Fine, AI Might Actually Revolutionize Language Learning

I'm excited!!

Just about a year ago I wrote urging everybody to chill out about chatbots in education. I stand by what I said. The emergence of cheap handheld calculators remains a useful parallel for thinking through the educational implications of chatbots. Expect some changes, but not fundamental ones, in most subjects.

OK—but I’ve been playing around with Google’s Bard over the past few days. And I want to give the robots credit, they are pretty much ready to improve language learning in a deep way.

First, a bit about me.

I know Hebrew, but not enough to read the news or novels fluently. I find this frankly embarassing given my background. Modern Hebrew was a core subject in the Jewish schools I attended. After graduating, I lived in Israel for a year and studied in an Israeli yeshiva. (I was in a program for Americans.) Since then I’ve basically allowed my Hebrew to atrophy. Over the past few months, this all has really started to bug me. Now I’m trying to get better.

At this point I can read children’s books in Hebrew easily, give or take a word or two on each page. Adult texts though are still a slog, especially fiction.

Frustrating enough, but my biggest limitation by far is my ability to speak and write in Hebrew. I was in Israel last year and my Hebrew was basically useless. In a way it would be easier to practice speaking if my comprehension was worse. It would certainly be less embarrassing. When I try to talk to Israeli friends I feel like an idiot, because to my own ears I sound like one. I can tell.

So, I’ve been waiting for this moment for a while. I want someone to talk to, without any of the judgment or shame. Every time I’ve tried out one of these chatbots, I’ve attempted conversation in Hebrew. They’ve been barely able to handle it. I figured I’d give Google’s chatbot a chance at this thing.

It was a very pleasant conversation. Its Hebrew is terrific. The hardest part was getting it to stick to asking questions. Bard kept trying to provide me with information. I was like, stop. I don’t want your information. I just want to talk. I’m not searching the web. Calm down.

I asked Bard, “what’s your name?” It tells me it’s an LLM. Then I asked it to ask me a question about my day. It obliged.

Not going to lie, the typing was pretty frustrating. I’m not awful with the Hebrew keyboard, but I’m not good enough that it’s seamless. I tried using vocal input, but it interpreted it as garbled English and that was useless to me. So, room for improvement.

My next idea was to work on translation skills. “I am learning Hebrew and would like to practice translating sentences into English,” went my prompt. “Would you please give me a sentence to translate?”

At first, it gave me the Hebrew along with the English. Eventually I got it to just give me the Hebrew, and then we were cooking.

I translated and asked Bard to evaluate my work. The chatbot was very supportive (too supportive). It pointed out a weird word choice I had made. (“Rolicking” was a guess on my part. “Thrilling” is definitely more correct.)

Whoops, I meant “stormy oceans.”

I tried it the other way too, asking for sentences to translate into Hebrew, and checking myself against their translation. Good! Fun!

I’m so jaded that it’s sometimes hard to tell when something is real. I think this is real and significant. It’s ready to help me learn.

• • •

We are in the middle of a hype cycle that is making it very difficult to think clearly about these machines, about what they do well and what they do poorly. Let’s try anyway.

What we have here is a technology that is good at understanding and generating typical human language—languages, actually. It’s very good at this.

But conversation has a complicated relationship with learning math. Just talking about fractions doesn’t necessarily help you learn anything. Conversation can drive learning—in structured situations and the presence of feedback—but it needs to be guided by someone who knows what they’re doing. Chatbots are great at chatting, but they’re not great teachers.

In fact, it’s tempting to say that chatbot tech will be useless for learning mathematics.

But what about when reading and understanding language is the target skill itself? When simply engaging with the chatbot on the chatbot’s own terms is good practice?

Let’s not let cynicism get in the way here—this is a new way to get better at learning a new language. This is going to be very useful to students and teachers of second languages. Less so for learners of other subjects. But could it still be a bit helpful?

I’m not an edtech guy. But if I were this is what I’d be thinking about. I’d be wondering about the moments in a mathematics (or whatever) classroom where simply having the discussion is itself an opportunity to learn. Or if there’s a way to add minor constraints or feedback that somehow make the conversation worthwhile.

There’s surely no replacement for chatting with another human. But thinking through my own experiences with language learning, maybe the chatbot can be preparation for human conversation.

In most subjects, that won’t be a huge change. But for language learning, I think it very well might be.