A Study of the Genetic Relation between Mawayana and Wapishana (Arawakan Family)

The Arawakan language family is one of the largest linguistic groupings in the American continent, with over 50 languages distributed from Central America (Garifuna) to the Brazil-Paraguay border (Terena) and maybe Argentina (Chané), and from Eastern Peru (Campan languages) to the Xingu river in Central Brazil (Wauja-Mehinaku, Yawalapiti), and to the Brazil-French Guiana border (Palikur). Most Arawakan languages are severely endangered: the vast majority of their over 500,000 speakers belong to one single language, Guajiro (Wayuu), spoken in the peninsula of La Guajiria, shared between Colombia and Venezuela.


Introduction
The Arawakan language family is one of the largest linguistic groupings in the American continent, with over 50 languages distributed from Central America (Garifuna) to the Brazil-Paraguay border (Terena) and maybe Argentina (Chané), and from Eastern Peru (Campan languages) to the Xingu river in Central Brazil (Wauja-Mehinaku, Yawalapiti), and to the Brazil-French Guiana border (Palikur). Most Arawakan languages are severely endangered: the vast majority of their over 500,000 speakers belong to one single language, Guajiro (Wayuu), spoken in the peninsula of La Guajiria, shared between Colombia and Venezuela. Within Arawakan, two languages are placed together in the most recent classifications, forming (together with other presumably extinct languages) a small sub-branch: Mawayana (aka.

The Data
For this study, the following sources of data were considered (note that the author's field data were given precedence: other sources were used to complement gaps or to confirm translations, but in general whenever a form was available in the author's data it was chosen for comparison, the assumption being that the other sources were more likely to contain errors): For Wapishana: -The author's field data on Wapishana (a list with a little over 500 lexical items, plus about 200 elicited sentences and two short texts), mostly obtained during the 2010 InField summer school workshops in Eugene, Oregon, from June 21 to July 30, with the help of a native speaker of Wapishana, Mr. Adrian Gomes, and three students: Ana Paula Barros Brandão (UT-Austin), Jessie Erikson, and Ian Lunger (UO-Eugene). More data were obtained by the author during two field visits to the village of Marurunau in Guyana, mostly with Mr. Gomes but also with other local people, in the context of collecting data on the Taruma language (new Wapishana lexical items or sentences were obtained as translations for Taruma lexical items). -Two published dictionaries of Wapishana: WLP 2000, with about 1,400 entries, and Silva, Silva & Oliveira (2013), with about 1,100 entries. -Melville, Tracy & Williams (2007) Santos (2006), a grammatical description of Wapishana with a 350-item word list; -Occasionally, other works on Wapishana grammar (TRACY, 1972;1974) were used, as well as local publications of texts and literacy materials, of which the most important is WLP 2001. Occasional use was also made of a New Testament translation (SB 2012).
For Mawayana: -The author's field data, ca. 600 words, 100 sentences and one short text, collected from a native speaker (Tenetene, a 60+-year-old female) in a field trip with Dr. Spike Gildea in 1997, to the Waiwai village of Mapuera, in the Nhamundá-Mapuera Indigenous Area in the State of Pará, in Brazil. -Howard 1985-Howard -1986, a questionnaire (the National Museum Questionnaire, developed by the Federal University of Rio de Janeiro) with about 400 lexical items mostly in phrases and sentences, collected by Catherine Howard in 1985-1986 with a 65-year-old female speaker (Čoowi) from the village of Kaxmi, at the Rio Novo, in the State of Roraima, in Brazil. -Carlin (no date), a short Mawayana text with Tiriyó, English, and Dutch counterparts, glossed and analyzed by the author; occasionally, also Carlin 2006 (sentence examples in a paper on grammatical borrowings from Cariban into Mawayana).

Transcription and Segmental Phonology
The transcription of language examples and lexical items in this article uses the IPA only; any material from sources having a different writing system are retranscribed with IPA symbols (note the use of dots to mark syllable boundaries). Table 2, a summary of Pidjanan segmental phonology, shows all the necessary IPA symbols.  [ˀdʲ] or even nonglottalized [b], [d], [dʲ]). Wapishana has also a normal plosive /d/, much rarer than /ɗ/ but still clearly distinctive (cf. the minimal pair /paraɗan/ 'to cut underbrush' vs. /paradan/ 'to speak', and analogous pairs like /pɨɗapɨ/ 'your house' vs. /pɨdaɾɨ/ 'your father'). Another interesting feature of Wapishana is the retroflex fricative / ʐ /, often realized with a simultaneous rhotic feature ([ʵʐ], [ʐʵ]).
The vowel system, on the other hand, is richer in Mawayana than in Wapishana: the latter has only 4 vowels (/a/, /i/, /u/, /ɨ/), while the former has, in addition, the mid vowel /e/. In both languages, /ɨ/ varies from [ɨ] to [ɯ] and /u/ from [u] to [o] in certain positions; in addition, Wapishana /a/ is raised, often all the way to [ɛ], when the preceding syllable ends in a palatal (/i/ or /j/) that also affects the intervening consonant: /kaiman/ 'good' =

Swadesh List and Shared Cognates
As an initial assessment of the degree of similarity and relatedness between Pidjanan languages, we consider the 100-word Swadesh list below (the version published in Swadesh 1971, p. 283).
In this list, cognate words are marked in bold, while non-cognate words are in normal type. Partially cognate words (i.e., words which contain non-cognate elements, as in, e.g., Portuguese nós 'we' and Spanish nosotros 'we', where the otros part is not cognate) are counted as fully cognate words, even though only their cognate part is written in bold. Notice that both languages are also (as is frequent in the Arawakan family) morphologically complex, which leads to many example words containing affixes. These are segmented out with hyphens and left in normal type. The percentage of fully or partially cognate terms (the cognate parts are in bold) is 44/100 or 44%. If duplicates (6 and 7, 27 and 28, 75 and 76) are counted as only one instead of two, the result is 41/97 or 42,3%. This is a rather low result for languages considered to be closely related in all classifications (see Table 1 above). The neighboring Cariban languages (Makushi, Taurepang, Ingariko) average 70-75%; even if one includes less closely related languages to the southeast of the Rio Branco (Waiwai, Hixkaryana), shared cognacy still remains at 50% or more. This suggests that, in this area (from the upper Rio Branco in Brazil eastwards into the Rupununi area in Region 9 in Guyana), Arawakan groups have lived for a significant longer period of time than Cariban groups, thus agreeing with current theories on the Arawakan and Cariban expansions.

Correspondences and Proto-Segments
In this section, we discuss the vowel (Sec. 5.1) and consonant (Sec. 5.2) correspondes found in the data listed in the Appendix 1 .

Vowel correspondences.
In this section, we examine all Pidjanan correspondences involving vowels. if there is no ellipsis at the end, the list is complete). A possible reconstruction for each correspondence, marked with a preceding asterisk (*), is given in the first column (PP = Proto-Pidjanan). Subscript numbers (*v1, *v2, *v3, etc.) identify different correspondences potentially reconstructible as the same proto-segment (*v). The reconstruction is discussed in the paragraphs following each table. From these correspondences, the four vowels *a, *i, *u, and *ɨ can be easily reconstructed: they all have numerous examples (*a1 occurs in over 63% of all cognate sets, *u1 in 35%, *ɨ1 in 26%, and *i1 in 24%) and are found in all environments.

Main Vowel Correspondences.
The two other possible vowels, a front mid *e and a back mid *o, are less clear. The plausible candidate correspondences, *e1, *e2 and *o1, have better explanations: *e1 has only six examples, all involving Wapishana a and i, either in the same syllable (CHILD: M kuɾenu, W kuɾaiɗaunaː) or in adjacent syllables (ARMPIT: M keːsu, W kiʃapu, PARTNER: M meːɾawa, W minaɨ-ɗaʔɨ, THIN: M metaɗa, W miɗa-ʔɨ), as does *e2 (CHILD/EGG: M ɗe, W ɗani, FISH SP. 1: M ɾiːtʃe, W ʐiːta-ɓa, PERSON: M ɗʲe(ː), W piɗan, SMOKE: M ise-sɨ, W iʃa-n). Those examples are better reconstructed as cases of *a and *i, with these vowels either merging (*ai > e is a common change cross-linguistically) or raising to e (ato-e or i-to-e raising; notice that a preceding i, as was pointed out in Sec. 3, is the environment in which Wapishana a raises to ae  e) in Mawayana. The only difficult example is *e1 TOUCAN SP. (M takwe, W tʃa:kui), in which the same i and a environment is still present, but the intervening ku / kw suggests a more complex story (maybe originally more than one syllable, e.g.,*ku-we). There is thus very little basis for reconstructing PP *e. Further instances of *a and *i are reconstructed instead.
Finally, *o1 and *o2 are even less clear. Two out of three examples, CICADA (M womu, W wamu) and MONKEY SP. 3 (M oʧɨ, W puwatɨ), have adjacent w's that suggest assimilation: PP *wamu > M womu, *puwatɨ > *puwotɨ > *puotɨ > * potɨ > *otɨ > oʧɨ (for Mawayana p-loss and *t > ʧ, see Sec. 5.2.1, 5.2.2). Only *o3 LEFT-HAND (M asuɓa-, W aʃaɓa-) remains. Even if one suspects that the labial ɓ may have influenced the preceding a, it would still be an idiosyncratic change. Note also that Mawayana o is quite rare and apparently non-phonemic (no minimal pairs for o and u have been found). It is currently analyzed as an allophone of /u/, so that oʧɨ and womu are /uʧɨ/ and /wumu/. Therefore, *o1 is reconstructed as *a and *o2 as *u. 2 5.1.2. Correspondences Involving Vowel Loss. Table 5 lists vowel correspondences in which one of the two Pidjanan languages lacks a reflex. For *a2, *i2, *u2, and *ɨ2, most of the cases result from one change, here called Mawayana diphthong reduction. These are cognate sets in which a Wapishana diphthong (or VʔV or VpV sequence, as in 1h-i; see Sec. 5.2.1, 5.2.6 on *p-and *ʔ-loss) corresponds to a Mawayana simple vowel, the same as the last vowel of the Wapishana diphthong (1a-c, f-g) except for *e2 cases, in which M e corresponds to W ai (1d; note 1e is irregular, probably because it is a monosyllable). 3 In these cases, *ai is reconstructed. Elsewhere, *a2, *i2, *u2, and *ɨ2 are reconstructed as *a, *i, *u, and *ɨ, respectively. Some of the *ɨ2 cases listed below can be explained by reconstructing, based on the Wapishana cognate, PP *wɨ, and by positing a (word-initial) assimilatory change *wɨ > M u(ː).

Idiosyncratic Vowel Correspondences
The remaining non-identical vowel correspondences, listed in Table 6 below, mostly involve idiosyncratic changes, often more than one per cognate set. In some cases, the words in the set may not even be really cognate. For specific details on changes within each set, see the Appendix.

Correspondences Involving Suprasegments
Finally, there are correspondences involving suprasegmental features like vowel nasality, vowel length and stress. Vowel nasality is not reconstructible: all examples have nasal vowels corresponding to oral vowels adjacent to n, thus suggesting PP *Vn or *nV sequences in which the consonant nasalized neighboring vowels and disappeared. 6 This means that *V͂ 1 and *V͂ 2 are actually subsets of *n2/*n4 and *n3, reconstructed as *ɲ and *n. For further details, see Sec. 5.2.5 below.
As for vowel length, most of the cases of Vː1 in Table 7 appear to be the result of compensatory lengthening, in that the corresponding Wapishana cognates have one extra syllable. These are reconstructed with an extra syllable and without length (5a-d). 7 The cases of Vː2, on the other hand, show Wapishana long vowels in all kinds of environments in the two last syllables of a word (5e-g). Sometimes the final long syllable alternates with an extra syllable (5g), sometimes it does not (5e-f). In the cases where synchronic alternation is attested, the extra syllable is reconstructed; if not, the long syllable is reconstructed. Turning now to stress, as was seen in Sec. 3, the Wapishana and the Mawayana systems differ in directionality: the former counts syllables from the end of the word, while the latter counts syllables from the beginning of the word. Note, however, that the latter corresponds exactly to the rhythmic stress system found in Cariban languages, among which Waiwai and Tiriyó (see Meira, 1998 for a Tiriyó case study). The few remaining Mawayana speakers live currently in Waiwai and Tiriyó villages in Brazil and in Suriname, where all conditions for borrowing are present (see Carlin, 2006 for examples of grammatical loans). Under these circumstances, it seems more likely that Mawayana has simply borrowed from Cariban. The Wapishana system is thus reconstructed to Proto-Pidjanan: counting from the last syllable of a word, secondary stress is assigned to every other CV syllable. Non-CV syllables are always stressed, and syllable counting restarts after them. Primary stress goes to the penultimate mora (i.e., to the final syllable if it is non-CV, or then to the penultimate if the final syllable is CV; in other words, to the first of all secondarily stressed syllables, counting from the end of the word). Mawayana unstressed vowels are never dropped, while Wapishana unstressed vowels often are; it seems better to reconstruct Proto-Pidjanan unstressed vowels as similar to their Mawayana counterparts, i.e., less salient, but not really strongly reduced or lost. Unstressed vowel loss is thus an ongoing Wapishana innovation.

Consonant correspondences
In this section, we examine all Pidjanan correspondences involving consonants.

Correspondences Involving Non-Alveopalatal Stops
The simplest consonant correspondences in Pidjanan are those involving the voiceless stops k and p, as listed in Table 8 below. The simplest of all is *k1, an identity (k : k) correspondence, immediately reconstructible as *k (e.g., ANT SP: M kuki, W kuki). It shows that this consonant was very stable in this branch: no significant changes occurred in its pronunciation or distribution. As for *k2 and *k3, they are clearly exceptional cases, possibly involving non-cognates or the result of partial analogy (see the corresponding examples in the Appendix).
The main correspondence reconstructible as *p is *p1. Note that it is not an identity (p : p) correspondence, which, since Mawayana does not have a distinctive p, is to be expected. In fact, *p1 shows that Proto-Pidjanan *p was simply lost in all positions in Mawayana. The *p2 example FISH (M kuwɨ, W kupaɨ) suggests lenition: PP *p > *b > *ß > Ø. 10 The examples below illustrate these changes that gave rise to *p1 and *p2 (note 6e-h, where p-loss interacts with Mawayana diphthong reduction, 6g, with a-to-e raising, and 6h, with one of the few cases of Mawayana o; see Sec. 5.1.1). 10 One additional supporting piece of evidence for lenition is the occurrence, in a M1 example, of an apparently optional w. So BLACK (PP *puɗɨ) is attested in this source as both uɗɨ-ɾe and wuɗɨ-ɾe. However, since w is phonemic in Mawayana, the lenited version of PP *p was probably not *w, which would have created confusion with pre-existing *w (e.g., in minimal pairs like the prefixes *pa-'3 refl.' and *wa-'1 pl.', the sources of the corresponding Wapishana prefixes in Table 17), but something like *ß.
The only remaining correspondence, *p3, is obviously irregular: its only example, ELBOW (M ɸatʃuɾi, W patuɾi), is also the only example with a bilabial fricative [ɸ] in the available Mawayana data. Since the neighboring Waiwai language does have a distinctive /ɸ/, found also in the word for 'elbow' (Waiwai aɸoɾesɨ), it seems best to ascribe this one instance of Mawayana [ɸ] to Waiwai influence. Given the Wapishana cognate, *p is reconstructed also in this case.

Correspondences Involving Alveopalatal Stops, Affricates, and Fricatives
Let us now consider the more numerous and complicated correspondences involving the alveolar consonant t, all listed in Table 9 below (except for those that involve the voiced stops d and ɗ; see Sec. 5.2.3). Curiously, the best candidate for reconstruction as *t, the identity correspondence *t1, has only three examples, BAT (M tamaɾiwa, W tamaɾiu), HIT (M ɾita, W ʐuɁita), and OLD MAN (M tauɾinu, W tɨnaɾɨnau), in all 149 cognate sets. This is surprising, since alveopalatal stops are often among the most frequent consonants in phoneme inventories cross-linguistically. One is thus led to consider the correspondences with the most examples, *t2 (6 examples) and *t6 (13 examples), as more likely candidates for *t (with *t1 being perhaps a special case, occurring, e.g., word-or morpheme-initially when the following vowel is *a; see the Appendix). The actual examples suggest that *t6 (7a-d) and *t2 (7e-h) contrast (cf. 7a and 7h, 7b and 7g, 7c and 7f, 7d and 7e). But which one is to be reconstructed as *t? Judging by reflexes outside of Pidjanan, *t6 is the best candidate, in that it is usually t (Bahuana hatuɾɨ, atuɾɨ 'caiman', atamɨna 'tree'; RAMÍREZ, 1992, p. 21-23), while *t2 is not (Achagua ʧeːma, Baniwa jeːma 'tobacco'; RAMÍREZ, 2001, p.755). Therefore, *t6 is here reconstructed as *t (with *t1 being also *t word-initially before *a), and *t2, given its reflexes (M t and W s), as *ʦ. According to this scenario, PP *t > M ʧ and PP *ʦ > M t in most environments. Obviously, *t > ʧ must have preceded *ʦ > t, or else all PP *ʦ's would also have become present day ʧ's. Note that the above examples (and the others in the Appendix) do not show *t or *ʦ adjacent to *i. Two correspondences could fill this gap: *t7 (8a-c) and *t5 (8h-j). Note that their examples either have a permanent adjacent *i (8a FIRE), or are frequently close to one (8b-c, h-j; the Mawayana third-person prefix ɾi(ː)-indicates a possible i-initial stem, whence the initial *(i) in the protoform). But they, and also *t3 (8d-g), seem to contrast with each other; cf. 8a and d; 8b/8c, 8f, and 8i. They are here ascribed to the proto-segments whose reflexes are most similar to theirs (i.e.,*t7 = *t, as ʃ ː t is most similar to ʧ : t = *t, and *t5 = *ʦ, as s : t is most similar to t : s = *ʦ). For *t3, a new proto-segment is necessary; *ʧ is proposed here. In other words, we reconstruct here a new alveopalatal affricate (*ʧ, from *t3) and two palatalized allophones, of *t (from *t7) and *ʦ (from *t5), found in the vicinity of *i (7h ʦamaʦama being an exception, probably motivated by sound symbolism; it is apparently also reduplicated).
(8) Examples of *ʧ (*t3; 8d-g), and palatalized allophones of *t (*t7; 8a-c) and *ʦ (*t5; 8h-j). Given the frequent historical connections between *t and alveopalatal fricatives and affricates, let us now take into account the remaining correspondences involving the latter two sound classes, as listed in Table 10 below. The best candidate for immediate reconstruction as *s is, of course, the identity correspondence *s1 (9a-c; note the apparent reduplication in 9a). As in the case of *t1 above, however, *s1 has too few examples for a fricative that is usually much more frequent in phonemic inventories; in addition, it appears only word-initially, followed by *ɨ or *u. Looking at *s2, we see that it is in complementary distribution with *s1 (it occurs near *a or *i, or, word-medially, near *u); it can thus be also reconstructed as *s. The next correspondence, *s3, is a little more complicated. It is a significant correspondence (11 examples in the available comparative data). It occurs word-internally, next to *ɨ (10a-d); it is thus in complementary distribution with *s1/*s2 and can be reconstructed as *s to fill the last gap in its distribution (see, e.g., 9b/10b BANANA, with both word-initial and word-internal *s). However, it is also in complementary distribution with *ɾ2 in Sec. 5.2.4 below, reconstructed as *ɽ. The reflexes of *s3 are just as compatible with *s and they are with *ɽ. For now, let us keep *s3 as its most likely phonetic realization, *ʐ. But is this *ʐ an allophone of *s, or of *ɽ? Phonetics supports *ɽ. After all, if *s were to be chosen as the source, then one would have to admit a change *s > W ʐ in intervocalic position. The retroflection can be explained via assimilation to the central vowel *ɨ, but not the voicing. Even though, cross-linguistically, intervocalic voicing is quite frequent, there is not much evidence for it elsewhere in Wapishana. Except for a few irregularities, voiceless consonants tended to remain voiceless (cf. the *s > ʃ change implied by *s2 in 9d-h). On the other hand, a look at non-Pidjanan cognates supports both *ɽ and *s. Looking at the Bahuana data, we see cases of ɾ (10b, d), suggesting a tap (*ɽ), and cases of ʐ (10a, c), suggesting a fricative (*s). It is as if an earlier (pre-Proto-Pidjanan) distinction between *ɽ and *s was neutralized to *ʐ in word-internal position, adjacent to *ɨ. 12 Taking all that into account, we propose (somewhat arbitrarily) to assign *s3 (*ʐ) to *ɽ, including cases that may later turn out to be from (PP or pre-PP) *s. 11 Ramírez (1992) used a capital R to represent the second, non-palatoalveolar Bahuana rhotic, which he described as a posterior velar tap. Later, Ramírez (2001) used IPA ɻ (retroflex approximant) for this sound, and also for the Wapishana rhoticized retroflex fricative here transcribed as ʐ, i.e., Ramírez apparently saw both the Bahuana and the Wapishana sounds as the same. For this reason, the same symbol ʐ is used here also for the Bahuana data. 12 Ramírez (1992, p. 16) mentions, for Bahuana, a similar neutralization of r (= ɾ) and R (= ʐ) to ʐ in a similar environment: between two ɨ's (e.g., /nɨ-ɾɨ/ = [nɨʐɨ] 'big').
The remaining correspondences, *s4, *s5, *ʃ1, *ʃ2, and *ʧ1, are all idiosyncratic cases. There is no evidence for a separate *ʃ (*ʃ1 and *ʃ2, and also *s5, probably result from *s in palatal environments), and *s4 and *ʧ1 suggest possible external influence. For further details, see the corresponding examples in the Appendix.
The remaining two correspondences, *ɗ3 and *d1, are more consequential, since they seem to suggest the reconstruction of a distinction between an implosive *ɗ and a non-implosive *d to Proto-Pidjanan.
(11) Examples of *ɗ3 (11a-c) and *d1 (11d-g). a. BELT: *ɽantawi > M ɾantawi-ɗa, W ʐaːɗawi b. NOSE: *(i)ɗiɓa > M ɾɨ-tiɓa, W ɨ-iɗiɓa c. THIN: *miCa-> M meta-ɗa, W miɗa-ʔɨ d. COMB: *mauCi > M muti-ɓa, W maudi e. FATHER: *Ca > M ɾɨ-ta, W ɨ-da-ɾɨ f. MOTHER: *Ca-ɾu > M ɾiː-taɾu, W ɨ-daɾu g. INAJÁ PALM: *puk-> M ukatɨ, W pukuɾidi Interestingly, most cases of *ɗ3 and *d1 have an adjacent *i, which suggests reconstruction as *ɗ (the cases of *ɗ examined thus far do not occur in this environment). There are, however, a few problems: PERSON *piɗaɲa (M ɗ j e(ː), W piɗana) seems to contrast with THIN (11c), and FATHER and MOTHER (11e-f) do not have adjacent *i's. The former case does not seem enough to warrant the reconstruction of a new proto-segment, but it is also difficult to choose between already available ones (*ɗ, *ɗ j and *t would all imply different reflexes). It seems best to signal this uncertainty by adding an undetermined consonant *C to the reconstruction (*miCa-). For NOSE (11b), *ɗ is reconstructed, with the adjacent *i's as the conditioning environment; for BELT (11a), *nt is reconstructed as the only example of a Proto-Pidjanan nasal cluster in the comparative corpus. As for FATHER and MOTHER (11e-f), there cleary is a common morpheme (M ta, W da), so that both cognate sets are effectively the same. There is therefore only one example of t : d not adjacent to a palatalizing *i, i.e., the reconstruction of a PP *d would have little basis. Taking also into account that the Wapishana d/ɗ distinction is somewhat problematic (W ɗ outnumbers d by a ratio of 6:1, and there are cases of variation), it seems best to reconstruct only an undetermined consonant *C, while waiting for more evidence on the history of FATHER and MOTHER. For COMB (11d), an undetermined *C also seems to be better (*ɗ would not be possible, since *ɗ3 in the same environment has already been reconstructed as *ɗ; see the contrast between *ɗ3 in NOSE 11b and COMB). It is hoped that further research will shed some light on the nature of these *C's (note that *C's reflected as *ɗ3, like 11c, are probably not the same as *C's reflected as *d1, like 11d-f).

Correspondences Involving Rhotics
In Table 12 below, we can see all Pidjanan correspondences involving ɾ. Of these correspondences, *ɾ1 is an identity and thus the best candidate for immediate reconstruction as *ɾ. It is also very frequent, occurring in 28% of all cognate sets (see, e.g., 12a-g, i). The next one is *ɾ2, occurring in a robust 15 (10% of all) cognate sets (see, e.g., 12h-l). In several pairs, *ɾ2 and *ɾ1 occur seemingly in contrast, in the same environments (e.g., 12g and 12k-l, 12c and 12i, 12f and 12j, 12b/12e and 12h), which indicates that they cannot have the same reconstruction. Given the retroflex quality of W ʐ, the best reconstruction for *ɾ2 would be a retroflex tap (*ɽ). For an additional allophone of *ɽ, see *s3 in Sec. 5.2.2 above. The remaining rhotic correspondences all appear irregular: *ɾ5, for instance, contains only one example with what seems to be the only case of metathesis in the available comparative data; *ɾ3, again with only one example, occurs at morpheme boundaries with an ablative morpheme *(ɾ)iki whose diachronic evolution is poorly understood; and *ɾ4 has two examples of what looks like irregular *ɾ loss in Mawayana (BITE: M auʧa, W aɾuːta, BIRD SP. 2: M maːtɨ, W maɾatɨ; cf. cases in which, in the same environmentsm PP *ɾ is not lost in Mawayana, like BIRD SP. 1: M anaɾu, W anaɾau, PARROT SP.: M waɾu, W waɾu, or LANGUAGE: M aɾa, W paɾa-da, FISH SP. 2: M atimaɾa, W aʧimaɾa). Thus, for *ɾ3, *ɾ4, and *ɾ5, partial or total non-cognacy cannot be excluded. To mark this uncertainty, a parenthetical *(ɾ) is reconstructed. Table 13 below lists all correspondences involving nasal consonants found in the available data. The simplest correspondences are *m1 and *n1, indentity cases with multiple examples; they can be immediately reconstructed as *m and *n (e.g., SUN: M kamu, W kamuː, BRAZIL NUT: M minɨ, W minaɨ, LEAF: M anaɓa, W anaɓa, AS: M ni, W niː, etc.). The next correspondence, *n2, shows cases of loss in Mawayana, as exemplified below. Notice that it occurs most often with an adjacent *i (13a-d, k-l), but there are exceptions (13e-g, i-j), including one, MEAT FOOD, that contrasts with *n1 as part of an analogous pair with SONG (13g-h); cf. also BREAST (13i), which has M ɗɨ in the possessed form (e.g., nɨ-ɗɨ 'my breast'). This contrast implies that *n1 and *n2 must have different reconstructions. Since *n1 is already *n, and given the often palatal environments where *n2 occurs, we propose to reconstruct *n2 as *ɲ, a consonant lost in Mawayana. Note that the *n4 example, TONGUE (13d), can also be explained via ɲ-loss: *n4 can thus also be reconstructed as *ɲ. Note that *ɲ-loss often, but not always, nasalizes adjacent vowels, sometimes leaving behind a palatal glide j, realized as [ɲ] (compare 13a-g and 13i-l). When nasalization happens, diphthong reduction does not (compare 13b and 13k). In Wapishana, *ɲ usually becomes n; the only exception, HEART (13e), is surprising (cf. 13d; see HEART in the Appendix). The remaining correspondences look irregular: *n3 contains two cases of word-initial syllabic n in Mawayana without counterparts in Wapishana (I: M n.nu, W ũ-ɡaɾɨ, LOUSE: M n.ni, W nai) and a Mawayana nasal-stop cluster corresponding to Wapishana vowel length (BELT: M ɾantawɨ-ɗa, W ʐaːɗawɨ), which suggests n-loss with compensatory lengthening and voicing of the remaining stop in Wapishana. Since these are the only cases in the available comparative data, it is better not to make too many inferences and simply note those as irregular developments of *n. As for *n5, its one example (TICK: M kunuɾiɓa, W kuɾinaɓa) contains what may be the only occurrence of metathesis in the available comparative data. For further details, see the Appendix.

Remaining Correspondences: Approximants and Glottal Stops
The last few consonant correspondences are listed in Table 14. It is obvious that *w1, an identity correspondence, is the best candidate for immediate reconstruction as *w (in, e.g., AÇAÍ: M waɓu, W waɓu, PAYMENT: M wina, W winipa, SALT: M ɗɨwɨ, W ɗɨwɨ). Likewise, *ʔ1 is the best (in fact, the only) candidate for reconstructing a Proto-Pidjanan glottal stop, even though it is not an identity correspondence. It reflects ʔ-loss in Mawayana, with subsequent diphthong reduction or, if the two adjacent vowels were the same, fusion (e.g., HIT: M ɾita, W: ʐuʔita, HAND: M kɨ-ɓa, M kaʔɨ, BIRD: M kuʧɨsa, W kutɨʔɨʐa). Note that the one *w2 example, ROPE (M jũwi, W inuʔi, inuɓi), includes a glottal stop and shows (dialectal?) variation within Wapishana. This variation is especially frequent when ɓ or ɗ are adjacent to the high vowels i, u (cf. W1 paritiɓi, paɾitiʔi 'ashes', ɗokoɾiɗi, ɗokoɾiʔi 'chest', ɗuɗua, ɗuʔua 'break', tuubuʐi, tuʔuʐi 'huge'). The Mawayana cognate jũwi can be readily explained if the glottal stop is reconstructed: as was the case for *ʔ1, it was lost, with M w resulting from the expansion of the preceding u to onset position (*iɲuʔi > *i.ũ.i > *jũ.i > M jũwi). Therefore, *w2 is really a case of *ʔ1 and is reconstructed as *ʔ. 14 Looking finally at the palatal glide correspondences *j1, *j2, and *j3, all three can clearly be reconstructed as *j. All three contain examples in which the Wapishana cognate shows loss of the final unstressed vowel, with the preceding glide being reduced either to the second element of a diphthong (*j1 PAPAYA: M maɓaja, W maʔapai) or to vowel length (*j3 HEART: M ĩkĩjã, W ɲɨkɨnɨː, *j3 KNIFE: M maɾe, W maɾiː, ɨ-marija-n), according to the surrounding environment: *j2 when the preceding vowel is *i, *j3 when it is *ɨ, and *j1 when it is neither; with ɨ, a tendency to avoid a final W ɨi diphthong, suggestive of the non-possessed marker -i, may have played a role). In Mawayana, *j is lost only in KNIFE, for which an explanation based on a-to-e raising and diphthong reduction can be readily found (PP *maɾija > *maɾi(j)e > *maɾie > M maɾe). Therefore, no contrast exists between *j1, *j2, and *j3, all reconstructible as *j. 15
Most non-palatoalveolar consonants did not change much either. The nasals *m and *n remained very much the same, except for occasional cases of loss with nasalization of neighboring vowels in Wapishana. The velar stop *k also remained unchanged, as did the voiced implosives *ɓ and *ɗ and the approximants *w and *j (except when the vowel following the approximant was lost, in which case it resyllabified with the preceding vowel, forming a diphthong). The labial and glottal stops (*p and *Ɂ) were lost in Mawayana but conserved in Wapishana.
It was in the palatoalveolar region that most of the action happened in Pidjanan history. The stop *t was conserved in Wapishana but changed to in ʧ Mawayana (except word-initially, followed by a, where it was conserved as t); the affricate *ʦ did not survive, becoming t in Wapishana and s in Mawayana. These two sounds were palatalized when adjacent to *i, becoming respectively ʃ and s in Mawayana; in Wapishana, even in this environment *t was preserved and *ʦ was fortitioned to t. The affricate *ʧ was conserved in Wapishana and fortitioned to in Mawayana. The palatal nasal *ɲ was lost in Wapishana (often nasalizing neighboring vowels) and depalatalized to n in Wapishana. The palatal tap *ɾ merged with *ɽ in Mawayana (resulting in ɾ). In Wapishana, *ɽ sibilantized to ʐ. (A few *ɽ may be historically derived from (pre-)PP *s surrounded by *ɨ's, an environment in which *ɽ and *s have apparently merged; judging by the reflexes, *ɽ was realized as * [ʐ] in this environment). The fricative *s was basically conserved in Mawayana but palatalized to ʃ in Wapishana.
As one can see, most changes were assimilatory (palatalization, raising, etc.), with a few cases of lenition and fortition thrown in for good measure. Only very sporadically did possible cases of metathesis or haplology surface.

Basic Proto-Pidjanan Morphology
At this point, a full reconstruction of Proto-Pidjanan morphological markers cannot be carried out. A few aspects of the basic morphology, however, can be preliminarily approached.
The person-marking system, shown in Table 17 in the Appendix, is mostly easy to reconstruct on the basis of what has been discovered so far about Pidjanan historical phonology. e.g., ɗe-sɨ 'son', ɗe-ɾu 'daughter'). Such remains, as well as the overwhelming evidence from outside of Pidjanan, warrant the reconstruction of gender distinctions to Proto-Pidjanan. The parenthetical *(ɾ) in 14c-d is based on the Mawayana masculine form, and also on overwhelming evidence from non-Pidjanan languages. It is true, however, that these *(ɾ)'s should have survived in Wapishana, which is why their reconstruction is only tentative. Finally, 14h (third-person plural) is more complicated: W nasal ĩ suggests PP *ni-, but the vowel does not agree with M -na. An undetermined vowel is reconstructed here, though one must admit that it is not clear that the two prefixes are actually cognate. The suffixes are clearly more difficult to reconstruct. First of all, because some forms are no longer attested (3S.FEM in Mawayana, though the final -ɾu in ɗe-ɾu 'daughter', ta-ɾu 'mother' is probably a remnant of this suffix; and the 1P and 2P forms in Wapishana, apparently now replaced by full pronouns). Secondly, the vowel is not always consistent (15b). Thirdly, there are apparently non-cognate forms (15g). Still, the existence of a suffix system, used on transitive verbs to refer to the object, must be reconstructed, even when the forms no longer exist (Wapishana 15e-f; the Mawayana forms are reconstructed by default), given the overwhelming agreement of non-Pidjanan languages on this topic. Further uses (the prefixes, besides marking verb subjects, are also used to mark the possessor on possessed nouns and the argument on postpositions; the suffixes also mark the subject of stative intransitive verbs) must likewise be reconstructed, even if in some cases they are no longer attested (Wapishana has apparently lost the use of the suffixes to mark the subject of stative intransitives, but Mawayana still retains it). In the realm of nominal possession, it is possible to reconstruct possession-marking suffixes zero and *-nɨ (17a-b); for inalienable nouns, it is possible to reconstruct a marker of non-possessed status *-i (17c; cf. also footnote 4). Finally, in the realm of verbal tense-aspect-mood morphology, it is possible to reconstruct a 'realis' / 'present' marker *-ɲɨ, from which both W -n, -nɨ 'realis' and M -e 'present' (actually, the raising of the stem-final vowel a to e): *-ɲɨ > W -nɨ regularly, and *-ɲɨ > *-ɲi (assimilação) > *-i (ɲloss); *-i then raises (and/or merges with) the stem-final vowel a, leading to M -e (see also footnote 5 above). 16 (18) Verbal TAM marker *-ɲɨ (> W -n(ɨ); *-ɲɨ > *-ɲi > *-i > M stem-final vowel change).
a. *(ɾ)ɨ-puːta-ɲɨ 'he blows' > M ɾ-uʧe, W ɨ-puːta-n APPENDIX This is a list of all Wapishana-Mawana cognates found in the available data. The cog-nate sets are alphabetized by (English) meaning. Each set includes the meaning (with a parenthetic 'to' if the meaning is verbal: e.g., GIVE (TO)), the reconstructed protoform (with a question mark when it is problematic), the cognates (M = Mawayana, W = Wapishana; variants are separated by commas), and comments. The meaning of a cognate is given only if it differs from the meaning of the set. If cognates include affixes, these are segmented out with hyphens; the target of cognacy, written in bold, is usually the stem. The most frequent affixes are listed in Table 17 below. ũ--n thematic -pa, -ta, -ka 2. sg.
ɨ-passive -kau 3 pl. ĩ-nomnlzr. -ʔu The default source for the data is the author's fieldwork. If there are disagreements between sources, or if a specific form is attested only in a source other than the author's fieldwork, then the source will be identified by a number: for Mawayana, M1 = Howard 1985= Howard -1986