Bikol clefts and topics and the Austronesian extraction restriction

Many Austronesian languages exhibit an extraction restriction whereby only one particular DP—the “pivot” argument, the choice of which is reflected by morphology on the verb—can be Ā-extracted. We show that such extraction restrictions can vary between different Ā-constructions in Bikol: local clefting is limited to the pivot, whereas topicalization can target pivots and non-pivot agents but not non-pivot themes of transitive verbs. Following the phase-theoretic, locality-based approach to such extraction asymmetries in related Austronesian languages, we propose that clefting and topicalization differ in the featural specifications of their probes, with clefting necessarily targeting the closest DP and topicalization simply seeking the closest topic constituent. Evidence for this approach comes from the behavior of long-distance clefting, which may target certain non-pivot arguments and involve gaps or resumptive pronouns. The inventory of different long-distance cleft types is explained by the possibility of embedded topicalization and hanging topic left dislocation feeding higher clefting. Our study strengthens the view that the classic Austronesian pivot-only extraction restriction is best characterized in terms of syntactic locality, rather than as a restriction on the grammatical function or morphological case of movement targets.

argument that we call the "pivot" and, among other properties, this designated pivot argument is the only DP that can be targeted forĀ-extraction. The nature of this extraction restriction-which has also been described by some authors as a "subjectonly" or "absolutive-only" restriction-has been a focal point for typological and theoretical discussions of extraction asymmetries (Schachter and Otanes 1972;Keenan and Comrie 1977;Aldridge 2004;Rackowski and Richards 2005;a.o.) and is also central to discussions of the notion of "subjecthood" in Austronesian and beyond (Keenan 1976;Schachter 1976Schachter , 1996Shibatani 1988;Guilfoyle et al. 1992;Kroeger 1993;a.o.).
In this paper, we describe patterns ofĀ-extractions-specifically, clefting and topicalization-in Bikol, an Austronesian language of the central Philippines closely related to Tagalog. 1 At first glance, Bikol exhibits a familiar Philippine voice system. In example (1a), the theme lalaki 'man' has been chosen as the designated pivot and therefore is in nominative case. Patient Voice morphology on the verb reflects that the nominative argument is the verb's theme. Local clefting is limited to this pivot argument, as in (1b-c), and must leave a post-verbal gap, as we show below. Clefting of the non-pivot agent eskwela 'student' in (1c) is ungrammatical both when retaining its original genitive case marker or changing it to nominative case. Local clefting thus manifests the basic pivot-only extraction asymmetry predicted of Philippine voice system languages. Intended: 'It's the student that killed the man.' In contrast, we observe that local topicalization in Bikol can target both pivots and non-pivot agents. Examples (2a) and (2b) are both grammatical and express the same proposition that 'The student killed the man.' In (2a), the pivot lalaki 'man' is topicalized to a pre-verbal position, whereas in (2b), the non-pivot agent eskwela 'student' is topicalized to a pre-verbal position. When the non-pivot agent is topicalized in (2b), its case marking changes to be in nominative case, resulting in a clause with two nominative phrases; however, (2b) unambiguously means 'The student killed the man,' and not 'The man killed the student.' (2) Grammatical topicalization of theme pivot and non-pivot agent a. Su 'The student killed the man.' The availability of non-pivot agent topicalization in (2b) is surprising against the backdrop of the widely-discussed pivot-only restriction onĀ-extraction in these languages. 2 In addition, the availability of two nominative-marked arguments in (2b) raises questions for the nature of nominative case and the interaction of voice marking and case in these languages, which we will address. The core of our proposal will be that clefting and topicalization involve probes with different featural specifications: clefting involves a head that probes for a focus that is the closest DP (see Branan and Erlewine to appear, a)-echoing Aldridge's (2004Aldridge's ( , 2017 proposals of a ϕ-probe for allĀ-extractions in related voice system languages-whereas topicalization probes straightforwardly for a topic feature. Following the work of Aldridge (2004Aldridge ( , 2008, Rackowski and Richards (2005), and others, the pivot argument in Austronesian voice system languages is the highest argument in the lower phase, in an (outer) specifier of vP. Due to their differing specifications, clefting cannot attract another DP past the pivot, whereas topicalization can skip the highest DP (the pivot) and attract a non-pivot agent occupying the inner specifier of vP. When the agent is itself the pivot, in Actor Voice (AV), it is the only DP at the edge of the vP phase. Probing obeys Phase Impenetrability (Chomsky 2000), explaining the unavailability of non-pivot theme topicalization, in (3). (3) Non-pivot themes cannot be topicalized *Su Intended: 'The man killed the student.' Support for our locality-based approach will come from the behavior of longdistance clefting. In contrast to local clefting which is restricted to pivots, as in (1) above, long-distance clefting can target non-pivot agents, as seen in (4). We argue that such examples involve a step of non-pivot agent topicalization within the embedded CP that makes the embedded non-pivot agent the highest DP within the embedded clause, which then counts as the closest DP target for matrix clefting. 'It's the student that the radio reported t killed the man.' This and additional data inform our description of the nature of the basic Austronesian pivot-only extraction restriction, obeyed in Bikol by local clefting. We argue that the observed "pivot-only" extraction restriction must be characterized in terms of syntactic locality, reflecting the attraction of the closest DP, rather than any requirement to attract pivots or even nominative DPs. We additionally discuss hanging topic left dislocation (HTLD), a non-movementderived form of topic with an obligatory corresponding pronoun. As a non-movement construction, HTLD can be used for any DP argument: pivots, non-pivot agents, as well as non-pivot themes of transitive verbs. Just as long-distance clefts can be fed by movement topicalization as in (4), allowing for non-pivot agent clefts, long-distance clefts can be fed by embedded HTLD. This results in long-distance clefts with resumptive pronouns, which have no restriction on the arguments they can target, unlike gapped clefts. This paper is structured as follows. Section 2 discusses the interaction of case marking and voice morphology in Philippine-type languages and how these properties manifest in Bikol. LocalĀ-extraction facts are presented in Sect. 3, followed by our core analysis in Sect. 4. In Sect. 5, we discuss long-distance clefting, which will motivate a locality-based characterization of the Austronesian extraction restriction. Along the way, we describe the two different types of pre-verbal topics, the organization of the vP phase edge, and the determination of morphological case in Bikol.

Case and voice in Bikol
In this section we introduce basic properties of Bikol morphosyntax that will be relevant for our study. Many Austronesian languages, including Bikol, exhibit a particular constellation of case marking, verbal morphology, and extraction interactions that have been termed a "voice system." A summary of these key properties, building on prior work such as Schachter (1976Schachter ( , 1996, Guilfoyle, Hung, and Travis (1992), and Ross (2009), is quoted in (5) from Erlewine, Levin, and Van Urk (2017) with minor modification. As is noted in these works, there is significant variation in the terms used for such systems in previous literature. 3 (5) Characteristics of Austronesian-type voice systems a. A privileged argument: One argument is designated the "pivot," and is realized in a particular morphological form and/or structural position, regardless of its original thematic role or grammatical function.
b. Articulated voice morphology: Morphology on the verb varies with the choice of pivot, including options for taking certain oblique arguments as pivot.
d. Marking of non-pivot agents: Non-pivot agents are morphologically marked, often coinciding with the form of possessors (i.e. genitive case).
(modified from Erlewine, Levin, and Van Urk 2017: 376) The core voice system properties in (5) are all readily observed in Bikol, although in the rest of the paper we will show that the facts surroundingĀ-extraction (5c) are in reality more complicated. In the rest of this section, properties (5a), (5b), and (5d) of the Bikol voice system will be presented. Data onĀ-extraction which partially supports the characterization in (5c) will be presented in the following section.
Canonical word order in Bikol is predicate-initial. Consider the examples in (6) below, which all express the basic proposition that 'The woman bought cheese at a store for Andrew.' In each example, there is one pivot DP in nominative case, in bold in (6), and voice morphology on the verb that correlates with this choice of pivot argument. The pivot can be the thematic agent (6a) or theme (6b), but can also be a non-core thematic argument such as a location (6c) or a beneficiary (6d) which is otherwise expressed as an oblique. Post-verbal word order is free; only one word order is given for each example here.

Andrew.
Andrew 'The woman bought cheese at a/the store for Andrew.' Bikol distinguishes three different cases-nominative, genitive, and dative-with a rich inventory of surface forms that vary based on animacy and number. The table in (7) covers all case marker forms in examples that we will discuss, involving singular noun phrases, as well as the corresponding third-singular animate pronouns and demonstrative pronouns, which are used for inanimate referents. See Mintz (1973: Ch. 2) andMcFarland (1974) for more detailed descriptions of these inventories. In examples throughout, we will simply gloss these markers as NOM, GEN, or DAT respectively. Genitive and nominative animate pronouns are second position clitics; see also Erlewine and Levin (2021 Non-pivot core arguments are generally in genitive case. In addition, specific nonpivot themes may appear in dative case rather than genitive case as in (8) Although the voice system allows for different arguments to be the pivot and hence nominative, in the canonical, predicate-initial word order, it is not possible for two arguments of the clause to simultaneously be nominative. This explains the ungrammaticality of (9) below, in contrast to (6b) above. The ungrammaticality of (9) with sa tindahan shows that the ungrammaticality of (9) is not due to a surface ban on adjacent nominative phrases. 4 Genitive inanimate common nouns are introduced by ning. Genitive animate common nouns can be preceded by kaso or ning. 5 Kaito is both the genitive and dative form of the demonstrative (Mintz 2019: 184). 6 The Tagalog equivalent of this Differential Object Marking is well-studied; see e.g. Schachter and Otanes (1972) and, more recently, Latrouite (2011a), Sabbagh (2016), and Collins (2019). From a broader, pan-Austronesian perspective, Chen (2017) describes such markers as variants of the case marker for non-pivot themes (in our terms, genitive) that encode definiteness or specificity, rather than as a distinct case.

(9)
Only one (post-verbal) argument may be nominative * Pig-bakal 'The woman bought the cheese (at the store).' It's worth noting that this "voice system" descriptively differs from familiar "voice" alternations in European (and other) language families. First, neither the Actor Voice nor Patient Voice appears to be morphologically or syntactically simpler on the surface, leading some authors to refer to such systems as "symmetric" voice systems; see especially Foley (2008). Second, in the Non-Actor Voices (NAV)-which some earlier works describe as "passives"-the agent argument continues to be a DP core argument of the clause, rather than a demoted oblique. The present paper will offer further support for the view that NAV agents are full-fledged DP arguments.
Finally, we note that there is a not insignificant tradition of describing Philippine languages as exhibiting ergative/absolutive alignment. See for example Payne (1982), De Guzman (1988), Gerdts (1988), Mithun (1994), and Aldridge (2004). Under this view, Actor Voice clauses are formally intransitive, with an oblique theme, and Non-Actor Voice clauses are formally transitive. Pivots are absolutive and NAV agents are ergative, with the case on non-specific AV themes then being a homophonous oblique. On this point, see especially Aldridge's (2004Aldridge's ( , 2012 ergative analysis for Tagalog, whose voice morphology and case facts parallel the Bikol facts above. The pivot-onlyĀ-extraction restriction is then an absolutive-only extraction restriction, which is also attested in other language families where the "ergative" designation is less controversial, such as Inuit, Mayan, and Salishan; see Deal (2016) and Polinsky (2017) for two recent overviews. This ergative hypothesis for Philippine-type voice system languages has however been controversial; see especially Chen (2017), Erlewine et al. (2017), and Kaufman (2017) for recent critical discussion.
In this paper we use the terms "nominative" and "genitive" for the two core cases in Bikol, as in the earlier examples in this section, and later present an analysis for Bikol case and voice in these terms. However, the empirical contribution of our paper as well as its theoretical import is logically separable from this choice. Our core proposal for Bikol extraction facts, in Sect. 4, in fact largely follows the syntax for Austronesian voice systems proposed in Aldridge's work. Lessons for the analysis of syntactic ergativity-to the extent that Philippine-type voice system languages should be described as ergative-will be presented at the end of Sect. 5.

Local clefts and topics
In this paper we discuss the clefting of DPs and two types of DP topic constructions in Bikol, which we refer to as topicalization and hanging topic left dislocation (HTLD). We limit our attention to dependencies with DPs, as the movement of non-DPs behaves quite differently in Philippine languages. 7 In this section, we specifi-cally consider local clefts and topics. In the interest of space, we will concentrate on extractions of agent and theme arguments of notionally transitive verbs from Actor Voice (AV), Patient Voice (PV), and Locative Voice (LV) clauses.
In our work we have also investigated DP wh-questions. As in many other Austronesian languages, ex-situ DP wh-questions are formally clefts (see Potsdam 2009 for an overview) and indeed the patterns we report here for local clefts (and longdistance clefts below) are replicated in DP wh-phrases.

Clefts
As noted above, it is often claimed that only the pivot can beĀ-extracted in voice system languages-famously described as a "subject-only" restriction by Keenan and Comrie (1977)-including in closely related Philippine languages (Kroeger 1991;Aldridge 2004;Reid and Liao 2004;Rackowski and Richards 2005). This characterization indeed holds for local clefting, our firstĀ-construction in Bikol. Clefts have two parts: the exhaustive focus or focus-containing phrase and the background (a gapped clause; "bg" below), separated by a nominative case marker. 8 Example (10) shows that only the agent pivot can be clefted out of an AV clause. Clefting the nonpivot theme in (10b) is ungrammatical, whether retaining the original genitive case marker ning or switching to nominative case. Example (11) similarly shows that only the theme pivot can be clefted from a PV clause, as we also saw in (1)  Intended: 'It's the woman that ate the cheese.' The ungrammaticality of the nominative pronouns in both examples shows that local clefts must be gapped; i.e. they must have a post-verbal pivot gap, not a corresponding pronoun. Post-verbal gaps will generally not be indicated in examples, due to the flexible post-verbal word order mentioned in Sect. 2. We also present clefts from Locative Voice (LV) in (12) below. (12) shows that only the locative pivot can be clefted. Clefting the non-pivot agent as in (12b) or the non-pivot theme as in (12c)  Intended: 'It's (the) cheese that the woman bought at the store.' From these examples, we see that local clefting can only target the pivot, the argument in nominative case and cross-referenced by voice morphology on the verb, and thus follows the claimed pivot-only restriction onĀ-extraction (5c). As noted above, DP wh-questions are also formed using clefts and therefore follow the extraction restriction observed in (10-12).

Topics
Next, we turn to topics in Bikol. We use the term "topic" to refer to DP arguments in pre-verbal position without an exhaustive focus interpretation; see footnote 9 below. Topics can be formed in two different ways in Bikol: topicalization and hanging topic left dislocation (HTLD). We will argue that topicalization involves movement, whereas hanging topics are base-generated high. In the following examples, topicsand corresponding pronouns, if any-are in bold.
The examples in (13) involve topicalization of their pivots. (13a) has topicalized an agent pivot from an AV clause and (13b) has topicalized a theme pivot from a PV clause. Topicalization is associated with no intonational break and cannot be resumed by corresponding pronouns.  (14) below. In (14a), the agent pivot 'woman' is topicalized from an AV clause, followed by an intonational break-indicated by #-with a corresponding post-verbal nominative pronoun =siya which encliticizes to the verb. In (14b), the theme pivot 'cheese' is topicalized from a PV clause, with a following pause and corresponding full pronoun. NOM.DEM 'The cheese i , the woman ate it i .' The obligatoriness of the intonational break and corresponding pronoun are generally one-to-one. (An exception is discussed in footnote 25 below.) Throughout this paper, we will give English translations with canonical word order for Bikol examples with topicalization, as in (13) above, whereas we give English translations with hanging topics with corresponding pronouns for Bikol HTLD, as in (14). We have chosen to do this to highlight the presence or absence of the corresponding pronoun in the Bikol sentences through their English translations. We should however reiterate that we are making no claims regarding the discourse status of these two constructions that we call "topics" here and, in particular, we make no claim that the informationstructural properties of these Bikol sentences match those of their English translations. 9 With these basic descriptions of the two forms of topics in place, we now consider which arguments can be targeted for topicalization and HTLD. Examples (13) and (14) above showed that both topicalization and HTLD can target pivots. Topicalization can additionally target the non-pivot agent of Non-Actor Voice clauses. This is 9 As Shibatani (1991: 116) notes, "the status of the initial topic construction in Philippine languages does not seem to be well established yet." Unfortunately we are unaware of substantial progress on this important question, three decades later, and especially as it concerns Bikol in particular. Here we follow Shibatani and others in referring to these preverbal phrases as "topics." In grammaticality and felicity judgment tasks, our speakers accept sentences with topicalization (13a, b) and HTLD (14a, b) in the same contexts as their corresponding post-verbal pivot clauses and comment that the choice is optional. The determination of the discourse effects of these topic constructions-if any-awaits further work, potentially requiring corpus and/or experimental methods. observed in the PV example (15), and we will see the same generalization extend to LV below. 'The woman ate the cheese.' ii. * 'The cheese ate the woman.' Note that the topic in (15) must be in nominative case, even though the corresponding post-verbal position is a genitive case position. Example (16) retains the original genitive case marker on the topic babayi 'woman' in (15), resulting in ungrammaticality. Recall that multiple post-verbal arguments cannot be in nominative case; see (9) above. On the surface, topicalizing a non-pivot agent as in (15) results in a string with two nominative phrases: the pre-verbal topic 'woman' and the post-verbal 'cheese.' However, (15) is unambiguous in its interpretation: the post-verbal nominative phrase is unambiguously the pivot of this PV clause and therefore the verb's theme, whereas the pre-verbal nominative topic is unambiguously the non-pivot agent. 10 Our proposal below will account for this restriction.
Although non-pivot agents of transitive verbs can be topicalized, non-pivot themes cannot. This is illustrated in (17) below, which attempts to topicalize the non-pivot theme keso 'cheese' from an AV clause. The sentence is ungrammatical with keso in nominative or its original genitive case. The generalization that topicalization can target pivots and non-pivot agents (15) but not non-pivot themes (17) also extends to clauses with transitive verbs in additional voices. Consider the options for topicalization from a Locative Voice (LV) clause in (18). Interestingly, in LV, where both the agent and theme are non-pivot arguments as determined by the choice of voice morphology, we continue to observe an asymmetry: topicalization can target the non-pivot agent (18b), again resulting in a structure with two nominative phrases, but cannot target the non-pivot theme (18c). The locative pivot can also naturally be topicalized, as in (18a). store 'The woman bought cheese at the store.' The one exception to the unavailability of non-pivot theme topicalization comes from unaccusative predicates. Example (19a) is a baseline that shows that the nonpivot theme of unaccusative 'fall' receives genitive case in Locative Voice; example (19b) shows that this argument can be topicalized, again requiring nominative case in place of its post-verbal genitive. In summary, topicalization-which we will argue below to involve movementdoes not follow a pivot-only restriction, unlike clefting in (10-12). Topicalization can target non-pivot agents and non-pivot unaccusative themes, although non-pivot themes are otherwise inaccessible. Topicalization is thus freer than clefting but not unrestricted.
Next we turn to hanging topic left dislocation (HTLD). We saw in example (14) above that HTLD can target pivots. In addition, HTLD can target non-pivot agents as well as non-pivot themes of transitive clauses as in (20)(21)  tindahan. store 'The cheese i , the woman bought it i at the store.' In these examples of non-pivot HTLD (20-21), the topics themselves are in nominative case, even though their corresponding pronouns are in genitive or dative case. 11 Like (15) and (18b) above, the resulting string has two nominative phrases, but each is unambiguous in its interpretation. The pre-verbal hanging topic must correspond to the post-verbal pronoun.
We conclude that there is no restriction on the DP arguments that can be targeted by HTLD. We will argue that this is because HTLD does not involve movement, in contrast to topicalization.

Summary
In this section, we presented data on clefting and two types of topics from local clauses in Bikol. Local clefting obeys the pivot-only extraction restriction. Topicalization can target pivots and non-pivot agents, as well as non-pivot unaccusative themes. Hanging topic left dislocation (HTLD) can target any core argument, including nonpivot themes of transitive verbs. These possibilities are summarized in (22) below for arguments of notionally transitive verbs.

(22)
Possible DP targets for local dependencies from transitive clauses pivot non-pivot agent non-pivot theme clefting (10a, 11a, 12a) * (11b, 12b) All non-pivot topics involve an apparent mismatch in case marking: the pre-verbal topic is in nominative case, instead of the genitive or dative case of its corresponding post-verbal gap or pronoun. In the next section, we present our analysis for Bikol voice and case, as well as the specific analyses for clefting, topicalization, and HTLD, with additional supporting data.

Proposal
In this section we present our analysis for the patterns of voice, case, and local dependencies in Bikol introduced in the previous section. A key point that we account for is the ability of topicalization to target non-pivot agents as well as pivots, but not non-pivot themes of transitives, in contrast to clefting which is strictly pivot-only and hanging topic left dislocation (HTLD) which is unrestricted. To preview our account, we will propose that topicalization is a movement construction that involves probing for a discourse feature [TOP], restricted only by Phase Impenetrability, whereas clefting involves a probe that seeks a focused DP that must be the closest DP to the probe (see also Erlewine 2018; Branan and Erlewine to appear, a). The pivot and non-pivot agent are the only DPs at the vP phase edge, and so there is no way to target non-pivot themes, even with a [TOP] probe, except where vP does not introduce an impenetrable barrier as with unaccusatives. In contrast, HTLD is a non-movement construction, unrestricted by Phase Impenetrability. We will also discuss the determination of morphological case in Bikol, explaining the appearance of multiple nominative phrases in some topic constructions.
Our proposal is presented in three parts. Section 4.1 presents our proposal for case and voice in Bikol. We present our analysis for the two topic constructions in Sect. 4.2, and for clefts in Sect. 4.3. We then further motivate our analysis of the Bikol clause periphery from patterns of multiple topics in Sect. 4.4. Note that all dependencies in this section will be local, accounting for the patterns presented in Sect. 3 above. We then discuss long-distance clefting in Sect. 5.

Voice and case in Bikol
We begin by presenting our framework for the voice system and morphological case in Bikol. For the voice system, we follow the shared insights of widely-adopted and influential phase-based approaches to voice systems in Philippine languages, drawing especially on the work of Aldridge (2004Aldridge ( , 2008 and Rackowski and Richards (2005). Under such approaches, the pivot DP is distinguished by being the highest DP in vP-the lower phase of the clause-in an (outer) specifier of vP. Agents are basegenerated in Spec,vP. In Actor Voice (AV) clauses, there is no movement to the edge of the vP phase; the agent pivot is base-generated as the only specifier of vP and remains the highest DP in the vP; see (23a). In Non-Actor Voice (NAV) clauses, a non-agent DP is moved to the outer specifier of vP, above the agent DP (23b). 12 Specifiers of vP are illustrated on the left in trees, but this does not reflect their word order, as we discuss below.
(23) The vP phase in AV and NAV clauses a.
Actor Voice: b. Non-Actor Voices: vP is a phase and therefore material within the complement of the phase head v will be inaccessible for syntactic operations from above (Phase Impenetrability), except in the case of unaccusative verbs (Chomsky 2000). In (23), this domain of impenetrability is illustrated with a double line. This approach predicts a basic asymmetry between AV and NAV clauses: in AV clauses, the vP phase edge has only one DP that is accessible for syntactic operations from above, whereas in NAV clauses, there are two. In the following subsection, we will propose that this is precisely what allows for topicalization to target only pivots (in AV and NAV clauses) and non-pivot agents (in NAV) of transitive clauses; these are the only DP constituents of the lower phase that can move out. See also Erlewine and Levin (2021) for a recent, additional argument for precisely this organization of the vP phase edge, based on the inventory of clitic pronouns in Philippine-type voice system languages, including Bikol. In the case of unaccusatives, the complement of v is accessible for probing and therefore non-pivot themes can be topicalized, as we have seen above. 13 Voice morphology is the realization of the head v, which the lexical verb V headmoves to. Aldridge (2004) and Rackowski and Richards (2005) differ in the precise mechanisms that relate the realization of voice morphology to movement of the pivot DP in NAV clauses. However, both of these approaches agree on the basic geometry for the vP phase edge in AV vs NAV clauses, reviewed in (23) above. We adopt this common proposal here. NAV clauses involve movement of the pivot DP to an outer specifier of vP whereas AV clauses involve no such movement, leaving the agent to be the highest DP in the phase and the only DP at in the vP phase edge. 14,15 13 Editor Julie Anne Legate points out that we predict that both internal arguments of a double unaccusative would be accessible for topicalization. At this point, we have not been able to identify appropriate predicates in the language to test this prediction. 14 Pivots of peripheral voices must either be generated as DPs via an applicative functional head without a lower source for structural case licensing, as Aldridge (2004) and Rackowski and Richards (2005) assume, or may originate as a prepositional/oblique object that is then severed from its preposition via prepositionincorporation (Baker 1988); the latter approach is discussed in Guilfoyle et al. (1992: fn. 7) and Kroeger (1990) for Tagalog and Bodily (2003) for Cebuano. For our current purposes, we leave this analytic choice open and concentrate on PV examples as illustrative of NAV syntax more generally. 15 As pointed out by our editor and a reviewer, and recently discussed extensively in Hsieh (2020: Ch. 7), non-DP extraction (in Tagalog) generally proceeds differently from DP extractions, and in particular does not interact with voice choice. The same appears to be true in Bikol; see for example (49) in Sect. 4.3 below for an instance of pre-verbal PP fronting which does not interact with voice and which behaves differently from pre-verbal DPs.
Post-verbal word order in Bikol is free, except for a requirement that complement clauses be rightmost. We adopt the proposal from Erlewine, Levin, and Van Urk (2020) (also in Erlewine 2018 for Toba Batak) that all linearizations of vP with the verbal complex (v+V) as the leftmost constituent can be generated. See also Fowlie (2013) and Branan (to appear) for similar proposals for Tagalog.
Next we turn to our proposal for morphological case determination in Bikol. Following Marantz (1991), we propose that morphological case in Bikol may be structurally assigned or realized with context-sensitive defaults. 16 In particular, we propose that default case in the vP phase is genitive and default case in the CP phase is nominative. The idea of genitive as a default case within some structures in Austronesian languages is developed by Chen (2018), Donohue and Donohue (2010), and Erlewine, Levin, and Van Urk (2020). 17 In addition, nominative can be assigned structurally by T, via Agree, as in Aldridge's (2004) analysis of absolutive in "T-type" languages.
The derivation of AV and NAV clauses as well as the determination of morphological case will be illustrated below. We begin with the transitive AV clause derivation in (24). Following the voice system proposal above in (23), the agent is base-generated in Spec,vP and no other argument is moved to the vP phase edge. We propose that T bears [PROBE:D] which assigns structural nominative case to its target. 18 As the agent is the highest DP in the vP-and, in this case, the only one accessible by Phase Impenetrability-[PROBE:D] on T necessarily targets the agent pivot, which receives nominative case.
As Hsieh points out, such behaviors are at first glance problematic for phase-based accounts of voice system languages and their extraction restrictions. In particular, prominent prior work such as Aldridge (2004) and Rackowski and Richards (2005) do not discuss any mechanism for moving non-DP constituents to the phase edge in order to feed their extraction from above. However, a recent theoretical proposal in Van Urk and Richards (2015) on the Nilotic language Dinka rectifies the issue. and Erlewine, Levin, and Van Urk 2015 observe that Dinka exhibits all of the characteristic properties of Austronesian voice systems in (5) above.) Van Urk and Richards (2015: Sect. 4) propose that v in Dinka is associated with two features, one Case/D-related and anotherĀ-related; when the two can be satisfied by a singleĀ-DP goal, the probes must target that one goal together, but when there is anĀ-non-DP distinct from the pivot DP, both that non-DP and the pivot can be moved to separate specifiers of vP, simultaneously. Voice morphology choice is only sensitive to the movement of DPs. This mechanism or similar allows for the principled movement of non-DPs through the lower phase edge without interaction with voice morphology. 16 Marantz (1991: 247) proposes that "unmarked case may be sensitive to the syntactic environment; for example, in a language GEN may be the unmarked case for NPs inside NPs (or DPs) while NOM may be the unmarked case inside IPs." See Levin (2015: Ch. 6) for recent discussion of Marantz 1991. Relevant for us is Levin's discussion which concludes that " [Marantz's] categories of unmarked and default case must be collapsed" (Levin 2015: 212). We use the term "default" here. See also Baker (2015) and Chen (2018) for recent work that further develops this idea of context-sensitive default case. 17 An anonymous reviewer further points out that this choice of genitive as default within the lower domain of the clause may have a diachronic origin, due to the lower portion of clauses in many Austronesian languages having its origin in nominal structures. 18 For explicitness, we use the notation [PROBE:F] to represent a probe for the feature [F]. This avoids the more common [uF] notation, which different linguists take to stand for "unvalued" or "uninterpretable." (24) Actor Voice clause derivation Any DP that is realized in the vP phase and lacks structural case-marking will receive default genitive case (Erlewine, Levin, and Van Urk 2020). This accounts for the genitive case on non-pivot themes in AV clauses. In addition, as noted above, specific non-pivot themes receive dative case through a separate process (see footnote 6) and therefore will not receive default genitive. The surface form computed for an AV clause with a non-specific theme is presented in (25). Recall that the linear order of constituents in the vP is subject to scrambling, with the only constraint being that the verbal complex be leftmost.
(25) An Actor Voice clause at PF "AV-V NOM=DP ag/pivot GEN=DP th " or "AV-V GEN=DP th NOM=DP ag/pivot " Next we turn to the derivation of transitive Non-Actor Voice clauses. This is illustrated with the tree in (26). As we introduced above, in NAV clauses, a non-agent DP moves above the agent to an outer specifier of vP. [PROBE:D] on T will find the highest DP, which is the pivot, and assign it structural nominative case. The non-pivot agent has not received structural case, so it will receive default genitive as it is in vP.

(27) A Non-Actor Voice clause at PF
"NAV-V NOM=DP pivot GEN=DP ag " or "NAV-V GEN=DP ag NOM=DP pivot " It's worth highlighting that the vP phase boundary is relevant here in two distinct senses. For purposes of probing and movement, the complement of the phase head v constitutes a distinct domain, inaccessible for higher probing (Phase Impenetrability), except where the verb is unaccusative. This boundary is indicated by the double line in the trees above. However, for purposes of linearization (scrambling) and default case calculation, it is the entire vP maximal projection, including its specifiers, that behaves as a unit in all clauses. Unless moved higher, specifiers of vP are linearized post-verbally and subject to scrambling together with all other vP-internal constituents. Non-pivot agents receive default genitive case, just as (non-specific) non-pivot themes do. We suggest that this distinction correlates with the timing of the relevant operations: probing is a narrow-syntactic operation and is sensitive to the double line (Phase Impenetrability), whereas linearization and default case determination takes place post-syntax, at PF, where the entire vP behaves as a single unit.
Finally, we discuss the calculation of morphological case for a constituent that moves out of vP. First consider the movement of pivots. Pivots receive structural nominative and will retain this structural case when moved. However, the situation is more complicated when a non-pivot DP moves. Due to Phase Impenetrability, this is only possible with non-pivot agents, as illustrated in (28), or with non-pivot themes of unaccusatives.

(28) Movement of a non-pivot agent
We propose that any DP without structural case that is pronounced in the CP phase will receive default nominative. Non-pivot agents have no source of structural case, so their morphological case realization will depend on the phase in which they are pronounced. If the agent stays within the vP phase, it appears with default genitive. But if the agent moves out into the CP phase, as in (28), it will appear in nominative case. The PF realization of a structure as in (28) is sketched in (29). This description also extends to moved non-pivot themes of unaccusatives, as in (19b).
(29) A moved non-pivot agent at PF "NOM=DP agent . . . [ vP NAV-V NOM=DP pivot ]" There are therefore two sources of surface nominative case in our proposal: structural nominative via Agree with T and default nominative in the CP phase. In (29), the post-verbal pivot DP bears structural nominative whereas the pre-verbal non-pivot agent bears default nominative by virtue of its position in the CP phase. As noted by Schütze (2001), identity between structural nominative and a default case in a higher domain of the clause (e.g. on topics) is cross-linguistically common. 19 A consequence of this proposal is that, despite there being multiple sources of nominative case, only one DP can bear nominative case and appear in a post-verbal position. This is the pivot DP which receives structural nominative from T while in the vP and therefore linearized post-verbally. For any other DP to bear nominative case, it must move out of the vP into the CP phase and therefore be in a pre-verbal linear position. This explains the impossibility of multiple post-verbal nominatives, as illustrated in (9) above.
The analysis for Bikol voice and case presented in this section derives the surface morphosyntax for basic AV and NAV clauses in Bikol that we saw in Sect. 2. In addition, two features of this approach will be important for the analysis of Bikol topics and clefts, which we turn to in the following sections. First, the new proposal that nominals in the CP receive default nominative will be important for deriving the case marking observed on topics. Second, two DPs are at the vP phase edge in NAV clauses-the pivot and the non-pivot agent-whereas only the agent pivot is at the phase edge in AV clauses. While this is a feature of previous phase-based accounts for voice system syntax in Rackowski (2002), Aldridge (2004), and Rackowski and Richards (2005), its consequences have not been fully discussed in previous work (except recently in Erlewine and Levin 2021). This organization of the vP phase edge will be crucial for explaining the differing extraction restrictions on clefting vs topicalization in Bikol.

Topicalization and hanging topic left dislocation
We showed in Sect. 3.2 that there are two topic constructions in Bikol: topicalization, which involves a gap and no prosodic break, and hanging topic left dislocation (HTLD), which has a corresponding pronoun and a prosodic break. Topicalization can target non-pivot agents as well as pivot DPs, but not non-pivot themes (except from unaccusatives), whereas HTLD can target any DP argument. In this section we present our analysis for these facts.
We propose two functional heads in the clause periphery, which we simply label Top 2 and Top 1 , with Top 2 c-commanding Top 1 . In Rizzi's (1997) terms, these can be thought of as heads in a split CP. This organization is illustrated schematically in (30): 19 An alternative approach would be to claim that all nominative is the result of default nominative in CP, and that Agree with T has the effect of making a DP target behave as if it is in the higher phase for the purposes of default case calculation. That is, even in cases where the pivot stays in the vP, the pivot will receive the CP's default nominative, due to its Agree relationship with the CP phase's T head. This alternative proposal has the conceptual advantage of describing all surface forms of nominative case as due to the same mechanism-default case in the CP-but with a new proposal regarding the effect of Agree with T. We will leave the full consideration of this alternative description for future work.

(30) Topic layers in the Bikol clause periphery
Topicalization is due to Top 1 : [PROBE:TOP] on Top 1 fronts any [TOP] goal it finds to Spec,Top 1 P. Top 2 generates hanging topics: a DP is base-generated in Spec,Top 2 P and binds a pronoun in its scope. 20 Any constituent in Spec,Top 2 P is followed by a prosodic break. In Sect. 4.4, we present data from multiple topicalization that supports the higher position for hanging topics.
The claim that topicalization involves movement while HTLD involves basegeneration and binding is supported by differences in island-sensitivity (Ross 1967 ].
'The student i , I saw the man [ island that killed them i ].' Further evidence for this movement / non-movement contrast comes from the interpretation of verb-argument idiom chunks (see e.g. Marantz 1984). Here we use two idioms for 'mumbling' and 'being a coward': We now turn to the explanation for the possible targets of topicalization. Probing is subject to Phase Impenetrability (Chomsky 2000); therefore, [PROBE:TOP] on Top 1 cannot probe into the complement of v and attract a matching goal. In AV clauses, this means that only the pivot agent can be topicalized (23a). In NAV transitive clauses, two DPs are potentially accessible for probing: the pivot and the non-pivot agent, which are both specifiers of vP; see (23b). Non-pivot themes are not accessible for topicalization because of Phase Impenetrability, except where the verb is unaccusative and therefore the complement of v is accessible for probing from above. This accounts for the patterns of topicalization documented in Sect. 3: pivots and non-pivot agents are the only DPs that can be topicalized from transitive clauses, with non-pivot themes of unaccusatives also topicalizable. [PROBE:TOP] will find the closest accessible target with the [TOP] feature. In cases of non-pivot topicalization, the non-pivot bears a [TOP] feature but the higher pivot does not. Because the pivot does not bear the feature that the probe seeks, it does not intervene for the topicalization of the non-pivot.
DPs without structural case in Spec,Top 1 P or Spec,Top 2 P will be realized with default nominative case; see (29) above. This explains the appearance of nominative case on non-pivot topics, as in (36), which correspond to a post-verbal gap or genitive or dative pronoun. In addition, pivots which receive structural nominative and are moved to Spec,Top 1 P also appear in nominative case. DAT.DEM 'The cheese i , the woman ate it i .' Because only pivots can be both post-verbal and in nominative case, the interpretation of such examples with multiple nominatives is unambiguous, as noted above.

Clefts
Recall that, unlike the possibilities for local topics, local clefting is strictly pivotoriented: Intended: 'It's the student that killed the man.' We propose that clefting is the result of attraction by a probe that seeks a target with both [FOC] and [D] features that must be the closest DP to the probe. Probing of this form is discussed and motivated in Erlewine (2018) and Branan and Erlewine (to appear, a), and following these works we notate this probe [PROBE:FOC+D].Āprobing that is restricted to the closest DP has been a component of some analyses of syntactic ergativity as in Aldridge (2004), but is also well attested in non-ergative languages, as shown in Branan and Erlewine (to appear, a). This probe will match the highest DP if it bears [FOC], but will not match a non- [FOC] highest DP and also cannot probe past it for a better match. 21 As discussed in Branan and Erlewine (to appear, b), such a probe can be described in Deal's (2015, to appear)  Pre-verbal DP exhaustive focus constructions-which we call "clefts" here-in Bikol and many related Philippine languages invite two types of analyses, schematized in (39) below. The first is a biclausal pseudocleft structure, in (39a), where the focused DP constitutes a predicate nominal in a null copular clause, and the cleft background is formally a headless relative clause. The pre-verbal nominative marker is then explained as the predicted nominative case marker for the subject of the copular clause. The second is a monoclausal structure where a functional head Foc extracts the DP focus out of the background clause. On this latter approach, we might associate the observed pre-verbal nominative marker su with the Foc head itself, with the motivation for this surface form being that the monoclausal structure (39b) historically derived from the biclausal pseudocleft structure (39a). Our core proposal for clefting in Bikol and its extraction restriction is in principle compatible with either analytical approach, as long as the probe involved is [PROBE:FOC+D] restricted to the closest DP and the background is not a full CP, as we discuss. Here for concreteness we adopt the latter, monoclausal description as in (39b), for two reasons. First, our preliminary look at patterns of relativization in Bikol suggests that it is not strictly pivot-oriented, suggesting that the extraction restriction of clefts cannot be attributed to that of relativization. 22 Second, the structure 21 Our proposal could account for the extraction restriction on clefting if we instead describe clefting as involving [PROBE:D]; this in turn would be functionally equivalent to Aldridge's (2004Aldridge's ( , 2017 proposal where, in her terms, extraction involves a [uϕ] probe. However, our description here of clefting involving [PROBE:FOC+D] that must target the closest DP better reflects the need of clefting to target a focused constituent, which must be independently enforced in some way under a [PROBE:D] analysis. 22 For instance, (i) and (ii) below are attested non-pivot agent and non-pivot theme relatives, respectively. We leave a complete investigation of patterns and mechanisms of relativization in Bikol for future work. of clefts, both above and below the focused DP, can be neatly modeled by positing a clause type that varies minimally from that in (30) in having the focus-attracting Foc head in place of Top 1 , as illustrated in (40) below. 23 (40) The Bikol clause periphery for clefts Below we will show that the cleft background clause-the sister of Foc, i.e. TP in (40)-has a structurally reduced left periphery as compared to matrix and embedded complement clauses, as is cross-linguistically common (see e.g. Belletti 2012), and in particular cannot include the Top 1 and Top 2 projections introduced in the previous section. A hanging topic can however be hosted above a cleft focus, as reflected by the Top 2 head in (40). Finally, we note that under our analysis, the su marker in clefts is synchronically the realization of the Foc head (40) rather than a case marker; we will nonetheless continue to gloss it as NOM throughout.
We first show that the cleft background cannot include a hanging topic nor movement-derived topic DP. As local clefting is limited to the pivot DP and topicalization can target the pivot or non-pivot agent of transitive verbs, the most plausible configuration to test would be with a pivot focus and non-pivot agent topic, as in (41) below. Example (41) is ungrammatical, with or without the pause and corresponding pronoun to make the topic a hanging topic.  Kroeger (1991) have observed that second-position clitics do not climb to fronted pre-verbal DP foci, but do climb to fronted non-DPs. Aldridge (2002Aldridge ( , 2004 in particular takes such facts to form an argument that DP clefts are biclausal, as in (39a). (But see also Kaufman 2010: 188, fn. 4 for critical discussion of this type of argument.) Parallel facts hold in Bikol; see (48) below. However, we note that clitics do not climb to pre-verbal topics (both movement-derived topics and hanging topics) in Bikol, and thus the domain of clitic climbing in both topic constructions as in (30) and in clefts as in (40)  ].
Intended: 'It's cheese that Andrew (, he) ate.' Additional, independent evidence for the cleft background not itself being a full clause comes from high adverbials. Consider the speaker-oriented modifier 'unfortunately' in (42). (43) shows that it can appear before or after a topicalized pivot.  (44). The availability of 'unfortunately' post-verbally shows that the adverb is semantically compatible with being in the cleft background, and thus that its ungrammaticality at the left edge of the cleft background is due to a syntactic restriction. The adverb can also appear above the cleft focus as in (45). These patterns further support our proposal that cleft backgrounds are not full clauses. This supports our proposal in (40), which includes a higher Top 2 head to host hanging topics above cleft foci. We propose that clefting involves movement of the focused constituent, predicting clefting to be island-sensitive. This is demonstrated with the adjunct island and relative clause island data in (47). The island-sensitivity of clefting here patterns with topicalization in (31)  Recall too that local clefts must have a post-verbal gap in the background clause, corresponding to the focus, further supporting their derivation via movement.
(See examples (10-11) above.) This detail will become important in Sect. 5, where we will see that long-distance clefts may have a resumptive pronoun in place of a gap. Finally, our description of clefts as triggered by [PROBE:FOC+D] predicts that non-DP categories such as PPs cannot undergo clefting. Although PPs can also be focus-fronted, as in (48), it is clear that this does not involve the same structure as the DP focus clefts described here. First, a pre-verbal focused PP cannot be followed by the su marker which is required for DP focus clefts. Second, second-position clitics such as the pronoun =ako in (48) can be hosted on these focused pre-verbal PPs, but second position clitics do not climb up to the DP focus of clefts, as has also been described for Tagalog (Kroeger 1991: 123-125); see also footnote 23. These facts suggest that non-DP fronting constructions are markedly different from DP fronting constructions; see also footnote 24. We leave their detailed study for future work and refer readers to Hsieh (2020) for recent discussion of such structures in Tagalog. A related prediction that our theory of clefting makes, as noted by an anonymous reviewer, is that clefting via [PROBE:FOC+D] should be able to skip intervening non-DP categories. This prediction is also borne out. Consider example (49), based on the cleft in (37) above, which shows that it is possible to have a locative PP before the verb within a cleft background: 'It's the man that the student killed at home.'

(48) Pre-verbal PP focus is not a cleft
Recall from (41) above that the edge of the cleft background cannot host DP topics. Therefore the pivot lalaki 'man' must have moved from a post-verbal position within the background clause, within vP. The pre-verbal PP is necessarily outside of vP. The grammaticality of (49) thus shows that clefting is unaffected by intervening PPs, as is predicted by our proposal that clefts are derived using [PROBE:FOC+D] on Foc. 24 We note that both aspects of our proposal for the structure of clefts-anĀ-probe that necessarily targets the closest DP (Erlewine 2018; Branan and Erlewine to appear, a) and the lack of topic projections within the cleft background (41)-are necessary to derive the strict pivot-only restriction on local clefting, reflected in (37-38). If we were to use a simple [PROBE:FOC] that is not limited to the closest DP, we would predict that a non-[FOC] pivot DP could be skipped, allowing the cleft to attract a [FOC]-bearing non-pivot agent or unaccusative non-pivot theme instead, just as we observed with topicalization via [PROBE:TOP]. At the same time, if the cleft background contained Top 1 or Top 2 , a topic could be built first, making a non-pivot argument the highest DP within the background clause. Subsequent clefting with [PROBE:FOC+D] would be predicted to be able to attract that non-pivot argument, fed by topicalization or HTLD within the background clause. Therefore, to derive the pivot-only restriction on local clefts, the background clause must not be a full CP, as we argued above.

Multiple topic constructions
We now return to the two topic constructions in Bikol and show that the analysis presented above in Sect. 4.2 is further supported by examples with multiple pre-verbal topics. This section serves to present these positive predictions of our account, as well 24 Example (49) also shows that, despite the unavailability of high adverbials and DP topics in cleft backgrounds (41-44), there is space to host locative PPs. This again reinforces the idea that the positions available for pre-verbal DPs and pre-verbal PPs in Bikol fundamentally differ, as we also noted above in relation to focused DPs vs focused PPs. See also footnote 15 for a suggestion regarding the analysis of non-DP movements. as to establish baseline behaviors that will become important for our discussion of patterns of long-distance clefting in Sect. 5.3 below.
We first consider the two grammatical PV examples in (50). Both topics in (50) are in nominative case, as is independently predicted for each topic construction. The generalization is as follows. In these sequences of two topics, the first topic is a hanging topic, with a prosodic break and corresponding post-verbal pronoun, whereas the second topic is the result of topicalization. Example (51) below shows that it is not possible to add a prosodic break after the second topic, with or without a break after the first topic, and regardless of the choice of post-verbal pronoun. This data in (50-51) supports our proposal that topics with a prosodic break and corresponding pronoun (hanging topics in Spec,Top 2 P) are structurally higher than topics with no break and no corresponding pronoun (movement-derived topics in Spec,Top 1 P). Let's consider the derivation of each of these PV multiple topic examples in (50). We first consider the derivation of (50a). Here there is a hanging topic binding an agent pronoun and a topicalized theme pivot. We therefore begin by constructing a PV clause with the full DP 'woman' with a [TOP] feature as the theme and a pronoun as the agent. Following movement of the pivot theme to an outer specifier of the vP, we result in a vP organized as in (52): 25 But note that the genitive clitic pronoun is optional in this example. We nonetheless analyze Pedro in (50a) as the result of HTLD, as it must be followed by its characteristic prosodic break, but with the pronunciation of its bound pronoun being optional. At this point we are unable to describe the precise conditions that govern this optionality, and leave this question open for future work.

(52) vP for (50a)
The rest of the clausal spine is built following the hierarchy in (30), beginning with the merger of T. [PROBE:D] on T will Agree with the closest DP, assigning babayi 'woman' nominative case. The agent pronoun is in vP so it receives default genitive case. Top 1 is merged and its [PROBE:TOP] fronts the pivot DP 'woman' to Spec,Top 1 P. Top 2 is then merged in and takes Pedro as its specifier, which binds the lower agent pronoun. The clause is complete once we merge the C head to form the root CP. The resulting hierarchical structure is as in (53a) below, together with its final linearized structure in (53b).
"NOM=DP i # NOM=DP th/pivot PV-V =GEN.3sg ag,i " ⇒ "si Pedro i , su babayi ginadan=niya i " 'Pedro i , he i killed the woman.' Both topics are realized in nominative case: babayi 'woman' bears structural nominative from T whereas Pedro receives default nominative in the CP. The hanging topic in Spec,Top 2 P is followed by a prosodic break. The post-verbal pronoun is genitive and thus appears in the =niya form. This results in the correct surface form attested in (50a/53b), and also derives the correct, unambiguous interpretation for this string.
Next we turn to the derivation of example (50b). This example is superficially similar to (50a) but with a post-verbal nominative pronoun in place of the genitive pronoun in (50a), resulting in a markedly different interpretation, 'Pedro, the woman killed him.' We begin by building a PV vP with a pronoun theme pivot moving to its outer specifier, above the [TOP]-marked agent DP babayi.

(54) vP for (50b)
We now build the higher phase. T is merged and [PROBE:D] assigns nominative case to the pivot pronoun, which is the closest DP goal. Next, Top 1 is merged and its [PROBE:TOP] moves the agent to Spec,Top 1 P. The Top 2 head is merged with its specifier, Pedro, which binds the theme pivot pronoun. After merging in C, we yield the structure in (55): (55) Final structure for (50b) a.
b. "NOM=DP i # NOM=DP ag PV-V =NOM.3sg th/pivot,i " ⇒ "si Pedro i , su babayi ginadan=siya i " 'Pedro i , the woman killed him i .' Both topic DPs receive default nominative case because they are in the CP phase. The lower pronoun is also nominative as it is the pivot and therefore received structural nominative, resulting in the post-verbal clitic form =siya. This results in the correct surface form in (50b/55b), with the correct interpretation. So far we've looked at multiple topics in a PV clause. Under our proposal both the pivot and non-pivot agent in a transitive NAV clause are at the vP phase edge and thus accessible for topicalization, and both arguments can be targeted for HTLD as well. This allowed for the two minimally contrasting examples in (50) above which are both grammatical but with differing interpretations. But now consider multiple topics in an AV transitive clause. Here we observe an asymmetry: example (56a) is grammatical with its post-verbal dative pronoun, whereas (56b) is ungrammatical with its post-verbal nominative pronoun. Intended: 'The student i , they i killed the man.' This asymmetry is predicted by our account. Following our proposal and the discussion of the PV examples in (50) above, the outer, hanging topic eskwela 'student' in (56) must bind the post-verbal pronoun, with the inner topic lalaki 'man' being moved from its base position. In an AV clause, only the agent pivot is at the vP phase edge and thus available for topicalization. In contrast, HTLD is not similarly limited as it does not involve movement. This together explains the grammaticality of example (56a). Example (56b) is ungrammatical because the non-pivot theme lalaki 'man' would have to be moved from within the lower phase, in violation of Phase Impenetrability. This asymmetry observed in AV clauses with multiple topics in (56) thus further supports both our analysis for the difference between topicalization and HTLD as well as our proposal for the syntax of the vP phase edge in AV and NAV clauses, following Rackowski (2002), Aldridge (2004), Rackowski and Richards (2005), and Erlewine and Levin (2021).

Summary
In this section we presented our proposal for Bikol clause structure, morphological case, topics, and clefts. Concentrating on the salient difference between the two movement operations of topicalization and clefting, we proposed a locality-based account for the differing extraction restrictions.
Our analysis builds on common Minimalist assumptions regarding the locality of syntactic operations. In particular, movement is subject to Phase Impenetrability and is triggered by a probe that must target its closest goal (Chomsky 2000(Chomsky , 2001; and many others). The pivot-only restriction on clefting is due to a probe that seeks a target with both [FOC] and [D] features and which must be the closest DP, as motivated further in Branan and Erlewine (to appear, a). Topicalization instead involves [PROBE:TOP] which can skip a non-[TOP] pivot DP to attract a non-pivot argument. Phase Impenetrability explains the inability of topicalizing non-pivot themes, which are not at the vP phase edge, except in unaccusatives where vP does not form a barrier for probing.
Finally, we note that under our view, nothing about clefting is inherently linked to pivot-hood. As proposed in Sect. 4.3 above, background clauses of a local cleft are simply structured so that the pivot is necessarily the closest DP to the probe. We predict that if there is a strategy for making a non-pivot DP closer to the cleft's probe, clefting would target this non-pivot DP instead. We will see that this is the case in the next section, where we consider long-distance clefts.

Long-distance clefts and the Austronesian extraction restriction
In this section, we take a closer look at the nature of the famed Austronesian pivotonly extraction restriction. We have seen that, in Bikol, this restriction is obeyed by local clefting but not by local topicalization or HTLD, so our approach will be to further study clefting in Bikol. At first glance, there are several different ways to characterize this type of extraction restriction: (57) Three characterizations of the classic Austronesian extraction restriction i. Pivot-only: Only arguments cross-referenced by voice ("pivots") can be extracted. ii.
Nominative-only: Only nominative arguments can be extracted. iii. Locality-based: Only the structurally highest argument can be extracted.
The challenge is to distinguish between these three different descriptions. Every clause has only one pivot, which is in nominative case. Assuming that a topic cannot be formed first (see Sect. 4.3 above), every clause also only has one nominative argument, which is the pivot. And assuming the basic proposal for the hierarchical structure of voice system languages (Sect. 4.1 above), the highest argument in every clause will be the pivot, in nominative case. Therefore, in basic examples of local clefting, these three descriptions in (57) are extensionally equivalent: In the background clause of local clefts, the pivot is the only nominative argument, and is structurally highest. The study of local clefts alone does not allow us to determine the correct characterization for the extraction restriction.
For this reason, in this section we study long-distance clefting in Bikol. We begin in Sect. 5.1 with some preliminary discussion of long-distance extraction in voice system languages. The core data on long-distance clefting will be presented in Sect. 5.2. Unlike in local clefts, long-distance clefting can target embedded non-pivot agents as well as embedded pivots, which forms an argument against the "pivot-only" characterization of clefting in (57i). We propose that, in such examples, embedded topicalization takes place first and feeds clefting. We support this approach, in Sect. 5.3, with additional data from the interaction of long-distance clefting and embedded topics. In the end, we will also be able to tease apart the "nominative-only" (57ii) and localitybased (57iii) approaches, solidifying our argument that the Austronesian extraction restriction exemplified by Bikol clefting must be described in terms of hierarchical structural configurations and the locality of syntactic operations.

Background: Voice systems and long-distance extraction
Just asĀ-extraction from local clauses is limited in languages with Austronesian-type voice systems, long-distance extraction is also similarly constrained. Descriptively, extraction out of an embedded clause in Bikol requires that the embedded clause itself be the pivot of the higher clause. In other words, long-distanceĀ-movement is always subextraction from a clausal pivot. This pattern has been well-documented in Tagalog since Kroeger (1991: Ch. 7), and is also a major point of discussion in Rackowski and Richards (2005).
The examples in (58) illustrate the grammatical long-distance clefting of embedded pivots. The pivots of the embedded clauses are the theme Andrew in (58a), where the embedded clause is PV, and the agent 'man' in (58b), where the embedded clause is AV.

A.
A. 'It's the man that the radio reported that t killed Andrew.' Notice that in both cases the higher verb 'report' is in PV, with its agent 'radio' in genitive case as expected. We can think of the complement clause 'that the man killed Andrew' as the pivot of the verb 'report' in PV, although CPs do not exhibit morphological case marking. The embedded clause's pivot is then subextracted to yield the grammatical cleft in (58).
The higher verb must be PV for this long-distance extraction to take place. Example (59) below minimally contrasts from (58a), with the higher 'report' clause now in AV, and the result is ungrammatical. The pivot of this higher clause is the agent 'radio,' instead of the complement clause. Under the probe-driven conception of movement adopted here, what is important for our purposes is that the highest DP within the embedded CP count as the "closest" for the cleft's [PROBE:FOC+D] in (58), instead of the agent DP of the verb 'report.' Different approaches could be taken, but for concreteness here we briefly present and follow the analysis of long-distance extraction from Rackowski and Richards' (2005) study of Tagalog. In grammatical cases of long-distance extraction as in (58), the complement CP itself moves to an outer Spec,vP above any agent DP. The verb is in the PV form, correlating with this movement of the theme to Spec,vP. This structure is illustrated in (30). Recall that vP will be linearized with the verbal complex leftmost, explaining the final word order. CPs are generally rightmost, due either to extraposition or their relative weight.
(60) An embedded CP in Spec,vP (PV), as in (58) ⇒ "v+V DP agent CP" Movement of the CP here "smuggles" the target DP above the higher clause's agent DP. In particular, Rackowski & Richards propose that the relationship between v and CP makes the CP transparent for probing from above. 26 The cleft's [PROBE:FOC+D] will thus search into the pivot CP, matching with the highest DP goal within. As discussed in Branan and Erlewine (to appear, b), such "smuggling" derivations (see also Collins 2005; Belletti and Collins 2021) may be made possible by a depth-first search procedure (or similar, as in Chow 2022) that looks into the contents of accessible specifiers before proceeding down the spine, or by first targeting the CP as a partial match (as per the intuition in the aforementioned works) and then searching further within it.
In contrast, if the higher verb is in AV as in (59), the complement CP will not move to Spec,vP. Due to Phase Impenetrability, it is impossible to probe into the CP that is inside the lower VP.
(61) An embedded CP in the complement of v (AV), as in (59) The licit and illicit patterns of probing from above for a goal in the lower phase of a transitive clause are summarized in (62). In simple cases of probing for a local goal, the goal must be in Spec,vP to be accessible for probing from above (62a-b) due to Phase Impenetrability, making pivots and non-pivot agents uniquely visible for probing from above. In cases where the goal is embedded within a CP, that CP itself must move to Spec,vP to escape Phase Impenetrability (62c-d) and to be made transparent for probing (see footnote 26).

Long-distance clefting
Long-distance clefting in Bikol differs from local clefting in two ways. First, longdistance clefting can involve a gap or a resumptive pronoun in the embedded clause, whereas local clefts must involve a gap. Second, gapped long-distance clefts can target non-pivot agents as well as pivots, unlike local clefts which must target the pivot. We begin with discussion of the latter property. We have seen in example (58) above that embedded pivots can be clefted long-distance. Example (63), repeated from (4) above, shows that embedded non-pivot agents can also be clefted long-distance. We propose that long-distance clefting of non-pivot agents as in (63) involves a first step of embedded topicalization, followed by long-distance clefting. First, we note that topicalization can take place within embedded complement clauses, moving a non-pivot agent to the embedded CP clause edge. Just as in topicalization in local matrix clauses, the non-pivot agent topic eskwela 'student' appears in nominative case in (64).

(64) Embedded topicalization of a non-pivot agent
Pig-balita lalaki. man 'The radio reported that the student killed the man.' The embedded non-pivot agent eskwela 'student' is now the highest DP in the embedded CP in (64). If we cleft from (64), [PROBE:FOC+D] will search into the embedded CP, as the higher verb 'report' is PV, and attract the highest DP in the embedded clause, if focused. This allows for the successful derivation of the long-distance nonpivot agent cleft in (63). Now recall that Bikol also has another way to form topics, hanging topic left dislocation (HTLD), associated with a prosodic break and a corresponding pronoun. Embedded CP edges can also host HTLD, as demonstrated with an embedded nonpivot agent hanging topic in (65). If the DP generated in this embedded hanging topic position bears [FOC], clefting using [PROBE:FOC+D] based on a structure as in (65) will yield a long-distance non-pivot agent cleft with a resumptive pronoun instead of a gap, which is indeed grammatical, in (66). 27

(65) Embedded HTLD of a non-pivot agent
Pig-balita Note that the cleft focus in (66) is not followed by the prosodic break associated with hanging topics (65). This is, however, predicted by our account, where the prosodic break associated with HTLD is tied to the pronunciation of a constituent in Spec,Top 2 P. In contrast to topicalization, HTLD can target all arguments, including non-pivot themes of transitive verbs. This predicts that an embedded non-pivot theme of a transitive verb can be clefted long-distance as long as it is fed by embedded HTLD, not topicalization, making the corresponding embedded pronoun obligatory. This is borne out in (67). 'It's the student i that the radio reported that the man killed them i .' We also predict that non-pivot themes of unaccusatives can be clefted long-distance, fed by embedded topicalization and thus leaving a gap. This is borne out, in example (68), based on example (19b) above. The availability of hosting both movement-derived topics and hanging topics at embedded clause edges reflects the fact that these embedded complement clauses are full CPs. This is in contrast to the background clause of clefts, which we argued in Sect. 4.3 to be TP which for instance cannot host the high adjunct 'unfortunately.' In contrast, 'unfortunately' is available at the edge of embedded clauses in long-distance clefts, again reflecting their full CP size: lalaki. man 'It's the student that the radio reported unfortunately t killed the man.' In this section we've concentrated on the possibility of topicalization or HTLD feeding clefting as a means of clefting embedded non-pivot arguments, but the same approach can also yield long-distance clefts of an embedded pivot DP. As predicted by this approach, long-distance pivot clefts as in (58) can also involve resumptive pronouns, which reflects embedded HTLD followed by clefting. 28 The patterns of possible long-distance clefting of the arguments of an embedded transitive verb, with an embedded gap or resumptive pronoun, are summarized in (71) below, together with the possibilities for different local dependencies (22) repeated from Sect. 3 above. As noted above, local and long-distance clefting differ in two ways: long-distance clefting can have a resumptive pronoun, while local clefting cannot, and long-distance clefting can target a greater range of possible DP arguments, also dependent upon the presence or absence of a pronoun.
(71) Possible DP targets for local and long-distance dependencies from a transitive clause pivot non-pivot agent non-pivot theme local clefting (gap) * * local topicalization (gap) * local HTLD (pronoun) long-distance clefting (gap) (58) (63) * ( 67) long-distance clefting (pronoun) (70) (66) Our proposal that embedded topicalization and HTLD can feed long-distance clefting predicts precisely this pattern in (71). Unlike the edge of a cleft background which is a TP (40), full CPs include the Top 1 and Top 2 projections, which can make a non-pivot the highest DP in the embedded CP, which can then be clefted.
We have already argued in Sect. 4 that topicalization involves movement, explaining why it is limited to pivots and a subset of non-pivot arguments and leaves a gap, whereas HTLD involves base-generation, explaining why it is not limited to particular arguments and involves a pronoun. Long-distance clefting with a gap can be derived by a first step of embedded topicalization, explaining why it is not strictly pivotoriented, unlike local clefting. Clefting with a resumptive pronoun is only possible long-distance, because it is fed by embedded HTLD, and consequently can target any DP argument. This derives the pattern in (71).
We conclude with discussion of a potential conceptual complication to this proposal. Under some approaches to information structure, topic-hood and focus-hood are expected to be mutually incompatible. However, here we reiterate that we use the term "topic" (and the corresponding feature [TOP] as the trigger of movement topicalization) descriptively to refer to fronting that is not interpreted with exhaustive focus semantics (see footnote 9 above). In particular, our core findings here would be unaffected if the movement here described as "topicalization" were instead thought of as a purely optional movement. 29 See also the other instances of topicalization feeding non-topicĀ-movement in other languages in footnote 27 above.

Long-distance clefting and embedded topics
We have argued that long-distance clefting can involve a first step of embedded topicalization or HTLD. This approach then predicts non-trivial interactions between long-distance clefting and embedded topics. We will discuss such patterns in this section.
First, we observe that topicalization and HTLD can simultaneously target an embedded clause edge, just as they can simultaneously target the edge of a simplex clause, as we saw in Sect. 4.4 above. 30 For ease of presentation, in this section we will use single and double underlines, respectively, for outer, base-generated hanging topics and inner, movement-derived topics, as well as their corresponding gaps. The two examples in (73)  i. * 'The radio reported that the man i , the student killed him i .' ii. 'The radio reported that the man i , he i killed the student.' b.
Pig-balita i. 'The radio reported that the man i , the student killed him i .' ii. * 'The radio reported that the man i , he i killed the student.' In (73a), the hanging topic lalaki 'man' is interpreted as the non-pivot agent, whereas in (73b), it is interpreted as the pivot theme. Just as we established above in Sect. 4.4 for unembedded multiple topic constructions, the generalization is that the postverbal pronoun unambiguously corresponds to the higher, hanging topic. See (52-55) above for the derivation of these patterns, which also apply to the embedded clauses in (73).
The question now is what options are possible when we build clefts from these structures in (73). On the surface, the resulting clefts in (74) appear as long-distance clefts of lalaki 'man' with a single topic at the edge of the embedded clause. The two examples in (74) again differ only in the choice of pronoun after the embedded verb, and each example is unambiguous in its interpretation. Descriptively, the embedded resumptive pronoun corresponds to the fronted cleft focus. port, with the head we have described as "Top 1 " in (30) bearing [PROBE:X] instead of [PROBE:TOP]. See Douglas (2018) for a recent proposal for topicalization via such a formal feature [X]. 30 In particular, recall that there are greater possibilities for multiple simultaneous topics in a Non-Actor Voice clause than in an Actor Voice clause; see (50) vs. (56). We therefore use examples with embedded PV clauses here. 31 The optionality of the genitive clitic pronoun in (73a) and (74a) below parallels its optionality in unembedded contexts as well, as we observed in Sect. 4.4. The notation " ? (=niya)" reflects that the use of the pronoun =niya is strongly preferred in these examples, as we would expect of HTLD. i. 'It's the man i that the radio reported that the student killed him i .' ii. * 'It's the man i that the radio reported that he i killed the student.' Notice that the interpretations of (74a, b) correspond one-to-one to the interpretations of examples (73a, b) above. That is, each example in (74) is unambiguously interpreted as a cleft of the embedded hanging topic 'man' from (73). We indicate this in (74) with corresponding gaps in the embedded hanging topic positions.
We can also be certain that the 'man' in the grammatical (74a, b) has indeed moved from the embedded clause as indicated, rather than being base-generated at the top. As we described in Sect. 5.1 above, movement out of an embedded clause is only possible if the embedded clause functions as the pivot of the higher verb. Long-distance movement in (74a, b) was possible because the higher verb 'report' is in PV. If the higher clause is instead in AV, both (74a, b) become ungrammatical: The unavailability of the (i) interpretation for the string in (74a) also teaches us that it is not possible to extract a post-verbal pivot across a pre-verbal topic. Consider a derivation where we begin with the embedded clause in (76a). If we were able to cleft the post-verbal pivot 'man' out of this embedded clause, across the pre-verbal agent topic 'student,' we would predict the availability of the structure in (76b) as a long-distance theme pivot cleft. This result in (76b) is string-identical to (74a) and would be predicted to have the unattested (74a, i) interpretation. Intended: 'It's the man that the radio reported that the student killed t.' We return now to the derivation of clefts from the embedded multiple topic structures in (73). The examples in (74) showed that clefting of the outer, hanging topics from (73) is possible. What about clefting the inner, movement-derived topics from (73)? This would result in long-distance clefts with embedded hanging topics, marked by their characteristic prosodic gap, with the cleft foci corresponding to gaps in the embedded inner, movement-derived topic positions. These hypothetical structures with their predicted interpretations are given in (77) Int.: 'It's the student that the radio reported that the man i , t killed him i .' The structures in (74) vs. (77) are presented schematically in (78) below. These patterns strengthen the argument that clefting necessarily attracts the DP that is highest and therefore structurally closest to the probe. From a structure with multiple embedded topics as in (73), it is only possible to cleft the higher, embedded hanging topic (74) and not possible to cleft the lower, embedded movement-derived topic (77).
(78) Long-distance cleft from an embedded clause with two topics a.
(74) b. * (77) Similarly, from a clause with one pre-verbal DP topic and one post-verbal pivot, it is not possible to extract the pivot across the topic (76). 32 Moreover, recall from the previous section that topicalization can feed clefting in cases where there is no embedded hanging topic. This configuration is repeated here in (79) from (72a) above. The ungrammaticality of (78b) therefore cannot be attributed to a general immobility of inner, movement-derived topics.
(79) Long-distance clefting can target a movement-derived topic (58,63) Such data helps us to distinguish between the "nominative-only" and localitybased characterizations of the Austronesian extraction restriction in (57) above. Although all DPs that can be clefted are nominative in their lower positions (pivots or topics), being nominative is not a sufficient condition to be clefted. That is, the proper characterization of the restriction on clefting cannot be that any nominative phrase can be clefted. Even if being nominative is a prerequisite for clefting-see e.g. Deal (2017) on deriving extraction asymmetries through case-discriminating probingonly the highest nominative DP can be clefted. Because the structurally highest DP within any clause will necessarily be in nominative case (either a pivot which receives nominative case from T, or a topic which is realized with default nominative in CP), a restriction of the cleft's probe to nominative goals is unnecessary. The extraction restriction inevitably must refer to locality (the closest DP), and any characterization additionally referring to the nominative case of targets must be rejected on grounds of theoretical parsimony.
A reviewer suggests that these long-distance clefting facts could also be addressed by taking cleft-formation to be topic-oriented, i.e. using [PROBE:TOP]. It is true that when there is an embedded topic, long-distance clefting targets the topic, and when there are multiple embedded topics, long-distance clefting targets the highest topic, as schematized in (78). In local clefting or with long-distance clefting without embedded topics, if we adopt the perspective that the pivot argument is necessarily a topic (see e.g. Chen 2017), the reviewer suggests that we could describe clefting as uniformly attracting the closest topic. Aside from conceptual challenges to this approach, where clefts with exhaustive focus semantics are paired with a syntax that picks out a topic feature, cleft-formation via [PROBE:TOP] also fails to account for the fact that intervening non-DPs, which may potentially bear a [TOP] feature, must be skipped; see (49) above. Cleft-formation thus must specifically target the closest DP, which bears anĀ-feature that accords with the semantics associated with the construction; we therefore argue that it is best described as the result of [PROBE:FOC+D] that must target the closest DP. In the conclusion below, we also discuss additional challenges that the data here pose to such proposals that conflate the notions of pivot and topic.
Our conclusion can also be translated into ergative hypothesis terms. As noted in Sect. 2 above, many works describe Philippine voice system languages such as Bikol as exhibiting ergative/absolutive alignment. This includes Aldridge (2004), whose influential approach to the basic clause structure of voice systems we have adopted here. In brief, for these authors, what we have described here as nominative case is better described as absolutive, andĀ-extraction in these languages exhibits "syntactic ergativity": in particular, an "absolutive-only" extraction restriction. If we were to adopt the ergative hypothesis as a mode of description, we would conclude that the syntactic ergativity observed in Austronesian voice system languages-evidenced in clefts in Bikol-in fact should not be described as an "absolutive-only" extraction restriction (pace e.g. Deal 2017). The appearance of this "absolutive-only" requirement on local clefts is due to the absolutive pivot argument being structurally highest in the cleft background clause. The source of this "syntactic ergativity" then is, again, best described as a locality-based effect: in particular, ofĀ-probing for the closest DP, which is attested in both ergative and non-ergative languages (Branan and Erlewine to appear, a).

Conclusion
In this paper we've described and analyzed patterns of clefting and topic formation in Bikol, an Austronesian language of central Philippines. Our analysis supports the view that the basic Austronesian "pivot-only" extraction restriction is best analyzed in terms of hierarchical configurations and the locality of syntactic operations.Ā-constructions that exhibit this "pivot-only" extraction restriction, such as cleft-formation in Bikol, involveĀ-probing that is restricted to targeting the closest DP (Branan and Erlewine to appear, a). This echoes Aldridge's (2004Aldridge's ( , 2017 earlier intuition that these constructions involve probing for [ϕ] or [D]. In local clefts in Bikol, the pivot is necessarily the highest DP, but in long-distance clefting, topicalization or hanging topic left dislocation (HTLD) at the embedded clause edge can feed the cleft with a pivot or non-pivot argument as its closest DP target.
The behavior of Bikol long-distance clefts also forms an argument against casediscriminating approaches to the Austronesian extraction restriction. Based on the interactions of long-distance clefting and embedded topics studied in Sect. 5.3, we conclude that consideration of syntactic locality is a necessary and sufficient condition for explaining the possible patterns for clefting in Bikol. Not only is there no preference for clefts to attract a "pivot," but it is both insufficient and unnecessary to describe clefting as subject to a case-discriminating (e.g. "nominative-only" or "absolutive-only") extraction restriction.
In contrast to clefting, topicalization in Bikol is not bound by the basic pivotonly restriction, even for local topics, but is also not completely unconstrained. The movement-derived construction of topicalization can target non-pivot agents as well as pivots, but not non-pivot themes except from unaccusatives. This is explained by the organization of the vP phase edge in Austronesian voice system languages: the agent is the only specifier of vP in Actor Voice, whereas in Non-Actor Voices, the pivot moves to an outer specifier of vP, resulting in two specifiers (Rackowski 2002;Aldridge 2004;Rackowski and Richards 2005;Erlewine and Levin 2021). vP is generally a phase for probing and extraction, with these specifiers of vP being the only possible targets for syntactic operations from above. Unaccusative vP however does not form a barrier for probing (Chomsky 2000), allowing the topicalization of their non-pivot themes. This contrasts with the behavior of HTLD, which can target any DP argument, including non-pivot themes of transitive verbs. Evidence from islandsensitivity and idiom interpretation motivates the view that HTLD does not involve movement, unlike topicalization and clefting.
Our study also secondarily contributes to the theory of case determination, offering new evidence for the domain-sensitive approach to morphological case determination as in Baker (2015), where "default" case (Marantz 1991) can take different forms in different domains of a single clause. In particular, we claim that genitive is the default case for DPs within the vP phase-as Erlewine, Levin, and Van Urk (2020) argue to be the case for many Austronesian languages and corresponding linearly to the post-verbal field-and nominative is the default case for DPs outside of vP in Bikol. Our evidence for this claim comes in particular from the behavior of movement-derived non-pivot topics, which are nominative but correspond to postverbal genitive positions. The behavior of hanging topics, which are base-generated high, shows that nominative is the default case in the higher domain of the clause, which accords with the observation that nominative is often the default case for similar hanging topic constructions cross-linguistically (Schütze 2001). Our analysis also offers a new, concrete theoretical approach to movement dependencies without case connectivity.
Finally, we conclude with a brief note on variation in the Austronesian extraction restriction(s), both within and between individual languages. We first note that some examples of non-pivot agent topics in other Philippine languages can be found in previous literature. The Tagalog example in (80) comes from De Guzman (1995) and shows that non-pivot agents can be topicalized-with or without a corresponding pronoun-but non-pivot themes cannot. 33 Pizarro-Guevara (2020) has experimentally confirmed that Tagalog speakers accept non-pivot agent topics but not non-pivot theme topics, and that local non-pivot agent topics are much more acceptable than local non-pivot agent clefts. See also Ceña andNolasco (2011) andHsieh (2019: 528, fn. 10;2020: Ch. 6;2021) for more discussion of licit non-pivot agent extraction in Tagalog. 33 Richards (2005) shows that non-pivot agents in Tagalog can also undergo another type of fronting called ay-topicalization, similarly resulting in structures with multiple ang-marked phrases, but non-pivot themes cannot. Miller (1988: 40-41) and Nagaya (2007: 366) give similar examples of non-pivot agent topics, also discussed in Kroeger (1991: 214-215), but the impossibility of non-pivot theme topics is not mentioned by these authors. kitchen Intended: 'The fish, mother will cook (it) in the kitchen.' We have found similar examples of grammatical non-pivot agent topics but ungrammatical or unattested non-pivot theme topics in the Philippine languages of Hiligaynon (Mithun 2019: 159), Limos Kalinga (Ferreirinho 1993: 68-71), Kapampangan (Mirikitani 1972: 154;Rowsell 1983: 57-58), Pangasinan (Benton 1971: 154), and Western Subanon (Blake 2020). Reid (1978: 36) also presents parallel examples of this form from Bontok, Ilokano, Ivatan, and Tagalog. The same is observed of topicalization in Seediq, an Austronesian language of Taiwan (Atayalic), but where topics appear in a dedicated clause-final position instead of pre-verbally (Aldridge 2004: 44-45); see also Erlewine (2014) for further discussion and Tsukida (2018: 320) for a similar example.
In each of these grammatical examples of non-pivot agent topicalization, the agent topic is in nominative case, resulting in a sentence with two nominative phrases, but the interpretation of the sentence is unambiguous, and the pronoun (if any) that corresponds to the pre-verbal topic is in genitive case. These properties are exactly what we have observed in Bikol non-pivot agent topicalization. Discussing such examples, Shibatani (1988: 133) notes that "only those preposed subject [pivot] topics... are associated with focus [voice] marking in the verbs... the (pure) topics do not control focus [voice] marking." In other words, such examples motivate a clear distinction between the notion of pivot (his "subject"), which is unique per clause and whose choice is cross-referenced on the verb, from nominative-marking, which may apply to more than one argument including preverbal non-pivot topics.
Such examples lead us to suspect that the availability of non-pivot topics, especially with non-pivot agents, may be quite widespread across Philippine-type voice system languages. At the same time, this possibility appears not to be universal across these languages: an anonymous reviewer notes that non-pivot agent topicalization is not tolerated in Malagasy, another language with so-called Philippine-type syntax, as recently documented in Xu (2019). Within the framework for extraction restrictions developed and defended here, we hypothesize that such variation reflects different featural specifications on the probes involved: if topicalization and clefting both involve the same probe specification, we would expect both constructions to exhibit the same extraction restriction.
The possibility of this cross-linguistic variation parallels the variation observed between differentĀ-dependencies within Bikol, where clefting involves [PROBE:FOC+D] limited to the closest DP (Branan and Erlewine to appear, a) and topicalization involves [PROBE:TOP]. AlthoughĀ-dependencies are often described as a natural class (see e.g. Chomsky 1977), subsequent work has also shown that there are important distinctions in this space as well (see e.g. Cinque 1990;Lasnik and Stowell 1991;Postal 1994). The featural specifications of the probes involved is one important way in which differentĀ-dependencies are distinguished.
Examples of the form in (80) in Tagalog and the many other languages cited above have largely been ignored in previous discussions of Austronesian syntax, but we believe that they are important data points that show that the characterization of allĀdependencies in these languages as strictly pivot-only is overly simplistic. The careful investigation of such extraction restrictions-both between different languages as well as between differentĀ-constructions in individual languages-will contribute to our broader understanding of the shape of possible variation inĀ-probing.