Taplio - Build a growth routine on LinkedIn
Mini tools
Razmik Badalyan
🙏 30 karma
2024-02-07
Language learning AI with memory
25
votes
0 answers
I have seen quite a few language learning apps that use AI, but still haven't encountered even one that has any of these features. If you have seen one, please let me know 🙌
0. The main feature I'm interested in is a long term memory of past conversations used to personalize teaching. e.g. Say today I learned the word "Comprehensible". The AI remembers this and, with the Spaced Repetition feature, reminds this word.
So, this is what I am looking for in the first place - a language learning tool with long term memory. The rest features are just examples of how I imagine this memory might be useful.
1. Generate learning readers based on users specific interests. e.g. I want to learn English to study physics. The app would generate readers on the topic of physics that are also tailored for learning English.
2. Summaries an original article to make fit for learning. Basically the same as the 1st point, but instead of generating a reader from the prompt app take a real article and rewrites it.
3. Generate readers based on the Comprehensible Input theory. To put it short when anywhere around 90% of words are known for the learner. The rest are new.
4. Consider words' rank when generating a reader. Words differ by how often they are used in a language. Say the word #100 is used much more often then the word #1000. So, when AI generates a reader it considers the words' ranks.
5. Consider learners vocabulary when generating a reader. e.g. If I know, give or take, first 4500 words, the 90% of the reader will be from that range. The rest will be mostly from the next, say, 500 words (words with a rank # 4501-5000). Also, considering the caveat that a learner's vocabulary has gaps (has vocabulary of 4500 words but doesn't know some words from the range 2000-2500) and outliers (has vocabulary of 4500 words but knows some words from the range of 10000-11000).
6. Use Spaced Repetition in the generated reader. So instead of using cards to foster retention of the new words the app uses the due words in the generated reader.
0. The main feature I'm interested in is a long term memory of past conversations used to personalize teaching. e.g. Say today I learned the word "Comprehensible". The AI remembers this and, with the Spaced Repetition feature, reminds this word.
So, this is what I am looking for in the first place - a language learning tool with long term memory. The rest features are just examples of how I imagine this memory might be useful.
1. Generate learning readers based on users specific interests. e.g. I want to learn English to study physics. The app would generate readers on the topic of physics that are also tailored for learning English.
2. Summaries an original article to make fit for learning. Basically the same as the 1st point, but instead of generating a reader from the prompt app take a real article and rewrites it.
3. Generate readers based on the Comprehensible Input theory. To put it short when anywhere around 90% of words are known for the learner. The rest are new.
4. Consider words' rank when generating a reader. Words differ by how often they are used in a language. Say the word #100 is used much more often then the word #1000. So, when AI generates a reader it considers the words' ranks.
5. Consider learners vocabulary when generating a reader. e.g. If I know, give or take, first 4500 words, the 90% of the reader will be from that range. The rest will be mostly from the next, say, 500 words (words with a rank # 4501-5000). Also, considering the caveat that a learner's vocabulary has gaps (has vocabulary of 4500 words but doesn't know some words from the range 2000-2500) and outliers (has vocabulary of 4500 words but knows some words from the range of 10000-11000).
6. Use Spaced Repetition in the generated reader. So instead of using cards to foster retention of the new words the app uses the due words in the generated reader.
Peter Liu
Apr 28, 2026
Hey! Your feature list really resonates with me — I've been building something along these lines.
Check out Read in Levels (readinlevels.com). Here's how it maps to what you described:
Points 2 & 3 (real articles + comprehensible input): We take real-world news and rewrite each story into 5 CEFR levels (A1–C1). So the same article exists at your exact comfort zone — roughly 90% familiar, 10% new, which is right in the Krashen i+1 sweet spot.
Point 1 (interest-based content): We publish 1 article daily across 8 categories — World, Science, Health, Tech, Business, Sport, Nature, Offbeat — so you can stick to topics you actually care about.
Point 4 (word frequency): Each level's vocabulary is deliberately chosen with frequency in mind. A1/A2 sticks to the most common ~1000–2000 words; higher levels progressively introduce lower-frequency vocabulary.
Reading + listening combined: Every article has audio narration, plus dictation and sentence recall exercises built in.
What we don't yet have — and I'll be honest about this — is the long-term memory and spaced repetition you're most excited about (your Point 0 and 6). That's the hardest part to build well, and it's on our roadmap. The idea of weaving due SRS words into a generated reading passage rather than flashcards is genuinely novel, and I think it's the right direction.
Would love your feedback if you give it a try!
Yngvi Karlson
Mar 15, 2024
Its not really designed specifically for language learning.. but we build Kin with advanced memory (mykin.ai). Its private and personal.. so you can try using it for language learning
Razmik Badalyan
🙏 30 karma
Apr 19, 2024
Hi, I tried Kin but it didn't work for me. I had reported the bug on the your Discord, but it was not resolved.
