91–100 of 137 posts
That would be my biggest fear about using a language model for foreign language study ... being taught something wrong or significantly unidiomatic (perhaps by example), and then having to unlearn it later on.
One of the things I notice right away about your app is that your learning arc is moving from saying simple things to saying complex things, as opposed to aligning with increasing grammatical complexity. One of the things Duolingo does well is start with simple verbs, simple nouns, then slowly introduce language complexity over time: new particles, new tenses, new grammatical structures, etc. So the learning arc follows the complexity of the grammar, not the complexity of what I'm trying to express. I'm not sure I'm communicating that well :-) But for Japanese, as an example, there is a series of proficiency tests that do a good job of laying out a rough learning arc, which makes it easier to scaffold someone in, meet them where they are, and then start building on the knowledge they already have. When I futz with LLMs it's largely been to box the LLM in at a particular language level and then use it to practice language concepts at that level: for example, use JLPT N4 vocab and grammar and let's have a conversation about my day at university using formal language. If there was some way to roughly align the Univerbal conversation against tiers in language learning and then help me progress along that arc, it would be even more valuable.
My actual expertise is in games for learning though, and I appreciate what I saw you doing with "goals" in the "meeting a new friend" stage. Gamification is primarily about engagement, keeping me motivated, keeping me involved, keeping me coming back. Duo does a ton of gamification and uses it to good effect even though I know it does drive some people bonkers. But am I about to break my 500-day streak? No I am not. Gamification works! :) With the capabilities that you already demonstrate you could probably walk closer to making it an actual game, if you wanted to: the primary difference being that you'd need to add learning elements in there beyond the suggestions you have now (which are great, btw). Think of the way Zelda teaches you a new control mechanic through the individual shrines and bosses (formative and summative assessment, essentially), and adds the control details to your options menu once you've learned them (like having a grammar guide or vocabulary list once you've demonstrated mastery). What your game would be would be something very different from Zelda, but... it is fun to think about!
Love what you're doing!
1) The assessment felt a little long.
2) There were many transcription errors which really was not a good experience. Are you using a quality transcription like insanely-fast-whisper?
2a) Why do I need to see my speech transcribed in real-time? It's distracting and leads to worse quality transcription.
3) It would be great to have accent coaching. From a transcription you should be able to figure out tone, inflection, and phonemes, and find the grossest errors to point out.
---
Gameification sounds like something that should be either left for later, or very subtle. It's a great product as is. Maybe later you could have different trouble words fly at the user like soap bubbles and you have to say the word twice to pop them. That might make the drilling/practice more fun. A reward could be a joke in the language they are trying to learn?
Another user left a comment about being distracted by seeing their words written out as they talk. I need that feature left in because on occasion the microphone thinks I've said one thing, and really I have said another. It's rare, and completely due to my accent, but it does happen
I like the idea, but when I started practicing some Spanish, I noticed it gave me incorrect grammar advice (it claimed that the subjunctive form of “creer” is “creen” - it was right that the subjunctive was needed, but that would be “crean”). That makes me nervous to use it for a language that I know less well like Korean because I won’t be able to tell if the grammar feedback I’m getting is accurate. If there’s a wa…
We're using LLMs and good ol'ML. These systems are never going to be 100% accurate. Then again humans are also not 100% correct and we're working hard on ironing out the kinks like the one you just discovered.
I certainly don’t point it out to be discouraging - on the contrary, I feel like I am the exact target audience for a product like this and would happily pay for it if it can reach or be near the trustworthiness of a human teacher.
It's great! I really like it. I would buy the pro subscription if it was a bit cheaper. 19.99 CAD/month seems a bit too much (at least for me).
This looks very interesting and I wish you guys the best. One suggestion - a nitpick usually that I wouldn't write, but because it's a language product, you want to give the right impression of your language skills: In English, uni- , the prefix, means 'one'. For example, unidirectional doesn't mean 'all directions', it means 'one direction'. [0] The name of the app means something like 'one language' or 'one word'.…