Philippine Clitic Pronouns and the Lower Phase Edge

Pronominal paradigms in Philippine-type Austronesian languages show a robust and curious gap: in transitive clauses, pivot arguments and nonpivot agents may have bound pronominal forms, appearing as second-position clitics, but pronominal nonpivot themes must be full, free pronouns. This gap is instructive regarding the organization of the lower phase edge. As cliticization involves a syntactic dependency between the host and argument position and all syntactic dependencies are constrained by phases, the gap is explained if pivots and nonpivot agents are specifiers of the phase head, making them the only DPs accessible for operations from outside of the lower phase.


Introduction
Since the identification of a cyclic boundary that separates a lower, thematic domain-often called the vP phase (Chomsky 2000(Chomsky , 2001 see also Chomsky 1986)-from a higher domain of the clause, more recent investigations have sought to articulate the fine structure of this lower phase edge (e.g., Pylkkänen 2008, Travis 2010, Legate 2014, Alexiadou, Anagnostopoulou, and Schäfer 2015, Harley 2017. Material within the lower domain is inaccessible for syntactic operations from outside of the domain, unless that material occupies the phase edge-the phase head and its specifier(s) (Chomsky 2000). One such operation is movement. Movement out of the lower phase must stop at its edge.
In this squib, we focus on identifying the precise position of the external argument with respect to the edge of this lower phase. Broadly, two views have been proposed in previous literature: (a) the external argument is generated as the specifier of the phase head, with any movements to the phase edge landing in another specifier position (e.g., Chomsky 2000, Nissenbaum 2000, Legate 2003, Aldridge 2004, Rackowski and Richards 2005 and (b) the external argument is generated in a projection below the phase head (Collins We offer a new argument for the structure in (1a), for a variety of Austronesian languages, from patterns of attested and unattested clitic pronouns. As we will review in section 2, in transitive clauses many Philippine-type Austronesian languages have two second-position clitic pronoun series, corresponding to "pivot" arguments and "nonpivot" agents; bound pronominal forms for nonpivot theme/patient arguments in such clauses are curiously absent. This fact is common to a wide range of Austronesian languages and has also been hypothesized for reconstructions of Proto-Austronesian. We contend that this paradigmatic gap is not accidental. Cliticization involves a movement relationship from a DP's argument position, within the lower phase, to the clitic's host in second position, outside of the lower phase. Under the organization of the phase edge as in (1a), to be elaborated in section 3, the pivot argument and nonpivot agent are precisely the only two types of DPs that are accessible for syntactic operations from outside of the lower phase. In section 4, we conclude by discussing an apparent counterexample.

Voice Systems and Clitic Pronouns
Many Austronesian languages exhibit what has been termed a voice system. Key characteristics of such voice systems are described in (2), taken from Erlewine, Levin, and Van Urk 2017:376. (2) Common characteristics of 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 grammatical function or thematic role. b. Articulated voice morphology: Morphology on the verb varies with the choice of pivot, including options for taking certain oblique arguments as pivots. c. Extraction restriction: Ā -extraction (wh-movement, relativization, topicalization, etc.) is limited to the pivot argument. d. Marking of nonpivot agents: Nonpivot agents are morphologically marked, often coinciding with the form of possessors (i.e., genitive case). (3). These sentences all describe Yuraw cooking taro, but they vary in word order and nominal and verbal morphology. In each example, one argument of the verb, which we call the pivot (in italics), is in sentence-final position and preceded by qu, which we gloss as nominative case. Voice morphology on the verb (boldfaced) correlates with the choice of pivot argument. Note that nonpivot arguments are also case-marked: nonpivot agents are genitive (also the case for possessors), whereas nonpivot themes are unmarked, glossed here as accusative. 1
cook-PV GEN Yuraw NOM taro 'Yuraw cooked taro.' (Patient Voice; PV) c. Hpuy-an na Yuraw sehuy qu knobuy. cook-LV GEN Yuraw taro(ACC) NOM kitchen 'Yuraw cooks taro in the kitchen.' (Locative Voice; LV) (Squliq Atayal: Erlewine field notes) Pronominals in Squliq Atayal can be expressed as free-standing pronouns or as second-position clitic pronouns. Clitic pronouns can 1 Many Philippine-type languages have previously been described as exhibiting an ergative/absolutive alignment. See especially Aldridge 2004 for such an analysis of Seediq (Atayalic), and see Chen 2017 and Erlewine, Levin, and Van Urk 2017 for overviews and discussion of the ergative hypothesis. The theoretical import of this squib does not change if the ergative hypothesis, especially that in Aldridge 2004, is adopted. S Q U I B S A N D D I S C U S S I O N 3 Linguistic Inquiry Corrected Proof https://doi.org/10.1162/ling_a_00374 ᭧ 2019 by the Massachusetts Institute of Technology be used for pivots (4a) and nonpivot agents (4b), but full pronouns must be used for nonpivot themes (5). Notice that the clitic pronouns are hosted by the auxiliary in these examples and therefore appear preverbally, unlike regular DP arguments. In particular, the first person nonpivot theme in (5) is a full pronoun, kuzing, following the verb. In examples without an auxiliary, clitic pronouns encliticize to the verb.
AUX‫ס‬GEN.1SG eat-PV NOM taro 'I am eating taro.' (Squliq Atayal: Erlewine field notes) (5) Wal‫ס‬simu m-ita kuzing. AUX‫2ס‬PL AV-see 1SG(ACC) 'You(PL) saw me.' (Squliq Atayal: Erlewine field notes) One way to explain this gap would be to claim that bound forms do not exist for accusative pronouns. But such an approach fails to explain the same gap in languages such as Tagalog, where both nonpivot themes and nonpivot agents may bear identical case markers. 2 This is seen in (6).
AV.PFV-see NOM man GEN bird 'The boy saw a bird.' b. Na-kita ng lalaki ang ibon. PV.PFV-see GEN boy NOM bird 'The/A boy saw the bird.' (Tagalog: Henrison Hsieh, pers. comm.) Although both nonpivot themes ('bird' in (6a)) and nonpivot agents ('boy' in (6b)) are in genitive case, corresponding pronominal forms do not behave the same. Consider the genitive proximate demonstrative pronoun nito 'this one', which exists both as a second-position clitic and as a full pronoun and can be used for both animates and inanimates. In (7a), nito is the nonpivot agent of a PV clause and can appear as a second-position clitic, hosted by negation, or as a full pronoun, in the postverbal field. In contrast, nito is a nonpivot theme in an AV clause in (7b) and can only occur as a full pronoun.
(7) a. Ang lalaki ang hindi ͕‫ס‬nito͖ na-kita NOM boy NOM NEG GEN.PROX PV.PFV-see 2 As an anonymous reviewer reminds us, there is, however, a further interaction with specificity for nonpivot themes: specific nonpivot themes may be in oblique (dative) case. See Latrouite 2011, Sabbagh 2016, and references there for discussion. However, as the reviewer also notes, this does not weaken the strength of the argument we develop based on nito in (7).

GEN.PROX
'It's the boy who didn't see this one.' (Tagalog: Henrison Hsieh, pers. comm.) The contrast in (7) shows that the lack of a clitic pronoun for a nonpivot theme cannot be reduced to surface morphological case. The proximate demonstrative takes the same genitive form nito as a nonpivot agent or nonpivot theme, but only the nonpivot agent can use the homophonous bound form, hosted by negation in (7a). The contrast here also forms a poverty-of-the-stimulus argument: given the optionality in nito placement available in (7a), what input leads the child to determine that the same form is only available as a full pronoun and not a clitic if used for a nonpivot theme (7b)?
Similar facts in closely related languages lead to the generalization in (8).

(8) Generalization
In transitive clauses, second-position clitic pronouns in Philippine-type languages are limited to pivot arguments and nonpivot agents.
A couple of clarifications are immediately in order. First, Philippinetype languages refers to those voice system languages with two or more different Non-Actor Voices (Blust 2010:307), which commonly have case markers and second-position pronominal clitics (Wolff 1996, Himmelmann 2002, Ross 2002, Blust 2010, 2013, Chen and McDonnell 2019. This comprises the Austronesian languages of the Philippines, most of Taiwan, northern Borneo and Sulawesi, and Madagascar. Second, the claim in (8) is not that Philippine-type languages necessarily have clitics for both pivots and nonpivot agents. For example, Malagasy has bound pronominal forms only for nonpivot agents; there are no clitic pronouns for pivots in the language (Paul 1996, Keenan and Polinsky 1998, Zribi-Hertz and Mbolatianavalona 1999. To our knowledge, the generalization in (8) has never been explicitly stated before, despite seeming to be common knowledge among Austronesianists. As Hsiu-chuan Liao (pers. comm.) notes, "It seems true that linguists working on Philippine-type languages simply assume that everyone knows that these languages have two sets of clitic pronouns: (1) genitive/ergative; (2) nominative/absolutive." For example, Billings and Kaufman (2004) offer an in-depth study of Philippine-type clitic pronoun patterns that discusses the genitive and nominative series without commenting on the consistent lack of other types. Descriptions of Philippine-type languages similarly describe two clitic pronoun series-one for pivots and one for nonpivot agents-without S Q U I B S A N D D I S C U S S I O N 5 Linguistic Inquiry Corrected Proof https://doi.org/10.1162/ling_a_00374 ᭧ 2019 by the Massachusetts Institute of Technology comment. Nonetheless, this gap has occasionally been mentioned in discussions of specific languages. In their survey of nine Formosan languages, Huang et al. (1999:167) give a table that explicitly indicates the lack of accusative (nonpivot theme) clitic pronouns for all nine languages.
The generalization is also supported indirectly by work on the historical reconstruction of Proto-Austronesian (e.g., Ross 2002, 2009, Blust 2015, Aldridge 2015. Ross (2002:36) in particular explicitly notes that reconstructions of the genitive and nominative clitic pronoun series are motivated, but there is again no accusative (nonpivot theme) clitic pronoun series.
The generalization in (8) also extends to languages with clitic doubling of full DP arguments. For example, in Nanwang Puyuma both nonpivot agents and nonpivot themes are in genitive case, marked by kan for personal names and kana for definite common nouns (Teng 2008, Chen 2017. But as Chen (2017:15-16) notes, the two types of arguments differ in their clitic-doubling behavior: nonpivot agents must be cross-referenced by a corresponding clitic pronoun on the verb, whereas nonpivot themes cannot be doubled by a pronoun. Crucially, we will maintain that cliticization and clitic doubling are, for the purposes of syntax, driven by the same operation(s). This is discussed in detail in section 3.
Additionally, as previously discussed in Erlewine, Levin, and Van Urk 2017, the key properties of Austronesian voice systems (2) can also be found in Dinka (Nilotic; South Sudan). Dinka clauses are generally verb-second, with an auxiliary or lexical verb in second position, preceded by the pivot. Argument cross-referencing in Dinka also obeys the generalization in (8). The verb or auxiliary in second position doubles a nonpronominal pivot with a prefix/proclitic. This is à-in (10). If there is a pronominal nonpivot agent, it will appear as a suffix/enclitic on the second-position head. This appears as -kù in (10b). In (11a), the nonpivot agent combines with the default PRF.PV auxiliary cí .. i to become cá.  Notably for our purposes, nonpivot themes must instead use full pronouns, as in (11b).
Finally, in all of these languages with two clitic series, a pivot and a nonpivot agent can be simultaneously cliticized onto the same host. See (12) for examples from Squliq Atayal, Tagalog, and Nanwang Puyuma. Example (10b) shows this for Dinka.

Proposal
The generalization in (8)-that in transitive clauses only pivot arguments and nonpivot agent arguments can appear as second-position clitic pronouns-can be productively understood as reflecting the organization of the lower phase edge. We adopt a phase-theoretic conception of Austronesian-type voice systems whereby the pivot argument is necessarily the highest DP in the lower phase of the clause (Aldridge 2004, Rackowski andRichards 2005), reflecting the intuition that the pivot argument occupies a designated and privileged position in the clause (Guilfoyle, Hung, and Travis 1992).
We propose that in Philippine-type languages with pivot and nonpivot agent pronouns that are second-position clitics, the agent is basegenerated as a specifier of the phase head. Here we refer to this phase as vP, but this is not crucial; for example, it may be VoiceP as in Legate 2008. In Actor Voice (AV), the external argument is the sole DP specifier at the phase edge (13a). In Non-Actor Voices (NAVs), the pivot DP moves to an outer specifier of the phase head (13b), which can be thought of as the effect of an EPP feature on v (Aldridge 2004(Aldridge , 2008 or object shift (Rackowski and Richards 2005). One famous property of voice system languages is their pivotonly extraction restriction (e.g., Keenan and Comrie 1977), in (2c). The organization of the lower phase edge in (13) potentially allows for higher probes to attract either the pivot or the agent in NAVs. We follow Aldridge's (2004Aldridge's ( , 2017 proposal by which the probe triggering Ā -movement to the CP edge will target the closest DP and therefore will be unable to skip the pivot in NAV clauses (13b). This probe stops probing after it finds one satisfactory goal: the pivot DP. This approach also predicts that, given a more articulated probe, extraction of a nonpivot agent may be possible. Nonpivot agent extraction is, for example, attested in Bikol (Erlewine and Lim 2019). Now, we turn to the derivation of second-position clitic pronouns, maintaining the following positions. First, recall that we adopt the view that cliticization and clitic doubling are, for the purposes of syntax, driven by the same operation(s), which we discuss below. Second, we posit that the second-position clitic pronouns discussed here are hosted structurally higher than the lower phase edge, on a head that could be called Aux or T. In cases where the clitic pronouns appear to be hosted on the verb, the verb itself has undergone head movement to Aux/T. Finally, and most importantly, we maintain that clitic doubling involves movement. Aux/T probes for accessible goals for clitic doubling, allowing for attraction of multiple goals. Clitics then move from the position occupied by the noun phrase they double, to the host Aux/T.
For concreteness, we illustrate the process of clitic doubling through the "Copy, Reduce, and M-Merger" derivation proposed and discussed in Harizanov 2014, Kramer 2014, Baker 2016 Kramer 2018, andSikuku, Diercks, andMarlo 2018. Under this approach, the entire noun phrase (DP) is first copied to a specifier of the hosting head (Aux/T) 3 -in (14a)-which then must undergo M-Merger (Matushansky 2006) to form a morphophonological word with its host head (14c). 4 (Step (14b) is discussed below.) . Reduce DP to D 3 Many other proposals for clitic doubling also involve movement, but vary as to exactly what moves: for example, the head of the doubled noun phase (e.g., Roberts 2010, Preminger 2019) or a clitic base-generated as a specifier of the doubled noun phrase (e.g., Torrego 1988, Arregi andNevins 2012). The derivation in (14) is provided for illustrative purposes, but the data presented here can be captured under any of these proposals so long as this movement is subject to Phase Impenetrability.
Similarly, an alternative family of analyses claims that clitics are basegenerated at the host site and establish a relationship with their doubled DP (e.g., Sportiche 1996; see also Travis 2006). If this relationship is also subject to Phase Impenetrability, the present data would also be amenable to such approaches. 4 As discussed by these authors-most explicitly by Baker andKramer (2016, 2018:1050)-particular specifier positions (or corresponding EPP requirements) can be specified to allow material of only a certain form. Requiring that specifiers, if present, undergo M-Merger with the head is one such requirement. The

. M-Merger of D and Aux / T
If the targeted DP is itself a D head without a restrictor-in other words, a weak pronoun (Postal 1966, Elbourne 2001)-it will undergo M-Merger with Aux/T directly (14a,c). The lower copy of the D/DP will then be unpronounced due to chain reduction, as the equivalent object is pronounced higher in the chain (see, e.g., Nunes 2004, Landau 2006. This results in a clitic pronoun with no double. If instead the targeted DP is a full DP with a restrictor, it must first be "reduced" to its D head alone as in (14b). Following reduction, the higher copy (now just a D head) and the lower DP will both be pronounced due to nonidentity, resulting in consistent clitic doubling, as in Nanwang Puyuma (9) or Kapampangan (C. Richards 1971, Mirikitani 1972. Baker (2016), Baker and Kramer (2016), and Sikuku, Diercks, and Marlo (2018:398) furthermore propose that languages vary in the availability of this "reduction" operation in (14b). 5 If a full DP is targeted for movement to the specifier of Aux/T but it cannot be reduced to a D head to feed M-Merger, we propose that the entire higher copy must be deleted at PF, resulting in the appearance of no clitic doubling at all. This parameter setting thus yields a language with clitic pronouns but no clitic doubling of full DPs, such as Squliq Atayal and Tagalog.
With these assumptions in place, the generalization on possible second-position clitic pronouns in (8) follows. First consider the derivation of NAV, (13b). The pivot argument (such as a theme in Patient Voice) and the nonpivot agent both occupy specifiers at the lower phase edge. Thus, both are accessible for syntactic operations from above such as clitic doubling. Next, consider the derivation of AV, (13a). Here, the agent is the pivot and is the only DP at the edge of the lower phase. Thus, the pivot agent is the only DP accessible for syntactic operations from above. Nonpivot themes-which may bear the same morphological case as nonpivot agents-remain within the complement of the phase head in AV clauses and are inaccessible for clitic doubling. If nonpivot agents occupied a position below the phase Linguistic Inquiry Corrected Proof https://doi.org/10.1162/ling_a_00374 ᭧ 2019 by the Massachusetts Institute of Technology edge, (1b), they would be unable to be clitic-doubled like nonpivot themes, contrary to fact. 6 Under our proposal, both Ā -extraction and clitic doubling involve probe-driven movements (Chomsky 2000). An important difference between these two processes is the specification of these probes as seeking one goal or multiple goals. Individual probes may be parameterized in terms of whether they stop after finding one matching goal or then continue, resulting in interaction with multiple goals (see, e.g., Hiraiwa 2001, Nevins 2011, Harizanov 2014, Deal 2015, Foley and Toosarvandani 2019. The Ā -probe in these languages seeks one matching goal-the closest DP, necessarily the pivot-and then stops. In contrast, the probe triggering clitic doubling in the languages discussed above seeks to interact with all accessible DP goals. 7 This accounts for the ability of pivot and nonpivot agent DPs to be simultaneously clitic-doubled, as shown in (12). The same has been explicitly proposed for multiple clitic doubling targeting the same host in Bulgarian: Harizanov (2014) claims that the head triggering clitic doubling will "have a property which forces any goals within its ccommand domain (subject to additional locality constraints, of course) to undergo movement" (p. 1066). This derives the contrast between the pivot-only restriction on Ā -movements attested in Philippine-type languages and the behavior of clitic doubling, which can target both pivots and nonpivot agents, but not nonpivot themes, which are inaccessible due to Phase Impenetrability.
Finally, we note that clitic doubling and Ā -extraction cannot simultaneously target the same argument. This is observed in Kapampangan in (15): the pivot DP 'man' is doubled by the third person clitic pronoun ‫ס‬ya in (15a), but when an argument in this position is wh-moved, it is no longer clitic-doubled, as in (15b).
(15) Clitic doubling does not target Ā -extracted DPs a. E‫ס‬ya masikan ing lalaki. NEG‫ס‬NOM.3SG strong NOM man 'The man is not strong.' b. Ninu ing e masikan? who NOM NEG strong 'Who is not strong?' (Kapampangan: C. Richards 1971:258, 276) Such facts are compatible with our account. Suppose the pivot is cliticdoubled, following the process in (14), followed by Ā -probing by C. The Ā -probe cannot attract the clitic pronoun, due to either some sort of general freezing of the specifiers of Aux/TP (see, e.g., Rizzi and Shlonsky 2007) or an antilocality constraint (see, e.g., Erlewine 2016, as suggested by a reviewer. 8 If C probes past the clitic pronoun to move the pivot DP across the coreferential pronoun, a crossover violation results (Postal 1971). See Baker and Kramer 2018 for much relevant discussion of such configurations.

Conclusion and Discussion
The absence of second-position clitic pronominal forms of nonpivot themes in Philippine-type languages follows from the organization of the lower phase edge. Nonpivot themes occupy a VP-internal position and are not visible to syntactic operations from outside of the lower phase. In contrast, pivot arguments, regardless of thematic role, and nonpivot agents occupy positions at the lower phase edge. Agents are base-generated there; nonagent pivots move there. This asymmetry between nonpivot agents and nonpivot themes is unexpected if agents are generated within the lower phase and not at its edge. By way of conclusion, in this final section we discuss an interesting potential counterexample: the behavior of clitics in Isbukun Bunun. Like the languages discussed so far, Isbukun Bunun has two clitic pronoun series. One series is a nominative series, used for pivots. However, the other-called the "default" series in Li 2010-marks both nonpivot agents and nonpivot themes. In (16a), the clitic ‫ס‬as marks the pivot theme and the default clitic ‫ס‬ku' marks the agent. In (16b), ‫ס‬ik marks the pivot agent, and the default clitic ‫ס‬su' marks the theme.
Realizing the nonpivot theme as a clitic appears at first glance to counterexemplify our generalization (8). However, further investigation reveals that these default clitics are not second-position clitics; thus, their behavior does not counterexemplify (8), which only applies to second-position clitics. Default clitics always encliticize to the verb, rather than appearing with any higher host. In this way, they contrast with nominative clitics, which must appear in second position-for example, encliticizing to negation in (17).
(17) a. Na‫ס‬ni'‫ס‬ik ma-ludah‫ס‬mu'. Our analysis for clitic pronouns in Philippine-type languages can in fact be easily extended to this behavior. If, in Isbukun Bunun, the host for clitic doubling of nonpivots is the phase head itself, we expect both agent and theme arguments to be visible for the purposes of clitic doubling with the default series. 9 There is no intervening phase boundary to block clitic doubling of the nonpivot theme.
For the sake of explicitness, we propose that both arguments in (17) are visible to the clitic-doubling probe on the lower phase head, resulting in "default" clitics, but that the clitic-doubling probe on Aux/ T probes only once (like the Ā -probe on C), interacting with the pivot but not with nonpivot agents. 10 Combined with a rule that arguments can only be clitic-doubled once per clause, in the highest possible position, 11 this proposal ensures the correct realization of a secondposition pivot clitic and a lower, verbal clitic for nonpivot arguments, regardless of thematic role. The default clitics in Isbukun Bunun-at first glance a counterexample to the generalization and approach developed here-thus can also be straightforwardly accounted for by our proposal. 9 We assume that a clitic-doubling probe on a head can target its specifier, as in Béjar and Rezac 2009 and much subsequent work. This allows for the phase head to clitic-double the agent in its specifier (1a). Alternatively, following the approach to clitic doubling adopted above (14), an agent pronoun could undergo M-Merger directly with the phase head. 10 We thank an anonymous reviewer for this suggestion. 11 Kenyon Branan (pers. comm.) notes that this straightforwardly follows from Kinyalolo's Constraint (Kinyalolo 1991, Carstens 2005 as applied to clitic doubling.