contexts, compatibility with instrument adverbs, and verb series (Massam & Roberge 1989).

The examples in (65) further show these observations. At the same time, they introduce another important indication. Note that the (cpaa) available variants foreground (Goldberg 2001) an eventuality characterizing the subject by describing an action computed as manner of behavior. That the manner component is central to this effect becomes crucially visible with defeasibility. Because a psych component (the emotional resulting state) is not necessarily entailed, its cancellation does not yield a contradiction.

Importantly, the fact that intransitive flexibility is favored by eventive (agentive/volitional) denotation, coinciding with the non-central role of the psych component (result/change undergone), offers a connection with the premise behind intransitive flexibility according to the MRC (discussed in Section 3). Martin’s corollary on defeasibility under agenthood/animacy captures these points (66), while efficiently accounting for the asymmetric productivity and distribution summarized in (57) above.

From here, a general observation follows. The full proposal is offered next.

3.2 Psych verbs: class-internal subtypes

In the matter of aspect and argument structure, psych verbs raise, as anticipated, a complex question. In addition to the many peculiarities concerning their syntax, the discussion on semantic aspects which are central to argument structure representation is also far from settled, including their Vendlerian type.³³

The most commonly found observation is that psych verbs enter the eventive-stative alternation (Rothmayr 2009; Alexiadou 2018), alternatively yielding eventive predications with animate subjects (agents), and states with inanimate subjects (experiencers). Other accounts (e.g. Arad 1998) focus on the availability of canonical syntax, as opposed to the special syntactic behavior known as psych effects (Belletti & Rizzi 1988; Arad 1998: 9), in agentive uses.

To solve this general puzzle, Arad (1998) proposes a threefold distinction between: agentive uses with an active participant (67)a, eventive uses with involuntary subjects (67)b, and stative uses describing a mental state (the typical psych reading, illustrated in (67)c).

On this view, there are no such things as predesigned psych verbs, but rather normal verbs participating in the expression of mental states. By setting (67)c apart, the proposal captures the fact that the syntactic peculiarities noted by Belletti & Rizzi limit to the psych use; thus, as soon as an agentive reading is forced, psych effects disappear (Arad 1998: 12).

In principle, the variation between psych and agentive representations is purely configurational and depends on the identity of the V head combined (v/V) in the derivation, producing (68) accordingly.

In a more recent study, Doron (2020) analyzes verbs with noncanonical syntax like interest, fascinate, disappoint, bore to argue that what makes these verbs different is that they express locative relations, in line with Landau’s (2010:11) original proposal. The idea that psych predicates are modeled on locative relations goes back to Jackendoff (1990), who proposes a conceptual semantics for dyadic psych predications like X frightens Y decomposed into an abstract locative representation.

This offers a possible solution. If psych verbs are grammatically different in several respects, and their semantic representation is conceptually modeled on abstract locative representations, the productive asymmetry would follow from the fact that a construction realizing only a causer of terminal (psych) location – conforming to the AV representation – is simply not possible in languages like English. In terms of event and argument structure this makes sense, as there would be no participant for the psych experience to be situated – in Jackendoff’s terms, no container (experiencer) in which the emotion (located object) can be situated. Yet, the question remains as to why the same structure is syntactically and semantically possible – and, moreover, fully productive – in many languages, like Romance and Greek.

We want to explore a different answer here. If we take up the productive asymmetry analyzed above, along with the verb-type difference including the (un)availability of alternative configurations and semantic (evental) representations, we can conclude that intransitive alternates are generally possible in languages like English for a specific verb type. More specifically, if we consider that: (i) the type of alternative allowed is the underlyingly transitive type (null/unrealized object); and (ii) its denotation takes an active participant in an agent-controlled action, it follows that intransitive alternations are productively available for psych verbs with canonical transitivity (null object selection) and not restricted to psych state predications – i.e., allowing manner-of-behavior computations.³⁴ Such conditions would accommodate the distributional facts noted above along with their crosslinguistic distribution.

The idea of distinct psych classes has gained prominence over the last decades (cf. Marin 2011 for summary). Recent works analyzing Mandarin, English, Spanish, Romanian, Greek, French, Finnish, and German psych verbs find a non-trivial contrast in transitivity, semantic entailments, and aspectual structure (Marin & McNally 2011; Marín & Fábregas 2020; Martin 2007; Alexiadou & Iordachoiaia 2014 among others). Specific differences are found in agentivity, eventivity, iterativity, and entailments associated with subject computation (e.g., pluractionality, Agent/Force, and volitionality, see Doron 2020, Marín & Fábregas 2020 and references therein) with syntactic effects. Notably, it is shown that a subset of verbs of the object-experiencer type allows passive and agentive constructions with zero-Change-of-State denotations [zero-CoS], illustrated in (70), while another subset does not.

Several studies find a strong link between eventivity/agenthood, passivization (canonical transitivity) potential, and defeasible causativity (Martin 2019, Martin & Schaffer 2017). In predications headed by a causative verb in which an action (behavior) concentrates the core denotation, the result component (the psych state) can be omitted or canceled without contradiction (recall (65) above). Crucially, this is not possible for all (psych) verbs.

In our case, we notice that psych verbs rejecting objectless variants in English are those having the psych state necessarily entailed, as in (71)b. Under a transparent event-to-syntax mapping (Structural Participant Condition, cf. (11) above) this makes sense, provided the (psych/result) entailment places a structural requirement: the argument realizing the mental state must be specified.

Apparently, in cases where the psych state is not a core part of the verb denotation, the corresponding (experiencer) argument can be omitted. Verbs crosslinguistically allowing intransitive variants turn out to be precisely those describing defeasible psych states (Martin 2019) in agent-controlled events.³⁵ Recall (65), repeated in (72) below:

Conversely, only an inner state that is necessarily entailed – hence, requiring structural representation – can be, for instance, tracked by verbs like keep.³⁶ This option fails in frame alternating (cpaa-friendly) verbs.

Agentivity plays a crucial role on defeasibility and zero-Change-of-State (cf. (66) above) conforming to the Agent Control Hypothesis.³⁷ Also here, eventive passives show a relevant contrast. Thus, while by-phrases identify agentivity/eventivity in defeasible (non-core) psych verbs, consistent with (71)a, with-adverbials guarantee that the animate subject in the objectless variant is automatically computed as an agent rather than as a cause (Martin & Schäffer 2017: 95). Similar cases, centrally combining an active Agent and a cancelable result are attested in the literature (cf. Copley & Harley 2018).

The immediate conclusion that follows is that the two criteria postulated by Arad to distinguish eventive (canonical transitive) uses of psych verbs and stative (typical psych denotation) predications – (I) whether there is an agent deliberately doing something; and (ii) whether there is change of state in the experiencer – stand in complementary distribution, with direct impact on argument structure alternations.

Our hypothesis is that Arad’s intuition on verbs which are not per se psych – those for which canonical transitivity is allowed but ‘disappears’ in psych uses (1998: 12) – refers to a lexically coded difference. It would be the case of verbs known for having both a psych and a “physical” use (cf. also Ramchand 2013); that is, verbs which are originally not psych, provided that they are not limited to psych denotation across all their uses – which is the standard criteria defining lexicalized content (Levin & Rappaport-Hovav 1998; 2000) – and for which the psych component is characteristically defeasible. Prototypical examples of this set include, precisely, verbs like disturb and bother (Arad 1998; Bouchard 1992) – i.e., the ‘normal’ verbs in Arad’s proposal allowing for canonical syntax and eventive/agentive denotations. These verbs are, in fact, those providing most examples of psych objectless alternatives across languages, especially those considered here.

By contrast, the distributional semantics of psych verbs suggests that whenever a subject cannot be interpreted agentively – i.e., in a manner-of-behavior predication – the verb will only yield a non-defeasible predication. This would be the case of verbs invariably appearing in non-eventive psych predications (Arad’s psych use), with non-agentivity and non-defeasibility restricting intransitive productivity and (non)canonical syntactic patterns accordingly.³⁸ In short, core (lexically-(pre)defined) psych verbs.

3.3 Summary of the proposal

Above, we contend that verbs crosslinguistically allowing objectless variants would be ‘normal’ verbs which could, given the semantic properties of their root, appear in psych predications. The distinction proposed between verbs allowing manner-like patterns and those that do not as a fundamental parameter to predict and explain intransitive alternations crosslinguistically is motivated by five facts, outlined next.

First, in the intransitive psych variants allowed in English, aspectual tests like compatibility with stop and What x did frames reveal a dynamic denotation (Dowty 1991). This contrasts with the atelic non-eventive (stative) patterns shown by non-alternating verbs.

Second, for both types incompatibility with culminative terminar ‘finish’ is predictable from the atelic nature of either (activity/stative) eventuality constructionally expressed. By contrast, the choice of terminative (stop-equivalent) verbs reveals, at least in Spanish, distinct termination senses. Whereas dejar may apply to stative cause(r)s, parar is distinctively associated with volitional dynamic predications (Marín & McNally 2011, Marín 2011, Marín & Fábregas 2015) suggestive of an agent-controlled type of termination. Consider the constraint, for alternating verbs, on volitional actors according to subject choice in (77).

Third, psych verbs not allowing av variants in English – i.e., those yielding monadic frames only in Romance (or Greek) – do not license agent-controlled termination, as (78) indicates.

(78) is coherent with the observation above on the subject being distinctively interpreted as inanimate (subject matter) argument in pure psych denotations (Rothmayr 2009, Alexiadou 2018). Therefore, whereas adverbials tracking the manner in which a volitional event is performed are fine for verbs with generalized (crosslanguage) intransitive alternation (i.e., the alternating type), the non-alternating type – the (sub)set with av productivity limitations – yields a middle reading rejecting agentive manner adverbials instead.

Imperatives – generally linked to agent-controlled actions – show a consistent distribution.

Fourth, in English-allowed intransitive constructions the quantifier scopes on the intensity/frequency of a (manner of) behavior, whereas the psych state can be cancelled without affecting the truthful occurrence of the event (cf.(81)a). Such conditions contrast with verbs not eligible for objectless (cpaa) alternation, which show restrictions expected from invariably stative, non-defeasible (core psych) verbs like bore and worry ((81)b).

Fifth, the distribution of accusative/dative clitics in (82) furnishes the intuition that, when used agentively, alternating verbs behave like normal transitive verbs (Arad 1998). Thus, accusative clitic marking in Spanish is only natural with agent subjects, yielding a dispositional activity (cpaa-like) construction with the expected pattern (Ackerman & Moore 2001). Instrument and control/purpose adjuncts, essentially describing the manner in which an event (activity) is performed, show a similar distribution.

In Greek, clitic doubling is allowed for actor subjects in eventive constructions but blocked for the psychological (inanimate subject) predication, mirroring the Spanish split in (82).³⁹

These facts support the existence of two structurally distinct verb types. Further support comes from works on agentivity and pluractionality in Spanish.

Marin & Fábregas (2020) also note the existence of two oepv subtypes. Taking amedrentar ‘intimidate’ and preocupar ‘worry’ as main examples, they build a principled distinction according to two conditions resembling the discussion above. These are: (i) the denotation of specific events performed by an Agent, and (ii) the reference to event iteration, tested with modifiers favoring pluractional readings. In addition, the construction must be episodic (2020: 117) and compatible with temporal adverbs setting identifiable points in time when the event occurred, as in (84). In this way, set verbs allowing eventive and agentivity-related structures (the amedrentar type) are set apart from those that do not (the preocupar type).

Crucially, these conditions apply to verbs with generalized intransitive alternation (cf.(79)-(81) above). Deonticity and episodicity show similar results for canonical manner verbs and psych types allowing objectless frames ((84)a), as opposed to non-alternating (proper psych) verbs ((84)b). Event iteration (repeatedly), and time-point adverbials tested by Marín & Fábregas support the distribution.

Dispositionality, habituality and other typical conditions facilitating unexpressed object constructions (Lemmens 2006; Mittwoch 2005) are also present in these examples. Such patterns instantiate the long-held MRC premise that as the action component is made prominent, the realization of the patient finds itself in a less-constrained situation. Lemmens (2006) specifically points out that a prominent action denotation coincides with de-emphasis – hence, structural triviality – of the object, as in (85).

Taken together, these observations combine with empirical patterns to produce a unified account of intransitive alternates across verb types (psych and non-psych), which is formalized next.

4.1 General situation

Given the general conditions that allow a verb to appear in ‘objectless’ frames, and based on all the discussion, data and theoretical arguments gathered above, the parameters in (86) provide a unified account of why intransitive alternatives are indeed more natural for some verbs than for others, and how this extends to psych verbs.⁴¹

Points like (86)c in particular introduce the central premise behind intransitivity alternation patterns and the MRC (Levin & Rappaport 1995; Levin 2006 i.a.; cf. (11) above). For some psych-rooted verbs of the manner-compatible type, this potential for frame flexibility comes with canonical transitivity (e.g. eventive passivization), eventivity, agency, and defeasibility – thus bringing the MRC and the Agent Control Hypothesis together. Similar remarks on the canonical transitive behavior of psy(chose)-rooted agentive/active predications are found in Arad (1998) and McGinnis (2000) among others.⁴²

An important finding, in this sense, is that contrary to what Arad argues, not all psy(chose) roots – i.e., those naming the emotion induced by a trigger –⁴³ seem to allow for agentive/active predications (her a-b options in (67) above), and not all are ‘normal’ verbs allowing for non-psych, agentive/eventive uses (i.e., allowing alternation between psych and ‘normal’ constructions). Consider, for instance (87)a (Arad 1998: 13) vs. (87)b.

The distribution of intransitive alternates would reflect a constraint on constructional possibilities according to specific semantic features (dis)allowed by the verb, with a distributional result predictable from the same criteria.

In languages like English, manner-compatible verbs with a psych component in the root (Croft’s psy-chose) would allow two different single participant construals: activity-type ‘objectless’, and anticausative⁴⁴/middle frames, as in (88). For verbs strictly denoting a psych state instead – i.e., showing no potential for active/agentive or defeasible readings – the manner-of-behavior intransitive (cpaa) is logically unavailable. Provided AVs are unavailable as well,⁴⁵ options eventually reduce to transitive/unaccusative frames – i.e., the transitivity alternation patterns typically associated with result verbs –, as (89) shows.

4.2 General analysis

Standard constructional approaches like Ramchand (2007; 2013) offer a coherent account for the asymmetries noted.

In this model, activities and accomplishments are associated with structures that do not include, as part of the verb’s core denotation, the lowest result projection. Ramchand’s clarification on non-default addition of the Res layer is crucial for three reasons: first, it agrees with the fact that the event structure is one requiring the realization of one participant: the external one (no result, no experiencer); second, in doing this, it accommodates a fundamental property like defeasibility; third, it preserves the constructional possibility to map a psy(chose) root onto an atelic dynamic event keeping the constant participant as initiator, but of a process (active agent), leaving open the possibility to add an optional (not verb-entailed) result state, provided this is lexically supported by the verb.⁴⁶

If the agentive manner/activity type of predication supported by the tests and observations above is correct, then verbs with a psy-chose root – ‘normal’ verbs not restricted to psych denotation – could appear in a manner-type (single event) representations very much like the one producing objectless eventive/agentive frames in cpaa/objectless occurrences ((47) above, repeated as (90)

In the same theoretical (Ramchandian) decompositional model, activities are standardly defined by (i) an external argument; (ii) an eventive core. The key is the referential identity of the DP. In (89), the argument in the specifier of the defective external-argument-introducing head shares referent with the central participant (actor) of the dynamic core (ProcP), hence retaining both (initiator/actor) roles (Ramchand 2007: 80). Since in this model Init’s argument is also interpretable as a holder, it could be alternatively (constructionally) computed as the holder of a characteristic property, provided this is understood as the behavior described by repeated deontic instantiations of the event. Consequently, the configuration captures the semantic nuance of the allowed (cpaa) alternation.⁴⁷ Now, being a psych-chose root, the verb could also appear in a typical psych predication. In that case, all ‘normal’ properties, along with canonical transitivity, ‘go away’, as Arad (1998) predicted.⁴⁸

On the other hand, there would be a set of roots producing the canonical ‘psych’ denotation. Such verbs lack all the properties just mentioned: they are stative (non-eventive, non-dynamic (Marin 2011; Marín & McNally 2011), fail to display canonical transitivity (Arad 1998 i.a.) and in many cases include an aspectual boundary directly related to the (non-defeasible) psych state inherently described.⁴⁹ Such properties, unparalleled by standard (non-debated causative) result verbs, result in unavailability of monadic structures with the subject as stative cause in English – which is instead possible, though limited, instead for standard result verbs, as shown above (Section 3). Taken as an indicator of structural representation, non-defeasibility of the (psych) state is coherent with both the strict(er) restriction on the realization of the object/experiencer and AV unavailability.

4.3 General summary & Results

If correct, the crosscut in productivity noted would be the aftermath of a specific restriction.

Languages like Romance and Greek would systematically allow monadic structures for both types of roots, including those invariably yielding core psych verbs, provided that free independent composition with the defective external-argument-introducing head is systematically allowed. This combinatorial possibility, also suggested by AV productivity with non-transitive IO experiencer psych verbs, yields a structure similar to (44) above, repeated in (91)b, where the aspectually-neutral head Init introduces the holder of the dispositional stative property, as stative/involuntary cause of the root-coded content (e.g., dehydration/worry). Direct complementation by rhematic material conforms to the essential stativity of the construction.⁵⁰

In languages like English, by contrast, independent composition with this head is severely restricted – presumably, to verbs with canonical transitivity and only to a certain extent.⁵¹ This confines psych roots (when allowed) to internal-argument-licensing frames. With AVs unavailable (English), roots producing ‘normal’ verbs with canonical transitivity may opt for either the dyadic or the ‘objectless’ (Levin’s intransitive) alternatives, following a manner-like (single dynamic, agentive event structure) pattern, conforming to (91)a, hence explaining the objectless construction allowed. For roots only realizable as pure (invariably stative, non-defeasible) psych verbs instead, AV unavailability reduces options to unaccusative/middle predications (cf. (91)c), coherent with unmarked or equipollent derivation, resembling the result pattern.

Ultimately, if manner-like (e.g., cpaa) alternations are basically transitive frames where the object remains semantically and/or syntactically unspecified (but licensed); and anticausative/middle alternations would not actually be argument structure alternations but rather Voice ones (Martin 2019), a final corollary would be that proper argument frame alternations are unavailable for English psych (oe) verbs or either type.

Above we mentioned some well-known arguments raised by Landau (2010). He proposed a classification of natural languages based on criteria like availability of verbal passives. Based on the facts discussed above, the present typology would look pretty much like this:

This distribution would explain crosslanguage productivity in a way amenable to the manner/result distinction in transitivity alternations, preserving a transparent syntax-to-event correlation. Strict complementary distribution is avoided by considering verbs with multiple uses and possible event representations; notably, verbs allowing ‘normal’ and psych predications.⁵³

The remaining questions can be, we believe, safely derived from the distinct properties of psych verbs, including non-canonical transitivity (Arad 1998), default stativity (Marín 2011) and unclear causative status (Neeleman & van de Koot 2012). On the question of why certain roots produce activity-denoting verbs with canonical transitivity and diverse event/argument representations, while others invariably hold a pych denotation, we believe differences like external vs. internal causation (McKoon & MacFarland 2000) are worth exploring.