Modeling in Historical Linguistics: Trisecting Computational Methods, Speech Communities, and Language Change1
Total Page:16
File Type:pdf, Size:1020Kb
Modeling in Historical Linguistics: Trisecting Computational Methods, Speech Communities, and Language Change1 Claire Bowern, Rice University Introduction §1. Three important issues facing Australian historical linguistics: 1. exceptionality 2. areality (and language contact, and their effects on reconstruc- tion) 3. reconstruction methods Today: State of the art and future directions in Australian prehistory; a look at the issues and some promising methods. §2. Outline: • a brief overview of Australian languages • the basis of claims of exceptionality in Australia • the contribution of areality and language contact and their ef- fects on reconstruction §4. Basis for classification • a case study of computational work in conjunction with the • Curr 1886: common lexical items comparative method. • Schmidt (1919): common lexical items, pronouns: ‘southern’ and ‘northern’ languages (but it’s unclear whether he means §3. Language Background: ‘Australian’ as a family genetic relationships) • 28 Phylic families • O’Grady et al. 1966a,b: Lexicostatistics + informal considera- • c. 250 languages tion of morphology (e.g. Yawuru and Karajarri): note that all • greatest diversity in North subsequent classifications are based on these works. • small and large families • Dixon 2002: Criteria unknown • Bowern and Koch 2004: Comparative method for assorted sub- Pama-Nyungan groups of Pama-Nyungan and Northern families. • 90% of Australia’s land mass Therefore: most of the (sub)groups haven’t been explicitly justified • c. 150 languages with anything more than a basic vocabulary list. This is work for the fu- > • 20 subgroups ture. • many singleton (unclassified) languages 1This work was funded in part by NSF CAREER BCS-0643517 “Pama-Nyungan Reconstruction and the Prehistory of Australia”. Many thanks to Luise Hercus and Gavan Breen for access to unpublished data for Karnic languages, and to Russell Gray for discussion of NNets. Networks drawn with Splitstree (ÛÛÛº×ÔÐiØ×ØÖeeºÓÖg). Exceptionality Age of the family: Pama-Nyungan and Proto-Australian §5. The origins of ‘exceptional’ claims about Australian languages There are claims (e.g. by Dixon) that Pama-Nyungan is an archaic fam- ily whose speakers spread out shortly after the initial colonisation of Aus- • nonconfigurational (e.g. Warlpiri; Hale 1983) tralia, and who have been in situ just about ever since. There are variations • syntactically ergative (e.g. Dyirbal; Dixon 1972) on this. • VC syllable structure (e.g. Arrernte; Breen and Pensalfini 1999) • unreconstructible or otherwise historically difficult (Pama- §9. Current distribution of languages vs colonisation Nyungan; Dixon 2001, 2002) • Humans have been in Australia for at least 40,000 years (Mul- That is, they exhibit structures which are different from languages else- vaney and Kamminga 1999). where in the world (no surprise – that’s true for all languages!) • It is unclear whether there has been more than one migration. §6. Specific historical claims: • If the current distribution is related, Australia is indeed ex- ceptional: • Australian languages have a dynamic which predisposes them – extreme linguistic conservatism. We would need to ex- to cyclic shift and extensive parallel development. (This is un- plain why these languages show the same sorts of similar- verifiable and unlikely; requires unscientific assumptions.) ities that we find in language families an order of magni- • the families are too old (> 40,000 years of settlement) tude younger. One way to explain this is by a combination • there’s been too much contact of cultural conservatism and extreme language contact. • we can’t get the right data – static populations... Data: exceptional? – little affected by climate change • This is directly contradicted by the archaeological record. §7. Data sources • Sutton, Evans, McConvell, Jones, etc: Pama-Nyungan spread • Linguists (most > 1955) linked to ‘intensification’ (Evans and Jones 1997, McConvell • Missionaries (much pre-1970) and Evans 1997, Sutton 1990) • Station owners (and other parties without training in linguis- tics) (19th Century; eastern Australia) Therefore whatever we say about initial colonisation in Australian lan- guages, we don’t need to take it into account when we are thinking about Note: very few native speaker linguists (exceptions include Laura lower level reconstruction. After all, the presence of Neanderthals in Eu- Dixon, Ephraim Bani and Leonora Adidi). It is a continent docu- rope is irrelevant to Indo-European and Finno-Ugric. mented by outsiders. §10. Is the time depth exceptional? This effects the amount of data we have for the languages, and it mul- tiplies uncertainties. These all feed back into difficulties in reconstruction • Probably not. We don’t know. (e.g. irregularity in sound change might be data error). • If a version of the McConvell/Evans/Jones/Sutton scenario is §8. Is the data state exceptional? true, there’s stuff that needs explaining, but Pama-Nyungan it- self isn’t exceptionally old. • Probably not especially so. • But it does affect reconstruction. 2 Areality and Contact §14. Example 2: Australia is multilinguistically diverse Claims of areality • Arnhem Land: community-wide multilingualism §11. Australia as a linguistic area A ‘hotbed of multilingualism’: • Lake Eyre Basin: key multilingual ‘translators’; widespread bilingualism • Breen 2007, Dixon 2001, 2002 • Western Desert: clan multilectalism but monolingualism • Heath 1978: linguistic diffusion in Arnhem Land • Dampier Peninsula: asymmetric bi/multilingualism • Dench 2001: it’s impossible to tell whether a given change is due to diffusion, parallel innovation, or shared innovation. Summary Claims: people borrow extensively from their neighbours, and they §15. Interim Summary borrow anything. • There’s interesting stuff going on It’s true that there are many areal phenomena in Australia. In East- • It’s complex ern Arnhem Land, exogamy has led to widespread multilingualism, to the • And it won’t be solved by single methods extent that it’s not unusual for older people to be fluent in five or more languages and for linguistically very diverse conversations with extensive §16. How to sort this all out? codeswitching to take place. It’s also true that there’s areal patterning at the local level for lots of features (Miller 1972 on Western Desert, my work • careful comparison using comparative method in Arnhem Land, see also Schebeck 2002). • areal analysis However, we need a much better idea of how geography, area and con- • quantification of results tact work in Australia before we can say for certain. It’s not the case that • engagement with the meta-arguments (e.g. plausibility) ‘anything goes’. If we assume that, we’re back to our circular exceptional assumptions. There’s no particular reason to assume that shared irregu- Now, this is all too complex for a single talk. larities (e.g. suppletive morphology) are easily borrowed in Australia. Karnic case study Two things stand out for further consideration: ideas of space, and pat- terns of multilingualism, In the time remaining I want to briefly present some results of recent re- search applying methods from computational biology to linguistic data. Desiderata for areal studies There are many issues that arise when formalising a heuristic (I along with §12. 1. A better model of geography: who was in contact? many other historical linguists regard the comparative method asa heuris- 2. A better idea of continent-wide multilingualism: who spoke tic (or set of problem-solving techniques) and not an algorithm). Work on what language(s)? this is at an early stage and we don’t know all the special issues which arise when using methods developed for different data. §13. Example 1: not all geography is created equal §17. • desert sand dunes funnel traffic in particular directions (cf Arabana-Wangkangurru; Hercus 1994) • river systems are important travel conduits in semi-arid areas • borrowing is sensitive to factors other than geography (geogra- phy is only one factor in variation) 3 Warluwarra N Guwa §20. Method: NNets Tobermorey Winton Jalarnnga H M a u Boulia y l l i • Coding cognate forms g R a . Pitta-Pitta n . R Longreach Alice Springs Antekerepenhe R . a • Code the correspondence sets for part of speech, semantic n i Mayawali Wangkajutjurru t n Bedourie a m SIMPSON Lake i a field D Lower Philippi DESERT Kunggari South Bilpamorea Karuwali Aranda Claypan • Compile networks for subsets of the data F Mithaka Windorah ink e . R Wangkangurru . R Tanbar R Karangura Birdsville o n o Lake o l • Not coded for archaisms vs innovation here s l Salt Yamma om u h B Lakes Yarluyandi Yamma T Ngantangara Quilpie • Multistate form-meaning correspondences SIMPSON Goyder Yawarra- Punthamara Lagoon n warka Marrgany o Durham t DESERT r u Ngamini Downs b r a Oodnadatta W Wangkumara GREYThargomindah RANGE §21. Expectations k C r Yandruwandha Lake Diyari e Antikirinya p Eyre o o STURT Naryilco Garlali Arabana C Badjiri DESERT Bulloo Downs Tree-like evolution Contact/diffusion Coober Pedy Lake Blanche Bulloo R. Overflow part of speech has no effect on part of speech subsets produce Wadikali Tibooburra Marree Pirlatapa . Guyani Leigh R tree different trees o Creek o . Malyangapa r R a g in 0 200 kilometres Lake P rl a semantic field has minimal effect semantic field affects topology Lake Frome D 0 100 miles Torrens independent of geography skewing correlates with geogra- phy §18. Some problems with Karnic data ... ... 1. Conflicting subgrouping (6 different proposals over the last 80 §22. Nominals only