5 Language Documentation & Conservation Special Publication No. 13 (July 2017) Documenting Variation in Endangered Languages ed. by Kristine A. Hildebrandt, Carmen Jany, and Wilson Silva, pp. 94+115 http://nlfrc.hawaii.edu/ldc/ http://handle.net/24750 Three speakers, four dialects: Documenting variation in an endangered Amazonian language Amalia Skilton University of California, Berkeley This paper offers a case study on dialect contact in Máíh!k̃ i (Tukanoan, Peru), with the goal of illustrating how documentation of variation can contribute to a general language documentation project. I begin by describing the facts of variation in one dialectally diverse Máíh!k̃ i-speaking community. I then argue that the outcomes of dialect mixing in this speech community can be understood only through a fine-grained analysis centering the dialectal composition of the communities of practice to which speakers belonged in early life. The coarse-grained identity categories used in most variationist analyses, such as age and gender, are less informative. After proposing a network theory interpretation of this finding, I discuss its implications for the role of (a) ethnography and (b) the European dialect mixing literature in research on variation in endangered languages. Second, I describe some surprising similarities between this speech community and those described in classic variationist literature. Like urban English speakers, Máíh!k̃ i speakers attach less indexical value to morphosyntactic than to phonological variation, and – although their language lacks a standard – engage in indexically motivated style-shifting. I discuss ways to adapt variationist methods to endangered language settings to capture these phenomena, then close with comments on the importance of documenting variation for conservation. 1. INTRODUCTION. This paper aims to illustrate some of the contributions which documentation of inter-speaker variation can make to a general language documentation project.1 Drawing on my own experience conducting fieldwork on variation in a critically endangered language, I argue here that studying variation is valuable – and in some cases, indispensible – to achieving the most common goals of language documentation projects. Where documentation materials will be used for language conservation activities, documentation of variation is necessary for the materials to accurately represent the internal diversity of the language. Likewise, where documentation will be the basis for 1 I owe endless thanks to all of the Máíhuna people who have shared their language with me, especially Féderico, Soraida, and Pedro López Algoba; Otilia López Gordillo; Trujillo Ríos Díaz; and Adriano Ríos Sanchez: m!́sárè góásàgànìkò bàìdʒì. For feedback, I thank Lev Michael; Christine Beier; Stephanie Farmer; audiences in Honolulu and Berkeley; the other participants in the “Documenting Variation” session, especially Sali Tagliamonte and the editors of the volume; and an anonymous reviewer from LD&C. I also thank Leanne Hinton and Justin Spence for thoughts relevant to the conclusion, and the Vicariato San José del Amazonas for logistical support during fieldwork. This research was supported by a Parker Huang Undergraduate Research Fellowship from Yale University, an NSF Graduate Research Fellowship, and NSF BCS-1065621. All errors and omissions are my own. CC Licensed under Creative Commons Attribution Non-Commercial No Derivatives License ISBN: 978-0-9973295-0-6 Three speakers, four dialects: Documenting variation in an endangered Amazonian language 95 grammatical description, variation can be crucial in weighing diachronic and synchronic explanations for generalizations (Rice 2006: 402-403). Yet, for several reasons, general documentation projects do not examine variation as often as they might (though see Rice 1989, ch. 4, and Campbell 2014, ch. 6, for two exceptions). It is unusual for students to be trained in both cross-linguistic descriptive fieldwork and sociolinguistics or dialectology. Field methods textbooks (e.g. Samarin 1967, Vaux & Cooper 1999, Bowern 2008) and language documentation literature (e.g. Austin & Sallabank 2011, Chelliah & DeReuse 2011) say relatively little about how to investigate and describe variation. And a survey of reference grammars suggests that many descriptive and documentary projects take much of their data from a small group of speakers who display only modest variation. Moreover, research on variation in endangered and poorly described languages faces particular obstacles. It is difficult to record casual speech, showing the full range of variation, from rusty or self-conscious speakers. Sociolinguistic variation in a non- industrial society may not follow the same social patterns that are highlighted in the sociolinguistic literature on the US and Europe. The linguist may be unfamiliar with the sociolinguistic variables, or may not expect variation in some areas of the grammar. Or, especially in the early stages of fieldwork, language ideologies may conceal variation from the linguist entirely. And while sociolinguists working in industrial societies have devised methods to address some of these problems, many common sociolinguistic research methods – for example, interview modules which involve reading passages or wordlists – are inappropriate in endangered language settings. Sociolinguistics has thus continued to focus primarily on varieties of English and other Indo-European languages, even as other subfields of linguistics have taken a strong cross-linguistic turn. As a consequence of this focus, the cross-linguistic and cross- cultural applicability of many sociolinguistic (and therefore also historical linguistic) generalizations remains to be tested (Stanford & Preston 2009; Stanford 2008, 2009). Against this background, this paper describes the methods which I used to carry out sociolinguistic documentation of Máíh!k̃ i, a severely endangered Amazonian language, in a field situation with massive variation produced by dialect mixing. My goals are two. First, I sketch some of the differences in the structure of variation between one Máíh!k̃ i-speaking community and some speech communities described in well- known sociolinguistic literature (such as Trudgill 1974, Milroy & Milroy 1985, Eckert 1988, or the studies summarized in Labov 2001). Like many communities where language documentarians work, the speech community which I discuss is small, non- industrial, and has been displaced from its traditional lands. My description of variation in this setting will therefore, I hope, give other documentarians a sense of where – in which linguistic features, and among which speakers – they may encounter variation at their own field sites. Second, I illustrate how general language documentation projects can incorporate documentation of variation, including by adopting methods from variationist sociolinguistics. The paper is organized as follows. The following section (§2) describes the speech community and language of study. §3 gives a case study of four sociolinguistic variables in this speech community, first summarizing the facts of variation (§3.1) and then arguing for an ethnographically grounded analysis of the social patterning of the variation (§3.2, §3.3). §4 discusses ways in which variation in Máíh!k̃ i departs, or DOCUMENTING VARIATION IN ENDANGERED LANGUAGES Three speakers, four dialects: Documenting variation in an endangered Amazonian language 96 interestingly fails to depart, from the kinds of variation discussed in the sociolinguistic literature cited above. These include the dominance of morphosyntactic over phonological variation (§4.1) and the presence of indexically motivated style-shifting (§4.2). §5 presents the methods which I used to integrate collection of data on and analysis of this variation into a general documentation project. In §6, I summarize and argue for the importance of studying variation to both language documentation and conservation generally. 2. BACKGROUND: THE LANGUAGE AND FIELD SITE. Máíh!k̃ i (Western Tukanoan, ISO-639: ore; also known by the pejorative terms ‘Orejón’ and ‘Coto’) is the traditional language of the Máíhuna ethnic group, an indigenous people of northwestern Amazonia. As shown in Figure 1, Máíhuna people traditionally occupied the area between the Napo and Putumayo Rivers, located in the extreme northeast of contemporary Peru. They subsisted by manioc horticulture, wild-gathering, hunting, and fishing. Unlike some well-known Tukanoan peoples of the Vaupés area (Sorensen 1967, Hugh-Jones 1979, Jackson 1983), and like other Western Tukanoan peoples, Máíhuna people traditionally practiced clan exogamy and language group endogamy, and did not stigmatize code-switching or lexical borrowing. FIGURE 1: Map of Máíhuna settlement areas by dialect. Inset shows location of Maihuna area within Peru. Base map by Farmer (2015: 5). DOCUMENTING VARIATION IN ENDANGERED LANGUAGES Three speakers, four dialects: Documenting variation in an endangered Amazonian language 97 The Máíh!k̃ i language is severely endangered, with ~75 speakers in the ethnic Máíhuna population of ~500. The speakers, most of whom were born before 1965, are scattered across five widely separated settlements, all but one within the traditional territory. All Máíh!k̃ i speakers speak at least some regional Spanish, and for most, Spanish is the dominant language of daily interaction. No children have fully acquired Máíh!k̃ i since the late 1970s. Máíhuna people distinguish three mutually intelligible dialects of Máíh!k̃ i, which they typically describe in terms of lexical differences. While speakers tend to refer to dialects by the names of the river basins where they are spoken, I refer to them here as ‘Northern,’ ‘Eastern,’ and ‘Western.’ Figure 1 shows the territory traditionally associated with each dialect. In 2014, the Northern dialect, associated with the Algodón River (a tributary of the Putumayo River), had 12 fluent speakers. The Eastern dialect, associated with the Sucusari River (a tributary of the Napo River), had 15 speakers. All of the other ~50 speakers of Máíh!k̃ i spoke varieties of the Western dialect, associated with the Yanayacu River (a tributary of the Napo). Historically, there was a fourth dialect, closely related to the Eastern dialect, associated with the Ampiyacu and Apayacu rivers (tributaries of the Amazon). This variety has never been documented and may no longer be spoken. I conducted the fieldwork described here between January and June 2014 in San Antonio del Estrecho, a multi-ethnic town of ~3,000 people located on the Putumayo River (a tributary of the Amazon), which forms the border between Peru and Colombia. Estrecho, as it is called, is one day's walk overland from the traditional Máíhuna settlement area on the Algodón River. At the time of my research, ~70 Máíhuna people, representing at least three-quarters of the total Máíhuna population of this region, had left the Algodón and come to live in Estrecho. Their migration formed part of a long, colonially mediated process in which Máíhuna people in the Putumayo basin have shifted from living in mobile bands in remote interfluvial areas, to living in fixed settlements on or near the main course of the Putumayo (Skilton 2014: 4-6.) My goal in working with the former Algodón speech community was to produce a grammatical sketch of Northern Máíh!k̃ i for a polylectal descriptive grammar, since the Northern dialect was the most endangered and least described of the varieties of Máíh!k̃ i.2 Local speakers, aware of differences between their speech and that of other dialect areas, supported this goal. Yet within a few weeks of my arrival, it became clear that there was no invariant ‘Northern dialect’ to take as the object of grammatical description. 3. THE SOCIAL PARAMETERS OF VARIATION. Máíh!k̃ i speakers born in the Algodón form a small and tight-knit speech community, but exhibit exuberant variation. Speakers who are closely related and have been coresident for most of their lives, such as siblings, display different reflexes of many sociolinguistic variables. In this section, I summarize the facts of phonological and morphological variation in Northern Máíh!k̃ i (§3.1) and describe the social situation which gave rise to this variation (§3.2). My account of the patterning of variation focuses on the life histories 2 SIL members published short descriptions of the Eastern dialect in the 1960s and 1970s (Velie 1975, Velie & Velie 1981). My collaborators Lev Michael, Christine Beier, and Stephanie Farmer had carried out several months of fieldwork on the Western dialect (Michael et al. 2014; Farmer 2015). DOCUMENTING VARIATION IN ENDANGERED LANGUAGES Three speakers, four dialects: Documenting variation in an endangered Amazonian language 98 and kin relations of speakers, rather than on their memberships in corporate groups3 such as age cohorts, genders, or clans. These more familiar sociolinguistic factors, I show in §3.3, capture almost none of the variation found in Northern Máíh!k̃ i. By this, I do not mean that different corporate groups are relevant in this setting – for instance, that socioeconomic status operates differently there (cf. Rickford 1986) or that clan is important and class is not (Stanford 2009). Rather, I aim to show that corporate group memberships in general are less relevant, and the details of life history more relevant, to analysis of variation in Máíh!k̃ i. 3.1 The facts of variation. Table 1 shows the values of four sociolinguistic variables for six Máíh!k̃ i speakers born in the Algodón basin, a place understood by speakers as forming a single settlement area. Two variables, (VHV) and (KʷA), are phonological – they represent sound changes which, at some point, affected the entire lexicon. The other two, (3SG.PST) and (go.1SG.PST), are morphological – they represent analogy-driven change in the form of specific morphological constructions.4 Both the phonological and the morphological variables are categorical, in the sense that I never observed forms intermediate between the variants shown.5 (VHV): (KʷA): ‘cook.VT’ (3SG.PST.NF/F) (go.1SG.PST) ‘descend’ Soraida gáè kòkò -aɨ/-ao sá-h! ̀ Pedro gáè, gáhè kòkò (text), kʷàkò (list) -aɨ/-ao sáí-b! ̀ Féderico gáè, gáhè kòkò (text), kʷàkò (list) -aɨ/-ao sáí-b! ̀ Trujillo gáè, gáhè kòkò (text), kʷàkò (list) -gɨ/-go sá-h! ̀̃ Adriano gáè kòkò -gɨ/-go sá-h! ̀̃ Otilia gáè kòkò -aɨ/-ao sáà-b! ̀ TABLE 1: Four sociolinguistic variables in Northern Máíh!k̃ i. 3 A ‘corporate group’ is a group of people that shares rights and duties, is relatively stable in membership, and outlasts the deaths or separation from the group of its members. For example, nuclear families, socioeconomic classes, and clans are all potentially corporate groups. Hayden & Cannon (1982: 133-134) discuss this term and its history in more detail. 4 Glossing in this paper follows Leipzig conventions (except that NF is ‘non-feminine’): 1 ‘first person’, 2 ‘second person’, 3 ‘third person’, E ‘Eastern’, F ‘feminine’, N ‘Northern’, NF ‘nonfeminine’, PL ‘plural’ PST ‘past’, SG ‘singular’, VT ‘transitive verb’, W ‘Western’. 5 §3.2.2 provides complete linguistic descriptions of the variables. In brief: the variable (VHV) represents the reflexes of roots with Pre-Máíh!k̃ i (PM) intervocalic *h, exemplified by the verb *gáhè ‘descend, go downriver.’ (KʷA) represents reflexes of roots with PM initial *kʷ, exemplified by *kʷako ‘cook.VT’. Máíh!k̃ i has two verb classes: a large open class, and the ‘i-class,’ a closed class which follows a different inflectional paradigm and displays root allomorphy governed by the first affix right of the root. (3SG.PST) represents the affixes which, on open-class verbs, mark past tense and 3SG subject agreement (forms on the left of the slash agree with non-feminine subjects, forms on the right with feminine subjects). (go.1SG.PST) represents the morphological construction, consisting of a root allomorph and agreement affix, used to mark past tense with (a) 1SG and (b) all plural subject agreement on i-class verbs, such as ‘go.’ DOCUMENTING VARIATION IN ENDANGERED LANGUAGES Three speakers, four dialects: Documenting variation in an endangered Amazonian language 99 Table 1 combines text and wordlist data (cf. §4). While there is intraspeaker variation in the phonological variables (VHV) and (KʷA), the data contains no evidence of intraspeaker variation in the morphological variables (3SG.PST) and (go.1SG.PST). The data in Table 1 exemplifies two general attributes of variation in this speech community: (a) the variants are numerous, and (b) for some variables, variants cannot be categorized into isogloss bundles even at the level of the idiolect. On (a), the six speakers in Table 1 display four different forms for the variable (go.1SG.PST). One of the four variants, sáh!̀ (Soraida), is identical to the form found in Western and Eastern Máíh!k̃ i. Another variant, sáh!̀ ̃ (Trujillo, Adriano) reflects a minor phonological change from this form – nasalization of the final vowel, likely due to rhinoglottophilia. Two variants – sáíb!̀ (Pedro, Féderico) and sáàb!̀ (Otilia) – represent significant morphological innovations relative to the forms with h. On (b), there are only two speakers in Table 1, Pedro and Féderico, who display exactly the same variants. Other pairs, such as Adriano and Soraida, or Trujillo and Pedro, display the same phonological variants, but different morphological ones. The extent of the variation shown in Table 1 is more surprising because the six speakers whose data is shown form a single, close-knit social network. At the time of my research, the speakers listed in Table 1 live in the same location, as they have for most of their lives, and communicate with each other in Máíh!k̃ i every day, though some pairs interact in Spanish as well. In addition, all of the people listed in Table 1 are close kin. Every speaker shown has a first- or second-degree kinship relation with every other one, and the list includes two sets of siblings (cf. Figure 2 below). In the terms of network theory as operationalized for sociolinguistics by Gal (1979) and Milroy (1980), all of the speakers in Table 1 have strong, multiplex ties – by kinship and by coresidence – with one another, and they have relatively few ties to fluent Máíh!k̃ i speakers outside the group. Small, close-knit networks have been claimed to suppress variation (Milroy & Milroy 1985: 364, Lippi-Green 1989: 217-218). But here, variation flourishes among the Table 1 speakers despite the strong and multiplex ties between them. 3.2 The source of variation: dialect contact. The variation shown in Table 1 is the consequence of almost a century of intensive dialect contact between Northern, Eastern, and Western Máíh!k̃ i in the Algodón basin, the traditional Northern Máíh!k̃ i dialect area. Máíhuna oral history indicates that Máíhuna people from the Eastern and Western dialect areas began moving to the Algodón in the 1920s, possibly brought by labor patrons. Migration from other areas to the Algodón accelerated in the 1960s, when a large number of young men from the Algodón began traveling to Eastern and Western Máíh!k̃ i- speaking communities to find wives, then bringing their brides – and, in many cases, the brides’ families – back to the Algodón with them. (This ‘spouse import’ was driven by severe demographic imbalances in the Algodón population. It is not a traditional Máíhuna practice.) Influential studies of koineization in European languages by Kerswill & Williams (2000), Trudgill et al. (2000), Trudgill (2004), and Kerswill & Trudgill (2005) have suggested that, in situations of intense contact between speakers of mutually intelligible dialects, dialect differences will be leveled – leading to the emergence of a uniform koiné – within one to two generations. Yet dialect contact in the social network which I describe here did not lead to the emergence of a koiné, even though it preceded language DOCUMENTING VARIATION IN ENDANGERED LANGUAGES Three speakers, four dialects: Documenting variation in an endangered Amazonian language 100 shift by at least five decades. Rather, each Máíh!k̃ i speaker born in the Algodón during the dialect contact period displays a different mixture of Northern, Western, and Eastern dialect features. In this section, I begin by tracing the kinship and coresidence relations of the speakers shown in Table 1 (§3.2.1). I then suggest, in §3.2.2, that the outcome of dialect mixing for each individual in this population directly reflects the dialectal composition of the gender-, kinship-, and coresidence-based communities of practice (Lave & Wenger 1991) to which he or she belonged in early life. In §3.3, I contrast the explanatory power of this very fine-grained, ethnographic analysis with the power of an analysis that focuses on the kind of coarse-grained identity categories, such as age and gender, that are often prominent in variationist analyses. My evaluations of the analyses in §3.2 and §3.3 are exclusively qualitative, since at the time of my research, only 12 Máíh!k̃ i speakers born in the Algodón were still living – a population too small for statistical analysis to be meaningful. 3.2.1 Kinship and place. Gender, age, and clan membership are central organizing principles in Máíhuna society and traditional ritual life (Bellier 1991). Yet when the speakers in Table 1 talk about the ways that they are different from one another and from other Máíhuna people, they do not treat these categories as most salient. What they see as relevant, instead, are the places with which a person and their kin are associated: where someone was born, where they lived as a child, and where their parents were born. These categories assume central importance in my analysis of variation as well, for it is kinship and place – as indices of the communities of practice to which speakers have belonged – which best explain the variation shown in Table 1. I therefore now summarize the kinship and coresidence relations among the speakers listed in Table 1. Figure 2 displays a genealogical diagram of the Table 1 speakers, overlaid with information about place. Each shape in Figure 2 represents a named individual who is listed in Table 1 or is the parent or spouse of a speaker listed there. Information in Figure 2 was collected in genealogical and life-history interviews with the speakers listed in Table 1 and confirmed, where possible, with missionary records. Squares represent men, circles women. DOCUMENTING VARIATION IN ENDANGERED LANGUAGES Three speakers, four dialects: Documenting variation in an endangered Amazonian language 101 = = = † † † † † † = = = = † Trujillo Adriano Otilia Pedro F´ederico Soraida BorninAlgod´on-Downriver BorninAlgod´on-Upriver BorninWesterndialectregion BorninEasterndialectregion DiedbeforeJune2015 † FIGURE 2: Kin relations among speakers in Table 1. Figure 2 shows that Adriano and Trujillo are siblings, as are Soraida, Pedro, and Féderico. Otilia is married to Adriano and is in a consanguineal relation to Soraida and her brothers through which she is classified as their sister. Because of this classificatory relationship, represented by the curved line, Otilia links the two sets of biological siblings into a single kin network. Soraida and her siblings are entitled to address Otilia as dòìkò ‘sister,’ her husband Adriano as bá!̀ ‘brother-in-law,’ and her children as h!́t̃!̀ ‘nephew’ and h!́t̃ò ‘niece.’ The contrast between blue and cyan shapes in Figure 2 represents historical coresidence relations among the speakers born in the Algodón. Six people here – Soraida and her brothers, whose variants are shown in Table 1, and three deceased people – were born far upriver on the Algodón. They lived in extremely remote locations in the headwaters of the Algodón and in the area between the Algodón and Napo until the early 1950s, when missionaries persuaded them to settle in a village on the more accessible middle Algodón. Otilia was born at this new settlement shortly after it was founded. Two other people on the diagram – Adriano and Trujillo, both also in Table 1 – were born in an extended-family settlement located downriver on Algodón, near the river’s mouth. This settlement, founded by Adriano and Trujillo’s father in the 1920s, dissolved in the mid-1950s. Its members then migrated to the village on the middle Algodón. Finally, the contrast between green, red, and blue or cyan shapes on Figure 2 visually represents the birthplace of each person. Except for Soraida, every Table 1 speaker has at least one first-degree relative – parent or spouse – born outside the Algodón. Adriano and Trujillo’s father was born in the Sucusari basin, where Eastern Máíh!k̃ i is currently spoken. Otilia’s mother was born in the Ampiyacu-Apayacu basin, which is geographically very close to the Sucusari basin, and likely spoke a dialect similar to Eastern Máíh!k̃ i. Pedro and Féderico are married to Western Máíh!k̃ i-speaking women from the Yanayacu basin. DOCUMENTING VARIATION IN ENDANGERED LANGUAGES Three speakers, four dialects: Documenting variation in an endangered Amazonian language 102 3.2.2 Communities of practice. In Table 2, almost all of the variation shown can be understood by reference to the three social factors displayed in the three left columns: (a) the speaker’s place of birth and childhood residence, (b) the dialects spoken by his or her parents (inferred from the parents’ birthplaces, as all of the speakers’ parents are deceased), and (c) the dialect spoken by his or her spouse. Table 2 re-displays a version of Table 1 expanded to include this information, as well as the forms found in the Eastern and Western dialects. Birth Parent Spouse (VHV): (KʷA): (3SG. (go.1SG. Group Dial. Dial. ‘descend’ ‘cook.VT’ PST) PST) Soraida Upriver N, N N gáè- kòkò- -aɨ/-ao sáh! ̀ Pedro Upriver N, N W gáè-, kòkò-, -aɨ/-ao sáíb! ̀ gáhè- kʷàkò- Féderico Upriver N, N W gáè-, kòkò-, -aɨ/-ao sáíb! ̀ gáhè- kʷàkò- Trujillo Downriver E (Sucu- N/A gáè-, kòkò-, -gɨ/-go sáh! ̀̃ sari), N gáhè- kʷàkò- Adriano Downriver N6 N gáè- kòkò- -gɨ/-go sáh! ̀̃ Otilia Upriver E N gáè- kòkò- -aɨ/-ao sáàb! ̀ (Ampi- yacu), N Eastern N/A N/A N/A áhè- kʷàkò- -gɨ/-go sáh! ̀ dial. (Sucusari) Western dial. N/A N/A N/A gáhè- kʷàkò- -gɨ/-go sáh! ̀ TABLE 2: Four sociolinguistic variables and three parameters of variation (3SG.PST). All finite verbs in Máíh!k̃ i bear inflectional suffixes which fusionally expone tense, sentential mood, and the subject features of person, number, and gender (feminine vs. non-feminine). This sociolinguistic variable relates to the segmental form of the affixes which mark past tense and third person singular agreement on the main class of verbs. In Pre-Máíh!k̃ i (PM), the third person singular non-feminine past tense (3SG.NF.PST) affix reconstructs to *-agɨ, and its feminine (3SG.F.PST) counterpart to *-ago. In the contemporary language, there are two reflexes of these forms: -aɨ (NF)/-ao (F), used only by speakers born in the Algodón, and -gɨ (NF) /-go (F), which is used by some speakers in the Algodón and all speakers in the other two dialect regions. Among the Table 1 speakers, place of birth within the Algodón basin has a clear relationship with this variable. Trujillo and Adriano, the two speakers born in the downriver settlement, are the only speakers here to use the -gɨ/-go variants. These variants are also found in Eastern Máíh!k̃ i, the dialect spoken by their late father, the founder of that settlement. All other speakers in Table 1 (and to my knowledge, all speakers born in the Algodón except for the former members of the downriver 6 Adriano and Trujillo, though full siblings, are assigned different values of Parent Dial because their father died when Adriano was approximately two years old and Trujillo, who is at least ten years older, was in his teens. DOCUMENTING VARIATION IN ENDANGERED LANGUAGES Three speakers, four dialects: Documenting variation in an endangered Amazonian language 103 settlement) use the -aɨ variant, which is found exclusively in the Northern dialect region. Trujillo and Adriano’s use of -gɨ therefore likely reflects that their input in acquisition included more Eastern Máíh!k̃ i and Eastern-influenced varieties – that is, their father’s speech and the speech of their older siblings, who would have been the brothers’ main caregivers – than Northern Máíh!k̃ i. The brothers’ variant of (go.1SG.PST), sáh!̀,̃ also closely resembles the form found in Eastern Máíh!k̃ i. This does not mean, though, that Trujillo and Adriano’s speech is Eastern Máíh!k̃ i. In addition to the morphological variables, the Eastern dialect as spoken in its historical territory also displays regular changes of PM root-initial *g > Ø and root-initial *gʷ > b. Adriano and Trujillo’s speech never displays *g > Ø, and it displays *gʷ > b in only one word. Instead, the brothers display – in all styles for Adriano, and in connected speech for Trujillo – the sound changes characteristic of Northern Máíh!k̃ i, root-medial PM *h > Ø and initial *kʷ > k, discussed below. These phonological variants likely reflect that the brothers’ language, and that of their older siblings, was influenced by the Northern Máíh!k̃ i speech of their mother and maternal relatives, who resided with them in the downriver settlement. It is unlikely that Trujillo and Adriano’s Northern variants are due to accommodation to Northern dialect speakers, since – among other members of their community – even extremely long-term accommodation has not led to change in the variants used in connected speech (cf. Labov 1981, which shows that even child migrants to Philadelphia do not fully acquire the Philadelphia short /a/ system). (go.1SG.PST). Like all Western Tukanoan languages, Máíh!k̃ i displays two classes of verbs: a large open class, and the small, closed ‘i-class.’ I-class verbs follow a different inflectional paradigm from the open class. They also exhibit root allomorphy governed by the suffix immediately following the root, varying between allomorphs of the form CVi (e.g. sáí), CV (sá), CVV (sáà), and CVni (sánì). This variable represents the morphological construction – the root allomorph and inflectional suffix – used to mark past tense and (a) 1SG and (b) all plural subject agreement on i-class verbs, which are exemplified here by the verb ‘go.’ In PM, this construction reconstructs as *sá-h!̀, with the i-class root appearing in its CV allomorph and bearing the inflectional suffix -hɨ. All Máíh!k̃ i speakers outside the Algodón basin either preserve the PM form of this variable, sá-h!̀, or display a variant, sá-h!̀,̃ which differs from the PM form only in that the final vowel has nasalized. In the Algodón, there are two more reflexes of (go.1SG.PST): sáà-b!̀, a form which reflects extension of the open-class inflection to the i-class, and sáí-b!̀, which reflects extension of the open-class inflection followed by leveling of the i-class root allomorphy. The social distribution of the reflexes of (go.1SG.PST) is explained by the intersection of place of residence and gender. The four speakers born in the upriver group display three forms for this variable: Soraida and Otilia show sáh!̀ and sáàb!̀, while the two men, Pedro and Féderico, use sáíb!̀. They share this variant, which is innovative relative to sáh!̀ and sáh!̀,̃ with an older, genealogically unrelated man from the upriver group (not shown in the table), Jorge. How did Pedro and Féderico come to use sáíb!̀ when Soraida, their sister, uses sáh!̀? The answer cannot lie in the earliest stages of acquisition – the same caregivers cared for all three siblings as young children, and all three would have encountered near- identical input in early acquisition. The variation also does not represent absolute gender indexicality, since speakers do not shift between the forms in reporting one another’s DOCUMENTING VARIATION IN ENDANGERED LANGUAGES