Three more dry runs for the Evolang conference tomorrow – everyone welcome, titles below!
Tue 25th March, 11.00-12.30, Room G.04, David Hume Tower (DHT Conference Room)
Caroline Kamps & Vanessa Ferdinand: The origins of regularity in language: why coordination matters
James Winters: Experimentally investigating the role of context in the structuring of the linguistic system over cultural evolution
Marieke Woensdregt & Willem Zuidema (Universiteit van Amsterdam): Neural networks, algebraic rules & human uniqueness
Tue 18th March, 11.00-12.30, M3 (room no. 1.05) Appleton Tower
Alan Nielsen: Motivated vs. conventional systematicity: implications for language learning and the structure of the lexicon
Marieke Schouwstra: About time: Semantic structure in emerging language
Monica Tamariz: Investigating the role of iconicity in the evolution of linguistic structure
Tue 11th March, 11.00-12.30, M3 (room no. 1.05) Appleton Tower
Kenny Smith
Eliminating unpredictable linguistic variation through interaction
Languages tend not to exhibit unpredictable variation. We explore alignment/accommodation during interaction as a mechanism to explain this cross-linguistic tendency. Specifically, we test the hypothesis (derived from historical linguistics) that interactions between categorical and variable users are inherently asymmetric: while variable users (of e.g. a grammatical marker) can accommodate to their partner by increasing their usage, categorical users should be reluctant to accommodate to variable partners, since this requires them to violate the rules of their grammar. We ran an experiment in which pairs of participants learnt a miniature language (featuring a potentially variable grammatical marker) and then used it to communicate. Our results support the hypothesis: variably-trained participants accommodate to their categorically-trained partners, who do not change their behaviour during interaction. More generally, interaction results in the elimination of variation: accommodation/alignment is a viable mechanism for explaining the absence of unpredictable variation in language.
Tue 4th March, 11.00-12.30, M3 (room no. 1.05) Appleton Tower
Marieke Woensdregt (Universiteit van Amsterdam)
I know what you did last iteration – Modelling the role of theory of mind in communication
Theory of mind – the ability to reason about the mental states of others – plays an important role in human communication. To be able to communicate successfully, it is important that both the speaker and the hearer have a model of each other’s knowledge and interests, and their common ground. Similarly, it has been argued that language is important for theory of mind, since it provides us with labels for and conversation about mental states. This plays a role in the development of theory of mind on the individual level, as is evidenced from experimental work with both typically and atypically developing children. But more recently it has also been hypothesized that such a positive feedback interaction between language and theory of mind has played a significant in the evolution of human language and social cognition.
To be able to explore the dynamics of such an interaction computationally, a model of communication is needed in which theory of mind plays an active role. In this talk I will present my proposal for such an integrative model, and the communication and learning dynamics it results in. The model I propose here is an agent-based model of a naming game, in which theory of mind-agents can learn a function that allows them to infer which meanings in the context are salient for the interlocutor, ultimately allowing them to communicate more successfully.
Tue 25th February, 11.00-12.30, Room G.04, David Hume Tower (DHT Conference Room)
Matt Spike
Lost in Transmission? The information dynamics of signalling games.
The ability to bootstrap learned communication systems without any apparent organisation has been shown in numerous experiments (eg. Fay et al. 2013) and models (eg. Steels 1999), and by the modern emergence of sign languages such as NSL. Various mechanisms have been suggested as driving this process, but most involve some treatment of reference and information. Shannon’s information theory (1948), with its concise measures of information and entropy, would seem to offer an ideal investigative framework: however, Shannon information famously ignores meaning. I propose a scenario which avoids this problem, in which an optimal system implies the preservation of reference. Following from this, a single measure of conditional entropy can be used to describe the overall dynamics of any signalling game, and can also diagnose whether any given system will converge on optimality. I will discuss the necessary ‘communicative bias’ in terms of several models of learning, feedback and contextual ambiguity, and (if there’s time) ask what some possible experimental applications might be.
Tue 11th January, 11.00-12.30, M3 (room no. 1.05) Appleton Tower
Ashley Micklos (UCLA) & Yasamin Motamedi
Gestural communication in the laboratory
Homesign systems and emerging sign languages offer insight into linguistic systems in their earliest stages of development. Recent research on Nicaraguan Sign Language and Al-Sayyid Bedouin Sign Language have allowed an understanding of the processes at work from the creation of homesign systems to the more developed languages that are in use today.
Although studies of gesture in the lab have thus far proved informative experimental studies involving communication via gesture alone are uncommon and there does not yet exist a real understanding of how to implement an interactive element into gesture studies.
We will present the results of a study that was intended to inform future experiments exploring the evolution of gestural communication, as well as exploring how interactional properties facilitate the evolution of these gestural systems. We will describe a communication task undertaken by pairs of participants who were asked to communicate using gesture only. We found that when the gestural systems created in these pairs were transmitted along diffusion chains, the gestures in the final generation of the chain took less time to produce than those at the beginning of the chain. Furthermore, the signs themselves showed greater predictability and we found that the systems used within pairs were more similar in the final generation that in initial generations.
Tue 28th January, 11.00-12.30, M3 (room no. 1.05) Appleton Tower
Vanessa Ferdinand
On the coevolution of culture and cognition
Human behavioral artifacts, such as language, music, institutions, laws, and technologies, all change over time. Why do we use different slang than our parents, why don’t we all know the same songs, and why do we update our software? One thing all cultural artifacts have in common is that they replicate by passing through a cognitive system; most commonly the human mind. I propose that the most salient driver of cultural change lies in our cognitive architecture, which is highly structured and biased in terms of what it can perceive, process, and produce. These biases can be understood as selection pressures, causing changes in cultural artifacts as they are transmitted over several generations of learners. To illustrate, I will present some of my research on linguistic regularization biases in adult learners and explain how these biases contribute to language change over time. I will also discuss research in the broader framework of iterated
learning, which is explaining how culture adapts to learners, but neglecting the other side of the story: how learners adapt to culture. Learning is an active process that restructures our brain by forming new neural connections and changing the strength of existing ones, both developmentally in childhood and throughout adult life. Current models tend to use static learners with fixed learning biases. Extending these models with adaptive learning algorithms will take on the complex coevolutionary nature of cultural evolution and provide us with better explanations for the changes we observe.
Tue 14th January, 11-12.30, M3 Appleton Tower
The lexicon as a dynamical system: Lexical competition and the evolution of phoneme inventories
Andrew Wedel (University of Arizona)
All human languages make use of small systems of signal categories, such as the sounds [p] and [b], in combination to compose meaningful lexical categories, such as the words ‘pat’ and ‘bat’. These perceptually contrastive, yet individually meaningless signal categories are often called phonemes. Over time, the properties of and number of categories within phoneme systems change, but little is understood about what drives and constrains this process. However, recent evolutionary models of language change that treat language as a complex dynamical system provide an exciting new set of testable hypotheses, based on the proposal that long-term language change is ultimately driven by biases operating at the level of individuals interacting with individual utterances. While there are potentially many such biases, we focus on a bias against lexical confusability as a model system. A bias against lexical confusability provides a particularly rich model because the lexicon of each language is distinct, with the result that predictions for individual languages are also distinct. Because of the large range of relevant time-scales, we use multiple approaches to test scale-specific hypotheses. In this talk, I will review (i) a cross-linguistic study showing that the probability of historical loss of a phoneme is inversely correlated with its role in disambiguating words, (ii) a corpus study of natural speech showing that phoneme tokens that play a larger role in lexical disambiguation are hyperarticulated, and (iii) a game-based laboratory model showing that phonemes become more or less hyperarticulated over the course of two days in correlation with their disambiguating role in the task. All these findings are predicted by the hypothesis that utterance-level micro-changes compound to produce macro-changes over generations.
Tuesday 17th, 11-12.30, Room 1.17 Dugald Stewart Building
Input variability as a mechanism for social structure determination of linguistic complexity
Mark Atkinson
Abstract: While non-linguistic features of a speech community have been shown to correlate with degrees of linguistic complexity, the explanatory mechanism(s) by which they could have such an influence has not yet been convincingly identified. Currently the most popular explanation appeals to the differences between adult and child learning. Larger languages, for example, are thought to attract higher proportions of non-native speakers, who then simplify their language through greater levels of analysis and learning error.
This is not the only possible explanation, however, and I will present the results from a recent experiment which finds support for the alternative candidate mechanism of input variability. 20 participants were trained on a morphologically-complex miniature language in two conditions: one in which aural input was provided by a single native speaker, and another in which they received their input from three speakers. Despite the training data being identical in both conditions, learners in the single-speaker condition demonstrated a more successful acquisition of the morphological system. As other aspects of the target language were learned equally well in both conditions, it does not appear that lower levels of input variability simply aid language acquisition in general.
This difference in performance between the conditions appears to be due to the participants learning their training data in systematically different ways. There are indications that the multiple-speaker condition promoted the more holistic learning of the training data utterances, so reducing the learners’ ability to generalise. I will describe the evidence for this and offer some possible explanations, before finally discussing some options for future research.
***NOTE UNUSUAL START TIME***
Tue 10th December, 11.30-1.00, B21 7 George Square
Functionalism and Its Discontents
Kevin Stadler
Functional explanations of language features have not only regained
appeal in the past decades by their reframing in evolutionary (read:
selectionist) terms, they’ve also recently been joined by newly
discovered relationships between linguistic features such as
grammatical and phonological complexity and non-linguistic features
such as the size of the speech community. Crucially, both functional
features in individual languages as well as probabilistic universals
(whether of the newfangled or the more traditional Greenbergian type),
while being synchronic in nature, require diachronic explanations to
go from description (or correlation) to explanation (or causation):
just because something is apparently functional does not mean we can
account for its existence or prevalence simply based on this function.
Functional explanations have a long history within the traditional
study of language change, and an equally long (and illustrious) list
of adversaries and fervent critics, from Jespersen over Weinreich et
al. and Lass to Haspelmath and back again. In this talk I want to
provide an overview of these critiques, both to raise awareness of the
problematic nature of selectionist explanations, but also to explore
possible ways out of the numerous functionalist traps.
Time allowing I will touch upon topics such as the actuation problem,
unmoved movers, Saussure’s firewall, ‘boring’ universals, frequency
effects, emergent grammar, process theories of language change and -
more fundamentally – the uneasy relationship between functional
accounts of language features and the arbitrariness of language.