Wide-focus subject-verb inversion in Ibero-Romance: A locative account

This paper explores the hypothesis that wide-focus subject-verb inversion in Ibero-Romance is a type of locative inversion, involving a null locative argument. Ibero-Romance displays fine-grained, systematic variation determined by verbal class and variety, offering evidence that Ibero-Romance neutral word order is SVO, rather than VSO as claimed by some null-subject accounts. It is proposed that ‘locative’ subject-verb inversion is a consequence of grammatically-encoded deictic features correlating with the semantic properties of the verbs involved. The locative element, available unequally across Ibero-Romance, can surface in different positions in the left periphery, yielding the variation encountered. The data indicate that the licensing of these constructions depends neither on the null-subject parameter, since this type of inversion also occurs in non- and partial null-subject varieties, nor on the unaccusative/unergative division, though in both cases a degree of correspondence exists.


Wide-focus inversion as locative inversion
Following Pinto (1997), Tortora (1997;2001) and Sheehan (2006;, we assume that a phonologically-null locative obtaining in initial position is capable of satisfying the EPP, analogous to locative inversion constructions elsewhere (cf. Eng. "Into the room walked John"). Drawing on Benincà's (1988) idea that verbs disallowing inversion appear to lack a locative argument, Pinto (1997) proposes that Italian intransitive verbs are of two kinds: inversion verbs which select an extra locative/temporal argument -in her terms, a covert clitic loc (a label which also subsumes the temporal nature of the argument) -, and non-inversion verbs, which do not. Inversion verbs generate differences in interpretation depending on whether the subject surfaces in a preverbal (4) or postverbal (5) position: Italian (Pinto 1997: 12) Dante è entrato. Dante be. prs.3sg enter.pst.ptcp 'Dante entered (into some place).'

(5)
Italian (Pinto 1997: 12) È entrato Dante. be.prs.3sg enter.pst.ptcp Dante 'Dante entered (here/into this place).' When the subject occurs postverbally with a wide-focus reading, the interpretation is deictic, implying locative or temporal proximity to the speaker. Existing work does not specify how the deictic interpretation is assigned, though Pinto (1997: 10) assumes it occurs through an unspecified pragmatic procedure. Sheehan (2006: 168) suggests that null locatives are prepositional phrases, though she does not indicate whether the responsibility for assigning a deictic interpretation lies with this element or not. She leaves the details of her hypothesis' implementation to future research. In what follows, we test the above hypotheses against a set of six closely-related varieties from the Ibero-Romance family. In particular, we question the validity of assuming a pragmatic procedure for the assignation of a deictic interpretation; instead, we take up Sheehan (2006)'s suggestion that the null locative may be prepositional in nature, and use recent work on the fine structure of prepositional phrases encoding spatial information (Cinque & Rizzi 2010) to account for the syntactic assignation of the locative's deictic interpretation. We distinguish between cases of overt locative inversion and cases of subject-verb inversion involving a covert locative by referring to the latter as 'locative' subject-verb inversion, with single quotation marks.
The paper is organised as follows: after setting out some preliminaries regarding methodology and empirical sources (Section 2), Section 3 presents the results of the data collection, summarising the word-order facts of the empirical evidence, and surveying the semantics of the verbal classes investigated. In Section 4, we put forward the analysis advocated here in terms of a covert locative element, which can satisfy the EPP in cases of 'locative' subject-verb inversion and which assigns the deictic interpretation of such sentences. Specifically, we examine the internal architecture of the proposed locative PPs, which may project more or less internal structure according to the spatio-deictic features these encode. We suggest that the word-order alternations observed with 'inversion' verbs are in fact dependent on the availability and subsequent selection by the lexical verb of a null locative prepositional argument, which is generated within the VP before moving to a higher, preverbal clausal position. Finally, we detail cross-varietal availability of the null locative PPs and how these surface in the TP/CP border and left periphery.

Materials and methods
Our investigation surveys wide-focus inversion across six Ibero-Romance varieties: Asturian (Ast), Brazilian Portuguese (BP), European Portuguese (EP), European Spanish (ES), Mexican Spanish (MS) and River-Plate Spanish (RS) (collectively referred to as Latin-American Spanish). All are consistent null-subject languages (NSLs) apart from Brazilian Portuguese, which we describe as a partial NSL (Holmberg, Sheehan & Nayudu 2009); viz., it shows a more limited distribution of null-subjects than consistent NSLs but displays many pro-drop characteristics nevertheless. 2 The varieties differ to certain degrees according to the prominence of their grammatical relations: European Portuguese is regarded as a subjectprominent language whereas Brazilian Portuguese is topic-prominent (and known to be more restrictive in licensing inversion constructions generally). Spanish shows greater word-order flexibility than either Portuguese variety and can be considered to be discourseprominent. Informal observation suggests that European Spanish allows more inverted word-order configurations than its Latin-American counterparts, an observation borne out in our investigation's findings. The data here suggest that Asturian patterns similarly to European Spanish in this respect, though further work is required to confirm such a claim.
Across the varieties, a total of 170 informants gave acceptability judgments on inversion constructions in their native language. Of these, the data analysed here represent the responses of 18 native Asturian speakers, 28 European Spanish speakers, 7 Mexican Spanish speakers, 15 River-Plate Spanish speakers, 3 20 Brazilian Portuguese speakers, and 15 European Portuguese speakers. 4 These informants were provided with a contextualised wide-focus structure in two versions, one with and one without an inverted order (as illustrated in (7) below). They were then asked to select which word order was most appropriate for the given context (out of the choices 'most appropriate for the context', 'appropriate for the context', and 'inappropriate for the context'), and, if both were acceptable, whether there was a difference in interpretation. We also worked closely with a smaller number of informants to gain a deeper understanding of the results obtained.
Wide-focus responses were elicited so that the "default" word order was obtained. 5 Usually, the method for eliciting such responses is to use a "what happened?", or equivalent, question, where nothing is presupposed. However, in Spanish, the complementizer que ('that') often precedes a reply to such questions, as in (6) Kato & Negrão (2000) discusses the relation between Brazilian Portuguese and the null-subject parameter. 3 Specifically, only speakers from Buenos Aires and Mar del Plata, Argentina. 4 This leaves 67 speakers whose responses were eventually excluded from the analysis, due to incompleteness (i.e. survey left unfinished) and/or unreliability of the data (i.e. misunderstanding of the task). Within the excluded group, the results of 21 informants for Asturian were left out as these speakers were natively bilingual in two languages (Asturian plus a second, usually Spanish, L1) and gave judgments which differed across the board from those whose first language was Asturian only (although these speakers also spoke Spanish as an L2). To control for default word order across all informants, only Ibero-Romance speakers who judged SVO over VSO orders in transitive sentences as acceptable were included (this group represented the vast majority), since on closer consultation with the relevant informants we found that the acceptability of the VSO order was dependent on discourse conditions distinct from those targeted by the questionnaire (e.g. speakers had in fact assumed a narrow focus interpretation for the subject; cf. also fn 1). 5 We follow Zubizarreta (1998: 1), after Chomsky (1971;1976) and Jackendoff (1972), in assuming that "the focus is the non-presupposed part of the sentence", and that "the presupposed part of a sentence is what the speaker and hearer assume to be the case (i.e. the shared assumptions) at the point at which the sentence is uttered in a discourse".
Use of que makes inversion sentences more acceptable (Sheehan 2006;Leonetti 2014). To avoid this effect, responses were elicited from contexts designed to extract a wide-focus sentence from the informant. The following example is the test used to elicit an out-of-theblue response for a transitive verb: This method proved a more reliable way of eliciting the desired judgment from informants than using the normal "what happened?" question. Moreover, we controlled for variation in interpretation by using the same contexts across varieties so that informants were asked to judge sentences arising from identical contexts. In the section that follows, we present the results of this data collection, summarising the microvariation observed in subject-verb inversion across the Ibero-Romance varieties under investigation.

Sentences without a word-order alternation
In all Ibero-Romance varieties surveyed, the subject must surface in initial position with transitive verbs in wide-focus contexts:  han (2006; 2010; 2015) for a comparative perspective on these languages' word orders. 7 There has been a decline in usage and narrowing of function of the present perfect in Latin-American Spanish (cf. Harris 1982;Squartini & Bertinetto 2000), hence the elicitation in the questionnaire of past simple responses in these varieties and the present perfect in European Spanish. Tense/aspect was otherwise controlled for, since these factors may affect judgments, although it appears not to be the case in this instance. However, if tense/aspect do affect the acceptability of inversion, then the fact that European Spanish distinguishes between the perfect and preterite past tenses, whereas the other varieties surveyed employ the simple past for both perfect and preterite meanings, may prove relevant. While here the licensing of inversion appears to correspond to the unaccusative/ unergative divide, Section 3.2 presents data which invalidate this assumption. Nonetheless, together with the unacceptability of wide-focus inversion with transitive verbs, the fact that inversion is generally disallowed with unergatives in wide-focus constructions indicates that Ibero-Romance basic word order is SVO. Preverbal subjects appear to be in the canonical argumental subject position, SpecTP, rather than a left-dislocated position, which suffices to disprove Alexiadou & Anagnostopoulou (1998). 10 Moreover, it is unclear how their approach would explain the infelicitous instances of VS orders, given that the verb's (pro-)nominal properties are meant to satisfy the EPP. In the absence of the verb being able to fulfil the EPP requirement, we assume that some other element must be present in a preverbal argumental subject position in order to derive a felicitous VS configuration in these Ibero-Romance varieties (the specific mechanics of which will be presented in §4). In addition, since the inversion data observed here includes Brazilian Portuguese, this type of 'free' inversion cannot only be a property of consistent NSLs. Accordingly, it cannot be a direct consequence of the null-subject parameter.

Ibero-Romance word-order alternations
A number of intransitive verbs in Ibero-Romance permit both SV and VS orders in widefocus contexts, corroborating Pinto (1997)'s bipartite division of intransitives into inversion and non-inversion verbs. However, the data presented here reveal that inversion occurs to differing yet systematic degrees across varieties. To present the complex variation in a comprehensible format, the inversion verbs allowing an SV/VS alternation are divided into four classes whose properties are detailed below. These groupings highlight a correspondence between the semantic features of inversion verbs and the systematic distribution of word-order configurations across Ibero-Romance. We will observe in §3.3 that not only do the verbal classes proposed below share key semantic features, but they also provide evidence for an Ibero-Romance "scale" of inversion with respect to the possibility and frequency of wide-focus VS structures (that is, verbs highest on this scale occur most often in an inversion configuration, whereas verbs lowest on the scale figure least often in a VS order). In turn, wide-focus VS structures systematically entail a deictic reading such that the event they describe is interpreted as anchored to the speaker's loco-temporal co-ordinates (an interpretation marked [+ deixis] in the following sections), contrasting 10 Sheehan (2006: 72) suggests that the burden of proof lies with the challengers of a preverbal subject position for Romance to substantiate their proposal that EPP satisfaction is parameterised. She claims that, in order for Alexiadou & Anagnostopoulou's (1998) analysis to be invalidated for a given language, one only needs to 'prove that i) the verb raises to I in said language, and ii) at least some preverbal subjects in said language do not occupy an A-bar [non-argumental] position ' (2006: 31). In her thesis, Sheehan provides substantial evidence that the basic word order of Romance is SVO, concluding that the empirical data falsifies the predictions made by analyses attributing null subjects to a parameterised EPP.
with SV structures whose interpretation is underspecified with respect to deixis (marked [ø deixis]), as illustrated for the Italian examples in (4-5). The regularity of this patterning of 'deictic' inversion structures relative to the verb's semantic properties suggests that said features are syntactically represented in these varieties; if this is the case, then the deictic interpretation of the inversion constructions appears to be syntactically encoded rather than derived inferentially from the utterance context, contra Pinto (1997), and in line with independent proposals (e.g. D 'Alessandro 2004;Giorgi 2010;Hill 2013) which argue for the formal encoding of pragmatic information in Romance. 11

Semantic features of inversion verbs allowing an SV/VS alternation
The relevant features pertaining to Ibero-Romance wide-focus 'locative' inversion are path, location and deixis, which we will propose in §4 to be formally encoded in Ibero-Romance.
The notion of path refers (here) to a telic trajectory which involves either a source (the start-point of the trajectory e.g. 'the bird fell from the sky') or a goal (the endpoint of the trajectory e.g. 'the bird fell to the ground'). The term location refers to a static spatial co-ordinate (e.g. 'the bird lay on the ground'). The concept of deixis, which anchors a sentence relative to a contextual index (the origo, or deictic centre, which is typically, but not necessarily, ego-centric), is variable in the strength of its meaning contribution in the relevant verbs. Thus, a strongly deictic verb such as venir/vir 'to come' obligatorily encodes a location proximal to the speaker (cf. fn 13), whereas other verbs occur in structures entailing a more general notion of deixis, linking the sentence only loosely to the utterance context (e.g. irse 'to leave', cf. §3.2.4). In §4, the 'variability' in the deictic strength of the structures under investigation is shown to be an effect of the compositionality of syntactically-encoded deictic meaning, which, following Sigurðsson (2004;, is understood here to be divisible into co-ordinates relative to an agentive interlocutor (typically the speaker), a patient interlocutor (typically the addressee) and the speech time, location and wider discourse context.
In the following sections, we base our classifications according to the presence of these key units of meaning -i.e. whether a verb is [± goal, source, location, deixis] -in the semantics of inversion verbs (for further discussion, cf. Corr 2012: 23-27). Anticipating the exposition of the theoretical analysis in §4, it is these pragmatico-semantic features which are encoded in the clausal architecture, triggering the movement of different sizes of complex null locative prepositional arguments that are proposed to be c-selected by inversion verbs.

Class A: +goal, +location, +deixis
Class A verbs (e.g. unaccusative venir/vir, 'to come') encode the key features of goal, location and deixis. With these verbs, both the inverted and non-inverted order are permitted across Ibero-Romance: However, the subject tends to surface post-verbally with such verbs in Asturian (17) and Spanish, whereas it tends to occur preverbally in Portuguese (18) Verbs in this class involve inherently-directed motion (Levin 1993;Fábregas 2007), which may be strongly deictic in the Ibero-Romance varieties considered here. 13 Comparison with Catalan, which we leave to future research, would be worthwhile in this respect, since its motion verbs, demonstratives, and locatives have different deictic paradigms from those of Spanish (Vann 1997: 307-8).
The relevant semantic features of Class A are motion to a fixed point (i.e. telic direction), location (the static location where the event is situated or to which, in this case, movement occurs) and deixis. However, given that not all inversion verbs are motion verbs, we prefer to capture the idea of directionality using the term path as discussed in §3.2.1 12 I qualify the notion of "tendency"/"preference" in §3.3. 13 Observe the contrast in acceptability between the following Spanish sentence and its English translation: (i) Spanish *¿Vengo a la biblioteca? come.prs.1sg to the library 'Shall I come to the library [location of addressee; speaker elsewhere]?' Thus, what constitutes a perfectly grammatical sentence in English is unacceptable in Spanish, if the speaker is not in the location of the addressee. The expression can only be rendered felicitous in Spanish by the verb ir ('to go'), which implies motion away from the speaker's domain only: (ii) Spanish ¿Voy a la biblioteca? go.prs.1sg to the library 'Shall I come to the library [location of addressee; speaker elsewhere]?' An anonymous reviewer suggests that llegar 'to arrive', unlike verbs such as venir 'to come'/ir 'to go', should not be classified as deictic, due to mismatches such as venir aquí/#allí ('to come here/#there'), ir #aqui/ allí ('to go #here/there') which contrast with the felicitousness of llegar aquí/allí ('to arrive here/there'). However, we maintain that there are a number of reasons to retain llegar in the same category as venir. For example, in practice, sentences containing llegar involve a proximal reading more often than a distal one (for example, in Mark Davies' Corpus del Español, there are 83 tokens of llegar aquí versus 47 of llegar allí in their C20 Spanish data). Moreover, llegar's interpretation is interchangeable with Class A venir in certain contexts in a way in which a verb encoding [+goal, + location] but not [+deixis] such as Class B entrar 'to enter' (cf. §3.2.2) is not, e.g. llegó/vino {*entró} de vacaciones a Puerto Rico ('He came {*entered} on vacation to Puerto Rico'). Since path and location are encoded in the semantics of all three, we suggest that the patterning of llegar with deictic venir rather than non-deictic entrar is to do with the directional meaning taking precedence, or being somehow 'stronger', in the former two, whereas the locational meaning takes precedence in the latter. This is reflected in the existence of the telic functional periphrases llegar a/venir a (cf. #entrar a), which, again, are synonymous in certain uses, e.g. Después de largas pretensiones, vino a conseguir la plaza 'After lengthy efforts, he came to achieve the job', Llegó a reunir una gran biblioteca 'He came to put together a large book collection' (examples from the dictionary of the Real Academia Española, respectively: <http:// dle.rae.es/?id=bXkUiz2>, <http://dle.rae.es/?id=NV155vl> [retrieved 9 February 2016]). While both orders are acceptable in River-Plate Spanish within this category, the VS order is preferred with the unergative llamar, 'to call' (23); conversely, the unaccusatives entrar, 'to enter' (24) and morir, 'to die' (25) are both favoured with the non-inverted order: (23) demonstrates that the patterning of inversion verbs does not strictly follow the unergative/unaccusative distinction in Ibero-Romance.
The variation in the degree to which Class B verbs permit wide-focus inversion is a function of the individual semantics of each verb. Llamar/ligar ('to call') denotes the transfer of information from one participant to another. The verb entails a notion of direction towards a participant within the speech act, a relation between speech participants that does not obtain in 'to die' or 'to enter'. Morir/morrer ('to die') plausibly involves movement away from one state (sc. being alive) to another (being dead) (Theresa Biberauer, p.c.): the change of state in morir/morrer is thus analogous to the change of location involved in entrar ('to enter'). This difference may be reflected in the temporal (rather than locative) interpretation of morir/morrer constructions (cf. Pinto's 1997: 23) example: È morto Fellini, 'Fellini has (just) died').

Class C: (-goal), +source, +location
The key semantic features of Class C verbs (e.g. unaccusative salir/sair 'to leave') are [+source, +location]. With this group, only the non-inverted order is permitted with Portuguese (26)  The absence of a goal feature in the semantics of Class C verbs is potentially crucial, since its presence or absence appears to correlate with tendency towards or away from the inverted order. While Class C verbs imply movement away from an unspecified location, there is a contrast between those with no deictic reference (e.g. salir/sair) and those which are also (subtly) participant-oriented (e.g. Spanish irse). In the latter example, the presence of the reflexive clitic se gives a deictic/indexical dimension to the verb, making irse plausibly participant-oriented in a way that salir/sair is not. Thus, in Spanish, irse (30) is more likely to figure in an inverted construction than salir (31) In fact, the use of the reflexive in a recipient/benefactive role often gives a participantoriented interpretation to a sentence in Spanish. Compare the difference in meaning between the following sentences with and without the clitic se: Spanish Até los cordones. tie.pst.1sg the shoelaces 'I tied (someone's) shoelaces.' (33) Spanish Me até los cordones. myself=tie.pst.1sg the shoelaces 'I tied my shoelaces.' MacDonald (2004: 5-6) understands such instances of se to be introduced as the complement of a null locative preposition, where the PP expresses that the direct object (los cordones 'the shoelaces', in this case) is in the same location as the denotation of the reflexive, i.e. the subject. The (loco-temporal) co-ordinates of the event are thus restricted to those of the subject, such that the interpretation of (33) is that the event of shoelace-tying occurs where the shoelaces (los cordones) are, and these shoelaces are located where the subject is (via the co-indexation of the clitic with the subject), and thus the shoelace-tying necessarily takes place at the speaker's location. If indeed a locative analysis is appropriate for se constructions (see also de Cuypers 2006;Armstrong 2013) and can be extended to lexicalised irse, the preference for inversion with irse over salir follows as a consequence of the 'extra' locative properties of the former (and lack thereof in the latter), a finding which lends plausibility to our hypothesis that a null locative is responsible for wide-focus inversion constructions.

Class D: +location (-goal, -deixis)
Of the varieties surveyed, only Asturian permits inversion with Class D verbs, such as the unergatives llorar/chorar ('to cry'), gritar ('to shout') and dimitir/demitir/renunciar ('to resign'): Why Asturian should permit inversion with these unergatives is not immediately clear, but what may differentiate non-inversion unergatives (e.g. bailar 'to dance', brincar 'to play') from inversion unergatives like llorar, gritar and dimitir is that the former class encodes non-directional movement but not static location whereas the latter can plausibly involve static location since no movement 14 is encoded in these verbs. Alternatively, an anonymous reviewer suggests that Asturian might categorise such verbs as containing source: Old Spanish llorar 'to cry', for example, could select an overt source, such as llorar de los ojos ('to cry from the eyes'). Whilst this option is unavailable in Modern Spanish, it is possible that Asturian might still preserve such 'substance emission' verbs as source verbs. Approached in this light, dimitir 'to resign' could be viewed as a figurative source verb (cf. dimitió del gabinete palestino 'he resigned from [out of] the Palestinian cabinet'). However, in this case, it would be necessary to explain why Spanish dimitir/renunciar encodes source yet does not allow wide-focus inversion (cf. #dimitió/renunció Putin 'Putin resigned').
The patterning of verbs which trigger subject-verb inversion in wide-focus contexts across Ibero-Romance are summarised in

Word-order "preferences" across Ibero-Romance
We have observed that both inverted and non-inverted configurations are possible across Ibero-Romance with Classes A and B. However, while Asturian and European Spanish also allow both orders for Class C, Latin-American Spanish and Portuguese do not, permitting only the non-inverted order with this class of verbs. Asturian alone licenses inversion with Class D verbs. Thus we can conclude that, while there is relative consistency in Ibero-Romance with respect to the verbs allowing an SV/VS alternation, intra-familial variation persists. Yet simple presence or absence of a null locative cannot account for the variation that obtains: that inversion occurs at all in these varieties suggests that the null locative proposed in previous analyses is available. However, an additional explanation is necessary to account for why inversion occurs with some verbs in some varieties, but not with other verbs in other varieties. Further, we observe that, when both the inverted and non-inverted options are available to a speaker, the speaker will display systematically-varying degrees of "preference" for one word-order configuration over another, depending on the verb used, the variety spoken, and the pragmatic interpretation the speaker assigns to the particular speech act. Thus, in (16), repeated here as (37), Portuguese speakers judge both SV and VS configurations as grammatical but they consistently prefer the inverted to the non-inverted order, despite ( By "preference" (an expedient label revised and qualified below), it is meant that, given an identical context for a particular verb, speakers will systematically show a tendency towards one configuration over the other, though both are felicitous, a phenomenon that recurs across Ibero-Romance wide-focus inversion constructions.
To illustrate, we observe that Ibero-Romance informants presented with Class A llegar/ chegar ('to arrive') judge both the VS and the SV order as felicitous for exactly the same context. However, Asturian speakers show a very strong preference 17 for the VS order (that is, speakers are very highly likely to favour the VS configuration) over the SV order; European Spanish speakers show a strong preference for (i.e. are highly likely to favour) the VS order over the SV order; Latin-American Spanish speakers show a weak preference for (i.e. are still more likely than not to favour) the VS order over the SV order; European Portuguese speakers show a weak preference for the SV order over the VS order; and Brazilian Portuguese speakers show a preference for the SV order, but the VS order remains a possibility, as captured in  The context in which the verb is elicited remains constant throughout, hence the systematic differences in preference cannot simply be attributed to speakers' interpretation of the given context: presumably, there are some interpretative differences that vary according to speaker, but if this were the only explanation then these would occur on an individual basis, but not systematically according to variety as we find in the empirical data. However, since "preference" means very little in the current framework, we can reasonably assume that the patterns observed do not reflect "preferences", but instead have some sort of grammatical explanation. Nonetheless, the term suffices for now, since it captures the empirical facts ( §4.3.3 will explain its relevance). The patterns of "preference" reveal complicated but subtle, systematic variation in Ibero-Romance. Whilst permitting both inverted and non-inverted orders, Class A verbs display a tendency towards inversion across varieties whereas Class C verbs show a tendency towards the non-inverted order. Other verbs pattern somewhere in between. The degree to which a given configuration is "preferred" varies according to both the verb and the variety. The differences and distribution are captured in Table 3, which illustrates a "scale" of inversion that is not only empirically-motivated but also has a semantic motivation in the features described in §3.  The scale above reflects the degree to which Ibero-Romance varieties license wide-focus inversion. Portuguese appears to be most restrictive in this respect, especially Brazilian Portuguese (a difference we would expect, as noted in §2). Spanish is more liberal in licensing inversion, European Spanish being the least restrictive and Mexican Spanish the most. Asturian exhibits the fewest constraints in its licensing of inversion, though word-order distribution is systematic and an SV configuration remains the norm outside inversion constructions. The variation's systematic nature warrants syntactic explanation. That optionality obtains with the verbs surveyed here fulfils the prediction that i) there are inversion verbs, and ii) the different word-order configurations involve an interpretational difference (i.e. they are not "truly" optional; cf. Biberauer & Richards 2006). We assume, therefore, that the alternation is dependent on the availability and subsequent selection of a null locative. 19 The degree to which each option is "preferred", however, necessitates a finer-grained analysis than simple presence or absence of a covert locative as postulated in previous work. 19 An anonymous reviewer suggests that additional evidence in favour of the 'locative' hypothesis is found in the contrast between the inversion and locative properties of individual-level and stage-level predicates (cf. Carlson 1977;Kratzer 1995 We agree that such facts lend plausibility to our 'locative' analysis of wide-focus inversion put forward here, noting in particular Gallego & Uriagereka (2009)'s proposal that stage-level predicates and the copula which selects them (e.g. estar in Spanish) involve more structure, specifically, a complex prepositional component that does not obtain in individual-level predication (see also Jiménez-Fernández 2013; Fábregas 2014). We leave the investigation and implementation of the unification of these phenomena to future research.

The locative argument as a null layered PP
On the basis of the observed correlation between Ibero-Romance inversion variation and the semantics of inversion verbs, we suggested in the preceding section that the spatial and deictic features identified are encoded in Ibero-Romance syntax. In this section, we draw on recent cross-linguistic work on the fine structure of prepositional phrases expressing spatial relations (Cinque 1999;Cinque & Rizzi 2010), which suggests that a single articulated configuration is reproduced universally, though different portions may be spelled out or omitted according to whatever constitutes the relevant information, a structure which is mirrored in the morphology of synthetic languages. 20 Our proposal is that Ibero-Romance displays a type of wide-focus inversion which is sensitive to spatial and deictic features. These features attract different sizes of locative PPs (decomposable into the same universal structure) to the TP/CP border and left periphery, depending on their availability in a given language. The appearance of these locatives in the C-domain is understood to involve movement to multiple argumental positions (cf. §4.3 onwards) rather than left-dislocation and is thus assumed to be capable of satisfying the EPP.

The structure of null locative PPs
The basic structure of the null locative PP combines and replicates previous analyses of spatial PPs (Koopman 1993;2000;Cinque 1999; den Dikken 2010; Pantcheva 2010; Svenonius 2010), which advocate the cartographic encoding of spatial relations within hierarchies of varying complexity. Relevant here is that these accounts propose unique projections for direction (or Path) and location (or Place), with the former layer dominating the latter, as in (38) Support for such an analysis is found in Terzi's work on null spatial prepositions in Greek. In (39), the Greek directional/locative preposition se (appearing in contracted form before the definite determiner) is optionally realised: Greek (adapted from Terzi 2010: 173) Pao (sto) spiti. go.imp.2sg (se.the) home 'I go home.' When se is absent, a null directional PP is assumed to obtain with the following structure: Greek (  Terzi argues that Greek null spatial PPs are associated with an unpronounced DP place, following Kayne (2004;. 21 Incorporating the silent place into the structure of the null locative could account for the subject-like properties locative inversion constructions 20 Observe that the word-for-word gloss of the following example from Lezgian, a head-final Dagestanian language, is the mirror image of its literal translation in English: (i) Lezgian (Riemsdijk & Huijbregts 2007: 341) sew-re-qh-aj bear-augm-behind-from 'from behind the bear' 21 On the basis of English, Greek and Italian data, Terzi (2010) proposes that, unlike null directional PPs, null source PPs are impossible due to their position in the extended projection of a spatial PP.
have been observed to display. We replace the Path projection with individual projections for Source and Goal following Pantcheva (2010)

SourceP
SourceP has projections for each of the proposed functional layers: Class C verbs (e.g. salir/sair 'to leave'), involving a source feature, are predicted to c-select for SourceP. As inversion is only licensed with these verbs in Asturian and European 22 The complete structure posited is as follows (the category below Ax(ial)Part is labelled K for case, often but not always overtly realized via a genitive marker cross-linguistically, such as English of): (i) Svenonius (2010: 144)  The availability of SourceP in some varieties of Ibero-Romance suggests that the lack of null source PPs in English, Greek and Italian is due not to the impossibility of covertly realising a directional source PP (as proposed in Terzi 2010), but to its absence in these varieties.

DeixisP
DeixisP, with functional projections for deixis, goal and location, corresponds to Class A verbs such as venir/vir ('to come'), whose semantic features in turn correspond to the PP's functional projections: As all varieties permit wide-focus inversion with Class A, it is assumed that DeixisP is available across Ibero-Romance. Therefore, the following configurations are possible in Ibero-Romance:

GoalP
The third type of locative PP is GoalP, which encodes goal and location in its functional projections: Since all varieties permit inversion with Class B verbs, the prediction would seem to be that all have GoalP in their lexical entries. However, we modify this hypothesis in §4.3.1, for reasons which will become apparent. For now, the occurrence of GoalP in Ibero-Romance is illustrated as follows: This class of unergatives which c-select for LocationP fits with the type of verb described in Pinto (1997), such as Italian abitare 'to live, dwell', with which wide-focus inversion is licensed only when the locative PP is overt (since, presumably, LocationP is unavailable in Italian also). This makes the prediction that i) Asturian should license wide-focus inversion with a null locative with verbs such as vivir ('to live'), and ii) in general, it should show a much greater incidence of inversion constructions in wide-focus contexts than the other Ibero-Romance varieties discussed here, since Asturian could potentially have "free" inversion with any relevant verb (presumably classified within Class D) c-selecting for LocationP. Nevertheless, intuition suggests there must be some limit to the number of verbs for which Asturian licenses wide-focus inversion as it would seem anomalous for this variety to exhibit so little restriction on the licensing of VS orders relative to other Ibero-Romance varieties. However, I leave investigation of such possibilities to future research.

Intermediate remarks
The foregoing account explains the similarity between overt and covert 'locative' inversion by unifying both constructions under one analysis. On the present proposal, the underlying structure is identical, the difference between the two constructions being whether the locative PP is overtly (55) or covertly (56)  This proposal is more elegant than existing coindexication accounts (Pinto 1997;Tortora 1997). In Pinto (1997), the null locative -the clitic loc, under her analysis -is present in both overt (57) and covert (58) locative inversion, responsible for EPP satisfaction, and must be coindexed with the overt PP in overt constructions: A simpler explanation is that, as already proposed above, Italian does not have Loca-tionP in its lexical entry, so must overtly realise this type of locative PP. Since the PP in locative inversion constructions must surface in clause-initial position, whether overt or covert (Pinto 1997: 57), we must assume that the EPP requirement holds. Moreover, the EPP can be satisfied by a non-subject element: by a PP (containing a DP) when overt, and by the null locative PP containing Kayne's (2005) place DP when covert. Thus whether a PP is overtly or covertly realised is partly dependent on the semantics of the verb in question, since inversion verbs must be able to c-select for the relevant kind of null locative PP. However, it is also dependent on the availability of the different types of null locative in a given variety's lexicon. By accounting for the distribution of wide-focus inversion constructions across Ibero-Romance in terms of the availability of the null locatives, the variation observed is reduced to cross-varietal differences in lexical entries. In this sense, the present proposal conforms to the Borer-Chomsky conjecture (viz. "all parameters of variation are attributable to differences in the features of particular items (e.g. the functional heads) in the Lexicon" Baker 2008: 3). If we wish to think in terms of recently proposed parameter hierarchies (Roberts 2012), then the type of variation attested in Ibero-Romance is to be found at the nanoparametric level, where the observed differences are reducible to the idiosyncrasies of the lexical items in the languages involved.

Wide-focus 'locative' inversion and the left periphery
The previous section (tentatively) predicts that DeixisP and GoalP are available across Ibero-Romance, with SourceP present in Asturian and European Spanish but not in the other Ibero-Romance varieties, and LocationP available in Asturian only. However, if DeixisP and GoalP are available in all varieties, and no other factors are considered, the systematic variation encountered cannot be fully explained. The analysis predicts that inversion constructions involving a null locative should only differ in three ways across Ibero-Romance, whereas the variation observed is much more complex. Consider the following: 24

ES RS EP
To come (  24 For simplicity, the analysis that follows concentrates on European Portuguese, European Spanish and River-Plate Spanish, since the variation between these varieties remains unexplained. Asturian is distinguished from the other Ibero-Romance varieties by the availability of LocationP, and is thus (temporarily) excluded. Mexican Spanish, whose word-order distribution patterns between River-Plate Spanish and European Portuguese (cf. Table 3,  §3.3), is excluded since some data remain unclear and potentially unreliable. Further investigation is therefore necessary before Mexican Spanish can be included in the present discussion. Brazilian Portuguese is excluded in order to control for variables, since it is known to be much more restrictive in the kinds of inversion it allows. However, in terms of the null locative's availability, we can assume that what can be said of European Portuguese also applies to Brazilian Portuguese, since both varieties license wide-focus inversion with the same types of verb (cf. From the above, we can observe that: i. When a SV/VS alternation is available in European Spanish, there is a strong "preference" 25 for a VS order with Class A verbs, a preference for VS with Class B verbs, and a strong preference for SV with Class C verbs; ii. When a SV/VS alternation is available in River-Plate Spanish, there is a preference for VS with Class A verbs, but a preference for SV with Class B verbs (there is no alternation available with Class C verbs); iii. When a SV/VS alternation is available in European Portuguese, the preference is for SV, irrespective of verb type; iv. There are systematic differences in inversion vs. non-inversion preferences which correspond to the verbal groupings (Classes A, B, C [and D]).
The preferences -a notion invalid in the current framework -cannot only be attributed to idiosyncratic differences in how informants interpret the context. Instead, the high degree of systematic variation necessitates a grammatical explanation. In what follows, we reconsider the availability of null locatives, proposing a four-way distinction across Ibero-Romance and concluding that the illusion of speaker "preferences" is an effect of the cross-varietal availability of the null locatives and where they can surface in the TP/CP border and left periphery.

Reconsidering the availability of the null locative
The hypothesis that GoalP and DeixisP are available across Ibero-Romance makes the wrong predictions as it suggests a three-way divide in wide-focus inversion patterning, a distribution in contradiction of the fine-grained variation observed. We therefore readjust our account as follows.
The original hypothesis assumes both GoalP and DeixisP are available in European Portuguese. However, suppose that, in a given scenario, a Class B verb such as entrar ('to enter') c-selects for a GoalP, resulting in an inversion construction. A deictic interpretation should be unavailable because neither verb nor locative argument encodes a deictic feature. Nevertheless, Sheehan (2006; has independently established that wide-focus inversion does entail a deictic interpretation in this context. As elsewhere, we assume an empirically-justified syntactic explanation is preferable to the postulation of a pragmatic procedure. Since deixis is always involved in the interpretation of null locative inversion in Italian (Pinto 1997), it follows that Italian only has one type of covert argument, which must be DeixisP. 26 Given the parallels observed between Italian and European Portuguese elsewhere (sc. unavailability of SourceP and LocationP; restricted inversion relative to Spanish; subject prominence), it is plausible that European Portuguese parallels Italian in that it also only has one null locative argument, DeixisP.
If this is the case, then this cross-varietal distribution of null locatives should be revised as follows: 25 Recall that this label is a provisional measure for capturing the empirical observations. 26 Note that Tortora (1997: 70) suggests that her null locative, a weak (in the sense of Cardinaletti & Starke 1999) pronominal argument pro-loc, is "the morpho-syntactic instantiation of the lexical semantic category location" or goal, but she does not elaborate further on this point (though she does propose that the reasons for the lack of a null locative that encodes source are conceptual in nature ). Like Pinto (1997), however, Tortora does not specify how the element receives a deictic interpretation to begin with (Tortora 2015 remains agnostic on whether the speaker-oriented value of her pro-loc is pragmatically or grammatically encoded). Thus her analysis points towards the idea of Italian having two different types of null locative, one encoding location and the other goal. Nevertheless, she never makes this suggestion explicitly. The implication of such an analysis, however, is that her goal pro-loc would not encode location, and would therefore surely not be a locative. The structural explanation below completes the account of Ibero-Romance wide-focus inversion.

A structural explanation
Intuition suggests that Ibero-Romance's fine-grained variation is a result of differences in where the locative surfaces at the TP/CP border and left periphery. Since deixis is a key element in 'locative' wide-focus inversion, an explanation of the variation observed would logically involve the C-domain, the area of the clausal spine associated with discourse. Yet deixis is the concept which has been least clearly defined in the proposal so far. Informants suggest that locative constructions can be more strongly or weakly deictic in the scope of their reference. 27 For example, in the following inversion construction with the Class A verb venir/vir, the sentence is unambiguously anchored in the speaker's domain, necessarily referring to the speaker's location: 27 Corr (2012: 33-35) provides further support to the notion of deictic reference described here, drawing on, amongst others, Gutiérrez-Rexach (2001)'s proposal that modern Spanish possesses a deictic clitic system derived from the strong adverbial proforms ahí and ahora. Gutiérrez-Rexach (2001), following Gutiérrez-Rexach & Silva-Villar (1998), assumes that modern Spanish has a system of locative and temporal clitics derived from the strong adverbial proforms ahí ('there', near speaker/addressee; cf. allí, 'there', far from both speaker and addressee) and ahora ('now'). The strong proform [a.í] becomes the locative clitic [áj], and the strong proform [a.ó.ɾa] becomes the weak temporal proform [áo.ɾa]/[á.ɾa] (cf. the above-cited works for evidence of these elements' clitic status). Following observations in Torrego (1989) and Bosque (1990), Gutiérrez-Rexach (2001: 159-160) notes that these weak proforms can occur in unaccusative constructions, as in (i a,b), but that they cannot in unergative (ii) and transitive (iii) constructions, contrasting with the availability of the strong proforms as illustrated in brackets in (ii, iii): ( Notably, the examples above with which Gutiérrez-Rexach illustrates the selectional restrictions which the locative clitics must obey are similar to the subject-verb inversion constructions under investigation here. Indeed, Gutiérrez-Rexach (2001: 161-162) goes on to observe that the weak locative [áj] "can be used in some occasions to deictically refer to a specific location, but in most of its uses its content is not that restricted. In general it denotes a contextually determined modifier linked to the speech time (t now ). Therefore, it can refer to locations but also denote a contextually relevant modifier, obligatory adjunct or argument function". This is, of course, reminiscent of the different types of deictic reference identified with the 'locative' inversion constructions investigated here.

(57)
European Spanish Ha venido María. aux.3sg come.pst.ptcp María 'Maria has come [here].' However, in an inversion construction involving a Class B verb, such as Spanish Murió la abuela ('Grandma died'), the deictic interpretation is less clearly defined: here, the event happened recently according to the speaker's timescale, but its temporal coordinates are less exact than their spatial counterparts in (57). If we accept that languages may have multiple subject positions (Diesing 1992;Kiss 1996;McCloskey 1997;Cardinaletti 2004;Ledgeway 2010), we can plausibly assume that, given its subject-like properties, the null locative may surface in more than one structural position, where each position is associated with a (subtly) different deictic/pragmatic interpretation. A corollary assumption is that the different types of null locative surface in a position encoding a corresponding feature.
Thus we would expect that in European Portuguese, which has only one null locative (DeixisP), only one corresponding position should be available. Likewise two, three and four structural positions should be available respectively in River-Plate Spanish, European Spanish and Asturian, reflecting the number and types of null locatives lexicalised in each variety. The availability of the null locatives, and their predicted corresponding structural positions, yields the number of "preferences" observed per variety (cf. Tables 3-5). However, since the different positions in which the subject or null locatives could appear have not been tested here, the analysis remains speculative. Nevertheless, the following section captures these intuitions in formal terms.

A formal analysis
The traditional formulation of the EPP -which we have thus far assumed the null locative is capable of satisfying -requires a preverbal subject to surface in SpecTP. However, various authors (for instance, Rizzi & Shlonsky 2006 for English; Sheehan 2010 for Romance) have suggested that, in locative inversion constructions, the locative PP surfaces in SpecFinP (where Fin is the lowest available position within a Rizzian split CP). Thus we assume that: i) the null locative does not necessarily have to surface in SpecTP in order to satisfy the EPP, and ii) the C-domain has its own internal structure (Rizzi 1997;Speas & Tenny 2003;Sigurðsson 2010). Sigurðsson (2010: 161) proposes that Fin "splits into two separate (but commonly indistinguishable) head features, a temporal one and a locational one, that is, Speech Time and Speech Location, S T and S L (the basic now and here of the utterance)". In his account, there are context-linkers and speaker and hearer features encoded in the CP which are "inherent features of the syntactic speech event" (2010: 162). 28 Context-linking is understood to be a general notion that involves setting the values of argumental and adverbial topics according to elements in the deictic or discourse context; dedicated projections for topics are thus replaced by the more general CL N , which licenses context-sensitive semantic variables: Between CL N and Fin are features for speaker (or logophoric agent, Λ A ) and hearer (logophoric patient, Λ P ). The structure assumed for the CP is thus: We can extend our previously-established analogies (cf.  CL N corresponds to SourceP and Class C verbs due to these elements' associations with a weaker or more generalised deixis. Λ A relates to DeixisP and Class A verbs on account of their association with (interpretations involving) speaker orientation. Fin corresponds to GoalP and Class B verbs since Fin either encodes Speech Time or Location, which relates to the interpretations of the pertinent verbs, and these c-select GoalP. 30 Furthermore, the correspondences allow us to predict the structural position corresponding to LocationP and the (Class D) unergative verbs encoding static location (only). We deduce that this position is either a) T, a logical conclusion since an argument surfacing in T can satisfy the EPP (i.e. in SpecTP); or b) a projection internal to, or corresponding to, Fin that encodes Speech Location (S L ) only, also a logical conclusion since LocationP involves a deictic interpretation. We opt for the second possibility, as this allows us to maintain the crucial deictic interpretation for the VS order, and link it to an independently-motivated locative feature. This gives us the following structural positions (if the second hypothesis, viz. that LocationP corresponds to S L , is correct, we reason that, since the structure of LocationP is a subset of GoalP's structure, S T must be higher than S L in the functional structure, as is also suggested in As CL N appears very far to the left within the CP, we assume that it is too high for the movement-triggering feature belonging to T to associate itself with. The SV/VS alternation is then only dependent on whether a null locative is present in the numeration, in which case it moves to SpecTP to satisfy the EPP, leading to the impression of a so-called "neutral" preference (SV/VS). Since Λ A and S T , S L are lower, T's movement-triggering feature can more plausibly associate itself with these discourse features. In European Spanish, these features are presumably bundled on Fin. Therefore, there are more factors involved in the SV/VS alternation, which is, firstly, dependent on whether these features are present. If they are, then T's movement-triggering feature will associate itself with them; if not, then the EPP will cause any available null locative (i.e. SourceP, DeixisP or GoalP in the case of European Spanish) to fill the canonical subject position. The multiple factors involved give the impression of a VS "preference". We attribute the stronger inversion preference of DeixisP over GoalP (and related inversion verbs) due to DeixisP actually encoding both Λ A and S T /S L (appearing on Fin), whereas GoalP only encodes S T and S L . 32 Note that our analysis does not necessarily preclude an SV order from involving a deictic interpretation: as indicated by the results of our the empirical survey ( §3), the interpretation of a non-inversion structure is underspecified with respect to deixis rather than being marked by its absence, contrasting with inversion structures which have a [+deixis] reading. Translated into theoretical terms, these empirical findings suggest that a null locative PP could be selected by an inversion verb but would not necessarily have to involve an inverted order, for example if the null locative is subsequently attracted to a discourse feature high in the left-periphery (i.e. a position from which the EPP could not be satisfied, thereby necessitating the subject to fulfil the requirement by movement to SpecTP, resulting in an SV order).
Indeed, in European Portuguese, a "neutral" preference obtains since the discourse features are scattered in the CP, and the EPP is therefore separated from discourse effects. DeixisP values its features via a long-distance agreement operation and does not surface in a higher position since Λ A presumably lacks a movement-triggering feature of its own. Thus the null locative satisfies the EPP in the usual way. Asturian is assumed to show essentially the same distribution as European Spanish. Whether it is S L (appearing on Fin according to Sigurðsson 2010) or T that corresponds to the fourth position available in this language 32 This involves re-analysing GoalP as including a deictic feature. Under such an analysis, DeixisP and GoalP would have similar layered structures (i.e. both would have a deictic projection) but they would encode different types of deixis; namely, DeixisP would encode Λ A , S T and S L , whereas GoalP would encode only S T and S L . In turn, LocationP, encoding a subset of GoalP's internal structure, would have a deictic projection for S L . We would therefore predict that SourceP encodes the CL N feature, which could be valued via a long-distance agreement operation (Agree, in Minimalist terms), in which case the null locative could both satisfy the EPP by moving to the subject position and value the CL N feature from a distance. This account is preferable, since it allows us to account for the deictic interpretation associated with 'locative' wide-focus inversion in all instances. Additionally, it allows us to account for the different types of weaker, stronger and speaker-oriented deixis that have been observed in Ibero-Romance (cf. §4.3.1). On this account, we would assume that it is GoalP rather than DeixisP that is available in Portuguese and Italian, since GoalP now also encodes deictic features (which would make more sense given the interpretation of VS structures in Portuguese is always (loco-)temporal but not necessarily speaker-oriented).
In fact, there is reason to think that it is not CL N but Λ P that is encoded in SourceP. Whereas Sigurðsson (2004;)'s logophoric patient (i.e. addressee) is lower in the functional structure than the logophoric agent (i.e. speaker), recent work on the encoding of speech-act information in the left-periphery (e.g. Lam 2014; Heim et al. to appear) has suggested that addressee is in fact encoded higher than speaker. If so, then we might wish to readjust our analysis so that SourceP corresponds to the Λ P feature rather than CL N : this would then link the distal interpretation of structures involving Class C verbs to a specifically speakerdistal deictic feature, and form a more natural opposition with Class A structures as encoding specifically speaker-proximal deixis (ie. via Λ A in DeixisP). The non-speaker but nonetheless deictic interpretations of Class B and Class D structures (i.e. S T and S L with the former but S L only with the latter) would also follow from this analysis. changes little, since EPP satisfaction is already possible from both these positions anyway. River-Plate Spanish is assumed to pattern somewhere between European Spanish and European Portuguese; that is, we expect fewer syncretic features on Fin than in European Spanish, but less feature-scattering across the CP than in European Portuguese.
If the above discussion is on the right lines, then the questionable notion of "preference" receives a grammatical explanation, dependent on the types of null locative available in a given variety and where these can appear in the TP/CP border and left periphery. 33

Conclusions
The evidence presented here -preverbal subject positions, a basic SVO word order, infelicitous instances of wide-focus VS constructions -indicates that null-subject accounts appealing to the pronominal properties of rich verbal inflection do not make the correct predictions for Ibero-Romance. Our findings are thus better supported by null-subject analyses involving an empty pronoun pro. Nonetheless, the wide-focus inversion phenomena investigated here cannot be directly related to the null-subject parameter, since Brazilian Portuguese, a partial NSL, licences these constructions under the same general conditions as those of consistent NSLs (sc. the availability of null locatives), in line with a lexicon-oriented approach to parametric variation. The formal analysis put forward to capture the systematic complexity of Ibero-Romance wide-focus inversion provides an elegant account of the variation observed, unifying overt and covert locative constructions under a single analysis, and suggests the potential for development across wider comparative domains. However, further investigation is necessary to establish its crosslinguistic validity.