Jennifer Culbertson has joined the LEC as a Chancellor’s Fellow. Jenny works on the relationship between biases in language learning biases and typological universals, using artificial language learning and computational modelling techniques. We are very excited to have her on the team – welcome aboard Jenny!
Tue 8th July, 11.00-12.30, 1.17 Dugald Stewart Building
Monica Tamariz
Cultural transmission: behaviours replicate; mental and material culture emerge
Human culture evolves, but it is also remarkably stable over time. In order to explain this stability, we need to understand how cultural information is transmitted from generation to generation. Human culture has been defined as mental information such as knowledge, beliefs, skills, values and attitudes, which is socially transmitted to new learners. However, empirically attested human social learning mechanisms such as imitation and over-imitation do not explain the transmission of mental information, but of behaviour. In order to bridge the explanatory gap between the imitation of behaviour and the stability of mental culture, I propose an evolutionary account of cultural transmission encompassing two distinct mechanisms. First, behavioural cultural patterns are replicated by new learners through the mechanisms of imitation and over-imitation as well as teaching. Second, mental (and material) representations are emergent phenomena arising in individual’s minds through the production of myriads of (replicated) behavioural patterns in context. The replication of behaviour therefore ultimately helps ensure the stability and constrains the variation of mental (and material) culture.
*** two LEC talks next week, note slightly unusual starting time for both ***
Tue 17th June, 11.30-13.00, 1.17 Dugald Stewart Building
James Winters
Heuristic-driven strategies in communication games and their implications for the emergence of structure in language
So far, both communication game and iterated learning experiments have focused on macro-level properties and their transitions. This normally involves how we go from one system state (e.g., an initially random holistic language) to another system (e.g., systematic compositional languages). Less attention has been paid to the individual, local interactions in these experiments, such as usage events in communication games, and how they give rise to different types of structure. One useful avenue of exploration is consider the role of heuristics: experience-based techniques for problem solving, learning, and discovery. With this in mind, the first part of the talk will consider four very basic heuristics participants can employ in communication game experiments: copying, reuse, modification, and invention. By considering how these heuristics are influenced by different stressors, such as memory, learning bottlenecks and the communicative context, we can make predictions on the likelihood of transitioning from one system state to another. This leads into the second half of the talk where I will present ongoing work into an experimental test of one of these transitions: going from an initially compositional system to an underspecified language. Time permitting I will also discuss plans for future experiments.
Tue 17th June, 14.00-15.30, 1.17 Dugald Stewart Building
Simon Kirby
Learning from learners: how social structure can shape linguistic structure
The cultural evolution of the basic design features of human language can be observed directly in the case of emerging sign languages. In particular, there are two broad contexts in which sign languages emerge: the deaf-community sign context, and the village sign context. The former involves deaf children being brought together in schools for the deaf, whereas the latter arises in typically rural communities with high rates of congenital deafness. Recent work in a particular sign language of the latter type, Al-Sayyid Bedouin Sign, reveals the evolution of a language lacking a fundamental design feature: duality of patterning.
In this talk, I will show using computational simulations (both Bayesian and connectionist) that we should expect the social context of an emerging language to have important consequences for the origins of structural design features such as duality of patterning. In particular, I will demonstrate that it is the school context that promotes the rapid evolution of systematicity at all levels in an emerging language.
I will also discuss the extent to which the very same underlying mechanisms can explain the apparent population-size/complexity correlations that have been found in the world’s spoken languages. I will tentatively conclude that it is not population size as such that is important for the emergence of systematicity, but rather the extent to which learners learn from other learners.
*** NOTE UNUSUAL DAY AND TIME ***
Mon 9th June, 14.00-15.30, 1.17 Dugald Stewart Building
Bart de Boer (Vrije Universiteit Brussel)
Cognitive adaptations to rapidly changing culture can evolve
This talk contributes to the study of what kind of cognitive adaptations to rapidly changing cultural traits can evolve. It introduces a mathematical approach (mean field approximation) and applies this to a very simple (evolutionary dynamics) model of co-evolution between biologically determined cognitive biases and cultural traits. Although the model is inspired by Christiansen, Reali and Chater’s model, it finds (in contrast to their results) that under a range of realistic assumptions about fitness and the interaction between biology and culture, cognitive adaptations to culturally changing traits can evolve. From the mathematical analysis, precise conditions under which this happens can be derived.
Although the material in the talk is mathematical in nature, I will make an attempt to present it so that it can be understood with only basic secondary school mathematics.
Tue 20th May, 11.00-12.30, 1.17 Dugald Stewart Building
Vanessa Ferdinand
What drives regularity in human language?
Languages evolve as they pass through generations of learners’ minds and adapt to selection pressures exerted by our cognitive architecture and learning biases. In this talk, I focus on one such pressure: our linguistic regularization bias, which drives learners to eliminate free variation in language. The nature of linguistic regularization is a subject of ongoing debate: some argue for domain-general causes such as memory limitations (Hudson Kam & Chang, 2009) and others argue for biases specific to the linguistic domain (Bickerton, 1984; Becker & Veenstra, 2003). I will present the results of a lexical frequency learning experiment which show that both domain-specific and domain-general factors independently contribute to regularization in adult learners. The domain-general component is then further explored in additional experimental conditions and analyses regarding concurrent frequency learning, cognitive load, and participant expectations / task framing. These results tell us how much regularity we expect to see after one generation of learning due to different components of our regularization bias. But how do these components contribute to the overall regularity of language? To address this question, participant behavior is extrapolated forward into evolutionary time (assuming the output of one learner becomes the input for the next learner). Using information-theoretic notions of similarity, I show that after one generation, participant behavior due to domain-specific regularization encodes more information about linguistic regularity, whereas after several generations of learners, behavior due to domain-general regularization encodes more. This raises important issues about the mapping of individual biases onto distributional features of human languages and demonstrates that evolutionary analyses provide more information about this mapping than the results of single-generation data alone.
Tue 13th May, 11.00-12.30, 1.17 Dugald Stewart Building
Carmen Saldaña
À la recherche des primitives, perdus: emergence and evolution of categories in syntax
Ever since Chomsky’s Three models for the description of language (1956), the field of linguistics has taken hierarchical processing and the underlying implementation of recursion to be elements of focus in the explanation of human language uniqueness. The minimal assumption in hierarchical processing is that languages are not syntactically monocategorical. Syntactic categories have been taken for granted to be primitives for derivations but up to date the field of linguistics is pretty much in the dark even for a theory of basic lexical categories (Baker, 2003). This theoretical void has been overlooked by the focus on labelling algorithms (Hornstein, Nunes & Grohmann, 2005; Berwick, Friederici, Chomsky & Bolhuis, 2013) in theoretical linguistics and the consequent black boxing of lexical items and the locus of asymmetry. The project I will present aims to engage in a truly minimal effort that tries to explain the emergence of functional morphology and different lexical syntactic categories through cultural evolution and therefore detach them from what has been referred to as narrow syntax. By doing so, we are committing to a strong theory of phases that takes category-less conceptual primitives as the basic input for derivation (bottom-up). We will then point to Edge-Features/combinatoriality and cumulative culture as locus of uniqueness in human language that allows our species to construct categories in the linguistic system, lexicalise, grammaticalise and more generally develop and acquire language. In order to test it empirically, we designed an experiment within the Iterated Learning Model framework (Kirby & Hurdford, 2002; Kirby, Cornish & Smith, 2008). Results support our hypotheses that cultural transmission amongst humans is enough to show the evolution of systemic syntactic categories. Because arguing for the emergence of categories from this precise dynamical complex systems point of view is arguing for a sufficiently general theory of form, I will also start to explore the relation between visual cognition following and language for the mechanisms behind visual cognition are a very good basis for conceptual primitives and exploring the dynamics of the construction of categories by the modern human mind.
Tue 6th May, 11.00-12.30, 1.17 Dugald Stewart Building
Monica Tamariz
Evidence for selection in the evolution of human communication systems
Human communication systems evolve, but the cultural evolutionary mechanisms that drive their evolution are not well understood. To better understand them, we constructed a parameterized mixed probabilistic model of the spread of communicative variants in four 8-person laboratory micro-societies engaged in a simple communication game. The parameters weight the contributions of selection pressures (Conformity bias and Content bias) against a neutral (Drift) evolutionary model. Drift alone explains the spread of communication variants in a minority of cases. Selection models, working in combination, explain the majority of the empirical data. The best fitting parameter setting includes an egocentric bias and a content-bias, suggesting that participants retain their own previously used communicative variants unless they encounter a superior variant, in which case the latter is adopted. This novel pattern of results suggests that 1) a theory of the cultural evolution of human communication systems must integrate drift and also selection dynamics and that 2) human communication systems are functionally adaptive complex systems.
Tue 22nd April, 11.00-12.30, room 1.17 Dugald Stewart Building
Sander Lestrade (Radboud University Nijmegen)
Modeling (the development of) argument-marking systems
A fundamental task for language is to provide the rules to map meaning to form. In the communication of an event, these rules should link semantic roles, such as agent and patient, to their grammatical functions, such as subject and object. Despite their importance for communication, grammatical strategies of argument marking (viz. word order, head marking, and dependent marking) are surprisingly limited, complex, and redundant. Why do languages not use a simple and straightforward means of encoding who did what to whom?
The starting hypothesis of my project “The exaptation of argument marking” is that natural languages in fact do not have dedicated means to map semantic roles to syntactic functions. Although some constructions are put to use for grammatical marking indeed, they originally evolved for other usages. From this, complex systems have developed in which interacting strategies jointly cover the meaning space. By explicitly taking into account the developmental history of the different strategies, it will be possible to provide a much deeper understanding of the constant rise and fall of grammatical argument-marking strategies in the languages of the world.
In this talk I’ll present the ideas, plans, problems and outcomes of this 3-year research project that I’ve just begun. I will focus on modeling issues.
The final session of dry runs for the Evolang conference – everyone welcome, titles below!
Tue 8th April, 11.00-12.30, Room 1.17 Dugald Stewart Building
Jon Carr: The cumulative cultural evolution of category structure in an infinite meaning space
Justin Sulik: Symbolisation and cognition
Bill Thompson & Catriona Silvey: The effect of communication on category structure
Three more dry runs for the Evolang conference – everyone welcome, titles below!
Tue 1st April, 11.00-12.30, Room G.04, David Hume Tower (DHT Conference Room)
Simon Kirby, Hannah Cornish, Kenny Smith: Systems emerge: The cultural evolution of interdependent sequential behaviours in the lab
Matt Spike: Minimal requirements for the emergence of learned signalling
Kevin Stadler: Momentum-based language change: a non-adaptive model of directional selection