CoEDLFest CI presentations 2017

For a robot to talk with people, it needs language skills, but less obviously, it also needs social skills and a sense of the physical and social context of conversations. What does it take to engineer a robot as a social technology that can interact in real time in human environments? In this project we are designing social robots for young children that are both engaging and safe to touch (safe for both child and robot). The talk will describe insights from the project to date about shared personal spaces and cognitive architectures that support real time social interactions and grounded language.

Each year the sheer number of “things” that are connected to the Internet grows exponentially. But the real value of the Internet lies in the number of connections not only among things but also people, processes and data. This is the Internet of Everything and it is linked to the growth in smartphones, third party software solutions (termed apps), big data analytics and other digital health technologies. Joining this technological revolution is a social revolution in healthcare. This revolution puts the patient or consumer at the centre of their healthcare journey; a philosophy that is frequently referred to as patient empowerment or patient participation. Our CoEDL projects leverage off these technological and social changes to investigate the application of digital technology to understanding and eventually assisting the communication, particularly the conversational, abilities of people with dementia.

Processing (talking, listening) is what language exists for, and through processing we can address the Big Questions in language research. What is universal, what is language-specific? (A question for all of linguistics, but processing offers a unique and informative window on it.) When is information flow bottom-up, and when top-down? (A question for processing of all kinds, but with obvious applications to language processing.) What is the balance of generalisation (abstract representation, rule deduction) and exhaustive registration (episodic representations)? (A question for cognition, but somehow particularly fascinating for language researchers.) Cross-cutting these issues we then also need to know how perception (the ins) and production (the outs) relate: in their structure, in their relevance to these questions, and in operation.

In many Aboriginal communities of north-central Northern Territory, the first language of most children and young adults is a variety of Kriol, an English-based creole language. We still know relatively little, however, about how infants and how children acquire their local variety of Kriol as a first language, especially in the first couple of years of life. In this talk I’ll map out our progress on investigating Kriol acquisition at Barunga community and how the projects of different team members relate to this overall goal.

As a CI within CoEDL I have been working in collaboration with CoEDL Affiliate Francesca Merlan on a project that addresses four of the biggest mysteries about human beings and their social worlds, namely: 1. The question of child language acquisition: how do children manage to learn, in a few years, systems so complex that no computer can yet model their use, yet starting from an initial state that is flexible enough to let them learn languages of wildly differing structures? 2. How to understand the nature and development of intersubjectivity – our human capacity for understanding and sharing intentions and perspectives with others? 3. How to account for the way people come to think, feel and act as human beings embedded within particular communities and ways of life? 4. The question of social reproduction and transformation, that is, how do more-or-less shared social and cultural formations at various levels and time scales get reproduced and transformed in everyday human interaction? All of these questions have been extensively studied, from many different viewpoints. What is special – maybe even unique – about our project is the way we bring the four questions together and try to show how the answers to each can be improved by studying all of them in relation to each other. We do so in dialogue with relevant studies by others on all four questions, and on the basis of ethnographic evidence from the Ku Waru region in the highlands of Papua New Guinea, where we have been doing anthropological and linguistic-anthropological field research at intervals since 1981. In order to provide a concise example of the work we are doing along those lines, in my CoEDL Fest presentation I will focus on a brief snippet of conversation between a Ku Waru toddler and her father, and examine it from all four of the perspectives listed above. I will then turn to a brief general account of the work done on the project so far, what remains to done, and how the project is benefitting from being carried out within CoEDL.

Reports in the media regarding the linguistic and educational outcomes of Indigenous children living in remote communities suggest there is a huge divide between them and most other groups of Australian children. However, we know little about how children living in remote areas acquire their languages, whether Indigenous or creoles. Given the complexity of the language ecology in remote Australia, we need to understand much better the educational challenges the children face which is usually in Standard Australian English, as well as the impact of the swathe of standardized tests administered routinely across the school year. In seeking to find answers, this paper explores some of these challenges.

Processing phonetic detail is crucial to understanding speech in native and foreign languages, especially in a world with increased speaker, accent and language diversity. I will present new acoustic analyses that shed new light on properties of infant-directed speech and predict developmental trajectories in monolingual and bilingual infants. I will discuss some of our ongoing discoveries about infant, non-native and bilingual use of phonetic detail to learn words through tracking statistics. These new findings will be explored within a computational model, which accounts for the role of phonetic detail in speech processing.

Children must process their language in order to learn it; therefore a complete explanation of language acquisition must incorporate a parsing mechanism that enables children to identify and extract linguistic patterns from their input. In my talk I will provide an overview and update to the Canberra Longitudinal Child Language project, a large-scale longitudinal individual differences study of children’s language processing. The study is tracking children’s language processing skills across time and linking them to their subsequent language acquisition, with the aim of moving towards more dynamic mechanistic explanations of the acquisition process.

With increasing migration across the globe, there has been an upsurge of interest in “ethnolects”, varieties of speech used by ethnically diverse groups of young people in modern metropolises. Major projects are underway, for example, in London, Berlin, Stockholm, Toronto and, through the Sydney Speaks project, Sydney. Sydney Speaks considers quantitative patterns of variation in phonetic, morphosyntactic and discourse features as observed in spontaneous spoken language by Sydney-siders of diverse social backgrounds. A contemporary (2010s) corpus will be compared with a corpus of 1970s Sydney speech (Horvath 1985), allowing the fine details of language change to be tracked in real time and in actual language use. In this talk, some of the key questions to be addressed in the project are presented, motivated by some findings from the 1970s data.

Prevailing theories of how we produce and perceive spoken language are largely based on a handful of well-studied languages with a focus on segmental features. At the same time, few would dispute that phonetic realization of an individual speech segment depends on where that segment is positioned relative to the entire linguistic structure of an utterance. The languages of our region provide a rich source of prosodic as well as segmental diversity ranging from stress and tone languages to those that show a complete absence of lexical prosody. Detailed exploration of a variety of languages and language contexts is essential to test and extend our existing typological models.

What can contemporary language variation and change in Australia tell us about the past? Language shift situations are often characterised as displaying high levels of variation, assumed to be symptomatic of grammatical fragmentation and a lack of systematicity. Elsewhere in linguistic theory, variation is treated as a crucial component of language evolution, representing the principled reorganisation of a linguistic system. The Ngumpin-Yapa language subgroup in Australia ranks as one of the highest world-wide for levels of borrowing which has created problems for its reconstruction. The contemporary situation for these languages is equally complex, involving rampant variation and rapid change. But unlike the historical situation, we have recorded data from multiple speakers across multiple generations which means we can observe changes as they unfold and test the principles underlying them. These real time observations have the potential to provide a window on ancient language change.

We need to create more and better records of the world’s languages and we need to be able to find the records that already exist. CoEDL provides an opportunity to locate and preserve primary records (tapes, manuscripts, videos and so on) that are otherwise going to be lost, sometimes going to the grave with their creators. Records are being produced in the course of research but from the cradle of their creation there may be no consideration of their future accessibility. Our project is curating these items, and has been running training workshops to help speakers and researchers create robust primary sources.

How does the possibility space of language structures, in all its diversity, get to be populated? To answer this we need to go beyond discovering what is there (food for shape), to how it got there, and why languages get to some points more often than others (food for evolution). ‘Psycholinguistic bottleneck’ explanations in terms of ease of processing and learning only give us part of the answer. And ‘system-internal’ explanations in terms of resolving competing functional selectors (e.g. preferred semantic role in new mentions vs re-mentions shaping ergative vs accusative marking) likewise only account for part of the picture. Particularly challenging, for all evolutionary accounts, is the question of how singularities or rara arise, since most explanatory machinery in terms of learning and processing focusses instead on accounting for cross-linguistically common patterns. My focus in this talk will be on the challenges of accounting for these rara in evolutionary terms, and the need to invoke a variety of ‘system-external’ explanations for how diversity arises. I draw on examples from ongoing CoEDL research that examine the effects of linguistic modality (speech vs sign), of cultural difference in what comes up frequently in conversation, and of cultivated divergence between small speech communities in multilingual contexts.

We need to have a community of scholars working on endangered languages who share knowledge about what constitutes good practice for creating replicable data and making replicable analyses with respect to: Drawing on legacy material and the analyst's own recordings; Creating single snapshot grammars versus several time-slice snapshot grammars versus composite grammars derived from different speakers (what is common to all of them and what is only attested from a few speakers); Representing three interlinked types of information: source of information frequency of occurrence, uncertainty of analysis; and Resolving tensions between faithfulness to the input (e.g. recording who provided what information) and getting rapid distribution of material in accessible formats.

Understanding how languages change – or stay the same – over both time and space is critical to historical linguistics. A robust theory of language change will give us a window on language and its structure, human (pre)history, and human cognition and psychology. Here we discuss the different perspectives our team are bringing together to deepen our understanding of language change, and the ways in which we are using evolutionary approaches to investigate macro-level patterns of language continuity and change, and micro-level case studies of linguistic history, and the links between these levels.

A central goal of the Shape program is to dramatically increase the description and documentation of languages in our region, arguably the most linguistically diverse region in the world. Such language documentation has the potential to have a significant impact on our understanding of linguistic diversity if it can be used to inform our models of language structure, and how language is used, processed and acquired. In this talk I showcase the multigenerational documentation of Murrinhpatha project as one model of how this can be achieved, discussing results already achieved and directions for future research.

The talk will begin with a brief overview of my main question and a few of its more intractable challenges: developing an account of the evolution, genetic and cultural, of the distinctive features of human social and cognitive life; locating the origins of language within the framework, and identifying ways of imposing empirical constraints to squeeze the space of plausible scenarios. I then turn to a more specific issue, the use of the sender-receiver modelling tradition as a workhorse for probing the initial expansion of communicative capacities, and I shall use that framework to suggest that there is an unidentified constraint on that initial expansion.