Evolang is over, but I have a backlog of posts to get out!

The idea that language change can be biased by the cognitive profiles of its learners has attracted a lot of interest (see Hanna’s post), and was a frequent topic of discussion at Evolang.  In the talk by James Winters and I, we urge a pluralistic approach involving statistical tests, models and experiments.  Here I describe some of the new studies relating to this presented at the conference.

Roland Mühlenbernd and Michael Franke discuss the network properties that characterise language contact.  They constructed an agent-based model where agents had to converge on a system of mapping two meanings onto two signals.  There were two evolutionary stable mappings:  Meaning 1 maps to word 1 and Meaning 2 maps to word 2, or meaning 1 maps to word 2 and meaning 2 maps to word 1.

Agents played communication games with others to settle on the mapping they use.  Mühlenbernd & Franke used two types of agent:  The rational agent which chooses the rationally best response and reinforcement learners based around a Polya-Urn model.  However, this factor didn’t make a significant difference in the results presented in this talk.

The main focus was the structure of the social network that determined which agents interacted.  Small world networks were generated using the Watts-Strogatz method which creates a variety of networks with certain degree and centrality features.  The social structure was held constant within each run, then the results over several runs were analysed.  Homogeneity always emerged, although the reinforcement learning maintained a higher number of ‘language regions’ (where connected agents used the same mappings) for longer.  These langauge regions tended to form in tightly connected regions of the network, not surprisingly.  Mühlenbernd & Franke looked at the properties of the langauge regions, including how early agents settled on a mapping (early vs late learners) and the strategies of agents on the border between dense communities (border agents).

Interestingly, there was a big overlap in late learners and border agents.  That is, people on the border between two communities tend to be late learners.  This offers an interesting take on the hypotheses linking second language learners and linguistic change.  Lupyan & Dale (2010) find a correlation between group size and morphological complexity.  They suggest that the cognitive profiles of L2 learners biases language change towards morphologically simpler languages.

This hypothesis is further supported by the work of Christian Bentz and Bodo Winter (not to be confused with James Winters, founder of this blog) also presented at this conference which shows that the ratio of L1 to L2 speakers of a language correlates with morphological features (number of cases, cast syncretism and case symmetry), while controlling for language family and geographic region.  However, as I suggested in my talk with James Winters, backing up a statistical correlation with another statistical correlation is not as powerful as running an experiment or a model.

Mühlenbernd & Franke’s model might provide some insight into this problem.  It shows that people who are the most likely to be in contact situations (on the borders of communities) are also more likely to be late learners.  If late learners in the model can equate to L2 learners, then this suggests a closer link than previously hypothesised between L2 learners and langauge change.  It would be interesting to think more about this dynamic.  However, Franke urged caution in interpreting the model in this way, since the concept of a ‘late learner’ is fairly abstract.

As a side note, Bart de Boer raised the intriguing idea you could use statistical analyses like Bentz and Winter’s to find exceptions to the rule.  Rather than them being problematic, perhaps by studying the causes of change in these exceptions, a clearer idea of the role of L2 speakers could emerge.

What’s clear is that there is an emerging body of work surrounding the linguistic niche hypothesis using statistical and modelling techniques.  Combined with Hannah Little’s experiment on this phenomenon, I’m wondering how long before we get a special issue on this subject.

Cedric Boeckx gave a remarkable plenary which tried to pull together the fields of cultural language evolution and biolinguistics, with surprising concessions on either side.  Boeckx started from a relatively uncontroversial part of Chomsky’s claim:  That aspects of language can be studied scientifically as part of biology.  However, Boeckx noted that Luria in 1976 was confident that ‘within a few years’ linguists would be interfacing with and contributing to findings from biology.  However, formal syntax has failed to carry out the biological commitment, and Boeckx wonders why linguists don’t have more to say about, for instance, the recent developments in the study of FOXP2.

Boeckx outlined his own position as minimalist, in the sense that a fully specified UG is not plausible.  We need to realise that biology is complex, and move beyond the classical model of Broca and Wernicke’s area as dedicated centers of language.  Also, Boeckx urged the audience to forget about the FLN/FLB distinction, since from a biological viewpoint this view is misleading:  Genes build neural structures, not behaviour (although linguists should note the richness of the range of aspects now thought to be part of FLB).

Instead, Boeckx suggests that the subject of study should be a set of formal properties.  Boeckx suggested the following, while emphasising that the particular terms were not important and it is just the concepts that he would focus on:

• An edge property:  This removes selectional restrictions on concepts in different domains and makes it possible to combine them.  For example, humans can pull together concepts from very different domains.  Also, lexical items have the property of being able to combine with other lexical items.
• Set formation or Merge:  The ability to combine lexical items.
• Cyclic transfer:  Elements are combined at different levels before being passed to other operations.  This allows recursion.

These specify a minimal specification of universal grammar for which might realistically find biological explanations.  Boeckx sees no problem with the idea that we share some of these abilities with animals.  An even bigger concession is that he believes that the particular structures of language (e.g. word order or pro-drop) can be explained by cultural evolution i.e. grammaticalism.  The minimal specifications are weak biases, but we need a cultural explanation.

Boeckx went on to suggest how the biological underpinnings of the minimal specification might be approached.  He promoted the concept of the ‘Global workspace’ as used by Dehanene and colleagues.  This approach suggests that cross-modular computation is the key to human cognition.  It focuses on distributed networks of neurons with long-distance connections which allow different modules of the brain to interact.  Humans are particularly good at integrating concepts across perceptual modalities or time.  Boeckx suggests that this ability is the biological basis for the edge property.  It allows different perceptions to be treated in such a way that they can be combined.  I was put in mind of synaesthesia and the work of Chrissy Cuskley on synaesthesia and language evolution.

Boeckx went on to suggest that the thalmus could act as a regulator of information exchange in this global workspace and cited some studies showing that it is sensitive to syntax and semantics, but not phonology.  The thalmus is ideally placed – right at the center of the brain.  Boeckx also suggested that humans have evolved to have a more regularly spherical brain, facilitating this workspace by placing the thalmus equidistantly from all brain areas (suggesting that earlier ancestors of modern humans had a more elongated brain).  However, he was skeptical that we could ever know if this was an adaptation for language.

This integrative approach is in close alignment with proponents of cultural evolution such as Simon Kirby, who sees the structure of langauge as emerging from cultural transmission, but biology as proving the platform for cultural transmission.  Boeckx’s approach differs a great deal to that of Massimo Piattelli-Palmarini, whose talk essentially told cultural evolutionists that they were wrong and should stop researching explanations that could not be true.  However, one commenter wondered if Boeckx’s concessions were a dangerous form of moderate liberalism – these arguments might leave both the cultural camp and the formalist camp believing that there is no conflict and actually lead to further isolation.  However, I welcome this impressive synthesis and hope that it’ll raise the profile of cultural transmission in the evolution of language.

Luke McCrohon suggests that tools from evolutionary biology can be applied to linguistic borrowing between languages.  McCrohon correctly points out that the descent of lexicons are far from tree-like, and there is a great deal of horizontal transfer (see also my post on analysing an etymology dictionary). Although it’s mainly nouns that are borrowed into a language, any feature can potentially be borrowed, according to Thmason & Kaufman (1988).  However, we tend to observe hierarchies of borrowing such that some types of words are borrowed more frequently than others.  For instance, Haugen notes that nouns are more likely to be borrowed than verbs, which are in turn more likely to be borrowed than prepositions.  McCrohon links this with a similar observation in biological evolution that certain types of genes are more likely to be borrowed.  Informational genes (that provide the basis for functions) are less likely to be borrowed than operational genes (that modify other functions).  Jain et al.’s (1999) complexity hypothesis suggests that, while all genes have the same probability of being copied, simpler genes are more likely to be copied faithfully since they have fewer constraints on the precise form they must take to be effective.

McCrohon argues that In a similar way, the explanation of the linguistic borrowing hierarchy might also reflect the increasing constraints on how a word can be used.  For instance, most nouns can be substituted by other nouns, while prepositions are highly restricted by context or domain.  Also, language-interal change might be affected by these restrictions.  Even if there is a more effective form than in the existing system, removing one form might have knock-on consequences for the whole system.  This inter-connectedness could have implications for how languages are likely to change.

Furthermore, this model might predict that words are equally likely to be selected for borrowing, but only certain types have a good likelihood of being successfully borrowed.  However, a commenter wondered about words that are borrowed to fill conceptual gaps such as new technologies.  Still, an interesting analogy between problems in biology and problems in linguistics.  And McCrohon is confident that his studies will also have something to give back to the biology community by studying how this problem applies to linguistics.

Andrew Smith asks what are Darwinian linguistic replicators.  He starts with Croft’s conception of the lingueme.  Croft says that linguemes are external manifestations: utterances including their full context.  However, this might mean that they are not observable, since we can’t observe the full context of an utterance nor the speaker’s intention.  Furthermore, this ignores the fact that meanings are different for each hearer.  So linguemes cannot be observed on the hearer’s side either.  Nikolas Ritt’s conceptualisation of the lingueme suggests that it is an entirely internal entity.  However, this means that we can’t observe the lingueme at all.  Furthermore, it ignores the fact that langauge is ostensive and inferential.  Smith advocates a view stronger than Mufwene’s position that meanings are re-constructed in the minds of hearers:  Hearers build their own knowledge and infer the meaning of speakers – this is a far remove from replicating anything in the speaker’s mind.   So the lingueme does not replicate faithfully.  In fact, we should not expect the lingueme to replicate faithfully, but be on the opposite side of the continuum to replicators.

Smith concluded with the paradox that linguemes must contain some aspect of meaning, but meaning is individual and not observable.

Monica Tamariz asked whether linguistic replicators needed to have an aspect of meaning.  Alternatively, Tamariz argued you could have replication of forms without replication of meaning.  Smith disagreed, seeing a pairing of form and meaning as an essential part of a linguistic replicator.

Smith pointed out that some priming effects demonstrated that people can re-create speaker’s individual voices in their minds, so would this count as faithful replication.  Smith replied that we shouldn’t expect linguistic replicators to be faithfully transmitted.

Luke McCrohan later suggested that perhaps you could have replication of a communicative event- that is, the lingueme is both the speaker’s intention and the hearer’s inference and the external form.

This was a refreshing and no-nonsense take on the linguistic replicator question.  But whatever the right answer, it demonstrates that evolutionary linguists are still struggling to reconcile language in the individual and language in the environment.  Nowhere is this clearer than in models, where typically addressing one aspect compromises the other.

This is written at 1am after a sake and sushi reception.  I have to praise the organisation of the conference so far!

Kicking off the workshop on Constructive approaches to Language Evolution (proceedings for all workshops downloadable here), Bart de Boer talked about the dangers of Fact-free science.  Maynard-Smith recognised of a certain kind of science that does not refer to outside phenomena, but merely concentrates on exploring models already established in the sub-field.  Constructive approaches and the Artificial Life approach was always susceptible to this criticism, but de Boer recognises that the initial enthusiasm for constructive models has waned while the skepticism has remained.   However, de Boer suggested that Maynard-Smith’s point should be a friendly warning to researchers in language evolution, rather than a criticism, since Maynard-Smith himself was subject to these kinds of criticism in the field of mathematical modelling.  de Boer emphasises that research should never loose sight of the research questions that motivated previous studies, and encouraged modellers to ask whether they were answering questions that other researchers were asking.

de Boer also talked about ‘Cargo cult science’ – a name derived from pre-industrial cults that believed in emulating the technologically advanced societies that they came in contact with would maintain the flow of new goods – a practice that goes through the motions of doing science, but doesn’t actually produce results.  For instance, a model shouldn’t just explain the data which it was built on, but should be expandable to explain other phenomena.

de Boer questioned whether the Iterated Learning Model experimental paradigms were guilty of this kind of cottage-industry science, wondering whether they study langauge evolution or how humans play certain types of games.  However, he did concede that it was a relatively new paradigm and at least it got modellers running experiments.  I asked whether this was a little unfair on the ILM, since part of the motivation of the ILM studies was to counter claims made in that pinnacle of fact-free science, formalist nativism.  That is, the ILM showed that you don’t need strong innate biases to get strong language universals in populations.  de Boer answered, quite sensibly, that these points had been made with the computation models already, but more importantly, there was no point in trying to convince those kinds of researchers – the real audience for researchers of cultural evolution should be biologists – de Boer pointed out that the most prestigious work on language evolution (in terms of journal prestige and citations) is largely by biologists, not linguists (e.g Nowak).  And to convince them, we need fact-free science.

It was a pity, then that some interesting modelling work by Reiji Suzuki and Takaya Arita (Reconsidering language evolution from coevolution of learning and niche construction using a concept of dynamic fitness landscape, also in the workshop proceedings) seemed to be suffering from this malady.  To start with, as Thom Scott-Phillips pointed out, the title doesn’t make sense, since niche-construction is essentially a type of coevolution.  Suzuki described model where individuals could affect each other’s linguistic inventories either directly through communication, or indirectly by contributing linguistic elements to a pool of linguistic resources, like an animal altering its adaptive landscape (e.g. beavers building dams).  Each individual had a phenotype space which was defined by several innate properties:  First, an initial phenotype.  Second, a learning variable where by an individual could bring its phenotype closer to the peak of the adaptive landscape.  Finally, a niche construction parameter by which individuals could pull the adaptive peak closer to or further away from their phenotype. Individuals inherited these parameters like genes.

A circular dynamic emerged where the population cycled through having many adaptive peaks, which increased the learning parameter, which lead to a single adaptive peak, which lowered the importance of learning, which finally pulled the single adaptive peak into many adaptive peaks, which increased the importance of learning, and so on.  While this was happening, the fitness of the agents was being ratcheted up by a series of steep increases, essentially a the Baldwin effect being repeatedly applied.  This is the first of a number of presentations about the Baldwin effect and coevolution (talk by Bill Thompson and poster by Vanessa Ferdinand).

While this is an interesting dynamic, when I asked how the concept of a shared environment or the ability to modify the adaptive landscape applied to language, there was not a clear answer.  I suspect that the distinction between individual interactions and modifying the external environment, which works well for animals building nests or dams, does not work so well for spoken language, because linguistic signals don’t persist in the environment.  However, the problem of how to represent the langauge of a community alongside individual behaviour is not an easy problem to solve.  Suzuki suggested that perhaps the model can be related to an earlier stage of language evolution, but we’ll have to wait for a better description of how this model can answer the questions that researchers in language evolution ask.

The 43rd Poznań Linguistic Meeting is holding a thematic session on Theory and evidence in language evolution research.  The call is still open, but the deadline is the 15th March.  From the conference description:

The aims of the session can be summarised as follows:

• to assess the present range of available evidence and to discuss the status of the new sources of evidence
• to assess the role of theoretical syntheses and holistic scenarios of language emergence and evolution
• to identify the ways in which linguistic methodologies can be made relevant to answering the ‘origins’ type questions,
• to identify the limitations of linguistic methodologies alone and thus directions of interdisciplinary collaboration
• to bridge the gap between conceptions of evidence in biology and linguistics

Evolang is busy this year – 4 parallel sessions and over 50 posters. We’ll be positing a series of previews to help you decide what to go and see. If you’d like to post a preview of your work, get in touch and we’ll give you a guest slot.

Sean Roberts & James Winters Constructing Knowledge: The nomothetic approach to language evolution
Session 2, Workshop on Constructive Approaches to Language evolution, 13th March

Recently, there’s been a surge in large-scale, cross-cultural statistical studies that look at the co-evolution of language structure are social structure.  These contrast with small-scale case studies on the one hand and computational models on the other.  Lupyan & Dale refer to this approach as ‘Nomothetic’ – looking for general patterns or laws.  For example, they find that the number of speakers of a language correlates with the morphological complexity of that language.  These approaches are cheap, fast and easy to perform.  They use real data, and they might reveal some interesting links that we might want to include in our models.  However, on their own, they have little explanatory power:  We know that group size and morphological complexity are linked, but the statistics don’t tell us why they are linked (see Hannah’s post and my comment, too).

Worse, the amount of data available on the internet and new statistical techniques mean that it’s possible to find some sort of link between any cultural traits (as this set of spurious correlations demonstrates).  For example, there is a robust link between linguistic diversity and the number of road fatalities in a country.  Does this mean that models of linguistic diversity should include a simulation of traffic accidents?  Probably not, but which studies should we pay attention to as modellers?

This talk discusses the new nomothetic approach and presents some criteria to keep in mind when conducting or reviewing a nomothetic study.  We conclude that nomothetic studies can work together with constructive, idiographic and experimental approaches to get a better picture of how language structure and social structure are linked.

You can read our paper in the online proceedings.

Levinson & Gray (2012) demonstrate how tools from evolutionary biology can help refine the way we look at human language and human cognition.  Phylogenetic techniques allow researchers to properly control for the fact that languages are related by descent.  More importantly, these tools allow the study of the full variation of linguistic structures, rather than assuming that the majority of linguistic structure is constrained by a limited set of Universal Grammar parameters.  This topic has been discussed before, by the authors and on this blog, but this paper is much more a manifesto for change.

Read the rest of this entry →