Obscure cognates and lexical reconstruction : notes on the diachrony of the Xinguan Arawak languages

The present paper discusses evidence supporting the claim that the common Xinguan Arawak noun for ‘hand’, *wɨʂɨku, is a shared lexical innovation of this subgroup and that, in addition, a formation for ‘finger’ derived from a nominal compound with the roots for ‘hand’ and ‘head’, *kapɨ-tɨwɨ, sets Waurá, Mehinaku and Kustenaú apart from other languages of the family, including Yawalapiti. The reconstructed Proto-Arawak etymon for ‘hand’, *khapɨ, is preserved in the Xinguan Arawak languages only in the form of obscure cognates, instantiating interesting developments in lexical semantics as well as a relatively uncommon sound change in Yawalapiti. The discussion incorporates and addresses the historical linguistic significance of the earliest documentation of the Xinguan Arawak languages, the material gathered by Karl von den Steinen in 1887-1888. The analyses and data discussed highlight serious shortcomings in some of the reconstructed forms and diachronic developments advanced by Payne (1991), such as the postulation of a shift *a > ɨ in Waurá and the reconstruction of syllable-final *h for the Proto-Arawak language, one of the most controversial aspects of his comparative study.


INTRODUCTION
The goal of this contribution is to discuss some phonological and lexico-semantic developments in the Xingu subgroup of the Arawak language family.The arguments and data featured here bear directly on the historical development of the three extant languages forming this group -Waurá, Mehinaku and Yawalapiti -though I will, on occasion, address issues of significance to the historical linguistics of the Arawak family at large1 .
After a brief and selective outline of the classification of the Xingu Arawak languages in the next section, I will start off discussing the fact that these languages agree in that they lack a semantically-matched reflex of a Proto-Arawak (henceforth, PA) reconstructed etymon, *kʰapɨ 'hand' (see Payne, 1991, p. 407).Next, I show that a form *wɨʂɨku 'hand' can be reconstructed for the Proto-Xingu (PX) language, and this is argued to be a lexical innovation of the Xingu branch, replacing *kʰapɨ in its original function (defined grossly by a particular semantic slot in vocabulary lists).However, in order to backup the claim that this constitutes a Xingu-specific lexical innovation I will first examine and ultimately reject Payne's (1991) proposed derivation of this innovative form from a different PA etymon and, in doing so, I will suggest improvements to some of the PA reconstructions presented by Payne (1991).PA *kʰapɨ is, however, preserved in the form of (slightly) obscure cognates in the languages of the Xinguan branch.This claim is backed up, in the formal side, by a consideration of general sound changes characteristic of these languages (some of which are discussed in detail in Carvalho, in press) and, in their semantics, by a consideration of parallel shifts attested cross-linguistically.A more encompassing picture of the Xingu Arawak languages will also be provided, as I discuss their earliest attestations in the form of vocabulary data collected by Karl von den Steinen between 1887 and1888 (published in 1894), of particular relevance for the material on the already extinct Kustenaú language.In a final section I provide a synthesis and discuss some conclusions and implications of the claims presented in the paper.

THE XINGU SUBGROUP OF THE ARAWAK LANGUAGE FAMILY
Though neither an extensive lexical reconstruction nor a phonological proto-system has been proposed for the Proto-Xingu (PX) language, it is commonly accepted that the extant languages of this branch, Mehinaku, Waurá and Yawalapiti, form a (cladistic) subgroup.Mehinaku and Waurá are basically co-dialects of the same language and it seems that the extinct Kustenaú (Kustenau, Custenau) language fits in this Mehinaku-Waurá cluster as well (see Seki, 1999, p. 419;Corbera Mori, 2012, 2015).At a slightly higher level of phylogenetic inclusiveness, Paresi seems to join the Xingu branch in a Paresi-Xingu subgroup.This proposed internal grouping has been based so far almost exclusively on geographic factors, cultural similarities between these groups and impressionistically-assessed degrees of linguistic proximity, though somewhat more objective (though not necessarily helpful) estimates based on shared lexical retentions have been advanced as well (see Payne, 1991;Urban, 1992;Aikhenvald, 1999, p. 67;Franchetto, 2001).I am currently engaged in a historical-comparative investigation of the (Paresi-)Xingu branch and some evidence stemming from this work, and from the investigations of other researchers as well, suggest lexical and phonological innovations -some of which will be discussed here for the first time -that might support the recognition of this internal classification.These are given below in a simplified cladogram and a related character state distribution.
The characters (a) to (d) in Figure 1 are binary characters, each defining a property that is either present '1' or absent '0'.In this very simplified tree constructed for expositional purposes, each stroke in the tree indicates an innovation '0 ⇒ 1' (that is, all characters are assumed to be in state '0' at the root of the tree).Character (a) describes a language having different vowels in second and third person prefixes -*i and *ɨ respectively.In Carvalho (in press) I propose tentatively that this may be a shared innovation of the Paresi-Xingu subgroup2 .Characters (b) and (c) are discussed in the present paper.Character (b) consists in the presence of a form *wɨʂɨku for 'hand', which seems to constitute a lexical innovation of the Xingu subgroup, while character (c) refers to the existence of a nominal compound of the roots for 'hand' and 'head' to convey the meaning for 'finger'.Finally, character (d) consists of the development PA *-ʧi > -i for the Absolute suffix reconstructed by Payne (1991), plus some consequent morphophonemic developments, as discussed in Carvalho (2015).
Figure 1 does not indicate diachronic developments that are restricted to particular terminal taxa.These, though relevant to a complete understanding of the history of these languages, are uninformative as far as internal classification is concerned.In Carvalho (in press) I discuss some of these developments, such as *p > h and *ɨ > e in Paresi, the merger of *ɨ and *i in Yawalapiti and the lexical innovation of kuʂu for 'head' in Yawalapiti.In the remainder of the paper my attention will be focused on characters (b) and (c), as well as in the Xinguan reflexes of a particular etymon reconstructed for the PA ancestral language by Payne (1991).Data on the extinct Kustenaú language will be addressed sparingly, as only limited data is available (see von den Steinen, 1894, p. 529-530, for the original data and Corbera Mori, 2008, 2015 for some considerations about von den Steinen's material).

OBSCURE COGNATES, LEXICAL SUBSTITUTION AND SEMANTIC SHIFTS
In comparative lists of 'basic vocabulary' for Arawak languages, the forms recorded for the Xinguan languages (i.e.Waurá, Mehinaku, Kustenaú and Yawalapiti) which appear in the meaning slot for 'hand' look utterly unrelated to the form attested in most languages of the family (see for instance Ramirez, 2001a, p. 619, for 100-item Swadesh lists).The widespread, non-Xinguan, forms are analyzed by Payne (1991, p. 407) as the reflexes of a PA etymon *kʰapɨ.An impression of the overall distribution of the attested forms and the distinct character of those attested for the Xinguan languages is offered by a simple perusal of Table 1 below.I have included languages from almost every currently accepted or tentatively proposed subgroup of the Arawak family, including extinct languages of the central Amazon and Solimões area whose forms for 'hand' can be straightforwardly interpreted as compatible with the proto-form reconstructed by Payne (1991) 3 .The list in table 1 is also the first one to include data from Yawalapiti, the most divergent of the Xinguan Arawak languages.
In the languages showing forms relatable to PA *kʰapɨ (i.e.all languages other than Mehinaku, Waurá, Kustenaú and Yawalapiti), deviations from the postulated proto-form are relatively well-understood, at least in the sense that regular developments explain the important deviations4 .Thus, debuccalization of *kʰ and *p to h are regular developments in Wayuunaiki and in Paresi, respectively.Likewise, *p > h regularly applies in Achagua preceding a high vowel (see Payne, 1991, p. 434, 442;Ramirez, 2001a, p. 449 for the evidence).For Baniva, there is independent evidence for the regular loss of PA *k/*kʰ seen in -apî 'hand'.See, for instance, PA *kʰiba > íipa 'stone'; *ka-'attributive' > a-(see Mosonyi, 2000, p. 503, 512 for the Baniva forms, Payne, 1991, p. 377, 419 for the PA reconstructions).Other Arawak languages show forms not plausibly relatable to *kʰapɨ and for this reason have not been included in table 1 (e.g.Yucuna -játeʔela (Schauer et al., 2005, p. 217); Iñapari -mujú (Parker, 1995, p. 52); Proto-Amuesha-Chamicuro *-tɨ (Parker, 1991, p. 182).Since this paper is focused on the Xingu branch of the Arawak family, I will have nothing else to add on these languages.I will come back, however, to forms for 'hand', 'arm' or 'shoulder' related to Campa -ßako/-ako (Heitzman, 1973), as these feature in one proposal to incorporate the Xingu Arawak items in a family-wide cognate set.
Assembling a number of cognate sets such as the one presented in Table 1 constitutes the natural starting point for a vocabulary-based method of lexical and phonological reconstruction like the comparative method.A crucial property of such cognates sets is their constitution around a common meaning or semantic label (see e.g.Wilkins, 1996), producing what Koch and Hercus (2013, p. 34) call 's-cognates', stressing their 'semantically-matched' nature.Though reliance on such semantically-matched cognate sets is a natural feature of the beginnings of comparative work on a given linguistic family, the natural progression in our understanding of the historical development of a group of languages calls for the identification of so-called 'obscure cognates' (semantically deviant cognates), for which cognacy can be established or at least plausibly claimed on the basis of formal (phonological, morphological) factors while prompting us at the same time to judge the effects of lexical semantic change.As Koch and Hercus (2013, p. 34) note: Such reliance on strict synonymous cognates is useful as the first step in language comparison (...) But for the further goals of comparative linguistics -describing the totality of historical changes affecting the languages, working out the genealogical relationship between the languages, and presenting the complete etymological evidence that justifies these historical relations -it is necessary to go beyond the s-cognates.
That lexical semantic change offers potential obstacles to cognate identification has been known for a long time, especially by those who insisted on the advantages of fixed meaning lists (basic vocabulary lists) for carrying out comparative work (cf.Wilkins, 1996, p. 264-265;Koch, 2003, p. 273).A situation such as the one described in table 2 -where one or more daughter languages lack a clear (that is, semantically matched) reflex of a reconstructed etymon -suggests the possibility of finding an obscure cognate elsewhere in these languages (see also Szemerényi, 1977, p. 315;Koch, 2003).In this particular case, a semantically-matched reflex of PA *kʰapɨ seems to be absent from the Xingu Arawak languages.Moreover, the fact that these languages, otherwise taken to be part of one and the same subgroup, have (apparently) replaced this patrimonial etymon with the 'same form' (i.e. one plausibly derived from a Proto-Xingu form) attaches further interest and relevance to this issue.I will propose here that there are indeed interestingly obscure reflexes of PA *kʰapɨ in the Xinguan languages and that the innovation of *wɨʂɨku for 'hand' may be advanced as a plausible shared lexical development of this branch.Before proceeding, however, I will comment on a flawed attempt to derive the form for 'hand' attested in the Xinguan languages from another PA etymon.
A MISGUIDED ACCOUNT OF WAURÁ wɨʂɨku Payne (1991, p. 393) includes the Waurá form wɨʂɨku 'hand' in a cognate set for which the gloss ' Arm (Hand, Shoulder)' is provided.In spite of the shortcomings of this proposal, which, as I show below call for the removal of the Waurá form from this cognate set, it is still an important finding that items with a widespread distribution throughout the family, appearing in particular languages with different meanings, likely go back to a single PA etymon.Some of the relevant forms are presented in Table 2.I list only four forms, as these belong to widely separated Arawak languages, far apart from each other in extant classifications of the family as well as in geographic terms, and in this sense suffices to strongly suggest that a form such as *-wako can be reconstructed for PA 6 .The shift *a > o and the weakening of *k to a glottal or zero are general, context-dependent developments in Terena (and to a certain extent also in Baure and the Mojeño dialects, its closest relatives), though the precise specification of the conditioning factors remains elusive.As for the two back vowels appearing in reflexes of the last syllable, Payne (1991) reconstructs for PA an opposition between *o and *u noting, however, that this opposition was lost in most daughter languages (see Payne, 1991, p. 440-441, 472, 476-478 on these points).Payne (1991) includes Waurá wɨʂɨku in this cognate set, and reconstructs a PA form *-wahku, where the aspirate's *h only function is that of accounting for the -ʂɨ-syllable of the Waurá reflex (as well as for a syllable final h in Chamicuro, though Parker (1991, p. 178) does not reconstruct this glottal aspirate for Proto-Amuesha-Chamicuro).This move, however, creates insurmountable problems.First and foremost, the mapping *h > ʂɨ is phonetically implausible and can be dismissed unless compelling independent evidence for this hardening process in the history of Waurá could be provided -but this is not the case.Second, the correspondence involving the vowels in the first syllable is equally suspect.Payne (1991, p. 473) lists this form as providing evidence for PA *a > ɨ in Waurá, though the environment for this change is presented by the author as *w__hk, which, in spite of the phonological formulation, is so parochial  Shaver (1996, p. 281) Table 2. Proposed cognate set for wɨʂɨku (after Payne, 1991).
6 I have chosen Nomatsiguenga as a representative of the Campa group in this case for it retains a more conservative form of this item, contrasting with other Campa languages/dialects where the root-initial consonant was lost (see Heitzman, 1973, p. 73).Note that there is in this language a seemingly morphologized alternation b ~ p in nouns like 'hand', which therefore appears as -pako in other contexts (see Shaver, 1996, p. 27).As to the degree of separation between these languages, Nomatsiguenga and Apurinã could be seen as relatively closer, to the extent that one believes in a version of the 'Pre-Andine' hypothesis that would join Campa and Piro-Apurinã-Iñapari in a single higher-level branch.Though I find this proposal worth looking at, there is no published evidence or argumentation on its behalf.See Payne (1991, p. 366) and Michael (2008, p. 212, fn.1), for comments.
(and lacking in phonetic motivation) that we end up having an item-specific, sporadic development.The other form presented as supporting the postulation of a development *a > ɨ in Waurá is the form for 'brother in law (of male)' for which Payne (1991, p. 397) (Payne, 1991, p. 390) in order to be clear on what I take to be the shortcomings of his proposal.First of all, an alternative account to the postulation of PA *a > ɨ in Waurá would ascribe the diachronic correspondence *a > ɨ in *ani > pɨhɨne to the effect of contextual harmony (*pɨhane > pɨhɨne), a process seen in many 'free variants' attested in both Mehinaku and Waurá such as [ɨnɨʂa] ~ [enɨʂa] 'man ' and [tɨnɨʂu] ~ [tenɨʂu] 'woman'.If this is the case, it is hard to see how this form and *wahku > wɨʂɨku could be described as supporting or instantiating one and the same development, unless additional, independent hypotheses are invoked, something that would make the burden of proof even heavier for Payne's proposed etymology.Most important, though, is the fact that it is far from clear that Waurá pɨhɨne fits in (1) as a bona fide cognate to forms such as Machiguenga ani 'brother in law' and Ashéninka aniri 'brother in law', all presumably reflexes of PA *ani.Payne's (1991, p. 473) factoring out of the sequence [pɨh] in the Waurá form (and in other cases as well) reflects a presumed independent morphemic status for the sequence in question, "even if no information on the primary sources supports this assumption" (see Payne, 1991, p. 390-391).As it happens, this is completely unmotivated in the Waurá case and no justification for this is offered at any point in Payne's text7 .Wrapping up the discussion, there is no independent support for the *a > ɨ shift required to derive wɨʂɨku from putative PA *wahku.
I conclude, therefore, that Payne's (1991) postulation of a development *a > ɨ in Waurá and, more generally, his claim that Waurá wɨʂɨku goes back to PA *wahku, are untenable, as the latter claim, in particular, demands the acceptance of phonetically implausible and ad hoc developments.Two implications of broader interest to the historical linguistics of the Arawak family follow from this.The first is that the PA etymon Payne (1991, p. 393) gives as ' Arm2 (Hand, Shoulder)' is better reconstructed as *waku, thus eliminating one instance of the controversial syllable-final *h he attributes to PA (see Payne, 1991, p. 455).An additional consequence is that, once we recognize that wɨʂɨku is not a reflex of a patrimonial, PA etymon, it may be considered an innovation characteristic of the Xingu subgroup.This is the topic of the next section.

ON PROTO-XINGUAN *wɨʂɨku
Of the three extant members of the Xingu branch of the Arawak family, only Waurá was sampled in Payne's (1991) comparative study.Though data on Mehinaku has appeared elsewhere (see Ramirez, 2001a, p. 619) no data on Yawalapiti has so far been considered, therefore precluding any systematic comparative study.Nevertheless, given the forms Waurá/Mehinaku wɨʂɨku and Yawalapiti wiriku, it is possible to reconstruct a Proto-Xingu (PX) form that is virtually identical to the form used by Payne (1991).As noted in Carvalho (in press), Yawalapiti innovated by merging *ɨ and *i as i, and the association of rhotic consonants with some degree of frication is a synchronic implementation rule in Yawalapiti, aside from being a recurrent feature of Arawak languages (see e.g.Santos, 2006 on Wapixana; Meléndez Lozano, 2011, p. 3 on Achagua andCaptain andCaptain, 2005, p. 8 for Wayuunaiki).Comparative data also support the equation of the Yawalapiti rhotic with a fricative in the other languages, as in M (ehinaku) keʂɨ : Y (awalapiti) kɨri 'moon'; M matukɨʂɨ: Y matukiri 'father-in-law' 8 .The evidence for this merger in Yawalapiti and, therefore, for the derived character of the high front vowels in wiɾiku is made clear by the existence of two regular segmental correspondences (Tables 3 and 4) 9 .
Based on the correspondences in tables 3 and 4, reconstructing a merger in Yawalapiti is the reasonable implication, as no pattern of complementary distribution in Mehinaku/Waurá calls for the postulation of a split in these languages.Therefore, a form such as *wɨʂɨku, with high central vowels as in Mehinaku/Waurá, is the probable reconstruction for the PX ancestral language.
The Kustenaú form <nuiriko> (cf.von den Steinen, 1894, p. 529) was plausibly analyzed in Corbera Mori (2015) as <n-uiriko>, <n-> standing for the 1sg prefix indicating possession.One might quibble that the consonantal    3 and 4 is to be found in Carvalho (in press).For instance, evidence shows that the word-final ɨ in the Mehinaku/Waurá form for 'blood' is the result of a recent innovation.Likewise, the forms for 'cloud' can be accounted for under the general change *e > ɨ that took place in Yawalapiti after the merger*ɨ, *i > i.
variant of person-number prefixes in Arawak languages, which normally have a Cv-form, is usually selected in the context of vowel-initial roots only, so that a better approximation of the Kustenaú form would be something like <nu-uiriko>, the sequence of a back round vowel and a homorganic glide being suppressed in von den Steinen's transcription.Moving on to a more substantive issue, the vowels in <nuiriko> could suggest a change *ɨ > i in this language as well, similar to the merger that operated in Yawalapiti.Though Corbera Mori (2008, p. 33-34, 2015) assigns the value [ɨ] to the symbol <ö> and [i] to <i> of the von den Steinen transcription system, on the basis of a comparison of the Kustenaú forms to modern Mehinaku/Waurá cognates, it is clear that the same kind of evidence shows that von den Steinen's transcription was not so Pairs like K (ustenaú) <n-utitai> : M (ehinaku) The discussion so far has established that a form like *wɨʂɨku can be reconstructed for the PX ancestral language and, given the comparative evidence presented in preceding sections, one may suggest that it has replaced the PA form *kʰapɨ as the lexemic realization of the meaning 'hand'.This would constitute evidence for a lexical innovation characteristic of this branch, though a more systematic investigation is necessary to eliminate the possibility that cognates of PX *wɨʂɨku are not found in other languages of the family, with a shifted or changed meaning/function.This would require careful bottom-up reconstruction within other branches of the Arawak family, something that is, for most branches, at best a near future prospect.Equally fruitful for this undertaking is an examination of the limited data available on already extinct languages, though the inherently limited and restricted nature of these materials places a rather strong constraint on evaluations of this hypothesis.In the meantime, I note Apolista <-pisi> (actually attested as <nipisi> 'my hand'; Payne, 2005, p. 243) as a possible cognate of the PX form 10 .Note that further research may justify analyzing the PX form as coming ultimately from *wɨʂɨ-ku, as many body part terms, both in the reconstructed PA language and in daughter languages, occur with a suffix having a form -ku/-ko, including, perhaps, the form *waku discussed in this section (see Payne, 1991, p. 393, 398, 401, 404 for examples such as 'arm2 (hand, shoulder)', 'chest (abdomen)', 'ear', 'flesh, meat').Until more is known on the morphology of PA and other intermediate proto-languages within the family, I submit the plausible hypothesis that *wɨʂɨku constitutes a shared lexical innovation of the Xingu Arawak languages. 10Payne (2005, p. 243) proposes that Apolista <-pisi> is a reflex of PA *kʰapɨ.As in the case of his proposed derivation of Waurá wɨʂɨku from a putative PA *wahku, I remain unconvinced and for similar reasons.His hypothesis concerning Apolista depends on assigning complex morphological structure to the form in question, which becomes then <pi-si>, where only the first morph would be a reflex of the PA etymon.This is an analysis brought up only by parochial convenience, as no independent evidence exists for this.The putative morpheme -si cannot be analyzed, for instance, as the Apolista reflex of the Absolute suffix *-ʧi which, as shown by Payne (2005, p. 244), is -shi.Regardless of this, the presence of this suffix would be unexpected here, as the form for 'hand' is attested in a possessive construction as <nipisi> 'my hand'.

REFLEXES OF PA *kʰapɨ IN XINGUAN ARAWAK
As already pointed out, general theoretical expectations on lexical change force us to consider the possibility of finding reflexes of PA *kʰapɨ elsewhere in the Xingu languages, most likely in derivative forms (see e.g.Szemerényi, 1977, p. 315).As a matter of fact, Payne (1991, p. 407) did include a form -kapɨ-for Waurá, indicating with dashes that it occurs in morphologically complex or derived forms, but without offering any additional comments.The qualifications necessary to identify this bound form in Waurá and other Xingu languages -as well as some unforeseen implications of these findings -are the subject of the present section.

'HAND', 'FINGER'
The particular reflex of the PA etymon is found preserved in the form for 'finger', which is, in both Mehinaku and Waurá, a derived noun that alternates between kapɨtɨwɨ and kapitiwi (see Corbera Mori, 2011, p. 209).This is a morphologically conditioned alternation, kapitiwi being the Absolute (non-possessed) form and -kapɨtɨwɨ occurring where an overt possessor is present.As shown by Carvalho (2015), the etymological form, the one reconstructible to Pre-Waurá/Mehinaku is *kapɨtɨwɨ, the fronting of ɨ to i being a result of the diachronic absorption of the Absolute marker -*ʧi reconstructed for PA (plus regressive vowel harmony, see Carvalho, 2015, p. 128-130 for details).Finally, the form can be morphologically parsed as a compound of *-kapɨ and *-tɨwɨ 'head'.
The preservation of *kʰapɨ as a (relatively) obscure cognate is also attested in Yawalapiti, though here we have evidence for the application of an interesting sound change.The Yawalapiti form for 'finger' is kariʦɨpu (<nukaritsobu> in von den Steinen, 1894, p. 533).The form can be plausibly analyzed as kari-ʦɨpu, with kari standing for the Yawalapiti reflex of *kʰapɨ and -ʦɨpu being a compounded nominal root or a classifier, following the general pattern attested in Arawak languages for the derivation of terms for 'finger' (see the following section for additional discussion).I have been unable so far to ascertain the meaning/function of -ʦɨpu, the analyses and data in Mujica (1992) and Bondim (1976) 11being of little help in this regard 12 .
The equation of PA *kʰapɨ and Yawalapiti -kari-might seem far-fetched due to its implication of a diachronic correspondence *p > r, one involving a major articulatory discontinuity, a change in labial to coronal place of articulation.Though von den Steinen (1894, p. 532) made no particular comments on the form for 'finger', he noticed that in many cases a labial stop p in Mehinaku, Waurá and Kustenaú corresponded to a coronal consonant, either t or r, in Yawalapiti: "Es sei auf die interessante Fälle von Lautwandel in verhältnis zu Mehinakú etc. aufmerksam gemacht.Meh.p erscheint mehrfach als t und r" ['One may note the interesting cases of sound change with regard to Mehinaku revealed by a comparison of the Yawalapiti data.Mehinaku p appears frequently as t and r'; my translation].A comparison of the data used by von den Steinen (1894) to back up this claim is given in the Table 5 (the corresponding segments appear in underlined).
The most notable aspect of the correspondence noted by von den Steinen ( 1894) is that it occurs in the context of a following front vowel, which may be present only in the Mehinaku and Kustenaú cognates (cf.'Blue').As I argue in Carvalho (in press), von den Steinen was correct in his suggestion of a sound change (Lautwandel) whose effect was to change p into a coronal stop in Yawalapiti, a change that I describe as the telescoped result of a palatalization of Pre-Yawalapiti *p to *pʲ, the palatalized allophone eventually merging with the reflexes of the rhotic *r.Though somewhat unusual, this change -or, better said, a diachronic correspondence mapping a labial stop to a coronal segment in a palatalizing context -is cross-linguistically attested in diverse language families, notable cases including Southern Bantu and the isogloss separating petak/tetak Czech dialects (cf.e.g.Ohala, 1978;Thomason, 1986).More tellingly, we find synchronic evidence for its action preserved in modern Yawalapiti morphophonemic alternations, copiously attested in Bondim (1976), such as puti ~ ruti 'thigh', piɲa ~ riɲa 'house', paka ~ raka 'to sing' and pɨhi ~ rɨhi 'penis', in all cases triggered by the prefixation of a person/number prefix with a front high (or palatal) vowel i.The preservation of these morphophonemic alternations in the modern language allows one to internally reconstruct the course of these changes and then validate and improve the proposed developments with the aid of comparative evidence (see Carvalho, in press for details).
Overall then, the existence of a change of diachronic coronalization of *p in Yawalapiti, revealed both by comparative data, such as that amassed by von den Steinen (1894), and by the existence of a synchronic alternation p ~ r, supports the claim that a reflex of PA *kʰapɨ is to be found in the root kari-, part of the form kariʦɨpu 'finger'.Note that the change in question was fed by the already discussed merger of *ɨ and *i as i.Only after this change was a form *kapi (< *kʰapɨ) produced and this was in turn changed to kari by the 'palatalization' of *p to r in the context of a following palatal vowel.Both the data and the explicit recognition of a segmental correspondence involving forms from the Xinguan Arawak languages constitute an important contribution of von den Steinen (1894), showing that the German explorer went beyond the 'mere' collection of vocabulary data and actually provided some insight on important issues.

'FINGER' = 'HAND' + 'HEAD': A WAURÁ-MEHINAKU INNOVATION?
I will briefly consider here the hypothesis that the particular formation of a form for 'finger' attested in Waurá, Mehinaku and possibly in Kustenaú as well constitutes a shared innovation.
The formation of the word for 'finger' derived from 'hand' is likely grounded on a lexical semantic metonymic relation between the two concepts, and is cross-linguistically recurrent (cf.Wilkins, 1996, p. 278).Though this is also the case within Arawak, the use of a nominal compound with the roots for 'hand' and 'head', as seen in Waurá and Mehinaku is, as far as I know, unattested elsewhere in the family.The form for 'finger' in Kustenaú, given in von den Steinen (1894, p. 529) is suggestive of a similar formation, and is included for this reason in the comparisons below.Most Arawak languages seem to employ the root for 'hand' plus a classifier, usually a reflex of the PA *-pi/*-api denoting elongated, thin or thread-like objects objects, or some other compound (Table 6).
The Wapixana form contains a classifier that Santos (2006) describes as denoting 'hard objects' and also 'flower'.The form for 'finger' in Baure is anomalous both synchronically (Danielsen, 2007, p. 153 calls it an 'inverse compound') as well as comparatively, for inverting the head-modifier relation typically seen in Arawak languages.The form wohis(o), with morphophonemic final vowel loss, is described as meaning both 'hand' and 'finger' (Danielsen, 2007, p. 154, fn. 129), though a (supposedly) unambiguous form for 'finger' may be derived as in Table 6, using a classifier denoting the 'pointed end' of objects.
For other languages, such as Palikur, we find a similar mechanism for deriving 'finger' from a root for 'hand' plus the modifying action of some form-based classifier for 'slender' or 'pointed' objects, though this is realized not by suffixation or composition with the root for 'hand', but in the classifier agreement system.Thus, a root -wak(u) is polysemic, meaning either 'hand' or 'finger'.Where it expresses the former meaning it triggers the appearance of a classifier -k/-ka/-bu (for flat, two-dimensional objects), for instance, in an agreeing numeral expression.If the same modifying numeral is marked instead with a suffixed classifier for 'three-dimensional, slender objects', whose marking is -t/-ta, then -wak(u) is understood as denoting 'finger' instead (Launey, 2003, p. 118-119) 13 .Likewise, in Baniwa (Ramirez, 2001b, p. 298), the suffixation of a form-based classifier for 'pointed objects', -hiwi, to a numeral that modifies a form such as pa-káapi 'somebody's hand', derives the notion of 'finger'.
The data in Table 6, which includes languages from all major branches of the Arawak family, suggests that there is something exceptional in the way Waurá and Mehinaku derive the form for 'finger'.Since it is currently accepted that a classifier of the form *-pi/*-api whose reflexes are attested in the Paresi, Iñapari and plausibly in the Lokono and Baniwa forms in Table 6 as well -can be reconstructed for the PA language (cf.Payne, 1991, p. 384), I propose that the use of the root for 'head' instead of this classifier in Waurá and Mehinaku for the purpose of deriving the form for 'finger' constitutes a plausible shared innovation.As Waurá and Mehinaku are independently known to be very closely related, being perhaps co-dialects of the same language, this proposal is not impressive as far as internal classification is concerned, though it does advance our understanding of the diachrony of these languages.

'HAND', 'FIGHT'
If the Yawalapiti form for 'finger', and the bound root kari, illustrate a rather uncommon sound change that makes the reflex of PA *kʰapɨ slightly harder to identify, an arguably more drastic semantic shift involving this inherited form is attested in all Xingu languages.The noun denoting the huka-huka, a wrestling-like fight of ritual importance in the Upper Xingu area (see Basso, 1973, p. 143;Agostinho, 1974;Carneiro, 1993, p. 422) is kapi in Waurá/Mehinaku (Corbera Mori, 2012, p. 149) and kari in Yawalapiti (Mujica, 1992, p. 15).I propose that this is a semantically-shifted reflex of PA *kʰapɨ, and I account for the formal and semantic deviations in what follows.
Starting with the Waurá/Mehinaku form, kapi, it is necessary to account for the change *ɨ > i, as in these languages no merger of the high central and high front vowels of the sort attested in Yawalapiti occurred.The explanation for this change lies in the account proposed in Carvalho (2015) for a separate set of patterns and comparative data.As the noun changed its category from a dependent, inalienable noun denoting a body part ('hand') to an independent, absolute noun ('huka-huka fight'), the occurrence of the noun in with an Absolute marker is a plausible consequence.One would have, then, at a Pre-Waurá/Mehinaku stage either a form *kapɨ-ʧi or *kapi (< *kapɨ-i), the latter if the semantic shift occurred after the changes that consisted in the loss of the affricate in the PA suffix *-ʧi (or some other, weakened reflex of this affricate) and the 'absorption' of the suffixal high front vowel, as discussed in Carvalho (2015).The change of -*kapɨ to kapi in Waurá/Mehinaku is, therefore, plausibly explained by its functional/semantic shift, its morphosyntactic consequences and the morphophonological developments involving the PA Absolute suffix *-ʧi (see Carvalho, 2015 for details).
For Yawalapiti, Mujica (1992) does not discuss the existence of Absolute morphology.In the data presented in Bondim (1976) 14 one does find what seem like remnants of an Absolute marking system that has been eroded, even more so than it was the case in Waurá and Mehinaku.Though a comprehensive account of the diachrony of the Yawalapiti Absolute-marking system will await for better (and badly needed) data on this language, one can note at this point that in all the paradigms involving dependent nouns attested in Bondim (1976), a general pattern consists in having the 'free form' showing word-final main stress, while possessed forms have penultimate main stress (a).A few other forms show either morphophonemic alternations (b) or some extra segmental material in the absolute forms (c).Examples include the following (for conciseness I present only 1sg possessive forms): (2) Remnants of an Absolute-marking system in Yawalapiti: (a) The pattern in (a) is identical to the one found in Waurá/Mehinaku nouns.As the languages in question have a regular pattern of main stress assignment to the penultimate syllable in the word, this can be accounted for by invoking the past presence of an additional syllable, the Absolute suffix, as discussed in Carvalho (2015)  15 .
Just like Absolute nouns (that is, dependent nouns employed without overt possessors), the Yawalapiti noun kaˈri 'huka-huka fight' has word-final stress, something that is consistent with hypothesis that it derives from a pre-Yawalapiti form with an added Absolute suffix, as proposed above for pre-Waurá/Mehinaku: preceding the loss of the Absolute marker, a form resembling *kaˈri-ʧi, with default penultimate stress, occurred.The stress-bearing syllable remained the same after the loss of the Absolute marker, yielding the typical, yet (synchronically) marked pattern of word-final stress that characterizes Absolute nouns.
On the semantic association between 'hand' and 'fight', there is plenty of evidence for its plausibility.A fairly traditional move consists in first placing the proposed change into one of the classical categories of diachronic semantics (cf.e.g.Geeraerts, 2010, p. 25-35).In our case, the association can be mediated through a metonymic relation: from 'hand-fight' or 'fight with/using the hand', the lexical item for 'hand' may end up being used to denote 'fight' (a semasiological change, sensu Geeraerts, 2010, p. 26), especially so, one supposes, once the inherited form for 'hand' has been replaced in its original function by a different word, as I claim was the case with PA *kʰapɨ and PX *wɨʂɨku.
Interesting parallels to the proposed change are attested in other language groups.In the well-trodden path of Indo-European historical linguistics, for instance, one finds in Latin the case of forms such as pūgnus 'fist' and pūgna 'fight' (also pugil 'fist-fighter'), with cognates elsewhere such as Greek púgme: (πυγμη') 'fist, fist-fight' (see de vaan, 2008, p. 499;Beekes andvan Beek, 2010, p. 1254).According to Beekes and van Beek (2010, p. 1254), pug-'fist' could be a noun originally, with Latin pūgna 'fight'/pugnāre 'to fight' being secondary derivations from the n-formation pūgnus.If this is the case, it provides a striking parallel to the proposed 'hand' > 'fight' shift proposed here for Xinguan Arawak 16 .
Similar changes involving what Wilkins (1996, p. 278) calls 'interfield metonymic changes', such as 'to slap' > 'palm of hand', can be shown, on the basis of comparative databases of semantic lexical change from diverse genetic groups, to be natural developments.I assume that lexical semantic change involves an intermediate stage of polysemy (cf.Wilkins, 1996, p. 268-271), so that a Pre-Proto-Xingu stage in which *kapɨ-ʧi (where *-ʧi stands for the Absolute marker) would mean both 'hand (unpossessed)' and 'huka-huka fight', the latter perhaps in particular syntactic constructions, is a plausible inference.With this much in mind, Wilkins (1996, p. 272) notes a similar pattern of polysemy involving a body-part noun and an action involving this noun, as in 'skin (noun)' and 'to flay', found in Dravidian, Indo-European, Bantu and Tibeto-Burman. 15Early observations by von den Steinen (1894, p. 527, 529, 531, 532) underscore the fact that all Xinguan Arawak languages had an identical distribution of main stress.In his Mehinaku, Waurá, Yawalapiti and Kustenaú vocabularies he remarks that "Wo der Accent nicht angegeben ist, wird die vorletzte Silbe betont" ["Where accent is not indicated, the penultimate syllable bears stress"; my translation].This is in agreement with the general finding of a default stress placement in the penultimate syllable of the word in Xinguan Arawak. 16A reviewer has brought to my attention the interesting case of Japanese karate, from kara 'empty' and te 'hand'.One might also add the Brazilian Portuguese expression sair na mão meaning 'to fight, brawl'.
I conclude that it is possible to plausibly account for the semantic and formal deviations implied by the hypothesis that Waurá/Mehinaku kapi and Yawalapiti kari, both nouns denoting the 'huka-huka fight', constitute semanticallyshifted reflexes of the PA etymon *kʰapɨ 'hand'.

SYNTHESIS AND CONCLUSIONS
This paper has advanced contributions to our understanding of the historical development of one particular group of languages -those usually grouped in the Xinguan branch of the large Arawak family -by on phonological, morphological and lexico-semantic changes in items denoting 'hand', 'finger' and related concepts.The argumentation supports the hypothesis that Proto-Arawak *kʰapɨ 'hand' has been replaced in the ancestral Proto-Xingu language by *wɨʂɨku in the expression of this particular lexical concept, but is preserved in these languages as a component of morphologically-complex forms denoting 'finger' and in a metonymically-related noun for a wrestling-like fight of central ritual importance for the peoples of the Upper Xingu cultural area.
I have attempted to build on past and ongoing research on these languages, stressing how the hypotheses proposed here build progressively on independent work on Arawak historical linguistics.More importantly, I have stressed the more or less tentative character of many of these claims, suggesting further directions for research that may eventually lead to the strengthening of their evidential basis or to their rejection.

Figure 1 .
Figure 1.Tree diagram and character-state distribution for the Paresi-Xingu languages.

Table 1 .
Forms for 'hand' in diverse Arawak languages.