Erich Round

Linguists seek insight from all human languages, however accessing information from most of the full store of extant global linguistic descriptions is not easy. One of the most common kinds of information that linguists have documented is vernacular sentences, as recorded in descriptive grammars. Typically these sentences are formatted as interlinear glossed text (IGT). Most descriptive grammars, however, exist only as hardcopy or scanned pdf documents. Consequently, parsing IGTs in scanned grammars is a priority, in order to significantly increase the volume of documented linguistic information that is readily accessible. Here we demonstrate fundamental viability for a technology that can assist in making a large number of linguistic data sources machine readable: the automated identification and parsing of interlinear glossed text from scanned page images. For example, we attain high median precision and recall (>0.95) in the identification of examples sentences in IGT format. ...

Full text: https://doi.org/10.1075/dia.20004.mac

Phylogenetic methods have broad potential in linguistics beyond tree inference. Here, we show how a phylogenetic approach opens the possibility of gaining historical insights from entirely new kinds of linguistic data-in this instance, statistical phonotactics. We extract phonotactic data from Pama-Nyungan vocabularies and apply tests for phylogenetic signal, quantifying the degree to which the data re ect phylogenetic history. We test three datasets: (1) binary variables recording the presence or absence of biphones (two-segment sequences) in a lexicon (2) frequencies of transitions between segments, and (3) frequencies of transitions between natural sound classes. Australian languages have been characterized as having a high degree of phonotactic homogeneity. Nevertheless, we detect phylogenetic signal in all datasets. Phylogenetic signal is greater in finer-grained frequency data than in binary data, and greatest in natural-class-based data. These results demonstrate the viability of employing a new source of readily extractable data in historical and comparative linguistics.

Causal processes can give rise to distinctive distributions in the linguistic variables that they affect. Consequently, a secure understanding of a variable's distribution can hold a key to understanding the forces that have causally shaped it. A storied distribution in linguistics has been Zipf's law, a kind of power law. In the wake of a major debate in the sciences around power-law hypotheses and the unreliability of earlier methods of evaluating them, here we re-evaluate the distributions claimed to characterize phoneme frequencies. We infer the fit of power laws and three alternative distributions to 166 Australian languages, using a maximum likelihood framework. We find evidence supporting earlier results, but also nuancing them and increasing our understanding of them. Most notably, phonemic inventories appear to have a Zipfian-like frequency structure among their most-frequent members (though perhaps also a lognormal structure) but a geometric (or exponential) structure among the least-frequent. We compare these new insights the kinds of causal processes that affect the evolution of phonemic inventories over time, and identify a potential account for why, despite there being an important role for phonetic substance in phonemic change, we could still expect inventories with highly diverse phonetic content to share similar distributions of phoneme frequencies. We conclude with priorities for future work in this promising program of research.

Linguistics, and typology in particular, can have a bright future. We justify this optimism by discussing comparability from two angles. First, we make explicit some of the logical underpinnings of typological sciences, linguistics included, which we believe are worth reminding ourselves of. Second, we give a brief illustration of comparison, and particularly measurement, within modern typology.

Feature stability, time and tempo of change, and the role of genealogy versus areality in creating linguistic diversity are important issues in current computational research on linguistic typology. This paper presents a database initiative, DiACL Typology, which aims to provide a resource for addressing these questions with specific of the extended Indo-European language area of Eurasia, the region with the best documented linguistic history. The database is pre-prepared for statistical and phylogenetic analyses and contains both linguistic typolog-ical data from languages spanning over four millennia, and linguistic metadata concerning geographic location, time period, and reliability of sources. The typological data has been organized according to a hierarchical model of increasing granularity in order to create data-sets that are complete and representative.

An automated method is presented for the commensurable, reproducible measurement of duration and lenition of segment types ranging from fully occluded stops to highly lenited variants, in acoustic data. The method is motivated with respect to the relationship between acoustic and articulatory phonetics and, through subsequent evaluation, is argued to correspond well to articulation. It is then applied to the phonemic stops of casual speech in Gurindji (Pama-Nyungan, Australia) to investigate the nature of their articulatory targets. The degree of stop lenition is found to vary widely. Contrary to expectations, no evidence is found of a positive effect on lenition due to word-medial (relative to word-initial) position, beyond that attributable to duration; nor do non-coronals lenite more than their apical counterparts, which freely lenite along a continuum towards taps. No signi cant effect is found of preceding or following vocalic environment. Taken together, the observed lenition, duration, and peak intensity velocities are argued to be inconsistent with a single, fully-occluded articulatory ‘stop’ target which is undershot at short durations, rather targets can be understood to span a range or ‘window’ of values in the sense of Keating (1990), from fully-occluded stop-like targets to more approximant- like targets. It is an open question to what degree the patterns found in Gurindji are language particular, or can be related to the organization of obstruent systems in Australian languages more broadly. Precisely comparable studies of additional languages will be especially valuable in addressing these questions and others, and are possible using the method we introduce.

Features are central to all major theories of syntax and morphology. Yet it can be a non-trivial task to determine the inventory of features and their values for a given language, and in particular to determine whether to postulate one feature or two in the same semantico-syntactic domain. We illustrate this from tense-aspect-mood (TAM) in Kayardild, and adduce principles for deciding in general between one-feature and two-feature analyses, thereby contributing to the theory of feature systems and their typology. Kayardild shows striking inflectional complexities, investigated in two major studies (Evans 1995a; Round 2013), and it proves particularly revealing for our topic. Both Evans and Round claimed that clauses in Kayardild have not one but two concurrent TAM features. While it is perfectly possible for a language to have two features of the same type, it is unusual. Accordingly, we establish general arguments which would justify postulating two features rather than one; we then apply these specifically to Kayardild TAM. Our finding is at variance with both Evans and Round; on all counts, the evidence which would motivate an analysis in terms of one TAM feature or two is either approximately balanced, or clearly favours an analysis with just one. Thus even when faced with highly complex language facts, we can apply a principled approach to the question of whether we are dealing with one feature or two, and this is encouraging for the many of us seeking a rigorous science of typology. We also find that Kayardild, which in many ways is excitingly exotic, is in this one corner of its grammar quite ordinary.

Typological datasets for quantitative historical-linguistic inquiry are growing in breadth, but a challenge is also to increase their depth, since advanced methods often ideally require many hundreds of traits per language. Using biphone transition probabilities from phonemicized vocabulary data, we extract several hundred high-definition phonotactic traits per language, for 17 languages in the Ngumpin-Yapa and Yolngu subgroups of the Pama-Nyungan family, Australia. We detect phylogenetic signal at a significant level (p < 0.001 for both subgroups), measured against a reference phylogeny inferred from basic vocabulary cognacy data. This contrasts with simpler, binary coding of biphones’ occurrence, which provides insufficient detail for the detection of phylogenetic signal. Thus, we demonstrate the viability of a new method in quantitative historical linguistics, and emphasize the inferential power to be harnessed from high-definition, trait-rich datasets for comparative research.

The contact history of the languages of the Eastern and Western Torres Strait has been claimed (e.g. by Dixon 2002, Wurm 1972, and others) to have been sufficiently intense as to obscure the genetic relationship of the Western Torres Strait language. Some have argued that it is an Australian (Pama-Nyungan) language, though with considerable infl uence from the Papuan language Meryam Mir (the Eastern Torres Strait language). Others have claimed that the Western Torres language is, in fact, a genetically Papuan language, though with substantial Australian substrate or adstrate influence. Much has been made of phonological structures which have been viewed as unusual for Australian languages. In this paper we examine the evidence for contact claims in the region. We review aspects of the phonology, morphology, syntax and lexicon of the Eastern and Western Torres Strait languages with an eye to identifying areal infl uence. Th is larger data pool shows that the case for intense contact has been vastly overstated. Beyond some phonological features and some loan words, there is no linguistic evidence for intense contact; moreover, the phonological features adduced to be evidence of contact are also found to be not specifically Papuan, but part of a wider set of features in Australian languages.

A revised model of Tangkic linguistic and cultural history is developed based on a reanalysis of relationships between six Tangkic languages in the southern Gulf of Carpentaria and drawing on recent archaeological and environmental studies. Bayesian phylogenetic analysis of Tangkic basic vocabulary was employed to infer the topology of the Tangkic family tree and define structural branching events. Contrary to previous models suggesting progressive colonisation and fissioning from mainland sources, the data support hypotheses that the modern configuration of Tangkic owes its form to pulses of outward movement from Mornington Island followed by subsequent linguistic divergence in both grammar and lexicon of the varieties. We also speculate that an extreme environmental event (c.800-400 BP) may have flooded low-lying coastal areas resulting in abandonment of some areas, a relatively short co-residence involving cultural and linguistic syncretism between neighbouring groups and then recolonization.

Debate over whether phonaesthemes are part of morphology has been long and inconclusive. We contend that this is because the properties that characterise individual phonaesthemes and those that characterise individual morphological units are neither sufficiently disjunct nor sufficiently overlapping to furnish a clear answer, unless resort is made to relatively aprioristic exclusions from the set of ‘relevant’ data, in which case the answers follow directly and uninterestingly from initial assumptions. In response, we pose the question: ‘According to what criteria, if any, do phonaesthemes distinguish themselves from non-phonaesthemic, stem-building elements?’, and apply the methods of Canonical Typology to seek answers. Surveying the literature, we formulate seven canonical criteria, identifying individual phonaesthemes which are more, or less, canonical according to each. We next apply the same criteria to assess non-phonaesthemic stem-building elements. The result is that just one criterion emerges which clearly differentiates the two sets of phenomena, namely the canonical accompaniment of phonaesthemes by non-recurrent residues, and this finding is not predetermined by our assumptions. From the viewpoint of morphological theory more broadly, we assume that any viable theory must find a place for lexical stems which are composed of a recurring, sound-meaning pairing plus a non-recurrent residue. Most phonaesthemes will occur in such stems. Consequently, theoretically interesting questions can then be asked about this entire class of lexical stems, including but not limited to its phonaesthemic members. Whether they are ‘part of morphology’ or not, phonaesthemes can contribute coherently to the development of morphological theory.

Kala Lagaw Ya is the language of the Western and Central islands of the Torres Strait. It exhibits an extremely complex pattern of ‘split argument coding’ (‘split ergativity’), which has previously been considered typologically exceptional, and problematic for widely discussed universals of argument coding dating back to work by Silverstein, Comrie and Dixon in the 1970s, and framed in terms of an ‘animacy’ or ‘nominal’ hierarchy. Furthermore, the two main dialects of the language, which centre around Saibai Island and Mabuiag Island, differ in the detail of their argument coding in interesting ways. In this paper we argue that once we take into account other typologically well-attested principles concerning the effect of markedness on neutralization in the morphological coding of grammatical categories, and in particular recent proposals about the typology of number marking systems, the Kala Lagaw Ya system falls into place as resulting from the unexceptional interaction of a number of universal tendencies.  On this view, the case systems of the two dialects of Kala Lagaw Ya, while complex, appear not to be typologically exceptional. This account can be taken as a case study contributing to our understanding of universals of argument coding and how they relate to forces affecting the neutralization of morphological marking. The reframing of the Kala Lagaw Ya facts then has broader implications: it reinforces the value of viewing complex patterns as the result of the interaction of simpler, more regular forces, and in so doing it also lends further empirical weight to the universals of argument coding which Kala Lagaw Ya was previously thought to violate.

Citation:
Macklin-Cordes, J. L. & Round, E. R. (2016). High-definition phonotactic data contain phylogenetic signal. Poster presented at the 90th Annual Meeting of the Linguistic Society of America. Washington, D. C.

See the extended paper:
Macklin-Cordes, J. L. & Round, E. R. (2015). High-definition phonotactics reflect linguistic pasts. Proceedings of the 6th Conference on Quantitative Investigations in Theoretical Linguistics. Tübingen: University of Tübingen. 5pp. http://dx.doi.org/10.15496/publikation-8609

Historical linguistic datasets are growing broader in terms of languages under study. However, a challenge is to also increase the depth of such datasets, since modern methods often ideally require hundreds of characters per language for statistical validity.
We extract many hundreds of high-definition characters from the phonotactics of two Australian language groups (Ngumpin-Yapa and Yolngu) and demonstrate that these contain phylogenetic signal. Thus, we demonstrate an important path towards power-intensive, modern methods.

Modern, large-scale typology, with its enormous datasets and batteries of algorithms rather than humans doing the comparisons, is more sensitive than ever to choices in how to code up language data. Thus we see increasing theoretical emphasis on the need for variables which robustly compare like with like (Haspelmath 2010, et seq.); which typologize language facts, not quirks of descriptive traditions (Hyman 2014); which decompose traditional variables into their finer-grained constituent notions (Bickel 2010, Corbett 2005, et seq.); and which attend closely to the logical relationships between those constituents (Round 2013). But how does this theory translate into the nitty-gritty work of actually building such variables? We offer a view from the coalface, as a complement to these more theoretical lines of thought, as we attempt to modernize, decompose and scrutinize one traditional typological variable: the presence or absence in a language of phonemic pre-nasalised stops. Our findings help to assess the practicality of some recent, theoretically-motivated proposals, and to nuance the theory itself.
The presence or absence of pre-nasalised stops, as a typological variable, has figured in several recent, large-scale typological studies (Dunn et al. 2005, Reesink et al. 2009, Donohue et al. 2013), however preliminary investigation (Round 2013) suggested that the variable performed poorly at comparing like languages with like. In response, we set about to develop a finer-grained set of ‘micro-variables’, which should encode richer information and perform better at comparing like with like. We coded them for 280 Australian and Papua New Guinean doculects (i.e., descriptions of language varieties; Cysouw & Good 2013), and along the way paid particular attention to the challenges we encountered.
A first finding is that, as a process which researchers undertake, the decomposition of typo-logical variables is iterative. The aim of decomposing a macro-variable is to tease apart some of the linguistic properties which it conflates, and which would lead to false comparisons of unlike with unlike. Our experience shows that in reality, it is likely that after a given round of decomposition, there will be further conflations that emerge and require addressing. To take an example, after we had created separate micro-variables which interrogate the structure of consonant clusters in word initial/medial/final position, we noted that these still conflate languages in which certain clusters are rare and those in which they are common. Accordingly, it would be wrong to view the building of micro-variables as a ‘fell-swoop’ process, or one which replaces ‘imperfect variables’ with ‘perfect variables’. Rather, it is a process for improv¬ing dataset design in an iterative fashion.
A second finding is that the logical dependencies between variables, including those which can be problematic for the ‘big data’ statistics coming into currency, may also be iterative, or tree-like in nature. For example, we find that micro-variables focusing on consonant clusters in certain positions funnel logically into a disjunctive ‘meso-variable’ focused on clusters in general, which then feeds along with other micro-variables into the macro-variable ‘are there prenasalised stops’.
Finally, we find strong evidence that preliminary suspicions about our pre-nasalised stop macro-variable were well founded. The macro-variable ‘are there prenasalised stops’ is primarily, and covertly, a variable about the size of consonant clusters, but one which (i) does poorly at grouping like languages with like, and (ii) is modulated by a second micro-variable which appears to us to act as a proxy for different schools of linguistic analysis, and not linguistic facts. Therefore, we strongly endorse a rapid shift towards micro-variate typology (Bickel 2010; Round 2013) if our aim is to reach better, clearer and deeper generalizations about human languages and human Language, from sound empirical data.

Australian languages are often cited for the unusual degree of homogeneity in their synchronic phonologies. In contrast, the fact that their diachronic sound patterns are equally if not more atypical has received little attention. We delineate the nature of the problem posed by sound change in Australian languages, and identify promising directions for the elucidation, and explanation of this uncommon state of affairs. 

THE POVERTY OF AUSTRALIAN SOUND CHANGE  In the comparative method, demonstrations of cognacy play a central role. A convincing demonstration requires regular correspondences, of which a significant number involve non-identical sounds. Non-identical sounds are necessary, since correspondences that are merely identical could result not only from shared descent, but from heavy borrowing. Potential cognates in Australian languages display a degree of phonological similarity which, to our knowledge, is simply not encountered in language families in other parts of the world. Potential cognates are often near-identical, and furthermore there is little recurrence of correspondences, as the number of cognate sets is only around 200. Given its atypicality, is not surprising that current theory provides few answers as to what to do with such data. Yet while most historical linguists will never face the problem of having an overwhelming majority of their sound correspondences being near identical, in the Australian case it is an issue which demands some kind of response, and presents both a puzzle and a challenge for theories of sound change.

THE POVERTY OF AUSTRALIAN PHONOLOGICAL DIVERSITY  What, then, are the tools we have to work with?  Synchronically, Australian languages display uniformity in static properties of their phonological systems – phonemic inventories, phonotactic constraints, morpheme structure conditions, and metrical systems. A point worth noting however, is that these metrics ignore the question of dynamic (morphophonological) alternations. Since the synchronic morphophonological alternations in any language typically have sound change antecedents, one might hypothesize that on a continent of absent sound changes, morphophonology should likewise be impoverished. In fact though, this is not the case.

THE POVERTY OF DIVERSITY RECONSIDERED  Results emerging from Round’s large survey of Australian languages’ morphophonology may provide new insight into Australian phonological diachrony and synchrony. If synchronic deletion and lenition processes reflect diachrony in at least most cases, then one effect of Australian sound changes is a tendency to preserve the typical Australian phonemic inventory and phonotactic patterns. Additionally, these new findings may shed light on the lack of observed changes in Pama-Nyungan roots. Butcher (2006) argues that post-tonic consonants in Australian languages occupy prosodically ‘strong’ positions. Assuming that these are resistant to changes such as lenition and deletion, a consequence is that the typical disyllabic Pama-Nyungan root will not contain the most common targets of sound change. Further exploration of links of this nature strike us as promising.

THE ROLE OF MULTILINGUALISM  Another question worth considering is whether the paucity of sound change and apparent high rates of lexical replacement may, in part, be the result of normal transmission in a multilingual context. Results emerging from an experiment by Ellison and Miceli investigating lexical choices in code-switching bilinguals show a statistically significant bias towards the avoidance of word forms that are shared (cognates or borrowings), if alternative, distinct word forms are available in the target language.  Simulations show that, diachronically, this would result in a fast depletion of cognate word forms from related languages in sustained, multilingual contact.  This type of methodology, combining experimental observation with simulations, could also be extended to the study of phonological categories, potentially yielding further insights into the problem of Australian synchrony and diachrony.

"Human languages can differ remarkably in the complexity of their morphophonology, and Australian languages are no exception. However, since the complexity of a linguistic analysis can also be due to the methods and assumptions of the analyst, when we are faced with a particularly complex analysis, it is informative for our understanding of both typology and linguistic practice to evaluate what complexity derives from the language itself, and what from the analyst's choices. We demonstrate the utility of this kind of enquiry by considering the morphophonology of Ngandi (Heath 1978).

Ngandi is an East Arnhem language, formerly spoken by at least five clans in the Rose River area, and one of five languages whose morphophonology was accorded a relatively complex analysis by Heath in a series of grammars written around 1980: Ngandi (1978); Ritharngu (1980); Warndarang (1980); Mara (1981) and Nunggubuyu (1984). Remaining within a rule-based paradigm, we demonstrate that some complexity is due to Heath's analytic choices and therefore can be reduced, though the language is undeniably rather complex itself. In particular, a more perspicuous analysis emerges when morphological conditioning is clearly partitioned from phonological conditioning. This is a practice which we strongly endorse when one's aim is to describe a language's morphophonology as perspicuously as possible. In addition, some complexity is reduced by reordering automatic, feature filling rules to a position late, rather than early, in the derivation, a principle which in various guises has been recognised by phonological theorists since pre-generative times.

Some specifics are as follows. Heath (1978) proposes (P-a) '[tense] feature filling', (Pb) 'morphologically conditioned hardening' and (P-c) 'lenition', which requires complex conditioning on (P-a) to derive a form like (1) where fG/ is underspecified for [tense].

(1) /yaŋ-Garu/ –(P-a)→ yaŋ-garu –(P-b)→ yaŋ-karu –(P-c)→ yaŋ-garu

By noticing that (P-c) serves to re-lenite those stops hardened by (P-b) in the same environment stated in the feature-filling rule (P-a), the analysis can be simplified by eliminating (P-c) and reformulating the 'feature filling rule' (P-a), now at the end of the derivation, as in (2):

(2) /yan-Garu/ –(P-b)→ yaŋ-karu –(P-a)→ yaŋ-garu

In addition, there is no need to characterise occlusive-final prefixes as conditioning hardening rule (P-b), since they are subsumed by the environment of (P-a) which always (re-)produces lenis morpheme-initial stops anyhow. The prefixes and stems which do trigger (P-b) are idiosyncratic and thus in any account of Ngandi, they will need to be specified lexically. Accordingly, following theories such as Lexical Phonology (Kiparsky (1982), Mohanan (1986)), we accord them a separate, morphologically determined stratum in which the hardening of (P-b) would apply; in other morphological contexts this stratum does not apply, and thus the final derivation can be simplified as in (3):

(3) /yan-Garu/ -(P-a)→ yaŋ-garu

Beyond the specifics of Ngandi, our results underscore the value in examining languages from multiple viewpoints. As with other, very well-studied languages, if our aim is to tease out the breadth and true nature of what is happening in the phonologies of human languages, then we will benefit from the availability of multiple analyses. By doing so, we can begin to clarify the impact of linguistic-analytic practice on our understanding of linguistic typology. "

Gender systems are endlessly fascinating, from those where meaning determines gender (Bagvalal), those where it is dominant but leaves intriguing loopholes (Mian) to those where form has an important role (Russian). Now it is time to integrate these systems into a fuller typology of nominal classification, taking in classifier systems as well as gender. Rethinking in this way leads us to take apart characteristics usually lumped together as defining gender, and those defining traditional classifiers. We then see that these characteristics combine in many ways. A canonical perspective proves helpful: we define the notion of canonical gender, and use this idealization as a baseline from which to calibrate the rich variety we find. It is then possible to approach the origin and nature of gender. Here Oceanic languages can provide a unique insight. Typically, a noun can occur with different possessive classifiers, depending on how the possessed item is used by the possessor: ‘my water (to drink)’ vs ‘my water (for something else)’. But in marked contrast, languages like North Ambrym (Vanuatu) typically have particular nouns occurring with one given classifier: water is just drinkable (Franjieh 2016). North Ambrym’s innovative system resembles a gender system: a noun takes a particular classifier regardless. We want to establish empirically whether gender systems can indeed emerge from possessive classifiers in this way. We must also uncover how and why languages would relinquish a useful, meaningful classificatory system, and move to a more rigid gender system. To this end we are running novel experiments to compare possessive classifier systems in six Oceanic languages of Vanuatu and New Caledonia, each with a different inventory size of classifiers — from two to twenty-three. Combining typology with psycholinguistics in this way promises to shed new light on how systems of nominal classification develop and function. The Oceanic data suggest that, in this instance, we find an interesting parallelism: diachronic change is running in the direction of canonicity.

Automated cognate alignment can help speed up the comparative method, however current implementations (List 2010, List et al 2018) are generic and phonetically course grained, so may miss detail and family-/area-specific information utilized in manual research. We demonstrate the advantages of adding family-specific historical phonological knowledge to automated processes. Using information on Australian sound changes (Dixon 2002) we assessed the alignments of 1,000 cognate sets across eight Gunwinyguan languages (Harvey 2003). Relative to a hand-curated gold standard, our aligner produced 25% fewer errors (36/1000) than the generic aligner (48/1000). Our method is straightforward and can be extended to other families.

For many revitalization projects, the only existing language data comes from word lists recorded by linguistically-naïve transcribers, yet there is a lack of standardized methodology to interpret these sources. We draw upon previous iterations of linguistic ‘reconstitution’ (Broadbent 1957; Dench 1999; Browne 2016) which themselves have drawn upon the historical-comparative method in order to transform the language data into a modern (i.e. phonemic) format, useable by linguists and language workers.
This paper aims to further refine the comparative reconsitution methodology by incorporating computer-assisted cognate alignment. This alignment takes identified orthographic ‘cognates’ as input and derives aligned correspondences. We tested our methodology on a collection of sources by linguistically-naïve English speakers recording Bunganditj (Pama-Nyungan, Australia: Blake 2003). Incorporating computational assistance dramatically reduces the duration of the reconstitution project, making it more suitable for revitalisation projects, while also increasing the accuracy of the results.

* Robots who read grammars *
Jayden L. Macklin-Cordes, Nathaniel L. Blackbourne, Thomas J. Bott, Jacqueline Cook, T. Mark Ellison, Jordan Hollis, Edith E. Kirlew, Genevieve C. Richards, Sanle Zhao, Erich R. Round

Poster presented at CoEDL Fest 2017, Alexandra Park Conference Centre, Alexandra Headlands, QLD, Australia. Hosted by the University of Queensland. 6 February 2017.
https://doi.org/10.6084/m9.figshare.4625248

* Abstract *
Linguistic typology has yet to undergo a computational revolution like that seen in other scientific endeavours. Nevertheless, we could soon be able to query the entire store of published knowledge on human languages when we do our research. To do so, knowledge must be represented in a machine-readable format. Here, we introduce a prototype ‘Grammar Harvester’, a set of processes for creating richly annotated, machine-readable versions of existing grammatical descriptions, starting from a scanned PDF. Further, we introduce ‘Finder’ and ‘Analyser’ Robots, scripts which automatically identify and compile information from harvested grammars using novel and existing ontologies of linguistic concepts.

* Author bio *
Jayden Macklin-Cordes is a PhD candidate at the Ancient Language Lab, University of Queensland. He is the lead investigator on CoEDL Transdisciplinary and Innovation Grant, 'A "data well" prototype for Sahul phonologies'. Erich Round (Ancient Language Lab director, UQ) and Mark Ellison (ANU) are collaborators on the same grant. Working hard on the project are Summer Research Scholars Sanle Zhao, Edith Kirlew, Thomas Bott and Nathaniel Blackbourne (all UQ), with further generous assistance from research assistants Genevieve Richards, Jordan Hollis, and Jacqueline Cook (all UQ).

Citation:
Macklin-Cordes, J. L. & E. R. Round, 2016. Reflections of linguistic history in quantitative phonotactics. Paper presented at the Australian Linguistic Society Annual Conference, Monash University, Caulfield, Australia. 7 December 2016. Doi: https://dx.doi.org/10.6084/m9.figshare.4299365

Abstract:
Advanced quantitative methods are at the cutting edge of historical linguistics, however these methods often ideally require many hundreds of data points per language. In order to generate reliable inferences at ever greater time depths, there is a need for typological datasets which are not only broader in coverage, but also contain a deeper store of information. We explore one avenue by extracting large numbers of high-definition phonotactic ‘traits’ per language. We show that these traits contain phylogenetic signal, thus demonstrating an important path towards high-powered methods of the near future.

Methodology: Languages may be compared in terms of which two-segment sequences they permit. Moreover, such biphones possess distinct lexical frequencies, which can also be compared. We examined whether such data contain information about family-tree structure, i.e., phylogenetic signal. Two standard statistics are used: D [1] tests coarse-grained biphone ‘permissibility’ data; and K [2] tests higher-definition transition probabilities.

We examined 2 subgroups of the Australian Pama-Nyungan family: 10 languages of Ngumpin-Yapa [3] and 7 of Yolngu [4], represented by phonemically-standardised lexicons from the CHIRILA database [5]. Phylogenetic signal is calculated with reference to phylogenies from C. Bowern (updated from [6]). Australian languages present a tough challenge, since phonotactically they are notoriously uniform [7­–9]. Moreover, Ngumpin-Yapa has some of the world’s highest borrowing rates [10–11]. Thus we hypothesized that the coarse-grained D test would fail. The key question is whether the high-definition K test succeeds.

Results: D attempts to reject two null hypotheses: that traits’ distributions are (A) too uniform to reveal structure present in the reference tree; and (B) random. We extracted 184 (Ngumpin-Yapa) and 164 (Yolngu) traits per language. We were surprised to reject both hypotheses for Yolngu (Stouffer’s Z>100, p=0.00): thus, even binary permissibility data revealed some phylogenetic signal. For N-Y only the second null hypothesis could be rejected (p=0.00), and further testing showed that when the subgroup’s outermost language was removed, even this failed. We conclude that binary phonotactic data contains weak phylogenetic signal at best; the Y result may represent statistical noise, and more subgroups should be tested.

K attempts to reject one null hypothesis: that no phylogenetic signal is present. A value K=0 represents random trait distribution relative to the reference tree; K=1 represents an exact match and K>1 indicates that outermost languages are even more distinct in the test data than in the reference tree. With 451 (Ngumpin-Yapa) and 541 (Yolngu) traits per language, we reject the null hypothesis in both subgroups (Stouffer’s Z=9.87; 17.6, p=0.00). In Ngumpin-Yapa, the confidence interval for K of [0.86, 0.92] indicates a very good match with the reference phylogeny, and in Yolngu, [1.15, 1.26] indicates an even stronger sorting of languages. Further testing, which removed the outermost language from both subgroups showed the result is stable: [0.81, 0.87] for Ngumpin-Yapa and [0.96, 1.00] for Yolngu.

Conclusion: As linguists attempt to up-scale efforts in quantitative historical linguistics, we demonstrate the significant potential of high-definition phonotactics, which permits the extraction of several hundred traits per language and has revealed phylogenetic signal in two Australian subgroups.


References:

[1]    S.A. Fritz and A. Purvis, “Selectivity in mammalian extinction risk and threat types: A new measure of phylogenetic signal strength in binary traits,” Conserv. Biol., vol. 24, no. 4., pp. 1042-1051, 2010.

[2]    S.P. Blomberg, T. Garland and A.R. Ives, “Testing for phylogenetic signal in comparative data: Behavioural traits are more labile,” Evolution, vol. 57, no. 4, pp. 717-745, 2003.

[3]    P. McConvell and M. Laughren, “The Ngumpin-Yapa subgroup,” in Australian Languages: Classification and the comparative method, C. Bowern and H. Koch, Eds. Amsterdam: John Benjamins, 2004, pp. 151-177.

[4]    Schebeck, Bernhard Dialect and Social Groupings in North East Arnhem Land, typescript, Australian Institute of Aboriginal and Torres Strait Islander Studies Library, Canberra, 1968.

[5]    C. Bowern, “Chrila: Contemporary and Historical Resources for Indigenous Languages of Australia,” Language Documentation and Conservation, vol. 10 http://nflrc.hawaii.edu/ldc/

[6]    C. Bowern and Q.D. Atkinson, “Computational phylogenetics and the internal structure of Pama-Nyungan,” Language, vol. 88, no. 4, pp. 817-845, 2012.

[7]    R.M.W. Dixon, The Languages of Australia, Cambridge: Cambridge University Press, 1980.

[8]    P.J. Hamilton, “Phonetic constraints and markedness in the phonotactics of Australian languages,” Ph.D. dissertation, University of Toronto, 1996.

[9]    B. Baker, “Word structure in Australian languages,” in The Languages and Linguistics of Australia: A comprehensive guide, H. Koch and R. Nordlinger, Eds. Berlin: De Gruyter Mouton, 2014, pp. 139-214.

[10] C. Bowern, et al., “Does lateral transmission obscure inheritance in hunter-gatherer languages?” PLoS One, 2011: e25195.

[11] P. McConvell, “Loanwords in Gurindji, a Pama-Nyungan language of Australia,” in Loanwords in the world's languages: A comparative handbook, M. Haspelmath and U. Tadmore, Eds. Berlin: Mouton de Gruyter, 2009, pp. 790–822.

[12] Y. Benjamini and Y. Hochberg, “Controlling the false discovery rate: A practical and powerful approach to multiple testing,” J. R. Stat. Soc. Series B (Stat. Methodol.), vol. 57, no. 1, pp. 289-300, 1995.

We report on a project which has created a digital version of lexical material on approx. 70 language varieties of Cape York, from the archival records of Bruce Sommer [1]. Our focus here is on methodology.

Background & aims:
Great strides have been made in preparing the lexicons of Australian languages in digitally readable and accessible form, however a notable gap so far is Cape York [2]. Bruce Sommer deposited lexical, grammatical and textual materials on some 70 language varieties of central and southern Cape York, comprising 4,950 pages of fieldnotes and summaries, and 203 audio tapes. Our aim was to key in Sommer’s handwritten and printed lexical materials, as a first step in the digital representation and eventual audio time-alignment of his invaluable archive.

Materials:
Fryer Library digitised Sommer’s print materials in 2014 and tapes in 2015. We identified 1,520 pages of lexical material. These wordlists range in length from 2 entries to 2635 (mean 485, median 255). Many are numbered, following the Hale–O’Grady 100-item list.

Methods:
Our work plan centred on simultaneous and collaborative data entry. Two researchers entered the same wordlist simultaneously into a Google spreadsheet, where the other’s activity is also visible. Each worker focussed on either the vernacular or English, but also provided constant checking of the other’s work, and assistance when necessary. The spreadsheet contained columns for: speaker, language, tape number, subheadings, page number, language form, notes on language form, English gloss, notes on English gloss, other text and notes on other text. Additional columns were added if wordlists become more complex: language form corrections, number, addi- tional language form columns for lists with two vernacular languages.

Challenges:
1. Legibility of handwriting was a challenge. To improve accuracy, researchers examined illegible entries together to reach agreement; if needed, other wordlists were consulted, to see if a word appeared elsewhere with a similar form. In rare cases where neither of these solutions worked, a note was entered.
2. Sommer used many abbreviations. These were gradually deciphered as our familiarity increased.
3. Some pages contained extensive corrections, annotations and/or margin notes; some had multiple languages or speakers. Extra columns were added for those documents.
4. Most of the materials were in IPA. This was entered using a convenient set of as hoc conventions to enable fast data entry, and then transposed into IPA afterwards. Having two researchers dealing collaboratively with challenges led to rapid and effective problem solving.
Analysis Cape York is a notoriously complex region [3]. Cross-linguistic datasets such as Sommer’s lexicons will make possible automated analyses which can detect diffuse patterns which challenge the observational and memory limitations of human linguists. We present some initial examples, including automated phylogenetic analysis [4]; network analysis [5]; and admixture analysis [6]. These do not replace expert manual analysis, but can increase productivity by rapidly highlighting areas deserving particular attention.

Methodological recommendations:
We cannot recommend strongly enough the method of collaborative data entry for this kind of data, which enables quick and effective detection and correction of data entry errors. It makes the task more collaborative, and hence enjoyable.


References:
[1] Sommer, B. 2003. Papers, 1964–2003 (item number UQFL476), Fryer Library, St Lucia.
[2] Bowern, C. 2016. Chirila: Contemporary and Historical Resources for the Indigenous Languages of Australia. Language Documentation and Conservation. Vol 10.
[3] P.Sutton(ed.) 1975. Languages of Cape York, Canberra: AIAS.
[4] Blomberg, S.P., T. Garland & A.R. Ives. 2003. Testing for phylogenetic signal in comparative data: Behavioral traits are more labile. Evolution 57:717-45.
[5] Bryant, D., & Moulton, V. 2004. Neighbor-net: an agglomerative method for the construction of phylogenetic networks. Molecular biology and evolution, 21(2), 255-265.
[6] Pritchard, J. K., Stephens, M., & Donnelly, P. (2000). Inference of population structure using multilocus genotype data. Genetics, 155(2), 945-959.

This article investigates the evolutionary and spatial dynamics of typological characters in 117 Indo-European languages. We partition types of change (i.e., gain or loss) for each variant according to whether they bring about a simplification in morphosyntactic patterns that must be learned, whether they are neutral (i.e., neither simplifying nor introducing complexity) or whether they introduce a more complex pattern. We find that changes which introduce complexity show significantly less areal signal (according to a metric we devise) than changes which simplify and neutral changes, but we find no significant differences between the latter two groups. This result is compatible with a scenario where certain types of parallel change are more likely to be mediated by advergence and contact between proximate speech communities, while other developments are due purely to drift and are largely independent of intercultural contact.