Author: Andres Karjus

March 31: Andres Karjus

Changing communicative needs and the structure of the lexicon: corpus-based and experimental approaches

Andres Karjus, CLE, University of Edinburgh

Tuesday, 31.03.2020
11:00 – 11:40
Room: [virtual Zoom talk]

Language change can be driven by a number of things, including social factors, evolutionary learning and communication biases, top-down language planning, or just random drift. In this talk, I will discuss the role of changing communicative needs in lexical change. The corpus-based component focuses on lexical competition and colexification. Here, competition refers to a situation where two or more words compete to be used in the same meaning space. Colexification refers to a word being used to refer to two or more meanings that are or could be otherwise expressed by individual words. I demonstrate how both processes, as well as changes in communicative need, can be detected and measured in diachronic corpora, using language-agnostic, unsupervised lexicostatistical approaches supported by word embeddings. Finally, I will discuss a work in progress artificial language experiment designed to investigate the effects of sense similarity and communicative need on colexification preferences.

slides | recording

February 18: Mark Atkinson

Inference, perspective taking, and cumulative cultural evolution

Mark Atkinson (University of Stirling)

Tuesday, February 18
11:00am – 12:30pm
G32, 7 George Square

Many human cultural traits become increasingly beneficial to their users as they are repeatedly transmitted, thanks to an accumulation of modifications made by successive generations. Learners typically modify the traits they observe in ways which make them more beneficial. But how do learners do this given that cumulative cultural evolution produces behaviours which are increasingly improbably as individual discoveries? And despite information bottlenecks which restrict the learner’s exposure to previously sampled behaviours?

I’ll present two investigations looking into how the effect of these bottlenecks can be mitigated. Firstly, in a series of experiments involving adult participants, we assess whether learners are (a) sensitive to cues of non-random production in the restricted information they observe, and (b) able to use those cues to infer and avoid other behaviours (behaviours the learner was not directly exposed to) which have already been sampled and rejected. Secondly, using adult participants, we assess the effect of the demonstrator intentionally selecting the restricted information to be observed by the learner. A follow up study with 5- to 10-year-old children then investigates how this intentional information sharing changes over development.

February 4: Mora Maldonado and Jennifer Culbertson

Ain’t no sunshine. A brief excursion into negative dependencies. 

Mora Maldonado and Jennifer Culbertson (CLE, University of Edinburgh)

Tuesday, February 4

11:30am – 12:30pm
Room G32, 7 George Square 

Languages vary quite a bit in how they express negation. Many languages use negative markers, such as English not, or negative indefinites such as nobody. In some languages, each of these negative elements clearly contribute one semantic negation (e.g. Dutch); in other languages (e.g. Spanish), negative elements seem to agree with each other when they appear together in a sentence, showing what is known as Negative Concord. The cross-linguistic variation in the behaviour of negative indefinites has drawn a lot of attention in the linguistics literature. In this talk, we will review the literature and propose some ideas for future experiments.

January 28: Fausto Carcassi

The role of learning in the evolution of semantic extremeness

Fausto Carcassi (CLE, University of Edinburgh)
Tuesday, January 28
11:30am – 12:30pm
Room S38, 7 George Square

Some words in natural language express categories that are extreme in the sense that they cover only an extremum of the underlying scale. For instance, the absolute adjective “full” expresses a category covering only the maximum of the scale of fullness. The evolution of extremeness poses an interesting puzzle, because extreme categories seem prima facie suboptimal for communication and difficult to learn. Some explanations of how extreme categories might come about have been proposed in the literature (Franke (2012), Lassiter & Goodman (2013), Qing & Franke (2015)). These previous accounts focus on the role of communication in the evolution of extremeness. We present work attempting to elucidate the role of learning in the evolution of extreme categories. We show in an IL model that even in the absence of a bias and pressure for communicative accuracy, extreme categories can evolve often.

January 21: Kenny Smith

Spoken iterated learning

Kenny Smith (CLE, University of Edinburgh)

Tuesday, January 21
11:30am – 12:30pm
Room S38, 7 George Square

We often run iterated learning experiments using written language, largely as a methodological convenience – participants are trained on text labels for objects or events, then they produce new text labels which we use as training data for the next participant in a chain of transmission. We have run similar experiments using gesture, but seldom using speech. Since written text might engage different learning and/or processing strategies than spoken language (and sign might be different again) there’s a potential concern that some of what we regard as core results might be dependent on our use of written text; we might also be missing out on studying interesting design features of natural language that represent adaptations to the communication of propositional structures over a rapid-fading serial channel like speech. I’ll present some preliminary results from two studies. In the first study we are checking that one of our core results, that structure emerges as a trade-off between learning and communication, still comes out in a conceptual replication in the spoken modality. In the second study we manipulate properties of the channel that participants communicate through (we filter out low and high frequencies) to see if this affects the contrasts employed in the languages that develop, specifically whether they adapt to avoid hard-to-hear cues and capitalise instead on the intact frequency regions.

December 10: Simon Kirby

Cumulative cultural evolution of systematic structure and grammatical complexity in humans and baboons

Simon Kirby (CLE, University of Edinburgh)

Tuesday, December 10
11:30am – 12:30pm
DSB, room 1.17

In this talk, I will present new analyses of some old iterated sequence learning experiments in humans and a newer version of the same experiment run with Nicolas Claidiere, Kenny Smith and Joel Fagot on a population of captive baboons. Mostly, I have been trying to figure out how to test the intuition that in these experiments cultural evolution is both making sequences simpler and making them more complex depending on what you think those terms mean. I’ll try and show that we can use different algorithms for compressing the data from these experiments to capture this intuition, and in doing so reveal that humans and baboons both exhibit cumulative cultural evolution, but in importantly different ways. I hope to convince you that this captures something fundamental about the products of cultural evolution including human language.

November 19: Asha Sato

The role of expectation and experience in the perception of iconicity

Asha Sato (CLE, University of Edinburgh)

Tuesday, November 19
11:30am – 12:30pm
Room G32, 7 George Square

Iconicity (a perceived resemblance between a communicative form and meaning) is generally accepted to be a subjective phenomenon: people do not always agree whether (or how) a particular item is iconic. One of the things that might influence how potentially iconic items are construed is previous experience and expectations produced by that experience (with one’s own language, or with iconicity). I will present an experiment using a novel artificial language modality based on point patterns, designed to test whether we can induce an expectation for iconicity in the lab and whether that expectation affects the perception of iconicity in new items. We find that participants trained on an iconic language are more likely to select an iconic label for novel ambiguous items than participants trained on an arbitrary language.

November 12: Jennifer Culbertson and Alex Martin

Adventures in Tharaka: A report on ongoing work from our ESRC project on word and morpheme order

Jennifer Culbertson and Alex Martin (CLE, University of Edinburgh)

Tuesday, November 12
11:30am – 12:30pm
Room G32, 7 George Square

 

In this talk I will present recent results from two strands of recent work involving speakers of Kîîtharaka, a Bantu language spoken in Tharaka-Nithi county, Kenya. This language has two properties of interest for the study of typological universals. First, its noun phrase word order is N-Dem-Num-Adj, an order which in our terms is “non-homomorphic”. This means that the (linear) order fails to transparently reflect what is typically hypothesised to be the underlying hierarchical structure in the NP, namely [Dem [Num [Adj [ N ] ] ] ]. This is unusual: the majority of the world’s languages have homomorphic NP orders. The main focus of the project is to test whether learners are biased in favour of homomorphic orders, therefore Kîîtharaka speakers are a prime test population. We will discuss the work we’ve done so far investigating homomorphism, and what we’ve just gotten start in Tharaka. The second properties of interest is that Kîîtharaka, like other Bantu languages, is predominantly prefixing (i.e. morphological affixes precede the stem). This is again in contrast to the majority of languages, which are primarily or exclusively suffixing. We report the results of an experiment testing whether Kîîtharaka speakers show the same suffixing preference that has been claimed on the basis of experimental data from English-speaking participants.

November 5: Fang Wang

Typology reflects learning biases for Cross-Category Harmony

Fang Wang (CLE, University of Edinburgh)

Tuesday, November 5, 11:30am – 12:30pm
Room 5.02, Bayes Centre

Typological evidence (Greenberg 1963, Dryer 1992) suggests that word order is correlated across different types of phrases, that is, heads tend to be ordered consistently relative to dependents (i.e., head-initial or head-final). This general tendency is called Cross-Category Harmony (Hawkins 1980). There is a long history of claims that these word order correlations are driven by cognitive factors– like a bias favouring consistency, or simplicity (e.g., Vennemann 1976, Chomsky 1988, Hawkins 1994, 2004, Pater 2012, Culbertson et al. 2012, Culbertson & Kirby 2016). At the same time, there are credible alternative explanations, including a common diachronic source for heads that are aligned (Givón 1979, Aristar 1991, Collins 2019). To link the typology to cognition, we need more than typological data (Kirby 1999, Culbertson 2012, Ladd et al. 2014). By using artificial language learning experiment, previous work has shown that harmony is preferred by learners acquiring the order of modifiers within the noun phrase (Culbertson et al. 2012), but prior work has largely neglected harmonic patterns that truly cut across categories.

In this talk, we will present two artificial language learning experiments testing Cross-Category Harmony between verb phrase (VP) and adpositional phrase (PP) and between VP and adjective phrase (AP). These two cases are critically different: there is strong typological evidence for the former but none for the latter (Dryer, 1992). Our results show that there is a strong learning preference for harmonic orders between VP and PP, regardless of whether VP order was VO or OV and whether the participants’ native language has hormonic order or mixed orders. By contrast, there is a weaker preference for harmony between VP and AP only when the order matches participants’ native language order. Our results suggest that an underlying cognitive bias for harmony may (at least in part) drive typology. We propose that the difference in the strength of preference for harmony across these phrase types might be attributed to the different degree of cross-category similarities.

October 29: Eric Round

An evolutionary science of phonology and morphology: What will it take to get there?

Erich Round (Max Planck Institute for the Science of Human History, Jena; University of Queensland)

Tuesday, October 29
11:00am – 12:30pm
G32, 7 George Square

Linguists are increasingly interested in explaining the observable diversity of the world’s languages in terms of the historical-evolutionary process by which they arise. But is that goal realistic? And what would it take to get there? To begin, we will need theories and models of evolutionary processes that spell out the factors involved in historical events and their probabilities of occurring within given time spans under a range of conditions. To then test those theories against empirical data, we will need phylogenies of language families, and if we are to evaluate hypothesised probabilities of changes we’ll need phylogenies with explicit branch lengths and quantifications of uncertainty. Currently, phylogenies that meet these desiderata can be generated by lexical phylolinguistic methods, but there is a wrinkle: those methods are based on cognacy judgements which in turn require an already existing hypothesis about the history of the language family, meaning that they are at least partly circular, and are dependent upon the manual comparative method, which is yet to be carried out on the vast majority of the world’s languages. What we might do in response? In this talk I explore two paths forward. I review emerging first results from research into phonotactic phylogenetics, which probes the phylogenetic signal contained within statistical patterns in the lexicon without requiring cognacy judgements. I then turn to the question of whether we could attempt to conduct core parts of the comparative method, namely the inference of sound change and morphological change, in an automated fashion. Until very recently such an idea was flatly unrealistic, but subsequent advances in statistics have opened new doors which were closed to us just a decade ago. The key, it appears, will be to build certain kinds of precise mathematical models of change within paradigms. I discuss what those models might look like and the kinds of science that their use might enable.