Hacker News new | past | comments | ask | show | jobs | submit login
Show HN: Learn a language quickly by practising speaking with AI (prettypolly.app)
325 points by cwbuilds on Aug 2, 2023 | hide | past | favorite | 255 comments
Hi guys,

Hope everyone is well. This app was borne out of my own frustration. I thought that I was terrible at learning languages at school, since I didn't become conversational in French after 5 years of study. However, I later traveled with some French friends and, in just under 3 weeks, I was able to hold a reasonable conversation. I realized that there's no substitute for speaking to native speakers.

I tried to adopt this approach for other languages, but it's much harder to find people to practise with when you aren't travelling. I started using iTalki to meet people from different countries and chat to them. It quickly became very expensive and time-consuming to schedule the calls, so I gave up.

I made PrettyPolly so that anyone can easily practice speaking 26 languages orally. The app uses ChatGPT (amongst other tools) to allow you to practice speaking whenever you want. It also generates a fluency score for each conversation so that you have an objective way of tracking progress.

It's free to use (up to 15 conversations per month). I've found that using it once or twice per day is plenty, and you'll be amazed at how much you will pick up in a week. I've added some FAQs here in case useful - https://www.prettypolly.app/learn

Would really appreciate any feedback. Let me know if you have any questions, issues or suggestions.

Thanks, Chris




It’s amazing, it worked brilliantly in Japanese although it got too difficult for me far too quickly and I was relying on the English translation it was showing.

Couple of things. I was promised a fluency score for my conversation but when I finally decided to press the button, I got a login/sign up page and I felt that that “ransom” feeling that a post the other day spoke about. If I’d been given a score, it wouldn’t have stopped me signing up and getting my own account to save the conversation I’d just had. I did sign up but only in the spirit of the fact you want feedback. If it hadn’t come from here I would have abandoned at this point and not come back. Sad but true.

So when I finally did get in, I still didn’t get my fluency score and that conversation was gone forever. I’m left now really wanting to know how I did, but I’ll never know.

I did a quick second conversation and then went straight for the score after a couple of rounds. My score was 0 :(

Btw I’d love it to be able to interrupt the AI because it was speaking away to me but I was ready to speak back but I had to wait or speak over it.

I really like it though and hope it gets smoother and better over time, because I know that I would be a paying customer for something like this that genuinely helped me improve my language skills.


Thanks for the feedback! Weird that it said you had a 0 fluency score. Had you spoken in that conversation?

Being able to interrupt the AI makes a lot of sense. Will add that to the roadmap.

Thanks for the kind words. Hope I can make it more helpful.


Yes I did speak twice, I didn’t really answer the question though, I just volunteered information to get something done quickly. So maybe 0 was accurate :)


This is really slick!

I really appreciate you having Japanese as a language choice. I've been dying to hone my japanese skills with AI in lieu of reading grammar/vocab books etc.

I definitely intend to use your product more, because it is definitely scratching that itch!

A couple of things that might help me personally would be the ability to cancel my current response and redo it, as frequently I will stutter or pause on a thought and wish I could restart over.

In that same vein I would love to see the output of what I'm saying in real-time. If only to gauge and understand what the application "thinks" I'm saying and to potentially correct and miscommunications from spoken input.

I would also love the chance to role play or have the tool help me learn contextual role playing. For example in Japanese keigo is a skill I just haven't mastered. I would love a tool to help put me in positions where I need to use keigo, and then be corrected where I'm failing.

Regardless, this is absolutely fantastic! Thank you so much for sharing! You're awesome :)


> For example in Japanese keigo is a skill I just haven't mastered.

Other than very common scenarios like introductions, spare yourself the headache.

If you are a diplomat or something similar, understanding it might be useful, but using it can be a minefield (e.g., you better know the hierarchy of everyone around you so that you use the appropriate language).

> I would love a tool to help put me in positions where I need to use keigo

In most if not all cases, if you find yourself in a situation that keigo is both standard and expected, one of two things will likely be true:

1. You will have been selected based on your ability to use keigo (e.g., customer facing roles that use keigo like banks).

2. There will be one or more people in your organization who will make sure you know what you need to know.

Most non-native speakers I have known who try to learn keigo use it in really unnatural and sometimes awkward ways.

I have known three people who “got it right”, and maybe a fourth. All are super fluent in “normal” Japanese and are lifelong students of the language.

1. Currently a professor of Japanese. PhD in Japanese from U Hawaii.

2. Polymath Cambridge grad who worked for Cambridge University Press. It was not uncommon for Japanese people to whisper to me about how freakishly good his Japanese was, especially when speaking about formal or technical topics.

3. Power nerd who went deep into the Japanese sword drawing martial art (iaidou). Also a professor.

4. Just a guess, but HN’s patio11 is probably good at keigo. I would listen to anything he has to say about the topic over anything I say.


I'm mystified by this comment. Keigo isn't some high-level PhD topic, it's used constantly in everyday usage. Even an intermediate speaker needs to understand some, or they'll be unable to follow many basic conversations. And sure, nobody is likely to yell at a foreigner for not using it perfectly, but given how common it is it's a very reasonable thing for even an intermediate speaker to want to practice.


See my reply here for more info:

https://news.ycombinator.com/item?id=36982980

> Keigo isn't some high-level PhD topic

It gets multiple book level treatment in Japanese (if you’re in Japan, please ask about the keigo section at a decent bookstore), and I am familiar with at least a few post-graduate papers on the topic.

Some additional comments:

1. Something like this link (http://keigo.livedoor.biz/archives/cat_14420.html) might be a good intermediate level of keigo. A lot of the less common keigo on those pages get butchered by non-native speakers when they try to use it (mainly because they shouldn’t be using it).

> given how common it is it's a very reasonable thing for even an intermediate speaker to want to practice

2. Unless someone is a diplomat, a 会社員 or in some sort of traditional arts, I would humbly suggest that they could go their entire lives without producing any keigo outside of memorized words and phrases (e.g., おはようございます) and could be seen as a highly fluent, high-functioning speaker by Japanese people. The basic link I provided in my first reply and the intermediate link in this reply are plenty to have receptive knowledge of, but one could argue that the important keigo in these lists simply falls under slightly more difficult memorized words and phrases (something acquired relatively early at lower proficiency) rather than advanced generative rules (something acquired much later at higher levels of proficiency).

3. Referring back to your “not a high level PhD topic” comment , I sure as hell would love to read some detailed breakdowns of top tier keigo use and keigo faux pas (especially in high level negotiations). I’ve never seen anything first-hand, but have I heard some stories. I’m pretty sure the best stories never leave the room they were uttered in.

To close, I’m not sure what your experience is with Japan and Japanese language, but I get the sense they you and I see keigo very, very differently.

Fwiw, I have been part of a training team that trained folks who were in “protocol positions” in Japan, and I can tell you that keigo was a constant sphincter-clinching aspect of that training (usually bypassed with an interpreter in high stakes Japanese-language situations, using English-English most of the other time in less formal situations, and using memorized words and phrases the rest of the time).


> It gets multiple book level treatment in Japanese (if you’re in Japan, please ask about the keigo section at a decent bookstore), and I am familiar with at least a few post-graduate papers on the topic.

There are entire PhD theses on aspects of English (or Spanish, German, Italian, French, Russian, ...) grammar too. That doesn't mean that people can not learn to use the aspect in question intuitively. It just means that it's difficult to provide an explanation for this behaviour.


> I get the sense they you and I see keigo very, very differently.

I'm using the term the same way wikipedia does, and all the links you posted - it's a broad category of usages including 尊敬語, 謙譲語, etc. It's not some rare thing - if you walk into a 7-11 or restaurant, literally every sentence spoken to you will include some kind of keigo. It sounds like you're using the term solely to mean the highest heights of formalism, but I don't know where you're getting that usage.


＞It sounds like you're using the term solely to mean the highest heights of formalism, but I don't know where you're getting that usage.

The keigo you refer to is the easiest keigo to learn, and much of it is learned as a beginning learner as memorized words and phrases.

In the context of the person I originally replied to, this type of content is either trivially easy to study/learn or not worth learning (e.g., most learners don’t really need to know that おはようございます is a highly stylized form of はやい — the memorized phrase and others like it are more than adequate until one is an advanced learner).

The keigo that would warrant using an AI speaking partner would, imho, lean towards the much more complicated aspects of keigo.

Note that I have actually referred this tool to folks who do training in keigo (and other language things). I’m not sure how the tool will do or what type of prompts need to be used to get the right type of engagement, but I think that there is potential on this front for advanced learners. For lower level learners, this tool is massive overkill.

Edit:

Note that most English-language web pages about keigo cover the simplest (and least interesting, imho) parts of keigo. Part of this is due to SEO, but part of it is probably due to ignorance about the topic.

For anyone interested, and if their Japanese is good enough, just search Amazon Japan for books on keigo. The topic is quite deep and can be incredibly fascinating. The content covered on English web pages just scratches the surface of the scope of keigo.


As a linguistic category it includes teineigo and set phrases, but when a JP learner says keigo they probably mean the stuff introduced the "keigo" chapters of JP textbooks, which starts with おっしゃる, 下さる, いただく, お～になる and so on. And again, those are everyday usages that are perfectly reasonable for even an intermediate learner to want to practice (and the person you replied to may be beyond intermediate). Sure there's also lots of other keigo that most learners would rarely encounter, but that's no reason to jump all over someone who says they want to practice it - you wouldn't tell someone not to bother with kanji just because there are characters they'll never need.


> … which starts with おっしゃる, 下さる, いただく, お～になる and so on. And again, those are everyday usages that are perfectly reasonable for even an intermediate learner to want to practice

Serious question…

Unless this “intermediate learner” works in a Japanese-language position (like a 会社員), is a diplomat, or is involved in a traditional art (or these types of relatively rare categories), in what context would they ever produce utterances using any of these words appropriately and naturally?

This is just not the type of language that intermediate learners need to produce very often. When they do, their organization will most likely train them beforehand, use an interpreter, or just not care.

I lived in Japan 8 years, and I worked in a Japanese-language environment. I’m pretty sure the times I had the opportunity to use any of the words you listed appropriately could probably be counted on my fingers and maybe extending to my toes.

Granted, I was not in a position of power (those folks used keigo all the time), and I was not directly involved in any highish stakes negotiations or business transactions (e.g., didn’t buy a house) but still…


> in what context would they ever produce utterances using any of these words appropriately and naturally?

Part-time job? Friends or in-laws they'd like to impress? Studying for JLPT? Just wanting to be able to speak everyday Japanese as it's spoken?

Respectfully, I really think you have a distorted sense of this due to whatever your background is. Per a quick google, some of the terms I listed earlier show up in JLPT from level 3. I'm no expert but I don't think N3 is considered high-level diplomacy territory. If you got by here without using such terms that's fine, but it's no argument against studying them - lots of people get by without learning kanji, etc.


Somebody wants to learn a language properly and you're saying "don't bother". I don't get it.


It's a matter of priority - you can get by fine without using keigo, and instead of learning all the weird intricacies of keigo you could go learn something more practical for day-to-day use first (even native Japanese speakers complain about keigo, e.g. use of manual keigo). Teineigo is perfectly sufficient in everyday use.

To put it another way, it's a bit like trying to to learn about how to write in the style of a judicial opinion when you're still learning how to talk to someone at the bar.


Well, OP didn't specify what their level was. Maybe they're already fluent? In which case learning more advanced things does make sense. Although I would never trust ChatGPT to use advanced Keigo properly (especially with the lack of context dictated by social settings), let alone to correct it and explain its usage


> Somebody wants to learn a language properly and you're saying "don't bother". I don't get it.

Yeah. I can see that. Let me see if I can expand.

1. Keigo falls into the category of “specific purpose” (SP) language. SP language can be relatively simple things like tourism language, or it can be very deep and technical like scientific, medical, engineering, etc. SP language is typically taught and learned on a need to know basis, and it’s totally possible to be highly fluent (usually measured in terms of general proficiency) and not know anything about SP language.

2. Keigo (and this is way more than basic formal and honorific language) is used in very specific situations. As a foreigner, in almost all cases that matter, there will be an interpreter or the foreigner will be trained in keigo.

3. For reference, I think that most of the keigo that I would recommend a non-native speaker learn can be found at this link (it’s a small and easy to learn list):

https://www.fluentu.com/blog/japanese/japanese-keigo/

Furthermore, I think that most of this language is more useful receptively (listening and reading) and will almost never need to be produced (spoken or written).

4. As far as proficiency goes, most keigo would be ILR 5 / CEFR C2, with the basics linked above being ILR 4 / CEFR C1 (used dynamically… much lower as memorized set phrases). When you’re at that level and needing keigo, you know you are lacking in keigo and make an effort to learn it. Before that would probably be a learning sequencing error (other than nerdy curiosity).

Beyond learning the content in the link above, most foreigners will not find themselves in the context of needing to know or use keigo. As special purpose language, it should really only be pursued if necessary. Otherwise, the knowledge learned will most likely be forgotten due to lack of use, or it will be used awkwardly and possibly inappropriately (as I mentioned before).

If it’s just a nerd itch to scratch, then go for it. For most people, I think there are more compelling aspects of Japanese to spend time on.


> Beyond learning the content in the link above

If you put it like that - OK. I was under the impression that you were implying "learn no Keigo at all beyond introductions (except Teineigo)", which would seem rather odd to me, as I've definitely heard 行ってまいります, Xでございます, etc. in Anime and movies.

If you're talking about the advanced uses that even many native speakers struggle with (unless properly trained), then yeah, maybe it's specialised vocab that you won't necessarily need.


> If you're talking about the advanced uses that even many native speakers struggle with (unless properly trained)

To me, this is the part that is keigo. The other stuff is relatively simple and early in terms of language acquisition.

Assuming that others share my definition of keigo may be assuming to much on my part.


IIRC, for Japanese people, even Teineigo is part of Keigo.

In learners' resources (where Teineigo is introduced usually right at the start), my impression is that Teineigo is not included, and that Keigo starts with Sonkeigo and Kenjougo. That's still introduced comparatively early, but much later than Teineigo (I think, it's in Genki 2, for example).


Thanks - glad it's useful for your Japanese. Do you like it being in the Japanese writing system or would you prefer phoenetical English? Wasn't sure which to do on that one.

Thanks for the feedback. The processing point has come up a couple of times so I'll definitely be adding that, and I'll look into doing some contextual scenarios. If your language is good enough, you could try prompting it to act as, eg a waiter in a restauraunt, but that's far from ideal for beginners!

Sure - realtime transcription makes a lot of sense. I'll look into it.


Absolutely, please keep Japanese output the way it is - it's the only approach that makes sense. Romanization loses a lot of information and makes things harder (except possibly for the absolute beginners), not easier.

However, it'd be extra cool if there'd be an option to display furigana - in case user had forgotten or doesn't know some particular kanji. Or maybe just throw an option to replay the audio.


Got it. Will learn about those and see what I can do.


>Do you like it being in the Japanese writing system or would you prefer phoenetical English? Wasn't sure which to do on that one.

Please do NOT use romaji. I honestly hate it when language-learning apps try to force romaji on me; it's not useful for learning the language. The written language in Japan is Japanese (hiragana/katakana/kanji), not romaji, so I just find it a distraction. Part of learning the language is learning to read it, not just to talk in it, and being forced to read it in its native script is far more useful than latin transliteration. That said, having furigana for unfamiliar kanji is really helpful too.


Got it. Thanks for the feedback.


> Do you like it being in the Japanese writing system or would you prefer phoenetical English?

Definitely Japanese writing system. Thanks.

Amazing tool.

Fwiw, if you’re looking to monetize, you may want to try to get a contract with the DoD. I would start with special operations (lots of money and urgent mission/training needs), and then move on to the Defense Language Institute.

Oral proficiency interviews (OPIs) are part of the qualifying standard, and everyone wants more practice.


Thanks for the kind words. That sounds like an interesting idea. Any chance you have a contact? Am UK-based so imagine it would be hard to get a meeting with the DoD from here. If so, I'd obviously be incredibly grateful, and here's my email: chris@prettypolly.app

Thanks again, Chris


Oh! I didn't know that. I'm really new to how to interact with AI. I'll try that out.

Please please please keep it in Japanese. Phonetic english for Japanese is awful. Hiragana only for Japanese is really not ideal.

The current output is absolutely perfect.

Also, I think it's really cool I don't need an account to try that out. Very cash money of you.


Ahah hear you loud and clear - no phoenetical words. Of course - hope it continues to help you learn!


Both. But perhaps also hiragana/katakana option too, like children’s book do when there is a kanji. Full on kanji is beyond me and I don’t expect to be able to pick it up just by seeing it in these conversations


nthing the consensus here - it's very good as-is, even though I maxed out my conversational ability fairly quickly.

The only changes I would make are to the UI, clicking the start and stop listening button feels awkward. Perhaps make it so you can click, talk, and release the button. And perhaps the option to select a few different voices.

Also intrigued to see where you go with Anki integration. Excellent work, you should feel very proud of what you've built so far.


Yes - Anki integration is one of the things I'm keen to do next. I'm convinced that any good language learning app needs spaced repitition.

Got it. Thinking of making it possible to control the mic by holding the space bar, although that wouldn't work for mobile obviously. What devices were you testing it on if you don't mind me asking?


I've just tried it on my desktop. I probably wouldn't use this mobile.


Cool, thanks for getting back


definitely Japanese writing system. Thank you!


Got it, thanks!


It needs a "Processing..." notification, I clicked "Start conversation" and it takes a while before it says anything. Same with "Start speaking" -> I said something, then clicked "Stop speaking" but the page seemed unresponsive. I repeated it, and when the transcript appeared I ended up answering the question, getting another question, and repeating the first answer.

Also after 3-4 sentences the questions got a bit "dull-person-at-a-party" small-talk... "What are you doing today?" / "I'm programming." / "What programming language are you learning?" / [I didn't say I was learning...] "Java." / "Oh, interesting. Java is a popular programming language. What do you like about Java?".


Thanks for the feedback. I'll add in a processing-style spinner. Sorry that it was a bit slow when you first used it. Do you mind if I ask what device you're using?


This my experience on an iphone 12. The UX doesn't really work for me yet but I absolutely love the concept, will keep an eye on it (would sign up for a mailist if you have one).


Thanks for giving me the device. I did have some trouble getting it working on iOS so maybe I didn't iron out all the bugs. I'll get working on a fix. I think we're already emailing so I can add that email to the mailing list if you'd like?


Thanks Chris. FYI I will pay for the app if you add Bosnian, Croatian or Serbian. It works pretty well.

Would be great to have hints for how to respond. Getting to the next level is important and we will get stuck once we hit our knowledge ceiling.

An idea for the hints is simply maybe showing a few of the words that might come next when the user clicks "hint".

I also wondered what the max fluency score is. Is it 100? I got 9 with saying the most basic of things in Italian.

I would make the text box way bigger too. I can barely see what has been said. Also my text and the response text show up at the same time. I would space this out more naturally.

Also it entered as my text "You: Untertitel der Amara.org-Community" which I definitely didn't say.


Thanks for the comment. Cool - those languages are on the long list so will hopefully be able to add them soon.

There's no max to the fluency score, although, in reality, it would probably be very hard to get anything over about 200. The fluency score is obviously fairly arbitrary and is calculated using the number of unique words and the rate at which you speak. I've found it to be a useful objective measure to compare yourself to previous conversations.


Yeah I really want to see my past text to see if it transcribed it correctly but can't with the current box size (web)


Okay, yeah that does sound annoying. I'll enlarge it


I've been meaning to brush up on Spanish and the app worked great to just hop in! The best part about using GPT is being able to talk about whatever is on your mind.

I noticed the transcription APIs clean up my mistakes (which is good for the transcript) but I wonder if there is a way to show the raw and corrected transcript to help guide corrections? Either way, it's already useful as is, looking forward to using it more.


Thanks for the comment and glad it's useful! Agreed - I love the flexibility to talk to it like it's almost just another human, but with vast knowledge. Okay - that makes sense - I can look into adding some more of the raw transcript. Thanks for the feedback.


Happy to help! To give some context: there was a GPT prompt which was something like "Act as a language tutor. I'll have a conversation with you, and you reply back. If I make a grammar mistake, keep the conversation going, but put the correction in a parenthetical statement."

I was using that for a bit, but typing in another language was enough friction (vs. speaking) that I didn't use it that much.

For this, audio is perfect -- all I need is the gentle reminders about what to fix, and why. GPT could probably have a parenthetical part in the official transcript too.

Anyway, hope that helps give context.


Agreed. There's a huge chasm between being able to read/write and being able to speak. Speaking is just more fun as well imo. I'll have a think about ways that it could provide some qualitative feedback. Thanks again


I'll second wishing I could see the raw transcript vs the cleaned up version (perhaps vertically over one another in different colors?), and also how fantastically magical it is at allowing me to come right in and converse naturally.

Great job!


You could do something two pass and collect forced alignment confidence scores similar to what is fine with whisperx. The OP would need to incorporate that into the app but it’s a pretty standard approach.


Thanks - I'll look into this


It's a very nice proof of concept that could revolutionise self taught language learning. It would need to provide lessons and this AI cools provided guided generated exercises, their corrections, explain the correction and check the pronunciation


Thanks for the feedback. I'll look into adding this.


I like the Anki shoutout.

I personally would be cautious learning an extra language through a tool like this though, especially for a huge list of supposedly supported languages like this. Chat-GPT gets basic concepts like verb suffixes wrong, that is, explaining those concepts and generating example sentences, when I was trying it out a few months ago. I wouldn't mind conversations feeling a bit awkward or unrealistic, (EDIT: as they do here), but I wouldn't trust AI tools atm to get sentence structure, spelling and pronunciation (!) all completely right which is imperative for such a tool to function (and this one didn't for the two languages I tested).

If we assume the AI is already at that level, it would be trivial to have it show translations to user's native languages instead of only to English and excluding a number of potential revenue-generators for the capitalist machine ("learners").

Your prompt leaked in the text box when I pressed the Start Conversation button multiple times, and the site seems to be getting HN-hugged-of-death right now.

Also, adding to the person who mentioned this being close sourced, if a "deep-rooted desire not to look foolish in front of others" was among my reasons not to talk to a stranger in an extra language, I don't think I'd be any more likely speaking into a microphone, alone, in a room, like I'm conducting my biannual mandatory TOEFL© test every two years.[1]

[1]: This one is actually worse because you're in a room with 10+ others simultaneously talking into a computer screen, not talking to a person but just having your voice recorded, and it's outrageous ETS is getting away with charging people like 200 bucks (getting to see feedback for your grade is extra)


Anki is such a great product.

That makes sense. I do understand the worries about learning from AI. It's worth saying that different software is used for transcribing what you say (rather than ChatGPT), and that's been trained on lots of data (including YouTube videos) of people speaking that language, not to say that it could never get it wrong. But I definitely hear what you're saying about ChatGPT sentence structure on the less popular languages.

Good point about open sourcing. I definitely agree.


> Chat-GPT gets basic concepts like verb suffixes wrong, that is, explaining those concepts and generating example sentences

This matters if you ask it for grammar. But you can get to a pretty advanced level before this starts to affect you when trying to have a conversation which is usually far more forgiving. And while ChatGPT certainly makes errors, even for a minor language like Norwegian it gets it close enough that I can get it to have a perfectly fine conversation in minor regional dialects.

I think the best thing to do is to pair something like this with a more traditional course for learning the specific rules, because starting to speak as early as possible is a great way of building a level of fluency that takes ages otherwise. Not least because ChatGPT will still understand you pretty well if you mix two languages within a sentence, so you can start talking almost right away and tell it to translate as needed.

You can already ask this app to translate things to your own language - I had it both translate French to Norwegian, and asked it a question in French which it translated to French before responding.


Not in agglutinative languages where verb endings considerably change the meaning of a sentence.


Even then you'll find a conversational environment far more forgiving, because of a combination of context and a feedback loop. There's a reason childhood acquisition of language works just fine without an explicit practice of grammar rules.

The point isn't that you wouldn't benefit from it doing better, but that it doesn't need to be anywhere near perfect to help a typical language student improve.


I actually think I disagree with this. Yes it'd be incorrect some of the time, but that is also true of humans. Provided it gets it right most of the time and is reasonably fluent, I don't think that minor errors would matter much.

This seems much more a tool for practicing the stuff you learn in another context to me, rather than a way of learning in itself. You can use it without the social pressure of stumbling slowly through sentences with another person.


Awesome app! I don't know exact steps to reproduce, but I was just alternating between selecting a language and starting a conversation a few times, and at one point the conversation box filled up with the prompt:

> You: You are a human who speaks Czech. In this conversation, you are interacting with a beginner who is learning your language. Your role is to help them practice and learn. Keep in mind that they will make mistakes. Your main goal is to maintain the flow of conversation by asking them questions about themselves, and make your responses as simple, short, and quick as possible. Please speak solely in the Czech language, without providing any translations or switching to English. Please start the conversation by saying something like Hi, how are you? or Hi, what's your name? Polly: Ahoj, jak se máš? Jak se jmenuješ?


Is that all it takes to create a language tutor these days? Then you have no excuses not to add Bosnian for my @op.

I will even give you the initial prompt (I am knowledgeable as my wife is Bosnian i ja priçam malo Bosankski)


I wish it were that simple. There's also speech-to-text and text-to-speech to get working. But I've seen your comment above and added it to the language-add list. Thanks again for the feedback


It's a bit intimidating to me..

Polly: สวัสดีครับ สบายดีไหม? คุณชื่ออะไรครับ? You: I don't speak Thai well. Can you speak more slowly and can you write in Latin script, please? Polly: สวัสดีครับ คุณพูดภาษาไทยได้ดีไหมครับ? ถ้าไม่เข้าใจสามารถบอกหรือถามฉันได้เลยครับ ผมพร้อมช่วยเหลือคุณในการเรียนรู้ภาษาไทยครับ

Not what I was hoping for..


Learning the script of any new non-Latin language is the most important thing to do, before you spend time trying to learn the language itself very much. (Casting Chinese aside on this opinion, YMMV)

You will generally not be accommodated in the future, and the time you spend learning via a translit is wasted time. Furthermore, learning the script is generally easier than it looks — it just always seems so intimidating, being faced with unpronounceable squiggles… but this is what our brains evolved to be good at.

Learn the script first. Olly Richards agrees: https://youtu.be/NCfmSD2qeUI


> but this is what our brains evolved to be good at.

Really? Personally I find having to learn a new written script is one of the biggest barriers to learning a new language. Despite years of exposure (admittedly at low intensity) I still struggle with reading anything in hiragana/katakana. To be clear I agree trying to learn a language using nothing but romanisations is almost certainly a bad idea, but I'm not sure I've ever felt the human brain had any special aptitude for learning written scripts - after all we've only been using them a few thousand years.


I guess I said it kind of the wrong way around. The written script has evolved to be used by humans. It's still evolving, of course, so it's not like it's a paragon of efficiency; but humans are good at pattern recognition, and the written word is exactly just that.

Learning to read, is learning patterns.


Unrelated to the site, but you really shouldn't be using latin script for thai.


I really like this! I’ve thought about making something similar but never got around to it. This is well done and the STT-TTS works great!

My foremost wish would be for my tutor to point out and explain my grammar mistakes. I can hold a conversation in German but still make grammar mistakes. The LLM corrects them when transcribing and it would be really nice for it to underline the mistakes with the ability to explain them. Maybe have a second LLM running with a different system prompt?


Thanks for the kind words. It's actually a deliberate choce to not have it correct the user. The reason is that it mimics trying to get your point across in real life. Over time, through speaking as much as possible, your grammar starts to improve by osmosis and exposure. The only real evidence I have for this is my own anecdotal experience.


I think that's good during the conversation, but an option to do a "review" after the fact that points out a selection of issues spotted during the chat would be great.


Yes, very good point - that seems very doable and useful


Wow, this is so cool! xD I think this is the strongest usecase for AI I've seen so far. I'd love some more details on how you made it - how are you handling the speech to text? It's so clean, I was amazed it could understand my stuttering mess lol.


Yeah this is great, shocking how reliable the speech-to-text is even for my unconventional accent.

Imho UX would be improved by a larger text box, more clear demarcation between "stuff I've said" and "stuff the bot said," similar to how iMessage or Whatsapp have two-column message layouts or blue-vs-grey or both.

Also maybe some "hold spacebar to activate mic" keybindings would not go amiss, similar to how Zoom or Gmeet makes the mic hot if you're muted but hold spacebar. That would save a lot of choppy/distracting clicking. (Or possibly just make the mic hot all the time when the bot isn't talking?)


Thanks - The larger text box has come up a few times so I'll fix that.

Good point on not being able to discern between what you said and what the AI said. Will change that when I get a chance.

Sure - I'll look into 'hold spacebar' type functionality. Thanks for the useful feedback.


Thanks! Speech-to-text is Whisper API, although that seems to have caused a couple of errors tonight so am considering moving away from it. Will have to have a think and do a lot of testing.

I feel your pain, I've also been a stuttering mess in Romanian.


maybe add a VAD in front of it? faster-whisper uses silero VAD.


Will look into it


I love the idea, and just had a successful conversation with it for a few minutes in Portuguese. Nice work!

For me in that language I need to work on understanding native speakers, so currently it speaks too slowly for what I personally need (it's too "easy"). I did ask it twice to speed up but it didn't seem able to - I guess it's hardcoded in your text-to-speech code? Is that something you can add an option for?


Awesome! Yes, it's hard-coded to basically be in 'beginner' mode at the moment, but I was considering allowing the user to select intermediate or advanced, which it sounds like would be more useful for you. Thanks for the feedback!


I would love to try it on advanced if you get time to do that. I honestly think I'd use it a lot. If adding it to the UI takes time I'd be absolutely happy with a URL query parameter prettypolly.app?level=advanced if that's easier.


It probs wouldn't take much longer to just add a slider or something too. If you don't mind me asking, what are you planning to use it for? Interested to hear some advanced use cases


Great. Oh, I'm just aiming to get much better fluency and ability to understand native speakers in real-life contexts. I've had basic bad foreigner level for a while now.


Everybody seems to be raving about this app but I'm running into one big issue.

Everytime I try to speak the system states I spoke the following: Subtitles created by the Amara.org community

sooooooo, guess this just doesn't work for me.


How many times did you try? It did this to me a couple of times but not every time.


As I'm at work not more then 10 min. But I'll give it another try this evening then :)


Please do. There's a flagged issue below where if you send any blank audio, it magics up some phrase that it's been trained on. Am working on a fix, but for now, it should work if you make sure there's audio in every response. If there isn't audio being transcribed, then please do let me know as that may be a problem with the app on certain devices with certain software versions. Thanks for the comment.


I'd really like to try this but every time I try it redirects me to https://www.prettypolly.app/payment-page and asks me to upgrade my subscription. I have yet to actually try it a single time. I've tried on several different days. Does the site actually work to try it for free or is the only way to try it to pay for a month?


Hi there, sorry about the poor user flow at the moment. We had to take emergency measures and change the signup process to ensure that premium members are able to get a service. You have to log in before you try to upgrade to premium. Here's the link to log in: https://www.prettypolly.app/login

Sorry for the poor UX in the login process. We'll be fixing this soon.


Amazing! I think the best way to learn a new language is to speak with someone and better if that person would tell you where you are wrong! I'm using Duolingo to learn but I'll start using your app as well. Fast feedbacks: - If you don't know what PrettyPolly is saying you close the conversation and that's bad for PrettyPolly (which is being less used) and for the learner (sad mood because you can't handle a conversation after days of learning) so maybe think about some difficulty levels? (I think you can handle it in the prompt of chatgpt) or I read in the comments about the translations. - When I click the button to stop my microphone, the user doesn't have any feedback of what it's happening, so maybe show a loading spinner and block all the buttons?

Btw well done! Can't wait to see updates!


Thanks for the kind words. Difficulty levels have come up a few times so I'll see what I can do.


Seems cool! I know basic Spanish from Duolingo but not enough to understand what it says to after a few rounds so my suggestion is that there is a side-by-side translated version of the conversation.


That makes a lot of sense! I've been awkwardly having a Google Translate window open with it, but agree that it would be much less onerous to have it built in. Thanks for the feedback.


The instructions feel unnecessarily convoluted to me when ChatGPT can do a very good job of doing the translations itself. I'd suggest you consider telling people to ask your app itself for help if stuck, and falling back to English if struggling to find words or if in need of translations.

Having tested your tool w/French, it in fact handles it perfectly well if I insert random English words in the sentences (and translates it into French), handles it fine if I ask it to explain words, and even if I ask it to repeat an explanation or sentence in English (hilariously, but I guess to be expected and provides nice continuity, it then spoke English with a French accent), so not much need to put things into Google Translate.

I sometimes do this with ChatGPT directly to practice language - just start asking it questions in another language and ask it to explain in English if something is unclear, or give it a short prompt to get it to correct my questions before answering.

The immersion you get when you can just start a sentence in the language you're practising and just switch "back" when you lack the vocabulary feels much better - you can be a lot more ambitious and throw yourself into sentences without a clue about how to finish them in the language you're practising.


Yeah, good point. I think the FAQs need a question on "How to chat with Polly," or something similar which will give examples of the type of prompts you can use, and that you can use your native language to help if you're struggling, like you suggest. Thanks for the feedback.


Maybe even just show a sample conversation that includes some of those methods. I think especially for rank beginners in a language the biggest problem will be overcoming the problem of starting to say something you don't know how to finish, and so just getting people aware they can fall back to English (or other languages, I spoke Norwegian to it in the middle of a French conversation) might make people more comfortable getting started.


Yep, sure. That's useful. Thanks again.


Nice, I'm looking for something I can learn a second language with and also teach my son English. In both our cases, I found this bot to be too fluent for the level we're at. It would be great if the bot spoke simpler shorter language, perhaps even limited vocabulary. I tried to guide it through speaking but couldn't get it simple enough.


Thanks for the feedback. Am thinking about ways to make it work for different levels (beginner/intermediate/advanced). If you want to receive a notification when this is integrated, feel free to sign up and you'll be added to the mailing list


This looks great thanks for sharing. I've been using a similar app on iOS and it's definitely been helping with my German. https://apps.apple.com/us/app/langs-ai-language-learning/id1...


I’ve tried a couple that I found on the app store.. langs is good, I like how simple and to the point it is. Doesn’t have a web interface though


Thanks for the comment and for sharing that app - I'll check it out.


When you're referring to Portuguese in your app, it seems like there's a blend of Brazilian and European Portuguese. It's crucial to differentiate between the two since they have distinct nuances. I'm presently working on mastering European Portuguese, so it would be beneficial if you could specifically incorporate that.


Got it, thanks for the tip! Will try to create 2 diffeent languages. I think this is a problem also for countries where they have different regional languages/dialects like China.


Hi ! Neat little app!

I was commenting elsewhere that I was super happy that Bard t had opened up foreign languages recently. That way I could get my kids to practice their foreign languages. But bard was not great, and prompt was not as intuitive as yours!

Click to talk. Click to stop. Great!

I'll be using your app. Great idea! Keep UX simple that even a 5 yr old can figure it out!


Thanks for the kind words. Sure - anything you didn't like about the UX?


Need: 1-click native tongue dictionary. How: the AI prompt clickable, specifically, a word should be clickable to find its meaning. Perfect scenario: clicking a word to automatically read out loud the word in the english meaning. Often I found myself seeing words I didnt know the meaning to, and its not possible to quickly iterate on it to get feedback. Yea im not going to use your app with multiple tabs on.

Want: hard to say if necessary, but perhaps look into making the "stop speaking" button redundant . Note however, i found your voice commands ux to be infintely better than bard. (Bard has the problem that its hard to make prompt inputs in foreign language unless you have some kind of extension installed, buttons poorly placed, tiny, etc)


I love it ! It's very cool. I was thinking of doing the same kind of thing too. I'll throw here some ideas that I thought would be cool also: - adaptative level: just gradually starts to add more vocabulary in the conversations. I would assume an LLM already knows by design the most common and the most rare worsts - remembers past vocabulary, and uses more the vocabulary you struggle with - remembers grammar concepts, ideally makes a brief summary of a new concept, and refreshes your memory if it's a concept you've already seen, but forgot.

I currently struggle learning Korean because so much of the language uses particles, and they all look the same to me. I think I would greatly benefit a tool that tells me "hey ! you've seen this concept before, look: ___" and by repetition give reinforce my grammar knowledge

This is how I would see the ideal language learning companion


Thanks for the feedback and suggestions. Will see what can be done.


Hmm, just had a weird speech-to-text hiccup.

I've asked to teach me how to construct past tense in Spanish. It gave me a few examples (Spanish and English translation right next to it, text-to-speech engine hadn't realized that some words were in English, though, and pronounced them in Spanish manner), then suggested I try to form my own sentences. So I've replied "Yo entendí sus ejemplos!" (uh, I hope I've got it right) but somehow it transcribed it as "You: Suscríbete para no perderte nuestros próximos videos." While I think I've stuttered and may be mispronounced something, it surely didn't sound anywhere close to that.

Being an LLM, it didn't bat an eye and asked me about the name of my channel :)

But I'm wondering if there was some request-response mismatch with the Speech-to-text API you're using and I've somehow got a phrase intended for someone else.


Thanks for the feedback. Yeah - it does seem to have hallucinated a little tonight. It normally happens when I time the microphone wrong and send some blank audio, then it seems to want to magic up some sort of phrase from nowhere. Am working on a fix for this. Any chance you could try again and let me know if it happens again? Ofc, no worries if busy. Thanks again


I tried it in Spanish, to get an idea of how it works. Pretty cool! Definitely worth polishing the interface and making it a native app. How hard would it be to add more languages? My current target languages aren't supported: Hungarian and Bulgarian.


Thanks - yeah lots of UI/UX work to do. I've added Hungarian and Bulgarian to the list of languages. If you sign up, you'll get an email when they're added. Thanks again.


Just tried a few languages and there seems to be a problem with German speech to text. When I say "Hallo" it just thinks I said "Untertitel der Amara.org-Community" :-)

English and Spanish seemed to be functioning OK though, but I really wanted to test my German.


The reason for this is quite interesting and is due to the data that Whisper (the transcription service) was trained on: https://github.com/openai/whisper/discussions/928


Thanks for the comment. Sometimes if you toggle the microphone quickly without saying anything, it will halucinate something. If you could, is there any chance you could try again in German and let me know if it's still not working?


Curious about this, why do you think it hallucinates? Isn't that supposed to be the direct output from speech recognition?

PS: Just happened with Spanish now as well, which was working fine. (You: Subtítulos realizados por la comunidad de Amara.org)


No, still not working. Also tried a bunch of other languages and they all seemed to work except Turkish, which also spits out something related to subtitles...


Ah darn - okay, thanks for the feedback. I'll find a way to stop it from doing this.


It would be amazing if the microphone detected say 15 second silence and responded automatically for a fully hands free experience


Yeah - this is a good shout


I just showed this to my partner and she really enjoyed it! I'll add a feature request of Slovenian as an additional language.

By signing up I'm hoping that you'll push out emails advising if / when new languages are added?

Wishing you all the best :)


Thanks! Great - I'll put Slovenian on the list. Yes - you've been added to the mailing list which will inform you of things that will help you learn, including new languages. Thanks for signing up! You too!


A couple notes from trying this, though people have probably already said them:

Would be nice to be able to pull down the conversation box so that it can fit more text in it (forgive my vocab, not a software guy lol).

Would also be nice to cancel my audio if I screwed up, as is common when learning a language. This leads to incorrect words/phrases ending up in the conversation and hurting the flow. For this matter, it would also be nice to undo things you have already said, though perhaps this is a limitation with the ChatGPT API?

All in all, really cool! Great job on this.


Nice work and congratulations on launching. I saw some people were mentioning Langs which is an AI conversational app that I built for iOS https://apps.apple.com/us/app/langs-ai-language-learning/id1....

Are you planning on building a mobile app at any point? i'd love to collaborate if you are :) also I like the fluency score idea, good stuff.


Thanks! Yes, definitely planning on building native apps. The first will probably be on iOS, since most issues seem to be coming from those.


Wow, I've seen many AI-powered language apps lately, but this is by far the best. The others, if they offered verbal conversation at all, really struggled with my stuttering while trying to find the right words. I'm impressed that yours doesn't have this problem at all. It even understands when I use technical terms in english.

I'd love if the AI was a bit more fleshed out, so I can actually ask questions without getting the "I'm an LLM by OpenAI..." lecture.


Thanks for the kind words and for the feedback. I'll see what I can do about stopping the AI from breaking character.


Hi Chris, I absolutely love it! This a https://i.kym-cdn.com/entries/icons/facebook/000/005/574/tak... moment.

The functionality is great, do some clean up and it will be an amazing product. E.g. instead of pressing start and stop, user could hold space while speaking.


Thanks for the kind words! Yes - definitely lots of cleaning up to do, and the spacebar idea has come up so I'll try to add that in!


Note: the website does NOT works if you call https://prettypolly.app/ (without the www). not sure it is on purpose

note: Cant find Esperanto, which is a useful language to get learning curve fast for others language, i guess it worth the add?

I wonder if there is a possible implementation to learn sign language (i'm thinking about LSF since i live in a french speaking area)


Yeah - sorry, hadn't had time to configure the root domain yet, but will do very soon. Thanks for reporting. I'll add Esperanto to the list of recommended languages!

I have no idea where to start with sign language, but I can do my best!


Very cool!

I could see this tool to be very useful in my Japanese studies. I only have two problems with it:

- AI feedback can sometimes be pretty overwhelming. Maybe you could implement a level selector (beginner, intermediate and expert?) so that the AI would switch to using shorter/simpler or longer/harder sentences accordingly?

- The speech to text works fine most of the time, but sometimes it gets a few words wrong. Could there be a way to correct the last sentence input?


Thanks for the feedback. Yes, the beginner, intermediate, advanced point has come up so I'll look at adding something like that.

Just to check that I understand, do you mean changing the text manually when it gets it wrong? If so, sure - I can look into it.


I think manually editing your last sentence would be cheating in the context of a speaking app. Entirely erasing the last sentence of the user (as well as the AI's reply, if any) and having him repeat the sentence he had in mind would be better because it would force him to speak more clearly.

(Also an undo button would be easier to implement than a whole system to edit previous sentences, right?)


Yeah, agreed that makes much more sense!


the app is great. Thank you for making this. The app encourages me to talk more about the topic, which I know sometimes human teachers don't do. I like the stop and start button so I can pause and think before I speak. I was speaking Mandarin and Japanese. It responds well when the my transcript is gibberish (my pronunciation is not perfect).I know some people will point out that it lacks phonetic feedback. But the point of apps like this is not giving feedback because to help learners build up confidence. Of course, some feedback in any form would be great. But this app is multilingual, so it's hard to you will have to prioritise some languages first. Overall, simple but effective. I know I'm being greedy, and I know it doesn't make sense when comparing with other traditional apps like Busuu or Pimsleur. But if you allow decent number of conversations with affordable price (netflix or spotify subscription price :)). I would consider buying it.


Thanks for using it! Glad it's been helpful. You get 15 free conversations per month when you sign up. If you want unlimited conversations per month, it's $4.50. Sound reasonable?


wow, that's more than reasonable. thank you, looking forward to your upcoming updates.


Great - glad to have you onboard and if you're signed up, you'll get the updates :)


I was searching for something like this as a complement to Duolingo.

Duolingo is good for memorization but not for actual using the language in real life. Spanish is spoken so fast that something like this really really helps to understand the spoken language at the speed fo native speakers.

I had been thinking of this use case for a while but had too many other things going on.


Chris, this is such a wonderful idea. Congratz to releasing this. There are so many ways how to improve this, first thing that comes to my mind is to improve readability of the dialogue. I would to have this as a typical messaging app layout, where you see in separate bubbles who said what. Keep up the great work and good luck!


Thanks. Yeah, that's a good idea and I'll see what I can do!


In the first paragraph is "Select your desired language in the app.". If you click on the link, it goes nowhere.


Thanks for reporting. Will fix this.


I kept speaking Croatian while I was practicing Spanish and it underatood me. What kind of magic is this? Really cool project!


Ahah the witches of the internet are pulling the strings


I was able to have a conversation in Mandarin, a language I have yet to utter a single word to a native speaker in since I'm self conscious about my accent. Really cool :)

If I had a suggestion, it would be to add a button by each line up have the TTS repeat saying aloud the sentence. Might be useful for Chinese/Japanese


Thanks for the feedback. I'll try to add something like this


Hey Chris! Cool app. Practiced a bit of German with it and it works well! I had the same issues as you with language learning and built a similar thing called https://www.langchats.com. Looking forward to seeing your progress.


Awesome - I'll check it out. Thanks for commenting


We made one too. https://www.languagereactor.com/chatbot

It's using the llama-2 13B model currently, running on gpus in the basement. LLama-2 is weak for languages outside of English, so we use machine translation, which is mostly ok but gets wierd occasionally when the model isn't aware that it's not 'speaking' English. The reason for not using chatGPT was that when we started it had a very stuffy character and didn't like to role play (think it's better now, text-davinci-003 was fine but much more expensive). Running models yourself can get potentially costs way down. chatGPT is used for the correction feature.

I check the logs occasionally, the feature doesn't have a lot of users. There's two issues I can see: it's hard to get the model to speak using simple language with a prompt only (probably a fine tune is needed), and, I haven't found good TTS for many languages (comparable to microsoft APIs) that we can run on our hardware. TTS is much better in Edge than in Chrome btw.

I've heard https://www.loora.ai/ is good, haven't tried it yet.


This is really super cool. I wonder if you could use elevenlabs or something similar to make the voice more legit because it feels at the moment like I’m speaking to a satnav (in French).

I also got fluency score None which, let’s be real, is pretty accurate. But I feel like I need a bit more to help me learn.

The microphone interaction is really nice.


I agree but 11labs can be too expensive something transient like a conversation. Maybe give users a choice, those who pay for it will get it. But personally it's not really a deal breaker because 11lab doesn't have Japanese and Mandarin which I'm learning, haha!


Aha thanks man. Yeah - agreed that there's lots of work to be done on the speech.

Makes sense - getting from 0 to 1 can be a bit of a struggle with this method, so could be good to find some ways to make that easier.


Really cool project, didn't test it much but tried German on it and since I don't really understand German that well it didn't go so well. Then again I just woke up so maybe I should've just googled translated the last answer and gone from there.

Just wanted to say this is real cool.


Can you add support for pinyin mandarin?


Yes, of course. I'll try to add this soon. Thanks for the input


Thanks for sharing, Chris!

How does something like this compare to direct ChatGPT usage? I didn't sign up, but I'm interested in LLMs for language learning and I'd like to understand what you've built compared to what I'd get by asking ChatGPT to be my Spanish pen pal tutor.


Thanks for the comment! The main difference is the use of speech instead of text. At school, I could read/write in French, but then was absolutely lost speaking to French people. If you want to speak a language, you have to speak it.

The fluency score is also quite useful for tracking your progress. I tried to get ChatGPT to give a fluency score, but it simply couldn't do it so I had to build some software to create a fluency score using rate of speech and the number of unique words you use.


Amazing app. Ux could use some adjustments like showing loading or clearly indicating when the AI is listening. That way I'm not stuck wondering if the words went through or not.

Oh, perhaps a hint or help button, so it'd give an example of how to respond


Thanks! Yes - lots of work to be done on the UX. A processing-style spinner is the top of my list to add this week.

I'll see what can be done about hints. Thanks for the feedback


This is great.

One thing that could be improved: sometimes when the AI gives a "fill in the missing word" task i.e a sentence with a "______" in it, the text-to-speech literally says out loud "underscore, underscore, underscore..."


Thanks! Yes, it's super annoying when it does that aha. In what context did it do it for you. Will try to find a fix.


Marvelous! One of the best applications of a ChatGPT backend that I've seen yet. It's even remarkable to talk in your own native language to ChatGPT, but there is just so much room to build out the next great language learning app here.


Thanks for the kind words!


It seems very nice, I tried japanese because I'm studying it, but it got difficult too quickly. In particular japanese needs furigana absolutely (the hiragana over kanji) or romaji (options to be activated/deactivated).


Since it is in the browser, you could use the Yomichan browser extension to enable furigana/translation on hover.


Got it. Will work out how to add this. Thanks for the feedback.


Hi Chris,

This is an exciting project you're working on, and I'm impressed by your use of AI in language learning. Speaking with native speakers is indeed an invaluable experience in picking up new languages.

We are developing a similar project called Speechy. While PrettyPolly aids in practicing languages, Speechy records, analyses, and provides feedback on speech, highlighting areas for improvement in pronunciation and grammar.

Our aim is to offer practical insights to enhance spoken language skills.

Currently available as a macOS desktop application, Speechy is in its alpha phase, and we'd be thrilled if you could be part of our testing journey. You can download the beta version at (https://speechy.ai)

Given the complementary nature of our projects, there might be potential for synergy.


Amazing app. I am trying to learn dutch with this. It pushes you to converse in the language of choice. I also tried to get translation in english. Could be a good feature to understand new words and form better sentences.


Thanks man! Yes, definitely - will look at adding translation


Thanks for sharing! I'm a bit confused with the fluency score, what does it mean exactly? It would be good to have some feedback and pointers for improvement, otherwise I don't see myself using it for long.


Thanks for the feedback. The number of unique words you use and the rate at which you speak contribute to the fluency score. I'm a little reluctant to give the full equation, because it prevents us from trying to game it.

That's useful feedback on pointers for improvement - I'll see what can be done.


Wow, I am pretty amazed at how well this works, at least for Spanish. Actually, I haven't really been convinced there were that many use cases for generative chat AI so far, but this convinced me!


Aha glad you're a believer now


Thank you for this! I’m enjoying the Chinese so far.

I’m unable to sign up, though. On mobile, I’m trying to hit the sign up button after filling out the form, but it’s not doing anything or giving me an error.


Glad to hear it! Do you have any preference on traditional Mandarin vs. simplified? It's simplified at the moment.

That's weird. Any chance you could ping me an email with the email you're using to sign up and I'll work out what's going wrong? chris@prettypolly.app. Thanks again.


I'd say you should offer both simplified and traditional in the language dropdown as there are many learners of both.


Thanks - that's useful


I was thinking about a piece of this problem earlier and I'm curious if you figured out element that I didn't:

How do you provide feedback on how someone mispronounces a phrase?


I think educators are too afraid in general of getting into some real accurate phonetics IMO. Lowest hanging fruit seems to be phonetic diagrams of where the tongue/lips should be. Perhaps normal in other peoples experience?


Phonetic diagrams aren't feedback though. You would explain how to do something correctly, but user wouldn't know what elements they need to change, unless they figure out on their own what they are doing right now. I know my teachers told me to do X when they heard me mispronounce something specific, but that's irl small group/1on1 experience.

Again, I'm curious if someone solved it. Theoretically this is within modern video/audio AI, but that doesn't mean we have the models or even organized data.


This is a great idea. It would also be nice to do this without the microphone input to practice writing. I'd also like the AI to correct my grammar if it's able to?


Hey thanks for the feedback. I'll be honest with you, if reading and writing is what your goal is, then you'll get better results speaking to ChatGPT in the OpenAI chat interface. If your goal is to be able to speak it eventually, you'll need some sort of speech version.

Sure - it's a deliberate decision to get it to not correct you mid conversation, because that mimics real life. However, I'm looking at having a way for you to get some qualitative feedback alongside the fluency score after each conversation.


Isn't your secret sauce here though the prompt? I've tried with mixed results trying to get it into the role of a language teacher.


Another thing I would want Chris is for the conversation prompt to be different in subsequent conversations. I'm sure you've thought about this.


Nice to see that a lot of people working on this topic!

Quazel was one of the first I saw: https://www.quazel.com/


This is very interesting. A question, how do less widespread languages fare against English? Is the quality output the same in i.e. Thai and English/Spanish?


Given the performance of tokenizers by language the answer must be “worse,” there are a variety of sources for this .. they typically focus on cost, but it’s also borne out in quality.

https://denyslinkov.medium.com/why-is-gpt-3-15-77x-more-expe...

https://arxiv.org/pdf/2305.13707

https://arxiv.org/pdf/2305.15425

A quick Kagi can find a lot more.


Mixed reviews so far. I've had one person say the Hebrew was brilliant, and one say that it was terrible. It seems to be slightly less accurate than English, but good enough for most purposes


Wow. This works much better than I had anticipated. How did you get it to work so fast? Usually there’s a huge delay from the LLM/TTS/STT processing.


Thanks so much! Blood, sweat and tears aha.


This is really cool. I can practice reading, listening and writing on my own. I wanted to practice speaking German.

I learned a few languages just through speaking with natives.


Awesome - glad you're finding it useful!


Generally if you are beginner and want to actually learn the language, you should focus on the input not the output.

Few problems I find with these kind of apps:

How is that different from just using chatGPT?

Why can't I just write my response instead of using mic?

chatGPT makes mistakes, as a beginner you can't spot them. (I talked to it in polish (my native language), it was making grammar mistakes)

Speech recognition is not the same as native listening to you, speech recognition software may "understand" you, but native would not and vice versa.

ChatGPT can't correct your pronunciation.

Replies generated are stiff and unnatural.

TTS can't model speech accurately (it lacks emotions etc.)


I personally like that it's using speech recognition.

First, chatting and speaking are not nearly using the same skills, training for one does not necessarily train the other and you can end up having a hard time to find the words you want on the spot.

Secondly, speech recognition while not perfect, does help to make you understood by a native speaker. Speech recognition is usually working best on what's considered some of the most neutral accents in the target language, which is as a foreign speaker, exactly what you want. Seeing the recognition failed is a clue that you might need to train again to speak those words.

> TTS can't model speech accurately (it lacks emotions etc.)

I do agree on this last part though and usually TTS lacks support for other accents.


I agree that chatting and speaking requires different skill set. However I would argue that it is even more of an argument to not use speech recognition here (or at least not to force it), because chatGPT is chatting and learner is speaking. Transcription will always lose some information (for example your tone can indicate sarcasm, but chatGPT can't detect it).

To the second point: whisper can be helpful, but how can you know if it fails because of you and not the software's error? I spoke in my native language with traditional accent and it still made mistakes, also it hallucinates. Additionally being understood by whisper doesn't mean, native will understand you.


I do agree that the text generated by ChatGPT isn't really "natural" but more verbose and text oriented, some things that a native speaker would not necessarily say and won't understand speech nuances. It's clearly not perfect, I'm really not claiming that it is.

It's a good tool that I'm going to use a few times per day though, there's no really substitutes to speaking to get better at it. I'm also using other methods and tools and this would be a minor addition to my learning schedule.

> I spoke in my native language with traditional accent and it still made mistakes, also it hallucinates.

I'm also in this scenario actually because I'm a native French speaker and I cannot make myself understood by Google or Siri at all because my accent is way too strong and far outside the training voices that they used.

It's kind of a paradox but it's less a problem for non native speakers in my opinion who are trying to pick the most common accents they can in order to be broadly understood.


That's understandable,these tools definitely can be helpful, but learners should know their limitations and problems.

Also I just think speaking to the actual native speaker is still much better practice, especially given tts quality. It even pronounces words incorrectly in Japanese (wrong pitch accent).


Oh yeah sure, there's no doubt about that, it's just that these ChatGPT tools have two massive qualities compared to native speakers, they are available at any time and timezone, day or night (even for just 2 minutes) and they never get bored, you can ask the most mundane questions over and over again to practice.


The Hebrew one is so wrong... both grammar and pronunciation.


Darn, okay - I'll work out how to improve this. Thanks for the feedback.


I really love this. Very cool. One step closer to my goal having zero human interactions in my day.

Just make sure to refine it so it’s a product not a feature.


Zero interactions in multiple languages is your goal by the sounds of it


Very good app, how so you sign up for premium service?


Thanks! Once you hit the conversation limit (15 per month) it will automatically take you to the upgrade page. If you really fancy paying, feel free to mash the start conversation button aha. Otherwise, you can just enjoy the free convos for a while :)


Underlined link not working FAQ: > How do I start using PrettyPolly to learn a language quickly?

1. Select your desired language in __the app__.


Ah thanks for reporting!


I think the app currently is under heavy load, so it is unstable and give 5xx.

When it has worked, it was amazing. I am ready to pay for it.


Yeah - I apologize for how slow it is right now. I didn't anticipate nearly this much traffic. I'm going to be making it faster and more robust over the next few days


Have made some changes now. Should be running as fast as it was yesterday, but let me know if any issues.


Did not log off on purpose but I am logged off now. Also, cannot login. Cannot use the app at the moment. BTW it is a paid account.


I can imagine how many native speakers and speaking club hosts will lose their jobs. Really great app, love it!


Awesome project! I had fun talking to it. Thanks for sharing. Can you give some hints as to what technologies you used?


Looks amazing! But will you be adding other languages like Marathi(15th most spoken) and Telugu(16th) if possible?


Yes, I'll add these to the list of other languages. Thanks for the feedback


Klingon would be amazing too


Nice work. Would be nice to for the app to modulate difficulty dependent on respnoses.

How did you get it to respond so fast?


That's a great idea! Some people suggested beginner, intermediate and advanced modes, but this sounds much more user-friendly.


Tried it in Korean and I quickly reached the pathetic limits of my Korean speaking abilities. Glad it was in 한글.


Do Cantonese please! It's such an underrepresented but widely used Chinese language...


Sure - will do. I had to create a shortlist for the MVP, but Cantonese will definitely be coming soon.


I can't find the app in the Japanese App Store, can you please release it there :)


If you're looking for an app that's similar you could try this one I'm using https://apps.apple.com/jp/app/langs-ai-language-learning/id1...


Afraid it's only a web app at the moment, so you can use it in your browser on mobile. There will be a native app soon, however


I'm getting 502 error. Congrats if this is because the demand is on fire :)


I think I got your email. Thanks for reporting that. Will get it sorted.


I tried German and managed to 502 the site. It's still down for me.


Sorry about that - the server was overloaded, but it's back on now.


Works now. Remarkably helpful. Unfortunately it gets exponentially difficult after the first few exchanges. But I think there is definitely something to this approach.

Maybe if there is a way to highlight individual words and get a sneak peak at translations? And maybe a side button to speak an English word and get the target language translation (just as an aid -- currently I'm keeping google translate open on another tab).


Cool project, I might recommend ordering the language list alphabetically


Thanks - yeah that probably makes more sense!


Or start with more common languages then switch to another type of grouping. For example, Swedish, Finnish, Norweigan, German, Dutch, etc


Learn a language quickly by practising speaking with a human


Seriously impressive! This has some big potential for sure


Thanks!


Yes I have a question, when is your next round!


Aha none planned! Let me get through the MVP launch first!


Hug of death? Error code: 502-backend


Yeah, apologies. Wasn't expecting nearly this much traffic. Have made it more robust so it should be running faster than ever now, but let me know if any issues


Will you make this open source please?


I can definitely look into it. Do you have any suggestions for how to do it well?


When open sourcing your work, look first into licensing and make sure to understand what each license offers and what implications it has.

Then you can chose to push it to github or similar, but please prepare a good readme.md file.


Upload the source code? :")


What is the fluency score based on?


Good question. I looked through literature to see whether anyone had come up with any good ways of measuring fluency, and there are a couple out there.

The key indicators that kept coming up were the number of unique words used and the speed at which you speak, so they're involved. I'm hesitant to give the full equation, because I thought it might be better to leave some mystery so we aren't consciously trying to game it while we learn.

It is by no means perfect, but I think it's useful to have some objective metric to track progress.


Cool, thanks for the response! I'm enjoying using the site.

One followup: does it take into account grammatical errors, incorrect conjugations and that kind of thing?


Glad to hear it!

Sort of. It will try to understand you even if you get the grammar wrong (a bit like a human would), but if it really has no clue what you're trying to say, it will ask you to repeat. I think it's best that way as it mimics learning in real life. If you keep getting your point across, over time, your grammar will slowly improve. The only evidence I have for this is anecdotal from learning Romanian with it myself.


seems to work great! it would be great to have pinyin for Manderin.


Thanks for the comment and thanks for the feedback! It's on the list of languages to add. If you'd like to get updates, feel free to sign up with a free account and you'll automatically be added to the mailing list.


It's very cool. Some things I noticed: fluency scores are always None for me. If you ask their name they give a different one to the diarised name! Thirdly they come in very strong, above my level so I find it quite hard to use.


Thanks for the feedback - that's all really useful. Do you mind if I ask what language you were using when the fluency score was none? Seems that it's broken for a few languages, so will fix it asap.


French


Great idea!


Thanks!




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: