Reconstructing phylogeny from linkage diffusion

Reconstructing phylogeny from linkage diffusion Evidence for cladistic hinge variation Jamin Pelkey

Ryerson University Linkage models of language diversification (Ross 1988, François 2014) represent the slow differentiation of closely related sister languages via dialect continua. Such historical relationships are said to prevent the reconstruction of branch‑internal phylogeny. A newly defined mode of linkage variation challenges this restriction. In cladistic hinge diversification, speakers of a geographically central variety mediate innovations between isolated extremes of a sub-branch, while all three daughter branches maintain evidence of their own exclusive innovations. The resulting pattern blends linkage relations with family relations. Following a contextual review, the paper presents supporting evidence for the distinction from the Phowa languages of southwest China (Ngwi < Burmic < Tibeto-Burman). Data analysis includes sociohistory, dialectometry and genetic linguistic components. The argument affirms both wave and tree models of language change, enabling an enriched understanding of focal, relic and transition areas and their influence on the leveling, development and diffusion of linguistic innovations. Keywords:  subgrouping; dialect continuum; wave model; tree model; diversification; Tibeto-Burman; Ngwi; Phula

1.  Introduction1 Faced with situations of intensive language contact, many linguists argue that the diffusion of innovations tends to scramble the phylogenetic signal beyond

1.  Fieldwork arrangements enabling the collection of data represented in this work were facilitated by the Language Commission of Yunnan Province, China, the Honghe Prefecture Ethnic Research Institute and the Wenshan Prefecture Zhuang Studies Council. Partial funding for data collection and research were provided by a La Trobe University Faculty

Diachronica 32:3 (2015), –. doi 10.1075/dia.32.3.04pel issn 0176–4225 / e-issn 1569–9714 © John Benjamins Publishing Company

 Jamin Pelkey

r­ecovery. Some find this principle to be true of sustained macro-level contact within and between large language families (e.g., Aikhenvald & Dixon 2001, Dixon 2002, Holden & Gray 2006, Labov 2007); others find that the principle holds true at micro-levels, within established sub-branches comprised of closely related sis‑ ter languages (see, e.g., Ross 1988, 1997, François 2011, 2014, Kalyan & François forthcoming).2 Still others beg to differ, arguing that in many intensive contact sit‑ uations it is indeed possible to tease apart diffusion and phylogeny, not only at the areal level but also among closely related languages (e.g., Greenhill 2009, Bowern et al. 2011, Bowern & Atkinson 2012, Bowern 2013). The present study describes a specific case of language diversification in which both sides of this debate are found to be at least partially correct. More importantly, findings described below may serve to move the dialogue into fresh territory. The paper develops a preliminary analysis initiated in Pelkey (2008, 2011a) to introduce a previously undefined pattern of language diversification in which speakers of a geographically central variety mediate innovations between two iso‑ lated extremes of a sub-branch historically, even though speakers from each of the three daughter branches maintain clear evidence of their own exclusive innova‑ tions. This is a pattern that emerges in the process of reconstructing historical relationships between a series of six sister languages spoken in the Sino-Vietnam borderlands. These languages belong to a newly defined sub-branch of South‑ eastern Ngwi (Ngwi < Burmic < Tibeto-Burman) identified in Pelkey (2011a) as the Phowa clade of Highland Phula. Following contextual orientation to theory, methodology, geography and typology, a comprehensive overview of the data in

Research Grant (Humanities and Social Sciences) along with Australia International and La Trobe University Postgraduate Research Scholarships. Parts of this paper were originally presented in draft form to the 46th International Conference on Sino-Tibetan Languages and Linguistics, Dartmouth College, Hanover, New Hampshire, August 7–10, 2013. My appreciation goes to several conference participants who provided helpful advice on the paper, David Bradley and James Stanford in particular. Finally, I thank the Diachronica editorial team and three anonymous reviewers for their thoughtful and extensive comments on earlier versions of this paper. 2.  Although the present study is primarily concerned with a specific instance of the latter problem, findings are likely to have broader implications for the former as well. Bowern (2013) has argued that, even though there are legitimate reasons for distinguishing between dialect contact within the same language and contact between speakers of distinct languages, contact between unrelated (or more distantly related) languages is ultimately no different than contact between closely related languages. What truly matters, in her estimate, is the nature and degree of sociohistorical motivation for sustaining contact; such factors can be highly variable regardless of the degree of genetic-linguistic relation.



Reconstructing phylogeny from linkage diffusion 

question is provided. This integrates insights from sociohistory (sociolinguistics, ethnology and ethnohistory), dialectometry (lexical distance networks using a reticulating algorithm) and genetic linguistics (comparative method and internal reconstruction), drawing on original fieldwork data. The pattern that emerges is defined as ‘cladistic hinge variation’. Findings have implications for our understanding of the nature of subgrouping and models of language diversification. Results of the analysis further demonstrate the importance of incorporating sociohistorical and geolinguistic evidence into comparative reconstructions and further affirm both wave and tree diagrams to be necessary for representing diachronic language relationships. More specifically, building on Ross (1988, 1997), the diversification pattern in question is argued to be a new sub-type of diachronic linkage blending tree-like family relations with wave-like chains and networks during language diversification. From the perspec‑ tive of historical dialectology, these findings enable an enriched understanding of focal, relic and transition areas and their influence on the transmission (and/or diffusion) of linguistic innovations. 2.  Theoretical and methodological context Before turning to the Ngwi branch relationships in question, it will be helpful to consider a number of theoretical and methodological issues that inform this study. The current section provides further background on linkage models of language diversification and offers a critical overview of wave vs. tree and diffusion vs. inheritance dichotomies in historical linguistics, past and present, with a special focus on the nature of subgrouping. Wave models of language change were originally developed in the late 19th century (esp. Schmidt 1872) as an alternative to family tree diagrams in order to represent data that were not useful for reconstructing historical ancestry. This prominently included contact-based variation, such as loan words and areal fea‑ tures. As a result, wave models came to represent diffusion phenomena, irrespec‑ tive of genetic relationship, and tree models continued to be used to represent direct (and nested) descent based on exclusively shared innovations. As Leonard Bloomfield went on to clarify (1933: 311–320), the wave/tree dichotomy actually rests on older assumptions of “uniform parent languages and their sudden and clear-cut splitting” (1933: 311), assumptions that were themselves determined by narrow applications of the comparative method. In reality tree-like patterns and wave-like patterns are both needed to model language change; but, with some important exceptions (esp. Thomason & Kaufman 1988), 20th-century approaches to language variation and change reinforced assumptions of a basic i­ ncompatibility

 Jamin Pelkey

between waves and trees, diffusion and phylogeny, contact and inheritance. Such dichotomies continue to be influential. Holden & Gray, for example, argue that diffusion in East Africa has made Bantu language inheritance “intractable to phylogenetic analysis” (2006: 30). Labov (2007) proposes that diffusion and inheritance involve two distinct pro‑ cesses of learning for individual speakers. Aikhenvald & Dixon (2001,3 following Dixon 1997) work from the assumption that diffusion is primarily a feature of sociohistorical equilibrium while the unambiguous branching of inherited inno‑ vations indicative of a family tree results only when this equilibrium is punctuated. Against this backdrop, Malcolm Ross’s (1988) discovery of ‘linkage’ rela‑ tionships in Oceanic languages would seem to provide a kind of synthesis of the standard dichotomy. Ross demonstrates that in many cases inheritance actually proceeds by diffusion, as genetically related language varieties slowly differenti‑ ate through dialect continua. But linkages, it is claimed, make the reconstruction of clade-internal relationships impossible. Linkage diversification calls for wave models instead of tree models and casts doubt on attempts to reconstruct subbranch relations (see also Ross 1997, François 2011, 2014, Kalyan & François forthcoming). Ross (1988: 7–10) argues that the dialect chains and dialect net‑ works typifying linkage diversification function as modes of ‘differentiation’ (the overlapping wave-like model), a process that necessarily precludes diversification by separation (the branching family tree model) since, for instance, “there is no way of knowing whether an innovation shared by all member languages of the linkage was present in the proto-language or has arisen since differentiation and subsequently spread through the linkage” (Ross 1988: 8). Naturally, few historical linguists would deny that tree-like patterns are still useful for describing abrupt splits that occur in migration scenarios, after which clade-internal contact is lost; but wave and tree patterns continue to be treated as either-or phenomena. Linkage processes that blend overlapping wave dynam‑ ics with linguistic phylogeny are not currently admitted as a potential mode of language diversification or subgrouping. This assumption is called into question by the model introduced in the remainder of this paper. The broader and more immediate concern in this section is to draw attention to the puzzling nature of the strict dichotomy itself. Many contemporary linguists favor wave and network representations over family tree approaches to historical relationships (e.g., Bossong 2009, Heggarty et al. 2010, Gray et al. 2010), and some even argue that the comparative method

.  Although this is the position of the volume editors, some contributors to the collection present evidence to the contrary.



Reconstructing phylogeny from linkage diffusion 

itself should be dissociated from models featuring splits and branches altogether (Kalyan & François forthcoming, François 2014). Such arguments are not new, of course, as can be noted in sources that span the century from Schmidt (1872) to Bailey (1973), but to force the question: should the tree model be discontinued as Kalyan & François (forthcoming) explicitly propose? Do overlapping isoglosses (whether inherited or borrowed) always curtail the definition of family relations? The core issues at stake in answering these questions are the nature and status of exclusively shared innovations, the relative chronology of these innovations and the comparativist’s ability to identify such phenomena at all. Contemporary arguments against the tree model focus on the importance of convergence in language history (Bossong 2009), the usefulness of explicitly mod‑ eling ambiguous variation in family-internal relationships (Heggarty et al. 2010) and the limitations of the tree model for representing ambiguous relations (Gray et al. 2010). If the trend continues, lower-level subgrouping within established families and their higher-order branches (i.e., the reconstruction of clade-internal splits and nested levels of exclusively shared innovations) may increasingly come to seem like a questionable pursuit. Current trends coincide with the proliferation of new quantitative methods for exploring language relationships (see outlines in Croft 2008, Levinson & Gray 2012, Barbançon et al. 2013), network analyses in particular. Since many of the basic principles of comparative reconstruction tend to be absent from statistical comparisons,4 the results of such analyses must either be validated with reference to the results of traditional comparative reconstruction or admitted as phenogram (typological) representations of language relations (and, thus, explored only ten‑ tatively for cladistic hypotheses). These points are illustrated in Gray et al. (2010), Greenhill et al. (2010), Pelkey (2011a: 278–285) and elsewhere. Quantitative approaches to historical relationships tend to work from the (tacit) assumption that all innovations or isoglosses are of equal relevance for sub‑ grouping. This, of course, is not the case, as general work on phylogenetic tree construction (e.g., Sober 1991) affirms. Cladistic parsimony determines not only that shared innovations (or ‘derived characters’) are more useful for subgrouping than shared retentions (‘ancestral characters’) but also that “rare characters are better evidence of relatedness than common ones” (Sober 1991: 213, emphasis in the original), due to the unreliable nature of homoplasy (parallel innovations and/or shared innovations due to convergence).

.  One exception is Yang’s (2012) innovative application of Levenshtein distance to Lalo subgrouping. Further exceptions (i.e., Kalyan & François forthcoming, François 2014) are discussed below.

 Jamin Pelkey

Even if the linguistic comparativist stops searching for genetic evidence at the level of cognate sound correspondences, not all regular patterns will be equally diagnostic. Since many regular sound changes can be attributed to ordinary pho‑ nological processes, and since these in turn can be attributed to parallel (vs. exclu‑ sively inherited) change, such innovations provide weak evidence for subgrouping. Unusual sound changes, irregular correspondences and whole paradigms provide stronger evidence (Meillet 1925 [1967], Greenberg 1957 [2005], Bailey 1974, Mati‑ soff 1978, 1983, Hamp 1998, Nichols 1996). In short, not all shared innovations are equally diagnostic, a principle that is often ignored in quantitative approaches to linguistic phylogeny. Some historical linguists have attempted to apply strict screening measures based on the comparative method prior to submitting data to an algorithm. Lead‑ ing the way in this regard are Kalyan & François (forthcoming; also François 2014) who name their method ‘historical glottometry’, a network algorithm rooted in dialectometry and glottometry and modified for analyzing linkage relations between languages already known to belong to a common clade. Historical glot‑ tometry goes beyond regular sound changes to consider a range of shared innova‑ tions, including morphological changes, syntactic changes, lexical replacements and irregular phonological changes (Kalyan & François forthcoming: 18). Ulti‑ mately, though, the algorithm is programmed to treat each innovation as equally diagnostic, such that a single lexical replacement holds as much weight as a sys‑ tematic morphological change. This feature of the algorithm persists in spite of the authors’ own admission that atypical sound changes and morphological innova‑ tions are “the most diagnostic of historical relations” (Kalyan & François forth‑ coming: 18). To their credit, perhaps this state of affairs is inevitable. The alternative runs the risk of circularity since the introduction of weights in a quantitative analy‑ sis too easily becomes an outlet for comparativists to justify their own intuitions. Due to such problems, constructing an algorithm capable of calculating relative degrees of diagnostic quality (for purposes of subgrouping) has thus far proven to be elusive.5 Some have even argued that the required statistical reduction would simply be impossible (Kessler 2001: 7, 32–33, 100). These facts do not, however, render qualitative evidence irrelevant. On the contrary, qualitative evidence con‑

.  As a reviewer points out, it might at least be possible to avoid artificial weightings by applying Markov Chain Monte Carlo (MCMC) methods to equally weighted innovations, thus allowing the algorithm to apply higher likelihood scores to regular changes than to irregular changes in the process of tree construction.



Reconstructing phylogeny from linkage diffusion 

tinues to be the most diagnostic of shared ancestry, at least in cases where the data are sufficiently rich and readily available for comparative purposes. As will be illustrated below (esp. §4.3), novel lexical innovations provide valid supporting evidence for subgrouping but are less diagnostic than paradig‑ matic innovations. Paradigmatic changes often have greater organizational scope, potentially extending across entire lexical classes; furthermore, they often feature higher usage frequency or greater usage productivity. As a result, unusual innova‑ tions within a paradigm are less susceptible to borrowing and less likely to arise independently. This is just as true of marked innovations that emerge in morpho‑ logical paradigms as it is of idiosyncratic innovations that occur in cognate tone systems (Nichols 1996: 64). Individual lexical innovations, by contrast, are more autonomous, less fully integrated into the overall system and more easily adopted by speakers of a closely related sister variety (or others). In short, the two kinds of evidence are qualitatively distinct.6 Novel lexicalizations can function as sup‑ porting evidence for innovations that are more diagnostic, such as morphological changes; but the latter provide strong evidence on their own terms. Applications of genetic linguistics in the current study affirm along with Nich‑ ols (1996) and LaPolla (2012, 2013) that valid subgrouping relies on the identifica‑ tion of shared innovations that are ‘individual-identifying’ (vs. type-identifying) in nature: typologically unattested (or unusual) innovations that also function in larger patterned sets or paradigms. Such innovation sets are preferred first of all because they are more diagnostic than correspondences that might arise through chance, normal internal change, areal contact and/or isolated loans. Just as impor‑ tantly, they are preferred because of the diagnostic leverage they provide for recon‑ structing relative chronology, duration of affiliation and ambiguous historical innovations. 3.  Typological and geographical context Data from the subgroup under consideration in this paper are based on original fieldwork undertaken during research projects spanning from 2003 to 2008. Data site selection, field methods and parameters for analysis are outlined in Pelkey (2011a: 1–95). This section introduces the general context of the language groups in question.

.  In much the same way that a faint whiff is qualitatively distinct from a strong aroma. In spite of the fact that both are ‘marked’ scents, the former is qualitatively weaker than the latter.

 Jamin Pelkey

The Burmic branch of Tibeto-Burman includes two principal sub-groups, Burmish and Ngwi (Bradley 1979, 2012). Ngwi speakers live primarily in south‑ ern Sichuan and throughout Yunnan Province of southwest China. In the his‑ torical past, Ngwi languages featured highly isolative typology, but contemporary Ngwi varieties are increasingly concatenative. The varieties maintain verb-final syntax, multiple classifier systems and have little inflectional morphology. Syllable structure is complex in terms of onsets, but most Ngwi languages have lost rhyme codas. The languages are tonal, and many also feature phonation contrasts, often as a reflex of *oral *stop *codas. In addition to other shared innovations, Ngwi languages differ from their Burmish counterparts in featuring a two-way tone split conditioned by initial voicing on stop final rhyme classes (Matisoff 1972). This engendered a suprasegmental system with five tone contrasts in the proto-­ language. As a result, mapping subsequent splits and mergers within cognate Ngwi tone systems tends to provide a rich source of subgrouping criteria, as Nichols (1996: 64) affirms for tone languages in general. Three sub-branches of Ngwi have traditionally been defined – Northern, Southern and Central. Due in large part to the emergence of fresh data from recent fieldwork, Southeastern Ngwi (first hypothesized in Shafer 1974 and later revived in Bradley 2002) is now defined as a fourth sub-branch (Pelkey 2008, 2011a), inclusive of the Nisu varieties, Sani, Axi, Azhe, Azha and the 24 Phula languages. All languages in this sub-branch show reflexes of a typologically rare series of lat‑ eral cluster onsets with four manners of articulation *tɬ, *tɬʰ, *dɮ and *ndɮ, each resulting from the merger of two proto-onset classes (Pelkey 2011a: 352–393), among other innovations. The Phowa clade of Highland Phula is one of four known macro-clades that descend from the Southeastern Ngwi ancestor language. Although a slow diver‑ gence appears to have been long underway, speakers of historical Phowa variet‑ ies are said to have undergone an abrupt split with their Highland Phula sister varieties (the Muji meso-clade) in the 16th century for reasons discussed in §4.1 below. The six known contemporary (grand)daughter languages of the Phowa meso-clade are Ani, Labo, Hlepho, Phukha, Khlula and Zokhuo. The Phowa clade now consists of some 107,000 speakers, inhabiting approximately 500 villages in seven counties of southeastern Yunnan Province, China, and two provinces of northwestern Vietnam. The contemporary distribution of Phowa clade varieties is illustrated in Map 1. Data from 12 Phowa clade dialects are considered in this paper. For ease of reference, each variety is assigned a three-letter abbreviation based on the name of the village in which the variety is spoken. These varieties are listed in Map 1 and further contextualized in Table 1, along with language of affiliation, administrative location and village name for each.



Reconstructing phylogeny from linkage diffusion 

12

5

KAIYUAN

YANSHAN

11

1

2

9

XICHOU

WENSHAN 10

MENGZI

3

4

7

MALIPO

MAGUAN PINGBIAN

Labo Phowa

1

CKB

7

LZC

Hlepho Phowa

2

DHN

8

MXC

Labo-Hlepho*

3

DXZ

9

MZC

Ani Phowa

4

FZK

10 SZW

Phukha

5

JJC

11

WBZ

6

LPC

12

WDP

Khlula Zokhuo

8

I CH

NA VI

*Transition Zone

E

A TN

M N

6

Sichuan 0

Ho Guizhou Yunnan

ng

he

20km

Riv er

Guangxi

CHINA VIETNAM LAOS

Map 1.  Current distribution of Phowa-clade languages marking locations of varieties analyzed

 Jamin Pelkey

Table 1.  Abbreviation key to affiliation and location of Phowa-clade varieties analyzed789 Variety Language

County

Township

Village

CKB

Hlepho

Wénshān 文山(县) Déhòu

德厚(镇)

DHN

Ani

Méngzì

西北勒(乡) Dàhēinéng 大黑能

DXZ

Zokhuo

Wénshān 文山(县)

FZK

Hlepho

Píngbiān 屏边(县) Xīnhuá

新华(乡)

Fēizūkè

菲租克

JJC

Labo

Kāiyuán 开远(市) Bēigé

碑格(乡)

Jiàjí

架吉

蒙自(市) Xīběilè Zhuīlìgāi7

Chěkǎbái

扯卡白

追栗街(镇) Dàxīngzhài 大兴寨

(Vietnam)8

LPC

Phukha

NA

LZC

Khlula

Wénshān 文山(县) Liǔjǐng

柳井(乡)

Lǎozhài

老寨

MXC

Khlula

Mǎguān 马关(县) Mùchǎng

木厂(镇)

Mǎxī

马西

MZC

Hlepho

Méngzì

鸣鹫(镇)

Méizichōng 梅子冲

SZW

Hlepho

Wénshān 文山(县) Bàxīn

坝心(乡)

Suǒzhīwān 所支弯

WBZ

Hlepho

Kāiyuán 开远(市) Bēigé

碑格(乡)

Wěibazhū

WDP

Labo

Kāiyuán 开远(市) Mǎzhěshào 马者哨(乡) Wūdūpí

蒙自(市) Míngjiù

尾巴猪9

乌都皮

Hlepho varieties comprise five of the 12 communalects under consider‑ ation below. This has as much to do with the relative size of the Hlepho speaker population as with the complex nature of Hlepho dialect geography. One-third (36,000) of the Phowa-clade population can be counted as first language speak‑ ers of Hlepho dialects (Pelkey 2011a: 415); furthermore, Hlepho dialect geogra‑ phy features a complex range of discrete isogloss sets, most of which are shared with Phukha. Of particular interest to the study at hand, roughly half of these distinctive innovation sets are also shared exclusively with Ani and Labo while the rest are shared exclusively with Khlula and Zokhuo. Facts such as these, coupled with the sociohistorical context from which they emerge, suggest a pat‑ tern of linguistic diversification that has until now been left undefined in the literature.

.  /gāi/ is the Southwest Mandarin pronunciation of 街, rendered /jiē/ in standard M ­ andarin. .  This language is defined in Edmondson (2003) and Fried (2000), drawing on data published online in Edmondson (2002), collected in Lào Cai Province, Bảo Yên Township, Vietnam. .  Full name: 尾巴猪小寨 Wěibazhū Xiǎozhài, often clipped to 小寨 Xiǎozhài in the local Chinese vernacular.



Reconstructing phylogeny from linkage diffusion 

4.  A distinctive pattern of language diversification in the Phowa clade The distinctive pattern in question appears to have resulted from a blend of cladeinternal family diversification (Ross’s 1988 ‘diversification by separation’) and clade-internal linkage diversification (Ross’s 1988 ‘diversification by differentia‑ tion’), requiring the application of both wave and tree models simultaneously for adequate reconstruction and subgrouping. The explanatory necessity of defining the pattern in this way becomes clear after examining clues from Phowa sociohis‑ tory, dialectometry and genetic linguistics. 4.1  S ociohistorical and sociocognitive evidence from Hlepho dialect geography The sociohistorical evidence for affirming a radical blend of phylogeny and dif‑ fusion in the diversification of the Phowa clade comes from recorded migration patterns, dialect intelligibility data, language attitudes, onomastic identities and reported marriage networks. Hlepho is the most internally diverse language in the Phowa clade and may be described as a dialect continuum in itself. Furthermore, the contemporary Hlepho speaking population shades into marginal varieties of Labo Phowa spoken to the northwest of the Hlepho distribution in ways that defy the definition of abrupt boundaries (Pelkey 2011a: 110–114, 158, 417, also below). Other indications suggest that this was once the case between certain Phukha- and Khlula-speaking populations living in the southeastern quadrant of the Hlepho distribution (see Map 1), but sister languages to the southeast have all but lost contact with each other by present day. Reported migration patterns in Chinese historical sources point to two primary population centers for languages now classified in the Phowa clade. The northwest‑ ern population center is said to have been established late in the 16th century deep in the mountains on both sides of the present-day Kaiyuan-Mengzi County border after valley dwelling populations reportedly fled following a failed revolt against a local tyrant (HHYC 2002: 50). The southeastern population center is said to have been established in Wenshan County by overland passage from the northwest, beginning a few centuries earlier (see composite discussion in Pelkey 2011a: 5–18). Onomastic identity patterns provide further useful context, particularly in the northwest corner of the present-day Phowa distribution. First it should be noted that all Ani, Labo and Hlepho speakers still affiliate under the endoethnonym10 Phowa (meaning, roughly, “the people”), but Phukha, Khlula and Zokhuo speakers .  Designating a macro-ethnonym or umbrella term that is internally generated and maintained by a majority of the population (vs. being externally applied by outsiders) over and

 Jamin Pelkey

no longer share this macro-identity. A further index of the historical linkage main‑ tained between Labo and Hlepho in particular is the use of autonyms and exonyms that function relative to the elevation of speaker and referent villages located in the Labo-Hlepho transition zone (as delineated in Map 1). In this particular region, Hlepho speakers tend to live toward lower, southeastern elevations while Labo speakers tend to live toward higher, northwestern elevations. The situation is nec‑ essarily gradient, however, since Phowa speakers living at higher elevations in this continuum are referred to as Labopho (“highlanders”) by those at lower elevations, while those at lower elevations are referred to as Digaopho (“lowlanders”). Thus, Phowa speakers living in villages at intermediate elevations may be called either “the lowlanders” or “the highlanders” depending on the location of the village in question relative to the situation of the speakers’ village (Pelkey 2011a: 110–114). Dialect intelligibility metrics across the Phowa clade correspond with major identity distinctions such that low-to-negligible intelligibility is shared between all major identity groups with the marginal exception of varieties spoken in the Ani-Labo transition zone and the strong exception of the gradient transition zone between Labo and Hlepho (Pelkey 2011a: 124–126, 132–133, 144–149). Notably, though, in the southeastern distribution of the Phowa clade, some Khlula speakers perceive Zokhuo to be a variety of Khlula while some Zokhuo speakers consider Khlula to be a variety of Zokhuo (Pelkey 2011a: 118). The two language groups are no longer in frequent contact, however, and share low-to-marginal intelligibility at best, an estimate that is based on both perceptual dialectology interviews and statistical metrics (Pelkey 2011a: 126–127, 133, 150–153). Though many sources of sociocultural evidence corroborate a link between the Hlepho continuum and the Ani-Labo micro-clade within Phowa, the best sociocultural evidence for a Khlula-Zokhuo connection with Hlepho-Phukha comes from an otherwise unusual marriage network that is still maintained between speakers of at least one Khlula variety in southeast Wenshan County (see LZC location in Map 1) and a variety of Hlepho located a two-day’s walk distant (80–100 km by foot paths), toward the middle of the Hlepho distribution (see CKB location in Map 1). This suggests not only an earlier period of contact during which the two varieties were more familiar but also a cultural mechanism whereby the diffusion of clade-internal innovations might have spread as the dialects of Phowa diversified into distinct languages. Slightly lower percentages of lexical ­distance between these two varieties (compare CKB and LZC in Table 2 below) are also accounted for in this connection.

beyond lower-level ethnonyms and autonyms used verbally to distinguish internal diversity within the macro-group.



Reconstructing phylogeny from linkage diffusion 

4.2  Statistical evidence from dialectometry Further evidence that the diversification of the Phowa clade proceeded by both splits and linkages comes from an application of dialectometry. Dialectometry encompasses a growing collection of distance-based statistical applications that seek to discover aggregate relationships in language variation linked to physi‑ cal geography (Nerbonne 2009). Such measurements lend themselves well to sifting through ambiguous overlap between dialect areas and dialect continua (­ Heeringa & Nerbonne 2001). Table 2 summarizes the lexical distance values of 12 Phowa-clade languages relative to a 200-item Swadesh-based wordlist, used for tabulating a rough count of shared cognates in basic vocabulary.11 Table 2.  Lexical distance percentages for the Phowa clade, marking Hlepho varieties Ani DHN 0.17

Labo JJC

0.26

0.15

Labo WDP

0.15

0.16

0.23

Hlepho MZC

0.18

0.09

0.17

0.15

Hlepho WBZ

0.28

0.26

0.28

0.24

0.28

Hlepho FZK

0.24

0.19

0.23

0.17

0.18

0.22

Hlepho CKB

0.41

0.38

0.40

0.37

0.38

0.29

0.30

Hlepho SZW

0.47

0.44

0.46

0.45

0.44

0.43

0.42

0.36

Phukha LPC

0.45

0.43

0.43

0.41

0.40

0.42

0.35

0.37

0.43

Khlula LZC

0.48

0.43

0.44

0.42

0.42

0.40

0.41

0.35

0.43

0.25

0.53

0.46

0.48

0.49

0.46

0.49

0.46

0.45

0.48

0.44

Khlula MXC 0.46 Zokhuo DXZ

.  The term ‘shared cognates’ is used loosely here, in keeping with standard procedures of calculating lexical similarity. To clarify, this includes shared retentions, shared innovations and in some instances split cognates (see further discussion in Pelkey 2011a: 120). Although recent loans are excluded, historical (phonologized, lexicalized) loans and areal loans are included, as are doublets and lexical mismatches due to semantic shift. Such evidence cannot be admitted as conclusive for purposes of genetic subgrouping since shared retentions, shared innovations and historical loans are all granted equal status in standard lexicostatistic calculations.

 Jamin Pelkey 0.1

Labo Phowa WDP

Khlula-Zokhuo Clade

Khlula MXC

Ani-Labo Clade Khlula LZC

Labo Phowa JJC Labo Phowa WBZ Ani Phowa DHN

Zokhuo DXZ

Hlepho Phowa MZC Hlepho Phowa CKB Hlepho Phowa FZK

Hlepho Phowa SZW

Phukha LPC

Hlepho-Phukha Clade

Figure 1.  Lexical distance network of Phowa clade varieties marking representative lects. Internal cladistic designations based on comparative reconstruction summarized in §4.3

Aggregate relationships implied in the results listed in Table 2 are output as Figure 1 using the Neighbor-Net reticulating algorithm (Bryant & Moulton 2004) in Splits­Tree 4 (Huson 1998, Huson & Bryant 2010) to produce an unrooted phe‑ nogram network.12 Line length corresponds with the relative confidence of a given split in the diagram. Reticulated patterns represent ambiguous splits. From the point of view of dialectometry this may be held to correspond with dialect continua and/or the slow divergence of varieties in space and time (Holden & Gray 2006). The gradient nature of the Hlepho continuum is visually apparent across the bottom of the phenogram, as is the relative distance separating Khlula-Zokhuo and Ani-Labo, which occupy opposite ends of the continuum. Hlepho varieties MZC and CKB are in frequent contact with Labo varieties in particular but are adjacent to both Ani and Labo varieties spatially. As discussed in §2, above, although phylogenetic relations may seem to be implied in the diagram, evidence from the comparative method and internal reconstruction is required in order to judge the degree to which the ­representation

.  To clarify, the Table 2 results were generated manually, external to the SplitsTree software. Results were then converted to a readable format and uploaded to SplitsTree in order to generate the Figure 1 network. My appreciation goes to Ken Manson for introducing me to these methods in 2007.



Reconstructing phylogeny from linkage diffusion 

might function as a cladogram. Nevertheless, even taken on its own terms, the network representation in Figure 1 provides further evidence that Hlepho func‑ tions as a kind of diachronic linkage, uniting two extremes of the sub-branch. The ‘clade’ labels in Figure 1, however, are applied based on a further layer of analysis, summarized in the next section using principles of comparative reconstruction. 4.3  Individual-identifying evidence from comparative reconstruction Data gathered primarily from varieties marked in bold in Figure 1 are used in this section to provide an overview of the evidence for reconstructing clade internal relationships between Phowa sister languages. The results illustrate further evi‑ dence for recognizing a distinctive pattern of diversification in the Phowa clade. A detailed account of exclusive, individual-identifying innovations (see §2 above) in the Phowa clade is provided in Pelkey (2011a: 313–330). Here I summarize the relevant innovation sets and then zoom in on two expanded datasets to illustrate what appear to be the most contradictory clade-internal sub-groupings. Table 3 groups together innovation sets that emerge from comparative and internal reconstruction. Innovations that are found only partially in a given lan‑ guage are listed in parentheses. As discussed above (see §§2–3), the best evidence for subgrouping rests on idiosyncratic splits and mergers within a demonstrably cognate tone system relative to the Ngwi branch in general (and the Southeastern Ngwi sub-branch in particular). Tone notation follows the Chao (1930) number‑ ing system in which /55/ is the highest relative register pitch and /11/ the lowest. Unequal numbers represent contour pitches (e.g., /13/ ~ low-mid rising contour). Items 4 and 12 are the focus of Tables 4 and 5, respectively. Some sets of innovations constitute stronger evidence for subgrouping than others. Also, some individual innovations within each set provide stronger evi‑ dence for subgrouping than others, particularly those that are more paradigmatic in themselves (see discussion in §2, above). Innovation sets 5–8 provide sufficient evidence for subgrouping the six known Phowa languages as a meso-clade within Highland Phula. Considering the broader sociohistorical context, it is unlikely that innovation set 5 would have occurred across communalects of the entire developing clade later than the diversification of Ani-Labo and Khlula-Zokhuo into distinct proto-languages. Regardless of the relative chronology of these two lexical innovations, however, it is much more important to note that innovation sets 6 and 8 are necessarily prior to innovation set 12. In fact, tone shifts in these environments quite possibly became the conditioning factors that led to the double flip-flop of tone classes that would eventually come to typify four of the six languages in the Phowa clade. In other words, not only is there sufficient evidence for subgrouping the

 Jamin Pelkey

Table 3.  Summary of 14 Phowa clade innovation sets (adapted from Pelkey 2011a: ­325–326) Ani

Labo

Hlepho

Phukha

Zokhuo

Khlula

1.

P

2.

P

P

O

O

O

O

P

P

O

O

O

3.

O

P

P

O

O

O

4.

P

P

P

P

O

O

5.

P

P

P

nd

()

O

6.

P

P

P

P

P

O

7.

P

P

P

P

P

P

8.

()

()

P

P

P

P

9.

O

O

P

P

O

O

10.

O

O

()

P

O

P

11.

O

O

P

P

P

P

12.

O

O

P

()

P

P

13.

O

O

P

O

P

P

14.

O

O

O

O

P

P

  1. * Tone 1: unmarked /55/ reflexes with splits to /33/ and /13/ on *voiced and *prenasalized *voiced *obstruent *initials, respectively; *Tone 2: dominant /21/ reflexes but unmarked /33/; reflexes of *prenasalized *voiced *obstruent *initials merge to /13/ in *Tones 1 & 2; *Tone 3: unmarked /21/ with /55/ splits in *ʔp, *ʔb and *voiceless fricative *initial reflexes; antithetical semantic shift of “idle” > “busy”, rendered lo¹³ ; “cricket” and “grape” or cognate with Ani ve⁵⁵ɬo³³dɨ⁵⁵lɨ⁵⁵ma²¹ and ɔ̱²¹ta̱²⁴ma⁵⁵, respectively.  2. Tone 1: /33/ reflexes of *ʔəm and *Cp initials; Tone 2: unmarked /33/ with /21/ reflexes of deaspirated *p initials; “steep”, “stone” and “husband” cognate with or Ani tɬʌ²¹, nɔ²¹ma⁵⁵ and za²¹va³³, respectively. Nasal deletion on reflexes of *ŋo¹ “cry”.  3. *Tone 2: /33/ reflexes of *s-prefixed *initials; *Tone 3: reflexes of *ʔm *initials split to /21/ and /55/.  4. * Tone 2: *voiced *onset *syllables condition split to mid/low-falling reflexes; *C-prefixed *voiceless *fricative initials condition split to /55/ (cognate with Phukha /35/).   5. “Flute” and “loc.upon” cognate with Ani si⁵⁵li⁵⁵ and mo⁵⁵tu²¹ respectively.   6. Tone H: low-falling reflex on *syllables with *k-l_, *k-r_ and *s-prefixed *initials.   7. Includes most ‘Highland Phula’ macro-clade innovations, but excludes Muji meso-clade innovations.   8. *Tone 1: Reflexes of *voiceless *obstruent *initials condition split to /21/ or /31/.   9. “Sugarcane” cognate with Hlepho sa³³bɯ³³ 10. *Tone 2: /33/ reflexes of *prenasalized *voiced *obstruent *initials; “bean” cognate with Khlula nu²¹mi⁵⁵ 11. *Tone 1: reflexes > dominant /21/ but unmarked /33/ with *voiced *obstruent and *prenasalized *voiced *initial reflexes conditioning split to /21/; *Tone 2 reflexes of *pre-glottalized *nasal and *voiceless *fricative *initials split to /55/; *prenasalized *voiced *initials condition merger of *Tones 2 & 3 to /33/. 12. D  ouble ‘flip-flop’ of high and low tone classes: (1) Tones 1 & 2 (*open *syllables), plus (2) Tones *H & L (*checked *syllables) 13. Tone 1: *r-y_ initials split to a rising tone; “all” cognate with Zokhuo ji²¹ha¹³ 14. Tone 1: /55/ reflexes of *ʔəm *initials; TC-2 /33/ reflexes of *Cs and devoiced *voiced *obstruent *initials; TC-H /21/ reflexes of regular *C-prefixed *voiced *obstruent *initials (*Cb) but /55/ reflexes of devoiced *Cb *initials; “nighttime” compound cognate with Zokhuo ʨʰi⁵⁵bə³³



Reconstructing phylogeny from linkage diffusion 

six Phowa languages as an exclusive meso-clade within Southeastern Ngwi, there is also sufficient evidence for defining an early split into two distinct micro-clade proto-languages at an intermediate stage historically, based on relative chronol‑ ogy of innovations. Even taken as a unit, these four innovation sets (5–8) are not as robust in terms of paradigm-quality organization (multiple paradigmaticity) as are Sets 1 (subgrouping Ani-Labo) and 14 (subgrouping Khlula-Zokhuo). Notably, then, the innovation sets that define the two extremes of the Phowa continuum are strongly indicative of exclusive shared ancestry, requiring the definition of at least two dis‑ tinct daughter clades within Phowa. Phukha, now spoken primarily in Vietnam (Edmondson 2003), is most clearly subgrouped with Hlepho due to the evidence of innovation Sets 9, 4 and 12 (see also Tables 4 and 5 below), though it appears that the variety split from Hlepho prior to the completion of innovation Set 12, as is illustrated in Table 5. This detail also provides plausible evidence for considering innovation Set 4 to have occurred prior to Set 12. Of the six languages, Hlepho is historically the “most fully changed isolect” (Bailey 1996: 143) and might be expected, therefore, to represent a historical lect of origin or ‘focal’ variety in some sense. Unlike traditional focal varieties, how‑ ever, Hlepho innovations are not unidirectional. Rather, different innovation sets are shared exclusively between Hlepho and two respective extremes of the Phowa clade: Sets 2–4 are shared with Ani-Labo to the exclusion of Khlula-Zokhuo, while Sets 10–13 are shared with Khlula-Zokhuo to the exclusion of Ani-Labo. Hlepho’s status as a focal variety may remain an open question, but it is clear that Hle‑ pho has served a historically mediating role, incorporating or initiating distinc‑ tive innovations shared with the two early-splitting micro-clades, Ani-Labo and Khlula-Zokhuo. To better illustrate the distinctive innovation pattern in question, consider the data in Tables 4 and 5.13 Table 4 presents strong evidence for subgrouping HlephoPhukha with Ani-Labo (see Table 3, Set 4, above) while Table 5 presents strong evidence for subgrouping Hlepho-Phukha with Khlula-Zokhuo (Table 3, Set 12). As Table 4 illustrates, unmarked reflexes for Tone 2 (*low-open) are maintained in spite of voiced-onset environment in Khlula and Zokhuo. The data show marked tone reflexes for the other four languages, however, signifying a historical split to a low-falling tone based on voiced-onset conditioning. The data in Table 5 illustrate an innovative tonal redistribution involving a flip-flop in pitch reflexes of Tone Classes 1 and 2. Tone 1 reflexes typically feature higher pitches than Tone 2 reflexes across the Ngwi branch. This status is p ­ reserved

.  Note: In Tables 4–5 nc indicates “no cognate” while nd indicates “no data”.

 Jamin Pelkey

in Ani-Labo but flip-flopped in the rest of the Phowa clade. In sharp contrast to the individual-identifying innovation illustrated in Table 4, which excludes KhlulaZokhuo, Hlepho (and, to a lesser extent, Phukha) participates in this innovation to the exclusion of Ani-Labo. The innovation can even be described as a double flip-flop since checked-tone classes (H and L) also feature pitch reversals in the rest of Phowa. Labo varieties actually show evidence of this latter pitch reversal, however; and in Ani the two checked tone classes have largely merged to a midlevel pitch reflex. Thus, strong evidence can be cited for subgrouping Hlepho-Phukha with both Ani-Labo and with Khlula-Zokhuo; and this is true in spite of the fact that the latter two pairs clearly constitute clade-internal sub-groups of their own. Nei‑ ther wave nor tree models are adequate to capture such relationships in isolation; rather, both seem to apply simultaneously. Table 4.  Phowa-clade tone system innovation excluding Khlula-Zokhuo: Low-falling reflexes of proto-voiced-onset syllables in Tone 2 Gloss

Proto-Ngwi Ani

Labo

Hlepho

Phukha

Khlula

Zokhuo

“thin”

*ba²

ba̱²¹

ba²¹

nd

ba¹³

bʌ⁵⁵

dɮa̱²¹ dɮa²¹ dɮa²¹

dɮa³¹

dɮa¹³

kʌ⁵⁵

“honeybee” *bya²

ba̱²¹

“son”

*ʒa²

za²¹

za̱²¹

za²¹

za³¹

za¹³

zo⁵⁵

“eat”

*dza²

ʣa̱²¹

ʣa̱²¹

ʣa²¹

ndza³¹

ʣa¹³

ʣʌ⁵⁵

“sky”

*mo²

mu²¹

mɯ²¹ m̩²¹

m̩³¹

ma¹³

m̩⁵⁵

“five”

*ŋa²

ŋa²¹

ŋa²¹

ŋa³¹

ŋa¹³

ŋʌ⁵⁵

ŋa²¹

“bone”

*ʃə-ro²

v²¹

vɯ²¹

vɯ²¹

ɣo³¹

vɯ⁵⁵

vɯ⁵⁵

“strength”

*ra²

ʁa²¹

ʁa²¹

ha²¹

nd

ʁa¹³

ʁʌ̱³⁵

“belly”

*wam²

ɔ̱²¹

ɣʌ²¹

nc

o¹³

o̥õ¹³

wu̱³⁵

“ride”

*dzi²

ʣi²¹

ʥi²¹

ʥi²¹

ʥiʡ³⁵

ʥi¹³

ʥi⁵⁵

Linkage diversification is most apparent in the development of the HlephoPhukha varieties relative to the two extremes of the Phowa clade. Whereas Phukha speakers ceased to function in a mediating role following their emigration toward Vietnam, Hlepho varieties remained in a geographically central location between Ani-Labo to the northwest and Khlula-Zokhuo to the southeast. The Hlepho vari‑ eties are not simply a dialect continuum between the two extremes, however. In addition to forming their own distinctive identity, as described above, Hlepho speakers show many distinctive innovations of their own.



Reconstructing phylogeny from linkage diffusion 

Tone Class L Tone Class H Tone Class 2 *low -checked *high -checked *low-open

Tone Class 1 *high-open

Table 5.  Phowa-clade tone system innovations excluding Ani-Labo: Double flip-flop of low-high pitch reflexes in checked and open syllable classes (shaded cells indicate pre‑ served reflexes)14 Proto-Ngwi

Ani

Labo

Hlepho

Phukha

Khlula

Zokhuo

Gloss

*C-gray¹

tʂɿ⁵⁵

ʦɿ⁵⁵

ʦɿ³³

tsɿ³³

ʦɿ³³

ʦɿ³³

“star”

*s-rwe¹

ɕə⁵⁵

sə⁵⁵

sə³³

ɕi³³

ɕi³³

ɕi³³

“yellow”

*ʔ-l(y)a¹

ɬa⁵⁵

ɬa⁵⁵

ɬa³³

ɬa³³

ɬa³³

lʌ³³

“tongue”

*po¹

pʰɯ⁵⁵

pʰɯ⁵⁵

pʰɯ³³

bu³³

pʰɯ³³

pʰɯ³³

“rooster”

*ʔ-mri²

me³³

mi³³

mi⁵⁵

miʔ³⁵

mi⁵⁵

mi⁵⁵

“tail”

*kwe²

ʨʰi³³

cçʰi³³

ʦʰɿ⁵⁵

tshɯ³⁵

ʦʰɿ⁵⁵

ʦʰɿ⁵⁵

“dog”

*tsa²

ʦʰa³³

ʦʰa³³

ʦʰa⁵⁵

tsha³⁵ [14]

ʦʰa⁵⁵

ʦʰʌ⁵⁵

“salt”

*(C)-na²

na³³

na³³

na⁵⁵

na³⁵

na⁵⁵

nʌ⁵⁵

“ear”

*C-grakH

ʥe³³

ʥɛ²¹

ʥɛ²¹

dja³¹

ʣə̝²¹

ʣa²¹

“cold”

*C-nakH

ne̱³³

nɛ²¹

nɛ²¹

nja³¹

nɑ̰̃¹³

nã̱²¹

“black”

*s-nökH

no̱³³

no̞²¹

no̞²¹

nu³¹

nu²¹

no²¹

“bean”

*(sə)-grokH

dʐo̱³³

ʣo²¹

ʣo²¹

dʑu⁵³

dʐɯ˞ ¹³

ⁿʣo²¹

“fear”

*lakL

le³³

lɛ³³

lɛ³³

lja³³

lɑ³³

la³³

“hand/arm”

*wakL

ve³³

vɛ³³

vɛ³³

via³¹

va̱³³

va̱⁵⁵

“pig”

*C-satL

χa³³

χa¹³

χa⁵⁵

xa³⁵

χɑ⁵⁵

χʌ⁵⁵

“kill”

*s-napL

na⁵⁵

na⁵⁵

na³³

na³³

nɑ³³

nʌ³³

“snot”

4.4  H  lepho-internal innovations In analyses carried out to date, most distinctive innovations have surfaced in the Hlepho variety discussed above (FZK: see Table 1, Map 1 and data analysis in §§4.1–4.3), a centrally located dialect spoken in Feizuke village of northern ­Pingbian County. Novel lexical replacements abound in the dialect, including pʰa²¹ “times”, sə³³ “wither”, ⁿdɮo⁵⁵ “flow”, and mbɿ¹³ “flowery” (a semantic shift from “beautiful”); and novel lexicalizations involving split cognates are even more

.  Rendered [tshan³¹] in the Edmondson (2002) wordlist but re-transcribed as /tsha³⁵m̩²¹/ based on accompanying audio recordings. Compare with the full Hlepho form /ʦʰa⁵⁵m̩²¹/ and the full Khlula form /ʦʰa⁵⁵m̩³³/.

 Jamin Pelkey

p ­ lentiful.15 More substantial innovations include the replacement of a paradigm set in the antonym pair deep/shallow with ʔmo³³ and ba²¹, respectively, not cog‑ nate with Proto-Ngwi *ʔ-nakL/*dim1 as would be expected (Bradley 1995: 9). Morphological innovations include the grammaticalization of χa⁵⁵ “kill” as a completive marker and unprecedented uses of the -za diminutive/nominal‑ izer at various stages of grammaticalization. The lexical source of the -za⁵⁵ suffix in Hlepho is “son”, cognate with a general -za suffix that is commonly used in Tibeto-Burman languages as a diminutive marker in lexical constructions. Fur‑ ther stages of grammaticalization are less common, however. As a result, languages that employ the morpheme as a more general suffix by extending its productivity through further stages of semantic bleaching often create idiosyncratic innova‑ tions in the process. Consider the following Hlepho (FZK) constructions with this in mind: (1) nɯ²¹ma³³za⁵⁵ “sand”, (2) lɛ³³ʦɯ³³za⁵⁵ “finger” and (3) nɛ²¹za⁵⁵ “eye‑ ball”. The first instance is a novel construction, but it is a standard Tibeto-Burman use of the -za morpheme as a diminutive since the construction nɯ²¹ma³³ is a compound meaning “rocks” (hence, “rocks-dim” = “sand”, qua: “tiny rocks”). This is a light stage of grammaticalization. The second example (glossed, “finger-dim/ nmlz”) illustrates a novel lexicalization employing -za in a slightly more bleached usage since this innovation involves a metonymic semantic shift from specific “pinkie finger” to general “finger”. The third example (“eye-nmlz”) is an instance of advanced grammaticalization since diminutive semantics are not focal or deri‑ vational in the construction. In short, the -za suffix has become more productive in Hlepho-FZK than in most other Ngwi languages. Although much work remains to more adequately document variation pat‑ terns across Hlepho dialects, it is already apparent that Hlepho varieties share exclusive internal innovations in wave-like patterns, and that FZK has served as a Hlepho-internal focal area historically. A Hlepho variety spoken in Meizichong village (MZC), of northeastern Mengzi County, for example, shares a novel lexi‑ calized reduplication for the adverbial “all” with FZK (rendered dɛ¹³dɛ²¹ in FZK and dɛ¹³dɛ¹³ in MZC). A third Hlepho variety spoken in Suozhiwan village (SZW) of central Wenshan County shows evidence of an unusual shared semantic merger between “wheat” and “barley”: zo³³ and ʐʴo³³ < *zu³ “barley” (in FZK and SZW, respectively), replacing *ʃa³ “wheat” cognates in the process (Bradley 1997). In .  Examples include kə³³sɛ¹³ “same”一样 (相同), bɛ²¹ʣi⁵⁵ʔmo³³ “arrogant” 骄傲, ⁿʣə²¹ⁿʣə²¹ “cool” 凉快, ʔnɛ²¹ʔŋə⁵⁵ “bright” 亮, ma⁵⁵ʦʰa⁵⁵ “bamboo basket” 箩筐, vɛ³³tɬɛ³³mi⁵⁵tʰa⁵⁵ “pointed knife” 尖刀 (lit., “pig stab knife”), “wheat flour” 面粉 zɛ¹³ʦə⁵⁵m̩³³, “charcoal”, za²¹ⁿdo³³va³³ “brother’s child (male viewpoint)” 侄子/侄女 (an unprecedented use of the -va suffix), sa³³ha²¹ “sister’s child (male viewpoint)” 外甥/外甥女, pɛ³³lɛ³³ “placenta” 胎盘, ɬa³³tʰə³³ “gallbladder” 苦胆, sə³³tɬʰɿ⁵⁵ “moss” 青苔, pɛ²¹nɛ²¹tɬo²¹ “grapes” 葡萄, a³³tʰo⁵⁵ʣɿ³³ka²¹ “dragonfly” 蜻蜓, a³³ɡɯ²¹lɛ²¹ɕɛ³³ma³³ “mantis” 螳螂 and more (Pelkey 2011b).



Reconstructing phylogeny from linkage diffusion 

both cases, the innovation in question is the only known evidence of such a merger in the Phowa clade (or elsewhere). Paradigm leveling provides a further source of linguistic evidence in support of Hlepho’s unique status relative to its sister languages. As demonstrated in Pelkey (2011a: 238, 313–330), Hlepho phonology is marked by general tone-leveling to a mid-level /33/ pitch in numerous environments across all five tone classes. In fact, paradigm leveling is often found in focal areas (high-contact transition zones), due to the simplifying effects of overlapping innovations and widespread diglos‑ sia. Because of this, Hlepho tone-leveling should be classified as a type-identifying change instead of an individual-identifying innovation. Nonetheless, given the broader context sketched out above, it is reasonable to say that this marked feature of Hlepho typology is a further index of its status as a unique diachronic linkage or cladistic ‘hinge’ between two more distantly related sister language subgroups.

5.  Diachronic linkage and hinge diversification in Phowa and beyond Defining a cladistic hinge scenario is necessary for understanding and integrat‑ ing many otherwise confounding facts of Phowa dialect geography. Recogniz‑ ing hinge diversification is necessary for explaining complex variation patterns shared between Hlepho speakers and speakers of their various sister languages within Phowa. Recognizing hinge diversification is also necessary for adequately subgrouping sister languages within the Phowa clade. Further discussion of the pattern and its implications are offered here to better support its admission as a general model for subgrouping in similar situations elsewhere in the world. 5.1  The explanatory necessity of a hinge model for Phowa clade subgrouping From the perspective of genetic linguistics, hinge diversification blends fam‑ ily relations and linkage relations in ways that have not been previously recog‑ nized. From the perspective of historical dialectology, cladistic hinge variation is a pattern of linguistic diversification that combines transitional dialect areas and dialect continua in ways that have not until now been clearly defined or docu‑ mented.16 As we find in the case of Phowa discussed above, three relic areas and three focal areas interact in a cladistic hinge scenario to the extent that the central region becomes a site of intensified clade-internal contact. As a result, changes are not simply unidirectional; rather, hinge varieties develop bilateral dialect

.  With this in mind, we might reasonably expect that many further patterns of diversification await identification within the traditional ‘dialect continuum’ rubric.

 Jamin Pelkey

c­ ontinua. The i­ dentification of a cladistic hinge pattern in Phowa further supports ­Bowern’s assertion that “Old dialect chains can be distinguished from extensive post-break-up language contact by the type of shared material and by the relative chronology of the changes” (2013: 428). In addition, since cladistic hinge varia‑ tion involves the diachronic transmission of individual-identifying innovations, it stands in sharp contrast to typical modes of diffusion such as areal Sprachbünde, convergence areas, dialect chains and transitional dialects (Chambers & Trudgill 1998: 110–119). Hock (1991: 441) suggests that “it seems preferable to determine the focal, relic, and transition areas for a given change on the basis of the specifically rel‑ evant evidence, rather than on the basis of preconceived notions as to what should be central or outlying/remote areas.” Letting the data do the talking is even more important when changes being considered are overlapping, individual-identifying innovations. To the degree that the data are rich, and approached from multi‑ ple perspectives, to that degree the analogies drawn and (tentative) conclusions reached are more likely to approximate linguistic reality. With this in mind, and with reference to the data examined above, I offer a summary list below of four criteria that are critical for identifying the Phowa cladistic hinge. 1. a common meso-clade, Proto-Phowa, is apparent based on shared ­individual-identifying innovations not present elsewhere in the macro-clade (Highland Phula). Evidence for this claim is summarized in Table 3, innova‑ tion sets 5–8 (Pelkey 2011a: 278–335 for further supporting evidence) and related discussion in §4, including sociohistorical, geolinguistic and genetic linguistic rationale for establishing the relative chronology of the innovations in question. 2. three micro-clades are apparent within the meso-clade: Ani-Labo, KhlulaZokhuo and Hlepho-Phukha based on evidence of their own exclusive inno‑ vations, as summarized in Table 3, sets 1, 14 and 4–12, respectively. Although Hlepho features many individual-identifying innovations (as discussed in §4.4), the Helpho-Phukha relationship is most apparent in the innovations these two languages share with Ani-Labo on the one hand and KhlulaZokhuo on the other. These internal subgroupings are supported by crucial sociohistorical and geolinguistic details discussed in §§3 and 4.1. Further sup‑ port is provided by network phenogram patterns in the dialectometry analysis in §4.2. 3. a strong split in shared innovations, adequate for subgrouping, emerges between the Hlepho-Phukha micro-clade and the two sister clades AniLabo and Khlula-Zokhuo, both of which share exclusive innovations with Helpho‑Phukha that they do not share with each other. Though there are



Reconstructing phylogeny from linkage diffusion 

many other supporting innovations, this is most vividly illustrated in a double flip-flop of low vs. high pitch reflexes in checked and open syllable classes (excluding Ani-Labo), as illustrated in Table 4, and a low-falling reflex on proto-voiced-onset syllables in Tone 2 (excluding Khlula-Zokhuo), as illus‑ trated in Table 5. These genetic linguistic splits are further validated and better interpreted by supporting evidence from Phowa sociohistory and geolinguis‑ tics summarized in §4.1. 4. speakers of the hinge variety are centrally situated. Hlepho speakers are geographically distributed between Ani-Labo speakers to the northwest and Khlula-Zokhuo speakers to the southeast. This provides a historical ‘mid‑ dle ground’ for ongoing contact and bi-directional sharing of innovations. The only way to account for overlapping innovation patterns shared with Phukha in Table 3 is to assume that there was little to no distinction between Hle‑ pho and Phukha prior to their separation into two distinct languages precipi‑ tated by Phukha’s emigration toward Vietnam. In other words, the ancestors of modern-day Phukha speakers must also have been located in this central region prior to their southward emigration. In order to represent this pattern of diversification as a subgrouping model, both tree and wave diagrams are needed. Figure 2 provides two illustrations of how this might be accomplished: the left hand diagram blends wave-dynamics with treelike splits; the right hand diagram mixes tree dynamics with wave-like linkages. In both cases, the Hlepho-Phukha variety is placed in the middle to illustrate its mediating role historically. Phowa Meso-Clade

Phowa Micro-Clades: Ani-Labo

Ani Phowa Labo Phowa

Hlepho- KhlulaPhukha Zokhuo

Phowa Cladistic Hinge (Dynamic wave model) Discrete Languages Transitional Dialects

Hlepho Phowa Khlula

Phukha

Zokhuo

Languages: Ani Labo Hlepho Phukha Khlula Zokhuo Phowa Cladistic Hinge (Dynamic tree model)

Historical Contact Ongoing Contact

Figure 2.  Phowa hinge diagrams: Dynamic wave and tree models for cladistic hinge ­diversification

 Jamin Pelkey

In addition to featuring individual-identifying linguistic patterns of their own (§4.4), Hlepho varieties feature much internal diversity (see Table 2 and Figure 1 for illustrations). As mentioned above, the Hlepho language still functions as a gradient dialect continuum in its own right. This is an important supporting piece of evidence for defining the clade hinge. Further adding to the linguistic evidence of a clade hinge definition is the paradigm leveling effect apparent across tone classes in Hlepho, which has resulted in a disproportionately high occurrence of mid-level pitch realizations in numerous environments across tone classes, as dis‑ cussed in §4.4. As outlined in §§4.1–4.2, and as indicated in the summary of criteria above, sociohistorical and statistical results provide further important evidence in iden‑ tifying hinge diversification. These sources of evidence range from onomastic identities to network phenogram patterns that would themselves be difficult (and perhaps impossible) to explain without reference to cladistic hinge variation. Con‑ versely, without these additional sources of evidence, it would be difficult (and perhaps impossible) to account for conflicting claims implicit in the linguistic data summarized above. As such, this paper further supports Anttila’s (1994) argu‑ ment that multiple-measure analyses, practiced in an interpretive, philological mode of inquiry, are better suited to the ambiguities of language variation and genetic linguistics than any single measure analysis can approximate (see Pelkey 2013, 2014 for further discussions and applications of this point). Evidence from social geography, history, culture and spatial distribution may be combined with ­individual-identifying evidence of genetic relationship and statistical distance measurements from dialectometry to establish the presence of a cladistic hinge.17 Gradient distinctions in identity and comprehension mark the transition zone between Hlepho and the Labo varieties to the northwest. Remnant marriage net‑ works that persist in spite of identity distinctions and comprehension barriers to the southeast mark a second historical transition zone between the Hlepho variet‑ ies and Khlula-Zokhuo. As long as marriage networks, trade events, festivals or other traditional social contracts are maintained, and as long as the original geographic situation is not disrupted by out-migration, the mediating role of the clade hinge may continue even after sister-clade varieties have become unintelligible due to d ­ ifferentiation .  More generally, the approach above has attempted to combine evidence from language variation and comparative reconstruction with evidence from sociohistory and areal typology using both qualitative and quantitative methods in an attempt to contribute to an emerging “paradigm in which the study of all types of linguistic variation – cross-linguistic (typology), intralinguistic (sociolinguistics and language acquisition) and diachronic (historical linguistics) – are unified” (Croft 1990: 258–259).



Reconstructing phylogeny from linkage diffusion 

and/or separation. We need only appeal to structural analogies that result from adult second language learning of a closely related variety, or to childhood simul‑ taneous bilingualism to supply an underlying mechanism for the innovation and spread of different features at different times and in different directions (i.e., to the exclusion of alternate innovations, times and directions). The former often occurs in market town situations such that older adults are more likely to comprehend (and speak) the sister-variety than younger adults. The latter often occurs when children grow up learning to speak two closely related varieties from birth: the variety spoken locally by family and friends and the closely-related mother-tongue of their in-married mother. Further discussion of such phenomena are provided in Trudgill (1989: 234, 1992: 197–199) and Croft (2000: 192), among other places. Given the large speaker population, broad distribution and wide diversity of the Hlepho varieties, it is plausible that historical innovations transferred to and fro across geographic linkage zones in the northeast and southwest regions of the Hlepho-Phukha distribution would have naturally been absorbed by other inter‑ mediate Hlepho varieties. Hlepho varieties that are relatively more isolated from clade-internal contact are also more centrally situated geographically (i.e., variet‑ ies such as FZK). These aspects of the situation would have naturally prevented the transfer of innovations from Ani-Labo to Khlula-Zokhuo. This is one reason for preferring a hinge analogy over alternative ‘mediator’ analogies such as conduits or bridges. There are others reasons for this decision as well. 5.2  Rationale for the ‘hinge’ analogy in light of related alternatives Both wave and tree models are required in order to represent features of diffu‑ sion and phylogeny present in cladistic hinge variation. Naturally, related models proposed elsewhere also seek to blend aspects of diffusion and phylogeny using different analogies. This section discusses reasons for choosing a ‘hinge’ analogy as opposed to other possible analogies via comparison with related patterns discussed elsewhere in the literature, including ‘bridge’, ‘linkage’ and ‘hybrid’ metaphors. To my knowledge, Hock (1991) comes closest to identifying what I am recom‑ mending as a ‘hinge’ pattern of linguistic diversification above, in a brief discus‑ sion of German dialect geography (Hock 1991: 435–450). Focusing on the status of Frankish varieties (including dialects spoken in Trier and Cologne) relative to Upper High German in the south (Alemannic and Bavarian) and Low German varieties to the north, he finds that the unique status of Frankish lies in the dual facts that (1) speakers share distinctive innovations with varieties of High and Low German spoken on either side but (2) also exhibit “a clear and separate identity as the consequence of exclusively shared common innovations” (Hock 1991: 448). Hock summarizes these dynamics with the claim that Central Franconian variet‑

 Jamin Pelkey

ies form “a bridge, as it were, between the High German dialects and Low Ger‑ man” (1991: 448). Hock’s metaphor is useful, and is at least partially validated by the data he presents; but to the degree that the situation he describes approximates the Phowa situation described above, it may be incompatible with major features of his selected analogy. After all, a single bridge would facilitate innovations, communi‑ cation or sharing between two land masses across an erstwhile divide, functioning much like a single conduit, while the phenomenon under consideration in the study above (and, apparently, in Hock 1991) would be better compared to three land masses interconnected by two bridges, the middle land mass being an island. Additionally, for the analogy to fit the data, traffic originating on either of the two outer land masses would seldom, if ever, manage to cross both bridges, while traffic originating on the central land mass might frequently cross both. In short, the bridge analogy quickly becomes unmanageable, or inelegant, particularly for describing the Phowa diversification pattern described above, but also, I would argue, for the pattern that emerges from Hock’s German data. Analogies that typify other models, such as ‘linkages’ and ‘hybrids’, fit the Phowa data in ways that are less problematic. Ross’s (1988, 1997) linkage analogy discussed above is at least partially applicable to the Phowa data since the ‘hinge’ distinction being proposed qualifies as a specific type of linkage. ­Current linkage sub-types are limited to diachronic networks and diachronic continua, however, while the new distinction in question cannot be purely classified as either. Croft’s (2000) ‘hybridization’ analogy, drawn from botany (i.e., interspecific botanical hybridization) is also partially applicable; but Croft applies the analogy to koi‑ néization processes and dialect mixing (2000: 209, 212, 230), two situations that range beyond clade-internal contact, particularly those in which two distinct ­sister-clades are mediated by a third centrally situated sister-clade. The hinge analogy is proposed (following a suggestion in Pelkey 2008, 2011a), then, because it is more functional than, and more inclusive of, other closely related analogies. In other words, it models more aspects of the phenomenon in question while incorporating useful aspects of related analogies in the process. A hinge, after all, provides a type of linkage between two distinct objects. It includes a central axis that is comparable to a bridge insofar as it connects two otherwise disconnected phenomena (but is not comparable to a bridge in the distracting ‘direct conduit’ sense critiqued above). A hinge is also a distinct object that comes to be integrated with two other objects, bringing them together into a kind of hybrid object, without being reducible to either object and without reducing either object to itself. In the analogy I wish to draw, then, the hinge is a geographically central set of communalects which mediate innovations from the (physical and genetic)



Reconstructing phylogeny from linkage diffusion 

extremes of a sub-branch diachronically, all the while maintaining (or eventually developing) a distinct heritage through the group’s own exclusively shared innova‑ tions. In the Burmic example above, this role is filled by centrally located Hlepho varieties, historically situated in the geographic center of the Phowa clade within Southeastern Ngwi. 5.3  D  efining hinge diversification in light of established categories of linguistic convergence Interrelations and distinctions between transitional dialects, dialect continua, areal Sprachbünde and diachronic linkages are discussed here to clarify some of the ways in which they both overlap with, and contrast with, the new mode of hinge diversification being proposed. The discussion concludes with a general summary of the hinge model. Sprachbünde or areal convergence zones involve the historical mixing of languages from a variety of families in a given region, leading to a “gradient of ­contiguous languages that share features as the result of contact” (Matras 2009: 266; also Heine 2011: 41). Languages in a given Sprachbund, such as the Sinosphere (­ Matisoff 1991), slowly come to share certain marked constructions and even come to influence each other typologically. With this in mind, ­Sprachbund ­convergence may properly qualify as a macro-level phenomenon geographically, ­taking ­language contact as its focus (Hieda et al. 2011). Transitional dialects constitute microcosms of linguistic convergence. In such scenarios, innovative variety A and conservative variety C are mediated by mixed variety B such that multiple sub-varieties of B may be identified: ‘mixed lects’, ‘fudged lects’ and ‘scrambled lects’ (Chambers & Trudgill 1998: 110–119).18 Change in this scenario is unidirectional and may or may not involve phylogeneti‑ cally significant innovations; furthermore, mixed lects are not discussed as being innovative centers (a.k.a. focal areas) in their own right. They are, rather, regional channels for change between a ‘focal’ area (inhabited by innovative variety A) and a ‘relic’ area (inhabited by conservative variety C). Dialect continua, also discussed as ‘dialect chains’, may be observed at either the micro-level or the macro-level geographically. In these scenarios speakers at the extreme ends of a convergence zone are typically contrasted with speakers at any two contiguous points along its length. Sociolinguistic and cognitive fac‑

.  Chambers & Trudgill (1998: 110-119) use the eastern British midlands between London and The Wash inlet to illustrate various features of mixed, fudged and scrambled lects that constitute a transition zone between innovative London dialects and more conservative northern dialects of British English.

 Jamin Pelkey

tors are also frequently emphasized over structural or typological gradience in the discussion of dialect continua since, while speakers of variety A may understand and identify with speakers of variety B, and while speakers of variety B may under‑ stand and identify with variety C, speakers of C may be unable to understand or identify with variety A and vice versa. As discussed above, linkage relationships are usually discussed as historical dialect continua (or historical networks of dia‑ lect continua). Both transitional dialects and dialect continua may contribute to the spread of exclusive innovations, and both are involved in cladistic hinge variation. Yet, the latter must be delineated for the following three reasons: (1) cladistic hinge variation involves the establishment of at least three distinctive clade-internal subgroups (or languages), (2) it involves bi-directional diffusion of innovations between the central subgroup and its two satellite subgroups and (3) it results in variation patterns that are more complex than transitional dialects and dialect continua combined. Concrete illustrations of these dynamics can be drawn from the language varieties discussed above. The gradient boundary between Labo and Hlepho qualifies their mode of diversification as a diachronic continuum (as discussed in Pelkey 2011a). Thus, the Labo-Hlepho relationship constitutes a micro-scale link‑ age of its own, and linkage relations (historical dialect chains and networks) are clearly involved in the Phowa clade data in question above. What this study illus‑ trates is that linkage relations can, in at least some cases, simultaneously involve diversification by differentiation (the wave-like relations of linkage proper) and diversification by separation (the classic family tree model). While Kalyan & François (forthcoming: 22) admit that tree-like affiliations are implicitly present in linkage relations (in spite of their stated goal of eradicating the tree model alto‑ gether), the current study identifies a mode of linkage variation in which wave and tree models are both clearly present: a mode of linkage defined as ‘cladistic hinge’ variation. To further test the explanatory power of the hinge model elsewhere in the world, an explicit outline of minimal criteria for identifying cladistic hinge varia‑ tion is needed. Cladistic hinge patterns of language diversification are currently expected to meet the following general criteria: 1. Lects belonging to three micro-clades (A, B, C) must affiliate with a common meso-clade based on shared innovations not present in the macro-clade. 2. Lects belonging to micro-clades A, B and C must (by definition) each show evidence of their own exclusive innovations. 3. Lects belonging to micro-clades A and C must each share exclusive innova‑ tions with B that they do not share with each other.



Reconstructing phylogeny from linkage diffusion 

4. Speakers of lects belonging to micro-clade B must be situated between A and C in physical space and/or must have been so situated prior to migration. 5. These patterns should be supported by sociohistorical and sociocognitive evi‑ dence such as reported marriage networks, folk histories, language attitudes, speech comprehension metrics and onomastic identity indices (ethnonyms, autonyms and exonyms). Furthermore, in a prototypical example, micro-clade B would be expected to func‑ tion as a dialect continuum in its own right, inclusive of one or more central hinge varieties that mediate between the two extremes. Hinge varieties may also be expected to show more evidence of paradigm leveling or structural simplification. While similar phenomena have been noted with much more preliminary analyses in German (Hock 1991: 435–450) and Bantu (Roth 2011: 100–117), the extent to which hinge diversification applies to other languages of the world (including Ger‑ man and Bantu) remains an open question. The question itself is simple and might be phrased like this: How often in the history of language diversification have two distantly separated sister languages been spoken on either side of a third closely related sister variety that features substantial dialect diversity and speakers engaging in frequent cladeinternal contact alongside both sister-language margins all the while remaining relatively isolated near its geographic core?

Among the many language families of the world, such a scenario is quite possibly common. It seems unlikely, at least, that the phenomenon would be restricted to a little-known cluster of Ngwi languages spoken in the Sino-Vietnam borderlands.

6.  C  onclusions Historical linguistic evidence discovered through the reconstruction of Phowa clade relationships suggests a previously undefined pattern of language diversifica‑ tion. I have proposed that this pattern be labeled ‘cladistic hinge variation’ or ‘hinge diversification’. Specifically, hinge diversification appears to be a unique sub-type of diachronic linkage, both encompassing and surpassing the two categories of link‑ age originally defined by Ross (1988: 8): networks and chains. Until now, networks and chains have been considered antithetical to diversification by separation, oth‑ erwise known as ‘family relations’, within a given sub-branch. In contrast with previous findings, the Phowa data presented above indicate that diversification by separation (the family tree model) is, in at least some cases, compatible with diver‑ sification by differentiation (the wave-like linkage model) – even for purposes of

 Jamin Pelkey

branch-internal subgrouping. These results further support Bowern’s assessment that diachronic linkage networks are actually “messy because of divergence pro‑ cesses; that is, it is not contact between related languages that directly produces ambiguities in discrete subgrouping, but rather conflicting language split” (Bow‑ ern 2013: 427). If splits are focal in linkage situations, splits must be reconstructed and modeled. This is particularly true in a cladistic hinge scenario. Since both branch-like phylogeny and wave-like diffusion are manifest in such cases, hinge relations cannot be adequately modeled without reference to both waves and trees. Simply put, cladistic hinge variation occurs when two early splitting ­sister-varieties, A and C, come to be spoken on either side of a centrally situ‑ ated, third sister-variety B; and when, over time, A and C come to share ­individual-­identifying innovations with B that they do not share with each other. As the three communalects diverge into distinct languages all three sister variet‑ ies (and their subsequent daughter varieties) also come to show evidence of their own individual-identifying innovations. These dynamics are illustrated using both wave and tree diagrams in Figure 3. Meso-Clade

Cladistic Hinge

A1 varieties A2 varieties

Micro-Clades: A Micro-

B

Discrete Languages Transitional Dialects

C B2 Hinge Varieties

C2 varieties

B1 Varieties:

A1

A2

B1

B2

Cladistic Hinge

C2

C1 varieties

C1 Historical Contact Ongoing Contact

Figure 3.  Idealized cladistic hinge diagrams: Dynamic wave and tree models

Cladistic hinge varieties show evidence of both clear diachronic splits and overlapping isoglossic diffusion at the level of individual-identifying shared inno‑ vations. Relationships between varieties cannot be reduced to a classic wave dia‑ gram because unambiguous splits are present diachronically, particularly between varieties A and C. Nor can they be reduced to a classic tree diagram because over‑ lapping innovations between A and B on one hand and B and C on the other would require variety B to be subgrouped with both A and C. Instead, both models



Reconstructing phylogeny from linkage diffusion 

must be applied simultaneously. The two diagrams in Figure 3 illustrate idealized cladistic hinge scenarios by blending the wave with the tree (left) and the tree with the wave (right). It is important to emphasize that the ‘hinge’ model emerges from observed patterns of variation in Phowa. In other words, the cladistic hinge distinction is itself a reconstruction of ambiguous variation found in the Phowa clade that is otherwise impossible to interpret. The evidence for defining Hlepho as a dia‑ chronic hinge within the Phowa clade comes from multiple sources. Sociohis‑ torical dynamics, statistical network patterns and comparative reconstruction all suggest that Phowa languages diverged into three historical varieties, two of which (Ani-Labo and Khlula-Zokhuo) split early, were distantly separated and had no further contact with each other. These micro-clades came to develop their own languages, identities and distinctive linguistic innovations. Speakers of a third his‑ torical variety, situated between the two (Hlepho-Phukha), remained in contact with both and came to share individual-identifying innovations with each that the geographical extremes did not share with each other. Although Phukha speakers later emigrated south, toward Vietnam, Hlepho speakers remained in their central location. As discussed above, Hlepho speakers came to develop their own lan‑ guage and identity (including much internal dialect diversity) and their own sets of individual-identifying linguistic innovations. This reconstructed narrative accounts for otherwise confounding results in ways that affirm both wave-like patterns and tree-like patterns. Tree-like patterns are evident in the linguistic data since Ani-Labo ≠ Hlepho-Phukha or KhlulaZokhuo (see esp. Table 3, innovation set 1) and since Khlula-Zokhuo ≠ Ani-Labo or Hlepho-Phukha (esp. Table 3, set 14). Wave-like patterns are evident since, in other ways, Hlepho-Phukha shares individual-identifying innovations with both Ani-Labo (Table 3, sets 2–4, and Table 4) and Khlula-Zokhuo (see Table 3, sets 10–14, and Table 5), innovations that the latter two do not share with each other. As for the relative chronology of innovations shared across the entire Phowa clade, it is unlikely that they occurred later than the separation of KhlulaZokhuo from Ani-Labo, not only for key genetic linguistic reasons discussed in §4.3, but also for important sociohistorical and geolinguistic reasons discussed in §§3 and 4.1. In other words, genetic linguistic findings are necessary for this analysis but not sufficient on their own to reconstruct the cladistic hinge pattern of branchinternal diversification. Other sources of evidence are needed for the validation and interpretation of genetic linguistic findings, as outlined in §4.1. Clues from extant geographic distribution and accounts of historical migration are crucial for reconstructing features of the historical narrative (such as time depth and rela‑ tive chronology), as are identity distinctions, speech comprehension metrics and

 Jamin Pelkey

information on the maintenance of long-distance cultural traditions such as mar‑ riage networks. Quantitative methods such as network phenogram algorithms from dialectometry, illustrated in §4.2, provide further supporting evidence. With these considerations in mind, Phowa-internal relationships can neither be reduced to the grey zone of a typical dialect continuum nor to perfectly dis‑ crete nodes representing abrupt splits on a family tree. On the other hand, both modes of language change are demonstrably involved. Neither reality cancels out the other. To choose exclusively between wave or tree models in such cases would obscure the actual nature of the diachronic relationships in question.

References Aikhenvald, Alexandra Y. & Robert M.W. Dixon (eds.). 2001. Areal diffusion and genetic inheritance: Problems in comparative linguistics. Oxford: Oxford University Press. Anttila, Raimo. 1994. Collaterality and genetic linguistics. In Michael Shapiro (ed.), The Peirce Seminar papers: Essays in semiotic analysis, vol. 2, 29–46. Providence, RI: Berghahn Books. Bailey, Charles-James N. 1973. Variation and linguistic theory. Arlington, VA: Center for Applied Linguistics. Bailey, Charles-James N. 1974. Naturalness in historical reconstruction and changes that are not natural. In Anthony Bruck, Robert A. Fox & Michael W. LaGaly (eds.), Papers from the Parasession on Natural Phonology, April 18, 1974, 13–29. Chicago: Chicago Linguistics Society. Bailey, Charles-James N. 1996. Conceptualizing dialects as implicational constellations rather than as entities bounded by isoglossic bundles. In Peter Mülhäusler (ed.), Essays on timebased linguistic analysis, 118–150. Oxford: Clarendon Press. Barbançon, François, Tandy Warnow, Steven N. Evans, Donald Ringe & Luay Nakhleh. 2013. An experimental study comparing linguistic phylogenetic reconstruction methods. Diachronica 30(2). 143–170. DOI: 10.1075/dia.30.2.01bar Bloomfield, Leonard. 1933. Language. New York: Henry Holt and Company. Bossong, Georg. 2009. Divergence, convergence, contact: Challenges for the genealogical clas‑ sification of languages. In Kurt Braunmüller & Juliane House (eds.), Convergence and divergence in language contact situations (Hamburg Studies on Multilingualism 8), 13–40. Amsterdam: John Benjamins. DOI: 10.1075/hsm.8.01bos Bowern, Claire. 2013. Relatedness as a factor in language contact. Journal of Language Contact 6(2). 411–432. DOI: 10.1163/19552629-00602010 Bowern, Claire, Patience Epps, Russell Gray, Jane Hill, Keith Hunley, Patrick McConvell & Jason Zentz. 2011. Does lateral transmission obscure inheritance in hunter-gatherer languages? PloS ONE 6(9), e25195. 1–9. DOI: 10.1371/journal.pone.0025195 Bowern, Claire & Quentin Atkinson. 2012. Computational phylogenetics and the internal struc‑ ture of Pama-Nyungan. Language 88(4). 817–845. DOI: 10.1353/lan.2012.0081 Bradley, David. 1979. Proto-Loloish (Scandinavian Institute of Asian Studies Monograph 39). London: Curzon Press. Bradley, David. 1995. Grammaticalization of extent in Mran-Ni. Linguistics of the Tibeto-­ Burman Area 18(1). 1–28.



Reconstructing phylogeny from linkage diffusion 

Bradley, David. 1997. What did they eat? Grain crops of the Burmic groups. Mon-Khmer Studies 27. 161–170. Bradley, David. 2002. The subgrouping of Tibeto-Burman. In Christopher Beckwith & Henk Blezer (eds.), Medieval Tibeto-Burman languages (International Association for Tibetan Studies Proceedings 9 and Brill Tibetan Studies Library 2), 73–112. Leiden: Brill. Bradley, David. 2012. The characteristics of the Burmic family of Tibeto-Burman. Language and Linguistics 13(1). 171–192. Bryant, David & Vincent Moulton. 2004. Neighbor-Net: An agglomerative method for the con‑ struction of phylogenetic networks. Molecular Biology and Evolution 21(2). 255–265. DOI: 10.1093/molbev/msh018 Chambers, James K. & Peter Trudgill. 1998. Dialectology. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511805103 Chao, Yuanren. 1930. A system of tone letters. Le Maître Phonétique 30. 24–27. Croft, William. 1990. Typology and universals. Cambridge: Cambridge University Press. Croft, William. 2000. Explaining language change: An evolutionary approach. Essex: Longman. Croft, William. 2008. Evolutionary linguistics. Annual Review of Anthropology 37. 219–234. DOI: 10.1146/annurev.anthro.37.081407.085156 Dixon, Robert M.W. 1997. The rise and fall of languages. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511486869 Dixon, Robert M.W. 2002. Australian languages: Their nature and development. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511612060 Edmondson, Jerold A. 2002. Phu Kha, Xá Phó, Mantsi, Coong, Sila, Lahu, Hani, and ProtoLoloish data List. Online: http://ling.uta.edu/~jerry/tbv.pdf (1 July 2005). Edmondson, Jerold A. 2003. Three Tibeto-Burman languages of Vietnam. In David Bradley, Boyd Michailovsky, Randy LaPolla & Graham Thurgood (eds.), Language variation: Papers on variation and change in the Sinosphere and in the Indosphere in honour of James A. Matisoff, 305–320. Canberra: Pacific Linguistics. François, Alexandre. 2011. Social ecology and language history in the northern Vanuatu link‑ age: A tale of divergence and convergence. Journal of Historical Linguistics 1(2). 175–246. DOI: 10.1075/jhl.1.2.03fra François, Alexandre. 2014. Trees, waves and linkages: Models of language diversification. In Claire Bowern & Bethwyn Evans (eds.), The Routledge handbook of historical linguistics, 161–189. New York: Routledge. Fried, Robert Wayne. 2000. A preliminary phonological sketch of Phu-kha, a Tibeto-Burman language spoken in northern Vietnam. University of Texas, Arlington, MA thesis. Gray, Russell D., David Bryant & Simon J. Greenhill. 2010. On the shape and fabric of human history. Philosophical Transactions of the Royal Society London, B: Biological Sciences 365. 3923–3933. DOI: 10.1098/rstb.2010.0162 Greenberg, Joseph H. 1957 [2005]. The problem of linguistic subgroupings. In William Croft (ed.), Genetic linguistics: Essays on theory and method, 47–58. Oxford: Oxford University Press. Greenhill, Simon J., Thomas E. Currie & Russell D. Gray. 2009. Does horizontal transmission invalidate cultural phylogenies? Proceedings of the Royal Society B: Biological Sciences 276. 2299–2306. DOI: 10.1098/rspb.2008.1944 Greenhill, Simon J., Quentin D. Atkinson, Andrew Meade & Russell D. Gray. 2010. The shape and tempo of language evolution. Philosophical Transactions of the Royal Society B: Biological Sciences 277. 2443–2450. DOI: 10.1098/rspb.2010.0051

 Jamin Pelkey Hamp, Eric P. 1998. Whose were the Tocharians? Linguistic subgrouping and diagnostic idio‑ syncrasy. In Victor H. Mair (ed.), The Bronze Age and Early Iron Age peoples of eastern Central Asia, 307–346. Washington, D.C.: Institute for the Study of Man. Heeringa, Wilbert & John Nerbonne. 2001. Dialect areas and dialect continua. Language Variation and Change 13(3). 375–400. DOI: 10.1017/S0954394501133041 Heggarty, Paul, Warren Maguire & April McMahon. 2010. Splits or waves? Trees or webs? How divergence measures and network analysis can unravel language histories. Philosophical Transactions of the Royal Society B: Biological Sciences 365. 3829–3843. DOI: 10.1098/rstb.2010.0099 Heine, Bernd. 2011. Areas of grammaticalization and geographical typology. In Osamu Hieda, Christa König & Hiroshi Nakagawa (eds.), Geographical typology and linguistic areas: With special reference to Africa, 41–66. Amsterdam: John Benjamins. DOI: 10.1075/tufs.2.06hei HHYC, Editorial Committee (eds.). 2002. 红河彝族辞典Hónghé Yízǔ cídiǎn [An encyclopedic dictionary on the Yi nationality of Honghe Prefecture]. Kunming: Yunnan Minzu Chubanshe. Hieda, Osamu, Christa König & Hiroshi Nakagawa (eds.). 2011. Geographical typology and linguistic areas: With special reference to Africa (Tokyo University of Foreign Studies, Studies in Linguistics 2). Amsterdam: John Benjamins. DOI: 10.1075/tufs.2 Hock, Hans Henrich. 1991. Principles of historical linguistics, 2nd edn. Berlin: Mouton de Gruyter. Holden, Clare J. & Russell D. Gray. 2006. Rapid radiation, borrowing, and dialect continua in the Bantu languages. In Peter Forster & Colin Renfrew (eds.), Phylogenetic methods and the prehistory of languages, 19–31. Cambridge: McDonald Institute for Archeological Research. Huson, Daniel H. 1998. SplitsTree: A program for analyzing and visualizing evolutionary data. Bioinformatics 14(10). 68–73. DOI: 10.1093/bioinformatics/14.1.68 Huson, Daniel H. & David Bryant. 2010. SplitsTree4, version 4.11.3. Downloaded from http:// www.splitstree.org/ (1 February 2011). Kalyan, Siva & Alexandre François. Forthcoming. Freeing the comparative method from the tree model: A framework for historical glottometry. In Ritsuko Kikusawa & Lawrence A. Reid (eds.), Let’s talk about trees: Tackling problems in representing phylogenetic relationships among languages (Senri Ethnological Studies). Osaka: National Museum of Ethnol‑ ogy, http://bit.ly/SK-AF_HG. (2 July 2015). Kessler, Brett. 2001. The significance of wordlists (Dissertations in Linguistics). Stanford: CLSI Publications. Labov, William. 2007. Transmission and diffusion. Language 83(2). 344–387. DOI: 10.1353/lan.2007.0082 LaPolla, Randy J. 2012. Comments on methodology and evidence in Sino-Tibetan comparative linguistics. Language and Linguistics 13(1). 117–132. LaPolla, Randy J. 2013. Subgrouping in Tibeto-Burman: Can an individual-identifying standard be developed? How do we factor in the history of migrations and language contact? In Balthasar Bickel, Lenore A. Grenoble, David A. Peterson & Alan Timberlake (eds.), Language typology and historical contingency: A festschrift to honor Johanna Nichols, 463–474. Amsterdam: John Benjamins. DOI: 10.1075/tsl.104.21lap Levinson, Stephen C. & Russell D. Gray. 2012. Tools from evolutionary biology shed new light on the diversification of languages. Trends in Cognitive Sciences 16(3). 167–173. DOI: 10.1016/j.tics.2012.01.007



Reconstructing phylogeny from linkage diffusion 

Matisoff, James A. 1972. The Loloish tonal split revisited (Research Monograph 7). Berkeley: Center for South and Southeast Asia Studies. Matisoff, James A. 1978. Variational semantics in Tibeto-Burman (Occasional Papers of the Wolfenden Society on Tibeto-Burman Linguistics 6). Philadelphia: Institute for the Study of Human Issues. Matisoff, James A. 1983. Linguistic diversity and language contact. In John McKinnon & Wanat Bhruksasri (eds.), Highlanders of Thailand, 56–86. Oxford: Oxford University Press. Matisoff, James A. 1991. Sino-Tibetan linguistics: Present state and future prospects. Annual Review of Anthropology 20. 469–504. DOI: 10.1146/annurev.an.20.100191.002345 Matras, Yaron. 2009. Language contact. Cambridge: Cambridge University Press. DOI: 10.1017/CBO9780511809873 Meillet, Antoine. 1925 [1967]. La méthode comparative en linguistique historique [The comparative method in historical linguistics]. Translated by Gordon B. Ford, Jr. Paris: Librairie Honoré Champion. Nerbonne, John. 2009. Data-driven dialectology. Language and Linguistics Compass 3(1). ­175–198. DOI: 10.1111/j.1749-818X.2008.00114.x Nichols, Johanna. 1996. The comparative method as heuristic. In Mark Durie & Malcolm Ross (eds.), The comparative method reviewed, 39–71. Oxford: Oxford University Press. Pelkey, Jamin. 2008. The Phula languages in synchronic and diachronic perspective. Melbourne: La Trobe University Ph.D. dissertation. Pelkey, Jamin. 2011a. Dialectology as dialectic: Interpreting Phula variation (Trends in Linguis‑ tics. Studies and Monographs 229). Berlin: De Gruyter Mouton. DOI: 10.1515/9783110245851 Pelkey, Jamin. 2011b. A Phula comparative lexicon: Phola, Phuza, Muji, Phowa, Azha (SIL Lan‑ guage and Culture Documentation and Description 18). Dallas: SIL International. Pelkey, Jamin. 2013. Analogy, automation and diagrammatic causation: The evolution of TibetoBurman *lak. Studies in Language 37(1). 144–195. DOI: 10.1075/sl.37.1.04pel Pelkey, Jamin. 2014. Diagnostic dialectology: Interpreting Ngwi variation in China’s Red River valley. In Alena Barysevich, Alexandra d’Arcy & David Heap (eds.), Proceedings of Methods XIV: Papers from the Fourteenth International Conference on Methods in Dialectology, 2011 (Bamberg Studies in English Linguistics 57), 236–248. Frankfurt: Peter Lang. Ross, Malcolm D. 1997. Social networks and kinds of speech-community events. In Roger Blench & Matthew Spriggs (eds.), Archaeology and language 1: Theoretical and methodological orientations, 209–261. London: Routledge. Ross, Malcolm D. 1988. Proto Oceanic and the Austronesian languages of western Melanesia. Canberra: Pacific Linguistics. Roth, Tim. 2011. The genetic classification of Wungu: Implications for Bantu historical linguistics. Langley, BC: Trinity Western University MA Thesis. Schmidt, Johannes. 1872. Die Verwandschaftsverhältnisse der indogermanischen Sprachen. ­Weimar: Böhlau. Shafer, Robert. 1974. Introduction to Sino-Tibetan, vol. 1. Wiesbaden: Otto Harrassowitz. Sober, Elliott. 1991. Reconstructing the past: Parsimony, evolution, and inference. Cambridge, MA: MIT Press. Thomason, Sarah Grey & Terrence Kaufman. 1988. Language contact, creolization, and genetic linguistics. Berkeley: University of California Press.

 Jamin Pelkey Trudgill, Peter. 1989. Contact and isolation in linguistic change. In Leiv E. Breivik & Ernst H. Jahr (eds.), Language change: Contributions to the study of its causes (Trends in Linguistics. Studies and Monographs 43), 227–237. Berlin: Mouton De Gruyter. Trudgill, Peter. 1992. Dialect typology and social structure. In Ernst H. Jahr (ed.), Language contact: Theoretical and empirical studies, 195–212 (Trends in Linguistics. Studies and Mono‑ graphs 60). Berlin: De Gruyter Mouton. Yang, Cathryn. 2012. Classifying Lalo languages: Subgrouping, phonetic distance and intelligi‑ bility. Linguistics of the Tibeto-Burman Area 35(2). 113–137.

Abbreviations DIM diminutive H high checked tone L low checked tone NC no cognate ND no data NMLZ nominalizer TC tone class Note: see Table 1 for key to  abbreviations for Phowa-clade language varieties

Résumé Les modèles des liens de parenté des langues (Ross 1988; François 2014) représente une différenciation lente des langues apparentées via le continuum dialecte. On dit que de telles relations historiques ont pour objectif d’empêcher la reconstruction d’une phylogénie intra-branche. Un type de variation des parentés, récemment défini, met à l’épreuve cette théorie restrictive. En diversification cladistique des charnières, les locuteurs d’une langue, centrés sur une zone géographique, arbitrent les innovations entre les extrémités isolées d’une sous-branche, alors que les trois branches sœurs prouvent l’existence d’innovations qui leur sont propres. Ce qui en résulte, fait que les relations entre les langues apparentées se fondent. Suite à une revue du contexte, l’article s’appuie sur l’exemple de la distinction des langues du Phowa du Sud de la Chine (Ngwi < Burmic < Tibeto-Burman). Les analyses de données intègrent des éléments sociologiques, historiques, dialectaux et géné‑ tiques de la langue. Le raisonnement affirme l’existence des 2 modèles de vagues et d’arbres d’évolution des langues, permettant ainsi une meilleure compréhension des zones de transition focales et du passé, et leur influence sur le niveau, le dével‑ oppement et la diffusion des innovations linguistiques.



Reconstructing phylogeny from linkage diffusion 

Zusammenfassung Verknüpfungsmodelle der Sprachdiversifizierung (Ross 1988; François 2014) stellen die langsame Differenzierung nahe verwandter Schwestersprachen über Dialekt-Kontinua. Derartige historische Zusammenhänge sollen die Rekonstruk‑ tion der brancheninternen Phylogenie verhindern. Eine neu definierte Art der Verknüpfungsvariation stellt diese Einschränkung in Frage. In der kladistischen Kupplungsdiversifizierung vermitteln die Sprecher einer geografisch zentralen Diversität Wortschöpfungen zwischen isolierten Extremen einer Sub-Branche, während alle drei Tochter-Branchen eine Nachweisliste ihrer eigenen exklusiven Wortschöpfungen führen. Das resultierende Muster vermischt Verknüpfungsbezie‑ hungen mit Familienbeziehungen. Nach einer kontextuellen Beurteilung stellt die Arbeit Nachweise für die Unterscheidung von den Phowa-Sprachen in Südwest‑ china (Ngwi < Burmisch < Tibetobirmanische) bereit. Die Datenanalyse umfasst Sozialgeschichte, Dialektometrie und genetisch-linguistische K ­ omponenten. Das Argument bekräftigt sowohl Wellen- als auch Baum-­Modellierungen des Sprach‑ wandels und ermöglicht ein bereicherndes Verständnis der Schwerpunkte, Restund Übergangsbereiche, sowie deren Einfluss auf den Verlauf, die Entwicklung und Verbreitung von Sprachschöpfungen. Author’s address Jamin Pelkey Ryerson University Department of Languages 350 Victoria St., JOR 5 Toronto, ON M5B 2K3 Canada [email protected]