Differential object marking in Western Malayo-Polynesian symmetrical voice languages

2.1 Major morphosyntactic distinctions

Several strategies for the differential marking of direct objects have been described in the literature. They include case marking (e.g., Iron Ossetic, Yaghnobi, Turkish), marking by an adposition (e.g., Spanish, Hindi, and several Pamir languages including Sanglechi), clitic doubling or pronominal reduplication (e.g., Romanian, Catalan), and agreement (e.g., Swahili). When the differential marking occurs on the verb, as in cases of clitic doubling, it is also common to speak of differential object indexing (DOI; Compensis 2022; Haig 2018). DOI occurs in some Austronesian languages, especially in the Oceanic branch and in southern Sulawesi (see Iemmolo 2011: 203–206 and Appendix 1), but we are only concerned with DOM proper in this article.

The two most common marking strategies are DOM by means of case marking and DOM via adpositions. For instance, in Parachi the preposition ma marks specific direct objects (Kieffer 2009: 699), non-specific direct objects remaining unmarked, see example (1). Otherwise, ma marks a) a specific location, time, and goal (e.g., ma dur ‘at the river’, ma dōwās ‘at 12 o’clock’, ma [žū qāter] suwār nhōšt ‘he sat down on/mounted [a mule]’), or b) recipients/addressees (as in [ma guγōn] āo dhaym ‘I shall give water [to the cows]’).

DOM via case marking is found, e.g., in (Iron) Ossetic, illustrated in (2).

Even though DOM is robustly attested for many modern Iranian languages, the two largest Iranian languages besides Persian, viz. Pashto and Kurdish, do not have DOM.

Iemmolo (2013), building upon de Hoop and Malchukov (2007: 1640, 2008), distinguishes between symmetric and asymmetric strategies in DOM (see also Chappell and Verstraete 2019). A symmetric DOM alternation involves two or more different overt markers (e.g., a genitive and an accusative case marker), whereas asymmetric DOM alternates between unmarked and marked NPs, as in examples (1) and (2) above. Example (3) shows a symmetric DOM alternation (direct object in bold once in accusative, once in genitive case) in Young Avestan.

Turning to the final point of this section, manifestations of DOM differ in accordance with the part of speech of the direct object. The distinction between (personal) pronouns, personal names, and common nouns is of major importance in this regard, with personal names sometimes aligning with common nouns, sometimes with pronouns, as further illustrated in Section 5. In DOM systems, the choice between two different constructions for direct objects is typically restricted to common nouns, as seen in the preceding examples. Pronouns typically occur only in one of these two constructions, often receiving the same marker as the subset of specific or animate common nouns. For instance, in Munji the preposition va serves as DOM marker, cf. (4b). Whereas the use of this preposition only occurs with definite common noun objects (cf. non-specific čoy in (4a), it is obligatory with personal pronouns. The preposition and pronoun have in fact become inseparable (va + ta ‘you’ > fta), cf. (4c).

In cases such as Munji, the same marker va is used for one subset of direct objects (i.e. pronouns and definite common nouns) while other direct objects remain unmarked. Importantly, however, the identity of formal marking across different parts of speech is not necessary for a DOM system. The essential feature of a DOM system is that at least one subclass of nominal expressions in direct object function differs in its overt marking from other subclasses (see also Sinnemäki’s 2014: 283 definition of restricted case marking and Bickel et al.’s 2015: 18–19 very detailed list of potentially relevant subclasses). With regard to marking variation across parts of speech, two options are common in addition to the one illustrated by Munji above.

The first option is to use different markers for common nouns and pronouns. Semnani, for example, exhibits an alternation between an unmarked (basic) and a marked (oblique) form of the 3rd person pronoun (e.g., ü ‘he/she’ vs. žö ‘he.obl’/žin ‘she.obl’, Bossong 1985: 31). Common nouns also distinguish a basic and an oblique form, but the actual forms are quite different. Pronominal obliques make use of a former preposition (ž- < až < *hača), while specific common noun obliques are marked with the suffix -in, as shown in example (5a). Non-specific common nouns remain unmarked, as shown in (5b).^[3]

The second marking option is for common nouns to show no special marking in object function, while personal pronouns are marked in a specific way when functioning as direct objects. This appears to be the case in Semnani’s close relative Lâsgerdi, where only pronouns have distinct forms when in object function, as seen in (6). The oblique form once again involves the fusion of a former preposition and a pronominal stem, but this is not crucial for the assessment of DOM. What is crucial is that pronominal direct objects behave differently from common noun direct objects.

As we will show in the following sections, the kinds of morphosyntactic differences in DOM illustrated here with Iranian languages are also found in WMP languages.

2.2 A brief note on the functions of DOM

Animacy and specificity/definiteness are usually thought to be the two main dimensions relevant for the occurrence of DOM. These dimensions are typically presented in the form of implicational scales, as in I and II, where occurrence of DOM for a particular expression type implies DOM for all expression types to its left.

1. Animacy scale:

   First/second person > third person pronoun > personal name > human > animate > inanimate

2. Specificity/definiteness scale:

   Pronoun > personal name > definite > specific > non-specific

   It appears to be the exception rather than the rule that only one dimension is relevant in a particular DOM system. In most systems, objects are differentiated with regard to both animacy and definiteness. In the following, we simply refer to the specificity scale, but this is to be understood as including both the animacy and the specificity/definiteness scales above.

   Two further scales of DOM-triggering factors can be represented as in III and IV, following Bárány and Kalin (2020: 2), based on Dalrymple and Nikolaeva (2011) and Næss (2004), respectively.

3. Information structure:

   Topic > Non-Topic

4. Affectedness:

   Affected > Unaffected

The exact cut-off point along the more complex scales is variable even among closely related languages and thus highly language specific, as is illustrated by the following example set in (7) from the two genetically closely related Pamir varieties Bartangi and Roshani. In both varieties, definite human arguments are regularly marked by the preposition az/as, but this also holds for non-human animates (animals) only in Bartangi.

Similar microvariation is observable in other languages and language families (e.g. Iemmolo 2011: 156–160 on Bantu languages, pp. 237–245 on different varieties of Italian).

The functions of DOM will not be further discussed in this paper as the sources for the languages investigated here do not provide much information in this regard. When using (in)definite or (non-)specific to characterize the referential meaning of a nominal expression, this is usually based on the translation given for a particular (set of) example(s) and does not reflect a proper semantic or pragmatic investigation.

4.1 Symmetrical voice

Voice is generally considered a mechanism that selects one semantic argument over others to become the privileged syntactic argument (PSA or subject) of the clause (Shibatani 1988; Zúñiga and Kittilä 2019, among many others). In most of the world’s languages, voice alternations go hand in hand with a decrease in transitivity. That is, the alternation between an active and a passive construction involves the demotion of the actor argument, and the alternation between an ergative and an antipassive construction involves the demotion of the undergoer argument. Symmetrical voice alternations as found in many WMP languages (cf. Foley 1998, 2008; Himmelmann 2005; Riesberg 2014), on the other hand, are alternations that link different semantic arguments to subject function without demoting the non-subject argument. Consequently, these languages exhibit two (or more) basic transitive constructions which are usually called actor voice (av) and undergoer voice(s) (uv) in the current literature. In AV, the subject of the construction bears the agent role, and the verb will be marked by AV morphology. In UV, the subject is an undergoer-like argument (patient, theme, etc.), which in turn will be signaled by UV morphology on the verb. Example (8) illustrates this for Totoli, a WMP language of Sulawesi, Indonesia. In (8a) the subject (the third person singular pronoun isia) is the agent, and the verb carries the AV (realis) prefix no n-. In (8b), the verb is marked by the UV (realis) prefix ni-, and the subject of the construction is the undergoer argument saginna ‘his/her banana’.^[8] In both examples, subject arguments and voice morphology are marked in bold in glosses and translation. Importantly, the non-subject arguments (saginna ‘her/his banana’ in (8a) and kami ‘we’ in (8b)) remain direct core arguments, which are generally restricted to immediate postverbal position, and both constructions are fully transitive.

The languages of the Philippines and adjacent areas of Indonesia usually exhibit two or three semantically distinct UVs that differentiate more fine-grained semantic roles of their subject arguments. Example (9) shows a symmetrical alternation between an AV construction in (9a) and three UV constructions in (9b–d) in Tagalog (again, subject arguments and the voice morphology are highlighted in bold). The three UVs differ in whether their subject is a patient/theme in (9b), a locative in (9c), or a benefactive in (9d). For languages like Tagalog, it is thus common practice to distinguish between patient voice (pv), locative voice (lv), and conveyance voice (cv) which are all subsumed under the more general umbrella term undergoer voice.

As the comparison of the Totoli and Tagalog examples reveals, the two languages do not only differ in their number of UVs. They also exhibit different argument realization strategies. In Totoli, core arguments are unmarked, whereas in Tagalog every noun in core function is preceded by a case marker (i.e., nominative ang, genitive ng, or dative sa in (9)). There is thus considerable variation regarding the formal and functional properties of the symmetrical voice systems attested in WMP languages. In the specialist literature, it is common practice to distinguish two basic types of WMP symmetrical voice languages. The “Philippine-type” languages behave like Tagalog and are characterized, inter alia, by more than three symmetrical voice alternations and overt case marking proclitics. The “non-Philippine-type” languages typically only have two productive voice alternations and lack overt case marking. The latter group is essentially negatively defined, reflecting the fact that it is much more heterogeneous than the former one. Non-Philippine-type languages have also been called “Indonesian-type languages”, but this term has recently been used in a number of different senses, so that “non-Philippine-type” is more precise (Himmelmann 2005: 111–114; McDonnell and Chen 2022: 13). The remainder of this section deals only with Philippine-type languages, as they provide the clearest evidence for those properties of symmetrical voice systems that impact the analysis of DOM.

In the preceding exposition of the basic features of symmetrical alternations we use the term ‘subject’ which, however, has been the object of a longstanding controversy in Philippine linguistics. In his influential paper, Schachter (1976) summarizes a debate that had already started in the late 1950s and asks, whether the Philippine subject is “topic, actor, actor-topic, or none of the above?” One common view at the time was that the nominative phrase constitutes the topic of the clause, while the actor phrase is always to be analyzed as subject. Under such an approach, alternations like in (9) do not involve a change of grammatical relations and consequently were not considered voice alternations. Instead, the term “focus” was adapted (see Blust 2002: 64) and constructions like (9a–d) were referred to as “agent focus”, “patient focus”, etc. Other terms used in the Austronesian literature are (verbal) “case”, “topicalization”, and “trigger” (see Blust 2002: 73–74 for an overview). Today, mainly following the arguments presented by Kroeger (1993a, 1993b), the common view in the Austronesian specialist literature is that the nominative marked argument constitutes the privileged syntactic argument and that it can indeed be called “subject” (Chen and McDonnell 2019: 117–118; Himmelmann 2005: 152–157; Kroeger and Riesberg 2024).

The bewildering diversity of analytical and terminological proposals does not only pertain to the nature of the voice alternations and the verbal morphology. It also extends to nominal marking. The terminological variation in the literature here is even worse than that for the voice alternations. According to Blust (2015: 437–39), there are (at least) 42 different terms for what we call “case markers” here, including “articles”, “attributive and object markers”, “auxiliary nouns”, “introducers”, “noun class markers”, “orienters”, “specifying nouns”, or “topic/goal markers”. It should be obvious that such heterogeneity makes for a major stumbling block when trying to include data from WMP symmetrical voice languages in a typological database. Non-experts hardly stand a chance of being able to distinguish between substantial grammatical differences and more superficial terminological differences. In the present work, we follow the currently most widely used terminology and speak of “case markers” (see Himmelmann 2021 for further discussion).

4.2 What is, and what is not, DOM in WMP symmetrical voice languages

In the previous section, it was shown that symmetrical voice alternations per se pose a number of analytical and conceptual challenges which have resulted in several competing analyses and terminologies which, in turn, make it difficult to identify related constructions across different languages. Here we address the issues that specifically pertain to the identification of DOM constructions.

When investigating DOM in languages that exhibit more than one basic transitive construction, the question arises which of these should be included in the analysis. None of the studies we scrutinize in this paper explicitly state which basic transitive construction(s) entered their typological sample. Neither Sinnemäki (2014) nor Witzlack-Makarevich et al. (2012) provide any information on this topic. Iemmolo (2011), too, does not explicitly state which constructions type(s) he included in his study. The examples from symmetrical voice languages he presents, however, are exclusively AV constructions.

In our investigation we only consider AV constructions to involve DOM for the following reasons. First, in UV the more patient-like argument occurs in subject function but DOM pertains to objects (i.e. core arguments that are not in subject function). Second, in AV the more patient-like argument fulfills the definition for direct objects given in Section 2, i.e., it is the more patient-like core argument of a prototypical two-place transitive predicate, the morphosyntactic marking of which differs from the marking of the single core argument of an intransitive predicate (which always occurs in the nominative ang-form). Third, there are different marking options for direct objects in AV. To illustrate this, we first have to take a closer look at case marking.

Case markers in Philippine-type languages usually come in two paradigms – one for personal names and one for common nouns – and they often distinguish three different cases, usually labelled “nominative”, “genitive” and “dative/locative”. Additionally, personal pronouns often exhibit distinct forms for each case. Table 2 shows the basic case markers and the three case forms of the first person singular in Tagalog.

The default pattern of case marking and voice alternations of a Philippine-type symmetrical voice language is as follows: Each verbal clause^[10] must contain one nominative argument and a voice marker on the verb marking the semantic role of the nominative argument in the clause (cf. the examples in (9) above). All other NP constituents of the clause are formally marked as genitive or dative/locative case, depending on their semantic role: genitive ng occurs with actors and experiencers in UV, and with patients, themes, and instruments in AV; dative/locative sa occurs with goals, recipients, locations, as well as with temporal adjuncts. The dative personal name marker kay is mostly used for recipients (and DOM).

As we will see in Section 5, both the formal and the functional properties of the case marking system are variable across Philippine-type languages, sometimes with important implications for DOM. While our sources use a wide range of labels for the case forms, we will generally speak of nominative, genitive and dative case forms, as just defined. In addition, we will use the label oblique for case forms that combine genitive and dative functions and thus occur with the full range of non-subject arguments, including agents, patients, themes, instruments, recipients, goals, and locations.

The different case forms for common nouns, personal names, and pronouns, as illustrated in Table 2 for Tagalog, do not per se constitute an instance of differential argument marking. That is, the mere fact that case marking for common nouns takes another form than case marking for personal names (e.g., genitive ng for common nouns vs. genitive ni for personal names) is not DOM. What does count as DOM, however, are those instances where nominal expressions of different classes are marked by different cases in the same grammatical context. This is illustrated in (10), where the common noun in (10a) receives genitive case, and the personal name in (10b) receives dative case.

Our definition of DOM, of course, also includes cases where a given noun class allows for alternating case markers in the same construction, as illustrated in (11). Here, both sentences contain the non-subject undergoer argument, dagaˈ ‘cat’, but this is either marked by genitive ng, as in (11a), or by dative sa, as in (11b). As reflected in the English translation, the differential use of the dative (instead of the genitive) evokes a difference in definiteness, in that dative marked phrases are unambiguously definite, while the genitive marked undergoer can be understood as either definite or indefinite (specific).

In concluding this section, a final note on the possibility of DOM in undergoer voice constructions is in order. Strictly speaking – taking the symmetry of the system seriously – actors in UV constructions such as (8b) and (9b–d) are direct objects in that they are non-subject core arguments of a transitive predicate marked differently from the single core argument of an intransitive predicate. But they are not the patient-like core argument, and hence any variation in the marking of the agent argument in object function (which in fact occurs in some WMP languages) would not count as DOM.

5.1 Type I: DOM for pronouns, personal names, and common nouns

Type I DOM is particularly common in Philippine-type languages. Since in general all common nouns and personal names in these languages are overtly case marked, Type I DOM is usually of the symmetric type, as illustrated in Section 5.1.1. However, there are, much more infrequently, also Type I languages of both the Philippine- and the non-Philippine-type, where nominal expressions are not overtly case marked in all grammatical functions and where, consequently, we find asymmetric manifestations of DOM, as illustrated in 5.1.2. For further details on variation and complications in determining DOM in Philippine-type languages, Appendix 2 discusses the Philippine language Cebuano.

5.2 Type II: DOM in pronouns and personal nouns only (Tboli, Toratán)

In a few languages in our sample (seven in all), DOM only occurs with personal names and pronouns. We briefly discuss two examples, which together give an idea of the range of variation that occurs among Type II languages.

In Tboli, common nouns occur in the unmarked base form in all core argument functions, including non-subject undergoer, as illustrated in (21a) for a non-specific undergoer and in (21b) for a specific undergoer. The latter often, but not obligatorily, are preceded by a demonstrative. Personal names in non-subject undergoer function are preceded by the oblique marker ke, as seen in (21c). Personal pronouns also occur in a special oblique form when in non-subject undergoer function (21d) and also kut in (21a)).

There is a minor exception regarding the pronouns. As in many Austronesian languages, eight distinct person forms occur in Tboli: in addition to singular and plural forms for first, second, and third person, there are also first dual and first plural inclusive forms. Seven of these eight person forms have the special oblique form illustrated above, the exception being first plural inclusive tekuy, which is used in all grammatical functions (Forsberg 1992: 22; Porter 1977: 35–36). The following discussion will show that in Type II and Type III systems of DOM, it is rarely the case that all personal pronouns have the same argument realization options. Similar exceptions for individual pronoun forms are generally not found for Type I systems.

DOM in Tboli is asymmetric as personal names and pronouns occur in a basic unmarked form that is found in most grammatical contexts, the oblique usage illustrated above representing the marked case.

A symmetric example of Type II DOM is attested in Toratán (Ratahan). Here, personal names and pronouns generally occur in one of three case forms: nominative, genitive and locative (Himmelmann and Wolff 1999: 31). These cases are overtly marked by proclitics (i=, ni=, and si=, respectively), except for the genitive of the singular pronouns which makes use of special enclitics. In non-subject undergoer function, personal names and pronouns are marked with locative si, as illustrated in (22).

Common nouns always occur in their bare form without case markers when functioning as subjects or as non-subject undergoers regardless of their specificity.

5.3 Type III: DOM for pronouns only (Kimaragang, Indonesian)

DOM for pronouns only occurs in two different variants, one found primarily in Philippine-type languages, the other in non-Philippine-type languages. We begin with the former, using the Borneo language Kimaragang as our main example.

Basic clause structure in Kimaragang is similar to Tagalog and other Philippine-type languages: verbs are marked for symmetrical voice, and core arguments are overtly marked by one of three case proclitics (nominative, genitive or dative). The major difference from Tagalog is that all non-subject core arguments, except for recipients, receive genitive marking (Kroeger 2005: 406–407). Example (23a) illustrates genitive marking for an indefinite non-subject undergoer, (23b) for a definite non-subject undergoer, and (23c) for a personal name in non-subject undergoer function.

Only pronouns in non-subject undergoer function require dative marking, as seen in (24).

Turning to non-Philippine-type languages showing Type III DOM, there are two crucial differences from the Kimaragang case just discussed. First, non-Philippine-type languages generally do not have overt case marking for common nouns in core argument functions (cf. Section 4.1), and the rare instances of overt case marking for personal names are restricted to the genitive. Second, DOM is not only restricted to personal pronouns in general, but rather more specifically to singular pronouns, and in some cases even to just the singular third person pronoun. Our main example is Indonesian, the standard variety of Malay spoken in Indonesia.

Indonesian generally does not show evidence for DOM. Pronouns, personal names, and common nouns in non-subject undergoer function usually occur in their base form, which is also used in practically all other grammatical functions such as subject, complement of preposition or possessor. In example (25), the first person singular pronoun aku ^[14] is seen in AV actor, AV undergoer and possessor functions, respectively.

However, singular pronouns (first, second, and third person) also have a short, clitic form which primarily occurs in possessor function, as in (26).

This short form also occurs in non-subject undergoer function (and after a small number of prepositions). As a consequence, there are two alternative constructions when the undergoer in an AV construction is a singular pronoun, one with a full pronoun and one with a clitic pronoun, as seen in (27).

There are thus two different marking options for pronominal non-subject undergoers, which formally constitutes an instance of DOM.

The functional difference between the two constructions, however, is not well understood. Sneddon et al. (2010) offer a few remarks on the use of the third person singular clitic =nya, which has a broad range of functions in Indonesian grammar, going well beyond the ‘normal’ scope of a possessive pronoun (e.g., use in nominalizations and for optional definiteness marking in NPs). Two points are of particular interest in the current context. First, unlike the third person singular free form dia, the clitic =nya may refer to inanimates, as shown in (28).

This usage shows that =nya appears to be more grammaticalized than its free counterpart dia. Second, example (28) also illustrates one typical usage context for the clitic, called “object topic-comment clause” by Sneddon et al. (2010: 290). Another very common usage is in object relative clauses (cf. also Arka 2021), as seen in (29).

Given these observations, it seems likely that the difference between the two constructions may relate to the topicality and/or prominence of the undergoer. In the clitic construction, the antecedent for the pronoun is typically mentioned in the immediately preceding context. Its referent is thus typically highly activated and discourse prominent. DOM here may thus be of the information structural type briefly mentioned in Section 2.2.

While details vary, similar alternations of full and clitic forms in non-subject undergoer function are attested across a range of non-Philippine-type languages. Totoli, a language spoken in Sulawesi, shows essentially the same distribution as Indonesian. In West Coast Bajau (northern Borneo), the alternation between free pronoun and genitive clitic for non-subject undergoers is only available for third person singular (obligatorily for 3rd person inanimate undergoers, optionally for animates). In Besemah (southern Sumatra), the alternation is also only found in the third person. In Karo Batak and Toba Batak (northern Sumatra), the genitive clitic is the only available option for third person singular non-subject undergoers. Strictly speaking, West Coast Bajau, Besemah and the two Batak languages thus constitute counterexamples to the specificity scale, as they show DOM (optionally or obligatorily) in third person (singular) pronouns, but not in first and second person.

5.4 Summary

Table 5 provides a comparison of the DOM assessments for WMP symmetrical voice languages in the literature (cf. Table 1 in Section 3) with our findings. The main result is clear: according to our assessments, DOM is found in more than half of the WMP languages listed in this table (13 out of 21), and hence more frequently than reported in the literature. As already noted in Section 3, our results probably still under-report the occurrence of DOM in WMP languages, because the available sources do not always provide sufficient evidence for deciding the issue (in addition to often being difficult to interpret).

The differences between our results and the ones reported in the literature depend of course in part on how DOM is defined, and which DOM-related phenomena are included in a particular database. For example, Type III DOM, which is limited to pronouns, is explicitly excluded by Sinnemäki (2014); hence the differing assessments for Indonesian and Riau Indonesian. Still, we would hold that the major factor underlying the differing assessments relates to the symmetrical voice systems characteristic for WMP languages, which makes it more difficult to identify DOM.

In this regard, it is important to note that all the phenomena that characterize typical DOM systems such as the Iranian ones presented in Section 2 also occur in WMP languages. Most importantly, the fact that the attestations of DOM in WMP languages almost perfectly adhere to the specificity scale provides strong support for the view that we are actually dealing with DOM. Furthermore, many instances of variation are similar to the ones documented for Iranian. For example, many central Philippine languages such as Tagalog and Hiligaynon provide clear evidence for DOM, while closely related Cebuano and Bikol do not (see also map in Appendix 1). Similar observations hold with regard to the types of systems (I, II or III) and their symmetry. Another similarity to the Iranian languages pertains to the fact that DOM forms of pronouns are more often fused when compared to DOM forms of common nouns and personal names. Likewise, some languages use the same DOM marker across different parts of speech; others have special forms for different parts of speech.

A more comprehensive table including all WMP symmetrical voice languages investigated for this study is given in Appendix 1. It supports our claim that DOM is robustly attested among WMP symmetrical voice languages. Philippine-type languages, especially the ones found in the central Philippines, tend to have Type I systems. A major exception is Philippine languages spoken in the northern half of Luzon which often lack DOM, contributing to their repeatedly noted special status among Philippine languages. Non-Philippine-type languages tend to have no DOM, with the conspicuous exception of Malayic varieties, both standard national language varieties and Trade Malay varieties. Otherwise, Type II and Type III DOM systems are primarily found in the transitional area between Philippine-type and non-Philippine-type languages (Mindanao in the southern Philippines, northern Borneo, and northern Sulawesi; see the map in Appendix 1).