17 January: Andy Wedel

Signal evolution within the word

Andy Wedel (University of Arizona)

Tuesday 17 January 2017, 11:00–12:30
1.17 Dugald Stewart Building

Languages have been shown to optimize their lexicons over time with respect to the amount of signal allocated to words relative to their informativity: words that are on average less predictable in context tend to be longer, while those that are on average more predictable tend to be shorter (Piantadosi et al 2011, cf. Zipf 1935). Further, psycholinguistic research has shown that listeners are able to incrementally process words as they are heard, progressively updating inferences about what word is intended as the phonetic signal unfolds in time. As a consequence, phonetic cues early in the signal for a word are more informative about word-identity because they are less constrained by previous segmental context. This suggests that languages should not only optimize the total amount of signal allocated to different words, but optimize the distribution of that information across the word. Specifically, words that are on average less predictable in context should preferentially target highly informative phonetic cues early in the word, while preserving a ‘long tail’ of redundant cues later in the word. In this talk I will review recent evidence that this is the case in English. Further, languages show a strong tendency to develop phonological patterns which support phonetic cue informativity at the beginnings of words, while reduce cue informativity later in words. I will argue that this typological tendency plausibly arises from this word-level phenomenon.

17 January: Kevin Stadler (pre-viva talk)

Direction and directedness in language change: An evolutionary model of selection by trend-amplification

Kevin Stadler (MPI Nijmegen)

Tuesday 17 January 2017, 15:00–15:30
3.11 Dugald Stewart Building

All human languages are constantly undergoing change. Linguistic conventions whose social and communicative meaning are understood by all speakers are gradually altered or replaced, whether by changing their forms, meanings, or by the loss of or introduction of altogether new distinctions. How do large speech communities go about re-negotiating arbitrary associations in the absence of centralised coordination?

This thesis first provides an overview of the plethora of explanations that have been given for language change. Approaching language change in a quantitative and evolutionary framework, mathematical and computational modelling is put forward as a tool to investigate and compare these different accounts and pressures in a rigorous fashion. The central part of the thesis investigates a relatively recent addition to the pool of mechanisms that have been proposed to influence language change: I will compare previous accounts with a momentum-based selection account of language change, a replicator-neutral model where the popularity of a variant is modulated by its momentum, i.e. its change in frequency of use in the recent past. I will discuss results from a multi-agent model which show that the dynamics of a trend-amplifying mechanism like this are characteristic of language change, in particular by exhibiting spontaneously generated s-shaped transitions. I will also discuss several empirical predictions made by a momentum-based selection account which contrast with those that can be derived from other accounts of language change.

Complementing theoretical arguments for the role of trends in language change, I will go on to present fieldwork data of speakers’ awareness of ongoing syntactic changes in the Shetland dialect of Scots. Data collected using a novel questionnaire methodology show that individuals possess explicit knowledge about the direction as well as current progression of ongoing changes, even for grammatical structures which are very low in frequency. These results complement previous experimental evidence which showed that individuals both possess and make use of implicit knowledge about age-dependent usage differences during ongoing sound changes.

Echoing the literature on evolutionary approaches to language change, the final part of the thesis stresses the importance of explicitly situating different pressures either in the domain of the innovation of new or else the selection of existing variants. Based on a modification of the Wright-Fisher model from population genetics, I will argue that trend-amplification selection mechanisms provide predictions that neatly match empirical facts, both in terms of the diachronic dynamics of language change, as well as in terms of the synchronic distribution of linguistic traits that we find in the world.

13 December: Marieke Woensdregt

Cultural selection vs. cultural attraction: What’s the debate?

Marieke Woensdregt (Edinburgh)

Tuesday 13 December 2016, 11:00–12:30
1.17 Dugald Stewart Building

The idea of this session is to have a bit of a mix between a talk and a paper discussion. You won’t be required to read any paper though!

The talk part will consist of me summarising a debate about the similarities and differences between a so-called ‘Darwinian’ approach to cultural evolution (work by e.g. Boyd, Richerson and Henrich) and a ‘cultural attraction’ approach (work by e.g. Sperber and Claidière). The Darwinian approach to cultural evolution is often characterised as locating the driving force of cultural change in selection among variants that are transmitted fairly faithfully from generation to generation. The cultural attraction approach on the other hand focuses on how the process of transmission itself can drive cultural change: by closely replicating certain variants while transforming others. Where the first approach relies on rather faithful ‘copying’ or imitation of variants, the second assumes reconstruction of variants each time they are transmitted.

I will discuss some of the theoretical views on whether this difference in approach is based on a real disagreement about how cultural evolution works, or whether it can be characterised as simply a difference in granularity, object of study or definition. After setting out this debate, including the attempts that have been made to reconcile the two approaches, we will move into our own discussion – focusing on how all this relates to our own work -, for which I’ll provide some discussion points.

If you do want to read up a little before coming, I’d advise this paper by Acerbi & Mesoudi (mainly the first 8 pages):

Acerbi, A., & Mesoudi, A. (2015). If we are all cultural Darwinians what’s the fuss about? Clarifying recent disagreements in the field of cultural evolution. Biology and Philosophy, 30, 481–503. doi:10.1007/s10539-015-9490-2

22 November: Bart Geurts

“The soul converses with herself”: Making sense of self talk

Bart Geurts (University of Nijmegen)

Tuesday 22 November 2016, 11:00–12:30
1.17 Dugald Stewart Building

People talk a lot. Much of that talk is addressed at others, but people talk to themselves as well, and though self talk has long been recognised as a key facet of our mental lives, its workings remain poorly understood. I argue that the phenomenon of self talk has profound consequences for our understanding of human communication, as well as the ways in which human cognition and cognitive development are grounded in communicative competence, and I present an account of communication which explains how that is possible.

15 November: Richard Blythe

Grammatical change in populations

Richard Blythe (Edinburgh)

Tuesday 15 November 2016, 11:00–12:30
1.17 Dugald Stewart Building

Language change is an inherently multiscale phenomenon, with structure and change at the population scale originating in the linguistic behaviour of individual speakers. Studies conducted separately at the two levels show certain parallels, for example, that individual biases observed in artificial language learning experiments tend to favour the structures that are more widely observed across the world’s languages. However, these can interact with a whole host of other factors, such as the communicative goals of the speaker, the human perceptual system, social network structure and so on, any or all of which could shape language structure and change on the path from the individual to the population. In this talk I will introduce an agent-based model that allows a wide range of factors to be included at the individual level and attempt to establish the relative importance of these factors in generating population-level changes that are consistent with the history of article grammaticalisation cycles.

8 November: Carmen Saldana

Level-specific learning biases in the regularisation of unconditioned variation: The case of morphological inflection and word order

Carmen Saldana (Edinburgh)

Tuesday 8 November 2016, 11:00–12:30
1.17 Dugald Stewart Building

Human languages contain very little unconditioned variation but occasionally, language learners are exposed to input that contains inconsistencies. Learners under those circumstances tend to reduce or remove such inconsistencies, in other words, they regularise their input. The product of such regularisation is a language that better conforms to the learner’s cognitive biases, i.e. easier to learn and use.

Regularisation behaviour has been documented extensively in natural language in different contexts (e.g. language acquisition and development, language change, and in emerging languages) and at different levels (e.g. morphology, syntax and the lexicon). Evidence from Creole typology suggests that in stages of observed pidginization followed by nativisation, where the pidgin system is highly inconsistent and linguistic performance is limited, linguistic levels might behave differently: morphological complex traits such as inflectional morphology seem to be highly simplified whilst syntactic traits such as word order tend to reproduce the input complexity more closely.

There is a vast literature on Artificial Language Learning on regularisation behaviour on various different linguistic units at different linguistic levels. Nevertheless no systematic comparison across linguistic levels has been published. In this talk we present an artificial language learning experiment that constitutes the first attempt to compare the effect of linguistic level (morphological inflection vs. word order) on regularisation behaviour. Adult learners in our experiment are exposed to a miniature artificial language featuring an inconsistent mixture of synonymous variants. This mix of variants models the uncertainty in the input to which learners are exposed to in periods of language change or in language formation. Not only are we interested to see how learners shift the input languages to languages that better conform to their biases but also to explore the extent to which that shift is comparable across linguistic levels. We therefore use this paradigm to test regularisation across levels with comparable complexities as well as its level-specific strength.

Our results do not comply with the creole typology predictions and word order seems to be more regularised than morphological inflection under certain circumstances and equally so under others. We will show that level-specific learning biases are at play in the acquisition of morphology and word order and will discuss how they affect the differences in regularisation behaviour obtained in our study.

1 November: Jasmeen Kanwal

Language-users shorten words in more predictive contexts when communicating in a miniature artificial language

Jasmeen Kanwal (Edinburgh)

Tuesday 1 November 2016, 11:00–12:30
1.17 Dugald Stewart Building

Zipf’s Law of Abbreviation, the observation that word length tends to be inversely proportional to word frequency, has been confirmed to hold across a wide range of human languages. However, a corpus study by Piantadosi et al. (2011) shows that, for many of these languages, there is in fact a stronger inverse relationship between a word’s length and its predictability in context. Behavioural experiments have shown that predictability in context can affect utterance length in terms of whether phonetic reduction occurs (Gahl and Garnsey 2004), and whether overt morphological (Fedzechkina et al. 2012) or syntactic (Jaeger 2010) markers are used. One study (Mahowald et al. 2013) directly addresses the question of whether predictability in context modulates word length at the level of the lexicon. However, this study is fraught with issues, which I will outline in the talk. I will report the results of a new miniature artificial language learning experiment designed to explicitly test the link between constraints on individual-level behaviour during communication and the structure of the lexicon. We show that, when predictability in context is varied across stimuli rather than bare frequency, language-users shorten words in more predictive contexts, but only when the pressures to communicate both accurately and efficiently are present. I will discuss how these results relate to the burgeoning literature on Uniform Information Density. Finally, I will present some possible ideas for a follow-up experiment.

25 October: Matt Spike (pre-viva talk)

Minimal requirements for the cultural evolution of language

Matt Spike (Edinburgh)

Tuesday 25 October 2016, 10:00–10:30
S38 7 George Square

Human language is a product of both cognition and culture. Any evolutionary account of language, then, must address both biological and cultural evolution. I look at how a cultural evolutionary perspective can shed light on two main questions: firstly, how do working systems of learned communication arise in interacting populations? Secondly, how do human communication systems take on their characteristic duality of patterning, i.e. systematicity at both a meaningless and meaningful level? A large, multi-disciplinary literature exists for each question, full of apparently conflicting results and analyses. I survey this work, find the commonalities, and tie them together to propose a minimal account of the cultural evolution of language.

18 October: Marieke Woensdregt

The co-evolution of language and mindreading

Marieke Woensdregt (Edinburgh)

Tuesday 18 October 2016, 11:00–12:30
1.17 Dugald Stewart Building

The hypothesis that mindreading (theory of mind) and language have facilitated or pushed for each other’s evolution has been put forward in several different ways on several different grounds. The evolution of explicit mindreading skills may have been important for the evolution of language because it allows us to express and recognize communicative intentions. The evolution of linguistic conventions may have been important for the evolution of mindreading because it allows us to make our mental states explicit, and transmit our understanding of mental states to others. The same arguments have been made on the developmental timescale – based on studies correlating mindreading skills with word learning ability, and studies showing that certain linguistic input can facilitate the development of mindreading.

In this talk I will present an agent-based model that explores under which circumstances language and mindreading can co-evolve – where language is implemented as lexicon-learning, and mindreading as the ability to learn another agent’s perspective on the world. Agents can learn both the lexicon and the perspective of another agent through Bayesian inference, but these two skills crucially bootstrap each other during development – meaning that one skill could not be learned if there was no ability to learn the other.

Over the course of iterated learning this co-development results in interesting evolutionary dynamics. Populations of agents that develop in this way are only able to establish an informative language if there is either a strong pressure in favour of expressivity, or a weak pressure that favours successful perspective-takers as cultural parents. Given such a weak pressure in favour of perspective-taking, an informative language can be established even when learners have a strong cognitive bias in favour of simple (compressible) languages. I will discuss these simulation results in the light of theories about the evolution of Gricean communication and social coordination.

14 October: Rachael Bailes (pre-viva talk)

An adaptationist psycholinguistic approach to the pragmatics of reference

Rachael Bailes (Edinburgh)

Tuesday 14 October 2016, 10:00–10:30
1.17 Dugald Stewart Building

This thesis formalises two broadly polarised approaches to the question of context integration in linguistic comprehension, and makes explicit the adaptationist particulars that each mechanistic account may imply. The Social Adaptation Hypothesis (SAH) holds that linguistic comprehension is performed by relevance-oriented inferential mechanisms that have been selected for by a social environment (i.e. inference-using conspecifics). In particular, the SAH holds that linguistic conventions are attended to in the same way as other ostensive stimuli, and comprehended on the basis of contextual information. The Linguistic Adaptation Hypothesis (LAH) holds that linguistic comprehension is performed by specialised cognition that has been selected for by a cultural, linguistic environment (i.e. language-using conspecifics). The LAH holds that linguistic conventions may constitute a privileged domain of input for the comprehension system, and in particular that the nature of linguistic representations can support comprehension without mediation by inferential cognition or contextual integration. The remainder of the thesis investigates referential comprehension with a series of four reaction time experiments, using a conversational precedent paradigm, relevant to the contrastive predictions of these two adaptationist accounts. Two additional production experiments measured the effect of visual context on whether speakers maintained their linguistic precedents. The broad question that covers all of these experiments is: how sensitive is the comprehension process to linguistic input qua linguistic input, relative to various other grades of contextual information? In light of the evidence presented and its limitations, I conclude that there is empirical support for the LAH, and that this alternative account should be considered as part of the unfolding conversation on pragmatics in evolutionary linguistics. More broadly, the thesis attempts to demonstrate that psycholinguistic processing is a valuable object of study for evolutionary linguistics, and that evolutionary theory can a useful conceptual tool in psycholinguistics.