 So, today I'm basically going to present almost myself because I'm going to present the way I work, right? So, given this, given that Nathan initially presented this as a methodological workshop, right? Where we present the actual way we do reconstruction in our particular fields, I felt that it was a good opportunity for me to show you how I do reconstruction, which is basically neogrammarian but with a twist. And the idea is that, well, the comparative method can be complimented, right? So, if you ever work with BASC as some of the people here or many other languages in the world, you would have probably read these famous sentences Nathan mentioned may lead today, right? Also, I never pronounced that name well. They said that it was impossible to reconstruct BASC and they may have a point. It's not easy, but it's certainly not impossible. But the point that they may have is that the comparative method finds most difficulties when dealing with language isolates. Not only, because, I mean, yes, it's difficult to reconstruct BASC, Ainu, Wabe, Burushaski, whatever isolate, given that they don't have a family. But it's also complicated if we are working with languages with small families like Korean or Japanese or languages with not as well documented as other languages we usually work with, right? So, the idea when doing reconstruction, phonological reconstruction, is that we are dealing with sound change, basically, right? Some correspondences, cognates represent sound change. And sound change is mostly, but not only, phonology, right? So if we just deal with phonological terms, we will be able to explain that much of a particular sound change. But then, I would argue that we need of many other sub-disciplines of linguistics in order to be able to give a full account of a particular sound change. In this case, I mentioned phonetics, social linguistics or contact linguistics, dichroony that would probably include philology here, and typology, philological typology, for instance. And this creates a number of, like, intersections or topics that come from each of those that are of interest in the study of particular sound changes. Those include phonological typology, sound change typology, historical phonology, phonetically based phonology, sound change in progress, contact linguistics, and so on. So, I will start with an example that is not the actual focus of my talk today, but it is nevertheless important for me because it's already done, and I think I was successful in working with this particular sound pattern. I'm talking about U-fronting in Basque. What is U-fronting? Well, U-fronting is U, this is U, becoming U, right? And I started working on this particular change philologically. I went to a text, this is OnSileCecovidia, a suburban Basque text from the 1666. And what you can see here, I hope you can see it from the back, is that we find two different ways of written down what in other Basque dialect is U, right? So here we find some examples in which U is used and others in which U is used to mark a sound that in other dialect does not need such a distinction. Well, this obviously tells us that at this particular time, this particular author required of two different graphematic ways of showing what is reconstructed as U because it wasn't U any longer, right? But although philologically we can get, because we know the French tradition and so on, we could imagine that this represents U and this represents U. We have to actually move on to another subdiscipline of linguistics which is phonology to understand what's behind this distribution. So what phonology tells us is that there is U and there is U and the crucial context that establishes the distribution between those two segments in suburban language is the following consonant, right? The consonant that follows the powers. From that, we create two sets of consonants. One of them is what I call the inhibitory context, a series of consonants that have mantine U as U and then the rest of the consonants which are the fronting context. If we think of phonological terms, what we could see is that here, right, the inhibitory context are all coronals. But at the same time, some of the fronting context, like T, D, the palatal, more surveillance, all those are coronals too. So it's not that easy to find a natural class that will account for this particular distribution. That's when we move on to yet another subdiscipline of linguistics and in this case, we move on to phonetics. And phonetics can help with our process of understanding what's behind this distribution and what has conditioned this sound change. In this case, we can go to physiological studies and find out that in order to produce these segments that we have classified as inhibitory context, we need to use the tip of the tongue in a very fine way, right? We require fine movements of the tip of the tongue. And in order to produce those fine movements of the tip of the tongue, we have to move the tongue body backwards into the region of U and far away from the region of U. So we can observe that there is a co-articulatory bias that makes us more likely to produce U for the segments that are actually inhibitory segments in BASC. And that could be applied to all the languages and in this case has resulted in a particular distribution within the history of the BASC language. So now that we mostly understand what's happening here, we can move on to technology and see that in Europe, this is a map from Juliet's work from probably a couple of years ago, in Europe, there are many languages that show U-fronting, right? And they do form an area. So there is an area in Europe of languages that show U to U. BASC is actually part of this area. It's the southwestern most language of this linguistic area that shares particular sound change. What does that mean? That it is very likely that contact played a role in this particular sound pattern. Now that we do know what happened here, we have a more complete understanding, we can go back to our data and be able to produce a better analysis, a more complete account of the sound pattern understand, right? So this is basically what I do. That paper was published a couple of years ago, if you're interested in it, and now I'm moving on to get another sound pattern that is found in the history of BASC. And in this case, the sound pattern of interest is what I call nasal aspiration. So nasal aspiration is the process whereby intervocalic N becomes H, right? Here we have an example of Aquitaine, which is the oldest attested variety of Uskarian language, right? Attested from the 1st to the 3rd century. And what we see is that what was written as N in the 1st to 3rd century shows up as H in modern BASC, but it is produced with audio nasalization in suburban BASC, something like saying, which is also true of reconstructed words such as Nani, which is Nahi in modern BASC, or a very old, long words such as anathem, which is ahate, superoan ahate, ahate, right? But, yeah, the phonological context of this process is clear, but what is the actual result of nasal aspiration, right? What is happening there phonologically? What is its status in the modern language? Is it still contrastive, or is it not? And is this sound change internal and phonetically natural, and thus it could have happened at any time in the history of BASC, or is it a sound change that's required of contact in order to develop the way it developed in BASC? So, first of all, I want to talk to you about how rare this contrast is, right? Because the sound pattern that I just described created a contrast between H and nasalized H. So, this opposition is actually quite rare, crossing with the two. If we think of the languages that have H, well, we could say that most of the languages in the world have H, right? 88% of the languages in the opposite database show H. Nevertheless, the languages that possess a nasal-glotal approximate, nasalized H, are much fewer, right? We don't find so many of them. We can find that in Grim, Lisu, and Piraha, for instance. But those languages don't show an opposition between the two segments. They just show nasalized H and no oral H. Only two languages have been argued to show a clear opposition between oral H and nasalized H. And those languages are Quengali, Avanto language of Namibia and Angola, and Seimat, a Austronesian language of the Admiralty Islands. There are at least two other languages, Aguaruna and Arabella, that could be analyzable with oral versus nasalized H contrasts. But the nasalized aspirate can be viewed as a predictable alophon of the villar nasal consonant as well. And that's the preferred analysis. So, yeah. Moving on to the methodology I used for this research. First, I mean, it's a three-fold methodology, which I'm going to represent here. First step is phonetic explanation, right? What this implies is that we will decompose the factors conditioned in this particular sound pattern to find their articulatory, acoustic, and perceptual biases that may condition the outcome. In the case of nasal aspiration, there has been research done on it since already 40 years ago, and we know that there is a property of speech, which is usually called rhinoglyophilia, which means that there is a relationship between aspiration and nasalization, in the sense that aspirated sounds can become nasalized and nasalized sounds can become aspirated. So this, I will talk about it shortly, but yeah, we find it in many languages and so on. Second, phonetic experimentation. Here we will devise experiments in order to get acoustic and articulatory data. And in some cases we will also prepare more specific experiments involving, for instance, natural contexts to find the kind of co-articulation that we expect would have resulted in a particular bias in sound judgment. Of course, this can be applied to any speaker, not necessarily speakers of Bach. We just need knife speakers that share the crucial features that are part of this sound pattern. Third, typological assessment. Here, not only the target segment and the outcome, but also the context where this sound pattern developed are of interest. And we will look at the languages of the world and especially the languages in contact to our target language in order to find out how common this sound pattern is. If we find that this sound pattern occurred all over the world in many different places in many different ways, then we would assume that it's a phonetically natural sound pattern, given that it also has a phonetic explanation. But if not, we may have to check whether contact played a role on this change and whether it was the combination of a constellation of features found in Bach and its contact language, in this case Gascon, which also shows this kind of development, but with a different outcome, which is loss of the end, played a role in this particular sound pattern. So now, what's the hypothesis after all that I have presented now? So the hypothesis is that N was more likely to unite than other nasal consonants in the same context. And two, that a language with nasal aspiration will have H as a category. Since the interpretation of a work and weekend N as nasalized H suggests that H is a variant of a category already existing in the language. Why is this important? As I said, Gascon shares the target and the context of this sound pattern, but crucially it does not show the outcome found in Bach. And if it ever was there, it's not preserved. Bach, on the contrary, had H as a category and H also occurred after nasal consonants, which means that was very likely assimilated in nasality to that consonant. That, as a consequence of that, this was an alophonic category in Bach prior to the sound. He already talked about this, so in case that this exact sound pattern does not show up in any other language of the work, we would have to assume that two languages with the right properties were in a contact situation and that produced the sound change we just described. So, as I said, Rhino-Glorophilia is the phonetic connection between aspiration and nasality and it has been accounted for acoustically as well as perceptually. It has been documented in many languages and those include, for instance, Thai, the Nepali language Haju, or Scottish Gaelic, in addition to Bach, for instance. Moving on, now we need to check phonetically what is actually happening here, right? We have described a Rhino-Glorophilia process, but we have to actually test that in the lab. So, we have 15 recordings from 15 towns in the Mixion region and they were recorded for dialectological description, as found in Camino's work. And we discarded some recordings because the speakers were very old and the recordings were also very old. So, sorry, not good enough. All the recordings are quite long, so they can be used for our research and they were recorded with a portable recorder in high quality, but in a very rural situation. So, there are noises, there are bells ringing, sometimes even animals. All the audio files were manually transcribed, forced aligned with the web app of Maus, and set for BASC of France, which I prepared myself so it can be used in the future for other research. And they were hand corrected by myself as needed. So, very shortly, I don't know whether I can play this, but it's not important. This is the word behe, right? As you can see, this is the H here. And it's difficult to see, because the H is really vocal. It's very vowel-like. The same is true for Ihisin, which shows an etymologically nasal stage. Very vocalic. And the last one, Berhoun, shows an H that was etymologically oral, but it is assimilated to a following N so that it's nasalized as a phonetic... as a result of phonetic articulation. Right. Now, the acoustic measurements. We measured 20 different features of speech. Why did we do that? Well, because different people have used, especially A1-P0, and that's not good enough. The results are very variable, and they don't really... They are not explanatory enough. So we went on and we developed a new methodology that used a lot of measurements. Then, well, the measurements were taken in Pratt & R, and then we also added MFCCs for extra information, and all the measurements were taken at 5 millisecond intervals. This is not really important. This is just methodological, but of the phonetic part. Then we submitted to a principal component analysis, and we retained the PCs that explained 80% of the variants, which is usually 11 to 13. So we used speaker-specific models and nasalized tokens that are as shown here, right? Vowels in contact to nasal consonants always, and oral tokens with vowels in contact with oral consonants, as training data set. So I don't know how much time I have, or... You still have 10 minutes. So 80% of this data set was selected for training, and we retained 20% to be able to compare with vowels, what we see in the HS. We applied the logistic regression models with the principal components score as independent variables, and we did that for each speaker. That's very important, because each speaker is different regarding their apparatus. So these are the categories we have. Asimilatory nasalized H, which means, as I showed you before, H that was intended to be oral, but is in contact with an nasal consonant. Etymologically nasalized H, which are the Hs that come from M, historically, oral H, which is non-nasal, and oral vowels, which is the 20% that we retained from the training data set. And the only thing that's really important for you, because I'm afraid I may have lost some of you now, this is the only thing that's important, the midpoint marks what's nasal and what's oral in our model. Everything above 05 is nasal, everything below 05 is oral. This is how it looks like in a word, okay? So this is how they come of the carnival, and you can see that this is the middle of the graphic, more or less. So the vowels following the nasalized H are really, really high. The nasalized H is absolutely high. And then the ones preceding are variable, but with nasalization. If you compare them to the vowels that follow T or R, they are below the midpoint of this graphic, okay? So now let's go to the... Well, if anyone is really interested in all these stats and the methodology, right now it's not under review any longer. It's already under revisions. Hopefully it will be published soon. And all the training data as well as all the code can be found online if anyone wants to apply it to a different dataset of any given language. So the global results. We aggregated the results, and now we have two nasal categories, as I said, assimilatory nasalized H and etymological nasalized H, and two oral categories, oral H and oral vowels. And these are the results. Okay, this is the midpoint, above is nasal, below is oral, okay? Now, the first half of the graphic. This is the training data. For this particular dataset, this is as oral as it gets, and this is as nasal as it gets, okay? It was done, as I said, with vowels that were unambiguously oral and unambiguously nasal, and that's the best a model can get. And now the rest. Here we have a simulated nasalized H, which means that it was etymologically oral, but the contextual assimilation has produced clear nasalization because it's way above the line. This is the Hs that historically come from N, right? So they are usually not written down as anything special in bus, but they are clearly unambiguously nasal. So we have a nasalized H here, right, phonetically. These are the oral Hs. This is a very interesting thing, I will talk about it later on, because we would imagine that those should be closer to oral vowels. Well, they're not. They are close to the midpoint, below the midpoint, but really close to it, it's not unambiguously oral or nasal. It's something in between. And the oral vowels, which are close to the training data, as we should expect, because they should be unambiguously oral. The difference here is not significant, so we can actually create a group between those two categories, but the difference between the oral vowels and the oral H is actually significant. Also, very importantly, the difference between those two groups is significant as well. This means that we can separate nasalized H from the rest, but at the same time, there is a difference between oral H and oral vowels. And we should look further into that to figure out what's happening here. So this mess is each of the speakers that I put in the analysis today. We have four more, but I didn't have the time to put them in the analysis yet. And what we can see here is that for most of the speakers, the trend is maintained, but not for all of them. For instance, this speaker does not assimilate the H in a simulatory context. So it does have a category which is etymologically nasalized H, but the assimilation which I previously assumed to be common to all speakers does not occur in this case. That's news, but it's interesting. Then some speakers show a more clear difference. This one, for instance, is very interesting because all his or her Hs are nasal, all of them. This is also the case for this one, and not for the rest. In this case, this has lost etymologically nasalized H, and as I said, this does not assimilate, but most of them do have etymologically nasalized H, and two of them nasalized all the Hs. This might seem surprising, but it goes in line with an observation by Iñaki Camino that maybe Julian has read who said that this opposition is actually merging. There is a merge in progress. So I did find some evidence for the loss of the opposition between nasalized H and H, and it is very interesting because historically it was the other way around for the rest of the vast dialects. They are merging in favor of the nasalized H. So oral H is becoming nasalized as a means of enhancing the aspiration with nasalization, and that's again right now a lot of people. Right. I didn't talk with Iñaki Camino about this, but he will probably be happy because that's what he heard and his impressionistic transcriptions of these speakers show the same that I have observed instrumentally. So we have started to understand what's happening with H in the mother language. We have seen the fine detail that we can infer that has historical consequences, and now we can go back in time and look at which is possibly more familiar to you correspondences and the influence of different assimilatory sound patterns in how we reconstruct the phonetic properties of a proto-segment. In this case, going on with H, we have observed how vocalic it was in the mother language. Everything was very vocalic and looked like vowels. If we look at this set of words, we would see that our internal H and nasalized H both produced the voicing of the to the, and who produced the realization of H and H also to the. This sound pattern, which was described in less detail that was already described by Walder some years ago, let us infer that at some point in time older intervocalic H was produced not as modern H and nasalized H, like a voiced, approximate like segment, but are real fricative. It was voiceless and it was different to the modern H because it added its voicing quality to voiced segments in order to produce this first fricative of the basque language. This is already, I mean, this is after Common Basque, so we are not talking about the state that Juliet has been presented today. This is sometime, let's say, the middle, the middle ages. But we do know more about it and maybe we could go back in time and it would still be the case that H was fricative like at those points in time and not as it is now. So, yeah, now we reconstruct a voicing process of H between vowels and sonorans. Right, this has implications too for our first statement that M became H intervocalically. The usual reconstruction that I assumed until now and I think, in regard to our assumes, implies a rhinoglophilic process such as intervocalic M becoming a voiced, nasalized H. But if it is the case that it was voiceless at some point in time, we may have to assume something like this, a more learning process that has different intermediate points. Well, I'm not actually completely sure about this, but I do think that the need of this stage conditions our middle steps of the reconstruction of this particular sound pattern because now we have evidence for this middle step that we didn't have before, voiceless, intervocalic, nasalized H. So, as conclusions, yeah, we get a better general understanding of the result and context of the reconstruction. Some patterns involved phonetic, phonology, typology, and the role of contact. And the reconstruction of some sound patterns can be revised once we understand fully what's happening around them. And also, we have found some evidence that we were not initially looking for, but it was already mentioned by a dialectologist work and we found evidence for it, so now we have something else to research. Thanks a lot.