Key Features
Tech Stack
I’m trying to learn to speak Chinese and not read it yet. The issue is most of the language learning apps have a focus on characters. I feel like I just want to see the pinyin. Maybe I don’t know what I need, but I haven’t found the right tool.
You listen to audio you don't understand yet, and over time your brain begins to pick up the patterns. It takes a lot of time but you can do it in the background, because that processing happens subconsciously. So you can get that time "for free".
I learned it from this guy https://alljapanesealltheti.me/index.html
But he got it from linguist Stephen Krashen and his Input Hypothesis of language acquisition. (i.e. that the way babies and kids learn languages, thru osmosis, works for adults too.)
I think the ideal solution is somewhere in the middle, starting with something like Pimsleur which is the same idea (audio and repetition) but more structured and focused, to give you that "seed" of vocabulary and grammar, before you flesh it out with the "long tail" of the language.
For adults learning a language, I think you need 3 things to be most efficient. You need to learn the grammar rules/structure, you need vocabulary, and you need lots and lots of content. The specificity of Pimsleur I think is a major blocker. It lacks both vocabulary and content, and there is often a better resource for explaining grammar. I guess maybe the first unit of each Pimsleur course is pretty ok for getting used to the mouthfeel of a language, though.
For Spanish, I got far more out of languagetransfer.org, which helped me understand the concepts of the language much more, and dreaming.com, which gave me lots of content. For Chinese, I haven't found a course I like, but I still think I got more from drilling characters (I made my own app, but something like hanzihero or just an HSK/TOCFL Anki deck is probably good) and using graded readers. I think spoken-first in Chinese is a little bit of a trap, because it's easier to remember things with the written characters, when the relationships between words is a bit more clear.
The gist of those methods is mass input + create SRS cards for sentences where only one word or grammar pattern is unfamiliar to you.
A similar but more relaxed approach is ALG (automatic language growth), where you start from very basic input with lots of visual aids and let the language “wash over you”: no taking notes, no creating flashcards, no dictionary lookups. Sounds crazy, but it works for a lot of people. It’s the method behind Dreaming Spanish, which was inspired by the teaching method at the AUA language school in Bangkok, where Dr. J Marvin Brown used Stephen Krashen’s ideas to create a Natural Approach course to teach foreigners Thai from zero to fluency.
My first concerns though:
1. How can the system know which words I already know.
2. To what degree will I misunderstand the meaning of words.
3. Somewhat related to 2, how inaccurate will be description / explanation of words be.
1. How does it know which words I already know? It doesn’t automatically. You provide that set. For example, if you’ve completed HSK 1, you can paste the HSK 1 word list into LangSeed and mark those as "known". From there, new explanations are constrained to that vocabulary. You can also paste in real text and mark the easy words as known, though that’s a bit more manual.
2. How much might I misunderstand word meanings? Depends on how advanced the vocab is and how large your known-word set is. I think of this as building intuition rather than giving dictionary-precise definitions. As you see words in more contexts, that intuition sharpens. This is just my experience from testing it over the last couple of weeks.
3. How inaccurate are the explanations? I tested it on Swedish (my native language). There are occasional awkward or slightly odd phrasings, but it’s rarely outright wrong.
More: https://triviumpursuit.com/childrens-books-in-words-of-one-s...
On the other hand, the Chinese writing system is logographic (or ideographic), unlike the English system which is phonetic. The most basic characters, such as 日 (sun), 月 (moon), and 山 (mountain), are essentially graphics (or pictures) of the objects themselves. that makes them very suitable for being represented by images. The emoji you are using is also very good.
I believe this method should be very effective for beginners in Chinese. However, once you have mastered the basic Chinese characters, you can learn about the structure of Chinese characters and then continue reading more materials to expand your vocabulary.
The real challenge is to expand your vocabulary through extensive reading, i'm actually working on a tool to solve this specific problem (https://lingoku.ai/learn-chinese), If you are reading English, it will insert Chinese text for you, if your are reading Chinese text, it will translate the text from Chinese to English then inject Chinese words into the translated text, thus improving your vocabulary while reading.
At least for me, there's large value in consuming bigger volumes of Chinese to get me used to pattern-matching on the characters, as opposed to only reading a smaller amount of harder characters that I'm less likely to actually encounter
it sounds like you are at a more advanced stage of learning Chinese, you have moved past simple vocab building and are focusing on flow and fluency reading. For your use case, that 'inverse' approach (Chinese with English safety nets) is definitely superior for pattern-matching, it's a different problem set, but a very valid one.
Appreciate your feedback.
I haven't dug into the github repo but I'm curious if by "guided decoding" you're referring to logit bias (which I use), or actual token blocking? Interested to know how this works technically.
(shameless self plug) I've actually been solving a similar problem for Mandarin learning - but from the comprehensible input side rather than the dictionary side:
https://koucai.chat - basically AI Mandarin penpals that write at your level
My approach uses logit bias to generate n+1 comprehensible input (essentially artificially raising the probability of the tokens that correspond to the user's vocabulary). Notably I didn't add the concept of a "regeneration loop" (otherwise there would be no +1 in N+1) but think it's a good idea.
Really curious about the grammar issues you mentioned - I also experimented with the idea of an AI-enhanced dictionary (given that the free chinese-english dictionary I have is lacking good examples) but determined that the generated output didn't meet my quality standards. Have you found any models that handle measure words reliably?
Not affiliated with Hacker News or Y Combinator. We simply enrich the public API with analytics.