THE RISE AND FALL OF APPROXIMANTS IN THE TUPARIAN LANGUAGES Текст научной статьи по специальности «Языкознание и литературоведение»

Andrey Nikulinf, Rafael Andrade*

+ University of Brasilia; andre.n.guzman@gmail.com * University of Sao Paulo; rafael.andrade.almeida@usp.br

The rise and fall of approximants in the Tuparian languages *

This paper addresses the evolution of the approximant series in the languages of the Tuparian branch of the Tupian family, native to the region comprised between the middle course of the Guapore/Itenez and the headwaters of the Machado/Ji-Parana (southern Rondonia, Brazil). It is shown that in addition to the approximant series of Proto-Tuparian (which, we argue, comprised *j, *w), some daughter languages created innovative approximants from a variety of sources, such as non-low vowels (*o/*i), post-oralized nasals (*mb/*nd/*rjg, by the way of *b/*d/*g), and hiatus-filling glides. The evolution of these sounds is discussed in great detail; in particular, we argue that at least some approximants have been historically forti-tioned in all Tuparian languages. A special attention is given to the subgrouping of the Tu-parian branch.

Keywords: Tuparian languages; Tupian languages; approximants; fortition; comparative method.

1. Introduction

This paper examines the phonological development of the approximant series throughout the reconstructed history of the Tuparian languages (Tupian family), a group of indigenous languages spoken in what is now the Brazilian state of Rondonia. We will argue that Proto-Tuparian inherited a series of approximants (*j8, *j, *w) from its ancestor, Proto-Tupian, which were later subject to massive fortition processes in the history of all contemporary Tuparian languages. In addition, we hypothesize that some Tuparian languages innovated at some point by creating approximants from two types of Proto-Tuparian sources: non-low vowels (*o, *i) and postoralized nasals (*mb, *nd, and *^g).

The Tupian language family is one of the most diversified and geographically disperse genetic units of South America. Its approximately 50 languages are spoken throughout a vast area which spans from the northern Amazon to the extreme south of Brazil and are classified into ten universally recognized low-level branches: Arikém, Tuparí, Mondé, Ramarama, Pu-ruborá, Mundurukú, Juruna, Sateré-Mawé, Awetí, and Tupí-Guaraní (Rodrigues & Cabral 2012). Recent studies have shown that Ramarama and Puruborá likely constitute a valid clade (Galucio & Gabas Jr. 2002), as do Sateré-Mawé, Awetí, and Tupí-Guaraní (Awetí and Tupí-Guaraní are more closely related to each other than any of them to Sateré-Mawé; Correa-da-Silva 2010, Meira & Drude 2015). From a geographic point of view, the genetic diversity within the family reaches its peak in what is now the Brazilian state of Rondonia, which has therefore been identified as the likely Urheimat of Proto-Tupian (Rodrigues 1958: 683).

The Tuparian branch — also known in earlier literature as Kanoé (Rodrigues 1958: 682), Mekens (Hanke et al. 1958: 188), or Makuráp/Macuráp (Loukotka 1963: 45, 1968: 122) after different members of the branch (both Kanoé and Mekens refer to the language now known as

* We thank an anonymous JoLR reviewer for comments which helped improve the manuscript. Nikulin is grateful to Ana Vilacy Galucio for discussions on the synchronic and historical phonology of the Tuparian languages. Andrade is further grateful to Luciana Storto and Denny Moore for providing original recordings of Makurap. The usual disclaimers apply.

Journal of Language Relationship • Вопросы языкового родства • 18/4 (2020) • Pp. 284-319 • © The authors, 2020

Mekéns or Sakurabiat) — includes the following languages. Wayoró (Glottocode [wayo128], ISO 639-3 [wyr]) is spoken in the Terra Indígena Rio Guaporé by three elderly speakers at the time of writing (Nogueira 2019: 3). Nogueira (2019: 4) also reports lexical and phonological differences between the varieties traditionally spoken by the Kupndiiriat and Ngwayoroiat groups. Tuparí (Glottocode [tupa1250], ISO 639-3 [tpr]) is spoken by 350 individuals in two reservations, Terra Indígena Rio Branco and Terra Indígena Rio Guaporé (Singerman 2018: 1). Mekéns (= Sakirabiat, Sakurabiat; Glottocode [saki1248], ISO 639-3 [skf]) is spoken by 14 individuals in the Terra Indígena Rio Mekéns (Galucio & Nogueira 2018: 96).1 It is subdivided into three dialects, including Sakurabiat/Guarategayat, Guaratira, and Siokweriat (= Kampé, now spoken by just one individual). Akuntsú (Glottocode [akun1241], ISO 639-3 [aqz]) is spoken by three individuals near the Omere creek (Aragon & Tavares 2019). Makurap (Glottocode [maku1278], ISO 639-3 [mpu]) is spoken by ca. 50 individuals in the Terra Indígena Guaporé (Galucio & Nogueira 2018: 96).

Until recently, the languages of the Tuparian branch had remained severely underdocu-mented. Tibor Sekelj documented short wordlists of Tuparí and Makurap during his 1948 expedition to the Rio Branco (Sekelj 1948). Emil-Heinrich Snethlage traveled around the region in 1933-4 and made notes on all Tuparian languages except Akuntsú (Snethlage 2015). Wanda Hanke visited the Mekéns in 1949 and also made some notes on the language (Hanke et al. 1958). Franz Caspar stayed with the Tuparí for several months in 1948 and 1955; based on his fieldnotes, a grammar sketch was prepared in 1958 (translated into Portuguese and published as Rodrigues & Caspar 2017).2 Fortunately, the situation has improved drastically over the last 30 years due to a documentation boom in Amazonian linguistics. The following recent works have been prioritized as primary sources of lexical data used in this study. For Wayoró, we rely on the works by Nogueira (2011, 2015, 2019). For Tuparí, we give preference to Singer-man's (2018) dissertation and to Alves's (2004) dictionary (especially when it comes to the position of the stress). As for Mekéns, Galucio's (1994, 2001, 2002, 2011a,b, 2014; Galucio et al. 2017; Alves & Galucio 2007) works have been consulted for the Sakurabiat and Guaratira dialects, whereas for the Siokweriat dialect the short appendix in Aragon (2014) was used. This latter work has also been our primary source for Akuntsú, though earlier works by the same author (Aragon & Carvalho 2007, Aragon 2008) as well as Gabas Jr. (2005) were also consulted. Finally, for Makurap we rely on Braga (1992, 2005) as well as on unpublished recordings by Denny Moore (collected in 2003 with the help of the consultant Alcides Makurap). In addition, lexical material has been extracted from the comparative works on Tuparian (presented below) whenever the relevant forms are not attested in our primary sources. For kinship terms in all Tuparian languages, we rely on Nogueira et al. (2019).

To this moment, however, few works have been dedicated to the phonological reconstruction of Proto-Tuparian.3 Moore & Galucio (1994) offer a pioneering proposal of the segmental

1 Although the language has been increasingly referred to as Sakurabiat in recent literature, here we reserve the label Sakurabiat for the dialect spoken by the Sakurabiat and Guarategayat groups. In contrast, the label Mekéns is used in a broader sense throughout this paper and covers the varieties spoken by the Guaratira and the Siokweriat (in addition to the one spoken by the Sakurabiat and the Guarategayat).

2 Other premodern sources on specific Tuparian languages include Anonymous (n/d, on Wayoro, apud Lou-kotka 1963: 46-7), Becker-Donner (1955, on Mekéns, apud Loukotka 1963: 48), Xerez (1946, on Makurap), and Lévi-Strauss (n/d, on the Kabisiana variety of Mekéns, apud Loukotka 1963: 48 and Lévi-Strauss 1950; see Nikulin submitted for the identification of Kabisiana as a Mekéns variety). We were unable to access any of these works.

3 Aragon & Cabral (2005) and Gabas Jr. (2005) also discuss the genetic relations within the Tuparian family (with special attention to the position of Akuntsu), but no claim is made with respect to the phonological reconstruction. Galucio & Nogueira (2018) reconstruct the evolution of the object focus construction in the Tuparian

phonology of Proto-Tuparian, which is based on a total of 124 cognate sets representing Way-oro, Tupari, Mekens, and Makurap; the respective reconstructed forms are also provided. Galucio & Nogueira's (2012) work by and large reinforces Moore & Galucio's (1994) reconstruction, differing from it mainly in that (i) the segment *nd(z) is removed from the reconstructed inventory; (ii) the phonological status of *b as a contrastive segment, treated as uncertain in Moore & Galucio (1994), is confirmed; (iii) the data of a fifth Tuparian language, Akuntsu, are taken into account. Furthermore, Galucio & Nogueira (2012) argue that the Proto-Tuparian segment *D (the ad hoc symbol used in Moore & Galucio 1994) should be interpreted as a voiced denti-alveolar stop *d. Galucio & Nogueira (2012) also address the reconstruction of aspects of Proto-Tuparian morphosyntax, including the person inflection and the morphosyntactic alignment, as well as derivational morphology. Due to the nature of the publication, the segment reconstructed by Galucio & Nogueira (2012) as *d is the only one to be supported with detailed discussion and examples. The cognate sets that were used to substantiate the reconstruction of all other segments are not presented. That way, the only published work on the phonology of Proto-Tuparian in which the reconstructed phonemes are illustrated with actual linguistic data is Moore & Galucio (1994), which predates the documentation boom of the Tuparian languages by a large margin.

An in-depth study of the historical phonology of the Tuparian group, besides being an interesting subject by itself, is crucial for our understanding of the diachronic development of the entire Tupian family (cf. Galucio & Nogueira 2018: 95). Although there have been pioneering attempts at a phonological reconstruction of Proto-Tupian (Rodrigues 2002, 2005, 2007), most subgroups of Tupian have been represented in them by one single contemporary language (Tupari for the Tuparian group, Munduruku for the Munduruku group, Yudja for the Juruna group) rather than by the respective intermediate proto-languages (with the notable exception of Proto-Tupi-Guarani). In other words, the comparative method has never been consistently applied to the Tupian family in a bottom-up manner. The situation, however, is likely to change in the near future, thanks to several recent and ongoing detailed, methodologically sound diachronic studies of low-level branches of Tupian (most recently Meira & Drude 2015 for Proto-Mawe-Guarani, Carvalho 2019 for Proto-Juruna, Picango 2019 for Proto-Munduruku, Carvalho forthc. for Proto-Tupi-Guarani). In this sense, this paper aims to contribute to the emergent field of diachronic Tupian studies in general by reconstructing parts of the consonantal system of Proto-Tuparian in some detail.

The International Phonetic Alphabet is used for representing data in this paper, with the following exceptions. The symbols r, ¡¡, d, y, and e stand for [r, §, 5, e], respectively. The coda consonants are considered to be underspecified for features other than place of articulation in all Tuparian languages (cf. Singerman 2016 for Tupari). Underspecified labial, dental/ alveolar, palatal, and velar consonants in coda are represented in small caps: P, T, C, k (cf. a similar analytical decision for another Tupian language, Aweti, in Drude 2009). The acute accent denotes stress in Tupari and Akuntsu and high tone in Makurap.4

languages. Nogueira et al. (2019) is an in-depth study of the Tuparian kinship terms, which also includes reconstructed forms; the phonological reconstruction in this work does not differ substantially from that of Galucio & Nogueira (2012).

4 The evidence for two level tones in Makurap (high and low) comes from our preliminary analysis of the Makurap recordings by Moore, wherein most words are conveniently accompanied by their whistled equivalents (cf. Moore & Galucio 1994: 122). The high tone occurs at most once in polysyllables, and its position interacts with morphology in ways that are currently poorly understood (e.g. pario [peH?i6L] 'hawk' ^ pario-cato [pBLri6HcaLlt:oL] 'harpia'). We do not mark the low tone explicitly. Makurap tokens taken from printed sources which do not transcribe the tonal distinctions (such as Braga 1992, 2005) are underlined.

The remainder of this paper is structured as follows. In section 2, we present some evidence for the subgrouping of the Tuparian group accepted in this paper. In section 3, we present the comparative evidence which supports the reconstruction of the approximant series for Proto-Tuparian as well as for the proto-languages of shallower genetic units, such as Proto-Core Tuparian and Proto-Corumbiara, and for an earlier stage of Wayoro. Specific sound changes required by our proposal are summarized in section 4. We conclude by a succinct discussion of our findings in section 5, followed by a list of abbreviations used in this paper.

2. Internal classification of the Tuparian group

In this section, we present evidence for a specific proposal regarding the subgrouping of the Tuparian group. Namely, we claim that (i) Wayoro and Tupari form a subgroup to the exclusion of other languages ("Wayoro-Tupari"), (ii) Mekens and Akuntsu likewise form a subgroup to the exclusion of other languages ("Corumbiara"), and (iii) all the aforementioned languages form a clade ("Core Tuparian") to the exclusion of Makurap.

Makurap vs. Core Tuparian. The claim regarding the binary split of Proto-Tuparian into Makurap and Core Tuparian has found extensive support in a number of published works (cf. the lexicostatistical assessment in Galucio & Nogueira 2012, Galucio et al. 2015: 238), even though little space has been allocated so far to the identification of innovations shared by the Core Tuparian languages. The most characteristic of them are listed below.

One such innovation appears to have affected the third person inflection pattern of the */j/-initial stems. In Makurap, a significant number of stems inflect for the third person by replacing their initial consonant (c- in oral environments; p- is nasal environments) with another consonant (t- both in oral and nasal environments), as in ceK 'house.POSS', pac 'tooth' ^ t-eK 'his/her house', t-ac 'his/her tooth' (Braga 2005).5 A plausibly cognate pattern is found in Tupian languages outside the Tuparian group, such as Munduruku (ddk-?a, nSj ^ t-dk-?a, t-Sj; Picango 2005), Kuruaya (l—> t-), Satere-Mawe (s—> h-), and most Tupi-Guarani languages

(*t-/*-r--> *ts-). This allows us to project the pattern attested in Makurap onto the Proto-

Tuparian level (we reconstruct PTpr *j-/*p—> *c-, where *p is the nasal allophone of */j/). All other Tuparian languages lost the archaic prefix *c- and now use reflexes of PTpr *i- in this function. For example, the third person of PTpr *ja?iP 'son, fraternal nephew (male ego)' (> Makurap caiP) is reconstructed as *c-a?iP (> Makurap t-aiP). In Proto-Core Tuparian, the unpossessed form yielded *da?iP (> Wayoro ndauP, Tupari ha?uP, Mekens/Akuntsu taiP); see 3.1 for PTpr *j > Proto-Core Tuparian *d. However, the third person form was not preserved as *c-a?iP but rather was substituted with *i-da?iP (> Tupari i-a?iP, Mekens i-taiP, etc.). The allo-morph *i- must have been extended through analogy from other consonant-initial stems.

Another clear innovation that identifies Core Tuparian as a valid genetic unit is the nasalization of the stops *p and *t in syllables with nasal nuclei, as shown in Table 1.

In turn, Core Tuparian is subdivided, in a binary manner, into Mekens-Akuntsu and Wayoro-Tupari. The former claim seems to be universally accepted (Gabas Jr. 2005; Galucio & Nogueira 2012), as Mekens and Akuntsu are remarkably close to each other and are reported to be mutually intellegible (Galucio et al. 2015: 237-8 even suggest that they are "co-dialects of the same language"). We propose the label Corumbiara for the clade which comprises Mekens

5 Braga (1992, 2005) transcribes the palatal obstruent of Makurap as tf. In Moore's field recordings of Makurap, we found its pronunciation closer to [c] or [cç]. Throughout this paper, it will be represented as c.

PTpr gloss Wayoro Tupari Mekens Akuntsu Makurap

*paT- to be tied — paT-

*paT-?a- to tie mara- para-

*pi:T hummingbird miT mi:T 'penis' mi:T — mi:T pi:T

*arapira woman aramira aramira aramira aramira arapija6

*-pia7 knee (ku}mia mia(K-?a} (nekiwa}mipa 'elbow'8 (a}mina (ka}pia

*pari(o) harpia — poT?a-mari — — pari(o}

*ti9 ashamed ni- ni- — — —

*atiP head — anuP 'brain' aniP anaP attP 'hair'

*tiK-10 to weave niK- niK- ni-a ni-a tiK-

*iiK11 spotted niK~niK niK — niK 'striped' —

*ti(:)K timbo vine ni:K ni(:)K — — tiK

*atiK12 worm aniK aniK 'leishmaniasis ulcer' Sio aniK aniP atiK

Table 1. Nasalization of *p, *t in the Core Tuparian languages in nasal environments

and Akuntsu. As for Wayoro and Tupari, the special proximity between these two languages has been suggested in Galucio et al. (2015) based on an application of two distance-based algorithms to the 100-word Swadesh lists of the Tupian languages (83.7% confidence rate), but this result was not replicated for other datasets considered in the cited work. In what follows, we identify several shared innovations which support the validity of both branches (Corumbiara and Wayoro-Tupari).

Corumbiara. The Corumbiara languages (Mekens and Akuntsu) share multiple lexical innovations which are unique to these two languages. The following are some examples thereof: PTpr *jge 'garden' is replaced with Mekens/Akuntsu tabiT 'garden'; PTpr *jgitaK ~ *jgitaK 'night' is replaced with *matso (Mekens matso(pi}, Akuntsu matfo); PTpr *mapi 'manioc' is replaced with *taPjVT (Mekens taPtsiT, Akuntsu taPtoT); PTpr *ekiP 'arrow' is replaced with *mapi

6 The correspondence between Core Tuparian *r and Makurap j is not known to be regular.

7 Hereinafter, the curled brackets denote material which is deemed not to be cognate despite not being de-monstrably segmentable in the contemporary languages.

8 The form is tentatively phonologized based on Snethlage's (2015: 518) attestation of <kina kiwaminja> 'Ellbogen' (likely the first person inclusive ki-(nekiwa}mipa).

9 This root appears to have been lost in Makurap, unless metia 'ashamed' (Braga 2005: 191) is somehow related. A voiceless dental stop is reconstructed in light of the external cognates (Proto-Mawe-Guarani *ti 'ashamed', Meira & Drude 2015: 292).

10 In Mekens and Akuntsu, the only attested forms of these verbs contain the theme vowel -a, which triggers the deletion of the stem-final -K. The underlying stem is expected to have the shape niK- in both Corumbiara languages, but the forms that could prove it have not been attested in the published works.

11 No cognate is attested in Makurap. A voiceless dental stop is reconstructed in light of the external cognates, such as Satere-Mawe tiK 'spotted' (Ribeiro 2010: 87).

12 The Mekens reflex is attested as atiK in Galucio et al. (2015: 266) but not in other sources on the language that we consulted. If the existence of this form is confirmed, it could be explained as a borrowing from Makurap atiK or from Karo atij, attested ibidem (the Makurap reflex is given there as ati, which must be a mistranscription, cf. Braga 2005: 184 and Moore's field data). Note that Akuntsu -P does not regularly continue PTpr *-K.

'arrow' (likely from Kwaza madi or Kanoe mapi; cf. Voort 2005: 386). Some further examples of lexical isoglosses specific to the Corumbiara languages are *tsaro 'yellow', *pi(:)K 'black', and *kicpiT 'fish', though we have been unable to provide an unequivocal Proto-Tuparian reconstruction for these specific concepts. In addition, the Corumbiara languages share multiple phonological innovations, some of which are exclusive to this subgroup (e.g. PTpr *t/*nd > *ts > Mekens ts, Akuntsu tf; PTpr *j/*c > *t > Mekens t, Akuntsu t; PTpr *i(?)V > *ijV > Sakurabiat itsV, Guaratira/Siokweriat iV, Akuntsu itV). Most phonological innovations that characterize the languages of the Corumbiara branch will be discussed in more detail in section 3.

Wayoro-Tuparf. Wayoro and Tupari are not as tightly related to each other as Mekens and Akuntsu, but nevertheless clear innovations shared exclusively by these two languages can be identified. For example, PTpr *mbo-ape 'fingernail' and *ojaT 'fire', whose reflexes are found in Makurap, Mekens, and Akuntsu, are replaced with Proto-Wayoro-Tupari *kmpa 'fingernail' and *akoP-k-aP 'fire', respectively (the latter is evidently an *-aP nominalization from a verbal derivative of *akoP 'hot'). Wayoro and Tupari are also unique in that they have high central rounded vowels /u u/ (Alves 2004: 41; Singerman 2016: 456; Nogueira 2019: 10), which correspond to /i i/ in Mekens, Akuntsu, Makurap, and many Tupian languages outside the Tuparian branch. In both languages, /u i/ pattern together in that they make up the environment for at least one phonological process (the diachronic assibilation *t > s _/u,i/ in Tupari; the morphophonological dissimilation /e/ ^ a _/u,i/ in Wayoro, cf. Nogueira 2015).

3. Proposal

In this section, we present the evidence which supports the reconstruction of the approximant series in Proto-Tuparian (3.1). We will also show how innovative approximants arose from various sources through multiple independent innovations in individual Tuparian languages (from non-syllabic vowels, 3.2), in early Wayoro (from postoralized nasals, 3.3), and in Proto-Corumbiara (3.4, as hiatus-filling glides). Note that in the contemporary Tuparian languages most of the segments under discussion have changed to some other sounds (either through fortition or through lenition). For example, PTpr *eji 'marico bag' and *wawo 'sweet potato' are reflected as Wayoro endu, jgwago; Tupari eu, wao; Mekens eti, kwa(:)ko; Akuntsu eti, kwako; Makurap eci, fiafid. For now, these examples should suffice to give the reader a flavor of the trends in the evolution of the approximants in the individual histories of the Tuparian languages. A detailed discussion thereof is deferred to section 4.

Note that in this paper we do not discuss the consonants that arose from combinations of an underspecified consonant in the coda position followed by an onsetless or a ?-initial syllable. In these environments, codas are commonly resyllabified as onsets in the contemporary Tuparian languages, which is usually accompanied by lenition (Braga 1992: 63-4 for Makurap; Galucio 1994: 991-2, 2001: 23 for Mekens; Singerman 2018: 372-3, 378-80 for Tupari, among others). An investigation of the approximants that may have developed in some languages through resyllabification of codas lies beyond the scope of this paper.

In Table 2, we summarize our proposal regarding the development of the Proto-Tuparian onsets in oral environments. Note that *[mb], *[nd], and *[qg] are held here to be allophones of underlying */m n q/ in oral environments (see 3.3).

In Table 3, we summarize our proposal regarding the development of the Proto-Tuparian onsets in nasal environments. Note that *[p] is held here to be the nasal allophone of an underlying */j/.

PTpr Mak PCT pre-PCor Mek/Aku PWT Tup pre-Way Way

*p p *p *p p *p p s-/-ps-A *p p

*[mb] [mb] *b *b/*p *p [mb]-/-p-

*P p *p *b b

*t t *t *t Mek ts, *t t, sB *t t

*[nd] [nd] *d Aku tj" *d [nd]

*j c *d t h-/-0-

*c t *c *c *c 0 *tj tj

*k k *k *k k *k k *k k

*[qg] fog] *g *g *Y fog]

*w P *w, *yC *gw, *gC kw, kC *w, *yC w, 0C *w, *yc fogW], [qg]C Kup p

*r r *r *r r *r r *r r

*? ? ~ 0 *? *? ~ *0 ? ~ 0 *? ? *? ~ *0 ? ~ 0

A = before i; B = before i or «; C = before a rounded vowel

Table 2. Proto-Tuparian onsets and their reflexes in oral environments

PTpr Mak PCT pre-PCor Mek/Aku PWT Tup pre-Way Way

*p p *m *m m *m m *m m

*m m

*t t *n *n n *n n *n n

*n n

*fo] n *n n n *n n, 0A *n n

*c t ?13 ? ? ? ? ? ?

*k k *k *k k *k k *k k

*qg qg *g *g *Y q

*w m *w, *yb *gW, *gB kw (Sak qw), k (Sak q)B *w, *yb w, 0B *w, *yb qw, qB

*r r *r *r, *nC r, nC *r r *r r

*? ? ~ 0 *? *? ~ *0 ? ~ 0 *? ? *? ~ *0 ? ~ 0

A = before i; B = before a rounded vowel; C = between front vowels

Table 3. Proto-Tuparian onsets and their reflexes in nasal environments

13 Although PTpr *c certainly occurred in nasal environments (as in *mic5 'curassow', *c-aC 'his/her tooth',

*c-eri 'his/her hammock' > Mak mito, t-aC, t-eri), we have not identified reflexes of any of these word forms in any Core Tuparian language.

A note on the reconstruction of PTpr *t

Although a detailed discussion of the reconstruction of Proto-Tuparian voiceless segments is beyond the scope of this paper, we deem it appropriate to briefly comment on our interpretation of the sound correspondence between Wayoró t, Tuparí t (s _/i,u/), Mekéns ts, Akuntsú tf, and Makurap t. Galucio & Nogueira (2012) claim that the correspondence set in question "clearly reconstructs as the affricate *ts, which becomes [+palatal] in Akuntsú, and loses the feature [+sibilant] in Wayoro, Makurap and Tupari, except before [i] in Tupari".14 We believe that the reconstruction should be amended to *t for four reasons.

(i) First of all, reconstructing *ts would imply an innovation (*ts > t) shared by Makurap, Tuparí, and Wayoró, though Makurap is not known to be closely related to Tuparí and Way-oró. No such problem arises if *t is reconstructed; in this case, we would only need to assume that PTpr *t yielded an affricate in the Corumbiara languages.

(ii) Note that what we reconstruct as *[nd] (the oral allophone of PTpr */n/) also yielded an affricate in Mekéns/Akuntsú and t (s before non-back high vowels) in Tuparí. In our current proposal, this is straightforwardly accounted for: all PTpr postoralized nasals (*mb, *nd, *qg) became voiced stops in Proto-Core Tuparian (*b, *d, *g), which subsequently merged with PTpr voiceless stops (*p, *t, *k) in the Corumbiara languages (yielding *p, *ts, *k) and in Tuparí (p/(p)s, t/s, k). No elegant explanation of the sort is available if one accepts the reconstruction of PTpr *ts.

(iii) There is no competing identity correspondence that could potentially involve PTpr *t in onsets, except for two isolated etymologies: 'chicha' (Wayoró tuero, Mekéns tiero, Akuntsú tieró) and 'daughter' (Tuparí haK, Wayoró, Mekéns, Akuntsú, Makurap taK). The former item is a Wanderwort (compare Arikapú t/uerd, Kanoe tsero; Voort 2005: 381, fn. 28, 2007: 138, fn. 4) and is thus likely to have diffused into at least some of the Tuparian languages through horizontal transmission. As for the word for 'daughter', the correspondence is unique and thus cannot back up alone the reconstruction of PTpr *t. In fact, it is possible to reconstruct *caK (no other examples for *c- in the word-initial position are known, so the reflexes h- in Tuparí and tin Wayoró could be regular).

(iv) Finally, the external correspondences of what we reconstruct as PTpr *t are dental/alveolar stops throughout the Tupian family (Karitiana, Karo, Proto-Mawé-Guaraní (*)t, Purub-orá d, as well as t alternating with n in the Mondé languages), even though admittedly affricate reflexes are also attested in non-palatalizing environments in languages such as Yudjá (tf-), Mundurukú (tf-/-dy), Kuruaya (tf-) and Proto-Tupí-Guaraní (*ts ~ *tj). At least Proto-Tupí-Guaraní demonstrably innovated its affricate from Proto-Mawé-Guaraní *t, as shown not only by external comparanda (Meira & Drude 2015) but also by the fact that the Proto-Tupí-Guaraní affricate *ts changes to *nd in the environment *V_ (cf. *tso 'to go' and its causative *mo-ndo 'to send'). Together, all these facts point to Proto-Tupian *t as the probable ancestor of what we reconstruct as PTpr *t, making our reconstruction more credible. For examples and a detailed discussion of the reflexes of Proto-Tupian *t, see Nikulin and Carvalho (2019: 276-8).

3.1. Proto-Tuparian approximants

In this section, we justify the reconstruction of three approximant phonemes for Proto-Tuparian: */p/, */j/, and */w/. All of these appear to have been straightforwardly retained from Proto-Tupian *]S, *j, and *w. Preceding nasal nuclei, */j/ was realized as a nasal stop *[p], which

14 In fact, the sibilant reflex in Tupari is conditioned not only by _i, but also by _u (as in suT 'peach palm', suT-'to cook' < PTpr *tiT, *tiT-).

is retained in all contemporary languages (unlike the oral allophone [j], which underwent major changes in all daughter languages; see below). In our reconstructions, we represent the al-lophones of */j/ as *j and *p, respectively, in order to highlight the fact that their default reflexes in all daughter languages are so different from each other that they are no longer syn-chronically analyzed as allophones of the same phoneme. As for the labiovelar approximant */w/, it is likely that it was phonetically nasalized in nasal environments (i.e., [w]); this is, however, not represented in our reconstructions, because in most contemporary languages the reflexes of *[w] and *[w] are reasonably similar to each other. The bilabial approximant */p/ is not attested in nasal environments, which may be a spurious gap, given that */p/ is an extremely rare segment in our corpus.

All Proto-Tuparian approximants, with the exception of the nasal allophone of */j/, were frequent targets of multiple fortition (and, to a lesser degree, lenition) processes, which operated in the history of each contemporary language to differing extents. These will be systematized in section 4. Nevertheless, the directionality of each sound change can be quite securely identified thanks to converging internal and external evidence, so that in each case one can be sure that the segments in question were indeed articulated as approximants in Proto-Tuparian. In subsections 3.2-4, we will discuss a number of correspondence sets for which the contrary holds: segments other than approximants (such as nasal stops or high vowels) are reconstructed for Proto-Tuparian, and it is shown that they gave rise to innovative approxi-mants in specific languages or subgroups of Tuparian.

PTpr *p. In Table 4, we show both secure etymologies which instantiate PTpr *]S. The original articulation is preserved in Wayoro only. In Tupari and Makurap, PTpr *j8 merged with PTpr *p, yielding Tup p (-Ps- before i) and Mak p. In the Corumbiara languages, PTpr *j8 is reflected as b. Despite the extreme scarcity of the relevant cognate sets, the reconstruction of PTpr *j8 gains some credibility in light of the fact that it corresponds to Proto-Tupi-Guarani *]S, as in PTG *tofia 'face' and *i§itu 'wind' (~ PTpr *jejSa, *ifiijo; cf. Mello 2000: 183, 207). Note that PTG *j8 is also a low-frequency segment (at least morpheme-internally): in Meira & Drude's (2015) corpus of Mawe-Guarani etymologies, it appears only in PTG *uru§u 'vulture' (~ PTpr *oroP?o), *jafioti 'tortoise' (no cognate in Tuparian), in the aforementioned *tofia 'face' and *ifiitu 'wind', as well as in *afiati 'maize', borrowed from a Cariban or other North Amazonian source (Rodrigues 1985: 389).15

PTpr gloss Wayoro Tupari Mekens Akuntsü Makurap

oral

*jeßa forehead — epa 'eye' — eba-pe cepa

CT only:

*jeßa-jopaP eye (*jopaP 'grain') eßa-paP — eba-opaP eba-paP —

*jeßa-pi face (*pi 'inner') — epa-Psi Sio eba-pi eba-pi —

*ißijo wind — uPsiö — — —

Table 4. Proto-Tuparian */|3/ (oral *p, unattested in nasal environments)

In our current proposal, PTpr *j8 is reflected as p in Makurap, thus paralleling the fortition and devoicing of PTpr *j to Makurap c (see below). We have also considered an alternative

15 In addition, Meira & Drude (2015: 295) give PTG *taßa 'village', but a more correct reconstruction would be

*taP (cf. Mello 2000: 195), of which *taß-a is an inflected form (the so called argumentative case). Its Tuparian cognate is *ja(:)P 'village'. Meira & Drude (2015: 294) also note that PTG *ißö 'to shoot' has a cognate in Aweti and reconstruct Proto-Aweti-Guarani *(?)ißö, but no cognate is known in Satere-Mawe.

scenario, whereby PTpr *fi would have been regularly preserved in Makurap as fi. This possibility is prompted by Nogueira et aUs (2019: 39, 41) reconstruction of two kinship terms: PTpr *afii 'father (vocative)' (> Way afii, Tup aPsi, Mek abi(-toP), Mak afia) and *afiatso 'grandfather' (> Way efiato, Mek abatso, Aku abatjo, Mak afiato). Regarding the former term, note that Makurap a is not a regular reflex of PTpr *i, which entails that Proto-Core Tuparian *afii is likely not to be cognate with Mak afia. As for the term for 'grandfather', there is evidence that the reconstruction should be amended to *joP-ato (literally 'father-big'), with an irregular development of the root vowel *o in all languages except in the Tupari compound mePsir-ob-ato 'father-in-law (female ego), lit. son's grandfather' (cf. ha?uP-b-ato 'father-in-law (male ego)', in which the same vowel was irregularly lost). That way, the bilabial approximant found in Way efiato or Mak afiato arose through resyllabification of a coda *p. As noted above, we are not concerned with such resyllabified approximants in this paper for lack of space.

PTpr */j/. The reader has already seen that we take PTpr *j and *p to be surface realizations of PTpr */j/ in oral and nasal environments, respectively. While *p retained its articulation in all daughter languages, *j shows more divergent reflexes. We assume that it preserved its palatal articulation in Makurap but became a voiceless affricate in this language (thus, PTpr *j > Mak c). In Proto-Core Tuparian, conversely, it appears to have preserved its manner of articulation but changed its place of articulation from palatal to dental (that is, PTpr *j > PCT *cf).16 In Wayoro, PCT *d became an underlying nasal stop /n/ (which surfaces as [nd] in oral environments), which parallels precisely other developments reconstructed for this language: PCT *w > Way /q(w)/ (see below in this subsection), PCT *o- > pre-Way *fi- > Way /m-/ (see 3.2), and PCT *i- > pre-Way *j- > Way /p-/ (see 3.2). In Tupari, one finds the reflex h- word-initially and -0- word-internally. In the Corumbiara languages, PCT *d became a homorganic stop t, as all other approximants (*fi > b; *w > k(w); *j > Mek ts/Aku t; see this subsection and 3.4).

The correspondence Way nd ~ Tup h/0 ~ Mek/Aku t ~ Mak c has not been previously claimed to continue the same underlying segment of Proto-Tuparian as the correspondence Way/Tup/Mek/Aku/Mak p. Important evidence for lumping them together comes not only from the fact that they occur in a complementary distribution (in oral vs. nasal environments, respectively), but also from the fact that their reflexes in Makurap show identical behavior when they occur as the initial segments of relational stems. Namely, both Mak c- and p- may derive relational stems from absolute ones (e.g. eK 'house', eri 'hammock' ^ c-eK 'house.POSS', p-eri 'hammock.POSS'; Braga 2005: 48sqq.). In addition, whenever these segments occur in the beginning of a relational stem (either derived from an absolute stem or underived), they are replaced with the 3rd person prefix t-. External evidence unequivocally shows that the correspondences Way nd ~ Tup h/0 ~ Mek/Aku t ~ Mak c, on the one hand, and Way/Tup/Mek/ Aku/Mak p, on the other hand, go back to a single consonant of Proto-Tupian. For example, in Karo, both correspond to j (e.g. Proto-Tuparian *jajo 'armadillo', *jaote 'peccary', *ja(:)ko 'lizard', *jao 'stingray', *pac 'tooth', *waktpa 'agouti', *p5kaT 'toucan' ~ Karo jajo, jate, ja?o, jaw, jaj, wakaja, jokan; data from Gabas Jr. 1999). In Proto-Munduruku, the regular correspondence is *d (e.g. Proto-Tuparian *jajo 'armadillo', *jaote 'peccary', *ja(:)ko 'lizard', *jakeK 'army ant', *pac

16 Note that unconditional dentalization of *j to d is known from the phonological histories of many Amazonian languages, such as Shiwilu (Kawapanan; Valenzuela Bismarck 2011: 279-80) and Guarasugwe (Tupi-Guarani < Tupian; Ramirez et al. 2017: 432). In Kubeo (Tukanoan), the reflex of Proto-Tukanoan *j is realized as [d] between non-high vowels (Chacon 2014: 65sqq). Several Cariban languages, such as Venezuelan Kari'na, Pemon, and Ma-kuxi, reflect Proto-Cariban *j as [d] at least in some environments. We know of no Amazonian language in which an opposite development (i.e., *d > j) would be claimed to have taken place in a non-palatalizing environment (an anonymous reviewer rightly remarks that outside the Americas, such a development is attested between vowels in many Turkic languages). This yields additional support for our hypothesis regarding PTpr *j > PCT *<J.

'tooth', *poK 'chigoe flea', *per 'faeces' ~ Proto-Munduruku *dajdo(?), *da^e(?), *da?o, *da?ik, *daj, *ddy, *dan; Picango 2019). In Proto-Tupi-Guarani, the default correspondence in the wordinitial position is *t- (e.g. PTpr *jajo 'armadillo', *jakeK 'army ant', *pac 'tooth', *pdK 'chigoe flea', *pdkaT 'toucan' ~ PTG *tatu, *ta?oK, *tac, *tuK, *tukar, cf. Mello 2000).

In Table 5, we list all known Proto-Tuparian tokens which instantiate PTpr *j and do not show significant irregularities in the development of this consonant in the daughter languages.

PTpr gloss Wayoró Tuparí Mekéns Akuntsú Makurap

oral (> Proto-Core Tuparian *S)

*japo: ~ *jambo: shimbillo (CT only) — hapó tapo: — —

*jajo17 armadillo ndato18 — tato tató —

*ja(:)ko lizard — hakó ta:ko — cáko

*jakeK19 army ant ? akeK hakék{e} takeK — —

*jai20 howler monkey ndau haú ta:?i tai-kóP —

*jaoZ1 stingray — — — — cáo

*jaote peccary — haote-?irí taotse taotfé cáóte

*jaP hair, feather ndaP haP taP a-táP caP

*ja:P village — ha:P ta:P — caP

*wejaP anteater (CT only) j^endaP — kmetaP witáP —

*jaT snake ndaT haT — — caT

*jaT?a bullet ant (WT only) ndara háT?a — — —

*ojaT fire — — otat otáT ócaT

*jac itchiness (WT only) ndac pe-ác hác-ka 'to scratch' — — —

Table 5. Proto-Tuparian */j/ (oral *j, nasal *p) (to be continued)

17 Although no cognate in Makurap is known, PCT *öaöo can be securely traced back to PTpr *jajo because precise external cognates are found all across Tupian (Karo and Puruborá jajo, Karitiana sosi, Proto-Mundurukú *ÖajÖo(?), Sateré-Mawé sahu, PTG *tatu; cf. Galucio et al. 2015: 262). Moore & Galucio (1994: 132) give Makurap tayto (in our transcription, tacto), but we could not confirm the existence of this form in our main sources on Makurap; moreover, it does not correspond regularly to the remaning forms. Mekéns tato is attested by Moore & Galucio (1994: 132) and Snethlage (2015: 520, <tatú> 'Tatú (Gürteltier)').

18 The expected reflex of the intervocalic PTpr *j would be *nd, not t. It is likely that that some sort of a dissimilation of the kind *NDVND... > NDVD... applied in this word, as in *rjgwarjgo > jg^ago 'sweet potato' (with an additional devoicing: *nd > *d > t; note that [d] is not part of the phonetic inventory of Wayoró).

19 The expected reflexes in Wayoró and Tuparí would be Way *ndakeK, Tup *hakeK. We surmise that Way akeK was borrowed from an early form of Tuparí (or a variety close to it), whereas in Tuparí some sort of expressive reduplication could have applied. The Mekéns reflex is regular but is attested only in Hanke et al. (1958: 212) as <takek>, where its regular relation to Mundurukú da?dk (< *öa?ik) and Tupinambá ta?oK is noted.

20 Although no cognate in Makurap is known, PCT *öai can be securely traced back to PTpr *jai because a precise external cognate is found in Karo (jai 'howler monkey', Galucio et al. 2015: 267). The Wayoró reflex is also attested as ndeu (Nogueira 2015: 617). Akuntsú -koP stands for 'red'.

21 Although no cognate in Core Tuparian is known, Mak cáo can be securely traced back to PTpr *jao because a precise external cognate is found in Karo (jaw 'stringray', Gabas Jr. 1999: 13). Galucio et al. (2015: 272) also cite possible cognates in Puruborá, Yudjá, and Xipáya.

PTpr gloss Wayoro Tupari Mekens Akuntsu Makurap

*ja?iP son, fraternal nephew (male ego) ndauP ha?uP taiP taiP caiP (also 'sperm')

*joP father ndoP ho(:)P toP toP coP

*jo(:)P22 to lie ndoP- — to P- to-a —

*jo-aP hammock (CT) ndo-aP o-aP ~ w-aP to-aP to-aP —

*jo?oP that (sitting) — ho?oP — — co:P

*jiri two (CT only) nduru-T huru 'pair' tiri tiri —

*j-eP leaf.poss23 ndeP heP teP t-eP ceP

*jeT name ndeT heT teT teT ceT

*jeT?oP24 rubber — heroP — tedo coroP

*j-eK house.poss nd-eK h-eK t-eK t-eK c-eK

*ji:T flower kuP-ndi:T 'forest' ? hi:T 'side dish' — — kiP-cir-eT

*ji:T-?a flower (CT only) ndi r-a hiT-?a tir-a25 tir-a —

*eji marico bag endu eu eti eti eci

nasal (> Proto-Core Tuparian *p)

*pa mother (voc.) pa pa a-tsi26 — pa

Sak makipa

*wakipa agouti ywakipa — Gua, Sio pakipa pakapa makip a

*tapa earring (CT only) tapa — — tjapa —

*kiripa nail (WT only) kiripa kiripa — — —

*paC tooth paC paC paC paC paC

*p5kaT toucan — pokaT — — pokaT

*p5taP flea — p5taP — — potaP

Table 5. Proto-Tuparian */j/ (oral *j, nasal *p) (to be continued)

22 Although no cognate in Makurap is known, PCT *do(:)P can be securely traced back to PTpr *jo(:)p (as opposed to **ndo(:)p) because a precise external cognate is found in Karo (jop- 'to live', Gabas Jr. 1999: 127). The noun *jo-aP 'hammock' is a nominalization of this verb ('lying place').

23 In most Tuparian languages, the only known term for 'leaf' is synchronically an underived relational noun., with no absolute equivalent attested. However, in Akuntsu one finds both an absolute (eP) and a relational (t-eP) form (Aragon 2014: 130), which is symbolized here by a hyphen.

24 The absence of a coda consonant in Akuntsu is irregular, as is the vowel o in the initial syllable in Ma-kurap.

25 The form is tentatively phonologized based on Snethlage's (2015: 520) attestation of <itira> 'Blute' (likely the third person i-tira). Hanke et al. (1958: 212) attests <otira> 'eyebrow' (o- is the 1sg prefix), which is likely a semantic offshoot of the same word. The semantic development from 'flower' to 'eyebrow' has also been reconstructed for the Macro-Je language Maxakali, where the compound kyC-dyT 'eyebrow' means literally 'forehead flower' (mi-dyT 'flower', literally 'tree flower'); see Nikulin and Silva (2020: 56).

26 The Mekens vocative term for 'mother' is identified by Nogueira et al. (2019: 41) as a fossilized compound of the original vocative term for 'mother' (PTpr *pa in our reconstruction) and the original referential term for 'mother' (PTpr *ti in our reconstruction, still found in Mekens as tsi), paralleling Mekens abi-toP 'father (voc.)' = abi 'father (voc.)' + toP 'father (ref.)'. Deriving the Mekens form from PTpr *pa-ti is unproblematic, because the loss of stem-initial PTpr */j/ is a recurrent phenomenon regular in polysyllabic relational stems, see below.

PTpr gloss Wayoro Tupari Mekens Akuntsu Makurap

*p5?5P27 powder p5:P p5?oP — e-p5P 'farofa flour' —

*p5K chigoe flea p5K p5K 'pimple' — — p5K

*peT *pe(:)T faeces, guts ashes peT (kujpe(:)T peT (kujpe(:)T peT pe T peT otaT-peT peT (kt}pe: T

*peT?a meat pera peT?a peera — pira28

*mapi manioc mapi mac29 — — mapi

*pi:K30 smoke pi:K (s}iK otaT-ni:K niK ocaT-piq-eT

Table 5. Proto-Tuparian */j/ (oral *j, nasal *p)

In a subset of relational stems, an unexpected correspondence occurs between Core Tu-parian vowel-initial stems and Makurap c- or p-initial stems. It is worthy of note that all such stems are, at the very least, disyllabic. We reconstruct their PTpr etyma as */j/-initial and posit a shared innovation for Core Tuparian, which consists in an almost regular loss of stem-initial */j/ at the left margin of polysyllabic relational stems. We have been so far unable to explain why some polysyllabic relational stems (such as PTpr *ja?iP 'son of a male ego', *peT?a 'meat') resisted the deletion of */j/. An anonymous reviewer suggests that the loss of */j/ should be viewed as a morphological — rather than phonological — change; however, at present we lack evidence for analyzing */j/- as a prefix (other than in the relational nouns *j-eK 'house.POSS' and *j-ekiP 'arrow.POSS', which are indeed derived from the absolute nouns *eK 'house' and *ekiP 'arrow'). The relevant cognate sets are given in Table 6 below.

PTpr *w. We reconstruct a labiovelar approximant *w for Proto-Tuparian. In Makurap, it is reflected as p before oral nuclei and as m before nasal nuclei (Makurap has no /w/). In the Core Tuparian languages, it usually retains its labiovelar articulation (Way /qw/, Tup w, Mek/Aku k™) except before rounded vowels, where one finds Way /q/, Tup 0,Mek/Aku k.31 It is thus possible that Proto-Core Tuparian */w/ had a velar allophone *[y], which occurred preceding rounded nuclei. In Wayoro and in the Corumbiara languages, both allophones (PCT *[w] and *[y]) underwent developments which affected most or all approximants in these languages: in Wayoro, they changed into underlying nasal stops /qw/ and /q/ (which surface as [qgw] and [qg] in oral environments), whereas in the Corumbiara languages they were

27 Although no cognate in Makurap is known, PCT *p5?5P can be securely traced back to PTpr *p5?5P because a precise external cognate is found in Proto-Munduruku (*c5m 'powder', cf. Picango 2005: 181).

28 The vowel of the first syllable is attested as high in Moore's field recordings ([;piH'ieL]), in Galucio et al. (2015: 253, piira?), and in Sekelj (1948, <nino>), but is not a regular reflex of PTpr *e. Braga (2005: 208) attests ptra. The expected form pera is in fact documented in one of our secondary sources (Moore & Galucio 1994: 133).

29 This token instantiates the loss of PTpr *p in Tupari preceding an i (that way, PTpr/PWT *mapi > *mai). The hiatus in the resulting form was resolved by reanalyzing [i] as a coda offglide [j], which is the surface realization of the underspecified palatal coda -C in modern Tupari (cf. Singerman 2016).

30 The Makurap form is from Moore's field data. The Tupari form appars to contain the fossilized third person prefix: s-iK < *j-iK < PCT *i-piK; this shows that the sound change *pi > i preceded the fortition of PCT *j in Tupari. In the Corumbiara languages, the root-initial is attested as n (instead of the expected *p) in all sources (Aragon 2008: 53, 2014: 108; Galucio et al. 2015: 259) except Aragon & Cabral (2005: 1537), who attest the Akuntsu word as otaT-piT, with the expected palatal onset but with an unexpected coronal coda.

31 The Kupndiiriat dialect of Wayoro appears to have p as its default reflex, though only one reliable datum is attested, piri 'agai' (Nogueira 2019: 4).

oral

*jai flour, crumbs — ha?u — — tfai

*ja:mbi crop seed afii a:Psí a:pi — cambi

*jape sharpened — ape — apé mbi-cape

*jato big — -ató atso -atjo cato

*jato to bathe ato- ató- atso- atjo- cato-

*jati pain, to hurt; sour ati así atsi atji cáti

*jo(?)a brother (female ego)32 okwa owá ~ o?á okwa okwa cóa

*jopaP grain — opáP 'maize' opaP 'maize' — copaP

*jopi?a egg upia33 oPsí?a opitsa opíta cópia

*joaC tail okwac oác okwac okwác péT-coaC

*jefia forehead CT only: — épa 'eye' — eba-pé cépa

*jefia-jopaP eye efia-paP — eba-opaP eba-páP —

*jefia-pi face — epa-Psi Sio eba-pi eba-pí —

*j-e-kiP arrow.poss — e-kuP — — c-e-kiP

*jei blood au éu ai34 e?i céi

nasal

*papi(-?a) nose api-a aPsí api-tsa apí-ta pápi

branch — aka — — nanga

Table 6. Loss of stem-initial Proto-Tuparian */j/ in polysyllabic relational stems

fortitioned to /kw/ and /k/, respectively. Tupari preserved *[w] without further changes, whereas the allophone *[y] yielded zero.

Unlike PTpr */j/, PTpr */w/ does not show radically different reflexes in oral and nasal environments. Only in Makurap does one consistently find different phonemes (j vs. m) as its re-flxes. More marginally, the Sakurabiat dialect of Mekens appears to sometimes have as the reflex of PTpr *w in nasal environments (e.g. ywae 'pot') as opposed to kw, which is found in oral environments in Sakurabiat and in all environments in Guaratira and Siokweriat (cf. Galucio 2001: 19). In our proposal, this is accounted for by positing a voiced stop stage in the development of the PTpr approximants in the Corumbiara languages, hence: *wae > *gwae > Sak qwae, Gua kwae, Sio kwa?e, Aku kwa?e.

The Tuparian etymologies which instantiate PTpr *w are listed in Table 7.

This concludes our presentation of the reflexes of the approximant series of Proto-Tuparian. In what follows, we present evidence for reconstructing innovative approximants for earlier stages of individual Tuparian languages and the proto-languages of low-level subgroups.

32 In all Tuparian languages except Akuntsu, the reflexes of this kinship term also denote a woman's cousin (son of a woman's paternal uncle). In addition, Makurap coaC is used for man's and woman's brothers alike (No-gueira et al. 2019: 42).

33 We have no explanation for the occurrence of u (as opposed to o) in this word.

34 We have no explanation for the occurrence of a (as opposed to e) in this word. Note the similarity of this (apparently irregular) development to the dissimilation of *ei, *eu to ai, au in Wayoro (Nogueira 2015).

PTpr gloss Wayoró Tuparí Mekéns Akuntsú Makurap

oral

*awa35 yam agma awa{té} akwa akwá —

*wara(:)c36 frog/toad sp. (CT only) jgwara(:)c warác-?a kwara:c kwarác —

*wari?a bat (CT only) jgwaria wári?a kwaritsa — _37

*waco alligator jgwacco wáo kmato kwató fiáto

*wako guan rjgwako wakó kwa(:)ko kwakó fiakó-peP

*wakara38 great egret — wakara-tó 'jabiru' — — fíakara

*wawo sweet potato jg^ago39 wáo kwa(:)ko kwakó fia^ó

*wa?i stone jgwai wá?i kwai Sio kwa?i kwa?í fiai

*waT-40 to go away jgwaT- waT- kwaT- kwaT- ? fiaT 'always'

*wac?a labret (CT only) — wác?a — kwacta —

*waK- to cry; sound — waK- kwaK kwaK —

*waK-toP-il to hear jgwaK-to-a — kwaK-tsoP- kmaK-tfoP- fiaT-to-a 'to look'

Table 7. Proto-Tuparian */w/ (oral and nasal *w) (to be continued)

35 Although no cognate in Makurap is known, PCT *awa can be securely traced back to PTpr *awa because a precise external cognate is found in the Juruna branch (Yudjá awa[?á] 'yam', Mondini 2014: 113). The expected reflex in Wayoró would be *aqgwa; it is unclear why the medial consonant is oral.

36 The vowel length is attested in the Wayoró and Mekéns reflexes by Galucio et al. (2015: 273). The Wayoró form is given with a short vowel in Nogueira (2011: 43, 52).

37 The Makurap term for 'bat' is fiacáriaC. Despite the obvious similarity to the Core Tuparian forms, there is no regular correspondence between them; Mak fiacáriaC could go back to PTpr *wajari(?)aC. It is unclear whether we are dealing here with an irregular development or with an indirect borrowing.

38 This token is a Wanderwort, as similar forms are found in many unrelated languages spoken all across the Amazon and even as far away as on the Caribbean coast of South America. Epps (2020, entry 'great egret') lists multiple languages of the Cariban (Carijona, E'ñepa, Wayana, Yabarana), Guahiboan (Cuiva, Sikuani, Macaguán), Arawakan (Paresí), Sáliban (Piaroa, Sáliba), Nadahup (Nadeb), and Tupí-Guaraní (Kokama, Wajapí) families and groups as having a Wanderwort of the approximate shape %wakara meaning 'great egret'. One may add other Arawakan (Wapixana wakara, Mehinaku wakala, Proto-Ta-Arawakan *wak'ara; Silva et al. 2013: 106, Corbera Mori 2008: 64; Nikulin & Muzykantova in prep.) as well as Tupian (Suruí-Paiter wakár, Zoró wakal, Ka'apor wakara; Bontkes 1978: 18, Lacerda 2014: 321, Caldas 2009: 304) languages to this list. Given the regularity of the correspondence between the Tuparí and Makurap forms, we deem it possible that * wakara was borrowed from an unknown source into Proto-Tuparian.

39 The expected reflex of PTpr *w before o would be *^g, not g. It is likely that that some sort of a dissimilation of the kind *NDVND... > NDVD... applied in this word, as in *ndando > ndato 'armadillo'. In fact, Nogueira (2019: 8, with a reference to her ongoing research) entertains the hypothesis that Wayoró [g] and [gw] could be even syn-chronically described as allophones of /q/ and /qw/.

40 Even if Makurap fiaT 'always' does not belong to this cognate set, the Core Tuparian verb cannot be regarded as a Core Tuparian innovation because precise external cognates are found across Tupian (Karitiana hot, Sateré-Mawé waT 'to go.pl').

41 This compound can be analyzed as 'sound-see'. The development *-Kt- > -Tt- in Makurap is unparalleled. The Wayoró and Makurap reflexes are attested only in their forms which contain the thematic vowel -a-, which triggers the deletion of the stem-final consonant. Clear cognates of PTpr *waK are found in other Tupian languages as well (e.g. Karitiana hok 'to play violin', Sateré-Mawé waK 'to cry').

PTpr gloss Wayoro Tupari Mekens Akuntsu Makurap

*wi42 blow — — — — pi

*awi- to enter (WT only) a/gu- au- — — —

*wo(:)43 thorn rgo: — ko44 ko 'fishhook' —

*a(:)wo bone — — a:ko — ao45

*wora sound, speech (?) rgora 'music' — — — pora-pi 'mouth'

*woroa- to look for (CT only) rgora- oroa- kora- kora- —

*aworo46 parrot — awro ~ aoro — — —

*woP red rgoP oP koP koP poP

*woT-kiP neck rgoT-kuP oT-kuP koT-kiP Sio kiT-kiP piT-kiP47 poT-kiP

*wetoK far rgwetoK ? toK kwetsoK — petoK

*wereP foreigner (CT only) rgwereP — kwereP kwereP 'dark' —

*wejaP anteater (CT only) r^endaP — kmetaP kmitaP48 —

*weP-49 to go up rgweP- — kweP- kweP- —

*wi ax — wi(:) kwi kwi pi

*ara:wi peanut (CT only) ara:gwi — ara(:)kwi arakwi _50

Table 7. Proto-Tuparian */w/ (oral and nasal *w) (to be continued)

42 Although no cognate in Core Tuparian is attested in our primary sources, Mak pi can be securely traced back to PTpr *wi because precise external cognates are found across Tupian (e.g. Karitiana he: 'to blow'). A likely cognate in Tupari, u- 'to blow, to play a wind instrument' is mentioned by Rodrigues (2002: 291), but we were unable to locate this form in our primary sources on Tupari, thus putting its existence in doubt.

43 Although no cognate in Makurap is known, PCT *wo(:) can be securely traced back to PTpr *wo(:) because precise external cognates are found across Tupian (Karitiana hi 'thorny tree', Satere-Mawe hu, PTG *ju 'thorn').

44 Attested by Wanda Hanke only (Hanke et al. 1958: 212) as <ku> 'thorn, needle'.

45 The absence of p is unclear. The word is frequently attested as cep-ao ([ceH.|3ao:?L] in Moore's data; cf. also Braga 2005: 162), with the relationalizer prefix ceP- (Braga 2005: 42-3).

46 Tup awro ~ aoro can be securely traced back to PTpr *aworo because precise external cognates are found across Tupian (Proto-Munduruku *aro, PTG *ajuru).

47 The development of PCT *wo into Akuntsu pi is not known to be regular. Note that the unrounding of *o is also attested in Siokweriat.

48 The development of PCT *e into Akuntsu i is not known to be regular. The word is attested as [wi'tap1] ~ [wi'ttap"'] ~ [wi'tdap"'] in Aragon (2008: 57), but is phonologized here with a /kw/ because this is almost certainly the same word as the one found in the hydronym KwitaP ki (Aragon 2014: 14), plausibly interpretable as 'anteater river'; the optional realization of /kw/ as [w] is independently attested by Aragon (2014: 57sqq.).

49 Although no cognate in Makurap is known, PCT *weP can be securely traced back to PTpr *weP because precise external cognates are found across Tupian (Karitiana hap 'to rise (of sun)', Aweti teP 'to go up'; Landin 2005: 10, Reiter 2011: 205).

50 The Makurap term for 'peanut' is attested as arapoiK (Braga 2005) or arapi:K (Moore's data). Despite the obvious similarity to the Core Tuparian forms, there is no regular correspondence between them. It is unclear whether we are dealing here with an irregular development or with an indirect borrowing.

PTpr gloss Wayoró Tuparí Mekéns Akuntsú Makurap

*ewiT51 honey, bee epgwiT ewíT ekwir-itsa 'bee sp.' Sio ekwiT ekwíT —

*wiT?i aqai Ngw rjgwiri Kup ßiri wíT?i kwiri kwirí ßirtfca}

nasal

*wämö?ä shaman — wämö?ä kwämöä kwämoä mämöä

*wäktpä52 agouti jwäküpä — Sak mäktp ä Gua/Sio paktpä pakäpä mäktp ä

*wäe53 pot jwäe wäe-töP-?a Sak jwäe Gua kwäe Sio kwa?e kwä?e —

*wäckiT plate (WT only) jwäckuT wäcküt — — —

*wt- to enter jü- — — — mt-

*wö pet (CT only) jO ? ö[äkiT] jö — —

*wi?iK leafcutter ant j^iK wi?iK — — miK

Table 7. Proto-Tuparian */w/ (oral and nasal *w)

3.2. Loss of syllabicity in high vowels

This subsection deals with the sound change whereby the high vowels */i o/ (and possibly their nasal equivalents) of PTpr became approximants when adjacent to vowels. Note that in the phonological systems of all Tuparian languages /o/ is analyzed as a high vowel, as there is no /u/. This type of sound change arguably recurred multiple times in the histories of the Tuparian languages, which is quite unsurprising given its naturalness. Its operation is most easily seen in the allomorphy patterns of the 3ncrf prefix (PTpr *i-) and of the 1SG prefix (PTpr *o-).

PTpr */i/ and */o/ were not affected by this process in the same fashion in the individual languages: while only Wayoro and Tupari show traces of the desyllabification of the reflexes of PTpr */o/, the front high vowel */i/ has been affected in all daughter languages. In Makurap, the PTpr 3ncrf prefix */i-/ before vowels yielded /p-/ (which surfaces as ndj- in oral environments and as p- in nasal environments), as shown in 1. Phonetically, this development must have proceeded through the stage *j (hence, PTpr *iV- > *jV- > */pV-/) and evidently postdates the specifically Makurap sound change *j > c.

(1) Makurap: /p-/ ndj- oral, p- nasal (Braga 2005: 50, 204; the glosses are ours) a. ndj-akare-T b. nd%-apiter-eT

/p-akare-eT/ /p-apiteT-eT/

3NCRF-head-POSS 3NCRF-sadness-POSS

'his/her head' 'his/her sadness'

51 Although no cognate in Makurap is known, PCT *ewiT can be securely traced back to PTpr *ewiT because precise external cognates are found all across Tupian: Karitiana e:t (< *ahit), Proto-Munduruku *eit, Satere-Mawe ewiT, Aweti ekiT, PTG *eiT.

52 The reflexes of this word in the Corumbiara languages show labial consonants (Sak m, Gua/Sio/Aku p) instead of the expected labiovelars (Sak yw, Sua/Sio/Aku km). That is, Proto-Corumbiara innovated by replacing *gwakipa with *bakipa. Currently we have no explanation for this (apparently idiosyncratic) development.

53 Although no cognate in Makurap is known (unless ie 'pot' is related), PCT *wae can be securely traced back to PTpr *wae because precise external cognates are found all across Tupian (e.g. Aweti ta?e, PTG *ja?e).

c. etetena kite pe p-o-a fiiT /etetena kite pe p-oP-a piT/ after_that people loc 3NCRF-give-TH all 'After that, he gave it to everyone.'

In the Core Tuparian languages, the PTpr 3ncrf prefix */i-/ was desyllabified before vowels as well, yielding PCT */j/. This must have happened after PTpr *j became PCT *d, because the reflexes of PCT *d and *j in oral environments are distinct in all Core Tuparian languages except Akuntsu (PCT *d > Way nd, Tup h-/-0-,Mek t, Aku t, whereas PCT *j > Way ndj, Tup s, Mek ts, Aku t). In nasal environments, however, there is no distinction between the reflexes of PTpr */j/ and those of PTpr */i/ before vowels in PCT, as both merge in PCT *p.54 That way, we believe that the inventory of approximants was augmented by one phoneme in PCT as compared to PTpr: first, PTpr *j was dentalized to PCT *d, leaving room for PTpr *i > PCT *j (before vowels). In 2-5, we show the prevocalic allomorphs of the 3ncrf prefix in each Core Tuparian language.

(2) WAYORO: /p-/ ndy oral, p- nasal (Nogueira 2019: 18) ndj-au-fia, p-indiakwa au-fia, ndj-ufiape au-fia, /p-au-pa p-Tniakwa au-pa p-upape au-pa 3NCRF-heal-vzR 3NCRF-food heal-vzR 3NCRF-beverage heal-vzR ndj-ato-a-P au-fia

p-ato-a-P au-pa/

3NCRF-bathe-TH-NMLZ heal-vzR

'He is healing it, healing her food, healing her drink, healing her bath.'

(3) TuparI: s- oral, p- nasal (Singerman 2018: 60-2)55

a. s-ope b. s-aT c. p-ope d. p-opo

3NCRF-thigh 3NCRF-grab 3NCRF-tongue 3NCRF-kill

'his/her thigh' 'to grab him/her/it' 'his/her tongue' 'to kill him/her/it'

(4) MEKENS: ts- oral and nasal (Galucio 2001: 35-7, 191)

a. ts-akoP b. fs-antP c. ts-o-kwe-a-T

/ts-akoP ts-aniP ts-(m)o-kWep-a-T/

3NCRF-be_hot 3NCRF-head 3NCRF-CAUS-climb-TH-PST

'hot (it)' 'his/her/its head' 'he made him climb'

d. aramira, aotse ts-opo ka:T i-tser-a-T /aramTra aotse ts-opo ka:T i-tseT-a-T/ woman man 3NCRF-beat and 3NCRF-leave-TH-PST 'The woman, the man beat her and she left.'

54 One could claim that Mekens has no merger: PTpr/PCT *p yields p in Mekens, whereas the allomorph of the 3ncrf prefix which occurs before nasal vowels is ts- (Galucio 2001: 35, fn. 6; 225, fn. 24) and not *p. We believe that in Mekens the allomorph s-, originally restricted to stems which start with oral vowels, has been analogically extended to all vowel-initial stems. That way, Mekens forms such as ts-dpo 'to beat him/her/it' (4d) are probably not cognate with Akuntsu p-op-a (5d) or Tupari p-opo (3d), but rather arose through analogy.

55 At different occasions, Singerman (2018) analyzes p- as a realization of /i-/ (p. 60-2) or of /j-/ (p. 371). For our current purposes, the choice between these two analytical options is irrelevant. Also note that in Tupari the al-lomorph i-, which was historically restricted to consonant-initial stems, may synchronically occur before vowels, as in i-eT 'his/her name' (Singerman 2018: 56), as a result of the elision of PCT *d (*i-deT).

(5) Akuntsu: t- oral, p- nasal (Aragon 2014: 46, 138, 177, 279)56 a. t-akoP te b. tatfo, tat/e tiri t-ajtji

3NCRF-be_hot FOC Tat/o Tat/e two 3NCRF-wife

'It is hot.' 'Tat/o and Tat/e were his two wives.'

c. t-anaP etfe kaP d. p-op-a

3NCRF-head diff wasp 3NCRF-beat-TH

'A wasp is on his head.' 'to beat him'

In some Core Tuparian languages, the innovative PCT *j has merged with segments whose ultimate source is different from *i. In subsection 3.4, we will show that Proto-Corumbiara innovated by creating transitional glides (as in PCT *pi?a > Proto-Corumbiara *pija), which have the same reflexes as PCT *j < PTpr *i (that is, Mekens ts, Akuntsu t). Similarly, the word for 'spider monkey' can be reconstructed as Proto-Corumbiara *jakiraP (> Mekens tsakiraP, Akuntsu takiraP).

Now we turn to the desyllabification of PTpr *o. This process is synchronically attested in Tupari, in which the 1SG prefix occurs as o- before consonants, but as w- before vowels (Singerman 2018: 42); it also coalesces with a following /o/ or /o/, yielding a long vowel. In Wayoro, the 1SG prefix also occurs as o- before consonants; before unrounded vowels, however, one finds the allomorph /m-/ (mb- in oral environments, m- in nasal environments), whereas before rounded vowels the zero allomorph occurs (Nogueira 2019: 11, 15, 150-1). This is shown in Table 8.

Wayoro (Nogueira 2019) Tupari (Singerman 2018)

before a consonant o- /o-/ o-?ußa 'my pot' o-piti:K 'I feel cold' o-rjgora 'to seek me' o- o-si 'my mother' o-kePk-a: 'I nursed' o-karaP 'toward me'

before a rounded vowel 00 /0-/ 0-upipe 'my port' 00omb-a: 'hit me!' o-o... ^ o:- o:P (o+oP) 'my father' ojaora (o+OjaoT+a) 'to answer me'

before an oral unrounded vowel mb- /m-/ mb-apiteP 'my ear' mb-ato-a-P 'my bath' mb-e-tJu:P-kwa-T 'I got wet' w- w-apaP?a 'my head' w-e-kiaraP-k-a 'I became happy' w-e-pak-a 'I woke up'

before a nasal unrounded vowel m- /m-/ m-ëjgu /m-equ/ 'my chicha' m-ämöC-kwa-T 'I dance fast' w-ekeT~?ekeT-k-aP 'my throwing up'

Table 8. The allomorphy of the 1sg prefix in Wayoro and Tupari

While the allomorphy pattern attested in Tupari can be explained away as a consequence of a recent natural sound change (*o > w _V[-rounded]), the pattern found in Wayoro requires a more elaborate diachronic account: positing a one-step sound change such as */o/ > /m/ would be an entirely implausible solution on typological grounds. Fortunately, there is independent comparative evidence which shows that at some point in the history of Wayoro all inherited word-initial (and some word-medial) approximants have become homorganic (underlying) nasals. We have already seen in 3.1 that PTpr *j and *w (> PCT *d, *w/*[y]) are reflected in Wayoro as nd, qgw/qw, and qg/q (underlying /n/, /qw/, and /p/); earlier in this subsection, it has been shown that the innovative PCT *j has yielded Wayoro ndj /p/. That way, it appears quite

56 Aragon (2014: 46, fn. 28) analyzes p- as an allophone of /i-/ but is explicit regarding its phonetic realization. For example, the example 5d is transcribed as [jü.'ba] ~ [jiö.'ba] in Aragon (2014: 46).

plausible that the allomorph /m-/ 1sg in Wayoro also continues an earlier approximant, which we reconstruct as pre-Wayoro *fi- and derive from PTpr/PCT *o- in the prevocalic position. Therefore, forms such as *fi-apiteP 'my ear' and *fi-ato-a-P 'my bath' are posited for the pre-Wayoro stage. Later on, *fi- would have undergone nasalization word-initially (in the intervocalic position, as we have seen in 3.1, it was preserved, as in PCT *efia-opaP > Wayoro efia-paP 'eye').

Note that the desyllabification of *o- before vowels cannot be considered an innovation shared by Wayoro and Tupari, even though it occurred in both languages in comparable contexts. First of all, the outcome of this process is different in pre-Wayoro (*fi-) and Tupari (w-). The second piece of evidence for positing two independent innovations is that the allomorph w- in Tupari occurs not only in originally vowel-initial stems, but also in stems which have diachronically lost their initial consonant (PCT *d, preserved as nd in Wayoro). For example, the consonantal allomorph occurs in Tupari w-eK 'my house' (Singerman 2018: 43), which goes back to PCT *o-deK (apparently by the way of pre-Tupari *o-eK). Its Wayoro cognate o-ndeK 'my house' (Nogueira 2019: 145, 165), which has not been affected by any process of consonantal loss, expectedly shows the vocalic allomorph o-.

The allomorphy patterns examined in this subsection are decisive in establishing the directionality of the sound changes which underlie the correspondence sets involving Wayoro nasal stops and non-nasal segments in other Tuparian languages. If one were to derive them from something other than approximants, it would be quite difficult to explain why Wayoro has /m-/ and /p-/ as the prevocalic allomorphs of /o-/ and /i-/, respectively. In our account, this is unproblematically attributed to a combination of two processes: the desyllabification of high vowels in the environment #_V (*i > *j in PCT, *o > *fi in pre-Wayoro) and the nasalization of approximants in Wayoro (*fi-, *d, *j-, *w, *[y] > /m-/, /n/, /p-/, /qw/, /q/).

3.3. Proto-Tuparian postoralized nasals and their development in Wayoro

It is possible to reconstruct three phonemic nasals for Proto-Tuparian: */m/, */n/, and */q/. In oral environments, they likely acquired an oral phase before an oral nucleus and thus surfaced as *mb, *nd, *yg (Galucio & Nogueira 2012). In nasal environments, */m/ and */n/ appear to have surfaced as *m and *n, which have been preserved as such in all contemporary Tuparian languages (in contrast, */q/ was likely postoralized even before nasal nuclei). Wetzels & Nevins (2018) classify the allophonic pattern of this type, which is known from many Amazonian languages, as nasal shielding. We call the allophones *mb, *nd, *yg postoralized in what follows. For our current purposes, it is essential that the Wayoro reflexes of the postoralized allophones are identical or similar to those of the Proto-Tuparian approximants in oral environments. More specifically,

- PTpr *fi and *mb merge in Wayoro as mb- (word-initially) or -fi- (between vowels);

- PTpr *j (> PCT *S) and *nd merge in Wayoro as nd;

- PTpr *w before rounded vowels (> PCT *[y]) and *yg merge in Wayoro as yg.

Based on the contemporary Wayoro reflexes, one may be tempted to attribute these mergers to a single sound change from the Proto-(Core) Tuparian approximants to Wayoro underlying nasals. Indeed, in 3.1-2 we have seen that most approximants of Proto-Core Tuparian became homorganic nasals in Wayoro: PTpr *j > PCT *d > Wayoro nd; PTpr *w > PCT *w/*[y] > Wayoro y(g)w/y(g); PTpr *o-, *i- before vowels > PTpr *fi-, *j- > Wayoro m(b)-, ndy. In contrast, the Wayoro reflexes of PTpr *mb, *nd, and *yg are identical to their reconstructed states, as in PTpr *mbo 'hand', *ndeT- 'to grind', *ygaP 'wasp' > Wayoro mbo, ndeT-, ygaP. At first glance, these sounds would appear to have been preserved intact in Wayoro all the way from Proto-Tuparian.

In this paper, however, we advance an alternative proposal. Namely, we hypothesize that the postoralized allophones of PTpr nasals (i.e., *mb, *nd, *yg) have been affected by a series of sound changes in Wayoro, which came full cycle to the initial state. The suggested evolution pathway of PTpr *mb, *nd, *yg in the Core Tuparian languages is as follows: (i) in PCT, they lose the nasal phrase and become *b, *d, *g; (ii) in the Corumbiara languages and in Tupari, they merge with PCT *p, *t, *k and yield Mek/Aku p, ts/tf, k, Tup p (s-/-Ps- before i), t (s before u/i), k; (iii) in pre-Wayoro, they lenite to *fi, *d, *y (and merge, therefore, with PCT *fi, *d, *[y] from PTpr *fi, *j, *w; the bilabial approximant in pre-Wayoro may also come from *o- as seen in 3.2); (iv) in contemporary Wayoro, they have been affected by the independently established nasalization process (*fi-, *d, *y > /m/, /n/, /q/). That way, the development from PTpr *mb, *nd, and *yg to Wayoro mb, nd, and yg is assumed to have proceeded in three steps (*mb/*nd/*yg > *b/*d/*g > *fi/*d/*y > mb/nd/yg), as opposed to a straightforward retention. It also entails that PTpr postoralized nasals and approximants first merged as pre-Wayoro approximants (and not as modern Wayoro underlying nasals).

Crucial evidence for our proposal comes from the development of PTpr *mb after an oral vowel in Wayoro: in this position, it is reflected as fi. Nogueira (2011: 45-6) documents forms such as o-fio 'my hand' and o-fii 'my foot' (< PTpr *o-mbo, *o-mbi), of which the uninflected forms are mbo and mbi, respectively (Moore & Galucio 1994: 133). Note that the environment which conditions the development of PTpr *mb in Wayoro is precisely the same that the one we have seen above for pre-Wayoro *fi from other sources (PTpr *fi or *o): it is reflected as /m/ word-initially (as in PCT *o-apiteP > *fi-apiteP > mb-apiteP 'my ear'), but is retained as /p/ after an oral vowel (as in PCT *efia-opaP > efia-paP 'eye'). It is, therefore, conceivable that PTpr *mb (as in *mbo 'hand' and *mbi 'foot') merged with other segments as pre-Wayoro *fi (as in *fio 'hand', *fii 'foot', *o-fio 'my hand', *o-fii 'my foot'), which was subsequently reverted to /m/ word-initially (and after a nasal vowel) by means of an independently reconstructed process (see 3.2), as in mbo 'hand' and mbi 'foot', but suffered no further changes after an oral vowel, as in o-fio 'my hand' and o-fii 'my foot'. For PCT, we reconstruct *b based on the fact that neither pre-Wayoro nor Tupari or Corumbiara show any traces of a nasal phase.

We find it likely that PTpr *nd and *yg have undergone in Wayoro a cycle of sound changes comparable to the one described for PTpr *mb in the preceding paragraph. For PCT, we reconstruct *d and *g: in Tupari and in both Corumbiara languages they merge with PCT *t and *k (thus paralleling the merger of PCT *b and *p in these languages), whereas in Wayoro they merge with PCT *d and *[y] as pre-Wayoro *d, *y > Wayoro /n/, /q/ (thus paralleling the merger of PCT *b and *fi) in Wayoro. That way, PTpr *ndeT- 'to grind', *ygaP 'wasp' are hypothesized to have developed into PCT *deT-, *gaP > pre-Wayoro *deT-, *yaP > Way ndeT-, ygaP.

PTpr */m/. In nasal environments, PTpr */m/ surfaced as *m and was preserved as such in all daughter languages. In oral environments, it likely had the allophone *mb, which was preserved in Makurap but suffered some changes in the Core Tuparian languages. As stated above, we believe it yielded PCT *b. In Tupari and in the Corumbiara languages, it merged with the reflexes of PTpr/PCT *p as Tupari p (assibilated to s-/-Ps- before i) and Mekens/Akuntsu p. In pre-Wayoro, PCT *b merged with the reflexes of PTpr/PCT *fi and PTpr/PCT *o (before vowels) as pre-Wayoro *fi, which yielded /m/ word-initially or after a nasal vowel and /p/ after an oral vowel. The Tuparian etymologies which instantiate PTpr */m/ are listed in Table 9.

PTpr */n/. In nasal environments, PTpr */n/ surfaced as *n and was preserved as such in all daughter languages. In oral environments, it likely had the allophone *nd, which was preserved in Makurap but suffered some changes in the Core Tuparian languages. As stated

PTpr gloss Wayoró Tuparí Mekéns Akuntsú Makurap

oral

*mbo hand mbo / -fío po po(-pi) po mbo

*(p)ombo- to kill, to beat (CT only) ombo- opó- opo- op-á —

*mboejoP- ~ *mboec/to-57 to know moendoP- pu(?)oP- poetoP- poetóP- mboeto- ~ mbieto-

*mboK?a tortoise (CT only) mboga póK?a poga pogá —

*mbi(-to) foot mbi / -fíi sitó pitso pi mbi

*ja:mbi crop seed afíi a:Psí a pi — cambi

*mbiro to have (CT only) mbiro -(P)síro piro — —

*mbiri?a58 trahira fish mbirija siri?á Sio piritsa biritá mbíria

*mbi?o horsefly — si?ó — — mbio

*mbiP to be afraid — — — piP mbiP

nasal

*ma- to put ma- ma- ma- ma- ma-

*mapi manioc map! mac — — mapi

*amana tayra amana — — — amana

*meT59 husband meT meT meT meT —

child, sororal

*mepiT nephew/niece (female ego) mepiT mePsíT mepiT mepíT mepiT

*mepir-epiT grandchild (female ego, CT only) mepir-epiT mePsír-ePsiT mepir-epiT mepir-epíT —

son-in-law

*mepoP (daughter's husband) mepoP mepoP — menoP

*ameko jaguar, dog ameko amekó ameko amekó ameko

*mico60 curassow — — — — mito

Table 9. Proto-Tuparian */m/ (oral *mb, nasal *m)

57 The Core Tuparian languages unequivocally point to PCT *boedoP- (in Wayoro, *d > nd nasalized the preceding vowels; in Tupari, *poeoP- was apparently simplified to pu(?)oP-). However, the expected Makurap correspondence would be *mboecoP- and not mboeto- ~ mbieto-.

58 The Wayoro reflex is from Galucio et al. (2015: 274), where it is given as mbirija, mbirija (with a transitional j /fi/). The expected reflex in Akuntsu would contain a /p/ and not a /b/; in fact, Galucio et al. (2015: 274) do give Akuntsu pirita (in our transcription, pirita), but our primary source has birita (Aragon 2014: 109).

59 Although no cognate in Makurap is known, PCT *meT can be securely traced back to PTpr *meT because precise external cognates are found all across Tupian (Karitiana man, Karo men, PTG *meT; Landin 2005: 16, Gabas Jr. 1999: 13, Mello 2000: 178).

60 Although no cognate in the Core Tuparian languages is known, Makurap mito can be securely traced back to PTpr *mico because precise external cognates are found all across Tupian (Karitiana mbisi, Proto-Munduruku *wito, PTG *mitu; Landin 2005: 16, Picango 2019: 140, Mello 2000: 182).

PTpr gloss Wayoró Tuparí Mekéns Akuntsú Makurap

oral

*ndi?a lip color — sú?a — — ndia

*(p)êndi mortar (CT only) endufdjaj 'pestle' êTsù-?a êtsi61 — —

*ndiri collared anteater — swru s wru — — ndiri

*ndo: mound, hill ndo: to-téT tso(:) — ndô-a

*ndeT- to grind ndeT- teT- — — ndeT-

*(j)aindi62 wife aindi — aitsi ajtfi —

nasal

*amana tayra amana — — — amana

*nako man — — nako63 nakô nako-fiiT 'boy'

*nô other no nô nô nô nô-T

*ândrê6A barred sorubim fish ânôrê anore — — anore

*në65 arm nê- (in compounds) (ajnê-tô 'shoulder' nê 'shoulder' — nê

*nê- to make nê- nê- — — nê-

*nêciK horsefly — nêuK — — nêtiK

Table 10. Proto-Tuparian */n/ (oral *nd, nasal *n)

above, we believe it yielded PCT *d. In Tupari and in the Corumbiara languages, it merged with the reflexes of PTpr/PCT *t as Tupari t (assibilated to s before i/u), Mekens ts, and Akuntsu tf. In pre-Wayoro, PCT *d became *d (merging with the reflexes of PTpr *j > PCT *ff), which yielded nd /n/ in modern Wayoro. The Tuparian etymologies which instantiate PTpr */n/ are listed in Table 10.

PTpr PTpr */q/ quite probably surfaced as before nasal and oral vowels alike. This contrasts with the pattern we reconstruct for */m/ and */n/, whereby the postoralized realization is found in oral environments only. One piece of evidence comes from Tupari,

61 Attested as <enzê> in Hanke et al. (1958: 204) and as est in Moore & Galucio (1994: 134).

62 PTpr also likely had the compound *ja?iP-ti 'wife' (literally, 'son's mother') preserved as Wayoro ndatP-ti and Mekéns tatP-si. In Tupari and Makurap, *(j)aindi and *ja?iP-ti appear to have contaminated: the former language has a?usi 'wife' (instead of the expected reflexes *aisi or *Ha?uP-siy, the latter has catp-ndi (instead of the expected reflexes *(c)aindi or *catP-ti). The irregularities in the correspondences have been noted by Nogueira et al. (2019: 46), where the reconstruction *ai(+)tsi is given. Although PCT *aindi has no exact cognate in Makurap, it can be securely projected to PTpr *(j)aindi because it influenced the shape of catP-ndi and because cognates are also found elsewhere in Tupian (Proto-Munduruku *tajtjï; Picanço 2019: 138).

63 Attested in Hanke (1958: 206) as <nanku> 'man'. Galucio et al. (2015: 251) give nakop 'man' instead.

64 Alves (2004: 145) claims that the Tupari form is borrowed from Makurap, but there would appear to be no formal reason to believe so.

65 The Wayoro and Mekéns forms are attested in Snethlage (2015: 518, 686) as Wayoro <onanto> 'Schulter', <unami a> 'Ellbogen' (likely o-nê-to, o-nê-mia, with the 1sg prefix o-) and Mekéns <kina, kina> 'Schulter' (likely ki-nê, with the 1incl prefix ki-). The Mekéns form is also attested as <unea> 'bras' (likely 1sg o-nê) by Claude Lévi-Strauss in his Kabisiana wordlist (apud Loukotka 1963: 48).

Mekens and Akuntsu, which reflect PTpr */p/ as k regardless of whether the nucleus of the syllable is oral or nasal (unlike what we saw above for PTpr */m/ and */n/, which show a conditioned split in these three languages). Based on the correspondence between Wayoro /p/, Tu-pari /k/ and Mekens/Akuntsu /k/, we may safely reconstruct PCT *g, which therefore differs from PCT *b and *d in occurring in oral and nasal environments indiscriminately. The second piece of evidence for reconstructing PTpr *qg as the only realization of PTpr */p/ comes from Braga's (2005) transcriptions of Makurap words, in which /p/ is transcribed as [pg] even in nasal environments: [pgem] 'breast', [pge'rej] 'to shut up, to be silent', [pg!i] ~ [pgai] 'knife' (Braga 2005: 195-6).66 That way, PTpr *qg (the only allophone of PTpr */p/) would have been preserved in Makurap. In PCT, it would have yielded *g, which was further devoiced to k in Tu-pari and in the Corumbiara languages (and merged with PCT *k, paralleling the merger of PCT *p/*b, *t/*d in these languages). In Wayoro, *g was probably lenited to pre-Wayoro *y (by means of the process which also lenited PCT *b/*d to pre-Wayoro *fi/*d) and later nasalized to modern Wayoro /p/. Unlike in PTpr (in our reconstruction), however, the fully nasal realization of Wayoro /p/ in nasal environments is compulsory (Nogueira 2011: 50-1). The Tu-parian etymologies which instantiate PTpr */p/ are shown in Table 11.

3.4. Innovative approximants in Proto-Corumbiara

There is good reason to think that the proto-language of the Corumbiara branch acquired innovative approximants via hiatus resolution, whereby glides were inserted in the environment *i,o,i_V. This includes both original hiatuses, retained from Proto-Tuparian, and new hiatuses, which arose as a result of elision of a PTpr glottal stop. Although the epenthesized segments are not phonetically approximants in the contemporary languages —rather, consonants such as t, s, or kw are found— we believe that these go back to erstwhile transitional glides, *j (inserted in the environment *i_V) and *w (*o,i_V), which were subsequently fortitioned. That way, the epenthesis in pre-Proto-Corumbiara could be characterized as a natural sound change, glide epenthesis (Blevins 2008: 84sqq.), which fed another natural sound change, approximant for-tition (independently established in subsections 3.1-2 above). The development pathway advanced in this subsection is thus essentially identical to Blust's (1994: 112-5) account of certain sound changes in a number of Austronesian languages (such as Chamorro), in which not only inherited approximants but also transitional/epenthetic glides have been historically forti-tioned, as in Chamorro pugwa? 'betel nut', gwidza '3SG' (from *buaq, *ia; cf. also Blevins 2008: 92).

As was already mentioned above, we suggest that the consonant originally epenthesized in the environment *i_V in pre-Proto-Corumbiara was *j. In 3.2, we saw that PCT *j- (from PTpr *i- before vowels) yielded Mekens ts, Akuntsu t. In the environment *i_V, however, a slightly more complicated situation is found: at least in some words, the Guaratira and the Siokweriat dialects of Mekens have 0 corresponding to ts in the Sakurabiat dialect (and to t in Akuntsu). For example, Galucio et al. (2017: 338, fn. 6) report that pia 'to wait' is the form used in the Guaratira and Siokweriat dialects, which corresponds to pitsa in the Sakurabiat dialect. The Tuparian etymologies which instantiate Proto-Corumbiara *j are shown in Table 12. The Proto-Corumbiara reconstructions themselves are not given in the table for reasons of space; we reconstruct *kije 'one', *pija 'liver', *apija 'nose', *opija 'egg', *pirija 'trahira fish', *gwarija 'bat', *(a)mija 'knee' (in the last word, Mekens and Akuntsu have p/n rather than ts/t, quite possibly due to the nasal environment).

66 We assume that these transcriptions supersede Braga's earlier claim, according to which /p/ is obligatorily postoralized before oral vowels only (Braga 1992: 45-7).

PTpr gloss Wayoró Tuparí Mekéns Akuntsú Makurap

oral

*jgaP wasp rgaP kaP kaP kaP rgaP

*jgapi(-?a)67 bullet ant — — — — jgápia

*jgi68 liquid, saliva jgu ku ki ki —

*jgiP louse ä-jguP kuP kiP kiP rgiP

*jgi?iT salt jgu:T69 ku?uT ki:T ? kic70 rgiT

*jgoP?i71 termite jgm kóP?i kobi kopí ? rg0ß-a

*rjgipi?a72 tick (CT only) jgupi?a — — kipíta —

*JgoT palm larva — koT — — rgoT

*jge garden rge — — — rge

*jgeaT sun, sky jgiaT 'sky' kiáT 'up' — — jgéaT

*jgeK73 caterpillar — — — — rgeK

*jgi-akoP sun (CT only) jgi-akoP ki-akóP ki-akoP ki-akóP —

nasal

*päjgä branch — äkä äkä74 — närgä

*kumjga75 scorpion ku(u)mjä kwíntká kimjä — —

*jgêP breast jêP kêP kêP kêP rgêP

*jgèT-*jgèT-rga- to sink to swallow (WT only) jêT-jêT-jga- kêT-ka — — —

Table 11. Proto-Tuparian */p/ (oral and nasal *rg)

67 Although no cognate in Core Tuparian is known, Mak rjgapia can be securely traced back to PTpr *rgapi(-?a) because an external cognate is found in Karitiana (nopi 'bullet ant', Landin 2005: 19). Karitiana n is a regular reflex of Proto-Tupian *r in nasal environments, but the mismatch between the nasality values of the first syllable in Makurap and Karitiana awaits further explanation.

68 Although no cognate in Makurap is known, PCT *rgi can be securely traced back to PTpr *rgi because an external cognate is found in Karitiana (rge 'blood', Landin 2005: 9).

69 Form attested in Moore & Galucio (1994: 134). Nogueira (2011: 40) documents a form with an initial k-, which could be a mistranscription or a borrowing from another Tuparian language.

70 It is unclear if this is an irregular reflex of *rgi?iT or a semantic extension of kic 'earth' (< PTpr *kic).

71 The Wayoro and Mekens forms are cited after Galucio et al. (2015: 272); the expected Wayoro reflex would actually be *rgofii. It is uncertain if the Makurap word is a precise cognate because of the vowel mismatch; it is possible that continues a derivative close to *rgoP?i-?a (compare Proto-Munduruku *kopia 'blood', Satere-Mawe nupi?a; Picango 2019: 138, Ribeiro 2010: 76).

72 The Wayoro form is from Galucio et al. (2015: 273).

73 Although no cognate in Core Tuparian is known, Mak rgeK can be securely traced back to PTpr *rgeK because an external cognate is found in Karitiana (rgak 'caterpillar', Landin 2005: 9).

74 The form is tentatively phonologized based on Snethlage's (2015: 520) attestation of <zanka> 'Zweig' (likely the third person ts-aka).

75 The Wayoro and Mekens are given as kiinija? and kinira in Galucio et al. (2015: 270) and as kinira and kitmra in Moore & Galucio (1994: 134); our phonologization is tentative. In Tupari, the vowel of the first syllable is irregularly diphthongized (*« > wi), assuming our primary source (Alves 2004: 204) records the word correctly; in Mekens, we would expect a /k/ rather than a /p/. The stem is likely inherited from Proto-Tuparian (as opposed to a Core Tuparian innovation), because there is a probable cognate in Karitiana: kennon 'scorpion' (Landin 2005: 13) < pre-Karitiana *kinVraT.

PTpr gloss Wayoró Tuparí Mekéns Akuntsú Makurap

*kie one (CT only) kie-T kíe kitse, kie{ka} Sio gie76 kite —

*pi?a liver pia si?á pitsa bíta pía

*papi(?a) nose apia aPsí apitsa apíta papi

*jopi?a egg upia oPsí?a opitsa opíta cópia

*mbiri?a77 trahira fish mbirija siri?á Sio piritsa biritá mbíria

*wari?a bat (CT only) rjgwaria wári?a kwaritsa Sio kwaritsa ~ gw- — (fíacáriac)

*-pía knee {küjmía mía{K-?a} {nekiwa}mípa 'elbow' {a}mína {ka}pía

Table 12. Proto-Corumbiara *j (> Sakurabiat ts, Guaratira/Siokweriat 0,Akuntsu t)

It is interesting to observe that in the variety of Mekens attested by Wanda Hanke the consonant in question is consistently transcribed as <z> (more rarely, <s>), which implies that she must have made her notes with speakers of the Sakurabiat dialect: <kitze> 'one', <-piza> 'liver', <ampiza> 'nose', <kurakura-rupiza> 'egg' (korakora 'hen'), <kwarisa, kwarisasu> 'bat' (Hanke et al. 1958). In contrast, Emil-Heinrich Snethlage's notes on Mekens typically have zero: <-pia> 'liver', <-ampia> 'nose', <-upia> 'egg'; one exception is <kizakatn> 'one' (Snethlage 2015: 518-20).

Curiously, not all instances of Sakurabiat ts ~ Guaratira/Siokweriat 0 reflect an erstwhile epenthetic glide. At least in two words, this correspondence continues PTpr/PCT *t > Proto-Corumbiara *ts. One example is the word for 'foot', which has the shape pitso in Sakurabiat, but pio in Guaratira and Siokweriat (Galucio 2001: 19); the Tupari cognate sito 'foot' shows that all these words continue PCT *bito (ultimately derived from PCT *bi < PTpr *mbi 'foot'). Similarly, Galucio (2011b:7, fn. 11) states that the 1incl pronoun is kitse in Sakurabiat but okie in Guaratira and Siokweriat (< PTpr *kite).

Now we turn to the environment *o,i_V. We believe that such hiatuses were resolved in pre-Proto-Corumbiara by means of the insertion of *w (merging with PTpr/PCT *w), which yielded Mekens/Akuntsu kw. Curiously enough, the Wayoro cognates of these words also have kw, which must be attributed to an independent innovation. The Tuparian etymologies which instantiate pre-Proto-Corumbiara *w are shown in Table 13.

PTpr gloss Wayoró Tuparí Mekéns Akuntsú Makurap

*jo(?)a brother (female ego) okwa owá ~ o?á okwa okwá cóa

*joaC tail okwac oác okwac okwác peT-coac

*o?e78 to wash okw-á o?é okw-a okw-a ó-a

*ia:c tapir ukwa:c — ikwa:c ikwác iác

Table 13. Pre-Proto-Corumbiara *w > Mekéns/Akuntsú kw

76 The forms kitse and kieka are cited after Galucio et al. (2015: 250); we would expect kitse to be from the Sakurabiat dialect and kieka from the Guaratira dialect. Siokweriat [gie] is cited after Aragon (2014: 310); we would expect it to be kie.

77 The Wayoro reflex is from Galucio et al. (2015: 274), where it is given as mbirija, mbirija (with a transitional j /fi/). The expected reflex in Akuntsu would contain a /p/ and not a /b/; in fact, Galucio et al. (2015: 274) do give Akuntsu pirita (in our transcription, pirita), but our primary source has birita (Aragon 2014: 109).

78 The stem without the thematic vowel is attested only in Tupari. Braga (2005: 196) gives the stem without the thematic vowel as oP, but it is not attested in any linguistic example and does not match the Tupari form.

Even though only one example is known for the environment *t_V, the *w-epenthesis must have been regular. No reflexes of PTpr *iar 'grass, lawn' (> Tuparí uút 'lawn', Makurap iaT 'grass') are attested in Wayoró or in the Corumbiara languages (we would expect them to have the shape *ukwaT in Wayoró and *ikwaT in Mekéns/Akuntsú). The correspondence between Wayoró ufiekoiP, Akuntsú iekó, Mekéns ieko, and Makurap íéko, all meaning 'king vulture', appears to be, therefore, irregular (especially regarding the Wayoró form), suggesting that this word has been diffused via horizontal transmission in the region.

4. Evolution in the daughter languages

In this section, we outline the evolution of the consonantal onsets of Proto-Tuparian in the daughter languages (except those which arose from erstwhile codas via resyllabification).

Proto-Tuparian to Makurap. The evolution of the Proto-Tuparian onsets in Makurap involves as few as four innovations: (i) fortition of the approximants and *j to Mak p, c (3.1);

(ii) loss of the velar articulation in *w, which became Mak fi/m (as per nasality of the nucleus);

(iii) fronting of *c into Mak t; (iv) consonantization of the prevocalic instances of *i into ndj/p (as per nasality of the nucleus), as in the third person prefix *i- (3.2). Note that PTpr *p remained in Makurap as p, but its phonological status appears to have changed: it is analyzed as the nasal allophone of */j/ in Proto-Tuparian, but as the nasal allophone of */p/ in Makurap due to the influx of ndj (from earlier *i before vowels). This is shown in Figure 1.

Proto-Tuparian to Proto-Core Tuparian. The Proto-Tuparian onsets have evolved in the following way in Proto-Core Tuparian: (i) the postoralized allophones of PTpr underlying nasal stops (*mb /m/, *nd /n/, *yg /q/) lose their nasal phase and become PCT *b, *d, *g (3.3); (ii) PTpr *p and *t nasalize to PCT *m and *n preceding nasal nuclei; (iii) *j dentalizes to PCT *d (3.1); (iv) in some polysyllabic relational stems, the stem-initial */j/ disappears; (v) the prevocalic instances of *i consonantize to j/p (as per nasality of the nucleus), as in the third person prefix *i- (3.2); (vi) *w loses its labial articulation preceding rounded vowels and becomes *y

(phonologically, still an allophone of */w/). Note that in PCT the voiced stops *b and *d occurred in oral environments only and were thus in a complementary distribution with PCT *m and *n. We tentatively analyze PCT *b/*m and *d/*n as allophones of underlying */b/ and */d/ conditioned by the nasality of the nucleus, but other solutions cannot be at present ruled out. The evolution of the Proto-Tuparian onsets in Proto-Core Tuparian is schematized in Figure 2.

Proto-Core Tuparian to Wayoro. For an accurate description of the evolution of the PCT onsets in Wayoro, it is useful to distinguish between two chronological stages, which we dub pre-Wayord and Wayoro. In pre-Wayoro, PCT *b, *d, and *g were lenited to *]S, *d, and *y, thus merging with pre-existing PCT *]S, *d, and *y (3.3). Note that *y is analyzed as an allophone of */w/ in PCT, where its occurrence was conditioned by a following rounded vowel. In pre-Wayoro, however, */y/ and */w/ are contrastive thanks to the influx of *y < PCT *g (compare pre-Wayoro *weP- 'to climb' and *ye 'garden'). Another minor source of pre-Wayoro *j8 is PCT *o before vowels (notably in the 1SG prefix *o-; 3.2).

After the completion of these processes, the approximant-rich sound system of pre-Wayoro suffered further changes. Namely, pre-Wayoro *]S, *d, *j, *y, *w were fortitioned to mb, nd, ndj, qg, qgw in oral environments and to m, q, qw in nasal environments (no examples are available for pre-Wayoro *d or *j in nasal environments); exceptionally, and *-j- between vowels yielded -§- and -dj- in modern Wayoro. Synchronically, Nogueira (2019) analyzes mb, nd, ndj, qg, qgw as allophones of underlying /m n p p pw/, which are thus derived via nasal shielding (Wetzels & Nevins 2018).79

79 Although we provisionally accept Nogueira's (2019) analysis in this paper, there appear to be serious arguments for alternatively analyzing Wayoro mb, nd, ndj, qg, qgw as allophones of voiced stops /b d d3 g gw/ derived via nasal venting (Wetzels & Nevins 2018). Note that mb, ndj, qg, and qgw may only occur word-initially and after nasal vowels; between oral vowels, [|3 d3 g gw] occur (Nogueira 2019: 8, 62, fn. 1). This distribution matches perfectly all four predictions made by Wetzels & Nevins (2018: 842) for nasal venting, including (i) the absence of prenasalized voiceless consonants; (ii) the absence of prenasalized fricatives; (iii) the decreasing susceptibility of consonants to nasal venting as one moves from velars to labials (as in pre-Wayoro *ayuP 'louse', *ewiT 'honey', *edu 'marico bag' > aqguP, eqgwiT, endu, but *afii 'seed' > afii); and (iv) greater prominence of the nasal venting in prosodic domain-initial position as compared to the intervocalic position (compare 'hand' > mbo, but *o-fio 'my hand' >

The epenthesis of kw in the environment o,u_V must have proceeded through the stage *owV/*uwV; that way, a glide was inserted and subsequently fortitioned. Both processes must have applied after the sound change *w > y(g)w, because otherwise we would expect tokens such as PTpr *ia:c > pre-Wayoro *ua:c 'tapir' to have participated in them, yielding *uwa:c > *ufjgwa:c (rather than the attested ukwa:c).

Finally, the change from PCT *c to Wayoro tf has not involved any phoneme splits or mergers and cannot be straightforwardly attributed to any specific chronological level. In fact, it is not even clear whether any sound change has been involved at all, because almost nothing is known about the articulation of PTpr/PCT *c other than that it was a voiceless coronal segment capable of changing to t, tf, or 0.In Figure 3 below, it has been provisionally attributed to the pre-Wayoro stage, but it should be kept in mind that other scenarios are also possible.

Proto-Core Tuparian to Tupari. The phonological history of Tupari is characterized by multiple mergers, which resulted in the loss of the original voice distinctions reconstructed for Proto-Core Tuparian, as well as by a number of palatalization processes. A full list of the innovations we could identify follows: (i) mergers of *p/*b/*fi > *p, *t/*d > *t, *k/*g > *k (3.1, 3.3); (ii) *p is palatalized before i to -Ps- (between vowels) or s- (elsewhere); (iii) *t is palatalized to s

o-fio). Future studies in Wayoro phonology should determine whether it is possible to posit the phonemes /b d d3 g gw/ for the language instead of /m n yi q |3/.

before i/u; (iv) *j is fortitioned to s (3.2) but its nasal counterpart p stays except before i, where it is deleted (3.1); (v) *d is debuccalized to h word-initially and disappears between vowels (3.1); (vi) *y (an allophone of PCT */w/) is deleted (3.1); (vii) *c is deleted at least between vowels; (viii) PCT *o is consonantized to w before unrounded vowels and fuses with following rounded vowels yielding long vowels (notably in the 1SG prefix *o-; 3.2). This is schematized in Figure 4 below.

Proto-Core Tuparian to Mekens/Akuntsu. It is convenient to present the evolution of the Proto-Core Tuparian onsets in the Corumbiara languages Mekens and Akuntsu by dividing the phonological history of these languages into two stages. At stage 1, a chain shift affects the voiced oral segments of PCT: the approximants *]S, *d, *y, *w become voiced stops (*b, *d, *g, *gw; 3.1), whereas the PCT voiced stops lose voice (*b, *d, *g > *p, *t, *k; 3.3). These processes are also fed by the glide insertion (PCT *i_V, *o/i_V > *ijV, *o/iwV; 3.4). At stage 2, *t becomes an affricate (Proto-Corumbiara *ts > Mek ts, Aku tj). Innovative Proto-Corumbiara *t (> Mek/Aku t) comes from two sources: fronting of *c > *t and devoicing of *d/*g/*gw > *t/*k/*kw. The latter process occurred in all Corumbiara varieties, with the proviso that in the history of the Saku-rabiat dialect it was bled by the nasalization of voiced stops in nasal environments (as in *baktpa, *gwae, *go > maktpa 'agouti', qwae 'pot', qd 'pet'; *d is not known to have occurred in nasal environments). In other dialects of Mekens and in Akuntsu, there was no nasalization of voiced stops, and the devoicing of *d/*g/*gw applied categorically.

Fortition and devoicing also affected pre-Proto-Corumbiara *j, which yielded ts in Saku-rabiat and t in Akuntsu. Based on these reflexes, one could be tempted to reconstruct Proto-Corumbiara */, which would have been subsequently devoiced to *c > Sak ts, Aku t. However, the fact that the Guaratira and Siokweriat dialects of Mekens have 0 corresponding to Sak ts, Aku t in the environment i_V, implies that the fortition of *j must have occurred only recently

in the history of the Corumbiara languages: while the sound change *ijV > iV is easily conceivable, it would be more difficult to account for a sound change such as *icV, *i/V, or *itsV > iV. A detailed reconstruction of the evolution of *j in the Corumbiara languages awaits further investigation.

Figure 5 summarizes our proposal regarding the development of the PCT onsets in Mekens and Akuntsu.

5. Conclusions

This paper has advanced the reconstruction of the evolution of a number of Proto-Tuparian segments in daughter languages, with a focus on the Proto-Tuparian approximants as well as on sounds which went through an approximant stage at least in some Tuparian languages. Compared to earlier works on the historical phonology of this genetic unit, our proposal stands out in taking into account its subgrouping, thus allowing us to identify crucial innovations restricted to specific clades (such as Core Tuparian and Corumbiara), as well is in considering important data sources which have become available only recently (notably Aragon 2014; Singerman 2016, 2018; Nogueira 2019).

Future research on the historical phonology of Proto-Tuparian should address issues such as the evolution of the Proto-Tuparian glottal stop (which is known to frequently disappear in the daughter languages under unknown conditions), the coda resyllabification patterns, the reconstruction of vowel length in Proto-Tuparian, and the reconstruction of the Proto-Tuparian prosody (which should minimally account for the contrastive stress in Tupari and Akuntsu and for the tonal patterns of Makurap).

It is hoped that this study will inform further comparative research of the entire Tupian family, and we believe that at least some of our results are of interest to the typology of sound change (cf. Blevins 2008) and phonological theory (cf. Wetzels & Nevins 2008).

Abbreviations

Aku = Akuntsú Gua = Guaratira Kup = Kupndiiriat Mak = Makurap Mek = Mekéns Ngw = Ngwayoroiat Sak = Sakurabiat Sio = Siokweriat Tup = Tuparí Way = Wayoró

PCor = Proto-Corumbiara PCT = Proto-Core Tuparian PTG = Proto-Tupí-Guaraní PTpr = Proto-Tuparian PWT = Proto-Wayoró-Tuparí

1/3 = first/third person

caus = causative

foc = focus

diff = diffuse locative

incl = inclusive

loc = locative

ncrf = non-coreferential

PL = plural

poss = possessed

PST = past

sg = singular

th = thematic vowel

References

Alves, Ana Carolina, Ana Vilacy Galucio. 2007. Análise fonética acústica das vogais orais da lingua Sakurabiat. Letras de Hoje 42(3): 27-42.

Alves, Poliana Maria. 2004. O léxico do Tuparí: proposta de um dicionário bilíngüe. Ph.D. dissertation, Universidade Estadual Paulista "Júlio de Mesquita".

Anonymous. No date. Palavras e frases da lingua Wyarú. Rio de Janeiro, manuscript (personal archive of Darcy Ribeiro).

Aragon, Carolina Coelho. 2008. Fonologia e aspectos morfológicos e sintáticos da língua Akuntsú. MA thesis, Universidade de Brasilia.

Aragon, Carolina Coelho. 2014. A grammar of Akuntsú, a Tupían language. Ph.D. dissertation, University of Hawai'i at Manoa.

Aragon, Carolina Coelho, Ana Suelly Arruda Cámara Cabral. 2005. A posigäo da lingua Akuntsú na familia lingüística Tuparí. In: Lúcia Maria Pinheiro Lobato; Stella Maris Bortoni-Ricardo; Ana Suelly Arruda Cámara Cabral; Heloisa Maria Moreira Lima Salles; Maria Marta Pereira Scherre; Daniele Marcelle Grannier (orgs.). Anais do IV Congresso Internacional da ABRALIN: 1533-9. Brasília: Universidade de Brasília.

Aragon, Carolina Coelho, Fernando O. de Carvalho. 2007. Análise acústica das vogais orais da língua Akuntsú. Revista da ABRALIN 6(2): 41-55. doi:10.5380/rabl.v6i2.52623

Aragon, Carolina Coelho, Luciana Keller Tavares. 2019. Aspectos da cultura material do povo Akuntsú. Paper presented at the 3° Congresso Internacional Povos Indígenas da América Latina (CIPIAL). Brasília: Universidade de Brasília. Retrieved from: <https://www.academia.edu/39865327/Aspectos_da_cultura_ material_do_povo_Akunts%C3%BA>, accessed on 02/07/2020.

Becker-Donner, Etta. 1955. Notizen über einige Stämme an den rechten Zuflüssen des Rio Guaporé. Archiv für Völkerkunde 10: 275-343.

Blevins, Juliette. 2008. Consonant epenthesis: natural and unnatural histories. In: Good, Jeff (ed.). Linguistic Universals and Language Change: 79-107. Oxford University Press.

Blust, Robert. 1994. Obstruent epenthesis and the unity of phonological features. Lingua 93(2-3): 111-139. doi:10.1016/0024-3841(94)90001-9

Bontkes, Willem. 1978. Dicionário preliminar Suruí-portugues, portugués-Suruí. Porto Velho: Summer Institute of Linguistics.

Braga, Alzerinda. 1992. A fonologia segmental e aspectos morfofonológicos da língua Makurap (Tupí). MA thesis, Universidade Estadual de Campinas.

Braga, Alzerinda. 2005. Aspects morphosyntaxiques de la langue Makurap/Tupi. Ph.D. dissertation, Université de Toulouse-Le Mirail.

Caldas, Raimunda Benedita Cristina. 2009. Uma proposta de dicionário para a língua Ka'apór. Ph.D. dissertation, Universidade de Brasília.

Carvalho, Fernando O. de. 2019. Revisitando o Proto-Jurúna: A reconstruyo das oclusivas orais. In: Edna dos Santos Oliveira, Eduardo Alves Vasconcelos, Romário Duarte Sanches (eds.). Estudos linguísticos na Amazonia: 215-236. Campinas: Pontes Editores.

Carvalho, Fernando O. de. Forthc. The Historical Phonology of the Tupi-Guarani Language Family (working title). Berlin: Language Science Press.

Chacon, Thiago Costa. 2012. The phonology and morphology of Kubeo: The documentation, theory, and description of an Amazonian language. Ph.D. dissertation, University of Hawai'i at Manoa.

Corbera Mori, Ángel. 2008. Aspectos da fonologia segmental Mehináku (Aruák). Estudos Lingüísticos 37(1): 63-72.

Corréa-da-Silva, Beatriz Carretta. 2010. Mawé/Awetí/Tupí-Guaraní: relagdes lingüísticas e implicagdes históricas. Ph.D. dissertation, Universidade de Brasília.

Drude, Sebastian. 2009. Nasal harmony in Awetí: a declarative account. ReVEL - Revista Virtual de Estudos Da Linguagem 3.

Epps, Patience. 2020. South American Languages. In: Claire Bowern, Patience Epps, Jane Hill, Patrick McConvell. Languages of Hunter-Gatherers and their Neighbors: Database. Retrieved from <https://huntergatherer.la. utexas.edu>, accessed in 19/07/2020.

Gabas Jr., Nilson. 1999. A Grammar of Karo, Tupi (Brazil). Ph.D. dissertation, University of California, Santa Barbara.

Gabas Jr., Nilson. 2005. Classificagao da língua Akuntsú. Estudos Lingüísticos 36: 105-10.

Galucio, Ana Vilacy. 1994. Fonologia segmental da língua Mekens. In: Anais do IX Encontro Nacional da ANPOLL (Caxambu, junho de 1994). Area de Lingüística, v. 2: 988-97. Joao Pessoa: Associagao Nacional de Pós-Graduagao e Pesquisa em Letras e Lingüística, 1994.

Galucio, Ana Vilacy. 2001. The morphosyntax of Mekens (Tupi). Ph.D. dissertation, University of Chicago.

Galucio, Ana Vilacy. 2002. Word order and constituent structure in Mekens. Revista da ABRALIN 1(2): 51-74. doi:10.5380/rabl.v1i2.52693

Galucio, Ana Vilacy. 2011a. Nominalization in the Mekens language. Amerindia 35: 237-60.

Galucio, Ana Vilacy. 2011b. Subordinate adverbial constructions in Mekens. In: Rik van Gijn, Katharina Haude, Pieter Muysken (eds.). Subordination in native South American languages: 25-44. Amsterdam: John Benjamins.

Galucio, Ana Vilacy. 2014. Estrutura argumental e alinhamento gramatical em Mekéns. In: Luciana Storto, Bruna Franchetto, Suzi Lima (orgs.). Sintaxe e semántica do verbo em línguas indígenas do Brasil: 167-96. Campinas: Mercado de Letras.

Galucio, Ana Vilacy, Antonia Fernanda de Souza Nogueira. 2012. Comparative study of the Tupari branch of the Tupi family: contributions to understanding its historical development and internal classification. In: Memorias del V Congreso de Idiomas Indígenas de Latinoamérica, 6-8 de octubre de 2011, Universidad de Texas en Austin. Austin: Center for Indigenous Languages of Latin America (CILLA). Retrieved from <https://ailla. utexas.org/sites/default/files/documents/Galucio_Nogueira_CILLA_V.pdf>, accessed on 02/07/2020.

Galucio, Ana Vilacy, Antonia Fernanda de Souza Nogueira. 2018. From object nominalization to object focus: the innovative A-alignment in the Tuparian languages (Tupian family). Journal of Historical Linguistics 8(1): 95127. doi:10.1075/jhl.16025.gal

Galucio, Ana Vilacy, Mercedes Guaratira Sakyrabiar, Manoel Ferreira Sakyrabiar, Rosalina Guaratira Sakyrabiar, Olimpio Ferreira Sakyrabiar. 2017. Sakurabiat. In: Kristine Stenzel, Bruna Franchetto (eds.). On this and other worlds: Voices from Amazonia: 331-51. Berlin: Language Science Press. doi:10.5281/zenodo.1008787

Galucio, Ana Vilacy, Nilson Gabas Júnior. 2002. Evidências de agrupamento genético Karo-Puruborá, tronco Tupi. In: XVII Encontro Nacional da ANPOLL: boletim informativo n° 31: 163. Porto Alegre: Universidade Federal do Rio Grande do Sul.

Galucio, Ana Vilacy, Sérgio Meira, Joshua Birchall, Denny Moore, Nilson Gabas Júnior, Sebastian Drude, Luciana Storto, Gessiane Picanço, Carmen Reis Rodrigues. 2015. Genealogical relations and lexical distances within the Tupian linguistic family. Boletim do Museu Paraense Emilio Goeldi. Ciencias Humanas 10(2): 229-74. doi:10.1590/1981-81222015000200004 Hanke, Wanda, Morris Swadesh, Arion Dall'Igna Rodrigues. 1958. Notas de Fonologia Mekens. In: Juan Comas (ed.). Miscellanea Paul Rivet octogenario dicata, vol. II: 187-217. Mexico City: Universidad Nacional Autónoma de México.

Landin, David. 2005. Dicionário e léxico Karitiana / portugués. 2nd ed. Cuiabá: Sociedade Internacional de Lingüística. Lacerda, María Conceiçâo de. 2014. Bekä Pamakube (lugar de aprender). Aprendendo com os Zoró: análise da identidade indígena através da experiencia das escolas nas aldeias do povo indígena Zoró. Ph.D. dissertation, Universidad de Salamanca. doi:10.14201/gredos.124106 Lévi-Strauss, Claude. No date. Vocabulaires de trois langues: Kabisiana, Kepkiri-uat et de Säo Pedro. Ms. Paris. Lévi-Strauss, Claude. 1950. Documents Rama-Rama. Journal de la Société des Américanistes 39: 73-84. doi:10.3406/jsa.1950.2381

Loukotka, Cestmír. 1963. Documents et vocabulaires inédits de langues et de dialectes sud-américains. Journal de la

Société des américanistes 52: 7-60. doi:10.3406/jsa.1963.2001 Loukotka, Cestmír. Classification of South American Indian languages. Los Angeles: Latin American Center,

University of California; Caracas: Sucre, 1968. (Reference Series, v. 7.) Meira, Sérgio, Sebastian Drude. 2015. A summary reconstruction of Proto-Maweti-Guarani segmental phonology.

Boletim do Museu Paraense Emilio Goeldi. Ciéncias Humanas 10(2): 275-96. doi:10.1590/1981-81222015000200005 Mello, Antonio Augusto Souza. 2000. Estudo histórico da familia lingüistica Tupi-Guarani: aspectos fonológicos e lexicais.

Ph.D. dissertation, Universidade Federal de Santa Catarina. Mondini, Juliana Nazatto. 2014. Yudja Utaha: a culinária juruna no Parque Indígena Xingu - uma contribuiçâo ao dicionário bilingue juruna-portugués. MA thesis, Universidade Estadual Paulista "Júlio de Mesquita Filho" (Araraquara campus).

Moore, Denny, Ana Vilacy Galucio. 1994. Reconstruction of Proto-Tupari consonants and vowels. In: Margaret Langdon, Leanne Hinton (eds.). Proceedings of the Meeting of the Society for the Study of the Indigenous Languages of the Americas, July 2-4, 1993, and the Hokan-Penutian Workshop, July 3, 1993, both held at the 1993 Linguistic Institute at Ohio State University in Columbus, Ohio: 119-37. Berkeley: Survey of California and other Indian Languages. Nikulin, Andrey. Submitted. O Kabisiana de Lévi-Strauss é um codialeto do Sakurabiat (Mekéns).

Nikulin, Andrey, Fernando O. de Carvalho. 2019. Estudos diacrônicos de línguas indígenas brasileiras: um

panorama. Macabéa - Revista Eletrônica do Netlli 8(2): 255-305. Nikulin, Andrey, Mário André Coelho da Silva. 2020. As línguas Maxakalí e Krenák dentro do tronco Macro-Jê.

Cadernos de Etnolingüistica 8(1): 1-64. Nikulin, Andrey, Serafima Muzykantova. In preparation. A phonological reconstruction of Proto-Ta-Arawakan. Nogueira, Antonia Fernanda de Souza. 2011. Wayoro êmêto: fonologia segmental e morfossintaxe verbal. MA thesis,

Universidade de Säo Paulo. doi:10.11606/D.8.2011.tde-22102012-093617 Nogueira, Antonia Fernanda de Souza. 2015. Dissimilaçâo em Wayoro (família Tuparí, tronco Tupí). In: Larissa Lima, Marília Ferreira, Eunice Braga, Jaqueline Reis, Zerben Aguiar (orgs.). Anais do IX Congresso Internacional da ABRALIN: 611-18. Belém: ABRALIN, PPGL/UFPA. Nogueira, Antonia Fernanda de Souza. 2019. Predicaçâo na lingua Wayoro (Tupi): propriedades de finitude. Ph.D.

dissertation, Universidade de Säo Paulo. doi:10.11606/T.8.2019.tde-14082019-101006 Nogueira, Antonia Fernanda Souza, Ana Vilacy Galucio, Nicole Soares-Pinto, Adam Roth Singerman. 2019. Termos de parentesco nas línguas Tuparí (família Tupí). Boletim do Museu Paraense Emilio Goeldi. Ciéncias Humanas 14(1): 33-64. doi:10.1590/1981.81222019000100004 Picanço, Gessiane Lobato. 2005. Mundurukú: phonetics, phonology, synchrony, diachrony. Ph.D. dissertation, University of British Columbia. doi:10.14288/1.0092991 Picanço, Gessiane Lobato. 2019. A fonologia diacrônica do Proto-Mundurukú (Tupi). Curitiba: Appris. Ramirez, Henri, Valdir Vegini, Maria Cristina Victorino de França. 2017. O warázu do Guaporé (tupi-guarani): primeira descriçâo linguística. LIAMES 17(2): 411-506. doi:10.20396/liames.v17i0.8647468

Reiter, Sabine. 2011. Ideophones in Awetí. Ph.D. dissertation, Christian-Albrechts-Universität zu Kiel.

Ribeiro, Maria de Jesus Pacheco. 2010. Dicionário Sateré-Mawé/portugues. MA thesis, Universidade Federal de Rondónia (Guajará-Mirim campus).

Rodrigues, Aryon Dall'Igna. 1958. Die Klassifikation des Tupi-Sprachstammes. In: Proceedings of the Thirty-Second International Congress of Americanists, Copenhagen 1956: 679-84. Copenhagen: Munksgaard.

Rodrigues, Aryon Dall'Igna. 1985. Evidence for Tupi-Carib Relationships. In: Harriet E. Manelis Klein, Louisa R. Stark (eds.). South American Indian languages: retrospect and prospect: 371-404. Austin: University of Texas Press.

Rodrigues, Aryon Dall'Igna. 2002. Correspondencias lexicais e fonológicas entre Tupí-Guaraní e Tuparí. In: Ana Suelly Arruda Cámara Cabral, Aryon Dall'Igna Rodrigues (orgs.). Línguas indígenas brasileiras: Fonologia, gramática e historia. Atas do I Encontro Internacional do Grupo de Trabalho sobre Línguas Indígenas da ANPOLL. Tomo I: 288-97. Belém: Editora da Universidade Federal do Pará.

Rodrigues, Aryon Dall'Igna. 2005. As vogais orais do Proto-Tupí. In: Aryon Dall'Igna Rodrigues, Ana Suelly Arruda Cámara Cabral (orgs.). Novos estudos sobre línguas indígenas: 35-46. Brasília: Editora UnB.

Rodrigues, Aryon Dall'Igna. 2007. As consoantes do Proto-Tupí. In: Ana Suelly Arruda Cámara Cabral, Aryon Dall'Igna Rodrigues (orgs.). Línguas e culturas Tupí, vol. I: 167-203. Campinas: Curt Nimuendajú; Brasília: LALLI/UnB.

Rodrigues, Aryon Dall'Igna, Ana Suelly Arruda Cámara Cabral. 2012. Tupían. In: Lyle Campbell, Veronica Gron-dona (eds.). The Indigenous languages of South America: a comprehensive guide: 495-574. Berlin/Boston: Mouton de Gruyter. (The World of Linguistics, 2). doi:10.1515/9783110258035.495

Rodrigues, Aryon Dall'Igna, Franz Caspar. 2017. Esbogo da gramática da língua tuparí. Brasília: LALLI/UnB.

Sekelj, Tibor. 1948. Wordlist Aruá-Makurap-Zaboti-Arikapú-Tupari. Ms. Retrieved from http://etnolinguistica.wdfiles. com/local--files/sekelj%3A1/Sekelj_Wordlist_Aru%C3%A1_Makurap_Zabot%C3%AD_Arikap%C3%B3_Tupar %C3%AD.pdf, accessed on 28/08/2020.

Silva, Bazilio da, Nilzimara de Souza Silva, Odamir de Oliveira (orgs.). 2013. Paradakary urudnaa: dicionário Wapichana/portugues, portugues/Wapichana. Boa Vista: EDUFRR.

Singerman, Adam Roth. 2016. Nasal harmony and phonotactic well-formedness in Tupari. International Journal of American Linguistics 82(4): 453-85. doi:10.1086/688603

Singerman, Adam Roth. 2018. The morphosyntax of Tuparí, a Tupían language of the Brazilian Amazon. Ph.D. dissertation, University of Chicago. doi:10.6082/ns2a-cv36

Snethlage, Emil Heinrich. 2015. Die Guaporé-Expedition (1933-1935): in Forschungstagebuch (aus dem Nachlass herausgegeben von Rotger M. Snethlage, Alhard-Mauritz Snethlage und Gleice Mere). Wien/Köln/Weimar: Böhlau Verlag. doi:10.7788/9783412218591

Valenzuela Bismarck, Pilar. 2011. Contribuciones para la reconstrucción del proto-cahuapana: comparacion léxica y gramatical de las lenguas jebero y chayahuita. In: Willem F. H. Adelaar, Pilar Valenzuela Bismarck, Roberto Zariquiey Biondi (eds.). Estudios en lenguas andinas y amazónicas. Homenaje a Rodolfo Cerrón-Palomino: 271-304. Lima: Fondo Editorial Pontificia Universidad Católica del Perú.

Voort, Hein van der. 2005. Kwaza in a comparative perspective. International Journal of American Linguistics 71(4): 365-412. doi:10.1086/501245

Voort, Hein van der. 2007. Proto-Jabutí: um primeiro passo na reconstruyo da língua ancestral dos Arikapú e Djeoromitxí. Boletim do Museu Paraense Emílio Goeldi. Ciencias Humanas 2(2): 133-68. doi:10.1590/S1981-81222007000200007

Wetzels, W. Leo, Andrew Nevins. 2018. Prenasalized and postoralized consonants: The diverse functions of enhancement. Language 94(4): 834-66. doi:10.1353/lan.2018.0055

Xerez, Sebastiao Moacyr. 1946. Pequeno vocabulário dos dialetos Macurape, Mondé e Caritiana. Manaus.

А. Никулин, Р. Андради. Развитие и падение аппроксимантов в языках тупарийской группы

В статье рассматривается развитие серии аппроксимантов в языках тупарийской группы (тупийская семья), изначально распространённых между средним течением реки Гуапоре/Итенес и верховьями Машаду/Жи-Парата (юг штата Рондония, Бразилия). Показано, что помимо исходной серии пратупарийских аппроксимантов (в нашей ре-

конструкции, *в, *] и *и>) в некоторых языках-потомках развились аппроксиманты и из других источников — из гласных ненижнего подъёма (*о/*г), посторализованных носовых (*mb/*nd/*rJg, через ступень *Ъ/*й/*^) и глайдов, заполняющих зияние. Детально обсуждается эволюция этих звуков; в частности, показывается, что во всех тупарийских языках хотя бы некоторые бывшие аппроксиманты подверглись фортиции. Особое внимание уделяется внутренней классификации тупарийской группы.

Ключевые слова: тупарийская группа; тупийская семья; аппроксиманты; фортиция; сравнительно-исторический метод.