Transcrobes: Free + Open Source language learning platform (for Mandarin)

[AntonOfTheWoods]

On 1/14/2022 at 3:52 AM, Jan Finster said:

Listrobes: whatever this is...I guess can choose if I know a word. Do, I really need to click through ~8000 words that I know manually. Why is there no "mark all as known" button?

 

Actually the reason this interface is for EXACTLY that purpose! Each of those smiley faces represents whether you know the word or not. If you open the config, you can set the default to "Happy", and then only click on maybe one or two you don't know. As you go through more vocab (which has a default ordering but you can also do this based on lists you import, change the order, etc), you will get to a point where it is more work to click on the words you don't know than those you do, so you can change the default to "Frowny", meaning when you click it goes to "smiley". 

 

Again, all of this should be clearly documented in the direct link you can click on in the top right corner (otherwise it's a bug). Because this can take a while and is not hugely fun, I wanted to make the interface as clean as possible.

 

Actually this interface gives you a LOT of power to enter thousands of words quickly. Getting the system to know you is crucial to the whole thing, so it needs to allow that as quickly and painlessly as possible. I have been in this for so long it is hard for me to see what is obvious and what is not though, so I am very interested in any and all feedback!

[AntonOfTheWoods]

On 1/14/2022 at 12:01 AM, Moshen said:

I strongly disagree.  The other day I was really interested to read an article in Chinese Wikipedia on a certain topic, and I started it, but I had to look up so many words (each with just one click) that it turned into an unpleasant chore, regardless of how motivated I was to learn about that topic. 

 

@Moshen This is EXACTLY why I started this project. I say this in the getting started but that is a huge fail that was unreadable... If the system knows what you know, then it can proactively gloss the words you don't know, and leave the rest unglossed so you get the benefits of reading in Chinese. That means you should only have to click rarely, when the system has not understood the word properly, or the grammar is too hard.

 

Here is a screenshot of Chinese Wikipedia *for me* using the browser extension (https://am.transcrob.es/#/brocrobes, available on the Chrome store https://chrome.google.com/webstore/detail/brocrobes/akeangohpdjllngpjiodmebgfhejbbpo?hl=en-GB?

 

 

You also get the "normal" stuff like a full popup dictionary entry on click, mouseover for all words (in case you want to make sure), full sentence translations (typically for if the grammar is too hard), synonyms, other contexts you have seen a word in, and more.

 

You have various glosses you can choose from: none (words are segmented and you get all the rest of the tools), English, Pinyin, English+Pinyin, Simpler words (if you already know a synonym it gives you that, otherwise English if you don't).

 

The same kind of reading interface is also available for reading EPUB ebooks and watching video. At the moment you need to find and download epubs and subtitle files (SRT and WebVTT supported for now, more later) as well as the video files. That is a pain, that is mainly oriented to technically savvy people (who know where and how to download such stuff). I will try and get the tools working as plugins to content platforms (Netflix, YouTube, Amazon books, etc.) but that is a massive amount of work, and for the moment it is just me doing everything, including a PhD.

 

[AntonOfTheWoods]

Here is an overview of the tools:

 

https://www.youtube.com/watch?v=UodaDI4XVf0

 

 

If you are more interested in the theory side, here is an introduction and demo that I presented at an online learning conference in Dec 2020. The first 7 minutes or so are very academic (it was an academic conference).

 

https://youtu.be/NYGFL3ArQxs

 

The demos start here: https://youtu.be/NYGFL3ArQxs?t=484

 

The interfaces have been upgraded since then (they were very unstable back then so I rewrote everything...) and I have added stuff but all of the principles are the same.

[AntonOfTheWoods]

On 1/13/2022 at 4:46 PM, markhavemann said:

I will say that the amount of reading required to have any real idea of what you're doing is putting me off. I love reading but I don't want to have to read thousands of words just to know if it's worth my time to even look into. 

@markhavemann

Is the video I updated the initial post with (at the top) a better quick intro?

[alantin]

On 1/14/2022 at 4:08 AM, AntonOfTheWoods said:

Here is a screenshot of Chinese Wikipedia *for me* using the browser extension (https://am.transcrob.es/#/brocrobes, available on the Chrome store https://chrome.google.com/webstore/detail/brocrobes/akeangohpdjllngpjiodmebgfhejbbpo?hl=en-GB?

 

Now I'm INTERESTED!

Due to all the talk about SRS I thought that this was unlikely to be something useful for me, but after seeing that screenshot I think this may be exactly what I was looking for (unsuccessfully) some time ago! I described my reading setup in another thread just yesterday and what I'm missing is actually and easy way to extend that out from reading text files into reading web pages. It seems that I would be able to import my known words to this thing in a csv file and then show pinyin for the unknown words right on the page using the browser plugin.

I'll have a look at this during the weekend. And you should definitely put that screenshot on the very top of your website!

Here's the description of my setup and reasoning for my reading practice:

https://www.chinese-forums.com/forums/topic/61839-tips-and-tricks-for-learning-chinese/?do=findComment&comment=484340

https://www.chinese-forums.com/forums/topic/61839-tips-and-tricks-for-learning-chinese/?do=findComment&comment=484357

[Jan Finster]

So, I am currently trying the chrome extension:

What do I need to put as server URL?

 

Trying to test it with Deutsche Welle (https://www.dw.com/zh/武汉暴发疫情两周年-病毒起源陷入罗生门/a-60420872)

 

I have been waiting 10 minutes and I only see the "loading" icon rotating. Is this normal?

 

[AntonOfTheWoods]

On 1/14/2022 at 7:14 PM, Jan Finster said:

What do I need to put as server URL?

Sorry, that is https://am.transcrob.es .

 

Because the platform is open source, and anyone can run their own server, I didn't hardcode that. I thought hardcoding it might also make it harder to get into the store.

[AntonOfTheWoods]

On 1/14/2022 at 7:14 PM, Jan Finster said:

I have been waiting 10 minutes and I only see the "loading" icon rotating. Is this normal?

If you have entered the server and a valid user/password, then at worst you should wait a few seconds (5-6) for the first "enrichment" to start appearing.Looks like I need to add better error messages.

 

Longer paragraphs can take a bit longer (10-20 secs) as the sentence parser takes significantly longer as things get longer. 

[AntonOfTheWoods]

On 1/14/2022 at 7:14 PM, Jan Finster said:

Trying to test it with Deutsche Welle (https://www.dw.com/zh/武汉暴发疫情两周年-病毒起源陷入罗生门/a-60420872)

 

This is what I get, with a click on "Wuhan" to show the popup. I don't seem to be able to make the screenshots very big unfortunately.

[Jan Finster]

Ok, thanks. This worked: 

 

Yes, pretty cool. I guess there are pinyin converting apps, but your app lets you have pinyin or translation. It might be worthwhile to have the pinyin or translation in a different colour (e.g. blue).

 

At first glance it tempts me to read text above my level ?

 

Lets stay with Deutsche Welle (https://www.dw.com/zh/武汉暴发疫情两周年-病毒起源陷入罗生门/a-60420872), how does this integrate now into the transcrobe platform (https://am.transcrob.es/)? How does it know what words of the article I know? How does it evolve to suggest appropriate texts for me?

 

 

[Moshen]

Honestly, I do not see how this revolutionizes *learning*.  It's a tiny bit more convenient, that's all.  I would not use this as a learning tool, only as a translation tool in a pinch.

[AntonOfTheWoods]

On 1/14/2022 at 8:17 PM, Moshen said:

Honestly, I do not see how this revolutionizes *learning*.

I can post some more of the theoretical stuff if you are interested but it might not mean much. The thing is that you quickly get network effects between tools if they are properly connected, and when you get a lot of accurate learner models, you can start predicting exactly the right level of cognitive load to fully maximise the learning value of every single interaction with the system. When you get an individual's data it is interesting. When you get a classroom's data it is very interesting. When you get many thousands of learners' data, you can start drawing out patterns that allow for very significant learning optimisations.

 

Example - I don't think you know word X but I know you know the constituent characters of it in several other words and that the contribution of the characters to the meaning of the word is relatively regular. If I detect your level of cognitive load is low, I should probably not gloss and let you guess *this word* in this context. There is quite a bit of evidence this is great for learning.

 

Without any machine learning at all, the system can already prioritise the words in your spaced repetition by the number of times you have seen a word - the words you have seen most and still don't know are probably the highest value for you *right now*.

 

But you also have all the tools to decide your own "highest value" order of priority. Say you have HSK5 next month, that is more meaningful to you *right now*.

 

I have hundreds of ideas like this in the pipeline.

[AntonOfTheWoods]

On 1/14/2022 at 8:17 PM, Jan Finster said:

Lets stay with Deutsche Welle (https://www.dw.com/zh/武汉暴发疫情两周年-病毒起源陷入罗生门/a-60420872), how does this integrate now into the transcrobe platform (https://am.transcrob.es/)? How does it know what words of the article I know?

If you already have lists of words you know you can import them (https://am.transcrob.es/#/imports) and set them as "known".

 

The listrobes tool https://am.transcrob.es/#/listrobes allows you to select from the most common words (there are default wordlists or you can create your own) and set words (up to 250 at a time) as known/unknown. You can set whether the default is known or unknown, so when you click on the "save" button you are telling the system about up to 250 words at a time. There are mouseovers for quick validation if you have a doubt.

 

Actually I just saw the help page for listrobes is incomplete but the YouTube video I posted has a demo of how it works.

[AntonOfTheWoods]

On 1/14/2022 at 8:17 PM, Jan Finster said:

It might be worthwhile to have the pinyin or translation in a different colour (e.g. blue).

Thanks for the suggestion, I have added it to the list (https://github.com/transcrobes/transcrobes/issues/11) and will have a look at how to implement that tomorrow!

 

Another user suggested making the font size smaller for the glosses and that has been implemented (hasn't been deployed for the Chrome extension yet but is available in the book reader and video player). Making both configurable will require a bit more thought though maybe.

 

 

[Jan Finster]

On 1/14/2022 at 3:11 PM, AntonOfTheWoods said:

Thanks for the suggestion, I have added it to the list (https://github.com/transcrobes/transcrobes/issues/11) and will have a look at how to implement that tomorrow!

It may actually be even better to put the pinyin and/or translation above or below the text. Otherwise reading is very interrupted and the eyes has to jump all the time. Other apps also put the pinyin above the text.

[alantin]

On 1/14/2022 at 5:16 PM, Jan Finster said:

It may actually be even better to put the pinyin and/or translation above or below the text. Otherwise reading is very interrupted and the eyes has to jump all the time. Other apps also put the pinyin above the text.

 

This would be great, but the font should be configurable or big enough to see the diacritics. My personal problem is a bad vision and I can't make them out if the font is too small.
 

  

On 1/14/2022 at 2:17 PM, Jan Finster said:

It might be worthwhile to have the pinyin or translation in a different colour (e.g. blue).

Actually, I would love something that tracked my ability to get the pronunciation right for a word completely separately from it's meaning. So for example showing the pinyin on the words that I'm unsure of about the pronunciation and separately gloss the words that I don't know the meaning for. I think that would be something unique!

[alantin]

@AntonOfTheWoods, I would love to answer your survey, but I got stuck on the second question. What am I supposed to answer if my family couldn't care less?
There doesn't seem to be a "neutral" option...

 

 

2. My family hopes that one day I will speak Chinese fluently. *

 

strongly disagree

strongly agree

[alantin]

I got the account set up, the plugin working, and my scv word list imported, but now I've been waiting for an hour or more for it to mark them known. Is there a way for me to know when it is going to begin this?

 

In imports it went through fine

 

[AntonOfTheWoods]

On 1/14/2022 at 11:50 PM, alantin said:

Actually, I would love something that tracked my ability to get the pronunciation right for a word completely separately from it's meaning.

You are touching on something *super* interesting here, for both theoretical and practical reasons - what does "knowing a word" mean. Very little practical research goes much above known/unknown but there are many dimensions to knowing a word, including meaning-related stuff, sound-related stuff, context-related stuff, production vs reception, etc. 

 

You need to start somewhere. My goal is for the project to become an open research project, where scholars from all over the world can provide ideas and code that we can test on (willing) learners doing real learning. The stuff that works well becomes part of the project for all learners.

[AntonOfTheWoods]

On 1/15/2022 at 2:27 AM, alantin said:

There doesn't seem to be a "neutral" option...

That's weird. Almost all questions have 5 levels (Likert). Here is what I see on Chrome on Android.

 

 

Desktop should be even better. Are you using Chrome?