2.1 Theoretical framework

Constructivist approaches propose that argument structure is not determined by verbs or predicates, but by their hierarchical position in the syntactic structure (Borer 2005; 2013; Ramchand 2008; Marantz 2013). Recent work has applied this perspective to Talmy’s typology of motion events, focusing on the distinction between Satellite- and Verb-framed languages, such as Folli & Harley (2020) and Acedo-Matellán & Kwapiszewski (2024). We briefly outline their respective approaches before situating our own analysis of Takituduh Bunun.

Folli & Harley (2020) argue that the distinction between S- and V-framing can be reduced to a head movement parameter. In V-framed languages, the Path-projection obligatorily moves to a verbalizing head v due to uninterpretable features on v. This blocks the insertion of Manner-roots as modifiers, since the head is already occupied. In S-framed languages, no such movement occurs, allowing for the co-occurrence of Manner and Path. In their framework, verbs lexicalize different bundles of syntactic heads. A motion verb like entrare (Italian) corresponds to [v+Path], while in a combination like run into (English) Path is spelled-out by an independent particle and the Manner is encoded by a lexical root licensed by the verbalizer ($[v+\sqrt{\mathit{\text{run}}}]$ ). The proposals adopt the view that head movement proceeds via Internal Merge followed by morphological merger (Matushansky 2006). An illustration of an S-framed structure within this analysis is given in (3), showing how the lexical root is concatenated with the verbalizer, while an illustration of a V-framed structure is given in (4), where instead the Path head is concatenated with the veralizer.

Acedo-Matellán & Kwapiszewski (2024) criticize Folli & Harley (2020) on both empirical and conceptual grounds. Working within Mirror Theory, they propose that V-framed languages lack a feature (w) that enables independent lexicalization of the Res(ult)/Path head. In S-framed languages, Res/Path carries additional features, allowing it to surface independently. They also employ a distinction between Strong and Weak S-framing languages. The latter incorporates Path as clitics or affixes, while the former lexicalizes Path as independent constituents. Their typology thus treats S-framing as the marked option, in contrast to Folli & Harley (2020). An illustration of their analysis for a resultative structure is given below, with a V-framed structure presented in (5) and an S-framed structure in (6). Since the Res-projection is situated in a specifier position in the S-framed structure, it must be receive a phonological realization from the rest of the verbal projections. It thus requires additional feature specification, marked with w and @ in the structure below, to allow it to be realized independently, thereby capturing the higher markedness of S-framed structures, since it requires additional feature specification.

The difference between Weak and Strong S-framed languages are captured based on the feature specification of the Res/Path-projection. For Strong S-framed languages, this projection carries both the features w and @, allowing it to linearize independently. In contrast, Weak S-framed languages lack the feature @ on the Res/Path-projection, meaning that while it will be able to linearize, it will not be realized as a phonologically independent word, and will instead clitize to the lexical verb. An illustration of a weak S-framed languages is given below, where the Path-element (ad-) is realized as a prefix onto the manner-of-motion verb (equitavit).

In sum, while there are differences in the details of the analyses, an emerging consensus is that Path (or Result for change-of-state predicates) is introduced as a distinct syntactic head situated in a subordinate position in relation to the event-denoting verbal projection, v. In V-framed structures, the two syntactic heads (Path and v) are concatenated as spelled-out by a single exponent, while the two syntactic heads are not concatenated and spelled-out independently in S-framed languages.

Our discussion of Takituduh Bunun follows in the vein of previous proposals. The specific clause structure for Takituduh is outlined in (8), where the Path element is the complement of a verb head v responsible for encoding the event. Locative elements are introduced as complements of Loc, which constitutes an additional syntactic head. Takituduh Bunun differs from the mono-verbal structure of prototypical S-/V-languages by allowing for serializing verbs, which can be captured by allowing for additional vP layers to be added to the structure, taking a lower vP as its complement (cf. Hopperdietzel 2022 for Mandarin Chinese and Samoan). Moreover, since we also consider causativization of motion events, a causative element can be added to this structure (Pylkkänen 2008). As additional verb phrases and the causative prefix are not present in all translocative motion clauses in Takituduh Bunun, these are marked in parenthesis in the structure in 8.

We argue that translocative motion events in Takituduh Bunun with the motion prefix u- can be accounted for by assuming a high degree of concatenation of the syntactic heads outlined above and that constraints derived from non-serializing languages (e.g., Folli & Harley 2020; Acedo-Matellán & Kwapiszewski 2024) cannot straightforwardly account for patterns found in Takituduh Bunun. As Hopperdietzel (2022) suggests, verb serialization and framing must be treated as orthogonal dimensions. The framework outlined here accounts for both the multiple verbalizations, and the extensive head movement observed in Takituduh Bunun. The remainder of the paper develops this analysis empirically, focusing on how the motion prefix u- verbalizes motion via syntactic concatenation.

2.2 Methodology

Data collection was conducted in 2022, in the Village of Zhongzheng, Nantou Country, Taiwan. Five native Takituduh Bunun speakers participated as informants (in the age span 60–80 years). They were all bilingual speakers of Takituduh Bunun and Mandarin Chinese, with some also exhibiting proficiency in Japanese and Southern Min. The first author conducted data collection in the homes of the participants. Data collection was carried out during short but frequent meetings, typically several times a week. These sessions were conducted in Mandarin Chinese.

To elicit motion event descriptions, we used video clips (Park 2020) and pictures (Mayer 1969) that are commonly used for elicitation in motion typology. Stimuli-elicited descriptions were complemented with translations and acceptability judgments. With respect to the former, the informants were asked to translate words, phrases and sentences from Mandarin Chinese to Takituduh Bunun. For acceptability judgments, native speakers were presented with sentences in Takituduh Bunun orally and then asked to judge the acceptability of the sentence, and were asked to repeat them out aloud.

2.3 Takituduh Bunun

Bunun is mainly spoken in the central and eastern parts of Taiwan. It is divided into a Southern and a North-Central Branch, the latter further divided into a Central and Northern branch (see 9). The different varieties display varying degrees of mutual intelligibility (De Busser 2009; Huang & Shih 2018; Li 2018).

Estimates on the number of Bunun speakers remain conjectural. De Busser (2009) approximates 25 000–32 000 speakers, but acknowledges that this may be an optimistic estimate. Huang & Shih (2018) states that the Isbukun variety is by far the largest variety, comprising around half of the speakers. Takituduh Bunun is spoken mainly in three villages in Ren’ai Township in Nantou Province, central Taiwan: Wanfeng (Sima’un in Takituduh Bunun), Fazhi (Bukai) and Zhongzheng (Qatu). The data presented here were collected in Zhongzheng (Qatu).

In terms of grammatical structure, Takituduh is predicate-initial with some variation regarding the linear order of Subject and Object, although VSO is the unmarked alternative (10). The Austronesian voice system is used in Takituduh Bunun, where the voice morphology of the verb correlates with the argument structure of the clause and the thematic role of the clause pivot (the pivot marker is glossed as nominative in our analysis). The primary distinction is between Actor Voice (AV), where the external argument is the clause Pivot, and Undergoer Voice (UV, comprising Patient Voice, Locative Voice and Circumstantial Voice), where an internal argument acts as a clause Pivot. Examples illustrating the distinction between AV and UV are given in (10a-b), with Patient Voice as the Undergoer Voice relevant for the paper as we do not discuss Circumstantial or Locative Voice here.

In (10a), the finite verb is in AV, which is connected to the agent thematic role of the nominative argument. In contrast, (10b) illustrates a UV clause, with the finite verb in Patient Voice and with the nominative argument having a patient thematic role. The clause pivot is marked by the case marker ca, glossed as nominative throughout this paper. In contrast, the case marker is fulfills several functions, including marking the agent argument in non-actor voice clauses and arguments beyond the agent in AV-clauses, as well as possession. The latter function is illustrated in (11a), where is signals the possessive relationship between the possessor (mother) and the possessed (clothes)². Of relevance for the current paper is that the same possessive structure is also used to provide further specification of a Region, as illustrated in (11b).

The Austronesian voice system has been widely discussed (see Chen & McDonnell 2019 for a recent overview, and Teng 2023 and Tsukida & Zeitoun 2023 for discussions specific to the Austronesian languages of Taiwan), especially in relation to features such as transitivity, argument structure and morphosyntactic alignment. Since the data discussed here are compatible with several different analyses of the Austronesian Voice System, we do not discuss the technical details of the Austronesian Voice System further. However, voice marking is still relevant in our discussion on the hierarchical structure of serial verb constructions, so we will return to this topic in subsections 3.3 and 3.4.

The motion prefix u- is reconstructed by Blust (2009) to proto-Austronesian as *mu-. However, we follow the proposal by Chen (2020) that the reconstruction is actually bimorphemic, consisting of the motion prefix u- and the Actor Voice prefix m-. This proposal is compatible with the Takituduh Bunun data, since the motion prefix u- can appear without m- (see sections 3.2 and 3.4 for examples and details). Reflexes of the motion prefix are found throughout the Formosan languages (The Austronesian languages of Taiwan), and it is only attested in one extra-Formosan Austronesian language (Cebuano) which is spoken in the Philippines (Chen 2020). The motion prefix is attested in other varieties of Bunun, including Takivatan (De Busser 2009) and Isbukun (Li 2018). In both varieties, the motion prefix can attach to both locative elements (sia in Isbukun, han in Takivatan) as well as nominal elements (De Busser 2009; Li 2018). However, to the best of our knowledge, the current paper constitutes the first attempt at a formal analysis of the clause and argument structure of this motion prefix.

3.1 Motion encoded via the motion prefix u-

3.2 Causation and the motion prefix u-

In the clauses we have discussed so far, no external causers were involved, and the subject in all motion clauses was the Figure. However, it is possible to add an external causer to motion prefix clauses with the causative prefix p-. This results in a clause structure with three actants, corresponding to Figure, Causer and Region (+ Landmark). This structure is crucial evidence in favor treating the sequence mu- as bimorphemic (see also Li 2018 for a similar analysis for the Isbukun variety of Bunun), consisting of an Actor Voice prefix m- and the motion prefix u-. A minimal pair illustrating caused and non-caused motion is given below, where the caused equivalent of (26a) is found in (26b).

As shown in the translation, both clauses encode translocative motion. Since the causativized clause is in Patient Voice, the internal argument that undergoes movement (i.e., the Figure) is marked by the nominative case marker, and the external causer is optionally marked by the case marker is. The Ground cungus is sipulan ‘backyard of the school’ cannot be preceded by the case marker is in the causative clause. As such, the case marking of the noun expressing Region is the same for both the cause and non-caused clause structures.

It is also possible to causativize the structure when the motion prefix u- is hosted by a locative noun. The causative alternative of (27a) is given in (27b). Since the causative clause is in Patient Voice, the Figure is marked with a nominative case marker, and the external causer argument is marked with an optional case marker is. Note that the Landmark sipulan (‘school’) in the causative clause may be preceded by the case marker is, just as was shown for non-caused clauses with locative nouns in the previous section.

As examples (26) and (27) show, the same case marking patterns for non-caused motion clauses carry over to their causativized counterparts. In clauses where it is the motion prefix together with the locative preposition that encodes translocative motion, its complement, the Ground, cannot be marked with the oblique case marker is, regardless if the clause structure is causativized or not. In contrast, in clauses where the motion prefix together with a locative noun encodes translocative motion, their complement (i.e., Landmark) can be preceded by the case marker is. We take this as evidence that the externally merged clause structures of the non-caused/caused pairs are the same, with the difference that the causative clause contains an additional projection that introduces causative semantics and the external causer argument. We begin by outlining the clause structure for example (26b), before moving on to discuss example (27b).

The presence of the causative prefix p- introduces both an additional argument (a causer argument) and causative semantics (causing the theme to undergo movement). Here we do not attempt to classify the causative prefix within the broader typology of causatives within generative constructivist frameworks (Pylkkänen 2008; Harley 2017, see also Nie 2020 for a relevant discussion on causatives in the related language Tagalog). Instead, we simply propose that the causative semantics are introduced via a separate verb head v_cau, and this head is spelled out by the causative prefix p-. Whether or not the Causer is introduced in the specifier of this projection or via a distinct Voice projection (i.e. Voice-Bundling vs. non-Voice-Bundling, Pylkkänen 2008) is not central to our main claim. As in the uncaused movement structures discussed in the previous section, the motion prefix u- in causativized clauses is the overt realization of Path⁰ and v⁰. The locative preposition can/han spells out a Path⁰, which is responsible for introducing the Ground. Both p- and u- are realized as prefixes, with the motion prefix situated closer to the locative preposition than the causative. The linear order of the prefixes thus directly reflects the syntactic order, where the syntactic head introduced higher in the structure (p-) is linearlized further away from the “root” (can in this example) than syntactic heads introduced in a lower position (u-), thus falling in line with the head concatenation analysis developed here. The morphologically complex p-un-can illustrated in (28) is the overt realization of v_cau, v, Path and Loc where this complex head can be seen as the result of iterative head movement, illustrated by the dashed lines in the tree structure in (28)⁷.

Through such an analysis, we can directly account for why the Ground cannot be marked with a case position. Like in the non-caused counterpart, the noun phrase denoting the Ground is actually not an argument of the motion verb, but rather the remnants of a prepositional phrase. The same analysis can be extended to clauses where the motion prefix is attached to a locative noun, illustrated in the tree structure in (29). The key difference is that the nominal also undergoes head movement. Thus, the morphologically complex word p-u-kumbu spells out v_cau, v, Path, Loc and n. As in the example in (28) above, the linear order of the two prefixes directly reflects the syntactic order.

As with non-caused clauses, the difference in the underlying structure is reflected by the case marking. As was shown for non-caused motion, the verbalized preposition cannot have the marker is before its complement, whereas for verbalized locative nouns the case marker is optional. We see the same pattern in the causativized counterparts, where the Region in clauses with pucan/puhan likewise cannot be preceded by is, whereas it can be preceded by an optional is in clauses with a verbalized locative nominal. We propose that this reflects the externally merged structure, where the Region in verbalized preposition clauses is the remnants of a PP, and the Landmark in verbalized locative noun clauses is the remnants of a noun phrase. Examples illustrating this with brackets and traces are given in (30), reproduced from (28) and (29), respectively.

Like with the non-caused translocation motion clauses with the motion prefix u-, the distribution of the case markers follows automatically from such an analysis. Another argument in favor of our analysis comes from word order. In Patient Voice causativized clauses with three actants, the ordering of the causand and the causee is variable, so it is possible for the causee (marked with the nominative case marker ca) to both precede and follow the causand (optionally marked with the case marker is). In (31a), the causee precedes the causand, whereas it follows the causand in (31b).

This type of variation in word order is not possible for the caused motion structures discussed here. These clauses exhibit the strict word order of Verb - Causer - Ground/Landmark - Figure. If the Figure were to be placed before the Ground/Landmark (as in 32a-b) with the same word order, then the structure would be ungrammatical.

We take this difference in word order between regular causative clause structure and cause motion clause structure to provide further evidence in favor of our analysis. This means that the Ground/Landmark constituents in the examples above are not arguments of the verb, but rather the remnants of a PP or an NP, respectively. This stands in contrast to regular causative clauses, where all three actants (causer, causee and causand) are all arguments of the verb. It is therefore not surprising that we see different word orders in the two clause types.

To summarize, we have shown that the morphologically complex path-verb containing the motion prefix u- is the result of extensive concatenation of syntactic heads, at least containing v⁰, Path⁰ and Loc⁰. This would make the structure characteristic of those found in a Verb-framed languages, as defined by Folli & Harley (2020) and Acedo-Matellán & Kwapiszewski (2024). However, in the next two sections, we show that these structures can combine with manner-of-motion verbs to form serial verb constructions that can be analyzed as a kind of satellite-framing.

3.3 Manner-of-motion and motion prefix u-

Manner-of-motion verbs can combine with constituents derived via the motion prefix u- in a productive way to encode translocative motion events. Manner-of-motion do not encode translocation motion when combined with a simple locative marker, as illustrated in (33a). This can be compared with example (33b), where a manner-of-motion verb combine with the motion prefix u- attached to the general preposition to encode translocative motion.

In (33b), there are two verbal elements: The lexical manner-of-motion verb malalia ‘run’, and the derived path verb muncan ‘go to’. This should thus be regarded as a multi-verbal clause. The ordering of these verbs is strict, with the manner-of-motion verb preceding the path verb. In clauses with two or more verbs, the first verbal element in the sequence is syntactically dominant, since it hosts the distinctive voice morphology of the clause, as well as imperative mood. These are two important criteria for identifying finite verbs in Takituduh Bunun (see examples (19) and (20) above, and examples below as well, illustrating this pattern). We return to this issue in the discussion and its implications for classifying Takituduh within motion typology.

Such multi-verb clauses are comprised of structurally higher vPs embedding structurally lower vPs. In the tree structures presented below, the different vPs are marked with subscript numbers to distinguish them. In these multi-verbal clauses, the same analysis of the motion prefix u- attached to the preposition as in clauses where it is the only verbal element. Thus, it spells out Path⁰ and v⁰, while the locative preposition spells out Loc⁰. Since manner-of-motion verbs in Takituduh Bunun cannot encode translocative motion by themselves, they are proposed to be the overt realization of a root and a classifying head v. Both verbs (malalia and muncan) can function as independent dynamic verbal predicates, so it is necessary for both to independently lexicalize separate v⁰ heads. The proposed analysis is illustrated in the tree structure in (34).

It is also possible for a manner-of-motion verb to combine with a verbalized locative noun, illustrated in (35a) below, where the manner-of-motion verb ‘crawl’ is combined with the verbalized locative noun ‘go out’ and in (35b), where ‘jump’ is combined with ‘enter’. The compatibility of manner-of-motion verbs with verbalized locative nouns is particularly clear in example (35c) with the verb ‘dance’, which by itself does not encode any directedness meaning (cf. Folli & Ramchand 2005). The same analysis of the verbalized locative noun (34) can be applied here: The manner-of-motion verbs in (35) lexicalize a lexical root and a classifying head v, while Path⁰ and an additional classifying head v is spelled-out by the prefix u-.

There are no overt markers of subordination present in these structures. Furthermore, such clauses share TAM-values, illustrated by the imperative mood marker in (36) below. When being the sole verb of a clause, it is possible for the Path verb to host the imperative suffix -a, as seen in (36a). However, in serial verb constructions with a path and a manner-of-motion verb, as in (36a) the imperative mood is only overtly marked on the leftmost verb, illustrating a syntactic hierarchy between the two verbs, while subsequent verbs lack any overt mood markers.

The structure also differs from subordinate modifier clauses, marked with maca and a, illustrated in (37). In this example, Path and Region are encoded in the subordinate ‘while’ clause, while the manner-of-motion verb is the sole verb of the main clause.

In light of the framework developed by Folli & Harley (2020), this is an unexpected pattern since they consider V-framing to be characterized by obligatory Path-to-v-movement, which prevents v from licensing a Manner root in its specifier functioning as a modifier. As such, languages with obligatory Path-to-v-movement, as we argue is the case for Takituduh Bunun, should not exhibit satellite-framed structures. However, the Takituduh Bunun data do not directly contradict any of the predictions made by their framework. In our analysis, Path is concatenated with and licensed by v₁ while the manner root is concatenated and licensed by v₂. As such, the data presented here do not contradict the predictions made by Folli & Harley (2020). However, it does show that obligatory concatenation of v and Path is possible for satellite-framed languages, provided that they are serializing languages, and that this cannot be the sole distinguishing feature between S-framed and V-framed languages.

Several criteria have been discussed for identifying verbs in Takituduh Bunun, including voice morphology and imperative mood. The morphologically complex verbs derived via u- exhibit both of these properties (examples (14) and (36a) show that they can host imperative mood, and in all mono-verbal examples with u- it hosts the distinctive voice morphology). As such, there are good reasons to believe that the morphologically complex words derived via the motion prefix u- are verbs. However, in multi-verbal clauses, neither of these diagnostics apply, since the clause-initial manner-of-motion verb would host both of these. This pattern is especially clear for the Patient Voice clauses discussed in 3.1, examples (19) and (20). Here, the distinctive voice morphology of the clause is clearly Patient Voice, as nominative argument is a Theme, not an external causer argument, as it would have been had they been AV clauses. As such, the voice morphology on the verb derived via u- is not distinctive voice morphology of the clause, but rather reflects a default value when no other voice form is assigned (Chang 2017). Multi-verbal clauses in Takituduh Bunun do not exhibit agreement or concord with the voice value. The fact that it is obligatory to mark all morphologically complex words derived via the motion prefix u- with a Voice value, either the distinctive voice morphology of the clause or the default Actor Voice, just like with lexical verbs (Bogren Svensson 2023), can be taken as a strong indication of their verbal status. We therefore conclude that the structures discussed here should be analyzed as containing multiple vPs, and that our conclusions regarding the short-comings of previous frameworks remain valid.

3.4 Causation, Manner-of-motion and the motion prefix u-

Caused translocative motion in clauses with manner-of-motion verbs in Takituduh Bunun can also be encoded with the help of a causative prefix. Two examples of causativized manner-of-motion translocative motion clauses are given below, one with a verbalized preposition (38a), one with a verbalized locative noun (38b). The distinctive voice morphology of the clause (Patient Voice in both examples) is hosted by the manner-of-motion verb, which is the left-most verb in the linear order and the hierarchically dominant verb. The motion verb is in Actor Voice, even though it is an Patient Voice clause. Actor Voice here can thus be characterized as a “default voice”⁸ (cf. Chang 2017; Wu et al. 2023). The causative is hosted by the manner-of-motion verb. As in their non-caused counterparts, the ordering of the verbs is strict, with the manner-of-motion verb obligatorily preceding the morphologically complex path verb. Since both clauses are in the Patient Voice, the Figure marked with the nominative case marker ca, and the Causer is optionally marked by the oblique case is for full nouns, or in the agent case form for pronouns.

The same analysis outlined in previous subsections can be applied to these examples. The manner-of-motion verb lexicalizes the hierarchically higher v⁰, while the motion prefix lexicalizes the hierarchically lower v⁰ and the Path⁰, and encodes translocative motion. Thus, there are two vP in the clause since both the morphologically derived path verbs and the manner-of-motion verbs can function as independent predicates. This kind of structure is possible if the two v⁰ are spelled-out by different constituents, as in the non-caused examples discussed in subsection 3.3 above. The v_cau, which is spelled-out by the causative, takes the entire multi-verbal structure as its complement. This yields the correct scope and semantic interpretation, since the Figure (‘child’ in both examples) is caused to carry out both eventualities denoted by the verbs (both ‘walk’ and ‘go to church’, both ‘jump’ and ‘go out’, respectively). The tree structure below provides an illustration of the clause structure. The only difference between the multiverbal clauses discussed above is that a v_cau is added to the structure, above the structurally higher vP. This prevents the causative prefix from being realized on the subordinate verb. The clause structure is illustrated in (39), with the example reproduced from (38a) above).

Example (38b) can be analyzed in the same way, as illustrated in (39), with the key difference between the two being that the Locative head is empty in such clauses, triggering movement of the (locative) nominal instead (as illustrated in (23) and (29) for non-caused and caused motion, respectively). As discussed for these examples, this is a necessary postulation because the motion prefix u- can alternate between being hosted by a preposition and a nominal.

4.1 Implications for the framing typology

Clauses containing both a manner-of-motion verb and a morphologically complex path verb can be understood as a kind of serial verb construction. Even though there are multiple verbs within a single clause, the two verbs are hierarchically ordered, with the manner-of-motion verb being syntactically dominant. When these two types of verbs combine, the manner-of-motion verb will host i) Imperative morphology, ii) Causative morphology and iii) the distinctive voice morphology of the clause⁹. Taken together, these features provide solid evidence that these structures are hierarchically ordered.

Regarding imperative voice morphology, it can be hosted by the stem derived via u-, but it must be hosted by the Manner-of-Motion verb if one is present in the structure, as illustrated in the examples below, reproduced from (14).

Just like the imperative voice suffix, the causative prefix can be hosted by the stem derived via u-, but it must be hosted by the manner-of-motion verb if one is present in the structure. Moreover, since the nominative argument (i.e. the clause pivot) is the theme argument and not the agent, it is possible to determine that Patient Voice is the distinctive voice morphology of the clause hosted by the manner-of-motion verb, reproduced from (38) above.

One proposal for dealing with serial verb constructions has been to extend to a tertiary typology for constructions where individual verbs have equal syntactic status, so-called “equipollently-framed languages”¹⁰ (Slobin 2004; Zlatev & Yangklang 2004). This proposed type has been exemplified with languages like Thai and Ewe that express motion events in serial verb constructions without any overt working marking of syntactic hierarchy. Together with the fact that all motion verbs can independently function as the sole verb of a clause, it has proven difficult to locate main verb status to one of the verbs, which, as we have argued, differs from serializing constructions in Takituduh Bunun. This is shown in (42) and (43) with serial verb constructions containing three independent motion verbs (Thai) and two verbs (Ewe).

However, Talmy (2016) argues that languages utilizing serial verb constructions for encoding motion events should not necessarily be classified as equipollently-framed, since criteria based on semantics, morphology, syntax and phonology can be used to identify a main verb even in serial verb constructions. Hopperdietzel (2022) follows a similar line of argumentation, arguing that resultative serial verb constructions can be both Verb-framed and Satellite-framed, depending on the language. The arguments used here to establish that the manner-of-motion verb was morphosyntactically dominant verb fall in line with the morphological criteria outlined by Talmy (2016) for identifying main verb status in serial verb constructions. However, in the constructivist framework adopted here, the morphological properties used to determine main verb status are also indicative of a syntactic hierarchy. We thus side with Talmy (2016) and Hopperdietzel (2022) in arguing that having serial verb constructions (in the sense of multiple verbalizations in a single clause) in translocative motion events does not necessarily entail that the structures are equipollently-framed. However, whether or not equipollently-framed languages can be dispensed with entirely remains an open question, and we only argue that having multiple verbalizations within a single clause (as was shown for Takituduh Bunun) does not necessarily imply that the structure is equipollently-framed.

4.2 Implications for formal approaches

A central claim in this study is that if a language has more than one available verbalizing head within a single clause, there are no inherent obstacles to verbalizing Manner, Path and Motion within the same clause. This presents challenges to previous formal accounts as their constraint (Folli & Harley (2020) treating V-framed as the marked counterpart, Acedo-Matellán & Kwapiszewski (2024) treating S-framed as the marked counterpart) are centered around preventing all three categories from being lexicalized via a single simplex verb. As such, there still are strong constraints against lexicalizing path, manner and motion within the same verb, the same constraints cannot be used to derive the S-/V-framed dichotomy within a serializing languages as Takituduh Bunun.

Within this binary typology, Takituduh Bunun exhibits both V-framed and S-framed structures. The former are the structures where the morphologically complex verb derived via the prefix u- is the sole verb of the clause, and the latter are clauses that contain the verb derived via u- in addition to a manner-of-motion verb. In line with the argumentation of Acedo-Matellán & Kwapiszewski (2024), the structures in Takituduh Bunun represents an additional way in which the seemingly more marked S-framed structures can be licensed. In the framework developed by Acedo-Matellán & Kwapiszewski (2024), S-framed structures require additional feature specification (using the labels @ and w, and the label w for Strong and Weak S-framed languages, respectively), and in Takituduh Bunun the S-framed structures require an additional verbalizing head. As such, the S-framed structures in both serializing and non-serializing languages appear to be the more marked alternative, since both need additional specification, verbalization of the Path element in the former, and additional feature specification in the latter, both necessary for the independent realization of the Path element.

If the V-framed/S-framed distinction ultimately reflects a morphological constraint (limiting how many semantic categories can be lexicalized in a single verb) why should this binary apply to languages with serialization? In Takituduh Bunun, S-framing results from structural hierarchy rather than lexical restrictions, suggesting that the typology may not be uniform across language types. While our findings align with the idea that lexicalization constraints shape framing patterns, they also highlight gaps in current theories, namely the need to also develop constraints for languages that distribute semantic categories across multiple verbs. More generally, the findings suggest that the S-/V-contrast may mask a deeper heterogeneity in how languages partition motion semantics across lexical and functional material with distinct morphological, syntactic, and lexical configurations across languages. Addressing this will require comparative work on other serializing languages.