Linguistic Typology

Using a sample of 116 languages, this article investigates the typology of comparative constructions and their distribution in Melanesia, one of the world’s least-understood linguistic areas. We present a rigorous definition of a comparative construction as a “comparative concept”, thereby excluding many constructions which have been considered functionally comparatives in Melanesia. Conjoined comparatives are shown to dominate at the core of the area on the island of New Guinea, while (monoclausal) exceed comparatives are found in the maritime regions around New Guinea. Outside of Melanesia adpositional and other comparative constructions including particle comparatives are most frequent in Austronesian languages. The unity of the conjoined comparative type in the core Melanesian area illustrates that, while morpho-syntactic profiles of Melanesian languages are heterogenous, significant convergence in the “ways of saying things” can be found across the region. Additionally, we find no cases of clause chaining constructions being used for encoding comparatives, even in canonical clause chaining languages of central New Guinea. Our findings thus offer no support for Stassen’s claim of a correlation between temporal chaining type and comparative construction type. Instead we suggest that an areal preference for mini-clauses may explain the dominance of the conjoined comparative in Melanesia.


Introduction
Melanesia, the geographical region centred on the island of New Guinea (Figure 1), is the linguistically most diverse area in the world with more languages per square kilometre than any other region (Ross 2017). Defined as the area in which Papuan languages and their influences are found, Linguistic Melanesia takes in around 1,500 languages, including many members of the Austronesian family (Schapper 2015;Donohue 2007). Typological studies rarely sample Melanesian languages with the depth that their diversity warrants. Papuan languages, in particular, are either left entirely unsampled, e.g. in Greenberg (1966Greenberg ( [1963), Müller-Gotama (1994), Haspelmath & Tadmor (2009), or significantly undersampled, see Hammarström (2009) on this deficit in WALS. As a result, there are very few linguistic domains for which typologies exist that provide a comprehensive picture of Melanesian languages and their various behaviours (Reesink 2013 is a notable exception).
In this paper, we present an in-depth typological study of comparative constructions in Melanesian languages on the basis of a sample of 116 languages of diverse genetic affiliations from across Melanesia and its surrounds, and offer revisions and refinements to the typology of comparatives on the basis of it (see Tables 1 and 2

for details of languages and their affiliations).
A comparative construction involves comparing two entities in terms of the degree of some gradable property relating to them, such as in the English sentence John is taller than Mark. Two examples of common ways of creating such comparisons in Melanesia are given in (1) and (2).  (1) Korowai (Greater Awyu, TNG) if-e=xa abül=efè xoŋgél=xayan; waf-e=xa this-TR=CONN man=TOP big=very that-TR=CONN abül=efè be-xoŋgé-tebo-da man=TOP NEG-big-be[RLS.2/3.SG]-NEG 1 'This man is bigger than that man.' (lit. 'This man is very big, that man is not big') (van Enk & de Vries 1997: 71) (2) Biak (South Halmahera-West New Guinea, AN) mansar i-ne i-ba syadi aya old.man 3SG.SPEC-this 3SG-big exceed 1SG 'This old man is bigger than me.' (lit. 'This old man is big exceeding me.' (van den Heuvel 2006: 109) The typology of comparative constructions has been investigated from different perspectives in several publications (Ultan 1972;Andersen 1983;Stassen 1985, andStassen 2013;Heine 1997: 109-130;Dixon 2008;Beck et al. 2009;Ansaldo 2010;Bobaljik 2012;Stolz 2013). However, only a rather limited number of languages have been looked at; the main typological work, Stassen (1985), is the basis for Heine (1997) and Stassen (2013), and is relied on indirectly by Dixon (2008) and Beck et al. (2009). And, as with other typological variables worldwide, the Melanesian language area has received little attention in these works. In Stassen (2013), for example, only 12 languages from the Melanesian area are included ( Figure 2).
The interest of comparative constructions such as in (1) lies in the fact that descriptions of Melanesian languages not infrequently note that comparatives are actually in fact absent and that this in turn reflects a cultural lack of interest in relative size. For instance, Longacre (1976: 110) in discussing the Papuan language Wojokeso observes from its lack of a dedicated, grammatical

Comparatives in Melanesia
comparative that " [it] follows that comparison is not a cultural or linguistic universal". Such claims often do not stand up to closer scrutiny of cultural details. Barclay (2008: 125), for instance, observes in the context of comparison constructions: Importance [in Dani culture], among other things, is measured in the community by the number of pigs a person owns. If a person is killed or hurt they must be adequately compensated and there is always pressure for the increase in the amount of bride price.
Similarly, Eipomek is described by Heeschen (1998: 212) as follows: "Though there is no grammatical category, the Eipo speakers easily make comparisons, and they like to do so." Language descriptions tend to focus on grammatical coding systems, and when there is little grammatical coding of a certain conceptual domain, expressions of these domains are not or hardly discussed. In many Melanesian languages there is not a great deal of grammatical coding of comparison. Instead, speakers tend to leave key elements of comparison implicit and they try to trigger comparative readings of their utterances in the mind of their addressees with minimal morphological and syntactic means. Juxtaposition of utterances connected only or mostly by relevance relations (as in (1)) is a prominent example of the minimalistic tendencies in the expression of comparison. The boundary between comparative strategies based on pragmatic relations between the components of the comparison and grammatical constructions with intonational, morphological and syntactic marked elements of comparison is not always easy to draw when sources supply little information and few examples. This paper seeks, in the first place, to make a contribution to the typology of comparatives and, in the second place, to advance the areal typology of Melanesian languages on the basis of a thorough survey of comparative constructions of languages in and around New Guinea.
Our paper addresses the problem of identifying comparative constructions in Melanesian languages where highly grammaticalised constructions are absent. We provide a cross-linguistically operable definition of a comparative construction (a "comparative concept" in Haspelmath's 2010 terms). This definition allows us to differentiate comparative constructions from other proximal constructions (such as superlatives and similatives) in a principled way and show that the Melanesian languages in our sample do exhibit constructions that qualify as comparatives cross-linguistically. Using the definition to circumscribe the domain of study, we turn to the issue of the variability of the constructions of the major types of comparative in (i) conjoined comparatives; (ii) exceed comparatives, and (iii) adpositional and other comparatives. We focus particularly on the variety displayed by conjoined comparatives, the dominant type of Linguistic Melanesia.
We observe a large number of types in the Melanesian area, as befits the diversity of the region, but at the same time identify a large-scale areal preference for conjoined comparatives centring on New Guinea. Around New Guinea we find that a second pattern, the monoclausal exceed comparative, is frequent. However the sample does not contain a single clear case of clause chaining constructions being used for encoding comparatives, even in canonical clause chaining languages of central New Guinea where clause chaining is the unmarked way to link sequences of clauses. Outside of Melanesia, adpositional and other comparative constructions including particle comparatives are most frequent in Austronesian languages. Our findings offer no support for Stassen's (1985) claim of a correlation between temporal chaining type and comparative construction type.
The structure of the paper is as follows. Section 2 presents our definition of the comparative construction. Section 3 discusses the data and sampling that underpin our study. Section 4 discusses the types of comparative construction found in and around Melanesia. In Section 5 we discuss the areal distribution of these types, highlighting the Mainland Melanesian type of conjoined comparatives ringed by a Maritime Melanesian type of monoclausal exceed comparatives. Section 6 discusses our findings on comparative constructions in terms of possible correlates with other linguistic features, taking into account especially Stassen's (1985) observations. Section 7 concludes.
2 What is a comparative?
Throughout the article, we make use of the terminological framework of Dixon (2008) for describing the elements of a comparative. 2 Using the English comparative John is taller than Paul as an example, we can define the elements as follows: 2 See Heine (1997: 110-111) for an overview of the different terminologies used by different authors. We selected Dixon's terminology because we found it to be the most accessible set of labels. Unlike other authors' names for comparative notions, it avoids terms which are used for other parts of grammar (such as, topic, pivot, comment), and does not rely on abbreviations, such as Heine's X, Y, D, M and Z.
(3) Comparee-that which is being compared (John) Standard of comparison-that which the comparee is being compared against (Paul) Parameter of comparison-the property of comparison (tall) Index of comparison-a marker of the comparative degree (-er) Mark of the Standard-a marker which serves to introduce the Standard into the clause (than) In (4) we present the definition of "comparative" that we use throughout this paper. Drawing on aspects of previous definitions (notably Stassen 1985, andStassen 2013;Dixon 2008), our definition is intended as a "comparative concept" in the sense of Haspelmath (2010), narrowly circumscribing the expressions caught by the typology and allowing for practical cross-linguistic comparison.
(4) A comparative is an expression that has the semantic function of differentially ranking two disjoint referents with respect to a scalar property that is explicitly expressed.
Our study excludes comparatives of equality such as that in (5a), and instead is limited to comparatives of inequality as in (5b). Comparatives of inequality and comparatives of equality are semantically similar in that both compare two referents with respect to a gradable property. However, as in the Bunaq examples in (5), they are rarely coded in the same way cross-linguistically (Haspelmath et al. nd), and therefore warrant separate typologisation.
(5) Bunaq (Timor-Alor-Pantar, TNG) a. Comparative of equality neto Markus g-utu legul uen~uen 1SG Markus 3-with tall REDP~one 'I am as tall as Markus' (Schapper fieldnotes) b. Comparative of inequality neto Markus g-o legul lesin 1SG Markus 3-from tall more 'I am taller than Markus' (Schapper fieldnotes) Our definition limits the study of comparatives to comparatives of inequality by stating that the grading of entities on a scale must be non-identical, meaning that they cannot equally possess the property as in a comparative of equality.
Perhaps the key aspect of the definition is that it requires that both referents (the comparee and the standard) be explicitly encoded in the expression. 3 This is because the significant ways in which comparatives differ cross-linguistically lie in the coding of the standard (as recognised by Stassen 1985). More importantly, without this requirement, the boundary between comparatives and other kinds of expressions is not always easy to draw. For example, consider (6) from Yonggom Wambon, which does not include any standard of comparison. Drabbe (1959: 119) writes "Concerning the expression of comparative and superlative, for example Juw e kamaeop te may mean 'he is big', 'he is bigger' or 'he is the biggest' dependent on the context". 4 (6) Yonggom Wambon (Greater Awyu, TNG) juw=e kamae-op 3SG=CONN big-ADJ 'He is big (bigger, biggest)' (Drabbe 1959: 119) The interpretation of this example as comparative is completely context-dependent: utterances such as this may receive comparative readings in the right context, but non-comparative readings in other contexts. With the two-entity criterion, however, we do not rely on context to draw the line between comparatives and superlatives.
The definition further requires that the standard and the comparee be disjoint, i.e. have distinct referents. This requirement is important in distinguishing topic constructions which have superlative readings versus those that have comparative readings. Consider the examples in (7) and (8) where the comparee is a member of the set of referents of the standard coded as a left-clefted topic and is either explicitly quantified by a modifier that means 'all' or contextually understood to be universally quantified in such a way that includes reference to the comparee: in (7) the comparee 'you' is one of the boys who constitutes the standard, while in (8) the comparee 'he' is included in the standard, namely (the set of people in) the village. These expressions have been interpreted by researchers as superlative rather than comparative.
(7) Marind (Fly River, TNG) otiv patur ipe, oh s-o meai ha many boy the you just-2SG handsome true 'the many boys (=all boys), only you (are) very handsome' (=you are the most handsome boy) (Drabbe 1955: 25) (8) Mian (Ok, TNG) ele bib olo lot teke=be this village this very long=DECL 'He is the tallest in this village' (lit., 'Given this village he is very long') (Fedden 2007: 121) By contrast, in formally similar constructions expressing comparative meaning, the topic has a cardinality of two, rather than a plural group. For instance, the Bauzi example in (9) has both comparee Markus and standard Lukas as topic, but then repeats the comparee as the subject of the graded property predicate. (See topic comparatives in Section 4.2.2 for more languages of this kind).
(9) Bauzi (East Geelvink Bay, NTNG) Markus-ti Lukas-ti ozo-m di Markus bisi Markus-and Lukas-and consider-IRR SIM Markus more kila-da am tall-PARTICULARIZER CERTITUDE 'Markus is taller than Lukas.' (lit., 'When one considers Markus and Lukas, Markus is the more tall one.') (David Briley pers. comm.) Obviously, the only constructional difference between (7-8) and (9) is the referential scope of the standard. However, the non-coreferentiality requirement provides a practical way to distinguish superlative from comparative. The requirement that it be possible for a standard of comparison to be explicitly expressed means that several Melanesian languages for which we otherwise have thorough descriptions are excluded from our sample. For example, the Manambu construction in (10) whereby -pək is suffixed to adjectival predicates 'as a comparative index' in Aikhenvald's (2008: 96) description, is not counted in our typology as a comparative construction because the 'standard of comparison is not overtly expressed, and the idea of comparison is understood from the context' (see Section 4.1 on the Manambu conjoined comparative which is counted in the typology).
(10) Manambu (Ndu, NTNG) vyaket-pək-a val good-COMP-LK canoe/car 'a better car (than other cars)' (Aikhenvald 2008: 96) In other cases, whilst the grammatical descriptions do contain information on what they term comparatives, they have no information on the coding of the standard of comparison, either explicitly in the text description or implicitly in the examples provided. Instead, descriptions tend to focus in the typologically less significant coding of the index of comparison, as is the case, for instance, in the grammars of Lavukaleve (Terrill 2003), Kambera (Klamer 1998) and Vaeakau-Taumako (Naess & Hovdhaugen 2011).
For practical purposes of classification, we also require the parameter of the comparison to be encoded explicitly in the comparative construction by means of an overt property word. This restriction impacts five languages in our sample, where the sources describe parameterless comparatives in addition to other comparative constructions. The descriptions of Usan, Korafe, Waskia, Iatmul and Sawai all set out a comparative construction involving an 'exceed' verb that includes no parameter. In (11a-e), the parameter of the comparison is to be inferred by the listener from the identity of the referents and is open to interpretation as is implied by the various translations given in (11). Similar excluded constructions are listed in the grammars of Mian (Fedden 2007: 122-123) and Awtuw (Feldman 1986: 169) (see Section 4.1.4 for more discussion).
Similarly, we exclude the Marind (Fly River, TNG, Indonesia) comparative using the transitive verb, -p-aheb, to compare two entities, with the compare and standard being subject and object respectively (12). This form of comparison falls outside the typology because from the description we do not know whether it is possible to include a specific parameter or whether this is a lexical verb especially for comparison of stature. By contrast, the Buru construction with dafu-k is included because, although the parameter can be elided and the nature of the comparison left to inference (13a), it can also be made explicit with a parameter (13b).
(12) Marind (Fly River, Papuan) nak-a-pahezab 1SG.SBJ-OBJ.NON-1PL-II.exceed.in.length 'I am taller than you/them.' (Drabbe 1955: 25) (13) Buru (Central Malayo-Polynesian, AN) a. Da dafuk ringe 3SG greater 3SG 'He is bigger/better/more skilled/swifter than him.' (lit. 'He beat him', 'He is greater than him'.) b. da haa dafuk ringe 3SG big greater 3SG 'He is bigger.' (Grimes 1991: 172) In sum, our definition is restrictive in that it does not allow for all phenomena that have been described as comparatives to be included, but at the same time this enables the exclusion of phenomena that are not exclusively comparative in meaning from the typology. The definition also excludes comparative constructions that are unproductive or lexicalised structures that cannot be used to compare items along any parameter. Finally, it excludes constructions with only vague descriptions that do not allow for full classification within the typology.

Data and sampling
This paper is based on a sample of 116 languages for which we had access to information on comparative constructions. Information on comparative constructions was farmed from grammars and grammatical sketches as well as by means of personal communication with fieldworkers of languages without published descriptions. Our sample is in essence one of convenience (Velupillai 2012: 50), based primarily on availability of data. Many Melanesian languages have only sketchy, incomplete descriptions. Frequently, descriptions of Papuan and Austronesian languages give no information on comparative expressions. Even where comparatives are given space in a grammatical description, they often lack sufficient information to be categorised in the typology, for instance, because there are no examples with parameters or standards (as demanded by the definition set out in Section 2). Recent examples of grammars of Melanesian languages with neither description of comparative constructions nor examples of them locatable within the grammar include Abui (Kratochvíl 2007), Oksapmin (Loughnane 2009), Usen Barok (Jingyi Du 2010) and Yeri (Wilson 2017). Our sample is geographically centred on New Guinea and the outlying islands to the east and west where Papuan languages are still extent today. This is "that area of the southwest Pacific, excluding Australia,… run[ning] from the easterly Indonesian islands of Halmahera, Timor, and Alor in the west, to the westerly island group of New Georgia in the Solomon Islands in the east" (Foley 2000: 357). Our sample also includes Austronesian languages that are spoken beyond the boundaries of Linguistic Melanesia, in the areas to the west, north and east. The geographical dispersal of languages in our sample is presented in Figure 3.
Of the 116 languages sampled, 68 are drawn from the Austronesian family (Table 1) and 48 are drawn from different Papuan families ( Table 2).
The Papuan languages sampled are drawn from across Linguistic Melanesia, encompassing not only the many different Papuan groups of mainland New Guinea, but also the Papuan outliers in the insular regions to its east and west. Of the languages sampled on the New Guinea mainland, 28 are drawn from the large, but very diverse Trans-New Guinea (TNG) family, as defined by Pawley & Hammarström (2017) while 26 languages are from non-TNG families. Our sampling of Trans New Guinea subgroups is by no means balanced. For instance, 6 of the TNG languages in our sample come from the Timor-Alor-Pantar family, a TNG subgroup which has had the sustained focus of a dozen researchers over a decade and is as a result much more fully described than most other TNG groups.
Austronesian languages within western Linguistic Melanesia belong to the non-Oceanic branches of the putative Central Eastern Malayo-Polynesian subgroup, namely Central Malayo-Polynesian and South Halmahera-West New Guinea. Austronesian languages to the east of New Guinea all belong to the Oceanic subgroup. Within eastern Linguistic Melanesia the Austronesian languages in our sample belong to the Western Oceanic linkage and Southeast Solomonic subgroups of Oceanic, while those outside the area the languages in our sample are from the Central Pacific, North and Central Vanuatu, and South Vanuatu-New Caledonia subgroups of Oceanic. 12 Austronesian languages outside of Linguistic Melanesia are included. Following the work of Stassen (1985), we regard the coding of the standard of comparison to be the main point of typological variation in comparative constructions. 5 Also like Stassen (1985), we use a mixture of grammatical and semantic criteria for identifying comparative types. The primary division between types we make, however, is based on the identification of grammatical structures; secondary divisions are then made within types on the basis of the semantics of elements involved in the comparative structures. The result is that we identify three main types of comparative in Melanesia: (i) conjoined comparatives (Section 4.1); (ii) (monoclausal) exceed comparatives (Section 4.2), and; (iii) adpositional comparatives (Section 4.3). Not all comparatives in our sample fit into this categorisation; they have been classed as "other" and are discussed in Section 4.4.
Many languages in our sample have more than one type of comparative. In some cases, one of the types is explicitly described in the source as due to calquing of a more dominant language. For instance, Meyah has both a conjoined comparative typical of most Papuan languages (14a) and an adpositional comparative that calques Malay structure with an explicit index and a 'from' preposition introducing the standard (14b).
a. Conjoined comparative efen mod aksa; Manu efen mod en-aksa guru 3SG.POSS house tall Manu 3SG.POSS house NEG-tall NEG 'His house is higher than Manu's house.' (lit. His house is tall, Manu's house is not tall.) b. Calqued adpositional comparative ofa aksa ekirsa jeska efen meka 3SG tall INDEX FROM 3SG.POSS father 'He is taller than his father.' (Giles Gravelle pers. comm.) 5 In the course of conducting the research for this paper, we considered other variables in comparative constructions, such as those pointed out in Dixon (2008). For example, we looked at the coding of the index of comparison by means of an index bound to the verb. This feature was found in western Austronesian languages in our sample (Cebuano, Chamorro, Karo Batak, Gayo, Madurese, Pendau and Tajio). So whilst a bound index marker was a variable of comparative constructions that did characterise a subset of languages in the region to the west of Melanesia, it was not a feature of Melanesian comparatives and therefore was not of primary concern.
Similarly, Lithgow (1989) argues that the Dobu comparative using ʹenega (a preposition meaning 'from', and a conjunction meaning 'than') is a recent phenomenon calquing the English than comparative (15b) that exists alongside an areally expected conjoined comparative (15a).
(15) Dobu (Austronesian, Papua New Guinea) a. Conjoined comparative tauna sinabwa-na ta tasi-na sinabwa wawasae 3SG big-3SG CONJ brother-3SG big very 'His brother is bigger than him.' (lit. 'he is big, but his brother is very big') b. Calqued adpositional comparative 6 tauna sinabwa-na ʹenega aboʹagu 3SG big-3SG FROM/THAN 1SG 'He is bigger than me.' (Lithgow 1989: 6) However, in most cases where we find multiple comparative constructions, no information is given about their relative status. As such, in our typology, we do not attempt to rank multiple comparative constructions as primary or secondary comparatives, as done by Stassen (1985) and other authors subsequently. For most languages in the area there are no corpora or descriptions with enough detail to allow for statements to be made about the relative frequency let alone primacy, in terms of naturalness or markedness (Stassen 1985: 27), of different comparative types where they are found. The areal distribution of languages with multiple types is returned to in Section 5 as it reflects on different zones of languages contact.

Conjoined comparatives
Conjoined comparatives involve two independent clauses, one with the standard and the other with the comparee as subject. We use the term 'conjoined' to cover simple juxtaposition, coordinate clauses linked by coordinators and asyndetic conjoining.
We identify four subtypes of conjoined comparative types in the Papuan and Austronesian languages of our database, listed here in order of most to least common: (16) a. Conjoined comparative with antonymous predicates (Section 4. The conjoined subtypes are not mutually exclusive; various languages in our sample make use of more than one conjoined comparative type. A number of languages in the sample with conjoined comparatives have biclausal EXCEED comparatives as an additional option. For example, all three conjoined subtypes are found in Mauwake but Mauwake also allows biclausal EXCEED comparatives (17).
(17) Mauwake (Madang, TNG) a. Conjoined comparative with antonymous predicates waaya nain gelemuta; oko nain maneka pig that small other that big 'That pig is smaller than the other one.' (lit. 'That pig is small, the other one is big') b. Conjoined comparative with positive-negative polarity Auwa uuw-owa eliwa; mua oko fain wia 1SG.father work-NMLZ good man other this no 'My father's work is better than this other man's.' (lit. 'My father's work is good, this other man's is not') c. Conjoined comparative with intensified parameter poka fain maala; ne oko maala akena stilt this long add other long very 'This stilt is longer than the other one.' (lit. 'This stilt is long but the other one is very long') d. Conjoined comparatives with exceed verbs poka fain maala; nain nomak-e-k, … stilt this long that EXCEED-PST-3SG 'This stilt is longer than that one, …' (lit. 'This stilt is long, it exceeds that (one)') (Berghäll 2015: 302-304) The first three subtypes (exemplified in 17 a-c) have in common that the standard and the comparee are both subjects, each in its own clause (derived-case comparatives in the terminology of Stassen 1985). The fourth type, the conjoined exceed subtype (exemplified in 17d), is different in that, while the comparee is a subject, the standard is an object. This means the subtype is a problematic hybrid (Stassen 1985: 48, Heine 1997). Yet, conjoined exceed comparatives always exist alongside conjoined comparatives of one of the other types with parallel subjects; we do not find languages where a conjoined exceed comparative is an alternative comparative construction to a non-conjoined comparative type. As a result, we take these conjoined exceed comparatives to be a subtype of the conjoined comparative and not the (monoclausal) exceed comparative (see Section 4.1.4 for more discussion of the distinctive features of conjoined exceed comparatives).
Where more than one subtype is available in a language (as in the Mauwake examples), they typically represent straightforward alternatives to one another in the expression of comparison. In a few cases, however, we observe specialised comparative meanings being associated with particular comparative conjoined constructions in individual languages. For instance, when used with an intensified adnominal property word, parameter intensification in Kobon can express a correlative comparative (18). In Eipo, parameter negation can be used to express 'permanent acts of comparison' (19).
(18) Kobon (Madang, TNG) kaj kub aij ji kaj kub yabɨl aij yabɨl pig big good and pig big very good very 'The bigger the pig the better. ', 'Very big pigs are better than big pigs.' (lit. 'Big pigs are good, very big pigs are very good') (Davies 1981: 93) (19) Eipo (Mek, TNG) yalea teleb; ya are malye; fi teleb gum stell.axe good stone.adze ART bad very good NEG 'A steel-axe is good, a stone-adze is bad, it is not at all good.' (Heeschen 1998: 212) Not only can more than one subtype be found in one language, but in a few languages, we find comparative constructions that mix elements from the different subtypes. For instance, Ternate constructs conjoined comparatives using antonymous predicates, as in (20a), but can optionally mark the second parameter with the borrowed intensifier lebe (used as an index of comparison in Malay), as in (20b) Similarly, Manambu has biclaual comparatives with predicates of positive-negative polarity, but can add the intensifier məyər to the negated clause, as in (21) Because antonymous predicates and positive-negative polarity consistently appear in the Ternate and Manambu comparatives we have examples of, we classify the languages as having the conjoined comparative with antonymous predicate and conjoined comparative with positive-negative polarity types, respectively. These examples, nonetheless, illustrate that the different conjoined comparative types should not be considered wholly discrete. In our database, the conjoined types are found in the languages given in Table 3. The most common type is that which uses antonymous predicates. Each Table 3: Languages with Conjoined comparative types.

TYPE
Papuan Austronesian ANTONYMOUS PREDICATES of these types is discussed in the following sections in terms of the constructional variation they display.

Conjoined comparative with antonymous predicates
The first conjoined type, with antonymous predicates in the two clauses, is by far the most commonly exemplified conjoined comparative type in Melanesia. It is found across the whole expanse of Papuan languages, both TNG and NTNG, as in: (22) Eipo (Mek, TNG) a-me are metek; or-me are wik DEM-child ART small DEM-child ART big 'This child is smaller than that child.' (lit. 'This child here is small, that child there is big') (Heeschen 1998: 212) (Bruce 1984: 422) (24) Idi (Pahoturi, NTNG) sos-a glä rmb dand skul-a kdh dand church-CORE EMP big COP.3SG school-CORE small COP.3SG 'The church is bigger than the school.' (lit. 'The church is big, the school is small') (Schokkin pers.comm.) It is also the most common type of conjoined comparative found in Austronesian languages of Melanesia. Examples include: Lae tu barego lelevagi; Saroa tu kei Lae TOP big very Saroa TOP small 'Lae is much bigger than Saroa.' (lit. 'Lae is very big, Saroa is small') (Tauberschmidt 1999: 38) (26) Kairiru (Oceanic, AN) numpuong nai nau spai a-mokin; nimpai nau sek yesterday sea some 3SG-good today sea bad 'The sea was better yesterday than it is today' (lit. 'Yesterday the sea was good, today the sea is bad') (Wivell 1981: 180) Typically, most languages with this type of conjoined comparative do not require an index. However, in the case of Wersing (27), the antonymic verbs of both clauses are marked by a prefix mi-. The prefix is a kind of index which makes it clear that the two entities are being compared to one another.

Conjoined comparatives with intensified parameter
The second conjoined type has an intensified parameter in the comparee clause. The intensifier functions as an index (John is tall, Pete is very tall > Pete is taller than John). The typical pattern is for the intensified comparee clause to be the second of the two clauses, as in: (28) (Tauberschmidt 1999: 38-39) Two languages have the comparee clause with the intensified parameter in the first clause, but this only occurs where the standard clause has its parameter marked with a limiter of sorts, such as di 'only' in Teiwa (30) or moli bakisi 'only a little' in Nakanai (31).

Conjoined comparative with positive-negative polarity
Just like the other conjoined subtypes, the third subtype distributes the comparee and standard arguments over the subjects of two clauses, but now the second predicate is not antonymic but the negated parameter of the first clause, as in Motu (32) (see also the Meyah example in (14a)).
(32) Motu (Oceanic, AN) una na namo herea; ina na dia namo this is good more that is not good 'This is better than that.' (lit. 'This is more good, that is not good') (Lister-Turner & Clark 1930: 32) Though not typical, the negated clause can also come first, as in Mekeo: (33) Mekeo (Oceanic, AN) l'faea au-ija bela aibaia; engaea au-ija belo that one-3SG beautiful not that one-3SG beautiful 'That one is more/less beautiful than that one.' (lit. 'That one is not beautiful, that one is beautiful') (Amala et al. 2015: 206).
Whilst the parameter is typically repeated in both clauses, it is possible in some languages for it to be elided where understood in the presence of the negator, as in the example in (34) (also Mauwake example in (17b).
(34) Lavongai (Oceanic, AN) a ei kekate vutulava; a ei aŋ suke kovek ART tree DEM big ART tree DEM other NEG 'This tree is bigger than that other one.' (lit. 'This tree is big, the other one is not') (Stamm 1988: 12)

Conjoined EXCEED comparative
The fourth conjoined comparative and least common subtype has the parameter introduced in one clause, 7 and an 'exceed' or '(sur)pass verb introducing the standard as its object in another. 8 For example, Gants (35) and Amele (36) encode the comparee as the subject of the parameter in the first clause and the standard as the object of the exceed verb in the second clause. The comparee is explicit in the first clause and the implied subject of the second clause. uqacecela; ija wol-te-na 3SG tall 1SG EXCEED-1SG/3SG-PRES 'He is taller than me.' (lit. 'He is tall, exceeding me') (Roberts 1987: 134) The same structure is found in the two Austronesian languages with conjoined exceed comparatives: 7 Biclausal exceed comparatives are likely more common than appears in this typology; see Berghäll (2015: 303) for a similar claim. The Usan, Korafe, and Iatmul constructions in (11) may well be able to be used with a separate clause introducing a parameter; it just happens that such examples are not included by authors in their descriptions. Indeed this is suggested in the description of one language, Usan, (Reesink 1987: 68). Nominalised parameter comparatives often involve exceed verbs and can be regarded as variants of biclausal exceed comparatives (see Section 4.4.4). 8 See Section 4.2 on the different kinds of verbs involved in exceed comparatives and how they can be identified and distinguished from other comparatives such as particle comparatives. In Hua (39) the comparee and standard are encoded as the subject and object respectively of an exceed clause followed by a clause that encodes the parameter.
(39) Hua (Madang, TNG) d-kaso-na; zaʹzaf-i-e 1SG-EXCEED-3SG tall-3SG-IND 'He is taller than me.' (lit. 'He exceeds me, he is tall') (Haiman 1980: 283) In all examples of conjoined exceed comparatives we have seen thus far, the two clauses in the comparative construction are linked by simple juxtaposition. There is however one language, Mauwake, where this exceed comparative construction is not always clearly conjoined: the exceed verb in (40a) is medial and comes in a chained form with -ep marking switch-reference and temporality (sequence/simultaneity); in the comparative construction in (40b) repeated from (17d), by contrast, the exceed verb is final and carries full inflection. Berghäll (2015: 192) describes Mauwake clause chaining as structurally intermediate between verb serialisation and main clause coordination, making (40a) appear problematic for the comparative construction types we establish here.
(40) Mauwake (Madang, TNG) a. Chained exceed comparative maa mane~maneka maa fain nomak-ep ik-ua thing REDP~big thing this EXCEED-SS.SEQ be-PST.3SG 'These things are bigger than these.' (lit. 'They are big things, exceeding these') b. Juxtaposed exceed comparative poka fain maala; nain nomak-e-k, … stilt this long that EXCEED-PST-3SG 'This stilt is longer than that one, …' (lit. 'This stilt is long, it exceeds that one') (Berghäll 2015: 303-304) The only other languages in which we find chaining with exceed verbs in this manner are for comparatives that lack a parameter coded explicitly with a property word. For example, in (41) Mian has the exceed verb gaisnenea medially with same-subject sequence marking followed by a final clause with a full inflected motion verb usn-'go up' (see (11a-e) for other language examples). Whilst this latter motion verb stands in for a parameter in the sense that the motion of moving up figuratively expresses height, it is not a property word and therefore is not taken to constitute a comparative construction under the definition set out in Section 2.  (Fedden 2007: 122-123) The Mauwake chained exceed comparative, whilst including an overt parameter in the form of a property word maneka 'big', denotes a comparison of stature; Berghäll (2015) does not include non-stature examples of the chained exceed comparative. As pointed out in Section 2, comparatives of stature frequently have their own lexicalised structures that cannot be used for comparisons along other parameters. This lack of productivity makes them unsuitable objects for crosslinguistic investigation.

(Monoclausal) Exceed comparatives
In this section we look at monoclausal comparative constructions which involve the use of a transitive verb with a lexical meaning which involves motion proceeding beyond (a specified point) to introduce the standard of comparison. In descriptions such verbs are glossed variously as 'exceed', 'surpass', 'pass', 'go further', 'move through', or in some cases, due to semantic shift, 'beat' in the sense of defeat, overcome or outdo.
In classifying a language as having an EXCEED comparative, we rely heavily on descriptions and the glosses they provide of morphemes/lexemes in comparative expressions. Where possible, we have sought to verify "exceed" glosses. This has involved seeking non-comparative uses of the claimed EXCEED verbs in which the item has motion semantics consistent with an EXCEED verb as defined above. For instance, the description of Tetun Fehan ( van Klinken 1999) presents clear examples of the verb liu denoting motion beyond a point, in both intransitive (42a) and transitive frames (42b), as well as in comparisons (42c).
(42) Tetun Fehan (Austronesian, Timor) a. ami ata liu lai 1PL.EXCL slave go.further PRIOR 'We lowly commoners will go on now.' b. …, sa'e liu ró ascend go.further boat '…, (she) got up into the boat.' c. nia kbít liu besi 3SG strong go.further iron 'It is stronger than iron.' (lit. 'It is strong exceeding iron') ( van Klinken 1999: 114) In other cases, "exceed" glosses in descriptions have had to be discarded and languages reclassified. For instance, in Holmer's (1996) description of Seediq, rmabang appears in comparative constructions with the gloss 'surpass' (43a). Rmabang does not ever refer to a motion and is in fact a stative derivation of rabang, a root meaning simply 'more' as seen with the causative derivation in (43b). Rmabang in (43a) functions, in fact, as an adverbial 'morely' modifying the following stative predicate malu 'good' (Arthur Holmer pers. comm.). Unlike in an EXCEED comparative, rmabang does not introduce the standard, rather daha ka does, with the result that Seediq is (re)classified as a particle comparative (see Section 4.4.1).
(43) Seediq (Austronesian, Taiwan) a. r<ma>bang malu qsiya lmiqu daha ka qsiya surpass good water forest than water <STAT>INDEX good water forest PART water Taihoku Taipei (Original glossing) Taipei (Revised glossing) 'The water in the forest is better than the water of Taipei.' (Holmer 1996: 52) b. p-rbang-un mu bobo na CAUS-more-PAT 1SG above 3SG 'I will add more later.' (Arthur Holmer pers. comm.) (2006) describe a comparative construction for Amis in which a verbal prefix ki-marks a property word, deriving a "transitive-like" verbal complex and licensing a second argument (the standard) in the locative case. They gloss this morpheme as 'exceed', as in (44). In Amis, -ki-, as an affixal verbal morpheme, does not fit well into any of Stassen's 'exceed' subtypes.

Similarly, Sung & Kuo
[…] we still consider -ki-as one 'exceed' type since the morphological complex -ki-V behaves exactly like a 'transitive-like' exceed as suggested by Stassen. Thus an attempt to associate Amis ki-with an EXCEED comparative type seems unwarranted and reclassification required. 9 Where we have found no independent corroborating evidence for an "exceed" classification and/or have no explicit reason to doubt a classification, we allow the author's characterisation of the comparative to determine the 9 Ansaldo (2010) makes the point that EXCEED comparatives are characterised by a combination of semantic and syntactic criteria. In arguing that Cantonese -gwo3 in comparatives is an incorporated exceed verb, he shows gwo3 also occurs as an independent verb meaning 'pass'. The absence of any such claim in the case of Amis, let alone demonstration, means that the semantic criterion for classification as an EXCEED comparative is not met. classification. For instance, Jackson (2014: 115) writes of Irarutu 'the verb sfar means 'surpass', and is found in comparatives …'. Although sfar is always glossed as 'more' (as in (45)) and is only evidenced in examples of comparatives and never with a motion meaning, we classify Irarutu as an "exceed" comparative in line with the description.
(45) Irarutu, (Eastern Malayo-Polynesian, Western New Guinea) matu adini=ro trro sfar matu adeji person DET=COP tall more person DET (Original glossing) person DET=COP tall EXCEED person DET (Revised glossing) 'This person is taller than that person.' (Jackson 2014: 113) We distinguish two types of EXCEED comparative in our sample: (i) serialised EXCEED comparatives (Section 4.2.1), and (ii) incorporated EXCEED comparatives (Section 4.2.2). In what follows, we use the standardised gloss EXCEED for the relevant morpheme/lexeme for ease of comparison.

Serialised EXCEED comparative
The serialised EXCEED comparative is the most numerous in our sample with 20 languages. Languages of this type include both Austronesian and Papuan and are located in the circum-New Guinea area where verb serialisation is common. Comparatives of this kind involve a stative verb or adjective expressing the parameter serialised with an EXCEED verb introducing the standard as its object.

(48) Dadu'a (Austronesian, Timor)
Maria matenek lii wawata sia seluk Maria clever EXCEED woman PL other 'Maria is cleverer than other women.' (Penn 2006: 128) In our sample, serialised EXCEED comparatives occur in six Papuan languages, from three separate families. The languages are Nedebang (49), Teiwa, Bunaq, Makasae (see example (52) below), Yawa (50) and Savosavo (51). The last of these differs in its construction in that in Savosavo the EXCEED verb which introduces the standard precedes the property word; all other languages, whether Papuan or Austronesian, have the same order, property word followed by EXCEED verb.

Incorporated EXCEED comparative
An incorporated EXCEED comparative has the EXCEED verb incorporated into the predicate that expresses the parameter of comparison. The incorporated EXCEED verb functions as an applicative and the standard of comparison is encoded as the object of the whole predication. This type is very minor, being present only in two languages, Makasae and Makalero, closely-related Papuan languages spoken in eastern Timor.
In Makasae, litaka 'pass, surpass' is used in serialisation with a preceding parameter introducing the standard of the comparison as its object (52a). A variant construction sees litaka incorporated as lita-into a preverbal position where it introduces the standard as the applied object of the parameter (52b). 'Our lives nowadays are better than our parents' lives.' (Correia 2011: 318) Makalero has almost little to no verb serialisation. Accordingly, there is also no serialised EXCEED comparative. The Makalero comparative is formed exclusively by means of EXCEED incorporation. The incorporated index may be either leto-(53a), or seti-(53b). The latter is cognate with Makasae lita-and litaka, while the former is an incorporated verb which introduces an object past which a movement is made (53c), consistent with an original independent verb meaning 'pass, surpass'.
(53) Makalero (Timor-Alor-Pantar, TNG) a. kuda ere ni-raku na'u leto-asan horse DEM REFL-friend just EXCEED-long 'This horse is bigger than its fellow horses.' (Huber 2011: 344) b. … tufuraa … nosa.siniora seti-felun … woman Holy.Virgin EXCEED-pretty '… a woman who is prettier than the Virgin Mary …' (Huber 2011: 344) c. fi la'a=ni Paifaka ueir seti-la'a … 1PL.INCL go=CONJ Paifaka river EXCEED-go 'We go and cross the Paifaka river …' (Huber 2011: 153) One Austronesian language, Dobu, appears to have an incorporated EXCEED comparative, as a variant on a serialised EXCEED comparative. Compare the clauses in (54). A stative verb agrees with the subject by way of a suffix (54a). In a comparative, based on the limited examples available, we find that this agreement does not always appear. In (54b) we see the EXCEED verb and the parameter are compounded together and as a complex take a prefix for the comparee and a suffix for the standard of comparison. In (54c), however, we find both the parameter and the EXCEED verb with separate agreement suffixes, consistent with a serialisation construction, though we have no information about what arguments these suffixes refer to.

Adpositional comparatives
An adpositional comparative is one where the standard of comparison is introduced by an adposition that has semantic functions beyond marking standard, such as comitative, instrument, and so on. Where a standard is introduced by an adpositional (-like) element whose exclusive function is to mark the standard, this is classified as a particle comparative (see Section 4.4.1).
Adpositional comparatives are divided into sub-types based on the semantics of the adposition introducing the standard of comparison. Table 4 summarises the types that we observe (following the so-called "schemas" identified by Heine 1997: 109-130). We establish the adpositional type, rather than Stassen's more restrictive semantic "locational" type, because, as can be seen in Table 4, there are over a dozen languages in Melanesia (~11% of our sample) whose standard encoding adposition cannot be characterised as locational (Section 4.3.1). Other subtypes are comitative (Section 4.3.2) and cross-categorical (Section 4.3.3).

Locational comparatives
A locational comparative involves the standard being encoded as an NP with a locational function. In all cases bar one (Amis, where a locative case marker is used) it is a locative adpositional that is employed to encode the standard of comparison. Locational comparatives are rare in Melanesia, and almost unknown in Papuan languages (Papuan separative comparatives all appear to be the result of contact with Indonesian). Locational comparatives are sporadic but widespread in Austronesian languages.

Comitative comparatives
A comitative comparative is where the standard of comparison is introduced into the clause by means of an adposition elsewhere denoting an accompanying entity (glossed as WITH in what follows). 11 Comitative comparatives are found in six languages in our sample. All languages are to the west of New Guinea, scattered over a large region of the Indonesian archipelago. Of these six, five are Austronesian, from east to west: Batuley, Balantak, Muna, Madurese, and Gayo. In the first three languages, the comparison is marked by the appearance of the standard of comparison in a comitative prepositional phrase without any further index of the comparison (59-61).
(59) Batuley (Central Malayo-Polynesian, Austronesian; Aru) sandal pasang en kat fei enon sandal pair DEM bad WITH DEM 'That pair of sandals is worse than that one.' (Daigle 2015: 155) (60) Balantak (Celebic, Austronesian; eastern Sulawesi) laigan-na balaki' tuu' tia laigan-ku house-3SG big very WITH house-1SG 'His house is much bigger than mine.' (van den Berg & Busenitz 2012: 125) (61) Muna (Celebic, Austronesian; eastern Sulawesi) no-bhala anoa bhe inodi 3SG-big 3SG WITH 1SG 'This old man is bigger than I am.' (van den Berg 1989: 142) 12 A similar pattern is found in Tobelo, the one Papuan language with a comitative comparative (62). A comparative is formed by means of the comitative/instrumental preposition de introducing the standard plus an index of the comparison. With a verbal parameter, the index is a suffix on the parameter, while with a nominal parameter, the index is encoded as a verb -holoi.
11 Often these adpositions are used to encode instruments, but not always. As such, we focus on the comitative component of these items. 12 Stassen (2013) lists Muna as a conjoined comparative language. We find no examples or description in van den Berg (1989) to support this classification.

Cross-categorical adpositional comparatives
There are also adpositions that do not easily fit into a subtype following Heine's (1997) schemas (labelled "cross-categorial" in Table 4). For example, the Kairiru preposition qe(qe) cross-cuts the semantic domains established in this typology, introducing the standard of comparison (63a), as well as both human accompaniers (comitative type, (63b)) and human sources (separative type, (63c).
(64) Paluai (Oceanic, AN) a. Standard of comparison yi menengan a-yi yi ta-yamyaman 3SG big OBL-3SG 3SG DEF-red 'It is bigger than the red one.' (Schokkin 2014a: 272) b. Location wo=tou si a-yi nisio purukei liliu 2SG=put come.down OBL-3SG other bowl again 'You put (it) into yet another bowl.' (Schokkin 2014a: 335) c. Instrument kei ta-i te-yo yi=la to nêktou a-yi mina-n tree POSS-3SG EMP-DEM 3SG=go CONT hold OBL-3SG hand-PERT Yi ta-almaru 3SG DEF-right 'He is holding his stick with his right hand.' (Schokkin 2014a: 286) d. Stimulus of emotion ip=to kaêrêt a-yi arona-n kauwat ta-ip 3PL=HAB afraid OBL-3SG way-PERT tradespartner POSS-3PL 'They used to be afraid of the ways of their tradespartners.' (Schokkin 2014a: 278) e. Theme of cognition verb ep=ma=tu mapwai liliu a-yi pwên 1PL.EXCL=NEG=stay know again OBL-3SG NEG 'We do not know about it anymore.' (Schokkin 2014a: 280) f. Duration yi=ki-sa ning an-sê=ong a-yi kunawaye-n pwên 3SG=IRR.SG-MOD see piece-small=1SG OBL-3SG life-PERT NEG 'She should not be able to see me again for the rest of her life.' (Schokkin 2014a: 371) Similar problems are presented by three nearby languages of New Britain and New Ireland (the Austronesian language Kove, and the Papuan languages, Sulka and Kuot). These all have adpositions for introducing the standard of comparison that encode a broad range of thematic roles), see examples in (65) and (66). Apart from the standard of comparison in the (a) examples, these adpositions introduce such roles as an unaffected patient (b), a stimulus argument of an emotion predicate (c), and a theme argument of a verb of speaking (d).

Other comparative types
A few minor types of comparative constructions are apparent in the region. This residual group includes the heterogeneous comparatives that have traditionally been referred to as "particle" comparatives (Section 4.4.1), comparatives where the standard is part of a topical phrase at the left-hand clause periphery (Section 4.4.2), comparatives where the standard is treated as the transitive object of the parameter (Section 4.4.3), and comparatives involving nominalisation of the parameter (Section 4.4.4).

Particle comparatives
Particle comparatives are constructions in which the particle introducing the standard of comparison is a dedicated marker and does not have adpositional functions beyond marking the standard. That is not to say that comparative particles do not have other functions in the language. As Stassen (1985: 189-197) observes, for many languages, the comparative particle can be identified with an item which performs coordinating or (ad) verbial functions.
Whilst particle comparatives are one of the major types of comparatives identified worldwide (Stassen 2013), they are a marginal type in Melanesia. In our sample, particle comparatives are only found in Austronesian languages (Table 6). Of these Austronesian languages, only two (Ambel and Buru) occur at the western periphery of Linguistic Melanesia; the remainder are outside of it.

Topic comparatives
In a topic comparative, both the standard and the comparee are presented as topics on the left-periphery. In the clause that follows, the comparee is repeated as the single argument of the property word. An adverb, such as an intensifier or an index, marks the property word to denote the graded nature of the property. We saw an example of a topic comparative in Bauzi in (9). Taba (67) and Puyuma (68), are the other two languages in our sample that use topic comparatives in our sample.
(67) Taba (SHWNG, Austronesian) mesel ne lo mesel da, i-dia lebe mlongan wall PROX and wall DIST DEM-DIST INDEX long 'That wall is longer than this' (lit., 'This wall and that wall, that one is longer') (Bowden 2001

Object comparatives
In this minor subtype, the parameter is the main clausal verb. The comparee is the subject of the parameter and the standard its object. Only two languages in our sample display this type, but with some constructional differences (see more variants on this type in the discussion in Dixon 2008).
In the Papapana comparative construction, the standard is encoded by the object of the parameter licenced by the transitive enclitic =i and an object marking enclitic (69a). This construction is allied to the causative construction in the language, which differs only in the addition of a causative prefix on the predicate (69b).   (Smith 2015: 142) The Buru comparative, similarly, allows the standard to be introduced without any marking following the parameter (70a). Unlike Papapana, however, Buru lacks object or transitivity marking; only the post-predicative position of the standard indicates its status as an object. This appears to be a zero-marked variant of the particle comparative in (70b).

Nominalised parameter comparatives
This is an infrequent and highly variable strategy for creating comparative constructions. The common factor is that the parameter occupies an argument slot within the comparative construction. Whether it is the standard or the comparee who is encoded as the possessor of the parameter depends on the metaphor of comparison that the individual language uses.
In Fuyug, the standard is the possessor of the parameter, which is introduced by the superessive postposition adad, while the comparee is the clausal subject (71). In Leti it is the comparee (72).

(71) Fuyug (Goilalan, TNG)
Mang ge Kipi hu=l inog adad and-engo Mang TOP Kipi 3SG=POSS big OVER stay-STAT 'Mang is bigger than Kipi.' (lit. 'Mang stays on top of the bigness of Kipi.' (Bradshaw 2007: 120) (72) Leti (Central-Malayo-Polynesian, Austronesian) Luksa tul-lu n-resi toms-e Lukas high-3.POSS 3SG-win Tomas-DEF 'Lukas is taller than Tomas.' (lit. 'Luke's bigness beats Tomas') (Aone van Engelenhoven pers. comm.) As mentioned in Section 4.1.4, Mauwake has a nominalised parameter variant of its exceed comparative construction. In this, the standard and the comparee are both encoded as possessors of a repeated parameter filling both subject and object argument slots of nomak-'exceed' (73).  (Berghäll 2015: 303) It is notable that nominalised parameter comparatives are based on semantic schemas of other comparative types. This, along with the little comparability the structures show in the coding of the standard, may be taken to indicate that this is not a type in its own right at all (cf. the diverse constructions that Dixon (2008: 799-800) identifies with a nominal parameter). Heine (1997: 113-114, 120-123) refers to such mixed type comparative as "blends" and observes that they are common in the languages of the world. In our sample, however, they appear rare and show no areal patterning or consistent semantic schema. We return to the issue of structural and semantic types in our discussion of the areal distribution of comparative types in the following section.

Zones of convergence
The comparative types discussed in the preceding section show striking distributional patterns around New Guinea. Figure 4 presents the results of our investigations cartographically. The areal patterns can be summarised as follows: a core New Guinea area in which only conjoined comparatives occur, a maritime Melanesian area circum-New Guinea in which monoclausal exceed comparatives are distributed, while beyond the boundaries of Linguistic Melanesian adpositional and particle comparatives are the norm.
Conjoined comparatives are distributed throughout Linguistic Melanesia. They occur both in the periphery and the centre of the area, and they occur in a wide range of unrelated families, both Papuan and Austronesian (e.g. Oceanic; NTNG Papuan: Sepik, Morehead-Maro, Pahoturi, Abun, Engan, East Bird's Head; TNG families such as Greater Awyu, Greater Ok, Mek, Lani). Conjoined comparatives are, however, unevenly distributed over Papuan and Austronesian languages. Conjoined comparatives occur in the majority of Papuan languages of the sample, that is, in 35 of the 48 sampled Papuan languages. Austronesian languages give the reverse picture: a minority have a conjoined comparative, only 16 out of the 68 Austronesian languages sampled. Of these 16 Austronesian languages, 13 are located within the boundaries of Linguistic Melanesia. All but one of these 13 are Oceanic languages spoken either on the coast of the New Guinea mainland or on islands in close proximity to New Guinea where Austronesian speakers settled in areas with Papuan speech communities around 3,200 years ago.
It is likely that the conjoined comparative construction is pragmatically available in almost every language. For instance, there is nothing ungrammatical about Lourens is tall, Antoinette is short in English, even though Lourens is taller than Antoinette would be a more usual English comparative construction. In accordance with this inference that many languages may have conjoined comparatives as pragmatic, but perhaps atypical comparative constructions, our sample takes in three Austronesian languages outside of Linguistic Melanesian (Amis, Samoan and Tuvaluan) that have conjoined comparatives alongside the areally more typical adpositional comparatives. However, within Linguistic Melanesia we find the exclusive appearance of conjoined comparatives: of the 35 Papuan languages with conjoined comparatives, 32 have it as the only option for expressing comparison; of the 16 Austronesian languages with conjoined comparatives, all 9 that have it as the exclusive option for expressing comparison are spoken on or near the New Guinea mainland. In short, what sets the core New Guinea area apart is the absence of non-conjoined types of comparatives, and it is this lack of alternatives which defines the area.
The monoclausal exceed comparative is characteristic of the maritime region of Linguistic Melanesia, that is the coastal and, most particularly, the insular regions around New Guinea (termed "Maritime Melanesia" in Figure 5). Whilst New Guinea Papuan languages in areas without a history of Austronesian contact have only conjoined comparatives, more than half of the sampled Papuan spoken languages off the mainland of New Guinea have a monoclausal exceed comparative, often alongside a conjoined comparative construction. Austronesian languages from Flores at the western extreme of Melanesian to Vanuatu and New Caledonia at its eastern fringe also evidence monoclausal exceed comparatives. This distribution of monoclausal exceed comparative corresponds to the region in which Papuan and Austronesian languages display significant reliance on serial verb constructions (rather than clause-chaining in the case of Papuan languages, or adpositional and applicative constructions in the case of Austronesian languages) for denoting complex events. Our primarily structural rather than semantic-based typology of comparatives is supported by the areal patterning of comparative constructions involving an EXCEED verb. Whereas the monoclausal exceed comparative has a distinctly insular distribution, the conjoined exceed comparative is characteristic of a small number of Papuan and Austronesian languages in a micro-region in the eastern half of New Guinea (Figure 6). 13 Moreover, the conjoined exceed comparative in these languages represents a variant on other kinds of conjoined comparative subtypes, and never alternates with a monoclausal comparative.
The somewhat broken distribution of the monoclausal exceed comparative is partly due to the presence of micro-regional patterns of adpositional comparatives in maritime Melanesia (Figure 7). West of New Guinea (black dashed lines), a comitative adposition is used to introduce the standard of comparison. East of New Guinea (grey dashed lines), we find a pattern of adpositions of very general ("cross-categorical") meaning (see Section 4.3.3) introducing the standard of comparison.
Locative and particle comparatives occur only in a few Papuan languages in the peripheral areas of Linguistic Melanesia where they have been in the vicinity of 13 Other than this, we find no areal skewing in distribution of the different sub-types of conjoined comparative. Austronesian languages for many centuries. Beyond this, in the areas of Oceania and Island Southeast Asia where there are no known Papuan influences, we find only particle and locative comparatives. This pattern is strongest in western Indonesia and the Philippines. In western Melanesia, the adoption of adpositional comparatives in Papuan languages under the influence of the lingua franca Malay has also weakened the areality of Melanesian comparative types. Three Papuan languages in our sample show such adoptions, Bunaq, Meyah and Abun. Speakers in these language communities all show a high degree of bilingualism with Indonesian/Malay. The separative constructions in these languages calque the Indonesian/Malay comparative construction with its explicit index of comparison and a FROM adposition. In Bunaq (illustrated in (5b) and Meyah (illustrated in 14), the calqued separative comparison exists alongside other comparative constructions that conform to the norms of the regions they occur in, a SVC EXCEED comparative in the case of Bunaq, and a conjoined comparative in the case of Meyah. In Abun (74), a presumably earlier EXCEED comparative has been adapted to create the modern separative construction: wai which marks now the index is historically derived from a verb 'to pass' (Berry & Berry 1999: 36). 14 14 David Gil (pers. comm.) notes a similar partial calque in the Malay variety of Atambua, the region of central Timor where Tetun is spoken. Here, he observed lewat dari EXCEED FROM used in a comparative construction; the verb lewat replicates the Tetun substrate of using an EXCEED verb for introducing the standard of comparison, while the preposition dari replicates the typical (Kupang or Standard) Malay pattern of introduction the standard of comparison with a FROM adposition.
(74) Abun (Papuan, Bird's Head) yenggras ne nggi wai kadit nan old.man DET strong EXCEED/INDEX FROM 2SG 'His house is much bigger than mine.' (Berry & Berry 1999: 36) Similarly, languages of central Maluku where exceed or comitative comparatives are expected show calquing of Indonesian/Malay constructions (Simon Musgrave pers. comm.) Buru, for instance, has a calqued separative comparative from the local Malay lingua franca using a borrowed index (lebé < Ambonese Malay lebe): (75) Buru (Austronesian, Central Maluku) gomi naa lebé beha fidi gomi leuk axe DEM INDEX heavy FROM axe precede 'This axe is heavier than the axe we had before.' (Grimes 1991: 369) The consistency with which the conjoined type is found in the core New Guinea area contrasts with the findings of Ross (2017: 806), who on the basis of a range morpho-syntactic features claims that it "is neither a geographic nor a typological area". Conjoined comparatives are, in essence, conventionalised lexical templates for "ways of sayings things" (Ross 2001). Many lexical features such as colexifications and lexical collocations show strongly areal distributions over the core New Guinea area (Schapper et al. 2016;Schapper 2017), indicating that conjoined comparatives are subject to convergence in similar ways. That the conjoined comparative is a reflection of "ways of saying things" in the New Guinea area is taken up in the following section.

Conjoined comparatives and their linguistic correlates
As we have seen, conjoined comparatives are by far the dominant type of comparatives in Papuan languages, occurring over the whole of the core area of Linguistic Melanesia. These conjoined comparatives also occur in a minority of Austronesian languages of Linguistic Melanesia, all in the contact areas with Papuan languages. In the centre and south coast of New Guinea, in areas without Austronesian contact, conjoined comparatives are the only type of comparatives in both TNG and NTNG families, the inner circle if you will.
(Yagaria family ;Haiman 1980;Reesink 2008: 883), Amele (Roberts 1987: 183) and Koromu (Priestley 2009: 88). Now in order to create utterances of the [(XP) V] form in the domain of comparison, speakers have two options, either to make arguments implicit or to distribute arguments of the comparison over sequences of mini-clauses. Making implicit standard arguments leads to the communication of comparative intentions by non-comparative structures that solely rely on relevance-guided inferential processing by their addressees to identify the standard. This has led linguists such as Drabbe to include non-comparative utterances in his description of comparison in Yonggom Wambon (see example 6). But when speakers need to make standards and compare arguments explicitly to perform the comparison, they distribute them over two conjoined or juxtaposed clauses, with the standard as the subject of one clause and the comparee as the subject of the other clause (78, repeated from 1).

(78) Korowai (Greater Awyu, TNG)
if-e=xa abül=efè xoŋgél=xayan; waf-e=xa this-TR=CONN man=TOP big=very that-TR=CONN abül=efè be-xoŋgé-tebo-da man=TOP NEG-big-be[RLS.2/3.SG]-NEG 'This man is bigger than that man.' (lit. 'This man is very big, that man is not big') (van Enk & de Vries 1997: 71) Stassen (1985 argues that "the choice of a particular comparative type in a language L can be predicted from the type (or types) of temporal chaining which that language permits." Stassen (1985: 46) defines a temporal chain as a semantic configuration of two tensed propositions that are presented successively, with two subtypes, simultaneous temporal chains (e.g. While John ate, Pete wrote a letter; John was eating and Pete was writing a letter when I came home) and consecutive temporal chains (Having eaten, John left; John ate and then he left). There are two basic syntactic strategies available to express temporal chains, balancing and deranking (Stassen 1985: 59-60). Balanced temporal chains have two predicates of the same rank and morphological type (Stassen 1985: 62 gives this example: After John had locked the door, he undressed). Deranking temporal chains have two predicates of unequal rank.
The conjoined comparatives (types 16a-d) in our database are what Stassen (1985: 22-23) calls derived-case comparatives where the case marking of the standard equals the case marking of the comparee. Stassen (1985: 138, universal 1A) claims that "If a language has a derived-case comparative, it must be balancing." However, in our database derived-case comparatives of the type (see 16a-c) occur frequently in clause chaining Papuan languages that are not balancing but deranking in the way in which they express temporal chains, with a medial clause that contains a non-finite sequence or simultaneity verb followed by a final clause with a fully finite verb of a different morphological type. Medial and final verbs are clearly of unequal rank. But such chaining linkage with medial and final clauses is never employed to encode comparatives in the clause chaining languages of our database. Instead, they encode comparison with juxtaposed or conjoined independent clauses with predicates of equal rank. 15 For example, Kombai uses clause chaining to encode temporal chains but these cannot be used to encode comparatives. In (79a-b) the medial verbs umora and umo are non-finite whereas the final verb xano is a finite verb, clearly a deranking strategy in the terms of Stassen (1985). Yet, Kombai comparatives are always encoded by two juxtaposed independent clauses (79c).
canoe this big that little-ADJ 'This canoe is bigger than that one.' (lit. 'This canoe is big, that one is little') (de Vries 1993: 36) Stassen's Universal 4 (1985: 138) claims that "If a language has a Conjoined Comparative, it must have a balanced simultaneous construction." This claim does not seem to hold either since languages like Kombai (and many other Papuan languages of central and south New Guinea) that have conjoined comparatives seem to encode simultaneous temporal chains exclusively or primarily with deranking strategies with non-finite medial verbs followed by final clauses with fully finite verbs.
15 There is one possible exception in our database to the rule that Papuan languages with clause chaining constructions do not use such constructions for comparatives, Mauwake (example 35a). This is not a clear case since Berghäll (2015: 192) describes this example as a construction halfway between clause chaining and a serial verb construction. See the discussion in Section 4.1.4.

Conclusion
This paper is part of the burgeoning study of Melanesia as a linguistic area and contributes to the typology of comparative constructions.
In our study, clear areal patterns are revealed by the distribution of comparative types. Concentric circles form around New Guinea, delineating regions of Austronesian-Papuan interaction. Our research confirms that 'core' New Guinea, the innermost of the concentric circles with no history of Austronesian contact until modern times, has only conjoined comparatives. But our findings add serialised EXCEED comparatives as a significant presence in the traditional Austronesian-Papuan contact zones in the coastal and island belt around New Guinea and adds the emergence of particle and adpositional comparatives in a few areas with intensive bilingualism with contact languages such as Indonesian and related regional Malay varieties.
The study of the languages of Linguistic Melanesia presents new ground for testing typological theories. As one of the few regions in the world where conjoined comparatives are common (to the exclusion of other types), Linguistic Melanesia offers the possibility of testing Stassen's (1985) claims that comparative types can be predicted from temporal chaining types. We found little to support this claim for core languages of Melanesia with conjoined comparatives. The areal pattern of conjoined comparatives on New Guinea, we suggested, instead possibly reflects a tendency in these languages to distribute arguments over series of mini-clauses.
The striking areality of the conjoined comparative type in the core New Guinea area counters the claims of some recent typological and descriptive work on New Guinea that seeks to emphasise the diversity rather than the unity of Melanesian languages. As a region with a large number of languages and population history going back perhaps 60,000 years, Linguistic Melanesia is a complex area. But as our study shows, there are still significant advances to be made in understanding the distributions of different typological features and what these can tell us about the history of the region.