Javanese (ISO 639-3 jav), an Austronesian language of Java, Indonesia, exhibits a di-marked construction traditionally described as a passive (Wedhawati & Arifin 2006; Robson 2014; Krauße 2017). This construction (1a) is characterized by an optional agent by-phrase and a theme argument in preverbal position—a position restricted to the syntactically prominent phrase in the clause. It also involves dedicated morphological marking (di-), which is absent in its active counterpart (1b), as is typical of passive constructions cross-linguistically (Shibatani 1985).

This construction shares key typological properties with passive constructions in neighboring Austronesian languages, such as Indonesian (ISO 639-3 ind) and Balinese (ISO 639-3 ban). Both are widely analyzed as exhibiting true passives involving A-movement of the theme to subject position (Wechsler & Arka 1998; Aldridge 2008; Arka 2008; Cole et al. 2008; Kroeger 2014; inter alia). The Indonesian passive (2) features the same verbal affix di- as the Javanese construction in (1); the Balinese passive (3) employs a different affix, but displays the same word order pattern.1

While these observations suggest that the Javanese di-construction is a typical Indonesian-type passive, new data from East Javanese, a lesser-studied dialect of eastern Java (including the major cities of Surabaya and Malang; Ras 1994; Krauße 2017), reveal a puzzling asymmetry that calls this analysis into question. In East Javanese actor voice clauses, a numeral quantifier (e.g. rolas ‘twelve’) can freely dislocate from its clause-initial host and intervene between the host and an aspectual auxiliary, such as the future auxiliary até in (4).

When a clause is marked in the so-called passive voice, quantifier float in the same pre-auxiliary field becomes unacceptable. This is shown in (5), where dislocating the same numeral quantifier rolas ‘twelve’ between the sentence-initial theme (e.g. ‘the tofu’) and the future auxiliary até yields semantic and grammatical consequences.

This voice-based asymmetry is unexpected if the di-construction (5) is a true passive. Under a widely adopted formal approach to Indonesian-type Austronesian languages (e.g. Aldridge 2008; Cole et al. 2008; Legate 2014; inter alia), actor voice (AV) exhibits nominative-accusative alignment, with T assigning nominative case to the highest caseless DP, as illustrated in (6a). The passive is analyzed as a subtype of AV, in which the theme A-moves to [Spec TP] to receive nominative case (6b), parallel to the external argument in transitive AV clauses (6a).

If this analysis holds for East Javanese, no asymmetry in quantifier float is expected between AV and the passive, as both involve the same A-movement step from the VoiceP phase edge to the subject position [Spec TP]. The availability of the pre-auxiliary landing site in AV but not in the passive, as indicated by the quantifier float asymmetry between (4) and (5), therefore remains unaccounted for under this analysis.

Building on this voice-based asymmetry, this paper pursues three goals. First, it demonstrates that the East Javanese di-construction is distinct from Indonesian-type passives (e.g. (2)–(3)) and involves no A-movement of the theme to the grammatical subject position. Second, it argues that this pseudo-passive construction should be unified with the object voice (OV) construction (7), which alternates with actor voice and together forms what is commonly referred to as an Indonesian-type voice system. Like the di-construction, the OV construction features a preverbal theme argument that is disallowed in actor voice, as shown below in (7).

Drawing on six diagnostics, we present novel evidence that the East Javanese OV construction, of which the di-construction is a subtype, is best analyzed as an object topicalization construction. On this view, the preverbal theme in (7) is a topic phrase occupying an Ā-position, as schematized in (8a). We then show that the East Javanese AV construction contrasts with OV in involving subject topicalization, as illustrated in (8b): the highest caseless DP first moves to the grammatical subject position ([Spec TP] in (8b)) and subsequently undergoes Ā-movement to a left-peripheral topic position. On this analysis, the contrast between the di-construction (i.e. an object topicalization construction) and AV in disallowing quantifier float in the same pre-auxiliary field (i.e. the presumed subject position, [Spec TP]) (4) follows straightforwardly.

The final goal of this paper is to examine how the East Javanese pseudo-passive sheds light on the developmental pathways of Indonesian-type passive constructions and, more broadly, on the evolution of syntax in Western Austronesian languages. Specifically, the coexistence of the Javanese pseudo-passive with morphologically identical di-marked passives in neighboring languages points to a plausible pathway whereby topicalization is reanalyzed as passivization through the grammaticalization of topics into subjects, in line with observations from other language families (e.g. Comrie 1988; Shibatani 2011). This account further addresses a longstanding question in Austronesian syntax concerning how true passive constructions emerged from symmetrical voice systems.

The remainder of the paper is structured as follows. Section 2 introduces the key properties of the East Javanese voice system. Section 3 presents evidence from six diagnostics that both the di-construction and the so-called object voice are best analyzed as instances of object topicalization. Section 4 compares the East Javanese system with those found in other Javanese dialects and Indonesian-type Austronesian languages, highlighting its relative conservatism, an underexplored dimension of syntactic variation among passive-like constructions, and the likely directionality of change. Section 5 concludes.

Unless otherwise noted, the data presented in this paper are drawn from primary fieldwork conducted between 2021 and 2025, comprising grammaticality judgment tasks and elicitation. Five native speakers of East Javanese were consulted, in addition to the first author’s native intuitions. All were born and raised in Surabaya and speak Indonesian as a second language; no discrepancies were observed in their judgments. The comparative data discussed in Section 4 are based on fieldwork with three monolingual Indonesian speakers born and residing in Jakarta, as well as three speakers of the North Aceh dialect of Acehnese.

Traditionally described as Indonesian-type, Javanese exhibits a three-way voice system (e.g. Suhandano 1994; Ogloblin 2005; Nurhayani 2014; Robson 2014; inter alia). The actor voice (9) is characterized by a homorganic nasal prefix (N-) and SVO word order. The preverbal field is occupied by a DP corresponding to the grammatical subject in accusative languages—i.e. the external argument in unergatives and transitives, or the internal argument in unaccusatives and passives. This DP follows any aboutness topic and precedes the aspectual auxiliary, while transitive themes remain postverbal.

The object voice (10) features a bare verb, flexible word order, and a strict person restriction on the agent. The theme DP may appear either in the pre-auxiliary field or postverbally, as indicated by the two triangle brackets in the examples. The agent is expressed as a person proclitic attached to the verb and is restricted to first or second person; third-person agents are ungrammatical in OV, as shown below.

The alleged passive construction (11) carries the prefix di-, traditionally described as a passive marker (Bintoro 1980; Badib 1980; Wedhawati & Arifin 2006; Robson 2014; Krauße 2017; inter alia; see also Van den Berg 2004 and Nurhayani 2014 for discussions of the etymology and diachronic development of di-), and cognate with the passive affix di- in Indonesian. We gloss this affix as ‘3/pass’, as it has also been analyzed as a third-person marker (e.g. Ramelan 1983; Suhandano 1994). Like OV, this construction is highly productive with dynamic bivalent verbs, displays word order flexibility, and allows the theme to surface either preverbally or postverbally, as in (11).

However, unlike the OV construction, the di-construction permits only third-person agents, as in (12a). The prefix di- likewise cannot function as a general passive marker co-occurring with first- or second-person agent morphology, as shown in (12b). This restriction is not unique to East Javanese; it has also been noted for Javanese more broadly (e.g. Ramelan 1983; Sofwan 2001), although some dialects do not exhibit it. See Section 4 for further discussion.

In addition to the third-person prefix di-, the agent may also be optionally expressed as a prepositional phrase (PP). When the PP is adjacent to the verb, the preposition ambe’ may be omitted, as in (13).

In short, both the passive and the OV construction employ an agent-indexing morpheme attached to the right edge of the verb stem, but they exhibit complementary person constraints, as summarized in (14). The labels “OV” and “passive” are used here in quotation marks, as we argue that the two are best analyzed as a single construction.

For clarity, we use the term ‘pivot’ throughout the remainder of the paper to refer to the phrase eligible for the preverbal position in each voice—namely, the boldfaced agent in AV (9) and the boldfaced theme in OV and the passive (10)–(11).

The syntactic status of the pivot—i.e. the syntactically prominent phrase eligible to appear in specific linear positions and participate in relativization—has long been a central question in the western Austronesian literature (e.g. Schachter 1976; Shibatani 1988; Kroeger 1993; Richards 2000; Pearson 2005; Chen 2025; see also Davies 1993 for a focused discussion of Javanese). Recent work further suggests that the answer may be language-specific (e.g. Pearson 2005; Legate 2014; Patrianto & Chen 2023a; b; Lohninger & Katochoritis 2025). In this section, we present new data supporting the view that pivots in East Javanese are topics (Ā-elements) rather than grammatical subjects (derived A-elements). On this view, the pivot of the di-construction is an object topic rather than a promoted theme subject. Evidence for this claim comes from six diagnostics, which jointly show that the theme in this construction behaves as an Ā-element with topic properties, while the agent retains its role as the binder of the theme pivot.

This view is not novel for Javanese. Cole et al. (2002), for example, posit that Javanese “subjects” (i.e. pivots) are Ā-topics that cannot remain in subject position, citing their definiteness constraint and the fact that wh-subjects, like ordinary pivots, must raise to a higher position rather than remain in the grammatical subject position. See also Poedjosoedarmo (1977), Suharno (1982), Davies (1993), Ewing (2005), Sato (2015), and Vander Klok (2019) for similar observations that Javanese pivots exhibit topic-like behavior.3 By contrast, Conners (2008) and Nurhayani (2014) analyzed pivots in Tengger Javanese and Yogyakarta Javanese as grammatical subjects. Despite these competing analyses, the literature to date lacks a unified set of diagnostics for a holistic assessment of the pivot’s syntactic status and its A- versus Ā-properties—a gap that this paper aims to address.

We now reconsider the structure of the di-construction (34), shown in Section 3 to involve Ā-movement of a non-subject phrase.

Recall that this construction shares three core properties with object voice: (i) both display a verbal affix indexing the person and number of the agent or initiator, (ii) both permit PP pivots, and (iii) in both, the pivot may appear either preverbally or postverbally. This shared configuration is schematized in (35).

We propose that both constructions involve a non-subject topic—either a DP or a PP—that Ā-moves directly from its θ-position. In both cases, a non-subject phrase bears a [top] feature and enters into an Agree relation with [utop], triggering Ā-movement to a left-peripheral topic position, as illustrated in (36). Following previous work on Javanese (Cole et al. 2002; Vander Klok 2019), we assume this position to be above FocusP and above the derived A-position in East Javanese triggered by Agree with [uD].

On this view, the di-construction is an object topicalization construction with a third-person agent, while what has traditionally been analyzed as object voice involves a first- or second-person agent. This analysis builds on earlier insights from Uhlenbeck (1975) and Poedjosoedarmo (1986), both of whom note that Javanese is a topic-oriented language in which discourse topic largely determines sentence structure.8 See also Asikin-Garmager (2017) and Khairunnisa (2022) for similar topic analyses of pivots in Sasak, Nagaya (2012) for Lamaholot, and Sirima (2025) for Mori Atas.

The theoretical assumptions behind this analysis are as follows. We assume that all finite clauses in East Javanese bear an uninterpretable topic feature on an Ā-head, which probes the closest phrase with a matching feature. Given Relativized Minimality (Rizzi 1990), this Agree relation skips any intervening DPs and PPs lacking a topic feature. A topic phrase therefore need not be the highest phrase in the clause to agree with [utop].

Where the highest DP in a clause also bears a topic feature, it is first probed by [uD] on T and raises to the grammatical subject position, before undergoing Ā-movement to [Spec TopP] following its Agree relation with [utop]. Following earlier work on Malagasy (Pearson 2005), we treat the AV affix N- as extraction morphology indexing the nominative case status of the subject topic.9

Where the topic of a clause is a non-subject phrase (either an object or a PP), that phrase enters into an Agree relation with [utop] and Ā-moves to [Spec TopP], while the highest phrase in the clause agrees with [uD] and A-moves to the grammatical subject position, as illustrated in (36). We propose that no overt extraction morphology is associated with non-subject topicalization, since the phrase undergoing Ā-movement does not bear nominative case.

The fact that the di-construction patterns with object voice in quantifier float—disallowing pre-auxiliary stranding, in contrast to actor voice—follows naturally from this analysis. This is shown in (38), which demonstrates a quantifier float asymmetry between AV (38a) and the two constructions under discussion (38b–c), paralleling the pattern described in Section 1.10

A further asymmetry in East Javanese quantifier float supports our analysis: the OV construction and the di-construction—but not AV—permit quantifier stranding in postverbal position, as in (39a–c). This follows from our analysis: in AV, the external argument is base-generated in preverbal position and undergoes A-movement, leaving no postverbal position available for stranding.11

There is independent evidence that the QF patterns discussed here represent genuine cases of quantifier stranding derived from Ā-movement (e.g. Sportiche 1988; Miyagawa 1989; Fitzpatrick 2006). First, as shown in the examples in Section 1 and above, QF in Javanese yields a non-exhaustive (i.e. partitive) reading, a hallmark of Ā-movement-derived quantifier stranding (Fitzpatrick 2006). Second, in postverbal positions across the three voices, the QF pattern follows the expected distribution: numeral quantifier stranding from the pivot is disallowed in AV but consistently permitted in OV and di-constructions. This pattern supports the presence of Ā-movement from the θ-position of the pivot in these constructions.

An immediate implication of this analysis is that the complementary person constraint attributed to the so-called passive and object voice is best understood as a consequence of terminological convention rather than a genuine syntactic distinction. The first- and second-person affixes traditionally described as ergative proclitics in OV, and the affix labeled as passive in the di-construction, are in fact subject clitics (or subject agreement) attached to the verb, as in (40).

Under (40), we assume that the alleged passive in East Javanese is a true transitive and involves no valency-decreasing operations. The fact that the agent/initiator in this construction consistently binds the theme, but not vice versa (section 3.2), follows from this analysis. In this view, di- is not a valency-decreasing affix but a third-person subject clitic encoding the nominative DP in [Spec TP]. This analysis is supported not only by existing descriptions of di- as a third-person affix, but also by the observation that (i) the di- construction is incompatible with first- and second-person agents, as in (41a), and (ii) in the absence of an overt DP/PP, all five native speakers we consulted interpret the construction as having a third-person subject, as in (41b).

We now turn to our proposal that East Javanese possesses an ordinary grammatical subject position, which is a derived A-position probed by [uD] and attracts the highest caseless DP along with nominative case assignment, as schematized earlier in (42) and repeated below. We assume that the prefixal person morpheme in the East Javanese OV construction (which comprises the di-construction) indexes the DP that A-moves to this position.

This proposal differs from traditional analyses of Indonesian-type object voice, which analyze the agent proclitic as an ergative argument that remains in its base-generated position in [Spec VoiceP] (Aldridge 2008; Cole et al. 2008; Legate 2014). We instead argue for a subject analysis, motivated by both empirical and theoretical considerations. First, since promotion to pivothood in East Javanese is a purely Ā-operation, the grammar must make available a derived A-position distinct from the pivot position. Second, the quantifier-floating facts discussed above indicate the availability of such a derived A-position in the actor voice; given that the AV and OV differ only in whether topicalization targets a subject or a nonsubject, there is no principled reason to deny the same position in the OV construction. Finally, binding diagnostics show that the non-pivot agent in OV systematically functions as a binder of the theme, indicating that it occupies a grammatical subject position. Together, these considerations necessitate the postulation of a grammatical subject position structurally lower than the pivot position.12

We now turn to a related question concerning the status of the by-phrase in the di-construction. Following recent analyses of comparable by-phrases in Balinese (Nomoto 2018), we propose that the agent by-phrase is best analyzed as an instance of clitic doubling or agreement doubling—a pattern disallowed for first- and second-person agents/initiators to avoid redundancy. This restriction reflects a broader person-based constraint in Western Austronesian languages, whereby transitive agents encoded through person-indexing on the verb cannot also be realised as full DPs when they are first or second person.13 Consider, for example, the same constraint in the Western Austronesian language Puyuma (43), as well as a parallel restriction in Standard Indonesian (44), where the di-construction disallows first- or second-person by-phrases.

Old Javanese texts suggest that the by-phrase in Javanese was historically a full NP, with the synchronic agent-introducing preposition ambè’ developing from the case marker ni (Poedjosoedarmo 2002). This marker is a regular reflex of the Proto-Austronesian genitive case *ni (Blust 2015) and has been shown to exhibit the distribution typically associated with structural nominative case in conservative Austronesian languages that retain it (Chen 2025; see also Rackowski 2002; Rackowski & Richards 2005 for earlier nominative analyses). This supports the present proposal that the East Javanese pronominal affix indexes the nominative argument, which was originally realized as a full DP and later reanalyzed as an optional by-phrase through the development of the preposition.15

Notably, similar diachronic developments—whereby a case-marked core NP is reanalyzed as an oblique—are attested in other Austronesian languages and are reflected in synchronic dialectal variation. Consider the data from two dialects of Puyuma below. In the more conservative Ulivelivek dialect (45a), non-pivot agents carry the genitive case marker ni; in the more innovative Nanwang dialect (45b), such agents instead bear the oblique marker kan/kana, which typically marks adjuncts, as illustrated in (45c) (Teng 2009). We suggest that the East Javanese di-construction may have undergone a similar development.

To conclude, the East Javanese data motivate an analysis in which the language exhibits a two-way voice system encoding topicalization, with no independent passive construction, as summarized in (46).

We suggest that the absence of a person-indexing proclitic (i.e. subject-indexing verbal affix) in the East Javanese AV construction is the result of an anti-agreement effect (Ouhalla 1993; Baier 2018). This effect is expected, given that the grammatical subject in a subject-topic construction undergoes further Ā-extraction to an Ā-position—an environment in which such effects are predicted to arise. We do not pursue this asymmetry further, as it lies beyond the scope of the present paper.

The present observations from East Javanese raise two further questions. First, can the passive constructions reported in neighboring Austronesian languages—such as Indonesian and Acehnese—also be accounted for under the current analysis? Second, do other Javanese dialects exhibit a similar Ā-oriented pseudo-passive construction that involves no A-movement of the theme to the grammatical subject position?

New comparative data suggest that the answer to the first question is negative. Our primary fieldwork confirms that pivots in Indonesian and Acehnese behave fundamentally differently from those in Javanese—not only in the binding patterns reported in the literature (Section 3.3), but also in their exemption from the definiteness constraint and their incompatibility with PP pivots, consistent with existing analyses of both languages. Consider the examples in (47)–(48).16 Together with the binding patterns indicating that the pivot functions as the subject, these findings reinforce the view that passives in Indonesian and Acehnese involve genuine A-movement to the grammatical subject position (Cole et al. 2008; Legate 2014).

The East Javanese case thus highlights surface-level similarities between object topicalization and passivization in languages lacking morphological case, while also revealing underappreciated variation within so-called Indonesian-type three-way voice systems. These systems are not structurally uniform but instead form a cline of voice systems in flux, in line with recent work highlighting syntactic variation among Western Austronesian languages roughly classified as Indonesian-type (Hemmings 2015; 2021; Himmelmann 2020; McDonnell & Chen 2022; Patrianto & Chen 2023a; inter alia).

Turning to the second question, a comparative examination of other Javanese dialects suggests that the East Javanese pattern described here may be conservative. Data from Tengger Javanese indicate that the di-construction pivot is not subject to a definiteness constraint. Consider (49).

In Yogyakarta Javanese, the theme pivot does not show reconstruction effects and cannot surface as a reflexive bound by the agent, as shown in (50). This contrasts with the binding pattern observed in East Javanese (51), and suggests that the dialect may possess a pattern parallel to Indonesian.

A further locus of variation is observed in Surakarta Javanese, where the di-construction is reported to exhibit no restriction on first- or second-person agents, as in (52), again aligning with the Indonesian pattern. Pesisiran Javanese likewise shows a relaxed person constraint: recent data indicate that the prefix di- may cross-reference a first-person by-phrase, as in (53).

To our knowledge, existing descriptions of these dialects do not address all the diagnostics discussed here, nor the quantifier float pattern; thus, whether the di-construction in these varieties is a true passive remains uncertain and warrants further investigation.

Nevertheless, the coexistence of A- and Ā-oriented di-constructions in Javanese suggests a possible derivational relationship between the two, and there is strong evidence that the East Javanese voice system is conservative. There is broad consensus in the Austronesian literature that Indonesian-type voice systems are more innovative than Philippine-type systems (Himmelmann 2002; McDonnell & Chen 2022), the latter of which can be traced back to Proto-Austronesian or a stage immediately following its split (Ross 2009). The binding pattern observed in East Javanese passive and OV constructions is consistent with patient voice constructions in Philippine-type languages such as Malagasy, Tagalog, and several Formosan languages. These languages feature a non-demoted agent and a theme pivot that exhibit the same binding behavior as the East Javanese passive/OV constructions (Tagalog: Rackowski 2002; Malagasy: Pearson 2005; Formosan languages: Chen 2017). The Tagalog examples in (54) illustrate this pattern: the pivot does not serve as a new antecedent for anaphors and is interpreted in its θ-position.

Second, the third-person constraint observed in the East Javanese di-construction—complementary to the person constraint in object voice, which is indexed by first- and second-person prefixes—suggests an economy-driven pattern that is likely to reflect an earlier system. Similar analyses have been proposed for Balinese and Malay/Indonesian, where the passive marker is argued to have originated as a third-person agent affix that was subsequently reinterpreted as a general passive marker (Van den Berg 2004; Arka 2008; Artawa 2013; Nomoto 2018; see also references therein).

Taken together, these observations suggest that many Javanese varieties may be shifting toward an Indonesian-style A-oriented voice system with a true passive, while East Javanese represents a more conservative variety that preserves Philippine-type syntax. East Javanese thus contributes to the growing literature on syntactic variation among Austronesian languages traditionally classified as Indonesian-type (see Ross 2002; Kaufman 2009; McDonnell & Chen 2022 for overviews showing that western Indonesian Austronesian languages do not form a coherent group, but instead exhibit recurrent patterns likely resulting from parallel innovations and language contact), and underscores that these voice systems should not be treated as a homogeneous group.

We have examined an understudied construction in East Javanese that has traditionally been analyzed as a passive, motivated by its optional by-phrase and typological similarities with true passives in neighboring Austronesian languages such as Indonesian and Acehnese (e.g. Cole et al. 2008; Kroeger 2014; Legate 2014). Six diagnostics instead show that the construction involves Ā-movement of the theme to a left-peripheral topic position, rather than A-movement to subject position, with the agent remaining syntactically active as the grammatical subject in a derived A-position. Its structural parallels and complementary person constraints with object voice further support a unified object-topicalization analysis, revealing that East Javanese exhibits a two-way, Ā-oriented voice system distinct from that of its neighboring languages. Together with the syntactic variation of the di-construction across Javanese dialects, these findings highlight the importance of distinguishing passivization from topicalization in languages without overt case marking and provide new evidence for a possible diachronic pathway linking Philippine-type and Indonesian-type voice systems.

The present analysis points to a potential diachronic pathway in which true passives develop from object topicalization through the reanalysis of topic-into-subject (Givón 1976; Comrie 1988; Shibatani 2011; inter alia). This grammaticalization pathway is attested in several language families (Indo-European, Pama-Nyungan, Japonic; see Shibatani 2011) and has recently been proposed for Western Austronesian languages (Patrianto & Chen 2023a; Lohninger & Katochoritis 2025). Further systematic investigation of the di-construction in other Javanese varieties and neighboring Austronesian languages, using the six diagnostics discussed in this paper (55), will help clarify this development.