Tuesday 26th May, 11:05-12:30, DSB 1.17
Properties of Hierarchically Primitive Languages
Geoffrey K. Pullum
PPLS, University of Edinburgh
There are families of formally definable `languages’ (sets of symbol strings) that have a descriptive and combinatorial complexity way down below the finite-state (FS) languages, which are commonly (but wrongly) taken to be the bottom of the hierarchy, a sort of baseline level of mathematical primitivity. In fact there are infinitely many infinite classes of stringsets that are proper subclasses of FS yet proper superclasses of the finite stringsets. And they have some linguistic interest. Any specialist in language emergence, prerequisites, learnability, or evolution should know something about them. I provide a tutorial on the properties of these low-complexity languages, and also argue that some of the psychological work on the pattern-learning abilities of animals would have been more relevant and valuable if it had been informed about this material.
Tuesday 19th May, 11:05-12:30, DSB 1.17
Carmen Saldana
Categorisation: the backbone of language
A cognitive function underlying language and its essential combinatorial power is a modified antecedent of a ubiquitous process of categorisation and extraction of similarities among vertebrates — a process that requires complex capacities such as induction, generalisation and abstraction. Far from being domain-specific then, cognitive mechanisms that underlie the components of language such as syntactic structure are based on processes of categorisation. These processes do not only include differentiation or discrimination but establishing interrelations amongst categories both paradigmatically and syntagmatically. Previous work in cultural evolution has shown how structure can emerge from the trade-off between learnability and expressivity in human communication systems. Nevertheless, the constructions obtained so far have been mono-categorical and have only included lexical items. The first objective in the work that will be presented is to increase the complexity of the culturally evolved systems in the laboratory towards basic sentential syntactic structure including different syntactic categories. By increasing the complexity of the meaning space to include basic motion events participants are forced to categorise more elements and they produce signals of higher complexity which resemble sentence structure. We ran two Iterated Artificial Language Learning experiments in the lab: the first one included iteration and an artificial pressure for expressivity; the second one, iteration and interaction. Multicategorical constructions emerged from both conditions with at least basic lexical categories being established in constructions with sentence structure. Both the artificial expressivity pressure and interaction enforce expressivity in the evolving languages, but we will show how interaction leads directly to languages optimising contrasts between meanings, whereas the artificial expressivity pressure leads to arbitrary addition of complexity to the form alone. The second and most important objective of the talk will be to discuss ideas for further work where I want look at the cognitive processes behind categorical changes in elements of linguistic systems through grammaticalization, in the evolution of surface grammatical categories such as adpositions.
Tuesday 12th May, 11:05-12:30, DSB 1.17
Nicolas Fay (The University of Western Australia)
Cognitive Biases Plus Social Interaction Drives the Adaptive Evolution
of Human Communication Systems.
Researchers have tended to focus either on the cognitive biases or the
social interactional processes that drive language evolution. Rarely
have these top-down and bottom-up approaches been combined. In this
talk I’ll present evidence that cognitive biases plus social
interactional processes drive the evolution of functionally adaptive
human communication systems. I’ll argue that a conservative
Egocentric-bias preserves sign variation by inhibiting the adoption of
signs produced by others during social interaction. An opportunistic
Content-bias encourages sign adoption on the basis of the intrinsic
qualities of the signs produced by others; if the sign encountered is
superior to previously used signs it is adopted. Together, an
Egocentric-bias, a Content-bias plus Social Interaction maximises the
chance that a population will converge on an optimal set of signs.
Congratulations to Dr Vanessa Ferdinand, who was awarded her PhD on 30th April. Vanessa’s PhD was supervised by Simon Kirby and Kenny Smith, and her thesis was titled “Inductive evolution: cognition, culture and regularity in language”. She is currently an Omidyar Fellow at the Santa Fe Institute.
Tues 28th April 11-12.30, Room 1.20, Dugald Stewart Building
Joe Fruehwald
Cohorts, Lifespans, and The Zeitgeist
I’ll be discussing three different kinds of time dimensions and how they relate to language change.
- Generational Time – Time defined in terms of generations, or birth cohorts, of speakers. An important component of language change is the differences between generations, independent from the speaker’s specific age, or the era in which they live.
- Lifespan – Time defined in terms of the lifespan of a speaker. The degree to which speakers remain stable in their linguistic system over their lifespan is a contested issue. This could be important if there is a characteristic age-linked pattern to which speakers participate in language changes.
- The Zeitgeist – Time defined in the normal way. To the extent that language change can be modeled as an innovation rapidly diffusing out to all current members of the speech community, the state of linguistic change would be best described in terms of the linguistic zeitgeist in the community at any point in time.
Current sociolinguistic approaches to language change treat (1), generational time, as the primary dimension to explore, with some noise factors introduced by (2), lifespan change. The linguistic Zeitgeist is essentially just ignored.
After coving a bit of background literature, I’ll walk through and attempt to tease a part these three dimensions, looking at three different language changes in the Philadelphia Neighborhood Corpus. The results show that the current sociolinguistic focus on generational change is well motivated. There is some marginal evidence for both Lifespan effects and Zeitgeist effects (the 80s was a weird time for /ow/), but otherwise generational time dominates the patterns observed.
Tues 21st April 11-12.30, Room 1.17, Dugald Stewart Building
Matt Spike
Rules and randomness: recognising and measuring the emergence of structure
Linguistic structure and possible explanations for its emergence are
common themes in evolutionary approaches to language. However,
probably because of the many different ways in which linguists employ
the word ‘structure’ (e.g. combinatorial, compositional, syntactic
etc.) and other related terms (e.g. complexity, construction,
pattern), it can be hard to pin down exactly which of these senses we
are trying to recognise as an emergent property. Along with this, we’d
also like a general measure of structure for application across both
experimental and modelling work.
I am going to contrast structural descriptions with structured
processes, and argue for the latter as a more intuitive and useful
conceptual tool. I’ll then take a quick look at how this applies to a
few well-known models and experiments, but will concentrate on using
experimental data from the Simon Game as a test-bed for various
candidate measures (entropy, n-gram entropy, Kolomogorov complexity,
statistical complexity, G-score). I’ll use this to propose that none
of the above measures are exactly what we’re looking for, but that we
can use a combination of approaches to look for traces of different
types of structured behaviour. Finally, I’ll have a more speculative
look at how the appearance of structure might precede it in actuality.
Tues 17th March 11-12.30, Room G32, 7 George Square
Hannah Little (Vrije Universiteit Brussel)
An extheremental investigation into modality and the emergence of structure
There is increasing evidence that physical aspects of a linguistic modality may affect the emergence of combinatorial structure. These effects may be because of the number of possible signal distinctions that can be made within different modalities, or because of an ability to recourse to iconicity within a modality, or because modality has an effect on the number of holistic signals that can exist before combinatorial structure becomes necessary. All of these possible hypotheses are not independent of each other, and are difficult to separate. Modality also will have a very significant effect on whether emerging structure will be characterised as combinatorial or compositional, or even something which lies outside of this binary.
How the size and topology of a modality’s signalling space interfaces with iconicity and structure is an important phenomena to understand, especially within a field extrapolating results created using artificial signal space proxies to language generally. If the topology of an artificial proxy can affect the structure that emerges using that proxy, then it is important to consider these as effects of modality, and how a modality interfaces with cognition. We must identify and understand these effects before we start attributing emerging structure purely to cognitive or functional mechanisms. I will outline experiments looking at how the size and dimensionality of a continuous signalling space might affect the discretisation and emergence of structure within signals. In the most recent experiments, participants generate signals using an infrared “Leap Motion” sensor (or “theremin”). I will discuss the following experiments (or “extherements”):
1. An investigation into how participants manage a mismatch in dimensionality between the signalling space and a meaning space, where I found evidence for iconicity where the dimensionality of signal and meaning spaces matches, but not with mismatches. I have also worked with collaborators (Kerem Eryılmaz and Bart de Boer) to develop measures for structure (and iconicity) within continuous signals, which I will briefly outline and present with reference to my results.
2. A further (ongoing) investigation into dimensionality mismatches, but with more dramatic mismatches, and a bottleneck on the number of meanings participants see within the experiment to encourage the adoption of more structural strategies.
3. An investigation into whether the ability to recourse to relative iconicity (by having continuous signal dimensions map to continuous meaning dimensions) may inhibit the emergence of combinatorial structure, or facilitate it where meanings do not differ continuously. Within this experiment, I found evidence that iconicity may be maladaptive for the discrimination of signals.
Further to the above, I will discuss how we can characterise the type of structure (combinatorial or compositional) that we see emerging within these experiments, referring to how that structure relates to the meaning space, and how the participants perceive the relation. Importantly, the binary nature of linguistic structure may be a false dichotomy, and modality may have affected how the different levels of linguistic structure initially emerged and coevolved.
I will also outline the design of my ongoing social coordination experiments, and future ideas.
Tues 10th March 11-12.30, Room G32, 7 George Square
Richard Blythe
Emergence of Typological Universals
Apparently equivalent variants of a linguistic variable (e.g. the different basic word orders) have a non-uniform distribution across the world’s languages, thereby suggestive of some non-equivalence among them. Various cognitive principles that apply during language acquisition or use are proposed as explanations for the prevalence of some variants over others, some of which have been observed in artificial language learning experiments. In this talk I investigate whether typological distributions and the course of language change in populations of speakers can be explained entirely in terms of biases in language learning that apply universally to all speakers in all speech communities, or whether some additional culture-specific factors are needed. I will suggest that evidence from grammaticalisation cycles of definite and indefinite articles point towards the latter conclusion: in particular, models that exclude culture-specific factors (combined with uncertainty in the strength of universal biases) predict a strong effect of speech community size on the dynamics of language change which is not apparently a feature of the empirical data for articles. On the other hand, models where learning biases work in concert with culture-specific biases seem to avoid this effect.
Tues 3rd March 11-12.30, Room G32, 7 George Square
Kenny Smith
Dendrophilia versus simplicity in sequence learning
Fitch (2014) presents the Dendrophilia Hypothesis, namely that “Humans have a multi-domain capacity and proclivity to infer tree structures from strings, to a degree that is difficult or impossible for most non-human animal species”. I think this is a really interesting idea, but I have the vague idea (to be fleshed out a little more in this talk) that it’s not really consistent with the existing experimental evidence from artificial language learning, which suggests that humans have real difficulties learning miniature languages which one might expect to be easy were Dendrophilia Hypothesis true as stated. I’ll try to present some of that evidence, together with some experimental work with artificial grammar learning in non-humans which suggests a slightly different hypothesis: all species prefer to infer simple patterns from strings, and humans are simply slightly less constrained than other primates in this regard. I think this accounts for the same observations as the Dendrophilia Hypothesis, but differs slightly in emphasis (emphasising reduced constraints on learning in humans, rather than a species-unique capacity) and highlights cognitive continuity between humans and other primates (maybe humans are only quantitatively different from other primates, rather than qualitatively different).
Tuesday 10th Feb, 11am-12:30, G32, 7 George Square
Yasamin Motamedi
The emergence of systematic structure in artificial gestural communication systems
Languages exhibit systematic structure: signals are not independent of each other but form part of a system. Previous work has shown that the emergence of systematic structure increases learnability of a system, and the pressures of transmission drive this emergence (Kirby, Cornish and Smith, 2008; Cornish, Smith and Kirby, 2013). Additionally, a link has been posited a between systematic structure and the arbitrariness of a systems signals; that is, as signs become more arbitrary, the systematic re-use of signs increases, aiding learnability (Theisen, Oberlander and Kirby,2010; Theisen-White, Kirby and Oberlander, 2011).
In this talk I will present a study that looks at how systematic structure arises in manual communication systems. I will present results that suggest the gradual emergence of systematic structure in the systems participants created. I will present analysis that attempts to understand which aspects of the sign are affected as systematic structure emerges, relating the results to structures found in natural sign languages.