Dear Students!
After this pretty intense excursion
into the realm of modeling we will continue next week with Rick Dale,
who will try to convince us that even though language does not fossilize (in the early stages), we can play detectives and find cues
for mechanisms and constraints that might have been important in the
evolution of language. There will be a little bit of modeling involved
again, but after Anne's crash course last week you are all pros now - at
least you have a more concrete idea now what modeling is all about.
Our presenters for this time are:
Keith - Christiansen-Dale 2003
also Keith - Dale-Lupian 2012
Stefano - Lewontin 89
Abby - Lupian-Dale 2010
You will find the papers again on the UMdrive, as usual!
have fun!
Uli
This comment has been removed by the author.
ReplyDeleteThe linguistic niche hypothesis is intriguing. I have never thought of language that way, maybe I should share this article with my morphology friend. Although there left some room to critique about, Dale & Lupyan, in general, demonstrated that the morphological diversity of a population positively correlates with its population size both in the computer simulation and in the real world.
ReplyDeleteInterestingly, Dale & Lupyan seems to derive much of their thinking, at least in their choice of demographic variable, from the assumption that language evolves via transmission of cultural artifacts. For me, there are clear uses for these demographic models in testing hypotheses for linguistic transmission and change, as I see language can be part of a cultural product. My major problem with the claim is theoretical: they hasn’t ruled out other historical-evolutionary explanations for these patterns.
Their choice and use of google.com/trends, though I haven't finished studying the description of n-gram analysis, does not sound fair to me esp. with regard to how they collected data and how they compared American vs. British’s use of “lighted” and “lit”. The main reason is that they did not control for the use of “lit” and “lighted” in verb inflection conditions only. I found that in most of the time (both in British and American), “lighted” is used restrictively as an adjective from inflecting its verb form-- it only goes with other instrument or tool that can be lighted, whereas “lit” is used widely--the top searched “lit” is in the context of “au lit,” a phrase that is obviously a borrowing from French, and can sometimes be a typo from “au lait”. The derived “lit vs. lighted” ratio in AE is 1.86:1 whereas in BE, there is not sufficient “lighted” data to use. The results somehow look puzzling to me and maybe can be even interpreted differently.
I am also curious about why the authors chose morphology to study language evolution? In order to answer myself the question of “Would this social niche hypothesis apply to other linguistic behaviors?” I went across some literature on other linguistic evolution models (e.g. Atkinson’s
Phonemic inventory closely related to demographic conditions such as population density and size). I am wondering why the author used population size only in this modeling. Is it because it correlates most strongly with morphological diversity, or is it because of other reasons?
This comment has been removed by the author.
ReplyDeleteI have a few technical questions about the papers that I will be presenting on tomorrow.
DeleteMy first question is on the Dale & Lupyan (2012) paper. On page 3 the question is presented, "Why would complex morphological systems arise in languages to begin with"? The authors give an explanation, but I'm still not sure I understand. Here is my understanding. The authors are arguing that complex morphological systems exist because languages begin within an esoteric niche. Because of this esoteric niche, an infant is around a language with multiple morphological markings (for relational roles, number, evidentiality, etc), which are consistently used in the same way by the adults. These multiple morphological markings being used by adults allow the infant to map more unique experiences (goings-on) onto these morphologically distinct meanings, which reduces the infants risk of overgeneralizing grammatical rules.
Is my understanding accurate?
My second question is on the same paper,on page 5 section 2.1. Here the authors are describing the computational model being used for the first simulation. The authors explain that,"...and when agent codes differ only by this much, they are deemed to be members of the same linguistic group, and will not change their bit-vector encoding" -- By this unchanged bit-vector encoding, are they saying that their hidden layer in the neural network is not changed?
I have another question on the same page. Why does D(Ai, Aj)≤ ϵ2 not bring about any modification of the agents' encoding? Why would't these agents try to meet in the middle? What is the assumption being made here?
The first paper we read "Language Evolution and Change" by Christiansen and Dale discussed the use of computational connectionist models in analyzing language evolution. They discuss ways in which models have been used to study the emergence of simple syntax, linguistic diversity, learning-based linguistic universals and linguistic change. What they found in each of these was a connection between language features, their changes and linguistic constraints that take place in the language training process.
ReplyDeleteThe two papers by Lupyan and Dale discuss the Linguistic Niche Hypothesis (LNH) which, simply put, proposes a connection between social and demographic environments and language structure. In a nutshell, the LNH hypothesizes that languages are shaped by learning biases such that a smaller social demographic is more likely to be shaped by the linguistic learning needs of its children and a larger social demographic will be shaped by the linguistic needs of its second-language learning adults as it spreads.
In Lewotin's paper we wade in to even deeper waters in his discussion of the evolution of human cognition and are inability to explain how such a complex and uniquely human feature was naturally selected over time. His section on "Linguistic Ability" is very similar to discussions we've had concerning previous papers on the evolution of non-human communication systems to human communication systems and their present similarities. After discussing at length the extent to which evolution can and cannot be explained in human cognition, Lewotin concludes by saying that the most important lesson his readers can take away from his paper is to dispell "the childish notion that
everything that is interesting about nature can be understood" (130). However, his discussion on the explanatory power of natural selection to investigate changes and features of human cognition support what the writers in the rest of our reading assignments are trying to do with the LNH. But, I suppose, at the end of the day, Lewotin would warn, that as simple as it may sometimes seem, it isn't.
The LNH isn't something I've ever heard of it, but it sounds plausible and makes sense. I would be interested in another serious of models that study child language-learning and whether or not the overspecified, more complex language actually empirically appear to be more effective and easier for them to learn than the more morphologically simplified languages. Obviously, they can learn them just fine, if there are native-born speakers of the larger, morphologically simplified languages. And I remember reading somewhere in one of the papers that the adverse seems to be true for adult-learners... that langues with more inflections and specifications are more difficult for them to learn than more morphologically simple languages.
I'll be very interested to here discussion in class on these papers and listen to Dr. Dale's presentation. Fascinating stuff.
Growing up in the frozen tundra of bratwurst, cheese, and beer known as Wisconsin, I suppose I was first peripherally exposed to the Language Niche Hypothesis when I discussing the various categories of snow during our frequent snowy winters. While I had only known of two categories—the category that meant I got off from school and the other where I would have to walk through it to get to get there—I remember the discussion often marveling at how Eskimos had nearly thirty words for different types snow. While I tended to think of my small Wisconsin town as being isolated enough that we should have developed our own vernacular for snow, this never seemed to happen. The four readings for this week discuss the aspects of social interaction and isolation as a factor in the lexical vs. inflective development within language. While I would have described a snowfall as snowblower snow vs. shovel snow vs. a “salt it and hope it melts on its own” snow, the difference proposed in these studies is that while my example illustrates a grammatical solution to describing the snow, in smaller population, all of the connotations would be built into the word itself—which I interpret as an inflective solution. I particularly liked that the Rick Dale articles approached the subject from a sociological and historical perspective, as well as utilizing neural network modeling. I believe that this type of approach takes in account is extremely effective in analyzing the subject from multiple angles and allows for each subject to establish a sort of checks and balance in the scientific process.
ReplyDeleteI enjoyed the article by Lewontin if not because I like the idea of questions that have no answers. The article outlines the departure from a purely genetic perspective on evolution of cognition to attempting to consider evolution as a combination of many factors. For example, Lewontin, is a proponent of natural selection and discussing how various environmental factors, such as in the example of the black moth, can effect evolution. I wish I had read this article before I read the Dale articles because it presented a comprehensive survey of the broader issues and perspectives involved with evolution of cognition.
My only question is how the spectral analysis software and neural networking software actually works….although I suspect that is a subject all within itself.
The Christiansen and Dale’s article mentioned word order in languages and the inconsistency of word order would result in the loss of the language. It’s said that English is a head-first language, while Mandarin is a head-final language. I found this statement interesting because I sometimes use head-first and sometimes head-first sentences in Mandarin. So I searched online and found this article http://naccl.osu.edu/sites/naccl.osu.edu/files/37_gao-q.pdf. The author indicated that the written form in Mandarin is more of the head-first order, while the spoken form is more of the head-final order. This tendency would probably result in learning difficulties for second language learners. I wonder if the modeling cares more about spoken form than written form. Does the written form matter in the model?
ReplyDeleteI also want to ask the same questions from last week: the value of the computational modeling. It’s my second time reading articles about computational modeling, so I am not sure if my understanding about modeling is right: The computational modeling simulates future or past phenomena, so after modeling, do you have implication for others. For example, a family speaking different languages immigrates to the U.S. To interact with people in America, they mostly speak English. The second generation can probably use very simple words at home and mostly speak English. The third generation is more likely to speak English only. Is there any model that simulates language changes and indicate “how” to preserve native languages of immigrants? The same question may be applicable to other field because sometimes I was puzzled after reading certain theories. What’s next after knowing this theory?
After reading both Lupyan & Dale (2010) and Dale Lupyan (2012), there seems to be one main question that remains unanswered. In some communities where the language is spoken by only a few speakers, why is the nature of the language often complex? Dale & Lupyan (2012) suggests that although complex morphological language systems are not common in the exoteric niche due to the learning difficulty they pose for adults, the same complex systems may ironically facilitate infant language acquisition. In other words, some morphological markings in complex systems may appear redundant; however, what appears to be redundant information to adult learners, may actually provide infants with additional cues to assist language learning. I find this to be an interesting way to address why some small communities can have such complex language systems. As the authors have noted it seems to be a paradox in that the language constructions posing learning difficulties in adults may actually benefit learning in infants. I often wonder though if one can predict the survival rate for complex language systems that are in small communities. The researchers propose that the more common and popular the language, the simpler its construction to facilitate its survival. Therefore, I would infer that a non-popular complex language system would not survive as long as popular simpler language systems.
ReplyDeleteSince I’m not sure I will have time to talk about this during my presentation tomorrow, I would like to comment on a consideration that Lewontin makes and that seems relevant to what has been discussed with the speakers of the seminar. Lewontin’s arguments don’t sound always very convincing to me, but this is a good point. We can formulate it simply by saying: Organisms, not traits, evolve.
ReplyDeleteAt Andrew Olney’s talk people were talking about the “evolution of language” and were saying things like “language and the brain evolve together”. These expressions may sound very intuitive, but I think they need to be clarified if we don’t want to accept unjustified assumption. Do we mean by “evolution of language” the evolution of the English language from the Middle Ages till today? In this sense, language continually evolves, today as well as in any other time of the past. Or rather, do we mean that language evolves from something other than language? But in this case we cannot simply say “language evolves”, we should say, “language arises”. To talk, without this qualification, about the evolution of language implies that there is the same thing, i.e. language, that is present in the ancestral species and in the present one, the only difference being that in the present species language has a more complicated form. This is, however, an unwarranted assumption because when we investigate the state of any particular trait in ancestors, we should keep in mind that it is possible that the trait simply does not exist in the ancestors and that it is a novelty. As Lewontin suggests, we should avoid the “temptation […] to stretch the description of the trait so that at least some rudimentary manifestation can be seen in ancestors”. When we investigate the evolution of a trait such as language, these are the two questions to be asked: How has the trait come into existence and changed? What are its effects on evolution?