 Okay. All right. So that's the title of my talk. And this was sort of my PhD thesis. And I'll try to give you a sort of an overview of the thesis, basically. But I thought I should start off by giving you a short introduction into the field of non-worthomology. So when speakers integrate the pronunciation of a foreign word in their native language, they can do one of three different things. They can adopt foreign sounds as innovations, which means preserving them in their original form. Or they can adapt them, which happens more often than adoption. And that means changing them to fit the native language. And that might involve substituting foreign sounds with native sounds, but sometimes, very rarely, but still sometimes involves deleting them altogether. And so that's the sort of, just the very general, the three strategies that you can encounter in non-worthomology. About the history of this field, the first systematic studies of borrowing or loanwords are from the 1950s. And then roughly from that period until the early 1990s, there was an interest in the subject among scholars of language contact, language change, and bilingualism. And as a consequence of that, there was a broad approach. Both structural and non-structural factors were included, and both adaptation and adoption were treated as possible strategies. But there was relatively little interest in phonetic details. But then in the 19s, especially with the onset of constraint-based phonological theories, we saw, we see an increased specialization, and more phonologists and phoneticians working in this field. But also because of that, the approach gets narrower. There's almost only focus on structural factors, especially the role of perceptual salience is a very common theme. And adoption is largely neglected as an alternative to adaptation. And only very few researchers explicitly discussed the role of bilingualism in the borrowing process. But since roughly 2009, there have been some gauge DTCs, which suggest that maybe the approach might be broadening again. And there are some superficial mentions of second language acquisition and bilingualism. So this is sort of the general historical outline of what happened. And I sort of came into the field in this recent phase. And I tried to sort of include, incorporate second language acquisition in a more thorough way into loanword phonology. So there are a few established models of loanword phonology that predict and explain what happens in loanword integration. One, which is actually a much more broader model of structural borrowing, is Sally Thomas's borrowing scale. And that can be one of the areas that it can be applied to is loanwords. And because it's about borrowing, contacting this change, it basically when we apply it to loanwords, it focuses on adoption as the main strategy. But then in the 90s, with the sort of perceptual wave of studies, we get the perceptual stance model. And then we have the bit later, the phonological stance model. And then there's also PMAP, which isn't really a model of loanword phonology, but it's a phonological model that has been applied to loanwords, to among many other things. So I thought that I will exemplify how these models work through some examples, so that you can see how my model differs from these. So I thought we can take the English loanword, English word cool, as an example, which is a very common slang loanword in many languages, for example, in Swedish. And in English it's pronounced cool with a velarized L. But there's a phonological rule in English that velarizes the L in coded position, and it's analyzed as underline the non-velarized. So the velarized L is not phonemic in English. It's an allophone. And standard Swedish doesn't have the sound ul, it only has the clear L. And the majority of native standard Swedish speakers pronounce the loanword with a non-velarized L as in cool, as opposed to cool. So why do we have this substitution of the velarized L with the clear L in this instance? And how would established models of loanword phonology explain this? Well, we can first look at the perceptual stance model, which doesn't only focus on adaptation in perception, but it's the big thing with this model. So here we have the phonemic underlying form on the left in English. And then there's the phonological rule, legalization rule, which derives the phonetic surface form cool with the velarized L. And this model says that the input to loanword phonology is the phonetic surface form. And then, as you can see here, there's an inaccurate surface mapping of velarized L on non-velarized L. And this model would say, well, the mapping is inaccurate because this velarized L has low salience. But it doesn't really say how do we determine salience. It's determined on ad hoc basis. So you look at something and if it looks like there's adaptation in perception, especially if there's deletion, then people say, oh, it must be low salience. But it's not really explained what's low salience, how do we predict it, how do we measure it. And also the mapping is inaccurate because this model presupposes that the borrowers or the majority of users are monolingual and they set the norms. But again, what do we really know about the contact situation? Was that really the case? And usually, these studies don't provide much information on the contact situation. It's just assumed that the monolinguals are the norm setters. In contrast to this, looks at focus on adaptation in production, not in perception. So when we have, in the previous one, when we have adaptation in perception, some people speak of misperception. Foreign sounds are misperceived and that's reflected in the loan work conversation. PMAP says there is no misperception. But similar to the perceptual stance model, it says the input is the genetic surface form. So that's a commonality between those. But then it assumes that the mapping is always accurate. Okay. And there we can ask, how can we assume that it's always accurate? What do we know about the contact situation? Who are the borrowers? Who are the users? It doesn't really say much about that because it's not a full-fledged loan word phonology model. But it would say that this form is accurately mapped to the native language and also stored in the lexicon of the native language of Swedish in this case. And then during production, the duration of the output in Swedish, there is a constraint saying the velarized L is not salient. Therefore, it doesn't reach the surface. It's eliminated in production. And it's based on perceptual feedback. Okay. So how do they determine in this model what low salience is? Well, they actually go a little bit further. They look at market-ness and psychological facts. And they try to operate with universal perceptibility scales. So there's a little bit more backing in terms of how we determine those salience. The phonological stance model is quite different from these two. The biggest difference is that it says the input is not the phonetic surface form, but it's phonemic. It's the underlying form. And in English, in this instance, it's very decisive because if the input is the underlying form, we don't deal with the velarized L at all. Okay. So there's no velarized L in the input then. And then it says there's accurate phonemapping of the L. And that's because the borrowers necessarily must be bilingual and they set the norm. However, few of them there are around. Okay. And here it's sort of assumed that, you know, how this language contact arrives to begin with if there are no bilinguals who would do the borrowing and so on, which is, of course, true of many cases, but not all cases. So it's also a bit problematic to assume that. So are the borrowers always bilingual? Or are bilinguals always native-like so that they have accurate mapping and access to the underlying form and so on? How about non-bilingual users? So maybe a bilingual, bilinguals borrow the word, but then it spreads to the rest of the community where there are monolinguals. So what happens then doesn't really say anything about that. However, one positive thing about the phonological stance model is that it allows for adoption. Okay. The previous two models, they say that adaptation is the only outcome. This one says adoption is the possibility. And here I have another word. Again, the English word web in the sort of internet context, which also is a common loan word in many languages like in Swedish. So here we have matching underlying and surface forms and the input is phonemic. So that's goes into Swedish. And there's accurate mapping because the borrowers were bilingual, as this model says. And then this model goes further and makes a prediction about when adoption will occur. It says if there's low degree of community bilingualism, if there are few bilinguals around in the community, then it will be adaptive. It will come out as web. However, if there are many bilinguals around, I forgot to say that Swedish doesn't have a sound. But if there are a lot of bilinguals around, then we will ignore the fact that Swedish doesn't have good. We will introduce it as an innovation and we will end up with web in Swedish. But one thing it doesn't say is that we can have variation. Some people might say web, other people might say web, which is exactly the situation we have in Swedish. Both pronunciations are tested. Okay. Thomas's borrowing scale is quite different. It has much wider scope because it's about borrowing, structural borrowing in general. But if you look at the phonological side, she has these four categories of contact, casual, slightly more intense, more intense and intense, increasing intensity. These are the names she has given to them. They're very obscure, but and so what happens between when the intensity increases is that you get more bilinguals, the bilinguals are more fluent in the borrowing community and there are more social and ethical factors favorable of borrowing. She actually makes the prediction based on the data she has looked at, she and Kaufman, that in level one, there is no phonological adoption. So you only adopt words or borrow words, but you don't borrow structures. When it gets to level two, you can borrow segments. Okay. And it doesn't matter which segment. If it's a segment, it should be borrowable. And then when you get more intense or intense contact, you can also borrow super segmentals, prosodic structures and so on. And there are no differences between different prosodic levels or types of structures. Okay. So this is the borrowing scale applied to a loanword phenology. So coming to my data, in my 2011 study, I looked at historical loanwords from Arabic used by elites in the Ottoman Turkey. And Arabic was a medium of instruction in schools. And I retrieved the data from contemporary dictionaries of Turkish, how these words are pronounced today, but also did historical reconstruction of the pronunciation in the 18th century. And then the other type of data I have elicited contemporary borrowings from Swedish. I gave bilingual Turkish speakers in Sweden a translation task, a text in Swedish which contained proper names. And I asked them to translate that text into Turkish. And when they did, they of course, couldn't translate the proper names. And the proper names have the structures. I was interested in them. And I looked at how they pronounce them, both unsuffixed and in suffixed contexts. And Swedish is, Turkish is spoken as a heritage language among children of migrants from Turkey to Sweden. So the structure I looked at in this study was El. So that's why I used a cool example in the beginning. So it was El, clear El after a back vowel, like in the word El. Okay. So Arabic and Swedish have the clear El in all positions. That's the only type of El they have. Turkish El, I mean, there are a few exceptions in Arabic, but there are very few. Turkish El is realized as clear El after front vowels, but as valorized El after back vowels. So it's the same other forms as in as in English, but distribution is different. So we get in Turkish, we get words like we have a front vowel followed by the clear El, but also and where you have a back vowel followed by. So that's what native words are like in Turkish. So the results regarding the Arabic loan words is that in contemporary Turkish, in 86% of all cases, the clear quality of the El, the original clear quality of the El in Arabic is preserved. It's adopted. So the clear El is adopted into a new position in Turkish as an innovation. Okay. And it's very similar in Swedish as well. Here I have a place named Spudiksval and Astamal. There we have the same, but in 78% of all cases. So there's a very clear preference for adoption in both cases, adoption of the clear El. So what would be the models I showed you in the beginning predict about these data? Well, as I said before, neither the perceptual stance model nor P-map allow for adoption. So they would wrongly predict adaptation to valorized El. Okay. So the conclusion we can draw from that is that models of loan morphology need to allow for adoption as an alternative. The phonological stance model states that the likelihood of phonemic adoption, because adoption only phonics can be adopted, okay, because the input is phonemic. Phonemic adoption, the likelihood of phonemic adoption is proportional to the degree of community bilingual. Okay. So if you have few bilinguals, you will get the low degree of adoption. If you have many bilinguals, you get the higher, okay. They say it's proportion. But when we look at the Ottoman context, we have very low degree of community bilingualism because it's only elites who went to school who acquired Arabic. So then this model would predict that the clear El should not be adopted, which is wrong. It's not what we see. In the heritage bilingualism case in Sweden, since we have higher degree of community bilingualism, the El should be adopted. So in that case, it makes the right prediction. Thomas's borrowing scales, that states that segmental adoption, like the El case here, requires an intensity of contact of at least level two. And in the Ottoman case, we have level two. And in the Sweden case, we have level three. So it makes the right predictions in both cases. Okay. So Thomas's borrowing scale fairs best with these data. So coming to my model, I say that adoption will occur if the speaker has the necessary linguistic competence in the donor language to accurately produce its foreign structures. Okay. And that's very similar to what Thomas says. She says, you can't borrow what you don't know. Okay. So if you can't say who or me in a certain position, you can't borrow it either. So that's very common sense. But also, the speaker has to have sufficient social linguistic incentive to borrow such a structure as an innovation into his or her native language. Okay. So there are two main concepts that I use. When one or both of these are lacking, adaptation will occur instead, which may happen either in already in perception or in production, which is similar to what we saw with perceptual stance modeling came up. But then I go further in deeper into linguistic competence. And I say linguistic competence is determined by the proficiency of the speaker in the donor language. Okay. Which in turn is determined by the non-structured conditions of the acquisition of the donor language and the structural properties of the foreign structure in question. It's perceptual silence. Okay. And I will explain what I mean by this. So what determines donor language proficiency and accuracy? This is where I bring in the sort of whole machinery from second language acquisition as probably the biggest innovation of my model, because other models don't really deal with second language acquisition at all. So when we look at an overview study of second language acquisition studies, they find that three factors, non-structural factors are decisive. One is the age of onset to the acquisition of the donor language, how old you were when you started learning it. Another one is language aptitude, which is how talented you are in learning new languages. And that I won't talk a lot about that, because when you look at these sort of very broad cases of loanwords, phonology like the Arabic loanwords used by almost everyone in Turkey nowadays and also in the past, language aptitude sort of loses its relevance because language aptitude is known to be normally distributed in large populations. Also the motivation to acquire this language has been found to be important. And some studies have also found this is more contentious. Some studies have also found that the degree of exposure to the target language plays an important role. So these are the non-structural factors that are important. Regarding the structural factors, I rely on the speech learning model, which is one of the leading models in the acquisition of phonetics and phonology. And the key structural factor in second language acquisition is perception accuracy in this model and it's affected by the second language. The second language sounds phonetic environment where it occurs and its similarity to existing first language sounds, okay, existing sounds in your native language. It might be a bit counter-intuitive to some of you who haven't done second language acquisition, but the more similar the L2 structure is to an existing L1 structure, the more difficult it is to be perceived accurately and ergo to be acquired because the foreign sound is misperceived as equivalent to a native sound, okay. This similar structures tend to be acquired more easily, but only after substantial exposure, okay. So beginners or people with less experience still struggle getting the collection accurate, but upon increased exposure pronunciation becomes more accurate. And this model also mentions a non-structural factor, namely the age of onset. It says the more similar the structure, the lower the age of onset needs to be to accomplish accurate acquisition, okay. So if you have a similar structure, you have to be a very young learner to be able to perceive it accurately and to pronounce it accurately. So how do we put these facts into a model of normal phonology? Well, the competence to adopt depends partly on the structure salience as we've seen, and I divide that into both its absolute perceptual salience in a particular environment, which means it's universal acoustic properties regardless of the first language of the borrower user or his or her acquisition background, okay. So it's not how a speaker of a certain language perceives the sound of a certain language, but it's basically that sound with acoustic measurements, you know, which are more or less stable cross languages. But also it's relative perceptual salience, which depends on the particular L1 and the particular L2 acquisition background, okay. So there's absolute perceptual salience and there's relative perceptual salience. Structures with low salience will then post greater acquisition difficulty in the donor language, which is the second language. Okay, so to exemplify that with some data, so we can apply that to the data we've seen. The heritage binding was in Sweden. They have acquired native-like proficiency in Swedish, and I've actually measured that in the study, and consequently they have the competence to adopt L, okay. And that's thanks to very early ages of onset, because they're basically surrounded by Swedish when they're growing up, even before school. And also high motivation for acquisition and high degree of exposure due to Swedish being the ambient majority language, okay. So that's how we can explain that they were able to acquire the competence to adopt this particular structure. How about the non-native-like ultimate elites then? Because we know there are comments in the literature of their non-native-like pronunciation of Arabic and so on. So how would they fare regarding the accurate production of the Arabic L? How salient would the vocaliate qualities of the bealer or non-bealer quality of this L be in this position for Turkish speakers? Well, in order to determine the absolute salience of the vocaliate quality of the L, one thing we can do is we can look at cross-linguistic observations or typological data. And in this case, I thought the phenomenon of lateral vocalization is a bit interesting. As you know, in some varieties in English, not the least in the city, milk is pronounced as milk. Milk, the L is bealerized to full. And also in some varieties of, for example, Bavarian, German, Austrian, German, the clear L can be vocalized to E. So you can have volum, pronounces voi. And that's a very stable pattern across languages. When a language has lateral bealerization, you can be certain that if it's a bealerized lateral, it will be vocalized to U or something like that, the back vowel. If it's a non-gularized L, it will be vocalized to E or something like that, which is a front vowel. Okay, so that's a very stable pattern. And even though the consummental quality, the laterality of the sound may disappear in this process, the vocaliate quality survives. And I think that suggests that it's a salient feature of the sound, its vocaliate quality. Yeah, yeah. How about relative salience? Well, since the clear L already existed as an elephant of Turkish, I'll be put a different distribution. Its absolute salience is not low, according to the speaker model. So it should be sufficiently dissimilar. Therefore, it should be salient for a Turkish speaker. And we can actually, the best thing to determine relative salience would be if we had a study on the acquisition of Arabic by Turkish speakers. Okay, but we don't have that. Okay, next best thing, we can pick a language pair with similar properties. As I said before, English has the same alphabets as Turkish, regarding the lateral. And Spanish only has one type of L, which is clear. Okay, therefore, I found a study on the acquisition of the Spanish L by speakers of English. And this actually confirmed my prediction that this is a segment whose pronunciation actors improves over time with increased exposure, as the speech learning model would predict. So this is how we can determine, these are some ways of how we can determine absolute salience and relative salience. Okay, so just giving you an overview of the salience related part we have seen so far. So on the left hand side, we have an native speaker pronouncing a certain sound. And now that enters the learners' second language. And let's use the Arabic Turkish case. So we have the Arabic L here, and it's perceived accurately as L. And also the learner learns that Arabic only has one type of L, and stored as the lateral phoneme and the L2 phonology of the learner who will then go on to borrow it. And it's produced accurately. We can say that the relative salience depends on what you have in your L2 phonology and what you have in your L1 phonology. Okay, so that sort of comparison between the two phonologies determines the relative salience, which in turn underpins perception. But also you have absolute salience regardless of the particular language pair you're looking at, which also influences your perception. But then you also have the non-structural factors. So age of onset directly influences your L2 perception, but also the degree of exposure and motivation may influence your perception and production. Okay, so then you have a, so this whole machinery here down here is what my model introduces based on SLA studies. So the vocaliate quality of L is absolutely salient, and also LF Rabakov is relatively salient for a Turkish speaker. Hence the acquisition of L in a new position in Arabic would require a substantial degree of exposure. And since the Ottoman Turkish elite had a lot of exposure to Arabic in school, they must have had, they must have had prior that accurately, ergo they must have had the competence to adopt the L after backgrounds. Okay, so another example regarding absolute salience. In another study I looked at borrowings, which in the original language, in the donor language had long segments, long vowels and long consonants, which were illicit in Turkish. And again, it's the same contact situation, it's the same languages as in the previous study I showed you. And prior to contact with Arabic in this case and partly Persian, Turkish did not have phonemic length distinctions, no primary length distinctions. And Arabic vowels and consonants as well as Swedish vowels have an acoustic duration in long segments that's at least 50% greater than the sharp counterparts. Okay, so a long segment can be differently long in different languages, sometimes also in different vowels and consonants and so on. But in the case of Arabic vowels and consonants and Swedish vowels, the long counterparts are at least 50% longer than the short ones. Okay, which suggests that long segments are dissimilar to short segments, so they should not have low salience, they should be relatively salient. Okay, so according to the speech learning model, their acquisition would then require substantial exposure. And therefore both the Ottoman elite bilinguals and the heritage bilinguals in Sweden should be able to acquire them accurately because they have substantial exposure. In contrast, the duration of long Swedish consonants is only 30% greater on average than their short counterparts, which suggests that long consonants in Swedish are similar to their short counterparts and therefore have low absolute salience. So these similar, so they would be long consonants and short consonants would be similar sounds according to the speech learning model and their accurate acquisition would then require low ages of onset to the acquisition of Swedish. And since we have that in the Sweden case, the heritage bilinguals in Sweden should have acquired them accurately. Okay, so we can in summary say that both borough groups must have had the competence to adopt all long segments despite these differences in absolute salience. When we look at the results of the study, and we have long segments in the closed word final syllable, so in the Arabic case, we have the initial Ottoman boroughs, the reconstructed pronunciation here, where the long vowel is preserved both in unsophics and suffix context. And then when we look at the contemporary data, we see that the length is preserved in unmarked positions, but it's shortened in marked positions, and we get that in 70% of all words of this type in Turkish. And it's very similar to what we see with long consonants from Arabic. Also in 70% of the cases in contemporary Turkish, on the right, we see shortening in marked positions and preservation of length in unmarked positions, which suggests that there's underlying length in these long words, at least in 70% of the long words in the dictionary. So we can say that the initial Ottoman elite boroughs, they are full adoption in all positions, whereas today's speakers, they have partial adaptation, they shorten vowels in unsophics context or when there's a suffix for the consonant following after the word. When we look at the Swedish data, it's quite similar. Long vowels are preserved in 67% of the cases, and remember that these, or maybe I didn't say that, but these are not phonemic in Swedish. But when we look at the long consonants, and they were the ones I said had low salience, low absolute salience, they are shortened in all positions in 80% of all cases. So these three are very similar, and this one really is very different. So although the initial borrowers had the competence to adopt all four long segment types, they only ended up adopting those that did not have low absolute salience. Therefore, high absolute salience must be a necessary condition for adoption. So what would the models again predict about these data? Well, as we saw before, no adoption in these two models, so they would make the wrong prediction apart from the Swedish consonant case. The phonological stance model would make the wrong prediction because it says that in the Ottoman case, there's very low degree of commutative bilingualism. So the long vowels and consonants should not be adopted, or adopted to a very low degree, which isn't the case. And with the Swedish case, it would predict that only the phonemically long Swedish consonants should be adopted, and that's exactly the opposite of what we have. The vowels are adopted, but the consonants are shortened. So this model fares really badly in these data. So we can conclude that phonemicity does not matter, because the input is clearly phonetic, even when the borrowers are native like bilinguals. What matters instead is the acoustic salience, regardless of phonological status. Thomas's borrowing scale would make the wrong prediction in the first case because we shouldn't have length adoption, which is super segmental, when we have level two contact. And in the Swedish case, we should have adoption, but we don't have it in consonants. So that's one, two. So here we can conclude that not all structures pertaining to the same level or type should be treated as equally borrowable or adoptable. Differences in salience between these structures of the same type or level need to be taken into account as they influence both acquisition difficulty and adoptability. And therefore social factors cannot trump salience as a structural factor. So in the Swedish case, we have the most favorable social factors, but still we get a case of adaptation length shortening. So how would we then analyze this? What happens with the Swedish consonants? Well, they're underlying the long and they surface as long. And we have native-like proficiency in the borrowers due to the heritage context. Therefore, they perceive it accurately and they map it accurately to their first language Turkish. And then this is stored else underlying long, but then it surfaces short. And that's because this long consonant is not salient. And that's because it has a relative duration that's 50, it's less than 50% greater than its short counterpart. So that's how my model explained this. So when perception is accurate, but absolute salience is low, adaptation takes place in production, a bit like as we saw in PMAP. So structures with low absolute salience will never be adopted. They will be adopted either in perception or later in production. So moving on to non-structural factors. In my third study, I looked at the integration of illicit word initial onset clusters, which featured in established loanwords from French and English, and whose phonetic forms are similar to one another, both across French, English and Swedish by virtue of the incognites or loans from one another. So these are words like standard and plan, which are quite similar in English, French and Swedish regarding the onset cluster, the coordination onset cluster. And Turkish words lack onset clusters altogether. So the question is then in loanwords of this type, whether they will be adopted as clusters as an innovation or whether they will be adopted through appendices. So if we had appendices, they would be like sutandart pilan, which depends on vowel harmony and so on. Okay, so here I have two main groups of Turkish speakers. I have 24 elite bilinguals with L2 English and Turkey. And I assume they would be dominant in their L1 Turkish because they've grown up in Turkey. And then we have 29 heritage bilinguals with L2 Swedish in Sweden. And since they grew up in Sweden, I assume they would be dominant in their L2 Swedish, which is a language with onset clusters. So the dependent variable is cluster adoption rate. So in how many percent of all instances do we have actual cluster pronunciation in loanwords of this type? And as independent variables, I had a competence-based variable, which was L2 cluster rate. So I looked at their pronunciation of their second language and how they pronounced onset clusters in that language as a percentage that the accuracy. And then I had two dominance variables. One was self-imported relative L2 proficiency. So I asked them to report on their proficiency in their first language and second language. And then I subtracted L1 proficiency from the L2 proficiency to calculate relative L2 proficiency as a measure of their L2 dominance. And I also looked at their L1 use. So the more they use their L1 Turkish, the less they would use their L2 and other languages. And that is also generally used as one measure of dominance in bilinguals, how much they use their different languages. So when we look at the results, when we look at the cluster adoption rates, you see that the elite bilinguals in Turkey, they have a cluster adoption mean of almost 55 percent. On average, in 55 percent of all cases, they have onset clusters when they could pronounce these long words. In the group in Sweden, it's much higher. It's significantly higher, 88 percent. It's almost full adoption there. And as I assumed initially, the bilinguals in Turkey were dominant in Turkish. That's why we have negative L2 dominance there. The bilinguals in Sweden were dominant in Swedish. So that assumption was verified. And also, the bilinguals in Turkey use twice as much Turkish as bilinguals in Sweden. Again, not very surprising. The differences in all variables were significant. So as the results of the dominance-related variables of relative L2 proficiency, and how one used clearly illustrate, the societal dominance status of the recipient language Turkish, as dominant language in Turkey and as a weaker language in Sweden, impacts individual speakers' language dominance, which in turn also impacts their adoption rates in bilinguals. Okay. So what would the different models say? Well, again, the first two don't allow adoption, which we have. The phonological stance model would say we should have very low cluster adoption rates in Turkey because there's very low degree of community bilingualism, which isn't true. We have to get 55 percent cluster adoption on average. But in the Swedish case, the prediction is correct. We have a lot of bilinguals and we have high adoption rates. Okay. So we can conclude that the degree of community bilingualism or similar macro factors such as societal dominance of a language are not sufficient in explaining the sociolinguistic incentive to adopt. Okay. Additional factors are needed and some candidates might be prestige of the domain language and attitudes towards the domain language. Okay. Thomas's sparring scale. So since clusters are prosodic structures, it would be predicted in this model that they shouldn't be borrowed, but they are. But then when the intensity increases to level three in the Swedish case, they should be borrowed and they are. So it's partly right and partly wrong. Okay. Among these 53 bilinguals in the two countries that I looked at, 31 turned out to have perfect cluster competence. So when we looked at their second language pronunciation of words with onset clusters, they had 100 percent accurate production of onset clusters. So in this group, we can exclude any competence-based effects. Okay. Anything we find, any differences we find among these 31 individuals must be due to incentive. Okay. So when we look at these, you see that most of them the dark dots there in Sweden, but we also have about seven individuals in Turkey who have perfect, perfect cluster competence in Turkey. Okay. And when we correlate that with their cluster adoption rate, we get the strong correlation, which is significant, 0.629. And we have on the right-hand side, we have four individuals who actually are full adopters. Okay. They use onset clusters in all of their normal words. Okay. Four out of 53 and out of the 31 with perfect cluster competence. So we can, and among the other others who didn't have perfect competence, there were no full adopters. Okay. And these full adopters, they have a minimum relative L2 proficiency of 2.8. Okay. So it seems that maybe a clear, clearly higher proficiency in the L2 is a requisite for full adoption. But then we also have these 10 individuals who are above that minimum line, but they don't have full adoption. Okay. But still, as you can see, the lowest of them has an adoption rate of just below 80%. So they still have higher adoption rates. Okay. So the higher a speaker's dominance score is in the cluster language, the more likely she or he is to adopt clusters in normal words. Okay. So if you have perfect cluster competence, then what makes a difference between these people? It seems that your relative L2 proficiency in a cluster language is a very good predictor. Okay. So going forward, so when to summarize, when the donor language is decidedly dominant as Swedish is in the Swedish, in the Swedish case, more recipient language native speakers tend to be bilingual. And they tend to have higher proficiency in the donor language and greater use of the donor language than the recipient language. So they tend to be more dominant in the donor language. Okay. So that's not very surprising. And in such cases, the donor language also tends to have higher prestige, which usually leads to positive attitudes towards a donor language in the general community. Okay. However, this may not always be the case. I'm saying tens and usually so I'm hedging a little bit there. So if you imagine that some donor languages such as in such as metropolitan languages in colonial contexts, they might be dominant among recipient language speakers as a heritage language. But and this is due to limitations in the use of that heritage language in the development of the speakers in that heritage language in the colonial context, where maybe they're excluded from schools and so on. Okay. This very domination can conversely give rise to negative attitudes towards borrowing from the donor language. And Sally Thomas and told me that she studied an Amerindian language, which borrowed words from other Amerindian languages, but not from English, because they have negative attitudes towards English because of its colonial domination in North America. In elite bilingualism, and also in monolingualism, you can have positive attitudes towards the donor language due to its high prestige in the context without that language at the same time being society, societally dominant. Okay. So you can imagine English in Turkey, for example, English has high prestige, but people are not dominant in English, very few people speak English at a good level. Okay. So therefore, it's best to keep dominance and attitudes separate as factors. They, in many cases, they go hand in hand, there are cases where they don't. Okay. So in my models, I say that the total sociolinguistic incentive to adopt is a composite of attitudes and dominance. Dominance is relatively predictable as we saw from the societal context. Okay. If the language is societally dominant, it produces these sort of proficiency and use patterns that we've seen while attitudes are more idiosyncratic in individuals. Okay. So I then go on to make this conjecture about three groups, monolinguals, elite bilinguals, and heritage bilinguals. I say, I just hypothesize that this is, let's say they're in the same society, and they're generally positive attitudes towards the donor language in that society. So that gives you some incentive to a certain level. And to go beyond that, you need dominance. And monolinguals don't have it because they're monolinguals. And the elite bilinguals have a certain degree of dominance, and heritage bilinguals have more because they're more dominant. They have a higher relative altitude proficiency. Okay. So this is a conjecture. So in the third study I'll show you, so we'll have the monolinguals and elite bilinguals in Turkey, and then we have the heritage bilinguals in Sweden. Okay. I think I'll just, okay. So these are the three non-structural factors I mentioned that influence the competence to adopt a certain structure, age of onset, real exposure, motivation from second language acquisition steps. Okay. And then I just said that we have attitudes and dominance as separate factors, inclusive incentive. And what I didn't say was that in order to borrow structures, you need to borrow words to begin with. Okay. So you need to be, you need to have a generally positive attitude towards borrowing. So you need to have foreign words to begin with. And then, so these are sort of the six factors. And then we can go a bit deeper and say, you know, are there any societal, so this is in individuals. On the societal level, do we have, what factors do we have that influence these? And I came up with the Pearson in the recipient language community. So if you're a purist, you don't borrow words from other languages. Okay. And then we have the societal dominance of the dominant language that we talked about. And that influences all of these factors. So if you have a societal dominant learning language, you will learn it before school or very early in any case, you will get a lot of exposure to it because it's all over you, all over society. And therefore you will have high motivation. And you will likely have or it will contribute positivity to your attitudes, then there might be, you know, counter influence from other things that influence your attitudes, but still it has some effect on your attitudes. And then of course on your dominance in the dominant language. And also, if you have one language that you borrow from, that's bound to influence your general attitude towards borrowing words from languages altogether. And then you have prestige of the dominant language. That's more based on language's symbolic status. Okay. So dominance is based on, you know, when you start learning it, how much exposure you get to it. And so how much you use it. So it's very sort of factual, stable factors, whereas prestige is a bit softer. Okay. It's, it can influence age of onset. So if it like in Sweden, if English has high prestige, people might open up bilingual kindergartens and things like that. So my things will say just one set. My thing is degree of exposure, because you get, you know, films with subtitles, a lot of English on TV, and so on. And that will influence your motivation, you know, you get you want to learn English because it's all around you. And also your attitudes, obviously. Okay. So, and we can even go deeper than that, one more level. We can say that, especially those of you who look into linguistic anthropology might find this interesting. So the purism in the community might be related to language ideology in general, what language means in that community, what it means for identity and so on. Societal dominance might have to do with economic, political, and even military dominance of the donor language speakers in that community, or more internationally. And prestige has to do with cultural dominance, sometimes called soft power. Okay. So something like, you know, we have a lot of English speaking TV programs on Swedish television, that's the soft power of the English speaking world in Sweden. Okay. So I'm coming to the end. So how about I've only spoken about bilinguals so far. And in the cases I've looked at, there are other cases as well, but in the cases I've looked at the initial borrowers were bilingual. And then we know that in some of those cases that those words were passed on to a larger monolingual community. So what happens there? So when we look at the L, we saw that comparing the reconstructed pronunciation and the contemporary one, the L was adopted by the initial borrowers bilingual borrowers, and it was also adopted by the monolingual users. Okay. So the adoption went through in that case. About the long segments in Arabic, we saw that the initial borrowers adopted them in all positions. But then when they were transferred to the monolinguals, they shortened them in the marked positions. Okay. So there was a change there. There was partial adaptation. Okay. In the Swedish case, we only have the initial borrowers. We don't have any monolinguals. And the third case, the onset clusters, we have high degrees of onset cluster adoption in both Turkey and Sweden. And in Turkey, I also had a control group with monolinguals. And even they had some adoption of onset clusters. They had a mean of 42%. Okay. So the Arabic L and long segments were adopted by the initial bilingual borrowers. But only the L and not the long segments were adopted by later monolingual users. Okay. So there's a differential transfer from bilinguals to monolinguals. In some cases, the adoption goes through. In other cases, something happens on the way. Why do we have this difference? Okay. And that's where we come into levels. Okay. So imagine this conjecture I had before. Let's say that we have two different necessary levels of minimum incentive. Okay. Let's say that the minimum level of incentive required for segmental adoption is much lower. Okay. Which means it's easier to have sufficient incentive to adopt segments than for super segmentals, especially purely prosodic adoption. Okay. And actually, Sally Thomason's borrowing scale has a very similar thing there. You know, with the levels of contact intensity, you know, when contact is low, very low, you don't borrow any structures. And when it increases one level, you can borrow segments. And when it increases further, you can start borrowing super segmentals. So this is very similar to that. Okay. So in the case of L, we have adoption across the board. Okay. In all groups, the L is adopted. Because it's a segment. And as a segment, the total incentive we have is above what's required. Okay. So that's what's required. And we have something about that. Okay. In the case of word initial onsets, there, okay, they are prosodic structures because what happens in onsets influence prosody, but they're not as purely prosodic as tone or stress would be or length. Okay. So they have a sort of intermediary position between segmental and super segmental. Okay. So in that case, they should be adopted by the bilinguals. Because it's sort of intermediaries, you should get some adoption by the monolinguals as well, which is what we have. And then the adoption of long segments. As we've seen, they're not adopted by the monolinguals, but they're adopted by the others. Okay. So this is sort of how I came up with this conjecture about these different levels of total incentive, because I had this differential transfer from bilinguals to monolinguals. And I thought this might be a, so it's very similar to Thomas's thinking. And when I have several levels, you only have two levels, segmentals and super segmentals. I think it's necessary to have several levels like this. Okay. And I call this the level condition. Okay. All right. So summarizing the model, the advantages over previous models are, first of all, this model is more interdisciplinary. It's very importantly, very centrally. It don't incorporate key findings from second language acquisition, but also from other fields that are relevant. It has wider scope than the recent models. It doesn't just look at adaptation, it also looks at adoption. It has a broader approach. It looks at both structural and non-structural factors. It has a high degree of formalization. It's not just a taxonomy of factors, but it's an ordered comprehensive model. And I couldn't give you the details, but if you look at my 2014 study, you can find a full account of an OT, optimality theory based account of how these innovations, these adoptions are incorporated into the chronological flexible. Thank you. So that was very information dense. Not just relevance for particular field, but you can work with me, but also for those of us who are interested in language content with regard to endangered languages, language endangerments, particularly the acquisition of local language or local language knowledge. I think this is something that's very relevant to language revitalization as well. We are on the ideological side. I was, like all the most of the models focused on the fact that whether people could or could not produce songs, but maybe, I mean, they could also choose not to. For example, that people could also choose not to reproduce something if they could. For example, in Germany, there's lots of English learners, and even people who are completely bilingual in English might not come in a German way for reasons that these models would not take into account like this, you know, where it might sound ridiculous to try to put an end and not integrate into German. What, how would you deal with that? How does it fit into your market? I think there you would have this sort of interplay between their individual, the speaker's individual dominance. So if the speaker were to report higher proficiency in English than German, then this model, my model would predict that they might be more willing to adopt a foreign pronunciation. But also if you, so what I didn't do in my studies, I didn't look at people's attitudes. You could, you know, do that in, you know, in the studies, you have direct measurement of attitudes through questionnaires or interviews, or you have indirect measurements through matched guides, tests, and similar techniques. So you could, you could do that. And that might be, you know, even if they're not dominant in English, they might have very positive attitudes towards English, let's say, because they live close to a US army base in Bonnberg or something like that. Or maybe not. I don't know. It depends. I don't know the dynamics of those terms. So it will be, you know, those two, so the dominance is it's more stable. You can sort of, by looking at the general context, you can sort of come up with some predictions, but the attitudes, you really have to go deeper into the speaker and look at their attitudes. Could it also be like language-wise? The case, let's say, for example, if I speak German and it's a cool, I would not do any theorized help, even though I'm quite capable of doing that. Then it would be about your attitudes. So maybe you would think that, oh, if I were to say, does this have a cool, then I would sound very pretentious or arrogant to the rest of my community. That's why you would avoid it. They would not be personal. I think that's yeah. And that's because, you know, they're not a high percentage of bilinguals in your language community. So then it goes back to societal dominance, depending on how you define bilinguals. The fact that you all, you can differentiate between different factors and even within the same community, the same boring, you can predict that in one word certain aspects get adopted in certain terms, depending on the intensity. But one factor that I'm kind of wondering that you don't seem to be foreground too much is the fact you're being understood. So the German case is a good example. When I got home and now I have the fur and wear as a difference, but German to be concerned, and I want to order whiskey, and I asked for whiskey, then they understand that if I ask it, and I say to them, I don't even know what I want. Has that happened to you? Oh yeah, all the time. Okay, so you say whiskey and they don't understand what you're saying. Okay. So it kind of forces you because the mother straight up has that rotation of course you can go along with it. What forces you to go along? But then the other thing is also, you see cases where for example in Welch, they borrow the shirt and jerks and they don't copy kids natively, and then they used to just adopt that. So the chips, they would adopt that as ships. Like Swedish. Yeah, but of course it's bilingualism becomes more rampant and nowadays they're all really proficient in English, and maybe a hundred years ago they were all in the lingual, so it didn't matter. But now there's this conflict, well if you adopt chin and you adopt chips, then suddenly you have the same thing, you can't keep it apart. So that seems to me like a treasure to say. Oh yeah, that's horrible then, even though everything else we still adopt because of other treasures. How does that sort of figure anymore? Yeah, I think in the Welch case there is, for example, Welch doesn't have the sound z, the tone s, and when the word zoo, the English word zoo is borrowed into Welch, in the south where English has a greater presence, more thorough presence, the z is borrowed, whereas in the north where Welch is stronger, it's pronounced as su, I'm told. So you get the community-wide differences north versus south, which also are parallel to other differences between the north and the south in Wales and so on, which you can also go hand in hand with people's confidence to adopt if people can say z or not. So I think it sort of goes back and forth between the individual level and the societal level. So in some cases, if you're not a strong individual that risks going against established norms, then you will follow either your own particular community, say in your region or in your social class, or you might go against that, you might follow the general language community or what your dictionary says or what you hear on TV news or something like that. It's a very dynamic sort of interplay between the individual level and the societal level, sort of like a negotiation almost. Which means it's very difficult to connect to. It is, yeah. I was wondering if adaptation adoption was such a strict binary or other practices where there's kind of an imperfect adoption. So one example of the theme of the Turkish way of claimant, which is part of the American claim, is there's an attempt to approximate a peculiar style. And so all the individual segments of the Turkish way of order are bound, indigenous to Turkish, but the combination of that way is not. That would be, I mean, I don't have to look at the particular instance, but I would say that's about how speakers with a certain acquisition background perceive the Arabic usually. If you take the syllable, it might very well be the case that they pronounce, they perceived, say in Ottoman Turkey, they perceived the eubularity not, say dorsality or whatever, not as a feature of the consonant, but as a feature of the vowel, because it does influence the vowel's production in Arabic too. So then they would, you know, perceive the q as their own k, whereas the eubularity would be realized through the vowel. But that's how their perception is. That's not a conscious, I wouldn't say that's a conscious strategy, but it's basically sort of what happens through the perceptual filter in their own acquisition of Arabic. And in that particular case, actually there's an interesting historical fact. The longer back you go in history of Turkish, the more the back a la foam of the k in Turkish is closer to the eubularity. So in the original period when those words were borrowed, maybe there was a perfect overlap between the eubular back a la foam of k, yeah. So this gentleman had a question. We have to be able to do it quite too far. That's one of the things that we wanted to do. Okay, so very quickly. And when you're in the kind of a reading system of the eubularity, or the change of the reading system, we have an impact on the dormitory system. So there's one example in my mind which is, you know, Beijing used to be techy, but now it is ever since Beijing. No one's been techy. Or people even say Beijing. Have you heard of that? Yeah, yeah, but not techy. No, not. Yeah, yeah. Yeah, I mean, that's a definite possibility. I've, you know, throughout my whole talk, I've only talked about oral input, but there's also written input. In my particular case, I have looked at the potential effects of the written form because the shift from Arabic based writing to Latin based writing in Turkish means that some things were more visible. Length was more visible earlier, and then it became less visible in the Latin alphabet, especially in the contemporary version. But that has no effect in my data. Okay, so there are other studies which show that there was an effect. But it's sort of, I suspect that that's maybe more, I mean, the written input is more important when you have so-called naive borrowing, where the borrowers are monolingual. They have no idea of what this language sounds like. They only have the written input to go with. Yeah. Much better. I was wondering, since that sailor and sailor is out trying to smite it somehow, so smite it becomes redundant. That's why I wanted to ask. And do you think that it would be possible to bring in beyond the chronological domain to the word, that it would even be considerable? And I mean, could you rank power relations in such a model? Power relations? Using the majority constraints. I can't go into the details, but if you look at my 2014 study, I think if you go on the website where the talk was announced, there's some references. So the 2014, there I actually go into that, where you actually how you re-rank your constraints in OT. And there I have a sort of sociolinguistically sensitive re-ranking of OT constraints. Okay, thanks very much. This is something about Linguistics and Hope. Sure, Linguistics and Hope will be this evening at 6pm. And I assume a group of people will come too, I guess. Yeah. So it's in London, it's true, if you want to go directly there. I think rather than go to the Institute of Education, it might be worth going straight in. So I think it's impressive to come to that as well. So yes, rather than having two lots of confidence, we'll just have the one. But it wants to continue the discussion in more informal circumstances.