A relative clause analysis of event existential constructions in Aklanon

This study is the first to document and analyze the ‘event existential construction’ (EEC) in Aklanon (akl), an understudied Central Philippine language. In EECs, the apparent correlation between Philippine-type voice and nominal case appears to break down. However, I argue that the ‘case-cleaving effect’ of EECs is superficial. I argue for an analysis wherein the existential maj selects a relative clause as its complement and allows optional possessor raising of a relative-clause-external agent, which obligatorily controls a DP-internal PRO, to topic position, producing the case-cleaving effect. This study presents a typologically unusual type of obligatory DP control and contributes to a richer understanding of cross-linguistic variation in Philippine languages, while allowing us to maintain the robust generalization that Philippine-type voice is correlated with the argument role of the topic-marked nominal.


Introduction
In Aklanon (endonym Inakeanon [ʔinakɰaˈnon]), an understudied Central Philippine language (akl; Western Visayan), Philippine-type 'voice' affixes generally reflect the argument role of a single discourse-prominent argument (henceforth 'topic').For instance, in (1), the perfective patient voice morphology on the verb indicates that the ro-marked argument is construed as the patient of the giving event: (1) Canonical Aklanon Patient Voice Clause gin-taʔó ʔit maʔéstra sa ʔuŋáʔ ro li ́bro.pv.pfv-give unm teacher dat child top book 'A teacher gave the book to the child.'However, this robust correlation between case and voice breaks down in so-called 'event existential constructions' (EECs).1 EECs seem to 'cleave' case and voice apart: the ro-marked argument is the semantic agent, despite the patient voice morphology on the verb, as in (2). (2) Aklanon Patient Voice EEC maj gin-taʔó sa ʔuŋáʔ ro maʔéstra.exist pv.pfv-give dat child top teacher 'The teacher gave something to the child.' (EEC) In this paper, I will argue that the case-cleaving effect of EECs in Aklanon is superficial and dissolves under an analysis of EECs in which: (i) the complement of maj is a relative clause with an underlying genitive agent, (ii) the underlying genitive agent is generated external to the relative clause and obligatorily controls a relative-clause-internal PRO, and (iii) the existential maj allows optional raising of the genitive agent to a higher position where it receives its ro-marking and produces the superficial case-cleaving effect.This paper is structured as follows: in Section 2, I will briefly describe the linguistic context of Aklanon, as well as some defining characteristics of Philippine-type voice and the distribution of the maj existential in Aklanon.This baseline understanding of maj will help inform the analysis of EECs in Section 3. I begin Section 3 by summarizing previous analyses of EECs, and then present the current analysis, divided into the three main claims and subdivided into the main arguments for each claim.Section 4 concludes.
The data throughout this paper has been collected via elicitation with one native, fluent speaker of Aklanon in the presence of her partner, a non-native, fluent speaker of the language.
The consultant has lived in the United States for several decades, and she speaks Aklanon regularly.
The Aklanon consultant and I have conducted regular elicitations since 2018, and during the pandemic, this has been done telephonically.To control for some intraspeaker variation, data was only included in this paper if it was successfully re-elicited during two (or more) elicitation sessions separated by significant periods of time (one or more weeks).Because the data in this paper comes from a single speaker, the claims made herein should be understood as describing a single Aklanon grammar that exists; future research will address the extent to which the findings presented in this paper generalize to the grammars of other Aklanon speakers.

Background
Aklanon is a Western Visayan Central Philippine language (Zorc & de la Cruz 1968) spoken by ±550,000 people in the province of Aklan on the island of Panay (Ethnologue 2021).The only formal work that has been done on Aklanon is descriptive (e.g.Zorc & de la Cruz 1968), lexicographic (e.g.Zorc et al. 1969), and historical.Zorc & de la Cruz's (1968) very thorough descriptive grammar has been foundational for the work presented in this paper.

Philippine 'voice'
The so-called 'Philippine-type voice system' or 'Philippine alignment' has been a rich and active field of research for many decades (Schachter 1976, passim;Shibatani 1988;Kroeger 1993;Richards 2000;Rackowski 2002;Aldridge 2004 i.a.;Rackowski & Richards 2005;Latrouite 2011;Chen 2017;2020).The essential generalization about Philippine 'voice' systems is that verbal morphology indicates the argument role of a single A'-extractable argument, known by convention as the 'subject,' 'pivot,' or 'topic' (Chen 2017:21-24).Although this system is widespread in Philippine and Formosan languages (among others, e.g.Malagasy), a long tradition of theoretical work on Philippine-type voice has been conducted on Tagalog, which is related to Aklanon. 2   Therefore, I will briefly compare Tagalog and Aklanon to show how these systems overlap in some ways, but not in others.
Tagalog and Aklanon belong to separate primary branches of the Central Philippine language family (Zorc 1986).Tagalog is typically described as having four 'voices': agent, patient, locative, and circumstantial.This is shown by Chen (2017:22, ex. 8): 3   (3) a. Tagalog Agent Voice Clause (Chen 2017:22, ex. 8) b<um>ili ang babae ng tela mula sa tindera para sa nanay.<av>buy pivot woman id.y cloth from df.y shopkeep p df.y mother 'The woman bought cloth from the shopkeeper for mother.' 2 Some notable work has been done on Malagasy (e.g.Pearson 2001;2005).Aldridge (2004) also includes Seediq (Northern Formosan, Taiwan) data, and Chen (2017) includes data from Amis (East Formosan, Taiwan), Puyuma (Formosan, Taiwan), as well as Seediq and Tagalog. 3 Glosses in (3) have not been modified from the original source; id is 'indefinite' and df is 'definite'; the use of pivot, x, y are explained as follows: 'The label "Pivot" indicates that the phrase is eligible to under go A'-extraction (relativization or pseudo-clefting); the label X stands for the marking on non-Pivot external arguments; the label Y stands for the marking on non-Pivot internal arguments and obliques.'(Chen 2017:16) b.Tagalog Patient Voice Clause (Chen 2017:22, ex Aklanon also distinguishes four 'voices,' but where Tagalog uses ng to mark non-topic arguments, Aklanon has two possible choices: ʔit (unm) or ko (obl).In my data, ʔit and ko alternate among non-topic arguments with a definiteness effect: ʔit is construed as indefinite, and ko, as definite.
For reference, a canonical paradigm using bakáe 'buy' in four 'voices' is shown below.4) is given in the orthography used by Zorc & de la Cruz (1968), where ‹e› represents [ɰ].
Additionally, the source gives these in a marked word order.In my data, clause-initial topics are consistently construed as focused.Morpheme boundaries and glosses are my own.
In languages for which Philippine 'voice' has been analyzed, such as Tagalog, it has been analyzed as vP phase-edge Agreement between T and the topic (Rackowski & Richards 2005), inherent ergative/structural absolutive case (Aldridge 2004;2006;2009), and topic agreement with C˚ (Chen 2017;2020).The extent to which such systems resemble canonical active/passive-type voices in Indo-European languages is an area of active research (cf.

references above).
The arguments for my analysis of EECs are not contingent on a particular analysis of Philippine-type 'voice,' and are therefore compatible with any of the aforementioned analyses.I opt to use conventionalized terminology and glosses (topic, unmarked/oblique, locative/dative), and in doing so, I do not intend to assume any particular view of case/voice in Aklanon.Table 1 shows the distribution of Aklanon case particles by argument role across all voices.Note that for each voice, the left column contains case markers for common nouns, the right column contains equivalent markers used exclusively for proper names and some terms of address.The topic markers are bolded in each column.

Existential maj
This section provides a baseline description of the distribution of maj in Aklanon.Zorc & de la Cruz (1968:215) describe three functions of this existential: (i) existential statements, (ii) indefinite statements, and (iii) statements of possession.This distribution roughly matches that of Tagalog ‹may› described by Sabbagh (2009) Third, maj requires its complement to be bare (7) -that is, the complement of maj cannot be ʔit/ko-marked, and changing the word order does not ameliorate this ungrammaticality. 6 Kaufman (2011:729) argues that Tagalog may can be reconstructed as an agent voice existential *k<um>a-i from Proto-Austronesian *ka exist + <um> av + -i det, supported by the fact that may assigns topic case to the possessor (in contrast to the Tagalog exclamative existential kay which assigns oblique case). 6 The availability of possessor raising to maj is important for the present analysis of EECs in Aklanon and will be discussed at length in Section 3.2.2.

Event existential constructions
The term 'event existential construction' was applied by Aldridge (2011) to refer to a 'type of existential [which] involves embedding of a clausal complement under an existential verb' (Aldridge 2011:1) EECs in both languages possess at least two defining characteristics.The first is case-cleaving, whereby the topic-marking on the semantic agent does not correlate with the voice of the embedded verb (otherwise 'the woman' in both examples above should be the patient of the giving event).The second is that the topic is obligatorily interpreted as an agent rather than a clausal possessor, i.e. (11) does not have the meaning 'the woman has a book that was bought (by someone else).'A further defining function of EECs is that they are the primary means to introduce indefinite, nonspecific arguments of verbs, such as 'some book' into discourse.This is shown in (12), which introduces an indefinite, nonspecific patient.Aklanon has no lexicalized indefinites, such as 'some,' 'someone,' or 'something.' 7   (12)  Aklanon Patient Voice EEC maj [gin-taʔó sa ʔuŋáʔ] ro maʔéstra.exist pv.pfv-give dat child top teacher 'The teacher gave something to the child.' (EEC) I will address each of these characteristics in the discussion to come. 8

Previous analyses of EECs
Very limited research has been done specifically on event existential constructions in Philippine languages.Schachter & Otanes (1972:276-280) present a meticulous description of this construction in Tagalog.Zorc & de la Cruz (1968:215, ex. 2.2c) provide an example of an Aklanon EEC in their discussion of 'indefinite statements,' but EECs are not otherwise described as a distinct construction therein.The first theoretical analysis of Tagalog EECs (under the name 'impersonal construction') was developed by Law (2010), followed by Aldridge (2011) andAdar (2013).
Any theoretical account of EECs must answer the following questions: (13) a.What assigns case to the topic-marked argument (i.e. the agent)?b.What is the complement of maj? c.How is the topic construed as the agent of the embedded verb?d.How does the EEC introduce indefinite, nonspecific arguments like in (12)?
In this section, I summarize the main arguments of the two main analyses of Tagalog EECs.

Law's (2010) account: Relative clauses
The analytical claim of Law (2010) is that the complement of Tagalog ‹may› in EECs is a relative clause and the topic-marked argument receives its case from the existential itself, not the  12) are of the form 'The teacher gave something …' or 'There is something that was given by the teacher…' For indefinite, nonspecific arguments, like 'some book,' her responses have more variation, including but not limited to: 'some book was given by the teacher…,' 'there is a book that was given…,' 'there is some book that was given…,' or 'the teacher gave a book…' If given, I opt for a translation without the existential frame (i.e.'The teacher gave…' instead of 'There is some X that the teacher gave…'), and if my consultant alternates between 'a' and 'some,' then I give both of these in my translations of the EEC.embedded verb.On this view, the nominal functions of ‹may› (existential, locative, possessive) can all be unified with the EECs, because the complement of ‹may› is always nominal -it is just that the nominal complement in EECs is a relative clause.To account for the function of EECs to introduce indefinite, nonspecific arguments like 'something, ' Law (2010) proposes that there is an indefinite, null, external head of the relative clause: 9 ( 14) ang guro.exist cooked.pvtop teacher 'The teacher has something that was cooked.' (Law 2010, ex. 45a) Law (2010) argues for this analysis using facts about pluralization, adverbs, extraction, and relative clause 'stacking.'For instance, he argues that the proposed null nominal head accounts for the fact that the seemingly verbal complement of ‹may› can be pluralized: ang guro.exist two-lk cooked.pvtop teacher 'The teacher cooked two things.'(Law 2010, ex. 46a) Law (2010) does not explain how the topic-marked argument is construed as the agent of the relative clause.On this point, he comments: 'I thus have no explanation for why ang guro 'the teacher ' [in (14)] is necessarily understood to be the Actor of the passive verb niluto 'cook', even though it is not the syntactic argument of the verb' (Law 2010:315, fn. 13).Aldridge (2011) argues against Law (2010) on two grounds.First, Aldridge (2011)  By virtue of the Complex-NP Island Constraint, the relative clause analysis predicts that argument extraction should be ungrammatical.This data is crucial for Aldridge (2011), who argues on the basis of ( 16) that 'the relative clause analysis must be rejected, because it cannot account for the lack of island effects in extraction from event existentials' (Aldridge 2011: 3). 9 Both Law (2010) and Aldridge (2011) treat the existentials ‹may› and ‹mayroon› as equivalent, but Law (2010:313) notes that they have slightly different syntactic behaviors.For instance, only ‹mayroon› allows wh-extraction from its complement.The same difference is found in Aklanon maj vs. maj ʔúnaʔ, where ʔúnaʔ is a demonstrative.I leave maj ʔúnaʔ for future work and focus my analysis only on the properties of Aklanon maj.

Aldridge's (2011) rebuttal: Raising from vP
Aldridge's second counterargument to Law (2010) is that the relative clause structure cannot derive the fact that what is asserted to exist in EECs is an event, not just an individual: 'the translations given by my consultants […] all have the event interpretation, not the nominal interpretation that Law gives' (Aldridge 2011). 10  Instead, Aldridge relates Tagalog EECs to 'modal existential constructions' (Šimík 2011, cf. also Šimík 2017; 2019) and proposes 'that the existential verb in Tagalog embeds a vP.The external argument raises to the edge of matrix vP' (ibid: 7).The exact structure that Aldridge proposes is in (17). 11Note that the DP [Abs] argument (i.e. the topic) raises out of the embedded vP complement and lands in external argument position of the higher vP, i.e. of ‹may› 'exist.'That is how the topic-marked argument receives its interpretation as the agent of the embedded clause.
Then, the embedded vP complement raises to the edge of the higher vP and the existential moves to T as a standard consequence of deriving V1 syntax.( 17) Aldridge's analysis draws on Côté (1999)'s work on an existential construction in Québec French.However, Adar (2013) argues that the defining characteristics of the existential demonstrated by Côté (1999) for Québec French cannot be shown for Tagalog.For instance, in contrast to Québec French EECs, Adar claims definite nominals (e.g. proper names) cannot be introduced in the complement of ‹may› in Tagalog, and also in contrast to Quebec French EECs, Tagalog EECs can be formed with individual-level predicates (Adar 2013:6, exs. 12b, 13).Adar thus argues that the Tagalog construction is not an event-introducing existential like the Québec French case.Adar (2013) speculates that the Tagalog EEC may be a type of pseudorelative (Cinque 1992).However, Adar (2013) does not propose a specific structure for Tagalog EECs, although he does address the grammaticality of extraction in ( 16) by appealing to Truswell's Single Event Condition (Truswell 2010;2011).I will continue to use Aldridge's (2011) term 'event existential construction (EEC)' despite Adar's (2013) objections. 11Movement arrows and boxes are my own addition.
In summary, for Aldridge (2011), the complement of ‹may› is a vP, and the topic-marked argument gets its topic-marking from ‹may›.The reason why the topic is construed as the agent of the embedded complement is that it raises directly from the embedded vP's external argument position to the external argument position of ‹may›.This raising operation thus produces on case-cleaving effect.However, while Aldridge (2011) strongly motivates her account with the Complex-NP Island Constraint, she does not attempt to account for Law's (2010) pluralization facts or the function of EECs to introduce indefinite, nonspecific arguments.I will show that Aldridge's analysis cannot be directly applied to Aklanon EECs.

The present analysis
I propose the following analysis of Aklanon EECs: (18) a.The complement of maj is a relative clause (a la Law 2010), which can be headless or headed, not a vP (contra Aldridge 2011).b.Case-cleaving arises because the topic-marked argument raises to its position from a DP-internal position which I label PossP; it is assigned topic by maj.c.The topic is construed as agent via control of a relative-clause-internal PRO.d.Clause-initial syntax is derived by VP-remnant raising of maj and its complement.
To preview my analysis, an EEC like ( 2 On the present analysis, the complement of maj is a (headless) relative clause, the topic-marked argument is base-generated in PossP and raises to a higher position, where it is assigned topic by maj, and is construed as the agent of this relative clause via control, in contrast to Aldridge (2011), who posits raising directly from the embedded external argument position.On the present analysis, the case-cleaving effect thus becomes a superficial consequence of the argument structure of maj in EECs.
The remainder of this section is divided according to three main claims: first, that the complement of maj is a relative clause, second that the agent originates as a possessor outside of the relative clause and raises to topic of maj, and third, that the topic controls a relative-clauseinternal PRO.

Claim 1: Complement of maj is a relative clause
This section will focus on arguments for that the complement of maj is a relative clause: The relative clause analysis captures that fact that relative clauses internal to EECs pattern together with relatives throughout the language. 12I propose to use the same structure to account for both.In the headless case, there is no overt head: In the headed case, the overt head occupies the NP position in the structure above.I assume that the linker ŋa is a head in C˚. (25) Structure of (23b) before V1 Movement top teacher exist book lk pv.pfv-give dat child 'The teacher gave a/some book to the child.'By contrast, for Tagalog, Aldridge (2011) proposes that the complement of ‹may› is an embedded vP in EECs.Her analysis mentions, but does not include, headed relative clauses like (23b), and would require modification to allow an attachment site for an external head of the relative clause.Consider again Aldridge's proposed analysis, where the vP complement of ‹may› is boxed.( 26) exist pv.pfv-give lk book dat child top teacher 'The teacher gave a/some book to the child.' (EEC) In (a) above, the head of the relative clause appears before the predicate.In (b), the head appears after the predicate.Aldridge (2003) analyzes these configurations at length in Tagalog, and argues that the structure analogous to (a) constitutes a head-external relative clause, and (b) constitutes a head-internal relative clause.In her treatment of EECs, however, Aldridge (2011) only includes EECs that are of the head-internal type and does not discuss the head-external case in (a).While noting that the Aklanon also appears to have head-external and head-internal relative clauses, and that this alternation obtains in EECs, I will leave more detailed analysis of the two relative clause types to future research.
On this structure, the complement of the existential is a vP, which contains an AspP projection that has moved to the outer specifier of vP. 13 The linker occupies v, and the internal argument occupies the boxed DP in SpecAgrP.Now try to apply this structure to the headed relative clause from Aklanon, repeated below: ( This data poses a problem for the structure in ( 26), because it requires that the head noun (the internal argument) líbro 'book' be attached to the left edge of vP or AspP, rather than its base-generated position in SpecAgrP, to derive the correct word order.How exactly to motivate this movement to the left edge of the phrase is unclear, and it is also unclear how to interpret the noun if it were in such a position.Finally, this vP analysis results in two undesirable outcomes: first, it obscures the fact that, in Aklanon, the headless vs. headed alternation found in EECs patterns together with relative clauses throughout the language; second, it would require that maj sometimes take a vP as its complement (i.e.headless relatives), or take a DP/NP or a vP with a nominal at its left edge in other cases (i.e.headed relatives).
By contrast, the relative clause analysis has explanatory power because it unifies the observed overlap between relative clauses in the complement position of maj with those that occur in other positions in the language.This analysis also accommodates the presence of nominal heads without requiring additional stipulations.On this analysis, maj always takes a bare nominal complement: if just a bare noun, then it expresses existence or clausal possession; if a bare relative clause, then it produces an EEC interpretation.
To summarize this argument, the fact that the complement of maj in Aklanon permits an overt nominal head motivates an analysis wherein the complement of maj is a relative clause, not an embedded vP.

Aklanon prohibits argument extraction from EECs
Recall that Aldridge (2011) strongly motivates her embedded-vP analysis of Tagalog EECs using the extraction fact in (28).The logic of her argument is: if the complement of ‹may› were a relative clause as Law (2010) argues, extraction should be ungrammatical by virtue of the Complex NP-Island Constraint; because extraction is grammatical in Tagalog, Aldridge (2011) argues that Law's (2010) analysis does not hold water. 13 Aldridge (2011) motivates this phrasal movement to account for certain word order facts in Tagalog.(28) Tagalog (Aldridge 2011:3, ex

Modifiers to the relative clause
This section shows that Law's (2010)  The grammaticality of ( 39) is difficult to reconcile with an embedded vP analysis because on such an analysis, the complement of maj would be vP or TP/AspP, and would therefore lack a nominal projection for maŋa to adjoin to.However, on the relative clause analysis, prenominal modifiers such as maŋa receive straightforward treatment as modifiers to the relative clause.In the structure below, I assume maŋa is an adjunct to NP, but this analysis is compatible with other theoretical assumptions, i.e. that maŋa may be in NumP or another In summary, pre-and post-nominal adjuncts reveal an asymmetry between the complement of maj and matrix clauses.This data has shown that these adjuncts are systematically available in complement position of maj, even though they cannot modify matrix verbal predicates.On the present analysis, this is because the complement of maj is a relative clause.

Constraints on long-distance relativization
Aklanon exhibits the 'subject-only constraint' (Schachter 1976) that is well-attested in other Austronesian languages (Clemens & Polinsky 2017).This restriction refers to the fact that only topics (on my terminology) are accessible for A'-extraction, such as relativization and pseudoclefting.Thus (48a) below shows the baseline to which we can compare the relative clause in (48b), where the relativized internal argument is grammatical if the verb is in patient voice, but not if the verb is in agent voice (48c).( 48 The subjects-only constraint applies also to long-distance A'-extractions, a pattern which has been described by Hsieh (2020)  The restriction imposed by the Subjects-Only and Matrix Verb Constraints makes a straightforward prediction for EECs: if the complement of maj is a relative clause, then it should be sensitive to them.The examples below show that this turns out to be the case.In (52a), both verbs in the relative clause are in patient voice, construed as a single chain of extraction with do ʔuŋáʔ 'the child' as the agent of both embedded predicates.But in (52b,c) it is not possible for there to be a voice mismatch between either the higher or lower verb in the relative clause.In the latter cases, the actual interpretation of such mismatches is that the higher predicate embeds the lower predicate, not that they form a single chain of extraction with a shared argument (constituency of embedded predicates bracketed). 15The present analysis captures these facts about long-distance extraction if we assume that there is successive-cyclic operator movement in the relative clause (52a).The simplified diagram below shows this proposal.The operator originates in the complement position of the lowest verb, kíwaʔ 'cut' and is first extracted to the outer specifier of vP.This corresponds to the pv morphology on the lower verb.The operator then extracts from the edge of the complement of the verb hínjoʔ 15 do is an allomorph of ro appearing after syllable-final /n/.
'request,' which corresponds to pv morphology on the higher verb and thus obeys Hsieh's (2020) Matrix Verb Constraint.The operator lands at the highest edge of the relative clause CP. (53) Simplified Structure of (52a) before V1 Movement This constitutes the fourth and final argument that the complement of maj in EECs is a relative clause.To summarize all four arguments briefly: I have shown that the complement of maj in Aklanon EECs patterns together with relative clauses throughout the language.The complement of maj shows headless/headed alternations, prohibits extraction, allows pre-and post-nominal modification to headless and headed relative clauses, and shows constraints on long-distance relativization that is consistent with the subjects-only constraint on A'-extraction more generally in the language.The next sections relate these arguments to claims about the argument structure of maj in order to derive the agentive reading of the topic.

Claim 2: Agent raises to topic of maj
Having established that the complement of maj is a relative clause in Aklanon, I will now turn to how exactly the topic-marked argument of maj (i.e. the agent of the relative clause) gets its topic-marking.In this section, I will develop an argument that the topic-marked agent of maj has raised to topic position from a base position as a genitive agent (i.e.possessor) of the relative clause: Recall from Section 2.2 that in possessive existential constructions, the possessor is licensed either as topic or genitive, shown in example (9) repeated below: Also recall that [gen NP lk NP], i.e. ʔákon ŋa kwárta in (9b), is a standard prenominal possessive phrase, 'my money.'This is relevant to the relative clause analysis of EECs because agents of non-av relative clauses can be expressed as genitive 'possessors' in Aklanon. 16The examples (55a-c) below show ʔána '3sg' as a genitive agent, and example (55d) shows, by contrast, that a genitive agent is not grammatical with an av relative clause. 16In Tagalog, the same has been discussed in detail by Hsieh (2020:157-72), who calls this construction 'genitive inversion.'This construction, at first glance, also shows structural similarities to so-called 'genitive relatives' that have been described in Polynesian, see Herd et al. (2011)  Importantly, the genitive paradigm, found as prenominal possessors and preposed agents of non-av relative clauses, are not licensed as direct arguments of verbs.The minimal pair below shows that the argument form of the pronoun has some additional morpheme, n-.I refer to the n-class of pronouns as 'postposed.' See Some evidence for this constituency comes from plural maŋa as an adjunct below the agent: (59) ʔána ŋa maŋa bákɰ-an 3sg.gen lk pl buy-lv.fut'(the places) where s/he will buy from' With these facts about genitive agents of relative clauses in mind, I would like now to return to EECs, to show that they also hold true for the complement of maj.In EECs, preposed genitive agents are only possible when the complement of maj is a non-av relative clause, a pattern which exactly parallels the relative clauses in ( 55 The examples (60a-c) show that non-av relative clauses with preposed genitive agents are grammatical as the complement of maj, while av relative clauses with genitive agents are ungrammatical (60d).Not only is this further evidence that maj takes a relative clause as a complement -it also explains one of the defining features of EECs shown at the outset of this paper: the case-cleaving effect.
On the present analysis, I argue that case-cleaving arises from a combination of factors: First, if the relative clause is not agent voice, a genitive agent is possible, and second, maj allows optional raising of genitive agents (or possessors) to a higher topic position (recall example 9).To capture the facts presented throughout this section, I apply the structure in (58) to EECs.For EECs with preposed genitive agents, I propose that the agent is base-generated in PossP.This structure is modeled below using (60a) as an example (see Section 3.2.3 for discussion of the relative clause structure and how the external agent receives its interpretation as agent of the relative clause; see 78 for the full structure of 62).( 62 I use the same structure to derive EECs with case-cleaving, but in these cases, the preposed agent in SpecPossP is possessor-raised to a higher position, such as the external specifier of vP or a higher projection.I remain neutral on the landing site of the raised agent, and, while I assume that this raising operation is optional and is unique to the predicate maj, I also do not currently have an explanation as to why maj allows optional possessor raising in the first place. 17(63) Simplified Structure of (61a) before V1 Movement top M. exist buy-pv.fut'Maria will buy something.' 17An anonymous reviewer suggests that possessor raising in the existential constructions receives a straightforward treatment if we assume that topic-marking is actually nominative case: on this assumption, the existential predicate still has an unvalued nominative case feature on C or T, which can be optionally valued by the genitive agent.This hypothesis will be considered in future work.
On this analysis, the selectional criteria of maj is the same across all existential constructions.In all cases, maj takes a bare nominal as its complement.However, what is unique about EECs is that maj takes a relative clause as its complement, and the argument structure of the relative clause feeds the argument structure of maj: non-av relative clauses in Aklanon generally allow preposed genitive agents that resemble external possessors, and these genitive agents can optionally undergo possessor-raising by maj, producing the case-cleaving effect.

Claim 3: Topic obligatorily controls a relative-clause-internal PRO
At this point, two main claims have been established: the complement of maj is a relative clause, and case-cleaving arises because the maj predicate allows possessor raising of the genitive agent to topic position.This final section ties these claims together to derive the agentive reading of the agent, which I have argued is not a direct argument of the relativized verb.Accomplishing this portends the perennial debate between control versus raising, and the option space of analyses contains at least a few viable hypotheses:  For now, I take the evidence in (67) to show that EECs satisfy Sundaresan's (2014) criterion (65ii), which motivates an OC PRO in external argument position of the relative clause in EECs.

Sloppy readings under ellipsis
The next diagnostic for control which can be readily shown for Aklanon is that under VP ellipsis, only sloppy readings of PRO are available.That is, in (69) below, the boxed clause can only be understood to mean 'Raul also cooked something' (sloppy reading), and not 'Raul is also such that Jose cooked something' (strict reading).& so also top R. 'Jose x cooked something and so did Raul y [PRO *x/y cook something].' (EEC) This is especially striking in contexts like (70), where the antecedent clause contains a possessive phrase which permits both a strict and a sloppy reading under ellipsis, even though only the sloppy reading is available to the agent of the elided clause.
(70) maj gin-taʔó sa ʔána=ŋ ʔasáwa si Gloría exist pv.pfv-give dat 3sg.gen=lk spouse top G. 'Gloria x gave something to her x partner,' ʔag maw man si María.and so also top M. 'and so too did Maria y [PRO *x/y give something to her x/y partner].'(EEC) The availability of only the sloppy reading of the external argument under ellipsis satisfies Sundaresan's ( 2014) criterion (65i) and thus provides further evidence for an OC analysis of Aklanon EECs.

Antecedent must c-command PRO
A familiar constraint on PRO is that its controller must c-command it, and this can also be shown in Aklanon EECs.In (71), for instance, the agent of the relative clause cannot be understood as This data shows that control in the EEC patterns in a way that is familiar from 'classical' types of control.

PRO alternates with overt DPs
The final observation that is crucial to understanding the issue of control in EECs is to understand that it is possible for overt external arguments to be expressed in the relative clause with canonical case marking.The relative clause analysis actually predicts this, since there is, in principle, nothing that blocks an external argument from merging as an argument of the relative clause.Example (72) shows three configurations (note that all three express clausal possession): in (72a), the possessor is genitive and the relative clause has a canonically-marked external argument; in (72b), the genitive possessor has undergone optional possessor-raising to topic and the relative clause has a canonically marked external argument; in (72c), the possessor has undergone optional possessor-raising to topic like (72b), but in this case, the agent of the relative clause is genitive.Note again that in (72), the topics are not construed as the agent of the relative clause, only as possessors.There is thus a cross-linguistic precedent for OC PRO to alternate with DPs.However, key differences between Aklanon EECs and the data presented in (75, 76) are that the relative clauses in EECs appear to be finite and unconstrained for tense-aspect and that this type of control involves control inside of DP.
Landau (2013, Section 5.6.2) discusses some variation that has been noted with respect to control into DPs and whether this involves pro or PRO.He states, 'A fundamental descriptive question is whether the null subject of nominalizations ever displays the strict referential dependence that OC PRO does.It turns out that the answer is not straightforward; sometimes it does, sometimes it does not' (Landau 2013:209).He then cites a study which shows four types of verbs in Catalan, Spanish, and Italian, some of which induce OC into their nominal complement, and some of which induce NOC, or a combination of both.Consider the following Catalan example of dedicar-se 'dedicate oneself to,' which induces OC into its nominal complement according to the author of that study: (77) Catalan OC in Nominal Complement (Landau 2013 Sichel (2010) argues against positing OC PRO in DP control contexts, opting instead for a coreferential relationship with pro.At present, resolving the debate between N/OC PRO or pro in DPs is well beyond the scope of this article, and further work will need to be done to explore how closely Aklanon fits into the typology of DP control in other languages.For now, however, based on the diagnostics presented in this section, I posit obligatory control of PRO in external argument position of the relative clause.I speculate that the presence of OC PRO in relative clauses is related to whether or not they have a genitive agent.That is, relative clauses which allow a genitive agent require OC of PRO into the relative clause by the agent.However, relative clauses which disallow a genitive agent will not have a PRO in external argument position (or at least, this analysis does not claim that, though it is possible).
To demonstrate the full analysis, the structure of an EEC with a preposed agent ( 62 For an EEC with case-cleaving, the same structure is proposed, but with the additional possessor raising of the agent to topic position of maj.Again, I remain neutral as to the landing site of the raised topic.In the structure below, I have not shown an additional step of VP-movement which yields the predicate-initial surface word order of ( 79 The analytical claim is thus that all Aklanon EECs, whether with or without case-cleaving, have the underlying structure with a genitive agent that controls into the relative clause.Those with case-cleaving have undergone an additional step of possessor raising, which the predicate maj allows optionally.
Obligatory control into a finite relative clause is certainly not theoretically 'classical,' especially not with alternations between PRO and an overt non-coreferential DP in agent position.However, this typologically unorthodox construction sheds light on our cross-linguistic understanding of what is possible with control, and future work on EECs in Philippine languages has the potential to reveal cross-linguistic connections, whether internal to the Philippine language family (e.g.

Conclusion
In this paper, I have proposed an analysis of event existential constructions in Aklanon in which the complement of maj is a relative clause.This was shown using data that reveals that (i) the complement of maj patterns together with other relative clauses throughout the language, (ii) is an island for extraction, (iii) can host nominal adjuncts, and (iv) is sensitive to the Subjects-Only and Matrix Verb Constraints.From this central claim, I presented data that showed both that maj allows optional possessor raising of a preposed genitive possessor, and that Aklanon non-av relative clauses generally allow the agent to be expressed as a preposed genitive 'possessor,' which then feeds the argument structure of maj and results in case-cleaving.Finally, I argued on the basis of several diagnostics that the topic-marked agent gets its agentive interpretation via obligatory control into the relative clause, a claim which merits further attention and analysis in the future, especially with respect to how this form of obligatory control patterns in a broader typology of control.
As Law (2010) argued for Tagalog, this analysis of Aklanon allows for a unified analysis of existential maj in Aklanon.In all cases, maj selects a bare nominal as its complement.However, in event existential contexts, this nominal is a relative clause.At first glance, EECs in Aklanon appear to violate a core generalization about Philippine-type languages -namely that nominal case is closely correlated with verbal 'voice.'However, as a result of this study, I have shown that this apparent case-cleaving effect is superficial.The relative clause analysis of EECs allows us to maintain this robust generalization about Philippine-type voice systems.This analysis also reveals a typologically rare instantiation of obligatory control within the DP domain, which merits more attention in future research.Furthermore, this is the first theoretical analysis which brings Aklanon data to bear on issues surrounding Philippine-type voice, control, and existential constructions.The introduction of Aklanon into this body of literature allows for a more nuanced understanding of EECs and of cross-linguistic variation within Philippine languages.

Preposed
{ʔit/ko kwárta} {ro maʔéstra}.exist unm/obl money top teacher Intended: 'The teacher has money.'Fourth, the possessive existential construction permits possessor raising.Compare the alternations in (8a,b) and (9a,b).In each case, example (a) shows a topic-marked (raised) possessor and (b) shows the minimal pair with a prenominal genitive possessor.In the (b) examples, note that the phrase [gen NP lk NP] is a standard prenominal possessive phrase in Aklanon.
ŋa kiwáʔ-on] do 16 ʔuŋáʔ.exist pv.pfv-request lk slice-pv top child 'The child [asked to slice something].' (EEC) b. #maj nag-hínjoʔ [ŋa kiwáʔ-on do ʔuŋáʔ].exist av.pfv-request lk slice-pv top child Intended: 'The child asked to slice something.'Actual: 'Somebody asked [that the child be sliced].'c. #maj gin-hínjoʔ [ŋa mag-ki ́waʔ ro ʔuŋáʔ].exist pv.pfv-request lk av-slice top child Intended: 'The child asked to slice something.'Actual: 'Somebody was asked (for permission) [that the child could cut (something)].' (Raised) Possession by a Pronoun maj kwárta akó.exist money 1sg.top 'I have money.'b.Internal Possession by a Pronoun maj ʔákon ŋa kwárta.exist 1sg.gen lk money 'I have money.' genitive agent raises to topic of maj from PossP, and from either position obligatorily controls PRO in the external argument position of the relative clause.Hypothesis 1b: The genitive agent raises to topic of maj from PossP, and from either position non-obligatorily controls PRO in the external argument position of the relative clause.Hypothesis 2: The genitive agent raises to topic of maj from PossP and corefers with pro in the external argument position of the relative clause.
Tagalog 'genitive inversions' in Hsieh 2020), in other branches of the Austronesian family (e.g.Herd et al.'s 2011 analysis of 'genitive relatives' in Niuean and other Polynesian languages), or beyond (e.g.languages highlighted in Landau's 2013 study).
-buy id.x woman id.y cloth from df.y shopkeep pivot mother 'The woman will buy cloth from the shopkeeper for mother.'

Table 1 :
. I give some examples of the Aklanon existential Aklanon Case Marking Paradigms for All Voices.Maj is unlike verbal predicates in Aklanon in at least four respects.First, it does not inflect for voice; 5 second, (6) shows that maj prohibits argument scrambling.If scrambling were permitted, we would expect that the order of the bare noun and the topic should be able to alternate freely.
1.1 Overt heads of the relative clause are possible Maria likes the book that the child gave to the teacher.'Theprediction for the relative clause analysis of EECs is that the same alternation between headless and headed relative clauses should obtain in the complement position of maj.It does, PossP t i [ NP ][ CP Op k [ vP PRO i gin-taʔó t k exist book lk pv.pfv-give dat child top teacher 'The teacher gave a/some book to the child.' (EEC) Aldridge's (2011) lk book dat child top teacher 'The teacher gave a book to the child.' (EEC) And the ungrammaticality of (29) is not due to a general prohibition against dative arguments or adjuncts being fronted, since that too is grammatical (in non-EEC clauses) -(31) shows frontingAldridge's (2011)argument against Law's (2010) analysis of Tagalog EECs as relative clauses thus does not carry over to Aklanon, and the two languages seem to differ significantly here.However, the ungrammaticality in (29) is not surprising if we consider data like (33), which shows that headless relative clauses in Aklanon generally do not permit arguments to scramble out of them.In (33b,c), for instance, it is ungrammatical for the dative argument of 'give' to be scrambled left of the headless relative.Taken together, the facts above present a strong argument in favor of the relative clause analysis of Aklanon EECs: the reason why argument extraction from EECs is ungrammatical in Aklanon is, in fact, due to the Complex-NP Island Constraint.
evidence from pluralization and other prenominal modifiers can be replicated in Aklanon, in addition to novel evidence from postnominal modifiers.Like Tagalog, Aklanon maŋa is a prenominal plural morpheme used as follows: PossP t i [ NP [ XP maŋa][ NP li ́bro top teacher exist pl book [ CP Op k [ C ŋa [ vP PRO i gin-taʔó t kThe prediction for EECs is that adjuncts like tuŋód sa Akɰán 'about Aklan' should also be able to modify the head of the relative clause.In EECs with an overt head, this is borne out; compare LK PV.PFV-give DAT child Similar support for this argument comes from postnominal modifiers.Consider the following sentence, which illustrates that nouns can host PP adjuncts: i maj [ DP [ PossP t i [ NP [ NP ][ PP tuŋód sa Akɰán]][ CP Op k [ C ŋa] as the Matrix Verb Constraint: for an overview and Otsuka (2010) on Tongan.Future work should discern the degree to which Polynesian genitive relatives resemble the 'genitive inversions' of Central Philippine languages like Aklanon and Tagalog.Thank you to Emily Drummond for making me aware of this connection!

Table 2
in the appendix which summarizes the preposed (genitive) and postposed (unmarked) paradigms in At a potluck dinner: 'I like what Maria cooked.'The overall distribution of preposed and postposed forms can be summarized as follows (data has been collected but examples have not been shown for all of these cells): Preposed gen, e.g.ʔána Postposed unm, e.g.nána Because the genitive forms are not licensed as direct arguments of verbs (56a), I propose that genitive agents of relative clauses are actually base generated in a projection within the relative clause DP, but not as a direct argument of the relativized verb.Pending future investigation, I label the projection which hosts the external agent PossP for now, assuming that it is part of the PossP ʔaj María [ Poss' [ Poss ŋa][ NP [ CP bákɰ-on]]]]].
the relative clause.This has been the case for all EECs presented thus far, but consider the EEC Maria is taking a carving class, and by the end of class, she has carved something.(Consultant'sComment:'Thenit's Maria who's the carver […] (67) means that she did the action herself.')b.#Context2:Maria recently got a piece of artwork from a local museum, so now she has carving (that someone else made).(Consultant'sComment:'No,because (67) means that Maria did it herself, not that she got a carving.')Theinfelicity of (67b) is evidence against an analysis that involves pragmatic coreference between the topic and pro, as well as evidence against a NOC PRO in external argument position of the relative clause.Note that the intended context (67b) is felicitous without voice morphology (68), Raʔúl 'Raul,' but only as the referent of the entire possessive phrase, ro tátaj ni Raʔúl 'Raul's father.' *x/y gin-ɰáhaʔ] [ro tátaj ni Raʔúl x ] y .exist pv.pfv-cook top father unm R. '[Raul's x father] y cooked something.'(EEC) One key takeaway from this data is that PRO in EECs appears to alternate with overt DPs.Sue i favored/insisted on [PRO i /Anna moving to Chicago today].
) is elaborated in (78).On this structure, the preposed agent, ʔaj María 'Maria' is generated in PossP, and obligatorily controls PRO in external argument position.