Estimated reading time: 3 minutes, 31 seconds
There are two things I don’t understand about DuoLingo: first, why each language has a completely different set of starter vocabulary. This is not just that there are some specific culturally relevant terms in some languages versus others, this is even that it takes them a long time to teach you simple things like thank you and please in some languages, which makes no sense to me. Most language learning I know assumes you are a tourist who needs to say basic words and puts you in expected situations where you need to use formal greetings and order food or such. Not so DuoLingo! In Arabic it teaches you early on about occupations (why is it necessary to learn the words for doctor and engineer early) whereas the Finnish one teaches you about a specific musical instrument and about saunas (when I went to Finland I understood the sauna obsession, but not so much that musical instrument which I never came across). Anyway, the Dutch one I’m taking now is teaching me all the animals, and I’m not talking cat and dog. I’m talking turtle and rhinoceros. Really? I need to know rhinoceras at A1 level Dutch? Why oh why?
The second thing I don’t understand about DuoLingo is something that I think relates to their use of AI. What I don’t understand is why they don’t have humans reviewing the output of the AI? And this is where the useful useless sentence things comes into play.
When I was first learning Arabic (Modern Standard Arabic, MSA, is not a language anyone speaks, so even native speakers of Arabic won’t intuitively understand or use its complex grammatical structures without direct instruction; almost impossible because you never use it in speech), my Arabic teacher taught us the concept of “gomla mofeeda” or literally, “useful sentence”. However, the literal translation does not apply here. In Arabic grammar, “useful sentence” means a grammatically correct sentence, and there are two types: the one that starts with a verb and the one that starts with a noun. She gave the example of the sentence “the sheep flies” and said that while this sentence is grammatically correct, it is not realistic. However, grammar-wise, we would consider it a “useful sentence”. I still rememer my child when learning Arabic and being asked to form a sentence containing the word tree and farmer, decided to say something like “The farmer is on the tree” and I was baffled… I was like, couldn’t you just say something more realistic, like “The farmer is under the tree”? But I think when you don’t think of language as particularly useful (for example, this exercise is not really that interesting for her) why bother to go beyond the very basic aspect of forming a grammatically correct sentence? My child’s behavior made me realize this exercise was not engaging her at all.
Anyway, back to DuoLingo, I just wanted to share a few funny sentences that keep coming up. I’ve been seeing things like “The apple speaks Dutch” and “Are you an apple?” and “The elephant drinks juice” and “The mouse reads” and “Are you not a sandwich?”. (well maybe mice read in fiction and apples speak in children’s books, but that’s not the average kind of thing you need to be able to say when you’re learning a new language… and I’ve never seen people talking to sandwiches before).
Am I wrong here?
I know, I know, DuoLingo is not the best language learning app out there, but even the other ones I hear people recommending don’t seem to work well for me for some reason. And I’m not like a serious language learning in anything other than French, and I don’t fully depend on DuoLingo for that one.
Funnily enough, when I search online sometimes for like 10 or 100 useful beginner phrases in any particular language, I also find really weird ones. I’m not sure why? I guess I could manually list all the words and phrases I want to learn in any language and put it through Google translate then check it against other websites, or also do it on ChatGPT and compare? Not sure.
Featured image from Pixabay, edited on Canva to emphasize the flying sheep
Hi Maha!
Excellent post, as ever. Thank you for everything that you’re doing. For background, I’m a librarian at a community college in the USA and do a lot of work around critical AI literacy, etc.
The first thing that came to mind regarding those funny example sentences you mention was that they sounded like bad translations. The connection between Google Translate, say, and Large Language Models like ChatGPT is fascinating, since they both leverage statistical predictions based on massive text examples and the former is something like a forerunner to the latter. (In this connection I highly recommend the article entitled “The Origins of Generative AI in Transcription and Machine Translation, and Why That Matters,” by Stone et. al. from the Critical AI journal, permalink below, Link1.) And, actually, Googe Translate started using LLMs to do its translations around 2016.
Anyway, this then presented me with a research question. The kind of cumbersomely large question that I have found it helpful to ask ChatGPT about as a first survey-the-scene step. So, I entered the following prompt: “How do apps like DuoLingo use AI at each stage of their development and use?” And sure enough, DuoLingo uses generative AI for more than just deployment and personalization…
Here’s the part of its answer that caught my eye:
“Generating Course Content: Duolingo now uses generative AI to rapidly create large volumes of course material, enabling them to: a) Launch 148 new courses at once, b) Translate existing courses into many base languages, c) Mass-produce example sentences, translations, and exercises. This replaced years of manual work and allows mass scaling of content creation.” (quoted from ChatGPT output, edited/reformatted for clarity by me)
To support these claims it cited a Verge article (link below, Link2), which in turn cited a DuoLingo announcement (link below, Link3).
As I would tell students: all of this would need further research, but this suggests a direction and provides a blurry snapshot. Quite interesting, though!
Link1: https://doi.org/10.1215/2834703X-11256853
Link2: https://www.theverge.com/news/658968/duolingo-language-courses-ai
Link3: https://investors.duolingo.com/news-releases/news-release-details/duolingo-launches-148-new-language-courses
When I was using it to improve my Polish it was obsessed with hats. Who is the woman with the hat? Whose hat is this? The small child has my hat! Who is the man next to the hat? Which woman has a hat?
The next time I tried it before becoming disinterested in the app again, it was obsessed with me learning variations of the phrase “she does not love you, she loves me!” Sometimes it was “she does not love me, she loves you” or it was a man who did or didn’t love me, but I would get a variation of this phrase in almost every lesson.
When on the lesson on units of time it threw in “period” as in “my period is really bad today,” (has the same double meaning as in English) I knew the app was really out to lunch.
Aside from that, it made sure I could say such useful phrases as “the elephant is big” and “the cheese is old” phrases that will obviously come in handy when I travel.