Lexically driven patterns of contact in alignment systems of languages of the northern Upper Amazon

2.1 The linguistic ecology of the NWA, with special attention to the foothill area

The NWA is characterized by a high linguistic (genealogical) diversity, comprising a number of small language families and (near) isolates, as well as (branches) of larger language families. Map 2 ^[2] indicates the genealogical units of the area, with each isolate counted as a separate genealogical unit.

Map 2

The languages of the NWA and their linguistic affiliations.

Three larger families are represented in the area: Tukanoan, Quechuan, and Arawakan. We zoom in on the former two languages in the next section and briefly introduce Arawakan here. Arawakan is one of the major language families of the continent, with an enormous geographical spread, spanning almost the entire northern part of the continent, with languages spoken between the Caribbean islands and southern meso-America to northern Argentina and southern Brazil, and from the Atlantic coast to the foothills of the Andes. The branch that is represented in the NWA is the Japura-Colombia branch in the classification of Hammarström et al. (2021), based on Aikhenvald (1999), with some 15 known languages. Where contact between Tukanoan and Quechuan has been understudied in the literature, there is extensive literature on contact between Arawakan and Tukanoan (e.g., Aikhenvald 2002, Stenzel 2015, Chacon 2017).

In addition to these larger families, a number of smaller families and isolates are represented in the NWA. The Chicham or Jivaroan family consists of five languages: Aguaruna, Wampis, Shuar, Achuar, and Shiwiar spoken in both Peru and Ecuador. This family has a relatively large number of speakers, with Shuar having over 60,000 and Aguaruna over 55,000 speakers (Kohlberger 2020, 15). The Zaparoan family (de Carvalho 2013) is a highly endangered language family with six known members located in Ecuador and northeastern Peru, of which three are extinct, and the remaining three nearly extinct (Hammarström et al. 2021). The Kawapanan (Cahuapanan) family of northeastern Peru (Valenzuela 2011) has two known languages, both of which are still spoken today, but also endangered to varying degrees. In addition, the area harbors a number of isolate languages including Camsá of southern Colombia, A’ingae (Cofán), spoken in Colombia and Ecuador, Urarina of northern Peru, and Wao Tededo (Waorani), an Ecuadorian language. This shows a linguistically diverse landscape with various potential contact scenarios in the NWA-foothill region. Naduhup, Witotoan, Barbacoan, Peba-Yaguan, and Ticuna-Yuri are small families at the fringes of the NWA. Tupian and Cariban are large families with a minor presence at the periphery of the NWA.

There are various subareas of intense contact within the NWA area that are mentioned in the literature. These include the Vaupés area (Aikhenvald 2002), the larger Upper Río Negro area, which includes the Vaupés area (Epps and Stenzel 2013), and the People of the Centre area (Gasché 2009). The best studied of these subregions is the Vaupés, which houses diverse communities living along the river system straddling the central Colombia–Brazil border area. There are roughly a dozen languages represented in this subregion, classified as belonging to five separate families: Tukanoan (namely, Eastern Tukanoan varieties), Arawakan, Nadahup, Kakua-Nukak, and a growing Tupí-Guaranían presence with the growth of Nheengatú populations (Epps and Michael 2017). According to most studies (Aikhenvald 1999, Epps and Stenzel 2013, among others), the Vaupés is classified as a zone of intense contact and cultural-linguistic diffusion across and within these families. This relationship of cultural and linguistic exchange in the Vaupés has likely been active for centuries, predating the arrival of the Spanish in the region, as sociopolitical shifts caused certain nonriverine communities to migrate to such fluvial zones (Chacon 2013).

Another NWA subregion which has received attention is subsumed under the label the People of the Centre (Gasché 2009). The groups that populate this region along the southern Colombia-northern Peru border speak languages belonging to the Witotoan family, the Boran family, and various southern Arawakan varieties. Although there are some works on the specific contact situation of this area (Seifart 2011, 2012, 2015), this area is more consistently described as a sociocultural contact area (Echeverri 1997, Gasché 2009).

The focus of this article is on the NWA-foothill area, which has been mentioned as a possible contact area (e.g., Wise 2011, Valenzuela 2015, Kohlberger 2020). The languages that have probably been in contact since before the colonial documentation in this area include the Western Tukanoan, Chicham, Zaparoan, and Kawapanan languages and the isolates A’ingae (Cofán) and Wao Tededo. Amazonian Kichwa probably developed in the colonial period (Muysken 2000). Although there are only a few lexical borrowings in the area (see, for instance, Velie 2008, 29 for Quechua borrowings in Máíhɨ͂̀kì), there are some structurally reoccurring similarities observed, such as the habitual construction as exemplified in (1) and the deontic construction as exemplified in (2):

These constructions show possible convergence among Tukanoan, Quechuan, and Chicham languages of the NWA.^[3] Morphosyntactic influence on Ecuadorian Quechua from Wao Tededo, Barbacoan and Chicham languages has also been suggested (see, e.g., Muysken 2011, Zúñiga 2015, Floyd 2022).

If we take a bird’s eye view of the alignment patterns found in the NWA-foothill area, we see that the area is dominated by nominative–accusative alignment systems (with the exception of Kawapanan languages),^[4] expressed in most languages by a combination of flagging and indexing.^[5] This resembles the dominant Andean pattern more than the surrounding Amazonian patterns, where ergativity and/or split-intransitivity are commonly found.^[6] In addition, differential argument marking seems to be a feature of the area. A number of languages (Western-Tukanoan, Witotoan, Kawapanan, and Zaparoan) have optional subject case markers, and Chicham and Witotoan languages have conditional object case marking depending on the person of the subject (Wise 1999, Overall 2007, Kohlberger 2020); object case marking in Tukanoan languages interacts with various factors: specificity (Kotiria – Stenzel 2013) and definiteness (Tukano–Bonfim Duarte and de Oliveira Lopes 2020). It is also possible that the lack or reduction (compared to southern varieties) of object indexing on the verb in northern Quechuan languages is a contact (substrate) effect. These elements, summarized in Table 1, suggest contact-induced diffusion of elements of alignment patterns in the languages of this area. In Section 2.2, we zoom in on the Tukanoan and Quechuan languages.

2.2 Tukanoan and Quechuan

One of the larger families that is confined to the NWA is the Tukanoan family (Tucanoan in the map),^[7] consisting of about 29 languages of which about 8 are not spoken anymore (Chacon 2014, 275). There is a major split between Eastern and Western Tukanoan languages, Eastern Tukanoan being the larger branch and Western Tukanoan being the branch that falls within the NWA-foothill area, the area of focus in this study. Eastern Tukanoan is concentrated in the eastern portion of the NWA, roughly between the Apoporis and Vaupés Rivers. The Western Tukanoan branch currently consists of Koreguaje, Siona, Sekoya, and Máíhɨ͂̀kì and shows its main concentration in the foothill area between the Upper Caquetá and Upper Putumayo Rivers, and with geographical outlier Máíhɨ͂̀kì in northeastern Peru. The branching within the family is shown in Figure 1.

Figure 1

A simplified version of the Tukanoan family classification by Chacon (2014, 282).

In the previous classifications (Mason 1950, Waltz and Wheeler 1972, Barnes 1999, 209), Kubeo and, in some cases, Retaurã and Tanimuka were analyzed as a separate branch, because they differ more from other Eastern Tukanoan languages. Mason (1950) analyzes Kubeo as forming a middle or central branch, because it has communalities with both Eastern and Western Tukanoan languages. However, as Gomez-Imbert (2011) points out, the “Middle” languages Kubeo and Retaurã (and Tanimuka) may differ from the other Eastern Tukanoan languages, because they have undergone more contact-induced changes due to their contact with Arawakan languages. Therefore, the deviation is not due to an earlier language split, but due to a specific contact situation. Furthermore, as Chacon (2014, 283) points out, previous classifications based their analysis on both retentions and innovations, whereas only shared innovations should be taken into account. When this is done, the central branch languages should be grouped into the Eastern Tukanoan branch.

Although there is no clear-cut outcome to the study of the ^[8]geographical origin of the Tukanoan people, it seems that they are from the NWA area. Aikhenvald (2012, 225) suggests that the Tukanoan people came originally from the Andes based on the nominative–accusative alignment system found in the Tukanoan languages. Chacon (2013, 2014) suggests the homeland of the Tukanoan family to be situated in the center of NWA from where they have spread westward and northeastward. This analysis is based on linking archeological patterns with common Tukanoan cultural patterns and on linguist reconstruction of vocabulary for basic material cultural elements.

Another language family that is found in the NWA-foothill area is the Quechuan family. The Quechuan family is originally an Andean language family that is predominantly spoken along the mountain range, from northern Chile and Argentina in the south to southern Colombia in the north. The more distant varieties might be mutually unintelligible, and the overall number of Quechuan dialects is still only approximately known (Adelaar and Muysken 2004, 168). The most widely applied classification of Quechuan languages was put forward by Torero (1964). It divides the family into Quechua I (Torero 1964), spoken in the highlands of central Peru, considered to be ‘the homeland of Proto-Quechua’ (cf. Adelaar and Muysken 2004, 180–1; Mannheim 1991, 9–12) and Quechua II (Torero 1964), including all the remaining varieties (Figure 2).

Figure 2

An adapted version of the Quechuan family classification by Adelaar and Muysken (2004, 184).

Since Quechuan languages are spoken predominantly in the Andean highlands, Quechua is generally regarded as connected to Andean identity (Adelaar and Muysken 2004, 180–3). However, in Peru, Ecuador, and southern Colombia, Quechuan languages are also spoken in the Amazonian foothills of the Andes, and even well into the lowlands. Uzendoski and Whitten Jr. (2014, 1) estimate that there are about 150,000 speakers of Amazonian Quechuan varieties in Ecuador. Although there is no consensus about how Quechua spread into the foothill and lowland areas, Quechua seems to have been used in this area at least since the colonial era (Muysken 2000), and since this time, Quechuan varieties have become the identity language of various peoples.

Tukanoan and Quechuan languages are not only found in the same area but also some of the languages from these families are in close contact for at least half a century. For instance, the Siona and Sekoya and Amazonian Kichwa people live in neighboring villages on the Aguarico River, and there are various intermarriages. These close contacts started at least in the middle of the last century (Ligia Criollo and Jaime Tangoy Pers. Comm.). There is also contact between the Quechuan Inga and Colombian Siona communities, which are on the same river. There is also some evidence for even older contact. Languages from both families were used as general languages of conversion during the ^[9]Spanish colony. Siona was the general language used by the Spanish in the Putumayo, Napo, and Caquetá areas (Triana y Antorveza 1987, 171–2). A variety of Quechua was more generally used as a language of conversion and was used in this capacity in the area where Western Tukanoan languages were spoken as well. Some evidence for this is the complaint by a Siona leader from the middle of the eighteenth century that the Siona people were taught Quechua to be sold into slavery, as reported in Chantre y Herrera (1901, 392), cited in Vickers (1976, 43). Additional evidence can be found in a colonial document, which includes grammatical information on a Western Tukanoan (WT) variety, a vocabulary WT-Spanish, and a doctrina cristiana in the WT variety and in Ynga (Quechuan), suggesting that both languages were used together in the same area (Cipolletti 1992, Floyd et al. 2021). So it seems that Western Tukanoan languages and Quechuan varieties have been in contact at least since Quechuan entered the NWA-foothill area, which probably happened before the Spanish conquest (Obrem 1980, cited in Muratorio 1998, 72–5).

Tukanoan and Quechuan show similar alignment patterns. The languages from both families show a nominative–accusative alignment system. As mentioned earlier, this is cross-linguistically the most common canonical marking pattern found around the world, but these systems are scarcer in South America and are found in the Andes and adjacent areas (Aikhenvald 2012, 225). Tukanoan languages show this alignment system in two ways: (1) both S/A are obligatorily cross-referenced on the verb and (2) objects can be marked with the object case marker -re. This object marker is not used on all objects, because these languages show differential or optional object marking, speakers can opt for the case marker -re or can leave the object unmarked. This same marker is used to flag indirect objects across the board in the Tukanoan family. All Western Tukanoan languages, with the exception of Máíhɨ͂̀kì, exhibit an additional object marking -ni, in alternation with -re, which is mainly restricted to animate objects.

Quechuan languages exhibit very similar behavior with respect to marking alignment. Both S and A show obligatory agreement on the verb, and the language has an object, or accusative case marker -ta. Quechuan languages also show differential object marking on some levels. In the case of some predicates, it is possible to use either the accusative marker -ta or the dative/allative marker -man. An example of these predicates is the motion predicates, which can both have an accusative marked goal or a dative marked goal. Quechuan languages, in addition, show some type of object indexing on the verb. This marking is in most languages restricted to speech act participants; however, in Ecuadorian Kichwa and closely related varieties, this marking is reduced to first person objects.

3.1 Framework

To consistently compare verbal lexicon across languages, we rely on the framework provided by the Valency Patterns Leipzig (ValPaL) project (Hartmann et al. 2013, Malchukov 2015), in combination with a cross-linguistically applicable semantic map of verb classes (Malchukov 2005). The ValPaL project identified 70 verb meanings that were taken to be representative of the verbal lexicon across languages (Malchukov 2015, 29). For each predicate, a set of the so-called microroles and macroroles is associated with each participant that the predicate introduces. Consider the following example of a typical transitive predicate concept that introduces two core participants^[10]:

EAT: A [microrole:eater] eats P [microrole:eaten food]

In this case, the macroroles are A (referring to the agent-like participant) and P (the patient-like participant). As can be seen in the aforementioned example, microroles are predicate specific in recognition of the potential for language-particular encoding patterns at the level of the individual predicate, rather than simply relying on traditional labels for the roles for each participant. However, by using the original terminology from the ValPaL project, traditional role-related labels are elaborated into the set of macroroles, which group microroles together into broader classes and allow for higher level analyses. By separating these two levels, the ValPaL approach presents an attractive means of arranging our data to search for lexical-level patterns as well as drawing broader conclusions regarding the alignment patterns in the area under discussion.

In the case of our project, we extend the ValPaL paradigm by categorizing each predicate entry a priori on the basis of its semantic type, following Malchukov (2005).

Malchukov proposes a hierarchy of verb types specifically for the purposes of cross-linguistic comparison, which is designed to predict the likelihood that a verb has a case frame that deviates from the canonical transitive case frame. At the same time, it aims to group verb types that are likely to show similar case frames, based on the semantic comparability of verb arguments. Malchukov’s proposal in fact contains three hierarchies, which all highlight different semantic deviations from the prototypical transitive situation type. On the basis of Hopper and Thompson’s (1980) characterization of transitive events, a prototypical transitive event has:^[11]

1. Prototypical agent: visible, salient, volitional, controlling agent cause, which initiates the event.

2. Prototypical patient: visible, salient, nonvolitional, noncontrolling patient effect, which registers the bulk of change associated with the event.

Based on these prototypical A and P arguments, Malchukov initially proposes two dimensions or subhierarchies that represent different types of deviations from these prototypes, as shown in Figure 3.

Figure 3

Two-dimensional verb hierarchy (adapted from Malchukov 2005, 81).

At the left end of Figure 3, effective actions (e.g., break, cut) involve a prototypical (controlling, volitional) agent acting on a prototypical (noncontrolling, affected) patient. The top line of the figure represents degrees of deviations away from the prototypical patient. Contact verbs (e.g., hit, pinch) involve a patient that is attained, but not necessarily affected in the sense that it undergoes a change of state. The next step on the top line is pursuit verbs (e.g., look for, aim at) which involve an even less prototypical patient, in that it is not (necessarily) attained. The nonagentive argument is more like a Goal, semantically. Motion verbs, finally, are typically one-place predicates, optionally taking an oblique second argument.

The bottom line depicts deviations away from the prototypical agent. This hierarchy is more complex than the one involving decreased patienthood, since a number of cases not only involve a nonprototypical agent but also a nonprototypical patient. A case in point is a verb like to see, which involves not only an Experiencer agent but also a nonaffected patient. Nevertheless, the main parameter structuring the lower hierarchy is decreased agenthood, which we focus on in the remainder of this paragraph. Perception and cognition verbs can involve a volitional, controlling agent (e.g., in verbs denoting conscious acts of perception or cognition, like to look (for), to listen (for), to actively remember or think), but often the A-like participant is semantically an Experiencer, as in to see, hear, forget, etc. The group of emotion predicates represent a step further away from the prototypical agent, in that it is not only a noncontrolling and nonvolitional agent but also arguably an (emotionally) affected one. This is also true (perhaps more so) for sensation predicates, which tend to be monovalent, but which, unlike motion verbs, are more likely to have a noncanonically marked Experiencer.

Malchukov (2005) makes further suggestions to expand the map, so that it can incorporate a greater part of the verbal lexicon. The full map is depicted in Figure 4.

Figure 4

Three-dimensional verb hierarchy (Malchukov 2005, 113).

First, an important addition is a third line, going from effective actions to reflexives, middles, and spontaneous events. The parameter governing this particular route is the relative distinguishability of the Agent and Patient participant, following the work of Kemmer (1993) on middle semantics. Kemmer’s (1993) semantic map is more fine-grained, but for the purposes of cross-linguistic comparison, Malchukov recognized three classes along this dimension: syntactic reflexives (I hit myself), semantic middles, and inherent reflexives and body actions including verbs of grooming, nontranslational motion, body posture, as well as symmetric predicates like ‘meet’. Spontaneous events are those events that occur seemingly without any identifiable agentive force (e.g., burst, explode). Two further types were added to the basic, two-dimensional map that was based on Tsunoda’s (1981) work: verbs involving affected agents and interaction verbs. The former are prototypical transitives in the sense that there is a volitional, controlling agent, and an affected patient, but the agent is also affected by performing the action. This group contains verbs of consumption (eat, drink) as well as other verbs like ‘put on’ or ‘take’ (Malchukov 2005, 111). Finally, Malchukov proposes a separate class of interaction verbs, broadly conceived, including verbs of communication (speak, converse, shout at) and social interactions of different kinds (marry, follow, agree). The lines between the different verb classes represent affinities between them. For instance, middle verbs, apart from connections on the transitivity scale to reflexives on the one hand and spontaneous events on the other, are connected to perception, cognition, and emotion predicates, which are cross-linguistically also often part of a group of middle verbs where the agent–patient distinction is collapsed (cf. Kemmer 1993),^[12] and they connect to verbs of interaction, which are, or can often be, symmetrical.

3.2 Database structure, language sample, and methodology

For the purpose of tracing the distribution of (non-)canonical case frames on the basis of lexical items, the organization of our database was inspired in large part by the ValPaL framework as outlined earlier. We adapted the original ValPaL model in two main ways to search for lexical patterns and to implement these data in testing whether Malchukov’s observations regarding semantic verb types and concomitant patterns of case frame selection hold for the languages of our sample. First, we expanded the total number of predicates from roughly 70 to 112 to have a representative number of predicates for each semantic class as per the extended semantic map presented in Malchukov 2005, replicated in Figure 4. The next and most crucial update that we made to the original model was to attribute each predicate to one of Malchukov’s types, independent of our language-particular analysis: for instance, EAT was classified as being of the “affected agent” type on the basis of the semantics of the verbal concept. A list of verbal concepts and their classification into the semantic classes, inspired by Malchukov 2005, is given in the appendix of this article. Up until this point, the top–down nature of this database is retained, and the modified ValPaL model could in theory be applied to any set of languages without further manipulation. With this groundwork in place, the database skeleton was ready, and we could begin to populate the charts with language-particular data sets.

When the database is populated with language-particular data sets, alignment patterns naturally emerge. Where each role is flagged in the expected way for a given predicate, the predicate is deemed to have a canonical case frame. Deviations from the canonical case frame can arise where any of its core participants does not receive the flag (or other relevant morphosyntactic operations) that it does in the context of a canonical frame. As has been explained earlier, the fact that we can pinpoint on which participant the deviant pattern arises is indispensable to connect shared patterns of noncanonical flagging or indexing to the parameter of the prototypical transitive properties associated with the participant given the semantics of the concept.

Data sets for each relevant language within our sample were developed on the basis of grammars and dictionaries and/or elicitation sessions with language consultants. As mentioned earlier, our sample focuses on Quechuan and Tukanoan languages of the NWA-foothill area. In addition, we sampled Quechuan and Tukanoan languages that fall outside this area as control languages, which help to distinguish genealogical and areal patterns. The selection of languages for our project was constrained in part by the availability of solid grammatical materials as well as the specialties of the researchers involved. Where possible, we aimed to supplement and reinforce the information available in the literature via elicitation with language consultants. These elicitation sessions consisted of translation tasks from Spanish to the language of interest. The sentences used in these translation tasks were preconceived using the set of stimulus sentences provided by ValPaL, extended with similar sentences for verbs not found in the original ValPaL list. We needed to add various concepts to the original list, such as be hungry and be sleepy because these were hypothesized to be relevant to alignment patterns in the region. In total, our study considered alignment patterns from 15 languages.

Table 2 enumerates the languages in our sample, classified as either falling within our interest area or not (i.e., the latter acting as a control in our study), along with the grammatical materials and dictionaries consulted to build up our datasets where applicable. Map 3 plots the geographical distribution of these languages.

Map 3

The geographical distribution of our language sample.

For each language in our sample, the first step was to find the relevant lexical item(s) for each of the predicates in our database. These forms were entered into a centralized mastersheet so that cognates and candidates for lexical borrowing could be identified away from the noise associated with language-particular flagging/indexing analyses. Once the lexical items had been found and entered into the corresponding cell in the mastersheet, the next step was to analyze the predicates for each language on an individual basis and in a separate analysis workspace. On the basis of pertinent examples or relevant descriptions in the pre-existing literature, and/or elicited examples, we identified the dependent marking pattern (i.e., the case frame) and the verbal indexing pattern associated with each core participant for every predicate within a given language’s analysis workspace. In certain instances where relevant examples could not be procured, alignment patterns were inferred as a last resort to level our analysis across the entire sample.

Once all of the data had been collected following the strategy outlined earlier, the goal of our analysis was to trace the distribution of noncanonical flagging patterns across the predicates and verb types in our data sets. In each language-particular workspace, we grouped predicates on the basis of their verb type and identified the flagging pattern used on A-like participants as well as P-like participants.^[13] In turn, we determined the percentage of predicates per type in our data set, which assign a noncanonical flag to the A-like participant and/or the P-like participant. In tandem with this type-level analysis, we also retained a list of predicates which exhibit deviant alignment behavior to determine whether these represent candidates for lexical borrowing.

4.1 Marking of P- and R-like participants across semantic classes

Figure 5 exhibits the percentage of accusative flagging found on the P-like argument against the semantic classes in Malchukov 2005 for our Tukanoan sample, separated by the languages within and outside of the NWA-foothill area (i.e., labeled “interest” and “control,” respectively). Certain concepts also invoke an R participant (i.e., Recipient or other indirect argument participant), and these are considered where relevant to our discussion. Interestingly, across both families figuring in our core discussion, these participants tend to pattern together as regards their flagging patterns. We discuss certain deviations from this generalization in what follows. In the following figure, and the others like it to follow, we have replicated the general configuration Malchukov’s tables, where the degree of shading indicates the extent to which we find deviancy from the expected flagging pattern. The percentage figures are presented separately for those languages within the putative contact area, namely, “interest” languages from the NWA-foothill region, and for those located outside of this area, namely, the “control” languages for our study. Table 2 presents a list of “interest” and “control” languages for both Tukanoan and Quechuan in our sample.

As can be seen, accusative marking is the clear majority pattern in all classes except those at the fringes of Malchukov’s map (motion, spontaneous, and sensation concepts), which are overwhelmingly intransitive cross-linguistically. In the case of sensation concepts, there is an exceptional concept, namely, hurt, which we analyze as taking an Experiencer A and a Stimulus P underlyingly. These concepts are considered in detail in Section 4.2, where we discuss the flagging patterns associated with Experiencer A arguments. The differences between the languages under investigation and the control languages are generally small. In our discussion of the differences between foothill Tukanoan languages and the control languages, we focus on the top line of the semantic map. Nonetheless, we begin by noting a few marginal instances of nonaccusative case marking on the P/R participant with the interaction predicates from various Tukanoan control languages. The examples in (3) demonstrate the flagging patterns found with talk in Desano and Kotiria, where the person spoken to is encoded with the comitative marker (i.e., roughly “with” in English).

Besides these marginal instances in the control languages, the P and/or R argument occurs with accusative flagging as predicted with all other verb classes where they are associated with prototypical transitive participants – for instance, with effective action, affected agent, perception, and cognition predicates. This pattern is pervasive across both the interest and the Tukanoan control languages in our data. Before discussing the variation that takes place at the top line of the semantic map, we compare the distribution of accusative flagging with P/R participants in Quechuan in Figure 6. The picture in Quechuan is largely similar, with slightly more instances of nonaccusative marking.

As is the case with Tukanoan, the majority of transitive predicates belonging to classes that are not along the top line in Malchukov’s semantic map select for a nominative–accusative case frame. So, the languages of both families generally follow the patterns predicted by the Malchukov model. Despite this general tendency to select a canonical nominative–accusative case frame on transitive predicates, we note a few discrepancies in the case of Quechuan. First, on par with Desano and Kotiria, we observe an instance of the interactive concept talk choosing a slightly different frame where the individual spoken to is encoded with comitative marker in Cuzco Quechua.

In the case of Quechuan, it is also relevant to note that, although the accusative marker -ta can extend to indirect objects in most cases (particularly where the referent is animate), just like the pattern which is noted for Tukanoan earlier, it is also possible in some varieties to mark the R participant with the dative/allative marker -man (or its cognate). This fact marks a stark contrast with Tukanoan where only the accusative marker -re can be used with the R participant, even in languages with a dedicated “goal” marker such as Siona and Koreguaje.

Another marginal instance of a nominative–oblique case frame arises in Upper Napo Kichwa. In fact, just as in the interaction predicates discussed earlier, there appear to be two case frames that can be selected for the concept think (about) – one where the entity or concept which is thought about is encoded typically with the accusative marker -ta, and another where it receives the ablative marker -manta. The sentences in (5) exemplify these two case frames.

Besides these instances of deviation from the canonical nominative–accusative case frame in Quechuan, just as is the case with Tukanoan, the majority of concepts found in the lower portions of Malchukov’s model select for the canonical frame as predicted by their prototypical transitive semantics.

We now shift our discussion to the semantic classes, which are less prototypically transitive and thus more naturally select for noncanonical case frames, namely, contact, pursuit, and motion predicates. It is not always clear whether a given concept is better classified as a contact or a pursuit predicate based on Malchukov 2005 (and others discussed in Section 3.1), so we discuss particular lexical entries in turn before returning to a general discussion of the patterns. Turning first to the Tukanoan languages, both foothill and control languages have verbs that were classified as contact verbs that take a nominative–oblique pattern; note that we have tracked the behavior of R participants in ditransitive verbs in Figure 6. These concern the same verb types across all languages in our sample: lean on, load, throw at, and stick to. These verbs do not necessarily pattern in the same way in each language within the sample, e.g., lean on takes the instrumental case in Kubeo, exemplified in (6a), but the locative case in Desano as in (6b). Stick to takes its goal-like argument in the locative case in Kotiria as in (6c), whereas it takes the goal marker in Siona as in (6d). Note that the case suffixes in (6a-b) are quite likely cognates. Although it might be because of this cognacy that these two languages feature different case patterns for this verb, this is beyond the scope of this article, and this claim is therefore neither made nor explored further here.

A particularly interesting case of divergent case frames on the same predicate in Tukanoan is found with throw (at). Contrast the instance in Kotiria in (7a), where the thrown object is encoded with an accusative flag and the goal of the throwing action takes the locative marker, and the frame in Tukano in (7b), where the thrown object takes an instrumental flag and the target of the throwing action is flagged with the accusative marker.

With regard to Quechuan languages, for contact verbs, the same verbs deviate from the nominative–accusative pattern as in the Tukanoan languages. This is across the board: in both the foothill languages and the control languages. In most Quechuan languages, either the locative or allative/dative case marker is used to mark the P/R participant. This is illustrated in (8) for Tungurahua Kichwa – note that this variety aligns its case frame selection for the predicate throw (at) with Kotiria (with accusative marking on the thrown object rather than the goal like in (7a)), as compared with the strategy shown for Tukano in (7b). For the sake of comparison, however, (9) demonstrates an instance from Upper Napo Kichwa with stick (to) where the animate goal-like P receives the accusative flag.

As suggested earlier, it is not always clear how to best classify these verbal concepts. For this reason, we have expanded what we found for each language across the four verbs that fed the variation discussed in the preceding examples with explicit role-marking correspondences for reference in Table 3. In the interest of unifying the terminology for our discussion of these predicates, we chose to collapse some sets of related markers under a single label where syncretism was observed, namely, ALL (‘allative’) subsumes dative, goal, and allative or directional marking, whereas INST (‘instrumental’) subsumes both instrumental and comitative marking.

To conclude our discussion of case-marking patterns associated with the P-like element, we shift to consider motion predicates. As per Malchukov’s semantic map, motion predicates are the least prototypically transitive semantic class as regards the prototypicality of their goal-like P argument, and this is where we ought to expect to find the highest proportion of noncanonical case frames. As Figures 5 and 6 demonstrate for the Tukanoan and Quechuan families, respectively, this fact is borne out in our data. Although motion predicates tend to select for noncanonical case frames in both families as predicted, there are notable differences between these families with regard to flagging patterns for these concepts. This fact is immediately clear when considering the distribution of accusative marking with motion predicates in Tukanoan (27% of instances in the foothill languages, and 44% of instances in the control languages) and those in Quechuan (50% for all varieties). Further details regarding the crucial differences across these language families are discussed later in this article.

Figure 5

Accusative marking on P/R participant in Tukanoan per verb type.

Figure 6

Accusative/dative marking on P/R participant in Quechuan per verb type.

To begin with the languages of interest in the Tukanoan family, the Western Tukanoan languages Siona and Koreguaje have a dedicated goal (or allative) marker -na. As discussed in Section 2, this marker is used differentially along the lines of differential subject/object marking in these languages, as well as used with all predicates that take a goal-like P, such as go. In addition, these languages also have a dedicated source (or ablative) marker -bi,^[15] which is used with motion predicates like leave. The same rich case system and marking patterns are found in the closely related Western Tukanoan language Sekoya (Johnson and Levinsohn 1990, 45–9). In the case of Máíhɨ͂̀kì, which has a reduced case inventory compared with the other Western Tukanoan languages, such goals are typically left unmarked altogether. The accusative marker arises where the goal is animate. However, it is not clear whether there is additional semantics associated with this role and whether it should therefore be treated separately, as in “I am going to (see and speak with) X” (Neveu 2012). For this reason, we included duplicate entries for these predicates to represent both patterns in our sample. Taken together, typical destination goal-like Ps with motion predicates do not straightforwardly receive the canonical flagging pattern associated with prototypical P-like arguments in any of the Tukanoan languages within the interest area.

Some of the control languages from our Tukanoan sample exhibit a comparable pattern with motion predicates to that discussed earlier for foothill varieties. Each of the Eastern Tukanoan languages studied presented a dedicated locative marker (or one of several such markers) on the P-like argument of motion predicates like go. These markers do not exhibit the same differential distribution as noted in the case of Siona and Koreguaje. Interestingly, Kotiria, Tukano, and Barasana optionally permit the specificity marker (identical in shape to the accusative maker -re, or perhaps the object marker itself) to appear stacked after the locative marker in such instances. Miller (1999, 61–2) also discusses a more restricted set of locative-accusative stacking in Desano. Such a stacking pattern is not noted in any of the Western Tukanoan varieties considered here. Stenzel (2014, 140) suggests that the particular configuration of the locative, in Kotiria -pʉ, and the optional -re marker might best be treated as encoding the allative case, and not as accusative marking as is found with more prototypical transitive verb classes. We encoded all languages with such patterns as selecting for a nominative–oblique frame with these predicates. In the case of Desano, most goal-like P arguments are flagged with the locative marker, as in (10a), whereas in certain marginal instances, this P is flagged with the accusative, as in (10b).

With respect to motion verbs, the picture is slightly different in Quechuan languages than in Tukanoan languages. The P-like argument associated with motion predicates can be encoded in two ways. As mentioned in Section 2.2, either the goal-like P receives the canonical accusative marker -ta as found with more prototypically transitive verb types or it can be marked with an oblique, namely, the dative/allative marker -man. The extension of accusative marking to goal-like Ps is common to all varieties of Quechuan (Adelaar 2017). As Adelaar (2017, 654) explains “characteristically [of Quechan and Aymaran generally], a verb selects an accusative case for the object or goal that is typical for its meaning, regardless of whether the latter is a direct object, an indirect object or a geographical goal.” In other words, the accusative case marking pattern in Quechuan extends into the class of motion verbs, which is unusual from a cross-linguistic perspective, although languages like English marginally allow for this pattern as well (as in to enter or to leave). Given that both a nominative–accusative and a nominative–oblique frame can be chosen in the Quechuan varieties figuring in this study, each relevant predicate received two entries; hence, the percentage representing the proportion of canonical flagging patterns with Quechuan motion predicates ends up as notably higher than those noted for the Tukanoan languages (comparing Figures 5 and 6, respectively).

4.2 Marking of A-like participants across semantic classes

When we zoom in on the flagging on the A-like participant, the languages in our sample very consistently mark A-like participants with a nominative marking pattern. This is true for both Tukanoan and Quechuan languages. Focusing first on the Tukanoan languages, Figure 7 neatly displays how they attribute structural nominative flagging to the A-like participant across all verb types, with the notable exception of sensation predicates, which are discussed at length later in this section. This pattern holds for languages within the putative contact area, namely, the Western Tukanoan varieties collapsed under the label “interest” in Figure 7, as well as for the Eastern Tukanoan languages considered as “controls.”

Figure 7

Nominative marking on A participant in Tukanoan sample per verb type.

The only nonsensation predicate where a noncanonical flag was found on the A participant in our Tukanoan sample is noted for a spontaneous predicate fall asleep in the control language Kotiria. This is an example of what we will later call a “suffer”-type predicate, also found in sensation predicates (see the following table).

For the sake of comparison, Figure 8 demonstrates the distribution of nominative marking patterns across the Quechuan varieties in our sample. Just as in the case of P-like participants discussed in Section 4.1, the pattern is more or less the same as in the case of Tukanoan, save for a few outlier predicates. Again, it appears that Quechuan has more, albeit only slightly more, outliers than does Tukanoan.

Figure 8

Nominative marking on A participant in Quechuan sample per verb type.

There are marginal patterns of noncanonical A marking with emotion and cognition predicates in Cañar and Upper Napo Kichwa, where the A-like participant is flagged with the accusative marker. First, consider the construction used in Upper Napo Kichwa as a natural translation for the cognition predicate remember.

Next, we turn to the case of the emotion predicate gushtana (like) as found in both Cañar and Imbabura Kichwa. In the case of Cañar Kichwa, it is possible to mark the A-like Experiencer with accusative -ta as shown in (13a), and this is the preferred strategy. In this case, both the lexical item and the case frame are borrowed from Spanish. However, it is also possible to mark the P-like stimulus argument with accusative marking using the canonical case frame (see 13b). This is the exclusive means of encoding the cognate predicate in the case of Imbabura Kichwa as shown in (14).

Returning to Cañar and Upper Napo Kichwa varieties, the same two languages present instances of noncanonical marking on the A-like participants of spontaneous events:

Conversely, it is relevant to acknowledge that Figure 8 shows no instances of nonnominative marking on A participants with nonsensation predicates in the control languages in the Quechuan sample of our study. On a general level, Quechuan languages, like Tukanoan and Chicham languages in the area, opt for the structural case marking strategy on the A-like participant.

In clear contrast to the other verb types, sensation predicates show the most deviant patterns with respect to noncanonical marking of A-like participants. This is again in line with Malchukov’s (2005) observation that sensation predicates are the most likely candidates for deviant encoding on the A-like participant (see also Section 3.1). This fact is borne out in our data in the case of all Tukanoan and Quechuan varieties found within our interest area and, although with less consistency, in our control languages.

To begin with, we consider the case frame associated with the exceptional sensation concept hurt. The valency of this predicate is unique when compared with the other sensation concepts in our study, since it presents both an Experiencer A and a Stimulus P. The remainder of the sensation predicates are intransitive, following the transitive-intransitive continuum in Malchukov 2005 (see Sections 3 and 5 for a brief discussion). We observed a split in encoding strategies of these predicates across our sample. Tukanoan languages are unanimous in flagging the A participant with the accusative marker, leaving the P argument unmarked, whereas all Quechuan varieties in our sample extend accusative marking to both the A and the P arguments. These observations are illustrated in the examples in (16).

We now shift our attention to the remainder of the sensation concepts, where we find the highest degree of variation across the languages studied. In our data, we note three competing strategies that dominate the formation of sensation predicates in our sample: the underived predicate strategy, the desiderative experiencer strategy, and the “suffer”-type experiencer strategy. It is important to note that none of these strategies require a noncanonical case frame in all cases. In addition, it is worth recognizing the fact that a language may exhibit more than one strategy, even for lexicalizing the same verbal concept, thereby forming (near) synonymous lexical pairs. This fact is explored in detail later in this subsection.

The first strategy refers to certain sensation predicates, which are best analyzed as having their own dedicated lexical items just like most predicates of other verb types. The lexical item associated with these predicates does not recruit any additional derivational morphology or straightforwardly combine with a light verb to represent the verbal concept. We simply refer to this as the underived predicate strategy. A handful of illustrative examples with accusative-marked experiencer A participants are presented in (17a) and (17b) and then with nominative experiencers in (16c).

Conversely, the desiderative experiencer strategy refers to a derivational operation where a dedicated desiderative marker is attached to the verbal (and sometimes nominal) root. In the case of many sensation predicates in the languages of our sample, this strategy is used to provide a natural translation for lexical concepts like “to be hungry,” “to be thirsty,” “to be sleepy,” etc. Below are examples of the desiderative experiencer construction with an accusative-marked Experiencer A participant in Tukanoan (18a) and Quechuan (18b), contrasted with a desiderative construct in Kotiria that does not trigger deviant marking on the experiencer A in (18c), for the sake of illustration:

The third strategy that is recruited for lexicalizing sensation predicates, which we have labeled the “suffer”-type experiencer strategy, makes reference to a set of sensation predicates that are lexicalized via the combination of a sensation element (sometimes this is clearly a verbal root, whereas in other cases, it may also be nominal) and a verbal head with a meaning along the lines of “suffer/hurt,” “be.sick,” and “die/kill,” This construction type is commonly found in the Tukanoan languages of our sample. Languages differ with regard to whether these constructions select for (non-)canonical marking on the A-like experiencer participant. The following examples demonstrate instances where the experiencer A is flagged with the accusative marker, as in (19), and others where it receives the canonical nominative flagging pattern, as in (20):

As demonstrated in example (20a) from the neighboring Chicham language Aguaruna, the ‘suffer’-type strategy extends beyond the primary languages considered in this sample. Further research is required to determine the extent of the distribution of this strategy in other areas throughout South America and elsewhere.

It is also worth noting that there may well be a miscellaneous category of nonproductive constructions that signal the meanings of the sensation predicates and select a noncanonical frame. Consider the following instances of clear (near-)synonyms with the aforementioned sensation predicates in our sample, but which do not form part of a commonly found pattern across the region. We collapse such constructions under the label “other” for the sake of the discussion at hand.

Taken together, the points earlier demonstrate the ^[22]complexity associated with the class of sensation predicates. Although it is not always the case that the predicates from this class select for a noncanonical case frame, there is certainly a high proportion of predicates which do in fact assign nonnominative flagging to the A-like Experiencer participant. Table 4 traces the precise distribution of the strategies alluded to in the discussion earlier (and their comitative case frames) across the languages in our sample.

In unpacking the variation captured in Table 4, the first striking fact is the sheer variety of constructions across and within the language families.^[23] There are essentially as many instances of derived predicates as there are underived predicates that serve as natural translations to the predicates at issue in the current study. Below, we discuss the distribution and details surrounding each of the major derivational strategies found in the area in turn.

First, regarding the “suffer”-type strategy, Table 4 shows that it is not found within our Quechuan sample. This type of lexicalization for sensation concepts appears to exhibit varying degrees of productivity across the languages of the sample. The concept be sick often involves a sensation verb, which selects for an accusative-flagged Experiencer. It seems that the bare predicate does not take an accusative A; an exception is Kubeo, as shown in (22a). The only other accusative-flagged A found with the natural translation of the concept be sick is found in Máíhɨ͂̀kì, exemplified in (22b). However, this instance combines the sensation of feeling sick with another generic sensation predicate. It is likely the case that the generic sensation verb is the culprit for triggering noncanonical flagging on the A participant, as discussed later.

Turning now to the desiderativization strategy, we note several crucial differences with respect to the “suffer”-type strategy. First, the desiderativization strategy is found in both Quechuan and Tukanoan languages, although, in the case of the latter family, this pattern appears to be restricted to the foothill languages. For the most part, this strategy is only found with the concepts be hungry, be thirsty, be sleepy, and not with the concepts be hot, be cold. A notable exception is found in Máíhɨ͂̀kì, where the desiderative is found with certain body-external concepts like be hot, as exemplified in (23).

In fact, the strategy of desiderativization goes beyond the sensation concepts discussed here and appears to be a productive process in many Tukanoan and Quechuan languages. Consider the following cases from two of the foothill Tukanoan languages where the desiderative is used with nonsensation concepts, and the A-like participant is marked with the accusative.

For the sake of clarity, we reiterate (18c) in (25) to demonstrate an instance of a productive desiderative morpheme in an Eastern Tukanoan control language Kotiria, where the A participant does not receive accusative flagging. Similar patterns are found in the other Tukanoan control languages.

Despite the fact that such desiderative morphemes, exemplified by Kotiria in (25), are not cognate with the marker -ɨa found in the Western Tukanoan languages, the story does not end here. In fact, several Eastern Tukanoan languages do have a cognate verbal item with the meaning want or desire (i.e., ɨá in Tukano per Ramírez 2019[1997], 92, 96; ɨ in Kubeo per Morse and Maxwell 1999, 71), just like Siona has a main verb yɨ’yɨ (want), and none of these trigger accusative marking on the A-like participant. This fact is suggestive of the desiderative lexeme ɨa itself being a retention from an earlier variety of Tukanoan, whereas the desiderativization strategy with noncanonical A flagging, discussed earlier, is likely a more recent Western Tukanoan innovation. In light of the fact that this strategy appears to be commonplace across all varieties of Quechua, and restricted only to the Western Tukanoan languages within the putative foothills contact area, we suggest that the D* strategy found in Siona, Koreguaje, and Máíhɨ͂̀kì (following Table 4) is a plausible candidate for Quechua-Western Tukanoan structural convergence. Siona presents a particularly interesting case where this strategy competes with the ‘suffer’-type lexicalization strategy that is common to all Tukanoan languages in our study, where the same A-like participant is accusative-marked in the former case, but canonically nominative-marked in the latter.