Typological features of Telugu: defining the parameters of post-Talmian motion event typology

ABSTRACT Recent research in motion event typology has moved beyond the binary Talmian division of “verb-framed” and “satellite-framed” languages and has established the existence of at least four distinct typological clusters, instantiated by, for example, Swedish (Germanic), French (Romance), Thai (Tai-Kadai) and Telugu (Dravidian). In this paper, we focus on characteristic features of Telugu, as a representative of the fourth cluster. In the study, 30 native Telugu speakers described video-recorded translocative events, in which the factors boundedness, viewpoint and causation were manipulated. Using the model Holistic Spatial Semantics, we show that Telugu speakers (a) preferentially used Direction verbs rather than Path verbs, (b) predominantly used case markers rather than verbs for encoding Path, (c) extensively used Landmark and Region expressions, and (d) frequently used Manner verbs in situations of “boundary-crossing” unlike speakers of typical “verb-framed” languages. We propose these features to be criterial of the fourth typological cluster mentioned above, a claim to be investigated in future research.


Introduction
When discussing the semantic typology of motion events, a traditional point of departure is the influential claim that languages fall into two types with respect to how they express the semantic category Path (Talmy 1985(Talmy , 2000. 1 On the one side are so-called "verb-framed" languages like Spanish, in which Path is said to be typically encoded by verbs like entró ("enter") as in (1). On the other side are "satellite-framed" languages like English, where Path is most commonly expressed by a so-called "satellite" to the verb (root), such as the particle/preposition into in (2). During the past two decades, however, this binary typology has been shown to be lacking both empirically and theoretically. Recurrent issues include (a) that the classification is unable to capture variation within a single language (e.g., Berthele 2013) and variation across languages supposedly belonging to the same type (e.g., Bohnemeyer et al. 2007;Ibarretxe-Antuñano 2009;Morita 2011;Matsumoto 2003;Lewandowski 2021); (b) the need to extend the number of types (Slobin 2004;Zlatev and Peerapat 2004;Fortis and Vittrant 2016); and (c) the vagueness in the definitions of the key concepts Path, Manner, and even Motion (Zlatev, Blomberg, and Caroline 2010;Croft et al. 2010;Imbert 2012;Blomberg 2014). (2) The bottle floated into the cave. (Talmy 2000, 227) Reviewing such problems and proposed alternatives, Naidu et al. (2018) and Zlatev et al. (2021) proposed the term post-Talmian as characteristic of the state of the field today. With this term, the authors acknowledge their indebtedness to the seminal work of Talmy, but mark that the field has moved beyond it. To justify this claim, Naidu et al. (2018) compared motion event expressions in Thai, an "equipollently-framed" language according to Slobin (2004), with those in Telugu, a Dravidian language spoken in southern India. Conducting an elicitation study based on "frog stories" (i.e., eliciting narratives based on a well-known picture book by Mayer 1969; cf. Berman and Slobin 1994), the authors revealed substantial differences between the languages in terms of the semantic categories that were most frequently expressed, and of the structural means by which they were expressed. Since neither Telugu nor Thai fit easily into the original "verbframed" or "satellite-framed" types, Naidu et al. (2018, 20) formulated the hypothesis that within a post-Talmian motion event typology, the Talmian types "will appear as only two such clusters, while serial verb languages like Thai (e.g., Ewe and Vietnamese) and languages like Telugu (e.g., Tamil and Finnish) will fall into clusters that are distinct from these, as well as from each other, thus giving us (at least) four distinct typological prototypes." Using the conceptual framework of Holistic Spatial Semantics, first proposed by Zlatev (1997), and subsequently revised and developed (Zlatev 2007;Blomberg 2014;Naidu et al. 2018), Zlatev et al. (2021) tested this hypothesis using systematic elicitation of different kinds of motion situations (Zlatev, Blomberg, and David 2010). They presented short video clips with human and animal actors to native speakers of Swedish, French, Thai, and Telugu. Given the similarities and differences between the motion event descriptions found in the four languages, the authors conclude that the proposal of "four distinct typological clusters in motion event semantics" has been confirmed "by demonstrating (beyond any reasonable doubt) that Swedish, French, Thai and Telugu motion event descriptions differ from one another to such an extent that each language can be seen as instantiating a distinct typological cluster" (Zlatev et al. 2021, 84).
In this approach to motion event typology, typological clusters are characterized by prototypical patterns in ways to express a (hypothetically) universal set of semantic categories (see Section 2). The first two clusters approximate the original Talmian types, with Swedish and French as representatives, but the expression patterns are much more complex than a specification of how Path (and Manner) are expressed. Notably, linguistic strategies to express motion vary depending on the type of situation. For instance, in uncaused-unbound motion situation types, the French speakers encoded Path seldom, and rather used Direction terms as in (3). In contrast, in uncaused-bound motion situations they expressed Path frequently.
(3) Un monsieur monte sur une colline en courant. (French) a man climb.up.PRS on a hill running Figure Motion+Direction Region Landmark Manner 'A man runs up a hill.' (Zlatev et al. 2021, 76) Thai (and other serial-verb languages like Ewe, Akan, Mandarin) represent a distinct cluster of languages in which the categories of Manner, Path and Direction are expressed by distinct serial verbs as in (4), while Path is commonly also co-expressed in prepositions. The latter also express Region, an important category which is generally neglected in Talmian typology (Zlatev et al. 2021).
(4) puying doen khâw pay nay hɔ̂ŋ (Thai) woman walk enter go in room Figure Motion Motion Motion Region Landmark +Manner +Path +Direction 'A woman walks into a room.' (Zlatev et al. 2021, 78) Finally, Telugu speakers rely not on any of these options, but rather on nominal case-marking as the predominant means of expressing Path, reserving verbs for Manner and Direction, as in (5).
Landmark Manner+Motion Direction+Motion Path:End FoR:VC 'A boy came to the school running.' (Naidu and Duggirala 2011, 186) On the basis of an elicitation-based study using the same methodology as the present study, Zlatev et al. (2021) showed how Telugu displays patterns that differ both qualitatively and quantitatively from those in the other three languages and can therefore serve as the hypothetical core of a fourth typological cluster. But what features can be used as criterial for this cluster? We explore this question in the present article with a more extensive data set than those used in Naidu et al. (2018) and Zlatev et al. (2021). 3 We do so in an inductive Manner in Sections 3 and 4, after first explaining our theoretical framework in Section 2. The relevance of the study is also enhanced by the fact that the many languages spoken in India, belonging to widely different language families, have received relatively little attention in motion event semantics. In the few studies that have been reported, Narasimhan (2003) concludes rather quickly that Hindi fits the Talmian "verb-framed" type, and Pederson (2006) does the same for Tamil (see also Slobin et al. 2011, 134). However, these authors ignored the difference between generic deictic verbs like those in the Tamil example (6) and the Hindi example (7), on the one hand, and Path verbs like entrer ("enter") and sortir ("exit") in French, on the other hand. 4 In (6), the most essential element of Path, specifying the endpoint of the motion event, is in fact expressed not in the verb, but in the dative case marker. 5  Pederson (2006) based his conclusions on the assumption that Tamil, like Spanish and French, follows the so-called boundary-crossing constraint 6 3 The present data for Telugu consists of that of the 20 speakers that was used by Zlatev et al. (2021), and 10 additional speakers. 4 The first author collected examples (6)-(9) by eliciting intuitions from five native speakers of Tamil and Hindi. 5 In (7), despite the fact that the dative case -ko is absent, the translocative interpretation is preferred to a locative one of 'going around' by native speakers of Hindi. (Aske 1989;Slobin and Nini 1994;Slobin 1997;Ibarretxe-Antuñano and Luna 2013;Fagard et al. 2013) in a particular way: Like many languages, Tamil has a largish class of Manner-of-motion verbs which by themselves do not indicate translational motion. If translational motion is to be indicated, a motion verb (principally "go" or "come") must combine with the Manner-of-motion verb (Pederson 2006, 415). This is hardly true, however, as native speakers of Hindi and Tamil fully approve of sentences in which a Manner-expressing verb combines with a Path-expressing case as in (8), and in which Path is not explicitly (overtly) coded, but can be inferred pragmatically (covertly), as in (9) These examples suggest that Pederson's analysis was biased towards Talmy's (2000) binary typology in several respects. First, the analysis ignores the difference between generic deictic verbs like those in the Tamil example (6) and the Hindi example (7), on the one hand, and Path verbs. Notably, Özçalışkan, (2013) regards such generic deictic verbs as neutral and therefore differentiates them from Path verbs. Second, the analysis bypasses the role of case markers, which are neither verbs nor "satellites" to the verb. 8 Third, the importance of Region nouns is not recognized. These are the kind of biases that we aim to avoid in our study of motion event expression in Telugu, by adopting the open-ended framework and methodology described in Section 2. Sections 3 and 4 provide 7 While example (9) is ambiguous between a locative ('ran inside the room') and a translocative ('ran into the room') interpretation, the context is likely to resolve this. For instance, if the boy is in a hurry to answer a call coming from the room, the translocative interpretation would be clearly preferred. 8 The notion of case marker is very controversial in the literature (Hudson 1995;Blake 2001;Schiffman 2004;Haspelmath et al. 2005;Butt 2006;Asbury 2008;Corbett and Noonan 2008;Malchukov and Spencer 2008;Iggesen 2013), and it is beyond the scope of the present paper to resolve it. For present purposes we consider case as a linguistic category, marked by a bound morpheme on nouns (thus excluding unbound morphemes like prepositions), and expressing the agentive (i.e., agent, patient), grammatical (i.e., subject, object), or spatial (i.e., source, goal) role played by the entity marked. We focus on the latter, given the context of the study. a qualitative and a quantitative analysis of the data, respectively, allowing us to spell out typological features of Telugu, and thus of the hypothetical "fourth cluster" in motion event semantics, to be tested in future research, as we explain in the discussion in Section 5. Section 6 concludes the paper.
Here we provide a summary with examples from Telugu. The framework derives its name from the claim that spatial meaning is expressed by the whole spatial utterance, and not by a specific form class like prepositions. Yet, this "holism" is only partial, since different parts of an utterance, belonging grammatically to different form classes like verb, adverb and case marker, map onto the following 10 semantic categories: 1. Figure g., eastwest, up-down) 6. Path: bounded translocation, with respect to Beginning, Middle and/ or End 7. Direction: unbounded translocation, along one or more "lines" defined by a FoR (e.g., upwards, towards the speaker). 8. Manner: various aspects of how motion takes place, such as bodily locomotion, use of vehicle etc. 9. Shape: the geometrical form of the trajectory of movement (e.g., zigzag, straight) 10. Cause + Agent: the force that makes the Figure move (e.g., ballistic, comotion) and the entity that is behind this force.
These categories are grounded in human embodiment, but not reducible to it, and are flexible enough to adapt to the conventions of any particular language (Zlatev 1997;Naidu et al. 2018). Further, the way these semantic categories map onto the form classes that express them can be (a) compositional (one-to-one), (b) conflated (many-to-one) or (c) distributed (one-tomany); these mappings are expected to differ extensively across languages.
Importantly, HSS makes a systematic distinction between the categories Direction (typically expressing unbounded motion as in up and come here), and Path (construing the event with respect to Beginning (e.g., from X), Middle (e.g., cross X) and End (e.g., to X)). The two categories have been shown to behave differently across languages (Zlatev et al. 2021) and therefore should not be conflated. For example, in (10) the speaker does not express Direction, but rather Path and Region. Figure  Region:Above+Path:End Cause+Motion 'A man threw the ball upward.' (Literally: "to the above") (Event 11, Participant 2) 9 Concerning motion event typology, HSS does not a priori assume a particular number of distinct language types. Rather, it endorses an openended methodology allowing the systematic examination of motion event expression using different methods of elicitation such as native speaker intuitions based on questionnaires (Zlatev 1997), narratives (Naidu et al. 2018), and motion event descriptions based on videos (Zlatev et al. 2021).

Materials and participants
The elicitation tool used in this study was developed in the Phenomenology and Typology of Motion (PATOM) project (Zlatev et al. 2021), for the purpose of testing and further elaborating the theory of Holistic Spatial Semantics. Out of the total of 52 clips, 38 showed a translocative motion event (i.e., where the Figure changes in average position relative to a specific Frame of Reference) while the remaining 14 showed a non-translocative motion situation where this is not the case (for example, a woman shaking a blanket). The clips were between five and eight seconds long, depending on how long it took a particular motion situation to end naturally.
For the present study, only the translocative motion events were analyzed and the other motion situations were treated as fillers, as shown in Table 1. Appendix B contains a specification of all the target stimuli.
The Direction of the target motion events was systematically balanced (towards/away from the camera; up/down), as well as their boundedness (bounded vs. unbounded), causation (caused vs. non-caused), angle of viewpoint (parallel vs. orthogonal to the motion of the moving object) and the actors' gender (male vs. female). Two examples are shown in Figure 1.
As participants in the study, thirty adult speakers (mean age 21.9 years; 15 female) were recruited from the undergraduate and postgraduate students registered at the University of Hyderabad, India. They were all native speakers of Telugu, with knowledge of Hindi, English or both from their school or college education. The participants signed a form of informed consent and were compensated for their participation.

Procedure and analysis
Each participant was tested individually by the first author, a Telugu native speaker himself. Written instructions were given, stating that after viewing each video clip, the participant should describe it in a way that is as natural and colloquial as possible. Each clip was played only once. The descriptions were audio recorded using a Sony ICD-MP3 recorder and later transcribed for analysis. The total number of clauses and motion event descriptions (see below) for the different event types are shown in Table 2.  The annotation was semi-automatized, following the procedure developed in the PATOM project (see Zlatev et al. 2021). The following steps detail how this was done for the present data.
First, the descriptions of the video clips were analysed in terms of clauses and motion event descriptions. The latter were operationally defined as expressions with motion verbs, consisting of: (a) one finite clause as in (10) above, (b) one participle clause with a non-finite verb as in (11), or (c) two clauses that had the same semantic Figure, as in (12). While finite Telugu verbs inflect for tense and agreement with the grammatical subject as in (10) and the second verb in (12), non-finite verbs like that in (11), and the first verb in (12) are marked with relevant tense-mood suffixes and are therefore regarded as defining clauses in Telugu grammar (Krishnamurti and Gwynn 1985;Jayaseelan 2004;Kissock 2014). Thus defined, Telugu motion event descriptions correspond to what is typically expressed by single clauses in most other languages and can therefore be used as a basis for cross-linguistic comparisons.
A list of the unique word forms in the transcripts were subsequently tagged with an English translation and a value for one or more HSS categories, providing a "lexicon" of the descriptions. The words in the descriptions were then automatically tagged on the basis of this lexicon, and subsequently checked and corrected when necessary. For example, in (12), ammāyi "girl" was correctly tagged as Figure, but in caused-motion expressions this category needed to be replaced by Agent. As a result of this analysis procedure, the data were analysed in terms of motion event descriptions, glossed and coded for the ten HSS categories. Of these, the following five categories, with their possible values, are of particular importance for the analysis in the following sections:

Qualitative analysis
In this section we describe patterns concerning the expression of each of the five HSS categories listed above, illustrated with examples from the present study. A list of all expressions for the categories Path, Direction, Manner and Region is given in Appendix A.

Direction
As pointed out earlier, the category of Direction applies only to elements of unbounded translocation, while Path concerns bounded translocation. The two categories were often combined in a motion event description, as shown in (13) and (14). Expressions of Direction were commonly Viewpointcentered: deictic verbs like waccu "come" and weḷḷu "go", Directional adverbs like kud _ i "right" and ed _ ama "left", or deictic adverbs like dūramgā "far-off". Direction could also be expressed with the help of an Object-centered FoR, combining a Region noun like muMdu ("front") with a Path case-marker, as in (15), or a Geocentric FoR as in (16), through verbs such as ekku "ascend" and digu "descend".

Path
As stated in Section 2, HSS allows for three kinds of mapping patterns between form classes and semantic categories: composition, distribution and conflation. While all three were attested in the data, composition was dominant for Path, as in all descriptions shown so far. In these examples, Path is exclusively expressed through the case markers nuMd _ i (ABL) and -ku/-ki ("to"), without conflating Motion. The form nuMci/nuMd _ i, exemplified in (13) above, is derived from the accusative case marker -n and the verb uMd _ i "having been there" (Krishnamurti and Gwynn 1985;Krishnamurti 2003, 241). Unlike -ku/ki, it is commonly, but not always, written separately from the nominal to which it applies. Nevertheless, we regard both forms as bound morphemes, as neither can be used separately. In addition, Telugu writing practice is not well standardized (Sproat 2003;Singh 2006) and cannot always be relied on as evidence for distinguishing between bound and free morphemes.
Encoding Path in case markers without conflating Motion clearly differs from the pattern in the so-called "verb-framed" languages like Spanish, where Path and Motion are typically conflated in the verb (e.g., Slobin 2004). There were some motion event descriptions with Path verbs in the data, as in (17), but even in these descriptions Path was also coded by the dative casemarker, and thus distributed over the verb and the case marker. When Path is distributed in languages like Spanish or French, this is over a verb which is dominant, and a preposition, more or less "required by the verb" as in (1). This is a pattern that is quite distinct from Telugu. 10 (17) oka wyakti mellagā tana gadi-lō-ki prawēśiMc-ā-d

Manner
Manner was expressed by finite verbs as in (18), or by verb-participles as in (19). Further, when either Path or Direction was expressed in the main verb, Manner was always expressed in a non-finite form as in (19). Any attempt to express Manner in the final, finite verb, and Direction in a non-finite participle, would be ungrammatical, as illustrated in (20)

Region and landmark
In line with the findings of Naidu et al. (2018), Region-expressing nouns (also known as "nouns of space and time", "nominal adverbs", or "adverbial nouns", see Krishnamurti and Gwynn 1985) were found to be essential for the motion event descriptions. Typically, they were compounded with a Landmark expression, as shown in many of the preceding examples and in (21). However, with the exception of -lō "in" they can also stand on their own as shown in example (10). Interestingly, together with Direction verbs like ekku ("ascend") it is not idiomatic to use the dative case marker -ku without a Region noun, as illustrated in (22). 12 Thus, the specification of Region is a precondition for Direction in such contexts. In contrast, the accusative case marker -ni may occur freely in such constructions, as shown in (23).

Summary
The qualitative evidence presented in this section shows three distinctive features in the way Telugu speakers describe motion events: • extensive use of generic Direction verbs • apparent dominance of case markers over verbs for expressing Path • common use of Landmark/Region expressions But are these features sufficiently frequent to be regarded as characteristic for Telugu, and hypothetically, the fourth cluster of languages that the present study aims to define? It is impossible to answer this question based on only a qualitative analysis, so we present a quantitative analysis of the data in the next section.

Quantitative analysis
In this section, we subject the three features established as typical for Telugu motion event descriptions to a statistical analysis. In addition, we address a fourth feature, claimed to be definitional for so-called "verb-framed" languages: the boundary-crossing constraint (see Section 1). Figure 2 shows how often Direction (mostly Viewpoint-centered, and predominantly expressed by deictic verbs), Manner, Path and Cause were expressed in the main verbs. Caused motion events were unsurprisingly characterized by the presence of a verb expressing Cause, but also by Direction and Manner verbs. Overall, Direction was expressed in the main verb approximately three to five times more often than Path or Manner. Path was expressed in the main verb only in the descriptions of uncausedbounded motion events. Across all four conditions, Manner was expressed more often than Path, but still considerably less often than Direction. The finding that Direction verbs in general, and deictic verbs in particular, are so dominant in Telugu motion event expressions, supports the proposal to distinguish it from languages like Spanish and French, on the one hand, and at least to some degree, from languages like Swedish and English, where deictic verbs may be a common, but hardly dominant strategy. Thus, we may regard this a criterial feature of Telugu and its corresponding cluster in an extended motion event typology.

Path expressed in case-marking
There were only 4 types and 18 tokens of Path-expressions through verbs in the descriptions, while Path was expressed by case markers 935 times (Figure 3). In fact, for three of the four event types, not a single Path verb was used. Another notable point in this context was the fact that whenever a Path verb was used, there was also a case marker to express it, making Path distributed, as in (17). A two-tailed paired t-test showed that the probability of choosing a Path-marking case marker over a Path-marking verb was statistically significant (t = 23.393, df = 29, p = 0.000).
A Telugu case marker that expresses Path can be combined with a main verb that expresses Path, Direction or Manner, as shown in Table 3. When Path is simultaneously expressed in the verb, it becomes distributed, as illustrated in (17). When the main verb expresses Direction, Path and Direction become co-expressed in the same description, as in (13). Finally, when a Manner verb is used as the main verb,  the so-called "boundary-crossing constraint" is violated. This is illustrated in example (24) and further discussed in Section 4.4.
(24) oka ammāyi toMdaragā oka sed _ -lōpali-ki parigettu-tuM-di. a girl rushingly a shed-inside-DAT run-PRS-3SG.F Figure  Landmark Manner+Motion Region:In Path:End 'A girl is running into the shed in a rush.' (Event 16,Participant 29) It should be noted that case markers cannot be regarded as "satellites" even in the Talmian framework, where the term was defined as "any constituent other than a nominal or prepositional phrase complement that is in sister relation to the verb root" (Talmy 2000, 222). What is less often noted is that Talmy, soon after stating this definition, qualifies how Path can be expressed in "satellite-framed" languages: Although the core schema in satellite-framed languages is largely expressed by the satellite alone, it is also often expressed by the combination of a satellite plus a preposition, or sometimes by a preposition alone. Such a "preposition" itself can consist not only of a free adposition, but also of a nominal inflection, or sometimes of a construction containing a "locative noun" (Talmy 2000, 222).
But this assumes that "satellites" somehow remain the main strategy for the expression of Path while other adnominal constructions are secondary. Apart from problems with blurring the category of "satelliteframing" such extension can be accused of assuming in advance that prepositions, and even more so case-marking are subordinate means when it comes to Path expression. As was shown above, this was clearly not the case for Telugu, which on the whole appears to favour an (ad) nominal pattern (nouns, cases) rather than an (ad)verbal (verbs, adverbs) pattern of Path expression (see Naidu et al. 2018). This is further discussed below.

Region and landmark
The feature of expressing Region and/or Landmark nominally was clearly characteristic for the Telugu motion event descriptions. There was not a single translocative motion event description without at least one of these two categories. Region was expressed by a set of dedicated spatial nouns, while Landmark was expressed by a noun or a pronoun. With the exception of the uncaused-unbounded event type, Region expressions (1022 tokens in total) were more frequently used than Landmark expressions (765 tokens in total) as shown in Figure 4. If we consider all instances of the five HSS categories under discussion (Path, Direction, Manner, Region, Landmark), and compare whether they were expressed (ad)nominally or (ad)verbally, as shown in Figure 5, the former form classes predominated for all types of translocative events. A two-tailed paired t-test also confirmed that the higher frequency of the (ad)nominals over (ad)verbals was statistically significant (t = 14.737, df = 29, p = 0.000).

Breaking the "boundary-crossing constraint"
As pointed out in Section 1, the so-called "boundary-crossing constraint" stating that a Manner verb cannot be used to describe a situation where a boundary is crossed is said to be criterial for "verb-framed" languages like French. For example, in (25), the typical meaning is that the person is running inside, rather than into, the garden.
(French) Marie AUX run-PTCP in DEF garden 'Mary ran in the garden.' While research has shown that in some allegedly "verb-framed" languages like Spanish and Turkish, the constraint is not always obeyed either, for example for rapid/instantaneous motion verbs such as Turkish daliyar "drive", or punctual vertical motion such as Spanish tirarse ("plunge/throw oneself") (Naigles et al. 1998;Özçalışkan 2013), Path verbs are clearly the norm in such languages.
In contrast, in our Telugu data, as in (24), Manner verbs in boundary crossing events including parigettu/uruku "run", dūru "penetrate/rush/ squeeze in", geMtu "hop", jāgiMg "jog", duMku "jump" were common. As shown in Figure 6, these were not as common as Direction expressions, but more common than Path expressions. Thus, we may regard this pattern as one more characteristic feature of Telugu, in addition to the three listed at the end of the previous section.

Discussion
While the study of Zlatev et al. (2021) was explicitly comparative, allowing the authors to contrast Telugu explicitly with Swedish, French and Thai, the present study allowed us to spell out in more detail four characteristic features of Telugu -and hypothetically, of the typological cluster to which it belongs. Using Holistic Spatial Semantics (Zlatev 1997;Blomberg 2014;Naidu et al. 2018), and focusing on how the semantic categories Path, Direction, Manner, Region and Landmark were expressed, we can generalize these characteristics in terms of the following four typological features: (a) Motion event descriptions require verbs, either finite or non-finite, and it is natural to focus on this parameter in typology. In Telugu, verbs predominantly express Direction (either Geocentric or Viewpoint-centered) rather than Path. Manner verbs play an intermediary role. (b) Limiting Path to the expression of bounded translocation (with the values Beginning, Middle, End), we established that it is predominantly expressed in Telugu through bound case-markers rather than verbs. In general, we consider that it is correct to regard Path as a central semantic category of motion events, but not as the pivotal criterion for typological classification, which is one of the reasons why we reject the notion of "framing", and the derivative categories of "verb-framed" and "satellite-framed" languages. We propose that these features define the prototype of a typological cluster that is distinct from those that approximate the original Talmian types (with Romance and Germanic languages being given as examples, even though it is mistaken to lump these like this, due to extensive differences; see, e.g., Berthele 2013), as well as from serial-verb languages like Thai. As stated at the onset, our main aim with the study was to establish the criteria for assigning other languages to this fourth typological cluster. While detailed future work is needed for this, we can so far tentatively propose that other Dravidian languages also fulfill these criteria, and hence belong to it. Tamil examples were presented in Section 1, but here we can add examples from  and  showing: (a) main Direction verbs in (26) and (28) (27) and (29). In other words, indications for all the proposed typological features.
(Malayalam) one boy room-DAT inside-ALL go-PST Figure  Landmark Region:In Direction+Motion Path:End 'A boy went into the room.' In addition, we may consider other, genetically unrelated, agglutinating languages such as Turkish, which is usually considered "verb-framed" in the Talmian tradition (e.g., Özçalışkan and Slobin 1999). We can observe strong similarities between Turkish and the examples given above, including the use of main Direction verb, Path expressed in the case marker and the presence of a Region noun (30). It is only the last feature that appears to differ, since Turkish is usually reported as being sensitive to the boundary-crossing constraint (Özçalışkan 2013) Yet, once we are freed from the straitjacket of a binary typology, and we operate with distinct clusters for, e.g., Spanish and Telugu, we could ask, to which cluster does a language like Turkish belong? Our proposed answer is that it would belong to the same cluster as Telugu, given the number of shared features. Other languages such as Finnish, which on the contrary have traditionally been classified within the "satellite-framed" construct, could very likely belong to this cluster as well, as recent work is suggesting (Tuuri 2021). FoR:GC 'The ball descended the hill while rolling.' (Furman 2012, 3)

Conclusions
Talmian motion event typology has contributed to many insights on motion expression across different languages over the past few decades. However, there is increasing agreement that the two original types are insufficient to capture all observed linguistic diversity (e.g., Fortis and Vittrant 2016). Using new empirical data from a language belonging to a language family that has been relatively unstudied (i.e., Dravidian), and a new theoretical framework, this article contributes to expanding the theoretical and empirical scope of post-Talmian motion event typology. While it is high time to leave Talmy's typology, we agree with Beavers et al. (2010, 332) that "any viable account should illuminate why Talmy's typology is so close to being right", and our approach does indeed help us in this way. Some, but not all, languages that have been analyzed as belonging to the "verb-framed" and "satellite-framed" types will correspond to two of an open number of typological clusters in a generalized post-Talmian motion event typology. Our main conclusion is that Telugu should be considered an example of a cluster that has earlier been conflated with others, due to the limits of an over-constrained typology. On the basis of the elicited material, complemented with native speaker intuitions, we conclude that this cluster can be characterised by four typological features: (a) preferential use of Direction verbs rather than Path verbs, (b) case markers to encode Path, (c) extensive use of Landmark and Region expressions, and (d) frequent use of Manner verbs in "boundary-crossing" situations.
We have suggested that these features help to distinguish Telugu from languages arguably belonging to distinct clusters, like French, Swedish and Thai (Zlatev et al. 2021), and at the same time help identify other languages that belong to the same cluster, both genealogically close ones like Tamil, and distant ones like Turkish and Finnish. Further detailed studies are necessary to support this claim and to understand the parameters of this typological cluster. Holistic Spatial Semantics, whose primary objective is to study the expression of space and motion from a usage-based and open-ended perspective, and the methodology that we use in the present study, should be able to contribute to this enterprise.
To sum up, further research is needed to fully understand the nature of the fourth cluster instantiated by Telugu. To achieve this goal, the typological features identified in the present study should be of help. Similarly, the other three clusters proposed by Zlatev et al. (2021) need to be complemented with the study of more languages, in order to understand the parameters of variation in post-Talmian motion event typology.