Introduction

1“Wallacea” typically refers to a zoogeographical area constituting a transition zone between Sundaland (the Malay Peninsula, Sumatra, Borneo, Java, and Bali) and Sahul (Australia and New Guinea) (see Map 1, Dickerson 1928). Biological Wallacea includes Sulawesi, Lombok, Sumbawa, Flores, Sumba, Timor, Halmahera, Buru, Seram, and many smaller islands of eastern Indonesia and independent Timor-Leste (Map 2). Defined by its mixing of faunal types from the Southeast Asian and Australian areas, it is demarcated in the West by the Wallace Line and in the East by the Lydekker Line, and cross-cut by the Weber Line. Wallacea is well known as an area of mega-biodiversity with very high numbers of species found nowhere else in the world. For instance, it is home to over 10,000 plant species and more than 1,100 terrestrial vertebrate species, many endemic (Coates & Bishop 1997, Myers et al. 2000, Schulte et al. 2003).

3Yet, until recently our understanding of the variety of languages within Wallacea was limited; our picture of many languages is based solely on rapidly elicited wordlists and sketchy example sentences (see, e.g., Collins 1982 on Maluku languages, or more recently, Hull 2005 on Makasae, Chlenov & Chlenova 2008 on Damar, Price & Donohue 2009 on Ansus) or early missionary grammars written with the categories of Latin and Greek grammar in mind (see, e.g., Geurtjens 1921, Maan 1951). Indeed, just a decade ago Tryon (1995: 12) wrote that the Wallacean region “perhaps remains the least known area in the Austronesian world today.” A surge in the production of modern linguistic descriptions across the region has meant there have been significant advances in knowledge of the details of individual languages, both Austronesian and Papuan.

4Modern syntheses of Wallacean languages have tended to focus at one and the same time on the diversity of morphosyntactic profiles they display (see, e.g., the papers in Ewing & Klamer, eds, 2010) and on their representing a progressive attenuation of the prototypical Austronesian linguistic profile towards a more Melanesian linguistic profile (see, e.g., Himmelmann 2005 for one statement of Austronesian characteristics with a “typical” Austronesian profile versus those with an “adjusted” Melanesian profile). In this paper, I argue for a more nuanced perspective on the languages of Wallacea. Rather than simply seeing Wallacean languages as diverse, I seek to show that despite the diversity there is a coherence to them, and that this coherence is not simply a matter of shared characteristics due to their being marginal Melanesian languages within a transitional zone between linguistic areas. Instead I propose that the languages of Wallacea display a distinct set of linguistic features that set them off both from the Southeast Asian Linguistic Area to the West (island Sunda) as well as the Melanesian Linguistic Area to the East (northern Sahul). The Wallacean Linguistic Area is focused further to the east than Biological Wallacea, consisting of Nusa Tenggara including Timor-Leste, Maluku, the Bird’s Head and Neck of New Guinea, and Cenderawasih Bay (Map 3). The nature and dispersal of the features shared by Wallacean languages, I argue, points to the existence of networks of seafaring agriculturists predating the Austronesian presence inWallacea.

5This paper is structured as follows. Section 2 looks at what a linguistic area is, what the features that best define it are, and what it tells us about the history of speech communities within it. Section 3 overviews the language scene within Wallacea. Section 4 contextualises Wallacea as the western-most region of Papuan linguistic influence within Island Southeast Asia. Section 5 presents the features that define not merely eastern Austronesian and Papuan in general, but specifically the Wallacean area. Section 6 looks at the prehistorical inferences that can be drawn from Linguistic Wallacea. Section 7 looks at the archaeological data and how it correlates with the linguistic data, pointing to Wallacea being a relic of an historical situation in which there were extensive pre-Autronesian interisland contacts and agriculture within Wallacea.

What is a Linguistic Area?

6A linguistic area (or convergence/diffusion area, Sprachbund, affinité linguistique) is a geographical region in which a group of languages, typically from different families, share a set of features in common not as a result of accident or inheritance from a common ancestor, but rather due to historical processes of language contact and convergence (Weinreich 1953, Thomason & Kaufman 1988: 91-97, Aikhenvald & Dixon 2001).

7Since the first identification of the Balkans as an area in which significant copying from one language to another had occurred due to long-standing bi- and multilingualism between speaker groups (Sandfeld 1930), many linguistic areas have been identified around the world. Some of the most well-described and best known are: the South Asian Linguistic Area (Emeneau 1956, 1974, Masica 1976), the Ethiopian Linguistic Area (Leslau 1945, 1952, Hetzron 1975, Ferguson 1976, Tosco 2000), the Meso-American Linguistic Area (Campbell, Kaufman & Smith-Stark 1986, Campbell 1992, van der Auwera 1998), the “Standard Average” European Linguistic Area (Whorf 1956, Haspelmath 1998, 2001), and the Mainland Southeast Asian Linguistic Area (Clark 1992, Matisoff 2001, Enfield 2005).

8With the discovery of more and more linguistic areas worldwide, the topic has generated a vast literature. In particular there is much debate about precisely what criteria must be fulfilled in order for the term linguistic area to be deployed justifiably (recent summaries of the problems include Dahl 2001, Thomason 2001, Stolz 2002, Campbell 2006). Nonetheless there is broad agreement about the features that establish a robust linguistic area. They are:

9Distinctiveness. Accidental similarity between languages can occur where a feature is cross-linguistically widespread and sharing such a feature does not necessarily signal any kind of historical connection between languages. Worldwide some linguistic features are common, while others are rare. As such, a feature that frequently occurs outside of a linguistic area has lower distinctiveness and accordingly provides weaker evidence for a linguistic area. The rarer a feature is cross-linguistically the higher its distinctiveness and the greater its value in defining a linguistic area.

10Consistency. Innovative linguistic features do not spread among the languages of a linguistic area evenly; some features spread farther than others, and if the feature originates in different places within the area there will inevitably be different patterns of spread. As such, not all languages within a linguistic area need exhibit all the features that are said to define the area. However, the higher the consistency with which languages in an area display a feature, the stronger the support for the area provided by that feature. Conversely, a feature displayed by a smaller proportion of languages in an area has lower consistency and provides weaker evidence for the area than do those of higher consistency.

11Demarcation. Shared structural features that characterize a particular linguistic area do not have to be confined to the area. This is for two reasons. Firstly, a language may have a feature that is inherited from an earlier ancestral language and this feature may be still present in its sister languages outside the area. In that case, the feature is obviously not confined to the linguistic area, but may be still used as diagnostic of the linguistic area as long as it can be shown to have spread widely from the first language(s) which had inherited the feature to other unrelated languages in the area. Secondly, speakers of some languages within a linguistic area are likely to have contacts beyond the boundaries of the area, and by that means there may be some restricted diffusion of the features to languages outside the area. So, the clearer the demarcation a feature shows within a linguistic area (that is the less leakage beyond the area), the stronger the support for the area provided by that feature.

12Cross-familial presence. The languages in a linguistic area need not be unrelated. However, with related languages, distinguishing changes due to drift from changes due to contact may be very difficult. Where languages are unrelated, it is easier to establish which features are the result of borrowing and diffusion rather than inheritance. It follows that the more unrelated families a feature appears in, the clearer it is that the feature has diffused and the better it is as a diagnostic for area.

The Wallacean Language Scene

15The Austronesian languages have their homeland in Formosa (modern-day Taiwan) and spread throughout Island Southeast Asia and into the Pacific from around 4500 BP (Pawley and Ross 1993, Pawley 2003). By 3500 BP, Austronesians appear to have reached Wallacea (Bellwood 2006). At exactly what level the more than one hundred Austronesian languages of Wallacea are related to one another is disputed. Blust (1982/1983, 1993) maintains that the Austronesian languages of Wallacea all belong to the Central-Eastern Malayo-Polynesian (CEMP) subgroup. This subgroup, he asserts, itself divides into two subgroups, Central Malayo-Polynesian (CMP) and Eastern Malayo-Polynesian (EMP). The EMP subgroup, as defined by Blust, divides into two subgroups again, South Halmahera-West New Guinea (SHWNG) and Oceanic (Oc). The CMP subgroup takes in the Austronesian languages of Timor-Leste and East Nusa Tenggara as far east as Bima on the eastern half of Sumbawa Island, as well as those of southern and central Maluku. The SWHNG subgroup includes the Austronesian languages of northern Maluku (south Halmahera and Raja Ampat) and Cenderawasih Bay as far as Warembori at the mouth of the Mamberano river, as well as Irarutu on the Bomberai Peninsula. Oceanic languages are found east of Warembori and fall outside of what we will define as Wallacea.

Linguistic Melanesia

17Understanding Linguistic Wallacea and its place at the cross-roads of the Austronesian and Melanesian worlds means having a broader picture of the linguistic patterns that define Austronesian and Papuan languages. In this section I show how Austronesian languages converge on the linguistic norms of Papuan languages the closer they are to New Guinea. The progressive replacement of individual Austronesian features by Melanesian ones results in concentric circles of linguistic features clustering around New Guinea, defining a sphere of Papuan influence which I shall refer to as “Linguistic Melanesia.” Austronesian languages caught within this sphere of influence I will refer to as “Melanesian Austronesian” languages. I will describe only a few of the most well-defined features of Linguistic Melanesia, progressing from largest to smallest in terms of the extent of their appearance in Austronesian languages around New Guinea.

Possessive Classification

19Possessive classification takes many forms in Melanesian Austronesian and has been assigned many different labels depending on the exact semantics of the possessive classes and the form(s) marking the possessive relationship (e.g., alienable-inalienable, dominant-subordinate, direct-indirect etc.). For instance, in Samoan, one class of nouns (illustrated by paopao “canoe” in 1a) is possessed with the possessive morpheme o, while another class (illustrated by naifi “knife” in 1b) is possessed with the possessive morpheme a. Nouns from the o-class cannot be possessed with the a possessive and vice versa.

Samoan (Austronesian, Samoa, Polynesia; Lynch, Ross & Crowley 2002: 43)

(1)	a.	le	paopao		o	Tavita
		art	canoe		poss	Tavita
	“Tavita’s canoe”
	b.	le	naifi	a	le	tule’ale’a
		art	knife	poss	art	young.man
	“the young man’s knife”

20In Ujir one class of nouns (illustrated by mata- “eye” in 2a) is possessed with a possessive suffix, while the other class (illustrated by juma “house” in 2b) is possessed with a free possessive morpheme.

Ujir (Austronesian, Aru Islands, Indonesia; own fieldnotes)

(2)	a.	mata-ng
		eye-1sg.poss
		“my eye”
	b.	kanang	Juma
		1sg.poss	House
		“my house”

21Possessive classification is typical of many Papuan languages, east and west of New Guinea. This is illustrated by Sulka and Bunaq. In each language, we see there are distinct forms for expressing the possessive relationship depending on the possessed noun.

Sulka (isolate, New Britain, PNG; Tharp 1996: 80)

(3)	a.	ko-nan
		1sg.poss-mother
		“my mother”
	b.	kua-rik
		1sg.poss-house
		“my house”

Bunaq (Timor-Alor-Pantar, Timor, Indonesia/Timor-Leste; own fieldnotes)

(4)	a.	n-iol
		1sg.poss-voice
		“my voice”
	b.	nie	zo
		1sg.poss	Mango
		“my mango”

22Mapping the distribution of possessive classification in Austronesian languages we see a clear skewing of this feature towards Melanesia. There are outliers to the west: Puyuma in Taiwan and Bidayuh languages in West Borneo (Adelaar 2005:25). They do not, however, diminish the clear dominance of the feature around New Guinea and out into Oceania. On the New Guinea mainland, possessive classification is often absent in Papuan languages of the central highlands and sometimes absent in those along the north coastal region. However, possessive classification is much more often present in Papuan languages than it is absent and is found in most Papuan families extending the full length of the Papuan area.

Complex Numerals below Ten

23The composition of numerals between “six” and “ten” becomes complex in many Austronesian languages in the proximity of New Guinea (Map 7). A complex numeral is one that is composed of other numerals by means of an arithmetical operation between component numerals (thus, an “additive” complex numeral may be composed of, e.g., 6+1 = 7; “subtractive” complex numeral, e.g., 10−2 = 8, and “multiplicative” complex numeral, e.g., 3x2 = 6).

24Proto-Austronesian (PAn), the common ancestor of all Austronesian languages, and Proto-Oceanic (POc), PAn’s most significant daughter and the most immediate common ancestor of all Oceanic languages, both had simplex numerals (numerals not composed of other numerals) and these are reflected in many modern day Austronesian languages across the full sweep of the Austronesian area. Table 1 sets out some examples.

Table 1 – Examples of Austronesian simplex numerals

	PAn	POc	Cebuano (Austronesian, Philippines)	Mentawai (Austronesian, Barrier Islands)	Longgu (Austronesian, Solomon Islands)
1	*esa/isa	*ta-sa/(sa)-kai	usa	sara	eta
2	*duSa	*rua	duha	dua	rua
3	*telu	*tolu	tulu	telu	olu
4	*Sepat	*pat(i)	upat	epat	vai
5	*lima	*lima	lima	lima	lima
6	*enem	*onom	unum	enem	ono
7	*pitu	*pitu	pitu	pitu	viu
8	*walu	*walu	walu	balu	walu
9	*Siwa	*siwa	siyam	siba	siwa
10	*sa-puluq	*sa[-ŋa]-puluq	napuluq	sapulu	tangavulu

25By contrast, in many Austronesian languages on and around New Guinea we find complex numerals of many different kinds. Table 2 sets out a few examples. Square brackets indicate the arithmetical operation used to form the numeral. We see that a single language may use more than one kind of operation to form its complex numerals and that the complex numerals need not be used for all numerals between six and ten.

Table 2 – Examples of Melanesian Austronesian complex numerals

	Rongga (Austronesian, Flores)	Mambae (Austronesian, Timor)	Onin (Austronesian, Bomberai Peninsula)	Pak (Austronesian, Admiralty Islands)	Lewo (Austronesian, Vanuatu)
1	(e)sa	id	sa	dih	taaŋa
2	ɹua	ru ~ rua	nuwa	huo	lua
3	telu	teul ~ tel	teni	loh	telu
4	wutu	fat ~ pat	fāt	alor	hat vari
5	lima	lim	nima	nuron	lima
6	lima esa [5+1]	lim nain ide [5+1]	nem	wono	orai [5+1]
7	lima ɹua [5+2]	lim nai rua [5+2]	tara sa [6+1]	arloh [10-3]	olua [5+2]
8	ɹua mbhutu [2x4]	lim nai telu [5+3]	tara nuwa [6+2]	arhuo [10-2]	orelu [5+3]
9	tara esa [10-1]	lim nai pata [5+4]	sa puti [10-1]	ardih [10-1]	ovari [5+4]
10	sambulu	sakul ~ sagul	pusua	songo	lua-lima [5x2]

26When we map the distribution of Austronesian languages with complex numerals for between “six” and “ten,” we again observe a striking clustering within Melanesia. Complex numerals are recurrent eastwards from Flores to New Guinea and persist throughout the Bismarck Archipelago and into Vanuatu and New Caledonia. Outliers are limited to the following six groups: (i) several Formosan languages with additive and multiplicative numerals; (ii) Malayo-Chamic subtractive numerals “eight” and “nine” (the large number of points on Map 7 merely reflects the wide dispersal of this subgroup’s daughter languages), representing a single innovation in their common ancestor; (iii) Enggano with its base-5 system; (iv) South Sulawesi subtractive numerals “eight” and “nine,” also representing a single innovation in their common ancestor; (iv) Makasarese additive “seven”; (vi) Ilongot quinary numerals for “six” through “nine,” and; (vii) a few Micronesian languages with subtractive numerals for “eight” and/or “nine.” By contrast there over 50 distinct innovations of complex numerals in Austronesian languages of the Melanesian area (Schapper & Hammarström 2013, Blust 2008).

27Papuan languages have notoriously varied counting systems including many rare types such as restricted systems and body-tally systems (Hammarström 2010). Crucially, while the actual nature of the systems are various, Papuan languages typically possess complex expressions for “six” to “ten” (Lean 1992, Galis 1955, 1960, Schapper & Klamer 2014).

Order of Noun and Numeral

29In most Austronesian languages outside of Melanesia, the numeral normally precedes the noun, as shown for Tagalog in (5).

Tagalog (Austronesian, Philippines; Gil 2013)

(5)	dalawa=ng	aso
	two=lig	Dog
	“two dogs”

30By contrast, in Austronesian languages further east, the numeral follows the noun. This is shown for Luang, Mbula and Araki.

Luang (Austronesian, Southwest Maluku, Indonesia; Taber 1999)

(6)	muanke’a	riy	wo’itu
	male	person	Seven
	“seven men”

Mbula (Austronesian, Umboi Island, PNG; Bugenhagen & Bugenhagen 2007)

(7)	me	tio	ru
	dog	1sg.poss	Two
	“my two dogs”

Araki (Austronesian, Vanuatu; François 2002)

(8)	racu	rapudo	rolu
	man	clf	Two
	“two men”

31The vast majority of Papuan languages have the numeral following the noun (over 90% in the sample of Dryer 2013a). Unsurprisingly, we again see a clear skewing of noun-numeral order in Austronesian languages towards Melanesia. There are only three outliers to the west: Ma’anyan, Western Cham and Pawnee. Otherwise the feature extends unbrokenly from Timor to New Guinea, and is almost invariably present in Austronesian languages of the New Guinea mainland and further as far as Vanuatu. The Austronesian languages of the Bismarck Archipelago, New Caledonia, Micronesia and Polynesia are in their majority numeral-noun order languages, but each region has a few exponents of the Melanesian noun-numeral order.

The Velar Nasal /ŋ/

32The next feature with a striking distribution in the Melanesian area is the absence of the velar nasal phoneme (Map 9). This is the sound /ŋ/, written in languages such as English and Indonesian with the letters ng as in English sing or Indonesian tangan.

33The vast majority of Austronesian languages have the velar nasal in their consonant phoneme inventories (more than 82% of Donohue et al.’s 2013 sample). Around New Guinea there is a clear concentration of Austronesian languages lacking the velar nasal. The lack of the velar nasal in Austronesian languages begins in Timor, moves through South-west Maluku, Central Maluku and the languages of the Bomberai Peninsula and Cenderawasih Bay. On the north coast of New Guinea roughly half the Austronesian languages lack the velar nasal, while on the Bird’s Tail of New Guinea almost all Austronesian languages lack it. A smattering of Austronesian languages without velar nasals are then found in the Bismark archipelago and northern Vanuatu. Beyond this, we find only two Austronesian outliers to the West (Nias and Enggano) in the Barrier Islands off the west coast of Sumatra, and five outliers in remote Polynesia.

34Papuan languages are divided roughly in half in terms of the velar nasal: present in 221 Papuan languages, and absent in 293 (57%). Viewed in isolation then the lack of the velar nasal in Melanesia does not seem remarkable; however, taking a larger areal perspective encompassing Mainland Southeast Asia and Australia where the velar nasal is near-universally present, the absence of phonemic ŋ becomes a highly marked feature of the Melanesian area (Anderson 2013). What is more, velar nasal-lacking Papuan languages are concentrated in the maritime and coastal regions to the West of New Guinea and from the central northern region down off the Bird’s Back and into the Bird’s Tail region, precisely the regions where velar nasal-lacking Austronesian languages are most found.

Order of Noun and Genitive

35The most famous and first recognised feature to define the Melanesian area is the so-called “reversed genitive” (Brandes 1884) (Map 10). This feature looks at the relative order of noun and possessor (genitive), as in English tail of the dog [noun preceding genitive] versus the dog’s tail [genitive preceding noun]).

36The typical Austronesian pattern outside of Melanesia has the genitive following the noun, as shown for Mualang and Boumaa Fijian in (9) and (10).

Mualang (Austronesian, West Kalimantan, Indonesia; Tjia 2007)

(9)	kisah	antu
		Ghost
	“story of a ghost”

Boumaa Fijian (Austronesian, Fiji, Polynesia; Dixon 1988)

(10)	a	liga-i	Jone
	art	hand-lig	John
	“John’s hand”

37By contrast, as with numeral-noun order, in Austronesian languages on or close to New Guinea, the order of the genitive is “reversed” and typically precedes the noun, as in Tokodede and Sudest.

Tokodede (Austronesian, Timor-Leste; own fieldnotes)

(11)	Luis	ni	dagana
	Louis	poss	Head
	“Louis’ head”

Sudest (Austronesian, Tagula Island, PNG; Anderson 2007)

(12)	anawang-gu	langgogwe
	ear-1sg.poss	Wax
	“my ear’s wax”

38Almost all Papuan languages have the noun following the genitive (95% in the sample of Dryer 2013b; there is an areally very restricted pocket of Papuan languages with noun-genitive order on the north-central coast). Austronesian languages with the reversed genitive are heavily concentrated in Melanesia; two languages in Micronesia, Ulithian and Puluwat, are the only outliers. The feature is dispersed throughout eastern Indonesia and into New Guinea where it is consistently present in Austronesian languages, but not beyond into the Bismarck Archipelago or further afield into the Pacific.

Order of Verb and Negator

39In this section, we are concerned with whether a language has a basic negative word (such as English not) that precedes or follows the verb it negates (Map 11). The order of the verb and negator can often be observed to switch in Austronesian languages within the Melanesian area.

40The typical pattern in Austronesian languages is to have the negator preceding the verb. This is illustrated in the following languages, as in (13) and (14).

Kemak (Austronesian, Timor, Indonesia/Timor-Leste; own fieldnotes)

(13)	Ua	tai	bue
	3sg	neg	sleep
	“He didn’t sleep.”

Ughele (Austronesian, Solomon Islands; Frostad 2013: 63)

(14)	...	mana	kai	doghor-i	rie	ka	ru.
		but	neg	see-3pl		3pl	part
	two
	“... but (their mother) didn’t see them.”

41Near New Guinea this pattern sporadically gives way to the negator following the verb, as in (15) and (16). The post-verbal negator in Austronesian languages begins in the islands just to the west of Timor (but skips Timor itself), moves through Central Maluku and the languages of Bomberai Peninsula and Cenderawasih Bay. Moving east, the postverbal negator is found in the Austronesian languages along the north coast of New Guinea stopping in the Madang district and West New Britain. Outliers are limited to two in our sample: Cham in the west and Vanikoro in the east (though more indepth sampling may reveal that postverbal negation extends as far as the latter).

Sawu (Austronesian, Sawu Island, Indonesia; Walker 1982: 47)

(15)	Wae	d’o	ke	ta	pe-hianga	roo.
	want	neg	part	asp	recp-friend	3pl
	“They do not want to be friends.”

Kaulong (Austronesian, western New Britain, PNG; Ross 2002: 401)

(16)	Eiak	men	ngi-n	som
	stone.axe	delim	tooth-3sg	neg
	“The axe isn’t sharp.”

42This feature in Austronesian languages is particularly striking as it is not seen under normal circumstances to co-occur with the SVO and VSO word orders typical of Austronesian, but correlates much more with SOV word order as is found in most Papuan families (Dryer 2013c). Still, compared with the other Melanesian features examined here, post-verbal negation has only diffused into Austronesian languages to a much more limited extent. This can be explained by the fact that Papuan languages are themselves very diverse in how they encode negation. In the sample considered here, only 43% of Papuan languages have postverbal negation. Apart from the preverbal and postverbal negators that we have looked at, there are also many other strategies evidenced including affixal negation and tonal negation (see Dryer 2013c for a full account of the other strategies). It should be noted that this variety in Papuan languages does not diminish the areality of the feature in Austronesian languages, since a Papuan language is statistically still more likely to have post-verbal negation than an Austronesian language (Reesink 2002, Vossen & van der Auwera 2013).

Summary

43In the previous sections we have seen that Austronesian languages converge on the linguistic norms of Papuan languages the closer they are to New Guinea. The progressive attenuation of individual Austronesian features to Melanesian ones results in concentric circles of linguistic features clustering around New Guinea, defining what I call “Linguistic Melanesia” which extends well beyond New Guinea. Map 12 sets out the area on the basis of the features considered here. We see Linguistic Melanesia begins in the area of Flores-Sumba-Timor reaches through New Guinea and into the Bismark Archipelago, and concludes in Vanuatu-New Caledonia. Interestingly, whilst in the west Linguistic Melanesia correlates roughly with the extent of Papuan languages, in the east the “effects” of Melanesian influence extend far beyond extant Papuan languages, but reach into Vanuatu and New Caledonia where no Papuan languages are found. In-depth analysis of the distribution of additional features (e.g., the appearance of person-number agreement on verbs, Himmelmann 2005; prenasalisation of plosives, Donohue & Whiting 2011; the absence of passive voice morphology; postverbal aspect markers, Reesink 2009; the use of serial verb constructions, Van Staden & Reesink 2009) will in the future contribute further to understanding and defining Linguistic Melanesia.

44The idea that Papuan languages have influenced the Austronesian languages in their vicinity is nothing new. For example: at least since Brandes (1884) the development of genitive–noun word order in place of the conservative Austronesian noun–genitive order has been attributed to Papuan contact; Galis (1960) already observed that many Austronesian languages to the north of New Guinea have exchanged the ancestral decimal system for quinary systems as found in many Papuan languages. However, the distribution of such features has rarely been looked at, as here, in a single picture with languages from both to the west and the east of New Guinea (notable exceptions are Donohue 2007 and Reesink et al. 2009). What is more, the features have not been brought together to define an area of “Linguistic Melanesia”. Rather, discussion of individual features has typically been scattered throughout the linguistic literature on the region and has remained inaccessible to the non-specialist, non-linguist. In taking a more holistic approach here we set the stage for understanding the place of, and features defining, Linguistic Wallacea.

Linguistic Wallacea

45That such “Papuan” features as we have looked at in the previous section appear for the first time in languages of Wallacea does not define Wallacea as a linguistic area. This is because these features are not areally well demarcated: that is, they are not limited in any significant way either to Austronesian or Papuan languages within Wallacea. This finding is contrary to previous statements of the region as a linguistic area, in particular Klamer (2002), Klamer et al. (2008) and Musgrave (2008). These works fail to look east as well as west when defining the area, instead they at best consider features only in comparison to the Austronesian languages further to the west, that is, outside the Melanesian area. The result is that features that are Melanesian rather than specifically Wallacean have been erroneously used to define the area (see, e.g., Ross’ 2003 and Donohue’s 2004 criticisms of Klamer 2002 or Vossen & van der Auwera 2013’s demonstration of the New Guinea-wide distribution of the negatives discussed in Klamer et al. 2008).

46In this section, I show that there are indeed linguistic features that can be used to define Wallacea as a linguistic area largely to the exclusion of both languages further east and west. Furthermore, the features I identify are for the most part highly distinctive, even though they are not necessarily consistently displayed across the area. All features cross familial lines, and in most cases cross not only the Austronesian-Papuan divide, but also the Papuan-Papuan divide, that is, appear in multiple distinct Papuan families. The cross-Papuan family presence of these geographically delimited features, I suggest, reflects an ancient linguistic area in Wallacea that predates the Austronesian dispersal, but that now includes the Austronesian languages due to the substratal influence of Papuan languages on them: in other words, because previous Papuan language-speaking populations have switched to Austronesian languages.

Semantic Alignment of Verbal Person Markers

Kamang (Timor-Alor-Pantar, Alor, Indonesia; own fieldnotes)

(17)	a.	Na-maitan-si.
		1sg-hunger-ipfv
		“I’m hungry.”
	b.	Markus kawailai-si.
		Markus stumble-ipfv
		“Markus stumbles.”
	c.	Markus na-tak-si.
		Markus 1sg-see-ipfv
		“Markus sees me.”

48Such splits as seen with semantic alignment are not phonologically determined, but are based on syntactic divisions which, often, at least partially reflect semantic features such as activeness, effectedness, control (volition), aspect and so forth (Donohue 2008a). For instance, in Wersing, the verb lailol “walk” normally occurs without a prefix where the subject is a controlling participant actively engaged in the act of walking (18a). It can, however, occur with a subject agreement prefix on the verb where the act of walking is as yet unrealized, with the participant poised to walk (18b).

Wersing (Timor-Alor-Pantar, Alor, Indonesia; own fieldnotes)

(18)	a.	Naida	lailol.
		1sg	walk
		“I’m walking.”
	b.	Naida	ne-lailol.
		1sg	1sg-walk
		“I’m about to get walking.”

49Semantic alignment is relatively infrequent in the languages of the world and therefore highly marked (Siewierska 2013), particularly in many of its manifestations within Wallacea (Donohue 2004: 231). Across the Austronesian and Melanesian worlds, the distribution of semantic alignment is visibly skewed towards Wallacea (Map 13). Within Wallacea semantic alignment appears repeatedly in Papuan and Austronesian languages around Yapen in Cenderwasih Bay and North Halmahera, before continuing into the Austronesian languages of central and eastern Maluku (Aru and Banda) where it is robustly displayed. In southern Maluku it occurs sporadically (Selaru and Luang), and subsequently disappears until reaching central Timor. Here it reappears in Bunaq and then is continued on in the related languages of Alor and Pantar. From there it extends westward through the Austronesian languages of the Solor Archipelago (Kedang, Lamalera, Lamaholot) and then ends in Sumba (Kambera). Outside of Wallacea there is a much more limited occurrence. To the west, there are only three Austronesian outliers, Acehnese, Nias and Mori Bawah. To the east, there are a collection of Papuan languages showing the feature, two Torricelli languages in northern New Guinea, Kewa in the eastern highlands, Anem-Ata of West New Britain, the Baining languages of East New Britain, two languages of north Bougainville, and Lavukaleve in the southeast Solomons. Despite its recurrence to the east, semantic alignment is unmistakably an areal feature of Wallacea (pace Donohue 2004, contra Klamer 2008b) because of its repeated occurrence in Austronesian languages bordering on Papuan languages showing similar systems. Elsewhere we do not find such cross-familial occurrences.

50Amongst the Austronesian languages of Wallacea, the person marking forms themselves are so diverse that they cannot easily be seen to originate in a single ancestral language (Adelaar 2005: 25, Donohue and Grimes 2008). The nature of the semantic alignment in the different Austronesian languages is also varied. For example, in Batuley (as in other Aru languages) there is a strictly lexicalised stative-active split: stative (intransitive) verbs have their subject indexed by a suffix (19a), while active (intransitive or transitive) verbs have their subject indexed by a prefix (19b); no verb can move from its assigned class.

Batuley (Austronesian, Aru Islands, Indonesia; Daigle 2015)

(19)	a.	Karaw-ing.
	thirsty-1sg
	“I am thirsty.”
	b.	Ang	ku-fan	fei	kai	tutui.
		1sg	1sg-fall	from	tree	top
		“I fell from the top of the tree.”

51By contrast, Kambera has a so-called “fluid-S” (Dixon 1994) split whereby the subject of an intransitive verb can be marked by different agreement clitic combinations, each expressing different semantic nuances. Compare the clitic marking on the verb meti in (20). In (20a) the subject of meti is encoded with the proclitic da=, a marking pattern that does not force a specific interpretation of the clause. However, in (20b) and (20c) the subject is encoded with the enclitic =na and =ya respectively, each carrying different mood, modality and aspectual connotations (see Klamer 1998 for discussion).

Kambera (Austronesian, Sumba Island, Indonesia; Klamer 1998)

(20)	a.	Jaka	nda	nyumu,	da=meti=ka	làti.
		conj	neg	you	3pl=die=pfv	in.fact
		“Without you, they would die/they would have died.”
	b.	Mbada meti=na=ka?
		already die=3sg=pfv
		“Is he dead already/has he died already?”
	c.	Jaka	nda	nyumu	meti=ya=ka	làti.
		conj	neg	you	die=3sg=pfv	in.fact
		“Without you, one (generic) would die/have died.”

52Such formal variety suggests that the appearance of semantic alignment in Austronesian languages in the region is the result of erratic diffusion. By contrast, as Donohue (2004 : 231-233) argues, semantic alignment appears to be an inherited trait in the Papuan families of Wallacea, which Austronesian languages in the region acquired either through substratum or contact.

Neuter Gender

53Across Wallacea, strongly semantic gender distinctions between neuter and nonneuter are commonplace (Map 14). The terms “neuter” and “nonneuter” refer here to gender categories in which there is a binary division of referents into classes according to their position on the animacy hierarchy. “Neuter” defines a class of referents lower on the animacy hierarchy (e.g., nonmale, nonhuman, inanimate), while “nonneuter” defines a class of referents higher up the scale (e.g., male, human, animate). For instance, in Maybrat, male (nonneuter) contrasts with non-male (neuter), marked by different verbal prefixes y- (21a) and m- (21b and 21c) respectively. Similary, Tobelo contrasts human (nonneuter) with non-human (neuter) by means of yo- (22a) and i- (22b) respectively.

Maybrat (Papuan, isolate, Bird’s Head, Indonesia; Dol 2007)

(21)	a.	Y-amo	amah.
	3.male-go		house
		“He goes into the house.”
	b.	Fnia	m-amo	aya
		woman 3.nmale-go		water
		“The woman goes to the river.”
	c.	T-tu	aya	m-amo	cerek
		1sg-pour	water	3.nmale-go	thermos
		“I pour water (making it go) into the thermos glass.”

Tobelo (Papuan, North Halmahera, Maluku, Indonesia; Holton 2003)

(22)	a.	Yo-honenge.
		3pl.hum-die
		“They [human] died.”
	b.	I-honenge.
		3.nhum-die
		“It/they [non-human] died.”

54Since such neuter gender distinctions are not typical of either Papuan or Austronesian languages across the board, the limited overall distribution can only be taken to represent an areal feature of Wallacean languages (Schapper 2010). Only four outliers are recorded: (i) the Formosan (Austronesian) languages with their human-nonhuman distinction marked on numerals (this is an inherited feature in these languages and thus is taken to constitute only one instance of the feature); (ii) Palauan (Austronesian) in Micronesia which has a human-nonhuman distinction in its pronouns; (iii) Tolaki (Austronesian) in Sulawesi which has a human-nonhuman distinction in numerous parts of its grammar, given the proximity of Tolaki to Linguistic Wallacea, this may represent “leakage” from the area, and; (iv) Kanum (Papuan) in Southern New Guinea which has an female-nonfemale agreement distinction.

55The area in which neuter gender is found begins with the Papuan languages of the western and northern Bird’s Head (Abun, Maybrat, Moi and Tehit) and Mairasi of the Birds Neck, extending west to the Papuan languages of the North Halmahera family and south to the Papuan languages of the Timor-Alor-Pantar family. Neuter gender in the Austronesian languages of Wallacea connects the disparate Papuan families together: the feature extends around the top of the Bird’s Head from languages such as Biak, Dusner and Waropen in Cenderwasih Bay to northern Maluku islands at the New Guinea beachhead in languages such as Matbat and Ambel of the Raja Ampat islands; from here it is attested broadly throughout Central Maluku, pervades the Aru languages and appears sporadically in the languages of southwest Maluku (e.g., Selaru) before halting in eastern Timor (Naueti), where Papuan languages with similar systems are found.

56It might be thought that the appearance of such systems in many Austronesian languages could represent shared inheritance. However, the forms marking the neuter/non-neuter gender distinction in the different languages cannot be reconciled as being part of a single innovation such as would characterize a sole ancestral proto-language. Compare, for instance, the subject agreement markers for neuter gender in Ujir (23) and Ambel (24). While Ujir only makes a neuter gender distinct in the plural, Ambel makes it in the singular as well. While Ujir -si and Ambel si- are both clear reflexes of the Proto-Malayo-Polynesian 3^rd person plural pronoun *sida (Ross 2006), they are likely to represent separate grammaticalisation events and are not due to common inheritance. This is particularly clear when we take into account the fact that Ujir uses suffixes on the verb, while Ambel uses prefixes, that the look-alike forms do not mark the same gender (animate in Ujir versus inanimate in Ambel), and that the other agreement forms that they occur in a paradigm with cannot be seen as having any shared inheritance.

Ujir (Austronesian, Aru Islands, Indonesia; own fieldnotes)

(23)	a.	Tamata	bangi-si.
		person big-3pl.an
		“The people are big.”
	b.	Juma	bangi-di.
		house	big-3pl.inan
		“The houses are big.”

Ambel (Austronesian, Waigeo Island, Indonesia; Laura Arnold pers. comm.)

(24)	a.	N-amdol.	b.	L-amdol.
		3sg.an-fall		3pl.an-fall
		“S/he falls.”		“They fall.”
	c.	An-amdol.	d.	Si-amdol.
		3sg.inan-fall		3pl.inan-fall
		“It falls.”		“They [inanimate] fall.”

57In short, the sporadic appearance of the gender distinction, the diversity of morpho-syntactic targets on which the distinction is marked, and the lack of uniformity in the forms expressing the distinction all point to neuter gender as having independently emerged multiple times in the Austronesian languages of Wallacea.

58Across the Papuan languages too, neuter gender distinctions take very different forms and appear to differ in their antiquity in the various groups. In the North Halmahera languages, for instance, the non-human gender form in the plural goes back to the shared ancestor, Proto-North Halmahera (Donohue 2008b). By contrast, in the Timor-Alor-Pantar languages, neuter gender appears in only a few languages and then in very different forms: on demonstratives in Bunaq (25); on verbal prefixes in Abui (26), and on some numeral and quantifying verb forms in Eastern Timor languages (exemplified on the basis of Makalero in 27).

Bunaq (Papuan, TAP, Timor, Indonesia/Timor-Leste; own fieldnotes)

(25)	a.	en	bari	b.	zo	bare
		person	this.an		mango this.inan
		“this person”			“this mango”

59Abui Atengmelang (Papuan, TAP, Alor, Indonesia; Kratochvíl pers. comm.)

(26)	a.	da	fokung	ha-nei	Mangrawali	ha-rul-i	mi
		3	gong	3.poss-name	M.	3-take.off-pfv	take
		he-tan-i.
		3.hum-release.pfv-pfv
		“He took off the gong named Mangrawali and gave it to them.”
	b.	ha-táng			baai	we-ong!
		3.poss-hand			also	3.nhum-make
		“Make his hands also.”

Makalero (Papuan, TAP, Timor, Timor-Leste; Huber 2011)

(27)	a.	amuni	ki=rial	b.	dila	ki=roual=ini
		person	attr=many.hum		frog	attr=many.nhum=art
		“many people.”			“the many frogs.”

60Such patchy, disconnected appearances of neuter gender in the TAP languages, as well as in the Austronesian languages in southern Wallacea contrasts with the proliferation of instances on and around the Bird’s Head. With this, that the Bird’s Head also shows the greatest genetic diversity in languages with neuter gender goes to suggest that neuter gender in Wallacea has the Bird’s Head as the centre of feature’s dispersal, and that the gender distinction in Austronesian languages is ultimately due from (a) Papuan source(s).

^#muku “banana”

62Various scenarios could be used to explain the presence of this banana term in the languages. Chance similarity is typical of sporadic lookalikes in far flung languages. In the case of ^#muku such an explanation is extremely unlikely given that we are dealing here with dozens of instances of a word in a single region. Shared inheritance, whereby languages have inherited the word from a shared ancestor, can also be dismissed as we know for certain that not all the languages involved are related. That leaves borrowing, either between languages or from some external source. As Denham and Donohue (2009) argue, “given that ^[#]muku does not occur in Austronesian languages away from the area of Papuan contact, a Papuan source can be deduced.” This reasoning is supported by the recent reconstruction of *mugu(l) to Proto-Timor-Alor-Pantar, the shared ancestor of the Papuan languages in the southeast corner of Wallacea (28 presents reflexes in modern TAP languages).

Reflexes of Proto-Timor-Alor-Pantar *mugu(l) “banana” (Schapper et al. 2014)

(28)	Bunaq mok; Makasae muʔu; Makasae muʔu; Fataluku muʔu; Oirata muː; West
	Pantar magːi; Teiwa muħui; Kaera mogoi; Blagar mol; Adang mɔʔɔi; Klon məgol;
	Kamang moːi; Wersing mulul; Sawila maːka.

63Since the breakup of proto-Timor-Alor-Pantar, the language from which all the Papuan languages of southeast Indonesia originate, predates the Austronesian arrival in the region (Schapper 2011b: 182), the reconstruction of this banana term points to it being a very old feature within Wallacea and not one dispersed relatively recently through Austronesian languages. This is confirmed by the early appearance of the term in Austronesian subgroups in the region. For instance, *muku reconstructs to the Proto-Aru subgroup of the Austronesian family (Rick Nivens pers. comm.). The appearance of ^#muku reflexes in three other Papuan groups in Wallacea confirms the early dispersal of the term and further points to Wallacea, as the region not only of the most appearances but also of greatest genetic diversity for the term’s coverage, being its centre of dispersal.

Synchronic Metathesis

Helong (Austronesian, West Timor, Indonesia; Bowden 2010)

(29)	a.	Auk	mali.
		1sg	smile
			“I smile”
		Auk	mail	lahin
		1sg	smile	yesterday
		“I smiled yesterday.”

Selaru (Austronesian, Tanimbar, Indonesia; Coward 1990)

(30)	a.	asw
		dog
		“dog”
		askwe
		asw	-ke
		dog-art
		“the dog.”

65Crosslinguistically, synchronic metathesis is a highly unusual process and accordingly has attracted much theoretical attention to the Wallacean languages which display it most robustly.

66Both kinds of metathesis, often in one and the same language, are found in many languages in the southeast corner of Wallacea (Map 16). The greatest concentration of metathesising languages lies in the islands just east of Timor. From here, the feature continues north and west, occurring sporadically in the languages of Tanimbar (see Selaru data in example 26) and Babar as well as western and central Timor (see Helong data in example 29) and Alor. Metathesis outliers are also found scattered through Austronesian languages in the islands to the east of New Guinea, and in several languages of Formosa and the Philippines. Only two Papuan outliers are known, Siane and Gizrra, in eastern New Guinea, both away from Austronesian contact. Due to its areally limited appearance within Wallacea, metathesis is certainly the weakest feature of Linguistic Wallacea in this characterisation. Nonetheless, it can legitimately be regarded as a circum-Timor areal feature due to the density of its occurrences and the fact that it crosses the Papuan-Austronesian divide. In no other region of Austronesia and Melanesia do we find anything like the concentration of metathesising languages as in southeast Wallacea, nor do we find any other region in which Papuan and Austronesian languages both show the feature.

67Although it is likely to be inherited in form and function on some low levels (e.g., amongst the Kisar-Luang languages, Engelenhoven 1995), inheritance cannot be used to explain all the appearances of metathesis in the Austronesian languages within Wallacea. Mambae, for instance, is a metathesising language: many nouns and verbs have two forms, a vowel-final and a consonant-final form, the latter appear in phrase-final and the former in non-final position, as illustrated with “rope” in (31).

Mambae (Austronesian, central Timor, Timor-Leste; own fieldnotes)

(31)	a.	kud	tali
		horse	rope
		“bridle” (lit. “horse rope”)
	b.	tail	mata
		rope	eye
		“trap” (lit. “rope eye”)

69This is particularly so when we take into account the feature’s appearance in three distinct Papuan languages of the Timor-Alor-Pantar family. Similarly, here inheritance cannot be used to explain the scattered appearance of metathesis, as the languages display distinct patterns. Compare the following. In Bunaq, metathesis is triggered by prefixation of roots with the shape CV₁V₂C where the first vowel is /i/ or /u/, as illustrated in (32). In Wersing metathesis is associated with suffixation or encliticisation onto roots with the shape CV₁CV₂ where the final vowel is /i/ or /u/, as illustrated in (29).

Bunaq (Papuan, TAP, Timor, Indonesia/Timor-Leste; own fieldnotes)

(32)	a.	sieʔ	b.	g-iseʔ
		rip		3an-rip
		“rip (it inanimate)”		“rip (him/her/it animate)”

Wersing (Papuan, TAP, Alor, Indonesia; own fieldnotes)

(33)	a.	ge-tati	b.	ge-tait-a
		3-stand		3-stand -real
		“(s/he) stands”		“(s/he) is really standing”

70In short, synchronic metathesis must be taken as an areal feature within Wallacea, albeit one with a limited distribution.

Summary

71In the preceding pages I have demonstrated the existence of a Wallacean Linguistic Area characterized by four distinguishing linguistic features (Map 17). This area, whilst contained within the western extremity of the Melanesian Linguistic Area, must be viewed as a distinctive area in its own right in which proximal Austronesian and Papuan languages share a set of distinctive properties in common to the exclusion of neighbouring regions to the west and east.

72Taken together, the features that I have discussed as defining the Wallacean language area have the region half way between the Bird’s Head and Timor as their geographical centre. At the peripheries are Flores and Sumba in the south and Halmahera and Cenderawasih Bay in the north. The extent of Linguistic Wallacea thus closely corresponds to the modern dispersal of Papuan outliers around western New Guinea and speaks to them as the origin of the area. This point shall be discussed further in what follows.

Prehistorical Inferences from Linguistic Wallacea

73The presence of each of the features that we have discussed across Wallacean languages can in theory be explained in one of three ways: (i) chance similarity, the features are the outcomes of independent parallel innovations in the various Wallacean languages; (ii) shared inheritance, the features were present in some inter-stage language from which Wallacean languages are all descended; or (iii) borrowing, features being copied from one Wallacean language to another.

74As we have seen in the discussion of individual features in the preceding sections, the first of these scenarios can be ruled out as extremely unlikely. In the case of semantic alignment and synchronic metathesis, this is because the features are relatively rare crosslinguistically, both generally in the world’s languages and specifically within the Southeast Asian region. Neuter gender is similarly marked because gender is not a typical feature of either Papuan or Austronesian languages. Finally, in the case of ^#muku, the sheer concentration of lookalikes in such a circumscribed area speaks against chance similarity.

75Shared inheritance cannot be the primary explanation since we know that the languages involved are from different families. Still, it could be speculated that the Wallacean features were copied from a long dead Papuan language into an ancestral Austronesian language from which all modern Austronesian languages in Wallacea descend. However, this scenario still would not explain what we observe among the Austronesian languages. Firstly, given that the postulated CEMP and CMP subgroups of the Austronesian tree are disputed, it is tenuous to say that all the Austronesian languages in the region belong to a single subgroup. Secondly, the differential appearance of Wallacean features makes a purely genetic explanation for them highly problematic. Moreover, the diverse forms that the features take make it impossible to reconstruct the features back to a single proto-language of the Austronesian tree.

76The third explanation would be to say that the features have been borrowed from language to language, either through contact or substrate. Borrowing fits the pattern of erratic diffusion that Wallacean features have in the Austronesian languages of the area. That the features often, as we have seen, appear to have some antiquity within small, low-level subgroupings of Austronesian languages, points to early adoption into the Austronesian languages from a Papuan substrate(s) present over the Wallacean area.

77Whilst this model deals with the appearance of Wallacean features in Austronesian language, an explanation remains to be found for their being shared across multiple unrelated, non-contiguous Papuan families. Some Wallacean features reconstruct to the proto-languages of some Papuan groups in the region. This early presence means that the features cannot have been transferred from Papuan to Papuan group by Austronesian languages, as the breakup of the Papuan families most likely predates the Austronesian arrival. So how did features become shared across Papuan groups at the peripheries of Wallacea, when they are broken up by intervening Austronesian groups?

78The obvious answer is that Wallacea is in fact an ancient linguistic area that was in existence before the Austronesian arrival. The problem with this is it does not accord with the conventional understanding of the prehistory of the region which, according to Bellwood (1997, 1998, 2006) and traditionally adopted by others was home only to sparse populations of simple hunter-gatherers prior to their being overwhelmed by expanding groups of farming Austronesians. The Wallacean linguistic area extends over a large archipelago of islands and thus can only have come into existence in the presence of a pre-Austronesian maritime culture connecting disparate speaker groups.

Archaeological Correlates of Wallacea

79The existence of Wallacea as a linguistic area is predicated on linguistic patterns observed in modern languages, but it is obviously the product of historical processes that have been played out over a long period by humans in interaction with one another. That means that if Wallacea did exist as a pre-Austronesian maritime culture, we should also find it reflected in the archaeological record. And indeed, whilst the traditional large-scale models of Island Southeast Asia have not been sensitive to it, archaeological work within Wallacea has increasingly eschewed the pre-Austronesian stereotype; the Papuan late Pleistocene is no longer viewed as necessarily a time of stasis in which economically simple hunter-gatherers were sparsely spread in contrast to the Austronesian Holocene as the time of rapid agriculturalization and technological change. Instead archaeology is progressively bearing out the conclusions we have drawn from the histories embedded in the languages of Wallacea.

81What is more, far from the putative hunter-gatherers of the conventional literature, the pre-Austronesian inhabitants of Wallacea also appear to have been agriculturalists (see Latinis’ 2000 characterisation of Wallacean arboriculture). In Timor there are signs of domesticated taro (Colocasia esculenta) cultivation from 10000 BP (Oliveira 2008: 387). Archaeobotanical evidence for the domestication of bananas in New Guinea and the movements of cultivars into Timor, Flores and then beyond has been dated to 4500 BP, thus preceding the arrival of Austronesians in Wallacea (Donohue & Denham 2009, Denham & Donohue 2009).

82The strong archaeological evidence which exists for extensive pre-Autronesian inter-island contacts in Wallacea, and for pre-Austronesian taro and banana cultivation in the same area, supports the inferences drawn in this paper regarding the formation of Linguistic Wallacea.

Conclusion

83By contextualising the areal features of the languages of Wallacea in terms of patterns found to the east as well as to the west of that area, this paper has provided a rigorous definition of Wallacea as a linguistic area. In linguistic terms as in biogeography, Wallacea lies at the crossroads of the Austronesian and Melanesian worlds, with its languages showing progressively more Melanesian features the closer to New Guinea they are spoken. However, the occurrence of Papuan features in Austronesian languages is not in itself limited to Wallacea. It also characterizes a much wider zone, referred to here as “Linguistic Melanesia,” which extends further east through New Guinea and out into the Pacific.

84What defines Linguistic Wallacea as such is a set of shared features more specific than those which characterize Linguistic Melanesia. These are, minimally: (i) semantic alignment, (ii) neuter gender, (iii) synchronic metathesis, and (iv) the occurrence of ^#muku “banana.” Previous attempts to define a Wallacean linguistic area under other names (“Eastern Indonesia,” “East Nusantara”) have erroneously included features which in fact define Linguistic Melanesia as a whole.

85The Wallacean features show signs of antiquity, and the fact that they are shared across unrelated and non-contiguous Papuan language families indicates that their origins lie in a period prior to the Austronesian language dispersal, among communities which spoke Papuan languages and were involved in sustained seaborne interaction across Wallacea. The wide and evidently early dispersal of ^#muku “banana” suggests that these communities also practiced agriculture. Archaeological evidence exists to support the proposition advanced here that in pre-Austronesian times, Wallacea was already inhabited by seafaring, agricultural peoples.