2.1 The historical and cultural context of the region

A language contact scenario can be thought of as “the organized fashion in which multilingual speakers, in certain social settings, deal with the various languages in their repertoire” (Muysken Reference Muysken2008d). This chapter is based on the premise that language is in large part like any other cultural trait, whose practice can be inherited, acquired, modified, or lost by any generation of speakers (Mace et al. Reference Mace, Holden and Shennan2005; Gray et al. Reference Gray, Greenhill and Ross2007, Gray et al. Reference Gray, Bryant and Greenhill2010). Therefore, an assessment of the non-linguistic record that encompasses details of the archaeology, ethnohistory, and ecological history of the region is relevant to our understanding of the language contact scenario: the hypothesis is that patterns of speaker behavior in dealing with non-linguistic cultural practices will shed light on how speakers may have dealt with the various languages that entered their environments, as well. This section provides a brief sketch of the historical and cultural context of the Isthmo-Colombian area, looking at evidence for practices related to subsistence, social organization, and trade, starting from the earliest known populations through to the character and consequences of an apparent watershed moment, roughly 1,500 years ago.

People have been living in the Isthmus of Panama since the late Pleistocene, some 12–10,000 years ago (see O’Connor and Kolipakam, this volume). Multidisciplinary research suggests that the Chibchan communities in the region today are in fact the genetic (Barrantes et al. Reference Barrantes, Smouse, Mohrenweiser, Gershowitz, Azofeifa, Arias and Neel1990, Melton et al. Reference Melton, Briceño, Gomez, Devor, Bernal and Crawford2007) and linguistic (Constenla 1991, 2012) descendants of the earliest inhabitants, having spread north into Nicaragua and southeast into Colombia, eventually occupying scattered territories of the Caribbean littoral and the drainage areas of the Cauca and Magdalena rivers. Although these groups apparently engaged in frequent conflict with each other and with non-Chibchan neighbors, they are also described as connected by a “diffuse unity” that encompassed belief systems and associated material practices (Hoopes and FonsecaReference Hoopes, Oscar, Fonseca, Quilter and Hoopes2003). Other studies discuss an “Isthmian Interaction Sphere” that extended from central Colombia to the Mexican Yucatan, in overlapping circles or down-the-line chains, within which commerce and the practice of other cultural activities flowed back and forth across stable local boundaries (MyersReference Myers and Browman1978, BrayReference Bray, Lange and Stone1984, CookeReference Cooke2005).

As will be presented in more detail below, Proto-Chibchan probably split from an older Central American stock nearly 10,000 years ago, while all but one contemporary Chibchan language developed from a core branch that emerged 3,000 years later, its speakers gradually filling the narrow isthmian region of Costa Rica and Panama. The diverse ecology of the region, with deep mountain valleys and rich coastal resources on both shores, encouraged long-term settlement of small groups that practiced a wide variety of subsistence strategies. People ate fish, shellfish, and birds, and evidence of early, local agriculture includes traces of bottle gourd, arrowroot, leren, and squash in Panama 9000–7000 BP (Cooke Reference Cooke2005). Raymond (Reference Raymond, Silverman and Isbell2008) notes maize, manioc, arrowroot, and yams on the Pacific coast, where skeletons show that maize was a primary ingredient of the diet by 6000 BP. The presence of both maize, from Mesoamerica, and manioc, from South America, demonstrates the impact of long-distance exchange in this cultural and commercial nexus.

By the time Chibchan speakers started to radiate away from the core Chibchan region, beginning some 5,000 years ago, a clear record of sustained human settlement had already emerged in multiple places. This development is seen in three archaeological sites mentioned repeatedly in the archaeological literature that nicely delimit our region of focus: Cerro Mangote, from pre-6000 BP, on the Pacific coast of Panama; Las Vegas, from 8500–4600 BP, on the Santa Elena peninsula in western Ecuador; and Puerto Hormiga, from 5100 BP, near the Caribbean coast in the estuaries of the Magdalena River in northern Colombia. Evidence at all three sites points to the existence of foragers and collectors who practiced a type of residential mobility (RaymondReference Raymond, Silverman and Isbell2008: 80–86). These mobile communities trekked in circuits, probably to achieve their attested varied diet of fish, birds, game, and other forest products, returning to the central bases to bury their dead in what Raymond notes as symbolic if not physically permanent homes (p. 81). As noted above for Panamanian societies, in Ecuador, too, we find evidence of very early plant domestication, with traces of squash and leren from nearly 10,000 BP, and maize and bottle gourds by about 8000 BP.

All three regions described above had ceramics fairly early. Pots were found at Valdivia (after 4500 BP) near the older Las Vegas site, and sand-tempered pots of lower quality were found at Monagrillo, around 4400 BP, near the older Cerro Mangote site. Distinctive large and globular ceramic bowls called tecomate, made of clay tempered with fibers, were found throughout Colombia at sites such as Puerto Hormiga, Turbana, Monsu, and San Jacinto, with the earliest tecomate dated pre-5000 BP (AllaireReference Allaire, Salomon and Schwartz1999: 679). Similar tecomate ware has been documented throughout the greater Central American region: from 4000 BP, on the Guajira Peninsula and the Pacific coast of Guatemala; from 3000–3500 BP, at Tronadora and Chaparron in Costa Rica; and from 3000 BP, at Momil in the Sinu lagoons of northern Colombia (Allaire Reference Allaire, Salomon and Schwartz1999). On the evidence of these ceramics, Myers (Reference Myers and Browman1978) made the case that overland trade routes could indeed have linked the coasts of Guatemala and Ecuador, and he also suggested that Puerto Hormiga ceramics looked more like those of the Orinoco and Amazon than like those of Ecuador. It is often quite difficult to prove an ultimate origin of particular ceramics; to complicate the question, the word tecomate appears to come from the Nahuatl word tecomatl, which describes this very type of pot. And yet, perhaps the Isthmo-Colombians traded tecomate pots for Saladoid and Barrancoid ceramics coming from the Orinoco region, as well as for jade from Guatemala, both of which entered the area by about 3500 BP.

Despite our increasing knowledge of the region through archaeological and ecological data, we cannot reconstruct with confidence any major social trends or patterns of dominance for much of prehistory. Scholars of the Isthmo-Colombian area do however note a major moment of change sometime around 500 CE (Bray Reference Bray, Lange and Stone1984: 331, Allaire Reference Allaire, Salomon and Schwartz1999: 707, Hoopes Reference Hoopes2005). The transition may have been motivated in part by increased consumption of maize, a much better crop for floodplain cultivation (Bray Reference Bray, Lange and Stone1984), or by climatic events, such as environmental catastrophe (Hoopes Reference Hoopes2005), and many think the sociocultural change was linked to a transition from jade to gold as the precious material of greatest cultural importance. Quilter (Reference Quilter, Quilter and Hoopes2003) describes the paradigm shift. Mesoamerican jade was difficult to obtain, and its hardness meant that working with it was a slow if relatively easy process. Once shaped, jade artifacts were quite durable. Gold was found locally, and the rendering process was complex and somewhat mysterious yet relatively fast. Furthermore, gold artifacts could be melted down and transformed into an entirely different object. The power of gold, as a commodity to be mined, owned, worked, traded, stockpiled, refashioned, and passed on to descendants, played a key role in the emergence between 300 and 600 CE of the ranking, inequality, and complexity that still characterized the societies encountered 1,000 years later by Europeans (e.g. Quilter and Hoopes Reference Hoopes, Oscar, Fonseca, Quilter and Hoopes2003, Hoopes Reference Hoopes2005). Many details of social development remain to be deciphered. As observed by Bray (Reference Bray, Lange and Stone1984: 307), early Spanish chronicles from the Gulf of Uraba, between Panama and Colombia, mention witnessing “a thriving business in slaves, fish, salt, cotton cloth, and live peccaries, as well as gold.” Bray continues, “It is worth noting that most of the products on this list will leave no archaeological trace and that pottery does not figure at all.”

Some of the societies that were powerful during the last millennium before the Conquest would be particularly relevant to the linguistic analysis in this chapter. Among these are two whose languages remain uncertain: the Zenu (or Senu), who erected astonishing raised fields over vast expanses of the San Jorge River basin in northwest Colombia, and the Quimbaya, renowned and prolific gold workers from farther south on the upper Cauca, whose gold work has been found throughout the Isthmo-Colombian area. There were also two important centers of Chibchan-speaking groups near the Magdalena River along the eastern edge of the area. The Muisca realm occupied the upper Magdalena, near present-day Bogota, at a site known to early Spaniards as El Dorado for its power and wealth, most notably in gold and emeralds. In the Sierra Nevada of Santa Marta near the mouth of the Magdalena, we find the Tairona civilization. The Tairona melded complex architecture with diverse agricultural practices in the design of terraced villages and fields at various altitudes, and they left a copious iconographic record of their rich religious life. Archaeological evidence suggests all these societies were chiefdoms, entities that demonstrate organization and hierarchy in political, religious, and economic activities, seen in artifacts such as large-scale public works, settlement layout, burial practices, and iconography (Bray Reference Bray, Lange and Stone1984: 331).

There is a growing body of literature from recent decades on what constitutes a chiefdom and on the many ways sociopolitical complexity can be manifest, and these definitions shape what we might expect from a given social scenario in terms of language contact. For example, Hoopes (Reference Hoopes2005: 6–9) discusses the literature on two modes of social power, known as network and corporate. A network type of chiefdom emphasizes such factors as individual power passed through hereditary lines, centralized chiefs, and commercial power achieved and maintained by military means. This type of scenario seems more likely to lead to language shift, as speaker communities are conquered and subjugated to enrich a central power. In contrast, the outcome of a corporate mode of social power may be more consonant with diglossia and language maintenance, as a corporate mode emphasizes the power of the office, non-linear inheritance, and control achieved through ritual and ideological means. In this context, the important leaders could have been not chiefs but priests and shamans, who exercised locally the power of a broadly shared worldview, expressed in common iconography and “routinized ritual” rather than through control of key resources (Hoopes Reference Hoopes2005: 31). Cooke (Reference Cooke2005: 31) proposes a type of mixed model, in which “above the chiefdom, there were larger, equally important social units – to judge from the ethnographic record, some kind of descent group or groupings of ethnias with closely related languages and memories of common origins, shared songs and praises, and conflicts between real and mythical personalities and social groups.” Thinking again of the linguistic outcome, we might envision the role of Latin as a language of worship that left space for the maintenance of local languages. Within the Isthmo-Colombian region, chiefdoms of Central America tended toward the network model, perhaps due to influence from Mesoamerica, and chiefdoms of Colombia tended toward the corporate model, a difference that would likely have consequences for patterns of language contact.

This section began with a mention of the “diffuse unity” said to characterize the Isthmo-Colombian area. Bray (Reference Bray, Lange and Stone1984: 336–337) argues the opposite side of the same coin: that despite constant contact and constant conflict, especially among close neighbors, individual cultures in the region remained distinct. He calls this phenomenon “conservatism in the face of opportunity for change” and proposes that while population stability may play a role, the true barriers to convergence were ideological:

The key notion here is that societies seem to have accepted the basic frameworks of new technologies, practices, or artifacts, and to have adapted and reproduced the new structures with locally relevant contents. With this notion in mind, I will summarize what this brief review of the non-linguistic literature on the Isthmo-Colombian area can bring to the question of the linguistic prehistory of the region, and especially to the investigation of the effects of language contact. Speakers of Chibchan languages have been in situ for millennia, with mostly Chibchan neighbors throughout Costa Rica and Panama and among mostly non-Chibchan speakers in scattered pockets from Nicaragua to central Colombia. There was ongoing contact and conflict, especially among nearest neighbors, which suggests a degree of bilingualism (or multilingualism) and intermarriage (which may have been forced, as an outcome of conflicts). Something happened around 500 CE that led to greater sociopolitical complexity that affected the entire region, shaped societies for the next 1,000 years, and may have involved the imposition of dominant languages, in the form of language shift or of diglossia. Throughout, individual cultures remained relatively distinct, but at the same time, societies did take advantage of new technologies and practices, accepting the frameworks and adapting details of the content to fit local needs.

When looking at language systems, stability may take different forms. Basic vocabulary is expected to be stable and to indicate family relations, while cultural vocabulary is expected to show more effects of cross-family borrowings that reflect the specific cultural and ecological context. A resistance to lexical borrowing is often interpreted as the conscious maintenance of a distinct social identity, a phenomenon documented in contact situations from the Vaupés (Aikenvald 2002; EppsReference Epps, Aikhenvald and Dixon2007a, Reference Epps, Matras and Sakel2008a) to Vanuatu (FrançoisReference François2011), and proposed for languages of Colombia as well (O’ConnorReference O’Connor2011). As was discussed in the introduction of this chapter, categories of structural features and their interpretation are less clear, but there are general expectations that stable features will reflect genealogy better than unstable features will. This chapter contributes a perspective from structural data that unpacks relative stability, or relative resistance to borrowing, using two types of metrics, one of which explicitly operationalizes the notions of abstract template and locally relevant content, as described by Bray (Reference Bray, Lange and Stone1984).

2.2 The Chibchan family

The Chibchan language family is by far the largest family in the Isthmo-Colombian Area, in number of languages and in geographic spread, and we know a great deal about the history of this family thanks especially to the work of Constenla Umaña (e.g. Reference Constenla Umaña1981, Reference Constenla Umaña1991, Reference Constenla Umaña, Campbell and Grondona2012) and Quesada (e.g. Reference Quesada1999, Reference Quesada2007). Citing evidence from phonological, lexical, and grammatical comparison and reconstruction, Constenla (2012: 418) suggests that the proto-language split around 9,700 years ago from a Lenmichí “micro-phylum” composed of the Lencan, Misumalpan, and Chibchan families. The Paya language, now spoken in northern Honduras, probably split from the proto-language some 6,700 years ago, leaving what is known as Core Chibhan (see Figure 4.1). Judging by the distribution and degree of present diversity, Constenla (2012: 419) presumes a Chibchan homeland in southern Central America and estimates that it was from this Isthmian homeland that other branches developed as speakers migrated north (Votic branch, ∼5325 BP) and east across northern Colombia (Magdalenic branch, ∼5225 BP), with another migration east by Kuna speakers some 4,800 years ago.

Figure 4.1 The Chibchan language family, after Constenla (2012: 417). Boxed languages appear in this study.

There were at least twenty-one languages, of which sixteen survive. Two of the extinct tongues would have been particularly useful for this study of language contact: the Antioquian languages Catío and Nutabe, thought to have been spoken between the Sinu and Cauca Rivers, near the Zenu and Quimbaya societies mentioned in Section 2. Sadly, they are virtually undocumented and therefore could not be included in Constenla's classification of Chibchan subgroups. Of particular interest in this chapter is the observation that “Tairona… seems not to be another language, but a variant of the still spoken Damana” (Constenla 2012: 391, citing older literature).

2.3 The languages in the study

The fourteen languages for this study, listed in Table 4.1, were chosen for their geographic location on the borders of the Chibcha sphere and because there were sufficient descriptive materials available for the data collection questionnaire, described in Section 3.

Table 4.1 Languages in this study

A fundamental goal of this paper is to determine if any particular subset of features can be identified as a good “trace of contact”: in other words, will any category of feature highlight areal relations among languages and speakers by occurring in patterns that correlate with geographic proximity of languages irrespective of genealogical relationship? Regional subareas are defined here (see Table 4.1) as a Northern group (languages 1–4), an Isthmian group (5–7), a Southern group (8–11), and an Eastern group (12–14).¹ The subset of features which best replicates these geographic subareas will therefore be claimed to contain the features most susceptible to the effects of contact in the given social scenario.

It should also be noted that, even in this small inventory of languages, assigning languages to areal groups is itself a matter for investigation and experimentation (see Map 4.1). We might expect Paya to group with the Jicaquean and Misumalpan languages. Jicaquean is a small family unconnected to any other, and not much is known about the history of the people. It is spoken in northern Honduras, along the Caribbean coast, and while it is likely a long-term neighbor of Paya, there is no known interaction. Misumalpan is a small family of languages spoken primarily throughout central and eastern Nicaragua, extending across the border into southern Honduras, and with a small pocket of speakers on the border between Honduras and El Salvador (Constenla 1991). The only representative in our dataset is Misquito, especially interesting for its historical extension all along the Caribbean coast of Nicaragua, where speakers could have had contact with Paya to the north and Rama to the south.

Rama, from the Votic branch of Core Chibchan, could pattern with the Northern group or alternatively with the Isthmian languages Teribe and Guaymi, while the position of the third Isthmian language, Kuna, could be expected to vary between its Chibchan Isthmian cousins and the Chocoan languages with which it has surely been in contact. The Chocoan language family is composed of two living language varieties, Waunana and Embera. Embera is itself described as a set of closely related languages or as a dialect continuum, each variant named for the region where it is spoken, and it is divided into Northern and Southern branches. The sample here includes one Northern Embera variant (called Northern Embera) and one Southern Embera (Epena Pedee). Chocoan languages are spoken today all along the Pacific coast of Colombia and into eastern Panama, and the speakers call themselves terms that translate as ‘mountain dwellers,’ ‘river dwellers,’ and ‘people of the wild cane’ (MortensenReference Mortensen1999: 1). Historically, they are known as a “flexible and expanding population” who have settled in regions vacated by other groups during the process of colonization (Adelaar with MuyskenReference Adelaar, Adelaar and Muysken2004: 56–57). Several extinct languages prominent in the discussion of the Isthmo-Colombian prehistory have been associated with Chocoan, though none of these has sufficient documentation to confirm the relationship. These include Cueva, which was spoken on the Isthmus between Kuna to the east and the rest of Isthmian Chibchan to the west, and the extinct Colombian languages Quimbaya, of the Upper Cauca Valley in western Colombia, and Sinúfana, of the Sinú region between the Sinu and Lower Cauca rivers near the Caribbean coast.

The Southern Embera languages may show areal similarities with a Southern group that contains Paezan and Barbacoan languages. Paez is the only language (or only surviving language) in the Paezan family, spoken on the eastern and western slopes of the Andean cordillera central in southwestern Colombia. Paez has had known contact with the surrounding Barbacoan languages Guambiano and Totoro and likely contact with Southern Emberan (Chocoan) languages of the nearby Saija, San Juan, and Cauca River systems. The Barbacoan languages are spoken in separate pockets scattered from the mountainous regions of southwestern Colombia to the coastal lowlands of northwestern Ecuador. The Barbacoan language in this study, Awa Pit, is spoken in the western foothills of the Andes along the Colombia–Ecuador border. This language is perhaps an odd choice for a study of areal contact, as the community is (and may have long been) known for a culture of “secrecy” and inaccessibility (CurnowReference Curnow1997; Curnow and LiddicoatReference Curnow and Liddicoat1998).

The final subregion in this study is the Eastern group, centered on Ika and Damana, Chibchan Magdalenic languages of the Sierra Nevada de Santa Marta, which may have had contact with the Northern Maipuran Caribbean branch of Arawakan languages. As mentioned previously, Damana may be the modern version of the language spoken in the powerful Tairona chiefdom, a factor which may impact its regional profile. Arawakan is a large family of around thirty languages spread geographically from Belize to Bolivia. The language of interest here is Guajiro, of the Guajira peninsula on the Caribbean coast at the border of Colombia and Venezuela.

Regretfully, no Western Cariban languages could be included in this dataset due to insufficient documentation. Cariban is a large family of forty to sixty languages, many extinct, and mostly spoken from the Orinoco basin of eastern Venezuela across the Guianas to the Amazon, and into central Brazil. There were Cariban speakers in the Magdalena River valley, and some scholars suggest that at least some of the unknown languages spoken throughout the Caribbean lowlands of northern Colombia were also Cariban. Constenla notes that between the Kuna and the Magdalenic group of Chibchan “there was a series of people of proven or supposed Cariban affinities, such as the Opon, the Muzo, the Panche, and the Pijao” (2012: 419). The surviving Cariban language closest to the Isthmo-Colombian region is Yukpa, a language cluster with scant documentation, spoken just west of Lake Maracaibo along the Colombia–Venezuela border. Relying on small descriptions in older sources, Constenla (1991: 60) described Yukpa as SV/SVO, with genitive and demonstrative before the noun and adjective and numeral after the noun. In more recent work, Flores (Reference Flores2002) finds that while constituency is varied, the basic word order of Japreira, one language of the Yukpa group, is SOV. Postpositions are illustrated as suffixes, and nominal constituent order includes Genitive-Noun, possessor-possessed, and both Adjective-Noun and Noun-Adjective.

A typological overview of the relevant language families, based on the Constenla (1991) binary-coded dataset, is presented in Table 4.2.

Table 4.2 Typological profiles, using features from Constenla (1991)

While some feature values seem rather widespread, such as SOV basic word order, postpositions, and case suffixes everywhere but in Arawakan, we can also see a certain amount of variation and indeed several features with “mixed” answers within the Chibchan family.

Interestingly, the history of the genealogical classification of some languages in the study has taken them from sisters to cousins to neighbors. The Chibchan family was identified as such by Uhle (Reference Uhle1890). Genealogical classifications involving the Chibchan, Chocoan, Barbacoan, and Paezan families include efforts by Rivet (Reference Rivet, Meillet and Cohen1924b) and Loukotka (Reference Loukotka1968), that grouped Barbacoan and Paezan inside Chibchan; by Greenberg (Reference Greenberg1987), that proposed a Chibchan-Paezan subgroup within the single family Amerind, placing Barbacoan and Chocoan inside the Paezan division; and by Campbell (Reference Campbell, Campbell and Grondona2012a), that registers the groups as four distinct families without known interrelation.

3.1 Stable vs. unstable

The sets of stable and unstable features used in this study were selected from a proposed stability ranking of structural features (Dediu and Cysouw Reference Dediu and Cysouw2013), itself based on a comparative analysis of eight individual approaches to calculating the stability of features archived in the World Atlas of Language Structures (WALS). The eight studies made use of different statistical analyses and operated under different definitions of stability. Some were based on the persistence of features within families, others on estimates of the evolution of feature values through time, and others measured patterns of persistence within the dataset without initial consideration of known language families. Every study devised an estimate of relative stability for each individual WALS feature, up to a total of 132 features, depending on the study. Dediu and Cysouw then converted the stability estimates into relative ranks from 0.00 (least stable) to 1.00 (most stable) to facilitate comparability, reported in their Table 1, and their Table 7 presents a composite ranking for the 62 features represented in all eight analyses, based on principal component analysis. For the present study, features 1–31 in the 62-item list were classified as stable, and features 32–62 as unstable. To supplement the dataset, 19 additional features were chosen from the 132-item list by averaging the seven relative rank scores reported for that feature. The resulting average score was then compared to the average score of the “cut-off” feature (that is, feature 31 in the 62-item list), calculated by averaging the same seven relative rank scores, in order to situate each supplementary feature in the appropriate stable or unstable category.²

The features used in this chapter were chosen with four parameters in mind: (i) relative position in the proposed stability rankings, (ii) presence in the Constenla (1991) dataset of typological features, (iii) presence in WALS, and (iv) likelihood of appearance in existing descriptive materials for the languages in question. The last parameter is subjective yet realistic, given the scarcity and brevity of materials on under-described languages of South America.

The Constenla (1991) dataset consists of binary indications (yes = 1, no = 0) of the presence of 42 morphological features and 39 phonological features in 76 languages of Mesoamerica, Central America, and northwestern South America. The first step in data collection was to incorporate all relevant information from the Constenla dataset for the 35 languages of the region in question, yielding data for 35 features in all languages. Next, all possible information from WALS (accessed 13 August 2012) was added to the dataset, providing information on 59 more features for only some languages (reflecting the uneven coverage in WALS). If WALS contained more recent documentation that contradicted information from Constenla (1991), the WALS feature value was used. These collections were especially fruitful for the subset of stable features. Published grammatical materials were then consulted to code the remaining features, resulting in data for 49 stable features and 49 unstable features for 14 languages. Certain pairs of questions, such as “Is the order of the NP Adj-N?” and “Is the order of the NP N-Adj?” gave only duplicate information for this set of languages, so these questions were eliminated. Nevertheless, dependencies remain in the data, in part because WALS features with multiple possible values were expressed in the questionnaire as multiple questions with yes/no answers. This left a final dataset for the 14 languages of 90 features each, composed of 44 stable and 46 unstable features, including five cells of missing data.

Initial hypotheses are that stable features will more often replicate genealogical patterns than areal patterns in the data, and unstable features will more often replicate areal patterns in the data. This is a conservative view, equating geographic proximity with the probability of language convergence and shared structures.

3.2 Template vs. contents: patterns that connect

The next step was to categorize each feature as template or contents according to its function within a language. Template and contents classifications are comparable to familiar structuralist categories of syntagmatic and paradigmatic relations, respectively. The label “template” appeals to formal, constructional properties, at the level of the phrase or the word. These features describe the sequence of forms in a construction (constituent order, the location of affix or adposition) and the quality of the form as bound or free (affix, clitic, free form). Template features indicate where and in what form a feature is expressed. In contrast, “contents” features indicate which value from a set of values will fill a given formal position. These features appeal to choices and computations made by the speaker, often in response to factors of social cognition, group identity, and cultural rules and preferences. Contents features include those relating to the sound system; to the choice of pronoun, based on gender, politeness, or inclusivity; to choices of nominal marking, based on animacy, inalienability, or other classificatory quality; and to the choice of verbal marking, based on the grammatical role, number, or referent of the participant.

The categories of template and contents were operationalized in part from descriptions of structural change (HeathReference Heath1984; Zavala 2002; WinfordReference Winford2003; Heine and KutevaReference Heine and Kuteva2005; Matras and SakelReference Matras and Sakel2007), all of which draw on distinct feature types commonly known as “pattern” (the formal construction, corresponding to “template”) and “matter” (the morphophonological content). The key difference between “matter” and “contents” is that “contents” indexes a semantic component rather than a specific morphophonological form, identifying the meaningful distinctions encoded without examining the forms.

The decision-based nature of contents features suggests these are more likely to be shaped and enforced by interaction in specific sociocultural contexts, in effect, computed for each utterance; the initial hypothesis is that contents features will reflect areal relationships. The stencil-like nature of template features suggests two possible outcomes. On a psycholinguistic basis, “template” could describe a property that would be backgrounded and perhaps not easily accessible and manipulable by speakers. In this case, the feature might be relatively stable and reflect inheritance more than contact. On the other hand, in the description of cultural borrowing patterns in Section 2, Bray (Reference Bray, Lange and Stone1984: 337) was quoted in Section 2.1 as observing “The more neutral the trait, the wider its distribution and the greater its chances of acceptance.” If we connect the pattern of linguistic change to the pattern of cultural change, extending the dynamics of cross-linguistic priming to cross-modal priming of non-material culture, then template features may reflect a more general, regional level of contact, while contents features reflect the most local social groups. In this sense, the template and contents features could possibly both reflect areal patterns in a set of nested levels.

Table 4.3 describes the resulting database of 90 features and two types of categorization, with intersecting subsets (and see appendix).

Table 4.3 Numbers of features in major categories and their intersections

4.1 Linguistic distance by feature category and pre-defined group

In this subsection, linguistic distances from the matrix calculated by SplitsTree are compared by family grouping and by the regional groupings defined earlier. The graphics represent, for each feature type, the median distance (bar inside the box), the standard deviation spread around the average distance (top and bottom of box), and the maximum and minimum values in the data (whisker tips). The distribution in this dataset did not have extreme outliers, so the median is close to the average value for each type. The lower the linguistic distance, the more similar are the pairs of languages.

Figure 4.2 addresses genealogical relations, illustrating linguistic distance by feature subset for all Chibchan languages (CH) vs. all non-Chibchan languages (NCH), with 21 unique pairs in each calculation, noted as (21).

Figure 4.2 Linguistic distances for Chibchan vs. non-Chibchan languages

The first observation is that average linguistic distance for “all features” is nearly identical between the two groups (although the spread is larger in non-Chibchan languages), perhaps reflecting the known similarity of structural features across the region.

The results also suggest the importance of separating the simple category of stable features into different types of stable features to identify Chibchan family relationship: the best predictors of genealogy in this dataset are stable template features, while the stable contents features show a higher average distance and much greater spread. Within Chibchan, contents features as a whole and in subcategories consistently show slightly higher linguistic distances than unstable features. This suggests that a classification scheme of contents vs. template has more consistently identified properties of relative mutability among features than has the stable vs. unstable metric.

No particular category of features signals any type of relationship among the diverse languages in the non-Chibchan group. Stable template features in this group display an extremely wide spread, suggesting these features are quite poor indicators of a general areal relationship for this set of languages.

The representation of linguistic distances in Figure 4.2 argues that the stable vs. unstable opposition is not useful in delineating the Chibchan languages in the dataset. Template features, particularly the stable template features, predict genealogical rather than areal relationships, and contents features seem most sensitive to change within Chibchan but make no predictions about contact relationships in the non-Chibchan group.

Figure 4.3 Linguistic distances for pre-defined areal groups

Figure 4.3 addresses areal relations, illustrating linguistic distance by feature subset among pairs of languages in northern, Isthmian, southern, and eastern groups. Data numbers are very small: groups with four members have six unique pairs for comparison, shown as (6), and groups with three members have three unique pairs for comparison, (3).

This discussion of Figure 4.3 examines the graphic from left to right within each regional group. The common sense hypothesis is that unstable features will be better than stable as predictors of areal relationship. This means that linguistic distances should be smaller for unstable features than for stable features – and this is only true in the Eastern group. In all other groups, the average distance of stable features is smaller than the average distance of unstable. This contrary outcome actually might make sense for Isthmian, where all three neighbor languages are also Chibchan, but in general we can say that unstable features do not consistently predict areal relations better than stable features in this dataset.

The prediction for the next two categories is that contents features will show local area while template features may reflect a larger area or may reflect genealogy. Results here are quite mixed. Contents distances are smaller than template distances in Northern and (only slightly) in Southern groups, while this is not true in Isthmian and Eastern groups. The outcome is again striking in Isthmian, where the spread among features is quite small and the template distance is much lower than the contents distance. The results in Eastern are harder to interpret, as the spreads are similar and the numbers are so small. The summary statement for these categories is that neither type consistently predicts clear areal relationships in the dataset.

The perspective from the four multi-class categories suggests no common predictor of areal relation for the four pre-defined groups. The results here simply subdivide the unexpected outcome of the stable vs. unstable features described above, with stable features in three of the four regions showing smaller linguistic distances, which could be interpreted as the feature sets most useful in delimiting each areal group. These are stable contents for Northern, stable template for Isthmian, and either of those for Southern. In the Eastern group, unstable contents and unstable template features show the smaller distances.

4.2 Geographic distance by feature category

This subsection departs from the linguistic distance matrix calculated by Splits-Tree and from pre-defined regional groups of languages. In this analysis, pairs of languages are compared using geographic distance to calculate a normalized median distance between languages that share values for each feature. There are 90 binary features, which yields 180 possible sets of shared values (pairs that share 1 and pairs that share 0, for each feature). Of the 180 possible sets, 162 were shared by at least one pair of languages. Distances between pairs of languages were calculated using point coordinates for language locations in the WALS database and an online calculator that used the haversine formula to calculate the shortest distance between two points over the earth's surface (the great-circle distance).

Normalized median distance was calculated as the median distance between all pairs that shared the value of a given feature divided by the median distance of all pairs with a value for that feature (see appendix). If all languages shared a value for a given feature, as was the case for example with feature S1 = 0, then the normalized median distance was 1. If no languages or only 1 language had a specific value for a feature, as with S1 = 1, then there was no pair and no median distance. Features S33, S34, U27, U34, and U35 contain missing data for one language each, and therefore the total number of languages with values is 13.

In all cases, the lower the normalized median distance, the closer the areal relation among the languages that share the value for the feature. Figure 4.4 illustrates the plotting of feature category by the number of languages that share values and the normalized median distance between pairs.³

Figure 4.4 Geographic distribution of feature values

The general impression is of a rather homogeneous distribution of feature types, with a possible cluster of unstable contents features between 0.8 and 1.0 normalized median distance. The spread is characterized in Table 4.4, which shows counts by feature type for cumulative ranges of the 162 features ranked by normalized median distance. The distribution of stable and unstable features in each range is strikingly balanced, which contradicts the hypothesis that unstable features should show clearer areal affinities than stable features. The highest proportion of unstable features (%U, at 55%) occurs at either end of the scale, where 11 of the 20 lowest and 11 of the 20 highest normalized median distances are between shared values of unstable features.

Table 4.4 Ranges of features and types ranked by normalized median distance (N = 162)

The proportions of template and contents feature types show a bit more imbalance, with the percentage of contents features in the low 60% bracket for most of the lower normalized distances. The highest percentage of contents features (%C, at 64%) occurs in the range of the 50 lowest normalized median distances. The final column of Table 4.4 shows the breakdown of stable and unstable secondary categorizations within the contents category, and indeed stable features outnumber unstable features throughout the range of lower rankings. These counts indicate that contents features and specifically stable contents features are best at capturing general areal relationships in this dataset.

4.3 Summary of quantitative assessments

The first subsection examined a matrix of linguistic distance among pairs of languages based on shared features, produced using SplitsTree4. Hypotheses of relationship among the languages were investigated in Section 4.1 with representations of linguistic distances by feature type within pre-defined groups. Here we saw that all stable features are not alike: while the comparison of all stable and unstable features was inconclusive in distinguishing the Chibchan family, the subset of stable template features was best at identifying the genealogical group. The perspective from linguistic distances in the four small regional groups was less fruitful, suggesting in fact that diverse categories of stable features were slightly better at characterizing individual groups. Hypotheses on the roles of unstable and/or contents features in reflecting local areas were not confirmed; instead, the role of stable template features as predictors of family relations was given a small measure of support in the results for the Isthmian group, composed of three Chibchan languages.

The second subsection took a more general approach to the question of areal relations, calculating a normalized median geographic distance between pairs of languages that shared values for each feature. The overall picture suggested a rather homogeneous distribution of feature type. However, a closer look at proportions of feature categories in graded ranges of distance revealed a small but consistent predominance of contents features among the lower distances, with more stable contents than unstable contents, in a trend that peaked and then faded as the distances increased. The consistently uniform distribution of stable and unstable features at every range of distance was surprising; this picture may reflect the negotiated nature of the ranking from which features and labels were drawn, and it certainly reflects the complex character of structural stability. The prevalence of stable contents over unstable contents features as predictors of areal relation was also surprising, or counterintuitive, as one might have logically identified unstable contents as the category of features most susceptible to change. This outcome invites further investigation of the impact of social constraints on linguistic change.

3.1 Phonology and morphophonology

Dixon and Aikhenvald (Reference Dixon and Aikhenvald1999) list the following phonological features, which are explicitly marked as being absent or having different values in the Andean area.

1. 1. one liquid phoneme, frequently a flap

2. 2. affricates outnumber fricatives

3. 3. presence of a high, unrounded central vowel

4. 4. presence of mid vowels

5. 5. contrastive nasalization of vowels

Andean languages, according to Dixon and Aikhenvald, typically have more than one liquid phoneme and a preference for fricatives over affricates in terms of numbers of phonemes. The high unrounded central vowel is mentioned by Torero (Reference Torero2002) as an Andean characteristic with limited extension, as it occurs in Mapudungun (central Chile) and the extinct northern Peruvian coastal language Mochica. He furthermore mentions that it is possibly reconstructable for Puquina, also extinct, which was spoken around Lake Titicaca at the present-day Bolivian–Peruvian border. It should be borne in mind that the range of the Andean area that Torero talked about has a wider extension than the area talked about in this chapter, as Torero's Andean area extended all the way down to the southern cone and included also the formerly spoken coastal languages. Since Mapudungun and Mochica fall outside the part of the Andes immediately adjacent to the foothill-fringe area, and because the high mid vowel is present in members of the most dominant western Amazonian families (Arawakan, Tupí-Guaranian, and Panoan), I take it up in the list of Amazonian features for this chapter. With respect to the mid vowels /e/ and /o/, there are also a number of Andean languages that have mid vowels as phonemes (see Torero Reference Torero2002: 524; Adelaar Reference Adelaar, Lubotsky, Schaeken and Wiedenhof2008a: 26), but the two most dominant language families of the Andes, Quechuan and Aymaran, have three-vowel systems containing only high and low vowels. Nevertheless, this feature should be considered with care, since Adelaar (Reference Adelaar, Lubotsky, Schaeken and Wiedenhof2008a) reports that some variants of Quechuan and Aymaran have developed phonemic mid vowels, possibly due to Spanish and Portuguese influence. Vowel nasalization is decidedly Amazonian, and does not occur in Andean languages.

Payne (Reference Payne, Dixon and Aikhenvald2001) adds a sixth morphophonological feature to the Amazonian list, nasal spreading, noting that the Tupí-Guaranian, Tucanoan, Jê, Panoan, and Makú families all show some form of this characteristic.

1. 6. nasal spread

Moving on to the Andean literature, Torero (Reference Torero2002) distinguishes between a number of levels of feature diffusion (general – wide extension – limited extension – restricted). In the questionnaire I consider the first two groups, plus a subset of the features with limited extension, to the extent that they are found in languages or language families that cover major parts of the Andes along the foothill region as it is considered here. These, however, will be used with some caution, and especially to shed more light on subareas within the general region.¹

General and widely extended

1. 1. presence of a palatalized nasal

2. 2. frequently closed syllables

3. 3. velar–uvular opposition for voiceless stops

4. 4. presence of retroflex affricate

Limited extension

1. 5. glottalization of stops (some Quechuan, Aymaran, Uru-Chipaya)

2. 6. aspiration of stops (some Quechuan, Aymaran, Uru-Chipaya)

3. 7. three-vowel system (Quechuan, Aymaran)

The palatal nasal, the velar–uvular distinction, and the retroflex affricate are mentioned as traits that distinguish the Andes from the Amazonian area (Torero Reference Torero2002: 523–524). Glottalization and aspiration of obstruents in Quechuan languages is limited to those languages that are situated in southern Peru and Bolivia. This is very probably an Aymaran substrate feature (see e.g. discussion in Büttner Reference Büttner1983 and Adelaar Reference Adelaar, Campbell and Grondona2012b). The three-vowel system consisting of phonemes /u/, /i/, and /a/ is also in particular a feature of Quechuan and Aymaran languages (although perhaps not historically – Adelaar Reference Adelaar, Campbell and Grondona2012b), and is rare in Amazonia.

Andean feature 2 requires a few extra remarks, first of all because it is not an entirely straightforward feature with respect to the Andean area, and second because it must be translated into a question to which an answer can be given in terms of discrete categories. Adelaar (Reference Adelaar, Campbell and Grondona2012b: 601–602) mentions that neither proto-Quechua nor proto-Aymara allowed complex codas in underlying form, but since Aymaran morphophonology contains complex deletion rules of phonetic material, surface forms can contain highly complex consonant clusters. Moreover, Adelaar mentions that proto-Aymara may have been more restrictive in terms of the kinds of elements allowed in the coda, although modern Aymaran languages seem to have acquired greater coda tolerance, possibly as a result of contact with Quechuan languages (see Cerrón-Palomino Reference Cerrón-Palomino2008: 47). In addition, many Amazonian languages do allow a few consonants in the coda (usually nasals or fricatives), but tend to have more severe restrictions on what can be present in the coda. Therefore, rather than looking at abstract syllable structure, I analyze the issue of closed syllables as the degree to which restrictions are placed on segments in the coda of a syllable – not counting phonologically deviating words like ideophones, interjections, etc. and looking at underlying syllable structure.² The answer to this question can be based on the percentage of phoneme consonants that can occur in coda position, ranging from 0 to 100, divided into three groups: A: 0–30, B: 31–60, C: 61–100. More Andean-type syllable structures will fall into categories B and C, with Amazonian-types in category A.

Since Andean characteristic 7 inherently contrasts with Amazonian characteristics 3 and 4, they can be collapsed. This leaves a total of twelve contrastive Andean and Amazonian phonological features for analysis: ten general features plus two which are more restricted (Table 5.3).

Table 5.3 The phonological features

3.2 Morphosyntax

Both Andean and Amazonian languages are by-and-large characterized by having verbs with a highly synthetic, agglutinating morphological structure. Although this is a salient feature, it is not contrastive. Nevertheless, a number of contrasting features can still be listed on the basis of the proposals by the different scholars.

The status of argument cross-referencing on the verb is unclear, since where Derbyshire and Pullum (Reference Derbyshire, Derbyshire and Pullum1986) claim that the tendency for Amazonian languages is to have a set of pronominal affixes for both subject and object participants, Dixon and Aikhenvald (Reference Dixon and Aikhenvald1999) claim that it is typically Amazonian to cross-reference only one core argument on the verb (which may differ according to context). Andean languages often cross-reference both subjects and objects on the verb, so this is potentially a contrastive feature. However, the three families with a large western Amazonian presence (Arawakan, Tupí-Guaranian, Panoan) differ with respect to this parameter. Whereas Arawakan languages usually have cross-reference markers on the verb for both subject and object, Tupí-Guaranian languages conform to Dixon and Aikhenvald's prototypical Amazonian situation in that they mark one core argument on the verb, and Panoan languages, finally, have “no, incipient, or little developed argument marking in the main verb or auxiliary” (Valenzuela Reference Valenzuela, Gildea and Queixalós2010: 68).

What is striking, however, is the number of Amazonian languages that have pronominal prefixes (see Payne, D. Reference Payne and Payne1990: 221). This may be part of a more general difference between Amazonian and Andean languages, in that large-scale Andean languages like Quechuan and Aymaran are exclusively suffixing, whereas in Amazonian languages, prefixing is more common and is present in almost all languages to different degrees (see e.g. Payne, D. Reference Payne and Payne1990; Dixon and Aikhenvald Reference Dixon and Aikhenvald1999: 9; Torero Reference Torero2002: 526).

Another opposing feature to do with person markers is the fact that isomorphism between possessors and one of the core arguments is common in Amazonia (Dixon and Aikhenvald Reference Dixon and Aikhenvald1999), and rather rare in the Andes. In Torero's data, this is limited to foothill languages Cholón and Cunza. The isomorphism feature can be extended to languages that do not have bound person markers by taking into account isomorphism on the basis of form parameters such as case marking or special forms of pronouns. I am more wary of basing isomorphism solely on positional encoding, since the possible variation is too limited. Therefore, languages that treat possessive and argument pronouns as the same only in terms of their position with respect to their head are counted as “non-applicable.”

In addition to verbal cross-referencing, many Andean languages employ rich case systems (Torero Reference Torero2002: 527), including core case markers, whereas Amazonian languages tend to have elaborate applicative systems (Payne, D. Reference Payne and Payne1990), and a rather small set of peripheral case markers (Dixon and Aikhenvald Reference Dixon and Aikhenvald1999: 8). This characteristic is hard to quantify, since it is difficult to tell what is a rich system and what is a restricted system. Iggesen (Reference Iggesen, Dryer and Haspelmath2012) classifies languages into nine categories. I will distinguish three categories in a less refined way: (A) small set of case markers or no case marking (0–4), (B) medium set of case markers (5–6), and (C) large set of case markers (>6), where the typical Amazonian profile is “small set of case markers” and the typical Andean profile “large set of case markers.” Aymaran and Quechuan languages moreover have accusative case markers. Core case markers are un-Amazonian, with the exception of Panoan languages, which often have an ergative marker.

Finally, an often mentioned trait of Amazonian languages is their tendency to have ergative alignment, or alignment systems with clear and substantial ergative elements (e.g. Derbyshire Reference Derbyshire1987). Although the range of the systems encountered in Amazonia is rather great and involves various types of split systems, fully accusative systems appear to be very rare in Amazonia (Dixon and Aikhenvald Reference Dixon and Aikhenvald1999: 8), so this feature can be contrasted with the Andean languages, such as Quechuan and Aymaran languages, as well as Barbacoan languages, Cunza, and Huarpe, which have accusative systems (Torero Reference Torero2002: 529). Encoding strategies I consider are constituent order, verbal cross-referencing, and case marking. For a language to be coded as accusative, at least one of these three must follow an accusative pattern, and the others cannot give a contrastive signal. I particularly look at NPs in simple clauses, and do not count as accusative any system that has a major alignment split (e.g. based on definiteness, semantic role, etc.).

In the nominal realm, possessive constructions can be contrasted. The typical Amazonian structure involves a head-marked construction, making use of bound person markers (see e.g. Dixon and Aikhenvald Reference Dixon and Aikhenvald1999: 8). The Andean type often involves dependent marking, sometimes in combination with head marking (Quechuan, Aymaran – see Torero Reference Torero2002; Adelaar Reference Adelaar, Campbell and Grondona2012b), sometimes not (Mochica, Huarpe, Barbacoan – Torero Reference Torero2002: 528). Puquina and Mapudungun both have Amazonian-type possessive structures in that they mark possessive relations on the head by means of person prefixes. Nevertheless, it is reasonable to contrast this feature in terms of Andean versus Amazonian in that the former tend to have dependent-marking strategies, and the latter not.

Another salient Amazonian feature is the presence of a noun class or gender system of some sort. Noun class systems are also encountered in some of the languages that Torero counts as being part of the Andean area (Mochica and Cholón), but as mentioned Mochica is a coastal language (on the west side of the Andes) and Cholón is considered here to be part of the FF area.

In terms of negation, Andean languages display several different strategies: a preposed particle, a suffix, or a combination of those. Torero (Reference Torero2002: 528–529) mentions that the first two strategies are also common in Amazonian languages, especially suffixal negation. So this feature is not contrastive enough to take up in the questionnaire.

Apart from the aforementioned subject and object cross-referencing and negation marking, a number of further traits are encountered to different degrees both in the Andean and the Amazonian area, and are therefore not contrastive: evidentiality, nominalized subordinate clauses, switch reference, phrase- or sentence-final particles or enclitics, inclusive–exclusive distinction, alienability, incorporation, and lack of passive. The above considerations leave us with a further seven contrastive morphosyntactic characteristics (Table 5.4).

Table 5.4 The morphosyntactic features

3.3 Constituent order

Especially Derbyshire and Pullum (Reference Derbyshire, Derbyshire and Pullum1986) and Derbyshire (Reference Derbyshire1987) give close attention to issues of constituent order. Among the Amazonian constituent order traits, they include O before S (Derbyshire and Pullum Reference Derbyshire, Derbyshire and Pullum1986) or O-initial (Derbyshire Reference Derbyshire1987) constituent order in the sentence and the combination of NA, Pr-Pd orders and postpositions. Torero (Reference Torero2002), on the other hand, mentions SOV clause order as an Andean trait with limited but still wide extension (Quechuan, Aymaran, Chipaya, and also true for Barbacoan languages – see Curnow and Liddicoat Reference Curnow and Liddicoat1998: 387), and AN and Pr-Pd orders as widely shared features. This results in two contrastive features (Table 5.5).

Table 5.5 The constituent order features

I have chosen the formulation of feature 20 as noted in Table 5.5 because asking for SOV order would have been too restrictive on the Andean-like languages, and asking for O-initial, too restrictive for Amazonian languages (in the sense that “no” as an answer would encompass many more logical possibilities). I will give more detailed information on word order below.

3.4 Lexicon

A final domain for which proposals have been made on the basis of which we can contrast an Amazonian profile with an Andean profile is the lexicon. One salient feature for Andean languages is a decimal counting system, shared by many of the languages and language families: Quechuan, Aymaran,³Puquina, Mochica, Cholón, Uru-Chipaya, Cunza, Huarpe, and Mapuche. Although we cannot contrast this trait as such with the Amazonian type of numeral systems, Dixon and Aikhenvald (Reference Dixon and Aikhenvald1999: 9) mention that there is generally only a small class of numerals in Amazonian languages. This means that we can set up an Andean–Amazonian contrast on the basis of elaboration of the numeral system, where a stable numeral system that goes to at least 10 (and that does not contain Spanish or Portuguese loans) is typically Andean, whereas smaller systems are typically Amazonian.

A final lexical characteristic of Amazonian languages is mentioned by Payne (Reference Payne, Dixon and Aikhenvald2001): the presence of an elaborate class of ideophones. Ideophones can be defined as “marked words that depict sensory imagery” (Dingemanse Reference Dingemanse2011: 25), i.e. they are words that typically show deviating characteristics, especially in their phonology and phonetics but often also in their morphological and/or syntactic behavior, that depict a situation in such a way that it evokes a perceptual sensation or perceptual knowledge. This goes well beyond the arguably universal onomatopoeia, as ideophones can depict at higher levels of abstraction, often involving perceptual modalities other than hearing, such as vision, taste, smell, touch, etc.

Table 5.6 completes the list of twenty-three contrastive features.

Table 5.6 The lexicon features

In the comparisons that are discussed in the next sections, I score the features for each of the thirty-two languages in the sample if the available data allow for it. By regarding the Andean profile and the Amazonian profile as “language” profiles, on a par with the profiles of the FF languages, I can calibrate the distance of an FF language to the Andean and Amazonian type.

4.1 Linguistic distance

Figure 5.1 represents the distance between languages by taking into account all of the twenty-three features discussed above, without any differences in weight for the features. The Andean and Amazonian profiles (which are maximally contrastive) are treated as if they were languages; they are boxed in the network. The distances between the languages are visualized in a Neighbor-Net network (Bryant and Moulton Reference Bryant and Moulton2004), a distance-based method that shows splits between languages, but also signals that go against proposed splits in the form of reticulation or ‘webbing’.

Figure 5.1 NeighborNet representation of the distances between the languages of the sample

A first major split we can observe is indicated by the vertical thick dotted line in Figure 5.1. The group of languages above the dotted line contains all the Arawakan languages as well as a few others, like Cholón, Muniche, Movima, and Záparo, and towards the right the isolate Yurakaré and Tupí-Guaranian language Yuki. The group below the dotted line contains the Quechuan, Tacanan, Panoan, and Jivaroan languages, as well as Aymara (Aymaran), Secoya (Tucanoan), Amarakaeri (Harakmbet), Cocama (Tupí-Guaranian) and the (semi-)isolates Leko, Waorani, Cofán, Mosetén Taushiro, and Urarina. For ease of reference, I will refer to the group above the dotted line as “Amazonian” and to the group below the dotted line as “Andean.”

If we contrast these two blocks, the binary features that contribute most to the contrast between them are (ordered according to contrast, the highest contrast first):

1. 1. presence of core case markers (0% of the ‘Amazonian’ group versus 89% of the ‘Andean’ languages);

2. 2. isomorphism of possessor and core verbal argument (91% Amazonian, 32% Andean);

3. 3. dependent marking for possession (9% Amazonian, 68% Andean);

4. 4. the presence or absence of an elaborate case marking system (64% of Amazonian languages have a small case marker inventory vs. 5% of the Andean group, and 18% of the Amazonian group and 84% of the Andean group have an elaborate case marking system);⁴

5. 5. the presence of gender/classifier systems (73% Amazonian, 16% Andean);

6. 6. accusative alignment (27% Amazonian, 74% Andean).

Both groups of languages show moreover a secondary contrast between languages to the left of the graph and languages to the right. This contrast is much less clear and seems more reminiscent of a continuum, or perhaps a tripartite distinction, and is indicated by the two thin vertical lines. If we take the most contrastive languages on the left–right axis, we can again distinguish an “Andean” group on the left, consisting of Arawakan languages Ashéninka, Nanti, and Yanesha’, and Cholón (Hibito-Cholón) and the isolate Muniche, as well as the Quechuan languages Imbabura and Cuzco Quechua, the Tacanan languages Cavineña and Ese Ejja, Aymara, and the isolate Leko. To the right of the graph we can distinguish an “Amazonian” group consisting of Yurakaré (isolate), Tupí-Guaranian languages Yuki and Cocama, the three Jivaroan languages Aguaruna, Achuar, and Shuar, Panoan Shipibo and Cashibo, isolate Urarina, and Amarakaeri (Harakmbet).

For this axis the most contrastive features are the following:

1. 1. basic adjective-noun order (73% of Andean, 0% of Amazonian);

2. 2. phonemic central high vowel (18% Andean, 90% Amazonian);

3. 3. the presence of more than one liquid phoneme (82% Andean, 10% Amazonian);

4. 4. nasal spread (0% Andean, 70% Amazonian);

5. 5. phonemic vowel nasalization (0% Andean, 60% Amazonian);

6. 6. phonemic palatal nasal (100% Andean, 40% Amazonian).

While the rather clear top-to-bottom split is dominated by morphosyntactic features, the more diffuse left-to-right split seems to be particularly based on phonological features (except for the first). This may at least in part reflect the fact that the phonological features seem to be more sensitive to diffusion through contact, probably through the incorporation of loanwords. If we split the phonological features from the morphosyntactic features,⁵ we can observe that the distributions of Arawakan and to a lesser extent Quechuan languages (together with Aymara) are rather diffuse in the network based on the phonological features (Figure 5.2), and much closer together in the network based on morphosyntactic features (Figure 5.3). In Figure 5.3, all the Arawakan languages in the sample are in the left “tail” of the figure (with Movima and Muniche). The two Quechuan languages and Aymara are identical with respect to the morphosyntactic features, and converge completely on the Andean profile. In Figure 5.2 on the other hand, the Arawakan languages are spread all over the network, while Quechuan languages and Aymara are still rather close to each other, if not so close as in Figure 5.3.

Figure 5.2 NeighborNet of distances between languages of the sample (phonological features only)

Figure 5.3 NeighborNet of distances between languages of the sample (morphosyntactic features only)

Another interesting difference that can be observed is the fact that Panoan (Cashibo, Shipibo) and Tacanan (Cavineña, Ese Ejja) languages, which are often regarded as being related in a deep sense (see e.g. Key Reference Key1968; Girard Reference Girard1971), are rather close together in the morphosyntactic representation, compared to the phonological feature representation. On the whole, then, the morphosyntactic picture makes the impression of representing a more conservative, genealogical picture than the phonological one. This can possibly be connected to borrowing of linguistic forms (especially lexicon), thus introducing new phonemes to the recipient language. Since grammar is generally assumed to be more resistant to borrowing than the lexicon, we might hypothesize that Figure 5.2 may be read as indicating patterns of (shallow) language contact, whereas Figure 5.3 may be reflecting either genealogical links or deep/intense contact.

In terms of features, the same major contributing factors that were identified for the top–bottom divide in Figure 5.1 are responsible for the main divide in Figure 5.3, which contrasts the same two groups of languages. The main contributing features of Figure 5.2 are nasal spread (79% for the Amazonian side, 0% for the Andean side), the presence of more than one liquid phoneme (5% for the Amazonian side, 79% for the Andean side), and the presence of a phonemic palatal nasal (32% for the Amazonian side, 100% for the Andean side).

4.2 Correlations with geographic factors

This chapter focuses on linguistic issues in the discussion on the foothill-fringe area. I will here touch on some possible geographic correlates, but it is clear that more in-depth research is necessary to give more detailed and definite answers to these matters.

The Andean side of the divide in Figure 5.2 suggests contact between Andean (Quechuan and Aymaran) languages and some of the languages spoken close to the Andes in northern Bolivia (Leko, Cavineña) and Peru (Nanti, Ashéninka, Cholón). From there towards the right of the graph the situation becomes more diffuse. There are a few probable contact pairs (Cofán and Secoya, perhaps Urarina with the Panoan languages – although there is quite a lot of reticulation, including the biggest divide in the graph), but there are more surprising positions: Záparo, Mosetén, and Amarakaeri, whose closest neighbors in the network are not their closest neighbors geographically. On the whole, then, the phonological graph seems to represent a rather specific Andean profile with some possible points of contact-induced change, and a much more diffuse Amazonian zone, where languages may share some traits but not others. The most widespread features in the Amazonian group are the presence of mid vowels (also quite common on the Andean side of Figure 5.2) and nasal spread (both 79 percent).

One clear exception to the pattern that phonology is less stable than morphosyntax is the fact that the Jivaroan languages are distributed much more diffusely in the morphosyntactic picture than in the phonological one. There is no straightforward explanation for this apparent anomaly. A suggestion towards an explanation may come on the one hand from the long-term and complex relations between Jivaroan groups and highland groups,⁶ and on the other hand because the Jivaroan groups “show a particularly strong ethnic consciousness” (Adelaar with Muysken Reference Adelaar, Adelaar and Muysken2004: 432). The first factor may contribute to the result that Achuar is found rather close to the Andean profile, and the latter may account for the fact that the Jivaroan languages pattern so closely phonologically, perhaps due to an ethnic consciousness that includes a resistance to lexical (conscious) borrowing.

Figure 5.3 shows a relatively homogeneous “Amazonian” group, containing all Arawakan languages, but also the isolates Movima and Muniche and, at more distance, Cholón and Záparo. The Movima case may be explained by (deep) contact of Movima with Trinitario/Ignaciano and also Baure. Other cases, such as the puzzling position in the midst of the Arawakan languages in the network of Muniche, Záparo, and Cholón, may require less straightforward explanations.⁷

Given the particular position of the languages in the sample, between two major geographical (and perhaps cultural) zones, a natural question to ask is whether we can find any correlations⁸ between the linguistic patterns and geographic variables. A first, simple question would be whether there is any correlation between linguistic distance and geographic distance.⁹ It seems that there is no such correlation, as can be observed in Figure 5.4, which shows geographic distance on the x-axis and linguistic distance on the y-axis. This confirms the observations made (i) that there are a few languages oddly placed in the graph, and (ii) that genealogical signals can be strong without necessarily coinciding with geographic proximity.

Figure 5.4 Correlation between linguistic distance and geographic distance

Two other geographic factors that seem intuitively important are elevation and river systems, as they are of consequence to how people travel, and perhaps they limit the range of contacts between peoples. Figure 5.5 shows the correlation between geographic elevation and linguistic distance. It should be read as follows: the greater the difference in elevation between a language pair, the greater the linguistic difference between these languages. In very general terms, this can be interpreted as identifying the Andean mountains as a barrier for contact, although the correlation is not very strong (r = 0.37).

Figure 5.5 Correlation between geographic elevation and linguistic distance

The idea that differences in elevations are barriers for contact is corroborated by the graph indicated in Figure 5.6, which is the correlation between elevation and proximity to the Andean profile. The x-axis indicates the height of the location where the languages are spoken, and the y-axis indicates distance from the Andean profile. As a general tendency, the higher a language is spoken, the more it conforms to the Andean profile.

Figure 5.6 Correlation between elevation and proximity to the Andean profile

A final geographic factor that I want to take into consideration is the river system. Rivers in South America form pathways along which people move around, are in contact with each other, and thus possibly influence each other. The foothill-fringe area as presented here can be said to consist of three major river systems:

1. 1. The northern basin, delimited in the north by the lower Napo and Aguarico Rivers and in the south by the lower Ucayali and Marañon Rivers, encompassing eastern Ecuador and northern Peru.

2. 2. The drainage basin of the upper Ucayali and Huallaga Rivers, covering north-central to southern Peru.

3. 3. The basin defined by the Madre de Dios and Mamoré Rivers, covering Bolivia and a small part of southern Peru.

The languages in the sample can be classified according to the river system they belong to (see above), and the average distance between their linguistic profiles can be compared to the average of the entire sample. However, it seems that the river systems do not have any impact on the average linguistic distance, as is shown in Table 5.7.

Table 5.7 The average linguistic distance per river system

One proviso that we should make with this result is that the genealogical diversity is greater in the Napo/Aguarico (ten families) and Madre de Dios/Mamoré basins (nine families) than in the Ucayali/Marañon basin (four families). This means that perhaps the picture should be adjusted somewhat and there might be a (relative) contact effect after all in the northern and southern basins. However, this is difficult to take into the equation, and must, moreover, await more detailed research.

4.1 QI and QII

It is clear that there are important differences between QI and QII varieties (Parker Reference Parker1963; Torero Reference Torero1964; Cerrón-Palomino 1987; Adelaar with Muysken Reference Adelaar, Adelaar and Muysken2004). It is also generally agreed upon that the split between them took place quite early, possibly longer than 1,500 years ago. The main question is whether this led to a sharp division, as assumed by the sources listed above, or a more gradual one, as argued by Heggarty (Reference Heggarty2007: 335). Summarizing the NeighborNet analysis of lexical data reported on in Heggarty (Reference Heggarty2005) and on a reanalysis of the data presented in Torero (Reference Torero and Escobar1972), he writes:

Inspection of the graphs presented in Heggarty suggests that the sources of contention concern three datasets:

1. (a) The northern varieties of Cajamarca and Ferreñafe (Torero) and Cajamarca and Inkawasi (Heggarty) do not cluster with the other QII varieties, but either with Central QI (Torero) or as a separate branch (Heggarty).

2. (b) The central variety of Pacaraos that was classified as QII by Torero (Reference Torero1964) is actually closer to QI in Torero's (Reference Torero and Escobar1972) data.

3. (c) The varieties labeled Yauyos are intermediate between QI and QII in both datasets.

Our data involve systematic morphosyntactic datasets that can help clarify these issues. The main question will be: Is the split between QI and QII so deep that we can speak of genealogical units or do we see a process of dialectalization in a network-like form with numerous early split-offs? We will focus on the position of Cajamarca, Pacaraos, and Yauyos.

4.2 Quantitative results for the Quechua languages

The primary technique we used involves distance matrices, representations of the percentage of shared features between any varieties in the sample.²

Applied to the three datasets we have, Noun Phrase, Argument expression, and Quechua 37 (the combined questions of Quechua and Andean), we observe that the internal variation within the Quechuan family is low when the analysis is based either on the Noun Phrase (GAD 0.11) or the Argument expression (GAD 0.12) questionnaire. The Noun Phrase feature database yields very few differences between the varieties (distances ranging from 0.05 to 0.20), suggesting that these features are relatively stable across the family. The Argument realization feature matrix results in a low distance on the average, but somewhat greater internal variation (range 0.05–0.30). The NeighborNet graphs associated with these figures present the dialects as relatively randomly grouped. This may be interpreted as an indication of the fact that the basic morphosyntactic layout of Quechuan languages has remained amazingly stable, assuming that geographic split and subsequent dialect formation started 1,500 years ago.

It may not come as a surprise that the global average distance jumps considerably to 0.41 (range 0.15–0.70) with the 37 combined Quechua and Andean features. The associated NeighborNet strongly contrasts the QI, QIIb, and QIIc dialects and the effect of these features is so strong that even if they are lumped together with the Noun Phrase and Argument expression features, the global average of this combination (Quechuaall) falls to 0.18 (range 0.05–0.30), but the associated NeighborNet gives the same clear picture: see Figure 6.1.

Figure 6.1 NeighborNet representation for the relative distances of the members of the Quechuan language family

The main branching between QII languages and QI languages is confirmed. The problematic variety of Pacaraos is close to QI, and problematic Cajamarca and Yauyos are close to QII. Ecuadorian Quechua, a sub-group within Quechua II, is always the outlier, with other lowland Quechua languages such as Peruvian Pastaza Quechua and Inga (Colombia) close by.

A first conclusion may be that the forms of morphemes may have changed (the 25 features in the Quechua database) but not the typological frame (Noun Phrase and Argument expression databases). Interestingly, variation is not random but clearly supports the classical division of Quechuan into QI and QII. In that way it provides answers to a number of questions that were formulated above. The split between the Quechua I, including Pacaraos, and the Quechua II dialects is deep and not likely the result of a slow dialectal spread. The northern Peruvian dialects Cajamarca and San Martin, and Yauyos Quechua as well, form a branch of the Quechua II cluster. This may be interpreted as support for the view that they are the result of early Huari expansion, as we will argue below. The loss of morphological complexity observed in the Ecuadorian dialects sets them clearly apart as a subgroup within QII. It is noteworthy that Kallawaya groups with them in this respect.

Kallawaya does not resemble the surrounding Southern Quechua varieties, in contrast with most remarks in the literature, namely that Kallawaya would simply be a relexified variety of Bolivian QIIc. Its intermediate position next to the Ecuadorian Quechua varieties probably results from processes of simplification in the Quechua varieties that were the input to Kallawaya, similar to what happened in Ecuador.

4.3 The QII cluster and Ecuadorian Quechua

The QII varieties of the southern branch (spread from Ayacucho in the Peruvian highlands to northern Argentina) form a genealogical unit together with Cajamarca and Yauyos. However, the distance matrices and the NeighborNet analysis in 4.2 suggest that northern QII languages (including Ecuador and some speakers in southern Colombia) form a group by themselves. The key question concerns the relation between the Peruvian QII varieties and those exported to Ecuador. The varieties of Ecuador have been linked to a lingua franca Quechuan form called lengua general, assumed to have been the Inca imperial expansion variety.

The morphological features of Ecuadorian Quechua, or Quichua, show that it is an off-shoot of QII varieties. It is related to early Chinchay (Torero Reference Torero1975) or “general Quechua” and has a few features of Cuzco Quechua as well. It was introduced into Ecuador in the Incaic period (see also the testimony of Cieza de León Reference Cieza de León1984 [1553]), and consolidated during the colonial period. Hocquenchem (Reference Hocquenchem, Heggarty and Beresford-Jones2012) argues convincingly that all ethnohistorical and archaeological evidence points to expansion of Quechua into Ecuador during the Inca conquests after 1450 CE. We know of no compelling linguistic arguments which would support an earlier expansion.

Quichua differs considerably from the Quechuan languages it is related to, such as Ayacucho Quechua (Muysken Reference Muysken1977; Reference Muysken and Arends2000b). It was consolidated as the community language of the runa peasants of the Ecuadorian highlands during the Incaic period, and particularly the Colonial period. Prior to the advent and genesis of Quichua, other substrate languages were spoken in Ecuador, notably Barbacoan languages in the north and Jivaroan languages in the south. There is some variation in Quichua that may be attributed to different substrates. Table 6.2 presents some of the main differences between Quichua and its Peruvian relatives.

Table 6.2 Morphosyntactic and phonological features that distinguish Southern Peruvian Quechua from Ecuadorian Quichua

The differences are striking particularly in the nominal domain. Compare verbal constructions (1a) and (1b), where the basic person indexing is maintained, to (2a) and (2b). Ecuador has lost nominal person indexing:

(1)

(2)

The consequences of this loss are also crucial in the nominalization and hence subordination domain; cf. the contrast between (3a) and (3b):

(3)

However, not all Amazonian varieties have lost person marking on nominals, in nominalizations, and in adverbial clauses; cf. the following examples from Waters (Reference Waters and Parker1996: 167–169), of equivalents in (a) Pastaza (Peruvian Amazonian Quechua) and (b) Napo (Ecuadorian Amazonian):

(4)

(5)

(6)

Thus the changes in the northern varieties (including Peruvian Amazonian), separating them structurally from all Peruvian Quechua varieties, cannot be uniquely attributed to the loss of nominal person marking.

4.4 Origin and history

Given the results from the NeighborNet analysis and the earlier literature, the scenario which we consider most likely for the origin and spread of Quechua contains the following elements.

Origin. As argued in Adelaar (Reference Adelaar, Campbell and Grondona2012a) and Muysken (Reference Muysken, Heggarty and Beresford-Jones2012a) Quechua emerged through interaction with Aymara. The precise details of the interaction are a matter of dispute and need further investigation. The overall typological profile of what we may think of as very early Quechua suggests affinities with languages in the north central Andes such as the Barbacoan and Jivaroan family. This reflects a possible northern origin for Quechuan.

Initial spread. The evidence regarding internal variation in the Quechua language family points to an initial dispersal from the central Peruvian highlands. First, varieties which later gave rise to QII moved towards the south (to roughly the Ayacucho area, associated with the Huari civilization). The original area remained highly differentiated, with some varieties later emerging as the current QI languages (including Pacaraos).

The Huari civilization. Beresford-Jones and Heggarty (Reference Beresford-Jones, Heggarty, Heggarty and Beresford-Jones2012b: 57–84) argue that the Huari horizon is associated with the expansion of Quechua.⁴ We agree with this and assume that later consolidation and spread of QII varieties was linked to the Huari civilization. In the Huari period terrace construction gained momentum, increasing the potential for maize cultivation through stone heat retention, but it requires substantial labor and thus population density. Huari settlements were very grid-like and regular in shape. Huari expansion was achieved through military rule rather than through extensive movement of people. We follow Adelaar (Reference Adelaar, Campbell and Grondona2012b) in the idea that Cajamarca and related Quechua varieties were an off-shoot from the Huari civilization.

The Inca Empire. The later spread of Quechua in the Inca period, both northward into Ecuador and southward into Argentina, can be linked to actual population movements. It is not evident that the Inca actually imposed Quechua on their subjects. There was very extensive movement of people, involving various types of subjects, during Inca rule, and in terms of numbers the mitmaqkuna were the most important. Mitmaqkuna were extended families or ethnic groups resettled by the Incas by force from their home territory to recently conquered areas.⁵ The Inca occupation of northern Argentina took place in the middle of the fifteenth century.

The Spanish occupation. According to Cook (Reference Cook1998) the Andean population was 9 million at the time of conquest, and this went down to 600,000 in 1620, an incredible devastation. The figure of 370,000 can be given for Peru in 1730, which increased to 1.5 million in 1876, and 2.9 million in 1940. At present the population is somewhere near its size at the time of conquest. Thus the current distribution of languages does not necessarily represent the original situation. Quechua and Aymara continued to spread into the foothill regions of Ecuador, Peru, and Bolivia during the colonial period. An overview is given in Figure 6.2.

Figure 6.2 The distribution of languages per region, over time (Q = Quechua)

Whether the Chavín and Huari/Tiahuanacu horizons need to be linked to language spread (Beresford-Jones and Heggarty Reference Beresford-Jones, Heggarty, Heggarty and Beresford-Jones2012b) is an open question. The Chavín horizon is very early, 900–200 BCE, and is believed to have been limited in its expansion. It pre-dates the generally believed presence of both Quechua and Aymara in central Peru, which began around 200 CE. As such, Chavín may be linked to pre-proto-Quechua, as has been argued by Heggarty (p.c.). On the other hand, linking Chavín with pre-proto-Aymara, as is done in Beresford-Jones and Heggarty (Reference Beresford-Jones, Heggarty, Heggarty and Beresford-Jones2012b), is not incompatible with (1) the idea that Quechua was an invading language coming from the north, maybe filling the gap after the fall of the Chavín cultural complex (Adelaar Reference Adelaar, Campbell and Grondona2012a), and (2) the spread of Aymara down the coast into the Nazca area, where it became the language of the Nazca culture (100 BCE–700 CE). This may also have been the avenue for the attested presence of Aymara in the highland area around 1200 CE, which makes it plausible that Aymara was one of the languages spoken in the Tiahuanacu realm, in addition to Puquina and Uru. There is no evidence that Quechua was present in southern Peru until the end of the Tiahuanacu era, and this raises the question why by 1300 CE Quechua was chosen as the imperial language by the (supposedly Aymara- or Puquina-speaking) Inca elite. The most plausible reason is that large parts of the south central area of Peru were Quechua-speaking by that time, making it likely that during the earlier Huari horizon (500–900 CE) Quechua was spoken in this area, probably in addition to Aymara. This squares with the idea that Quechua- and Aymara-speaking groups, in a herding and agriculture symbiosis, moved southward from north central Peru into the Ayacucho area between 300 and 600 CE.

5.1 Aymaran

Apart from two very small and rather distinct older branches of the family in central Peru (Jaqaru and Cauqui), Aymara is spoken in a contiguous area in the south of Peru and northwest of Bolivia, with very little internal differentiation, as far as we know. Given the many similarities and limited structural distance between Quechua and Aymara, questions concerning the relation involve both ultimate relatedness and further contacts between the two families.

Ultimate relatedness. How similar are Quechua and Aymara, and what light does this shed on their genealogical relationship? By both proponents and opponents of such a relationship it is accepted that if they are related, the moment they split dates back to a period at least 2000 BCE if not much more. Adelaar with Muysken (Reference Adelaar, Adelaar and Muysken2004: 34ff.) argue that the evidence points to a separate origin for the two language families, and we have no reason to assume differently here. So, unless clear genealogical links of either one of the languages with languages spoken to the north would make a scenario of the movement of either language family to Central Peru plausible, for our current state of knowledge of the early history of the central Andes, this does not make much of a difference. However, from a linguistic point of view, it does since it is a kind of a test case up to which points languages may converge or split.

Aymaran, QI, and QII. Adelaar assumes that Quechua coexisted in a (pre-) proto-form with (pre)proto-Aymara speaking populations, to which this proto-Quechua adjusted its form. Much later this transformed Quechua began its spread during the Huari horizon, both to the north (to the central highlands where it became a superstrate on Aymara dialects and to the coast) and to the south east (Cuzco). However, this expansion can only have involved QII varieties. There are at least two points in which Aymara and QI strongly differ from QII: directionals in verbal derivation and verbal plural cross reference marking. The Aymaran system is richer than the QI system but both have directional and aspectual suffixes. QII manipulates a number of these suffixes but only one has retained a directional meaning, apart from an aspectual one like the others; -yku, -rqu, -rpa. This does not look like a real innovation but more like a slow loss of complexity. Plural marking is a parallel case. It is an innovation in the QII dialects to create an extra slot for number marking in the verbal matrix and to get rid of the fairly complex event plurality as it is marked in the Aymara and QI dialects. Since Cajamarca, San Martín, and Yauyos share this innovation we must assume that they were later split offs from Huari (Ayacucho) varieties rather than early remnants.

Substrate. Another intriguing aspect of the Quechua/Aymara contact is the fact that Aymara may have disappeared without leaving many traces when overlaid by Quechua. One of Adelaar's arguments for assuming much wider Aymara presence are a few Aymara lexical elements present in the north-central Quechua speaking area where Aymara for time immemorial has disappeared, but a much more recent case is the disappearance of Aymara in southern Bolivia where it was widely distributed well into the colonial times and where it disappeared without leaving noticeable traces, for example in the Northern Potosi area.

Has the spread of QII southward led to Aymaran substrate influence in more southern varieties such as Cuzco and Puno? Regarding Aymaran influence, it has often been assumed that the glottalization and aspiration of initial stops in the QII varieties found in Cuzco and Bolivia (where historical sources suggest the earlier presence of Aymara) may reflect an Aymaran substrate. Likewise, the use of separate lexical dependent clause markers in southern varieties of QII, such as chayqa ‘that’ and hina ‘like’ may possibly be linked to Aymaran influence. However, a systematic exploration of possible syntactic convergence of southern varieties has not yet been undertaken.

5.2 Uru

The Uru languages Chipaya, still spoken, and Uchumataqu, which survived until recently, were spoken in parts of the Altiplano and the Lake Titicaca basin, mostly on the Bolivian side (Adelaar with Muysken Reference Adelaar, Adelaar and Muysken2004: 362–363). There is no evidence of population movements for the group as a whole, although it is clear that the range of communities where Uru languages were spoken is much wider than the present, quite reduced aquatic zones along the lakes and rivers of the Altiplano. There is evidence of earlier structural influence of Quechua on the Uru languages, and of current Aymara influence (see e.g. Muysken Reference Muysken, van der Voort and van de Kerke2000a), suggesting the possibility of metatypy (Ross Reference Ross1999, Reference Ross and Brown2006). Nonetheless the Uru languages are clearly distinct from Quechua and Aymara structurally.

5.3 Cholón

Cholón was spoken until fairly recently in the upper Huallaga valley in northern Peru, and together with related Hibito may have occupied a larger area in the Andean foothills. Alexander-Bakkerus (Reference Alexander-Bakkerus2005) has documented a number of borrowings from Quechua, both lexical and morphological. In spite of this, Cholón appears as a clearly separate entity in the NeighborNet trees.

5.4 Mochica

Mochica (Muchik) or Yunga was traditionally associated with the Moche or Chimu culture in coastal northern Peru. Mochica is typologically quite distinct from the other languages in the area, and has been linked to the Mayan languages (Stark Reference Stark1972), without general agreement so far.⁶ Some of the “Mochica” were probably Muchik speakers, but there were multiple competing polities in Moche (Kaulicke Reference Kaulicke, Heggarty and Beresford-Jones2012). Cerrón-Palomino (1987) has shown multiple contacts between Mochica and Quechua. This may have led to structural convergence, but Mochica remains clearly distinct.

5.5 Arawakan, Puquina, and Kallawaya

Although it is still very much debated, for the sake of exposition we will operate on the assumption here that elements of what came to be known as Puquina came from an Arawakan language, with an early presence in the Altiplano. Relations between the Quechuan, Aymaran, and Arawakan language families are very complex and may span the last two thousand years. We may distinguish at least three stages in the interaction.

First of all, the possibility has been suggested that Arawakan peoples participated in the culture of Tiahuanacu (300–1000 CE) near Lake Titicaca. However, Tiahuanacu may not have been exclusively Arawakan. Archaeologists Eduardo Machicado, Paul Goldstein, and Sarah Baitzel argue that Tiahuanacu was heterarchical rather than hierarchical, as can be seen in the archaeological remains. There were, for example, different styles of cranial modification. In Tiahuanacu itself there were specialized workshops with different food consumption patterns, suggesting multiethnic constituency. Tiahuanacu expansion was through people, with non-contiguous settlements and several co-existing styles in diasporic enclaves, which suggest that settlements were themselves multi-ethnic.

Further north in the Andes, the presence of monkey imagery in Nazca cultural representations (the famous Nazca lines are dated 400–650 CE) may likewise be linked to Amazonian, possibly Arawakan, influence in coastal Peru. Monkeys were not present on the coast, but they appear in the lines in the desert as well as in pottery motifs.

It is clear that there were Arawakan words in Cuzco Quechua, like unu ‘water,’ as well as the term for month, -quiz. This suggests an early influence on Inca civilization from Arawakan societies. Arawakan words have also been found in the Chilean coastal language Mapuche, further to the south, and Bertonio's Aymara dictionary contains words from Puquina. Although it is clear that there was a strong Arawakan influence in the Andes starting at least 1 CE, this does not imply necessarily that there were large Arawakan-speaking populations. This is compatible with the idea that Puquina, identified above as one of the important languages in southern Peru and Bolivia in the pre-Inca and Inca periods, contains a substantial Arawakan component but is not a fully Arawakan language. Puquina is most likely one of the important languages of Tiahuanacu.

Rodolfo Cerrón Palomino argues that Puquina was the early language of the Incas. His evidence for this comes from the names of the first members of the dynasty, from the names for the rituals in Inca civilization, and from the practice of sun worship. Cerrón Palomino also assumes that the term Colla, now used for the Altiplano Aymaran populations, originally referred to Puquina.

Cesar Itier's work on Amarete Quechua in northern Bolivia may reveal still more traces of Puquina vocabulary. In the colonial period there were intensive relations between the Arawakan language Amuesha and local Quechua in the central Peruvian foothills. Currently, Quechuan and Arawakan Campa languages are in close contact in southern Peru.

Of course the most famous example of Puquina–Quechua interaction is Kallawaya, the lexicon of which contains a number of Puquina elements, and the grammar of which, even though primarily Quechua, also has some unusual features, particularly in the early sources (Muysken, Reference Muysken, Crevels and Muysken2009). Kallawaya is the (almost) extinct ritual language of the Charazani region (northern Bolivia). Katja Hannss is currently exploring the Kallawaya lexicon, attempting to find more Puquina roots. A typical example of the differences between Kallawaya and Quechua is found in (7), cited from Oblitas Poblete (Reference Oblitas Poblete1968: 40). Bold elements are Quechua endings.

(7)

In our interpretation, Kallawaya may be an example of a language changed by processes of metatypy (through which Puquina was gradually restructured under the influence of Quechua), and then its structural frame was fully replaced by Quechua. Quechua underwent relexification with words from Puquina and other languages. Table 6.3 summarizes a proposed history of the complex historical relationships of the region.

Table 6.3 Putative historical development of language use in Kallawaya villages in the Charazani region

2.1 The Amazonian pioneers

Since the first voyage of Christopher Columbus and his followers, South American landscapes, and particularly the Amazon region, have been thought of as the ultimate example of pristine wilderness, encompassing a unique example of rich biodiversity with little human disturbance. Informed by research in anthropology, archaeology, historical ecology, and soil science since the 1980s, the scientific community has slowly adjusted this image towards a view encompassing substantial human influence in the species composition of the world's largest area of tropical rainforest. Since the discovery of Balée (Reference Balée, Descola and Taylor1993: 231) that up to 12 percent of the Amazonian ecosystem is of anthropogenic origin, scholars have noted that much of the “pristineness” of Amazonian forest is actually an effect of the demographic collapse of the indigenous populations following in the wake of the European colonization. Furthermore, archaeological investigations reveal large-scale earthworks, subsistence systems, and settlements, confirming the hypothesis that the sparsely populated Amazonia of the historical period is a relatively recent anomaly when contrasted to the socio-economic development of the region during the last 3,000 years.

Human subsistence strategies in Amazonia have for at least 9,000 years involved domesticated crops (Piperno and Pearsall Reference Piperno and Pearsall1998: 4; Oliver Reference Oliver, Silverman and Isbell2008: 208). When small bands of hunter-gatherers at that time began the domestication of bitter manioc (Manihot esculenta crantz), it marked the starting point of a landscape modification process that was to continue until the demographic collapse following the European colonization some 8,500 years later.

By 7000 BP¹ the indigenous societies along the lower Amazon and the present Brazilian Atlantic coastline were producing ceramic vessels and shell middens, forming the earliest centers of pottery production in the New World. Along the middle and lower Amazon, the archaeological sites of Dona Stella, Pedra Pintada, and Taperinha show initial signs of horticultural activities between 8000 and 7000 BP (Roosevelt et al. Reference Roosevelt, Lima da Costa, Machado, Michab, Mercier, Valladas, Feathers, Barnett, da Silveira, Henderson, Silva, Chernoff, Reese, Holman, Toth and Schick1996; Piperno and Pearsall Reference Piperno and Pearsall1998: 4; Petersen et al. Reference Petersen, Neves and Bartone2004), and at Taperinha forest clearing is indicated by 7000 BP, and clearly documented at Lake Geral, a site located in the same region dated to 5760 BP (Bush et al. Reference Bush, Miller, de Oliveira and Colinvaux2000). These early signs of food production and accomplished material culture soon spread through a regional exchange system operating along the coastline between the mouth of the Amazon and Orinoco Rivers (Eriksen Reference Eriksen2011: 127f). Along the coastline of present-day Colombia, Venezuela, and Guyana, the location of a number of shell mounds with a characteristic lithic assemblage labeled the Ortoiroid series, sharing similarities with the above-mentioned sites of the Amazon river region, indicates the establishment of a wide-reaching exchange system already at this point in time (Boomert Reference Boomert2000: 74).

Sometime between 6500 and 5250 BP, the art of ceramic manufacture was exchanged between the lower Amazon and the Guyana coastline, an event marked by the establishment of the Late Alaka phase of the latter area (Evans and Meggers Reference Evans and Meggers1960; Roosevelt Reference Roosevelt1997: 360; Plew Reference Plew2005: 13). These two areas continued to share similarities when the Mina phase (5500–4000 BP), another archaeological complex producing crude ceramics and shell mounds, was established along the lower Amazon and at the coastline south of the river mouth (Simões and Araujo-Costa Reference Simões and de Araujo-Costa1978; Roosevelt Reference Roosevelt1997). Without losing ourselves in details of the early indigenous material culture, subsistence strategies, and exchange systems of northern South America, it is safe to say that much of the early accomplishments of these socio-economic categories took place through the sharing of important achievements between different groups separated by rather large geographic distances, a mechanism in itself indicative of the character of the future to come.

2.2 The birth of the Arawakan matrix

The complexity of pottery production grew steadily along the Guyana coastline and the Orinoco River, a process leading to the establishment of technologically more complex and stylistically elaborated wares in the form of the Saladoid² and Barrancoid³ series along the Orinoco River by around 3000 BP. By that time, agriculture also was substantially intensified in the same region. Along the Orinoco, the Saladoid and Barrancoid producing societies developed a technology for soil fertilization based on the addition of ash, charcoal, and domestic waste to the soil, with increased microbial activity and improved fertility as the result (Oliver Reference Oliver, Silverman and Isbell2008: 211; see also Arroyo-Kalin et al. Reference Arroyo-Kalin, Neves, Woods, Woods, Teixeira, Lehmann, Steiner, Winkler Prins and Rebellato2009 for technical specifications). This process created sustainable conditions for high-intensity food production, and the black, fertile soils (also known as terras pretas or Amazonian Dark Earths [ADE]) spread widely across Amazonia between 900 BCE and 1500 CE (Lehmann et al. Reference Lehmann, Kern, Glaser and Woods2003; Glaser and Woods Reference Glaser and Woods2004; Woods et al. Reference Woods, Teixeira, Lehmann, Steiner, WinklerPrins and Rebellato2009). Apart from the addition of charcoal and ashes to the agricultural lands, burnt tree bark (Licania sp.) was also being added to the ceramics as a potent tempering material for increased solidity of the vessels. West of the Orinoco River, on the seasonally sedimentary soils of the flooded savannas of the Llanos, proper drainage of the soils was a bigger challenge than lack of available nutrients. In this area agricultural intensification took place through the construction of elevated cultivation surfaces, so-called raised fields or camellones, improving agricultural conditions by elevating parts of the otherwise flat landscape for the improvement of soil conditions, drainage, water management, and nutrient production in order to stimulate agricultural productivity (Denevan Reference Denevan1970; Darch Reference Darch1983; Erickson Reference Erickson, Balée and Erickson2006: 251).

The refinements of agricultural technologies and pottery production were not isolated technological advancements, but, as we will argue below, part of a cultural package that was just beginning its march across Amazonia. Interestingly, the cohesive links of this cultural package were not the technological advancements or the surplus production (even though they were both intrinsic parts of it) but language, and more particularly languages of the Arawakan language family. At the time of European contact⁴ at least sixty Arawakan languages were spoken from the Greater Antilles in the north to northern Argentina in the south, and from the mouth of the Amazon in the east to the eastern Andean slopes in the west (Grimes 2009 lists fifty-nine documented Arawakan languages, not including several extinct ones) (Map 7.1). Arawak-speakers across South America and the Caribbean are united by two main factors: (1) the genealogical relationship of their languages, i.e. their descent from a common proto-language, and (2) a shared set of cultural features, i.e. both material and non-material attributes.

By 1500 CE the Arawak-speaking groups inhabited numerous seasonally flooded environments of the South American tropical lowlands with raised field agriculture or similar technologies, including the Llanos of Venezuela and Colombia (Spencer and Redmond Reference Spencer and Redmond1992), the Llanos de Moxos of Bolivia (Erickson Reference Erickson, Balée and Erickson2006), Marajó Island at the mouth of the Amazon (Schaan Reference Schaan, Silverman and Isbell2008), and the Guyana Littoral (Versteeg Reference Versteeg, Silverman and Isbell2008). They were also known as the moderators of an elaborate set of ritual ceremonies with the use of sacred musical instruments and chanting as essential ingredients (Izikowitz Reference Izikowitz1935; Hill Reference Hill and Santos-Granero2009). Apart from this, Arawak-speakers such as the Taino of the Greater Antilles, the Lokono of the Guyana Littoral, the Manao of the central Amazon, the Achagua and Caquetío of the Llanos, and the Moxo of the Llanos de Moxos (just to name a few) were well-known traders carrying out exchange between various ethnolinguistic groups (Eriksen Reference Eriksen2011: 275).

2.3 The timing of the geographic expansion of the Arawakan matrix

The main ingredients of the Arawakan cultural package were first being brought together in the Orinoco region around 900 BCE. By that time, we find high-intensity landscape management systems in the form of raised fields and terras pretas for agricultural production and causeways for transportation, water management, and possibly also including ritual functions. Also present were ceramic artifacts rich in painted and plastic decoration, that is to say features indicative of a rich ceremonial life similar to that documented from Arawak-speaking communities of the historical period (Santos-Granero Reference Santos-Granero1998; Heckenberger 2008; Hill Reference Hill, Hornborg and Hill2011). Interestingly, the rich ceremonial life of the Arawak-speakers of the historical period was essentially constructed around two themes: (1) the presence of elaborate techniques for physically and ritually domesticating their surrounding landscapes, and (2) the use of fire in processes of landscape domestication and during other types of ritual ceremonies.

As described above, by-products of fire such as charcoal and ashes were an essential part of the subsistence strategies and ceramic manufacture of the Orinoco Region already by 900 BCE. The elaborately decorated ceramics of the Saladoid and Barrancoid series act as indicators of the elaborate ceremonial life of the Orinoco communities, and, interestingly, the rich ethnographic record of Arawak-speaking communities across Amazonia shows that the use of tobacco smoke by Arawakan shamans is considered an essential aspect of ritual ceremonies, including healing processes. Thus, an image of a cultural package capable of transforming the landscape into a high-productive resource, while at the same time providing a powerful ceremonial life for its members, now arises through the archaeological, anthropological, geological, and historical records.

The combination of high-intensity landscape management strategies and a rich ceremonial life would prove to be highly successful during the centuries to follow. By 400 BCE the first evidence of terra preta farming appears in the central Amazon, and shortly thereafter the first earthworks of the Llanos de Moxos represents the initial signs of landscape modification in this region. Judging by the great differences in terms of ecology between the habitats colonized by the subsistence strategies of this emergent regional system, there was a great deal of adaptation available within the communities involved in this process. In much of the central and lower Amazon, large surfaces of fresh sediments rich in available nutrients annually drained from the Andes, forming great conditions for high-productive agriculture, were available. These so-called várzeas eliminated the need for raised fields or terras pretas in many areas of central Amazonia and, interestingly, Versteeg (Reference Versteeg, Silverman and Isbell2008: 305) notes how the raised fields can be compared to artificial várzeas that are subjected to controlled inundation, bringing nutrient-rich sediments to the elevated surfaces during parts of the year.

By 200 BCE Barrancoid pottery and terra preta farming were present along the Ucayali River in the Peruvian Amazon (Lathrap Reference Lathrap1970: 117; Eden et al. Reference Eden, Bray, Herrera and McEwan1984: 126), and at around 100 BCE archaeological dates of the huge earthwork complex of Acre, northwest of the Llanos de Moxos, begin to cluster (Saunaluoma Reference Saunaluoma2010: 106). The geometrical earthworks of Acre, also known as geoglyphs, are perhaps the most visually stunning example of the ceremonial aspects of earth-moving that had been crystallizing across Amazonia from about 900 BCE. The Acre geoglyphs so far discovered consist of more than 200 (an estimated 10 percent of the total number [Mann Reference Mann2000]) geometrical figures carved out of the soil by ditches and walls extending up to 3 meters deep and 11 meters wide. The size of the earthworks measures up to 300 meters across and their frequency is up to 5 geoglyphs per km² (Hornborg et al. Reference Hornborg, Eriksen and Bogadóttir2013).

The cultural associations of these earthworks are so far unknown, but the presence of pottery related to the Barrancoid series (Saunaluoma Reference Saunaluoma2010: 94), their dating, and the engineering skills and focus on soil moving among the Arawak-speaking communities on the nearby savannah of the Llanos de Moxos makes an association with the Arawakan cultural complex highly plausible. As for the functions of the geoglyphs, little is known, but suggestions that they were used for fortification purposes have been made. Although this may be true for the circular structures, particularly those dated to the late pre-Columbian period when military conflicts were expanding across the lowlands (see below), it is unlikely that the communities erected up to five elaborately constructed geometrical structures with low ditches with little or no defensive capabilities per square kilometer because they feared an external threat in the form of spears, arrows, or war clubs. On the contrary, the fortified villages of the Arawakan cultural complex documented from the late pre-Columbian periods are semi-circular structures, located on high ground, adapted to the local topography and surrounded by palisades (Rebellato et al. Reference Rebellato, Woods, Neves, Woods, Teixeira, Lehmann, Steiner, WinklerPrins and Rebellato2009).

Interestingly, new research in the upper Xingú area, another region of Amazonia populated with – and culturally dominated by – Arawak-speakers, is finding support for the notion that Arawak-speakers sometimes devoted themselves to strictly ceremonial domestication of their landscapes. During the late pre-Columbian period, the upper Xingú area had developed integrated patterns of centers organized in multi-ethnic, or “galactic,” clusters populated by up to 2,500, and perhaps as many as 5,000 inhabitants (Heckenberger Reference Heckenberger, Balée and Erickson2006: 330; Reference Heckenberger, Silverman and Isbell2008: 955). These multiethnic confederations, referred to as an early form of urbanism by Heckenberger et al. (Reference Heckenberger, Silverman and Isbell2008), were integrated by wide road-like causeways resembling the elevated causeways of the Llanos de Moxos, which facilitated cultural, linguistic, and material exchange within and between regions. The landscapes were domesticated by the creation of circular villages with a central plaza and radial road networks with perfectly straight passages connecting a multitude of such population centers to each other in a regional system.

In Arawak-speaking areas where earthworks and other landscape-altering techniques were less prevalent, other strategies of landscape domestication were employed. The Yanesha’, an Arawak-speaking group of the eastern Andean slopes in present-day Peru, apply an intricate system of “topographic writing” in order to maintain an intimate relationship to their landscape (Santos-Granero Reference Santos-Granero1998). Topographic writing is the concept Santos-Granero uses to describe how individual place names (topograms) are connected to extensive systems (topographs) and reiterated, for instance through chanting in ritual ceremonies in order to strengthen the ties to the local and regional landscape (p. 128). Such ritual place-naming is also well documented from the northwest Amazon Arawakan people (Vidal Reference Vidal2000, Reference Vidal, Hill and Santos-Granero2002; Hill and Chaumeil Reference Hill, Hornborg and Hill2011; Wright Reference Wright, Hill and Chaumeil2011) and from Arawakan groups in southern Amazonia such as the Paresi (Schmidt Reference Schmidt1917: 21f). Santos-Granero (Reference Santos-Granero1998: 132, 139) refers to the landscape domesticating process of topographic writing among the Yanesha’ as a form of proto-writing, also present among other Amazonian groups such as the Paéz (to which it likely diffused through contact with nearby Arawak-speaking communities), a linguistic isolate between the Marañon and Napo Rivers, and the Arawak-speaking Kurripako (Wakuénai) of the northwest Amazon (Hill Reference Hill and Hill1996: 153f; Reference Hill, Hill and Granero2002: 235f; Reference Hill and Santos-Granero2009: 250).

Returning to the archaeological material, by 300 CE a new ceramic style, labeled the Amazonian Polychrome tradition, had developed out of Barrancoid material along the middle and lower Amazon (Hilbert Reference Hilbert1968; Lathrap Reference Lathrap1970: 156f; Eden et al. Reference Eden, Bray, Herrera and McEwan1984: 137; Myers Reference Myers, Glaser and Woods2004: 79; Petersen et al. Reference Petersen, Neves and Bartone2004: 9; Eriksen Reference Eriksen2011: 181). At the time of European contact, the Arawak-speakers of Marajó Island at the mouth of the Amazon, the Aruã, were still manufacturing an undecorated variant of the Marajoara phase (one of the most elaborate phases of the Amazonian Polychrome tradition)⁵ labeled the Aruã phase, when they were encountered by the Europeans (Brochado and Lathrap 1982: 53).

Once again, the significance of burning is reflected in the anthropomorphic burial urns typical of the Amazon Polychrome tradition, an inventory indicating secondary urn burials in which the cremation of the corpse and the storing of the ashes in the urn were central components. In many instances, even the pottery itself included ashes in the form of caraipé temper utilized in the Ipavu phase in the upper Xingú (Heckenberger Reference Heckenberger1996: 136f), the Guarita phase in the middle Amazon (Petersen et al. Reference Petersen, Heckenberger, Neves, Alofs and Dijkhoff2003: 252), the Mazagão phase in Maracá (Meggers and Evans Reference Meggers and Evans1957: 596), the Koriabo phase of the Guyanas (Boomert Reference Boomert, Delpuech and Hofman2004: 259), and, together with crushed sherds, in Marajoara (Brochado and Lathrap Reference Brochado and Lathrap1982: 50). The burial urns, also well known among the northwest Amazon Arawakan people and those in the Llanos de Moxos − the Moxo and the Baure − (Mann Reference Mann2000; Métraux 1948c) and Arawakan groups in the Bolivian Chiquitanía, like the Paunaka (Métraux Reference Métraux and Steward1948b), were often stored in caves that could be visited and inspected (Chaumeil Reference Chaumeil, Fausto and Heckenberger2007: 250ff), indicating close ties to the ancestors, and the importance of maintaining a strong relationship to deceased relatives, shamans, and political leaders – a custom typical of the Arawak-speaking communities of the historical period. Another way of maintaining a close link to the ancestors was through ritual consumption of their remains, as illustrated by the Arawak-speaking Guayupe and Sae, who cremated their ancestors and drank their ashes mixed with beer (Kirchhoff Reference Kirchhoff and Steward1948: 387f.).

The process of burning was also a central aspect of other Arawak-moderated rituals performed in their sphere of influence. At religious ceremonies performed in the northwest and southwest Amazon and in the upper Xingú area, tobacco smoke is a central element in healing-ceremonies conducted by the Arawakan shamans, who blow the tobacco smoke on the patient's body in order to eliminate the patient's symptom (Hill Reference Hill and Santos-Granero2009: 249, 259; Hill and Chaumeil Reference Hill, Hornborg and Hill2011). The blowing of tobacco smoke on patients has also been reported of the Bolivian Arawakan groups of the Paunaka in shamanic ceremonies (Danielsen, own observation) and it is also the way the shaman gets in contact with the spirits of the deceased among the Baure (Riedel Reference Riedel2012). According to Goldman (Reference Goldman and Steward1948: 789), smoke was also blown during funerals, reflecting the association between this custom and the importance of deceased ancestors.

Via the historical and contemporary ethnographical sources, we find another interesting connection between, on the one hand, shamanistic blowing of smoke, and, on the other, the ritual wind instrument also utilized by Arawakan shamans during religious ceremonies. Among contemporary Arawak-speaking communities of the northwest Amazon, the upper Purús River (the Apurinã) and the upper Xingú area, ritual wind instruments play a central role in annual ceremonies and during rites de passages.

Apart from the apparent association to shamanic blowing, the sacred flutes of the Arawakan people also had a striking connection to landscape and fire worth exploring further. According to the legends of the Arawak-speakers of the northwest Amazon, the earth was created from the remains of a mythological hero, Kúwai, after his body had been destroyed in a fire. Besides being the ancestor from whose body the world of humans was created, Kúwai also provided material for the ritual wind instruments used in religious ceremonies. The Yuruparí flutes are artifacts directly derived from the bones of the mythological hero and thus representatives of the ancestors (Steverlynck Reference Steverlynck2008: 580). In the words of Robin Wright (Reference Wright, Hill and Chaumeil2011):

Overall, the sacred wind instruments of the Arawakan people were one of their most central characteristics. Sacred flutes have been known to occur among a number of Arawak-speaking groups of the northwest Amazon, including the Achagua, Baniwa, Bare, Cabiyarí,⁶Kurripako, Maipure, Yucuna,⁷ Pasé,⁸Resígaro, and Yumana, and they also occur among neighboring non-Arawak-speaking groups who maintain close sociocultural contact with the Arawakan groups (Chaumeil Reference Chaumeil, Fausto and Heckenberger2007; Wright Reference Wright, Hill and Chaumeil2011). Chaumeil (Reference Chaumeil1997, cited in Steverlynck Reference Steverlynck2008: 579) points to the connection between the sacred flutes complex of the northwest Amazon and the use of ceremonial trumpets by Taino shamans of the Greater Antilles.

Map 7.1 The reconstructed geographical dispersal of the Arawakan and Tupian language families at the time of European contact. For complete references, see Eriksen Reference Eriksen2011: 12

Chaumeil (Reference Chaumeil, Fausto and Heckenberger2007: 265) notes how Arawak-speaking groups dominate the sacred flutes complex throughout Amazonia, and Wright (Reference Wright, Hill and Chaumeil2011) identifies the sacred flutes as an important element in the expansion of Arawakan languages. Arawak-speaking groups located outside of the northwest Amazon who also use sacred flutes include the Apurinã of the Purús River; the Baure and Moxo in the Llanos de Moxos; the Paresi further west; and the Mehinaku in the upper Xingú. Other groups belonging to the same complex include some Tupian-speaking groups such as Cocama and Omagua, Mundurukú, Tupinambá, and Kamayurá. In the upper Xingú, the complex also spread to the Carib-speaking Kalapalo and Bakairí, who were “Arawakanized” by their Mehinaku, Kustenau, Yawalapití, and Waurá neighbors (Chaumeil Reference Chaumeil, Fausto and Heckenberger2007: 266).

During female initiation rites among the Arawak-speaking Kurripako (Wakuénai) in the northwest Amazon, the sacred flutes are used during up to six hours long ceremonies of chanting during which an enormous series of place names along the rivers of northern South America are reiterated (Hill Reference Hill and Hill1996: 153f; Reference Hill, Hill and Granero2002: 235f; Reference Hill and Santos-Granero2009: 250). These place names represent nodes in an exchange system once dominated by Arawak-speakers, but they are also part of a geographic network with strong mythological connotations. This exchange network, known as the Kúwai route (borrowing its name from the creator), represents both a trade network constructed on the basis of physical travels over centuries, but also a collection of mythological places where Arawakan shamans head on their transcendental journeys during séances. Thus, the sacred flutes, the Kúwai routes, and the relationship to the ancestors and the mythological past form a trinity of inseparable components that collectively contributed to a strengthened identity and sociopolitical status of the Arawak-speaking communities.

Like many indigenous cultures around the world, for Arawak-speaking communities the physical and religious aspects of the landscape form a whole. The landscape functions as a single meaningful unit, steadily present in the life and minds of its inhabitants. However, the Arawakan domestication of the landscape was not only meaningful to the Arawakan people themselves but also part of a vast socio-religious and economic exchange system that affected the lives of all inhabitants of northern South America between 1000 BCE and 1000 CE. Together with intensive agriculture, an effective exchange system, and an advanced sociocultural and religious concept based on social hierarchies and ancestor worship, Arawakan languages (which formed intrinsic parts of these three concepts) expanded across an enormous geographic territory on mainland South American and in the Caribbean. The diversity and the power of the Arawakan groups led to the complete or partial adoption of the Arawakan cultural matrix and associated languages by many indigenous groups between 1000 BCE and 1000 CE. At the time of the European arrival Arawakan languages were widely spoken and northern South America showed extensive domesticated landscapes (Map 7.1).

Arawakan culture also came to influence the Andean region substantially, as indicated by the presence of a large number of lowland products brought to the highlands via the Arawak-controlled trade routes (Eriksen Reference Eriksen2011: 164). Along the trade routes of the eastern Andean slopes, an ethnolinguistic group known as the Kallawaya transported lowland products with pharmaceutical or hallucinogenic characteristics to the Andean cultures (Rowe Reference Rowe and Steward1946: 239; Wassén Reference Wassén1972: 63; Lathrap Reference Lathrap1973: 180f; Taylor Reference Taylor, Salomon and Schwartz1999: 199; Eriksen Reference Eriksen2011: 78). The Kallawaya tongue was a mixed language based on the Arawakan language Puquina, the isolated Chipaya language, and Quechua (Gordon Reference Gordon2005, Hannss p.c.). Puquina was a high-status language spoken among the Inca elite (Torero Reference Torero2002; Dudley Reference Dudley and Alexiades2009: 146). Along the eastern Andean slopes in present-day Peru, a number of Arawakan languages are still spoken, and advanced systems for ritual domestication of the landscape have been documented among the Yanesha’ of that region (Santos-Granero Reference Santos-Granero1998) (Map 7.1). These groups bear testimony to the incredible ability of the Arawakan matrix to maintain its relevance for its users during the sociocultural changes on-going for centuries.

2.4 The expansion of the Arawakan family from a linguistic perspective

As noted, the Arawakan language family has expanded over a very large area of South America, more than other language families (Map 7.1). A suggested Arawakan language classification is presented in Table 7.1, mainly based on geographic proximity, but also on grammatical features, summarized from Aikhenvald (Reference Aikhenvald, Dixon and Aikhenvald1999a), Walker and Ribeiro (Reference Walker and Ribeiro2011), and Danielsen and Terhart (forthcoming).⁹

Table 7.1 The Arawakan language family

The internal classification of Arawakan languages is difficult to establish (see Facundes Reference Facundes, Hill and Santos-Granero2002), and some reasons for this are discussed in 2.6. This section focuses more on the character of the Arawakan linguistic family as such. Instead of taking the lexicon as the basis of comparison, which was done in other studies (Payne Reference Payne, Derbyshire and Pullum1991, Walker and Ribeiro Reference Walker and Ribeiro2011), we take grammatical features as our point of departure here and compare our findings to those of lexical comparisons. The similarities of the Arawakan languages in some key linguistic features suggest that the expansion happened rather quickly (see Danielsen et al. Reference Danielsen2011). The personal paradigm, e.g., is similar in many respects in most Arawakan languages, formally as well as functionally (see Payne, David L. Reference Payne1987), and it has often served as the first characteristic for assigning the genealogical relationship to a language (Gilij Reference Gilij1780–84). The proto-system of person marking was presumably a “Latin-type” paradigm (see Cysouw Reference Cysouw2003: 107): 1sg, 2sg, 1pl, 2pl, 3sg with a gender distinction (masculine/non-feminine and feminine), and 3pl (also applied as a general pl suffix). Person markers are employed to mark the possessor on nouns (prefix), subject on verbs (S_A, prefix), object(s) on verbs (suffixes), and the subject on stative or non-verbal predicates (S_O, suffix).¹⁰ In addition, free pronouns are derived from these bound personal forms, as well as certain adpositions marked for person. This general grammatical system can be claimed for all Arawakan languages, and the differences lie mainly in the specific lexemes that make use of a certain kind of marking, such as which nouns are actually part of the category inalienably possessed (with person marking), and which verbs belong to the (active) set with S_A marking or the (stative) set with S_O marking. There may also be striking differences in the SAP (speech act participants, i.e. first and second person reference) marking system versus the 3rd person in some languages, and the SAP forms tend to be more stable. Here also certain functions only hold for sub-clusters of the language family, such as gender marking or derivation of adjectives or nouns by means of the same personal forms (suffixes) for 3rd person. If we model the Arawakan language family as a NeighborNet splitsgraph only with respect to the person marking system (forms and functions), we get a rather plausible picture of the geographic distribution of the languages and possible migration routes (Figure 7.1).

Figure 7.1 The distribution of the personal paradigms in Arawakan languages¹¹

The star-like splitsgraph in Figure 7.1 shows no clear northern versus southern branching of the language family with respect to the features examined, contrary to what Table 7.1 suggests. There is a tendency that Northern Arawakan and Southern Arawakan languages group together and a few sub-clusters can be observed. The same is true for the lexicon, as analyzed in Walker and Ribeiro (Reference Walker and Ribeiro2011: 2563). This means that there are some shared Arawakan features almost equally distributed: the personal paradigms to some extent and some of the conservative lexicon. The reason for this presumably goes back to the Arawakan exchange network that functioned for a long time. If the languages had spread like Tupian, we would presumably be able to see certain clear expansion groups, which is not really the case (compare Figure 8.1 in Chapter 8 on the Tupian expansion). However, some groups and subgroups within the language family can be found clustering in the graph in Figure 7.1: Some Northern Arawakan languages cluster, like Lokono with Island Carib, Garifuna, and Paraujano, but also with Resígaro in this graph. While Bare and Tariana appear closely related, Piapoco and Achagua have been separated according to their person marking systems. So, there is some confusion and a geographically unclear picture of northern and central Amazonian Arawakan languages. A similar observation was made about the lexical relations (Walker and Ribeiro Reference Walker and Ribeiro2011: 2563). The Purus group and Campa Arawakan cluster nicely, and Bolivian Arawakan languages, the Baure, Moxo, Pauna languages, more or less as well. The less Arawakan character of Apolista, Yanesha’, and Chamicuro, presumably due to the Andean influence these languages have undergone, is reflected in their relatively isolated positions in the splitsgraph. This may be a sign of an earlier migration into the region than the other Arawakan languages of the Andean foothills. The complicated position of Resígaro within the Arawakan language family has been discussed in the literature before (Payne Reference Payne1985). Its odd position among the Northern Arawakan languages in the graph − and also in Figure 7.2 − is probably just a sign of a connection at some point in history; this may have been in times when the Arawakan web stretched from the Caribbean to the Andes. The Taino language, excluded from the analysis in Figure 7.1 due to the lack of sufficient data, has also lately been described as being on the one hand closely related to Northern Arawakan languages, such as Island Carib, Lokono, Guajiro, Piapoco, and Achagua (cf. Granberry and Vescelius Reference Granberry and Vescelius2004: 56). On the other hand, it also seems to have certain characteristics found in South-Western Arawakan languages, in particular the Campa group, namely certain nominal (classifying) root formatives (cf. Granberry and Vescelius Reference Granberry and Vescelius2004: 94). Is this another hint at traces of the times of the interaction over such distances? The Northern Arawakan language Palikur appears right among the Southern Arawakan languages, from which it is geographically far away. However, being a Brazilian Arawakan language, Palikur seems to display some connection to other eastern and southeastern (Brazilian) Arawakan languages. The latter have not been claimed to form any particular subgroup, but in Walker and Ribeiro (Reference Walker and Ribeiro2011: 2563), these languages are included under the name “Central Brazil,” since they cluster in their lexical analysis. In Table 7.1, we call this assumable group South-Eastern Arawakan. The Arawakan languages of the Xingú group (Paresi, Saraveka, Waurá) and possibly those of the Terêna subgroup (Terêna, Kinikinau, Mehinaku) may well be a loose intermediate group with characteristics similar to their northern and northwestern neighbors as well as to their southern genealogical neighbors. The findings in Granberry and Vescelius (Reference Granberry and Vescelius2004: 55 ff.) also seem to point in this direction, and Walker and Ribeiro (Reference Walker and Ribeiro2011) demarcate Palikur (and Marawan) as a subgroup named “Northeast.” This, however, remains to be further substantiated.

Figure 7.2 Minimum spanning network of the Arawakan language family (also taken from the NeighborNet algorithm, Huson and Bryant Reference Huson and Bryant2006)¹²

To get a better picture of the possible expansion of Arawakan languages, the same feature matrix as used for the NeighborNet splitsgraph in Figure 7.1 can be reduced to a Minimum Spanning Tree (MST), as given in Figure 7.2.

The graph in Figure 7.2 provides us with possible routes of the dispersal of the respective Arawakan languages. Even though this is only one interpretation of the given data as a dispersal route, this scenario has plausibility.¹³ The position of Island Carib and Garifuna shows an excursion of the Arawak-speaking people into the Caribbean Sea, probably at an early stage and starting off from Maipure and Palikur. Another expansion of Arawakan could have led to the northern coast with Lokono at its end. As already mentioned above for Figure 7.1, Resígaro may be a remnant of the Arawakan expansion towards the Andes and a sign that there was still regular exchange at that time between the east (Resígaro) and the west (Lokono and others) of Amazonia through Arawakan peoples.¹⁴ In the south, we may conclude that the Moxo languages Trinitario and Ignaciano came into the area through Baure. There is some evidence for the fact that Baure came into the region earlier due to its relatively conservative character. It is here also suggested that Apolista and Yanesha’ are part of one migration route, starting off with Chamicuro, which is the most northern member of this loose group of Andean-influenced Arawakan languages.

2.5 The fragmentation of the Arawakan matrix

By approximately 800 CE, the Arawakan languages and the associated cultural complex, here labeled the Arawakan matrix, had reached their maximal territorial extent. By that time, Arawakan languages from the Greater Antilles to northern Argentina and from the Atlantic to the Andes were united by a large and complex regional exchange system. The Arawak-speaking communities had by that time accumulated considerable land-based capital in the form of agricultural earthworks, aquacultural facilities, and infrastructure that was attractive to other indigenous Amazonian groups.¹⁵ Since the early centuries of the first millennium CE, another major ethnolinguistic formation came to the fore in the Tupian language family, which had begun expanding out of its point of origin in the Brazilian state of Rondônia (Rodrigues Reference Rodrigues1964). Up until then, the geographic distribution of the Tupian family had remained very restricted, despite early internal branching (Eriksen and Galucio, this volume).

The geographic expansion of the Tupian languages took place very differently from the spread of the Arawakan languages. While the Arawakan languages were part of a complex exchange system with strong mythological and ceremonial underpinnings, the Tupian language family was part of an expansionistic military culture. Where the Arawakan societies prioritized ancestry and descent as the bases for political power, the Tupians based their social hierarchies on feats on the battlefield (Eriksen and Galucio, this volume).

Particularly among the communities of the Tupí-Guarani branch, the groups developed an effective ability to absorb cultural traits and technological elements from neighboring groups in order to strengthen their own social status, military power, and agricultural production (Eriksen and Galucio, this volume). Due to these abilities of the Tupian cultures, it was inevitable that the encounter between groups speaking Tupian and Arawakan languages along the shores of the middle and lower Amazon around 700 to 800 CE would lead to extended periods of conflict (Map 7.1). Military aggression was in evidence from the start and the remains in the archaeological record bear witness of burned villages, destroyed palisades, and ultimately a change in village layout from the circular villages of the Arawakan communities to the linear settlements of the Tupí-speakers documented from the historical period (Heckenberger Reference Heckenberger2005: 56; Rebellato et al. Reference Rebellato, Woods, Neves, Woods, Teixeira, Lehmann, Steiner, WinklerPrins and Rebellato2009: 22, 29). The military conflict between Tupí- and Arawak-speaking communities along the main river ultimately led to Tupian control of a large section of the Amazon River from the mouth of the Amazon to the tributaries in eastern Peru by 1200 CE (Map 7.1).

In addition to this, other Tupí-Guarani languages had expanded along the Atlantic coastline, circumscribing the Macro-Jê speakers and restricting their distribution to the Brazilian highlands. The three other expanding branches of the Tupian language family, Mundurukú, Mawé-Sateré, and Yuruna expanded in the area immediately south of the middle and lower Amazon River, contributing to a strong dominance of Tupian languages in southern Amazonia during the late pre-Columbian period (Eriksen Reference Eriksen2011). Overall, the expansion of the Tupian family replaced the Arawakan dominance in many areas of Amazonia and led to the fragmentation of the previously pan-Amazonian character of the Arawakan regional exchange system (Map 7.1).

However, the sociocultural processes through which Tupian languages replaced Arawakan were sometimes more complex than the predatory ethos of the Tupí-speaking groups would lead us to believe. According to several linguists (Cabral Reference Cabral1995; Jensen Reference Jensen, Dixon and Aikhenvald1999: 129; Adelaar with Muysken Reference Adelaar, Adelaar and Muysken2004: 432), the structures of the Tupí-Guarani languages (Omagua, Cocama, and Cocamilla) of the upper Amazon indicate that they represent a language shift¹⁶ from some non-Tupian language(s) to Tupinambá. This indicates that a new cultural pattern, including both language and material culture, was adopted in the region about 1200 CE. Included in this cultural package was polychrome pottery, locally developed into the Napo and Caimito phases. Epps (Reference Epps2009: 599) has suggested that Cocama and Omagua represent two different language shifts from Arawakan languages to Nheengatú, the Tupinambá-based lingua franca still spoken in the northwest Amazon.

As a result of the Tupian expansion at the expense of the Arawakan languages, a change in land use also followed. While the Arawakan communities had been heavily sedentary, relying on their earthworks, terras pretas, and aqua-cultural facilities for long-term subsistence, the military apparatus of the Tupian groups acted as an incentive for more mobile subsistence strategies. Many ethnolinguistic groups of the Tupí-Guarani, Mundurukú, Mawé-Sateré, and Yuruna branches launched annual war expeditions up the major rivers and tributaries. These expeditions, some of which were documented by the early Europeans of the continent, lasted for months and required access to easily transportable food resources. As a result of this, the Tupian groups along the main river came to rely heavily on short-ripening maize-varieties that were grown on the annually flooded várzea areas. The Arawakan groups also produced a considerable food surplus, e.g. beer made from maize and manioc, to be consumed during religious ceremonies. Thus, the Tupians could rely on their ability to steal food during their war expeditions and their dominance over tributary populations for part of their subsistence (Santos-Granero Reference Santos-Granero2009).¹⁷

It would take another millennium before Amazonia once again experienced an alteration of the landscape similar to the one that took place during the Arawakan expansion from roughly 1000 BCE until 1000 CE. Between 1000 and 1500 CE, the region suffered conflicts and warfare, and the most important socio-economic progress took place in the Andes. During this period, the landscape alteration processes were less intensive than during the first millennium CE. As a consequence of the demographical collapse among the indigenous populations that followed in the wake of the European colonization (an event that eradicated perhaps 90 percent of the population in Amazonia), the anthropogenic environments of Amazonia underwent a reforestation process that in most areas resulted in an advance of the tropical rainforest at the expense of previously maintained grounds. The image of the reforested Amazonia (a process that was completed in most areas of the lowlands before any Europeans entered) has contributed strongly to the image of Amazonia as an ecosystem with little human historical influence.

2.6 The fragmentation of the Arawakan language family from a linguistic perspective

A subdivision into Northern versus Southern Arawakan is not as straightforward as suggested in Aikhenvald Reference Aikhenvald, Dixon and Aikhenvald1999a (Danielsen et al. Reference Danielsen2011, cf. also Walker and Ribeiro Reference Walker and Ribeiro2011: 2563). This is supported by the fact that no major branches can be shown for the language family (cf. Figure 7.1). As we have shown in Section 2.4, Arawakan languages are much alike, at least with respect to selected linguistic features such as the personal paradigms and the lexicon. This then results in the star-like splitsgraph in Figure 7.1. Some other features are in tendency more Southern Arawakan, such as the morphological complexity of the verb and applicative marking types on verbs.¹⁸ Taking all linguistic features into consideration, however, the picture is much more blurred. The splitsgraph in Figure 7.1 demonstrates that there are not many major splits between groups of languages of the Arawakan family, and the distances between them are relatively even, and much more balanced than a main subdivision into Northern versus Southern Arawakan would suggest. This tells us something about the nature of the Arawakan expansion: Firstly, the Arawakan matrix must have remained intact for quite some time, so that linguistic features could still be exchanged (and spread throughout Amazonia to other languages). This explains why general Amazonian features (see Derbyshire and Pullum Reference Derbyshire, Derbyshire and Pullum1986: 19; Dixon and Aikhenvald Reference Dixon and Aikhenvald1999: 8–9) mostly reflect Arawakan features, and why it is almost impossible to find an Arawakan feature that is not also Amazonian or vice versa. Secondly, the expansion of Arawakan was neither unidirectional, nor did it happen in one stroke. Walker and Ribeiro (Reference Walker and Ribeiro2011: 2566) have suggested a more southern point of departure of the Arawakan expansion − western Amazonia, in the area of the Apurinã − than others have come up with before (the Caribbean coast in Aikhenvald Reference Aikhenvald, Dixon and Aikhenvald1999a: 75; the Upper Amazon referring to Lathrap Reference Lathrap1970 and Oliver Reference Oliver1989 in Aikhenvald Reference Aikhenvald, Dixon and Aikhenvald1999a: 75). Usually, we take the area of most linguistic diversity within the language family as the probable homeland, as in the case of Tupian. However, the diversity must represent the source of divergence within the family. In the case of Arawakan, diversity may mean local linguistic interaction with other unrelated languages and is not directly related to the different migrations of Arawakan languages. The area of the northern Amazon and the Caribbean coast are both examples of intensive language contact, in particular between Arawakan languages and languages of other stocks. Therefore, linguistic diversity alone may not be always taken as the key evidence for a homeland. Later language contact is probably the reason why an analysis of general linguistic features of Arawakan languages gives us the picture it does in Figure 7.3 (see Danielsen et al. Reference Danielsen2011).

Figure 7.3 Structural analysis of thirty-one Arawakan languages¹⁹

In Figure 7.3, which is again a star without any major branching, we see that general linguistic features are shared by some subgroups within Arawakan − indicated by the dotted lines − but a great number of the languages appear to be simply mixed in the graph. This fact is the main reason why it has been complicated till now to come up with a decent internal classification of the language family. In contrast to Figures 7.1 and 7.2, Resígaro now occurs more closely to the languages that are also geographically closest, such as Tariana, and not to the Caribbean Arawakan language Lokono. Thus, local contact effects are stronger than possible historical genealogical relations that may be restored from the personal paradigm. Examples of languages that underwent language contact with genetically unrelated languages and are therefore grammatically quite different from other Arawakan languages are the following:

1. Garifuna (see Escure Reference Escure, Escure and Schwegler2004): Arawakan with Cariban (and European languages); Garifuna is a language of Arawakan origin (Island Carib) with substantial interaction of Cariban (at the time during colonization) and English- and French-lexifier pidgins and Creoles at the time during and after colonization)

2. Tariana: Arawakan with Tucanoan (AikhenvaldReference Aikhenvald, Dixon and Aikhenvald1999b, Reference Aikhenvald, Aikhenvald and Dixon2001, Reference Aikhenvald2002); Aikhenvald has done a detailed study of language diffusion in the Vaupés area, and these strong effects of language contact hold for all the languages in this region, not only Arawakan.

3. Resígaro: Arawakan with Bora (SeifartReference Seifart2011); Resígaro has in particular changed its nominal morphology and underwent great grammatical changes under the influence of the Bora language; the lexicon and the verbal morphology remained more Arawakan.

4. Yanesha’ (WiseReference Wise1976) and other Andean foothill Arawakan languages (see Table 7.1): Arawakan with Quechua; Yanesha’, Chamicuro, and Apolista were influenced by Quechua, and they are therefore grammatically distinct from other Arawakan languages in the same area, like e.g. the Campan Arawakan languages.

5. Paunaka: Arawakan with Bésiro (Macro-Jê) from the Chiquitanía (Danielsen and Terhart forthc.); the Paunaka language shows regional grammatical constructions in the morphology of borrowed verbs that are typical for the Chiquitanía, and Paunaka has also had lexical influence from Bésiro more recently.

6. Moxo: Arawakan with possibly Bésiro (Macro-Jê) or already extinct language(s) of the area; the Moxo languages only show a particular non-Arawakan pattern in the personal paradigm form for 3rd person with speaker gender distinction (see Danielsen Reference Danielsen2011, Rose, p.c.) that is not Arawakan.

The long list of reported cases of language contact of Arawakan languages with other languages demonstrates why it is difficult to base an internal classification on the same features for all languages. While some Arawakan languages have been influenced in their nominal morphology, others have changed their verbal morphology, and again others the personal paradigms or the lexicon, the results of the fragmentation of the Arawakan matrix after their wide expansion. A more detailed comparative analysis is needed of the grammatical features of the languages that Arawakan languages came into contact with before a more comprehensive analysis of the fragmentation process can be carried out.

3.1 Linguistic distance based on lexical similarity analysis

Galucio and colleagues (to appear) present the results of a lexicostatistical and phylogenetic study based on the analysis of the Swadesh list of 100 diagnostic words considered to be most stable over time (Swadesh Reference Swadesh1955) for all the nineteen Tupian languages outside the Tupí-Guaraní family and for four Tupí-Guaraní languages (Guaraní, Parintintim, Tapirapé, and Urubu-Kaapor). Their study shows the degree of distance across Tupian languages, confirms the two more recently established branches of Ramarama-Puruborá and Mawé-Awetí-Tupí-Guaraní, and also supports the internal structure of each branch of the family based on historical-comparative methods. In the case of Tuparí and Mondé, the two most diversified branches outside Tupí-Guaraní, the phylogenetic similarity tree agrees exactly with the independent internal classification of these branches (Moore Reference Moore2005; Galucio and Nogueira Reference Galucio and Fernanda Nogueira2012). We took their study and extended it to include a more complete set of languages from the Tupí-Guaraní branch. Using the NeighborNet algorithm implemented in Splits-Tree4 (Huson and Bryant Reference Huson and Bryant2006), we generated an unrooted network expressing a distance measure among the Tupian languages on the basis of lexical similarity in the basic vocabulary for thirty-one Tupí-Guaraní languages and dialectal varieties and the nineteen languages from the other Tupian branches already established by Galucio et al. (to appear). The distances between the languages based on the percentage of shared lexical items are shown in the NeighborNet representation in Figure 8.2.²

The analysis is not intended to show the historical development of these languages but rather the degree of distance between them, based on lexical similarity that may also reflect the result of horizontal transfer. It is nonetheless remarkable that the major clusters of languages that surface from the distance measure shown in the graphic are comparable to the proposed path of historical development for the Tupian languages, on the basis of the comparative method (cf. Figure 8.1). The NeighborNet representation places Awetí as the closest language to the Tupí-Guaraní cluster, followed by Mawé, and together forming the Mawetí-Guaraní larger cluster, which is consistent with the proposed path of evolution in the history of these languages (Drude Reference Drude, Dietrich and Symeonidis2006; Correa da Silva Reference Correa da Silva2011; Drude and Meira, to appear). The six other lexical clusters (Juruna, Arikém, Ramarama-Puruborá, Mondé, Tuparí, and Mundurukú) and their sub-splits correspond exactly to the more recent genealogic classification of these languages, as clearly seen in the Tuparí and Mondé branches (Moore Reference Moore2005; Galucio and Nogueira Reference Galucio and Fernanda Nogueira2012).

The linguistic cohesiveness of the Tupí-Guaraní branch is also prominent in the network representation. Horizontal transfers due to contact and borrowing may be responsible for a great number of the synchronic resemblances in the Tupí-Guaraní lexicon, not all of them due to retention from a common ancestor language. Nonetheless, as expected from the known history of these languages, the thirty-one Tupí-Guaraní languages are closer to each other than to any other language in the Tupian family.

However, the splits inside the Tupí-Guaraní cluster do not correspond exactly to classifications of the Tupí-Guaraní branch based on phonological criteria (Mello Reference Mello2000; Rodrigues and Cabral Reference Rodrigues, Cabral, Cabral and Rodrigues2002) or on a combination of lexical, phonological and grammatical criteria (Dietrich Reference Dietrich1990). There is an overall absence of well-delimited lexical clusters inside the Tupí-Guaraní group in Figure 8.2. Among the few specific clusters that surface from the quantitative lexical comparison are the Kawahib languages (Parintintim, Tenharim, Amondawa, and Uru-eu-uau-uau) that are classified as dialectal variants (Sampaio Reference Sampaio1997); the Yuki-Sirionó cluster of two closely related Tupian Bolivian languages; the Wayampi-Emérillon cluster of languages spoken in the same geographic area in French Guiana; the Língua Geral Amazônica³-Urubu-Kaapor grouping, for which there have been claims of mutual influence through contact; and a Guaraní cluster that includes most of the languages in Rodrigues and Cabral's subgroup I of Tupí-Guaraní (2002) but also includes Guarayo, spoken in Bolivia. The Cocama-Cocamilla language⁴ appears close to Xeta. The Cocama lexicon, including the core vocabulary, is primarily Tupian (Cabral Reference Cabral1995), but it also shows lexical traits of Arawakan, Panoan, and Quechuan origin, in addition to Portuguese and Spanish (Muysken Reference Muysken, Campbell and Grondona2012b).⁵

Figure 8.2 NeighborNet representation of lexical distances among Tupian languages

3.2 Linguistic distance based on structural similarity analysis

For the structural analysis, we designed a preliminary questionnaire of twenty prominent typological features, divided between phonology, morphology and syntax.⁶ Due to the availability of data, our structural sample is smaller than the lexical sample. It consists of thirty languages, including eighteen from the Tupí-Guaraní branch and twelve from the other nine branches of the family. The features were coded on the basis of published material, complemented with direct verification with specialists working on particular languages. The current location of the analyzed languages is shown in Map 8.2. Only two features are identical for all the languages in the sample. All Tupian languages have the order possessor-possessed in the possessive phrase, and all have noun-postposition order in the noun phrase, which is consistent with the general head-marking characteristic of the family. With the exception of Cocama-Cocamilla, which has a causative suffix -ta, all other languages have a causative prefix of the form mV- (<Proto-Tupí *mõ-). The phonological features show few splits throughout the family. The vowel inventories go from four (Língua Geral Amazônica) to seven vowels (Karo, Puruborá and Tenetehara-Tembé), with most of the languages showing five or six vowels. With respect to suprasegmental phenomena, languages of two branches (Mondé and Mundurukú) have contrastive tone, three languages (Juruna, Karitiana, and Karo) have pitch-accent systems, and the remaining languages, including Xipaya, the sister language of Juruna, have stress-only systems (cf. also Storto and Demolin Reference Storto, Demolin, Campbell and Grondona2012).

Map 8.2 Current location of the languages in our structural sample

Preliminary results of the structural analysis can be visualized in the network representation in Figure 8.3. The discrepancies between the lexical and grammatical analyses are remarkable. The only languages that seem to cluster in both analyses are Kaiowa with Mbyá-Guaraní, and Karo with Puruborá, although there is little evidence for the latter in the grammatical outcome, probably due to the gaps in the Puruborá data.⁷ The other lexical clusters that correspond to the known genealogy of the Tupian languages are not found in the structural analysis. The lexical major split opposing a large Mawé-Awetí-Guaraní branch on one side to all the other branches (cf. Figure 8.2) is not present in the structural analysis, either. A split in terms of Eastern and Western languages (Rodrigues Reference Rodrigues, Cabral and Rodrigues2007) that opposes the language branches spoken in the Rondônian region (Arikém, Ramarama-Puruborá, Tuparí, and Mondé) to the branches spoken outside that region (Mawetí-Guaraní, Juruna, and Mundurukú) does not surface in either the lexical or the grammatical analysis.

Figure 8.3 NeighborNet representation of structural distances among Tupian languages

The output of the structural distance measure does not compare favorably with known genealogical relations. The closely related Juruna branch members Juruna and Xipaya are adjacent in the lexical measure but do not cluster together in the structural analysis. On the other hand, languages that belong to distant genealogical branches such as Makurap (Tuparí) and Asuriní do Tocantins (Tupí-Guaraní) or Mekens (Tuparí) and Yuki (Tupí-Guaraní) show high measures of grammatical similarity. Cocama-Cocamilla and Língua Geral Amazônica constitute another unorthodox cluster in Figure 8.3, sharing fourteen of the twenty analyzed features, including SVO as the basic order of clausal constituents. As the most prominent expansion varieties of Tupian, these two languages played an important role in the contact scenario (see Section 5).

5.1 The material culture of the Tupian expansion

Most attempts to reconstruct the Tupian expansions have relied heavily on pottery as a marker of the pre-Columbian spread of Tupian languages (see e.g. Lathrap Reference Lathrap1970; Brochado Reference Brochado1984; Noelli Reference Noelli, Silverman and Isbell2008; Neves Reference Neves, Hornborg and Hill2011). Lathrap (Reference Lathrap1970) argued that the dispersal of the Arawakan and Tupian families took place from roughly the same point of origin in central Amazonia. He suggested that Arawak-speaking groups were associated with Barrancoid pottery (also known as Incised Rim), while the Tupian groups could be traced through the presence of ceramics associated with the Amazonian Polychrome Tradition.

The Barrancoid tradition originated in the Orinoco Valley around 900 BCE (Cruxent and Rouse Reference Cruxent and Rouse1958, Reference Cruxent and Rouse1959; Sanoja Reference Sanoja1979; Sanoja and Vargas Reference Sanoja and Vargas1983; Barse Reference Barse1989; Oliver Reference Oliver1989; Roosevelt Reference Roosevelt1997; Boomert Reference Boomert2000; Gassón Reference Gassón2002), expanded into the central Amazon by 400 BCE (Heckenberger et al. Reference Heckenberger, Petersen and Neves1999; Rebellato et al. Reference Rebellato, Woods, Neves, Woods, Teixeira, Lehmann, Steiner, WinklerPrins and Rebellato2009), and progressed further into the upper Amazon (Evans and Meggers Reference Evans and Meggers1968: 17, 81; Lathrap Reference Lathrap1970: 109; Brochado and Lathrap Reference Brochado and Lathrap1982: 12) and southern Amazonia (Lathrap Reference Lathrap1970: 159; Heckenberger Reference Heckenberger2005: 56; Saunaluoma Reference Saunaluoma2010: 94) during the centuries around 1 CE (see Eriksen Reference Eriksen2011 for an inclusive account of the Barrancoid expansion). The timing and expansion of the Barrancoid tradition correlates strongly with the Arawakan linguistic dispersal, supporting their association (Eriksen and Danielsen, this volume).

The historical record on a correlation between Tupian languages and Amazonian Polychrome pottery is less straightforward. The association between this pottery style and Tupian groups originated because the first Europeans who entered Amazonia found polychrome pottery among several indigenous groups of the Tupí-Guaraní branch, including the Omagua, Cocama, and Cocamilla of the upper Amazon (Salazar Reference Salazar, Silverman and Isbell2008: 264), the Tupinambá of the Atlantic coastline south of the mouth of the Amazon (Brochado Reference Brochado1984: 283–297), and the southern Tupí-Guaraní groups from the Sao Paulo area to the Andean foothills of Bolivia, among whom a variant of the Polychrome tradition with corrugated decoration was widespread (Howard Reference Howard1947; Métraux Reference Métraux and Steward1948b: 411). The earliest confirmed datings¹² of the tradition are from the Marajoara phase of Marajó Island at the mouth of the Amazon (Meggers and Evans Reference Meggers and Evans1957; Roosevelt Reference Roosevelt1991; Schaan Reference Schaan, Silverman and Isbell2008), from around 300 CE (Brochado and Lathrap Reference Brochado and Lathrap1982: 51), but dates closer to 750 CE are more frequent (Boomert Reference Boomert, Delpuech and Hofman2004: 259).

Around 500 CE, polychrome decoration also began to develop in the central Amazon, alongside ceramics of the Barrancoid tradition and later of the Paredão phase (700–1200 CE). These are two ceramic styles associated with Arawak-speaking groups (Neves Reference Neves, Hornborg and Hill2011: 48), who by that time were integrated into a vast regional exchange system encompassing large parts of the northern South American lowlands (Hornborg Reference Hornborg2005; Eriksen Reference Eriksen2011; Neves Reference Neves, Hornborg and Hill2011: 41). Arawak-speaking groups dominated sociocultural exchange in Amazonia during this period, demonstrating status through different but interrelated ceramic styles. The complex historical record of Amazonian Polychrome pottery illustrates the multifaceted character of its association with potentially multiple ethnolinguistic groups.

The Arawakan expansion began in the Orinoco area around 900 BCE (Eriksen Reference Eriksen2011) and continued along the Amazon River, via the Madeira onto the Llanos de Moxos, and further into the upper Xingu between 500 and 1 BCE. The vast Arawakan linguistic and cultural dispersal was integrated through a regional exchange system that circumscribed the Tupian family in Rondônia on all sides but the east, so when the Tupian family slowly started to expand, it was indeed eastward (cf. Noelli Reference Noelli, Silverman and Isbell2008).