Direction and associated motion in Tibeto-Burman

3.1 Sampling process

The 23 languages used for this study were chosen with an eye to geographic and typological diversity, and with the goal of building a sample that represents distinct sub-groups within Tibeto-Burman.^[3] With regards to the latter point, it is important to note that there is a lack of consensus on the phylogenetic structure of Tibeto-Burman (Genetti 2016). Rather, there are multiple proposals of distinct (yet partially overlapping) structures. Following van Driem (2005, 2011, we take a maximally agnostic “fallen leaves” approach, grouping together those sub-families where there is strong evidence of genetic relationships and leaving for future work the final determinations of how these fit into higher-level arborial “branches.”^[4] The 23 languages of our sample represent 18 of the “fallen leaves” sub-families as identified by van Driem. The languages and their sub-families in our sample are given in Table 1.

Geographically, the languages range from central Nepal in the west to Sichuan province in the east, and cluster in the Himalayas. Their distribution can be seen on the map in Figure 1. Most of these languages are spoken in hilly – even mountainous – environments, but not exclusively; some are spoken on the flat river deltas south of the Himalayan range.

Figure 1:

Geographic location of languages (for color version, see https://orcid.org/0000-0003-2456-5025).

In broad typological terms, Tibeto-Burman languages run the gamut from isolating structures to significant polysynthesis, with the full spectrum of morphological complexity between. Our sample contains languages from across this typological spectrum.

3.2 Coding process

For each of the languages sampled, we examined reference grammars looking for evidence of DIR or AM encodings, and categorized these by their morphosyntactic properties, sorting them into three broad types: affixes and clitics, particles, and multi-verb constructions, the latter being used as a cover term for serial-verb and verb-auxiliary constructions that express DIR and AM meanings (see Anderson (2011) for a typological comparison of auxiliary verb constructions and other complex predicate types). For serial-verb constructions, we limited our study to asymmetrical SVCs, which allow only a restricted set of verbs in one of the verbal slots (Aikhenvald 2006: 3). These are thus similar to auxiliary verbs, which differ from restricted serial verbs in being syndetic, receiving some overt marking of their relationship with the primary verb (serial verbs are asyndetic and – within Tibeto-Burman – unmarked). In all cases, we included in our dataset elements that appear to productively combine with other verbs.

We included in our analysis some encodings that have a broader array of functions than DIR and/or AM. For example, the Karbi proclitic nang= is not only used for cislocative and associated motion. It also extends into person-marking in complex ways, and has both temporal and more general semantic extensions, such as “orientation toward a reference point” where no motion is involved (Konnerth 2015). In other languages, some DIR or AM encodings also function as tense/aspect markers or convey other meanings resulting from metaphorical extension. In these cases, while we included the overall encoding in our dataset, we excluded the examples where it had a function other than DIR or AM.

Central to our analysis is the concept of subsystems,^[7] created when one or more morphemes with Directional and/or AM meanings occur in a particular grammatical position, either affixal, cliticized, particle, or grammaticalized verb (serial or auxiliary). Of the 23 languages, six had two grammatically distinct subsystems and one (Eastern Kayah Li) had three (one prefix, one particle, and a set of serial verbs). Thus, our dataset covers 31 subsystems in all. The number of encodings in each subsystem ranged from one to nineteen.^[8] The grammatical classifications of subsystem and the number of morphemes in each is shown in the middle column of Table 2. Only six subsystems had dedicated AM markers. The numbers of encodings in these systems were comparatively small, with Daai Chin having the most at four.

The third column of Table 2 indicates whether the morphemes for a given subsystem exclusively or primarily encoded DIR, exclusively or primarily encoded AM, or encoded both.^[9]

For each subsystem, we compiled examples for each encoding, taking them from grammars, associated text collections, or fieldnotes. The majority of examples are from connected discourse, typically narratives.^[11] While we limited ourselves to one example for many of the encodings, we explored in more detail when the systems looked especially rich and when extensive discourse samples were available (e.g., Dhimal). Through this process we compiled a database of 238 examples, which served as the basis for this study.^[12]

Once we compiled the dataset, we coded each example for the features listed in Table 3.^[13]

As the coding of examples is central to our analyses, a few of these features and how we coded them merit discussion.

Regarding morphosyntactic type, we based our coding of systems as Affix/Clitic, Multi-V, or Particle on the descriptive terms used by grammar writers. However, we found that different linguists writing about the same morphemes sometimes describe them in different ways,^[14] or even that in the same language morphemes are analyzed as affixes in some examples and as serial verbs in others.^[15] There can also be variation between the descriptive analysis and transcriptional practice with respect to whether morphemes are represented as bound. For example, Sangdong transcriptionally represents Kadu serial verbs as a single word, linking them with a hyphen, although he describes them in the text as “juxtaposed” (2012: 171).

The primary reason for this lack of consistency is that what have been traditionally been called “directive” markers in Tibeto-Burman have a common grammaticalization pathway from motion verb to affix to lexicalization, a process that DeLancey (1985) has shown to be cyclic in the history of the family. The transitional nature of these encodings means that their status as verb or affix is not always clear. This falls out naturally from the graduality of grammaticalization (Lichtenberk 1991). Also, in some languages there is evidence of polygrammaticalization, whereby a single morpheme is the source of multiple grammaticalization chains (see Craig 1991 for an original account of this phenomenon). In coding these forms, we prioritized more extensive and recent works over older and shorter ones (Apatani), descriptive text over transcriptional practice (Kadu), and structures that appeared to be prevalent over those that appeared to be rare (Lamkang). However, we acknowledge that the act of coding itself requires us to place each example into one of three discrete categories, when in fact, the category boundaries are porous and some examples lie along a cline from verb to affix (Schiering et al. 2010). As a result, we do not make claims or draw conclusions based on the morphosyntactic classification of these constructions.

In coding examples for the motion category, we coded the primary verb as a motion verb if translational motion is a semantic component of the verb (‘go’, ‘run’, etc.). Transitive verbs that involve translational motion of the O, such as ‘put’, or ‘carry’, were coded as motion verbs, as were ‘cut (away)’, ‘saw (off)’, and ‘tear (out)’, as they entailed the movement of some piece of the O. The verb ‘give’ was coded as a non-motion verb, as its primary sense is change of possession and change of location is not entailed. Verbs that denote metaphorical or fictive motion were excluded from this study. We also coded motion verbs for type of motion; this is more fully discussed in Section 5.2.

Perhaps the most important coding involved the categorization of examples as DIR and AM. Examples were coded as DIR if the primary verb was a motion verb and if the relevant encoding provided additional information on the trajectory of motion. Examples were coded as AM if the encoding added a semantic component of translational motion that would not be present in the absence of that encoding. Belkadi (2016) discusses cases where a single encoding may have DIR interpretations in some examples and AM interpretations in others, depending on semantic and pragmatic factors (see also Bourdin (2005)). This is true for our data as well; we discuss this in detail in Section 5. We follow Guillaume (2016: 9) in considering such cases to be bona fide instances of AM.

We have included in our dataset examples of “purposive AM,” which we define as translational motion undergone for the purpose of carrying out a subsequent action. An example is the Dhimal “intentive” suffix, which is in paradigmatic alternation with the other AM suffixes in the language; see (10):

King notes that the intentive differs from the “distal” (i.e., prior AM) suffix, with which it paradigmatically alternates, in that “completeness of the movement does not necessarily entail completeness of the action… only that the subject went with the intention” (2009: 174). Lovestrand and Ross (in preparation) similarly differentiate purpose from prior AM in that with the former the occurrence of the event predicated by the non-motion verb is only implied (and so is cancellable), while in the latter it is entailed. In both cases the translational motion is asserted to occur and is associated with another verb event, even if that event is not realized. It thus meets the definition of AM.^[16] We do not include in our dataset examples of non-AM purposive constructions, which allow non-motion verbs as the main verb (e.g., ‘collected wood to light a fire’). These typically involve adverbial clauses in the languages of our sample.

A few encodings were categorized as AM although they might be considered non-canonical. One such case is the Eastern Kayah Li prefix -kə, which Solnit (1997) glosses as subject.moves. This morpheme explicitly indicates that the subject of the primary verb undergoes translational motion that would otherwise be absent, as shown in (11):

Solnit compares this to Ɂa phjá thɛ ‘he picks it up’, which lacks the prefix. Here, only the object undergoes motion. We interpret this as AM, although it differs from other cases in that it attributes the translational motion signaled by a grammaticalized verb to a particular argument of the main verb, rather than being a motion encoding itself. The result is translational motion of that argument that would otherwise be absent.^[17]

Somewhat related are two suffixes in Karbi. One is -pōn ‘on the way’, which indicates that the action of the verb takes place while the subject is in motion, e.g., lāŋ ēn-pōn [water take-take.away] ‘to take water as one moves along’. This is canonical AM, but it makes an interesting contrast with another suffix -dūn, which “profiles an event against a background of motion” (Konnerth 2014: 255) e.g., lāŋ pī-dūŋ [water give-join] ‘give water (to someone) as they move along’, where the figure is not the subject of the primary verb. Indeed, the figure doesn’t have to be an argument of the verb at all, as can be seen by thùi-dūn [wrap-join] ‘to wrap something for someone else to take as they go’. In both cases, the suffixes provide a translational motion component that would otherwise be absent, but they differ with respect to the attribution of the motion to the subject or to another participant.

Lastly, we classify encodings meaning ‘leave behind’ as AM. These occur in three of the languages of our corpus (Camling, Dhimal, and Karbi). A Dhimal example is given in (12); the morpheme in question is the “relinquitive” (relinq) -dhi:^[18]

This suffix clearly adds a translational motion component. When the verb is transitive, it profiles the fact that the O does not move while someone else does. When the verb is intransitive, the S of the verb does not move, although someone else does:

While the primary semantic force of the suffix is to indicate lack of movement of the absolutive argument, it does this by contrast to the motion of others (the A of a transitive verb or an un-named participant (non-argument) of an intransitive), resulting in separation (King 2009: 179–181). This is a quite distinctive type of AM.

We now turn to deeper analysis and exemplification.

5.1 Overlapping DIR and AM systems

There are at least three distinct ways by which a language can have both DIR and AM. First, a language can have discrete grammatical subsystems, with one set of encodings specialized for AM and another set specialized for DIR. Five languages of our sample are of this type. Second, a language can have structurally ambiguous examples that produce AM readings under one structural analysis and DIR under another. We only have one clear example of this type. Finally, one or more encodings in a subsystem may have DIR interpretations under one set of semantic and pragmatic conditions, and AM under another. Ten of the languages in our study are of this third type.

The first type, a language with discrete grammatical subsystems for encoding DIR and AM, is exemplified by Kyomkyo rGyalrong (Prins 2011). The language has two distinct set of prefixes: six DIR prefixes which occur outermost in the verb, as well as a distinct pair of prefixes which Prins calls “viewpoint” (see also Jacques 2013 for Japhug Rgyalrong). The forms are related to ‘come’ and ‘go’, and the examples typically translate as Prior AM and frequently have a purposive reading. Example (26) has prefixes from both sets. The perfective orientation prefix na- is the first element of the verb; the viewpoint prefix ʃi- follows the person index:

The second type of language with both DIR and AM is one where these interpretations result from underlying structural ambiguity. Our only apparent example of this is Lamkang, which has both prefixal directionals and a serial-verb construction. Example (27) has two possible meanings, one resulting from a structural analysis of directional prefix + verb, and the other from an analysis of “a sequence of two finite verbs which can also be interpreted as serial actions.”

In this language, serial constructions can have both verbs marked with tense-aspect morphology, but this can also be dropped, as in (27). In these languages, syncretism is common, as is polygrammaticalization, whereby a lexeme can develop into a grammatical marker, while still retaining a separate lexeme form and function (see Craig 1991 for an extensive study of polygrammaticalization in Rama). Two distinct structures, one resulting from reanalysis of the verb as a prefix, and the other with it retaining the status of a verb, underlie a single string of morphemes, leading to two distinct yet related semantic interpretations.

The third type of language with both DIR and AM is one where individual encodings allow different interpretations based on semantic and pragmatic factors. An example is Darma, where the auxiliary verb, r̃a ‘come’ has DIR interpretations with a motion verb ‘dance’, as in (28), and AM interpretations with a non-motion verb ‘meet’, as in (29):

Belkadi (2016) provides detailed discussion of similar cases in a number of African languages. She suggests that AM readings might arise in contexts where DIR readings are not available due to the lexical semantics of the primary verb and pragmatic factors (2016: 63). This assumes that DIR readings are basic and AM a secondary function that arises in particular contexts, a situation that Voisin (in preparation) classifies as DD-AM, where DD stands for “deictic directional.” A language like this in our sample is Daai Chin, which has a set of six preverbal directional particles. AM interpretations appear to arise sporadically, as in (30).

So-Hartmann writes of this example that juk- implicates downward motion that does not include the agent (2009: 285). Since the act of giving does not typically involve downward motion of the O, the prefix invites the pragmatic inference of translational motion over a distance, confirmed by So-Hartmann’s translation of the verb as ‘sent down’. So here the directional takes on an AM interpretation under utterance-specific semantic and pragmatic conditions.

A very different situation is found in Dhimal. This language has a set of five suffixes that King (2009) calls “deictic motion,” and which convey different types of AM. Of these, three can take on DIR interpretations under specific semantic conditions. Consider the suffix -pu (labeled “distal”). King writes that when this suffix is used with telic events “the action of the main verb takes place away from the referent and after the motion has occurred” (King 2009: 171). This results in a Prior AM interpretation, as shown in (31):

With atelic events, the action of the primary verb and the motion are simultaneous. If the primary verb is a non-motion verb, a concurrent AM interpretation is produced, as shown in (32). If the primary verb is a motion verb, then a directional interpretation is produced, as in (33).

A similar three-way split can be found with the “indeterminate motion” suffix -gil. With telic events, “the action is performed on an object, causing it to move in a direction away from the referent” (King 2009: 177), while with atelic events, “the event occurs in multiple locations and at multiple times in a haphazard, back and forth manner” (King 2009: 178). All the examples with motion verbs, however, have DIR translations. Examples are in (34) to (36).

Of the other three AM suffixes in Dhimal, two consistently have AM readings (one is a “leave-behind” construction, and the other is purposive AM). A third, the venitive suffix -pa, takes on DIR meanings with motion verbs.^[22]

We see that the Dhimal subsystem is preponderantly one of AM (i.e., is AM-DD in the terms of Voisin (in preparation)); only three of the five suffixes can have DIR readings, and these only occur with motion verbs. The range of AM meanings conveyed by Dhimal is complex, and critically depends on verbal semantics. However, the factors are identifiable and systematic. This is thus different from the Daai Chin case, where the subsystem is preponderantly a DIR system, with AM interpretations arising sporadically.

Thus we can see that the interplay between AM and DIR works in both directions: encodings that are primarily DIR can derive AM interpretations, and those that are primarily AM can derive DIR interpretations. Underlying all of this is verbal semantics: some classes of verbs have a natural affinity for DIR and others for AM.

5.2 Direction, associated motion, and motion-verb type

So far, we have been treating “motion verbs” as a unified lexical class. However, different types of motion verbs have different levels of affinity for DIR or AM readings, suggesting a more fine-grained analysis is advisable, a point made by Belkadi (2016). To examine this in cases where the interpretation as DIR or AM is based on semantic or pragmatic factors, we categorized all of the motion verbs by type, including (following Belkadi) path-of-motion (e.g., ‘come’, ‘ascend’, ‘insert’, ‘drop’), manner-of-motion (e.g., ‘hop’, ‘fly’, ‘run’), and causative-motion verbs (e.g., ‘pull’, ‘chase’, ‘toss’). We also included a category for “general motion” (‘move’, ‘roam’).^[23] The number of examples in each motion type, by AM/DIR, is presented in Table 5. For comparison, examples with non-motion verbs were included as well.^[24]

These figures demonstrate that AM encodings do occur with motion verbs, although DIR encodings with motion verbs are much more common in our data set.

Our data contains four examples of verbs of general motion – i.e., those that do not lexicalize path, manner, or causation. Of these, two occurred with maximally general motion verbs (translated as ‘move’ and ‘move around’). In both cases, the encodings provided path specifications and the resulting interpretation was DIR, as in the Eastern Kayah Li example in (37):

Not surprisingly, motion verbs which explicitly lexicalize a lack of direction (‘roam’, ‘wander around aimlessly’, counted as “general motion” in Table 5) also do not occur with DIR encodings in our data. As indicated in Table 6, our dataset contains two such examples with AM; both have concurrent AM markers that redundantly indicate the lack of a distinct motion trajectory.^[25]

The remaining three categories of motion verbs are those discussed by Belkadi: those that lexicalize manner of motion, path of motion, and causative motion. She proposes a ranking of motion-verb types based on how likely they are to trigger AM or DIR meanings: path-of-motion verbs > manner-of-motion verbs > causative-motion verbs. She predicts that verbs to the right of this ranking are more likely to trigger AM, while those to the left are more likely to derive directional meanings (2016: 64).^[26]Table 6 shows the number of examples of each type in our dataset that occur with AM and with DIR encodings:

AM encodings are most likely to occur with causative verbs, as Belkadi’s ranking predicts. These examples are all transitive and typically involve tandem motion of both the A and the O; an example from Lamkang is in (39):

The only example of AM with a manner-of-motion verb in our dataset is a Karbi example with a verb-stem meaning ‘steer’; this is given in (40):^[27]

There are seven path-of-motion verbs with AM in our data. These are of mixed types. Two are Dhimal relinquitive “leave-behind” examples, which allow a semantic disjunct between the primary verb and the motion event resulting in someone or something being left behind. In (41) a causative relationship between the event of the primary verb and the motion event can be inferred, but it is indirect; one event (falling out) causes the other to occur (they are left behind):

Another path example has a concurrent AM suffix and distributive aspect:

The remaining example of this type is that with the somewhat unusual Eastern Kayah Li kə- ‘subject moves’ prefix exemplified in (11) above.

In sum, causative-motion verbs have the most robust pattern of occurrence with AM encodings, while path-of-motion verbs that occur with AM are somewhat idiosyncratic, and manner-of-motion verbs are rare.

Looking at the DIR column of Table 6, we see that examples with DIR encodings are most likely to have path verbs, followed by causative, followed by manner. In addition, manner verbs are much more likely to occur with DIR than AM. Thus, although we must be tentative due to the small number of examples, overall the Tibeto-Burman data seem to confirm Belkadi’s proposed ranking, especially with regards to causative verbs being the most compatible with AM and path- and manner-verbs being more compatible with DIR.

6.1 Prior AM

Of the three temporal relationships, prior AM examples are the most semantically consistent. Most denote movement to a location with the intention of carrying out the action of the verb (‘go to VERB’) or indicate movement to a location followed by the action of the primary verb (‘go to VERB’). Accordingly, the figure in Prior AM examples is limited to the nominative category (A or S). In most examples, the motional component indicates a simple trajectory towards or away. However, we do find lexicalization of path in some encodings, as in the following examples from Apatani and Tangkhul Naga:

6.2 Concurrent AM

Examples of concurrent AM can be classified into three types. One denotes iterative actions that are carried out as the nominative argument (A or S) moves from place to place; these are the verbs which lexicalize lack of direction. An example is the Dhimal suffix -gil, which King glosses as “indeterminate motion” (im):

The second type of concurrent AM denotes continual or iterative movement as one or more participants move along a trajectory toward or away from the deictic center. While the path may not be specified precisely, a path is implied. In these examples, we do not find lexicalization of either manner or more complex path trajectories.

While the figure in this type is typically the S or A of the primary verb (45), one also finds tandem motion where both the A and the O undergo movement, as in (46). Karbi provides an interesting contrast with the suffix -dūn, which explicitly denotes that the figure is other than the A or S (non-nominative). In (47), the figure is the two objects.

The third type involves semantically rich encodings. The following Kadu example has a semantically complex V2 in a serial-verb construction, although it is bleached in this concatenation. There is a causative relationship between the action of the primary verb and the motional component of the serial verb, unlike in the examples above:

6.3 Subsequent AM

Subsequent AM is the most semantically diverse of the three temporal categories. One notable subtype is the “leave behind” category. These indicate cessation of tandem motion, but profile the participant left behind, rather than the figure, that continues to move.

Here the first action temporally is that of the primary verb – the giving of the flutes – that is then followed by the movement of the A away, resulting in the O being left behind. However, the figure does not have to be an argument of the primary verb. In (50), the primary verb is intransitive and the S of that verb does not move, but is left behind as some other unspecified participants move away:

In addition to “leave behind” examples, several languages have subsequent AM constructions where the motion is toward or away from another deictic center at the conclusion of the action of the primary verb. With these cases the figure is either A/S or tandem:

With subsequent AM, it is also possible for the figure to be the O. These either involve encodings translated as ‘send’, as in (53), or causative motion where the action of the primary verb results in the movement of O, as in (54):

In sum, although the Tibeto-Burman data of our sample do not confirm the proposed hierarchy of prior motion > concurrent motion > subsequent motion, we do find distinct semantic patterns in these three types of temporal relationships that have not been previously noted in the typological literature. Prior AM examples involve simple motion along a path, typically towards or away, although other paths are attested in our data. In our examples, the figure is consistently either A or S. Concurrent AM examples most commonly encode iterative activity that occurs along an indeterminate path, or iterative or continuous activity that occurs with motion along a determined path. In both cases, the figure is A/S or tandem. There are also some examples of concurrent AM with semantically rich encodings of varying types, some of which involve movement of the O. Finally, the examples of Subsequent AM are the most diverse semantically, although many are either “leave behind” examples or motion towards a deictic center at the conclusion of the action of the primary verb. Subsequent AM examples show the most variation of the figure: A, S, O, tandem, and even participants that are not arguments of the primary verb.