Baring the bones: the lexico-semantic association of bone with strength in Melanesia and the study of colexi ﬁ cation

: In this article I demonstrate that there is a pervasive lexico-semantic association BONES ARE STRENGTH in the languages of Melanesia, but that its linguistic expression is highly varied; languages are scattered along a lexical-to-clausal cline in their expression of the association between bone and strength, with a large number of language-speci ﬁ c idioms based on the association to be observed in Melanesia. I argue that the striking areality of this lexico-semantic association is readily missed in top-down approaches to lexical semantic typology that rely, for instance, on databases of word lists, or on narrow search domains limited to the meanings of simplex lexemes.

domains of phonology and morpho-syntax, linguistic typology has developed an increasingly detailed picture of recurrent linguistic features, of correlations between linguistic features, and of correlations between linguistic features and non-linguistic phenomena. The typological investigation of the lexicon has, however, lagged behind the study of these other elements of linguistic structure (despite several wellknown and highly cited studies on a range of semantic fields, including Andersen 1978;Berlin and Kay 1969;Talmy 1985;Viberg 1984), both in the number of studies and the diversity of lexical phenomena treated. It is debatable whether the lack of typological research into the lexicon is due to it being regarded as merely "an appendix of the grammar, a list of basic irregularities" (Bloomfield 1933: 274), or as simply too massive and diverse, especially in comparison to the more constrained parameters of morphosyntax. However, a new wave of lexical typological research in the last 15 years has broadened the cross-linguistic study of the lexicon beyond the small number of semantic fields dealt with in earlier studies through systematic analysis of lexico-semantic associations and morphological motivations for a wide range of lexemes (e.g., Koptjevskaja-Tamm 2015;Koptjevskaja-Tamm and Vanhove 2012;Urban 2012;Vanhove 2008).
Within this recent body of research, the colexification approach established by François (2008) has gained wide currency among typologically oriented linguists. Sidestepping "the vexing problem of distinguishing between polysemy and vagueness" (Haspelmath 2003: 231), the colexification approach constructs expandable etic grids of meanings on the basis of semantic distinctions lexicalized in languages, regardless of the emic status of the meanings in any one language. Since François (2008), there have been several studies in lexical typology which make use of colexification (most notably, Urban 2012). As a methodology, it has produced robust and innovative results, not least of all in the domain of areal linguistics (e.g., Koptjevskaja-Tamm and Liljegren 2017;Schapper et al. 2016;Schapper 2017Schapper , 2019Urban 2009Urban , 2010, and the articles in this issue). The emergence of colexification as a new tool for lexical typological research has been concomitant with the growth of digital databases of lexical material from across the languages of the world, chiefly for the purposes of historical-comparative linguistic research (e.g., WOLD, Haspelmath and Tadmor 2009;IDS, Key and Comrie 2015;ASJP, Wichmann et al. 2018). Inevitably, these databases have been co-opted for use in colexification studies concerned with identifying patterns or skewings of patterns on a global-scale (e.g., Gast and Koptjevskaja-Tamm 2018;Jackson et al. 2019;Östling 2016;Pericliev 2015) and given rise to collated databases for automized detection of colexifications, most notably the various incarnations of CLICS (Version 1: List et al. 2014, Version 2: List et al. 2018, Version 3: Rzymski and Tresoldi 2020). In this contribution, I draw attention to the pitfalls of simplistic implementations of the colexification approach using databases based on naïve wordlists that require complete identity between cells for a colexification to be detected.
Colexifications are, at their broadest, lexico-semantic associations 1 that can manifest themselves linguistically in different languages through different formal means, including the extensional range of a single lexeme's semantics, complex lexical items such as compound words and phrases, or larger constructions, formulae and idiomatic expressions. I address this issue here through an examination of the lexico-semantic associations of lexemes denoting 'bone' in the Melanesian Linguistic Area. Centred on the island of New Guinea, Linguistic Melanesia is defined as the area in which Papuan languages and their influences are found (Ross 2017;Schapper and de Vries 2018;Schapper 2020). It takes in upwards of 1,000 languages across 20 to 40 language families; Papuan languages from numerous unrelated families dominate at its core, but a large number of Austronesian languages are also taken in, particularly in the insular and coastal regions on and around New Guinea. 2 In this article I demonstrate that there is a pervasive lexico-semantic association BONES ARE STRENGTH in the languages of Melanesia, but that its linguistic expression is highly varied; languages are scattered along a lexical-to-clausal cline in their expression of the lexico-semantic association between bone and strength, with a large number of language-specific idioms being based on the association observed in Melanesia. I argue that the striking areality of this lexico-semantic association is readily missed in top-down approaches to lexical semantic typology that rely, for instance, on databases of word lists, or on narrow search domains limited to the meanings of simplex lexemes.
The paper is structured as follows. Section 2 discusses the colexification approach and associated issues. Section 3 presents an overview of common patterns of cross-linguistic colexification for the target meaning of this article, 'bone'. Section 4 discusses the numerous lexical patterns and expressions underpinned by the lexico-semantic association BONES ARE STRENGTH in the languages of Melanesia, highlighting problems in capturing the cross-linguistic similarities in these when using a narrowly applied colexification approach. Section 5 argues for the areality of the association of bone with strength in Melanesia and suggests that the diversity of its manifestations in the languages of Melanesia fits in with the 1 Following Koptjevskaja-Tamm and Liljegren (2017), I use the term "lexico-semantic association" as a cover term for colexification, lexico-constructional parallels and shared phraseology. 2 The core of the Melanesian Linguistic Area is the island of New Guinea. Greater Melanesia takes in a large area to the east and west of New Guinea, extending from the island of Sulawesi in Indonesia to the western fringe of Polynesia and stopping just short of the islands of Fiji, Vanuatu and New Caledonia (Schapper 2020).

Colexification of bone and strength in Melanesia
well-known structural convergence-lexical divergence model for the area. Section 6 concludes.
2 What is colexification? François (2008) articulates a practical method for detecting shared lexical packaging of senses across languages based on the simple concept of 'colexification': "A given language is said to colexify two functionally distinct senses if, and only if, it can associate them with the same lexical form" (François 2008: 170). Confusion about what "functionally distinct senses" are has been the basis of criticism of the colexification approach. Although generally positive about colexification as a development in lexical typology, McConvell and Ponsonnet (2013) observe that it is often impossible on empirical grounds to decide whether two meanings found in two different languages are one or two senses. However, François (2008) does not require us to distinguish; the phrase "functionally distinct senses" is intended to refer to any semantic distinction potentially lexicalized in a language, with no regard to whether the speakers of any one language view the meanings associated with a lexeme as psychologically distinct senses, contextual readings of the same sense, or accidental homophones. Stating a colexification, therefore, involves forming an etic grid on the basis of the semantic distinctions lexicalized in a sample of languages, without regards to the emic status of the meanings associated with a lexeme in any one language. For instance, Table 1 represents an etic grid of three meanings 'fire', 'firewood', and 'tree', lexified distinctly in four Papuan languages: Kalam colexifies all three meanings; Kamang colexifies 'fire' and 'firewood', but has a distinct lexeme for 'tree'; Eipomek colexifies 'firewood' and 'tree', but has a distinct lexeme for 'fire'; and, Yimas has distinct lexemes for all three (cf. Schapper et al. 2016). Burenhult et al. (2017) implicitly criticize the colexification approach, suggesting that in practice it is little more than the comparison of translation equivalents. In their study of 'forest', they state that looking for possible translation equivalents for it is "a simplistic and selective approach to lexical semantics, and one that does not do justice to the complexities of language-specific lexical systems" (Burenhult et al. 2017: 456). Urban (2012: 41) counters these kinds of criticisms by pointing out that "while there is not necessarily a set of lexical items whose semantics corresponds to, say, English words such as mountain, river, etc., there are lexical items that allow [one] to refer to these configurations in the landscape, and it is not least the variation in the other semantic areas covered by the respective expressions that are […] important." Because colexification studies are still very much in their infancy, there is a need for a discussion and evaluation of the comparability of the lexical structures under study. I draw attention here to two particular issues, neither of which are new in the broader semantic typology literature. However, to my knowledge, they have not been raised in any significant way in the emerging colexification literature, but should be kept in mind in the development of future colexification studies.
The first issue surrounds the morphosyntactic analysis of colexifications. Two subtypes of colexification are identified by François (2008): (i) strict colexification where the exact same form in synchrony can be used for senses A and B (e.g., Kamang ati for 'fire' and 'firewood'), and (ii) loose colexification 3 where there is a connection between the two either diachronically (e.g., Waskia tama 'fire' and tam 'firewood' are obviously historically related, but not by any identifiable regular, synchronic process, Malcolm Ross pers. comm.) or morphologically (e.g., Klon adaʔ 'fire' and adawel 'firewood', a compound of 'fire' and 'burn (intr.)'). 4 In identifying loose colexifications, Urban (2012) correctly makes no distinction between compounds (e.g., noun + noun) and phrases (e.g., noun + adjective), treating both as types of modifier of the head noun. As we shall see in this paper, 3 Schapper et al. (2016) refer to the presence of colexifying elements in compounds as "subcolexification". That is, sub-colexification for them is a special kind of loose colexification which involves compounding of lexemes, as opposed to marking with grammatical morphemes. Urban (2012) uses "analysability" to characterize morphologically complex expressions formed by any language-specific syntactic or morphological process of word-formation. 4 "Loose colexification" covers much of what has been called "heterosemy", the situation in which different but related senses of a given lexeme are associated with distinct morphosyntactic contexts (Lichtenberk 1991). Note, however, that François (pers. comm.) regards "pure heterosemy" as strict colexification. For Evans and Wilkins (2000), these heterosemies are situations in which there is no morphological marking or derivation on the lexeme to distinguish different senses (e.g., Kómnzo mni with feminine agreement for 'fire' and mni with masculine agreement for 'firewood', or instances of zero conversion like hand as body part noun and hand as a verb). the same lexico-semantic association can manifest itself on the level of the simplex lexeme, the morphologically derived lexeme, the phrase, or even the clause. François (2008: 170 fn 4) appears to recognize this issue in his important article, writing that senses can be shared not simply between lexemes but also constructions. I will show that identifying lexico-semantic associations on the different levels of linguistic structure is important for the detection of shared patterns in different languages and language areas.
The second issue concerns the need to attend to the contexts of a lexeme's use. These are important not simply for explaining the existence of colexifications, but also in detecting ongoing semantic extensions leading to colexifications. Evans and Wilkins (2000) observe that when a lexeme's semantics are first extended, the new meaning often comes into existence in a regular context that allows an inference of a second meaning to arise and, only given sufficient frequency, to later become lexicalized. They rightly point out that attention to these "bridging contexts" is essential: "speech participants do not detect any problem of different assignments of meaning to the form because both speaker and addressee interpretations of the utterance in context are functionally equivalent, even if the relative contributions of lexical content and pragmatic enrichment differ" (Evans and Wilkins 2000: 550). Looking for uses of target lexemes in context is particularly important for languages where linguistic documentation is relatively basic and the lexicon has not been the subject of close semantic analysis. In the course of the present study, contextualized attestations of 'bone' in many cases revealed the relevant, often repeated, lexico-semantic association of BONES ARE STRENGTH and extended senses, even though not listed as a sense of 'bone' by individual linguists or linguistic reference works. In addition, by examining the morphosyntactic contexts in which lexemes denoting 'bone' were used with a contextual association to strength in different languages, a picture of the staged emergence of the lexico-semantic association of BONES ARE STRENGTH was constructed.

Colexification patterns of 'bone' worldwide
Before turning to the Melanesian data with which we will interrogate these problems, it is useful to have a picture of world-wide variation with which to establish the areality of the linguistic phenomenon in question. The meaning 'bone' is observed by Urban (2012: 621-622) to have relatively few lexical associations crosslinguistically, compared with the other nominal meanings covered in his study. This view is largely borne out by the synchronic colexification patterns discovered in my own investigation as well as through an examination of the CLICS databases 5 and the Database of Semantic Shifts (Zalizniak et al. 2012). Only a small number of synchronic colexifications for 'bone' are found repeatedly the world over, as will be shown in this section. 6 We will see that the rarity of an association of 'bone' with strength makes it a good candidate for an areal linguistic feature.
The most common lexical associations of 'bone' are far and away contiguitydriven; colexification of 'bone' is overwhelmingly with (parts of) the body. These colexifications are found sporadically the world over and rarely, if ever, show any strong areal or genetic skewing. Specific bones of the body that colexify with 'bone' in the world's languages are 'shin' (Kartujarra, Catio, Guayabero, Central Tunebo and eight Atlantic-Congo languages: Chuka, Gwere, Kiseri, Mabia, Meruimenti, Rukiga, Sambaa and Zanaki), 'spine' (Rumanyo, Upper Chehalis, Maca, Kumyk, Takia) and relatedly 'back' (Tabriak, Lau, Walade, Vano, Flinders Island, Kalali, Mirniny), 'rib' (Tuamotuan, Alorese, Kula, Paliu, Bolewa, Mang'an B), 'shoulderblade' (Upper Chehalis, Yavitero, Avar), 'ankle' (Karata), 'collarbone' (Mlabri) and 'jaw' (Kimbu), as well as 'tooth' in Oro Waram, Mapudungun and several Madang languages (Isabi, Ogea, Siane, Uyajitaya). Colexification of 'bone' with 'leg' is synchronically attested in Kartujarra, Guayabero, Central Tunebo, Epena, Ese Ejja, Mundurukú, Lengua, Kobol, Moere, and most well-known of all, in many of the Germanic languages, while 'bone' colexifies with 'foot' in Siraya, Embera Tadó, Amto and Nagatiman. We find 'bone' and 'head' colexify in Momuna and three unrelated languages of New Guinea (Konda, Duna, Iau). Chorote, several Daghestanian languages (Andi, Archi, Avar, 5 Note that when I refer to "CLICS", I mean the first, second and third editions of CLICS collectively. While CLICS 2 and CLICS 3 have expanded data sets and thus greater coverage, CLICS 1.0 allows the user to see all colexifications no matter how infrequent their attestation in the database. The interfaces in CLICS 2 and CLICS 3 set a threshold of three attestations for a colexification to be outputted by a user's query. While such a threshold might be a principled way for a researcher to draw a line between accidental homonymy and actual polysemy, for an interface to mask data in this way is extremely unfortunate. It means that using the interface developed by the CLICS team actually prevents the researcher from making connections between colexifications that are recurrent in the data, but might be glossed in slightly different ways, perhaps as an artefact of the way a wordlist was collected. A case in point is the colexification of 'bone' with 'stick' or 'branch' in different languages; clearly both reflect a biological analogy between the bone of a human/ animal and the woody parts of plant. 'Bone'/'stick' colexification is shown in CLICS 2 because it has four attestations, but 'bone'/'branch' colexification is not because it only has two occurrences in the data, as reflected in CLICS 1.0. 6 The reader is referred to the appendices for further data used for this section, as well as ISO 639-3 codes. Appendix I is my database on the association of 'bone' with strength in Melanesian languages, but also includes examples of other associations of 'bone' that are common in Melanesia. Appendix II contains summaries of the data from online databases that were used. Urban's (2012: 621-622) discussion of associations 'bone' does not always include the forms of the items he discusses and so is not represented in Appendix II.
Colexifications with 'bone' based on biological analogies are attested widely cross-linguistically. These colexifications with 'bone' mostly refer to a homologous structure in animals and plants, and occasionally, inanimates. While most vertebrates have bones much like humans, there are other animal parts which are also treated as homologous structures, particularly in non-land animals. This is seen in the varied colexification data: 'bone' colexifies with 'horn' in Yakut, Highland Chontal and Itzaj; with 'fishbone' in Buli and Basque, but 'fishscale' in Lezgi (Mikrakh), 'projecting spike on the fins of certain fish' in Rotuman, and 'dorsal fin' in Mwotlap; with 'shell' (e.g., of a mollusc, turtle) in San Lucas Quiaviní, Qawasqar, Nama and Hawaiian, and; with 'egg' in Yangulam (as well as 'fruit'), Kwoma and Ngoreme (as well 'reed' and 'thorn' in the latter, see below).
What colexification patterns capture as homologous structures in plants are similarly diverse. 'bone' colexifies with various woody or structure-giving parts of the plant, including with 'stalk' in Ngaanyatjarra, Sora, Arabela, Lengua, Maxakalí and Yanomámi, with 'branch' in Seri and Yuwana and, relatedly, with 'stick' in Secoya, Garus, and Mungkip, with 'stem' in Aghu, with 'tree trunk' in Northern Emberá, with 'tree' and 'wood' in Colorado, with 'tree' and 'stick' in Tsafiqui Pila, and with 'tree' in Qawasqar and Kalkutung. Less frequently, 'bone' colexifies with reproductive parts of the plant, typically, as contained in hard casings; most common in this type is the colexification of 'bone' with 'seed', found in 11 languages (Sinhalese, Karajá, Rumai, Bahinemo, Mende and a cluster of six Austronesian languages). Related colexifications are 'bone' with 'stone (of a fruit)' in Modern Greek, Latin, Lahu Shi, Mongolian (Khalkha), Middle Persian (Pahlavi), Khmer and Yakut and with 'hard inner shell' (e.g., of a coconut or mango) in Melanesian languages such as Bosavi, Mende and Kâte, to name a few. 7 We also find 'bone' colexifying with 'thorn' in Jita and Mwotlap (along with many other meanings in the latter), with 'reed' in Totoró and Sema, with 'reed' and 'thorn' in Ngoreme, and with 'midrib' or 'central nerve' in reference to a leaf in Aisi, Mwotlap and many other Melanesian languages.
In Melanesia and Oceania, it is common for 'bone' to colexify with one or more items used as supporting frameworks. For example, Buang seḳë is glossed 'bone, forearm, framework, supporting structure, building materials'. Selepet has emet tȃn-ŋe house bone-3SG for 'frame for house' and manam tȃn-ŋe banana bone-3SG 'prop, support for a banana tree', while Kâte has fic siec house bone 'framework of a building' and jȃc siec tree bone 'poles for a wooden structure, poles for the structure of a bush house'. Tuamotuan ivi colexifies 'bone' and 'ridgepole', while Mwotlap hyē colexifies 'bone' with 'ridgepole roping' which is described by François (2020) as the 'diagonal reinforcement supporting the roof structure linking the tie beams to the ridge beam [of a house]'.
The lexical associations of 'bone' we have seen thus far have all been with nominal meanings that have concrete referents, but 'bone' is also observed to have associations with property concepts. These are highly infrequent. The meaning 'bone' colexifies with 'dry' in Burarra, Guraní and Meyah, with 'hard' in Ecun Buyang, Wappo and Bororo, and finally, with 'strong' (or 'strength' in some unspecified cases noted by Urban 2012: 622) in Ngambay, four Papuan languages of New Guinea and Ontong Java, an Austronesian language of Melanesia. The geographic clustering of the latter five languages suggests areal convergence may be at play, and it is this possibility that is taken up in the subsequent sections of this paper.

Lexico-semantic association of bone with strength in Melanesian languages
In what is now a canonical case of language contact, Ross (1996Ross ( , 2001 describes "metatypy" on the basis of two unrelated Melanesian languages, Takia and Waskia. These have come to have radically parallel linguistic organization as a result of Takia speakers' bilingualism in Waskia. A major component of metatypy is semantic reorganization whereby "equivalent lexical items in Takia and Waskia have the same range of meaning, closed sets of morphemes have similar membership and semantic structure, and complex lexical items, whether compound words, phrases, or larger formulae have been reformulated so that their component morphemes are the same as their Waskia equivalents" (Ross 2001: 144). Among the many features of semantic equivalence between them, Ross (2007) names Takia tatu and Waskia sokel as sharing the meanings 'bone' and 'strength'.
A lexico-semantic association of bone with strength in fact goes well beyond Takia and Waskia; it is found widely in Melanesia, being expressed through a large variety of means across all levels of semantic organization named by Ross, both within and across languages. The similarity of the great variety of lexical expressions constructed at different morphosyntactic levels can, I argue, be captured using the formula BONES ARE STRENGTH. By encapsulating the Melanesian lexicosemantic association in this way, it is my intention to abstract away from languageparticular idiosyncrasies in expressions and to capture the association that sees the having of bones as giving strength on the broadest level. Before turning to an examination of the Melanesian data, it is important to note the partial character of the documentation of most languages in the area. This study is based on a survey of 180 languages in and around the Melanesian region. Data was gathered from dictionaries, grammars, texts, and through consultation with field linguists. However, even the lengthiest sources were often rather minimal in terms of the lexicographic detail they provided. Relevant collocations or complex expressions that would provide evidence for a lexico-semantic association of bone with strength were frequently not included in lexical materials. Or, as is the case for Takia and Waskia, an association may be noted in glosses, but examples illustrating contextualized use of the different senses were lacking. 8 In total, 75 of the surveyed languages, all within the Greater Melanesia area, evinced a clear synchronic association of bone with strength. For those that did not, it is suspected that more complete materials, particularly texts, would in many cases yield a different result.
In what follows, the reader will encounter numerous lexical patterns and expressions underpinned by the association BONES ARE STRENGTH. Data was harvested from sources by identifying item(s) meaning 'bone' and then creating lists of all lexemes, phrases and expressions that involved the 'bone' lexeme. 9 This "wide net" data collection strategy revealed, in addition to a large data set for the BONES ARE STRENGTH association, linguistic expressions that are underpinned by several other recurrent lexico-semantic associations involving 'bone' in Melanesia. 10 Only a portion of the amassed materials will be used to illustrate the large variety of the linguistic guises in which the BONES ARE STRENGTH association finds expression. For a complete set of the compiled materials by language, including details of all languages sampled, textual examples (where available), and sources, the reader is referred to Appendix I.

'bone' = 'strong' and the pathways to colexification
The Melanesian association BONES ARE STRENGTH can be manifested in an apparent strict colexification of 'bone' and 'strong'. In some of these languages, the colexification is heterosemic, with the different senses being associated with different morphosyntactic contexts. For example, Leti ruri has the sense 'bone' in nominal contexts (1a), but 'strong' in verbal contexts along with the expected morphology, i.e., in 9 In a few cases I also looked at particular target passages from translations of the New Testament, e.g., Matthew 12:29, Luke 16:3, John 6:18, Romans 15:1. 10 In addition to the many linguistic expressions underpinned by the most frequent association BONES ARE STRENGTH, there are two other recurrent associations involving 'bone' encoded by diverse expressions that came out during data compilation. The first is CONSISTING OF BONES IS BEING THIN/HAVING ONLY BONES IS BEING THIN. This very widespread association shows frequent extension from the meanings of 'thin', 'skinny' and 'bony' to others such as 'unhealthy', 'weak' and 'wasted'. Examples of the sorts of expressions coding the association include: Aghu bigi da only bone 'thin (of person)' (van den Heuvel 2016), Mali Baining slēp pēt viso bone LOC AUX 'skinny, bony' (Stebbins and Tayul 2012), Motuna koona nong bones only 'anorexic, unhealthy, very thin (person)' (Masa Onishi pers. comm.), Orokolo ukiila with bones 'bony, in a wasted condition'. There is a further, but much rarer, extension of the HAVING ONLY BONES IS BEING THIN association to refer to the elderly, e.g. Oksapmin hän tam sän person bone container/top 'old man' (Lawrence 2006), Komnzo ausi kam∼kam woman REDUP∼bone 'old woman' (Christian Dӧhler pers. comm.). Note that the HAVING ONLY BONES IS BEING THIN association, where the presence of bones indicates thinness/weakness, is not to be confused with the negative polarity version of BONES ARE STRENGTH, although their translated meanings on occasion converge. In a few languages, LACK OF BONES IS LACK OF STRENGTH is extended from 'weak' to 'skinny', as in Mende hava-mba bone-without 'be skinny' (Nozawa 2006), but here it is absence of bones (rather than their presence) that gives rise to extension to 'skinny'. The second recurrent, but rare association was THE BONE IS THE IMPORTANT/MAIN PART. This association is found, for instance, in the Mek languages, for example, Yale yoo ina bone path 'main path, important way', Eipo asik yok-nang hamlet bone-people 'people of a central hamlet (one with a men's house, not a newly founded garden village)' (Volker Heeschen pers. comm.). predicative function marked by either a stative verb suffix (1b) or a dynamic verb prefix (1c), and in attributive function with dependency marking reduplication (1d).
(1) Leti 11 a. pipruri pipi-ruri goat-bone In other sampled languages, however, we find no morphological marking associated with different syntactic contexts in which the meanings 'bone' and 'strong' are found. The Papuan language Ambulas typifies this pattern: apa can be used nominally to mean 'bone' (2a) and predicatively to mean 'strong', either as a stand-alone predicate (2b) or with do-support (2c). Apa is also used adnominally with the meaning 'strong' (2d).
(2) Ambulas a. apa dé tékwiyaa-k bone 3SG broken-PST 'The bone was broken.' b. wani nyéga apa yamarék yate dé nyéknwu DEM leaf bone NEG and 3SG go.slack:3 'That leaf is not strong and it goes slack.' c. apa mé ya! bone IMP do 'Be strong!' d. apa du bone man 'strong man' In each of these syntactic appearances, apa behaves in accordance with the word class properties that its respective meanings would predict. That is, in its nominal appearances it behaves as a noun would be expected to, while predicatively and adnominally it shows behaviour consistent with a stative verb/adjective. As such, Ambulas apa can be taken to represent a straightforward case of strict bone/strong colexification. Ambulas is unique in this, due to its unusual word order profile: like most other Papuan languages it has SOV and Genitive-Noun orders, but atypically for a Papuan language it combines this with Demonstrative-Noun, Adjective-Noun and Relative clause-Noun orders. This means that there is no way in Ambulas to distinguish between adjectives and other kinds of noun modifiers based on their ordering relative to the noun. This is, however, not the case for the vast majority of Melanesian languages, where most modifiers, apart from possessors, follow the noun. A case in point is Aisi which, at least superficially, has strict bone/strong colexification. Nominally, dagar has the meaning 'bone' and by extension 'strength' (3a), but it can also appear meaning 'strong' predicatively (3b) and adnominally (3c). In Aisi, an adjectival modifier would normally follow the noun (Don Daniels pers. comm.). In (3c) adnominal dagar appears before the head noun ki 'speech' as a possessor would, thereby indicating that dagar synchronically has the status of a noun. If this analysis is correct, the predicative appearance of dagar in (3b) would need to be analysed as an equative constructive (literally, 'your livers are not bones').
( The fact that Aisi dagar does not have the expected syntax of an adjective in its adnominal use, suggests that conversion is not taking place and that dagar is still a noun. In another language with bone/strong colexification, we do have positive evidence that conversion takes place. In Pamosu, a modifying noun precedes the head noun, as in (4a). However, when kenav 'bone' is used with the meaning 'strong' it unexpectedly follows the head noun (4b). Pamosu has a further restriction: kenav does not appear predicatively with the meaning 'strong'; the adjective pelat 'strong' is used in predicative contexts (Ian Tupper pers. comm). These features set the Pamosu colexification apart from that presented by Aisi (3c).
The morphosyntactic differences in these appearances of 'bone' lexemes with the sense 'strong' goes back to the diversity of possessive structures and their functions in Melanesian languages. I posit that the starting point for the emergence of the colexification of 'bone' and 'strong' is the contextual use of 'bone' in an adnominal possessor construction to mean 'strong'. This can be illustrated with the Ekari expression in (5a) where mitoo 'bone' modifies meː 'man'. Here, the modifying noun precedes the head noun, consistent with a possessive noun + noun compound structure (as in 5b). Notably mitoo meː is the sole known expression in Ekari where mitoo is used with a contextual relationship to strength, but it is presumably from such a construction that the pattern we see in Aisi (3c) above was extended to modifying more referent types. 12 (5) Ekari a. mitoo meː bone man 'strong man' b. boko mitoo breast bone 'breast bone' Alternatively, the 'bone' lexeme that contextually comes to denote 'strong' may begin constructionally as the possessed item. It is common in parts of Melanesia 12 A likely progression for the contextual extension of 'bone' to the meaning 'strong' is of male > human > animate > inanimate. In fact, it is quite common in my data that 'bone' can only be used with an association to strength for animates. See, for example, Bilua's use of piza 'bone' to mean 'strong', which is limited to human referents, while in Buin kaana-raake 'bone-with/have' is used for 'strong' of both humans and animals (discussed further in Section 4.2).
for attribute constructions to have the morphosyntax of possession: the noun denoting the class of referent behaves as if it were the possessor and the attribute as if it were the possessed (Ross 1998). We see in (6a) that Jabêm têkwa 'bone' denoting 'strong' has the identical morphosyntax to the possessed item leguʔ 'tail' in (6b). Such a construction is the likely source for the use of Pamosu kenav after the head noun (in an adjective-like position) to mean 'strong'. The important point for the reader to take away here is that there are subtle differences between languages in the morpho-syntactic behaviour of 'bone' lexemes when expressing the meaning 'strong' that are indicative of the pathways along which colexifications emerge. While these details do not impact on their status as colexifications for the model as developed by François (2008), they do have implications for how we detect and collect data on such lexico-semantic associations. Because the morphosyntactic contexts in which 'bone' and 'strong' colexify in these languages are highly varied, elicitation situationssuch as when collecting a word list or completing a questionnaireare unlikely to be sufficiently detailed for heavily contextualized or restricted colexifications of 'bone' and 'strong' to come out, especially where they exist alongside a more general property word meaning 'strong', as is the case with most of the languages examined in this section.
(8) Buin -raake a. inna-raake blood-raake 'mess of blood, bloody' b. koni-raake lie-raake 'lying, deceitful' c. ina ruka iko-raake wife:1SG man:PL DEM.M-raake 'my wife with all those men' d. ako noke mokokotu minno-paaroi karapana-raake DEM some spear follow-TAM poison-raake 'some spears accompanied by poison' Section 4.4 deals with the different extended senses of BONES ARE STRENGTH that are present in the surveyed language data. e. mareki-raake axe-raake 'with an axe' f. ne ia-raake nonumoi tentemu 1SG.EMPH DEM.SG-raake one.thing only 'I have only this one' The problem for a comparative study here is that it is not clear that there is a truly principled way to separate out instances using morphology, like Buin kaana-raake bone-with = 'strong', from languages such as those in (9). In these, phrases composed of a free adposition meaning 'with' plus a noun 'bone' are used to express 'strong'. The adpositions have a range of functions that overlap with those of Buin -raake, introducing comitative and instrumental participants etc., and thus differ to a large extent only in their boundness. This appears to be a case of "what is one language's morphology is another language's syntax". From the point of view of the current study, if we were to draw a strict line at the simple lexeme or even morphologically derived lexeme as the sole unit of investigation, to the exclusion of phrases, then the lexico-semantic association of bone with strength that both the languages in (7) and those in (9) share would not be captured. If we do regard syntactic units such as adpositional phrases 'with bone' = 'strong' alongside complex lexical ones as comparable analytical units, then the question quickly arises as to whether it is principled to exclude non-phrasal predications of similar meaning. Numerous sampled Melanesian languages express 'strong' by means of a clause in which a form meaning 'bone' is the subject of the clause with an 'exist' verb as predicate; the person whose strength is at issue is encoded as the possessor of the subject 'bone'. Examples of this type are given in (10) It seems apparent that clauses such as those in (10) are actually just another linguistic form of expressing what is captured by derived lexemes in (7) and adpositional phrases in (9). Aiton (2016: 138-142) even points out in relation to the Eibela example (10c) and other similar constructions in that language that these clauses are readily used adnominally and show signs of lexicalizing into adjectives. 15 In short, the lexico-semantic association BONES ARE STRENGTH is expressed in the languages of Melanesia not simply through simplex or complex lexemes, but also by constructions that are phrasal and even clausal. It is by no means novel to regard the boundary between grammar and lexicon as fluid (see, e.g., Fillmore et al. 1988). Nonetheless, in the context of increasing moves to automatize colexification studies, it is important to recognize the limitations of a lexeme-based view of the lexicon.

Idioms associating bone and strength
Phrasal and clausal constructions expressing the association BONES ARE STRENGTH qualify as idioms in that they are polylexemic expressions of a single semantic unit that cannot be understood as having the literal meaning of their constituent parts.
In comparing the different expressions across languages, we can see that the constructions used in many languages are striking similar. For example, Table 2 presents the constructions expressing LACK OF BONES IS WEAKNESS, the negative polarity version of the BONES ARE STRENGTH association, that are made up of the lexeme 'bone' plus a negative morpheme (and, where necessary, a verb). In each language, the exact make-up of the construction is different according to the specific languageinternal resources available. Their meanings (as represented here by glosses taken over from individual linguistic descriptions) are also not exactly same across languages, but are broadly similar enough to be regarded as instantiations of the same lexico-semantic association.
The lexico-semantic association of bones and strength is also widely expressed in the languages surveyed with constructions of the shape 'bone' + 'weak/soft' (Table 3) and 'bone' + 'strong/hard' (Table 4). Again, the meanings expressed by these constructions are not identical, but can still be traced back to a common association, BONES ARE STRENGTH. More specifically, we could describe these expressions as representing a figurative seating of strength in the bones. Melanesian languages (and perhaps all languages, according to Enfield and Wierzbicka 2002) typically have a range of fixed expressions that convey emotions and personal characteristics involving a figurative seating in the body. These "seats" manifest themselves in linguistic constructions involving, often lexicalized, body part + predication constructions, such as is the case in Tables 3 and  4. Typically, one body part will be the main seat of emotions/traits in a language; it is frequently the case, however, that particular semantic clusters of emotions/traits (e.g., anger, violence, etc.) will be seated in their own distinct body part. 16 For many of the languages sampled here, idioms expressing the association BONES ARE STRENGTH and LACK OF BONES IS WEAKNESS were not simply a matter of negated (i.e., 'bone exist' = 'strong' vs. 'bone not exist' = 'weak') or antonymous constructions (i.e., 'bone strong/hard' = 'strong' vs. 'bone weak/hard' = 'weak'). Rather, a large number of  idioms of unique or near-unique formation were attested. For example, while Mehek 'with bone' = 'strong' (11a) finds close parallels in numerous languages in the sample, the antonymous idiom 'empty bone' = 'weak' (11b) is unique, even though it is clearly similar in import to 'bone' + negative morpheme = 'weak' expressions in Table 2.
(11) Mehek a. yefa kana b. yefa timba bone with bone empty 'be strong' ' be weak' In fact, there is a huge amount of variation in the form of idiomatic predications involving 'bone' to express 'strong' and 'weak'. Examples in (12) and (13)  ' be weak' I do not claim that the different idioms presented in this section are interchangeable in all contexts. The variation in glosses associated with 'bone' + predication constructions plus the fact that numerous languages surveyed exhibit multiple 'bone' + predication constructions make clear that there are semantic differences between the various idioms. For our purposes, however, it is enough that they possess a common image component on the lexical level: these idiomatic expressions use lexemes for 'bone' to figuratively stand for strength as part of a common association which I have expressed as BONES ARE STRENGTH. In Dobrovol'skij and Piirainen's (2010) typology of idioms, this means the idioms discussed here have a "symbol-based motivation", that is, they exploit 'certain cultural conventions based on the ability of material objects to "stand for" some non-material entity' (Dobrovol'skij and Piirainen 2010: 79). In this Melanesian case, bones are shared cultural symbols of strength. Whilst the association is shared, the ways to express it are language-specific; a handful of recurrent constructions with 'bone' lexemes exist alongside a large number of unique idioms involving 'bone'. Recognizing the shared association despite the diversity of idiomatic constructions is crucial in order to identify the areal pattern here.

STRENGTH
In the previous sections, we have focussed on the use of the BONES ARE STRENGTH association with respect to physical strength in humans and animals (i.e., HAVING BONES IS HAVING PHYSICAL STRENGTH). Whilst this is certainly the most widely found sense of the association, the different glosses we have seen accompanying idioms have hinted at the numerous semantic extensions found in the data. In this section, I illustrate the patterns of extension of the BONES ARE STRENGTH association that came out of the gathering of contextualized attestations of 'bone' lexemes. The reader is referred to Appendix I for the complete dataset. Figure 1 presents a tentative semantic map of the various semantic connections between the senses that are expressed by BONES ARE STRENGTH. The ordering of senses on the map represents my own intuitions about what constitutes a semantically plausible ordering based on my reading of the data. I also choose for simplicity's sake in the visualization to link the senses I distinguish together in relatively linear chains. In reality the relationships between senses are more web-like with many connecting lines running between different senses. According to common semantic properties, I group the senses reflecting the BONES ARE STRENGTH association into three clusters, which will each be discussed briefly in what follows.
The first cluster in the lower left-hand corner of the map pivots on the association HAVING BONES IS HAVING PHYSICAL STRENGTH. Used typically in reference to humans, the association has three recurrent semantic extensions in the data. The first extension is reflected as HAVING BONES IS HAVING LARGE SIZE, with further, less frequent extensions to being fat or fully developed: In Melanesia, exhibitions of physical strength are key to political power and authority; so-called "big men" do not govern by inheritance or command, but influence the group through individual example (Sahlins 1963). Accordingly, extensions to a person's ability (or their speech's ability) to command authority or have controlling political power by reason of character (i.e., HAVING BONES IS HAVING POLITICAL POWER) are also in evidence in the data (16). The third cluster, located on the right-hand side of the semantic map, involves extensions of the association of 'bone' with physical strength to include physical exertion and intense activity (broadly, HAVING BONES IS EXERTION). Senses in this cluster are frequently expressed with verbal and adverbial uses of 'bone' lexemes, such as those illustrated in (18). ' able to carry out heavy work' In short, the lexico-semantic association of BONES ARE STRENGTH, in addition to taking a wide-range of forms, has a wide-range of semantic extensions in the languages of Melanesia. These connections become clear when contextualized attestations of 'bone' lexemes are gathered together. Without these collected together in a large database, it would not be possible to observe and show the recurrent semantic extensions of 'bone' lexemes across Melanesia.

Areality in Melanesia
If we map the surveyed Melanesian languages for the colexification of 'bone' and 'strong', the result is an extremely scattered distribution (Map 1). Strict bone/strong colexification (see Section 4.1) is found in only a handful of the languages surveyed. Loose bone/strong colexification, where 'strong' is derived from 'bone' by a morphological process occurs in a further 12 languages. In short, only taking into account these narrowly defined colexifications would mean that the areality of this broader areal feature in the form of a linguistic association is not captured. By contrast, the lexico-semantic association BONES ARE STRENGTH is found in 75 of the 180 languages sampled across Linguistic Melanesia (Map 2). The core area for the lexico-semantic association is in northeast New Guinea, with numerous outlying groups to the east and west. In the core area, the lexico-semantic association is dispersed across unrelated language families, including Sepik, Ndu, Ulampo, Torricelli, Border, Sentanic, Nimboran and numerous members of distinct Trans-New Guinea subgroups. Outside this core, there appears to be a loose cluster in the Bird's Head and Bird's Neck regions, Sulawesi and around Timor in the far west, and a group of local centres in the Bird's Tail and Island Melanesia, where both Austronesian languages and Papuan languages of diverse families display the association. The skewed nature of the distribution, combined with the occurrence of the association over numerous language families, indicates areal pressures at play. Similar areal distributions have been found for other linguistic features in Melanesia, most notably the prenasalization contrasts described by Donohue and Whiting (2011). 17 Discussing Manambu, Aikhenvald (2008) suggests that the association between bone and strength has arisen in that language under influence of Tok Pisin, the lingua franca of Papua New Guinea. Tok Pisin has the linguistic association of bone and strength, as exemplified by the expressions in (21) 17 The reasons for this distribution are not understood. It may be an artefact of the level of documentation for western New Guinea. Outside of the Bird's Head at the western extreme of New Guinea, there is very little modern documentation of the non-Trans New Guinea languages of western New Guinea. At the same time, there is a relatively limited picture of the linguistic prehistory of New Guinea in general, but the western half of New Guinea is particularly unknown. The large areas covered by quite closely related languages of the Trans-New Guinea family in western New Guinea is suggestive that region is a relatively recent spread zone.
appears to be regional variation in the Tok Pisin expressions using 'bone' in association with strength. For instance, bun i dai 'bone die' and bun pinis 'bone finish' for expressing loss of physical strength appear to be used in Tok Pisin varieties on the north coast of New Guinea, where similar expressions exist in local languages. This is suggestive of local languages influencing the forms of the expressions in Tok Pisin. This view is reinforced by the diversity of constructions and expressions across languages in Papua New Guinea. Languages that are known to be in contact, like Manambu and Kwoma languages (Bowden 1997: 337-339, Aikhenvald 2008, bear little similarity to one another in the way they express the association. Both languages use the lexeme meaning 'bone' to denote the property concept 'strong, hard': the Kwoma form hapa-ga is derived morphologically by means of the similative suffix -ga 'like' (22a); Manambu, however, does not use its similative suffix, but instead has a phrase consisting of ap 'bone' plus the verb tə 'have' to mean 'be strong' (22b). (26) Kalam a. tŋi aybone put/form 'be strong, energetic', 'be fierce, aggressive' tŋi kls bone hard/strong 'physically strong and active, energetic' tŋi sayn bone soft/weak 'weak, frail, lethargic' tŋi tmey bone bad 'very strong, very fast (of a person)' Kobon b. le lbone put 'be strong, exert oneself' le l-agbone put-NEG 'be weak' le mɨd-agbone be-NEG 'be weak' Bilingualism with Tok Pisin is very recent in much of PNG, and although its influence has been in many cases radical (Kulick 1992), we would expect greater similarity in the constructions and expressions relating bone and strength, both between Tok Pisin and local languages and between individual local languages, if Tok Pisin was the ultimate source for the association. The lack of similarity in the language-specific expressions points to substrate influence from local Melanesian languages as having given Tok Pisin its association of bone and strength and not the other way around. How did this association come to be spread over such a wide area? Melanesia has been characterized as an area in which there is widespread convergence in structural features of language but divergence in lexicon (see different descriptions of this setting in, e.g., Foley 1986: 283;Ross 2001;Thurston 1987Thurston , 1989. Laycock (1982), for example, writes that because many adjacent languages in Melanesia have very similar morpho-syntactic structures, learning another language is largely a matter of learning new lexicon-"that is, the two languages involved are very close to being the same language with a different set of labels". This situation is thought to have arisen out of documented patterns of language contact whereby long-term, stable multilingualism between speech groups leads to similar structural features being used in both languages, while the emblematic function of words, as the most salient markers of identity for speech groups, acts as a brake on adoption of foreign word-forms. In addition to a simple prohibition on borrowing lexemes, there appears to also have been processes in Melanesia whereby speech communities in frequent contact with one another deliberately enact changes to their languages, such as spontaneous innovation of distinctive word-forms or complex language structures to mark their identity as distinct from their neighbours (Laycock 1982;Schapper 2020). The shared lexico-semantic association discussed here is parallel to this structural convergence-lexical divergence model of the Melanesian Language Area: the association of bone with strength represents the underlying "structural" template that has undergone convergence, while the "lexical" forms by which the association is expressed are highly divergent, with each group having language-specific means for capturing the association.
A final question to be considered is: how significant is the areal association of bone with strength in Melanesia in comparison to other areas of the world? We saw at the end of Section 3 that colexification of 'bone' with 'strong' or with the related meaning 'hard' was, whilst rare, by no means exclusive to Melanesia. Sporadically in European languages, we find bone-hard/strong collocations like those discussed for Melanesian languages in Section 4.4., for instance, German knochenhart bone-hard 'very hard, very tough' and Russian krepk-oj kost-i hard-F.GEN.SG bone-GEN.SG 'physically strong, sturdy' (cf. English rock-hard 'very hard' where the association is with rocks not bones). In many Semitic languages, 'bone' shares a single triconsonantal root with a range of meanings related to strength, a relationship that likely goes back to Proto-Semitic (Kogan and Militarev 2003: 293): Amharic aṭənt 'bone' and aṭəntam 'bony, robust, strong-boned'; Tigre ʕáṣəm 'bone' and ʕəṣṣum 'strong'; Arabic ʕađˤm 'bone' and ʕađˤi:m 'be great, powerful, important, awful'; Hebrew ʕecem 'bone', ʕocma 'power', ʕacum 'huge', and heʕecim 'empower' (among others). It is reasonable to assume that the association of Hittite ḫaštāi 'bone' with 'strength' (Kloekhorst 2008: 325) is the result of contact with Semitic. Nonetheless, I am yet to find a similar concentration of linguistic expressions of the association in such a large number of language families over such a wide geographic region as in Linguistic Melanesia.

Conclusion
In this paper I have argued that there is a widespread association between bone and strength in the Melanesian Linguistic Area. The striking areality of the BONES ARE STRENGTH association is readily missed in automatized approaches to lexical semantic typology and crude applications of the colexification methodology. Because in most languages there are not straightforward instances of colexification at the level of simplex lexemes, the complex expressions and idioms observed are unlikely to show up in word lists. There has been a recent surge in studies that make use of lexical databases, often not built for the purpose, in order to detect lexical patterns of interest to areal linguistics. As interesting as those studies are in their findings, and as useful as those databases are proving, this paper has illustrated the limitations of simplistic applications of the colexification approach and highlighted the necessity of careful investigation of typological patterns in the lexicon that go beyond the simple lexeme as a unit of analysis and that seek out contextualizing attestations to show recurrent semantic relationships in the data.
The colexification approach was meant to sidestep some of the most persistent complications for comparison of lexical semantics that had been raised and to provide a functionally grounded approach to comparison of meaning. In this paper, I have argued that by just looking at lexemes for 'bone' in isolation, we cannot observe pertinent morphosyntactic differences, let alone capture the shared similarities present in the Melanesian language data. It was shown that languages are scattered along a lexical-to-clausal cline in their expression of the lexico-semantic association of bone and strength, not to mention the large number of languagespecific idioms based on the association. Future work in developing the colexification approach, particularly its automatized implementations, will need to work to incorporate multi-morphemic and poly-lexical data in order to enhance the robustness of its results. Including figurative language in all its complexity, as witnessed by the BONES ARE STRENGTH association discussed here, will be particularly essential, since such complex linguistic structures are expected to accumulate more culturally significant information than simple structures. And it is the crosslinguistic study of these aspects of the lexicon encapsulating cultural variation that are most valuable for the discovery of areal skewings in linguistic patterns.