 Okay, welcome everyone. My name is Joey Levestre and I'm a post-sector fellow at SOAS University of London, and thanks everyone for joining this linguistics webinar this morning or this evening depending on where you are. We have two speakers for today's webinar. One is Lindell Bromheim, who's probably the first biologist to speak at SOAS linguistics webinar. She's affiliated with the Research School of Biology at the Australian National University, and the Macro Evolution and Macro Ecology Lab where they study macro evolutionary patterns and mechanisms. Lindell has a long list of publications and many awards, including the Eureka Prize for interdisciplinary scientific research, which was one with other people in collaboration with our other speaker Felicity Meakins, who is a linguist at the ARC Center of Excellence for Dynamics of Languages and the School of Languages and Languages and Cultures at the University of Queensland. Felicity is well known for describing languages of Northern Australia, including working at four dictionaries, two grammars, two ethnobiologies, and was recently recognized with the Linguistics Society of America's Kenneth Hale Award for Linguistic Fieldwork. I also know Felicity's work through her collaboration with Jennifer Green and Fanny Turpin on a textbook understanding linguistic fieldwork, which I would recommend to anyone teaching fieldwork. But I invited both of these speakers here today because of their work on a paper in Nature Ecology and Evolution called Global Predictors of Language, Engagement, and the Future of Linguistic Diversity. So I look forward to hearing more about that research and related topics from them today. So with that, let me hand it over to you to speak. Thank you so much for joining us today. Thanks very much. Now, how do I get the shed screen? Can everyone see the slides? There should be that green icon again. You'd have to click that again to go through that. Right. Is that working? Yes. Right. And this is a picture of Malonglo River, which runs through Ngunnawal Country as a marker for me to acknowledge the traditional owners of the land that I'm coming to you from, which is Ngunnawal Country, the area surrounding Canberra, the Australia's national capital. So I pay my respects to the traditional elders, the traditional owners, the Ngunnawal Ngambri Ngago people. So, as has been pointed out, it might be a bit unusual to have a biologist speaking to a linguistics series. I'm going to start today with looking at broad scale patterns of diversity and endangerment and then Felicity is going to follow on with some micro level stuff on changing populations. But we actually are following in a long tradition of biologists and linguists working together. This goes back a long way, particularly to the start of evolutionary biology. Darwin famously recognised that there are many similarities between language evolution and biological evolution, and very often we might ask the same questions of both or use the same tools to understand both. And in fact, Darwin's work inspired August Schleicher who'd already been working on language evolution. This is his groundbreaking Indo-European phylogeny. And like Darwin, many of the early evolutionary biologists actually use language evolution as an exemplar to make the Darwinian hypothesis more convincing. One of the problems with evolution of species is it's so slow that you don't really see it happening. And so many people like Darwin and Lyle and others use language change as a convincing case for evolution, and indeed Schleicher here is saying that in many cases language evolution will form a better way to demonstrate the Darwinian hypothesis than species will. So biologists and linguists have always had this common ground recognising common principles and we can also share some useful analytical tools. So it's certainly right for interdisciplinary exchange. And the reason that I got involved in language evolution studies is because I found that often as a biologist I was asking similar kinds of questions to those interested in language change might ask. For example, how does individual level variation contribute to population change? What influences rates and patterns of change? Why are there some hotspots of diversity? And in particular biologists have spent decades developing analytical tools to answer these kinds of questions because very often the way you approach these analytically is actually a lot more complicated than it first appears. And today we're going to illustrate this with some population models of change over time which Felicity will talk about and then I'm going to look at these more macro ecological and macro evolutionary comparative methods. And the reason we use biological tools for these is because it's easy to get misled if you don't use the right analysis. So take for example the very broad scale diversity of languages over the globe. Now if you show this map of number of languages per unit area to a biologist it would be instantly recognisable to them because it looks like a map of biodiversity. So for example if you plotted bird species diversity over the globe, obviously it's not exactly the same but you'll see that the areas of high diversity for species will tend to also be areas of high diversity for languages. So why is that? If we want to ask questions about what's causing these broad scale patterns we do need a fairly sophisticated analytical toolkit otherwise we can be led astray by non-causal relationships. And to illustrate some of the possible pitfalls, let's consider one hypothesis that has been put forward linking biodiversity and language diversity which is that biodiversity causes language diversity. Sorry I'm having trouble with my screen. For a particular kind of biodiversity which is parasite diversity. Because parasites show the same sorts of biodiversity patterns as other species. So what are parasites got to do with language diversity? Well a hypothesis published by a number of people including Randy Thornhill and Corey Fincher is that if you live in an area with a high parasite load well populations that will limit their contact with their neighbors will also limit the chance of infection from neighbors. And therefore this will promulgate a cultural isolation which will lead to language diversity through isolation of communities and you can see that if you plot the number of languages per country against the number of infectious diseases you get this striking relationship. But the problem is that parasites like most species diversity show a latitudinal diversity gradient there's more of them in the tropics and languages also have a latitudinal diversity gradient and of course anything with a latitudinal diversity gradient will correlate with everything else that varies with latitude. And so it's actually very easy to generate these significant correlations because so many variables correlate together. Now it's not just a matter of throwing latitude into the analysis or something like that. And the next example I'll show you will hopefully illustrate that problem. Now this example may not immediately be obvious why it's connected to language diversity but stay with me because it will connect it eventually. Right so why is it that when we think about spicy food we associated with hot countries whereas cold countries often have blender food. Well the hypothesis that was published a few decades ago was that if you live in an area with high temperature, then you have high risk of foodborne infection. And so any culture that adds antimicrobials to their cooking will do rather better and many spices have antimicrobial properties. Okay so that's all well and good and indeed if you collect recipe data from around the world and you add up the average number of spices per recipe you do indeed see this very clear relationship. However, if you ever see a graph where the data points are countries or areas or cultures plotted against some kind of environmental thing. The alarm bells should ring why because you can clearly see from this that there's clusters on this graph related cultures cluster together nearby countries cluster together. And in fact that's all you need to generate this relationship the fact that a near your nearby neighbors tend to have similar cultures similar cuisines and similar environments to you is enough to generate this relationship. And if you take into account the relatedness between cultures and the proximity between them. There is no significant association with temperature and spice above and beyond what you get for free just by the fact that nearby cuisines tend to be more similar. So there's no significant relationship between temperature and spices nor is there between spice use and parasite load. However, there are some significant correlates of spice use above and beyond the relationship you get for free from close relatives and near neighbors. For example, spice use is significantly related to the number of per capita deaths on the road and this is important because it's already been shown that linguistic diversity is related to road accidents. So you can see we're connecting these things up now language diversity is related to spice use. So we have a new hypothesis which is if you live in a hot country, you eat a lot of spicy food it causes you to have traffic accidents and somehow this generates language diversity. Okay, so obviously this is a silly example. But the reason I put it in here is because it's actually very, very easy to generate significant correlations between cultural traits, linguistic diversity, environmental traits, economic traits, all of these things correlate together. So the reason I'm putting these examples up is to show you that we can't use simple analysis because if we do we will get led astray by non causal relationships. So we have to use a slightly more sophisticated analytical toolkit. So here's another example. Which is that one hypothesis that's been put forward for I'm sorry I can see a hand raised there. Does someone want to ask a question. I don't have the chat so it's in the. Shall I go ahead and. Just make time for questions at the end. Yeah. Yeah. Okay, so we'll leave that one to the end. So anyway, here's another example of a hypothesis for language diversity and this is that if you live somewhere with a long growing season so you can grow or gather food for more of the year. Then the hypothesis is that smaller groups can be self sufficient. They can exist in a small area that don't need to rely on the neighbors. Therefore you can jam more independent cultures into an area and this leads to language diversity. But again, whenever you see a graph where there's countries plotted against some environmental variable alarm bells should ring again you can see here there's very strong geographic and cultural clustering. Why is this a problem because we can imagine there's a whole lot of things for example that these countries up in this top end of the graph the Melanesian countries differ in a whole lot of ways, not just language diversity and climate. But many other aspects as well and they'll differ from the the cultures down the other end of the graph the Middle Eastern countries which have short growing seasons and low language diversity. So we can never treat countries or areas or grid cells as independent data sets, because neighboring countries and related cultures will be similar in so many different aspects that it will be very hard to tease apart causal relationships. I just one more example here to show you how that this applies even when you're using grid cells as opposed to countries. So New Guinea of course is an area of great language diversity it's also an area of high species diversity. And if you plot language diversity and species diversity on a map you see similar patterns but intriguingly, if you plot threatened species and threatened languages you get opposite patterns. So the threatened mammal species tend to be more concentrated in the highlands perhaps because there's a long period of agriculture and hunting at high density. Whereas threatened languages tend to be more on the low lands why well it could be malaria which is not tend doesn't tend to be in the highlands it could be the influence of colonizers on the low land areas. In any case there are these two distinct patterns mammals in the highlands threatened mammals in the highlands threatened languages in the lowlands. Now, if you take every one of these grid cells as an independent test of the association between threatened species and threatened languages, then every time you draw a grid from the lowlands, you'll find that it has high language loss, low mammal threat. Every time you take a square from the highlands you'll find the opposite and so you will generate significant relationship just by sampling the same observation over and over and over again, even if there's actually no connection between the two. And in fact when you take into account the spatial connections between the grid cells, there is no relationship. So these examples are just there to illustrate that when we look for patterns of diversity or patterns of endangerment, we have to include as many possible covariating features as we can. We've got to account for the fact that neighbors will tend to be more similar to each other, and we have to account for the relatedness between cultures. So taking these into account, we wanted to start by asking what drives global patterns of language diversity because we've got to understand the patterns of language diversity before we can tackle patterns of language endangerment. And of course we have to control for covariation proximity and relatedness. So we gathered variables that represent global variables that represent the ease of movement across the landscape. So altitude, roughness, river density. We gathered variables that are climatic variables known to correlate with either species diversity or language diversity. We've got variables relating to the size of speaker populations and also the population density of every point on the globe, and also indicators of biodiversity. So well described species groups that allow us to map biodiversity. I'm happy to talk about the analysis later on, but for the sake of expedience, I'll go straight to the results. We do indeed find support for the ecological risk hypothesis because mean growing season is the best correlate of language diversity of all of those variables we tested. But it doesn't seem to be because all populations are smaller in areas with mean growing season, but the minimum population size of a speaker group is lower. So it suggests that a long growing season allows smaller groups to persist. Now surprisingly, we didn't find any support for the variables representing isolation, the difficulty of movement over the landscape, except for river density. Now, river density has previously been suggested as a universal driver of language diversity, the idea being that it chops up groups. However, we didn't find any association between average population size and river density, but again between the minimum viable population size and river density. So it suggests that rivers aren't acting to chop populations up, but like growing season are acting as a resource to allow smaller populations to persist. Okay, so using these kinds of analyses where we allow for relatedness and proximity, we can show that there's no direct connection between biodiversity and language diversity. Instead, it's an indirect link because both of them are associated with the same sorts of environments. So now that we have the patterns of diversity, what can we learn about language endangerment and loss. So there's around 7000 spoken languages in the world, and various estimates are that around half of them are endangered. And to put this in some kind of context, you know, about 40% of amphibian species, about a quarter of mammal species, and about 14% of birds, although I think that's a bit higher now, I consider endangered. So if you plot endangered languages on a map, you can see that there are some areas with many more endangered languages. But also, these tend to correlate with the areas of high diversity. That's not surprising. If you've got more languages, if there's smaller speaker populations, if there's more of them per unit area, you're going to have more endangered populations, more endangered languages. So instead of you plot the proportion of languages in a given grid cell that are endangered, you see a very different pattern. And here you can see there's very strong areas of loss around the globe, most notably North America and Australia, but others as well. And that tells us that there's some effects on language diversity that affect many different languages in an area. So we want to explore in as much as we can what some of those variables are. Obviously we can't include all possible influences on language, evolution, language endangerment, but we're trying to get a general picture. So we constructed a database of about six and a half thousand spoken L1 languages. We were not able to include signed languages because there isn't enough information, comparable global information on them. I'd be happy to discuss that afterwards. For a variety of reasons, we use the eGids language scale, which is based on intergenerational transmission. And we gathered as many predictive variables as we could. It has to be globally available language data for representing the transmission of language to children, representing the factors that cause people to shift from using the L1 language to another language. And aspects of language policy as well as the correlates of environment and environmental change. So an overview of these variables. We're gathering information on each language where it is how many speakers it has on the kind of landscape it exists in to represent movement of the landscape, whether it's in a country with that has one of the major world languages as an official language. Environmental variables that we know correlate with language diversity. Variables representing the degree of biodiversity loss and also land use such as built environments or agricultural land and the change in those environments over time. We have socio economic markers such as GDP and life expectancy and as much as possible we gathered information on educational systems so this is quite hard to get globally comparable data on. We also use language distribution data to describe how many other languages each language comes into contact with and how many other threatened languages there are in the area. So to cut a long story short, we identified a number of globally significant factors that correlate with language endangerment over the whole world. There's also many of the factors are regionally significant as you'd expect you're not expecting the same patterns of endangerment over the whole world. Now I'm not going to talk about those regional factors which I think need a lot more analysis but just to give you some examples. So for example, temperature seasonality climate correlates with language endangerment in Europe, whereas cropland correlates with language endangerment in West Africa. So there are regional differences, but today I just want to run through some of those globally significant predictors. So not surprisingly the greatest predictive endangerment is whether the language is being learned by children how many L1 speakers there are. But just to point out that L1 speaker numbers isn't the whole story with language endangerment. You can have languages with a large number of L1 speakers that are endangered because they're not being actively learned by children. You can have very small languages with very few L1 speakers that are not endangered because they are actively being learned by children. So here's some examples of languages that are spoken really only in one or two villages in New Guinea and yet are strong and vital because children are learning them. So one of the other predictors around the globe of language endangerment is road density. You might think, well, this is just a stands for economic development and change in the environment, but we don't think it does because we've got a lot of other variables that represent economic developments such as GDP, shifting urban population. And those ones are not significant predictors of language endangerment, but road density is so we think this probably represents population movement. The roads connect previously isolated communities to say regional centres, they allow the movement of people for employment or school. And this might be why areas of road density have higher rates of language endangerment. It's not just about language contact though, because the number of autochthonous languages that are in contact with the language is typically associated with language endangerment, but it's actually a negative association. The more languages in contact, the less endangered a language seems to be. So simply being in contact with other languages is not in itself a threatening process. However, one of the predictors of language endangerment is how many other languages are endangered in the same area. So there are clearly area wide threats to language diversity. One of those threats that covers where there's a shared factor over a whole area is the average years of schooling. So languages in countries with higher average years of schooling have greater rates of language endangerment. Now, I just want to emphasise we're not saying here that education is bad or that education necessarily has to endanger languages, but clearly the fact that there is a correlation between years of schooling and Indigenous language loss shows that in some cases the positive aspects of education are coming at a cost of Indigenous language vitality. And if you think this is a slightly odd result, it is actually backed up by micro level scale, community level scales and in fact Felicity will talk about this in the Australian context in the second half of this talk. So many of these global original predictors are things that will actually change over time. Some of them we can't predict like education policy, but some we can. For example, we can predict if a language is not being learned by children, then when the current L1 speakers have died, there will be no more L1 speakers unless something changes. So we can use the age distribution of speakers to predict that demographic shift in L1 speakers. We can use current patterns of land use change to predict future land use change, and we can use models of climate change to predict change environmental variables. So when we do that we move the world forward 40 years and then 80 years and say how many languages do we expect to be endangered or to have been to have gone silent. We come up with the fairly broad brush but depressing statistic that we're likely to see a tripling of language loss, even just in the next generation. And in particular, our broad estimate would be obviously there's no way that you can come up with precise estimates but our broad guess would be that we could lose 1500 languages which might cease to have L1 speakers by the end of the century. Now most of these currently have little or no documentation, but they do most of these currently have living L1 speakers. So we still have expert knowledge holders alive for many of these languages. So it's not too late to act to intervene to support communities to lessen this fairly horrific future picture of language loss if we don't do enough to support communities to encourage child learners. Okay I will now hand over to Felicity who I thought was coming to you from the Brisbane River but is actually Yarra side I think. Yep, no that's right so I'm actually on Wurundjeri country in Melbourne. But many of you might know that we've just had a very large flood in Brisbane so looking at a picture of the Brisbane River is quite relevant at this particular time. So I'm going to zoom in now so Lindell's talk very broadly about language endangerment across the world and what I want to do is just look at one specific situation which Lendell and I in collaboration with Jacques and a gringy collaborator Cassandra algae have looked at. So this is a particular case, which I'll describe in a minute, which is gringy which is spoken in Australia. So what we're looking at is a case study of language shift across three generations in a single speech community this is the gringy speech community to look at which social factors contribute to language endangerment. And we're also going to look at whether all language feeds features are equally prone to loss so we know for instance from the literature that productions generally lost faster than comprehension so speak people understand a language longer than they speak it. We understand from the literature that grammar is more susceptible to language change than words and nouns are easily more easily adoptable than verbs so in cases of for instance borrowing or code switching. So we're going to take a multi varied approach so this means looking at more than one variable at a time and use three different methods for revealing language patterns and language shift and change. So we use discriminant correspondence analysis DCA is these aren't specific to biology but you know we're still talking about collaborations between biologists linguists. But we're interested in this case and looking at the degree to which speakers from different groups which are defined by particular social factors have distinct language use. These mixed models which quite a few linguists will be familiar with so this isn't specific again to biology, but to look at how these social factors influence patterns of change, and then we're interested in using right Fisher models so these are specific to this situation. Genetics in this particular methods that sharp bar has. I guess specified for languages to compare the relative rates of gain and loss of language variants within speaker populations over time. So based on Northern Australia to give a bit of background for people who don't know the situation in post what in colonial Australia let's say there were 300 to 400 languages that were spoken before colonization. There's around 14 now that have been spoken continuously since colonization is a lot of language revival that's going on and some of that is pretty extraordinary actually. And the other thing that's going on is a number of newly emerged hybrid varieties. So one of those endangered languages is one that I've worked on for the past years gringy. Okay, so we're in Northern Australia. Let me know if there's still problems. So gringy is quite endangered there's an English based career language which has spoken right across Northern Australia. Many people indigenous people within Northern Australia shifted to this particular Creole language for Gringy people there's a new language which has emerged from this which is called Gringy Creole which language is referred to as a mixed language. So you can see this language change happening within the single language community across three generations so we have someone like topsy dot dodd seen here she's seeing songs woman in the community and someone I've worked extensively on. On the documentation of Gringy with the grammar the songs and the dictionary. She's bilingual she speaks Gringy and Creole she actually speaks other indigenous languages as well and she also codes which is between those languages. And that code switching has formed the basis of Gringy Creole the mixed language so her daughter and the next generation down Deborah speaks Gringy Creole, which is the mixed language coming from the code switching but speaks it with quite a lot of variable use her daughter now the carer who was off speaking age not just a baby still again speaks Gringy Creole but with much less variable use. So if we think about Gringy Creole we can think about it is the adoption of some Gringy words and grammar some Creole words and grammar. So some innovative features and some of those Creole features have actually become a little bit lighter or acrolectilizes we talk to say in contact with English and there's just quite a lot of variation. So if we look at this sentence you can see the big dog slept next to the table. You've got Creole features like that. And one which is magic title market clearly comes from one in English, you've got been a past tense market which comes from being in English table is quite obvious. You've also got Gringy words in there so you've got what a wolf a dog mug and for sleep and some innovations so this innovation here you've got the combination of a locative marker longer which comes from a long, ultimately, which is from Creole. It's a Gringy locative marker but this double mark construction makes an innovative construction, but then a Gringy Creole speaker might equally say a sentence like bigger just which is a heavier kind of Creole word for big. And they might just use the Gringy locative marker for instance without that Creole locative proposition. So the corpus that we're drawing from is the largest corpus of an Australian language to date. The data set that we've extracted from this corpus is over 20,000 data points we've got 78 people who are from three generations who are also coded for the family that they come from, where they live in the community and their exposure to Gringy and their Western education that's a kind of proxy for their exposure to English. We've got 174 variables with a whole bunch of variants and if you put this in comparison for instance to a lot of sociolinguistic studies where there's often just single variables study, you can sort of get the idea of the extent of this study where we're studying 174 variables simultaneously. So we've got production variables versus comprehension variables so things that people have said versus comprehension tests that people have done lexical variables versus grammatical variables and noun variables versus variables to answer some of the questions that I'll get back to. So this particular intergenerational corpus partly comes from conversational data, but to ensure that speakers have had the opportunity to express lexical features or morphosyntactic features using their variant of choice we've also used picture prompt books. So, a lot of you might be familiar with frog stories for instance, and also director matcha tasks and I'll give you an example of one of these director matcha tasks, partly by way of introducing you to Gringy collaborator Cassandra algae who's in this video as well as Samantha Smala, who was one of the participants. Okay, so we had a lot of pictures which directors then gave sentences to, and I'm picking out this particular picture again of the dog sleeping next to the table. So we can express this using a Gringy case marker you can use a Creole preposition or you can use this innovative double marked construction, and in fact speakers do different things so some speakers use pattern A where they just use the case marker. Other speakers use pattern B where they just use the Creole preposition. Other speakers use pattern three where they sometimes use the Creole case marker and other times just use the Creole preposition. And some speakers use pattern four where they just use the Creole preposition or they just use the double marking and these patterns are really important. Because they form the basis of what we've done to a lingua type. So recording which language variants and individual users for each language feature is what we call an individual's lingua type and it's an analogous to an individual's genotype. So genotype doesn't describe the whole genome and in a similar way, the lingua type just uses a broad sample of language features in a way that describes the sample of the ideolect of a speaker. So unlike classic variations sociolinguistic studies which just track one variable at a time the lingua type provides an unbiased sample as an individual's characteristic pattern of the language use at a particular point in time. And this allows us to study the process of the incorporation of particular language features from one language into another, which we see in this kind of language shift context. So getting back to our questions we're interested in what shapes language change are the particular kinds of patterns that are predictable and whether there's particular factors that increase language endangerment other ones which might preserve heritage languages. And we're also interested if you remember in the kinds of variants which are prone to loss of language contact. So for instance whether people understand language longer than they speak a language whether they use words much longer for instance and they can use grammar from a language and whether they use a lot more nouns than they might use verbs in in language change situations. And again we're using DCA and mixed models, which are general statistical modeling but also right Fisher models that are specific to population genetics and biology which is part of the idea of this kind of interdisciplinary work. So let's look at the first analysis which is the discriminant correspondence analysis. And this particular analysis is allowing us to ask whether speakers from different groups have similar linguotypes. And if you have a look in this PC space, each speaker's position is determined by their linguotype and speakers who are closer to each other have similar linguotypes. And what we find is all the sort of social demographic factors that we've tested for have a significant influence on language change over time. But the largest influence is generation and this shows a very strong role of peers and shaping language variation. So we found that there were 36 variables that were shown to generate differentiate sorry generation one which are the oldest speakers from generation to the sort of middle speakers my generation sense and generation through the child speakers and then 51 speakers that differentiated these adult speakers from the child speakers here. Okay, so then another important question was whether social fact which which of those social factors affect patterns of variant use and for this we're using mixed models. So the most significant finding that we had from this was that formal schooling, which is conducted in English this is important is a significant factor on the loss of guarantee variants even amongst the older generation so even grandparents, who have this generation here with a higher level of education are more likely to lose gringy language variants so this is the G minus. And they're also less likely to use the G plus variants. So actually quite a significant finding probably less surprising using the mixed modeling is that exposure to gringy is a significant factor on the retention of gringy variants it's a bit of a no brainer but if you want to talk about that later we can talk about it. So now we're using right Fisher models which are specific to biology to determine which aspects of gringy and most vulnerable to loss and their relative rate of loss. So what's particularly interesting in this is that unsurprisingly people, the people's ability to speak particularly lexical and grammatical variables is lost 11 times faster than their ability to understand it. We know this from the literature and this particular finding supports literature. What doesn't support the literature which is perhaps more interesting and surprising is that it's not. These are non significant results is that grammar isn't lost significantly faster than lexicon and that nouns aren't adopted faster than verbs and this is kind of surprising particularly if you have a look at the code switching and borrowing literature. So, I want to continue a little bit with the right Fisher models because they're specific to biology. So, one of the questions that we're interested in as well. Yeah, there is language shift going on to creole why is there a greater uptake of some creole variants and colonization is the obvious answer to this and throughout our talk we don't want to diminish the devastating effects of colonization. So, when we drill down to it, we want to ask questions like, why is it that the creole form for the variable variable butterfly for instance, largely replaces the gringy form which is Mully Mully that the gringy form of the variable dog while ago which was seen before is really resilient. So one of the questions that we had which we're looking at within the literature is that morphological complexity for instance often is shown to be affected by language contact. So we see a significant reduction morphological complexity. Most of these studies usual single use single variants and what we're wanting to do with this study again is to use multiple variables that are chosen for the factor variation. And this avoids the issue of circular argumentation so picking variables that are undergoing simplification which then demonstrates the story you're trying to tell. And then we're using right Fisher models and what we're doing is evaluating whether the adoption of elements from either gringy or creole is completely random, or whether it's biased towards one of the parent languages and we've already seen that it's biased towards creole, or whether it's driven by morphological simplification. So if we go back to our example of the dog sitting next to the table we can code these variants variables for sorry these variants for morphological complexity and I can talk about how we coded these later if you're interested. But for now just to say that case in this case because it's spatial case was coded as being medium level complexity because it's not contextual case it's can it's inherent case. Sorry inherent flexion, the preposition is coded as low complexity because it's a single word it's not part of a complex word and double marking is actually sorry coded as meeting complexity as well not high complexity. Again I'll talk to, I can answer that question a bit later if you have some, some things you want to ask. So the first thing we found is not surprisingly there's a bias towards creole. So when you're looking at these plot this plot as it's emerging generation one, the oldest adults I'm going to reveal generation to the adults and then the children. What you should be looking for here is the proportion of orange which is gringy changing to the proportion of blue which is creole. So you can see that as you go to the adult generation now so we're getting younger you see proportionately more creole which is blue. And again, as the child generation we see more blue which is again creole so there's a clear bias towards creole both across the lexicon which is this strip here and then the grammar which is this strip here. Okay, more interestingly, what we find is there's no bias towards simple variants so complex creole variants are frequently adopted rather than their simple gringy alternatives. Simple creole variants are almost no like are also no more likely to be adopted into the contact language gringy creole than complex creole variants. And complex gringy variants also have a greater rate of adoption than both simple and highly complex creole variants which is a slightly less straightforward story. Okay, so in general there's no bias towards simplicity. So just leaving it here and leaving that question a little bit open, I just want to conclude in saying that linguistics has had a long history of using biological methods to map both macro level linguistic variation which Lyndall has talked about for instance in diversification and all of the pitfalls that have been associated with that a little bit more recently. Right Fisher evolutionary models have been used in population genetics to map biological variation. And the work that we've doing is really one of the first applications of right Fisher evolutionary models to linguistic data. So using this kind of modeling does require linguists to scale up and use multiple variables and their expression. So to use variants to map across an entire language so this is our concept of a linguistic type and this is a little bit different from traditionally what variations linguists have done that we want when we do use multiple variants we can avoid some of the pitfalls of some of the theorizing on language change and endangerment when we just use individual variables. So it's been a long time since this generalized claim that simplification occurs through contact. So I'm going to leave it there so that we've got time for questions since I think we've only got about 10 minutes left for questions but I'll just leave you with. The person who's the person who's done most of the mathematical work with this joint work and also Cassandra algae who's been really crucial in the collection of the grungy data for this particular project. So I'll stop sharing now and we're back to the audience. Okay, thank you for this to do for an excellent talk very interesting very informative very well organized and excellent research you've been doing obviously for many years behind the scenes so it's great to see some of the pieces of that. We've got time for discussion so you can use the raise hand function in zoom if you'd like you could also just know in the zoom chat if you prefer they'd like to ask a question by writing the word question or put a queue or you can also write out your question in zoom so is there anyone that would like to get us started. Okay, go ahead, Stella. Thank you. It's really really interesting. I don't have English adjectives to express how interesting this was, but I'm intrigued by the fact that the simplicity did not work out as expected. I guess that what is simple for a linguist is not simple for an 80 speaker. Yeah, so what we did with this particular analysis was to specifically address the question of morphological simplicity so morphological complexity has been looked at from the point of view of both absolute morphological complexity and relative morphological complexity. I think what you're talking about from a speaker's point of view is relative morphological complexity, which is indeed a really interesting question, and not one that we could answer with the data that we had so what we looked at was absolute linguistic complexity so we're looking at, for instance, the participation of morphines within paradigms, the participation of morphines as both being inherent inflection, for instance, whether they were independent words whether they were part of complex words that sort of thing. I think that's a really good question but what we wanted to do was not address specifically what is morphological complexity because as you as you note that's been raised a lot in the literature, but to use existing criteria for absolute morphological complexity and see whether that related to language change and endangerment. So I hope that answers your question. Yeah, thank you. And I'd just like to add in there that from a biologist's point of view, I'm pretty sure it's the same for both biologists and linguists that the easiest way to start a fight as mentioned the word complexity, that there's never going to be something that we all agree on. And what we've tried to do with all of these analyses we've been talking about is we've never got a perfect solution. But what we've tried to do is find something that's tractable, something we can do so in none of the analysis we look at can we possibly give the whole answer, but we try to find something that we can measure something we can put a quantitative value to so we can begin to answer that question, but it's never satisfying in the sense that we never managed to capture all of the things that we want to in any of these analyses but we try to make a start on it. Jonathan, did you have a question. Yeah, I did but I'm nearly out of time so I kind of hesitated and asking but I will anyway just okay. So I really enjoy this paper as well and I suppose I was a hinge on one of the last things that Felicity you were saying that the pop up variation of sociolinguistics is well warranted here. The issue of that particular paradigm right is that they're trying to understand variation and change through social embedding and evaluation and, you know, things like indexicality and invariably you end up having to zoom in quite significantly, and you lose a lot of this macro detail because I mean my half baked questions you would have been do you think that kind of paradigms approach is useful in what you're trying to do in it in this particular way, understanding, you know, linguistic diversity and loss. Yeah, absolutely I think these kinds of approaches, you know have to operate in parallel so I don't mean to say, you know to chuck the baby out with the bathwater. But just to say that what happens often in studies of language change is that we particular we particular pick sorry a particular feature, and we build a whole story around that feature and often we could pick that particular feature because we can see that there's something going on. And what we're trying to do with this is just pick linguistic variables for the fact of being variables as opposed to particular directions of change and what we're trying to do is come up with a more neutral model of this. But I guess that then what you want to do is find the things that are really doing something quite interesting. And then down on those, and that's actually already what various and so selling this are doing. So yeah, like I say we don't want to chuck the baby out with the bathwater, we're just trying to sort of pull back from trying to make bigger statements and global statements about language change just based on single variables. Okay, thank you very much. I have my own questions but if anybody else has a question or comment or give a space for that. I wanted to ask a question about the first study and I'm wondering if you can count on the predicting language diversity and loss. If you can comment on the difference between your approach to this and the previous approaches, like the previous study, the catalog of danger languages, or the assignments of Lewis study predicting language loss. I think they predicted slightly even higher numbers of language loss over the next century compared to yours. Do you comment on some of the differences, both the methodology and results. I think there's a number of differences. And we've tried to say what can we learn from the way that people have studied species endangerment that's useful now it's never going to be exactly the same thing and you end up doing something quite different. But we're trying to for a start avoid circularity. So this is why we're not using L one speaker population size or distribution as markers of endangerment so some previous studies have used those as their measure of endangerment but that's a bit circular. And so we tried to get away from that by using an endangerment scale. But we, you also need to take diversity patterns into account because if you remember I put up a map that shows that the greatest losses of language are the places where they're the most languages so we need to look for patterns above and beyond that. But I think in some ways we're looking for different things as well. We're looking for the correlates that allows to predict language loss. As far as the actual numbers I wouldn't take any particular set of numbers as being particularly robust because obviously we can't get a forward prediction that we're trying to get a probabilistic estimate. So differences in numbers of endangered languages I think are not something worth getting exercised about because that's going to change if you change the endangerment scale. The L one speaker population sizes can't all be accurate because they're constantly changing so at any given time we can only take a snapshot some languages are knowing better than others. So I guess what distinguishes our approach is the number of variables that we've included. I'm not aware of any other language endangerment studies that used anything more than say L one speaker population size or range. So we've tried to include as many factors as we can obviously we can't include everything because we can only put in things that we can get globally comparable data for that's particularly frustrating for things like education policy. Or for historical factors such as colonization or wars or massacres. The other thing that distinguishes us is there has never been a study that is allowed for the fact that similar languages will be similar in many things nearby languages will be similar in many things so you have to take the amount of similarity you get just from being nearby and related you have to look into account to look for patterns above and beyond that. So sorry, but I probably rambled on a bit to that answer the question. Yeah, no definitely that's great. Someone else has a question do feel free to raise your hand and put it in the chat but just on a related note to that. If I'm correct I think your study relies on the e-gids ratings pretty heavily. 6a I think has been stated to be sort of the default one where if they don't know they should put in the default one. So then by default you know if we don't have information and language gets marked as safe. So am I correct that that would sort of make your estimate even more conservative of an estimate because of that. Yeah, we've definitely earned on being conservative and we've at every point in the analysis we've done that so for example with our future predictions. We've only taken the languages that are already endangered and said how much worse situation will they be in future. We haven't allowed any of the languages that are currently stable to enter that. And so it has to be conservative. And the reason we do that is because we need to have a good basis for making those predictions. Now if we have an e-gid score of seven, what that's telling us is it's not being learned by children. So if you're seven and above, according to the information we have children are not learning it you're going to run out of L1 speakers we can use demographic information and just emphasize this is not proper demography. This is effectively back of the envelope demography we're using the age structure for regions we're using the number of L1 speakers, we're making very basic predictions so a proper demographer with proper information will do this in a much better way. But all we can do is these broad brush estimates and that is why I would emphasize that I wouldn't focus on the numbers. But the broad patterns and the areas of concern and we all know that without intervention we are going to see a lot of languages go quiet, we wouldn't predict exactly which ones. But we can say it's going to be a lot unless we do something and I don't think that message has really got through enough to people outside the endangered language community yet just what kind of language loss crisis where we're facing. So I hope that this helps bring a little bit of attention to the problem we could face in future if we don't do something about it. Julia you have a hand raised. Yes, thank you. Sorry, sorry. Yeah, thank you kind of a kind of answered really what I was going to say anyway thank you. Basically about it being broad brush. Because practically every linguist I know it says oh well well, ethylogue is great but for the language I'm studying it's not correct. So I think you can really only take it as you say it like as a broad overview kind of overall indicator. As you say that a lot of the figures are I think are pretty much back of the envelope there's no reliable, basically no reliable definition what is a speaker for example how good you have to be. For example that's one of the and we know that the L one speaker numbers aren't great but what we're doing is we're using the best that we can get our hands on. And it's clearly not it's not good enough but it's the best that we have, and we can go through the database and find examples where it doesn't look right and there's a few languages where we adjusted it based on other sources. It's simply, unfortunately one of the costs of doing a global analysis is you're going to lose experts understanding of each of those languages, because it would take an expert to look at every single one of them to make that correct. So we hope we're getting a broad picture but I would never use this sort of analysis to predict the trajectory of any particular language. So it's also this geographic biases so for example the Australian languages I think could really do with a lot of improving the the information. Yeah, absolutely and that's partly why we're also pairing this brutal work with work on at least one and we're expanding this to other individual language situations and there's another question in the. That from Lidiana about whether speakers conscious about speakers consciousness so are they aware of the fact that they're mixing complexifying language. So just a quick answer to that which is a preview of work that we're doing currently at the moment we are looking at social linguistic salience as a factor in the gringy creole situation about whether people are choosing particular gringy or creole variance over the question that's probably the the closest we can get to answering questions about consciousness, not so much around complexifying language but around variant choice so that's just a quick answer that question. Thank you for the city and George you have your hand up that maybe this will be our final question of the session. Okay thank you. Can everyone hear me. Yes, I just want to say I really, I enjoyed both talks and I'm very excited to read many of the papers. I want to go back to the point that we were discussing about, you know individual languages, the expert versus people want to do global studies and everything I'm very appreciative of all the studies. Linda that you that you've conducted. I want to discuss the, the e get scale, just a little bit the endangerment scale. So we are currently. So I do work in Indonesia and we're currently challenging the claim that transmission to children is a marker of kind of endangerment. And one of the reasons for that is that we've found in the island of alor in Indonesia, as well as other parts of Indonesia and more broadly in Malaysia. The concept of delayed vernacular production, which is that the language, the vernacular is not passed on to children, but the children acquire it in kind of post adolescence or early adulthood. And this is completely missing from the, you know, from the language endangerment scales and so on and we want to kind of find a way to incorporate models like that because we are finding. So I found it for a buoy by doing an in depth study. We also find it in passing in footnotes on various language grammars and so on. The language doesn't mean pass on children, but they do end up picking it up and becoming fluent speakers in adulthood. So we're looking at what effect I can have on endangerment and on language change. That's fantastically interesting. I'd love to learn more about that. And again, yeah, there is this very strong focus on L one speakers. And I think one of the reasons that there's a focus on L one speakers is because it's tractable. It's something that you can point to and say this is the first language, although even in some multilingual communities identifying exactly as your L one might not be straightforward either. And there's, again, we you'd emphasize language revitalization. I think there might have once been the idea that if you're not L one, you could never be, you know, an L one speaker and yet we're now looking at and especially in Australia. There's a really strong language revitalization movement. Now, of course, they are not going to exactly recreate the language that was spoken a couple of generations ago, but they're going to reclaim language that going to reclaim important elements of it. So I agree with you that there's this binary scale L one L two. And, and it can't represent everything on the ground. And in that case, we need a new way of measuring that and reflecting it in the endangerment status. Now I'll just say why we used e kids as opposed to the other scales that are around. We looked at all of them. And we, we mapped all of them. And of course, in most cases they, they agree with each other. The practical reason we used e kids is it's available for more, more languages. So for something like the UNESCO rating, they tend to only evaluating endangered languages and we for our analysis, we need a rating for every language, not just the endangered ones for something like a yes. Again, it's just available for fewer languages. And it also, I think they're starting to iron this out now, but there's also some issues with AS where within the language hierarchy where you have umbrella languages or dialects. The AS score sometimes gets. So you'll use the score from the higher level to apply to the lower level and things like that. So we just, we did investigate all of these things and we had to choose e kids because if we wanted to do a global scale it was really the only viable option we had, we would have just cut, we would have had to buy a sample of languages otherwise. That isn't to say there won't be a better way of doing in the future. So, I would hope that people look at the study we've done and go, Well, hang on, that's not right, because it's missed this, and then they go out and do it better. So I think we've done a better job than previous because we've brought more in, but obviously we haven't brought enough in. And obviously there's better ways of doing it. So I hope this is just a stepping stone and we get better analysis after this. Yeah, and just to add to that. If you look at individual language situations, the one that you're talking about in particular actually sounds very interesting so we've just had a study that's actually come out for Gringy Gringy Krill that looks at long acquisition. And so what we're trying to do is work out whether some features of the language are due to long acquisition or whether they're due to language shift. So that's trying to compare actual change over time, and also looking at snapshots at a speaker population so I'd be happy to send you that paper, and as Linda said, this is all about, you know, a situation where languages are being lost but actually in some really inspiring situations that are going on. I've just been down in Northern New South Wales talking to Gumbangia mob. They now have a first language acquisition school. I'm so inspired by the work that they're doing down there. And I'm not injured here and not injured you sorry in South Australia are doing similar things so there are L1 L2 speakers who are now emerging from its really severe language endangerment situation so I think we don't want to see this is, you know, a nail in the coffin that really throws support support at those places where there's enormous amounts of energy and to, you know, forestall the process of language silencing and really support languages where people are doing enormous amounts of work. Thank you George for the question. Thank you Lindell and Felicity for agreeing to be with us this morning this evening your time I know it's probably past work hours for you and staying extra late to be able to speak to us. Research obviously very interesting for those of us who study language from multiple different perspectives and has some real chance to have positive impact on the speakers of these indentured languages as well so very meaningful work in that sense. So thank you so much for the work you're doing and for sharing with us today. Thank you very much and please do get in touch if you've got any more questions or want to discuss this further we're very happy to talk about it. Yeah, absolutely. Thanks for your attention. Thank you everyone.