3.1.1 The pronominalization of the PPC and its grammatical category

A long-debated question refers to the syntactic category of the PPC, whether a DP with some sort of spurious preposition or a real PP. For Catalan, Bel (2008) emphasizes the parallelisms between PPCs and direct objects, on the basis of different kinds of evidence,⁸ and argues that PPCs are argumental NPs with a preposition, governed by the verb (see also Demonte 1991 for Spanish). PP pronominalization, however, reveals that PPCs are substantially different from argumental DPs. Rosselló (2008) uses this evidence to show how PPCs behave in a different way from indirect (and direct) objects (see also Kuhn & Mollica 2013 for Italian).

Consider, first, the class of clitic pronouns used as proforms for indirect objects and PPCs. Indirect objects, when denoting humans, are represented by dative clitics, li in (14).

PPCs headed by any preposition except de ‘of, from’ are represented by the locative clitic hi. Interestingly, even when a human indirect object and a PPC are headed by a ‘to’ (cf. ‎(15)b), they pronominalize differently.

Thus, PPCs pattern with bona fide adjunct PPs. As the adjunct PPs illustrated in (16), the PPCs in (17) appear as the partitive clitic en when headed by the preposition de ‘of, from’, and as hi elsewhere.

That said, our study makes the novel empirical observation that a few IPVs reject the pronominal expression of their PPC: abundar en / de ‘abound in’, constar de ‘consist of’, and, for some speakers, consistir en ‘consist in’.

As also observed by Rosselló (2008), a second difference between DP arguments and PPCs relates to the fact that, while accusative and dative arguments, when pronominal, require clitic doubling (cf. ‎(19)a, b), PPCs bluntly reject it (cf. ‎(19)c).

Note that, in this respect, PPCs behave as VP adjuncts, again (a benefactive in (20)):

Assuming the theory that takes doubling to emerge from a structure where the clitic is a D and it is in some basic relation with the surface DP argument (Torrego 1994; Uriagereka 2005; Gallego 2012), the exclusion of both PPCs and adjuncts from this kind of doubling reveals that they are PPs, not DPs.

From the above considerations, we can conclude that, despite not being VP adjuncts (as shown, again, by the VP-substitution test), PPCs—with the exception of those of constar de ‘consist of’ and abundar en / de ‘abound in’—retain the properties that are common to all PPs, including adjunct PPs.

3.1.2 Argument status of the PPC and its omissibility

As discussed in §1, PPCs do not survive the fer-ho ‘do so’ substitution test, clearly confirming that they are not VP adjuncts. The question still remains whether they are arguments or something else. Traditionally, it has been claimed that PPCs are argumental since they cannot be omitted (e.g., Fabra 1956 or Bel 2008 for Catalan; Neeleman 1997 for Dutch; Cano Aguilar 1999 for Spanish). However, Rosselló (2008) and GIEC (2016), for Catalan, and Martínez García (1986) and Demonte (1991) for Spanish, observe that there are IPVs that can drop their PPC, raising back the question about its argument status. For instance, Demonte (1991) argues how a subclass of IPVs like discrepar de ‘disagree with’ and alardear de ‘boast about’ can drop their PPC both in referential and non-referential contexts, while another subclass containing verbs like consistir en ‘consist in’ and prorrumpir en ‘break out in’, cannot drop their PPC in any context.

When we take a look at the PPC drop facts from the perspective of Catalan, the resulting scenario is far more complex. This is due, again, to the fact that Catalan features PP pronominalization. We focus on two different non-anaphoric constructions featuring the omission of the PPC: (i) the sluicing construction and (ii) the stative habitual construction.⁹

The sluicing construction (Ross 1969; Merchant 2001; 2006), as implemented in (21)-(23), involves a contrastive coordination between a conjunct showing the IPV without its PPC, and the sluiced conjunct, with a wh-PPC. Any chance of interpreting the first conjunct as involving a contextually retrievable PPC is made impossible by the fact that the second conjunct asks, precisely, about the reference of the PPC. This forces a pure activity reading in the first conjunct, which is possible in (21) and impossible in (22) and (23).

Like aŀludir a ‘allude to’ (Class D) pattern constar de ‘consist of’ (Class A) and caure en ‘fall into’ (Class B). The two subclasses Di and Dii pattern identically with respect to this test.

A second test involves the use of the verbs in a stative, habitual construction without the PPC. In this construction, the verb identifies a property of the subject, rather than describing an activity. Interestingly, this test distinguishes Class Di verbs, (24a), from Class Dii verbs, (25a), the former patterning, in this case, with Class C verbs, (24b), and the latter with Class A and Class B (25b) (caure en in the intended IPV interpretation).

A comparison of the two constructions just examined determines three classes of IPVs as far as PPC omissibility is concerned. Verbs of Class A, Class B, and Class Dii disallow PPC drop in both constructions. Class C verbs allow PPC drop in both constructions. Finally, Class Di verbs do not admit the sluicing construction, and accept the stative habitual construction. This fine-grained scenario is summarized in Table 1.

Going back to the argumenthood status of the PPC, in this paper we argue that the PPCs that turn out to be droppable in the habitual construction (Class C and Class Di verbs) are adjuncts. They cannot be VP adjuncts, however, since, as pointed out above, they do not survive the fer-ho ‘do so’ test. We propose, therefore, that they are low adjuncts, as suggested in Gallego (2010; 2012), but of different sorts. As for those PPCs that cannot be dropped in any context (Class A, Class B, Class Dii), they will be argued to be either predicates or what we will call copredicates, on the basis of other syntactic evidence. The result is a three-way syntactic distinction for the PPC: adjunct, predicate, or copredicate. The details of the analysis are in §4.2 and §4.3. One of the major contributions of this paper is thus to show that classifications like Demonte’s (1991), which involves a two-way predicate-argument distinction, developed as a predicate-adjunct distinction by Gallego (2010), are insufficient to account for all cases.

We end this subsection by observing an interesting correlation between the morphological format of the IPV and its (in)ability to drop the PPC in the sluicing construction: prefixed verbs very generally do not accept PPC drop in this construction (cf. a-ŀludir a ‘allude to’, co-ŀlaborar amb ‘collaborate with’, etc.), whereas unprefixed verbs may or may not allow it (parlar de ‘talk about’, optar per ‘opt for’).

In §4.1.2 we offer an explanation of this correlation and also of the different behaviour of verbs like coŀlaborar amb ‘collaborate with’ (Class Di) and verbs like aŀludir a ‘allude to’ (Class Dii) as regards PPC droppability in the habitual constructions.

3.2.1 “V_3sg que V_3sg”

Rosselló (2008: 1920–1921) observes that unergative verbs in Catalan can be used in a reduplicative construction of the format “V_3sg que V_3sg”, with the interpretation of an unceasing eventuality, (27). Unaccusative verbs reject this construction, ‎(28).¹⁰

The application of this test to our corpus clearly distinguishes two classes of IPVs.

The construction is licensed by Class C verbs (29a) and also Class Di verbs like coŀlaborar amb ‘collaborate with’ (30). This is expected, since these verbs allow their PPC to be dropped only in habitual contexts (see Table 1).

By contrast, Class B and Class Dii verbs behave like Class A verbs (29b).

3.2.2. Arbitrary pro with an existential interpretation

On the basis of Italian, Belletti & Rizzi (1988) point out that unaccusative verbs, when set in the 3^rd person plural and used with a null subject, do not license an arbitrary (and human) interpretation of such pro subject.¹¹ Similar observations are found in Jaeggli (1986) for Spanish, and Bartra Kaufmann (2008) and Rosselló (2008) for Catalan. Rosselló (2008: 1891) provides the following contrast between unergative dormir ‘sleep’ and unaccusative entrar ‘go in, enter’.

However, neither Bartra Kaufmann nor Rosselló observe that a generic, rather than existential, interpretation of the null indefinite subject makes it acceptable with unaccusative verbs, as pointed out by Borer (2005: 118–119) for Hebrew, Italian, and Spanish. Catalan, we note, behaves in the same way as these languages.

Bearing these qualifications in mind, we apply it to our list of Catalan IPVs. We illustrate the divide with parlar de ‘talk about’ (Class C) and constar de ‘consist of’ (Class A).¹²

Class C (33a) patterns with Class Di and Class Dii, whereas Class A (33b) patterns with Class B.

4.1.1 Argument and event structure

We follow a theoretical tradition stemming from Hale & Keyser’s (1993; 1998; 2002) proposals on argument structure and including Mateu (2002) and Acedo-Matellán (2016). Building, specifically, on Acedo-Matellán (2016), argument-event structure is directly represented in the syntax. Roots, the units encapsulating conceptual content and a phonological representation, are inserted in the structures, but they do not determine them (Borer 2005). This allows a natural treatment of the verbal elasticity exhibited by verbs like patir per/de ‘worry about/ suffer from’, as indicated above. Argument structure configurations are thus articulated on functional heads denoting the grammatically relevant aspects of the semantics of linguistic expressions.

First, the head v expresses eventualities. In this study we assume v to come in two varieties (as in Ramchand’s 2008 two-fold ontology; see also Arad 1999): v_e, expressing events, and v_s, expressing states. The two heads correspond, respectively, to the two sorts that Maienborn (2007) proposes for events and Davidsonian states, on the one hand, and Kimian states, on the other hand. In this sense, we agree with Kratzer (1996) that SL predicates involve an eventive head and IL predicates do not, but we add, based on Maienborn’s (2007) treatment of Kimian states like resemble, that IL predicates must have some eventuality head that can be located in time and referred to with anaphora.¹⁴

Second, the head Voice (Kratzer 1996) introduces the external argument of an eventuality, as is standardly assumed.

Finally, in the lower part of the configuration, we can find Place, a head articulating the predication of a property with respect to an entity. A further projection, Path, can merge on top of a PlaceP. In combination with Place, Path expresses a transition. Place and Path, encode the notions of Central and Terminal Coincidence relations, respectively (Hale 1986; Hale & Keyser 2002; Mateu 2002). We note that the relationship established by Place can be satisfied immediately, by Spec-PlaceP, or via ‘‘delayed gratification’’ (Hale & Keyser 2002: 227), that is, by a later merged specifier (for instance, Spec-VoiceP). For a more recent minimalist implementation of this idea, see Wood (2015: 176).

The eventive head v_e is able to take different configurations as complements: PathP, PlaceP, a nominal expression, or a mere root. Each selection impinges on the final interpretation of the eventive head. Thus, v_e taking a PlaceP as complement denotes an event where a predicative relation is maintained for an interval of time (Jaque 2014). In turn, when taking PathP, the event encoded by v_e is a change of location or state, whereas when taking a nominal expression or a bare root, it is an unergative or a transitive creation/consumption event (Dowty 1991; Ramchand 2008). In turn, v_s combines with PlaceP to denote states, VoiceP merged or not on top of v_sP.

We briefly illustrate this configurational model with a few intransitive examples (adapted from Acedo-Matellán 2016: 33). In the unergative structure in (35), the head Voice introduces the external argument and the root is selected as the argument of an eventive verbal head v_e.

The stative unaccusative structure in (36) involves the projection of a stative verbal head v_s that takes a predicative relation PlaceP in the complement position.

Finally, the structure in (37) projects an eventive v_e that combines with a Path-Place structure that represents the eventive change of state. In this case, the figure argument, la nena, undergoes a movement from Spec-PlaceP to Spec-PathP. As in Acedo-Matellán (2016), la nena is understood both as a figure with respect to the beach and also as a measurer of the transition introduced by Path (Verkuyl’s Generalization; see Verkuyl 1972; Tenny 1987; Borer 2005).

Turning to roots, we follow Marantz (1997), Borer (2005; 2013), and Embick & Noyer (2007), among others, that, even if they do not come with categorial features, they acquire them in syntax, by merging with functional heads as either complements or modifiers. With Acedo-Matellán (2010; 2014) and Acedo-Matellán & Mateu (2014), we assume that roots receive thematic interpretations by virtue of their syntactic position. For instance, a root like FEIN set-merged as the complement to the eventive head v_e becomes non-distinct from a nominal category, which we represent below with Borer’s (2013) “C=N” formalism. Thematically, it is interpreted as an incremental theme, in parallel to the interpretation of a fully fledged nominal (feina ‘work’) in the same position.

Being properly categorized by virtue of their embedment under functional structure, roots may be able to project. In particular, they may take phrasal adjuncts, and we will see an example of this in §4.3.2.¹⁵ Note that their ability to project does not entail that roots can take complements. We think that they cannot, due to the independent fundamental assumption that they lack, of themselves, any proper semantic denotation, namely, that of an unsaturated proposition enabling them to take arguments (see Beavers & Koontz-Garboden 2020 and Ausensi 2021, among others, for a different view).

4.1.2 The morphosyntax of prefixation in Intransitive Prepositional Verbs

It has not gone unnoticed that many prepositional verbs in Romance display a (Latinate) prefix, such as a-ŀludir a ‘allude to’, which contrasts with e-ludir ‘avoid’. Gallego (2012), for instance, observes this for Spanish and argues, following Marantz’s (2001; 2005) syntactic approach to morphology, that the PPC of prefixed prepositional verbs is a double of the prefix, in the same way that DPs are doubles of clitic pronouns in clitic-doubling languages such as Rio de la Plata Spanish. In Gallego’s theory (which follows Torrego’s 1994, and Uriagereka’s 2005 approaches to the doubling of pronominal clitics), doubles originate as adjuncts of the head they “corefer with”. Thus, DPs in clitic-doubling structures would be adjuncts to the clitic, and PPCs would be adjuncts to the verbal prefix, the clitic or prefix eventually incorporating into an upper head. The problem with Gallego’s analysis is that PPCs of prefixed prepositional verbs do not always behave as doubles. Typically, doubles are optional. Thus, DPs doubling a pronominal clitic can be freely omitted.

However, as we showed in §3.1.2, not all PPCs are optional, and, crucially, those of the Dii Class of prefixed prepositional verbs never are.

By contrast, verbs that can drop their PPC unproblematically are typically unprefixed (Class C).

Finally, prefixed verbs of Class Di, such as coŀlaborar amb ‘collaborate with’, allow PPC drop in contexts like the habitual construction or the V que V construction.

We explain the causal relation between the presence of a prefix and the obligatoriness of the PPC in verbs of Class Dii by hypothesizing that the prefix and the preposition heading the PPC originate as one and the same item in the underlying syntactic structure: the Place head. In verbs like aŀludir a ‘allude to’, Place heads a constituent that is a complement of the eventuality head, and head movement may bring a copy of this head to the upper verbal heads (v and Voice). We endorse the Distributed Morphology hypothesis of Late Insertion (Halle & Marantz 1993; 1994), by which syntactic structures are endowed with phonological content via Vocabulary Insertion at PF after Spell-Out. Thus, at Spell-Out, the higher copy of the Place head of aŀludir is spelled out as the prefix a-, and the lower one as the preposition a (see further discussion in §4.2, §4.3.1, and §4.3.3). The prefix and the preposition turn out to be, thus, contextual co-allomorphs, the particular phonology determined, at Vocabulary Insertion, by the structural environment (Embick 2010).

As will be shown in §4.3.2, for Class C verbs that admit PPC drop in all non-anaphoric contexts, like somiar amb ‘dream of’, we propose that the PPC is an adjunct to a nominalized verbal root. Under the common assumption that head movement cannot operate from adjunct position, we expect the preposition not to be able to head-move to the verbal head in these cases, and hence, we expect these verbs not to be prefixed—at least not with a prefix identifiable as cognate with the preposition heading the PPC, as in a-ŀludir a ‘allude to’.

Finally, as regards Class Di verbs, we assume the same structure as the one proposed for verbs of Class Dii, that is one involving an adpositional projection, whose head is the source of the prefix. The difference between the two subclasses lies, we claim, in the status of the PPC. While for Class Dii verbs the PPC itself corresponds to that adpositional configuration, for Class Di verbs the PPC is just an adjunct of the same vP-internal configuration. Thus, we claim that Gallego’s (2012) adjunct approach to the PPC of prefixed verbs can only be applied to Class Di verbs.

In the remainder of this section, we deal with each class of verbs in our typology of IPVs.

4.3.1 Unaccusative Stage-Level IPVs (Class B)

The SL IPVs of our corpus testing out as unaccusative, i.e., with an internal argument subject, are verbs like caure en ‘fall into’ and conviure amb ‘live together with’. Importantly, all these verbs strongly disallow PPC drop in any context.

The disallowance of PPC drop is expected under a view of unaccusativity whereby the subject is an internal argument originating as a specifier of some vP-internal projection and the PPC is precisely what articulates the rest of the projection of which the subject is the specifier. We find two subtypes of unaccusative SL IPVs, according to their eventive-aktionsart properties: those that involve telicity, understood as a transition, like caure en ‘fall into’, and those that are atelic, like conviure amb ‘live together with’. These two verbs, for instance, show a different interpretation with a durative adverbial. Only the latter allows a non-repetitive reading.

IPVs like caure en involve a PathP encoding the transition traversed by the internal argument, i.e., the surface subject. Those like conviure amb ‘live together with’ involve a PlaceP, i.e., they establish some kind of predicative relation between two entities that is maintained during the interval in which the event introduced by v_e holds.

An observation is in order with respect to these representations. We assume the verbal root to originate in different loci. In Path unaccusatives, we take it to be an adjunct to Path. It is well known that transition predicates in languages like Catalan, which belongs to the verb-framed class, cannot express the manner in which the event takes place (Talmy 2000; Mateu & Rigau 2002; Acedo-Matellán & Mateu 2015). In our terms, the root could not be adjoined to the eventuality head, precisely because this clashes with the obligatory incorporation of the Path head into v in these languages. Thus, we take the root to originate as a modifier of Path, describing the type of transition involved (in this case, a metaphorical downward trajectory: caure ‘fall’).²³

As far as morphology is concerned, verbs of Class B are expected to allow the prefixation of the Place head, since the prepositional structure corresponding to their PPC is merged as a complement to the eventuality head. Thus, see con-viure ‘live with’, in (55b), where Place is realized as con- in the final position, as adjoined to v_e, and as the preposition amb ‘with’ in the tail position.

4.3.2 Non-unaccusative Stage-Level IPVs allowing PPC drop in non-anaphoric contexts (Class C)

The SL IPVs of our corpus testing out as non-unaccusative verbs that allow PPC drop in non-anaphoric contexts are verbs like parlar de ‘talk about’.

In §3.1.1, we have shown that PPCs of verbs like parlar de ‘talk about’ in Catalan pattern more like adjuncts than arguments. As a matter of fact, it proves quite difficult to determine the adjunct/argument-hood of the PPC on tests other than sheer droppability. It is well known (Den Dikken & Lahne 2013) that adjuncts are opaque to PP-extraction. This should produce a difference between those PPCs that can be dropped, if they are adjuncts, from those that cannot be dropped. However, PP-extraction from all sorts of PPCs, droppable ‎(56a) or not ‎(56b), is deviant.

Complementarily, for those verbs that admit a non-finite clausal PPC, DP-extraction seems quite fine. This is independent, again, of whether the verb is of the PPC-drop kind ‎(57a)‎ or not (57b).²⁴

Extraction facts are, thus, inconclusive as to whether droppable PPCs are adjuncts.

Demonte (1991), for Spanish, proposes that for IPVs belonging to her Class 2, like discrepar de ‘disagree with’, the PPC is a spurious PP and corresponds in fact to a direct argument, the preposition being a mark of structural case. The fact that the direct object of many transitive verbs can be dropped ‎(58) would seem to support Demonte’s hypothesis that her Class 2 verbs are underlyingly transitive.

However, the pronominalization facts examined in §3.1.1 cast doubt on the idea that the PPC is a direct object. Gallego (2010), who tries to accommodate Demonte’s (1991) two-fold classification of prepositional verbs to Hale & Keyser’s (2002) theory of lexical syntax, claims that verbs that drop their PPC are unergatives, the PPC being a low adjunct. We concur with this author, who, however, does not offer the syntactic representation of the PPC as an adjunct.

Recall that, much as it may be droppable, the PPC does not survive a context where the VP has been replaced by a proform (see example (2) in §1). Under our theoretical assumptions, unergative predicates involve the projection of v_e and of Voice, v_e taking the verbal root as argument, and the predicate having a creation/consumption semantics, (59).

The projection of v_e ensures that an event is interpreted to take place. In turn, this event consists in “making talk”, as related to the root merged at Compl-v_eP. By virtue of this merger, the root is interpreted as non-distinct from a nominal category, and as an incremental theme. The question is where the PPC attaches, when parlar is used with a PPC. We know that, being a PPC, and not a VP-adjunct, it cannot attach outside v_eP. The only site left is the very root of the verb. In fact, we take the PPC to be an adjunct to the nominalized root, further specifying its content (in this case, the subject of the talking).

The root PARL merges with the PP de la Maria, PARL projecting. When PARLP is merged with the eventive head, the root (and its projection) is automatically categorized as a nominal, with the interpretation of an incremental theme. The PP is interpreted as a modifier of such nominal: “Marcel made talk about Maria”. The root ends up incorporating into v_e, and the whole to Voice.

Since the PPC in this class of verbs is merged as an adjunct, it cannot constitute a source of the prefixation of the adpositional head, as in the other classes of verbs (see §4.2, §4.3.1, §4.3.3; see also §4.1.2). Thus, this class of verbs, allowing PPC drop in all non-anaphoric contexts, are predicted not to be prefixed. This prediction is born out, i.e. Class C contains mostly unprefixed verbs.²⁵

4.3.3 Non-unaccusative IPVs disallowing PPC drop in activity contexts (Class D)

This class of verbs, to our knowledge, has not been previously identified as such (cf., for instance, Demonte’s 1991 and Gallego’s 2010 studies on Spanish). It contains verbs that take an external argument and embed an adpositional structure that is the source of the prefix, but with no direct object, like aŀludir a ‘allude to’ and coŀlaborar amb ‘collaborate with’.

IPVs of Class D in Catalan, we argue, are non-native verbs that are learned as associated with certain non-Romance traits. Starting with verbs of the Dii class like aŀludir a ‘allude to’, which do not allow PPC drop in any context, we contend that the PPC itself corresponds to an adpositional configuration that articulates the VP. Thus, we take these verbs to involve a Voice head, which accounts for the presence of an external argument, a v_e, and thus their eventive nature, and also a prepositional configuration, either PathP or PlaceP, where PlaceP does not contain a specifier. Since PlaceP in this case does not project a specifier, it does not function as a full predicate, but as a mere copredicate, forming a complex predicate with the eventive head.²⁶ Its requirement of establishing a relation with an entity is met by “delayed gratification” (Hale & Keyser 2002: 227), when the specifier of Voice is merged. The external argument of the event, the agent, is therefore also interpreted as a metaphorical figure (Wood 2015: 176). Importantly, we take the root to be an adjunct to the eventive head, unlike what is generally the case in verb-framed languages like Romance.

These verbs do not feature any nominal at Spec-PlaceP that could function as a measurer of the event via movement to Spec-PathP. In spite of that, our theory predicts that, if they feature a Path head, a transition is interpreted to take place. This, we note, is shown by the fact that verbs like aŀludir a ‘allude to’ license a repetitive reading in the presence of a durative adverbial.

By contrast, one must conclude that other verbs of this class do not project PathP. For instance, disposar de ‘use’ is not forced to be interpreted repetitively in combination with a durative adverbial. Rather, it allows a single event reading.

For verbs like disposar de ‘use’, therefore, we propose the same configuration as for aŀludir a minus the projection of PathP.

Verbs of Class Dii are predicted to allow prefixation. This is because they involve a non-adjunct vP-internal adpositional configuration whose head may successively raise to the eventive head, where it is interpreted as a prefix. The prefix and the preposition heading the PPC are different realizations—contextual co-allomorphs—of the same Place head: a- and a in the case of aŀludir a ‘allude to’, and dis- and de in the case of disposar de ‘use’.

Turning to verbs of Class Di, they do allow PPC drop in certain non-anaphoric contexts like the habitual construction. We take this to indicate that, as with verbs of Class C, the PPC is an adjunct. If that is the right analysis, two questions emerge: what portion of the structure is the PPC an adjunct to? And what then is the source of the prefix of, for instance, co-ŀlaborar? Here we would like to adopt Gallego’s (2012) general approach to prefixed prepositional verbs in Spanish: the PPC is an adjunct to the prefix itself, and functions as a kind of double, specifying the general meaning of the prefix (see §4.1.2). In our terms, we take the PPC of Class Di verbs to be an adjunct to the VP-internal adpositional projection. In turn, this VP-internal projection contains a root that corresponds to the prefix, which ends up adjoined to the verb, (65).

As mentioned, the adjunct amb diversos lingüistes ‘with different linguists’ further specifies the impoverished interpretation of the PlaceP whose complement is the very root of the prefix: CO ‘with’. Unlike with verbs of Class Dii, the PPC, as an adjunct, can be dropped, as in the habitual construction. This brings Class Di closer to Class C. The difference between these two classes lies in their underlying configuration. Class C verbs involve, by themselves, an unergative structure licensing an activity interpretation, and the PPC, as an adjunct to the root, just further specifies its content (see §4.3.2). By contrast, the absence of the PPC in Class Di verbs does not leave such a configuration, able to license an activity interpretation. This is the reason for the difference in grammaticality status of verbs of the two classes in the sluicing construction, where an activity interpretation is induced.

By contrast, the habitual construction, being generic, is compatible with the lack of referential specification, and hence, with the drop of the PPC adjunct.