[Music]
About 30 km south of Brussels, there is a small pedestrian university town called Louvain-la-Neuve. Here, at the Université Catholique de Louvain, you'll find Europe's most renowned research environment for learner corpus research. Since Multiwrite uses learner corpora, we decided to go to Louvain-la-Neuve to visit this environment, learn from the experts and get inspired.
We sat down to have a chat with Professor Fanny Meunier, who has been working at the university in Louvain-la-Neuve and with learner corpora for more than 25 years. We asked her to tell us about the different corpora that exist at their university and how they use them for educational purposes. And what are the possibilities and pitfalls of using corpora in the field of language pedagogy? And last but not least, how do new AI tools change the field?
[Music]
Eva Thue Vold: Hello Fanny and welcome to the Multiwrite podcast.
Fanny Meunier: Hello Eva.
Eva Thue Vold: So some of our listeners might not be familiar with the concept of learner corpus research. So perhaps you could start by explaining to us: What is a language learner corpus?
Fanny Meunier: Okay, so for people who don't know much about corpus linguistics, I would describe a corpus as a huge mass of language, spoken language, written language, including all sorts of text types, registers and styles produced by all sorts of people who can have various levels of language proficiency, so from kids to older adults, and who are all learners of that language. So that's what they have in common: they're all learners of the language, so that means the language that they are producing is not their mother tongue.
Another specificity of corpora is that the language collected is stored and organized to be made available for electronic consultation. That means that all the variables that I mentioned earlier, like age, proficiency, or text type, can be used as filters to obtain the type of language you're looking for.
And perhaps two other things: additional metadata can be inserted - I won't go into details but - into the corpus so for instance a proficiency level of the people who have produced the language or other languages that they know. And also a third specificity
is that the learner corpus can be annotated by people who would like to analyze it, as for instance in the case of your Multiwrite project, you have feedback on texts that can be inserted in the corpus for further analysis. So that's basically how I would describe it.
Eva Thue Vold: Yeah perfect, and here at the University of Louvain you are renowned for your research in in this area. How many learner corpora do you have here at the university and what languages are represented?
Fanny Meunier: Okay, so I'll try to give you an idea because I don't know all the initiatives related to learner corpora here at UC Louvain. But I know, of course, quite a few initiatives. We now have an online catalogue of corpora collected at UC Louvain. I can give you the link if you want, if people want to have a look at it. But I guess also the catalogue is incomplete because not all the corpora listed are publicly available. Some people haven't inserted references to their corpora. But I had a look at it yesterday evening and I found 18 learner corpora.
Eva Thue Vold: Wow.
Fanny Meunier: Not all are publicly available, but quite a lot are. And the languages represented: English, French, Dutch, Spanish, and the corpus types are numerous. So I had a quick look, I'll just give you a few keywords. So we have corpora for English for specific purposes. We have corpora for, for instance, the International Corpus of Learner English, which is well known, is for English. It's general essays and it represents 25 different mother tongue backgrounds. So we have a Spanish narrative spoken corpus. We have learner academic writing in French and English. We have a corpus of, I didn't know about that one but (sorry for coughing), for short answers written by learners of English and graded with CEFR level. We have interactions in Spanish, we have a corpus which is a corpus of process, a learner corpus, that means based on keystroke logging and screencasting, so to see how learners write and what sort of things they do. So their writing process is recorded. We have student interviews. We have a corpus of immersion versus, let's say, traditional second language classes in Dutch and in English. We have a Tunisian lecture corpus, so it's lessons taught in English but by Tunisian lecturers. We have a longitudinal corpus of learner English. We have narration, so this one I didn't know either, it's a corpus of Dutch and French - native and learners - based on narrations on the basis of excerpts from the modern times by Charlie Chaplin. So you see you have all sorts of things, spoken interlanguage, student translation corpus, we have a corpus of native and non-native teacher talk and a French interlanguage database.
Eva Thue Vold: Yeah, that's impressive.
Fanny Meunier: They all have their little names, fancy names and acronyms, but I don't want to list all that, but you see the types.
Eva Thue Vold: But that's a broad range of different corpora and the spoken mode and the written mode and many different aspects.
Fanny Meunier: Many different aspects and that's what makes it really interesting.
Eva Thue Vold: And that must be an incredible resource.
Fanny Meunier: Yes, that's a very useful resource and as I said, I mean, for many research teams here that are working with corpora, that's really good.
Eva Thue Vold: Yeah. How many are you here that work with corpora?
Fanny Meunier: I'd say that it's difficult to give a precise number of researchers because it's an ongoing process. So sometimes you have a project and a corpus is collected. And also I want to add, people don't work exclusively with corpora, but it's one part of the data, one type of data that they can use. Then people go, but the corpus is still being used by other people. But I'd say working on a daily basis with corpora, perhaps I'd say around 20 would be a nice average count, sometimes more sometimes less, depending on the times but quite a few people. And as I said, in various research from various research teams and you could see that on the basis of the languages that are quoted.
Eva Thue Vold: Yes, because that's interesting as well that you don't only have English corpora. It's a lot of languages represented actually.
Fanny Meunier: I think it started with English and when I started working here I was appointed by Sylviane Granger and I think that she's the one who started the collection of corpora here at UC Louvain with the International Corpus of Learner English. But since then, numerous corpora, also native corpora and learner corpora have been collected in various research teams.
Eva Thue Vold: So I know that you have been for a long time interested in the connections between second language acquisition and language education. So what do you feel that learner corpus research can bring to the field of language education and language pedagogy?
Fanny Meunier: So perhaps to answer your question, I think I will first expand a little bit and put the learner part between brackets and say that corpora in general can bring a lot to the field of language pedagogy. I had a look, I mean, that was for a recent talk, but if you have a look at, I took that book as an example, the Routledge Handbook of Corpora, that was for English, English language teaching and learning, but simply to have a look at the key aspects that were addressed or that could be addressed with corpora and I'll just quote, not the headings, but you know the keywords, so you've got vocabulary you've got phraseology, so be it in terms of how those things are acquired or how they can be taught, you have grammar, grammar and discourse, pragmatics, speaking, the teaching of social conversation, the teaching of culture and intercultural communication also with the help of corpora, materials design of course, textbook evaluation, ESP, that was ESP but it could be language for specific purposes or language for academic purposes, writing, listening comprehension as well, feedback, of course, teaching, all sorts of, I mean, all those aspects, but related to the teaching part. Also data-driven learning, but for younger learners, which is perhaps less common than for older learners, autonomous language learning, and you name them. So there's quite a lot of aspects that can be sort of supported with corpora, be they native or learner corpora. So quite vast.
Eva Thue Vold: And you work yourself in teacher education, so do you use, or how do you use corpora in teacher training?
Fanny Meunier: Okay, so you also work in teacher education, so you know that it's a vast domain and we have to cover numerous areas that have nothing to do with corpora, often within a limited number of teaching hours.
Eva Thue Vold: Yes.
Fanny Meunier: So discovering the curriculum, general pedagogical principles, classroom interactions, focus on skills, differentiated learning, etc., where corpora don't play a part, at least not a big part. But if I take our own program here, I'd say that in our case, and that may vary from one university to another, but most students have been introduced to corpora during their bachelor's, and also sometimes their master studies of course and so we don't really have to introduce corpora to them or corpus linguistics because they're familiar with at least with corpora, what you can do with corpora, but if I want to narrow down to their use in teacher education, maybe I could point three aspects and things that we do here, for instance. The first one would be input. So for instance, future teachers here are introduced to corpus-based materials. And so we show them how corpora can influence or can be integrated in the creation of teaching materials. So if I take a concrete example, I, for instance, introduced them to a textbook series which is called Grammar and Beyond, and which is based on corpora. And then we show them what is the added value of having corpora included in the creation, so from the start of teaching material. So that's one thing. Another example is that we show them how corpora can be used for awareness-raising exercises, like comparing what native speakers produce and what learners produce and to show them… But I would like to say it's really awareness-raising and not so much vocabulary or grammar learning or teaching. It's just to show them that they can see those differences and that it can help them. And perhaps a third example is for course planning. So we show teachers, future teachers, how they can use corpora, for instance, to create form-focused exercises and take authentic examples from their own learners' productions so that it's more aligned with what they do instead of taking examples from, you know, exercises from textbooks which are not directly related to what their learners have produced. So specific... so pedagogical corpora, so that might be useful. So I think those three things it's already a lot.
Eva Thue Vold: Yeah, it is and if you are a teacher and your class writes a text assignment then the texts will themselves create some kind of corpus that you can use as a teacher
Fanny Meunier: Yes, they can be really used as a corpus and… building progressively, and of course with the time and resources that teachers have but they can also see it as, I think I once used the term sustainable resource, because teachers spend a lot of time correcting and giving feedback, and sometimes if they can keep that in a corpus, it's a great resource for teachers as well. So they can use it in those three ways.
Eva Thue Vold: So we have been talking a lot now about the benefits of using corpora and all the possibilities that exist, but certainly there must be some challenges or some drawbacks as well. So what would you say are the biggest challenges with learner corpus research?
Fanny Meunier: So when I, yes it's a challenge all the way, I'd say. It's a challenge all the way. So, corpus collection itself is a challenge and you have collected corpora yourself.
Eva Thue Vold: It's time consuming.
Fanny Meunier: It's time consuming. It's very complex. You need to get permissions, especially when, you know, for learner corpora and especially with younger learners. So, you need to get the permissions. You need to collect the data. Then you need to organize the data, to sort the data, to annotate the data. You don't want to lose the data. So it's a first challenge. So corpus collection. The second challenge, I think, is access to corpora. Because although corpus tools are becoming more accessible, if you want learners and teachers to access them, the tools have to be user-friendly, easily accessible, ideally free of charge because, you know, if schools want to access those. And they must be robust at the same time. So you need to have people behind this creating the tools so it's, it's time and cost and accessibility is also an issue. And then of course, well, basically the last part is the use of corpora. So I've always pleaded for...and I'll go on doing that, a pedagogically relevant use of corpora, especially if we want the teachers and the learners to use them, because for researchers, you may have very specific linguistic aspects you'd like to analyse, and that's extremely fine. And so it doesn't necessarily have to be, you know, pedagogically relevant, it can be relevant for research. But if we want teachers to use them, I think teacher-researcher collaboration is essential from the start. Because corpus collection is often prompted by research teams, and that's quite normal. They have some budgets and some projects, etc. But in order to ensure a pedagogically relevant use of corpora, I believe teachers should be involved from the start. Otherwise, we run the risk of having what I would call research-led options for corpus use, which are not necessarily what teachers want or need. And I'm not saying that research can’t prompt new usages, and that's fine, but we also need to know what teachers need and what they can do on a daily basis. Otherwise, it's just there, but it's not used. So there's little point in that.
And perhaps the second challenge, so I said it's a challenge all the way, but we have, let's say recently, a new challenge, I would say, is the competition with artificial intelligence-based tools, which should be taken into account. So if you take some typical examples that are presented in Applied Learner Corpus Tasks, so for instance, for awareness-raising purposes again, or like provision of concordance lines related to trying to find out which word structure or grammar structure is used after this or that word in this or that context, and if you want to compare the differences between native speaker usage and learner usage, you have to do a lot of things. Well, first you have to introduce learners and teachers to the corpora, to the corpus tool they need to access the tool, they need to do the exercises, so all in all you will spend quite a few hours doing that. If you ask, let's say, ChatGPT, what are the typical prepositions that are used after the verb agree, for instance, and you can ask another prompt and say: provide example sentences for each typical use, and you can even say: I want the example sentences to be related to business, medicine, economics, teaching, whatever. I want them to be easy to understand. I want them, they can be more complex. You get the answer in 55 seconds. But honestly, you do, in 55 seconds. So I tried it and I compared it with what we get when we use native corpora. So not learner corpora, but native corpora, and so if you ask teachers and learners alike what they would rather use in the long run and what makes sense and what is readily usable in their classrooms, I guess I do not have to be a fortune teller to imagine what the answer will be. So they might use chat GPT if you ask learners: okay, spend two hours looking for the prepositions that nicely follow the verb agree or ask chat GPT, they will ask chat GPT, that's for sure. I mean, maybe not in class if you force them, but outside classroom they will. So I think the challenge is to… the challenge is really to use corpora in a way that makes sense for teachers and learners. And if you look at learner corpora, there's a lot that can be done with learner corpora. And one thing that's always puzzled me is that we didn't spend enough energy, I think, on working on the content of learner corpora. Because so far, at least that's my... that's my impression and I may be wrong, because I know some people focus on content, but not enough, I think. Learner corpora are a goldmine of content. Learners express ideas, they debate things, they provide suggestions on any topic that we cover with them in class or outside class or related to essay topics that we give them. And so far, much of Learner Corpus research has focused on form. So focus on form, vocabulary, grammar, etc. As I said when I listed the keywords, there is research on social interactions, on pragmatics, but I think there's a lot more can be done on that. And to do that, we need to go for content. And so I think one thing that learner corpus will always be extremely useful for is this idea of content. And also, one other thing is, sometimes people say, well, corpora is a mass of linguistic data, but I'd rather say it's language data. And it's also content data. So we need to go for that. So content is one thing. Another thing that I find particularly interesting is the type of work that you're doing, for instance, on teacher feedback, because this is something that is really interesting. So focus on the feedback. What is the feedback that teachers give on the basis of the productions of learners? And it's also very interesting to compare that feedback, teacher feedback, with automatic feedback that you can get with AI tool. And you see the added value of humans there. So some things maybe can be done by machines and teachers can concentrate on some other things. So I think it's important, if we look at this perhaps challenge or threats that people see with artificial intelligence tools, is not to forget that there are very interesting things you can do with corpora and do those things. Do those things that are different from what the machine can do. And in one of the talks that I gave recently, I went back to one of the quotes that, I think it was Nick Ellis who said that corpora were very useful to go to the heart of the language, and to look for patterns, etc., so regularities. But machines today can do that.
Eva Thue Vold: Yeah.
Fanny Meunier: Because they've been helped with corpora.
Eva Thue Vold: Of course.
Fanny Meunier: So corpora have fed the machine.
Eva Thue Vold: That's what they have learned from.
Fanny Meunier: So machines have learned from corpora. And so maybe if machines can do some of those things, then we should go to the periphery. And the periphery of corpora is… content is one thing. And perhaps it's weird and I see it as a contradiction, because when we introduced corpora, we said the value of corpora is because it's a whole text and it's not just isolated sentences. And at the same time, we didn't focus enough on the content of that text. So I think there's room there. And also maybe another thing is multilingual corpora and this comparison and trying to... I mean, the work you're doing is very good in that respect, because you're trying to see which languages influence each other, how learners can build on their vast linguistic repertoire, so I think that's very important. And so perhaps, so again I’ll say that word because it's dear to my heart, but we must rely on pedagogy and take into account the learners and the teachers needs and we shouldn't fall prey to developers' bias or to the machine's bias, whatever, if it's a corpus tool or an AI tool. But we must really use corpora in research and teaching to invest the human added value. And I think that's... Perhaps I'll finish on that. The human added value of corpora.
Eva Thue Vold: Yes, exactly. So the technological development has really changed the field.
Fanny Meunier: Yes, it has changed. And corpora have fed this development.
Eva Thue Vold: Exactly. So that wouldn't have been possible without corpora.
Fanny Meunier: No, that wouldn't have been possible without corpora. And corpora will still be used for that as well. But I guess we must not be afraid, we must sort of focus on what is useful because otherwise it's... you don't want to be slave to the machine, you want to use it.
Eva Thue Vold: So we will still use corpora in an educational or for educational use but... in a somewhat different manner than before.
Fanny Meunier: I think we need to use them in a somewhat different manner and to really exploit the full potential. And if things can be done automatically, more easily, more accessibly, in less time by other machines, well let them do it, but let's do other things
Eva Thue Vold: Yes, then we have more resources and time to do what we need humans to do.
Fanny Meunier: Yes, what we need humans to do and what we need to use humans for.
Eva Thue Vold: Exactly. Yes, well thank you, Fanny, thank you for coming.
Fanny Meunier: You're most welcome.
Eva Thue Vold: We are so happy that you could come and be our guest in this podcast episode and share with us your thoughts on learning corpus research.
Fanny Meunier: Thanks for asking me.
[Music]