Tense-aspect conditioned agent marking in Kanakanavu, an Austronesian language of Taiwan

2.1 Morphological voice marking

Kanakanavu differs from prototypical Philippine-type languages in its voice system as it exhibits a binary voice opposition, instead of a more commonly found four-way voice distinction.^[2] As can be seen in (1), a semantically transitive event such as SEE, which involves both an agent (semantic experiencer) and a patient (semantic stimulus), can be packaged in either the agent-voice or patient-voice construction.^[3] In the agent-voice construction, the clause-initial predicate position is occupied by an agent-voice verb, such as cʉmʉ´’ʉla /c<um>ʉ’ʉla/ ‘see (agent voice)’ in (1a). In the patient-voice construction, a patient-voice verb is involved instead, such as cʉ’ʉ´le /cʉ’ʉla-i/ ‘see (patient voice)’ in (1b).^[4]

There is one crucial morphological difference between agent-voice and patient-voice verbs in Kanakanavu: agent-voice verbs may or may not be morphologically marked for their voice status, whereas patient-voice verbs are always morphologically marked. An example of an agent-voice verb marked by an agent-voice marker is the verb cʉmʉ´’ʉla /c<um>ʉ’ʉla/ ‘see (agent voice)’ shown in (1a), where the agent-voice marker <um> is involved. Depending on the specific morphological verb class, agent-voice verbs may be marked by other agent-voice markers, such as prefix m- or prefix mu-, or not involve any voice marker at all. Examples of some agent-voice verbs in the language are listed in Table 1.^[6]

Examples of some patient-voice verbs in Kanakanavu are provided in Table 2. In addition to always being marked for voice, patient-voice verbs are also always marked for tense-aspect-mood. All the examples included below are indicative verbs occurring in the perfective, where the perfective patient-voice marker -ai is involved^[7] (reducing to -i when combining with some stems ending in the low vowel /a/).^[8]

Furthermore, every patient-voice verb has a morphologically related agent-voice counterpart. The morphological relationship can be symmetrical. In this case, a given verb stem combines with a patient-voice marker to form a patient-voice verb, and the same stem can be marked with an agent-voice marker to form the agent-voice counterpart of the verb. For example, the patient-voice verb íme /ima-i/ ‘drink (patient voice perfective)’ involves the (perfective) patient-voice marker -ai attached to the stem ima; its agent-voice counterpart míma /m-ima/ ‘drink (agent voice)’ involves the same stem, but is marked by the agent-voice marker m- instead. On the other hand, the morphological relationship may also be asymmetrical: a patient-voice verb is formed by attaching a patient-voice marker onto a verb stem, and the stem can occur as a (voice-unmarked) agent-voice verb without any overt morphological marking. For example, the patient-voice verb po’ocípe /po’ocip-ai/ ‘cook (patient voice perfective)’ involves the (perfective) patient-voice marker -ai combining with the stem po’ocipi, and the stem can occur as an agent-voice verb on its own without any overt affixation: po’ocípi /po’ocipi/ ‘cook’.^[9] Some more examples of the two types of voice alternation are provided in Table 3.

2.2 Lexical effects on voice alternation

A well attested feature of western Austronesian voice systems is that not all verbs can undergo voice alternation (e.g., Starosta 2002, Huang 2005), and Kanakanavu is no exception. Generally speaking, in Kanakanavu, voice alternation is possible for verbs that package semantically transitive events, where both an agent and a patient are involved in the verb semantics, as has been exemplified in (1).^[10] Another example of semantically transitive event packaged in either an agent-voice clause or a patient-voice clause is provided in (2). Here, the verb stem ala ‘take’ is involved, and the voice alternation is between the agent-voice verb umâla /um-a-ala/ ‘take (agent voice imperfective)’^[11] and its patient-voice counterpart in the perfective alái /ala-i/ ‘take (patient voice imperfective)’:

Still another example can be seen in (3), which involves the verb stem vua ‘give’ derived as the agent-voice verb móvua/ mu-a-vua/ ‘give (agent voice imperfective)’ and its patient-voice counterpart vʉ´n ‘give (patient voice imperfective)’:

Many agent-voice verbs that package semantically intransitive events, such as ’ʉmʉ´can /’<um>ʉcanʉ/ ‘rain (agent voice)’ and musúcaru /musucaru/ ‘slip’, do not show patient-voice counterparts. Some semantically intransitive agent-voice verbs, however, do, and the usage of the derived patient-voice verbs would allow for a transitive construal of the event in question. An example is the motion verb mukúsa /m-u-kusa/ ‘go (agent voice)’, whose patient-voice counterpart (in the perfective) is ukúse /u-kusa-i/ ‘go (patient voice perfective)’. As evidenced in discourse, the agent-voice form of the verb is typically used when the semantic goal is an inanimate location (4a), while the patient-voice form is typically used when an animate goal is involved instead (4b). In the latter case, the verb is typically interpreted as ‘approach’, suggesting that the animate goal is construed as a patient involved in a transitive event.

The agent-voice verb mácaca /maacaca/ ‘laugh’ also shows a patient-voice counterpart, whose imperfective form is pacacʉ´n /paacaca-ʉn/ ‘laugh (patient voice imperfective)’.^[12] As can be seen in (5a), whereas usage of the agent-voice verb implies that only a semantic experiencer (i.e., “laugher”) is involved, the interpretation of its patient-voice counterpart, as in (5b), clearly shows that the packaged event (laugh) is presented as being directed at an additional patientive participant (the one being “laughed at”).

Importantly, whereas voice alternation is possible for the aforementioned two cases, this may not be the case for other agent-voice verbs with similar semantics, where considerable lexical idiosyncrasies are observed. As can be seen in Table 4, a patient-voice form is found for the motion verb – ivátu ‘come’ – but not for the verbs mucán ‘leave (agent voice)’ or móca ‘walk (agent voice)’, both of which indicate motion; furthermore, agent voice is also the only possibility for the stems tangi ‘cry’ and macai ‘die’, whereas the stem mati’avasʉ ‘stick one’s tongue out’ can derive as both agent-voice and patient-voice forms.

3.1 The PSA

Different from prototypical Philippine-type languages, where PSA status^[13] is indicated by case markers (Himmelmann 2005, Chen and McDonnell 2019), in Kanakanavu PSA status of a phrase is morphologically marked only when it occurs as a pronoun.^[14] Pronominal PSAs obligatorily occur in what is referred to as the unmarked case in this article.^[15] The case forms across all person-number categories are listed in Table 5. Unmarked-case pronouns in Kanakanavu fall under two categories – free/independent and bound/clitic – except in the third person, where only free pronouns are found. Note that there are interchangeable forms across free and bound pronouns in the first-person singular.^[16] In the first-person plural, there is also a distinction made between inclusive and exclusive forms, a phenomenon widely found across Austronesian languages.^[17]

As in other western Austronesian languages, a major function of voice in Kanakanavu is to map different semantic roles onto the PSA of the clause. In patient-voice clauses, it is always the patient that serves as the PSA. This can be seen in the patient pronouns in (6) occurring in the unmarked case.^[18]

In agent-voice clauses, on the other hand, the semantic role depends on whether the verb encodes a semantically intransitive or transitive event. When a semantically intransitive event is involved, the single participant involved serves as the PSA, and the semantic role can vary: in (7a), the PSA represents a semantic experiencer, and in (7b), it represents a semantic theme. If an agent-voice clause encodes a transitive event, then the PSA always assumes the agent role, as can be seen in the pronouns in (7c–d).^[19]

The PSA is the highest ranked grammatical relation in the clause. Syntactically, it also represents the only phrase that can be relativized on. Therefore, if a head noun of a relative clause assumes the patient role, the subordinate verb occurring in the relative clause is restricted to occur in patient voice, as in (8a). If the relativized participant assumes the agent role, then the relative clause always occurs in the agent-voice construction, as in (8b).^[20] The latter pattern aligns syntactically with relative clauses which package semantically intransitive events, where agent-voice verbs are used and the single participant is relativized on (8c).

3.2 The non-PSA agent and other non-PSA participants

In addition to the PSA, a specific type of non-PSA grammatical relation in Kanakanavu deserves special attention – the non-PSA agent in patient-voice clauses. In many western Austronesian languages, the non-PSA agent is consistently case-marked (often in the genitive case, see De Guzman 2000). In Kanakanavu, however, the non-PSA agent occurs in two alternating cases when occurring as a pronoun, which is conditioned by the tense-aspect value of the clause.

Tense-aspect is highly grammaticalized in Kanakanavu: all indicative verbs show a perfective form, which contrasts with at least one of two non-perfective forms: the imperfective and the perfect. In patient voice, all perfective verbs are marked by the perfective patient-voice marker -ai. Patient-voice verbs also always show an imperfective form, which is marked by the imperfective patient-voice marker -ʉn. Some patient-voice verbs lack a perfect form, but for those that do show a perfect form the perfect patient-voice marker ni- is involved.^[21] Table 6 includes examples of indicative patient-voice verbs and their tense-aspect forms. Notice here that the verbs that lack perfect forms appear to have rather stative, less dynamic semantics. Section 4.2 will return to this phenomenon when discussing how the non-PSA agent is interpreted in the usage of stative patient-voice verbs.

Tense-aspect marking typically corresponds to different types of temporal construals on the packaged event. The different temporal construals can be exemplified by elicited patient-voice clauses involving the stem kʉʉnʉ ‘eat’ as in (9). First, the perfective–imperfective distinction is based on whether explicit reference is made to the internal temporal structure of the event (Comrie 1976). The perfective (kʉ´ne /kʉʉn-ai/) serves to indicate that no such reference is made, where the event is simply presented as having occurred (9a). In contrast, the imperfective (kʉ´nʉn /kʉʉn-ʉn/) presents events characterized by different temporal structures, such as progressive, habitual, or future events; in (9b), it serves to indicate the latter two. Similar to the perfect in a variety of languages, the perfect in Kanakanavu may fall under several semantic types. The example in (9c) (nikʉ´n /ni-kʉʉnʉ/) serves to indicate either a continuing state resulting from a prior event (i.e., the “stative” perfect) or an event that had happened not long ago from speech time (i.e., the “hot news” perfect) (Ritz 2012). While it is worth further investigating the functional differences across the tense-aspect categories marked on the verb, for the purpose of this article the discussion that follows will focus on the agent that occurs across the three patient-voice clause types when realized as a pronoun. In the perfective (9a), it occurs in the oblique case. In non-perfectives (9b and c), it occurs in the genitive case.^[22]

Different from the unmarked-case pronouns, all oblique pronouns are free/independent. Their forms across person-number categories are listed in Table 7. Note here that in the third person category, there is no further distinction between singular and plural, which is nevertheless present in the unmarked-case pronouns.

Importantly, the oblique is also the obligatory case form for non-agent event participants that are also not serving as the PSA. Some examples can be seen in (10), where the oblique pronouns may represent a range of non-agent roles, including a semantic goal (10a) or a semantic stimulus (10b) in agent-voice clauses. An example of a semantic recipient occurring as an oblique pronoun in patient voice is shown in (10c).^[23]

The genitive case forms across person-number categories are listed in Table 8. Similar to the unmarked-case pronouns, a mix of bound and free pronouns can be found in the genitive case. Similar to the oblique pronouns, there is also no further distinction made between singular and plural in the third person. As can be seen in (11), the genitive case may also be used for representing possessors in the nominal domain.^[24] Again, this is only found when possessors are realized as pronouns.

The oblique case-marking of the pronominal non-PSA agent in the perfective may suggest that it aligns in syntactic status with other non-PSA participants which are also non-agents, since the latter also occurs in the oblique case when realized as pronouns, as has been shown in (10). However, at least one piece of evidence exists to show that the two are syntactically distinct: only the former can serve as the obligatory controller of an omitted agent in a complement clause.^[25] In (12a), for example, the third-person oblique pronoun ’inía, which realizes the non-PSA agent of the matrix verb alái /ala-i/ ‘take (patient voice perfective)’, serves as the controller of the omitted agent in the complement clause that follows. In other words, the omitted agent in the complement clause is obligatorily interpreted as being coreferential with the non-PSA agent (’inia ‘they’) in the matrix clause. In contrast, the oblique second-person singular pronoun kasúa in (12b), which is a non-PSA recipient in the matrix clause, cannot serve as the controller of the omitted agent in the complement clause. Notice here that the complement verb is causativized (’apakʉ´n /’apa-kʉʉnʉ/ ‘make eat’) so that the agent in the complement clause (assuming the causer role) is coreferential with the PSA (first-person singular pronoun =ku), instead of the non-PSA oblique recipient (second-person singular pronoun kasúa) in the matrix clause.

As suggested earlier, the PSA can also serve as an obligatory controller in control constructions. Another example can be seen in (13a), where the PSA is realized as a lexical noun phrase kana’uá=maku ‘my brother’. Control is also a property shared by the non-PSA agent in non-perfective clauses, although it occurs in the genitive case when realized as a pronoun. This can be seen in (13b), which involves the third-person genitive pronoun ké realizing the non-PSA agent.

In this regard, at least two types of grammatical relations can be identified in the Kanakanavu verbal clause based on case-marking and other syntactic properties, as summarized in Table 9. The PSA is the most well-defined: it is consistently case-marked in the pronominal domain, represents the only participant of the clause that can be relativized on, and may control an omitted agent in a complement clause. The agent in patient-voice clauses is a non-PSA participant that nonetheless can serve to control an omitted complement-clause agent. It is, however, case-marked differently when occurring as a pronoun depending on the tense-aspect value of the clause: oblique in the perfective but genitive in non-perfectives. Finally, other non-PSA participants lack any of the syntactic properties described previously, so whether they represent an identifiable type of grammatical relation in Kanakanavu requires further investigation.^[26] They are non-PSA in the sense that they cannot be relativized on; they are also crucially differentiated from the non-PSA agent in that they do not control omitted agents in complement clauses.^[27]

Two interesting mismatches are therefore identified between case-marking and behavioral properties of participant types in Kanakanavu. On the one hand, both the perfective non-PSA agent and other non-PSA participants are marked in the oblique when realized as pronouns; however, only the former can serve as the obligatory controller. On the other hand, although the pronominal non-PSA agent is case-marked differently depending on verbal tense-aspect marking, control is a property that is unanimously shared whether it is the oblique/perfective agent or the genitive/non-perfective agent that is involved.

4.1 Dynamic verbs and referential prominence of the agent

The two types of non-PSA agent exhibit one crucial difference when dynamic verbs are involved, which is concerned with how they are referentially interpreted when left unexpressed. This is a property that is rather elusive when only elicited data are considered, but can be shed light on by data from spontaneous speech.

To begin with, non-perfective patient-voice verbs are commonly observed to be used in discourse with the non-PSA agent omitted from mentioning. In this case, the omitted non-PSA agent is generally interpreted as having generic reference. In (16a), for example, the imperfective patient-voice verb alémin /alooim-in/ ‘forget’ is used to indicate that the PSA – tapcaláke ‘the Tapcalake people’ – shall not be forgotten, not by a specific agent (semantic experiencer), but by everyone. Notice here that no explicit (pro)nominal expression is present to indicate the (generic) semantic experiencer. Another example can be seen in (16b), where the perfect patient-voice verb nelupáca /ni-ulupaca/ ‘use’ is again only accompanied by the semantic patient – váva kalávung – with the generic/unimportant agent backgrounded and omitted from mentioning.^[30] In both cases, the agent, if present and pronominal, is expected to occur in the genitive case.

Perfective patient-voice verbs, on the other hand, are almost always used with an overt agent in discourse. When left unexpressed, which very rarely occurs in discourse but can be exemplified in (17), the agent is always understood as a “zero” participant, representing a tracked, topical referent that has been mentioned in prior discourse. The zero agent is represented in the data below with an empty-set symbol to signify that despite being unexpressed, it is still centrally involved in the interpretation of the perfective patient-voice verb.^[31] Thus, a major distinction between the non-perfective and perfective patient-voice verbs concerns whether leaving the non-PSA agent unexpressed implies a backgrounded interpretation. Whereas this is the case for non-perfective verbs, as seen in (16), perfective patient-voice verbs are always interpreted as involving the agent as a referentially prominent participant.

Interestingly, the “zero” behavior of oblique agents is also exhibited by the PSA. In addition to representing a syntactically prominent/privileged participant, the PSA often assumes topical discourse status as well, being continuously tracked by the speaker in discourse. An example of zero PSA occurring together with a zero non-PSA agent has been shown in (17). Two more examples of zero PSA can be seen in (18). Here, the unexpressed semantic agent ‘they’ in (18a) and the semantic patient ‘him’ in (18b) represent tracked referents already mentioned in prior discourse:

On the other hand, the backgrounded status of the non-perfective non-PSA agent is shared by other non-PSA participants, which are also often left unexpressed when not saliently involved in event interpretation. This can be exemplified by the semantic patient (cooked food/ingredient) that is implied in the meaning of the agent-voice verb po’ocípi ‘cook’, but is not explicitly expressed due to its generic/backgrounded status in (19a). Here, it is the general event of cooking that is of interest to the speaker.^[32] This is in contrast with the example in (19b), where the same event is presented as involving a specific patient. In the latter case, the verb occurs in the patient-voice construction, and the specific patient is realized as a zero PSA of the clause.

A detailed investigation of the behaviors of the grammatical relations in discourse, therefore, reveals that the two differentially marked non-PSA agents are indeed functionally different in terms of how they are interpreted when left unexpressed. Whereas the non-perfective agent (genitive-marked when pronominal) allows a generic/backgrounded interpretation when omitted from mentioning, the perfective agent (oblique-marked when pronominal) is always interpreted as a topical, continuously tracked zero referent when left unexpressed in discourse. In this regard, the two types of non-PSA agents can be further analyzed as aligning differently with other participant types. The perfective non-PSA agent aligns more with the PSA, the most syntactically privileged argument in the verbal clause, as both are interpretable as zero/tracked referents in discourse. On the other hand, the non-perfective agent aligns more with other non-PSA participants because of its ability to be omitted from mentioning due to backgrounded status in discourse (Table 10).

4.2 Stative verbs and the semantic interpretation of the agent

The natural discourse data examined in the previous section show that despite being syntactically equivalent in control properties, the two types of non-PSA agents (differentially case-marked when occurring as a pronoun) show very different behaviors when left unexpressed in discourse. The data investigated previously, however, are all based on the usage of dynamic verbs, which are found to frequently show tense-aspect alternations in natural, spontaneous speech. Stative verbs, on the other hand, are not as frequent in discourse. However, elicited examples indicate that they behave very differently from dynamic verbs in terms of what effects the perfective/non-perfective alternation may have on the interpretation of the non-PSA agent.

As suggested in Table 6, repeated in Table 11 for ease of reference (with the verbs slightly rearranged), patient-voice verbs with stative semantics in Kanakanavu converge in one important morphological property: they tend to lack a perfect form. The first three patient-voice verbs shown below are characterizable as having rather stative meanings, and they all lack perfect forms. This is in sharp contrast to the other three patient-voice verbs that follow, which have rather dynamic meanings. The latter consistently show a three-way tense-aspect distinction.

Different from what is observed for dynamic verbs, omitting the agent in stative verbs has more implications for how the verb itself is interpreted. This can be exemplified by two patient-voice verbs that encode emotional/psychological states. The imperfective forms of the verbs – alakulácʉn ‘(be) angry (patient voice imperfective)’ and kacangcangálʉn ‘(be) happy (patient voice imperfective)’ – are shown in (20). Here, the genitive pronouns realizing the non-PSA agent represent the semantic experiencers involved in the stative situation. On the other hand, there are also semantic stimuli involved; they are designated as the PSA and are realized as lexical noun phrases in the examples below.

When the genitive pronouns are omitted, as in (21), interestingly, completely different interpretations of the verbs result: alakulácʉn ‘(be) infuriating’ and kacangcangálʉn ‘(be) pleasant’.

Omitting the agent in the usage of the imperfective stative verbs can be argued to result in completely different situations expressed by the same verbs. First, the two interpretations – one with overt agent and the other without – are concerned with predication directed toward different participants. The overt-agent interpretation (20) indicates an emotional/psychological state that is present in the semantic experiencer (i.e., ‘(be) angry’ and ‘(be) happy’), whereas the omitted-agent interpretation (21) presents a property that characterizes a semantic theme (i.e., ‘(be) infuriating’ and ‘(be) pleasant’). Second, the former concerns a transitory state that the agent is presented as experiencing, while the latter presents a state that is rather long lasting or a stative property exhibited by the theme participant.^[33] The semantic differences can even be argued to result in completely different lexemes. An example list of stative verbs that illustrate the overt-agent/omitted-agent based alternation is given in Table 12.^[34] Again, this is drastically different from what is observed for dynamic verbs, where leaving the agent unexpressed in the imperfective implies that the same event is concerned, but with a generic/non-specific agent involved.

A comparable effect is also found when the same stative patient-voice verbs undergo alternation to occur in their perfective forms. Consider again the verb alakulácʉn, which has a stative interpretation ‘(be) angry’ in the imperfective, as in (20a), repeated in (22a) for ease of reference. When the verb occurs in the perfective (alakuláce), interestingly, an obligatory dynamic interpretation results, leading to the meaning ‘scold’, as in (22b).

In this case, the semantic difference can be considered even more drastic than what is found for that between the overt-agent and omitted-agent usages of the imperfective patient-voice verbs discussed earlier. Here, the perfective now encodes a dynamic event instead of a stative situation, and the semantic experiencer that is represented by the genitive-marked pronominal agent in the imperfective corresponds to the (oblique-marked) volitional instigator (i.e., a prototypical semantic agent) in the perfective. The two interpretations also crucially differ in telicity: the stative interpretation (in the imperfective form) represents an atelic situation, whereas the dynamic interpretation (in the perfective form) represents a telic one.

Importantly, there is considerable lexical idiosyncrasy in what type of reinterpretation results in the imperfective/perfective alternation. A similar case can be seen in (23), where the perfective patient-voice verb – amanʉ´nge ‘treat well’ indicates a dynamic event performed by a volitional agent (23b). The event expressed in the perfective verb form can be considered to be motivated by the psychological state indicated by its imperfective counterpart – amanʉ´ngʉn ‘(be) happy/pleased’ (23a):

Sometimes, however, the perfective verb form introduces an inchoative interpretation instead, indicating a change of state caused by entrance into an emotional/psychological state by the experiencer.^[35] This can be seen in (24), where the perfective verb form kacangcangále is interpreted as ‘get/become happy’ (24b), whereas its imperfective counterpart kacangcangálʉn does not imply any indication of entrance into the expressed state (24a).

Notice that the type of semantic reinterpretation resulting from the perfective/imperfective alternation is arguably lexically idiosyncratic. Although kacangcangálʉn ‘(feel) happy (with)’ can be considered semantically similar to amanʉ´ngʉn ‘(feel) good (about)’, the reinterpretations led to by their perfective forms are very different. A list of patient-voice stative verbs and their imperfective and perfective interpretations is given in Table 13.

Evidenced from the alternations undergone by both dynamic and stative patient-voice verbs, it can be concluded that interpretation of the non-PSA agent not only interacts with referential status in discourse but is also interconnected with (stative) verb semantics, which can result from either the omission of the agent or alternation between the perfective and imperfective forms of the verb. It is important to recall, nevertheless, that syntactic evidence (as discussed in Section 3.2) still suggests the two types of differently marked non-PSA agents are equivalent in the grammatical sense: they are both the grammatical relation that represents the agent in patient-voice clauses, and they both serve as the obligatory controller in clause combining. At any rate, the findings presented in this section show that agent marking and agent interpretation interact closely with verb semantics, where lexical effects are especially strong when stative verbs are considered.