A Few More Cool AI Things

An end-of-year review you can use, learning to train and converse with my new robot assistant, and a creepy Black Mirror-esque moment.

‘Twas a week before Christmas, and all through my feed, teachers were tweeting about ChatGPT.

In my first post, I had just barely dipped my toes into the OpenAI playground and its possibilities for teachers. I wrote about using it in an impromptu fashion to create texts relevant to the people in the classroom and what they want to talk about. The idea was to plant a seed and start a conversation in the language teaching twitter sphere and it certainly did.

There are a lot of open questions about plagiarism, what it means to be a teacher, and whether or not this technology will ultimately be a help or a hinderance. There was debate about the quality of the content produced, some finding it lifeless and some finding it wonderfully coherent and rich in the kinds of “chunks” and structures they would like their students exposed to. All in all, there is still a lot to learn and a lot to talk about. Personally, I’m still on the side of “very excited.”

In the last couple of weeks, I’ve had some very interesting experiences, ranging from mildly interesting to very thought provoking.

End of Year Review (mildly interesting)

The least mind-blowing of uses, yet perhaps still interesting and illustrative to other teachers, was the answers to some end-of-year discussion questions that I like to use in December. The answers were good models; thought provoking and rich with language.

The activity went as follows:

1. I put the questions up, without answers, on a Jamboard.
2. Learners familiarized themselves with the questions and we discussed any unknown vocabulary
3. I read one of the short texts, but it couldn’t yet be seen. After listening, they had to decide which question I just read the answer to.
4. I then pasted the answer onto the board in the correct place and learners read the answer.
5. We focused on vocabulary and form as needed.
6. I repeated steps 3-5 until the questions were finished.
7. Since my sessions are only an hour, learners had the week to think about/write their own answers to these questions.
8. We discussed, compared, and explored their answers in a follow-up session, followed by setting goals for 2023.

If you want to use this, you can make a copy of the Jamboard. Page one contains the questions, page two the answers to be read and copy/pasted onto the board one at a time by the teacher one by one, and page three is an answer key.

Creating Meaning-focused Form Activities (pretty cool)

Most of my grammar teaching is reactive, but I don’t think proactive grammar teaching is always without value. I really like “structured input” activities for this purpose. The idea behind structured input is that learners don’t learn and apply explicit “rules” when learning language, but they acquire the forms through interaction with input.

In structured input activities, the input is manipulated with the learners’ input processing strategies in mind. Essentially, in order to complete the activity, which is always based on doing something with the forms in a meaningful context (choosing, ranking, etc.), they have to actually understand the form, therefore facilitating meaning-form connections.

There is a good overview here, and another here, so I won’t go into too much detail about them, as the purpose here is to show how they can be created with OpenAI playground, utilizing both its language capabilities and knowledge of the world.

One thing a learned in trying to create more complex tests and activities is that providing good examples of what you are looking for can ensure better output. So I first tried to create a structured input activity by prompting OpenAI to essentially copy the form of a Spanish structured input activity to be used in English. Notice that I’m also instructing the AI to pay attention to particular properties of the task.

The first attempt was unsuccessful, as only present simple sentences were included.

I didn’t give up there. One thing I’m learning is that you aren’t starting from scratch each time you ask the AI to do something. It’s conversational. Tell it what you want in more detail.

That’s better. After some formatting and changing the wording that I didn’t like in the activity introduction, this activity was ready to go.

Now you might be asking, with all that time-consuming back and forth, why not just create the activity yourself? And that is where it gets interesting. Once it knows what you want it to do, it will remember.

Here is my second attempt, focusing on a totally different form, and it took a fraction of the time. My prompt was “Create a structured input activity. The context is how New York has changed since 2000. Learners have to process present simple and present perfect sentences to complete the activity.”

Now, the above facts about New York are not earth shattering, but a thought did occur: provided we are careful and double check the quality of the information we are given, this can simultaneously provide us with information about the world to add meaningful contexts to meaning-focused form activities.

It still requires a lot of thought on behalf of teachers: What contexts are meaningful and interesting to my learners? What is a reasonable context for this particular form? How do I best prompt the AI to give me what I want? But it is sure faster than doing research on NYC and then trying to design an activity like this afterwards.

Teaching AI to Modify Texts (potentially very exciting)

One of the most obvious uses of OpenAI for language teaching is its ability to produce level-appropriate input. Dual-language readers (text with each page or paragraph also provided in the L1) can be created quickly. Graded readers are simple: just ask it to simplify a text. It’s good at mimicking all sorts of genres, and is good at taking pragmatics into consideration. But what about working with texts in more advanced ways?

While simplified texts certainly have their uses, particularly for extensive reading, what about when we want learners to be exposed to particular language in authentic texts, increasing the chances that they will learn it incidentally?

Enter elaborated input.

Michael Long argued that elaborating input is superior to simplifying it. Florencia Henshaw does a great job of unpacking an article by Long here.

To put it simply, Long argues that while authentic texts can be too difficult for learners, constantly simplifying them impoverishes them of the very input that learners need in order to learn. Here’s an example from Florencia’s video:

In the simplified version, the word “unemployment” is removed entirely, robbing learners of exposure to the word and a chance of learning it incidentally. In the elaborated version, which includes redundancies that might sound strange to a native or proficient speaker, learners get exposed to the word.

This is fairly easy for teachers to do, but can be very time consuming, particularly with longer texts. Wouldn’t it be great if a teaching assistant could do that part for you? That was my goal in my next exchange with my AI assistant.

This provided a really good example of the importance of prompting appropriately and providing the AI with tons of detail regarding what you want. I was frustrated and almost gave up for a while, but kept pushing until I got what I wanted. What impressed me the most was that by the end, the AI could “remember” what I “taught” it to do, and it will now elaborate texts for me instantly. Here is my exchange, which, I have formatted instead of using screenshots for improved readability:

Me: The following text:

“The only witness just caught a glimpse of the driver as he fled the scene, so she could only provide the police with a rough description.”

has been rewritten as:

“The only person who saw the accident, the only witness, was a woman. She only caught a glimpse of the driver, just saw him for a moment, because he fled the scene, driving away without stopping. As a result, she could only provide the police with a rough description of the victim, not an accurate one.”

This process is known as elaborated input, and it is a way of creating texts for language learners. Potentially unknown words and phrases, such as witness, caught a glimpse, and rough description, were not replaced but provided with more context so that language learners can still be exposed to the new vocabulary. In other words, the text was not simplified but elaborated. Can you do the same with the following text? Do not remove any of the existing words. If a word is at an intermediate or above level of English, expand on the word within the same sentence as I have in the example above.

“Discussions of optimal types of spoken and written input for language learning have traditionally focused on the relative merits of authentic and linguistically simplified spoken and written texts. I will argue that elaborated input and, in particular, modified elaborated input, constitute better options, with tasks, not just texts, functioning as important input sources. Modified elaborated input, potentially coupled with bimodal presentation, has many positive features, especially, but not only, for programs seeking to increase students’ opportunities for enhanced incidental learning.”

AI: “Discussions about the most suitable types of spoken and written input for language learning have usually cantered on the potential benefits of both authentic and linguistically simplified spoken and written texts. In this paper, I will put forward the argument that elaborated input, and specifically, modified elaborated input, are better options, with tasks, not just texts, serving as essential sources of input. Modified elaborated input, possibly combined with bimodal delivery, has many positive attributes, especially, but not only, for programs that aim to give students more chances for enhanced incidental learning.”

Me: You didn’t do as I asked. I would like the original vocabulary to remain in the text. For difficult words such as optimal, don’t replace the word but keep it in the text, adding more words and descriptions to help the reader understand the words.

AI: “Discussions of optimal types of spoken and written input for language learning have traditionally focused on the relative merits of authentic and linguistically simplified spoken and written texts. I will argue that elaborated input and, in particular, modified elaborated input, constitute better options, with tasks, not just texts, functioning as important input sources. Modified elaborated input, potentially coupled with bimodal presentation, has many positive features, especially, but not only, for programs seeking to increase students’ opportunities for enhanced incidental learning.”

Here, optimal means the best, the most advantageous, and the most suitable. Modified elaborated input, in this case, refers to a type of input that has been tailored and adjusted slightly to make it more accessible for language learners. Bimodal presentation is a method of presenting information that combines both visual and auditory elements. The goal of this type of input is to give students more opportunities to gain knowledge and language skills through incidental learning, which is the acquisition of knowledge without any conscious effort.

Me: Instead of providing the expanded context for the difficult words after the text, can you include them in the text within the sentences in which they occur?

AI: “Discussions of optimal, meaning the best, the most advantageous, and the most suitable, types of spoken and written input for language learning have traditionally focused on the relative merits of authentic and linguistically simplified spoken and written texts. I will argue that elaborated input, which is a way of creating texts for language learners by providing more context for potentially unknown words and phrases, and, in particular, modified elaborated input, which is a type of input that has been tailored and adjusted slightly to make it more accessible for language learners, constitute better options, with tasks, not just texts, functioning as important input sources. Modified elaborated input, potentially coupled with bimodal presentation, which is a method of presenting information that combines both visual and auditory elements, has many positive features, especially, but not only, for programs seeking to increase students’ opportunities for enhanced incidental learning, which is the acquisition of knowledge without any conscious effort.”

Me: Good. How can I ask you to complete the task above for future texts?

AI: For future texts, please ask me to rewrite them using the process of elaborated input. This means that instead of replacing any potentially unknown words and phrases, provide more context for them within the same sentence so that language learners can still be exposed to the new vocabulary.

Me: Will you remember how to rewrite using elaborated input?

AI: Yes, I will remember how to rewrite using elaborated input.

Me: Rewrite the following text using elaborated input.

 “If a friend invites you to their party, there’s an obligation for you to invite them to a future party you are hosting. If a colleague does you a favor, then you owe that colleague a favor. And in the context of a social obligation people are more likely to say yes to those who they owe. One of the best demonstrations of the Principle of Reciprocity comes from a series of studies conducted in restaurants. So the last time you visited a restaurant, there’s a good chance that the waiter or waitress will have given you a gift. Probably about the same time that they bring your bill. A liqueur, perhaps, or a fortune cookie, or perhaps a simple mint.”

AI: “If a friend invites you to their party, there’s an obligation, a sense of duty or responsibility, for you to invite them to a future party you are hosting. If a colleague does you a favor, then you owe that colleague a favor, meaning you should return the kind gesture. And in the context of a social obligation, people are more likely to say yes to those who they owe. One of the best demonstrations of the Principle of Reciprocity, the idea that one should offer something in return for a favor, comes from a series of studies conducted in restaurants. So the last time you visited a restaurant, there’s a good chance that the waiter or waitress will have given you a gift. Probably about the same time that they bring your bill. A liqueur, a sweet alcoholic drink, perhaps, or a fortune cookie, a cookie with a message inside, or perhaps a simple mint, a small, hard candy.”

Good AI. So now I’ve trained my digital assistant to produce structured input activities and create modified elaborated input texts, to some degree.

Though a lot of the discussion around AI’s ability to create content is interesting, I am now much more excited by the capabilities of AI to support teachers with this type of task. Teachers, if anything, need to know their stuff even more to utilize this technology effectively. Garbage in, garbage out.

This platform will spit you out a CELTA quality PPP lesson, a language analysis, and a grammar-based syllabus in an instant, so maybe this technology will support the transition to more thought out, SLA-backed ways of teaching and training teachers, and free up time for teachers to focus on their learners.

More than ever, it’s important that teachers understand how languages are learned and that they adhere to principles that are informed by theory, research and practice.

Good Idea, AI. (a creepy, thought-provoking moment)

For my last act, something much less about pedagogy and materials.

Based on what I’ve written so far, readers might be surprised to hear that I actually spend a majority of my time just talking with students.

For my context, teaching small groups and one to ones in an embedded corporate setting, seeing the same people week after week for months or years on end and being looped into their daily correspondence, meetings, and happenings of the day, a dogme-inspired, materials-light, just-in-time approach is my default (I’ll write more about that in a future post).

I created this ages ago and I have been waiting for an opportunity to use it.

Last week, I was discussing plans and goals for 2023 with a pair of B1 learners, virtually. One woman said that she and her husband would be turning 47 and 53, respectively, in the same week. Their ages would then total 100. The plan was to have a big summer birthday party together.

So I asked my AI sidekick to crank me out a short text about this birthday party and I included a few tidbits I know about Anja (she makes a mean lasagna, apparently). In bold are forms and chunks for talking about the future that I wanted to draw their attention to, as well as a few terms that I thought might be unknown, such as engraved and to turn 53.

Anja and her husband are getting ready for their joint birthday party. Anja will turn 53 next year, and her husband will turn 47. Together, they will be 100 years old. Anja is busy making preparations for the party while her husband is outside setting up the lights and decorations. Anja is going to make her signature dish – a delicious lasagna. She’ll also be baking a chocolate cake for dessert. Anja’s husband is going to put up the balloons and streamers and might even buy a piñata for the kids who are coming to the party. The couple is excited to see all their family and friends at the event.

Anja is also looking forward to giving her husband a special present – a watch engraved with the words, “100 Years Together”. The couple is counting down the days until their special day. They can’t believe that they are going to be 100 years old together. It’s a milestone that they will always remember. The birthday party is sure to be a night to remember. Anja and her husband are looking forward to celebrating their special day with all their loved ones.

At the end of the session, I displayed the story and we read it together. The ladies in the group were cracking up and really engaged. And also amazed at my ability to simultaneously write such a thing while being fully engaged in the lesson (I did tell them about/show them OpenAI afterwards, which also amazed them).

We discussed some of the future forms and then the dual meanings of “100 Years Together” — that, their ages added together equaled 100 years, but also it could mean that they have spent 100 years together.

Then, Anja realized that she actually really liked the idea…and that she was going to be buying her husband an engraved watch that says “100 years together.” No piñata, though.

OK, so it’s not as crazy or dystopian as a black mirror episode. But there’s something eery about the fact that, due to a fun little story produced by AI in a language class, someone is getting an engraved watch next year.

I will certainly be asking it about gift ideas this Christmas.