ArticlePDF Available

The Higher Phylogeny of Austronesian and the Position of Tai-Kadai

December 2004
Oceanic Linguistics 43(2):411-444

DOI:10.1353/ol.2005.0012

Authors:

National Institute for Oriental Languages and Civilizations

This paper presents a new higher phylogeny for the Austronesian family, based on three independent lines of evidence: the observation of a hierarchy of implications among the numerals from 5 to 10 in the languages of Formosa and in PMP; the finding that the numerals *pitu '7', *walu '8', and *Siwa '9' can be derived from longer additive expressions meaning 5+2, 5+3, and 5+4, preserved in Pazeh, using only six sound changes; and the observation that the phylogeny that can be extracted from these and other innovations—mostly changes in the basic vocabulary—evinces a coherent spatial pattern, whereby an initial Austronesian settlement in NW Taiwan expanded unidirectionally counterclockwise along the coastal plain, circling the island in a millennium or so. In the proposed phylogeny, Malayo-Polynesian is a branch of Muic, a taxon that also includes NE Formosan (Kavalan plus Ketagalan). The ancestor language, Muish, is deemed to have been spoken in or near NE Formosa. Further evidence that the Tai-Kadai languages, contrary to common sense, are a subgroup of Austronesian (specifically: a branch of Muic, coordinate with PMP and NE Formosan) is presented.

The An settlement of Taiwan with the MP and TK migrations [a]: The pre-Austronesians, from NE China, expand southward along the SE China seaboard in the 5th and early 4th millennia BCE: they cultivate rice, foxtail millet, exploit marine resources, practice tooth evulsion. [b]: the Nanri and Pingtan islands, from which the top of Mt Xueshan (3884 m., at center of 200-km radius visibility circle) can be seen, 27 are reached. From there one group crosses to Taiwan c. 3500 BCE, while [c] the rest continues expanding in a SW direction towards the Pearl River Delta. [1]: location of earliest An (PAnspeaking) settlements on Taiwan. [2]: location of Pituish, [3] location of Enemish, [4] location of Walu-Siwaish, [5] location of Muish, [6] Tai-Kadai migration, [7] Malayo-Polynesian migration, c. 2000 BCE.

…

Figures - uploaded by Laurent Sagart

Content may be subject to copyright.

Content uploaded by Laurent Sagart

Content may be subject to copyright.

HAL Id: halshs-00090906

https://halshs.archives-ouvertes.fr/halshs-00090906

Submitted on 4 Sep 2006

HAL is a multi-disciplinary open access

archive for the deposit and dissemination of sci-

entic research documents, whether they are pub-

lished or not. The documents may come from

teaching and research institutions in France or

abroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, est

destinée au dépôt et à la diusion de documents

scientiques de niveau recherche, publiés ou non,

émanant des établissements d’enseignement et de

recherche français ou étrangers, des laboratoires

publics ou privés.

The higher phylogeny of Austronesian and the position

of Tai-Kadai

Laurent Sagart

To cite this version:

Laurent Sagart. The higher phylogeny of Austronesian and the position of Tai-Kadai. Oceanic Lin-

guistics, 2004, 43 (2), pp.411-444. �halshs-00090906�

THE HIGHER PHYLOGENY OF AUSTRONESIAN

AND

THE POSITION OF TAI-KADAI1

Laurent Sagart

CNRS, Paris

This is a modified version of a paper presented at the workshop on "Les premiers

austronésiens: langues, gènes, systèmes de parenté", Paris, May 5, 2004. Thanks go to

Sander Adelaar, Peter Bellwood, Bob Blust, Isabelle Bril, Alexandre François, Jeff Marck,

Estella Poloni, Lawrence Reid, Malcolm Ross, Alicia Sanchez-Mazas and John Wolff for useful

discussion.

Abstract

This paper presents a new higher phylogeny for the Austronesian family,

based on three independent lines of evidence: the observation of a hierarchy

of implications among the numerals from 5 to 10 in the languages of

Formosa and in PMP; the finding that the numerals *pitu '7', *walu '8' and

*Siwa '9' can be derived from longer additive expressions meaning 5+2, 5+3

and 5+4, preserved in Pazeh, using only six sound changes; and the

observation that the phylogeny which can be extracted from these and other

innovations -mostly changes in the basic vocabulary- evinces a coherent

spatial pattern, whereby an initial Austronesian settlement in NW Taiwan

expanded unidirectionally counterclockwise along the coastal plain, circling

the island in a millennium or so. In the proposed phylogeny, Malayo-

Polynesian is a branch of Muic, a taxon which also includes NE Formosan

(Kavalan plus Ketagalan). The ancestor language: Muish, is deemed to have

been spoken in or near NE Formosan. Further evidence that the The Tai-

Kadai languages, contrary to common sense, are a subgroup of Austronesian

(specifically: a branch of Muic, coordinate with PMP and NE Formosan) is

presented.

This paper presents a new higher phylogeny of Austronesian based on

strictly cladistic principles: each node will be supported by linguistic

innovations. This is not a new approach: the phylogeny in Starosta (1995)

was based in morphological innovations, and those in Blust (1999) and Ho

(1998) were based primarily or entirely on phonological innovations, mostly

mergers. Here I will use innovations drawn almost exclusively from changes

in the basic vocabulary. Using this methodology I will construct a tree-like

phylogeny for the higher (non-MP) part of Austronesian phylogeny, and give

further evidence for the claim, made earlier (Sagart 2001; in press, a) that the

Tai-Kadai languages are a subgroup of Austronesian. Before proceeding with

the main issue, I need to make some methodological remarks.

1. Methodological remarks

Phonological mergers are convenient features in subgrouping, because one

can be sure that they are innovations. In that respect they fit the basic neo-

grammarian requirement that subgrouping be effected on the ground of

shared innovations rather than shared retentions. At the same time,

phonological mergers are consequences of regular sound changes; regular

sound changes in turn are known to spread along social networks which

routinely cross-cut dialect boundaries, and even, through bilingual speakers,

language boundaries: witness the spread of Parisian /r/ in parts of 17th- and

18th century Europe (Trudgill 1974:162); in Taiwan, the merger of the PAn

phonemes *d and *z which has affected Basai and Kavalan but not

Trobiawan, Rukai but not Tsouic, and all other Formosan languages save

Taokas, Siraya and Favorlang: that collection of languages is not a taxon by

anyone's subgrouping: spreading by contact must have played a major role.

A long list of sound changes having spread across language boundaries,

sometimes over very large expanses of land, such as the spread of tonal

contrasts in East Asia, could be presented.

The propensity to spread over dialect or language boundaries is not a curious

idiosyncrasy of certain sound changes: it is the way regular sound change

habitually works. That is why phonological isoglosses

normally

overlap in

dialect maps, and why phylogenies constructed from phonological mergers

tend to be ambiguous and inconclusive, except of course in situations where

spreading in made impossible by the geography, as in the eastern Pacific

island world. In such regions, phonological mergers are useful evidence in

constructing linguistic phylogenies.2 In the case of language taxa having

evolved on relatively extensive land masses, like Taiwan, it is preferable to

construct phylogenies on those types of innovative characters that are least

likely to be transferred through contact. The most useful and readily

available are morphological changes and especially lexical replacements in

notions belonging to the core vocabulary: personal pronouns, numerals,

body part terms and the like.3

The main difficulty with morphological and lexical changes is how to be sure

that one is dealing with innovations. For morphology, Starosta (1995) used

the principle that innovation can be equated with complexification, but this

does not appear to be a reliable principle, as it appears that morphological

processes can be lost without leaving any lexicalized traces. With the lexicon,

it is sometimes claimed that lexical innovations cannot be identified prior to

subgrouping. This, if true, would render them useless in subgrouping. Yet

principles allowing the identification of lexical innovations prior to, rather

than as a result of, subgrouping, exist. Here are examples of situations

where this is possible:

• When a word can be shown to be phonologically reduced from an

expression consisting of several words in a related language, it is a

natural inference that the reduced form is an innovation.

• It is often the case that when two etyma compete for a certain

meaning, the more etymologically transparent of the two is the

innovation.

• when a root occurs with two meanings in different languages, and one

of the meanings is clearly the older one. See the discussion of 'moon'

in section 4.1.

• In the case of closed lexical systems, it is a fair bet that an analogically

leveled form is innovative and that its non-leveled counterpart is a

preservation.

2 I am grateful to Malcolm Ross for this important caveat.

3 I am

not

claiming that lexical innovations in the basic vocabulary do not spread at all: I am

only claiming that when we select lexical innovations from the basic part of the vocabulary,

we minimize the risk of selecting characters that have spread by contact, without eliminating

that risk.

• At times linguistic geography allows us to tell which of two competing

forms for a certain meaning is old, and which is innovative (see fn. 14).

In what follows I will make use of these principles to investigate the higher

phylogeny of Austronesian.

2. The distribution of the numerals 5-10

Nearly consensual reconstructions for the PAn numerals are *isa or *esa '1',

*duSa '2', *telu '3', *Sepat '4', *lima '5', *enem '6', *pitu '7', *walu '8', *Siwa

'9', *puluq '10'. An interesting situation can be observed with the numerals

from '5' to '10' (Table 1): throughout Taiwan, a reflex of *puluq '10' implies

the presence of a reflex of *Siwa '9',4 which implies the presence of *walu '8',

which implies the presence of *enem '6', which implies the presence of *lima

'5', which implies the presence of *pitu '7', while the reverse implications do

not hold. PMP has reflexes of all numerals from 5 to 10, in conformity with

the Formosan implicational hierarchy. In diagrammatic form:

puluq >> Siwa >> walu>> enem >> lima >> pitu

where '>>' means 'implies the presence of'. This is shown in Table 1.

The numerals 5-10 can be used as lexical criteria (or 'characters') to classify

Austronesian languages. The data in Table 1 show that they are are mutually

compatible in the sense of Meacham and Estabrook (1985): that is, they are

all compatible with the same phylogenetic tree. A natural explanation for a

distribution of the kind shown in Table 1 is that we are dealing with a

sequence of nested innovations: specifically, that PAn had neither of these

numerals, and that they arose one after the other, in succession, with *pitu

arising first, then *lima, in a language that already had *pitu; then *enem, in

a language that already had *pitu and *lima; and so forth.

4 In some varieties of Rukai, such as Oponohu (as cited in Ferrell 1969), there is a reflex of

*puluq but none of *Siwa: presumably a reflex of *Siwa was displaced by the Rukai-speciic

form vaŋatə 'nine'.

The alternative is to suppose that the PAn numerals for 5-10 were lost in

various Formosan languages, especially on the West coast and in the center,

in such a way that the languages which lost *pitu '7' were a subset of those

which lost *lima '5'; that those which lost *lima were a subset of those which

lost *enem 'six': and so forth. It is difficult to think of a reason why this

would happen. The odds of it occurring by accident do not seem high either.

The idea that the numerals under discussion are not PAn is certainly

paradoxical, considering that the consensus opinion favors the opposite

view. Yet one should keep in mind that assignment of *lima, *enem, *pitu,

*walu, *Siwa and *puluq to the PAn level crucially relies on the fact that all

are reflected in PMP, combined with the assumption that MP is a primary

branch of An (or even two primary branches, in Dyen's view). If however, as

argued by Harvey (1979, 1982), Reid (1982), Starosta (1985, 1994, 1995,

1996, 2001), Ross (1995), Benedict (1995), Bellwood (1997), Ho (1998), Ho

and Yang (1999) and Sagart (2002), the MP languages are part of a taxon

which also includes languages of the Formosan east coast, then the

testimony of PMP counts for little, and assignment to the PAn level should be

decided primarily on an etymon's distribution in the languages of Taiwan. If

so, there is no particular reason why the six numerals under consideration

should be PAn words.

Assuming that the consensus forms of the numerals 5-10 are post-PAn

innovations immediately raises the question of their etymology. There is a

chance that cues to the origin of these words can still be found in the

numeral systems of the languages where the innovative forms are not

reflected.

3. New etymologies for pitu, walu and Siwa.

What forms of the numerals 5-10 do we find in the Formosan languages that

do not have the familiar forms ? Most common above '5' are analytic forms.

In Pazeh we find additive forms (discussed in detail below): 6=5+1, 7=5+2,

8=5+3, 9=5+4; additive forms also in Saitaoyak and Tarumyan, two varieties

of Saisiat recorded by Ino Yoshinori (Ino 1998), the word for '7':

saivuseaha

made up of

saivusa

'6' and

aha

'1'. Multiplicative forms of the kind of 2x3,

2x4 are common for '6' and '8': Sediq

materu

, Thao

katuru

are based on *telu

'3' and Sediq

maspat

, Thao

kashpat

'8' are based on Sepat '4'. In addition to

katuru

and

kashpat

, Thao has longer forms

makalh-turu-turu

'6' and

maka(lh)-shpa-shpat

'8'. In Saisiat too the form for '8' is based on '4':

Tarumyan

kaspat

, Saitaoyak

makaspat

. Favorlang

maaspat

< *ma[k]aSpat '8'

agrees with Saitaoyak. Taokas

ma-hal-pat

, Siraya

kuixpa

similarly appear to

be based on '4'.5 Various languages have subtractive forms for '9': Sediq

maŋali

'nine', imperative of

maŋal

'take', apparently through 'take [one out of

ten]', as noted by Pecoraro (1977); perhaps also Saisiat-Saitaoyak

ra:ha

'nine'

(Ino 1998) which contains

aha

'one'. An interesting set consists of Thao

tanacu

, Favorlang

tannacho

, Taokas

tanaso

'9', which point to an earlier

*[st]a[nŋ]aCu. The first syllable might reflect *sa- 'one', in which case we are

perhaps dealing with a subtractive form.6

In Taiwan, the prevalence of analytic forms for numerals above five is

essentially a west coast phenomenon. This is striking, because the west coast

faces the continent and can be suspected of being the area of the earliest An

settlement and the place where the first An diversification took place.

It is not in principle impossible that analytic forms could have arisen

secondarily on the west coast, displacing earlier short forms. The

development of analytic forms based on 'five', displacing older reflexes of the

numerals 5-10, occurs in various languages of the Philippines, New

Caledonia and north Vanuatu, such as Ilongot (Reid 1971), Nêlêmwa (Bril

2002: 381-82) and Mwotlap (François 2001: 344), for instance.

Another interpretation of the facts is possible: PAn had a numeration system

with stable words for numerals up to '5', and no stable words for '6', '7', '8',

5 The alternation in these multiplicative forms of ma- and ka- prefixed forms is interesting.

Where both occur (Thao, Saisiat-Saitaoyak, Favorlang), it is always in the order ma+ka. The

final consonant in Thao makalh- unambiguously reflects *R: this is perhaps the same

formative as in Rukai ma- < *maR- 'dual' (Li 1975: 14, 74, 261), south Paiwan mag- 'dual,

plural' (Elizabeth Zeitoun, p.c., 2002). In Thao (Blust 2003:113, 115) prefix makalh- is

attested only with numerals (including makalh-tanacu a maqcin '90') but maka- derives

verbs meaning 'to resemble X, produce X, from X, in X, to X' out of nouns. It is possible that

the PAn prototypes of the multiplicative forms for '6' and '8' are retained in the long Thao

forms makalh-turu-turu '6' and maka(lh)-shpa-shpat '8'. If so, the corresponding PAn forms

would be *makaR-telu-telu '6' and *makaR-Sepat-Sepat '8'. They would be verbal in origin,

perhaps something like 'to be from N (maka-) doubled (-R)'.

6 This form could contain a verb of taking, perhaps the same etymon as in Pazeh

asu

'bring'

< *aCu. The intervening nasal could be the segment that sometimes attaches to the end of

the numeral 'one', as in Pazeh

adang

'1', Bunun (ishbukun: Imbault-Huart 1893:256)

tashang

'1': the prototype would then be *sa-ŋ-aCu 'one brought (towards ten)'.

'9'. Expressions for the corresponding notions were made up on the spot

using additive, multiplicative and subtractive strategies. The analytic forms in

the west coast languages are their fossilized descendants. The familiar

disyllabic forms of the numerals arose one after the other in successive

daughter languages of PAn, gradually displacing all the old PAn analytic

expressions.

I will argue that the latter explanation is true, by proposing new etymologies

for '7', '8' and '9'. I will take as my starting point the Pazeh numerals for 6-9,

from Li and Tsuchida (2001):

xaseb-uza

'6'

xaseb-i-dusa

'7'

xaseb-a-turu

xaseb-i-turu

'8'

xaseb-i-supat

'9'

The Pazeh word xasep means 'five' and the forms from 6 to 9 are additive:

5+1, 5+2, 5+3, 5+4. Li and Tsuchida (2001) give two alternating forms for

'eight': one with linker -a- and another with linker -i-. They give

xasep-a-

turu

a separate entry and list

xaseb-i-turu

under

xasep

'five'. In the Pazeh

texts edited by the same authors, only

xaseb-a-turu

occurs (Li and Tsuchida

2002: 49, 53). I regard

xaseb-a-turu

as the primary spoken form and

xaseb-

i-turu

as its analogically levelled variant. This will explain why the -i- form

tends to appear when numerals are elicited as part of a word list (Ferrell

1969; Lin 2000:159). Only 'eight' shows this variation. The numerals for '7'

and '9' attest only -i-. This idiosyncratic fact will prove significant.

Change of final -p to -b in

xasep

is regular (Blust 1999:326, rule I).7 Pazeh

xasep

has cognates in other West coast languages: Favorlang

achab

(drawn

from Ferrell 1969), Saisiat

a:seb

(Yeh 2000); these two reflect *RaCep, if we

suppose that, as in Pazeh, final voicing is secondary in Favorlang and Saisiat.

Taokas

hasap

(drawn from Ferrell 1960) appears to reflect *qaCep. The

numerals added to *RaCep in the Pazeh words for 6, 7, 8, 9 are the PAn

words for 1 (*esa), 2 (*duSa), 3 (*telu), 4 (*Sepat), all with regular

developments.

7 "A rule of intervocalic voicing that affects voiceless stops before a morpheme boundary but

not within a morpheme".

The full additive forms found in Pazeh are not observed elsewhere, but

shorter forms in some west coast languages are indicative of their former

existence in other dialects/branches of An:

• The final three syllables of Pazeh

xasebaturu

, the additive form for '8',

are paralleled in Luilang '8'

patulu-nai

(where -

nai

is detachable, compare

'7'

in-nai

and '9'

satulu-nai

, with

- = '1' in Luilang).

• One dialect of Siraya (Tsuchida, Yamada and Moriguchi 1991, point M2)

has

sipat

'nine': 8 given the phonetic proximity with PAn *Sepat 'four', this

is almost certainly based on an additive 5+4 form.

Let us now return to the Pazeh paradigm, leaving the word for '6' aside:

xasebidusa

'7'

xasebaturu

'8'

xasebisupat

'9'

Under the standard explanation, any resemblances between the analytic

Pazeh expressions for '7', '8', '9' and the familiar words *pitu, *walu and

*Siwa must be fortuitous. My suspicion that the standard explanation does

not account for the facts was raised by the observation that the Pazeh word

for '7' contains the sequence -

bidu

-, close to *pitu, the familiar word for '7':

x a s e b i d u s a

p i t u

That -b- in the Pazeh form is phonologically a /p/ makes the resemblance

more specific.

8 This form

sipat

'9' was obtained by Ogawa from a nonnative policeman in Kalapo. Siwa-

type forms for '9' are not common in Siraya. From another policeman in the same locality,

Ogawa recorded

ra-siwa

'9' (dialect M3 in the same book). Probably the -p- was in the

process of being lenited and final-t was weak.

Looking now at the Pazeh word for '8', we notice that it contains consonants

of the same point of articulation, and the same vowels, as *walu '8', all in the

correct sequential order, even though a syllable intervenes between them:

x a s e b a t u r u

w a l u

The -r- in Pazeh

turu

'three' reflects PAn -l-, which improves the goodness

of fit with *walu. Note also that the intervening syllable -tu- had schwa

(orthographically 'e') in proto-Austronesian: PAn *telu '3'. Suppose that

schwa fell, a -tl- cluster would result, which could then easily simplify to -l-,

given the hostility of An phonology to consonant clusters.

With '9' we also find in the Pazeh words several of the phonetic ingredients

that enter in the familiar reconstruction *Siwa, again in the right sequential

order:

x a s e b i s u p a t

S i w a

Pazeh s- reflects proto-Austronesian S-, which again improves the

comparison. Moreover, the correspondence between Pazeh /pa/ and

'consensus PAn' /wa/ is the same as in '8' (remember that in the Pazeh word

for 'eight' xasebaturu, 'b' is phonologically a p). Least satisfactory is the

match between Pazeh u and PAn i in the first syllable, but note that Pazeh u

reflects PAn schwa in this case (PAn *Sepat 'four'), and that of all PAn vowels,

schwa is the most sensitive to contextual influence.

These facts suggest that the 'consensus PAn' forms for '7', '8' and '9' arose

as reduced forms of earlier additive expressions. I will now describe a simple

evolution path leading from the latter to the former. Let us first go back to

the list of Pazeh numerals and reconstruct the PAn forms that they would

derive from, based on the accepted sound correspondences between Pazeh

and PAn (Blust 1999; Li and Tsuchida 2001), supposing that the additive

forms existed in PAn:

*RaCep-i-duSa '7'

*RaCep-a-telu '8'

*RaCep-i-Sepat '9'

I first introduce two

arbitrary modifications to this paradigm:

1. I change RaCep-i-duSa to RaCep-i-tuSa. There is independent evidence

for an old variant *tuSa of '2' (Amis

tusa

, Puyuma

towa

, Thao

tusha

) with

initial *t- instead of *d- perhaps on the analogy of *telu '3'. The arbitrariness

is in stipulating that the *tuSa variant was used in the PAn form which

underlies *pitu '7'.

2. I arbitrarily9 assign stress to the penultimate syllable of '7' and to the final

syllable in each of '8' and '9'.

In the resulting paradigm I write stressed vowels in bold type:

*RaCep-i-tuSa '7'

*RaCep-a-telu '8'

*RaCep-i-Sepat '9'

Now I submit this paradigm to six sound changes (Table 2).

There are several possible variants of this derivation: in particular, changes 4,

5 and 6 are not crucially ordered with respect to one another.

Although they are arbitrary, the changes supposed here

9 I assume PAn generally had final stress. A reason why '7' could have its penultimate vowel

stressed would be if -a in *tuSa were the ligature, which became attached to the original

word for '2', which ended in -S. This in turn could explain why in Pazeh /a/ is optional

(indeed, mostly absent) between

dusa

'2' and a following noun:

dusa daali

'2 days',

dusa

rakihan

'two children',

dusa saw

'two persons',

dusa ilas

'two months',

dusa isit

'twenty' (two

tens); compare

adang a daali

'one day',

turu a rakihan

'three children',

supad a saw

'four

people',

supaz a isit

'forty',

xaseb a saw

'five people',

isid a ilas

'ten months', etc. This is

apparently not because of a constraint on sequences of like vowels, as we find noun phrases

tula a daran

'path of an eel' (Li and Tsuchida 2002: 82).

• are natural changes: assimilations, cluster simplifications, schwa

deletions, lenitions, stress-conditioned prunings, not outrageous

changes like p > r, or l > m, or i > q;

• affect at least two forms (except for changes 1 and 6): two changes (#

3, 4) affect the entire paradigm (three forms). By definition,

ad hoc

changes would affect only one form. The relatively marked lenition -

pa- > -wa- affects two forms;

• do not change the vowels, except for schwas: this is a general

tendency of later An phonetic evolution;

• do not affect the points of articulation of the consonants.

The changes in Table 2 did not apply to the entire vocabulary. There is ample

evidence that, for instance, PAn /pa/ normally remains /pa/ in PMP, and that

the schwas of PAn are usually retained in PMP. I am arguing that these

changes took place when expressions of four or more syllables were reduced

to disyllables as a result of the 'drive to disyllabism' which has been at work

throughout early An history. Such reduction could have taken place when

long forms came to be treated prosodically as prosodic feet, rhythmically

equivalent to canonical disyllabic feet. Compression of the phonic material

into the narrow temporal confines of a rhythmic foot would have provided

the basis for the lenitions, schwa-deletions and procrustean prunings in

Table 2.

I am well aware that the sound changes I am hypothesizing lack the support

of recurring sound correspondences. That is unavoidable, since the drive to

disyllabism could only have affected a small number of expressions

simultaneously: even supposing that the changes happened before our very

eyes, the number of examples for each of the changes involved would be too

small for sound correspondences to be established anyway. What Table 2

establishes is that phonetic evolution from the long to the short forms is

possible and that it only requires the application of a small number of natural

sound changes.

At the same time, lack of support from sound correspondences leads to the

question as to whether the resemblance between long and short forms is not

accidental, especially considering that I have made use of two

ad hoc

stipulations and six arbitrary sound changes. I have not conducted a proper

probabiIistic study of this question,10 but I believe that the resemblances are

meaningful, for the following reasons.

First, there exists independent evidence for change #4 Prune left:

• Thompson (1873) records some words from a now-extinct variety of

Pazeh. The numerals from 6 to 9, cited here as reproduced by Imbault-

Huart (1893:319) are

boudah

'6',

bidousut

'7',

bitouro

'8',

bissoupat

'9':

leaving aside the curious word for '7', these are clearly left-pruned

outcomes of the long forms in Li and Tsuchida (2001).

• Luilang

patulu-nai

'8' (Luilang -

nai

is detachable, as already noted)

appears to be the left-pruned outcome of *RaCep-a-telu.

• Siraya-Makatao

sipat

'9' (point M2 in Tsuchida, Yamada and Moriguchi

1991), which contains *Sepat 'four' in a slightly altered form must be

the left-pruned outcome of an earlier additive expression, although in

this case there is no evidence that the pruned-off part was *RaCep.

Second, the irregularity in Amis

falu

'8', earlier *balu, instead of expected

walu, can now be explained: the lenition of -pa- in *RaCepatelu went

through βa-, which was reinterpreted word-initially as ba- in pre-Amis. At

the same time, -pa- in '9' was fully lenified to -wa (Amis

siwa

'9'), perhaps

because it was in intervocalic position, a facilitating context for a lenition.

Third, the present interpretation explains an interesting detail in the

distribution of forms across Formosan languages: no language has at the

same time a reflex of *pitu and a transparent additive from based on *RaCep

'5' for another numeral. This is because a language that has a reflex of *pitu

is at least a stage-5 language in terms of Table 2: at stage 5, additive forms

based on *RaCep have already been reduced to disyllables.

Above all, the etymologies proposed here relate the familiar *pitu, *walu,

*Siwa paradigm to another paradigm: the Pazeh additive forms for '7, '8' and

10 I have counted the number of changes required to convert the same three putative PAn

analytic forms into the same three familiar numerals, only

paired differently

(7>8, 8>9,

9>7): I have found that at least 10 changes, some strange or unmotivated, are required,

most applying to only one form.

'9'; rather than to forms unrelated to each other, reflected in different

languages.

The alternative requires one to accept that:

• the resemblances between the long and short forms for 7, 8 and 9 are

accidental;

• /f-/ in Amis

falu

'8' is irregular;

• the fact that An languages reflecting *pitu can have multiplicative and

subtractive forms, but not additive forms based on *RaCep, is

accidental;

• the fact that the three etymologies proposed for '7', '8' and '9' form a

paradigm in Pazeh is accidental.

Table 1 shows that (except for some varieties of Rukai, see fn. 4) *walu and

Siwa occur in the same languages. Before we examine the question in more

detail, let us first pause to consider what the PAn numeral system may have

been. The numerals for '1' to '4' were *esa, *duSa~tuSa, *telu, *Sepat; '5' was

*RaCep (the 'standard PAn' form *lima is transparently from 'hand', while

*RaCep is opaque; only *RaCep occurs in the additive forms: so *lima must

have displaced *RaCep in early post-PAn times; see below). There is the

possibility that final -a in *esa and *duSa~tuSa is a captured ligature. There

was also a word for '10': Pazeh

isit

, Luilang

isit

, Favorlang

tsxiet

, Taokas

(ta)isid

, Papora

metsi

, Hoanya

(miata)isi

. An approximation of the PAn form is

#(sa)-iCit, where sa- = '1'. That word was later displaced by *puluq, as we

will see. Between '5' and '10' there were no stable numerals. Table 3

describes the analytic forms found in Formosan languages for numerals 6-9.

Additive and multiplicative strategies generally appear to have been favored

over subtractive ones for '6' to '8', but subtractive formations are more

common for '9': all numbers were open to additive formation, and both even

numbers to multiplicative formation. With additive expressions, only those

involving the highest lexicalized numeral (*RaCep 'five') were in use (to the

exception of Saisiat

saivuseaha

, from saivusa '6' plus

aha

'1', where the origin

saivusa

is problematic).

Let us now return to the question of why *walu and *Siwa appear only in

Formosan languages that have *pitu. The shift from the PAn numeration

system to the post-PAn system was effected, I have proposed, through the

reduction to disyllables of full PAn additive forms, as shown in Table 2: but

while the old additive forms disappeared even as they were being

phonetically reduced, this did not eliminate multiplicative and subtractive

competitors for each numeral. Since *pitu appears at stage 5, the common

ancestor of all the languages which show *pitu must be the stage-5

language. By Table 2, in that language, one had innovative forms *pitu,

*watlu and *Siwa for 7-8-9, but competing multiplicative and subtractive

forms were still available for '8' and '9'. In contrast, for '7', there were no

multiplicative and subtractive competitors. For this reason, all the daughters

of the stage-5 language have a reflex of *pitu, but only some of them show

*walu and *Siwa. Those which do not (Atayal, Sediq, Thao, Favorlang, Taokas

and Siraya) all have a multiplicative form for '8', and either a subtractive

form or a form of unknown origin for '9'. One last consideration is that the

combination in one language of a multiplicative form for 'eight' and an

additive form for 'nine' is never observed, probably because both forms

would then end in a reflex of *Sepat 'four'. On the other hand, the absence of

languages showing simultaneously the additive form for 'eight' and a

subtractive form for 'nine' is regarded as coincidental.

The three innovations discussed above have qualities that make them ideal

subgrouping criteria:

• their etymology is known and direction of the innovation is certain (in

contrast, the etymologies of *enem '6' and *puluq '10' are not known);

• the risk that they might have spread by contact is minimized by the

fact that they belong to the (relatively) basic vocabulary;11

11 Although borrowing of numerals is rare in Indo-European languages, numerals, especially

the higher numerals, are not immune to borrowing. This is true especially in east Asia, where

sets of Chinese numerals have been borrowed by Thai, Be, Miao-Yao, Bai, Baonan among

others; where Chamorro has borrowed the entire set of Spanish numerals (Topping 1973:

166); where Qau Gelao has borrowed its numerals from 2 to 10 from Yi (Edmondson and

Thurgood 1992) etc. In many (though not in all) cases, the lending language is a state

language and the borrowing language has been subjected to the pressure of the lending

language within the confines of that state, in the context of a monetary economy. That the

power of a state is often behind the transmission of numerals by contact suggests that

• the risk that each of these innovations was made several times

independently is almost nonexistent, given the complexity of the six-

stage process in Table 2 and the fact that the resulting cognate sets

obey the habitual An sound correspondences (in contrast, the *lima '5'

< 'hand' innovation consists of a mere semantic shift: it could have

occurred several times independently).

I will therefore use the *pitu, *walu and *Siwa innovations as the backbone of

a new higher Austronesian phylogeny, shown in Figure 1.

I call PAn the language spoken by the first neolithic settlers of Taiwan, from

the moment they set foot on the island to the moment when their language

broke up into two or more dialects some generations later. I call 'Pituish' the

hypothetical daughter language of PAn in which *pitu was first innovated

and, having no multiplicative or subtractive competitors, became the sole

word for 'seven'. In Table 2 this corresponds to the stage-5 language. Pituish

is ancestral to all An languages except Pazeh, Saisiat and Luilang. I assume

Pituish already had *Siwa for '9' and a form which I take to have been [watlu]

for '8'; but these two forms are not expressed by Pituish's descendants

Atayalic, Favorlang, Thao, Taokas and Siraya because of competing

multiplicative forms for 'eight' and subtractive forms for 'nine'. In another of

its descendants, however, watlu had become walu, and it and Siwa eliminated

their multiplicative and subtractive competitors, thereby becoming the only

words for '8' and '9'. I call the language where this occurred 'Walu-Siwaish'. I

regard this language as ancestral to all An languages except Pazeh, Saisiat,

Luilang, Atayalic, Favorlang, Thao, Taokas, Siraya, as well as Papora and

Hoanya. Although these two appear to have reflexes of *walu and *Siwa, I

suspect they are loans from southern Tsouic, as they show the change w >

zero, a regular development in Kanakanabu and Saaroa (compare

Kanakanabu

(h)a:ru

'eight',

siya

'nine'). Note also initial h- in the Kanakanabu

word for '8', reflected in Papora

ma-hal

. Taokas

mahalpat

looks like a blend

of a multiplicative form and Papora

ma-hal

borrowing of numerals would perhaps not have been very frequent in the early neolithic

communities which spoke PAn.

Mention must be made of the situation in Rukai. Rukai has a reflex of *walu

eight' but has

baŋatǝ

for 'nine', an isolated innovative form of unknown

origin. Since Rukai has been shown to belong to Rukai-Tsouic by Tsuchida

(1976) and since all Tsouic languages have reflexes of *Siwa, it is likely that

baŋatǝ

has displaced a reflex of *Siwa as a Rukai-specific innovation.

That PMP shares the *pitu, *walu and *Siwa innovations with other *Walu-

Siwaic languages, as shown in Figure 1, indicates that PMP is not a primary

branch of PAn, but part of Walu-Siwaic, a branch of Pituic. This is in

agreement with Harvey (1979, 1982), Reid (1982), Starosta (1985, 1994,

1995, 1996, 2001), Ross (1995), Benedict (1995), Bellwood (1997), Ho

(1998), Ho and Yang (1999) and Sagart (2002).

4. Enriching the phylogeny

I will now enrich the phylogeny in Figure 1 with additional lexical and

morphological characters that are compatible with it. I will also show how the

principal sound changes that have affected Formosan languages are to be

understood in the present framework. Before I proceed, I need to state that I

accept Tsuchida's proposal that Rukai and the three Tsouic languages (Tsou,

Kanakanabu, Saaroa) subgroup together on the basis of his documentation,

which includes several uniquely shared lexical innovations for basic

meanings, notably 'leg', 'nose', hand', shoulder', 'star' and 'river' (Tsuchida

1976:11-12). Atayalic, a taxon consisting of Atayal and Sediq, is self-evident

and has been silently accepted in Figure 1. Likewise, NE Formosan, a taxon

consisting of Kavalan and Ketagalan is accepted on the basis of the

documentation in Blust (1999), who cites a uniquely shared irregular

dissimilation in *susu 'breast' > sisu, and in Li (2001) who cites uniquely

shared innovations for the meanings 'tooth', 'eyelash', 'spider' and 'unhusked

grain/cereal'.

4.1. Fitting more lexical characters into the phylogeny

I now turn to some well-known cognate sets which are often assigned to the

PAn level. I will claim that they are post-PAn innovations. Although their

etymologies are not all known, they all have an 'opposite number' (by this I

mean another Formosan etymon of the same meaning) which I take to be the

PAn word they have displaced. This opposite number is represented only in

the higher regions of my phylogeny, while the innovative word is

represented in the lower regions (including, in most cases, PMP), with a cut-

off point somewhere in between. This distribution forms the basic ground on

which the innovation is recognized. Some vagueness in the cut-of point, or

overlap in the distributions of the competing forms (apparent resurgence of

the old form below the cut-off point) is tolerated: it is recognized that

displacement of a lexical item by another is normally effected through a

phase during which both words are competing in the language, so that

daughters of that language may randomly reflect one or the other, and a

degree of overlap between the distributions of the two etyma in a

phylogenetic tree may result.

A first group of additional compatible innovative characters is furnished by

the other numerals in Table 1, that is, by '5', '6' and '10'.

• *lima 'five'. The 'consensus PAn' word for '5' is *lima, which is also the PAn

word for 'hand'. Here the older meaning is presumably 'hand'. This in

itself is an indication that *lima is innovative as '5'. A second indication of

the innovative character of *lima is the fact that *RaCep, never *lima, is

used in the old additive expressions for '6', '7', '8' and '9'; moreover, in

those languages that have additive expressions based on '5' for any of '6',

'7', '8' and '9', the word for '5' is never *lima. This suggests that *lima did

not mean '5' at the time these expressions were created. That the *lima

innovation took place later than PAn is shown by the fact that its opposite

number *RaCep is (1) etymologically opaque, as befits an inherited word,

and (2) is distributed only in the upper region of my phylogeny:

specifically in Pazeh, Saisiat and Taokas (Luilang has an obscure form:

(na)lup). To the exception of Taokas and Favorlang, the innovative form

*lima is universal in Pituic as '5'. One explanation is that the descendants

of Pituish included on the one hand the ancestor(s) of Taokas and

Favorlang, and on the other hand Limaish, where *lima was innovated:

Limaish then broke up into Atayalic, Thao, Siraya, Hoanya, Papora and

Walu-Siwaish. Alternatively, *lima already had the meaning '5' in Pituish;

but Pituish had not eliminated *RaCep, so that there were competing

forms for 'five'; of its daughter languages, Taokas and Favorlang

eliminated *lima, while the others eliminated *RaCep. Either explanation is

compatible with the tree in Figure 1.

• *enem 'six'. This is the PAn form habitually reconstructed in this meaning.

Benedict (1995:400) has gathered intriguing evidence that this etymon

was phonetically more complex. He reconstructed *ʔəmləm (based on

forms like Makatao

ulum

and Bunun-Ishbukun

ʔabnum

'six'. Benedict's *l-

is equivalent to PAn *N). The etymology of this word is not known but

there are good reasons to suppose that it too is a post-PAn innovation: as

shown above, some of the highest languages in our phylogeny have

additive or multiplicative forms for 'six': Pazeh

xaseb-uza

(5+1), Sediq

ma-teru

, Thao

ka-turu

, etc.: the rest have forms of obscure origin:

Luilang

na-tsulup

, Favorlang

nataap

etc. All other Formosan languages,

including Siraya, together with all east coast languages and PMP, show

reflexes of *enem. In our phylogeny this translates as an innovation from

an unknown source, taking place in a language ('Enemish') ancestral to

Siraya and Walu-Siwaish. A variant story has *enem arising in Pituish and

coexisting with older additive or multiplicative forms; being eliminated in

most daughters of Pituish, but eliminating the older forms in Siraya and

Walu-Siwaish. However, as we shall see below ('year'), another lexical

innovation supports an Enemish node, and I shall here tentatively accept

the Enemish node story.

• *puluq 'ten'. This is the 'consensus PAn' word for '10'. In Formosan

languages, it occurs exclusively in Rukai (some varieties), Paiwan, Puyuma

and Amis, all of which are Walu-Siwaic languages, and outside of Formosa

in PMP. Its opposite number is the entirely pre-Walu-Siwaic set given

above as #sa-iCit, which must reflect the PAn form. Saisiat has

ranpon

, an

isolated form and apparently a local innovation. However the situation is

more complex than this simplified account would suggest, as a third

cognate set is reflected in Sediq

mahal

, Tsou

máskə

, Kanakanabu

ma:nə

Saaroa

ma:ɬə

and Bunun (Ishbukun)

masʔan

. An approximation for this set

is #masehaN.12 This etymon appears to occupy an intermediate position

between the other two in our phylogeny. I propose the following account:

the PAn word was #(sa-)iCit; Pituish innovated #masehaN which competed

with #(sa-)iCit, and finally displaced it in Atayalic and in Walu-Siwaish; all

other Pituish languages which have not otherwise innovated retain

*(sa-)iCit. Walu-Siwaish then innovated *puluq which competed with

#masehaN, with the result that Bunun and the Tsouic languages reflect

12 Tsuchida (1976) reconstructed 'Proto-South-Formosan' *masʔaL '10'

#masehaN, while the rest of Walu-Siwaic reflects *puluq —barring ulterior

innovations of the kind of Kavalan

betin

and Ketagalan

labatan

, of

course—.

The other lexical innovation which supports an Enemic taxon is the following:

• *CawiN 'year'. The PAn word for 'year' was given as *kawaS by Blust

(1999). This is undoubtedly correct, as reflexes of *kawaS occur

exclusively in the higher regions of our phylogeny: Pazeh, Saisiat, Atayalic,

Thao. Another form for the same meaning: *CawiN has reflexes in Rukai,

Kanakanabu, Saaroa, Paiwan and Bunun (Tsuchida (1976:145), to which

Sander Adelaar (p.c. 1999) adds Siraya

tawil

‘agricultural season, year’. In

terms of my phylogeny *CawiN displaced *kawaS as 'year' in a language

ancestral to Siraya and Walu-Siwaish, in other words, in Enemish. An

Enemish *CawiN would normally give *tawin in PMP, and this appears to

be the form reflected in some Central Cordilleran languages and in

Ilokano:

tawen

(Lawrence Reid, p.c., August 2, 2004), although a

competing form *taqun is more widespread as 'year' in Malayo-

Polynesian.

I will now add other lexical innovations from other areas of the basic lexicon.

First I will discuss two lexical innovations supporting a sub-taxon of Walu-

Siwaic which I call 'Muic', based on one of its innovations. Muic consists of NE

Formosan (Ketagalan, Kavalan), PMP, and a language I call FATK (discussed in

section 5). The sharing of these items by Ketagalan (though not by Kavalan)

and PMP has been noticed by Paul Li (1995) who misunderstood their

innovative character and took them as evidence that Ketagalan migrated to

Taiwan around 2000 years ago.13

• *-mu '2sg-genitive'. Blust (1977) has argued that a politeness shift

replacing the PAn 2sg-genitive pronoun *-Su with the former 2pl-genitive

*-mu is a characteristic innovation of PMP. In a more recent paper (1995)

he acknowledged that -Su did not disappear as 'your' (sg.) in MP

languages, but maintained that -mu is a MP innovation. The coexistence

in MP languages of reflexes of -Su and -mu as 2sg-genitive pronouns

13 In a more recent paper, Li (2001) apparently abandoned the idea that the Ketagalan

migrated to Taiwan in a separate migration. His new views are close to those in Blust (1999),

although he does not say so himself.

probably means that both existed side by side in PMP: presumably -mu

was a polite form. Paul Li (1995) citing an unpublished text recorded by

Asai, has pointed out that Trobiawan, a Ketagalan language of north

Taiwan, shows that very innovation (

tama imu

'your father'). It is unclear

from Li's account whether Trobiawan

imu

is a polite form and whether

Trobiawan also reflects -Su as 2sg-genitive, but Basai (the other

Ketagalan language) has

isu

for 'your' (sg.) (Tsuchida, Yamada and

Moriguchi 1991:257). This indicates that proto-Ketagalan had variation

between

isu

and

imu

for 'your' (sg.), with

imu

presumably the polite

variant. The appearance of -mu as polite 2sg-genitive must be regarded

as an innovation of the common ancestor of Ketagalan and MP. Since, as

mentioned earlier, Kavalan and Ketagalan form a taxon ('North-East

Formosan'), and since that taxon must be part of Muic, the question arises

as to why Kavalan shows a reflex of -Su, not -mu, as 2sg-genitive. It is

possible that Kavalan eliminated the polite pronoun in favor of the non-

polite form.

• *manuk 'bird' is reflected in PMP *manuk and Ketagalan

manuk(ə)

(Basai),

manukka

(Trobiawan), but in no other language of Taiwan; its opposite

number *qayam 'bird' is widespread in Formosa (including in Kavalan

where 'bird' =

alam

). Yet *manuk did not displace *qayam, which is

reflected in MP languages in some cases still as 'bird', but more often as

'domesticated animal'. My understanding is that *qayam was the PAn word

for 'bird', including the meanings 'wild bird' and 'fowl, domesticated bird';

that *manuk first arose in Muish, from an unknown source, as a hyponym

of *qayam meaning specifically 'wild bird': *manuk and *qayam then

coexisted in Muish and PMP as 'wild bird' and 'domesticated bird'

respectively. Kavalan, a Muic language, abandoned *manuk, keeping

*qayam as the only word for 'bird'. Later, in some WMP languages, *qayam

expanded its meaning to 'domesticated animal' in general, leaving

*manuk free to shift to 'domesticated fowl, chicken', or not.

Together, these two items ('2sg-genitive', 'bird') argue for the existence of a

language ancestral to Ketagalan and MP, which I call Muish. Note that neither

of these innovations displaced an earlier form of the exact same meaning:

the -mu innovation resulted in the addition of a polite pronoun where none

existed, and the manuk innovation in the creation of a hyponym of the

general term for 'bird'. Both *-Su and *qayam continued existing side by side

with -mu and manuk in the Muish proto-language. Unlike displacing

innovations, which are irreversible, these two could be reversed. Judging

from the meager evidence at hand, Kavalan seems to have abandoned both

innovative forms, keeping only reflexes of *-Su and *qayam, thereby

reversing the -mu and manuk innovations. But we should remember that

absence of evidence is not evidence of absence. More detailed descriptions of

moribund Kavalan are needed to establish this point.

The main lexical innovations and the phylogeny they support are shown in

Table 9.

I will now describe another lexical innovation whose discontinuous

geographical distribution across Formosan languages apparently challenges

my phylogeny.

• *bulaN 'moon'. There are two competing items in Formosa for 'moon':

*qiNaS and *bulaN. *qiNaS is limited to Formosa but *bulaN is regularly

reflected in PMP *bulan 'moon'. Of the two, *qiNaS is clearly the older,

being the only word reflected above Pituish: Pazeh

ilas

'moon, month',

Saisiat

ʔilaŝ

'moon'. In Pituic, it is also reflected in Sediq

idas

'moon,

month', and in Favorlang

idas

'moon': but interestingly, it also occurs in

the Walu-Siwaic language Paiwan:

qilas

'moon' (Puyuma/Katipul

qilas

cited

by Ferrell 1969:94 is clearly a loan from Paiwan). The other etymon:

*bulaN, is reflected in some Pituic languages: Thao

furaz

(Blust 2003),

Papora (Hajyovan and Vudol)

voda

(Ino 1998), perhaps also Hoanya

(Taorakmun)

pu:loa

(Ino 1998), and in Siraya, Tsouic, Bunun, Amis,

Kavalan and Ketagalan. The peripheral and discontinuous geographical

distribution of *qiNaS in the northwest and south of the island, in contrast

to the compact and more central distribution of bulaN is interesting

because on the one hand it forcefully argues that *qiNaS is the old form,14

14 J. Gilliéron was to my knowledge the first scholar to make use of the principle that a

feature found in at least two distinct zones at the periphery of a linguistic area is older than

the feature which occurs between them. The principle was explicitly formulated by Dauzat

(1922), overgeneralized by Bartoli (1925), endorsed by Bloomfield (1935) and more recently

by Chambers and Trudgill (1998:94) In Bloomfield's words:

Especially when a feature appears in detached districts that are separated by a

compact area in which a competing feature is spoken, the map can usually be

while on the other hand the discontinuity itself calls for an explanation. I

propose the following. The PAn word for 'moon' was *qiNaS. It also meant

'month'. In Pituish *bulaN was innovated as 'moon'. It coexisted with

*qiNaS, the basis for their coexistence being that *bulaN meant 'moon' not

'month'. This coexistence lasted through Enemish and Walu-Siwaish, until

Muish where *bulaN finally eliminated *qiNaS. Later on, among those

pituic, enemic and walu-Siwaic languages which had both, there was a

tendency for *bulaN 'moon' to generalize its meaning to 'month' and thus

compete with *qiNaS, displacing it in central Taiwan, and separating

Paiwan from the western and northwestern languages.

We have collateral evidence to reinforce the presumption that *bulaN is the

innovation. In his dictionary, Buck (1949:54) showed that those Indo-

European languages which do not show reflexes of PIE *mēnes 'moon' in the

meaning 'moon' have independently replaced the inherited PIE word with

other words, "most of them from the notion of 'brightness'": thus Gk

σελήνη

from

σέλας

'light, brightness', Irish

gealach

'moon', also 'brightness', from

geal

'bright, white' etc. Further on (1949:1054) he states that "most of the

words for 'white' come from the notion of 'bright'". Thus 'moon' and 'white'

are normal semantic outcomes for words meaning 'bright'. Reversing Buck's

observations, we may state that when a language has identical forms for

'moon' and 'white', there is a presumption that they have a common source in

an older word meaning 'bright/brightness'. PMP had *bulan 'moon' and

*bulan 'white' (Blust, ACD). We may therefore presume that these two words

have a common source in an older word for 'bright(ness)', even though such

a word has not been reconstructed.15 This, then, supports the view that

*bulaN 'moon' is a post-PAn innovation, and that *qiNaS is the PAn word for

'moon'.

In this section I have enriched the simple phylogeny in Figure 1 with

additional lexical characters and dealt with an apparent piece of lexical

interpreted to mean that the detached [districts] were once part of a solid area. In

this way dialect geography may show us the stratification of linguistic features.

(Bloomfield 1935: 340)

15 Note Saisiat

bolalas

'white' (Ferrell 1969).

counterevidence. I will now examine whether any morphological innovations

can be fitted into the model.

4.2. Fitting morphological innovations in the model

Six of the eight morphological characters discussed in Starosta (2001) are

found both in Saisiat and/or Pazeh and elsewhere in Taiwan. They must be

PAn features by our phylogeny:

• Ca- verbal reduplication for deriving instrumental nouns (examples see

Blust 1998), seen in Pazeh, Saisiat, Thao, Siraya, Paiwan, Puyuma,

Amis, and MP. Starosta assumes it is a post-PAn innovation because of

its absence in Rukai and in the Tsouic languages.

• Ca- verbal reduplication marking non-completive aspect (Pazeh, Thao,

Atayalic, Bunun, Tsouic, Rukai, Puyuma, Amis, MP). To these, add

Siraya (Sander Adelaar, p.c. August 12, 2004), Kavalan (Chang Yung-li

2000:59). Starosta points out that under Blust's flat tree, this

distribution must be regarded as the result of independent innovations

or of independent losses. In the present phylogeny it is simply a PAn

process, lost in Paiwan.

• CV- verbal reduplication marking future or imperfect (Pazeh, Tsou,

MP).

• Sa- prefixation marking 'instrumental focus' (Pazeh, Rukai, Amis).

• Sa- prefixation deriving instrumental nouns out of verbs (Pazeh, Rukai,

Amis).

• Si- prefixation marking 'instrumental focus' (Pazeh, Saisiat, Bunun,

Paiwan, MP). In Starosta's phylogeny this is a post-PAn innovation, and

(unlike me) he does not have to explain the absence of this process in

Rukai-Tsouic languages.

Fitting these onto the tree in Table 4 is only a matter of assuming the

requisite extinctions. For instance with the first item in the list (Ca- verbal

reduplication for deriving instrumental nouns), I have to suppose that this

PAn process was lost once in proto-Rukai-Tsouic and once in proto-Atayalic,

and that any lexical traces it might have left there were later displaced by

other forms.

Only two of the processes discussed by Starosta can be post-PAn innovations

in my phylogeny, and are therefore potentially informative for early An

subgrouping:

• a-prefixation marking future in verbs (Thao, Tsouic, Rukai, Amis).

• paŋ-prefixation deriving instrumental nouns out of verbs (Amis, MP).

The first of these two is not seen in Saisiat or Pazeh, though its absence in

these languages could be the result of a loss or of incomplete description. If

neither possibility applies, the process can be an innovation of Pituish. In any

case we need to suppose several independent events of loss of this feature at

later times.

The second process is most likely a post-PAn innovation, as its absence in all

of Pazeh, Saisiat, Thao, Atayalic, Favorlang, Siraya, Paiwan and Rukai-Tsouic

can hardly be fortuitous. In Starosta's phylogeny as in mine, it is meaningful

that Amis, an east coast language, is the only Formosan language to show

this feature. In Blust's, it is coincidental. In the phylogeny in Table 4, the

paŋ- deverbal instrumental derivation has to be an innovation of Walu-

Siwaish; unless independent evidence for a lower-level taxon including Amis

and Muic appears.

4.3. Fitting phonological innovations into the model

In the present proposal, for reasons explained above, the suspicion that the

sound changes which have formed the basis of several previous attempts at

classifying An languages might have spread by contact has relegated them to

a secondary place for classification purposes. Their contribution to our

understanding of early An phylogeny is limited, because of the risk that they

might have spread to already individualized languages. In this section I will

show that the principal sound changes having affected Formosan languages

are better explained as areal events than as phylogeny-defining events. I will

discuss five important mergers:

1. the merger of PAn *C into *t in Ketagalan, Kavalan, Amis, Siraya, Bunun

and PMP.

2. the merger of PAn *j into *n in Ketagalan, Kavalan, Amis and Siraya;

3. the merger of PAn *N into *n in Ketagalan, Kavalan, Kanakanabu,

Bunun and PMP

4. the merger of PAn *ŋ into *n in Favorlang, Papora, Taokas and Thao.

5. the merger of the PAn phoneme called S2 in the tradition of Dahl and

Ho with S1 in Amis, Bunun, Puyuma, Kavalan and PMP.

The merger of *C and *t is seen in Siraya, Amis, Bunun, Kavalan, Ketagalan

and PMP. The NE languages: Kavalan and Ketagalan, were not in contact with

the rest in recent historical times, but this appears to be due to the intrusion

of Atayalic on the east coast in the Yilan region a few hundred years ago

(Mabuchi 1954). If so, the precursors of all of Siraya, Bunun, Amis, Kavalan

and Ketagalan were in contact a few hundred years ago, and may well have

been contiguous at the time the merger occurred. That PMP also underwent

the change indicates that its Formosan precursor was located within the zone

where the change occurred. This means that the change occurred before

4000 BP. However, PMP and Kavalan-Ketagalan did not inherit the change

from Muish, because one muic language: FATK (see section 5) did not merge

*C and *t. The conclusion must be that the change spread to Kavalan-

Ketagalan and to the Formosan precursor of PMP between the breakup of

Muish and 4000 BP. This is more parsimonious and realistic than supposing

three separate occurrences of *C => *t, one in Bunun, one in PMP and one on

the East coast (so Blust 1999: 46, 52).

Blust (1999:46) regards the merger of *j and *n as the defining innovation of

his 'East Formosan', a construct comprising Ketagalan, Kavalan, Amis and

Siraya. This, again, cannot be a monophyletic taxon in my phylogeny. In

order to account for the facts under my phylogenetic assumptions, a similar

scenario as for *C => *t would have to be supposed, with the innovation

arising on one coast and spreading to the other, this time leaving out PMP. In

this case however, the hypothesis of a spread from one coast to the other is

less easy to maintain than with *C => *t, because geographically

intermediate Bunun did not undergo the change, and it is not clear that

Siraya was ever in direct contact with the east coast languages that have

undergone *j => *n.16 Supposing that such direct contact existed earlier on

between the precursors of Amis and Siraya would be a leap of faith. Another

16 In their study of Ogawa's material on Siraya varieties, Tsuchida and Yamada (in Tsuchida,

Yamada and Moriguchi 1991, for instance on pp. 55, 57, 60, 65, 89, 107 etc.) identified a

small number of Amis loans into a variety of Taivoan -a dialect of Siraya- spoken in the

village of Dazhuang 大庄. This is a southern outlier of Siraya, however. No loans to other

varieties of Siraya were identified.

troubling element with this change is that it affects simultaneously the place

and mode of articulation of its target phoneme, and involves the highly

unusual process of spontaneous nasalization: in a word it is a highly

unnatural change and I strongly doubt that such a merger ever occurred in

Taiwan. Suffice it to say here that another interpretation of the facts is

possible: the correspondence identified as PAn *j was a palatal nasal in PAn,

not a stop. The main innovation with this phoneme was its shift, under

systemic pressure, to a voiced palatal stop, in all the languages of Taiwan

(including the ancestor of PMP) with the exception of two conservatives

zones: Amis and Kavalan-Ketagalan on the east coast and Siraya on west

coast, where the phoneme preserved its nasal character: in these two areas

independently, again under systemic pressure, the palatal nasal merged with

*n, as part of the general process of loss of palatal sounds. It is this merger

which gives the appearance of a palatal stop merging with *n. A full

discussion of the phonetic value of the PAn phoneme identified as *j requires

a reconsideration of the PAn consonant system, however, and I must reserve

it for another occasion.

The merger of *N into *n affected Bunun, Ketagalan, Kavalan and PMP, but

not Amis, and Kanakanabu but not the rest of Tsouic. Lack of contact

between these languages in historical times again seems due to the late

intrusion of Atayalic, and a single event followed by geographical spread can

explain the membership of this change.

The fourth change is the merger of n and ŋ. This is one of the innovations

characterizing Blust's Western Plains group (Blust 1999), which comprises

Papora, Hoanya, Favorlang, Taokas and Thao. Here the position of Hoanya is

uncertain. On p. 44 Blust does not list n/ŋ as a merger in Hoanya, yet he

places Hoanya next to Papora as a Central Western Plains language on p. 45.

However this may be, all these languages were in contact in historical times

and the basic requirement is met for a sound change to spread by contact.

There is indeed some evidence to show that contact played a role in the

distribution of this feature.17 Note however, that there is nothing in the

present phylogeny to contradict Blust's Western Plains group.

17 There seems to be a difference between southern Hoanya (self-designation Lloa,

represented by Taorakmun in Ino's data) and northern Hoanya (self-designation Arrikun:

represented by Savava in Ino's data): thus 'ear', PAn *Cariŋa, Savava

sangera

, Taorakmun

The fifth change is the merger of the sounds identified as S1 and S2 in the

tradition of Dyen, Tsuchida, Dahl and Ho Dah-an in Amis, Bunun, Puyuma,

Kavalan and PMP. An example of a word with S2 is *kaS2uy 'wood, tree'.

Again, a full discussion of the PAn consonant system must be reserved for

another occasion, but some initial remarks on S2 are in order here, as the

existence of S2 as a separate phoneme is controversial: Blust does not regard

it as distinct from S1. The reflexes of S1 and S2 are however markedly different

(Dahl 1981:35; Ho 1998:165). On the other hand, while S2 is typically viewed

as a sibilant with a different point of articulation from S1, it is curious that S1

and S2 are never reflected as different sibilants in the same language. In fact,

with some exceptions in Bunun, the only languages which reflect S2 as a

sibilant are Amis, Bunun, Kavalan and PMP, precisely those which merge it

with S1.18 In all the languages where S2 is not merged with S1, its reflexes are

identical with those of PAn *h (a.k.a. *H1). This suggests that the

correspondence called S2 defines a subset of PAn *h- (probably *h- before a

high front vowel) which palatalized to ɕ- after PAn, thereby merging with *S1

in an unbroken contact area on the east coast covering Amis, Bunun, Kavalan

and the Formosan ancestor of PMP. Due to the small number of forms

including S2, it is unclear whether Ketagalan participated in this merger. At

any rate the membership of this sound change, and of the other sound

changes discussed in this section as well, is adequately explained by

geographical contact.

Overall, the contribution of sound changes to the subgrouping of Formosan

languages is not great. The most useful information they provide is

geographical: what languages were in contact at the time when a change

took place. WE can use this information to probe the location of the

Formosan precursor of PMP: that language must have been spoken near the

precursors of Bunun and Kavalan-Ketagalan, since these are the languages

with which it shares the most sound changes. This probably indicates an

east-coast location.

sa:rinna

(Ino 1998). This suggests that the change spread to Hoanya without affecting all its

dialects.

18 Puyuma also merges the two but the result is zero. It is not clear whether S1 and S2

converged towards zero in Puyuma or first merged as a sibilant.

The enriched phylogeny in Table 4 summarizes the discussion of lexical and

morphological innovations in this section.

5. The position of Tai-Kadai

The observant reader will have noticed that Table 4 includes two languages

named 'FATK' and 'FAMP'. These acronyms means 'Formosan Ancestor of

Tai-Kadai'19 and 'Formosan Ancestor of Malayo-Polynesian'20 respectively.

With respect to the former, the claim is being made that, contrary to common

sense, the Tai-Kadai languages are descended from an East Formosan

language: in effect, that they are a branch of Austronesian, and specifically,

a subgroup of Muic.

The modern Tai-Kadai languages are spoken in parts of south China,

Vietnam, Laos, Thailand, Burma and Assam. Ostapirat (2000)21 distinguishes

three branches: Hlai in Hainan, and on the mainland Kam-Tai and Kra, but in

a more recent paper (in press), he suggests a different classification, with a

northern branch (Kra and Kam-Sui) and a southern branch (Hlai and Tai). The

Tai-Kadai languages outside of south-east China and adjacent areas of

Vietnam are all within the Tai subgroup, and are very homogeneous. This is

due to the historically well-attested expansion of Tai speakers in the late first

and early second millennia CE. The area of highest diversity is in the north-

east part of the Tai-Kadai domain: in Hainan Island, in northern Vietnam, and

in the Chinese provinces of Guangxi and Guizhou. This is presumably where

the TK homeland was located (although some of the original diversity must

have been lost to Chinese in Guangdong province).

On the ground that they share a remarkable set of very basic vocabulary

items (personal pronouns, numerals, body part terms, basic verbs), Benedict

(1942) proposed that the An and TK families are coordinate within an

19 In Sagart (2001; in press, a), FATK was called 'AAK'. I will from now on use the

synonymous, but more specific term FATK.

20 I introduce the acronym FAMP to distinguish the pre-migration language from PMP proper,

which I believe was spoken in the Philippines.

21 Ostapirat (2000) uses the name 'Kra-dai' for Tai-Kadai

'Austro-Tai' macrophylum. Yet, because of his exuberant methodology, his

proposal did not meet with the full approval he expected, in particular from

Austronesianists. Yet, as argued in Sagart (2001; in press, a) and Ostapirat

(in press), sets in Benedict's basic vocabulary comparisons can be isolated

which exhibit strong phonetic regularities,22 as shown in Table 5. This kind

of evidence virtually eliminates the possibility of chance resemblances.

For the three words in Table 5 Benedict reconstructed Proto-Austro-Tai

*maplay 'die', *mapla 'eye' and *mamlok 'bird': the medial clusters then

evolving to *C and *N in PAn, and either to pl-, ml- or to t-, n-, in Tai-Kadai.

To him, this was proof that a language ancestral to both PAn and PTK was

needed to explain the sound correspondences between them. However, both

Sagart (in press, a) and Ostapirat (in press) reject this interpretation. Sagart

argues that the TK forms are better accounted for on the ground of cluster-

less forms like PAn maCa, maCay, or PMP manuk, based on an explanation

originally proposed by Haudricourt (1956). Thus, in the case of 'eye':

maCa > mCa > pCa > pta > pla ~ ta

Sagart observed that in general, most An-related TK forms can be adequately

explained on the ground of PAn, or even PMP.

To remove any lingering misconceptions that the resemblances between Tai-

Kadai and Austronesian are due to chance, I give below comparisons

involving Buyang, a recently-described Tai-Kadai language from the Kra

branch, spoken near the China-Vietnam border (data from Li Jinfang 1999).

While TK languages generally reflect An disyllables as monosyllables (either

by losing the first syllable, or by collapsing the two), Buyang is remarkable in

that it preserves several An disyllables as disyllables (Table 6). Observe that

in these words, the first syllable is reduced: the vowel is always /a/, the

syllable is toneless ('tone zero'), and the inventory of initial consonants is

22 Ostapirat (in press) has greatly clarified the sound correspondences between An-TK,

although he cautiously refrains from characterizing the An-TK relationship as genetic as

opposed to contact-induced.

limited to a few: m- (for An m- and w-), q- (for An q- and k-), t- (for An C-

and t-). Yet Buyang shows that An words in Proto-TK were still disyllabic.

It is noteworthy that among the best An-TK comparisons are personal

pronouns and numerals, as shown in Table 7 and Table 8:

What is remarkable about the vocabulary shared by Tai-Kadai and

Austronesian is not so much the very basic nature of the shared elements as

the paucity of credible comparisons in the cultural vocabulary, notably the

vocabulary of agriculture (Sagart 2003), the names of domestic animals, and

the vocabulary of house-building. I have argued (Sagart 2001; in press, a)

that this is not compatible with Thurgood's explanation in terms of An loans

to Tai-Kadai (Thurgood 1994): Tai-Kadai could not plausibly borrow

principally basic vocabulary from An. The only realistic explanation left is

genetic, as Benedict thought, although the explanation I have proposed is

different from his.

Benedict regarded TK as a very old taxon, with an ancestral language of a

comparable age to PAn, or even older. Misled by exaggeratedly early

archaeological dates for bronze in North Thailand (Solheim 1971), he

characterized proto-Austro-Tai as the bearer of the high culture in early east

Asia, and the source of many loanwords to Chinese, in particular in the

domains of metallurgy and agriculture. It is now crystal-clear that the loans

went the other way (Sagart 1999 for metal names). Benedict located the

Austro-Tai homeland in south-eastern China. However, Ostapirat's recent

reexamination of the family, independently supported by Peiros's

glottochronological study, shows the date to be considerably more recent,

not earlier than 2000 BCE (Ostapirat, in press) or 1800 BCE (Peiros

1998:15).23

23 Peiros did not take the Kra languages into account in his calculations, however.

Inspection of Table 6, Table 7 and Table 8 shows that where PAn and PMP

disagree, Buyang sides with PMP. On the ground that they share the *-mu

'2sg-genitive' and *manuk 'bird' innovations with PMP, I have argued (Sagart

2001; in press, a) that the Tai-Kadai languages are a subgroup of

Austronesian, closely related to PMP, rather than a related but separate

language family. Table 7 now shows that Buyang also sides with PMP against

PAn in the matter of numerals. The inescapable conclusion, then, must be

that TK is a subgroup within An. The interesting question with the An-TK

relationship, in my view, is not so much 'why are there so many An words in

TK ?' as 'why are there so few ?'. 24 My tentative answer (Sagart 2001, in

press, a) is that TK evolved on the mainland out of the Formosan

Austronesian language I call FATK; and that, once on the mainland, it

underwent intimate contact with, and extensive relexification from, a local

language having not left any other descendant (although macrophylic

connections to AA or MY are a distinct possibility). In the course of that

period of contact, a large part of the original An vocabulary of TK was lost,

with only the most basic part of the vocabulary resisting relexification.

Under the present theory of An subgrouping, sharing the *-mu and *manuk

innovations with PMP and with Ketagalan quite definitely makes FATK a muic

language. It is interesting to consider the predictions made by this claim. If

Tai-Kadai truly goes back to a muic language, and barring, of course, ulterior

innovations, then we should expect it to have all the post-PAn innovations

discussed earlier in this paper, such as the short forms of the numerals '7',

'8' and '9'. It would be devastating if Tai-Kadai reflected PAn etyma which I

have claimed had already been displaced by newer words in muic: for

instance if Tai-Kadai reflected *RaCep-i-tuSa as 'seven', *RaCep as 'five',

*(sa-)iCit or #masehaN as 'ten', *kawaS as 'year', *qiNaS for 'moon' etc. On

the contrary it would support my theory if Tai-Kadai reflected *pitu as

'seven', lima as 'five', *puluq as 'ten', etc.

The predictions on Tai-Kadai of the present theory of An phylogeny are

verified, a shown in Table 9. The original Tai-Kadai numerals are preserved

24 This is the question asked by Peiros (1998:103): while he recognized the basicness of the

list of the list of An-TK lexical comparisons, and noted it was indicative of a genetic

relationship, he was puzzled by its brevity, and by the failure of attempts to enlarge it.

only in the Hlai and Kra branches: in the Kam-Tai branch they have been

replaced with Chinese numerals. In Table 8 we have already seen the

numerals of Buyang, a Kra language: the resemblance between the Buyang

and MP numerals is obvious. The Proto-Kra numerals from five to ten, as well

as the words for 'your' (sg.) and 'bird' as reconstructed by Ostapirat (2000),

are listed in Table 9.

The sound correspondences with PAn are only beginning to be elucidated

(Ostapirat in press), and a full demonstration that all the segments of P-Kra

words regularly correspond to An words cannot yet be made. Even so, one

can make some preliminary observations.

The PK word for 'five': *r-ma matches the last syllable of *lima and probably

the first consonant too, as there are parallels for An l :: PK r ('eight'). There

are also good parallels for An m :: PK m and An a :: PK a. Ostapirat

reconstructs PK *x-nəm 'six', where 'x' is an element aiming at accounting

for the unexplained alternation between high and low tones in this set. It may

have been a glottal stop. Benedict (1975: 212) similarly reconstructs *nəm

'six' for proto-Hlai in Hainan. This is close to An *enem. Ostapirat (2004)

gives parallels for An *n- :: TK *n- and An *-m :: TK *-m. Ostapirat's

reconstruction for 'seven': *t-ru, is problematic. The r- in it is nowhere

reflected as a r-type sound, and Hlai indicates a simple voiceless t-. I believe

an alternative PK reconstruction for 'seven' is *(C-)tu (where C is a voiceless

stop), meaning that PK had an alternation between *C-tu and *tu, with a

majority of languages reflecting the latter, but Paha *C-tu > ʔd- > ʔr- >

ðhuu

. If so, *(C-)tu is a likely match for *pitu: Ostapirat (2004) gives other

examples of *t :: *t and *u :: *u.25 In Ostapirat's PK reconstruction for

'eight', -ru matches the last syllable of *walu, with the same An l :: PK r

correspondence as in 'five'. Whether or not Kra initial m- regularly matches

An *w- is unclear. We have seen that in Buyang, m- in the first syllable of '8'

matches An m- and w-, but it is unclear whether this is true of the other Kra

languages. The Kra language Laqua prefixes mə- to all of 'seven', 'eight' and

25 Weera Ostapirat (p.c., January 2004) indicates that the reconstruction *C-tu for '7' had

been proposed in his PhD dissertation. Although he abandoned it in his book (Ostapirat

2000), he now considers it preferable on Kra-internal grounds.

'nine' (Benedict 1975: 212), perhaps as a result of leveling. For Proto-Gelao

(a subgroup within Kra) Ostapirat (2000:122) reconstructs initial *wr- in

'eight', which appears to agree well with An *walu. Proto-Gelao *wr- is not

the regular outcome of PK *m-r-: it could represent the original Kra onset of

this word. Finally, Proto-Kra *s-ɣwa 'nine' is an attractive match for *Siwa,

especially since the correspondence An *S- :: PK s- has parallels (notably

*duSa 'two' : PK *sa A 'two'). The correspondence An *-w- :: PK ɣw- has a

parallel in PAn *duwa 'come, go' (Pazeh

dua

'go', Puyuma

Zowa

'come'): PK

*ɣwa C 'go'. With 'ten' the final consonant correspondence An -q :: PK -t

seems off but Ostapirat (in press) treats it as regular following /u/.

Tai-Kadai, then, has very plausible candidate reflexes for the post-PAn

innovative forms in Table 9. I know of no case of a PAn word being displaced

by an innovation after PAn and before Muish, and at the same time having an

attractive TK comparison. This supports the view that TK is a daughter

language of PAn, and more specifically a muic language. FATK cannot be a

MP language because, as discovered by Ostapirat (in press), TK preserves the

distinction between PAn *C and *t, and, at least in some words ('two'), has a

sibilant reflex of PAn *S.

Ostapirat further argues that, contrary to my claim, TK cannot have its origin

in an early east Formosan language, because all east coast languages merge

*C and *t. He considers that, if the relationship between Tai-Kadai and An

really is a genetic one, Tai-Kadai must be coordinate with the whole of

Austronesian, as Benedict thought, or, at least, with a higher-order taxon

than East Formosan-PMP. I maintain, based on the evidence of lexical

innovations shown in Table 9, that Tai-Kadai is part of Muic. As indicated

earlier, I believe that the merger of *C into *t spread to Ketagalan and to the

Formosan ancestor of PMP but failed to reach FATK, after the breakup of

Muish. In fact there is a tiny bit of evidence to suggest that FATK underwent

a merger which cannot be traced back earlier than Walu-Siwaish, and which

no language of the west coast exhibits, to wit, the merger of S1 and S2:

Buyang has

ɕa

'two', corresponding to S1 in PAn duS1a 'two', and

ɕui

'fuel',

corresponding to S2 (my *h) in PAn *kaS2uy (my *kahiuy) 'wood'.

Recognition that TK is a subgroup of An opens new perspectives for the

reconstruction of PAn, notably in the area of final laryngeals, which are

reflected in TK tones.

I am claiming, therefore, that the tree in Table 4 is an approximation of the

higher phylogeny of the An family, including its newly-recognized subgroup

Tai-Kadai.

6. The An phylogeny in space

Supposing that the implicational hierarchy in Table 1 was accidental, and that

the etymologies proposed for *pitu, *walu and *Siwa in section 3 were

fanciful, we should not expect the phylogeny proposed in Table 4 to result in

any kind of recognizable geographical pattern. Yet a clear pattern emerges.

Approximate geographical locations for PAn, Pituish, Enemish, Walu-Siwaish

and Muish can be determined based on the location of their direct

descendants in historical times. The most likely location of PAn is in the

region of Luilang, Saisiat and Pazeh, in the north-west of Taiwan. Pituish

must have been spoken in the western plains somewhat to the south of PAn,

being ancestral to Atayalic, Favorlang, Taokas, Thao, Papora and Hoanya.

There is a tradition that the present-day northern location of Atayal and

Sediq was reached following migrations from west-central Taiwan (Mabuchi

1954): Blust (2003:6) reports that Thao was previously spoken near Jiayi in

the western lowlands, reaching Sun-Moon Lake only 300-350 years ago.

Enemish must have been spoken more to the south on the West coast,

towards the area occupied by Siraya in recent times. Walu-Siwaish may have

been spoken near the southern tip of the island, or on the south-east coast.

Muish was probably further north along the east coast, as suggested by the

location of its modern descendants Kavalan and Ketagalan in the NE of the

island, and the observation made in section 4.3. that FAMP must have been

spoken in contact with the precursors of Kavalan-Ketagalan and of Bunun.

This model shows a consistent geographical pattern: early Austronesian

speakers settling Taiwan progressively in a counter-clockwise movement,

starting from the north-west, then expanding southward along the west

coast, and reaching the southern tip of the island before finally settling the

east coast from south to north, as shown in Figure 2.

The progression of the early ANs in Taiwan is an illustration of the "wave of

advance" model (Cavalli-Sforza and Ammerman 1984), which describes the

progression of neolithic settlers in areas not yet touched by agriculture. In

Taiwan, the cumulative character of the linguistic innovations at each stage

leaves no other explanation than the progression of such a wave of advance,

gradually encircling the island (although minor population movements,

leaving no linguistic traces, must certainly have occurred).

One may wonder why there was no clockwise progression from north-

western Taiwan around the northern tip of the island towards the east coast.

I conjecture that the large and possibly malarial freshwater lake or swamp

which at the time occupied the Taipei basin rendered movement in that

direction unattractive.

Although the location of Pituish, Enemish, Walu-Siwaish and Muish cannot be

established with a high degree of precision, the general pattern is clear: a

gradual, unidirectional encirclement of the island by Austronesian speakers.

Apparently the main direction of movement was along the coastal plains.

This implies that, given a choice, the early Austronesians preferred to expand

into the coastal plains. This pattern is consistent with what archaeology and

linguistics tell us about their mode of subsistence, which combined

exploitation of marine resources, including fishing, with hunting and

gathering and cultivation of rice and millet. We may suppose that population

movements into the mountains, as with the Saisiats, Atayalics, Thaos,

Tsouics and Bununs, were generally late, and made under pressure. Such

indeed is the pattern observed in the rest of the Austronesian world (Blust

1999:53). The pattern of progression from the west to the east coast is

moreover consistent with archaeological dates for Ta-Pen-K'eng sites, which

are older on the west coast than on the east coast. It is tempting to imagine

that the Nan-kuan-li people, who were active near Tainan c. 5000-4500 BCE

(Tsang, in press) spoke a form of Enemish, while the Yuan-Shan people, who

were active north of Taipei from around 4500 BP (Bellwood 1997:215), spoke

a form of Muish.

The geographical stability over time of the initial settlement pattern is

striking. Most modern languages are still spoken or were still spoken until

recently in the area of the meso-language they are descended from. A major

factor in this is the geography of Taiwan, where the central mountain range

very effectively prevents contact and migration between the east and west

coasts.

Finally, under the present interpretation, FAMP and FATK, the two Muic

languages whose speakers left Taiwan to settle other regions, were probably

located in the north-east or north of the island, where the last available

agricultural lands had been. The MP and TK migrations out of Taiwan thus

appear motivated by the need to find new agricultural lands. It is probably no

coincidence that the site of Yuan-Shan near Taipei, in the region where

Ketagalan was spoken until the early 20th century, has significant

connections to the earliest neolithic of the Philippines (Bellwood 1997: 215).

7. The time scale of the early An settlement of Taiwan

The primary evidence for the time scale of the Austronesian settlement of

Taiwan comes from archaeology. Bellwood's estimates for the date of the

initial Austronesian settlement of Taiwan, inferred from the earliest Ta-Pen-

K'eng radiocarbon dates (plus a few hundred years for good measure), and

from the earliest neolithic sites in the Cagayan Valley in the northern

Philippines, are c. 5500 BP for the former and ca. 4000 BP for the latter

(Bellwood 2004). During that period, the daughter languages of PAn were

presumably confined to Taiwan. While these are provisional dates, they

provide approximate external limits between which the full settlement of

Taiwan must have taken place: in the present framework, the initial

settlement of the Philippines, presumably by PMP speakers, could not have

occurred until Muish had already broken up into its three components.

Likewise, the initial Tai-Kadai settlement on the mainland, by FATK speakers,

could not have occurred before the breakup of Muish. It is relevant here to

recall that Ostapirat estimates the date of PTK to be no older than 4000 BP,

simultaneous with the Cagayan dates.

That two sound changes: merger of *C and *t and merger of *n and *N,

spread to FAMP but not to FATK (Ostapirat, in press, claims these four initials

were distinguished in proto-TK) indicates that a significant amount of time

elapsed between the break-up of Muish and the MP migration out of Taiwan,

even though it is impossible to say precisely how much. That FATK failed to

undergo these changes means that either it was located further north along

the east coast than FAMP and the changes, spreading from the south,

stopped at FAMP; or alternatively that the TK migration was earlier than the

MP migration.

8. Archaeology and language: some conjectures on pre-Austronesian times

I have argued elsewhere (Sagart, in press, b) that the pre-Austronesians

spoke a language related to Sino-Tibetan, and that they reached Taiwan from

a location in NE China where millet and rice were cultivated, and where ritual

evulsion of the upper lateral incisors in boys and girls was practiced. The

eastern China seaboard region north of the Yangzi estuary, from north

Jiangsu to north Shandong, is the one area in East Asia where the distribution

of these three traits overlaps in the period before the arrival of the

Austronesians in Taiwan: thus both rice and millet were cultivated in Xihe in

north Shandong (Wright 2004) c. 8000 BP and in Longqiuzhuang in the lower

Huai basin c. 7000-5000 BP. Tooth evulsion is attested from 6500 BP on in

Shandong and north Jiangsu (Han and Nakahashi 1996). We may surmise that

before they reached Taiwan, the pre-Austronesians were expanding

southward along the coastal plains of central-eastern China in Jiangsu,

Zhejiang and north Fujian. We can expect that archaeological sites with rice,

Setaria, tooth evulsion, and a technology intermediate between the

Dawenkou culture of north-east China and Ta-Pen-K'eng of Taiwan will

eventually appear there.

If this scenario is correct, it is likely that the passage to Taiwan did not

exhaust the pre-An population of the Fujian coast. More likely, this

population continued expanding along the coast in a south-westerly

direction towards the Pearl River delta, even after a group of them had

crossed to Taiwan. Their archaeological traces SW of Fujian are perhaps seen

in the Pearl river delta, although direct evidence of agriculture there has so

far not appeared; Hedang in the Pearl River delta, with tooth evulsion

(Higham 1996:84), c. 3000-2000 BCE, may be one such site. In Taiwan,

Tsang (in press) describes the newly excavated site of Nan-kuan-li near

Tainan in south-west Taiwan, where a team led by him recently discovered a

neolithic culture having rice, millet, and practicing ritual tooth ablation

around 5000-4500 BP. In the same paper he argues that the Ta-Pen-K'eng

culture, as seen in Nan-kuan-li near Tainan, "has close affinities with the

Neolithic cultures of Hong Kong and the Pearl River Delta". I disagree with

Tsang when he concludes that "The Pearl River Delta of Kuangtung is most

probably the source area of the Tapenkeng Culture in Taiwan". I think it more

likely that both cultures are descended from a common precursor on the

Fujian coast. Pearl River delta sites having affinities to Taiwan TPK like

Hedang are also probably too early and too far east to be ancestral to the

Tai-Kadai-speaking cultures.

9. conclusion

I have presented an explicit account of the early phylogeny of the

Austronesian family. The new phylogeny is tree-like. A salient characteristic

is that out of a majority of nodes, only one branch leads to further branching

(Table 4). This makes Formosan phylogeny similar to Malayo-Polynesian

phylogeny. Non-branching nodes can be associated with stay-at-homes, and

branching ones with out-migrating groups. PMP has been shown to be part

of a taxon that also includes languages of the NE Formosan Coast, as well as

Tai-Kadai (as proposed in Sagart 2001; in press, a). That taxon itself is part

of a larger taxon including languages of the East coast and south Taiwan.

These proposals have been made on the ground of the convergence of three

independent lines of evidence: (1) the implicational hierarchy with the

numerals 5-10, shown in Table 1; (2) the systematic resemblances between

the consensus numerals for 7-8-9 and the corresponding numerals in Pazeh,

described in section 3; and (3), the geographically coherent and processually

realistic spatial pattern of settlement shown in Figure 2. (2) is obviously

independent from (1) and (3); and (1) and (3) are independent from each

other because one could have an implicational hierarchy which did not result

in a coherent spatial pattern. None of these independent sets of facts actually

'proves' the phylogeny in Table 4: rather, that phylogeny makes sense of

them all: while the same facts have to be regarded as coincidences under

other phylogenies. In effect, (1), (2) and (3) are three independent verified

predictions of the phylogenetic hypothesis in Table 4. Because it makes more

verified predictions than earlier hypotheses, it should be preferred.

References

Adelaar, K. A. 1997. Grammar notes on Siraya, an extinct Formosan

language.

Oceanic Linguistics

36, 2: 362-397.

Bartoli, M. 1925.

Introduzione alla neolinguistica. Princìpi, scopi, metodi

Geneva: Olschki.

Bellwood, P. 1997.

Prehistory of the Indo-Malaysian archipelago

. Honolulu:

University of Hawai'i Press.

-------- 2004. Taiwan, Batanes, Cagayan, Talaud, Maluku, Marianas, Lapita

Gardens, and all stations to Polynesia. All aboard the Austronesian Express.

Paper presented at the workshop on Human migrations in continental East

Asia and Taiwan

, Geneva, June 10-13, 2004.

Benedict, P.K. 1942. Thai, Kadai and Indonesian: a new alignment in

Southeastern Asia.

American Anthropologist, n.s.,

44: 576-601.

-------- 1972.

Sino-Tibetan: a Conspectus

. Cambridge: University Printing

House.

-------- 1975.

Austro-Thai: language and culture

. New Haven: HRAF Press.

-------- 1995. Extra-Austronesian evidence for Formosan etyma. In

Austronesian studies relating to Taiwan

, ed. by Paul J.-K. Li, C.H Tsang, Y.-

K. Huang, D.-A. Ho and C.Y. Tseng, pp. 399-454. Symposium series of the

Institute of History and Philology, Academia Sinica 3. Taipei: Academia

Sinica.

Bloomfield, L. 1933.

Language

. New York: Holt.

Blust, Robert A. 1977. The Proto-Austronesian pronouns and Austronesian

subgrouping: a preliminary report.

University of Hawai'i working papers in

Linguistics

9, 2:1-15.

Blust, R. 1995. Sibilant assimilation in Formosan languages and the Proto-

Austronesian word for 'nine'.

Oceanic Linguistics

34:443-453.

-------- 1998. Ca- Reduplication and Proto-Austronesian grammar.

Oceanic Linguistics

37:29-64.

-------- 1999. Notes on Pazeh phonology and morphology.

Oceanic

Linguistics

38:321-365.

-------- 2001. Malayo-Polynesian: new stones in the wall.

Oceanic

Linguistics

40:151-155.

-------- 2003.

Thao Dictionary

. Nankang: Institute of Linguistics

(preparatory office), Academia Sinica.

Bril, I. 2002.

Le nêlêmwa (Nouvelle-Calédonie): Analyse syntaxique et

sémantique

. Collection "Langues et Cultures du Pacifique", n° 16. Paris:

Peeters.

Buck, C. D. 1949.

A dictionary of selected synonyms in the principal

European languages

. Chicago: Chicago University Press.

Carstairs, D. 1899

edn.

Chinese-English dictionary of the vernacular or

spoken language of Amoy, with the principal variations of the Chang-chew

and Chin-chew dialects.

London: Presbyterian Church of England.

Cavalli-Sforza, L.-L., and A. Ammerman. 1984.

The Neolithic Transition and

the Genetics of Populations in Europe

. Princeton: Princeton University Press.

Chambers, J.K. and P. Trudgill. 1998.

Dialectology

. Cambridge: Cambridge

University Press.

Chang, Yung-li. 2000.

Gemalan yu cankao yufa

. Taipei: Yuan-liou.

Dahl, O. Ch. 1981.

Early phonetic and phonemic changes in Austronesian

Instituttet for sammenlingnende kulturforskning Serie B: Skrifter LXIII. Oslo:

Universitetsfor-laget.

Dauzat, A. 1922.

La géographie linguistique

. Paris: Flammarion.

Edmondson, J. and G. Thurgood. 1992. Gelao reconstruction and its place in

Kadai. Paper presented at the 25th International Conference on Sino-

Tibetan Languages and Linguistics, 14-18 Oct. 1992, Berkeley, California.

Egerod, Søren. 1980.

Atayal-English Dictionary

. Scandinavian Institute of

Asian Studies monograph 35. London and Malmø: Curzon Press.

Ferrell, R. 1969.

Taiwan Aboriginal groups: problems in cultural and

linguistic classification

. Monograph No. 17, Institute of Ethnology,

Academia Sinica. Nankang: Academia Sinica.

-------- 1982.

Paiwan dictionary

. Pacific Linguistics C-73. Canberra: A.N.U.

François, Alexandre. 2001.

Contraintes de structures et liberté dans la

construction du discours. Une description du mwotlap, langue océanienne

du Vanuatu.

Doctoral dissertation. Paris: Université Paris-IV Sorbonne.

Fujian Provincial Museum. 1991. Brief report on the excavation of the

Keqiutou site in Pingtan, Fujian [in Chinese].

Kaogu

1991, 7:587-599.

Han, K.X. and T. Nakahashi. 1996. A Comparative Study of Ritual Tooth

Ablation in Ancient China and Japan.

Anthropological Science

104,1:43-64.

Harvey, Mark. 1979. Subgroups in Austronesian. BA Honours thesis.

Canberra: A.N.U.

-------- 1982. Subgroups in Austronesian. In

Papers from the Third

International Conference on Austronesian Linguistics

, vol. 2:

Tracking the

travellers.

(Pacific Linguistics C-75), ed. by A Halim, L. Carrington and S.A.

Wurm, pp. 47-99. Canberra: ANU.

Haudricourt, A.-G. 1956. De la restitution des initiales dans les langues

monosyllabiques : le problème du Thai commun.

Bulletin de la société de

Linguistique de Paris

52:307-322.

Higham, C. 1996.

The Bronze age of southeast Asia

. Cambridge: Cambridge

University Press.

Ho, Dah-an. 1998. Taiwan Nandaoyu de Yuyan Guanxi.

Chinese Studies

16,

2:141-171.

Ho, Dah-an and Yang Hsiu-fang. 1999. Nandaoyu yu Taiwan Nandaoyu.

Introduction to a collection of monographs on Taiwan languages

. Taipei:

Yuanliou.

Huang, Lillian. 2000.

Beinan yu cankao yufa

. Taipei: Yuanliou.

Imbault-Huart, C. 1893.

L'île Formose

. Paris: Ernest Leroux.

Ino, Y. 1998. Xun Tai Ri Cheng. In

Ino Yoshinori Fanyu Diaocha Shouce

, ed.

by T. Moriguchi, pp. 13-201. Taipei: Southern Materials Center.

Li, Jinfang. 1999.

Buyang Yu Yanjiu

. Beijing: Zhongyang Minzu Daxue.

Li, P. J.-K. 1975.

Rukai texts

. Nankang: Institute of History and Philology

special publications No. 64-2.

-------- 1995. Formosan vs. non-Formosan features in some Austronesian

Languages in Taiwan. In

Austronesian Studies Relating to Taiwan

, ed. by

Paul Jen-kuei Li, Dah-an Ho, Ying-kuei Huang, Cheng-hwa Tsang, and

Chiu-yu Tseng, pp. 651-681. Symposium series of the Institute of History

and Philology, Academia Sinica, no. 3. Taipei: Academia Sinica.

-------- 2001. Basai yu de diwei [in Chinese].

Language and Linguistics

2,2:155-171.

Li, P. J-K, and S. Tsuchida. 2001.

Pazih dictionary

. Nankang: Institute of

Linguistics (preparatory office), Academia Sinica.

-------- 2002.

Pazih texts and songs

. Nankang: Institute of Linguistics

(preparatory office), Academia Sinica.

Lin, Yingjin. 2000.

Bazehai yu

. Taipei: Yuanliou.

Mabuchi, T. 1954. Migration and distribution of the Formosan aborigines [in

Japanese].

Minzogaku Kenkyu

18,1-2:123-154; 18,4:23-72.

Matisoff, J.A. 1997.

Sino-Tibetan numeral systems: prefixes, protoforms and

problems

. Canberra: Pacific Linguistics B-114.

Mazaudon, M. 1985. Dzongkha number systems. In

Southeast Asian

linguistic studies presented to André G. Haudricourt

, ed. by S. Ratanakul,

D. Thomas and S. Premsrirat, pp.124-157. Bangkok: Mahidol University.

-------- 2002. Les principes de construction du nombre dans les langues

tibéto-birmanes. In

la pluralité

, ed. by J. François, pp.91-119. Paris:

mémoires de la Société de Linguistique de Paris, nouvelle série, tome XII.

Meacham, C., and G. Estabrook. 1985. Compatibility methods in systematics.

Annual Review of Ecology and Systematics

16:431-446.

Ogawa, Naoyoshi, and Erin Asai. 1935.

The myths and traditions of the

Formosan native tribes

. Taipei.

Ostapirat, W. 2000. Proto-Kra.

Linguistics of the Tibeto-Burman Area

23.1.

-------- in press. Kra-dai and austronesian: notes on phonological

correspondences and vocabulary distribution. In

The peopling of East Asia:

Putting together Archaeology, Linguistics and Genetics

ed. by L. Sagart, R.

Blench and A. Sanchez-Mazas. London: RoutledgeCurzon.

Pecoraro, F. 1977. Essai de dictionnaire taroko-français.

Cahier d'Archipel 7

Paris: S.E.C.M.I.

Peiros, I. 1998.

Comparative Linguistics in Southeast Asia

. Canberra: Pacific

Linguistics C-142.

Reid, Lawrence A. 1971.

Philippine minor languages: Word lists and

phonologies.

Oceanic Linguistics Special Publication, 8. Honolulu:

University of Hawaii Press. xii

-------- 1982. The demise of Proto-Philippines. In

Papers from the Third

International Conference on Austronesian Linguistics

, vol. 2:

Tracking the

travellers

, ed. by A Halim, L. Carrington and S.A. Wurm, pp.

210-216.

Canberra: Pacific Linguistics C-75.

Ross, Malcolm D. 1995. Some current issues in Austronesian Linguistics. In:

Darrell T. Tryon. ed.)

Comparative Austronesian Dictionary, part 1, fascicle

: 45-120. Berlin, New York: Mouton de Gruyter.

Sagart, L. 1999.

The Roots of Old Chinese.

Current Issues in Linguistic

Theory, 184. Amsterdam: John Benjamins.

-------- 2001. Comment: Malayo-Polynesian features in the An-related

vocabulary in Kadai. Paper presented at the workshop "Perspectives on the

Phylogeny of East Asian Languages", August 28-31, Périgueux.

-------- 2002a. Sino-Tibeto-Austronesian: an updated and improved

argument. Paper presented at the 9th International Conference on

Austronesian Linguistics, Canberra, January 8-11, 2002.

-------- 2002b. Gan, Hakka and the Formation of Chinese Dialects, in

Dialect Variations in Chinese

, ed. by Dah-an Ho, pp. 129-154. Taipei:

Academia Sinica.

-------- 2003. The vocabulary of cereal cultivation and the phylogeny of

East Asian languages.

Bulletin of the Indo-Pacific Prehistory Association

23.

Taipei papers, Volume 1:127-136.

-------- in press, a. Tai-Kadai as a subgroup of Austronesian. In

The

peopling of East Asia: Putting together Archaeology, Linguistics and

Genetics

, ed. by L. Sagart, R. Blench and A. Sanchez-Mazas. London:

RoutledgeCurzon.

-------- in press, b. Sino-Tibetan-Austronesian: an updated and improved

argument. In

The peopling of East Asia: Putting together Archaeology,

Linguistics and Genetics

, ed. by L. Sagart, R. Blench and A. Sanchez-Mazas.

London: RoutledgeCurzon.

Solheim, Wilhelm G. 1971. New Light on a Forgotten Past.

National

Geographic Magazine

, 139,3:330-39.

Starosta, Stanley. 1985. Verbal inflection versus deverbal nominalization in

PAn: the evidence from Tsou. In

Austronesian linguistics at the 15th Pacific

Science Congress

, ed. by Andrew Pawley and Lois Carrington

Pacific

Linguistics.

-------- 1994. Rukai-Tsouic: subgroup or treetop ? Paper presented at the

7th International Conference on Austronesian Linguistics. Leiden: Leiden

University.

-------- 1995. A grammatical subgrouping of Formosan languages. In

Austronesian Studies Relating to Taiwan

, ed. by : Paul Jen-kuei Li, Dah-an

Ho, Ying-kuei Huang, Cheng-hwa Tsang, and Chiu-yu Tseng, pp. 683-

726. Symposium series of the Institute of History and Philology, Academia

Sinica, no. 3. Taipei: Academia Sinica.

-------- 1996. The position of Saaroa in the grammatical subgrouping of

Formosan languages. In

Pan-Asiatic Linguistics: Proceedings of the Fourth

International Symposium on Languages and Linguistics, Volume III

, ed. by

Suwilai Premsrirat, pp. 944-966. Salaya, Thailand: Institute of Language

and Culture for Rural development, Mahidol University at Salaya,

-------- 2001. Reduplication and the subgrouping of Formosan languages.

Paper presented at the International Symposium on Austronesian Cultures:

Issues relating to Taiwan: Taipei, 8-12 December 2001.

Thompson, J. 1873. Notes of a journey in southern Formosa.

Journal of the

Royal Geographical Society

XLIII:97-107.

Thurgood, G. 1994. Tai-Kadai and Austronesian: the nature of the historical

relationship.

Oceanic Linguistics

33:345-368.

Topping, D. 1973.

Chamorro reference grammar

. Honolulu: University of

Hawaii Press.

Trudgill, P. 1974.

Sociolinguistics: an introduction

. Harmondsworth: Penguin.

Tsang, Cheng-hwa. 1995. New archaeological data from both sides of the

Taiwan straits and their implications for the controversy about

Austronesian origins and expansion. In

Austronesian Studies Relating to

Taiwan

, ed. by Paul Jen-kuei Li, Dah-an Ho, Ying-kuei Huang, Cheng-hwa

Tsang, and Chiu-yu Tseng, pp. 185-225. Symposium series of the

Institute of History and Philology, Academia Sinica, no. 3. Taipei: Academia

Sinica,

-------- in press. Recent discoveries at a Tapenkeng culture site in Taiwan:

implications for the problem of Austronesian origins. In

The peopling of

East Asia: Putting together Archaeology, Linguistics and Genetics

ed. by L.

Sagart, R. Blench and A. Sanchez-Mazas. London: RoutledgeCurzon.

Tsuchida, S. 1976.

Reconstruction of Proto-Tsouic phonology

. Study of

languages and cultures of Asia and Africa monograph series #5. Tokyo:

Tokyo Daikokugo Daigaku.

Tsuchida, S. Yamada, Y. and T. Moriguchi. 1991.

Linguistic materials of the

Formosan sinicized populations I: Siraya and Basai

. Tokyo: The University

of Tokyo, Linguistics Department.

Tung, T.H. 1964.

A descriptive study of the Tsou language

. Nankang:

Institute of History and Philology, Academia Sinica.

Wright, H. 2004. Early Austronesians and their neighbours on the East Asian

mainland: the archaeological context.

Paper presented at the workshop on

Human migrations in continental East Asia and Taiwan

, Geneva, June 10-

13, 2004.

Zeitoun, E. 2000a.

Lukai yu cankao yufa

. Taipei: Yuan-liou.

-------- 2000b.

Bunong yu cankao yufa

. Taipei: Yuan-liou.

Zhang, Yongli. 2000.

Gemalan yu cankao yufa.

Taipei: Yuan-liou.

Table 1: implicational hierarchy of the numerals 5-10 in Formosan languages

and in PMP. Gray cells: presence of the etymon. Sources: Amis: Wu (2000); Atayal:

Egerod (1980); Bunun: Zeitoun (2000a); Favorlang=Babuza: Ferrell (1969); Hoanya: Ferrell

(1969); Kanakanabu: Ogawa and Asai (1935), as cited in Ferrell (1969); Kavalan: Zhang

(2000); Ketagalan-Basai: Yamada, Tsuchida and Moriguchi (1991); Luilang: Ferrell (1969);

Paiwan: Ferrell (1982); Papora: Ferrell (1969); Pazeh: Li and Tsuchida (2001); Puyuma: Huang

(2000); Ruka (Budai): Zeitoun (2000b); Saaroa: Tsuchida, as cited in Ferrell 1969; Saisiat: Ino

(1998: Saitaoyak); Sediq: Pecoraro (1977), Siraya: Adelaar (1997); Taokas: Ferrell (1969);

Thao: Blust (2003); Tsou: Tung (1964).

pitu

'7'

lima

'5'

enem

'6'

walu

'8'

Siwa

'9'

puluq

'10'

Luilang innai (na)lup (na)tsulup patulunai satulunai isit

Saisiat saivuseaha rrasu saivusa makaspat ra:ha ranpon

Pazeh xasebidusa xasep xasebuza xasebaturu,

xasebituru

xasebisupat isit

Favorlang naito achab nataap maaspat tannacho zchiett

Taokas yweto hasap tahap mahalpat tanasu (ta)isid

Atayal pituʔ imagal cziuʔ spat qeruʔ lpuu

Sediq pito lima mataro maspat maŋali maxal

Thao pitu rima ka-turu,

makalh-

turu-turu

kahspat,

maka(lh)-

shpa-shpat

tanacu maqcin

Siraya pǐttu rima nəm kuixpa matuda saat kǐttian

Hoanya pito Lima (mi)nun (mi)alu (a)sia (miata)isi

Papora pitu nema (ne)nom mahal (me)siya (me)tsi

Tsou pítu eímo nómə vóeu sío máskə

Saaroa (k)upito (k)ulima (k)ənəmə (k)ualo (k)usia (ku)ma:ɬə

Kanabu pitu rima nəm (h)a:ru si:ya ma:nə

Bunun pitu' hima' nuum vau' siva' mas'an

Rukai pitu Lima eneme vaLu baŋatə maŋeale

Paiwan pitju lima enem,

unem

alu siva puluq

Puyuma pitu Lima nem waLu iwa puLu

Amis pitu lima 'enem falu siwa polo

Kavalan pitu rima 'nem waru siwa betin

Ketagalan pitu tsjima anəm wasu siwa labatan

PMP *pitu *lima *enem *walu *siwa *puluq

0. start = PAn 1. schwa > i

after -iC

2. pa > wa 3. delete

remaining

schwas

4. prune left

of pretonic

syllable

5. prune right

of stressed

vowel

6. tl > t

7 (5+2) RaCep-i-tuSa RaCepituSa RaCepituSa RaC_pituSa _pituSa pitu_ pitu

8 (5+3) RaCep-a-telu RaCepatelu RaCewatelu RaC_wat_lu _watlu watlu walu

9 (5+4) RaCep-i-Sepat RaCepiSipat RaCepiSiwat RaC_piSiwat _Siwat Siwa_ Siwa

Table 2: derivation of *pitu, *walu and *Siwa out of PAn analytic forms

5+ additive 6+ additive multiplicative subtractive

six

5+1 ___ 2x3 no exx

seven

5+2 6+1 ___ no exx

eight

5+3 no exx 2X4 no exx

nine

5+4 no exx ___ 10-1

Table 3. The PAn numerals from 6 to 9: analytic forms in Formosan

languages

PAn

Pituish

Walu-Siwaish

Luilang

Pazeh

Saisiat

Atayalic

Thao

Favorlang

Taokas

Siraya

Papora

Hoanya

all other An lgs

*pitu

*walu

*Siwa

Figure 1. Higher An phylogeny based on three characters, with innovations

(gray area) at each node

PAn

Pituish

Enemish

Walu-Siwaish

Muish

Luilang

Pazeh

Saisiat

Atayalic

Thao

Favorlang

Taokas

Papora

Hoanya

Siraya Tsouic

Paiwan

Rukai

Puyuma

Amis

Bunun

NE-Form.

Kavalan +

Ketagalan

FATK FAMP

TK PMP

pitu '7' enem '6'

CawiN

'year'

walu '8'

Siwa '9'

pang-V > N

instr

-mu '2sg-gen.'

manuk 'bird'

Blust 1999

Li 2001

Blust 1977a

Blust 1995a

Blust 1995b

Table 4: Enriched higher An phylogeny based on basic-lexical and

morphological innovations (gray area at bottom). Taxon names are in bold type.

To the exception of the NE-Formosan languages Kavalan and Ketagalan, no claim is made

that the languages whose names are in the same column (for instance Luilang, Pazeh and

Saisiat) form a taxon. Dotted arrows indicate an oversea migration. 'FATK' = Formosan

Ancestor of Tai-Kadai; 'FAMP' = 'Formosan Ancestor of Malayo-Polynesian'.

PAn PMP Tai Lakkia

die

maCay matay ta:i1 plei1

eye

maCa mata ta1 pla1

bird

manuk nok8 mlok7

Table 5: regularity of sound correspondences in some An and TK basic

vocabulary items

Buyang PAn MP

die

ma0 tɛ54 maCay matay

eye

ma0 ta54 maCa mata

bird

ma0 nuk11 qayam manuk

ma0 ðu312 _____ walu

head

qa0 ðu11 quluh quluh

louse

qa0 tu54 kuCu kutu

fart

qa0 tut54 qetut

raw

qa0 ʔdip54 qudip

bear (n.)

ta0 mɛ312 Cumay

cover (v.)

ta0 qup11 WMP ta(ŋ)kup

Table 6. Disyllabic An words in Buyang

Buyang PAn PMP

ku54 -ku -ku

thou

ma312 -Su -mu

Table 7. Personal pronouns of An and Buyang

Buyang PAn PMP

ɕa54 duSa

tu54 telu telu

pa54 Sepat e(m)pat

ma312 RaCep lima

nam54 _____ enem

tu312 _____ pitu

ma0 ðu312 _____ walu

va11 _____ siwa

put54 sa-iCit puluq

Table 8. Numerals of An and Buyang

PMP P-Kra (Ostapirat 2000)

five

lima r-ma A

six

enem X-nəm A

seven

pitu t-ru A / C-tu A26

eight

walu m-ru A

nine

Siwa s-ɣwa B

ten

puluq pwlot D

2sg

-mu mǝ A/B

bird

manuk ɲok D

Table 9: Post-PAn lexical innovations and proto-Kra

26 see fn. 25.

Figure 2. The An settlement of Taiwan with the MP and TK migrations

[a]: The pre-Austronesians, from NE China, expand southward along the SE China seaboard

in the 5th and early 4th millennia BCE: they cultivate rice, foxtail millet, exploit marine

resources, practice tooth evulsion. [b]: the Nanri and Pingtan islands, from which the top of

Mt Xueshan (3884 m., at center of 200-km radius visibility circle) can be seen,27 are

reached. From there one group crosses to Taiwan c. 3500 BCE, while [c] the rest continues

expanding in a SW direction towards the Pearl River Delta. [1]: location of earliest An (PAn-

speaking) settlements on Taiwan. [2]: location of Pituish, [3] location of Enemish, [4] location

of Walu-Siwaish, [5] location of Muish, [6] Tai-Kadai migration, [7] Malayo-Polynesian

migration, c. 2000 BCE.

27 I thank Christophe Coupé who in May 2004 calculated the visibility distance of the Xueshan using a

formula used by the French Navy. That the top of the Xueshan is visible from these islands has since

been confirmed to me by Prof. Tsang Cheng-hwa, from personal experience (p. c., June 2004).

Reconstructing the case-marking and personal pronoun systems of Proto Austronesian

Chapter

Jan 2006

Malcolm Ross

Contrasting modes of cultural evolution: Kra-Dai languages and weaving technologies

Preprint

Full-text available

Aug 2024

We investigate and compare the evolution of two aspects of culture, languages and weaving technologies, amongst the Kra-Dai (Tai-Kadai) peoples of southwest China and southeast Asia, using Bayesian Markov-Chain Monte Carlo methods to uncover phylogenies. The results show that languages and looms evolved in related but different ways, and bring some new insights into the diaspora of the Kra-Dai speakers across southeast Asia. We found that the languages and looms used by Hlai speakers of Hainan are outgroups in both linguistic and loom phylogenies, and that the looms used by speakers of closely related languages tend to belong to similar types. However, we also found differences at a deep level both in the details of the evolution of looms and languages, and in their overall patterns of change, and we discuss possible reasons for this.

A Voice System in Search of an Identity: The Multiple Functions of the Patient Voice Construction in Formosan Languages *

Article

Full-text available

Apr 2024

Shuanfan Huang

Philippine-type languages, including Formosan languages spoken in Taiwan, are known to lack the grammatical category of subject representing convergence of topic, actor and pivot and do not have a pivot system governed by the exigencies of topicality and linkage patterns under coreference. In this study we presented the Formosan solution to 'voice' by undertaking a systematic examination of patient voice constructions in four Formosan languages. Three distinct discourse functions of patient voice constructions were distinguished based on evidence from discourse linkage patterns, namely active transitive, notional passive, and pragmatic inverse. The PV constructions in Formosan languages are not true voice constructions in the traditional sense, since PVs in these languages are neither active, nor inverse, nor passive, precisely because they can be all of them, given appropriate discourse context. These empirical findings pose a challenge to the mainstream views on voice marking and call for a rethinking of the typology of the voice systems in the world' languages.

Genetic formation of Sui populations in southwest China

Article

Full-text available

Feb 2024

The Sui people living in Guizhou province have a unique ethnic culture and population history due to their long-time isolation from other populations. To investigate the genetic structure of Sui populations in different regions of Guizhou, we genotyped 89 individuals from four Sui populations using genome-wide single nucleotide polymorphisms arrays. We analyzed the data using principal component analysis, ADMIXTURE analysis, f-statistics, qpWave/qpAdm, TreeMix analysis, fineSTRUCTURE, and GLOBETROTTER. We found that Sui populations in Guizhou were genetically homogeneous and had a close genetic affinity with Tai-Kadai-speaking populations, Hmong-Mien-speaking Hmong, and some ancient populations from southern China. The Sui populations could be modeled as an admixture of 33.5%-37.9% of Yellow River Basin farmer-related ancestry and 62.1%-66.5% of Southeast Asian-related ancestry, indicating that the southward expansion of northern East Asian-related ancestry influenced the formation of the Tai-Kadai-speaking Sui people. Future publications of more ancient genomics in southern China could effectively provide further insight into the demographic history and population structure of the Sui people.

‘Liangdao Man’ from Physical and Cultural Anthropological Perspectives

Chapter

Apr 2025

Chung-yu Chen

The end of the Ice Age marked the beginning of a new era—the Holocene epoch—which begins ca.

Formosan languages and Proto-Austronesian morphology

Chapter

Jan 2024

Dialek Melayu Sabah sebagai Penanda Identiti Orang Sabah

Chapter

Full-text available

Dec 2023

Chapter 5. A single origin of numeral classifiers in Asia and the Pacific: A hypothesis

Chapter

Dec 2023

Linguists have long been interested in systems of nominal classification due to their diverse functions as well as cognitive and cultural correlates. Among others, ongoing research has focused on semantic, functional and morphosyntactic properties of complex systems such as co-occurring gender and numeral classifiers. Such approaches have typically focused on the languages of north-western South America and Papua New Guinea. This volume proposes to fill in a gap in existing research by focusing on Asia, based on case studies from languages belonging to a wide range of families, i.e., Austroasiatic, Austronesian, Dravidian, Hmong-Mien, Indo-European, Mongolic, Sino-Tibetan and Tai-Kadai as well as the language isolate Nivkh. Gender and classifiers in these languages are approached within several different perspectives, i.e., functional, typological and diachronic, thus revealing complex patterns in their lexical and pragmatic functions as well as origin, development and loss. Describing and analysing such properties is a unique and innovative contribution of the volume.

Taiwan Plains Austronesian Latin scripts and documents from the 17th to the 19th century: A bibliographical survey

Article

Dec 2023

Chung-pui Tai

In this paper, we aim to describe the evolution, from birth to extinction, of the Latin scripts of the Taiwan Plains Austronesians, which record the Siraya, Taivoan, Makatau, Favorlang and Basay languages, based on a comprehensive bibliographical survey of the documents, surviving and lost, written in these scripts. We first discuss the background of, motivations for and process of creating the scripts. Using the Siraya script as an example, the paper demonstrates how different usage layers can affect the spelling and document format of a script. From a survey based on published Dutch and Spanish archival materials and other records from the 17th to the 19th century, we compile a total of 51 items of Plains Austronesian Latin documents. A comparison of these items reveals the text formation process of the documents. We suggest that the contribution of Daniël Gravius in the Siraya translation of the Gospels of Matthew and John and the Christian formulary may not have been significant, although he was often later credited as the author. The survey also provides clues when searching for ‘lost’ items, such as the Siraya ABC Book. In addition, we collate Plains Austronesian Latin script accounts in Chinese records. These provide strong evidence of the use of Latin script in Makatau villages during the Qing dynasty. The records also reveal that during the 18th century, there were Plains Austronesian people who knew the Latin script, called ‘teachers’. It may have originated from the schoolteachers of the Dutch colonial period. After the Dutch withdrew, these ‘teachers’ became responsible for the villages’ accounting and administrative documents. The education of Plains Austronesian children was then conducted by schools established in the Qing dynasty, with Han Chinese teachers recruited to teach the children Chinese classics.

Linguistic atlas of Asia and Africa III

Book

Full-text available

Nov 2023

[Full text available at the doi above] This volume, Linguistic Atlas of Asia and Africa, Volume III, is a sequel to Linguistic Atlas of Asia, Hituzi Syobo, Tokyo, 2021. This volume covers 'system of 'sibling' terms', 'alignment', and 'numeral systems'. In addition to these chapters, the volume contains bibliographical information on the data sources covering the three volumes of Linguistic Atlas of Asia and Africa.

Contraintes de structures et liberté dans l'organisation du discours. Une description du mwotlap, langue océanienne du Vanuatu

Thesis

Full-text available

Dec 2001

Alexandre François

(Doctoral thesis, in French) *English abstract* _Structural constraints and freedom in the elaboration of speech: A description of Mwotlap, an Oceanic language of Vanuatu_ Mwotlap (Motlav), an unwritten Austronesian language belonging to the Oceanic subgroup, is spoken by about 1800 people living in northern Vanuatu – Melanesia, South Pacific. This description of Mwotlap grammar addresses various domains and issues, all of which are topics relevant to current functional and typological linguistics: phonology and morphology; syntactic categories; reference tracking, spatial deixis, possession and quantifiers; verb serialisation and valency; aspect and mood categories; discourse pragmatics and speech acts. Each grammatical structure is described synchronically, and situated historically. Many cases of grammaticalisation, and other processes of syntactic and semantic change, thus come to light. Under the pressure of the cognitive and structural constraints active in the speaker's mind, the power of linguistic innovation may even give rise to spectacular upheavals. _____ *Français* _Contraintes de structures et liberté dans l'organisation du discours. Une description du mwotlap, langue océanienne du Vanuatu_ Langue de tradition orale parlée par 1800 personnes au nord du Vanuatu (Mélanésie, Pacifique), le mwotlap (motlav) appartient à la branche océanienne de la famille austronésienne. Au fil d'une description grammaticale aussi complète que possible, le mwotlap nous permet d'aborder diverses problématiques dans l'esprit de la linguistique générale et typologique : phonologie, morphologie ; catégories syntaxiques, opposition verbo-nominale ; construction de la référence, déixis, possession et quantification ; valence, diathèse et séries verbales ; opérations aspecto-modales ; pragmatique du discours et argumentation. Chaque fait de langue est à la fois décrit en synchronie, et replacé dans son contexte historique. Se dessinent ainsi de multiples parcours de grammaticalisation, des processus de réanalyses formelles ou sémantiques ; et parfois, sous la pression des diverses contraintes cognitives ou structurelles, la puissance de l'innovation conduit à de spectaculaires bouleversements. François, Alexandre. 2001. Contraintes de structures et liberté dans l'organisation du discours. Une description du mwotlap, langue océanienne du Vanuatu [Structural constraints and freedom in speech elaboration: A description of Mwotlap, an Oceanic language of Vanuatu]. Doctoral thesis in Linguistics, Université Paris-IV Sorbonne. 3 volumes (1078 pages, 472,000 words).

Grammar Notes on Siraya, an Extinct Formosan Language

Article

Full-text available

Dec 1997

Karl Alexander Adelaar

This is an attempt to unravel the grammar of a gospel text in Siraya, an extinct West Formosan language. It includes a discussion of the historical setting, the spelling, phonological features, function words, relation markers, morphosyntax of the verb, and the use of deictic verbs as prepositions. Regarding verbal morphosyntax, special attention is given to case marking suffixes, verbal classifiers, compound verbs, and anticipating sequences. Verbal classifiers are lexical elements prefixed to a root (a verb, adverb, or noun) with which they constitute the overall meaning of the resulting verb (mattäy- 'talking, saying' + vli 'reciprocating, doing in return'→ mattäy-vli 'to answer'; mattäy- + rĭx 'mind' → mattäy-rĭx 'talk to oneself'). Compound verb constructions are constructions in which auxiliaries assume the functions of adverbs in English. These auxiliaries form an open class and are in fact the head of the verb phrase, as they carry most of the marking. An anticipating sequence is an element of a verb that is prefixed to the preceding auxiliary.

3. Some current issues in Austronesian linguistics

Chapter