West-coast Walu-Siwaish (master post)

(view clickable tree)

(jump up one level to Walu-Siwaish)

The West-coast Walu-Siwaish languages are Papora and Hoanya, two extinct languages, each of them significantly diverse. Their last speakers were interviewed mainly by Japanese linguists in the late 19th and early 20th centuries. We possess short vocabularies for a number of locations in rough, not fully systematic phonetic transcriptions. Tsuchida (1982) includes a collection of these forms,  now accessible online through the Hoanya and Papora pages of the Austronesian Basic Vocabulary Database (ABVD; here and here) of Greenhill et al. (2003-).

Despite the difficulties inherent in establishing cognacy given the nature of the data, the Hoanya and Papora words for ‘fire’ and ‘black’, present us with two uniquely shared western Walu-Siwaish innovations. 

A correspondence between Hoanya and Papora voiced coronals  recurs over these comparisons, which include ‘fire’ and ‘black’:

  Hoanya #1-1a Hoanya #1-8 Papora #1-3 Papora #2 
  dz z d d
dzapu zapū dapu dapū
rain mudzas _
modad _
road, path dzalan _ dalan  
black mavidzu mavizū abidu avedoo

‘Fire’. Blust’s Austronesian Comparative Dictionary (ACD; here) assigns the Hoanya and Papora words for ‘fire’ to PANBlust *Sapuy ‘fire’. In their Austronesian Basic Vocabulary Database (ABVD), Greenhill et al. (2003-),  do the same, listing the forms  under cognate set #1 (here).   However this will not do, since Blust’s PAN *S goes to s- in Hoanya (*Sikan ‘fish’ > sikan, *Suaji ‘younger sibling’ > suazi, *Sepat ‘4’ > supat) and in Papora as well (*Siwa ‘nine’ > siya, *daqiS ‘forehead’ > ddes). Conversely none of the other Hoanya-Papora words with the same correspondence as ‘fire’ (see the above table) goes back to PAN *S. Either there has been an irregular sound change turning PAN *S into some voiced coronal in the Papora-Hoanya word for ‘fire’, or, more plausibly,  the Hoanya and Papora forms for ‘fire’ are cognates of another word: PMPBlust *dapuR ‘hearth’, even though this last form has not hitherto been observed in Formosan languages. Tsuchida (1969) reconstructs *Z in ‘rain’ (*qu2ZaN), one of the words which exemplify the Papora-Hoanya correspondence (above table). Allowing for the uncertainty due to the lack of  cognates in other Formosan languages, I will write PWS *[Z]apuR. Taking a hint from Wolff (2010) I gloss the form as  ‘cooking fire’. The best solution is to assume that *[Z]apuR displaced Blust’s *Sapuy as ‘fire’ exclusively in western Walu-Siwaish.  For the final consonant, Papora-Hoanya examples of PAN final *-R are few, but *-R falls in Hoanya #1-1b matulu ‘to sleep’ < *ma-tuduR  (this etymon is not found elsewhere in Papora or Hoanya).

‘Black’. The PAN word reconstructed by Blust as *qudem ‘black’ has a broad distribution in Taiwan; but it was displaced, exclusively in Hoanya and Papora, by a word reconstructible as *abid[z]u. There is an apparent cognate in Mongondow: mo-bidu ‘blue’, so that *abid[z]u can be assigned to PWS. However there was competition between *abid[z]u and *qudem in PWS: it is only in western Walu-Siwaish that the former diplaced the latter. In Mongondow mo-bidu has a doublet mo-biru ‘blue’. This mo-biru belongs to a widespread set which Blust (here) treats as borrowed from Malay. Although he admits that in Mongondow “it would normally be assumed that where variants in d, r appear the d variant is historically primary”, he assumes “for the present”  that “Mongondow mo-bidu is a result of analogical back-formation from a loanword with r“. However,  the Hoanya-Papora cognate convincingly shows that d, not r, is primary: indeed, Mongondow d is a match for Papora d, Hoanya dz, as in ‘road, path’ : Papora dalan, Hoanya dzalan, Mongondow daḷan. Only the biru-type forms are part of the loan distribution identified by Blust.


Blust, Robert A. and Stephen Trussel. Ongoing. Austronesian Comparative Dictionary (ACD). Online at http://www.trussel2.com/acd/

Greenhill, Simon, Robert Blust, and Russell Gray. 2003-. Austronesian Basic Vocabulary Database (ABVD). Online at  https://abvd.shh.mpg.de/austronesian/.

Tsuchida, S. (1982) A Comparative Vocabulary of Austronesian Languages of Sinicized Ethnic Groups in Taiwan, Part I: Western Taiwan. Memoirs of the Faculty of Letters, No. 7, University of Tokyo. Tokyo, Japan.

Wolff, John U. 2010. Proto-Austronesian phonology with glossary. Ithaca: Cornell University Press.

Cite this post as: Laurent Sagart, "West-coast Walu-Siwaish (master post)," in Sino-Tibetan-Austronesian, 18/11/2020, https://stan.hypotheses.org/1065.



Limaish (master post)

(view clickable tree)

(Jump up one level to Pituish)

(jump down one level to Enemish)

Limaish is a daughter of Pituish. It is defined by two innovations:

1. *lima ‘five’.  This originates in *lima, *qa-lima ‘hand’. *lima displaces PAN *RaCep, the main word for ‘five’ on the northern half of the west coast:  Saisiyat, Pazeh, Favorlang, Babuza, Taokas. (cognate set #2 in the ABVD). *lima is virtually universal in the rest of Taiwan, Malayo-Polynesian and—outside of Chinese loanwords—Kra-Dai.

2. Limaish languages are also characterized by the appearance of a special series of numerals used with human referents. There is no evidence for such a series in the languages upstream of Limaish: Taokas, Favorlang, Babuza, Pazeh, Kaxabu, Kulon, Saisiyat, Luilang. The Limaish human numerals are derived out of the main series by means of Ca- reduplication, where ‘C’ is a copy of the base’s initial consonant, and –a is the plural marker of human and personal nouns. Because this series is initially a plural one, at the time it arises, a numeral for ‘one’ cannot be part of it. In the immediate daughters languages of Proto-Limaish, Thao and Atayal, tata and qutux, both ‘one’, respectively, serve to count things and humans. Here is a list of Limaish languages with Ca-reduplicated numerals for human reference:

Thao (Blust 2003): ta-tusha ‘two’, ta-turu ‘three’, ʃa-ʃpat ‘four’, ra-rima ‘five’, pa-pitu ‘seven’ (here)

Atayal (Mayrinax): ra-rusa? ‘two’, ta-tuu ‘three’, sa-spat ‘four’, a-ima ‘five’ (here)

Siraya (Adelaar 2011)  da-ruha ‘two’, ta-turo (-u) ‘three’, pa-xpat ‘four’, ra-rĭma ‘five’, pa-pĭto (-u) ‘seven’.

Bunun (Isbukun) ta-cini (?) ‘one’, da-dusa ‘two’, ta-tau ‘three’, sa-sapaat ‘four’, a-ima/ha-hima ‘five’, a-apnum ‘six’, pa-pitu ‘seven’, va-vau ‘eight’, sa-siva ‘nine’, ma-masan ‘ten’ (here)

Rukai: Tanan Rukai does have a separate series for counting humans (and animals), but instead of Ca- reduplication, the series is marked by ta- prefixation (here).

Kanakanabu: tacíni ‘one’, tasúsa ‘two’, tatúlu ‘three’, sasúpatə ‘four’, lalíma ‘five’, nanəmə ‘six’, papítu ‘seven’, laláalu ‘eight’, sasía ‘nine’, mamáanə ‘ten’ (here).

Saaroa: tsa-tsiɬi ‘one’, sa-sua ‘two’, ta-tulu ‘three’, a-upatɨ ‘four’, la-lima ‘five’, a-ɨnɨmɨ ‘six’, pa-pitu ‘seven’, la-la-alu ‘eight’, sa-sia ‘nine’ (here)

Tsou: Tsou has a vestigial Ca-reduplicated series, limited to sasupatə ‘four’ (here)

Kavalan has a series of numerals for counting humans, but replaces Ca-reduplication by prefixation of kin-, after the break-up of Eastern Walu-Siwaish.

The Formosan Puluqish languages: Amis, Puyuma, Paiwan, have either lost the series (Amis), or mark it with a prefix (Puyuma mia-, Paiwan ma-). Nevertheless Puluqish languages outside of Taiwan, like Itbayaten and Chamorro, tell us that Proto-Puluqish did have a human counting series marked by Ca-reduplication:

Itbayaten: da-doha ‘two’, la-lima ‘five’, pa-pito ‘seven’, wa-waxo ‘eight’, sa-siyam ‘nine’ (here)

Chamorro: ta-to ‘three’, fa-fiti ‘seven’, gua-gualo ‘eight’, sa-sigwa ‘nine’ (here; fa-fiti is incorrectly assigned to ‘six’).

Most Malayo-Polynesian languages lack an independent series of  numerals for human reference but occasionally forms with Ca-reduplication crop up among a language’s numerals, e.g. Tagalog dalawa ‘two’, tatlo ‘three’.

The Kra-Dai languages do not preserve the left edge of words well enough for us to detect a Ca- reduplication with confidence.


Adelaar, Alexander K. 2011. Siraya, Retrieving the Phonology, Grammar and Lexicon of a Dormant Formosan Language. Berlin: De Gruyter Mouton.

Blust, Robert A. 2003. Thao Dictionary. Nankang: Institute of Linguistics (preparatory office), Academia Sinica.

Web site:

Eugene Chan, Numeral Systems of the World’s Languages. Online at https://mpi-lingweb.shh.mpg.de/numeral/. Accessed May 25, 2020.

Cite this post as: Laurent Sagart, "Limaish (master post)," in Sino-Tibetan-Austronesian, 24/05/2020, https://stan.hypotheses.org/914.

Southern Austronesian *sa-puluq-at in numbers 10 to 19

(view clickable tree)

Jump to Southern Austronesian master post

This post returns to a subject discussed in a 2010 conference paper (here). New facts are presented and the interpretation is slightly modified.

Several Philippine languages belonging to the South-Central Cordilleran subgroup in northern Luzon: Kankanay, Kalinga, Bontok, Ifugao, Balangaw, Inibaloi, have a conjunction reflecting Proto-South-Central Cordilleran *et [ət] ‘and, and then’. In the Central Luzon group of Philippine languages, Kapampangan, not in direct contact with the Cordilleran languages, has at ‘and’, reflecting either *et [ət] or *at. Tagalog, a Central Philippines language, has at ‘and’, which can only reflect *at, unless it is a loan from neighboring Kapampangan at ‘id.’ (personal communications from L. Reid and R.D. Zorc, April 2007). Kapampangan is known to be a source of loanwords to Tagalog. Among the Central Cordilleran languages reflecting *et ‘and’, Isinai, Kalinga and a few others use a phonologically reduced form of that morpheme between pulu ‘ten’ and a following unit numeral. Thus in Kalinga, ‘ten’ is simpulu < *sa-n(g)a-puluq, but ‘eleven’ is simpu:lut osa (Tryon et al. 1995, Vol. 4, 43-45). That last form is analyzable as /sin-pulu=t osa/, where /=t/ ‘and’, /osa/ < *əsa ‘one’. Similarly, ‘twelve’ is simpu:lut duwa /sin-pulu=t duwa/, where /duwa/ < *duSa ‘two’ and /=t/ ‘and’. Kapampangan, the only Central Luzon language with a reflex of *et, does not use it in compound numerals. Although Tagalog may have borrowed the conjunction from Kapampangan, it does use a reduced form of at in compound numerals above ‘20’, in a very similar way to Kalinga and Isinay:


twenty’ < *daduha ‘two’ + nasal linker + *puluq ‘ten’

dalawampút isá

twenty-one’ < *daduhatwo’ + nasal linker + *puluq ‘ten’ + =(a)t ‘and’ + *esa ‘one’

dalawampút dalawá

twenty-two’ < *daduhatwo’ + nasal linker + *puluq ‘ten’ + =(a)t ‘and’ + *daduha ‘two’

The puluq-t-numeral form cannot easily have been borrowed from Cordilleran into Tagalog, as the languages are not in contact: if Tagalog borrowed the conjunction from Kapampangan, Tagalog shows a development parallel to Cordilleran in its use of the conjunction.

There are two possible accounts of the development of the conjunction itself in the Philippine languages. In the first, the conjunction’s oldest form in the Philippines was *et [ət], regularly evolving to at in Kapampangan, and loaned to Tagalog as at. Under that account, the conjunction, inherited only in the South-Central Cordilleran and Central Luzon groups, reconstructs to the northern clade of Philippine languages in the phylogeny of Gray, Drummond and Greenhill (2009), which this post follows for Malayo-Polynesian. Use of the conjunction between *puluq and a following numeral in Cordilleran and in Tagalog must be the result of parallel innovations.

In a second account, the conjunction’s oldest form in the Philippines was *at. Kapampangan and Tagalog inherited it in that shape, while in the Cordilleran languages the presumably unstressed vowel underwent centralization, to [ət]. Under that account, Tagalog being part of the southern clade, both *at ‘and then’ and the *puluq=at+numeral construction reconstruct higher, to the ancestor of the southern and northern Philippine clades.2

Whether as *et or *at, the Philippine conjunction is reminiscent of the form ato ‘and, with’ in Amis (central Amis: Fey 1986, Namoh 2013) and of the conjunction katu in Paiwan. In Nataoran Amis (Nataoran Amis: Bril fieldwork), atu may occur in numeral expressions (Isabelle Bril, p.c., 11 February 2014):

a cacay-en atu la-tusa ku ni-pabeli itakuan-an tu pawli

HORT one-UV and half NOM PERF.NMZ-give me-OBL ACC banana

Give me one and a half banana (lit. ‘let it be one and a half the giving a banana to me’)

Both Namoh (2013:136) and Bril treat ato~atu as originally composed of a ‘and’ plus =to ‘oblique common noun marker’, drawing a parallel to aci ‘and + personal noun’ (<a ‘and’ + =ci ‘personal noun marker’). The oblique common noun marker to itself is analyzable into t- oblique and u or o common noun marker.

In a northern Amis numeral system recorded by Tsuchida (here), tu, a shortened form of atu, serves in numerals between 11 and 19 as a linker between the locally innovated word for ‘ten’: sabaw , and the following unit number, e.g. sabaw tu tulu ‘13’, sabaw tu lima ‘15’.

Similarly, in a Paiwan dialect recorded by Chang Hsiou-chuan (here), a conjunction katu occurs between *sa-puluq and a following unit number, e.g. tapuluʔ katu ʔunəm ‘16’. katu is analyzeable as katu ‘and-oblique noun marker’, or possibly kaatu ‘andoblique noun marker’.

While *sa-puluq, the Puluqish word for ‘10’ ends in *-q, which correspoponds to -k in Kra-Dai, the Kra-Dai word for ‘10’ ends in *-t, as if it came from *pulut. The Philippine developments suggest an explanation. The following scenario is based on the hypothesis that the oldest Philippine form of the conjuction was *at. It is also possible to explain the facts under the view that it was *et, but the explanation is slightly more complex.

Comparison of Amis and Paiwan leads one to posit a conjunction *atu ‘and’ between *sa-puluq ‘10’ and a following numeral. In PSA the presumably enclitic conjunction was reduced to *at through syncope. Proto-Malayo-Polynesian and Proto-Kra-Dai, the direct daughters of PSA, inherited both *at and the compound numeral construction with it. Proto-Malayo-Polynesian in turn transmitted these to its immediate daughter Proto-Philippines, but the conjunction and construction were lost in Malayo-Polynesian outside of the Philippines. For this to occur, a single loss is required as such a node is supported with 98% posterior probability in Gray, Drummond and Greenhill (2009). Soon after Kra-Dai branched off, and independently from developments in the Philippines, *at ‘and’ was further reduced to /t/, and the resulting *sa-puluqt was reduced to *sa-pulut in PKD. This then evolved to Proto-Hlai *apu:c, Proto-Kra *pwlot. The outcome in Proto-Kam-Tai is unknown because these KD languages have abandoned Austronesian numerals for Chinese ones.

Cite this post as: Laurent Sagart, "Southern Austronesian *sa-puluq-at in numbers 10 to 19," in Sino-Tibetan-Austronesian, 07/05/2020, https://stan.hypotheses.org/881.

Southern Austronesian lexical innovations

(to clickable tree) 

(up one level to Southern Austronesian)

Each of the lexical innovations listed below associates a Malayo-Polynesian form with reconstructions from one or more Kra-Dai branches. The MP side of the comparison either has no Formosan cognates, or cognates with distinct semantics. If it has no Formosan cognate, a Formosan etymon of the same meaning, partly or fully displaced by the MP etymon, exists.

1. PSA *baqbaq ‘mouth’.  Tsuchida’s *ŋuθuq ‘mouth’ (1976:130) is identical with Blust’s *ŋusuq ‘mouth’ and Wolff’s *ŋucuq ‘snout, beak’. This is the highest-reconstructing  Austronesian etymon for ‘mouth’ (humans and animals). It is reflected in Tsouic, Batanic and Oceanic (cognate set #2 here). In MP  *baqbaq ‘mouth’ (cognate set #1 here; see also here) is more widespread. In Taiwan Wolff recognized its precursor in Bunun vaqvaq ‘chin’ (2010:757); Bril (p.c., April 13, 2020) mentioned northern Amis babaq ‘jaw’. The most likely scenario, then, is that *baqbaq, a word for ‘chin’ or (lower ?) ‘jaw’ in a Formosan precursor of PMP, shifted to ‘mouth’ in PMP, competing with *ŋuθuq, ultimately displacing it in Cebuano, Malagasy, Old Javanese and other languages.

A reflex of *baqbaq is the word for ‘mouth’ in the Kam-Tai branch of Kra-Dai: Proto-Tai (Pittayaporn) *pa:k D, Proto-Kam-Sui (Thurgood) *pa:k 7, Proto-Ong-Be *ɓa:k D1.  For the  correspondence of PAN *-aq to Kam-Tai *-a:k, compare ‘otter’, PAN  *Sanaq : Proto-Tai *na:k.  Normally a singleton PAN *b in word medial position ought to evolve to Kam-Tai initial *b or *ɓ, not to *p: *p- in this word actually corresponds to PMP *b in the word-internal *-qb- cluster.

2. PSA *biRáq ‘kind of taro’.  An etymon *biRaq has been assigned to PAN in two meanings: ‘leaf’ and ‘taro/wild taro’. Evidently these are related, as the taro’s leaf is remarkable. Blust reconstructs the meaning as ‘taro’ (here) but as shown by Wolff (2010) Saisiyat bilaʔ, Tanan Rukai bia, Kavalan biɣi, Puyuma bira, all mean ‘leaf’. In MP,  the meaning ‘leaf’ has disappeared, all languages indicate ‘taro’ or a meaning derivable from ‘taro’.

In the Kra-Dai languages, *biRaq appears as ‘taro’, never as ‘leaf’: Proto-Kra *p-ɣak D (Ostapirat), Proto-Hlai (Norquest) *ra:k, Proto-Tai (Pittayaporn) *prɯək, proto-Kam-Sui (Thurgood) *ɓra:k 7.  The vowel in the Proto-Tai form points to a precursor with penultimate stress, such as reduplicated *biRaq-bíRaq. Kra-Dai and MP share the loss of the meaning ‘leaf’.

3. PSA -ŋel ‘deaf’. Blust reconstructed PAN * Culi ‘deaf’ (here), Wolff *tilu ‘id’. This word persists in MP, but MP languages add words for ‘deaf’ having root *-ŋel (here), possibly somehow connected with *deŋeR ‘hear’ (Puyuma, Paiwan, MP).

The Kra-Dai languages reflect *ŋel: Proto-Kra (Ostapirat) *ŋel C ‘deaf’.

4. PSA *píntu ‘door’. A PMP word for ‘door’, reflected as Malay pintu, Tagalog pinto, pintu-an reconstructs to PMP as *pintu, even though the cognate set is not recognized by either of Blust or Wolff. Formosan words for ‘door’ often include root *-Neb ‘close’ (Blust; here) so *pintu is clearly innovative. Its etymology is unknown. Metanalysis of a compound including root *-pit ‘hold by pinching’ is a possibility (Wolff 2010; see also Blust’s evidence for this root here).

The Kra-Dai word for ‘door’ is cognate with PMP *pintu. Buyang, a Kra language, has ma tɔ  A2 ‘door’, where ma is the expected reflex of an Austronesian first syllable beginning in a labial (e.g. *walu > ma ðu ‘eight’, *buŋaH1 > ma ŋa C ‘flower’, *maCa > ma ta A ‘eye’), and   A2 reflects an unstressed *-ntu, i.e. a paroxytone PSA prototype *píntu. ‘A2’ is the low variant of tone A: it shows that initial t– went through a voiced stage, as it evolved out of  PSA *-nt-. PSA plain medial *-t- would evolve to Buyang t- with a high-series tone. Compare PSA *taŋkup ‘to cover’, PMP *ta(ŋ)kup or *tuŋkup id., Buyang ta qup D2 id. Proto-Lakkja (Theraphan) also has a low tone in ‘door’: *tɔ: A2.

5. PSA *qi(d)zúR ‘sputum, saliva’. The earliest recontructable Austronesian form for ‘sputum’ or ‘saliva’ is PCEWS *ŋalay, a form reflected in Rukai-Tsouic, Amis, Puyuma and Bashiic (Tsuchida 1976:230). The corresponding reconstruction in Blust’s system: *ŋajay is erroneous, as the medial consonants in Paiwan ŋadjay and –h– in Yami ŋahay are not regular outcomes of PAN *-j-. In MP, *ŋalay is not found except in several Bashiic languages (Yami, Itbayaten, Ivatan…). In KD, Proto-Tai *la:j A, Proto-Lakkja *lei A reflect *ŋalay, and Ferlus (REF), unaware of the Austronesian comparison, reconstructs the Proto-Kam-Sui initial as *ŋl-.

PMP innovates *qi(d)zuR ‘saliva, spittle’ (Blust *qizuR), reflected in the Philippines (Ibaloi, Southern Cordilleran), Toba Batak, Javanese. In KD, the Kra branch reflects *qi(d)zuR: Buyang qa tu B (remember that PSA *-R goes to tone B in KD).

From what precedes, PCEWS *ŋalay and PMP *qi(d)zuR are in competition in Philippine and KD languages, and nowhere else. This fit well with the idea that the competition existed from PSA through an early time in the diversification of MP.

6. PSA qa-sáuŋ ‘canine tooth’. The highest reconstructable word for ‘canine tooth’ is *waqit (Blust, ACD), a Formosa-only etymon reflected in Atayal and Puyuma. *bangeliS (Also Blust, ACD), with reflexes in Kavalan and MP, is specifically a word for ‘tusk’. As ‘canine tooth’, *sáuŋ  displaced *waqit in Philippine languages. This became the word for ‘tooth’ in the Kra branch of KD: Proto-Kra *C-tʃuŋ A ‘tooth’ reflects *sáuŋ with a prefixed element, which Buyang qa ɕɔŋ 54 ‘tooth’ shows to have been a velar or uvular: *ka-sáuŋ or *qa- sáuŋ.

7. PSA *sapeléd, *ma-sapeléd ‘astringent’. Blust (ongoing) reconstructed *qasepa ‘astringent’, reflected in Puyuma and MP, and thus assignable to Puluqish. Only in Puyuma is the meaning ‘astringent’: all MP forms (Iban, Malay, Javanese) mean ‘stale’, ‘insipid’, ‘tasteless’. Subsequent to this semantic shift, the Puluqish word was displaced by a new word: PMP *sapeled, with reflexes in the Philippines, Malay and Mongondow (reconstructed in Blust, ongoing). A stative ma-prefix is moreover in evidence in Bikol, Hanunoo. In Kra-Dai, Buyang phat 54 ‘astringent’ reflects a prefixed variant of *sapeled, possibly *ma-sapeled. The Buyang vowel points to a stressed vowel in the PSA prototype, thus *Pfx-sapeléd. Austronesian  vowels except for the last one are  lost in the evolution to Kra. For the reduction of the -pl- internal cluster to -p- in Buyang, compare PSA *sa-puluq-Vt ‘ten’, Buyang put 54 id. Buyang voceless aspirated stops originate in voiceless stops preceded by a voiceless obstruent, here /s/. Proto-Tai ʰwɯət probably has the same origin as Buyang *phat. Of the two words for ‘astringent’, it is the innovated one which occurs in KD.

8. PSA *buŋáH1 and *bujak ‘flower’.  Which was the earliest Austronesian etymon for ‘flower’ ? Despite Blust (ongoing) and Wolff (2010), not *buŋaH1—I reconstruct final H1 in this word based on the correspondence between Aklanon final -h and Proto-Kra-Dai tone C. When examining the semantics of *buŋáH1 one must evidently leave aside languages (Atayal, Seediq, Thao, Amis, Puyuma, Thao) where the referent is ‘sweet potato’, a south American domesticate spread across the Pacific in prehistoric times. Such forms are in all likelihood loanwords, spread from an unknown source. This leaves only the two Southern Tsouic languages Kanakanabu and Saaroa. Inherited reflexes of *buŋaH1,not meaning ‘sweet potato’,  have been reported to occur there. During fieldwork on Saaroa in 2014 by myself and Hsu Tzefu, I elicited vuŋavuŋa < *buŋa-buŋa as ‘ear of foxtail millet’ (here). Wolff (2010) gave Saaroa vuu-vúŋa < *buu-buŋa ‘flower’. For Kanakanabu we have conflicting evidence. Wolff (2010 sub *buŋa) cited vuŋávuŋu, which seems to be vuŋ < *buŋ reduplicated, with final echo vowel; he notes a variant vuŋávuŋ. This does not reflect  *buŋa well—where is final *-a ? more probably it reflects *buŋ-a-buŋ. A root connection to Proto-Philippine *sabuŋ ‘flower’ (here) is possible. Blust (here) cited Kanakanabu buŋa-buŋa ‘flower’, the language source hyperlinked to Tsuchida (1976), but I could not find that form there.  Reconstructing Proto-Southern Tsouic (PST)  from Blust’s Kanakanabu buŋa-buŋa ‘flower’ and my Saaroa vuŋavuŋa ‘ear of foxtail millet’ one obtains PST *buŋa-buŋa ‘outgrowth of a plant’. The meaning ‘flower’ is too narrow.

‘Outgrowth of a plant’ also characterizes the meaning of *buŋáH1 in Philippine languages well: ‘fruit’, ‘seed’, ‘sapling’, ‘outgrowth’, ‘result’ are common meanings. ‘Flower’ is not among them, evidently because *buŋáH1 as ‘flower’ was displaced by *bujak (here). In MP languages from Malagasy to Oceanic where *bujak has not become the word for ‘flower’, the meaning of *buŋáH1 has narrowed down to ‘flower’.

What, then, was the earliest AN word for ‘flower’ ? very probably *buRay, a reconstruction of Dyen’s accepted by Blust   (here). This is based on Saisiyat and Paiwan forms, with the possible addition of Atayal and Kavalan.   In the post-Formosan phase of Austronesian history, *buRay was displaced as ‘flower’ by innovative *bujak and a semantically narrowed version of *buŋáH1. The former took hold in the Philippines, the latter in the rest of Malayo-Polynesian. Kra-Dai reflects both *buŋáH1 and *bujak.

A direct cognate of *buŋáH1 is seen in the Kra branch: Buyang ma ŋa 11 (tone C) ‘flower’, Proto-Kra (Ostapirat) *hŋa C id.. Preinitial ma in Buyang is regular for any labial initial in PSA. *h in Ostapirat’s reconstruction *hŋa C serves to account for the high-series tones in certain Kra varieties, but a reconstruction with *pŋ- or *ɓŋ- would serve the same purpose (the KD reflexes of PSA *b and *d are high-toned in KD), while accounting better for the Buyang pre-initial.

Proto-Tai *ɓlo:k (Pittayaporn) ‘flower’ is a cognate of *bujak .  PSA *j merges with PSA *d as PT *ɗ: *mújiŋ ‘nose’, PT *ɗaŋ A id. ; *púja ‘navel’, PT *ɗwɯ A. When forming a cluster with a preceding labial stop (as inside a trisyllable, where syncope of the second vowel occurs), the stop is retained and *ɗ (like *t-) lenites to l-: PSA *(Pfx-)punti ‘banana’, PT pli: A ‘banana blossom’; PSA *qapejúH2 ‘gall’ : PT *ɓli: A id. (change of *u to *i after *j is regular).

In order to have cognates of both *buŋáH1 and *bujak, KD needs to have branched off at a time when both words were in competition. That would have been the case between the moment when  *buRay was displaced (by *buŋáH1 and *bujak) and the separation between Proto-Philippines and the rest of MP: in other words, around the time of the out-of Taiwan event, c. 2000 BCE.


RETRACTED: PSA *sedút ‘to suck’. Austronesian etyma for ‘to sip, suck’ assignable to pre-MP times are mostly onomatopoetic forms that begin in a sibilant and end in -p: Blust’s PAN *SiRup, *sepsep, *supsup, Wolff’s *siɣup, *cep, *cepecep , *cipecip, etc.  Blust’s PWMP *sedut ‘to sip, suck’ is a minor form with a  patchy but geographically widespread distribution in Sarawak, Lombok and New Guinea.

Kra-Dai has a cognate of *sedut: Proto-Hlai (Ostapirat) *iru:c ‘to suck’, Saek du:t 6, Siamese du:t 2.

This comparison is retracted because of Amis hedot ‘to ingest noisily, making sucking noises’ (Namoh). This apparently reflects *Sedut,  distinct from *sedut (which would give Amis #cedot), but the Kra-Dai forms just cited could reflect either.


Dyen, Isidore. 1995.Borrowing and inheritance in Austronesianistics. In Paul Jen-kuei Li, Cheng-hwa Tsang, Ying-kuei Huang, Dah-an Ho, and Chiu-yu Tseng, eds., Austronesian studies relating to Taiwan:455-519. Symposium Series of the Institute of History and Philology, No. 3. Taipei: Academia Sinica].

Norquest, Peter. 2015. A Phonological Reconstruction of Proto-Hlai. Brill.

Ostapirat, Weera. 2000. Proto-Kra. Linguistics of the Tibeto-Burman Area 23.1:1-251.

Pittayaporn, Pittayawat. 2009. The phonology of proto-Tai. Unpublished Cornell U. dissertation.

Theraphan L.-Thongkum. 1992. A preliminary reconstruction of Proto-Lakkja (Cha Shan Yao). Mon-Khmer Studies 20: 57-89.

Thurgood, Graham. 1988. Notes on the reconstruction of Proto-Kam-Sui. In Jerold A Edmondson and David B. Solnit (eds) Comparative Kadai: Linguistics studies beyond Tai, 179-218. Dallas: Summer Institute of Linguistics and the University of Texas at Arlington publications in Linguistics.

Tsuchida, S. (1976) Reconstruction of Proto-Tsouic phonology. Tokyo: Study of languages and cultures of Asia and Africa monograph series N° 5.

Wolff, John U. (2010) Proto-Austronesian phonology with glossary. 2 vols. Ithaca: Cornell Southeast Asia program publications.

Cite this post as: Laurent Sagart, "Southern Austronesian lexical innovations," in Sino-Tibetan-Austronesian, 07/05/2020, https://stan.hypotheses.org/875.

Rukai-Tsouic (master post)

(Jump up to Central Walu-Siwaish)

(Jump down one level to Tsouic)

Rukai-Tsouic consists of Tsouic and Rukai. Rukai-Tsouic innovations are detailed below.

  • Morphologically, the Rukai-Tsouic languages replaced CV-reduplication as a marker of the non-human series of numerals with prefixation of *u-, a common noun marker in PAN (details here).

This is not a uniquely shared Rukai-Tsouic innovation, however, because like Rukai-Tsouic, Kavalan (North Formosan) also replaced CV-reduplication with u-prefixation in the nonhuman series. However u-prefixed numerals have not been recorded for the extinct North Formosan languages: Basai and Trobiawan—what is known of them. I am provisionally assuming that CV-reduplication and u-prefixation were competing marking strategies for the nonhuman numeral series in the common ancestor of Rukai-Tsouic and Kavalan—Proto-Central-Walu-Siwaish in the present model, this situation giving rise to two independent displacements of CV-marking by u-marking, in Rukai-Tsouic and in Kavalan. *u-prefixed numerals were abandoned in Puluqish.

Whether or not Bunun, Rukai-Tsouic’s partner in Central Walu-Siwaish, participated in this innovation cannot be ascertained, as the Bunun languages have lost the distinction between serial and non-human numerals.

Lexically, Rukai-Tsouic is supported by at least eleven uniquely shared innovations.

  • Ten can be drawn from Tsuchida (1976:11-12): *ɬa ‘and’,  *S16i ‘because’, *tukuɬu ‘heart’ (anatomical), *Dakəraɬə ‘river’, *CaŋəRaɬə ‘star’, *-bali ‘to smell’, *kəku ‘leg’, *S16iqipi ‘shoulder’,  *nətənə or *tətənə ‘lungs’, *qaputu ‘hammer’. In the same list, Tsuchida proposed Rukai-Tsouic *ramuCu ‘hand’: but this word means ‘finger’ in Rukai. We are dealing with a Tsouic innovation changing *ramuCu from ‘finger’ to ‘hand’. I have not retained *ŋuRuq2u ‘nose’ which seems problematic because the only cited Tsouic form means ‘nasal mucus’.
  • *cáni ‘one’, given by Tsuchida as Tsouic-only, but observed by myself (2013, here, § 3) as a component in ‘ten’ in an extinct Rukai variety recorded by Ogawa, (Li and Toyoshima 2006 language #63) : ucani pulu. This is the supporting evidence for Proto-Rukai-Tsouic *cáni ‘one’: Tsou coni (‘1’, serial and non-human), Kanakanabu cani (serial), u-cani (non-human), Saaroa caani (serial), u-cani (non-human), Rukai #63 u-tsani pulu ‘10’ (one-ten; pulu ultimately goes back to a loan from Paiwan). It is possible that the PRT form was *u-cáni, belonging to the non-human counting series; if so Tsou coni, Kanakanabu cani, Saaroa caani, all ‘one’ (serial counting) are analogical creations of proto-Tsouic. Tsou does not have an u- prefixed form because it does not distinguish between a serial and a non-human series in counting. See this post for the etymology of PRT *cáni ‘one’.

In Rukai *u-cani coexists with inherited reflexes of *isa/*esa ‘one’: thus Rukai #63 has isa ‘one’ (presumabily the serial counting form) vs. u-tsani ‘one’ (presumably non-human).

in Tsouic  *isa/*esa were eliminated as a result of the generalization of *cáni.


Li, Jen-kuei and Masayuki Toyoshima (eds). 2006. comparative vocabulary of Formosan languages and dialects, by Naoyoshi Ogawa. Asian and African lexicon series 49. Institute for Languages and cultures of Asia and Africa, Tokyo University of Foreign Studies.

Sagart, L. (2013). Is Puyuma a primary branch of Austronesian ? A rejoinder. Oceanic Linguistics 52, 2: 481-492.

Tsuchida, S. (1976) Reconstruction of Proto-Tsouic phonology. Tokyo: Study of languages and cultures of Asia and Africa monograph series N° 5.

Wolff, John U. (2010) Proto-Austronesian phonology with glossary. 2 vols. Ithaca: Cornell Southeast Asia program publications.

Cite this post as: Laurent Sagart, "Rukai-Tsouic (master post)," in Sino-Tibetan-Austronesian, 18/04/2020, https://stan.hypotheses.org/719.

Tsouic (master post)

(to clickable tree)

(up one level to Rukai-Tsouic)

Tsouic is a group of three Austronesian languages spoken in south-central Taiwan: Tsou, Kanakanabu and Saaroa; plus Rukai. Within Tsouic, Kanakanabu and Saaroa form a southern branch, coordinate with Tsou.

Shared innnovations of Tsouic in sound change and morphosyntax were detailed in Sagart (2014 ) (here, sections 5.2 and 5.2, pp. 19-20).

At least 57 Tsouic-only lexical items reconstructed to Proto-Tsouic by Tsuchida (1976) are probably innovations (see the list here, section 5.3, pp. 20-21).

Proto-Tsouic *ramuCu ‘hand’, from PRT *ramuCu ‘finger’, displaced PAN *qa-lima ‘hand’. This word belongs to the most basic vocabulary: the likelihood that it was borrowed from another language, or that it is a retention from PAN is practically nonexistent.

The unusually large number of innovative Tsouic words implies a very long period of almost complete isolation in the southern part of the Formosan ridge —perhaps three thousand years—between the separation with Rukai and the breakup of proto-Tsouic.

Tsouic languages also share the metathesis of *pataS ‘tattoo, write’ to Proto-Tsouic *tapaSə (Kanakanavu tapásə, Saaroa taa-tapa-a, Tsou ta-tpos-a ‘pattern, design’).

In addition: ki-Verbs are lost; reflexes of *esa ‘one’ are eliminated from the numeral system by the generalization of the innovative form *cáni ‘one’ (jump to Rukai-Tsouic master post).


Sagart, Laurent (2014) In Defense of the Numeral-based Model of Austronesian Phylogeny, and of Tsouic. Language and Linguistics 15(6) 859–882.

Tsuchida, S. (1976) Reconstruction of Proto-Tsouic phonology. Tokyo: Study of languages and cultures of Asia and Africa monograph series N° 5.

Cite this post as: Laurent Sagart, "Tsouic (master post)," in Sino-Tibetan-Austronesian, 18/04/2020, https://stan.hypotheses.org/756.

Central Walu-Siwaish (master post)

(to clickable tree)

(Jump up one level to Walu-Siwaish)

(Jump down one level to Rukai-Tsouic)

Central Walu-Siwaish (CWS) includes Rukai-Tsouic and Bunun.  Two innovations have been identified, both of them in the numeral system: the displacement of *isa/*esa/*sa ‘one’ by *Ca-CiNi in the human counting series; and the displacement of *iCit by *ma-sa-N as ‘ten’.

The merger of PAN *H1 and *H2 in word-final position comes close to a shared innovation, being attested only in Tsouic (Saaroa) and Bunun, but as Rukai reflects both phonemes as zero, the possibility cannot be excluded that the mergers of *H1 and ¨H2  took place independently in Saaroa and Bunun.

1. Displacement of *isa/*esa/*sa ‘one’ by *Ca-CiNi in the human counting series. In Proto-CWS, a word *Ca-CiNi was innovated to serve as ‘one’ (human): Bunun tatini ‘one’ (human) Tsou cihi ‘one’ (human), Kanakanabu: ta-cini (where ta- has replaced Ca- to form a run with ta-susa ‘2’ and ta-tulu ‘three’ (all human), Saaroa: ca-ciɫi ‘1’ (human). *Ca-CiNi is a human-counting series Ca- reduplication of *CiNi, reflected in Takivatan Bunun (de Busser 2009) as tini means ‘alone, on one’s own’.

A system with two contrasting series of numerals (in addition to the serial counting series), one with Ca- reduplication for human reference and another with CV- reduplication for non-human reference, arose in Austronesian languages between the Limaish and Enemish nodes. That system is seen in Siraya, Kanakanabu, Bunun (Isbukun, here (https://mpi-lingweb.shh.mpg.de/numeral/Bunun.htm)), Puyuma; it continued into Malayo-Polynesian. One recurrent problem with the CV- vs. Ca- marking of numerals is that it failed to produce a distinction in those whose first vowel was /a/, especially ‘1’ for which Siraya had sa-saat in both series. As a result there was a tendency in Walu-Siwaish languages to innovate distinct forms of ‘1’ to be used in reference to humans and nonhumans.

2. Displacement of *iCit by ma-sa-N as ‘ten’. PAN *iCit ‘ten’ is attested in most languages of the west coast of Taiwan: Luilang, Pazeh, Favorlang-Babuza, Taokas, Papora, Hoanya. Siraya kyttian is probably *ka-iCi(t)-an, with nominalizing *-an suffix, suggesting *iCit vas originally a verb root.

The central Walu-Siwaish languages replace *iCit with *ma-sa-N as ‘ten’: Bunun (Takituduh, Li 1988) ma-cʔan; Tsou máskə, Kanakanabu ma:nə, Saaroa ma:ɬə, Proto-Tsouic *másaɬə (Tsuchida 1976). Proto-Rukai has *poɭoko ‘ten’ (Li 1977), apparently related to the Puluqish numeral for ‘ten’ (here), but this is evidently a loanword, as already pointed out by Pauk J.K. Li and E. Zeitoun, since Rukai treats PAN *-q as zero.

*ma-sa-N ‘ten’ consists of *sa, short form of *isa ‘one’, with the circumfix *ma-…-N ‘times’ (?). The glottal stop in Bunun ma-cʔan is already present in the word for ‘one’: Takituduh tasʔa. Bunun geminates the vowels of monosyllables and inserts a glottal stop between the vowel’s two copies (Wolff 2010:169). In compounds ma-cʔan and ta-sʔa, the vowel’s first copy has been lost.

Although superficially similar, Atayal (Mayrinax) maɣalpuɣ ‘ten’ (which includes lpuɣ ‘to count’) and Seediq (Paran) maxal id. have medial fricatives which cannot reflect *s.


de Busser, Rik. 2009. Towards a grammar of Takivatan Bunun, Selected Topics. PhD thesis, LaTrobe University, Bundoora, Australia.

Li, Paul Jen-kuei. 1977. “The Internal Relationships of Rukai.” In Li, Paul Jen-kuei. 2004. Selected Papers on Formosan Languages. Taipei, Taiwan: Institute of Linguistics, Academia Sinica.

Li, P. J.-K. 1988. A c omparative study of Bunun dialects. BIHP 59,2: 479-508.

Wolff, John U. 2010. Proto-Austronesian phonology with glossary. Vol. 1. Ithaca: Cornell Southeast Asia program publications.

Cite this post as: Laurent Sagart, "Central Walu-Siwaish (master post)," in Sino-Tibetan-Austronesian, 18/04/2020, https://stan.hypotheses.org/748.

Southern Austronesian (master post)

(to clickable tree)

(Jump up one level to Puluqish master post)

Within Puluqish, the Malayo-Polynesian (MP) and Kra-Dai (KD) languages form a southern clade,  consisting of all Austronesian languages spoken south of Taiwan. I use the term ‘Southern Austronesian’ (SA). Shared innovations cannot be grammatical since KD preserves morphology indirectly at best and grammar has aligned on Chinese, but they can be phonological or lexical.

phonological innovation

Puluqish linker *atu ‘and’ reduced to *at after *sa-puluq in numbers 10 to 19 (here)

Lexical innovations (here)

Cite this post as: Laurent Sagart, "Southern Austronesian (master post)," in Sino-Tibetan-Austronesian, 13/04/2020, https://stan.hypotheses.org/679.

iwa vs. siwa ‘nine’ in Puyuma

(jump up to Puluqish)

The Walu-Siwaish word for ‘nine’ is reconstructed as *Siwa with capital *S. This is expected to give iwa in Puyuma, where *S evolves to zero. As recorded by Li and Tsuchida (here), Nanwang Puyuma indeed has iwa, but Thomson 1875 gives siva for the ‘tribe at Pilam’, which is 卑南, modern Beinan, Nanwang Puyuma.  Cauquelin’s dictionary of the same dialect gives both iwa and siwa, the latter  used in serial counting. Puyuma initial s- reflects PAN (and PWS) *s.

In an insightful paper, Blust (2001) stated that Formosan languages reflect *Siwa while Malayo-Polynesian languages reflect *siwa. He argued that the irregular shift of *S to *s in ‘nine’ in PMP was motivated by the proximity of *sa-puluq ‘ten’, which begins with *s-.  The existence of siwa in Puyuma shows that the shift from *S to *s in ‘nine’ is older than PMP. If Blust’s analogical explanation is correct, then Puyuma, or an ancestor of Puyuma, must once have had *sa-puluq  for ‘ten’. This is precisely what the etymology I have recently presented for *puluq ‘ten’ implies.


Blust, R. 2001. Malayo-Polynesian: new stones in the wall. Oceanic Linguistics 40, 1: 151-155.

Cauquelin, J. 2015. Nanwang Puyuma-English dictionary. Institute of Linguistics, Academia Sinica, Taipei, Taiwan.

Thomson, J. 1875. The straits of Malacca, Indo-China, China or ten years’ travel, adventures and residence abroad. London: Sampson Low, Marston, Low and Searle.

Cite this post as: Laurent Sagart, "iwa vs. siwa ‘nine’ in Puyuma," in Sino-Tibetan-Austronesian, 11/04/2020, https://stan.hypotheses.org/664.

Puluqish (master post)

(To clickable tree)

(Jump up one level to Eastern Walu-Siwaish)

(Jump down one level to northern Puluqish; to southern Austronesian)

Puluqish (Sagart 2008) is an Austronesian subgroup defined by the innovative numeral *puluq ‘ten’. It includes three languages of southern and southeastern Taiwan: Amis, Puyuma, and Paiwan, plus Malayo-Polynesian and Kra-Dai.

A second innovation of Puluqish has now been identified: the use of prefixes *paka- and *maka- to indicate abilitative meaning (here).

Critics of Puluqish had underlined the lack of an etymology for *puluq (e.g. Winter 2010: 284): this apparent opacity seemed to indicate that *puluq is as ancient as the lower Austronesian numerals *isa ‘one’, *duSa ‘two’ etc.  The etymology of *puluq has now been determined (here).

The new etymology implies that the original Puluqish word for ’10’ was actually *sa-puluq (one-puluq) rather than just *puluq. The *sa- component was lost in Amis and Puyuma but preserved in Paiwan, Malayo-Polynesian and Kra-Dai. The irregular phonetic shape of one word for ‘nine’ in Puyuma confirms the loss (here).

The internal structure of Puluqish has been clarified: lexical evidence that Amis and Puyuma form a subgroup within Puluqish has been described (here).

Similarly, lexical evidence has been presented (here) that Malayo-Polynesian and Kra-Dai form a subgroup: Southern Austronesian (SA).

Puluqish clashes with Blust’s East Formosan (Blust 1999) and with Ross’s Nuclear Austronesian (Ross 2009; now abandoned by him). See my criticisms of these constructs here and here.


Blust, Robert A. 1999. Subgrouping, circularity and extinction: some issues in Austronesian comparative linguistics. In: Elizabeth Zeitoun and Paul Jen-kuei Li (eds.) Selected Papers from the Eighth International Conference on Austronesian linguistics, 31-94. Taipei: Institute of Linguistics (preparatory office).

Ross, Malcolm D. 2009. Proto Austronesian verbal morphology: a reappraisal. Austronesian historical linguistics and culture history. A festschrift for Robert Blust, ed. by Alexander Adelaar and Andrew K. Pawley (eds), 285-316. Canberra: Pacific Linguistics.

Sagart, L. (2008) the expansion of setaria farmers in East Asia: a linguistic and archaeological model. In Sanchez-Mazas A, Blench R, Ross M, Peiros I, Lin M, eds. Past human migrations in East Asia: matching archaeology, linguistics and genetics, 133-157. Routledge Studies in the Early History of Asia, London: Routledge.

Sagart, L. 2014. In Defense of the Numeral-based Model of Austronesian Phylogeny, and of Tsouic. Language and Linguistics 15(6) 859–882.

Sagart, L. 2015. East Formosan and the PAN palatals. Paper presented at the 13th International Conference on Austronesian Linguistics, Taipei, July 18-22, 2015.

Sagart, Laurent.2014. In Defense of the Numeral-based Model of Austronesian Phylogeny, and of Tsouic. Language and Linguistics 15(6) 859–882.

Winter, Bodo. 2010. A note on the higher phylogeny of Austronesian. Oceanic Linguistics 49:282‒87.

Cite this post as: Laurent Sagart, "Puluqish (master post)," in Sino-Tibetan-Austronesian, 09/04/2020, https://stan.hypotheses.org/651.