Dutch (change-of-)location verbs display mixed transitivity properties.1 Consider the verb vallen ‘fall’. By itself, this verb is unaccusative: it disallows personal passivization and selects the auxiliary zijn ‘be’, not hebben ‘have’, in the perfect (Hoekstra & Mulder 1990; Levin & Rappaport Hovav 1995). Yet derivatives such as binnen+vallen ‘invade’, aan+vallen ‘attack’, and over+vallen ‘raid’ license direct objects and form personal passives; however, only overvallen obligatorily selects hebben in the perfect. Other (change-of-)location verbs that take particles and/or full PPs as complements show similar behavior.2

Our goal is to reconcile these contradictory properties by relying on the distinct functions of v and Voice, thereby revisiting core issues of argument structure.

Starting with our basic assumptions, we adopt Distributed Morphology’s principle that lexical items are underspecified roots interpreted by their syntactic context. Accordingly, verbs consist of acategorial roots embedded in syntactic structures which determine their behavior (e.g., Harley 2013; 2017). Various properties emerge from the functional heads dominating the root; we speak of unaccusativity, telicity, or dynamicity as ‘verbal’ characteristics simply for brevity.

We recognize the functional categories v and Voice (Pylkkänen 2008; Harley 2013, i.a.). The Dutch data challenges both a bundled approach to vP/VoiceP and a split approach whereby both vP and VoiceP are always simultaneously present. We build upon the proposal that some structures have both transitive vP and VoiceP, while others only have VoiceP (Burukina & Polinsky 2024; 2025). This approach captures the similarities and differences between ‘fully transitive’ verbs (containing both projections) and ‘partially transitive’ verbs (containing only VoiceP), including patterns of auxiliary selection and the (in)ability to form subject-oriented passive participles.

The paper proceeds as follows. Section 2 presents the transitivity properties of the Dutch verbs in question. Section 3 outlines our proposal: not all (change-of-)location verbs are structured identically, the core difference being the presence or absence of vP. Section 4 demonstrates how this accounts for auxiliary selection and participial formation. Section 5 discusses agentivity, showing it does not obligatorily follow from syntactic structure.

The simplex Dutch motion verb vallen ‘fall’ is unaccusative. It selects zijn ‘be’ in the perfect, and its past participle can be used as a prenominal attributive modifier of its subject (Hoekstra 1984; Broekhuis et al. 2015).

In the presence of certain prepositional elements, such as the particles binnen ‘inside’, aan ‘on’, or over ‘over’, vallen is used transitively. Similar to the causative vellen ‘fell’ (2d, 3b),3 it licenses a direct object (2a-c) and forms personal passives (3a).

As (2a–c) show, derivatives of vallen do not form a homogeneous group regarding auxiliary selection in the perfect. Overvallen patterns with other transitives (including vellen) in combining with hebben.4 Aanvallen strongly favors hebben though zijn is possible for some speakers. Binnenvallen is similar to its base unaccusative vallen in allowing and, for many speakers, preferring zijn.

Consider other examples combining (change-of-)location verbs with a preposition-like particle and/or a full PP:5

Examples (2a) and (4a) feature transitive versions of unaccusative verbs that preserve certain properties of their intransitive counterpart—specifically, their auxiliary selection patterns. By contrast, (2c) and (4c) contain ‘fully transitive’ verbs. The verbs in (2b) and (4b) fall between these extremes, showing more flexible properties.

These patterns are summarized in Table 1.

The Dutch data presents a challenge: how can the grammar transform simplex unaccusatives into transitive, passivizable verbs while preserving some key properties of unaccusativity? We offer an explanation in the next section.

We now consider two properties sensitive to v*’s presence: auxiliary selection and use of the past participle as a prenominal attributive modifier.

The most distinctive property of ‘partially transitive’ verbs, such as binnenvallen ‘invade’, is their selection of zijn ‘be’ in the perfect, which sets them apart from ‘fully transitive’ complex verbs.12 Our analysis captures this contrast effectively.

In Dutch, selection of zijn in the perfect is governed by three factors: (a) dynamicity of the base verb, (b) telicity of the VP, and, as formulated traditionally, (c) absence of an external argument/transitivity (Hoekstra & Mulder 1990; also, Zaenen 1993).

All events denoted by the positional and change-of-location verbs illustrated in section 2 qualify as dynamic. Vallen ‘fall’ is unaccusative and denotes a telic event, allowing it to combine with zijn. In contrast, overvallen ‘raid’ is transitive and atelic, combining only with hebben. Binnenvallen and aanvallen present a more complex picture. Unlike overvallen, the transitive binnenvallen for many speakers requires zijn-selection, suggesting that telicity plays a crucial role, with the subject reaching a terminal point. However, this would incorrectly predict that aanvallen, which denotes an atelic event, should require hebben, just like overvallen. While aanvallen does combine with hebben, some speakers also allow zijn for this verb. This means telicity alone cannot be decisive, as it fails to capture the subtle differences between these verbs. Similarly, analyzing binnenvallen, overvallen, and aanvallen as sharing the same transitive structure does not adequately distinguish them.

Our approach reformulates condition (c) above in terms of the absence/presence of v*. With vallen and binnenvallen, there is no v* (hence zijn-selection). With aanvallen, v* may be present (allowing hebben-selection), but this verb can also appear without v*, alternating between two structural configurations in (11b) and (12b), hence, either zijn- or hebben-selection. The idiomatic expressions in (4) follow the same patterns.

An aspectual auxiliary, introduced above VoiceP, is thus sensitive to two factors: the presence/absence of v* and the predicate’s telicity. Both affect the combination of V with a particular aspectual head as V undergoes head raising to v*/Voice.13

The auxiliary selection patterns correlate with another important property—distribution of participles. Unlike unergatives, Dutch unaccusative verbs allow their past participles to modify their logical subject—a key diagnostic of unaccusativity, illustrated in (13):

Past participles formed from all complex transitive verbs can modify the logical object. However, when it comes to modifying the logical subject, only ‘be’ selectors—such as binnenvallen, aanvallen, and the idiomatic (in de rede) vallen—permit prenominal attributive use:

The details of Dutch past-participial grammar lie beyond this paper’s scope; for an overview, see Sleeman (2011). For our purposes, what matters is that the acceptability of some predicates in (14) and (15) cannot be reconciled with analyzing these verbs as ‘fully transitive’. Past participles of regular transitive verbs cannot function as attributive modifiers of their subjects. Consider causative vellen ‘fell’: while de gevelde bomen ‘the felled trees’ (with the participle modifying the object) is perfect, *de (bomen) gevelde houthakker is ungrammatical when intended to mean ‘the lumberjack that felled trees’ (with the participle modifying the logical subject).

Our approach to argument structure offers a straightforward way to accommodate the contrast in (14)–(15). As discussed in section 3, overvallen, but not binnenvallen or aanvallen, obligatorily contains v*P. It is v*’s presence that leads to the loss of logical-subject modification.14 A uniform approach whereby all transitives contain either both v*P and VoiceP or a bundled vP/VoiceP would struggle to account for the contrast presented here.

To summarize, auxiliary selection and participle distribution can be successfully explained by the presence/absence of v* in the structure.15 In contrast, alternative analyses whereby all transitive verbs have identical thematic domain size and structure cannot accommodate the relevant data in its entirety.

All transitive verbs in section 2 are compatible with agent-oriented modifiers, as illustrated in (16) with expres ‘purposely’:

Agentive modifiers such as expres are permitted even with intransitive verbs exhibiting all the hallmarks of unaccusativity, such as vallen,16 hence such adverbs cannot diagnose the presence of an Agent in syntax:

Similarly, agentive verbs that are normally unergative typically exhibit unaccusative properties when combined with a directional PP marking the endpoint of an event. In (18), the presence of uit het raam ‘out the window’ causes springen ‘jump’ to shift from selecting hebben to selecting zijn, and its participle can now modify the logical subject. But even as an unaccusative, springen is still compatible with agentive modifiers, (18b).

These empirical facts argue against full isomorphism between syntax and semantics. The agentive interpretation associated with expres does not correlate one-to-one with the presence of v* (see also Rooryck & Vanden Wyngaerd 2011 on the lack of correlation between agentive adverbials and syntactic projection of an Agent). When both v* and Voice are present in syntax, an Agent is structurally licensed. However, agentivity can also emerge without v*, as a pragmatic inference (see Williams 2015 for inferential thematic properties); such inference accounts for the acceptability of ‘purposely’/‘deliberately’ in (16a), (17), and (18), which lack v* to assign an Agent role.

In the absence of structural licensing, pragmatically inferred agentivity can easily be suppressed by adding a modifier such as onbewust ‘unknowingly’. Such modifiers are compatible with binnenvallen and aanvallen (19a,b) but marginal at best with overvallen (19c), where the logical subject receives a combined Agent+Theme role, the Agent part structurally licensed by v*.

Likewise, in the absence of v*, pragmatically inferred agentive interpretation can be cancelled:

We have examined Dutch (change-of-)location verbs whose mixed transitive-unaccusative properties challenge existing approaches to argument structure syntax. Complex predicates such as aanvallen ‘attack’, binnenvallen ‘invade’, and overvallen ‘raid’—all built on the simplex unaccusative vallen ‘fall’—exhibit contradictory transitivity properties that neither bundled approaches (conflating vP and VoiceP) nor split approaches (assuming both projections are always present) can fully account for.

We propose that understanding these mixed properties requires recognizing structural variation within the verbal domain: some constructions contain both vP and VoiceP, while others contain only VoiceP. Correspondingly, certain complex verbs behave as ‘fully transitive’, while others show only ‘partial’ transitivity. The structural distinction between VoiceP + v*P and VoiceP alone explains the observed properties, including auxiliary selection patterns and the distribution of subject-oriented participles.

An alternative approach would argue that all complex verbs are structurally uniform, with both vP and VoiceP always present, and that variation arises from different prepositional combinations. However, this approach fails to explain why auxiliary selection differs even with identical P-elements, as shown by (4b) and (4c) above.

Our analysis further captures the auxiliary selection facts and extends to other languages (cf. English overrun, overcome). It supports a flexible view where functional projections’ presence varies systematically, offering new insights into the architecture of the verbal domain.