2.1 Syntactic and Semantic Criteria

There is general consensus that the main characteristic of verbal periphrases is that [V2] lexically determines most of the argument structure and selectional restrictions of the whole construction, whereas [V1]—as a hallmark of the periphrastic pattern—contributes the sort of abstract meaning usually associated with functional categories in the realms of tense, aspect, and modality. (3a–d) shows that the lack of thematic roles and the expletive subject lexically determined by Fr. pleuvoir ‘to rain’ extends to all periphrastic patterns in which it appears as [V2]:

(3)

This characteristic is more often than not interpreted as a symptom for the auxiliary status of [V1]. Now, in formal approaches, auxiliaries are notoriously difficult to define as a cross-linguistic category (cf. Bonami, 2015; Remberger, 2006), and even in a single language, the adequate analysis of the syntactic structures containing them is subject to debate (Abeillé & Godard, 2002). Rosen (1997, p. 112) proposes to characterize auxiliaries as a lexically designated class of verbs that inherit their subject. This provides a necessary condition for auxiliary status. However, it may cover items that widely differ from each other in some relevant syntactic properties (as, for instance, seem-verbs vs. modal auxiliaries in English). Moreover, some [V1]s in periphrases are, arguably, control verbs insofar as they select their own subject. In most functionalist approaches, auxiliaries are loosely characterized as a gradient category, which emerges diachronically when some lexical verbs gradually lose their full verbal properties and autonomous word status in specific constructions (Heine, 1993, apud Squartini, 1998, pp. 19–20).

In a tradition going back at least to Fontanella de Weinberg (1970) and Dietrich (1973), a full set of diagnostics is applied to distinguish verbal periphrases from free syntactic combinations of two verb phrases. They involve mainly different substitution possibilities for the constituent headed by [V2], including the possibility of null-complement anaphora (Olbertz, 1998), as well as different symptoms of monoclausality, called transparency or restructuring effects in the generative tradition (Cinque, 2001a; Rizzi, 1978; Wurmbrand, 2001). Sections 2.2 and 2.3 briefly address these issues; section 2.4 is then devoted to the properties and types of [V1].

2.2 The Syntactic Status of the Constituent Headed by [V2]

In free syntactic combinations, infinitival complement clauses and gerundial or participial adjunct clauses allow for different substitutions not available for periphrastic combinations. Free syntactic combinations ((4a)) and periphrases ((4b)) contrast with respect to substitution by finite complement or adjunct clauses ((5a) vs. (5b)), by wh-words ((6a) vs. (6b)), or by anaphoric items ((7) vs. (8)):

(4)

(5)

(6)

(7)

(8)

Substitution of the constituent headed by [V2] by neutral strong pronouns or demonstratives (those normally representing propositional contents) is in principle excluded ((9b)), but clitic substitution shows some degree of variability across patterns and across languages ((10a,b)):

(9)

(10)

In fact, the anaphorization possibilities for the constituent headed by [V2], although variable, differ both from those for nonfinite clauses and from those of the complement of bona fide auxiliaries, namely, the ones entering into the composition of compound tenses (the descendants of Lat. habere and esse, as well as Cat. anar ‘to go’ in anar + Vinf, known as the ‘periphrastic past’, and Rom. voi ‘will’ in voi + Vinf, the ‘periphrastic future’) (see ORE of Linguistics article “Morphological and Syntactic Variation and Change in Romanian,” 2021). Unlike the latter ((11a), (12a)), a number of verbal periphrases allow for null-complement anaphora (Zagona, 1988), in which the contextually given [V2] constituent is simply not pronounced ((11b), (12b)):

(11)

(12)

Olbertz (1998, pp. 94–103) and Bosque (2000) argue that the possibility of null-complement anaphora correlates with the status of [V1] as a control verb contributing an external argument of its own (see section 2.4). Notice that according to this analysis, some periphrases have to be assigned two different syntactic structures, depending mostly on the animate or agentive status of their subject.

It is sometimes assumed that restricted substitution possibilities indicate that the constituent headed by [V2] is not a complement of [V1], that is, that the constituent structure is not [[V1] + [VP2]] (substitution tests being the elementary tests for constituency). This assumption is questionable: What these tests show is rather that the constituent headed by [V2] does not have the (semantic) status of a clause and therefore cannot be replaced by neuter pronouns. Arguably, the tests also show that when the periphrastic pattern is altered by substituting a pronoun or an adverb for its fixed nonfinite form, [V1] is prevented from making its particular semantic contribution to the periphrasis.

2.3 Restructuring, Argument Sharing, and Monoclausality

A wide consensus has been reached on the monoclausal status of verbal periphrases (Abeillé & Godard, 2002, 2009; Cinque, 2001a, 2001b). This consensus is partly based on the existence of restructuring effects, in which an argument of [V2] is treated syntactically as if it were an argument of [V1]. Extensively treated for the first time in Aissen and Perlmutter (1976/1983) for Italian and Spanish, restructuring effects comprise most famously clitic climbing and passivization.

In clitic climbing, an argument is cliticized on [V1], although it is subcategorized by [V2], as in (13a,b):

(13)

Clitic climbing is obligatory with bona fide auxiliaries, as illustrated in (14a,b), and it is possible for (most) verbal periphrases across the Romance languages.

(14)

Although Contemporary French does not allow for clitic climbing in verbal periphrases, this possibility still existed in Classical French (cf. (13a)).

In the reflexive-passive construction, [V1] agrees in person and number with the DP corresponding to the patient argument of [V2] and thus marks it as its subject:

(15)

Restructuring phenomena are for the most part optional.² Clitic climbing is known to be excluded when the sequence [V1 + V2] is disrupted by some intervening element, as for instance the parenthetical clause in (16b):

(16)

The label restructuring itself presupposes a double possibility of analysis for these constructions, the first one involving the [V2] phrase as a co-constituent of [V1], the second, restructured one, involving a verbal complex [V1+V2] (Hernanz & Rigau, 1984; Rizzi, 1978; Zubizarreta, 1982, p. 161). However, more careful attention to the overall behavior of periphrases reveals that their constituent structure is subject to a great deal of variation (Abeillé & Godard, 2009), thus depriving the verbal complex solution from much of its initial attractiveness. Verbal periphrases are defined by the fact that the two verbal forms involved share their arguments, but the consequences of shared argument structure for constituent structure are far from clear.

A major impulse for the monoclausal analysis of verbal periphrases came with the proposal by Cinque (2001a, 2001b), according to which [V1]s—or rather, the periphrastic patterns they head—are to be treated as exponents of a putatively universal array of functional categories. Cinque’s approach—independent of its exact implementation in terms of functional categories—has an important immediate consequence: The constituent headed by [V2] in the periphrastic pattern is a VP (which in the VP-internal subject hypothesis includes the subject of [V2]), not a full-fledged clause. This, in turn, converges with the idea that restructuring phenomena arise when the [V2] constituent is defective, not being overlaid by a TP of its own (Wurmbrand, 2001). For a later cross-linguistic treatment of restructuring that builds on Cinque’s theory, see Grano (2015).

Despite the relevance of restructuring for the syntactic understanding of verbal periphrases, it should be kept in mind that restructuring constructions and verbal periphrases are roughly overlapping but not identical categories. First, restructuring may occur in constructions that do not exhibit the peculiar substitution patterns of periphrases. Thus, clitic climbing is quite usual with some directed motion verbs in combination with infinitival purpose clauses ((17a)), as well as with verbs of the try/manage type ((17b)). But these verbs allow for substitution of the [V2] constituent by neutral pronouns ((18a,b)) and by a finite clause ((19a,b)):

(17)

(18)

(19)

Second, some particular periphrastic patterns exclude restructuring by virtue of their syntactic properties. This is the case of impersonal patterns (Kayne, 1989)—in which the inflection of [V1] is fixed as 3sg (20a,b)—and of patterns in which [V1] is inherently reflexive, as in It. mettersi + a +Vinf (21a,b):³

(20)

(21)

de Andrade and Fischer (2017) discuss an apparent diachronic reduction of the frequency and scope of restructuring phenomena in Ibero-Romance, most notably a decrease in the frequency of clitic climbing in the modern as opposed to the medieval varieties. They analyze this diachronic change as a reflection of separate and independent changes in word order between older and modern varieties, which do not affect periphrastic status as such. According to Loporcaro (2008, pp. 134–135), absence of clitic climbing in modal periphrases is an innovative feature, restructuring being obligatory, for instance, in the more conservative Sardinian varieties. For a quantitative diachronic study of clitic climbing in Spanish, see Davies (1997).

2.4 The Status of [V1]

As stated at the beginning of section 2, verbal periphrases are characterized by the fact that the best part of the thematic and argument structure and the selectional restrictions on arguments are determined by [V2], which thus qualifies as the main predicate of the construction. [V1]s are sometimes loosely classified as auxiliaries or semiauxiliaries, the latter label being motivated by the clear distributional differences they exhibit with regard to bona fide auxiliaries in the same language. Occasionally, [V1]s are referred to as modifier verbs. In a strictly semantic interpretation, a modifier produces an object of the same semantic type as the object it applies to; such an interpretation has been argued to be adequate for a large subset of verbal periphrases (Laca, 2004a).

The idea of a desemantized [V1], which in the periphrastic construction fails to exhibit the lexical meaning it has elsewhere, pervades research on periphrases since its very inception (Squartini, 1998, pp. 21–23), and traditional studies relied on this as the single criterion for periphrasticity (cf. Gougenheim, 1929, among many others). Most prominent among [V1]s are items that—when acting as full verbs—are verbs of posture and motion, such as Pt. estar, from Lat. stare ‘to stand’ (22a), or andar ‘to walk about’ (23a). Many of these verbs also exhibit uses as a copula ((22b)) or a light verb ((23b)), which reveals interesting parallels between auxiliarization and copularization processes (Laca, 2000):

(22)

(23)

But not all periphrastic patterns involve a [V1] that has, synchronically, a full verb as a counterpart. Phasal verbs, such as begin, cease, stop, and the like (superlexical verbs for Smith, 1991, or verba adjecta for Dietrich, 1973), do not denote a type of event but a part of the temporal structure of any event; they are not predicates of events and lack an argument structure of their own. Moreover, fully auxiliarized or copularized verbs (see ORE of Linguistics article “Copulas in the Romance Languages”) such as Sp. haber ‘have’ or estar ‘be’ lack occurrences as full verbs, and the same holds for the descendants of Lat. *potere ‘can’. This notwithstanding, patterns with such verbs as [V1] are also considered periphrastic.

The status of [V1] with regard to the control-versus-raising distinction has received some attention in the literature (Lamiroy, 1987, for French; Olbertz, 1998, pp. 94–103, and Bosque, 2000, for Spanish; Cinque, 2001a, for Romance). There is a growing tendency to consider all periphrases as raising constructions. Thus, Cinque (2001a) observes that partial control (Wurmbrand, 2001) is impossible in periphrases, and he concludes that strict control is in this case nothing but a symptom of the raising status of such constructions. However, it is not clear that the analysis of [V1] as the realization of a functional head necessarily presupposes that it is a raising verb. In some influential approaches, little-v projections are assumed to introduce the external argument—or an agentive thematic role. Now, little-v arguably belongs to the functional domain. It would be thus perfectly possible for some [V1]s to express a functional head while introducing their own subject, which controls the subject of [V2].

In fact, the control-versus-raising distinction seems to be actually orthogonal to the question of periphrastic status. The behavior of verbal periphrases with regard to the substitution patterns for the constituent headed by [V2] and the restructuring phenomena discussed in sections 2.2 and 2.3 differs from both that of free control and that of free-raising constructions. In any case, a subset of the periphrastic patterns exhibits clear restrictions of its own with regard to the thematic role and/or animacy of its subject (cf. (24a,b)), which may be an indication that in these cases, [V1] selects its own subject, which acts as a controller for the subject of [V2]:

(24)

In sum, verbal periphrases are a descriptive category whose core is constituted by patterns that involve a [V1] making little or no contribution to argument structure and forming a single clausal domain with a nonfinite constituent headed by [V2]. These patterns are the result of diachronic processes of grammaticalization, in which some originally lexical, biclausal combinations are recruited for the expression of functional meanings. Borderline cases comprise at the one extreme compound tenses, which are paradigmatically integrated along with synthetic forms into the verbal inflection system and in which [V1] exhibits all the properties of an auxiliary. At the other extreme, there are constructions with superlexical [V1]s that do not denote types of events but dispositions or intentions toward any event (verbs of the try/manage type). In contrast, with constructions involving superlexical [V1]s denoting parts of temporal structures (phasal or aspectual verbs), constructions with verbs of the try/manage type, although subject to restructuring phenomena, also exhibit the typical substitution patterns for nonfinite clauses in free syntactic combinations.

3.1 Problems for the Classification of Nonpassive Verbal Periphrases

In most descriptive grammars, verbal periphrases are traditionally grouped by the nonfinite form of [V2] into infinitival, gerundial, or participial periphrases. This principle of grouping is not very helpful in comparative Romance studies, since it blurs rather obvious semantic near-equivalences among different Romance languages and even among different regional varieties of one and the same language, such as those exemplified in (25a,b) and (26a,b):

(25)

(26)

A classification by semantic categories is therefore more adequate (see Coseriu, 1976, for a first proposal in this sense). In fact, most researchers agree that verbal periphrases should be classified along the lines of the TAM categories they express, and the rough outlines of a subdivision into temporal, aspectual, and modal periphrases are consensual. However, this grouping poses two problems of different scope. First of all, certain periphrases—first identified by Dietrich (1973)—have rather discourse-level functions, ordering eventualities in a narrative sequence as in (27a) or ordering them in scalar domains based on expectations as in (27b), without affecting their temporal structure in the least (cf. García Fernández & Carrasco Gutiérrez, 2008):

(27)

Second, as is to be expected from expressions involved in grammaticalization processes, one and the same periphrastic pattern at the same synchronic stage may be the exponent of more than one category (cf. the notion of syncretism among periphrases in Coseriu, 1976). Thus, for instance, Sp. ir + a + Vinf has temporal uses which can be assimilated to an expression of future tense ((28a)), others in which it can only be aspectual ((28b)), and others still in which it conveys a modal or scalar meaning ((28c)) (cf. Bravo, 2008; Bravo & Laca, 2011):

(28)

In the same vein, Fr. devoir + Vinf is first and foremost a modal periphrasis, but in some uses in the imparfait, it expresses future-in-the-past, as an alternative to the conditional (Il a dit qu’il devait sortir/qu’il sortirait ‘He said that he would go out’), and it is preferred over the conditional for the expression of scheduled or planned events (Vetters & Barbet, 2006).

Whereas the divide between the modal and the temporal-aspectual realm is comparatively easy to draw—with the modal realm defined by the notions of possibility and necessity—a strict distinction between aspectual and temporal periphrases depends on the specific assumptions made about the notion of aspect. Given the notorious difficulties surrounding the definition of aspect, and the fact that aspect is a relative newcomer in the description of Romance languages and formal linguistics, there are no generally accepted criteria for this distinction. Thus, for instance, Sp. ir + a+ Vinf was assigned an aspectual and not a temporal meaning in example (28b) on the grounds that tenses setting Reference Time after Speech Time are in general impossible in conditional antecedents. The same pattern was classified as temporal and not aspectual in example (28a) on the grounds that haber + Vptcp (perfect aspect or relative anterior tense) cannot normally appear below an expression of aspect. But such facts are subtle and open to discussion.

3.2 Aspectual Periphrases, Syntactic Layers, and Semantic Types

As a way out of this conundrum, certain studies have explored distributional facts that may reflect semantic distinctions. Periphrastic patterns may combine with each other, as in (29a), but not all logically possible combinations are grammatical ((29b)) (cf. Cinque, 2001b):

(29)

These distributional facts have been taken to indicate that verbal periphrases belong to different structural layers. Olbertz (1998) was the first to exploit this idea. She assigns periphrases to the different clausal layers postulated in Functional Grammar, thus distinguishing among expressions of derived predicates, inner and outer aspect, participant- and event-oriented modality (predicate or predication operators), polarity, truth-commitment (proposition operators), and so forth. In the generative tradition, Cinque (2001a, 2001b) argues for total order among the distinct functional heads, which, in his view, are expressed by periphrastic patterns. Cinque takes this total order to reflect a universal functional architecture overlaying verb phrases.

Building on a critical assessment of some of the predictions made by Cinque’s hierarchy, and on the distinction between viewpoint aspect and eventuality modification drawn by Smith (1991), Laca (2004a, 2004b) has advanced the hypothesis that temporal-aspectual periphrases distribute over at least two layers of structure. In the lower, innermost layer, periphrases are v/VP modifiers, which change or further specify the temporal structure of the situation described in the VP. In the higher, outermost layer, periphrases contribute a time-relational operator defining a relationship between the time of the described eventuality and a further interval—which can be conceived of as the “interval of visibility” by Smith (1991) or Assertion/Topic Time by Klein (1994) and Demirdache and Uribe-Etxeberria (1997).

The distinction is based on four correlated differential properties: linear position, possibilities of co-occurrence, selectional restrictions, and tense restrictions. In contrast with the few time-relational periphrases, such as Fr. aller + Vinf, venir + de + Vinf, and être + en train de + Vinf, the much more numerous eventuality-modification periphrases (a) never precede time-relational periphrases ((30a,b)), (b) can combine rather freely with each other ((31a,b)), (c) exhibit selectional restrictions as to the temporal structure of the eventualities they combine with ((32a,b)), and (d) lack any tense restrictions ((33a,b)):

(30)

(31)

(32)

(33)

The issue of selectional restrictions and tense restrictions will be taken up in section 4.

It should be stressed that a periphrastic pattern may have a double status with regard to this distinction. The clearest case is that of Ibero-Romance acabar + de + Vinf, which is ambiguous between a “terminative” and a “recent past” reading, as shown by the two possible translations in (34) (Laca, 2005; Olbertz, 1998).

(34)

(35)

(36)

This situation, in which one and the same periphrastic pattern clearly functions at two different layers of structure, seems to be relatively exceptional in the temporal-aspectual realm. As we will see in section 3.3, it arises with most modal periphrases and constitutes a central issue in the analysis of modal verbs. .

3.3 Modal Periphrases, Syntactic Layers, and Semantic Types

Modal periphrases are semantically easy to recognize, since they centrally involve the notions of necessity and possibility. The most widely discussed issue in this domain is the distinction between two major types of interpretations, known as epistemic and nonepistemic modalities. This distinction is claimed to correlate with different layers of structure, with epistemic modals occupying syntactically higher positions than nonepistemic modals (cf. Cinque, 2001a; Falaus & Laca, 2020; Olbertz, 1998; van der Auwera & Plungian, 1998). Modals in epistemic interpretations express that the propositional content they embed either follows from (in the case of necessity) or is compatible with (in the case of possibility) the information or evidence available to an epistemic agent, normally the speaker. By contrast, modals in nonepistemic interpretations deal with what is necessary or possible according to a set of given circumstances and to more or less explicit bodies of rules or laws determining the way things are or should be (Kratzer, 1981/2012). The prototypical case of nonepistemic modality is deontic modality, the ascription of obligation and permission. The overwhelming majority of modal periphrases in Romance are apt to express both epistemic and nonepistemic interpretations. Thus, the pattern Sp. tener + que + Vinf is an expression of epistemic necessity (deduction) in (37a), but it is the expression of a necessary condition for reaching a goal (teleological modality) in (37b):

(37)

Notice that in the case of (37b), the modal periphrasis is preceded by a temporal/aspectual one. In such combinations, epistemic interpretations are excluded. In a tradition going back to Picallo (1990) and extending to Hacquard (2009) and Homer (2013), this distribution is given a structural explanation: Whereas nonepistemic modal periphrases are modifiers of verb phrases, epistemic modal periphrases attach at a higher level of structure, at the very least above aspectual modifiers. A correlation between semantic ambiguity and differences in the syntactic height of the periphrastic pattern—which we exemplified in section 3.2 with Ibero-Romance acabar + de + Vinf—is thus generally assumed for the realm of modal periphrases. In the approach by both Olbertz (1998) and Cinque (2001a), modal periphrases are interwoven into the architecture postulated for temporal and aspectual periphrases, in a hierarchy distinguishing several subtypes of nonepistemic modalities.

Section 2.4 suggested that the raising-versus-control distinction cuts across the realm of periphrases, with a majority of the periphrastic [V1]s exhibiting the typical lack of restrictions on their subject, which characterizes raising verbs. It might seem tempting to correlate epistemic modalities with raising and nonepistemic modalities with control structures (cf., for instance, Gavarrò & Laca, 2002, p. 2720, for Catalan). However, the correlation between nonepistemic modalities and control does not hold (cf. Motapanyane & Avram, 2000; Wurmbrand, 1999): Obligations or dispositions are perfectly possible in combination with expletive ((38a)), passive ((38b)), and inanimate subjects ((38c)), which presuppose raising structures:

(38)

Moreover, although it is true that modals in epistemic interpretation do not allow for null-complement anaphora ((39a))—a possibility that only seems open to control structures ((39b)) (cf. section 2.2)—nonepistemic modals with nonagentive subjects ((39c)) do not allow null-complement anaphora either (cf. Authier, 2011; Olbertz, 1998, pp. 94–103):

(39)

Periphrastic patterns containing the descendants of *potere and debere, as well as those with verbs of possession as [V1] (for instance, Cat. haber + de + Vinf, Port. ter + de/que + Vinf, and Sp. tener + que + Vinf) exhibit all the properties characterizing verbal periphrases (cf. section 2). The auxiliary status of these [V1]s seems uncontroversial. By contrast, there is less consensus as to the status of some patterns that are undoubtedly modal from the point of view of their semantics but do not fully behave syntactically as periphrases (Cinque, 2001a). At one extreme, we find patterns with volitional modals, which are subject to restructuring but are headed by a [V1], which seems to contribute its own argument structure. At the other extreme, we find patterns with epistemic/evidential seem-verbs, which contribute no argument structure of their own but are rarely subject to restructuring.

The patterns with volitional modals as [V1], such as It. volere + Vinf or Sp. querer + Vinf (see the class of want-verbs in Grano, 2015; Remberger, 2010) exhibit some of the properties characterizing verbal periphrases. So, for instance, they show restructuring phenomena, as illustrated by clitic climbing in (40a) and by the reflexive passive in (40b):⁴

(40)

Moreover, like modal auxiliaries with animate subjects, they give rise to null-complement anaphora:

(41)

However, unlike what happens in prototypical verbal periphrases (cf. section 2.2), the constituent headed by [V2] can be replaced by a demonstrative ((42a)), and it can be coordinated with a noun phrase ((42b)) or with a finite object clause ((42c)). This suggests that the constituent headed by [V2] is a full argument of [V1] (cf. RAE-ASALE, 2009, p. 2127):

(42)

Over and above selecting for a [V2]-complement, which behaves as an argument clause, volitionals clearly select for an animate subject argument. Their contribution to the argument structure of the construction is thus more important than that of [V1]s in typical periphrastic patterns.

Epistemic/evidential seem-verbs, as, for instance, It. Sembrare and Sp. parecer (Alcázar, 2018; Cornillié, 2007; Haegeman, 2006), like bona fide modal auxiliaries, contribute no argument structure of their own, but they rarely, if at all, show restructuring effects:

(43)

3.4 Grammaticalization and the Typology of Periphrases

The overall picture that emerges from the past three decades of research on grammaticalization in the TAM domain (as initiated by Bybee & Dahl, 1989, and Bybee et al., 1994) confirms the initial hypothesis that functional exponents of TAM across languages crystallize on a small set of categories with analogous, if not identical, semantic profiles: future, past, progressive, perfect/anterior, and so on. It is therefore natural that an important part of research on periphrases focuses on those whose semantic contribution can be associated with one of these cross-linguistic targets for grammaticalization processes.

Such periphrases may be conceived of as representing stages on one or the other of the grammaticalization paths for which there is robust cross-linguistic evidence (Deo, 2015). Periphrases expressing perfect aspect or anterior relative tenses, progressive aspect, and prospective aspect/future tense are among the best studied (Fleischman, 1982; Squartini, 1998). These innovations enter into a complex competition with older (at the relevant synchronic state) synthetic forms. Competing synthetic forms may be older expressions of the specific meaning of the innovation, as, for instance, the synthetic anteriors based on the Lat. perfect stem versus the new periphrastic anteriors of the type habere/esse + Vptcp, or the Romance synthetic futures versus the new periphrastic futures of the type ire + Vinf in Spanish, Portuguese, and French. But the older synthetic forms may also be unspecified as to the meaning of the newly created form, as in the case of the contrast between simple tenses and the new periphrastic progressives of the type BE + Vger / prep + Vinf (It. and IbR. estar + Vger, Pt. estar + a + Vinf, and Fr. être + en train de + Vinf). Since these complex competition situations give rise to all sorts of descriptive problems, involving, among other things, dialectal and register microvariation, it is not surprising that they have attracted more than the lion’s share of attention in research. It is, for instance, impossible to do justice to the bulk of publications devoted to the periphrastic expression of futurity in Romance (see, however, Bravo, 2008, and Aaron, 2010, on Spanish; Schrott, 1997, on French; Poplack & Malvar, 2006, on Portuguese; and Pana Dindelegan, 2013, § 2.2.1.2.2 on Romanian).

One of the most fertile ideas advanced by Squartini (1998, pp. 74–90) is that aspectual-temporal periphrases grammaticalize along a cline necessarily including a stage in which they are the expression of a particular type of temporal structure or Aktionsart. That is, before becoming expressions for grammatical aspect or tense, periphrases are modifiers of temporal structure. Thus, for instance, at the beginning stages of the grammaticalization of progressives, a number of competing patterns for durative or gradual change Aktionsart are recruited. Only in later stages is one of them selected as the single exponent for the new aspectual category. The other candidates normally subsist in the language as event-modification periphrases, associated with a rather specific temporal structure.

Such a process has taken place in Ibero-Romance and Italian. Next to the periphrastic progressives, these languages exhibit similarly built patterns with motion verbs (the descendants of Lat. ire, ambulare, *andare, and venire) and the gerund, which express temporal structures analogous to those of frequentative or gradual-change lexical verbs (Laca, 2006; Squartini, 1998). Such periphrases are only very rarely used in Modern French and Modern Romanian (Laca, 2004b; Niculescu, 2014). Modern Romanian preserves a copular periphrastic progressive a fi + Vger exclusively in the so-called present presumptive (Pana Dindelegan, 2013, p. 40). Modern French has recently recreated a progressive periphrasis être en train de + Vinf after the decay of the old system of periphrases with the present participle/gerund (V-ant) (Gougenheim, 1929).

From a syntactic perspective, grammaticalization has been characterized as “upward reanalysis” (Roberts & Roussou, 2003), a process by which material originally merging at the lexical level is reanalyzed as merging at the functional layer and further at progressively higher stages in the functional layer. This characterization squares well with Squartini’s idea of an intermediate Aktionsart stage in the development of time-relational periphrases. Sometimes the Aktionsart stage and the time-relational stage coexist synchronically for a periphrastic pattern, as illustrated by the case of Ibero-Romance acabar + de + Vinf discussed at the end of section 3.2.

As for modal periphrases, given the generally assumed differences in configurational height between nonepistemic and epistemic modality, the natural assumption would be that they start their functional life at the lower layer of ability and obligation and then acquire epistemic uses (cf. van der Auwera & Plungian, 1998). Although this seems on the whole to be the right picture, at least one periphrastic pattern, Sp. haber + de + Vinf, arguably exhibits predominantly epistemic uses from the very start (Garachana, 2017a).

4.1 Tense Restrictions

There is a widespread but rarely explicitly discussed assumption that the more auxiliary-like [V1] is, the higher the likelihood of tense restrictions (Bonami, 2015). The possibility of exerting selectional restrictions, in turn, is an indication of at least partial lexical status. Bona fide auxiliaries, as, for instance, Cat. anar ‘go’ in anar + Vinf (the periphrastic past), tend to show tense restrictions (this particular pattern in Catalan only exhibits present morphology) and no selectional restrictions whatsoever.

As to tense restrictions, they undoubtedly exist for the set of time-relational (including temporal) periphrases. Thus, Fr. venir + de + Vinf, aller + Vinf and être en train + de + Vinf, It. stare + Vger and solere + Vinf, and Sp./Cat. acabar + de + Vinf (in the relevant, recent past meaning) and soler + Vinf are impossible in the compound tenses and in the perfective past. Some other periphrasis, as, for instance, Sp. ir + a + Vinf or Sp./Cat. estar + Vger, change their properties when occurring in the compound tenses and in the perfective past, in such a way that they may be suspected of losing their time-relational character (Bravo, 2008; Laca, 2005; Squartini, 1998). Curiously enough, Pt. acabar + de + Vinf (in the relevant, recent past meaning) appears either in the present or in the perfective past without any ascertainable difference as to temporal location (acabou de sair=acaba de sair ‘s/he’s just left’).

Analyses of tense restrictions on [V1] have to answer a fundamental question as to whether the observed defectivity is arbitrary or motivated (Bonami, 2015). In arbitrary defectivity, the paradigmatic gaps are not amenable to explanation, whereas in motivated defectivity there is a deeper rationale to the gaps. The fact that the tenses excluded by time-relational periphrases form a semantic natural class (they are perfective or perfect tenses) has been taken by Laca (2005) as an indication of motivated defectivity. However, the fact that for some of these periphrases the restrictions also extend to nonfinite forms ([V1] in the infinitive is impossible for Sp./Cat./It. soler(e) + Vinf, for IbR. acabar + de + Vinf, for Sp./Pt. ir (+a) + Vinf, and for Fr. venir + de + Vinf, aller + Vinf) may hint at arbitrary defectivity.

Differences in tense restrictions for modal periphrases along the epistemic/nonepistemic divide are a central topic in contemporary semantics (cf. Portner, 2009, ch. 5, and the references therein). For the Romance languages, research has concentrated on the interaction between modals and compound tenses (Demirdache & Uribe-Etxeberria, 2008; Hacquard, 2009; Mari, 2015), so that we still know relatively little about the interactions involving other temporal or aspectual periphrases. This could be a promising field of inquiry, which might shed light on both the semantics of different temporal-aspectual periphrases and the issue of different flavors of modality, particularly in the nonepistemic domain.

4.2 Selectional Restrictions

Selectional restrictions bear first and foremost on the temporal structure of the event described in the constituent headed by [V2]. Their analysis offers important clues as to the semantics of specific periphrastic patterns and as to subtle meaning differences between closely related patterns. Furthermore, progress of a periphrastic pattern along a grammaticalization path is normally evidenced by a loosening of selectional restrictions, whereas the arrested development of a periphrastic pattern manifests itself in tighter selectional restrictions, which may acquire the status of lexical collocations (Garachana, 2016). In the next paragraphs, we give some examples of the relevance of selectional restrictions.

Phasal periphrases denoting the initial or final subinterval of an event, such as Fr. commencer+ à+Vinf or cesser + de+Vinf, obviously require that it is possible to distinguish initial or final subintervals for the event distinguished and are thus incompatible with nondurative event descriptions such as achievements and semelfactives. In fact, incompatibility with cesser + de+Vinf and their ilk is the most reliable test for distinguishing achievements from accomplishments. Periphrases built with motion verbs and the gerund, such as It. andare+Vger, may be identified as expressions of pluractionality (frequentative or gradual aspect) on the grounds of their incompatibility with states and with once-only events (Laca, 2006; Squartini, 1998).

As for the meaning differences detectable by selectional restrictions, Squartini (1998, p. 130) shows that the closely related It. stare+Vger and stare+ a+Vinf differ in that the former, but not the latter, is compatible with achievements, which is an indication of its status as a relatively advanced expression for progressive aspect. Bertucci (2011) systematically exploits the differences in selectional restrictions in his detailed description of aspectual periphrases with começar ‘begin’, continuar ‘continue’, deixar ‘quit’, parar ‘stop’, passar ‘pass’, voltar ‘resume’, and acabar/terminar ‘finish’ as [V1] in Brazilian Portuguese. Rosemeyer (2016) compares the diachronic evolution of the pattern tornar(e)+ a+Vinf in Ibero-Romance and Italian and shows on the basis of their selectional restrictions that they first grammaticalize as restitutives and only in later stages acquire a repetitive meaning.

Tense restrictions and selectional restrictions give important clues as to the semantics of periphrastic patterns. They should be distinguished from restrictions that are constitutive for the makeup of the periphrastic pattern itself. These comprise the type of complement subcategorized by [V1], as well as cases of fixed person and number inflection for [V1].

Even if periphrases are often loosely referred to by their [V1], it is important to bear in mind that they are constructions. When the same [V1] subcategorizes for different types of [V2] constituents, the periphrastic patterns that arise are normally different constructions with different meanings. Thus, for instance, Fr. finir + de + Vinf ‘finish Ving’ is a phasal periphrasis denoting the culmination of a telic event and is therefore incompatible with [V2] constituents which cannot be interpreted as durative and telic event descriptions. By contrast, Fr. finir + par + Vinf ‘end up Ving’ is an ordering or scalar periphrasis, which may in principle apply to any type of event. In the same vein, Sp. ir + a + Vinf is a time-relational periphrasis expressing prospective aspect or future tense, whereas ir + Vger is an eventuality modification periphrasis expressing gradual change. Only in some rare cases is there apparently free variation between two different prepositions, as in Fr. continuer + à/de + Vinf, or between two different nonfinite forms, as in Pt. estar + a + Vinf / estar + Vger, the former being preferred in European and the latter in Brazilian varieties.

As to fixed inflection, some modal periphrases expressing necessity, such as It. bisogna + Vinf, Cat. cal + Vinf, Fr. il faut + Vinf, and Sp. hay + que + Vinf, are impersonal expressions that have become fixed in the third person singular (cf. Benincà & Poletto, 1997; Garachana, 2017a).