Differential place marking beyond place names: Evidence from two Amazonian languages

Not all spatial adjuncts behave alike. In some languages, certain spatial adjuncts display different marking or different combinatorial possibilities than others. Recent functional-typological studies make two claims about this differential place marking phenomenon: (1) it is primarily motivated by noun semantics, opposing place names and other nouns; and (2) it is primarily realized as a contrast between zero-marking and overt marking of spatial adjunct function. We evaluate this view against new fieldwork data from two Indigenous Amazonian languages, Dâw (Naduhup) and Ticuna (isolate). In Dâw and Ticuna, differential place marking is conditioned not only by noun semantics, but also by the perceptual properties of noun referents and by morphosyntactic and semantic properties of verbs. Further, the phenomenon is realized not only by alternations between zero and overt marking, but also by alternations among overt markers and alternations in the set of markers with which the adjunct can combine. These findings suggest that differential place marking – like other differential case phenomena, such as split ergativity – is conditioned by a cross-linguistically diverse suite of factors within and beyond the noun phrase.


Introduction
Not all spatial adjuncts behave alike.For example, in languages such as Yucatec (Mayan), Kilivila (Austronesian), and Tswana (Bantu) (Stolz et al. 2014: 109, 119, 145-147), place names are zeromarked in spatial adjunct function, while other noun phrases receive overt marking.Prior studies of this phenomenon, where some spatial adjuncts are marked differently from others in the same semantic role -for example, with vs. without a preposition -have christened it differential place marking (DPM; Haspelmath 2019).These studies suggest that DPM is driven primarily by the lexical semantics of the adjunct noun phrase, especially its status as a place name, and is realized primarily as alternation between zero-and overt marking of the spatial adjunct function (Stolz et al. 2017: 456;Haspelmath 2019: 319).Under this analysis, DPM -with its simple form and its close relationship to lexical semantics -looks like a reflex of noun categorization, analogous to the differential plural marking of mass vs. count nouns.
In this study, we propose an alternative view of DPM: not as noun categorization, but as a form of differential case marking.Presenting fieldwork data from two Amazonian languages, Dâw (Naduhup, Brazil) and Ticuna (isolate; Brazil, Colombia, Peru), we show that DPM can be conditioned not only by the lexical semantics of the noun phrase, but also by properties of the verb and clause, such as lexical aspect and transitivity -much as split ergativity, another kind of differential case marking, can be conditioned either by noun or verb properties.We further demonstrate that DPM can be realized through alternations between overt markers of equal complexity, as well as through alternations between zero and overt marking -again, much as differential object marking can be realized via alternations between overt accusative and overt oblique cases with object-marking functions, as well as between overt and zero markers.In other words, while DPM is sensitive to lexical semantics, it is not simply a reflex of noun categorization or the semantic role associations of nouns (cf.Haspelmath 2019).Rather, DPM is a differential case phenomenon which departs from other forms of differential case marking only in that it affects adjuncts rather than arguments.
The paper is structured as follows.In §1 we summarize the main arguments made about DPM in the functional-typological literature.In §2, we introduce the study languages and data sources.Over the following two sections, we examine the conditioning and form of differential place marking in Dâw ( §3) and Ticuna ( §4).In §5 we discuss our findings in the light of prior approaches to DPM, and in §6 we conclude.

Approaches to DPM
DPM is often conditioned by properties of the place-marked noun, and it often involves alternations in the length or complexity of marking.Some analyses of DPM focus on the conditioning role of nouns, treating the phenomenon as a form of noun categorization.Other, complementary mediated by role-referent association (hence, role association; Haspelmath 2021) -the concept that a noun's complexity of marking in a particular semantic role varies inversely with its degree of semantic association with that role (often conceptualized as frequency in the role).In figureground constructions, these authors argue, what-nouns typically appear as figures.When they instead appear as grounds, they deviate from their usual semantic roles and therefore require more explicit place marking.In contrast, where-nouns require less marking because ground is their default semantic role (Rybka 2014: 41).

Alternation analyses
Approaches to DPM which focus on the form of the alternating place markers, such as Haspelmath (2019) and Stolz et al. (2014), depart from categorization-based accounts in several ways.First, while categorization-focused analyses of DPM focus on the distinctness of marking between noun categories, alternation-focused analyses emphasize differences in the length and complexity of marking.This emphasis is evident from authors' comparisons of ground phrases formed with place names vs. with common nouns.For example, Stolz et al. (2017: 455) write that the marking of place name grounds is "less complex," while Haspelmath (2019: 317) states that typologically, place names bear "shorter" marking and are more likely to appear zero-marked.
Within this literature, these comparisons are important because they support larger claims about correspondences between form length and role association (Haspelmath 2019: 331).
Second, while categorization-focused analyses see DPM as conditioned primarily by the realworld properties of nominal referents, alternation-based analyses posit that it is also conditioned by the lexical properties of nouns.These authors acknowledge that referent properties like animacy can impact DPM, but treat lexical category -status as a place name -as the most important influence on a noun's DPM behavior: place names are more likely to display zero-marking than any other noun type (Haspelmath 2019: 319;Stolz et al. 2017: 456).Despite this difference, alternation-focused analyses, like categorization-focused ones, still view role association as the most immediate cause of DPM.In this framework, the zero-marking of place names in ground function is explained as showing that -because of these nouns' place-denoting semantics -they frequently appear in the ground role (Haspelmath 2019: 315-317).Thus, despite their lesser emphasis on the real-world basis for categorization, alternation-based analyses still see DPM as driven by a form of noun categorization -i.e., categorization as an item that more vs.less frequently or prototypically appears as a ground.
Last, alternation-focused accounts examine a different kind of data than categorizationfocused ones.Categorization approaches describe the DPM systems of specific languages, while alternation approaches describe typological trends.Furthermore, where categorization approaches analyze the behavior of individual nouns, alternation approaches offer explanations for the cross-linguistic behavior of entire noun categories.

DPM vs. differential argument marking
Under both alternation-and categorization-focused approaches, DPM centers on alternations between more vs.less complex marking of spatial adjunct function; is controlled by lexical semantic properties of the noun; and is mediated by role association.This view strongly resembles functional-typological analyses of differential object marking (Bossong 1991).Haspelmath (2019: 314) makes the analogy explicit, arguing that DPM and asymmetric differential object markingsuch as the alternation between zero-marking of inanimate and overt marking of animate objects in Spanish -are driven by the same universal pressure toward more explicit marking of less expected semantic roles.Haspelmath's (2019) analogy is insightful for reconceptualizing DPM as a form of differential case marking, rather than a reflex of noun categorization.Yet if we view DPM through the lens of differential case marking of arguments, we predict that the phenomenon will display much greater diversity in both conditioning and form than observed in prior studies.While existing research views DPM as conditioned only by the lexical semantics of the noun, differential object marking can also be conditioned by the definiteness and topicality of the noun phrase (Aissen 2003) and the TAM properties of the entire clause, as in Palauan (Austronesian; Woolford 2000).Similarly, differential object marking -like DPM -is most often realized, and sometimes even defined (e.g., by Aissen 2003), as alternation between zero and overt markers.But it can also be realized by symmetric case alternations, for example between an overt accusative case and an oblique case with certain objectmarking functions (Iemmolo 2013: 397;Malchukov & de Swart 2008: 347).Differential subject marking phenomena, such as split ergativity, display an even more typologically diverse array of conditioning factors -not only noun phrase properties, but also clause properties like tense, aspect, mood, and main vs. dependent clause type -and realizations (Hopper & Thompson 1980;de Hoop & de Swart 2009).
In other words, if DPM resembles the differential case marking of arguments, we expect that it will be conditioned by a cross-linguistically diverse suite of factors.These will include the lexical semantics of the noun (i.e., status as a place name or denoting a place), but may also -as with differential argument marking -include information-structural features of the noun phrase, as well as TAM and other features of the clause.Similarly, if DPM is fully analogous to differential argument marking phenomena, asymmetric alternation (e.g., between zero and overt marking) will be typologically most common, but symmetric alternation will also be present.In the remainder of this study, we demonstrate that the DPM systems of Dâw and Ticuna lend strong support to these predictions, and less support to an account of DPM as noun categorization.

Working definitions
Because studies of DPM use widely varying terminology, we offer the following working definitions before introducing the language data.We define differential place marking as any phenomenon where two spatial adjuncts with the same semantic role display different morphological possibilities for marking of the adjunct role.We consider adjuncts to have the same semantic role if they are both goals, both sources, or both locations.We refer to proper nouns denoting places interchangeably as place names and toponyms.Following Stolz et al. (2014: 42), we refer to nouns that denote places -landscape terms, landform terms (e.g.mountain, valley), and place names -collectively as place nouns.Following the convention of DPM literature (Stolz et al. 2014), we refer to all other nouns, including the names of animates, as common nouns.We avoid the terms what-noun, where-noun, and topo-noun (Haspelmath 2019: 322) because they do not map well onto the noun categories proposed in our analyses.

Dâw
Dâw (ISO-639: kwa) is a Naduhup language spoken by 142 people.Speakers live in a single community near the town of São Gabriel da Cachoeira in northwestern Brazil.
All Dâw examples are from language documentation conducted by the second author and the Waruá community.Most data come from landscape descriptions and personal and traditional narratives.Elicited data was collected using standardized stimuli, such as space games (e.g.Ameka et al. 1999), images (e.g.Bowerman & Pederson 1992)  The Ticuna consultants ranged in age from 36 to 74 years and all spoke Ticuna as their sole native language.Six spoke Spanish as an L2, while two had only passive knowledge of Spanish.

Ticuna
All Ticuna data is publicly available in the California Language Archive (CLA).Below each example, we identify the consultant who produced the example, then cite the archival source of the example by CLA accession number and filename.Transcriptions of Ticuna use IPA, except that IPA [ɾ] is written <r>.Numerals represent lexical tone; 5 is the highest tone.

DPM in Dâw: Between boundedness and markedness
DPM in Dâw is realized mostly by alternation in the combinatorial possibilities available to nouns, involving categorical alternation between markers only to a limited extent ( §3.2).Furthermore, while DPM is controlled by noun semantics, the semantic classes involved partition the lexicon much more finely than seen in previous categorization-based analyses.
The Dâw examples in (1)-(2) provide initial evidence for DPM.Here, the alternation between configuational postpositions such as ked and the generic locative marker rid is rooted in the semantic class of the noun heading the spatial adjunct.As we will explore in §3.1, in Dâw human, inanimate, and place-denoting nouns can combine with a rich inventory of configurational postpositions, such as ked in (1).However, these classes of nouns cannot be zeromarked, nor can they combine with the generic locative marker rid.In contrast, spatial adjuncts headed by a toponym, as in (2), are compatible only with the generic locative rid.
(  , ailla:254700, 20130723_historia_MFM.wav, 6:50-7:30) As we discuss in more detail below, the postpositions heading the adjuncts in (1)-(2) differ in meaning and with respect to the types of arguments they select.Configurational postpositions such as ked express many different spatial relations (with equivalents to in, at, on, and so on).In contrast, rid lacks any configurational meaning.It collapses the roles of location, goal, and source and contributes   ailla:254700, 20130723_historia_MFM.wav, 0:40-1:10) Thus, Dâw displays three place-marking strategies: configurational postpositions (1), the generic locative marker rid (2), and zero-marking (3).Two properties of the nominal referent control the alternation between these strategies: perceptual boundedness and size.
Perceptual boundedness is defined by a referent's intrinsic perceptual properties.Firstorder entities -three-dimensional, self-moving or manipulable entities, like humans, animals, or prototypical objects (e.g.baskets) -show clear perceptual boundaries (Lyons 1977).In contrast, place nouns, including landscape or landform terms such as forest and sky, usually lack such boundaries (Cablitz 2008;Smith & Mark 1999).Furthermore, some geographic formations, such as ponds, cliffs, and lakes, occupy an intermediate position between these poles.They have clear perceptual boundaries, but are not manipulable.
DPM in Dâw displays a correlation between perceptual boundedness and markedness: referents which are larger and less perceptually bounded have more restricted marking possibilities.Common nouns (those denoting smaller objects, with clear perceptual boundaries) can be marked with a variety of configurational postpositions.Landscape terms denoting larger but bounded referents are incompatible with most of these postpositions, instead allowing only one configurational postposition each.Only landscape terms that lack perceptual boundaries at all (e.g.sky) can occur zero-marked.Toponyms occupy a special place within this continuum.Though landform terms, such as nâax 'river,' can combine with postpositions, toponyms -even toponyms formed with landform terms, such as wâan nâax 'Curicurari river' -combine only with rid.

Nouns that combine with configurational postpositions
Dâw displays 14 spatial postpositions.They mark common nouns and some place nouns in spatial adjunct function.Semantically, they express configurational relations between the figure and ground, equivalent to English prepositions such as on, in, under, at, and below.In this section, we explore the properties which make nouns compatible with spatial postpositions, examining how size and degree of boundedness influence this compatibility.
Recall from §3 that first-order entities (human and object-denoting nouns) usually have clear perceptual boundaries.In line with this, figures can participate in a range of spatial relations with first-order entity grounds.For example, consider the inanimate noun yeg 'hammock.'4)-( 6) point to a first correlation between boundedness and markedness.First-order entities -which have clear perceptual boundaries and extension in space -have very extensive combinatorial possibilities, potentially combining with any spatial postposition in Dâw.
At the same time, some geographic entities, such as islands, ponds, and mountains, have fairly perceptible boundaries.Dâw nouns denoting these referents can also combine with configurational postpositions.However, their combinatorial possibilities are restricted, usually to just one postposition that expresses the default spatial relationship between a figure and the relevant geographic entity.Manmade landscape terms -such as taaw 'town,' kaaw 'manioc garden,' and xam 'soccer field' -provide a prominent example of this restriction.All of these landscape terms appear as complements of the postposition wâ' 'on,' as illustrated by ( 7). (7) [taaw wâ'] ground daad têen rid-i' town on write now 3pl-foc 'Now they are studying in the town (São Gabriel da Cachoeira) [landscape term].' (NMS, ailla:254700, 20130725_narrativa_NMS.wav, 3:40-3:55) In the topographic context of Northwestern Amazonia, towns, gardens, and fields can be understood as clearings within the dense tree cover of the rainforest, and their boundaries are clearly perceptible to Dâw speakers.At the same time, these landforms typically show large extensions in space -much larger than human referents or objects.This restricts the possible spatial relations between a figure and these grounds.
Like manmade landscapes, islands and river beaches also show clear perceptual boundaries delimited by water.As a result, figures are usually understood to be wâ' 'on' these places, as illustrated in (8).In sum, for nouns compatible with at least some postpositions, DPM is controlled by ontological properties of the nominal referent, primarily size and boundedness.It is differential in that nouns differ in their combinatorial possibilities -how many different postpositions they may combine with -and not in that nouns differ in the length or complexity or marking.Furthermore, while properties of the nominal referent clearly impact DPM, animacy (treated as important to DPM by Haspelmath 2019: 323) does not.Humans and other animate referents pattern together with inanimates, reflecting their shared status as first-order entities.

Nouns that combine with the locative marker rid
Besides the configurational postpositions, Dâw also displays a generic locative marker rid.This item can mark all place nouns functioning as spatial adjuncts, but does not mark common nouns in this position.Thus, compatibility with rid is the most notable morphosyntactic cut-off point between common nouns and place nouns in Dâw.
The generic locative rid is 'generic' because it can mark all three spatial roles on place nouns: location in static spatial scenes (10)  Nouns that are compatible with the generic locative rid function as grounds per se, since they are permanently anchored in space.Furthermore, their specific ontological properties -lesser perceptual boundedness, larger size, less crisp spatial boundaries, and immobility -all imply that these nouns function more readily as grounds (Rybka 2015;Talmy 1983).
Many landscape and landform terms can combine either with (one) spatial postposition or with the generic locative marker, as exemplified in ( 14).This interchangeability does not alter the spatial relationship that holds between figure and ground.However, the alternation between configurational postpositions and the generic locative marker again seems to be conditioned by the perceptual properties of the place noun.Landscape and landform terms that denote referents with clearer perceptual boundaries, such as kaaw 'manioc garden,' 'mõr 'pond,' and paas 'mountain,' are preferentially marked with spatial postpositions.In contrast, landscape and landform terms that denote referents with less clear perceptual boundaries and larger size -such as xaay 'forest' and pox 'sky' -allow only marking with rid or zero-marking (cf.§3.3).( 14 Dâw also exhibits a small class of nouns that represent homophonous substance/landform pairs, including nâax 'water/river,' çax 'earth/territory,' paas 'stone/mountain,' and xeew 'sand/beach' (Obert 2019).When these nouns denote a substance, they are marked with a configurational postposition, as in ( 15), but when they denote a landform, they bear the generic locative marker rid, as in ( 16).Thus, for this specific class of nouns, DPM is meaningful -it contributes to the semantic disambiguation of grounds in locative constructions via coercion of mass nouns to count nouns.The DPM behavior of the substance/landform pairs also shows that Dâw does differentiate morphosyntactically between object/substance/human nouns and place nouns.['liw' mĩi'] ground mãay çeeb pee bug dâw êe' Kariwa.Creek in.liquid not.bechange.placego.upriver there Dâw.people dub 'Wasn't it at the Kariwa creek that the Dâw moved upriver?' (CMM, MIRJ, daw_22112017_ko.wav,0:40-1:10) Comparing toponyms to their corresponding landscape or landform terms -as in Solimões River vs. river -reveals a syntactic contrast.( 19) shows that a locative adjunct headed by a toponym, Tumbil 'Tumbira community,' is marked with the generic locative marker rid.In contrast, (20) shows that the corresponding landscape term nĩr xoot 'community' is marked with the postposition wâ' 'on'.This suggests that the ontological properties of toponyms are not the only factors responsible for their DPM behavior -since the toponym in ( 19) and the landscape term in (20) have similar conceptual properties but distinct marking.Rather, the alternation here is syntactically motivated, driven by word class: toponyms can combine only with the generic locative marker rid, while landform terms can be marked with either rid or a configurational postposition.
In sum, DPM in Dâw primarily affects nouns' compatibility with the generic locative marker rid.Real-world properties of the nominal referent seem to influence compatibility with this marker.
Nouns that denote places with larger size and less clear perceptual boundaries may combine with rid, while common nouns (animate and inanimate) cannot.These observations illustrate that, in line with the DPM literature, Dâw place nouns receive different spatial marking than common nouns.
At the same time, these findings also contrast with some prominent ideas about the realization of DPM.DPM in Dâw is realized by different, but not shorter, marking of place nouns.This departs from two typological generalizations about DPM: (1) that whenever place nouns and common nouns are marked differently, place nouns receive shorter marking (Creissels 2009;Haspelmath 2019: 319), and (2) that whenever a language displays zero-marking of spatial adjuncts, the zero-marking will apply to toponyms (Stolz et al. 2014: 291).At a conceptual level, the Dâw data also demonstrates that DPM alternations can affect not only the syntagmatic complexity of place marking -that is, the length or morphological composition of the marker(s) with which the noun combines -but also its paradigmatic complexity, or the number of possible markers with which the noun can combine.

Nouns that display zero-marking
Differential place marking in Dâw can also involve zero-marking.Zero-marking applies only to four landscape terms: xaay 'forest', pox 'sky,' tuu 'ground', and xoot 'place.'All four nouns denote large referents which lack clear perceptual boundaries.As with the landform terms in §3.2, these properties prevent the nouns from combining with a configurational postposition.But in contrast to the landscape terms, these four nouns predominantly occur zero-marked in our corpus.They also allow, but do not require, marking with rid (21, 22).[tuu] ground ✓(rid) and.then man knock.downlie.on.ground rep woman-obj ground loc 'And then, they say, the man knocked the woman down on the ground.'(DFS, ailla:254700, 20130725_historia_onca_paje_DFS.wav,3:00-3:20)

Interim summary
DPM in Dâw is motivated primarily by the ontological properties of noun referents.Evidence for this claim comes from nouns' varying compatibility with configurational postpositions vs.
the generic locative marker vs. zero-marking.As summarized in Figure 1, the cut-off point between common nouns and place nouns in the DPM system of Dâw is compatibility with the generic locative marker rid.While human and object nouns are not compatible with rid, all nouns denoting places can combine with it.We can thus observe a relation between boundedness and markedness: the larger an entity is and the less perceptible its boundaries are, the more restricted are the marking possibilities of the noun which denotes it.
Furthermore, DPM gains semantic significance in contexts where marking can alternate between postposition and rid marking, as we have seen for the substance-landform pairs above.
In these cases, change in marking leads to systematic changes in interpretation of the noun: more generic marking with rid supports a landform reading of the noun, while postpositional marking leads to a substance reading.
On the one hand, as Figure 1 shows, all types of place nouns in Dâw pattern together in one way: they are compatible with the generic locative marker rid.Because this aspect of the system divides place nouns from all other nouns, it supports the core claims of role association theories.
But on the other hand, different classes of place nouns pattern apart in their compatibility with zero-marking and configurational postpositions, and this split is not simply conditioned by status as a toponym.
Thus, the DPM system of Dâw expands prior role association theories in two ways.First, this system provides evidence that place nouns are not a homogenous class.Fine-grained semantic distinctions can trigger distinct marking of subclasses within the class of place nouns.Second, while the data shows that place nouns and human/object nouns are marked differently, the marking of place nouns is not shorter or less morphologically complex.
Though our Dâw findings challenge some aspects of role association theories, they support these theories' premise that DPM is motivated primarily by noun semantics.This contrasts sharply with the DPM system of Ticuna, which we discuss next.

DPM in Ticuna: Controlled by verbs, not nouns
While role association theories suggest that DPM is conditioned entirely by properties of the noun, this is not the case in Ticuna.Instead, in this language, DPM is controlled primarily by morphosyntactic and semantic properties of the verb.Verb-controlled DPM in Ticuna shows that a cross-linguistically adequate account of DPM cannot rely entirely on semantic or morphosyntactic properties of nouns.Rather, DPM is sensitive to both properties of the place-marked constituent and properties of the clause which contains that constituent.
We discuss three forms of verb-controlled DPM in Ticuna: DPM controlled by the verb's conjugation class; DPM controlled by the verb's lexical aspect; and DPM controlled by the verb's transitivity.

DPM is not controlled by nouns
As context for our discussion of verb-controlled DPM, we first provide background on the general system of place marking in Ticuna.When Ticuna nouns appear as adjuncts, they always combine with either the allative case enclitic =wa⁵ or the locative case enclitic =gu².(We label =wa⁵ as allative and =gu² as locative because of their distribution in intransitive motion clauses, discussed in §4.4.The names should not be taken as claims about the markers' meaning or distribution in any other type of clause.) The same requirement for overt case marking with either =wa⁵ or =gu² applies equally to all spatial adjuncts.It does not distinguish between the roles of location (23), goal (24, 25), and source (26); nor does the requirement for case marking distinguish between place names ( 23 (LWG, 2015-06.049, tca_20180718_lwg_ahs_tak.wav, 5:41) Zero-marking of spatial adjuncts never occurs in our corpus materials, which contain over 110,000 words.It is also judged unacceptable in elicitation, including for place names ( 27). ( 27) [Kɨ³ʔtʃi³tu¹*(wa⁵)] ground na⁴ʔũ⁴³ Kɨ³ʔtʃi³tu¹=*(wa⁵) na⁴=ũ⁴³ Cushillococha=*(all) 3sbj=go:sg.sbj'They (sg.) went to Cushillococha [place name].' (DGG, 2015-06.040, tca_20170823_dgg_ahs_elicit_001.wav, 9:25-9:46) Since nouns from all DPM-relevant semantic categories behave the same in place marking (23-26), we conclude that Ticuna does not have DPM controlled by noun semantics.

DPM is controlled by the morphological class of the verb
Ticuna has an extensive system of verb inflection classes.The inflection classes sort verb stems in several orthogonal ways: by the form of the subject proclitic, by the form of third-person object marking, and by the syntax of spatial adjuncts.In this section, we are concerned just with the classification which sorts verbs by the syntax of spatial adjuncts.
The spatial adjunct classification divides verbs into three classes, shown in Table 1.Two classes of verbs require a spatial adjunct, realized as either a free noun phrase or a proclitic.They contrast in the case which they assign to the adjunct.The other class of verbs does not require a spatial adjunct.
Verbs that do not require a spatial adjunct, such as the transitive verb ɟau¹ʔ 'take, acquire,' are grammatically acceptable with only their subject and object arguments.The citation forms of these verbs include subject and object proclitics only, as in ( 28). ( 28) tʃa³na³ɟa¹ʔu² tʃa³=na³=ɟau¹ʔ 1SgSbj=3obj=acquire 'I took it.'(Anon., 2015-06.041,tca_20170530_ecp_ahs_elicit_001.wav,1:17-1:30) Verbs that require a spatial adjunct are not grammatical with only subject and object arguments.
In citation form, they take an additional proclitic before the subject proclitic.For some obligatory spatial adjunct verbs, such as ta̰ ¹ 'discard,' this proclitic has the form i⁵ = (29).For others, such as the homophonous verb 'bury,' the additional proclitic has the form i² = (30).
( 29 Likewise, if a verb takes the additional proclitic i² = in its citation form, then when it appears with a spatial adjunct, it assigns the locative case marker =gu² to the boadjunct.As before, the i² = proclitic deletes, as shown for 'bury' in ( 32).
Thus, the best analysis of verbs such as 'bury' and 'discard' is that they always require a spatial adjunct with a particular case value -locative or allative.In clauses that include a noun phrase adjunct, such as ( 31) and ( 32), this requirement is satisfied by the presence of the casemarked noun phrase.In the citation forms of these verbs, the requirement is satisfied by the i²=/i⁵= proclitic.These proclitics effectively act as expletive pronouns which absorb the verb's requirement for a spatial adjunct.The i⁵ = proclitic is an allative case pronoun, and the i² = proclitic is a locative case pronoun.
As the above examples suggest, whether a verb root requires a spatial adjunct, and what case it assigns that adjunct, is an essentially morphological property.Verb roots' phonological form, other morphological properties, and semantics do not categorically predict their spatial adjunct syntax.For example, the verb roots shown in ( 31) and (32), ta̰ ¹ 'discard' and ta̰ ¹ 'bury,' are homophonous, belong to the same subject and object inflection classes, and both require a spatial adjunct.Yet 'discard' assigns the allative case to its adjunct, while 'bury' assigns the locative case.
As in some other languages with inflection class, such as Yucatec (Bohnemeyer 2002), inflection class membership -including membership in the obligatory adjunct classes -does correlate with certain semantic properties of verb roots.For example, all intransitive verbs of posture, such as 'sit,' 'stand,' and 'kneel,' belong to one of the obligatory spatial adjunct classes (though they do not all assign the same case to their adjuncts).But this regularity does not extend to other stative verbs: for example, both ŋe²ʔma⁴ 'be in a place' and pe⁴³ 'live in a place' are stative verbs with space-related meanings, but pe⁴³ is an obligatory spatial adjunct verb and ŋe²ʔma⁴ is not.Thus, since semantic features underdetermine roots' inflection class membership, it is better analyzed as a morphological than a semantic feature of the root.

DPM is controlled by the lexical aspect of the verb
Beyond morphological class, other properties of the verb also affect DPM in Ticuna.Another key DPM-controlling property of the verb is lexical aspect.Much as in English (Vendler 1957), Ticuna verb stems can be divided into four lexical aspect classes: states, activities, accomplishments, and achievements.Lexical aspect influences inflection class assignment and has many other morphosyntactic effects, controlling the enclitics, adverbs, and adjuncts with which the verb can combine.
One of the most conspicuous effects of lexical aspect in Ticuna is on the syntax of spatial adjuncts.When a state, activity, or accomplishment verb combines with a ground-denoting spatial adjunct, that adjunct is marked with the allative case =wa⁵.This is shown for a stative verb in (33), an activity in (34), and an accomplishment in (35) (other language-internal tests establish the lexical aspect of these verbs). (33) [  (Anon., 2015-06.041, tca_20170629_ecp_ahs_elicit_002.wav, 38:24-38:40) In contrast, when an achievement verb -a verb denoting an instantaneous change of statecombines with a spatial adjunct, that adjunct is marked with the locative case =gu².This is shown for three different achievement verbs in ( 36)-( 38).It is clear that all three verb stems in ( 36)-( 38 (DGG, 2015-06.040, tca_20170626_dgg_ahs_elicit_001.wav, 33:48-34:22) The spatial adjuncts in all of (33)-( 38) have the same semantic role, acting as grounds.Similarly, the nouns which head the adjuncts all belong to the same semantic category of human-made landform nouns.Yet despite these semantic similarities, the spatial adjuncts are not all assigned the same case.( 33)-( 35) take the allative case, but (36)-(38) the locative case.
As in §4.2, the contrast in case marking between (33)-( 35) and ( 36)-( 38) is not a matter of nominal semantics.All of the nouns in these examples belong to the same semantic category of manmade landform nouns, and some of them are actually the same nouns (e.g., 33,34,36).
Given these similarities among the nouns, the contrast in case marking between (33)-( 35) and ( 36)-( 38) must arise from a property of the verb.
But what property?None of the verbs given in (33)-( 38) are verbs which display obligatory spatial adjuncts.Thus, the contrast between (33)-( 35) and ( 36)-( 38) cannot be due to the same property of the verb -a requirement for a spatial adjunct marked with a specific case -which motivated the DPM seen in §4.2.Instead, this contrast in place-marking is due to a different property of the verb: the lexical aspect contrast between achievement verbs, which cause place adjuncts to be marked with =gu² (36)(37)(38), and all other lexical aspect classes of verbs, which cause place adjuncts to be marked with =wa⁵ (33)(34)(35).
Intransitive verbs of motion, such as ã¹ 'paddle (a boat)' in (40), mark ground adjuncts with the locative case marker =gu².Using the allative case marker =wa⁵, which marks the ground of most non-motion events (e.g.33-35), is ungrammatical. (40) [na⁴ʔta⁴a²[✓gu²/*wa⁵]] ground tʃi³ʔã¹ na⁴=ʔta⁴a²=[✓gu²/*wa⁵] tʃi³=ã¹ dflt.poss=lake=[✓loc/*all]1SgSbj=paddle 'I paddled in the lake (ground = locative).'(DGG, 2015-06.040, tca_20170807_dgg_ahs_elicit_002.wav, 0:31-0:59) In contrast to intransitive verbs, transitive verbs of caused motion and placement mark ground adjuncts with the allative case marker, =wa⁵.Using =gu² to mark the ground of a transitive verb of caused motion or placement leads to either ungrammaticality or an interpretation of the adjunct as a goal rather than a ground.(41) illustrates this for ku³¹ʔ 'kick,' a transitive verb of caused motion.[locative interpreted as goal] (LWG, 2015-06.042, tca_20170623_lwg_ahs_elicit_002.wav, 35:10-36:25) (40) and (41) show that intransitive verbs of motion mark grounds with the locative case, while their transitive counterparts mark grounds with the allative case.Exactly the opposite contrast appears in marking of the goal of motion.Intransitive verbs of motion mark their goals only with the allative case =wa⁵.Using the locative case =gu² is ungrammatical (42).( 42 det(IV) bag(IV) 'I dropped off (lit.put) a bag at your house (goal = locative).'(DGG, 2015-06.040, tca_20170626_dgg_ahs_elicit_002.wav, 27:10-27:49) To summarize, intransitive vs. transitive verbs of motion display precisely opposite case-marking on spatial adjuncts.Intransitives mark ground adjuncts with the locative case (40) and goals with the allative (42).In contrast, transitives mark ground adjuncts with the allative case (41) and goals with the locative (43).This transitivity-controlled form of DPM is distinct from the other forms of verb-controlled DPM seen in the previous two sections.It differs from the DPM observed with obligatory spatial adjunct verbs ( §4.2) because not all verbs of motion are obligatory spatial adjunct verbs (some are, but none of those described in this section).Similarly, it differs from the DPM observed with achievements ( §4.3) because not all verbs of motion are achievements.Thus, transitivity-controlled inversion in place marking is not due to the other DPM phenomena in the language.Rather, it represents a separate, third form of verb-controlled DPM.
Since Ticuna is not morphologically ergative and lacks other evidence of syntactic ergativity, this finding is surprising theoretically.In areal perspective, however, it is less surprising.While this may be the first report of transitivity-controlled DPM in a language that is not ergative, it is not the first report of transitivity-controlled DPM in Amazonia.DPM of spatial adjuncts in transitive vs. intransitive clauses is attested in several Panoan languages, including Shipibo-Konibo (Valenzuela 2005), Kashibo-Kakataibo (Zariquiey 2018: 160), and Yaminahua (Neely 2019: 305-307).However, in Panoan languages the differential place marking (1) is controlled by the orientation of the adjunct to an argument, rather than by the transitivity of the clause as such, and (2) involves addition of extra (agreement) marking to the case-marked adjunct, rather than alternations in the case marking appearing on the adjunct.

Interim summary
This section has shown that Ticuna displays three distinct forms of DPM which are controlled by verbs rather than nouns.DPM can be controlled by the morphological class of the verb ( §4.2), by the lexical aspect of the verb ( §4.3), and for verbs of motion and placement, by the transitivity of the verb ( §4.4). Figure 2 provides a diagram representing these multiple forms of verb-controlled DPM.
Role association theories, as reviewed in §1.1, attribute DPM exclusively to the (semantic) properties of ground-denoting nouns.In contrast, this section has shown that -while much of DPM across languages is controlled by noun properties -DPM may also be controlled by the verb.Moreover, the Ticuna system shows not only that verbs as a class can control DPM, but also that many different features of the verb or verb phrase can control the phenomenon.The language displays DPM controlled by lexical aspect, a semantic feature; by inflection class, a morphological feature; and by the transitivity of verbs of motion, a property which involves crossed morphosyntactic and semantic features.Furthermore, contrasting with the predictions of role association theories, these forms of DPM are completely insensitive to properties of the noun.
The same noun, in the same role, displays different marking with different verbs, indicating that these phenomena are driven exclusively by verb features.

Discussion
We now consider the impacts of our findings on the two research questions proposed in §1 about what motivates DPM across languages, and about how it is realized.

Motivation of DPM
Recall from §1.1 that previous analyses, using both categorization-and alternation-based frameworks, view DPM as motivated by noun semantics and mediated by role association.Under these analyses, properties of the nominal referent determine each noun's degree of association with the semantic role of ground.Differential association with the ground role then leads to differential marking of that role.Many of our findings for Dâw support this analysis.As we showed in §3, DPM in Dâw is motivated largely by properties of the nominal referent.However, the relationship between Dâw nouns' semantic properties and their DPM behavior is much more complex than predicted by either type of previous analysis.In this language, the referent properties that control nouns' DPM behavior are not coarse-grained features such as animacy, but fine-grained ontological properties such as relative size and perceptual boundedness.As a result of these finer divisions, DPM in Dâw distinguishes at least six different classes of nouns, in contrast to the two to four classes posited by earlier analyses (Cablitz 2008;Haspelmath 2019: 323).
At the same time, other findings of this paper complicate the analysis of the motivations of DPM outlined above.Though noun semantics do control DPM in Dâw, in Ticuna most DPM is controlled by properties of the verb, such as transitivity and lexical aspect.Per §4, this verbcontrolled DPM is completely insensitive to properties of the ground-denoting noun: with different verbs, the same noun displays different marking.In contrast, prior analyses in both categorization-and alternation-focused frameworks treat DPM as controlled by the grounddenoting noun, without considering effects of the verb.
While these findings contrast with traditional analyses of DPM, they are consistent with research on differential argument marking.Just as languages vary in the factors which control argument marking or define transitivity (Hopper & Thompson 1980), they also differ widely in the conditioning of place marking: DPM is controlled primarily by verbs in Ticuna, but primarily by the perceptual and ontological properties of noun referents in Dâw.This position leaves a place for role association -which may still create the link between nouns' semantics and their DPM behavior -but does not treat it as the only possible motivation for the phenomenon.Rather, exactly as split ergativity can be conditioned either by lexical properties of the agent, such as animacy, or by properties of the verb, such as aspect, DPM can be controlled by either noun or verb properties.

Form of DPM
Role association-based theories of DPM, whether in alternation-or categorization-based frameworks, emphasize the length of spatial adjunct marking.They propose that association with a given semantic role is inversely related to the length (or complexity) of marking for that role.Thus, the nouns most associated with the ground role display the shortest or least complex marking for that role.This analysis clearly explains DPM alternations between longer and shorter (or zero) markers.
On the other hand, it cannot explain DPM systems that involve alternations between markers of equal length and complexity.Yet in both Dâw and Ticuna, most DPM belongs to this type.DPM in Dâw primarily involves alternations in the combinatorial possibilities available to nouns, rather than alternations in markers per se.Where there are alternations in markers, as between configurational postpositions and rid, the alternating markers are all equal in length ( §3.1).
Similarly, DPM in Ticuna consists entirely of alternations between two case markers of equal length ( §4).Thus, DPM in these languages does not concern the syntagmatic complexity of case markingthe number of morphemes or segments involved in a noun's marking for the ground role.Instead, DPM in Dâw concerns the paradigmatic complexity of marking, instantiated by the number of possible markers with which a noun can combine.And in a further contrast with traditional analyses, DPM in Ticuna does not involve contrasts in the complexity of marking at all, but rather equipollent contrasts between markers of equal complexity.These findings show that in terms of form, DPM cannot be understood as simply less explicit or shorter marking of nouns more frequently associated with the ground role.Alternations between shorter and longer marking are part of the phenomenon, but so are equipollent and paradigmatic alternations.
This diversity of realizations represents another parallel between DPM and the differential case marking of arguments.In differential argument marking, as in DPM, asymmetrical contrasts between zero and overt marking are typologically most common (Bossong 1991).But differential argument marking also includes symmetrical alternations, for example between overt accusative and partitive cases in the differential object marking system of Finnish (Kiparsky 1998).
Likewise, differential argument marking may be realized via paradigmatic alternations as well.
For example, the Pama-Nyungan language Umpithamu (Verstraete 2010) displays a split ergative system where pronominal agents are always nominative, inanimate NP agents are always ergative, and animate NP agents alternate between nominative and ergative marking depending on information structure.This kind of multiply split ergativity can be seen as analogous to the DPM system of Dâw: in both systems, nouns contrast in the combinatorial possibilities of marking -the number of different cases (for split ergativity) or postpositions (for DPM) which they allow -rather than in the form or acceptability of a single overt marker.In other words, DPM mirrors differential argument marking in form as well as motivation, and the form of both phenomena varies substantially across languages.

Conclusion
This study has examined the DPM systems of two unrelated languages spoken in northwestern Amazonia, Dâw and Ticuna.Both languages' DPM systems contrast with the claims about DPM made in previous typological literature.While prior work claims that DPM is conditioned exclusively by noun semantics, we show that it can also be conditioned by syntactic and semantic properties of verbs.Similarly, while prior work views DPM through the lens of alternations between more and less syntagmatically complex place markers, we show that DPM can also be realized by alternations in paradigmatic complexity, as well as by equipollent alternations between markers of equal complexity.Together, these findings show -contra Haspelmath (2019), Landau &Jackendoff (1993), andStolz et al. (2014) -that DPM is not exclusively conditioned by the role associations of the ground-denoting noun, or exclusively realized by alternations between markers of differential complexity.Rather, like differential argument marking (Bossong 1991;Aissen 2003), DPM can be conditioned by properties of either the noun or the clause, and can be realized with either symmetrical or asymmetrical alternations.
We reach these novel conclusions because we choose to avoid some common assumptions about the scope and form of DPM.For example, while other analyses of DPM do not include markers encoding configurational relations (Haspelmath 2019: 316), our analysis of Dâw includes configurational postpositions.This decision allows us to describe an additional form of DPM in the language -the contrast in paradigmatic complexity between nouns which combine only with configurational postpositions, and nouns which can combine with both configurational postpositions and other place markers.Similarly, most studies of DPM consider only clauses headed by intransitive verbs of motion and location.Our analysis of Ticuna, on the other hand, also includes transitive verbs, as well as verbs outside the semantic domains of space and motion.
Because of this decision, we can describe the system of verb-controlled DPM -which would not be evident from intransitive motion verbs alone, as in Ticuna these verbs all display the same DPM behavior.
Looking to the future, we encourage other researchers studying the grammar of space to treat DPM as analogous to differential argument marking, with the same possible diversity of motivation and realization.Because so much is known about differential argument marking, this analysis makes many testable predictions.For example, since information-structural features of the noun phrase often condition differential argument marking, we predict that information structure can condition DPM as well, and since pronouns pattern apart from noun phrases in many differential argument marking systems, we predict that they may also pattern apart in DPM.To test these predictions, much more research on DPM -in individual languages, and in typological perspective -is necessary.
Ticuna (ISO: tca) is a language isolate spoken by 38,690 to 70,000 people (Lewis et al. 2014; ISA 2021).Most speakers live along the western course of the Amazon/Solimões River in western Brazil, southern Colombia, and northeastern Peru.Ticuna examples in this paper are from the first author's fieldwork in Cushillococha, Peru.Positive examples come primarily from landscape description interviews (Kita 2001).These are non-metalinguistic interviews where speakers describe the location of landmarks in their community and how the landmarks have changed over time.Negative examples and minimal pairs come from elicitation.Landscape description interviews were conducted monolingually in Ticuna, while elicitation was conducted bilingually in Ticuna and Spanish.Eight people, four women and four men, contributed most of the Ticuna examples.Angel Bitancourt Serra, Yaneth Candido Guerrero, Deoclesio Guerrero Gomez, and Sotil Suarez Gonzalez contributed both landscape description interviews and elicited examples.Lilia Witancort Guerrero and an anonymous consultant contributed elicited examples only; Lucinda Gomez Cordero and Hortensia Coello Guerrero contributed landscape description interviews only.All of these consultants agreed to the disclosure of their names, except that one chose to remain anonymous.
no information about the spatial relation between the figure and ground.The alternations between the configurational postpositions and the generic locative rid therefore represent a form of DPM.Exceptions to the pattern of marking with postpositions in (1)-(2) do exist.Three Dâw landscape terms can occur zero-marked as spatial adjuncts: xaay 'forest' (in (3), tuu 'ground,' and pox 'sky.'This indicates a first division among place nouns -nouns which require postpositions, such as the toponym baal 'Manaus' 'in (2), contrast with nouns which allow zero-marking, such as xaay 'forest' in (3).

Figures
Figures can participate in a variety of spatial relations with specific parts of this ground, such as containment (4); attachment (5); and non-contiguity (6).Each relation is marked with a different configurational postposition.

Figure 1 :
Figure 1: Combinatorial possibilities for place marking of each category of nouns in Dâw.
and goal (11) or source (12) in motion events.As this range of readings indicates, rid lacks any configurational meaning.All place nouns -whether landscape terms, landform terms, toponyms or spatial adverbs -are compatible with rid.In contrast, object and human nouns are not compatible with the marker for any spatial role.(13)illustrates this for the role of location; goal and source behave the

Table 1 :
Ticuna verb inflection classes defined by spatial adjunct syntax.If a verb takes the additional proclitic i⁵ = in its citation form, then when it appears with a spatial adjunct, it assigns the allative case marker =wa⁵ to that adjunct.The i⁵ = proclitic ) are achievements because they have other morphosyntactic properties that are unique to achievements in Ticuna, such as being incompatible with phasal verbs like ɨ²gɨ⁴