I'm working on https://yap.town - an SRS based language learning app.
I would say it combines the best parts of Duolingo and Anki. Anki is great for memorizing words, but you don't see the words in the context of novel sentences. Duolingo is great for exposure to new sentences, but it's oriented around "lessons" and SRS is an afterthought. (Duolingo is also not designed for people serious about learning a language IMO, it's too easy and goes too slowly.)
Had to do quite a bit to get it to work well.
1. At first you would think that if you know all the words in a sentence, that should be enough to understand the sentence. But it doesn't work like that. For starters, words can have multiple meanings. The french word "bois" can mean "(you) drink" or "wood". You want to learn these separately. I trained an NLP model (a gemma3 finetune) that I use to understand the manner each word is used in each sentence: https://huggingface.co/collections/anchpop/lexide-nlp-models
2. Even then, what about a sentence like "you'd better not"? Even if you know the words "you" "had" "better" and "not", you still won't really get this. So I use the wiktionary "multiword terms" category for each language to get a huge list of terms like "'d better" , "you better believe it", etc, and teach these in addition to individual words. And then I only show sentences where you know all the individual words as well as all the terms.
And I'm not planning to get rich off of it haha. Right now there's no monetization at all. If lots of people use it to learn a language and avoid wasting their time on duolingo, I'll be happy
I would say it combines the best parts of Duolingo and Anki. Anki is great for memorizing words, but you don't see the words in the context of novel sentences. Duolingo is great for exposure to new sentences, but it's oriented around "lessons" and SRS is an afterthought. (Duolingo is also not designed for people serious about learning a language IMO, it's too easy and goes too slowly.)
Had to do quite a bit to get it to work well.
1. At first you would think that if you know all the words in a sentence, that should be enough to understand the sentence. But it doesn't work like that. For starters, words can have multiple meanings. The french word "bois" can mean "(you) drink" or "wood". You want to learn these separately. I trained an NLP model (a gemma3 finetune) that I use to understand the manner each word is used in each sentence: https://huggingface.co/collections/anchpop/lexide-nlp-models
2. Even then, what about a sentence like "you'd better not"? Even if you know the words "you" "had" "better" and "not", you still won't really get this. So I use the wiktionary "multiword terms" category for each language to get a huge list of terms like "'d better" , "you better believe it", etc, and teach these in addition to individual words. And then I only show sentences where you know all the individual words as well as all the terms.