YT YouTube2Text

YouTube2Text → Transcripts → Tarek Assassi English for specific purposes “Generative artificial intelligence in domain specific l

Tarek Assassi English for specific purposes “Generative artificial intelligence in domain specific l — Transcript

by Valerii Marenichenko · 17,846 words · 2,534 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00the University of Alger. Um I'm I'm a
0:03professor at the English language and
0:05literature department and currently I'm
0:07also the head of the department at my
0:10own university in Algeria. Um my work
0:13usually goes through um English for
0:16specific purposes um as as a sub field
0:19of teaching English as a foreign
0:21language and applied linguistics in
0:24general. Um I welcome everyone and uh I
0:28hope you enjoy um the data we're going
0:30to provide and we will try to make
0:32things go very smoothly so you can
0:35follow along. Um let me just um share my
0:39screen here.
0:52All right. So I think uh for now you can
0:55see my screen.
0:57>> Yes, we can see.
0:58>> Yes, it's okay.
0:59>> Okay. Thank you very much. So um again
1:02um thank you very much professor Valerie
1:04for your uh invitation. Um I have I have
1:08checked um most of the presentations or
1:11the lectures you have um invited of
1:13course our colleagues from all over the
1:15world and I really appreciate um the
1:18topics you are dealing with and the
1:20objective behind this kind of internship
1:22which is which is really great. Um so as
1:26you can see the title here is not as as
1:29complicated as it seems. uh we will try
1:32to focus on generally speaking um
1:35education or the generative artificial
1:38intelligence in education. So usually
1:41students that we teach even
1:43postgraduates they have um let's say
1:47misunderstanding of the concept of AI.
1:50So usually when they talk about
1:52education they say AI. However, it is
1:55very important to distinguish AI from
1:58generative AI or from generative
2:01artificial intelligence because AI is a
2:03very big and very uh huge uh science and
2:07and uh kind of development procedure we
2:10can see nowadays. Um so we will focus on
2:14generative artificial intelligence
2:15because this is what matters to us as
2:17educators as as professionals of this
2:20specific domain. Uh at the same time uh
2:23since some of you might be interested in
2:25in becoming educators or um interested
2:28in education either by teaching or by
2:30doing research in education I think
2:32generative artificial intelligence has
2:34become very important nowadays uh given
2:37of course how helpful it is how how
2:40simplifying the path through AI or
2:42generative AI is today. Uh however we
2:44always have of course ethical
2:46considerations etc which we need to pay
2:48attention to. So the main um let's say
2:51the main objective of this uh
2:53presentation or this lecture is to help
2:56future educators to realize first the
2:59importance of uh generative AI um and at
3:03the same time um whatever we're going to
3:06provide today is not only related to uh
3:10foreign language education. It means for
3:13example prompt engineering how to
3:15evaluate the prompt how to provide to to
3:18get and provide feedback for gen AI.
3:21This is or this specific concept that we
3:24will discuss today can be replicated in
3:27different areas. So do not do not think
3:30like this is only for um language
3:32education. You can even if you are
3:34interested in be becoming a data
3:36analyst, if you are interested in
3:37becoming an administr administrator. So
3:40everything here is is relatable. So you
3:42do not have to worry that this is only
3:44for language education. Uh at the same
3:47time um talking about education in
3:50general might be a bit confusing. So we
3:52chose English for specific purposes or
3:54language or or specific uh language
3:57teaching. So as you can see um if we
4:01talk about English for specific
4:02purposes, it is an interesting how do we
4:06say process in language teaching. It is
4:09very highly demanded nowadays and
4:12requested by many professionals in other
4:14domains and um any foreign language
4:17teacher or any foreign language uh let's
4:20say educator will insist that um English
4:24is a lingua frana means the language
4:27used by individuals from different
4:29linguistic backgrounds. It is the
4:31international language. It is the
4:32language of science and technology
4:34nowadays. everything new as a discovery
4:36or everything new as uh as as an
4:39invention is shared in the English
4:41language to reach u a bigger audience.
4:44Um so the focus on English for specific
4:47purposes is intentional of course not
4:49because only it's my research interest
4:52but at the same time because it's highly
4:54recommended and highly required in
4:56today's job market or international job
4:58market. Um so as educators of the
5:01English language it is very important to
5:04pay attention to the learner because
5:05this is one of the most important
5:07aspects of education today or language
5:09education more particularly. Um however
5:13um some will say that ESP is a different
5:17field or English or teaching English for
5:19specific purposes is a bit different is
5:21a bit challenging is is a bit difficult
5:23but as a matter of fact it's easier
5:25because teaching ESP it means you are
5:27not going to teach the whole language or
5:29the English language you are going to
5:30teach a very specific chunk of the
5:32language that these learners actually
5:34need. So they can need it of course for
5:37or they may need it for academic
5:39purposes if they are students,
5:41researchers, professors or they may need
5:44it for occupational purposes such as for
5:46example a lawyer, a pilot, a doctor. So
5:50it would be easier for them of course to
5:52learn that language they need not all
5:55the the English language. And in ESP we
5:58always see that in different ESP courses
6:01that these learners will be highly
6:02motivated. So you do not have to worry
6:05about okay my learners are bored um they
6:07are going to give up the session. So ESP
6:11learners in theory of course they are
6:13always motivated. So this will help you
6:15develop your competencies. This will
6:17help you develop your teaching materials
6:19and teaching skills um because having a
6:23motivated learning learner sorry is
6:25always very helpful in having what we
6:27call um a successful and efficient teach
6:31and learning experience. So um first I'd
6:35like you to take a closer look at this
6:38specific statement here. So we say if AI
6:41or generative AI can generate a language
6:44lesson in 30 seconds does that mean that
6:46it can design a good professional
6:48communication course. So if you consider
6:52this because we know that generative AI
6:54or AI bots or chat bots can do really
6:59like very uh impressive things nowadays.
7:02You just have to ask a question and you
7:04will receive everything. However, if um
7:07you ask this chatbot to design a lesson,
7:10we have to consider that this lesson is
7:13it really efficient or not given that
7:16this chatbot does not know who I am as a
7:19teacher and does not know who my
7:21learners are because um ESP relies on
7:25your learners specifically. Usually when
7:28you you design an ESP course for the
7:30first time, the first thing you need to
7:32do is to collect data about your
7:34learners. Who are they? Uh what's their
7:36profession? Uh what do they need English
7:39for? What are their necessities? What
7:41are their lacks? Uh what do they want to
7:44learn from the language? Where are they
7:46going to use the language? So all of
7:48these data are extremely important for
7:50us to collect and analyze. Then we start
7:52designing the lesson. So if you take
7:55like 30 seconds to think of this, you
7:57will see that yes, it can design a
8:00generative uh or let's say more generic
8:04lesson that is a bit general that is not
8:07very specific to my learners. So in this
8:09case we can say that as far as
8:12generative AI it is very important to
8:14use generative AI but we always have to
8:17pay attention to what it produces or
8:20generates as educators or or as course
8:23designers.
8:25All right. So um as you can see here the
8:28um the results we may have from um from
8:32AI can be of course very fluent.
8:35However, fluency is not fitness for
8:38purpose. As we as we have seen that I
8:41can design lessons. Uh I have very good
8:44lessons. Uh they tackle very specific
8:46language skills or language aspects such
8:50as vocabulary, grammar, pronunciation,
8:52etc. that annoying.
8:55>> However, it won't be as efficient as we
8:58expect because all of these data or all
9:01of these language aspects produced by AI
9:04are not really uh designed specifically
9:07for my learners. So um and this takes us
9:11to that saying we have in in um using
9:14gen AI when we say the more specific you
9:17are uh with with these chat bots the
9:21better results you will have. So second
9:23point materials ignore learners needs um
9:27level and prior knowledge as we said um
9:30of course the um generative AI or chat
9:34bots in general do not know who are my
9:36learners specifically that's why I
9:38always recommend the use or the users of
9:40of AI to be more elaborate with a
9:43chatbot provide as much data as you can
9:46um even with this process you will
9:48minimize the threat but you are not
9:50going to eradicate the threat of of um
9:52generic generic materials. Um also
9:56professional realities, genres and
9:58conventions are absent and this is what
10:00you can see when you try to ask Chad GPT
10:04for example for something. You will see
10:06that uh the answers seem generic and
10:09designed for everyone but not for me
10:11specifically and not for my uh learners.
10:14So when we say genre for example here we
10:16are referring to that type of language
10:18used in a in a specific profession. So
10:21um we will try to um share let's
10:25[clears throat] say two or three
10:26examples but we will try to focus more
10:28on aviation because aviation and
10:30medicine are two fields in ESP that are
10:34critical. It means usually it's a matter
10:37of life of or death breakdown of
10:39communication. It means it's it's a big
10:41deal. Uh so when we say genre we are
10:44referring to that kind of language used
10:45by those professionals. When you are
10:48listening to two doctors speaking to one
10:50another the way they speak the language
10:52they use the the competencies they share
10:54with one another is what do we call the
10:56genre of the language. Uh number four
10:58tasks are activities but they are not
11:01not not authentic communicative events.
11:04So in ESP it's not like general English.
11:07So some of you or many of you of course
11:09you have attended English language
11:11courses and um you can see that you have
11:14some kind of exercises related to
11:16vocabulary to grammar to pronunciation
11:19punctuation etc. In ESP we try not to
11:22focus on this. We focus on something we
11:25call competencies. Uh for example as a
11:29doctor for example we teach them a very
11:31specific competency they need in their
11:33job. as an example we can say write in a
11:35prescription. So the the lesson will not
11:38be grammar, vocabulary, pronunciation
11:41etc. So we will directly focus on this
11:43task and in applied linguistics in
11:46dactics we always call this task based
11:49teaching and with several tasks at the
11:52end we have something we call
11:54projectbased teaching. So at the end we
11:56have a a whole project based on
11:58different tasks we dealt with during the
12:01unit or the set of lessons we designed.
12:04So it is important to uh when we teach
12:06ESP we always focus on competencies
12:10writing it prescription for example for
12:12doctors um for nurses for example
12:16describing a condition to uh to a
12:19patient. So it's not vocabulary or
12:22grammar but how to describe that
12:23specific condition. for pilots for
12:26example um discussing an issue uh or an
12:30incident or an accident reporting data
12:34all of these we call them competencies
12:36and they are very simple of course you
12:38can just same time you can ask ch you
12:41can ask chat bots what are the main
12:43competencies pilots need doctors need
12:46lawyers need and you will base your of
12:49course your teaching on these um
12:51competencies
12:53um some of my of course uh students and
12:56even doctorred students they ask me how
12:58can we describe a competency? So a
13:01competency is very simple. A competency
13:03is you take a language skill and you
13:06take um a task in that job. For example,
13:10if we say writing a prescription. So
13:13prescription is something they do in
13:14their jobs. These doctors and writing is
13:17a language skill. So when you mix them
13:18together basically you have what do we
13:20call a competency as a professor for
13:23example lesson design you are teaching
13:25at university or in secondary school one
13:28of your tasks is to design a lesson for
13:30example I am teaching mathematics in in
13:32in secondary school or in university um
13:36I need to design this I I'm teaching
13:38with my own language of course um then I
13:42want to design this lesson in English so
13:45basically you have the lesson design as
13:47a competent So you will help your
13:49learners who are ESP learners of course
13:52to be able to design a lesson. So this
13:54is what do we call a competency. So
13:56language in this case your focus you do
13:58not have to worry. Um for example I'm
14:00teaching pilots I do not have to be a
14:02pilot uh because ESP is related to
14:05teaching the language with time of
14:07course you will learn many things about
14:09aviation. If you are teaching um
14:12engineers for example, you do not have
14:14to be an engineer because ESP is
14:16teaching the language. Um some students
14:19may say okay what if they ask me things
14:22about their specialty. Well, it's not
14:25your specialty and they know they know
14:27you are an English language teacher. So
14:28you do not have to worry about being
14:30knowledgeable about all of these topics.
14:33Uh finally of course we have accuracy is
14:35not assumed. uh we have to always make
14:38sure that the content we are providing
14:40for our our learners is accurate enough
14:43to not say 100% accurate because you are
14:46not a domain specialist you are not a
14:48pilot you are not a doctor so you are
14:50teaching a language so you can of course
14:53always verify this and uh you can seek
14:56help from of course subject specialists
14:59to help you uh design more accurate uh
15:03lessons as far as knowledge of the
15:05domain is concern concern.
15:07Um, okay. So, um, I'm not I'm not sure
15:10if you have any questions. So, if anyone
15:13has questions, of course, you just
15:16intervene whenever you want and we'll
15:18try to simplify things more.
15:21Um,
15:23all right. So uh the central aim uh we
15:27have here is
15:29exactly the value of generative AI in
15:31domain specific pedagogy which we know
15:34that it lies in transforming generic
15:36language instruction into authentic
15:38needsdriven adaptive and professionally
15:41relevant learning experiences. Um when
15:45we uh have some for example we have some
15:47examples of our colleagues who are
15:50university professors who are pilots who
15:53are doctors and lawyers uh they always
15:57seek these ESP courses but they do not
15:59really know uh what is an ESP course
16:01because they know English is a language
16:03I want to learn English and that's it.
16:05So when they call us and they say okay I
16:07joined an English language course in
16:09such school and they say I literally
16:13gave up after two to three sessions and
16:15when we ask them a question why they say
16:18I do not feel that this uh course is
16:21efficient and it's not suitable for me
16:24they say they talk about different
16:26topics they they uh they talk about
16:28different language points we are a
16:30heterogeneous classroom everyone is from
16:33a different domain the level might not
16:36be very close. So all of these problems
16:38will demotivate the learner and they
16:41will not be able to pursue their studies
16:43and develop their linguistic skills.
16:45However, in ESP, we always create what
16:48do we call an authentic learning
16:50experience. Uh in ESP you have to
16:53remember that any ESP course designed
16:55especially for occupational purposes uh
16:57the learners must be from
17:00at least a close by uh field. For
17:04example, I have taught pilots. Okay, all
17:06of them are pilots. I have taught
17:08engineers. Some of them are mechanical
17:11engineers or others are electronics in
17:14into electronics. But all of them are
17:15actually engineers. So, uh they are not
17:18exactly they do not have exactly the
17:20same job but their jobs are quite
17:22similar. So, this is also okay because
17:24it's very difficult sometimes to have um
17:2815 learners all of them from exactly the
17:31same field. So this is not really a bad
17:33idea. So you can you can manage. So if
17:36you have uh how to say if you have um a
17:39group of pilots or a group of engineers,
17:41you will create what do we call an
17:43authentic relevant learning experience.
17:46So they can feel that all of them have
17:48the same objective. all of them have the
17:50same job and you create that um that
17:54environment that is very helpful that is
17:56very interactive uh that is very
17:59authentic to their domain and they will
18:01feel definitely that they belong to the
18:04classroom so they are highly motivated
18:06they they are very interactive they ask
18:08questions they they interact with you so
18:11this is very important in ESP and this
18:14is u one of the best solutions if we may
18:16say to uh learning English language for
18:19specific purposes. So we have to realize
18:22now with the use of AI that uh
18:25generative AI generates options. Uh it
18:28does not generate uh perfect solutions.
18:30So the educator in this case has to
18:33design the learning. So you you receive
18:34data from generative AI but you have to
18:37evaluate, you have to think, you have to
18:39uh learn at the same time. Then of
18:42course you provide the learning
18:43experience for your learners. Um number
18:46two promoting is or sorry prompting is
18:48instructional design but not magic
18:51wording. So do not take it for granted.
18:53Not uh when you design for example a
18:56prompt uh do not expect the the results
18:59to be as specific as you expect.
19:03However, the more specific you are and
19:05we will see some examples. The more
19:07specific you are with your prompt, the
19:09better always.
19:11Now um as for human validation here is
19:14not optional. It is pedagogy itself. Uh
19:18you are a pedagogue. You are an
19:20education uh professor or professional.
19:23Uh you have to realize that um your
19:27human touch is essential to the teaching
19:30learning experience. uh if and if if we
19:35take a look at the very amazing advances
19:38we have in AI uh at first we have
19:43noticed I think during the first or
19:45second year of developing this new
19:47technology
19:49uh we have noticed that it's not very
19:51authentic or AI is not very authentic it
19:54is not very interactive and um it is not
20:00really u how to say specific to what I
20:03ask for. However, with these advances,
20:06especially during like 2025, 2026, we
20:10have seen that the major advances went
20:12directly to this specific point, which
20:14is interactiveness in education. So, AI
20:17has become extremely
20:19um interactive, which is amazing, which
20:21is great. However, uh if this was the
20:24case, I believe we can get rid of
20:27teachers. However, this is not the case.
20:30Up till now we still have teachers. We
20:32still have teachers with their touch. We
20:34still have to humanize the classroom.
20:37Even if we are having like now an online
20:39classroom, um it is still I mean the
20:43teacher or the educator in general is
20:45still high highly needed and precious to
20:48the teaching learning experience.
20:52Now um if now we try to dig deeper into
20:56um ESP and and generative AI. So ESP or
21:01English for specific purposes is not
21:03general English plus technical
21:05vocabulary. And this is a misconception
21:07we have with many educators um not only
21:11in [clears throat] in um in the southern
21:14part or the non-native let's say uh part
21:17of the world. Uh however even with
21:20native speakers so even native speakers
21:22they still not um they are not very well
21:25aware of ESP and how different it is
21:27from uh general English. So the purpose
21:32um in this case of ESP
21:34um so we try to identify professional
21:38and academic learning goals uh it is
21:41very important to do this in the very
21:42beginning of the uh teaching process. Um
21:46let's say you get a call like you are an
21:49ed educator or teacher you get a call uh
21:51for a job they say we have a group of
21:54learners from such field and we want you
21:58to uh teach them English okay they are
22:00not going to say ESP because they do not
22:02know this however if you tell them who
22:05are the learners and this this is like
22:08from very personal experience um they
22:11always get confused in the beginning
22:12they say okay I'm not sure let me just
22:15check etc.
22:16So learners profile is very important. I
22:19have to know who are the learners. I
22:21have to sit down with them. I have to
22:23talk to them. I have to discuss their
22:25issues, their problems. So the more I
22:27learn about my learners, the better the
22:30learning experience will be. Uh also the
22:33genre itself. So the texts and events.
22:36Um this is what do we mean by genre
22:38which is which are those communicative
22:41events we have in the in the job. So uh
22:44you can just go to this place where they
22:47are working if they are engineers or
22:50doctors, pilots etc. and you try to
22:52learn about the job itself. Here
22:55remember we are talking about inservice
22:57professionals means professionals who
22:58are actually working. Now you may ask
23:01does this work with pre-ervice
23:04professionals? It means someone who is
23:06studying engineering and he will become
23:08an engineer. Yes, that works as well. It
23:11will not be as how do we say as
23:13interactive as you expect because this
23:15person or this student did not start
23:18working yet but still they know that
23:20they are going to be engineers. So they
23:22have an idea about engineering about the
23:25job which we call in ESP the target
23:28situation. So they have an idea about
23:30this. So when we say general we are
23:32referring to those texts those those um
23:35words expressions um let's say
23:39conversations that happen in the target
23:42situation um so two engineers talking to
23:46one another. So this communicative event
23:49they are two engineers talking to one
23:51another about a specific topic in a
23:54specific place in a very specific time
23:56or on a very specific time. So this is
23:58what do we call a communicative event
24:01and this is what's important to us of
24:03course uh to be teaching in the ESP
24:06course of course they are doing it in
24:08their native tongue so we will do this
24:10in the classroom in English and this is
24:12ESP this is as easy as it gets um next
24:16we have discourse conventions so how
24:19those texts are structured so uh you
24:22know that sometimes in um in different
24:24fields they have to go through a process
24:28For example, if if we teach a group of
24:31engineers, I have taught in in a group
24:34of engineers in in in a water factory,
24:37the water that we drink, of course, uh
24:40mineral water. So, they were a group of
24:42engineers. Every single one of them was
24:44the head of a specific department. So,
24:47one of them, for example, the um
24:49logistics, maintenance, production,
24:53electronics, etc. So at first it seemed
24:56a bit difficult to teach them because
24:57everyone has a specific task. However,
25:00if we want to make this um more
25:03interesting, we have to be a bit
25:05creative. So what what should we do in
25:07this case? We create what do we call
25:09scenarios. So for example, I say uh we
25:12have a pan or we have a problem um with
25:16production as an example. Suddenly we
25:19have an electronic or sorry an electric
25:22issue. So basically production will
25:24stop. So the head of the department of
25:26production has to interfere. We have a
25:29problem for example with um um spare
25:33parts. So someone who is in who is the
25:35head of the logistics department has to
25:38order this specific part. Uh maintenance
25:41has to react as well. They have to do
25:43something. Um HR maybe they have
25:46something to do with this. So they have
25:48to interfere. If they if they do not no
25:50problem they just say we do not have to
25:53interfere the administration has to do
25:55something as well. So you see it's a
25:57whole process. So what happens next?
25:59What happens second? What happens third
26:02fourth etc. So when we teach using this
26:04process everyone will be interested
26:07because they are actually waiting for
26:09their contribution. So this might be
26:11more interactive in teaching ESP.
26:15Um okay. So as far as the tasks are
26:17concerned now um what learners must do
26:20with the language. So we have to
26:21understand that teaching English for
26:23specific purposes um does not really
26:27have to do with acquiring the language.
26:29So we acquire the language then we use
26:31the language. That's why ESP is pretty
26:34limited. When we say limited it means it
26:36becomes easier for me to teach and
26:39easier for the learners to learn. We do
26:41not have to learn everything. I just
26:43learn what I need and whatever is needed
26:46is going to be used in the target
26:47situation is going to be used in the
26:49job. So if something is not really
26:53feasible for them means they cannot
26:55really learn it they are not going to
26:57use it. If um something is too difficult
27:00for them so logically speaking they do
27:03not really need it. So you do not have
27:04to teach it. You do not have to worry
27:06about this specific language skill or
27:08this competency. So always ESP relies on
27:11logic. If something sounds logical, do
27:14it. It does not sound logical, yeah,
27:16just just avoid it. Now um as far as
27:19knowledge is concerned, u the content
27:22and practices behind the language, um
27:25some
27:26some students who have already graduated
27:29my course, uh the course is called an
27:32introduction to ESP and it is an
27:34undergraduate course means they are
27:36trained to be ESP teachers. uh sometimes
27:39they ask the question and say they say
27:41this is hap this happened more before AI
27:45they say how can I know what happens in
27:47their job um I'm not a pilot or I am not
27:50an engineer so what should I I'm not a
27:52businessman so we always say make things
27:55very easy for them we say okay ask them
27:57just ask these professionals what is
27:59your job what do you do etc but now with
28:02generative AI it became way easier you
28:05can just ask jp and they will and it
28:08will definitely share very interesting
28:11practices with you. However, your human
28:13touch is always required. So you have to
28:15check and recheck etc.
28:19Now um
28:21so as far as the first step of of um how
28:25do we say of of ESP or teaching uh
28:27specific language or it can be even
28:30other languages can be um I don't know
28:33Ukrainian for specific um purposes you
28:36can teach uh French for specific
28:37purposes. So usually we say language for
28:40specific purposes. So what learners can
28:42do now versus what target situation
28:45demands. So here we have to be careful.
28:47Anything that is not needed will not be
28:50taught. Um this definitely will limit or
28:55um how do we say it limits the effort
28:57you are going to do. So you do not
28:59you're not going to spend a lot of of
29:01energy on your design. Uh according of
29:05course to Hutchinson Waters in this this
29:07has begun in 1987.
29:10Um they they describe these needs as
29:12necessities, lacks and wants. So my
29:15learners they have specific necessities.
29:18It means things they really need and
29:20they are going to do in their jobs. For
29:22example, I can say um language aspects
29:26such as learning how to ask a question,
29:28learning how to explain. Uh then we have
29:31lacks. It means something they know but
29:35they are facing problems with. Here I
29:37can just give them a a small test like a
29:40speaking test, a writing test. then I
29:42can detect what are the lacks
29:45and the wants are things that they
29:47really want to learn but maybe they are
29:50not needed so I can consider them but if
29:53I feel like these wants are not really
29:56needed in the job so I can drop them so
29:58no problem now you can say um how can I
30:01really uh get all of these or this data
30:05so it's very simple um let's say
30:07theoretically speaking
30:10needs analysis is the First step you do
30:13in in before you start teaching. Usually
30:16we do it through data collection tools
30:18such as questionnaires, interviews,
30:20tests, observation etc. Um this the only
30:24problem with this is that it takes time.
30:28Now when we say it takes time, we're not
30:29talking about years. We're talking about
30:31a week for example 10 10 days. So it's
30:33not a big deal. However nowadays with
30:36generative AI it became a bit easier.
30:39You can speculate the needs, the
30:41necessities like swans, but you cannot
30:45rely 100% on generative AI. Why? Because
30:50uh chat bots, they do not know your
30:52learners. Uh so they can share or chat
30:56bots can share with you some of the
30:57insights, some of the necessities, lacks
31:01or wants you may [clears throat]
31:03experience with your learners, but not
31:06with every specific learner. not with
31:09learner one and two and three that you
31:11already met and you can see and you can
31:13interact with etc. So um doing what do
31:17we call a dual experience uh doing them
31:20both because generative AI does not take
31:23much time. You [clears throat] can do it
31:24like in five to 10 minutes. You can
31:26finish the whole job in 10 minutes. Uh
31:29and doing it like face to face would
31:31definitely help you design a better a
31:34better course. So how can we do this? Of
31:36course um how or how can generative AI
31:40help you? uh it can help you through
31:42drafting questionnaires, interview
31:44protocols and task inventories. Um this
31:48when you when we do it manually uh takes
31:50a bit of time. So you can ask uh chat
31:54GPT for example to do it for you. Then
31:56you check or recheck evaluate the work
31:58then you distribute it to your uh
32:01audience be it like learners,
32:03professionals, uh administrators,
32:07um ESP teachers if they have ESP
32:09teachers. So all of these we call them
32:11sources of information. These are the
32:13people who will uh be contributing to
32:16the design of the course. Um um namely
32:19we usually have five means when I
32:21conduct needs analysis I have to ask
32:24five let's say um five sources of data
32:29or information. Number one is always the
32:32learner. Of course the learner is the
32:34most important part. Uh number two is
32:36the ESP teacher in case we have an ESP
32:39teacher which is not always the case. Uh
32:41number three is the professional. So if
32:44I am teaching pre-ervice engineers, I am
32:46going to ask a real engineer, someone
32:49who is working. Uh we have policy makers
32:51as well. They are very important. The
32:53head of the department um the head of
32:56this company etc. They are important
32:58because they will help us with mostly
33:01with logistics when to teach at what
33:04time which area who are the learners
33:08they are the ones who are going to pay
33:10you obviously. So all of these are are
33:12very important to the ESP course as
33:14well. So usually we collect data from
33:16all of these people then we analyze data
33:19and at the end we come up with something
33:21we call competencies. Now u competencies
33:24of course as we said before are very
33:26simple like the the task we do in the
33:30job plus a language skill. So, as we
33:32said, writing a prescription
33:35um check or or answer in or um fill in a
33:38checklist because you know a pilot
33:41before they operate any flight, we
33:44always have two pilots, the the
33:46commander and the co-pilot, uh we have
33:48two checklists. They have to check the
33:50airplane from the inside and the
33:52airplane from the outside. So, they have
33:54a checklist they have to read and keep
33:56checking. Okay, this is okay, this is
33:58okay, etc. So this can be one of the
34:00most important lessons. Um also pilots
34:04for example they need to
34:06report an incident. For example let's
34:09say a pilot is on cruise means they are
34:12flying of course and there is I don't
34:16know an incident there is u um a very
34:19sick passenger. So what should they do?
34:21They they have to report this incident
34:24to the tower or to the company etc. So
34:27how can they do this? By saying these
34:29expressions, these specific expressions.
34:31And this is my lesson. Why? Because this
34:34is needed in the future. So they will
34:36need to learn it because it happens in
34:39the um of course in in in um in the
34:43target situation. Um you may ask the
34:45question, do I have to design all of
34:48this? Um in a way, yeah, yes and no.
34:52I'll give you a very simple example. For
34:54example, when we check aviation English,
34:57maybe you can see this. Um, it is I'm
35:01not sure it's it's a bit blurry. So, it
35:03it is a book called Check Your Aviation
35:05English. Of course, I can share it with
35:07you later on if you want. Um, so
35:10teaching teaching aviation English, we
35:12can rely on readymade courses or course
35:14books. Um, you are you have a group of
35:17of pilots, you can just go online, you
35:20can buy a book called aviation English.
35:22It it is it is feasible and it is good.
35:25However, that book was not designed
35:27specifically for your learners. This is
35:29the small issue we have with readym made
35:31course books. I have for example a group
35:33of of farmers for example. Why not? I
35:37have a group of streamers. I have a
35:40group of influencers. I have a group of
35:42hairdressers. All of them can actually
35:44attend ESP courses. Do I find like every
35:48single course book ready online? Yeah,
35:51it is possible. Um, maybe not possible
35:54with new jobs or so-called jobs like
35:57streamers, like influencers because they
35:59are current and contemporary jobs, but
36:02with other jobs, yeah, it is possible.
36:05So, you can download that book, buy that
36:07book and you can use it with your uh
36:09students. However, is it always
36:12feasible? Not always because that book
36:15was is purely generic. It deals with
36:18standard or standardized
36:20um uh target situations. It is very
36:24good. It can help you but you always
36:25have to pay attention to your learners
36:27because their needs are different. Their
36:29lacks are different. Their target
36:31situation is the same but linguistic
36:33background is different. So being uh
36:36being from for example Ukraine or from
36:38Estonia, from the United States, from
36:40Japan is not the same thing. we have
36:43different linguistic lux because there
36:45is always something we call the how do
36:48we say the mother tongue interference
36:50uh mother tongue interference can be
36:52with vocabulary with grammar with
36:54pronunciation and that's why if you
36:56listen to some recordings online for
36:59example on YouTube when two pilots to
37:02talk to one another or one pilot talks
37:04to um the tower or air traffic
37:06controllers
37:08uh the the pronunciation sometimes is
37:10difficult to understand uh because a
37:13Japanese pilot is not like a French
37:16pilot. They have diff different accents.
37:19Um and this is one of the main problems
37:22we have for example with aviation
37:23English nowadays. Um
37:26those standardized tests uh focus only
37:30on on native speak or or native speech
37:33sorry which is wrong actually. It is
37:35very wrong. Uh and this happens with all
37:37aviation English tests because to become
37:40a pilot you have to acquire three
37:42certificates. This is all over the
37:44world. You have to get your license as a
37:47pilot. You have to get your uh health
37:49license as well or health certificate
37:51that you are healthy and you have to get
37:53the English language certificate. So if
37:55you cannot pass the test of the English
37:58aviation English, you cannot be a pilot.
38:01So these tests all of them I have
38:03checked many many of them of course and
38:05I have discussed with many many
38:06educators as well most of the recordings
38:09are native speech and the target
38:12situation or the job itself is 75 to 80%
38:17of pilots all over the world are not
38:19native speakers. So I cannot learn or
38:22train my ear um for for uh let's say
38:26with with native speech and when I am a
38:28pilot I am speaking to a Chinese pilot
38:31and to an American pilot then to a
38:33British pilot. So it is quite so the
38:36balance between the two is not uh is not
38:38is not really good. So this is one of
38:40the main problems we have with high
38:43standard or high risks tests. Um so how
38:47can AI actually contribute as we can
38:51said as as we can say of course with
38:53drafting questionnaires interview
38:55protocols task inventories so you can
38:58save a lot of time you do not have to do
39:00this manually so just ask JPT to do this
39:03for for you then you of course you check
39:06how [clears throat] how good the work is
39:09um number two you can simulate
39:11stakeholders perspectives for
39:12triangulation
39:14um triangulation usually means I'm not
39:17sure if you are familiar with the uh the
39:19concept but triangulation means I get
39:21different data from different people
39:23using different um tools just to make
39:27sure that there are no there is no bias
39:29there is no subjectivity if I get data
39:33from one learner for example and design
39:35the whole lesson based on that learner
39:38this would be wrong okay because
39:40especially if there is a group so I have
39:43to make sure to get data from different
39:46learners so I can make sure that I cover
39:49most of their needs not just the needs
39:51of one learner. So this is what you mean
39:53by triangulation. At the same time when
39:55we u when we refer to data collection
39:58tools such as questionnaires,
39:59interviews, observation, tests etc. I
40:03cannot rely only on one. So for example
40:05I cannot design questionaire for ESP
40:08learners and ESP teachers and
40:10professionals. No, it would be better to
40:12diversify because a questionaire is very
40:15limited. They just check and answer in a
40:17written form. But an interview is more
40:20important in case you can be face to
40:22face with this person. They can they can
40:25say more. They can show more. So
40:28interviews provide a larger set of data
40:30and a more accurate set of data. And we
40:34do not have to forget of course the
40:35facial expressions, the hand gestures,
40:38the paral linguistic features of
40:40language which are very important and
40:41they say a lot about this person.
40:45Uh finally it speeds up analysis. So it
40:47does not replace the replace sorry the
40:50an the analyst but it speeds up the
40:52process. Um one of the most important
40:56benefits or advantages of chat bots is
40:59timesaving.
41:01So uh instead of analyzing by hand or
41:05using non let's say nonAI powered tools
41:10such as SPSS such as uh some formulas on
41:13Excel you can rely on Chad GPT it gives
41:17you like not only CHP of course but any
41:20chatbot they can actually help you with
41:23analyzing data in minutes especially
41:25large quantities of data. However can I
41:28rely 100% on that? No, this is just to
41:32speed up the process and to save time
41:34and energy. However, I will need to
41:36recheck of course uh the data later on.
41:40Um okay, as we said before concerning
41:43the u um the target situation, for
41:46example, the target situation of a pilot
41:49is in a cockpit or in in um in a cockpit
41:53on an airplane. So this is the target
41:55situation. uh a doctor for example the
41:58target situation is the emergency room
42:00or the operation room for for surgeons
42:02of course um an engineer is a factory um
42:06I don't know a biologist in a laboratory
42:08so this is what do we call of course the
42:10target situation so as we said genre
42:13what do we mean by genre is the typified
42:16rhetorical uh response to a recurring
42:18situation means the kind of language
42:21reaction used in a laboratory the kind
42:23of language and reactions used in in a
42:26in an emergency room etc. So
42:28authenticity here is very important. So
42:30the task that we teach in our classroom
42:34has to mirror a real communication
42:36event. Um some some students they say um
42:40is it possible to teach outside of the
42:42classroom? Yeah, it is possible. It is
42:45it actually it's better if if you have
42:47the chance to teach an ESP course for
42:50biologists in a laboratory. Yes, do it.
42:52It's perfect because they can see
42:54everything in front of them. They are
42:56actually in the target situation. So
42:58they can practice the language in uh the
43:00target situation which is which is very
43:03helpful for your learners and very
43:04helpful for you to learn more about
43:06their about their jobs. Um so uh next of
43:12course we have uh an example. Now we
43:14will try to practice together um of
43:17course together in a way that suits you.
43:20Uh if again if you have any questions
43:22any any um how do we say any
43:25interventions go ahead no problem. So
43:28how do we use generative AI across
43:30pedagogical uh cycles or here we have
43:33six applications again these ones are
43:37related specifically to specific
43:41language teaching. However is this
43:43applicable to other domains? Absolutely.
43:47You can rely on these of course with a
43:48few tinkles, change changes etc. with
43:51tinkering with some of the words and
43:53some of the steps. You can actually do
43:56this and um you can rely on this process
44:00in basically many many jobs not only
44:03language education even if you are into
44:06education but in management yeah the
44:08these will definitely work with you or
44:10for you. So um as far as language uh
44:14design or specific language teaching
44:16design uh number one of course you can
44:20uh use generative AI for needs analysis
44:23means you analyze the needs of the
44:25learners using chat GPT or chat bots.
44:28Number two it can help you with
44:30curriculum and syllabus design. It means
44:32designing those lessons for every single
44:34session. Number three, you can it can be
44:38extremely helpful with specialized
44:40materials because if you are teaching
44:43medical doctors, you are not a doctor.
44:45How can I know about those difficult
44:48words, expressions about the target
44:50situation in hospitals? Um, it's
44:53difficult for me. No. Um, chat bots can
44:56be extremely helpful with this. They can
44:59simplify the job for you. Um it can also
45:02help with differentiation and
45:03personalization which is very important.
45:06So uh usually the language used by
45:09doctors and nurses even if the target
45:11situation is the same uh the language is
45:14not the same. So the doctor a doctor
45:16speaking to another doctor is different
45:18from a doctor speaking to a nurse is
45:21different from a doctor speaking to a
45:22patient. So the level of technicality
45:25will be risen of course or or lowered
45:28when we move down. A doctor speaking to
45:31another doctor is very technical. A
45:34doctor speaking to a nurse is less
45:36technical. But a doctor speaking to a
45:38patient is basically non-technical. So
45:41the doctor has to be extremely extremely
45:44simple simplifying giving examples using
45:47plain English or plain language in
45:49general. So there would not be breakdown
45:52of communication. And this is um that's
45:56why I focus a lot on aviation and
45:58medicine or medical English for medical
46:00purposes because it is what we call a
46:04high stakes uh field. Uh breakdown of
46:07communication is forbidden especially in
46:10aviation and and uh and medicine.
46:14Um AI also can help or generative AI can
46:17help with assessment and feedback. This
46:19is very important. You can um collect
46:22the data like tests for example or
46:24quizzes from your learners. You can
46:26upload them to uh a chatbot and ask for
46:29an analysis for example or some feedback
46:33and it will definitely be very helpful.
46:36And finally within the test uh sorry the
46:38teach and learning experience it can
46:40really help you with simulations and
46:41role play. I have tried this and
46:44basically I wouldn't be able to do this
46:47in that short period of time because it
46:50saves you a lot of time. First you do
46:51not know the target situation but when
46:53we ask uh for example any chatbot to do
46:56this for us create simulations uh create
46:59roleplay exercises in medical English.
47:02So chat bots can give you really
47:05interactive and amazing um practices
47:09that you can use in the classroom.
47:11However, one thing that we need to
47:12consider is the the the context you are
47:15teaching with because sometimes I teach
47:17in
47:18we can we can call them poorly equipped
47:21classrooms. Uh if if your classroom is
47:24poorly equipped, you have to explain
47:26this to to the chatbot. So it can
47:28provide you with simulations and role
47:30plays that are applicable to your
47:32classroom.
47:34So all of all of these are applications
47:36we can uh rely on using chatbots. Now
47:40how can we use this? We can ask or we
47:43can apply each and every single step
47:47but um we have to pay attention to these
47:50points we have below. First we have to
47:53check with the challenge. So we have to
47:55see what is challenging about needs
47:57analysis. What is difficult about needs
47:59analysis? Then how can generative AI
48:03contribute to this task to needs
48:06analysis? Then I'll give you an example
48:08of of using generative AI for needs
48:11analysis. Then I'll share with you a
48:13prompt very simple prompt that you can
48:15copy and paste. And finally we will
48:18discuss the risks. What are the risks I
48:20may find when I use um generative AI in
48:23needs analysis. And finally of course we
48:26have the teacher role. What is your role
48:28as a teacher at the end of using uh
48:31generative AI for this task. So let's
48:34start with number one.
48:36So using generative AI in needs analysis
48:40um as we have said before needs analysis
48:42is very simple means it's about
48:44collecting data analyzing data and
48:47filtering data that's it what do I mean
48:50by analyzing data or collecting data you
48:53just ask questions for learners for for
48:56professionals etc then you try to filter
48:59data this is needed this is not needed
49:01etc and finally to analyze data it means
49:04you come up with competency
49:06I have asked a few doctors, a few
49:08professionals in in in medicine etc. I
49:11have realized that my learners will need
49:13to learn how to explain um for example
49:17medical uh substances, how to ask
49:20patients about their condition, how to
49:24inform nurses about processes, how to
49:27write a prescription,
49:29um how to to to consult using the phone
49:33for example or a video call if the
49:35person is is far away. So all of these
49:38are competencies. So lesson one, lesson
49:40two, lesson three and that's it. How can
49:42I know about these competencies? You
49:45have two ways. Number one is to ask
49:47these people what is your job? What do
49:49you do? Notice them. Go to a hospital.
49:51Just notice them what what they are
49:53doing. Or the easiest path is to ask
49:56Chip. So what are the competencies that
49:59I need to teach in my ESP class for
50:02medical doctors? So chip will say number
50:04one, number two, number three, etc. So
50:06this is the easier path of course.
50:09Uh now the challenge we have in needs
50:11analysis no the challenge we have in
50:14regular process of needs analysis it is
50:16time consuming uh because it requires
50:19triangulation of learners tasks
50:21stakeholders etc. So it takes a lot of
50:23time so in this case I can use chpt for
50:27example to to uh to simplify uh the
50:30path. Uh now how can generative AI help?
50:34It can help you draft instruments which
50:37we call them data collection tools such
50:38as interview schedules, task
50:40inventories, questionnaires, tests. So
50:44all of these tools we use them to
50:46collect data. So chart GPT as an example
50:49can help you with this. Um again I use
50:52charge GPT but um for just as an example
50:55you can use any uh platform or any tool.
50:58They are of course all very very
51:00helpful. uh as an example here uh
51:03stakeholder role play to pilot interview
51:06questions. Uh one of the data collection
51:08tools we use is an interview and you
51:11have to design the questions you have in
51:13an interview. Then you have to go to
51:14this person and you have to ask the
51:16questions and record the answers and you
51:18go home and you transcribe those answers
51:22and analyze them. So you see this takes
51:24a lot of time. So in this case I can ask
51:28um generative AI for for for uh how do
51:32we say for for the exact questions I may
51:35ask then I can modify of course and when
51:38I ask these questions of course I get my
51:41answers via uh a phone call facetoface
51:45interview video call then I do not
51:48really have to transcribe I just take
51:50those recordings I upload them then
51:53generative AI will analyze and give me
51:55the data. So you can see it's very uh
51:57very simple. Now one one prompt we can
52:00use is I can ask generative AI to draft
52:04a 10 item needs analysis questionnaire
52:07or interview for nurses preparing for
52:10international uh sorry placement for
52:12example the B1 or B2 uh level um let's
52:17say the common European framework of
52:19references of course. So here mentioning
52:21the level is is extremely important. So
52:24the risk we may find is that generic
52:26items will be generated and these will
52:29miss the local realities. So your your
52:32your job here is when you receive the
52:34data from generative AI, you have to
52:36validate every item against the real
52:39target situation. So you have to think
52:41about the target situation and see if
52:44whatever you have got from generative AI
52:48is really mirroring the target situation
52:50or the needs or not. So this is for
52:53needs analysis. Now the how can how can
52:57we integrate generated AI in curriculum
53:00and syllabus design. The challenge we
53:02have with designing the syllabus the
53:04lessons the objectives etc is that we
53:08may find it difficult to align
53:10objectives and the content and the
53:12sequences and the assessment. All of
53:14these are in syllabus design. So when
53:17you are an educator you are not relying
53:20on a very readymade course book you are
53:23designing your own course book you will
53:25find some problems in aligning objective
53:28which objective comes first which thing
53:30I'm going to focus on more which thing
53:32I'm going to focus on more on less how
53:35am I going to sequence which lesson
53:37comes first which ones come second how
53:39many units I'm going to have u how can
53:43gradation which is uh work gradation we
53:47mean from simple to complex um
53:52frequency of occurrence is important as
53:54well. What do we mean by frequency of
53:56occurrence? If u let's say when we talk
53:59about aviation English most of the
54:03skills or let's say the skills they use
54:06the most in aviation are listening and
54:08speaking.
54:10So they really rarely write or read. So
54:14this is what the target situation tells
54:16us. So when I am teaching I cannot teach
54:20them using reading and writing all the
54:22time because they will be bored. They
54:24know that in the job I will need to
54:26listen and speak a lot not read and
54:28write because as a pilot I will need to
54:31write a report and read a checklist but
54:35rarely. I mean I'm not going to write a
54:37report at the end of every single
54:39flight. This does not work. a report you
54:41write it when you have an incident and
54:43accident and it's very rare. However,
54:46speaking and listening happens at every
54:48single flight. So this is what they mean
54:50by frequency of occurrence. The things
54:52that or the aspects that happen a lot in
54:55the target situation I am going to teach
54:58them more. Simply put, this is the
55:00easiest way to to manage uh course
55:02design. So um so all of these points you
55:07can see within the challenge uh label
55:09here are the problems we have when we
55:12design a course. So how can generative
55:14AI help? It can draft the syllabus maps
55:17for you. It can design the course
55:19outlines. It can help you with objective
55:21banks. So if you upload what you came up
55:25with in um in step number one which is
55:29needs analysis all of that data if you
55:32upload it to generative AI tools or or
55:34or chat bots it will definitely help you
55:37with this with alignment objectives it
55:39will say okay put this objective first
55:42and this is second with the content they
55:44it will definitely help you simplify the
55:47the content move from something simple
55:49to something complex. it will help you
55:51with the sequencing. So Judge GPT for
55:54example can say uh okay it would be
55:57better if you discuss this point two or
56:01three times not only once because it
56:03happens a lot in the target situation.
56:05So you need to remind your learners
56:06about this and it will definitely help
56:09you with designing assessment because
56:11sometimes we may commit a mistake. uh I
56:14design a lesson that is very interesting
56:16that has a very specific objective but
56:19at the end within the assignment the
56:21assignment does not really match what I
56:23have taught. So generative AI can help
56:27you with providing options then you
56:30check those options and see which one is
56:32more suitable to the objective of the
56:34session. So as an example uh you can
56:37backward design modules or module map
56:40with task based sequencing. So you can
56:42ask for um how do we say uh a very
56:46specific lesson in which it relies on
56:49task based quizzes or applications of
56:52the language and it will be easier for
56:55you of course to save time and energy um
56:58asking chaptt to do this for you as as a
57:01prompt. uh you can see here propose a
57:0412week syllabus for academic writing for
57:06doctoral students task based outcome per
57:09outcomes per week and you share of
57:12course what you want to have in your
57:13course design and it will definitely
57:15give you something really interesting
57:18however we always have the risk of
57:20illusions so uh we may have
57:22hallucinations we may have plausible but
57:25misleled uh tasks or or or
57:28um how do we say exercises so Your your
57:32job here is here sorry is to um to see
57:36this product against what you had before
57:39in needs analysis is can we actually see
57:42that there is a match or not. If there
57:44is something missing if there is
57:45something that is not feasible that is
57:48not valid that is not reliable just
57:51remove it and change it to something
57:52else that is compatible with the needs
57:55of your learners. Uh now with number
57:58three, how can generative AI help you
58:01with specialized materials? This is
58:03actually one of the most interesting
58:05points because as we said before, you
58:07are a language specialist. You are not a
58:10pilot and you are not um how do we say
58:13you are not a lawyer. Um you may find
58:16problems in in actually designing
58:19authentic materials. So authentic texts
58:22are copyrighted. So maybe they are not
58:25available online. they are scarce or in
58:29some jobs of course not all of them or
58:32they they might be level inappropriate.
58:35Um when we teach ESP usually we try to
58:38explain the difference between material
58:40sorry authentic and nonauthentic. So
58:43when we see a nonauthentic material is a
58:45book like the one I showed you before.
58:46it's called English for aviation
58:50or if um for example we use English for
58:54lawyers
58:55language for lawyers so whenever we find
58:57the word English or the word language or
58:59the word grammar or the word vocabulary
59:02this is what we call a nonauthentic so
59:04it's not original to the domain it is
59:07original to teach in a language but what
59:10do you mean by authentic for example if
59:12I am teaching English for lawyers
59:14I have a group of lawyers I am teaching
59:16them English. If I use the constitution
59:19as an example or examples from the
59:21constitution, the constitution is an
59:24authentic material. Why? Because it was
59:26not designed to teach English. It was
59:29designed for these people. I'll give you
59:31another example. If I am teaching a
59:33group of uh farmers for example, um I
59:37have I think another book for yeah
59:39English for agriculture. So if I find I
59:42find the word English for agriculture,
59:44it must it means this book was designed
59:46to teach English. So it's nonauthentic.
59:49However, if the book is called modern
59:52agriculture now, this book was not
59:54designed to teach English. This was
59:56designed to teach agriculture. So for
59:58me, this is authentic as a material. I
1:00:01can use it to teach my class. However,
1:00:03it would be difficult for me and for my
1:00:06learners. Why? Because it was not
1:00:08designed to teach the language. So the
1:00:10text might be level inappropriate. It
1:00:13might be difficult to read because this
1:00:15was designed by someone who knows
1:00:18already English and he is specialized in
1:00:20the domain. So it would be difficult for
1:00:22me even as an English language teacher
1:00:24to understand those concepts. Maybe
1:00:26agriculture is not as difficult or but
1:00:29we're talking about more technical
1:00:32fields such as electronics, such as uh
1:00:36renewable energies etc. So it would be
1:00:38difficult for you to create what do we
1:00:40call a level appropriate text or a level
1:00:43appropriate lesson. So generative AI in
1:00:46this case can be extremely helpful. It
1:00:49can give you what do we call scaffolded
1:00:51versions of authentic genres. So it
1:00:53moves smoothly from something very
1:00:55simple, something very specific to
1:00:57something more challenging to not say
1:00:59difficult. Uh it can help you with
1:01:02scenario generation as well. It can give
1:01:04you scenarios that you do not even know
1:01:06that they exist in this field. So you
1:01:08can read them and you can understand the
1:01:10field more. So as an example in aviation
1:01:13English of course. So radio telephony
1:01:15exchange uh nursing handovers in
1:01:17medicine and business pitches or pitch
1:01:19in in uh in in business or English for
1:01:22business or businessman. uh as a prompt
1:01:25you can write create a B1 or B2 level
1:01:28pilot controller exchange routine then
1:01:32non-rine situation the non-rine
1:01:34situation is hydraulic failure for
1:01:36example um use standard standard
1:01:39phrasiology
1:01:41so standard phrasiology we use it in
1:01:43routine situations like taking off
1:01:45landing on cruise etc uh and plain
1:01:49English we use it for non-rine
1:01:50situations so this is a simple prompt
1:01:54that you can of course replicate in
1:01:56different fields.
1:01:58So um
1:02:00the risk we have in using uh generative
1:02:02AI for this task is that in it can of
1:02:06course invent procedures and present
1:02:08them as very confident procedures. Here
1:02:12uh it is a risk. U does it really affect
1:02:16your teaching learning experience? Maybe
1:02:18yes but it's not like it's not fatal to
1:02:21a certain degree. So you do not really
1:02:23have to uh worry a lot because as a
1:02:26teacher at the end as you can see you
1:02:27have to verify the data you have to
1:02:30verify this against in aviation against
1:02:32against the IKO docs you can find them
1:02:35online IKO stands for international
1:02:37civil aviation organization so you can
1:02:40find these uh docs online they are free
1:02:42and available and you can see if these
1:02:46pieces of data are correct or not uh for
1:02:49medical English you can you can verify
1:02:51with clinical protocols also you can
1:02:53find them online um or real genre
1:02:56samples. If you want to have something
1:02:59easier, you can just check with a
1:03:01professional in the domain. It would be
1:03:03easier for you.
1:03:06All right. For um now using generative
1:03:08AI in differentiation and
1:03:10personalization of learning um one of
1:03:13the main challenges of being an educator
1:03:16is we say is to cater to learners needs
1:03:20or individual needs.
1:03:22Usually we have like great classes with
1:03:2410 to 15 learners. It's a great number
1:03:27of of learners. You can be interactive
1:03:29etc. But usually we have that every
1:03:32learner has a very specific learning
1:03:34preferences. They like to learn a
1:03:36certain way. So for you as an educator
1:03:39or a teacher or a professor to teach
1:03:41them the way they like is a bit
1:03:43difficult because everyone is special.
1:03:46So one class many levels mixed needs and
1:03:49this is or even mixed learning
1:03:51preferences. This is the challenge we
1:03:53have in personalizing learning
1:03:55experience. So how can generative AI
1:03:58help? It can create um how do we say
1:04:03same core tasks at multiple levels which
1:04:05is very good. It can alter alter sorry
1:04:08create alternative supports. Um
1:04:11sometimes we get stuck like I have some
1:04:13materials but I need to simplify the
1:04:16material. So yeah I can pick pick
1:04:19something from an authentic material
1:04:20which is very difficult um like a video
1:04:23online or or a text from a very specific
1:04:27book and I can upload this to to
1:04:29generative AI and say please make this
1:04:33material an A2 level material. it can it
1:04:36can definitely help you with this and
1:04:38create more versions for you which is
1:04:40which is very helpful for you of course
1:04:42so as an example uh we have one aviation
1:04:45scenario which I may feel that it's a
1:04:48bit difficult so A2 support versions
1:04:51plus a B2 stretch version so you begin
1:04:54with something simple with A2 level then
1:04:56you give them something a bit more
1:04:58challenging so generative AI can help
1:05:00you with this with the same material it
1:05:02can create two scenarios which is which
1:05:05is very actually helpful. Uh the prompt
1:05:08can be you can of course use this adopt
1:05:10this handover scenario. This is in
1:05:13medical English. Adopt [clears throat]
1:05:15this handover scenario for the CFR A2
1:05:18level with sentence frames. Then a B2
1:05:22extension requiring uh inferences. So in
1:05:25this case they will create or ch as an
1:05:28example will create two exercises. one
1:05:30of them is simpler than the other. Of
1:05:33course, one of them is uh for warm up
1:05:36then something is a bit challenging. So
1:05:38they will feel or your learners feel
1:05:39that they are um into the classroom and
1:05:42they will be very motivated to work
1:05:44more. Um the risks here are accidental
1:05:47simplification of content itself.
1:05:50Um it is a risk but it's it does not
1:05:53really affect you badly as as a teacher.
1:05:56So you may have like very simple content
1:05:59about aviation or about medicine. Uh
1:06:02learners will grasp it very quickly but
1:06:05this is not a problem for me because I
1:06:08do not really pay attention to the
1:06:10content. I pay attention to the language
1:06:12because again I have to remember that I
1:06:14am an English language or a language
1:06:16teacher not a specialty domain teacher
1:06:19or a course or specialty course teacher.
1:06:22So this is not very like a really big
1:06:24risk. No, for you you can of course
1:06:27differentiate language load um avoid
1:06:30discip sorry make sure that there is
1:06:32some kind kind of dis disciplinary rigor
1:06:35um but mostly I will focus more on
1:06:38language if the language is learnable in
1:06:42a way if it's simple simpler to grasp if
1:06:45learners are motivated and interactive
1:06:47I'm I'm good to go I don't have to worry
1:06:49about the uh the content itself
1:06:53okay um number Five, assessment and
1:06:56feedback. Uh, how can generative AI help
1:06:59me in or outside of the classroom as far
1:07:01as assessment and feedback for my
1:07:03learners or my learners is concerned? So
1:07:06the challenge is that assessment and
1:07:08feedback is has to be timely means you
1:07:10have to keep um your continuous
1:07:13evaluation
1:07:15you have to keep do doing it actually
1:07:17continuously so you can manage the
1:07:19teaching learning experience and um as
1:07:23professor um Valerie asked me already to
1:07:25to to to provide or design an assessment
1:07:29um I have seen that it is a good idea
1:07:31you have to design beforehand but at the
1:07:33same time it's always better to check
1:07:36what your learners grasped during during
1:07:38the course of course or the lecture. So
1:07:41you can design something that is
1:07:43for example achievable at least or
1:07:46something that is uh close to what they
1:07:48understood. So um generative AI can help
1:07:52you develop rubric aligned draft
1:07:55feedback exemplar generation. uh it can
1:07:58give you like um um a more interactive
1:08:02test or a more interactive assessment
1:08:04procedure. So this is this is very
1:08:06interesting actually and I encourage you
1:08:08to use this um whatever your specialty
1:08:11is. As I said before this is not only
1:08:13for language learning but for all uh
1:08:16fields this actually can help. So the
1:08:18prompt here is using this rubric comment
1:08:21on task achievement organization
1:08:24register quote the text for each
1:08:26comment. Um so with every um let's say
1:08:30uh how do we say like like like task it
1:08:34can give you comments about the task you
1:08:37have designed or even that AI has
1:08:39designed. So you can know exactly how to
1:08:42uh record the task, how to give grades,
1:08:46how to be more more importantly how to
1:08:48be fair to learners because fairness is
1:08:51one of the most important aspects of of
1:08:53of assessment in evaluation. So the risk
1:08:56here is that there is some kind of
1:08:59confident wrong judgment. So here it's
1:09:03it's always your work to verify the
1:09:06samples to decide uh to disclose uh the
1:09:10institutional policy. Maybe your
1:09:11institution asks you to do um
1:09:14assessments a certain way that the AI
1:09:18has provided and it's not really
1:09:20interesting. So you have to manage or to
1:09:23revise um at least the type of
1:09:26assessment or the type of grading
1:09:29provided by generative AI. Uh finally we
1:09:33have sim simulations role play. Uh these
1:09:35are actually very interesting as well.
1:09:38Um so the mo the [clears throat] most
1:09:40important challenge we have in this
1:09:41aspect is that classrooms cannot
1:09:44reproduce professional interaction. So a
1:09:47classroom is not a cockpit. A classroom
1:09:49is not an emergency room. A classroom is
1:09:52not a factory. So to recreate these
1:09:56scenarios in the classroom can be very
1:09:58challenging. So in this case, how can
1:10:00Chad GPT or any platform work help you
1:10:03is that it can provide um tasks or
1:10:07scenarios that are applicable to the
1:10:09classroom which I really recommend of
1:10:11course uh especially if you have what do
1:10:14we call underresourced classrooms or or
1:10:17schools. This can be really helpful. As
1:10:20an example, you can create a non-rine
1:10:22ready telephon patient explaining
1:10:24symptoms. Um, first in aviation and
1:10:27second in in in medicine. Um, in
1:10:30business uh classes, you can design
1:10:32difficult client negotiations. In
1:10:36academic English, you can design thesis
1:10:38defense viva.
1:10:40You can create these scenarios on chat
1:10:42GPT then apply them into the classroom.
1:10:45So um generative AI can actually help
1:10:47you even with what you should bring to
1:10:49the classroom with you as as physical
1:10:51materials. Even if you do not have
1:10:54something you can ask Chip to uh provide
1:10:58an alternative suitable to you. Uh as an
1:11:01example act as an air traffic
1:11:03controller. This is in aviation. Reply
1:11:05only via radio telephony or radio. Give
1:11:08me vectors that are information about
1:11:11the flight. Then unexpected runway
1:11:14change. So this is a scenario of an um
1:11:16how do we say of of an aviation or a
1:11:20flight situation. So you can check this
1:11:23with a GPT and it can give you this
1:11:26scenario and you can take the scenario
1:11:28as it is share it with your um pilots or
1:11:31the pilots you are teaching and of
1:11:33course create a whole scenario a whole
1:11:35assess h sorry assignment around this
1:11:38scenario and they will be very
1:11:40interested because it's their job. Of
1:11:42course uh the risk is unscripted realism
1:11:45drift. In some cases the scenarios might
1:11:49not be very realistic and applicable.
1:11:52They cannot be realistic because they
1:11:54are not real. So you have to recheck
1:11:56this and they cannot be applicable
1:11:58because they are very difficult to apply
1:12:00in the classroom. So in this scenario we
1:12:04can be as teachers we can define the
1:12:06scenario monitor debriefs or for example
1:12:11um um
1:12:13let's say calibrate uh the um the
1:12:17assignment or the scenario to meet the
1:12:19expectations of of the learners. So you
1:12:22have a job to do as a teacher. You are
1:12:24not going to rely 100% on uh on on chat
1:12:27bots.
1:12:29Okay. Um what we have next here is
1:12:33prompting as a pedagogical design. So
1:12:35what what we mean here is that we will
1:12:38try to help you design the prompt
1:12:40itself. Now we have talked about how can
1:12:44um generative AI help me with those
1:12:48steps. Uh all of these steps of course I
1:12:51have to use prompts. Now in this last
1:12:54part of the presentation we will try to
1:12:57to learn how to design a prompt because
1:12:59as we said in the beginning the more
1:13:01specific the prompt is the more specific
1:13:04and better results I will have from
1:13:07chbpt as an example. So prompt is not a
1:13:10request is not only a request but it is
1:13:12a specification.
1:13:14So a generic prompt will give you
1:13:16generic output. It means um a simple um
1:13:22undetailed
1:13:23prompt will definitely not be a good
1:13:25idea because it will give you what do we
1:13:27call a generic output. So
1:13:31do not expect garbage in gospel out.
1:13:33This does not happen in on on on these
1:13:36platforms. So the prompt should carry at
1:13:38least some of these um concepts we have
1:13:42here. So number one it should include a
1:13:44context. It should include a domain. It
1:13:47should include learners, who are the
1:13:49learners. It should include also
1:13:51objectives, tasks, pedagogy means
1:13:54teaching and learning. Uh constraints or
1:13:56problems. Uh output format, what kind of
1:14:00output out output you expect. So you
1:14:02have to to to be sure do you want it
1:14:04like in a text, in an audio, in a video,
1:14:07in a word file, in a PDF file, in a
1:14:09PowerPoint presentation. Uh and finally,
1:14:12validation. You have also to make sure
1:14:15that there is some kind of validation
1:14:17before you validate it yourself. So
1:14:20writing the prompt means making your
1:14:23design decisions or decisions explicit
1:14:26because the more you discuss uh prompts
1:14:29with your um or your chat or chatbot the
1:14:33more the chatbot will learn about you.
1:14:35So I do not recommend doing this only
1:14:38once. So do it twice, three, four times
1:14:40just so the chatbot learns more about
1:14:43you and your learners.
1:14:45So weak prompts or simple prompts are
1:14:47not lazy wording but they are unfinished
1:14:50thinking means you have not finished
1:14:52thinking yet. You have to come up with
1:14:54something more specific.
1:14:56Now uh these are nine prompt design
1:14:58frameworks. Again you can just take a
1:15:01look at them. However in some cases you
1:15:04do not have to use all of them. So you
1:15:07do not for example in every prompt I
1:15:09have to make sure to use all of these
1:15:13pieces of data in my prompt because in
1:15:14some cases I can I can neglect one or
1:15:17two no problem. So the first thing as we
1:15:19said you have to share the context.
1:15:22Context is usually called setting
1:15:25setting plus people. Setting is time and
1:15:28place and people it means who are these
1:15:30people or the topic you are discussing.
1:15:32So the context you have to explain it
1:15:34from the very beginning to the chatbot.
1:15:37Number two, you have to explain the
1:15:38domain. So if you are teaching an ESP
1:15:41course from the very beginning, you have
1:15:44to explain what domain it is. And do not
1:15:46just say medicine. You have to say what
1:15:49kind of medical subfield it is. So if
1:15:53it's a general practitioner is not like
1:15:55a specialized doctors. It's not like a
1:15:57surgeon. It's not like a nurse. It's not
1:16:00like a medical administrator. It's not
1:16:02like a biologist, um, a laboratory
1:16:05assistant. All of a pharmacist for
1:16:07example. So all of these people belong
1:16:09to the medical field. Even medical
1:16:11researchers are concerned. So make sure
1:16:14do not just mention the domain and you
1:16:16stop. Always mention what is the exact
1:16:19domain you are dealing with because the
1:16:21data later on of course will be more
1:16:23specific and easier for you to grasp.
1:16:26Number three. Number three, sorry. Who
1:16:28are the learners? You have to explain
1:16:30are they native speakers, non-native
1:16:33speakers? Um do you have a a mixed uh
1:16:38linguistic uh background? Um because
1:16:41having
1:16:42two, three, four learners from different
1:16:45linguistic backgrounds, you have to
1:16:46explain this to to to the chatbot
1:16:49because the the output later on will be
1:16:54better in meeting those specific needs.
1:16:56Um, in pronunciation, for example, if
1:16:59you have five to 10 pilots, if they
1:17:02belong to different linguistic
1:17:03backgrounds, they will or the chatbot
1:17:06will understand that, for example,
1:17:08someone who is Spanish or speaks Spanish
1:17:11will have a problem with the sound V
1:17:13because the Spanish cannot really
1:17:15pronounce the sound V. If, for example,
1:17:19they are from the Middle East, people in
1:17:21the Middle East cannot pronounce the
1:17:23sound P. So, they always say B. So in
1:17:26this case if pronunciation is important
1:17:28in this field it would be better to
1:17:31explain this of course to the chatbot so
1:17:33they can take this into account. Grammar
1:17:35same thing there are some languages um
1:17:38in which their grammar is very extremely
1:17:42different from the English language
1:17:43grammar. So you have to explain this as
1:17:45well. Uh of course not explain the
1:17:47language but just give the background of
1:17:49your learners. um the communicative
1:17:52tasks that you are expecting from of
1:17:55course the chatbot you can explain this
1:17:57as well. The pedagogical approach. So
1:17:59for example, if you are teaching an
1:18:01online course is not an on-site course.
1:18:03So you have to be careful. If it's
1:18:05blended learning online and on-site,
1:18:07it's it's different as well. There are
1:18:09things that I cannot do with you. Now if
1:18:11we were in a classroom together, we may
1:18:14be more interactive because I can see
1:18:16the learners. I can feel the learners. I
1:18:18can feel the energy. I can see the
1:18:20learners if they are uh interactive or
1:18:23not. So here you have to explain this as
1:18:25well. Um the constraints what problems
1:18:29do I have in my classroom? Is my
1:18:31classroom well equipped? Uh do I have
1:18:33bad internet uh connection? So all of
1:18:36these issues you have you have to
1:18:38explain them as well. Uh output
1:18:40requirements. Uh what do I want as a
1:18:44kind of output? um do I want them to be
1:18:48um specific competencies, specific
1:18:50lessons um uh how do we say graded
1:18:54lessons like gradation from simple to
1:18:57complex etc. So all of this the more you
1:18:59share of course the better. And finally
1:19:01validation criteria you can share the
1:19:04validation criteria you you are going to
1:19:06use later on. So they can or the AI bot
1:19:09can take it into account. So as again as
1:19:13we said before you do not need all nine
1:19:15every time but uh the more you use the
1:19:18better of course when you design your um
1:19:20prompts.
1:19:22Okay. So here we have two examples. One
1:19:25of them is a very generic prompt. Create
1:19:27an English lesson about communication in
1:19:30aviation.
1:19:31It's not a bad prompt, but it's it will
1:19:35definitely give you something that is
1:19:37too generic and um you cannot really
1:19:39apply it in a classroom. So what I
1:19:42recommend is using the nine points we
1:19:44have talked about. Not all of them of
1:19:46course but I always recommend using the
1:19:49word act or act as because this is a
1:19:52very important word that will definitely
1:19:55learn what is your uh profile as as an
1:19:58educator. So act as an aviation English
1:20:01instructor familiar with the AKO
1:20:02language proficiency requirements.
1:20:05Um so for example Chad GPT in this case
1:20:08will definitely go to the AKO um
1:20:11website. Uh it will definitely download
1:20:15those uh documents. It will analyze them
1:20:17and it will definitely take into account
1:20:20those language requirements or required
1:20:22by this uh huge of course organization.
1:20:27Then uh design of course professional
1:20:29pilot controller communication. Uh
1:20:32design a a 45 minutes or minute B1 to B2
1:20:36activity for traininey traininee pilots.
1:20:39So here the profile is important. We
1:20:42said trainee pilots. We did not say
1:20:44pilots. So we know that these pilots are
1:20:46not operational. They are not working
1:20:48yet. They are still student pilots. And
1:20:51this is totally different from teaching
1:20:53operational pilots or inservice pilots.
1:20:56So the objective is to understand and
1:20:59respond to routine and non-rine radio
1:21:02telephony exchanges. Uh include
1:21:05scenarios uh listening tasks,
1:21:07phrasiology, speaking activity, rubrics,
1:21:10etc. Uh distinguish standardized
1:21:13phrasiology from plain English because
1:21:15in aviation it's different. Uh do not
1:21:17invent procedures. It means stay
1:21:20authentic. Do not speculate. Do not give
1:21:23me hallucinations about um scenarios
1:21:27that do not really exist in reality. And
1:21:30finally, flag content needing
1:21:33validation. This is also very important
1:21:35to put at the end of your prompt. Why?
1:21:38Because if uh the chatbot does not
1:21:42really understand something, it will
1:21:44flag it for you. It will tell you,
1:21:46please check this if it's suitable or
1:21:48not. So it's easier for you to check
1:21:51rather than having to check everything
1:21:53once again. So you are going to save
1:21:54some time. Now um it can go the same way
1:21:59with different fields. We have health
1:22:01care management, public administration,
1:22:04academia, tourism. Uh in a healthcare
1:22:07you can deal with nursing handover or
1:22:10those small you know charts that include
1:22:12uh data or information about the
1:22:15patient. uh in management we can discuss
1:22:19difficult conversation simulation with
1:22:21debrief rubric. This is a very specific
1:22:24competency uh that chart GPT will give
1:22:26you and you use it uh in in how do we
1:22:30say in a prompt to make sure that you
1:22:32want um materials based on this prompt.
1:22:35Uh public administration which is also a
1:22:38very u let's say demand demanding job.
1:22:42So plain language policy brief for
1:22:44citizens. This is a very good competency
1:22:48to develop or that needs to be developed
1:22:50for these people who are of course
1:22:52public administration officers. In
1:22:55academia, you can
1:22:57u design courses about doctoral writing,
1:23:00reviewer, response letter, genre etc. So
1:23:03this is also can be used in in these
1:23:05prompts. And finally in tourism one
1:23:08competency which is complaint handling
1:23:10dialogue or register shift practices um
1:23:13which which might be interested
1:23:15interesting a lot in in tourism or being
1:23:18a touristic guide or tourist guide. Um
1:23:21so the domain may change as we said
1:23:24before here we have five domains but the
1:23:28nine elements do not. So this is
1:23:30applicable those nine points we
1:23:32discussed here these ones are um usable
1:23:37valid for all fields not only not only
1:23:40in language teaching but you can use
1:23:41them even in if you are asking chpt to
1:23:45help you revise your lesson for example
1:23:48you can use these nine of course and
1:23:50they can be very very helpful.
1:23:53All right. So, um the human and AI
1:23:57pedagogical collaboration. No. Um as you
1:24:00as you have seen up till now, um I think
1:24:04it's it's it's clear to say that um AI
1:24:07cannot do everything and the humans can
1:24:10really rely on AI to uh to to produce
1:24:13more and to save a lot of time and
1:24:15energy and resources sometimes. So the
1:24:18scenario for example here B1 and B
1:24:20traininey pilots routine and non-Rine
1:24:22situations can actually
1:24:26can can actually be done by AI alone and
1:24:29can actually be done by by humans alone.
1:24:32If AI does it, it will definitely be or
1:24:35there will be definitely gaps. If humans
1:24:38do it, it will take a lot of time and a
1:24:41lot of collaboration with subject
1:24:42specialists. So the collaboration in
1:24:44this case actually is very interesting
1:24:47because first it will definitely be just
1:24:50an initial request you are discussing
1:24:52with CH GPT. Then you have a generic
1:24:55output that you need to of course check.
1:24:58Uh you need to diagnose to see what has
1:25:01been flagged that you have to validate.
1:25:05Uh then you provide something we call
1:25:06context sensitive prompt. Now the um the
1:25:11AI bot has learned a lot. You are going
1:25:14to evaluate then finally you are going
1:25:16to validate and modify. So here um if I
1:25:21tell you that
1:25:23um learning those nine points and using
1:25:27them to uh create prompts is not enough
1:25:30because you are going to use those nine
1:25:33points and you still will have to learn
1:25:35more. You still have to validate, you
1:25:37still have to modify etc. So it is a
1:25:40continuous effort but it is um a lesser
1:25:43effort than doing it [clears throat]
1:25:44sorry than doing it yourself.
1:25:47Okay. So now how can we actually uh
1:25:50sorry how can we ensure that there is
1:25:52collaboration? We have four or five very
1:25:54simple uh steps. So step number one,
1:25:57instead of course of using a very simple
1:26:01generic prompt such as make an aviation
1:26:03English language or sorry English
1:26:04listening exercise, the output here will
1:26:07be a dialogue about a flight to Paris, a
1:26:11multiplechoice quiz, a word list,
1:26:14discussion in pairs, etc. So you may ask
1:26:17the question, is this actually usable or
1:26:19not? Does it really help my learner
1:26:22learn a specific um competency or not?
1:26:26So if you have this output, the second
1:26:29sorry the the next step you will use is
1:26:31try to see what kind of vi violations we
1:26:36have in comparison to the nine steps uh
1:26:39you know process we have explained
1:26:41earlier. So number one if we take a look
1:26:44here uh it is uh a listening test or
1:26:48assignment. So they will listen and they
1:26:50will actually have a quiz, a word list
1:26:53or multiplechoice questions. Does this
1:26:56really happen in in when when a pilot is
1:26:59working? Is a pilot going to listen then
1:27:01they are going to check? No, it does not
1:27:03really happen in aviation. So the first
1:27:05problem we have is that there is no
1:27:08radio telephone genre. We do not have a
1:27:11language of pilots that is used in the
1:27:13target situation. We do not have it in
1:27:15this task. So this task is not really
1:27:17good. Number two, the second problem we
1:27:20have is that we do not have domain
1:27:22constraint or phrasiology or plain how
1:27:25do we say plain language. So the
1:27:27language itself does not reflect the
1:27:29language used in the target situation.
1:27:32Uh the third uh problem we have with
1:27:34this uh prompt
1:27:37is that um it is a quiz and a quiz does
1:27:42not really help the learner do the
1:27:45language. they will help they will help
1:27:46the learner remember the language
1:27:49acquire the language. So in ESP this is
1:27:52not a good idea. I do not care if the
1:27:54learner really learns or memorizes
1:27:57words. Does this learner know how to use
1:27:59those words? If not it means my teaching
1:28:03learning experience has failed. Uh
1:28:06number four uh the fourth problem we
1:28:08have is that the level is unknown. I did
1:28:11not mention the level and even the
1:28:14output given by chart GPT was not u was
1:28:17not leveled was not designed for a
1:28:19specific level and finally nothing to
1:28:21verifi verify against. So as you can see
1:28:24here uh the output provided by ch has
1:28:28has not flagged anything for for ch this
1:28:32is a perfect uh proposition and for me
1:28:36this is not the case. So as you can see
1:28:39I have used a generic uh prompt. I have
1:28:42received a generic output and all of
1:28:45these are problems
1:28:48uh wi with the generic output I I have
1:28:50uh you know I have received from CH GPT.
1:28:52So my recommendation is uh the more
1:28:55specific you are the better in every
1:28:57single domain. You are a teacher. You
1:28:59are a doctor using cha GPT. You are uh I
1:29:02don't know a data [clears throat]
1:29:03analyst. You are an information and
1:29:05technologies um information technology
1:29:08specialist. It doesn't matter what's
1:29:10your job, what's your occupation, what
1:29:12is your status or professional status.
1:29:16It is important to go through those
1:29:18nine. Of course, I'm not asking you to
1:29:20remember them or to memorize them, but
1:29:22just consider them when you design your
1:29:24next prompt. Okay. Um finally how to
1:29:28create a very sensitive uh you know
1:29:31prompt based on the mistake we have
1:29:33committed earlier with the generic
1:29:35prompt. So act as as we said before
1:29:39always use this very important act as an
1:29:42aviation English instructor familiar
1:29:44with the AKO language proficiency
1:29:46requirements. Create a listening based
1:29:49activity for B1 or B2 traininee pilots.
1:29:53a pilot controller exchange 90 seconds
1:29:56routine clear for takeoff. So these
1:29:58these details you do not have to
1:30:00memorize them but just ask Chad GPT to
1:30:03to help you with this. Uh use only
1:30:06standardized AKO phrasiology. Mark any
1:30:09plain language as segments. Then provide
1:30:12tape scripts, five comprehension
1:30:14questions requiring interference, a
1:30:1610-minute speaking role play with roles
1:30:18and a four-level performance rubric. So
1:30:20you can see you can ask everything in
1:30:22one prompt. No problem. You do not have
1:30:23to keep asking one by one. It can
1:30:25definitely analyze the prompt even if
1:30:28it's I don't know a thousand words long.
1:30:31So you do not have to worry about this.
1:30:33And finally flag any content that
1:30:35requires verification. As we said before
1:30:37at the end I always begin with act as
1:30:40and I finish with flag. Give me any
1:30:43problems you you may face. So, uh it
1:30:45would be easier for you, of course, to
1:30:47recreate um how do we say um a better
1:30:52version or a more specific version of of
1:30:55uh the prompt. And finally, of course,
1:30:57when you finish um finish, you know,
1:30:59dealing with the flagged problems, etc.,
1:31:02uh of course, you are going to validate
1:31:05means read again, check again, and see
1:31:07if everything is into place. For
1:31:10example, here um check against element
1:31:13number five flagged items that we have
1:31:15asked of course GPT to to do at the end
1:31:19of our prompt. It will for example flag
1:31:22that phrasiology is verified against
1:31:25AKO. So you do not have to pro to to
1:31:27worry about the content. Um bird strike
1:31:31which is how do we say in aviation it's
1:31:35it can be like um a small problem. It's
1:31:39not a big problem in aviation operations
1:31:41but it is still a problem. It means that
1:31:42the pilot has to react to this. Uh so
1:31:45the event is plausible but sequence
1:31:47edited to standard communication flow.
1:31:49So you see here chajp has reacted in a
1:31:53certain way when you asked it to flag a
1:31:55problem. So it said that birdri bird
1:31:59strike sorry might be um an incident and
1:32:03in aviation an incident means you cannot
1:32:06only use phrasiology. Phrasiology again
1:32:08to make things very simple is it's coded
1:32:10language. It's not very plain language
1:32:13that like the language we are using now
1:32:15for example um an example of radio
1:32:19telephoneony or or phrasiology is say
1:32:21again. When a pilot says say again it
1:32:24means he or she is needing more or to to
1:32:28repeat the u how do we say the message
1:32:30from the ATC or another pilot. So they
1:32:33do not have to say would you please
1:32:35repeat what you have said because in
1:32:37aviation talk it's seconds you do not
1:32:39have to tell your life story you have to
1:32:41be very quick. Um so this is what we
1:32:44mean by phrasiology. It's it's formulaic
1:32:46language. It's pre-fabricated chunks of
1:32:48language. Um also we have um it might be
1:32:52a problem with comprehension questions
1:32:54rewritten to require listening not tape
1:32:57tape script reading. So you see here um
1:33:00it flagged that actually uh this task
1:33:04needs to be listening and speaking. So a
1:33:06transcript is not really necessary in
1:33:09this case. So here you are going to to
1:33:11think again and ask it again. Please
1:33:13remove anything that is written or or uh
1:33:16not necessary as reading and writing. We
1:33:18are focusing on listening and speaking
1:33:20only. So here the teacher of course
1:33:22directs, evaluates, modifies and finally
1:33:25of course you are going to validate the
1:33:27um the content you have generated at the
1:33:30end of course in in my opinion if you do
1:33:32this uh you will end up with a course
1:33:35book your own course book uh and you can
1:33:38even like sell it in the future why not
1:33:40it's a course book designed for aviation
1:33:42English or medical English or whatever
1:33:44of course um how do we say whatever uh
1:33:47field you are dealing with Now finally
1:33:51um one important point as well which is
1:33:53what AI or generative AI cannot um do or
1:33:58replace. Number one is contextual
1:34:00understanding of these specific learners
1:34:03my learners in this institution.
1:34:06Remember um generative AI does not know
1:34:09who you are does not know your learners
1:34:11their background and does not know your
1:34:13institution. So these three points are
1:34:15extremely important to at least share
1:34:18with uh the bot in the very beginning.
1:34:22So um these bots cannot really uh get
1:34:26into uh like personal professional data.
1:34:30So it's better to do this yourself. Um
1:34:33and with problems with privacy etc. I do
1:34:36not really recommend to share everything
1:34:38because this is very important for you
1:34:40and your personal or p private data as
1:34:43well. Professional judgment about what
1:34:46matters and why. Um
1:34:49[clears throat] as much as we believe as
1:34:51humans that we are objective,
1:34:54we might be mostly subjective in many
1:34:56things. So u chipt relies on large
1:35:00language models and these large langu
1:35:02language models are um written and
1:35:05spoken productions of other people. So
1:35:08can actually ch it really be subjective
1:35:11in many cases yes. So we have to be
1:35:14careful with what what kind of
1:35:16professional judgment
1:35:18chat bots provide. So here we have to be
1:35:21um very careful with this and we are
1:35:23going to decide as as humans. We do not
1:35:25let bots really decide about this
1:35:28pedagogical decision-m under real
1:35:30constraints. As we said before, um these
1:35:33bots do not know your your situation,
1:35:36your context, your institution, your
1:35:37classroom. So, it would be better to um
1:35:41how do we say to calibrate as we said
1:35:43before the uh the contents according to
1:35:46what you have or where you are. Uh
1:35:49validation for special specialized
1:35:51content. um bots can really hallucinate
1:35:54which is which is very risky sometimes.
1:35:56So it's always better to validate. Do
1:35:58not rely 100% on content generated by uh
1:36:02bots. And um uh this is a very
1:36:05interesting point. Human interaction as
1:36:07we said humanizing the classroom is u
1:36:11has has never been more important
1:36:13because before with the ra the rise of
1:36:16technology, the rise of computers,
1:36:19internet etc. many teachers were very
1:36:22afraid of the classroom being
1:36:24dehumanized. And uh after the or during
1:36:27the the covid pandemic this question was
1:36:30um was asked like I don't know many many
1:36:33many times uh would it be would the
1:36:37future be only with online classes would
1:36:39we be able to get back to the classroom
1:36:41and humanize the classroom more uh would
1:36:45we be just numbers like okay the
1:36:47classroom is online we have never met
1:36:51etc. So there there were many concerns
1:36:54about humanizing the classroom. Nowadays
1:36:56I think these questions are way more
1:36:58important. Why? Because with generative
1:37:01AI we have moved further from humanizing
1:37:04the classroom. So this is something I I
1:37:07really um appreciate as as a teacher as
1:37:10a professor and u as a pedagogical
1:37:13coordinator as well. I have I am in
1:37:15charge of teachers and students. So I
1:37:17try to spread the word and say please
1:37:20humanize the classroom even if you are
1:37:22using technologies. So I really
1:37:24recommend this as well unless of course
1:37:26it is [clears throat]
1:37:28not feasible like what we are doing now.
1:37:31Uh so that would be helpful but we do
1:37:34not over rely on on on these
1:37:35technologies and dehumanize the
1:37:37classroom. Um finally we have
1:37:40institutional legal and ethical
1:37:42responsibility. Uh this is very
1:37:44important. It depends on your
1:37:45institution if it allows you to use such
1:37:47technologies or not. And ethically
1:37:50speaking, it's not bad, okay, to use
1:37:52generative AI and declare that you have
1:37:55used generative AI for specific um
1:37:58tasks, it's not a bad idea, but ethical
1:38:01considerations must be taken into
1:38:02account just so you do not have legal
1:38:06issues in the future. Um and and that's
1:38:09it. Um again of course just as a summary
1:38:13um generative AI can help you automate
1:38:17uh can help you augment your your um
1:38:19your performance. So by automation here
1:38:22we are talking about uh providing data
1:38:25etc saving time creating alternative
1:38:28phrasing first draft feedback. So all of
1:38:30these are totally legitimate. You do not
1:38:33have to worry about this. It's totally
1:38:34ethical. So there there are no problems
1:38:36with that with augmenting your um what
1:38:40do we say your performance it can help
1:38:42you with scenario design material
1:38:45adaptation rubric aligned for your
1:38:47assessment. Um also this is totally
1:38:50ethical. You do not have to worry about
1:38:51this as well. Now with human oversight I
1:38:56think generative AI can be very helpful.
1:38:59We are humans. We may we may forget
1:39:00something. We may miss some important
1:39:03details. So, generative AI can really
1:39:05help you with this, especially if you
1:39:07are struggling with grading. What kind
1:39:09of grading I I should use to this uh how
1:39:12do we say this test? How should I grade
1:39:14writing? How should I grade a speaking
1:39:17test? It can help you with needs
1:39:19analysis, interpretation as well. With
1:39:22ethics or ethical issues, it can explain
1:39:24to you what you should do and what you
1:39:26should avoid. Uh so, all of these
1:39:28actually can be very helpful using
1:39:31generative AI but with a human touch. um
1:39:34as well.
1:39:36Um okay, I think I have taken more than
1:39:40uh I should. Um just a final word here
1:39:44concerning the five principles uh from
1:39:46experimentation to responsible
1:39:48integration. So five principle we should
1:39:50abide to, we should respect, we should
1:39:53um take into account. So we have to
1:39:56design first tool second. So AI serves
1:39:59as just as a learning design. It's not
1:40:01everything. Authenticity is the
1:40:04standard. So um we try to get to keep
1:40:07the teaching learning experience very
1:40:08standard. We have to validate everything
1:40:11at the end of of of using generative AI.
1:40:13Uh we have to be very transparent with
1:40:16learners that um we are using generative
1:40:19AI for specific purposes but not for
1:40:22everything.
1:40:23uh protect human agency and actually use
1:40:26generative AI for helping yourself
1:40:29develop professionally, not just to take
1:40:31your job or to do your the job for you.
1:40:34And um as a pro as a practitioner
1:40:37checklist, you can use this of course to
1:40:39check your work or you the use of
1:40:42generative AI. Are the needs identified?
1:40:44Are the general authentic? Are the
1:40:46levels set? Are the constraints stated?
1:40:49etc. Of course, all of these can be used
1:40:51just to check your uh production at the
1:40:54end. So the question we may leave with
1:40:57is if a AI can generate the content,
1:41:00what makes the educator experience or
1:41:02sorry expertise indispensable? Uh as you
1:41:05as as we have said AI can write drafts
1:41:08but educators they are the ones who
1:41:10designed the learning experience. uh the
1:41:12value so was never in the text itself
1:41:15that AI produced but it was always the
1:41:18judgment behind it. So uh these are some
1:41:21of the references. Of course I will
1:41:23share the template or the whole
1:41:25presentation with professor Valerie and
1:41:27uh he can of course share this with you.
1:41:30Again thank you very much. Sorry if I
1:41:32took much more time than expected. I
1:41:36appreciate your time and um if you have
1:41:38any questions, inquiries, I would be
1:41:40more than help more than happy to to
1:41:42answer any.
1:41:44>> Yeah, thank you. Thank you Dr. Asasi.
1:41:46Thank you for your excellent
1:41:49presentation. It was full of practical
1:41:53approaches. Now we uh we understand
1:41:56better this u this um part of research
1:42:00and your research. Uh but I have one
1:42:03question actually one question it's
1:42:06really important for me for example um
1:42:09you know uh in different countries we
1:42:13have straight A students failing
1:42:16students with uh uh bad um
1:42:21how do you call it basementment of in
1:42:23vocabulary in uh in practice of English?
1:42:27uh can they start ESP English without
1:42:31bass or uh maybe it's better to to get
1:42:35some uh like a base vocabulary
1:42:40base English practice. How do you think?
1:42:42>> Yeah, that's a really interesting
1:42:44question. Yeah. So, um usually ESP
1:42:47follows EGP. Uh what do you mean exactly
1:42:50is that any ESP learner is expected to
1:42:53have what do you call basics of
1:42:54language. Why? Because ESP we do not
1:42:57teach language for the sake of
1:42:59acquisition but for the sake of use.
1:43:02Anything that is taught in ESP should be
1:43:05practical. That's why in ESP we do not
1:43:08really teach we do not really focus on
1:43:11grammar, vocabulary, pronunciation but
1:43:13we focus on competencies. Today's lesson
1:43:16is how to write a prescription. Today's
1:43:18lesson is um how to solve a breakdown of
1:43:22communication between a pilot and and a
1:43:25controller. Uh however, these
1:43:28competencies require uh a very basic
1:43:31level. Now uh what do we exactly mean by
1:43:34a basic level? some um I I have many
1:43:38actually I have many colleagues or many
1:43:40um colleagues I mean professors from
1:43:43other departments who are studying
1:43:45English with us in our department even
1:43:47if this is not really a good idea
1:43:50because in our department we are
1:43:51training English language teachers so
1:43:53they do not really need these courses
1:43:55but um their expectations are always
1:43:58high. They say um okay I can speak basic
1:44:03English I can write basic English but I
1:44:05have those lapses I have those gaps I am
1:44:07not very fluent in ESP this is not
1:44:10required in ESP I do not want you to be
1:44:12very fluent I do not want you to be uh
1:44:15like um I have that kind of ortorial
1:44:18style that I can speak without any slips
1:44:20gaps or mistakes I have perfect
1:44:22pronunciation I have perfect grammar no
1:44:25in ESP this is not required at all so by
1:44:28basic um how do we say by basic level we
1:44:33mean that you know how to speak very
1:44:36simple language you you know basic um
1:44:40how do we say uh English words if we may
1:44:43say uh you have basic pronunciation so
1:44:47it's not really highly requiring like or
1:44:49requiring as far as level is concerned
1:44:52so ESP learners can can have very basic
1:44:55level because in ESP sometimes we have
1:44:57like a very sub field of ESP called
1:44:59English for vocational purposes. And um
1:45:03the difference between vocational and
1:45:05occupational is that occupational usual
1:45:08usually is designed for what we call
1:45:10white collar jobs like a professor, a
1:45:13doctor, a lawyer and vocational purposes
1:45:16is designed for blue collar jobs like a
1:45:18mason, like a carpenter. It means that
1:45:21person who has a less like developed job
1:45:24but it is an important job. But why did
1:45:27we differentiate between them? Because
1:45:29occupational English is a bit more
1:45:32developed than vocational English.
1:45:35Vocational English like a handyman, a
1:45:38plumber, for example, they need English,
1:45:40but it's a very basic English. Their
1:45:42communication requirements are not very
1:45:44high. So, uh it is an ESP course, but
1:45:48it's not very high demanding. It's very
1:45:50simple. So uh if you say that we have
1:45:53some pre-ervice professionals or
1:45:56students who have like a very basic
1:45:58level and can they actually attend an
1:46:00ESP class for me? Yes, it is possible.
1:46:04However, the ESP class will begin as an
1:46:06academic class then move to an
1:46:08occupational class. Why? Because in ESP
1:46:11we have two main branches. We have
1:46:12academic English and occupational
1:46:14English. For academic English is
1:46:16designed for researchers, professors and
1:46:18students. So what is the status of my my
1:46:22learners? They are students. Okay. So I
1:46:24begin with first academic English simple
1:46:27English like uh how to study, how to
1:46:29pick up a book, how to take notes, how
1:46:31to attend a class. Then a bit by bit if
1:46:35my students are pre-ervice engineers, I
1:46:37will move to ESP. If my u my learners I
1:46:41know them like they are studying
1:46:43engineering, of course they are going to
1:46:44be engineers. They are not going to be
1:46:46pilots or they are not going to be
1:46:47lawyers. So I know about the future. I
1:46:50know about the target situation. So yes,
1:46:52I think it is possible and ESP does not
1:46:55really uh it's not very high demanding.
1:46:58It's not a demanding. That's why it's
1:47:00it's fun to teach ESP because the
1:47:03learners will see that their level is is
1:47:05um is developing very quickly because
1:47:07they are first they are aware of their
1:47:09needs and they are um very interested
1:47:13and motivated to develop their their
1:47:16English because they need it in the
1:47:17future. So uh for for this case I think
1:47:20it's it's totally feasible and um
1:47:23actually they will if if the content and
1:47:25the teaching learning experience is
1:47:28related to their job or their future job
1:47:30they will see that their level is
1:47:32developing very quickly and you can see
1:47:34as an instructor as well.
1:47:38>> Thank you. Thank you so much. And if you
1:47:39have some social community uh where you
1:47:42can uh you you can share your courses.
1:47:46Yeah, we can we can share it with our
1:47:48participants for the future
1:47:50communication. Yes, if we have of
1:47:52course. Uh
1:47:53>> yeah, that's that's a very good idea as
1:47:54well. Yeah, because I I have um I have a
1:47:57couple of students who are working on a
1:47:59on a startup now. Uh they are working on
1:48:01a project for their master's
1:48:03dissertation. So they are working on a
1:48:05dissertation academically and they are
1:48:06working on their company. Um which is a
1:48:09platform for ESP teachers and ESP
1:48:11learners. It is is a bit you know
1:48:15complicated in the beginning because
1:48:17these two students want to do everything
1:48:19all at once which is uh they are very
1:48:21motivated etc. Technically speaking we
1:48:24have of course we have I have a
1:48:26co-supervisor who is an IT professor so
1:48:30they will be helpful with the technical
1:48:31stuff. Um but I think yeah creating
1:48:34communities like these especially on
1:48:36social media on uh on educational
1:48:39platforms such as mood for example they
1:48:41are very important to share uh to share
1:48:44materials to share experiences to ask
1:48:47questions. So creating a community I
1:48:49have actually mentioned this in the last
1:48:52two or three papers I have published
1:48:54among the recommendations it's always uh
1:48:56creating a community of practitioners
1:48:58and ESP students all over the world. I'm
1:49:01not only talking about my context but I
1:49:03think it's it's a great idea actually.
1:49:05Yes.
1:49:07>> Yes. So thank you. I I have no we have
1:49:10don't we don't have questions but maybe
1:49:14uh if if you don't mind we can
1:49:17communicate in the future. Yes. If we
1:49:20have questions or you you will have some
1:49:22uh some ideas uh to cooperate. So uh I
1:49:27really uh I really appreciate your
1:49:29efforts. I uh on behalf of uh all
1:49:33participants I think I I would like to
1:49:37express uh our gratitude for your time
1:49:40dedication. Uh so thank you so much and
1:49:44you can see in the chat many many people
1:49:47uh really delighted uh to to hear from
1:49:50you to know uh all these ideas. So thank
1:49:53you so much.
1:49:55>> I appreciate it. Thank you very much
1:49:56again for uh the invitation and um I'm
1:49:59at your service anytime. If if you have
1:50:01any um any questions, of course, I will
1:50:04share my my um my email, my um social
1:50:08media, you know, profiles. If you have
1:50:10any questions uh I'll be more than happy
1:50:12to to to provide you with materials to
1:50:14provide you with uh um maybe if there
1:50:17are some people who are interested in
1:50:19books in in articles I can share many
1:50:22things with you and uh as long as we as
1:50:26as long as we prepare um this platform
1:50:29that we are developing uh today uh I
1:50:32will be happy to to share it with you
1:50:34first so you can have a look and uh you
1:50:37will be the first to use uh these this
1:50:40platform and um hopefully we we can meet
1:50:43soon or in different of course occasions
1:50:46and collaborate more and again thank you
1:50:49very much professor and I appreciate
1:50:51everyone's time and interest uh thank
1:50:54you very much and see you uh on another
1:50:57better occasion hopefully
1:50:59>> yeah so so thank you and we wish you a
1:51:02great success in in your job in your
1:51:05research in your uh your efforts to to
1:51:09get some results in in your sphere. So,
1:51:12thank you so much. Uh and now we we can
1:51:15finish but this is not a finish. We we
1:51:18are open for future uh collaboration.
1:51:21So, thank you so much. Thank you.
1:51:23>> Thank you. And see you soon. Thank you
1:51:25very much.
1:51:25>> Yeah. Thank you. See you soon. Goodbye.

About this transcript

This page contains the full transcript of Tarek Assassi English for specific purposes “Generative artificial intelligence in domain specific l by Valerii Marenichenko, generated from the public captions YouTube serves with the video. The transcript has 17,846 words across 2,534 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.