language evolution – Page 3

Language Evolution or Language Change?

You sometimes hear people complaining about the use of the term “language evolution” when what people really mean is historical linguistics, language change or the cultural evolution of language. So what’s the difference?

Some people argue that evolution is a strictly biological phenomenon; how the brain evolved the structures which acquire and create language, and any linguistic change is anything outside of this.

Sometimes this debate gets reduced to the matter of whether there are enough parallels between the cultural evolution of language and biological evolution to justify them both having the “evolution” label. George Walkden recently did a presentation in Manchester on why language change is not language evolution and dedicated quite a large chunk of a presentation to where the analogy between languages and species fall down. It is true that there are a lot of differences between languages and species, and how these things replicate and interact, and of course it is difficult to find them perfectly analogous.

However, focussing on the differences between biological and cultural evolution in language causes one to overlook why a lot of evolutionary linguistics work looks at cultural evolution. Work on cultural evolution is trying to address the same question as studies looking directly at physiology, why is language structured the way it is? Obviously how structure evolved is the main question here, but how much of this was biological, and how much is cultural is still a very open question. And any work which looks at how structure comes about, either through biological or cultural evolution can, in my opinion, legitimately be called evolutionary linguistics.

Additionally, in the absence of direct empirical evidence in language evolution, the indirect evidence that we can gather, either through observing the structure of the world’s languages, or by using artificial learning experiments, can help us answer questions about our cognitive abilities.

Furthermore, Kirby (2002) outlined 3 timescales of language evolution on the levels of biological evolution (phylogenetic), cultural evolution (glossogentic) and individual development (ontogentic). All of these timescales interact and influence each other, so it’s necessary to consider all of these levels in language evolution research, and to say work on any of these timescales is not language evolution research is not respecting the big picture.

So what’s the difference between language change and language evolution? As with almost everything, it’s not a black and white issue. I would say though that studies looking at universal trends in language, or cultural evolution experiments in the lab, are very relevant to language evolution. What I’d label historical linguistics, or studies on language change, however, is work which presents data from just one language, as it is hard to make inferences about the evolution of our universal capability for language with just one data point.

Figure 1 from: Kirby, S. (2002b). Natural language from artificial life. Artificial Life, 8(2):185-215.

A Note on Memes and Historical Linguistics

When I began my most recent series of posts on memes, I did so because I wanted to think specifically about language: Does it make sense to treat words as memes? That question arose for a variety of reasons.

In the first place, if you are going to think about culture as an evolutionary phenomenon, language automatically looms large as so very much of culture depends on or is associated with language. And language consists of words, among other things. Further, historical linguistics is a well-developed discipline. We know a lot about how languages have changed over time, and change over time is what evolution is about.

However, words have meanings. And word meanings are rather fuzzy things, subject to dispute and to change that is independent of the word-form itself. Did I really want to treat word meanings as memes? That seemed rather iffy. But if I don’t treat word meanings as memetic, then what happens to language?

But THAT’s not quite how I put it going into that series of posts. Of course, I’ve known for a long time that words have forms and meanings. I don’t know whether it was my freshman year or my sophomore year that I read Roland Barthe’s Elements of Semiology (English translation 1967). That gave me Saussure’s trilogy of sign, signifier, and signified, the last of which seemed rather mysterious: “the signified is not ‘a thing’ but a mental representation of the ‘thing’.” Getting comfortable with that distinction, between the thing and the concept of the thing, that took time and effort.

That’s an aside. Suffice to say, I got comfortable with that distinction. The distinction between signifier and signified was much easier.

And yet that distinction was not uppermost in my mind when I thought of language and cultural evolution. When I thought of memes. When I approached this series of essays, though some papers by Daniel Dennett, I thought of words they same way Dennett did, the whole kit and caboodle had to be a meme. It was the sign that’s the meme.

That’s not how I ended up, of course. That ending took me a bit by surprise. Coming down that home stretch I was getting worried. It appeared to me that I was faced with two different classes of memes: couplers and the other one. What I did then was to divide the other one into two classes: targets and designators. And to do that I had to call on that thing I’ve known for decades and split the word in two: signifier and signified. It’s only the signifier that’s memetic. Signifiers are memes, but not signifieds.

It took me a couple of months to work that out, and I’d known it all along.

Sorta’.

What does that have to do with historical linguistics? Historical linguistics is based mostly on the study of relationships among signifiers, that is, relationships among the memetic elements of languages. Which makes sense, of course.

But… Continue reading “A Note on Memes and Historical Linguistics”

CogSci 2013 – the others

It’s over but there’s loads of cool papers I didn’t cover. Below is probably not a definitive list because there are SO MANY papers here, but here’s a good flavour of of the more language evolutiony offerings. In no particular order.

Language and Gesture Evolution by Call, Goldin-Meadows, Hobaiter, Liebal and Tomasello

Abstract:

In humans, gestural communication is closely intertwined with language: adults perform a variety of manual gestures, head movements and body postures while they are talking, children use gestures before they start to speak, and highly conventionalized sign systems can even replace spoken language. Because of this role of gestures for human com-munication, theories of language evolution often propose a gestural origin of language. In searching for the evolutionary roots of language, a comparative approach is often used to investigate whether any precursors to human language are also present in our closest relatives, the great apes, because of our shared phylogenetic history. Therefore, the aim of this symposium is to present recent progress in the field of language evolution from both a developmental and compa-rative perspective and to discuss the question if and to what extent a comparison with nonhuman primates is suitable to shed light on possible scenarios of language evolution.

Individuals recapitulate the proposed evolutionary development of spatial lexicons by Carstensen and Regier

Abstract:

When English speakers successively pile-sort colors, their sorting recapitulates an independently proposed hierarchy of color category evolution during language change (Boster, 1986). Here we extend that finding to the semantic domain of spatial relations. Levinson et al. (2003) have proposed a hierarchy of spatial category evolution, and we show that English speakers successively pile-sort spatial scenes in a manner that recapitulates that proposed evolutionary hierarchy. Thus, in the spatial domain, as in color, proposed universal patterns of language change based on cross-language observations appear to reflect general cognitive forces that are available in the minds of speakers of a single language.

Systems from Sequences: an Iterated Learning Account of the Emergence of Systematic Structure in a Non-Linguistic Task by Cornish, Smith and Kirby

Abstract:

Systematicity is a basic property of language and other culturally transmitted behaviours. Utilising a novel experimental task consisting of initially independent sequence learning trials, we demonstrate that systematicity can unfold gradually via the process of cultural transmission.

Experimental insights on the origin of combinatoriality by Roberts and Galantucci

Abstract:

Combinatoriality—the recombination of a small set of basic forms to create an inﬁnite number of meaningful units—has long been seen as a core design feature of language, but its origins remain uncertain. Two hypotheses have been suggested. The ﬁrst is that combinatoriality is a necessary solution to the problem of conveying a large number of meanings; the second is that it arises as a consequence of conventionalisation. We tested these hypotheses in an experimental-semiotics study. Our results supported the hypothesis based on conventionalisation but offered little support for the hypothesis based on the number of meanings.

Combinatorial structure and iconicity in artificial whistled languages by Verhoef, Kirby and de Boer

Abstract:

This article reports on an experiment in which artificial languages with whistle words for novel objects are culturally transmitted in the laboratory. The aim of this study is to investigate the origins and evolution of combinatorial structure in speech. Participants learned the whistled language and reproduced the sounds with the use of a slide whistle. Their reproductions were used as input for the next participant. Cultural transmission caused the whistled systems to become more learnable and more structured. In addition, two conditions were studied: one in which the use of iconic form-meaning mappings was possible and one in which the use of iconic map- pings was experimentally made impossible, so that we could investigate the influence of iconicity on the emergence of structure.

Linguistic structure is an evolutionary trade-off between simplicity and
expressivity by Smith, Tamariz and Kirby

Abstract:

Language exhibits structure: a species-unique system for expressing complex meanings using complex forms. We present a review of modelling and experimental literature on the evolution of structure which suggests that structure is a cultural adaptation in response to pressure for expressivity (arising during communication) and compressibility (arising during learning), and test this hypothesis using a new Bayesian iterated learning model. We conclude that linguistic structure can and should be explained as a consequence of cultural evolution in response to these two pressures.

Communicative biases shape structures of newly acquired languages by Fedzechkina, Jaeger and Newport

Abstract:

Languages around the world share a number of commonalities known as language universals. We investigate whether the existence of some recurrent patterns can be explained by the learner’s preference to balance the amount of information provided by the cues to sentence meaning. In an artificial language learning paradigm, we expose learners to two languages with optional case-marking – one with fixed and one with flexible word order. We find that learners of the flexible word order language, where word order is uninformative of sentence meaning, use significantly more case-marking than the learners of the fixed word order language, where case is a redundant cue. The learning outcomes in our experiment parallel a variety of typological phenomena, providing support for the hypothesis that communicative biases can shape language structures.

Regularization behavior in a non-linguistic domain by Ferdinand, Thompson, Kirby and Smith

Abstract:

Language learners tend to regularize unpredictable variation and some claim that is due to a language-specific regularization bias. We investigate the role of task difficulty on regularization behavior in a non-linguistic frequency learning task and show that adults regularize variable input when tracking multiple frequencies concurrently, but reliably reproduce the variation they have observed when tracking one frequency. These results suggest that regularization behavior may be due to domain-general factors, such as memory limitations.

Learning, Feedback and Information in Self-Organizing Communication Systems by Spike, Stadler, Kirby and Smith

Abstract:

Communication systems reliably self-organize in populations of interacting agents under certain conditions. The various fields which model this – game theory, cognitive science and evolutionary linguistics – make different assumptions about the learning and behavioral processes which are responsible. We created an exemplar-based framework to directly compare these approaches by reproducing previously published models. Results show that a number of mechanisms are shared by the systems which can construct optimal communication. Three general factors are then proposed to underlie any self-organizing learned system.

A robustness approach to theory building: A case study of language evolution by Irvine, Roberts and Kirby

Abstract:

Models of cognitive processes often include simplifications, idealisations, and fictionalisations, so how should we learn about cognitive processes from such models? Particularly in cognitive science, when many features of the target system are unknown, it is not always clear which simplifications, idealisations, and so on, are appropriate for a research question, and which are highly misleading. Here we use a case-study from studies of language evolution, and ideas from philosophy of science, to illustrate a robustness approach to learning from models. Robust properties are those that arise across a range of models, simulations and experiments, and can be used to identify key causal structures in the models, and the phenomenon, under investigation. For example, in studies of language evolution, the emergence of compositional structure is a robust property across models, simulations and experiments of cultural transmission, but only under pressures for learnability and expressivity. This arguably illustrates the principles underlying real cases of language evolution. We provide an outline of the robustness approach, including its limitations, and suggest that this methodology can be productively used throughout cognitive science. Perhaps of most importance, it suggests that different modelling frameworks should be used as tools to identify the abstract properties of a system, rather than being definitive expressions of theories.

And last but not least…replicated typo’s very own Sean Roberts with A Bottom-up approach to the cultural evolution of bilingualism

Abstract:

The relationship between individual cognition and cultural phenomena at the society level can be transformed by cultural transmission (Kirby, Dowman, & Griffiths, 2007). Top-down models of this process have typically assumed that individuals only adopt a single linguistic trait. Recent extensions include ‘bilingual’ agents, able to adopt multiple linguistic traits (Burkett & Griffiths, 2010). However, bilingualism is more than variation within an individual: it involves the conditional use of variation with different interlocutors. That is, bilingualism is a property of a population that emerges from use. A bottom-up simulation is presented where learners are sensitive to the identity of other speakers. The simulation reveals that dynamic social structures are a key factor for the evolution of bilingualism in a population, a feature that was abstracted away in the top-down models. Top-down and bottom-up approaches may lead to different answers, but can work together to reveal and explore important features of the cultural transmission process.

CogSci 2013: Communication Leads to the Emergence of Sub-optimal Category Structures

Next up is Catriona Silvey, Simon Kirby and Kenny Smith, who use an experiment which gets participants to categorise shapes from a continuous space either in a communicative condition or a non-communicative condition. You can read it here (you should, it’s really awesome, and I only describe it briefly below): http://www.lel.ed.ac.uk/~s1024062/silvey_cogsci.pdf

Silvey et al. are interested in how semantic categories emerge. They set up an experiment in which participants were given a continuous semantic space:

In one condition single participants had to simply divide the space up into different labeled categories. In another condition, participants had to play a communication game. Participants communicated in pairs using the same semantic space. In both conditions, participants were given nonsense words to label their categories. Participants could assign as many categories as they liked.

In the communication condition one participant was the sender and one was the receiver. The sender chose a word to communicate a given shape from the grid above, and the receiver had to chose which shape they thought the word referred too. Participants were then given feedback as to whether the receiver had chosen the right image. They were then given a score based on how close the chosen shape was to the original given shape. The sender and received swapped roles every trial. Together they labelled every shape 4 times throughout the experiment and the last label for each shape chosen by both participants were taken for analysis.

An optimal categorisation strategy within this game would be to give every shape its own label, however, memory constraints are likely to stop participants using this strategy. Given that you will score more if shapes are closer, it was expected that participants would use small, clustered and equally sized categories in order to optimise getting the right shape, and if not, maximising their score.

In the non-communicative experiment, participants arranged the categories in fairly balanced chunks that would have served relatively optimally for the communication game. However, despite expectations, participants in the communicative condition behaved sub-optimally and did not maximise their communicative success (their score in the game) in that their categories weren’t clustered or optimal in colour or size. This could have been caused by the communicative condition having extra pressure for the learnability of categories, as well as a pressure for communicative success, which the non-communicative condition did not have. The authors argue that this is possibly a demonstration for how real languages arrive at suboptimal categories, e.g. where words vary as to whether they represent a very small category or represent a much broader part of the semantic space.

Memes as Data: Targets, Couplers, and Designators

FLASH! The active brain-hopping meme is to culture as the preformationist germ seed is to biology. More in a later post.

While these posts have been quite critical of Dennett’s use of the concept of information, that should not be taken to mean that I see no use for such a concept. There is, of course, the technical usage established by Shannon and Weaver, but that is of limited use in these discussions. The challenge is to make use a less specialized, indeed an informal, concept. I intend to do so by identifying memetic information with the formal concept of data, as employed in computer science, and in particular with the concept of a parameter.

The Wikipedia defines parameter in this way:

In computer programming, a parameter is a special kind of variable, used in a subroutine to refer to one of the pieces of data provided as input to the subroutine. These pieces of data are called arguments. An ordered list of parameters is usually included in the definition of a subroutine, so that, each time the subroutine is called, its arguments for that call can be assigned to the corresponding parameters.

Memes are arguments in the sense of that definition. The parameters for which they are values are components of “subroutines” in the mind, where, for the purposes of this discussion, the mind is considered to be a rich and sophisticated computational device whose characteristics are unspecified. Continue reading “Memes as Data: Targets, Couplers, and Designators”

CogSci 2013: The Impact of Communicative Constraints on the Emergence of a Graphical Communication System

The Annual Conference of the Cognitive Science Society is happening next week with quite a but of language evolution stuff going on. On the run up I’ll post a few evolutionary linguisticsy papers which will be presented. If you’re presenting something please feel free to send me it to be covered here or write a guest post and I’ll post it up!

Firstly, Bergmann, Lupyan & Dale are presenting a paper called “The Impact of Communicative Constraints on the Emergence of a Graphical Communication System”. The paper can be seen here: http://www.tillbergmann.com/papers/cogsci_squiggle_bergmann_dale_lupyan.pdf

The authors present an experiment in which participants had to communicate using graphical symbols (called squiggles) of faces. The experiment worked as a communication game but also had generational turnover as in iterated learning experiments. They investigated the effect of different features of the input faces and the effect of changing the comprehension condition and how these affected the structure of the symbols that emerged.

The study used Dale & Lupyan’s (2010) squiggle framework but used human faces as the input. Participants were to “squiggle” the face in just 5 seconds to prevent them from building representations which were too detailed and prevented participants from writing words etc. In the first experiment they found that participants used strategies such as using hair shape/length, face shape and features to differentiate between faces and different strategies were more common in some gender/age categories, e.g. women were more likely to be defined by their hair.

The second experiment was the same as the first one but differed in the “listening” round, where participants were to choose which original face a squiggle represented. In experiment 1, participants had to choose between any two faces, in experiment 2 they were to choose between an “opposite” face. Faces differed in being male or female and young or old. So in experiment 2 young females were always pitted against old males, and old females pitted against young males. This was to see if the environment in interrupting the squiggles effected how they were represented and interrupted. This experiment found that squiggles in experiment 2 had less detail in order to be successful, and the amount of complexity fell over generations. I suppose this is an effect of the differentiating features between faces being more salient in the second experiment where faces always differ in both age and gender. This shows that the context in which communication occurs can shape the structure and complexity of that communication.

How Do We Account for the History of the Meme Concept?

First, in asking THAT question I do not intend a bit of cutesy intellectual cleverness: Oh Wow! Let’s get the meme meme to examine it’s own history. My purpose would be just as well served by examining, say, the history of the term “algorithm” or the term “deconstruction,” both originally technical terms that have more or less entered the general realm. I’m looking at the history of the meme concept because I’ve just been reading Jeremy Burman’s most interesting 2012 article, “The misunderstanding of memes” (PDF).

Intentional Change

Second, as far as I can tell, no version of cultural evolution is ready to provide an account of that history that is appreciably better than the one Burman himself supplies, and that account is straight-up intellectual history. In Burman’s account (p. 75) Dawkins introduced the meme concept in 1976

as a metaphor intended to illuminate an evolutionary argument. By the late-1980s, however, we see from its use in major US newspapers that this original meaning had become obscured. The meme became a virus of the mind.

That’s a considerable change in meaning. To account for that change Burman examines several texts in which various people explicate the meme concept and attributes the changes in meaning to their intentions. Thus he says (p. 94):

To be clear: I am not suggesting that the making of the active meme was the result of a misunderstanding. No one individual made a copying mistake; there was no “mutation” following continued replication. Rather, the active meaning came as a result of the idea’s reconstruction: actions taken by individuals working in their own contexts. Thus: what was Dennett’s context?

And later (p. 98):

The brain is active, not the meme. What’s important in this conception is the function of structures, in context, not the structures themselves as innate essences. This even follows from the original argument of 1976: if there is such a thing as a meme, then it cannot exist as a replicator separately from its medium of replication.

Burman’s core argument this is a relatively simple one. Dawkins proposed the meme concept in 1976 in The Selfish Gene, but the concept didn’t take hold in the public mind. That didn’t happen until Douglas Hofsadter and Daniel Dennett recast the concept in their 1982 collection, The Mind’s I. They took a bunch of excerpts from The Selfish Gene, most of them from earlier sections of the book rather than the late chapter on memes, and edited them together and (pp. 81-82)

presented them as a coherent single work. Al- though a footnote at the start of the piece indicates that the text had been excerpted from the original, it doesn’t indicate that the essay had been wholly fabricated from those excerpts; reinvented by pulling text haphazardly, hither and thither, so as to assemble a new narrative from multiple sources.

It’s this re-presentation of the meme concept that began to catch-on with the public. Subsequently a variety of journalist accounts further spread the concept of the meme as a virus of the mind.

Why? On the face of it it would seem that the virus of the mind was a more attractive and intriguing concept whereas Dawkins’ original more metaphorical conception. Just why that should have been the case is beside the point. It was.

All I wish to do in this note is take that observation and push it a bit further. When people read written texts they do so with the word meanings existing in their minds, which aren’t necessarily the meanings that exist in the minds of the authors of those texts. In the case of the meme concept, the people reading The Selfish Gene didn’t even have a pre-existing meaning for the term, as Dawkins introduced and defined it in that book. The same would be true for the people who first encountered the term in The Mind’s I and subsequent journalistic accounts. Continue reading “How Do We Account for the History of the Meme Concept?”

Dennett Upside Down Cake: Thinking About Language Evolution in the 21st Century

About two years ago Wintz placed a comment on Replicated Typo’s About page in which he lists several papers that make good background reading for someone new to the study of linguistic and cultural evolution. I’ve just blitzed my way through one of them, Language is a Complex Adaptive System (PDF) by Beckner et al (2009)*, and have selected some excerpts for comment.

The point of this exercise is to contrast the way things look to a young scholar starting out now with the way they would have looked to a scholar starting out back in the ancient days of the 1960s, which is when both Dennett and I started out (though he’s a few years older than I am). The obvious difference is that, for all practical purposes, there was no evolutionary study of language at the time. Historical linguistics, yes; evolutionary, no. So what I’m really contrasting is the way language looks now in view of evolutionary considerations and the way it looked back then in the wake of the so-called Chomsky revolution—which, of course, is still reverberating.**

Dennett’s thinking about cultural evolution, and memetics, is still grounded in the way things looked back then, the era of top-down, rule-based, hand-coded AI systems, also known as Good Old-Fashioned AI (GOFAI). In a recent interview he’s admitted that something was fundamentally wrong with that approach. He’s realized that individual neurons really cannot be treated as simple logical switches, but rather must be treated as quasi-autonomous sources of agency with some internal complexity. Alas, he doesn’t quite know what to do about it (I discuss this interview in Watch Out, Dan Dennett, Your Mind’s Changing Up on You!). I’m certainly not going to claim that I’ve got it figured out, I don’t. Nor am I aware of anyone that makes such a claim. But a number of us have been operating from assumptions quite different from those embodied in GOFAI and Language is a Complex Adaptive System gives a good précis of how the world looks from those different assumptions. Continue reading “Dennett Upside Down Cake: Thinking About Language Evolution in the 21st Century”

Bleg: Do Memes Matter to You?

For those doing or training to do academic research on linguistic and/or cultural evolution: Do memes matter to you?

I’ve got the impression that the issue that I’ve been chewing on recently, the appropriate account of memes of, if you prefer, the cultural analog of the biological gene, is mostly a theoretical one and has, so far, little bearing on empirical issues. However, I’ve also got the impression that most of the work on cultural evolution in the past decade or so has been empirical, either analysis of real-world data of one kind or another, or running simulations, and that the appropriate definition of meme doesn’t matter. You count what you can count. What matters is the quality of the raw data and the quality of the analysis.

If that is so, who cares about memes?

The Memetic Changeover: When and Why?

This is going to be quick, I hope, and dirty, I’m sure. What I’m up to is taking the first crude steps toward an argument about why putting memes in the head makes culture unintelligible.

Dawkins’ central insight, and the only reason to think about memes at all, is that, properly understood, properly cultural evolution is a regime where the beneficiary of successful cultural change (see my post, Roles in Cultural Interaction) is some kind of cultural entity rather than the organism that exhibits, uses, creates, that cultural entity. Call this the memetic regime. In gene-culture coevolution, by contrast, it is the organism that benefits from successful cultural change.

That is, in the regime of gene-culture coevolution, cultural inheritance is simply mode of behavioral inheritance that is different from, and more rapid, than ‘ordinary’ gene-mediated behavioral inheritance. All of animal culture is inherited in this regime. And this regime remains active in human life as well, though it is swamped by the memetic regime.

The question I’m asking is when, and why, in human prehistory did the memetic regime emerge? Stone tools emerge in the archeological record roughly 2.5 million years ago. Finely crafted hand axes–if that, indeed, is what they are–show up 1.5 million years ago. The shapes of these hand axes are conserved over 100s of thousands of years. They don’t change.

On the one hand, these artifacts indicate a level of craftsmanship beyond that we see in any animal. But they don’t show evidence of rapid and directed change. Do they exist fully within the regime of gene-cultural co-evolution? The question is not, of course, simply about the tools and axes themselves, but about the entire way of life in which they are embedded.

I don’t, of course, know the answer to that question. But if they are pre-memetic, then when did the memetic regime emerge, and why?

One obvious inflection point would be the emergence of language as we know it, which seems to have happened between 200,000 and 50,000 years ago. If that’s when the change happened, why? What is it about language that facilitated that change? Continue reading “The Memetic Changeover: When and Why?”