Show HN: Phrasing – learn every language, to any level (phrasing.app)
Today I'm sharing a demo of my our language learning tool, Phrasing. It's a tool born from the language acquisition hypothesis, too many hours in an anki slog, and a strange desire to always be learning obscure languages.
The method is simple:
1. type in a show
2. learn the most important words
3. watch the show/acquire the words
4. review the words when needed in the future.
On top of that, we're trying to:
- do some novel things with spaced repetition (no more anki slog)
- expand the sort of content you can learn from (I want to to refresh my French by reading The Stormlight Archive)
- make an insanely beautiful tool for all languages (I want to learn Sanskrit and Hawaiian and such)
I think we're off to a great start so far, and I'm happy to be able to share what we have already! We've taken some of our core features, and ripped them out to put them on a playground for HN to explore. There's so much more to come though, this is just the beginning.
I'll be here all day to answer any questions. Thanks for checking it out and have a wonderful day <3
EDIT: This link was meant as a demo so hacker news has something to click around on (as per the rules of Show HN). The main marketing page can be found at https://phrasing.app/ - I think that's causing some confusion
125 comments
[ 4.5 ms ] story [ 186 ms ] thread"Phrasing works by finding you two documents, in two different languages, and connecting them sentence by sentence with eachother. While we're add it, we clean things up, score them, and analyze their relevance in each language."
EDIT: was able to confirm that my supabase instance can't connect to the auth service at the moment. 99% uptime but of course the 1% downtime is when you launch ¯\_(ツ)_/¯
Most of the app we try to encourage you to use professionally translated works to learn whichever language you're studying, so try to do as we say not as we do :D
> Una carrera comparativamente segura y próspera con algún prestigio básico automático es peligrosamente tentador para alguien joven, que no ha pensado mucho sobre lo que le gusta realmente.
If that's what's being presented for the world's 2nd most spoken native language and 4th most spoken overall, I shudder to think what the ostensible "Polish" equivalent phrasing is like:
> Porównawczo bezpieczna i dostatnia kariera z pewnym automatycznym prestiżem jest niebezpiecznie kusząca dla młodej osoby, która nie zastanawiała się zbytnio nad tym, co naprawdę lubi.
It is a bring-your-own materials type of app, so if you bring an authentically translated work, you avoid this problem. We try not to do any translation ourselves (although we of course allow it for convenience if you choose it)
> in realtà ad essi sono state dette tre bugie: le cose che sono state insegnate a scuola riguardo al lavoro non sono il vero lavoro;
> actually they've been told three lies: the stuff they've been taught to regard as work in school is not real work;
the first sentence is ok, but the latter is different: the Italian version says "the things they have been told about work is not the actual work".
This is an actually different meaning of the sentence.
And the first Hungarian example is weird too, e.g.
> Kivágnak az iskolából, vagy új utakon jársz?
has some greek translation where "τολμηρά" appears and the english translation gets "boldly carving a new path".
The literal translation would be "you walk new paths", I do not believe there is any boldness or carving involved. These are just different meanings.
I doubt you will get many conversions based on current landing page.
> Try clicking around on some of the words below - and make sure to give the audio a listen!
OP I clicked on the text and bunch of tags popped up. I wasn't sure what those tags are or what I am supposed to do after that.
What's confusing? It's clearly a language learning app, with various features that it seems like the landing page leads you through. The HN link is to the playground, which also does explain things.
Where are you getting stuck?
Perhaps these may help OP improve the page.
edit - I see the comment has been updated to be softened, which I think is a good choice.
If you have any lingering questions you're left with after seeing the home page, please let us know. That way we can add those to a list of questions we're trying to get sure get answered in any future updates!
Like Rust or Go.
:)
Cleaning up the grammatical information to make it a bit less information dense is definitely on our todo list though.
I had very good results many years ago polishing the last steps of my French while in France. My tutor used very novel and interesting short articles, one example was a short article about shrimp found in a cave with no light. At the time this was a novel bit of information and quite interesting on its own. I think an effort to find novel and educational information to learn from is an effort well worth the time.
Please add Albanian. I should learn this language but I am not motivated enough. I would definitely give your tool a proper test run.
When it comes to educational content, one of the things I'm most excited for is to have the stuff I'm learning hooked up into an SRS _and_ in another language. Not only will I remember things better because I'm learning it in another language, but I'll learn the other language better because I'm interested in the content.
Although I'm most interested in turning stories into language learning content. Instead of 'Jack was in the park in summer'', something like "John Snow stood on the wall in winter" I would expect to elad to much stronger memories.
Great idea. It could be simple cloze deletion from the content pasted in, or are you thinking to use AI to create simplified expressions with similar semantics?
For extra credit, analyze which subjects each person is learning the best from, and give them more of that. Eventually you should be able to find patterns and know when to switch things up by finding each person's learning cadences... romance, romance, chemistry, nature, romance, etc.
Out of curiosity I tried chatGPT with a simple: "please give me a list of ten interesting facts about the life of polar bears". It was ok. I got better results with: "can you rewrite that list, providing more unique, novel facts about polar bears? Things most people would not know?"
"Toujours produire" découvrira le métier de ta vie de la même façon que l'eau, aidée par la gravité, trouve le trou dans ton toit.
If you replace the first two words by "tu", and correct a minor mistake in the third one (add a "s" at the end), it has a meaning but otherwise not. I guess we cannot trust it to learn a language then.
Not saying it's correct, but it's definitely only going to be as good as it's inputs. If we gave it a bad translation, that's on us! Sorry
I built something similar, Manabi Reader, currently only for Japanese on iOS/macOS but expanding to more languages shortly.
I built it as a native app so that it behaves like a web browser - no sharing your data with providers, offline friendly, no worry about what copyrighted material users might consume. It tracks the words and kanji you read and learn, showing your incremental progress against JLPT levels.
I’ve built an AI app for BYOK OpenAI plus local LLMs that I’m integrating into Manabi Reader as a free feature to add these kind of LLM features
It has a flashcard app or lets you mine expressions directly into Anki
https://reader.manabi.io
I do not worry that language acquisition will not work for everyone if we can make it accessible enough though. Blame the tools (Phrasing included if you want), not the learner!
I'd be really interested in trying this out if you had Bengali as a language option. It's a notoriously underrepresented language by language learning publishers, despite being the 6th most spoken language in the world [1] with a very rich, 1000+ year history of poetry and a people that literally fought a war in some part for their language [2].
Again, congrats on the launch -- now comes the hard part!
[1] https://en.wikipedia.org/wiki/List_of_languages_by_number_of...
[2] https://en.wikipedia.org/wiki/Bengali_language_movement
Our target audience is definitely polyglots/advanced learners at the moment - mostly just due to time. We'll definitely get to the point where everything is explained - the only thing we're lacking atm is time :)
We've got some ideas to remove the limit on languages at some point this year. In the meantime I wonder if it's possible to use Hindi or Kannada TTS with Bengali. I'd be happy to work with you to try and get it working! It's very important for us to eventually support all languages.
> our Humane Spaced repeition system. Based off of the popular FSRS, our Spaced Repeition System will adapt to your review style. Take a few weeks off? No problem. Want to study ahead one day? No problem. Took a 2 year vacation? We're jealous, but we've got you covered.
I really resonate with this. I find that after a long break, I have forgotten a lot less than Anki’s algorithm believes.
Is there some reasonably scientific reference for this? I mean, I’d be interested to read someone’s study or theories here.
anki uses fsrs but afaik it still "punishes" you for taking breaks
And it also supports to postpone your backlog.
FSRS would indeed be more interesting, as it seems to be heavily “inspired” by SM-18.
Once you go too far away from the 'due date', the algorithms decent quickly to "0% chance of recall", where in actual memory it's more similar to a half-life.
It's been a couple years though since I've worked with that algorithm
That’s all, just wanted to give them a shoutout.
———
I’m also an avid language learner myself, and I feel like what has worked for me has been to study a wide range of words with Anki, but then also try to use them in conversation. Often times I wouldn’t remember the words I studied, but it would be “on the tip of my tongue”. Then the speaker would fill in the word, and from then on I would never forget the word.
Finally, I think you guys are also headed in the right direction with the idea of “expression”, because I feel it is often difficult to map words 1 to 1 with a language you already know. Oftentimes the expression will be said differently “my name is” vs “to me the name is”. The idea is to capture a unit of meaning in a set expression and to be able to combine the expressions to communicate.
I disagree with this. The hardest part about learning a language is finding content and resources in the target language that are both engaging and comprehensible at your current level. It's hard to find books, shows, or conversation partners, all the way through that critical awkward stage from beginner to vaguely proficient.
The parts of the language you need to learn will reveal themselves to you through your interaction with it. And then, either your brain will figure it out intuitively, or you'll be able to look things up if they remain a mystery.
Perhaps being equipped with a base set of phrases will increase your range in the very beginning, but that does involve offloading a lot of comprehension work to your brain's CPU, rather than letting it be handled by your brain's in-built language processor GPU. I'm suspicious about any such strategy for language learning.
Context clues that accompany the language input are a much more powerful way to make language comprehensible, compared to memorizing phrases and adding strain to your already busy brain.
The idea here is to load your RAM with enough information so you can go and interact with the language, and you have enough critical 'hooks' to start to boot up your GPU. Then once you've done that, you should have some good memories that you can start to commit to long term storage, should you choose
(did I go too far with the computer analogies? I feel like it broke down).
Either way, I think we say in several places, no app is going to replace the learning that happens from interacting with the content. _That's_ where you actually learn. Our sole purpose is to get you to be able to interact with it as fast as possible, with as little gymnastics as possible.
Loading it up into Phrasing tells me that the most important words to the episode were: conservitorship, inheritance, deep submerge vehicle, treason, philanthropies, ransom, nerve gas, and anaphylactic shock.
If I’m going to spend any time on flashcards or SRS, these are the cards I want to learn. Once I know these words, not only could I follow half the episode even if I didn’t speak Italian, but the other 99% of the language in the episode has something to “latch onto” while I’m passively watching it.
I fully expect 95% of my progress in Italian to come from watching the tv show. But I expect to be able to significantly improve the speed of that progress with Phrasing.
Barnkanalen (Barn = children, kanalen = "the channel")
Meanwhile, I have hardly ever had the chance to speak or write it. I seriously doubt that I've "output" more than a hundred sentences. Perhaps I've spoken some gibberish sentences now and then on my my way to work, but never for any practical purpose. Yet my comprehension is doing okay!
The first few months were intense, I remember. My prioritizing was detrimental to a couple of other aspects of life. Every day I wanted to get in some "active study", i.e. watching the current few episodes of FIA over again, trying to understand more this time of the latest one I was working on. And any free moment I had, walking somewhere or going shopping, I'd pop my earbuds in and listen in a more laid back way.
In the very beginning I also figured that my ear and brain needed practice at simply differentiating all these weird new vowels and sounds, so I played French noises almost constantly, including at night and when working. In hindsight, I think only the night time sounds were a smart decision.
Yet despite my dedication being a lot and the project affecting my life, I was always somehow ashamed of my lack of conviction and effort, when hearing about other learners doing "AJAT" or equivalents (All Japanese, All the Time). There are crazier stories our there!
The sad thing also is that I don't think it would be necessary to go so hard at ot to accomplish what I have. An hour or two a day of more optimal input (what I claim is so hard to find), might get you a really long way in the same amount of time. I do not know. My experiment, however, was testing whether I could acquire a language without speaking it, which it did confirm at least to me.
Children's books? Children also learn from scratch. I think an adult reading books targeted to progressively higher ages would do reasonably.
assuming we are talking about children's content in general (and not content for babies), you will find that it is not much more comprehensible than the adult stuff.
a language learning beginner is nowhere near the command or vocabulary of a child.
I think part of the problem is that adult learners learn things in a sequence. They learn one part of grammar, then the next, then the next, so that there are grammar rules and structures that you are almost completely unaware of until pretty far into your learning. Whereas a native child is exposed to the entire language all at once. They don't fully understand everything, but they are aware of all of the different structures. Children's books know this and are written with it in mind, as opposed to things like readers, which restrict their grammar to a particular level. I have a much easier time reading an A2 reader than any actual childrens book.
Nothing to do when I was a kid, so I would not rule it out.
Getting to a middle-school or YA level though, and increasing through reading levels there, is probably the best thing you can do for your language (and phrasing was built specifically to make this level accessible to you before your language level is high enough).
PS: Childrens books work on Phrasing too, so this isn't inherently biased in that respect, just my genuine opinion.
The trick is rather to make consistency easy. My own strategy there is aligned with most language enthusiasts: Make your input and conversations engaging! Make them so engaging you partake in those activities independently of the language, or better yet, in spite of the language. Make engagement automatic!
As an example, I've come to adore the French YT channel "ScienceEtonnante", and now watch anything published by it ASAP. I started watching it because I have a masters in physics, and found much of the jargon very easy to pick up as it's often similar to English. That left only the particularities of French itself for my brian to figure out. Now that I can simply lean back and enjoy the show, that is also what I do! It has become my favorite YT channel.
Is it working for everyone else? Is this a cloudflare config problem?
I would love to try this. I am currently on my 7th language so I am always excited to find language learning apps. So far none of them has beaten a good old language course and consuming media in the language.
Maybe the website has received a bit too much love but I am trying to search for a TV show and the UI is not reacting at all (I entered "How I Met Your Mother" in the textbox below "How to do what you love" on the playground page).
A bit of a strange question I have is who created the pictures for your updates page? The style reminds me of Raycast marketing material which I find incredibly beautiful. I really like the optics, is there a name for this style?
All the best!
In the actual application, you're correct, you just type in the show name and it will look it up for you. That's just not this page :)
With regards to the design, I think it's been dubbed 'Aurora UI'. We have a large figma files with dozens of similar sources, I think raycast is one of them haha
The /playground link I posted to HN was meant to give HN something to click around on. The root index page is meant to be the primary marketing materials.
With such powerful forces leading us astray, it's not surprising we find it so hard to discover what we like to work on. しかし息子が木から落ちたり、娘が妊娠させられたりしたときには、そうした結末に対処せざるを得ない。 規律
Actually they've been told three lies: the stuff they've been taught to regard as work in school is not real work; 子供たちが「自分はこの人たちとは違う。 自分はこの世界に向いていない」と思ったとしても彼らを責められまい。
Still, we don't want to stop our users from learning any language with it, even if it's not well supported. But I've been very transparent with any of our beta testers that Chinese/Japanese are not going to be a great experience (yet!)
We’ve made sure to be transparent with any of our beta testers about the limitations. We also agree that more docs would be better, and they will be coming. We been discussing internally a way to grade languages… but any conversation we’ve had in that area ends up in the same place: we need users to help us grade the languages.
We very specifically do not want to focus in on just a couple languages. There are hundreds of tools out there dedicated to individual languages, and we don’t feel we have anything unique to contribute there.
Sorry if it came across disingenuous! That was not our intention. As we work with beta testers, and learned of all these languages, we will develop more documentation around the precise limits :)
What I want as a learner is for someone else to give me good, accurate content; I definitely want to avoid any content in the target language which isn't native-sounding, because who knows how long it will take those bad habits to break.
One idea I had for content was something like this:
1. Start with two languages; say, English and Chinese (the first of which I'm a native speaker, the second of which I'm learning); and focus on English speakers who want to learn Chinese, and Chinese speakers who want to learn English.
2. Start with some basic content in both languages; from Tatoeba, from wikipedia, whatever. And hire a few native speakers of both languages to review, validate, and rate content.
3. Give new users some amount of credit for free; enough for them to use it for a few months.
4. Once they're out of credit, they have two choices: They can either pay a flat monthly fee, or they can contribute back to the content: Writing new content, reviewing & validating content, rating content, etc. And people could do whatever mix they want.
5. Expand to other languages as you grow.
I definitely see the attraction of trying to use automation to expand as far and wide as possible, but I think that ultimately it's more important to start by doing a few things well than by doing 50 things poorly.
Phrasing English: I am not suited to this world.
Phrasing Japanese: それなのに大人たちは、自分たちのやっている事が好きだと言い張る。
Actual meaning: Despite this, adults insist that they like what they do.
Phrasing English: It's hard to find work you love; it must be, if so few do.
Phrasing Japanese: 数十億人のうち、おそらく二、三十万人といったところだろう。 好きな仕事を見つけるのは難しい。
Actual meaning: Probably 2-300,000 out of billions of people. It's hard to find a job you like.
This is literally what created a drastic decrease in the overall quality of Duolingo. I wouldn't trust an app like this if it's relying on machine translations, AI or not. Especially when it comes to learning phrases and real-world examples. Translation is hard. Machine translation is even harder. Machine translation for languages with limited sample sizes (e.g. "Sanskrit an Hawaiian") is nearly impossible.
The only way to make this thing pass an even remotely reasonable bar for quality would be to hire humans to train it but at the rates required to keep this thing profitable that borders on exploitation.
We also have lots of ideas for 'Humans in the loop', verification, etc. But while we use AI for _a lot_ of things, we're trying to stay as far away from 'generate some Sanskrit' or 'translate <english text> to Sanksrit'
That being said, we've done exactly this, polished the landing page, and are literally working our way through all our friends and family again to refine it again. Another couple rounds and I think we might make it all clear! :D Indeed none of us are marketers so we're learning as we go :)
If you look at the waybackmachine though I think it's definitely trending in the right direction!
Huh? lol
Some people elsewhere in the comments are saying it's just a weird English sentence so it has a weird French sentence. My French is no good enough to critique on it's nativeness though
(Paul Graham is a very poor writer, and his image doesn't mean much to Europeans, who don't live much in individual, shoddily-made houses the way Americans do)
A time-limited example or something similar would be really useful. Maybe just one that you pick. Whatever it may be, I'd love to see how it actually works before paying money for it. Assuming it dubs videos, how does it handle multiple speakers?
Sometime this week or next we will work on recording videos of the whole experience, and making them available. Then another thing to do this quarter is automatic payments, payment portals, trials, etc.
We didn't think we should delay 'launching' any longer though.
I do like the UX tho, and find it a good idea that you created a HN-related landing page. Seems like a good amount of work went into this, congrats!