Proto-Tupi-Guarani had no palatalized velar stop

O Proto-Tupi-Guarani não tinha uma oclusiva velar palatalizada

Fernando Carvalho fernaoorphao@mn.ufrj.br

Universidade Federal do Rio de Janeiro, Brasil

The goal of this paper is to resolve one of the still open issues in the phonological reconstruction of Proto-Tupi-Guarani (PTG), the shared ancestor of the largest branch of the Tupian language family. I will show that the palatalized velar stop *k^j (whose status has been recently called into question; Meira & Drude, 2015, p. 282, fn. 7), can be eliminated from the reconstructed PTG inventory, and the relevant correspondences can be more insightfully analyzed as the result of language specific developments in Kayabí, Kagwahiva, Tenetehára, Guarayu, Kamayurá, and Ka’apor. The paper is organized as follows: After a presentation of the current standing of this question (‘the current view’), I will discuss the segmental correspondences in a representative sample of ten, well-attested TG languages, based on which a PTG plain velar stop *k can be straightforwardly reconstructed. Next, I will show that overlapping correspondences with diverging reflexes in a subset of these languages can be accounted for by invoking language-specific developments of the same PTG *k, with no need for an independent and contrasting PTG velar stop (‘PTG *k and its reflexes’). All the relevant correspondences have been extracted from cognate sets that appear in the Appendix to the paper. In the section entitled ‘some implications’ I briefly discuss how this finding eliminates the need to postulate an unmotivated split of Proto-Tupian **k^j into PTG *k and *k^j. Finally, the section ‘conclusions’ is devoted to the synthetic presentation of the findings in the paper.

In her overview of the then current understanding of the Tupi-Guarani language family, Jensen (1999, p. 139) notes that PTG *k^j is reconstructed for three morphemes: *ik^jé ‘to enter’, *k^jér ‘to sleep’ and *k^jé ‘here, near speaker’. According to her, the change in the reconstructed forms - previously uniformly reconstructed with *k - was deemed necessary to account for the Guarayu form *k^je ‘sleep’ in Hoeller’s (1932) data. The contrast between *k^j and *k is reconstructed for PTG by Mello (2000) and by Rodrigues (2007). Mello (2000) reconstructs *k^j in *-k^jer ‘sleep’ only, while Rodrigues (2007) has *k^j in *-k^jer ‘sleep’ and *-ejk^je ‘enter’. Cognate sets in languages other than Guarayu would presumably support this proto contrast, such as Kayabí set ‘to sleep’ and se ‘to enter’, and the change *e > i in this context in Parintintin: kir ‘sleep’ and ki ‘here’. Meira and Drude (2015, p. 281), in a paper focused on the comparison between PTG and its two closest relatives, Awetí and Mawé, note that *k^j has an uncertain status at the PTG level, being reconstructed only preceding *e in works such as Mello (2000) and Rodrigues (2007). The authors offer a convenient summary of the status of the phonological problem:

Mello has only four cases of PTG *ke: *kerap ‘to close’, *keramu ‘to snore’, *purake ‘electric eel’ and *ukeʔi (doubtful) ‘sister/brother-in-law’ (the latter apparently related to Man’s Older Brother). Mello claims that *k and *k^j have different reflexes in Siriono, Apiaka, Kayabí, Urubu-Kaapor and (sometimes) Tembe, but, in his data, (a) these languages are all missing in the sets for *kerap and *ukeʔi; (b) only Sirionó occurs in the *keramu cognate set, where it has the same reflex (kenãmu with k) as in *k^jet (> ke, also with k); and (c) in *purake, Tembe and Urubu-Kaapor both occur with k (murake, purake), while in *k^jet only the Urubu-Kaapor reflex is different (ʃer with ʃ), while the Tembe reflex is simply ker, with the same k as in *purake. There is thus almost no evidence in Mello (2000) to support a distinction between PTG *ke and *k^je

The situation is, in fact, more difficult for the proponents of this contrast than the Meira and Drude (2015) quote above suggests. First, note that the number of supporting etymologies falls from four to three, once it is recognized that the PTG etymon meaning ‘to snore’, Mello’s (2000, p. 172)*keramu ‘roncar’ [to snore], is not independent from the etymon *k^jer ‘dormir’ [to sleep] (Mello, 2000, p. 176), but is likely a reflex of the derivative *ket-amu ‘to snore (while) sleeping’, as shown by Old Guarani aquerambu ‘roncar’ [to snore], ambu ‘ronquido’ [snoring sound], tayaçu apĭîmbu ‘de puerco’ [pig’s snoring sound] (Restivo, 1893 [1722], p. 482), and Old Tupi Xequerambû ‘roncar, o que dorme’ [to snore, he/she who sleeps], Xeambû ‘roncar o porco’ [to snore, the pig] (Drumond, 1952, p. 108)1. As suggested below, the ‘sleep’ and ‘snore’ sets where phonologically segregated in Mello’s (2000) reconstruction only because his set for *keramu ‘to snore’ fails to include cognates from some languages such as Kayabí, which, in his view, are critical for reconstructing *k^j, while these same languages do contribute witnesses to the ‘sleep’ set. Second, *kenaβ ‘fechar’ [to close] (Mello, 2000, p. 172) is very doubtful and not clearly reconstructible for PTG (see next section). Third, the supposed Ka’apor reflex ʃer ‘to sleep’ is a non-existent ghost form (more on this below). Fourth, as suggested by Meira and Drude (2015) and demonstrated in the remaining of this paper, the other sets are not problematic at all, pointing to language-specific developments and not to the independent reflexation of a separate PTG segment.

Before proceeding, however, I would like to highlight a generalization about the sound structure of PTG that has not been so far explicitly commented upon, but which is relevant for the evaluation of the issue at hand. This generalization will also provide a background for the synthesis of the current understanding of the putative contrast between *k and *k^j.

An examination of all extant proposals on the reconstruction of PTG etyma (Lemle, 1971; Schleicher, 1998; Mello, 2000) reveals that the sequence *ki is not reconstructed, as shown in Table 1, where examples for each of the reconstructed sequences *kv (where v = any vowel) are given for each PTG source2.

Table 1
Vocalic contexts for PTG *k in published comparative reconstructions.

As noted by Meira and Drude (2015), this putative contrast between PTG *k and *k^j is attested only in the context of a following *e, which strongly suggests that this palatalization is a secondary effect of the contextual front vowel, the only PTG front vowel that was found in this context. The PTG etyma in Rodrigues and Dietrich (1997, pp. 273-274) which exemplify the contrast are: *ɨkeʔɨr ‘brother of man, younger’, *ɨke ‘side of the body’ vs. *k^jer ‘sleep’, *ek^je ‘go in’. Mello (2000, pp. 163, 172, 176, 184, 191) gives only *k^jer ‘sleep’ for PTG *k^j, as opposed to *k in the same context (that is, preceding *e) in the forms for: *keramu ‘to snore’, *oken ‘door’, *ike ‘to enter’, *kenaβ ‘to close’, *purake ‘electric eel’ and *ukeʔi ‘brother/sister-in-law’. Other studies give only *k, as in Lemle (1971) and Schleicher (1998), where *ker ‘sleep’ is the only case of a *ke sequence. Jensen (1999, p. 139) presents *ik^je ‘enter (to)’, *k^jer ‘sleep (to)’ and *k^je ‘here, near the speaker’, as putative examples of PTG *k^j but does not discuss explicitly the existence of contrasts.

We will employ a sample of TG languages for addressing this specific aspect of PTG sound structure. The set of languages compared, given below in Table 2, includes languages for which relatively significant documentation is available, and which comprehensively represent the internal diversity of the family, as indicated by their classification within the two major, extant proposals on the internal classification of TG languages (that of Rodrigues, 1984/1985, later updated in Rodrigues & Cabral, 2002, and that of Michael et al., 2015)3.

Table 2
Position of languages used for reconstruction in each of the existing internal classifications.

The relevant correspondences, identified for the cognate sets featuring in the Appendix to this paper, are given in (1) below. Each correspondence is followed by the semantic glosses that identify the cognate sets featuring the correspondence in question5.

The correspondence in (I) is the main (identity) correspondence that establishes PTG *k. The correspondences in (II) and (III) are the two correspondences that have been accounted for by postulating a separate PTG segment *k^j. These correspondences are not only attested in fewer sets than is the case with (I), but as noted above, also happen to be contextually very limited, and occur in contexts that are complementary to those of the identity correspondence (I). The identity correspondence for *k is attested in a variety of vocalic contexts: initially preceding u (Woman) and ɨ (Knife); medially between e_ũ (Tongue), a_u (Hot), u_u (Long), a_a (Cayman), a_ã (Head), e_a (Look for), o_õ (Swallow), a_ɨ̃ (Wet), u_a (Kill), u_ɨ (Salt) and (ɨ)_o (Dig). The three etymologies that support correspondences (II) and (III) show the presumed reflexes of PTG *k in a single context: that of a following *e, as shown in Table 3, where the most important reflexes are highlighted by cell shading.

Table 3
Cognate sets instantiating correspondences II and III.

Correspondences (II) and (III) are jointly distinct from (I) due to a series of ‘palatal’ reflexes in Tenetehára, Kayabí, Kamayurá, Guarayu and Ka’apor. Given the complementary distribution of these correspondences, both (II) and (III) are best reconstructed as reflecting *k, just like (I), with special, context-specific developments taking place in the diverging languages. Note that the two upper rows in the table show that *ɨke ‘side of the body’, which has never been reconstructed with *k^j, shows, nevertheless, the same reflexes as *-k^jer ‘sleep’, which is reconstructed with PTG *k^j in every study that recognizes the distinction. If *k is reconstructed in all these cases, the following developments are implied for each of the five languages:

For Kayabí, a search through Weiss’ (2005) dictionary reveals that ke is an unattested sequence, which supports the regular operation of *k > s /_*e. Note that in ‘side of the body’, which is not reconstructed with *k^j in the extant literature, Kayabí has *k > s, exactly as it does in the cases of ‘sleep’ and ‘enter’, both of which are usually reconstructed as having *k^j (see section ‘the current view’). This shows that Kayabí offers no evidence for the recognition of two distinct PTG velar stops.

The facts of Guarayu are the same as those of Kayabí, although the languages have phonetically distinct reflexes for *ke. Jensen (1999, p. 139) claims that the postulation of PTG *k^j was motivated, in part, by the existence of k^je in Alfred Hoeller’s data on Guarayu. The problem is that there is no ke in Guarayu and that all cases of PTG *ke show up as k^je in the language (cf. ìquie ‘die Seite des menschlichen Körpers’; aquie ‘Ich schlafe, ruhe’; aiquie ‘Ich trete ein’; Hoeller, 1932, pp. 90, 102, 210). A search in Danielsen et al. (2019) shows that in all cases where their data has ke, the same form in the Hoeller (1932) materials has k^j <quie>. It seems that ke → [k^j] is a purely allophonic process in Hoeller’s Guarayu, one that affects the pronunciation of loanwords too, as in kesu ‘cheese’ (< Spanish queso), where Hoeller (1929, p. 88, quoted in Danielsen et al., 2019) registers a variant <quiezu> ‘Käse’. There is no obstacle then for the postulation of *ke > k^je in the language, with the implication that Guarayu k^j offers no evidence whatsoever for the postulation of a separate PTG proto-segment6.

For Kamayurá, *k > ts /*i_*e only within morphemes, which makes it difficult for assessing the regularity of the development since the environment is very specific. There does not seem to be any other currently reconstructible PTG morpheme, other than *-ike ‘to enter’, where a sequence *-ike- is found. That the development did not take place inter-morphemically is shown by the fact that Kamayurá -ket ‘to sleep’, when prefixed with the Set II third person marker i-, retains the velar stop as such (see Seki, 2000, p. 343, for an example). This restriction to tautomorphemic contexts does not seem to be unique in the family, as noted below for Tenetehára, and it is active even in languages where the effect is simply variation in the existence or not of secondary palatalization k → [k^j]. This seems to be the case of Old Guarani, where optional palatalization takes place in the reflex of *-ike ‘to enter’ (cf. e.g., yque ~ quié ‘entrar’ [to enter], aiquie ‘yo entro’ [I enter], Teiquîe ~ teique ‘entrar’ [entry]; Montoya, 1639, p. 376), but not in the reflex of *-ket ‘I sleep’, when it is preceded by the Set III7 first person singular prefix wi- (cf. aque ‘yo duermo’ [I sleep], but: guiquebo; Montoya, 1639, p. 330).

For Tenetehára, Jensen (1999, p. 139) argues that the medial affricate in -iʧe ‘to enter’ must be a reflex of *k^j, and not a contextual, palatalized reflex of *k conditioned by the preceding *i. As evidence for this claim she cites the diachronic correspondence ikó < *-ikó ‘to be in motion’, which would be evidence that *i had no general palatalizing effect upon a following *k in Tenetehára. Note, however, that the two cases are not entirely comparable, and that the palatalization *k > ʧ in Tenetehára could have applied only when preceded by *i and followed by *e, thus making the existence of -ikó in the modern language unsurprising. Moreover, it is not clear what the source for this presumed form -iko in Tenetehára is. While often given as a separate entry, for instance, as iko ‘morar, viver, ser, estar’ [dwell, live, be, stay] (Boudin, 1978, p. 73), the -i in this case results from diphthong formation whenever a preceding prefix vowel is added (as a-iko ‘eu moro’ [I dwell], u-iko ‘êle está’ [he is] (Boudin, 1978, p. 73), and it reflects, in fact, an underlying e, which is present when no preceding vowel occurs, as in the third person form hêkó- (Boudin, 1978, p. 60). Although the verb in question does have a third person ikó, rather than -ekó, when used as a positional auxiliary, this fact carries no weight in rehabilitating Jensen’s proposal. As noted by Bendor-Samuel (1972, p. 130) for the Guajajára dialect of Tenetehára, the verb ikó has a third person i- in this function, and it is not implausible that the apparently root-initial i- in this case is just the third person prefix in question (that is: *i-eko > iko). Finally, see that, as in Kamayurá and Old Guarani, palatalization of *-ke by a preceding *i occurs only morpheme-internally.

The more attentive reader may have noticed yet another development possibly tied to the reflexation of PTG *k. The Kagwahiva forms in Table 3 display a diachronic correspondence *e > i for the vowel following *k8. Jensen (1999, p. 139) appeals to this Kagwahiva development *e > i as evidence for the presence of an earlier secondary palatalization in the preceding *k, that is, as evidence for *k^j. However, Kagwahiva shows *ki both in sets that have been analyzed in the literature as evidence for PTG *k^je, such as ‘sleep’, and in sets that have been reconstructed as *ke, such as ‘side of the body’, and thus offers no evidence whatsoever of separate and contrasting reflexes (see the etymologies in the Appendix). It is likely that PTG *ke [k^je] > ki in Kagwahiva, with the precursor phonetic palatalization of *k preceding *e being not only phonetically natural but attested elsewhere in family, as noted above for Guarayu. Further evidence for this intermediate stage with phonetic palatalization [k^je] as a condition for the change is the independent evidence for *e > i in the context of a preceding palatal approximant *j, as in -nhi’ig̃ ‘speak’ (Betts, 2012, p. 188), from PTG *-jeʔẽŋ ‘to speak’ (Schleicher, 1998, p. 352), -kyhyij ‘afraid’ (Betts, 2012, p. 156) < *ʧɨkɨje ‘fear’ (Schleicher, 1998, p. 341) ‘fear’ and in the reflexive prefix ji- (Betts, 2012, p. 121) < *je- ‘reflexive’ (Jensen, 1998, pp. 515-516)9.

Correspondence (IV) differs from the identity correspondence (I) only in the Ka’apor reflex ʃ alternating with k. As noted in Meira and Drude (2015) quote in the section ‘the current view’, Ka’apor ʃ has been suggested as this language’s reflex for the presumed PTG *k^j, in contrast to *k > k. Any discussion of the potential evidence offered by Ka’apor reflexes for the reconstruction of PTG *k^j must consider a well-known innovation specific to Ka’apor which consists of the palatalization of *k to ʃ when preceded by *i (Silva, 1997, pp. 49-50; Jensen, 1999, pp. 139-140). This produces alternations in the case of *k-initial PTG roots/stems, which show ʃ in their third person forms alternating with k- elsewhere in their paradigms. Table 4 presents diachronic correspondences between PTG nouns and their reflexes in Ka’apor, illustrating the effects of the Set II *i- prefix on the initial *k-.

Table 4
Diachronic correspondences for PTG *i-k- > Ka’apor i-ʃ-.

Mello (2000, pp. 257-313) gives two cases where Ka’apor would have a ʃ reflex for a PTG velar stop, one in the reflex for his PTG *k^jer ‘sleep’ and the other in the set for PTG *kɨʔa ‘dirty’10. First, note that the claim that Ka’apor has ʃ as a reflex of PTG *k in the form for ‘sleep’, as in the Mello (2000, p. 176) etymology for his PTG *k^jer ‘sleep’, is factually incorrect: The form attested is -ker, as in u-ker ‘ele dorme’ [he sleeps] (Kakumasu & Kakumasu, 2007, p. 141). In agreement with the development PTG *i-k- > i-ʃ-, what Ka’apor does have is a derivative of -ker ‘to sleep’ which shows the expected palatalization when preceded by the Set II third person prefix i-: i-ʃerai ‘ele sonha’ [he dreams], as opposed to ihẽ kerai ‘eu sonho’ [I dream] (Kakumasu & Kakumasu, 2007, p. 193). It is possible that Mello (2000) has incorrectly coded the form for ‘dream’ in the ‘sleep’ set, but one cannot be sure about it, as the cognates in Mello’s (2000) etymologies are not sourced. For the set for ‘dirty’, the existence of the third person ʃiʔa ‘it is dirty’ (Kakumasu & Kakumasu, 2007, p. 43) suggests an error in the same direction. Therefore, the supposed evidence for PTG *k^j in the form of a Ka’apor reflex ʃ in the set for ‘to sleep’ (see section ‘the current view’) is non-existent. Finally, see that in correspondence (III) the reflex of PTG *-ike ‘to enter’ has the expected ʃ reflex in Ka’apor for medial *-k-.

Two etymologies call for separate discussion since they apparently breach the pattern of complementary distribution observed for the correspondences (I) and (II-III). These are the terms for ‘husband’s sister’ and ‘elder brother’, which were included in correspondence (I) in (1). The two involve etyma with *ke sequences, just like the sets for correspondences (II) and (III) (see Table 3). However, the recognition of sporadic and language-specific developments, in addition to missing forms (due either to poor documentation or actual lexical replacement), allow one to account for this exceptionality without invoking an additional PTG proto-segment. The relevant cognate sets appear in Table 5, again with cell shading highlighting the most noteworthy data.

Table 5
Cognate sets displaying unexpected correspondences for PTG *ke.

The Kayabí reflexes are the first to strike the eye: The expected reflex of PTG *ke in the language is se, not ki. For PTG *-ukeʔi ‘husband’s sister’ (see Carvalho & Birchall, 2022), one finds a Kayabí form -ukiʔi ‘cunhada da mulher’ [woman’s sister-in-law] (Weiss, 2005, p. 109). The Kayabí have, however, in historical times, lived in a region geographically close to that of the Kagwahiva, in the Upper Tapajós river, with which they display cultural and historical affinities (Aguilar, 2017; Menendez, 1989, pp. 6-7). Since the development *ke > ki evidenced by the Kayabí form is a regular Kagwahiva development, the best explanation, for the moment, is that Kayabí -ukiʔi is a Kagwahiva loan, even though the form seems to have been lost in Kagwahiva itself.

The same unexpected sequence ki is again attested in the Kayabí reflex of *-t-ɨket-ʔɨt ‘elder brother’. In this case, however, Kayabí, Wajãpi and Kagwahiva show a sporadic vowel metathesis: *-t-ɨket-ʔɨt > KAY -reki-ʔɨt : WAJ -lɛkɨʔɨ : KAG -rekɨʔɨr. Although sporadic, metathesis is not unparalleled within TG, having targeted at least two other etyma: *-kɨpɨ-ʔɨt ‘younger sister, female Ego’, which has a reflex pɨkɨ-ʔɨt in some languages (Carvalho & Birchall, 2022), and *tsɨkɨje ‘to fear’, with reflexes such as Kaiowá kɨhɨje (adapted from Schleicher, 1998, p. 341; see the etymologies in the Appendix of the present paper for comments on this particular etymon).

As noted in the section ‘the current view’, there are four cognate sets that are usually addressed in discussions of the issue of PTG *k^j, but that have not been discussed here so far: ‘electric eel’, ‘door’, ‘to close’ and ‘to snore’. Since these are offered as cases of (non-controversial) PTG *ke, they will not add any evidence for reconstructing *k^j and, for this reason, they will be only briefly discussed here.

PTG *keramu ‘to snore’ (e.g., Mello, 2000, p. 172) is, as noted before, a derivative of *-ket ‘to sleep’. Inspection of the relevant etymology in the Appendix reveals that the reflexation of *k in this set is identical to that of *-ket, and hence, offers no evidence for a separate reflex. One can only speculate on the reasons that have led Mello (2000) to reconstruct an apparent contrast in the initial stops of *keramu ‘roncar’ [to snore] (Mello, 2000, p. 172) and *k^jer ‘dormir’ [to sleep] (Mello, 2000, p. 176), though the lack of a Kayabí cognate for in the former set, versus the Kayabí cognate with s- in the latter, have mislead him into recognizing two separate correspondences.

The three other sets, although often reconstructed for PTG, have distributional problems, and these will be addressed here for the sake of completeness. They have not been included in the etymologies featuring in the Appendix. A form like *oken is often reconstructed for the meaning ‘door’ in PTG (Rodrigues & Dietrich 1997, p. 273; Mello, 2000, p. 184; Meira & Drude, 2015, p. 292), though the cognates are restricted to Old Tupi Oquẽna ‘porta’ [door] (Drumond, 1953, p. 83), Tenetehára uken ‘porta’ [door] (Harrison & Harrison, 2013, p. 157), Guarayu oquienda ‘die Türe’ [the door] (Hoeller, 1932, p. 159), Old Guarani oquȇna ‘puerta’ [door] (Restivo, 1893 [1722], p. 455) and Ka’apor huken ~ hukwen ‘porta’ [door] (Kakumasu & Kakumasu, 2007, p. 96). That is, the form seems essentially restricted to the non-Amazonian TG languages and to languages that are, in some internal classifications of TG languages, suggested as having a rather close relation to Old Tupi: Tenetehára and Ka’apor (see e.g., Michael et al., 2015; Gerardi & Reichert, 2021). Quite telling is the absence of a cognate in Kayabí (-‘okwat ‘porta’ [door] – Weiss, 2005, p. 165) and in the Kagwahiva lects (where an extension of -juru ‘mouth’, or, like Kayabí, of -kwat ‘hole’, is used instead; see Betts, 2012, p. 125)11. Although consideration of a larger sample of languages (cf. Xingu Asurini ukina ‘porta’ [door], Pereira, 2009, p. 85) and of external, non-TG evidence (Meira & Drude, 2015, p. 292) make a PTG provenance for this set virtually safe, it offers no other insight on the reconstruction of the *k^j-k contrast.

As noted before (‘the current view’), *purake ‘poraquê’ (Mello, 2000, p. 191), the name of a kind of fish or electric eel, is one of the forms traditionally discussed in the literature where PTG *k would be attested preceding *e. There are, however, both formal and distributional issues. Formally, the existence of forms with initial m (Tenetehára murake ‘poraquê’; Harrison & Harrison, 2013, p. 113) matching forms with a supposedly etymological p- (Tocantins Asurini poraké ‘poraquê’; Cabral & Rodrigues, 2003, p. 194), often with both attested in the same language (Kagwahiva mburaki, puraki ‘electric eel’; Betts, 2012, p. 170) calls for adequate explanation. See that m : p correspondences, often with doublets in the same language, are expected in cases of Class Ib dependent nouns, where m- seems to code an unspecified possessor of the noun in question (Jensen, 1998, pp. 500-501, 1999, pp. 152-153). However, purake/murake, in languages that do have this item, is an independent noun, hence the correspondence cannot be accounted for in these morphological grounds. Second, the set lacks cognates in languages such as Kamayurá, Old Guarani and Guarayu and, although limited documentation prevents a simple inference of historical hypotheses, this is enough to command caution. There are other formal properties that call for explanation, such as Wajãpi having ɨ unexpectedly matching u in the other languages – see pɨlakɛ ‘Electropharus electricus’ (Grenand, 1989, p. 92) –, and the coexistence of two forms, pura and puraque in Old Tupi (see Cardim, 1925 [1583], p. 88; Marcgrave & Piso, 1648, p. 151).

Finally, the set for PTG *kenaβ ‘fechar’ [to close] is very limited in distribution already in Mello (2000, p. 172). Examination of comparative data reveals that there are a number of semantically close yet formally irreconcilable sets across TG languages, with some languages participating in multiple sets. Thus, an etymon #pemĩm is suggested12 by Old Tupi aipemim ‘cercar assi’ [to enclose] (Drumond, 1952, p. 70), Ka’apor jupimi ‘fechar o olho’ [to close eyes] (Kakumasu & Kakumasu, 2007, p. 117) and Kamayurá -pemi ‘fechar’ [to close] (Seki, 2000, p. 317), while Mello’s (2000)*-kenaβ is somehow13 related to Old Tupi Açoquendab ‘fechar porta’ [close door] (Drumond, 1952, p. 136), Tenetehára ukênaw ‘fechar, tapar buraco’ [close, close a hole] (Boudin, 1978, p. 282), Old Guarani oñoquendá ‘cerrar ventana o puerta sin llave’ [to close window or door without a key] (Restivo, 1893 [1722], p. 207). Tenetehára u-wàpytym ‘fechar’ [to close] (Harrison & Harrison, 2013, p. 183) and Wajãpi ɔ-wapɨ ‘fermer’ (Grenand, 1989, p. 59) suggest a third form with the same broad meaning. The fact that a single language, such as Old Tupi or Tenetehára, can participate in more than one set with semantically similar cognates suggests that independent etyma with meanings such as ‘enclose’, ‘close’, ‘cover with lid’ got confounded, either due to semantic extensions and replacement in some of the languages, or because the relevant sources are too coarse in the semantics of the material included. Be that as it may, Mello’s (2000)*-kenaβ, if accepted as a PTG etymon offers, at best, another instance of PTG *ke, and no evidence whatsoever for a PTG velar contrast in this context.

The proposal that PTG had a single velar stop *k offers not only the best account for the relevant comparative correspondences but also eliminates inconsistencies from the previous reconstruction with a *k - *k^j contrast. Jensen (1999, p. 139, fn. 22) noted, for instance, the anomalous character of the diachronic correspondence PTG *k^jer > Tenetehára ker ‘sleep’, since PTG *k^j predicts, in her account, a reflex ʧ in the language. No such anomaly exists under the current proposal.

In addition, there are implications of the findings reported here for our understanding of the diversification of the Tupian language family. Rodrigues (2007, pp. 180-181) reconstructs **k^j for the Proto-Tupian (PT) parent language, but this implies an unmotivated split in the PTG reflex: while **k^j merges with **k in **ɨk^jet > *-ɨker ‘irmã senior da mulher’ [older sister, female Ego], it is retained in **k^jet > *k^jer ‘dormir’ [sleep], in both cases the same phonetic context of a following **e > *e yields an unmotivated bifurcation of PT **k^j (see also Rodrigues, 2005, p. 40; Rodrigues & Cabral, 2012, pp. 505-507). The present reconstruction of PTG eliminates this unmotivated split. If PT must be reconstructed with a **k - **k^j contrast, PTG offers no special evidence in this respect, and the contrast was likely merged already at the Proto-Maweti-Guarani level (see Meira & Drude, 2015).

This paper has shown that there is no need to reconstruct a contrast between a plain velar stop *k and a palatalized velar stop *k^j for the parent language of the Tupi-Guarani family. All diachronic divergences from reconstructed etyma can be accounted for as conditioned developments of PTG *k, and one sporadic development, represented in (3) as diachronic replacements in specific segmental sequences:

The relatively lengthy discussion presented here in order to deal with one very specific issue on the reconstruction of PTG shows that a proper understanding of TG historical phonology requires more attention to detail and a more careful treatment of the comparative data than has been the case so far. If further progress in our understanding of the historical development of TG languages is to be attained, the practices of relying on a superficial treatment of correspondences, or what is worse, on a few supposedly conservative languages that are taken as proxies for PTG, should be left behind as features of the past of comparative TG historical linguistics.