Interested in creating a course for Kabyle from French

[Bouaziz-aitd]

Hi everyone,
First of all, thank you for the committed people who are working on this project. This is really a huge work that's been deployed through the platform.
I was searching for such a platform since a while now trying to find a way how to build an online course for my native language Kabyle (a branch of the more general Tamazight language family). This is a native language from North Africa. I recently came across LibreLingo and found the project very interesting and I thought it would be a good platform for teaching Kabyle for French speaking people as North Africa is overly french speaking.
I had a quick read through all the LibreLingo documents, I went through GitHub repos and I started playing a bit with Git as I was new to this. I also read through a number of the discussions in the community space and a little about the issues discussed in the LibreLingo discussion group in Element and I am unsure if Kabyle can be rolled out on this platform for several reasons:

• Kabyle language writing relies on a few non Latin characters and I read that LL does not support those.
• Kabyle language does not benefit from any text to speech tools to make the audios.
  At this stage of my understanding, I am unsure if it would be possible to go around some of the hurdles and achieve the creation of an interesting course. Kabyle language is rapidly declining so it is in danger to die and that is why I am committed to do something through the LL platform.
  It should be noted that Kabyle is among the languages that are selected in the Common voice and Tatoeba databases. I though read some discussions about the difficulties to bring-in the information from these DBs and I do not know if any progress was made in this context.
  To make the long story short, my question is : Can someone tell me if I should engage in such endeavor and start working on the course (Kabyle from French) or it is not worth it? The least least that can be done is laying out a course without audios to start with and implementing any audios later on.
  In order to substantiate the issue of having non Latin letters these are examples of letters in use in some words: aɣrum: bread, aɛebbuḍ: abdomen, ečč: eat, awi-t-id: bring it here.
  Thank you for your consideration and help.

[kantord]

Hi @Bouaziz-aitd!

Can you please tell me more about this part?

Kabyle language writing relies on a few non Latin characters and I read that LL does not support those.

That might not be an issue if it's a few characters, as we have a feature to add a virtual keyboard to the user interface that contains special characters (for example this is used for adding accents in Spanish even when the user has an English keyboard with no accents). This should be enough to solve the problems if a language mainly relies on Latin characters but has a few added ones. But it's probably not enough when the language uses an entirely different writing system.

Edit, I see you added this concrete example:

In order to substantiate the issue of having non Latin letters these are examples of letters in use in some words: aɣrum: bread, aɛebbuḍ: abdomen, ečč: eat, awi-t-id: bring it here.

so that could be solved with the special characters feature, basically treat those as "accent characters" even when they are completely independent characters such as ɣ. That should not be a problem unless maybe the number of these special characters is very high.

All that being said, in theory a course can be created, but at the current moment there won't be additional features for teaching the writing system itself and there aren't any special feature currently built for this. The other problem is that a lot of users, especially when just starting a course, will not have a way to use 2 different keyboards for typing in 2 different languages, hence we have the "special keys" feature in Spanish.

Kabyle language does not benefit from any text to speech tools to make the audios.

This is a problem for many courses and there is a plan to solve it by relying on real audio recordings rather than TTS. This could actually have some advantages, albeit it can also be an added difficulty. While the software support is not ready for this yet, there is no reason to not start developing a course because of this, as there is a way to disable audio support and start testing the course without audio. That is not ideal, but it should allow you to start developing and even testing the course while the software support catches up

[Bouaziz-aitd]

Thank you Kantord for your response and clear guidance. It is really good news to know that we take a stab on the preparation of the course.
Yes the virtual keyboard will do the job. I saw that for Spanish course indeed. There are a few letters beyond the regular letter from a standard keyboard. here they are:
č ḍ ǧ ṛ ṣ ṭ ẓ ɛ ɣ

As far as audio is concerned, I think we could use the common voice dataset as it is based on recordings. I contributed to some extent and offered my voice to Common voice development for Kabyle (kab). I though have to admit that my technical knowledge of how fetch that information and I am afraid that I can only rely on competent people in the group.
I have read some of the exchanges you had in the group about the issue using the dataset since those are phrases. I did not catch very well where would be the issue. For you information the translation platform Glosbe has managed to get single words in the audio pronunciation. you may have a look at this where one word can be listened to. https://fr.glosbe.com/fr/kab/pain.
I think they use the Common voice dataset, so if we find out how they did this we can certainly do it for the our courses.
Sorry if I misunderstood the real issue with Common voice.

I have a few questions for you if you could help:

1. Should I start building the course in a word document like what I saw for Basque course? Is there a template I should use in order to anticipate for a seamless integration in LibreLingo platform?
2. I read through the LibreLingo documentation and I am still unsure about how all the example are introduced in the platform. Any guidance will be highly appreciated.
3. I assume the detailed technical discussions are done on the Element (Matrix) platform? so can we create a specific discussion thread for Kabyle? I intend to potentially bring in some additional help.

Thank you in advance for your guidance.

[kantord]

Thank you Kantord for your response and clear guidance. It is really good news to know that we take a stab on the preparation of the course. Yes the virtual keyboard will do the job. I saw that for Spanish course indeed. There are a few letters beyond the regular letter from a standard keyboard. here they are: č ḍ ǧ ṭ ẓ ɛ ɣ

As far as audio is concerned, I think we could use the common voice dataset as it is based on recordings. I contributed to some extent and offered my voice to Common voice development for Kabyle (kab). I though have to admit that my technical knowledge of how fetch that information and I am afraid that I can only rely on competent people in the group. I have read some of the exchanges you had in the group about the issue using the dataset since those are phrases. I did not catch very well where would be the issue. For you information the translation platform Glosbe has managed to get single words in the audio pronunciation. you may have a look at this where one word can be listened to. https://fr.glosbe.com/fr/kab/pain. I think they use the Common voice dataset, so if we find out how they did this we can certainly do it for the our courses. Sorry if I misunderstood the real issue with Common voice.

I have a few questions for you if you could help:

1. Should I start building the course in a word document like what I saw for Basque course? Is there a template I should use in order to anticipate for a seamless integration in LibreLingo platform?

2. I read through the LibreLingo documentation and I am still unsure about how all the example are introduced in the platform. Any guidance will be highly appreciated.

3. I assume the detailed technical discussions are done on the Element (Matrix) platform? so can we create a specific discussion thread for Kabyle? I intend to potentially bring in some additional help.

Thank you in advance for your guidance.

I think there isn't specifically a major problem with Common Voice, I think it's just about needing to build a system that can fetch and manage all that data. There is some work on this already, I mean, the TTS is built in a way that it could be fairly easily replaced with a different system. That different system might just need a way to automatically match the audio with sentences. As long as the audio source has both audio and text, this process can be automated. I prefer to automate it in order to make it easy to enforce consistency with the data and avoid errors such as the text not matching the audio.

By 2), all the examples, you mean all examples of words and phrases in general, or you mean, all examples for translating a word or phrase specifically? In addition to checking the documentation you might want to take a look at an existing course, for example the Spanish course https://github.com/LibreLingo/LibreLingo-ES-from-EN/tree/main/course

3. most technical discussions are usually done in GitHub issues, and Matrix is being mostly used for quick technical questions, quick help or discussion of ideas. If you want to bring in help (meaning, more people working in the Kabyle course for example) it could be a great option to create a channel for this in matrix and I could add that as part of the LibreLingo group in Matrix.

[kantord]

I guess as a next step, I could create a repo for your course in GitHub with some example content. Then you could edit the readme file to add links, such as a link to the matrix group and any information that you think other people collaborating to the course will need. Then you could for example use the sample content (taken from the spanish course) and change it to have the content in Kabyle. After that we can deploy that course to see if everything works as expected, and you get a link that you will be able to use as a minimalist demo you can show to other people who might be interested in contributing

[Bouaziz-aitd]

Thanks for your quick reply and appologies for not being quick at the begining. I am still trying to figure out all the mechanics of the structure. sorry if there are any stupid questions. I promise when I get it it will go smoothly and I will lead this to finish.
Are the YAML files all built by filling in the file structure in a text editor? Should I type all the text I have to include even if I have it in a separate document?

I guess as a next step, I could create a repo for your course in GitHub with some example content.....

Yes. I think that is the way to move forward.

[kantord]

sorry if there are any stupid questions.

I guess there is not stupid questions, these questions signal the weak points of the system or the documentation. I guess in this case it could be nice to solve them with adding some GUI or sth like that though 😅

Are the YAML files all built by filling in the file structure in a text editor?

Yes, for now they are built by manually editing these files in a text editor, or directly on GitHub. Actually, GitHub provides an interface to do this. And if you make a pull request, there are automatic checks made to find mistakes that could break the course. Of course the next step here is to have a GUI that automates this process and allows people to edit this data in a GUI interface, but that is not available at the moment

Should I type all the text I have to include even if I have it in a separate document?

I am not sure I understand this question well, could you please rephrase it or add more context?

[Bouaziz-aitd]

Understood! Your last question is answered in your previous response.

[kantord]

I have created a course repository: https://github.com/LibreLingo/LibreLingo-KAB-from-FR

and invited you to become a maintainer of this repository, @Bouaziz-aitd. If you accept the invite you should be able to manage this repository including inviting more members, approving merge requests and such

[Bouaziz-aitd]

Thank you. Let me have a look and get used to it. I will read again the documents to make sure I understand what to do efficiently. I assume at this stage there is nothing harmful to the system should make any mistakes.

[kantord]

yes, just by working on this repo you won't break anything, especially not anything outside if this course

[Bouaziz-aitd]

Hi kantord, I made an update of the Introduction module changing to kab from fr.
Before spending too much time working on the rest of files, it would be good if we had a chance to see the outcome on the web interface at this stage. Is there any way to make the result visible behind the scenes without showing it to the public? How can we do that? I need to work a bit more on mastering the structure of the YAML files such as the alternative translations and the mini-dictionnary. I am eager to know how all this comes together. Thanks

[kantord]

Hi kantord, I made an update of the Introduction module changing to kab from fr. Before spending too much time working on the rest of files, it would be good if we had a chance to see the outcome on the web interface at this stage. Is there any way to make the result visible behind the scenes without showing it to the public? How can we do that? I need to work a bit more on mastering the structure of the YAML files such as the alternative translations and the mini-dictionnary. I am eager to know how all this comes together. Thanks

Sure @Bouaziz-aitd! I will deploy this on the web interface.

I am trying to build a database of course maintainers and their contacts in order to be able to contact them quickly regarding their course. Currently it seems like email is the best form of contact for this. Is there an email address you would consent for being used for these purposes? Keep in mind that this database will be public and stored in the GitHub repo, and other people will be able to make copies of this database, and the email will be accessible just like the email you use for Git commits. If you don't like this idea or you would not want to participate in this, could you please provide some input at this issue? I'm also looking for alternatives: #2320

[kantord]

I noticed that you have a failing pipeline here: https://github.com/LibreLingo/LibreLingo-KAB-from-FR/actions/runs/4560209501/jobs/8044942542

note that such CI failures (for example, incorrect data in the yaml files, or syntax errors in the yaml files) prevent the course from being deployed as it is not possible to export the course with incorrect data. Hence, the deployment system will always automatically deploy the last version that doesn't have a failing CI. At the moment that is my last commit.

I'm working on enabling the deployment for your course, but since my last commit is basically example data from the Spanish course, it will mean that your changes won't be visible until you fix the failed pipeline. You can take a look at the output on of the GitHub actions checks to find out what exactly went wrong.

Btw, the tests to prevent such problems are executed on all merge requests as well, so if you protect your main branch in the GitHub repo, it will prevent such mistakes from being merged

[Bouaziz-aitd]

If you don't like this idea or you would not want to participate in this, could you please provide some input at this issue? I'm also looking for alternatives: #2320
@kantord, yes I am willing to help being a contact. Please keep in mind that I am new in this so I am not fully functional to be able to tackle the technical issues. Concerning the privacy of email, I read through #2320 and I am a bit concerned about getting my email address spamed. So I might create a new email address for this purpose. I have two ideas I see if they fit the purpose:

1. Maybe create a google account where can restrict the list of emails that can make to the inbox and filter the rest. I am assuming the emails would come from the selected internal collaborators of LibreLingo. I read that organizations that have gmail acounts can control the inbound and outbound email flow via permissions.
2. If these exchanges are only intended to be between you as the manager and the maintainers, you may consider create a private group in Matrix where messages can be posted. The notification from Matrix platform will serve as a messenger to everyone's inbox. I already have the notification switched on and I get any messages that are posted in there.

If these are not viable then you may use this email address: [email protected]

[Bouaziz-aitd]

@kantord I am lost in trying resolving the failures from the CI. As an example when I look at the output errors from the check of Preferences skill I see "keyError: 'Kabyle' "
So I am not sure what are the errors in the yaml files. I though have to inform that I changed in the mini-dictionary section Spanish for Kabyle and English for French. Are these keywords? I reverted back to the original and it does not solve the issue.
I get also a bunch of messages in the cmd screen that seem are related to the Python program running in the back. Your help will be appreciated.

[kantord]

ok, let me take a look if I can figure it out.

in the meantime this is the deployment, now it's working https://librelingo.app/course/kabyle-from-french of course, it shows the last passing version. btw the deployments are automatically done once every hour, so once you merge changes, they should take max 1h to automatically show on the web interface as well

[kantord]

ok, so I merged some fixes for the problems that were clear to me. there are some further problems that seems like could be related to a bug, or at least the error reporting is deficient and it's not easy to debug exactly what happened. I will work on fixing that but in the meantime in order to allow you to make some progress, I have commented out all of the skills and modules from the yaml files which effectively disables temporarily.

Now the next error that shows up is in the animals skills and it seems to be just related to the mini dictionary: https://github.com/LibreLingo/LibreLingo-KAB-from-FR/actions/runs/4568004879/jobs/8062518572

it seems like at least in this case the reported error makes some sense (I hope), please take a look if you could fix it.

[kantord]

once you fix the mistakes there, you can start enabling other modules or skills again and see what kind of errors (if any) happen in them

here's where you would remove "#" characters to enable modules:
https://github.com/LibreLingo/LibreLingo-KAB-from-FR/blob/56eb67647edca5b8ed6e61e5c387a9e7297c734d/course/course.yaml#L37-L40

and here's one example of where you would enable skills in a module:
https://github.com/LibreLingo/LibreLingo-KAB-from-FR/blob/main/course/basics/module.yaml

[Bouaziz-aitd]

Hi @kantord. I just navigated a bit through the course online and I understand that you deployed the original course in Spanish. Correct?
I understand that now when I fix the issues in a given module or skill it should update to the online view. Correct?
I will attempt to make small changes just to prove it works. Then I will tackle the work more consistently.
Thanks again.

[Bouaziz-aitd]

@kantord super good news! you may have seen that the course has updated and I was able to correct what was failing in the skill animals. I did not understand that the mini-dictionary requires to have words from the vocabulary used above it. I am still unsure if all the words cited must be listed.
you should note that the presentation text in the exercise is still in English. I don't know where it is stored in the skills or modules. I will for it but will easier if you told where it stands and will translate it.
Thanks for your help.

[Bouaziz-aitd]

I finished rewriting two other skills Food and Clothes but they failed validation.
Might be still the mini-dict is still causing trouble but I updated module.yaml removing the # comment but it failed also. This is a simple change I made. I suspect there is still something in the compiler that is failing.
In the meantime, can you please briefly explain what needs to be included in the mini-dict?

[kantord]

the purpose of the mini dictionary is to explain the meaning of new words. when you teach a new word in a phrase instead of teaching it explicitly as a word, the user will need to know it's meaning. They are able to find out the meaning by hovering over the word in the web interface. Because of this purpose, any word used in phrases that has not been taught elsewhere in the course needs to be included in the mini-dictionary and it will not pass validation otherwise, since it would result in a broken state for the user

[kantord]

Hi, the other problem was seemingly kinda a syntax error or schema error in a yaml file, I have pushed a fix to your repo so now you have a different error that is only about the mini dictionary.

For schema errors, there is a current proposed (and partially implemented) way of solving them: #2438

It seems like this is a syntax problem or rather, not a real syntax error but a typo that results in a different syntax tree than intended. I can try to improve the error reporting in that case, it could actually be solved by the JSON schema thing, but I guess the ultimate solution to avoid such problems will be to implement a GUI

[Bouaziz-aitd]

the purpose of the mini dictionary is to explain the meaning of new words. when you teach a new word in a phrase instead of teaching it explicitly as a word, the user will need to know it's meaning. They are able to find out the meaning by hovering over the word in the web interface. Because of this purpose, any word used in phrases that has not been taught elsewhere in the course needs to be included in the mini-dictionary and it will not pass validation otherwise, since it would result in a broken state for the user

I assume multiple words (small sentence) are not allowed then in one item, such as for ex:

• Por favor: S'il vous plait
  I will make a few tries and see what happens.
  Thank you

[Bouaziz-aitd]

Hi @kantord, I managed to make skill "Clothes" to work as I see it in the web app. You may notice that I made lots of try and error until I solved the failure. I thought I understood the structure but I think there is still some nuances that I do not necessarily understand very well.
I worked skill "Food" but I still get an error that seems to be related to some string fromatting but the error message is not clear.
In addition to that, I released the skill in the module.yaml but it throws at me a few errors. hopefully this is because the "Food" skill is not working yet.
Can you please have a quick look at the Food verification errors? thanks

[Bouaziz-aitd]

Hi @kantord, I spent a good deal of time resolving the failed skills. I read again the LibreLingo (LL) documentation and made plenty of tests step by step to figure out how all this works. I think the explanation documents need to be a bit more specific to better guide and avoid mistakes that would take a lot of time to resolve. It is clear that the fix format required by the YAML files is a big constraint. This brings me to the early times when I was dealing with fix formats in Frotran IV with very little explanation in the error messages. This being said, I managed to resolve other skills and I hope that now I am better equipped to move forward the other skills of the course.
I though have a few questions if you kindly help answering them. In some cases words such as "l'eau" are not accepted in the mini-dictionary. I followed the indications in the LibreLingo documentation putting those in "{}" and it looks like it solved the issue in the most of cases but in some others I got the quotes "" printed in the exercise in the web app. Can you please confirm that the format for sentences is "{}"? What could be the issue behind those quotes I seen in the final product?
I would like to better structure the kab from fr course before I move too deep in the development. As you know every language has its specifics and some examples may be pertinent for one language and not for another. So I would like to tailor the course to kab through some additional examples. I am wondering if have any listing of the images contained in the LL database. I have seen the list of the files but it would be good to get any explanatory listing if you have it.
BTW, one thing I noticed when working with the commits for resolution is that : someone should always work on resolving the skill before you change the module yaml. I had always had to make a slight change even when it is not required in the module.yaml and reload in order that the new changes take place. Reloading and old version without change even if the syntax is correct will not take place in the final result on the web app. Maybe it is included somewhere in the documentation but if not, I think it is important to include it somewhere in the LL docs.
I think soon I will be needing to include the audio. Should we load the mp3 files onto the database? I will need to understand how to do that. Should we import the Common voice files? How we do that?
At this time, I understand that the course dictionary is populated through the skills. Is there any documentation on the structure of the this dictionary? I would like to understand how it is built in order that I work on the skills more effectively to built the database that a learner needs to integrate through the learning process. Is the use of past learned vocabulary used randomly in the course? Any guidance to understand the flow structure?
One thing I have noticed through the use of the web app is that when I wanted to cancel my learning session, LL platform does not bring me to the main module's menu. It pushes me to the test page. I don't know if this is done intentionally?
One last thing, I have mentioned earlier is about the the English text that is used in the questions on the web app. Where should we switch that to French?

Sorry about this long email and I always appreciate your guidance and huge effort you deploy to making this project happen.

[kantord]

I read again the LibreLingo (LL) documentation and made plenty of tests step by step to figure out how all this works. I think the explanation documents need to be a bit more specific to better guide and avoid mistakes that would take a lot of time to resolve. It is clear that the fix format required by the YAML files is a big constraint. This brings me to the early times when I was dealing with fix formats in Frotran IV with very little explanation in the error messages.

Thanks for the feedback on this. It is a very good observation that the format seems to be a big constraint, and also that the error messages are deficient. Like I said, the current path I see to improve this is adding a GUI and better error reporting. Unfortunately the constraints are related to actual data constraints, especially when it comes to the mini dictionary. So I think the only way to improve this in the data structure/schema itself would be to somehow make the constraints more intuitive, basically like making the correct data be the only possible data? 🤔 Not sure, do you have some kind of additional suggestion to this?

Also, regarding improving the documentation, could you please help me with it? I have been trying to improve the documentation, however I think I have a lot of blinds spots, basically since I designed the system I am struggling to approach it with the same perspective as a person with no prior knowledge would, and inherently assume that some of the things that are not intuitive at all should make sense to other people. If you could either edit some of the documentation files, or what I should write there, that would be an amazing help

[kantord]

so, just by putting the curly braces, the quotes magically appear?

[Bouaziz-aitd]

Apologies, this is my mistake. I reviewed and I realize that I included in the phrase : "{l'eau}" without declaring it in the Mini-dict. This is normal as it is considered as part of the text and is displayed as is. The strange thing is that it was not flagged by the compiler as a non-defined word?!?.
I also realize that when you put the "{}" on the right hand side of the ':' in the dictionary it will take it as the definition of the left hand side of the word that is defined. No error is prompted then. The ex below works fine:

• infirmières: "{timaragin n umsuji}"

The sentence is write without any curl braces like this:

• Phrase: Timaragin n umsuji zedɣent dagi
  Translation: Les infirmières habitent ici

[kantord]

The strange thing is that it was not flagged by the compiler as a non-defined word?!?

yes, that is indeed strange

[kantord]

I also realize that when you put the "{}" on the right hand side of the ':' in the dictionary it will take it as the definition of the left hand side of the word that is defined. No error is prompted then. The ex below works fine:

yes I think here it's important to say that the definition is either a single string, or a list of strings, and they are displayed to the user without any manipulation

[Bouaziz-aitd]

I agree!

[kantord]

I am wondering if have any listing of the images contained in the LL database. I have seen the list of the files but it would be good to get any explanatory listing if you have it.

yes, there is a list: https://github.com/LibreLingo/LibreLingo/blob/main/docs/image_attributions.csv

though long term the idea here is to be able to link the unsplash files directly. at the moment it's not possible but it's a known issue

[kantord]

BTW, one thing I noticed when working with the commits for resolution is that : someone should always work on resolving the skill before you change the module yaml. I had always had to make a slight change even when it is not required in the module.yaml and reload in order that the new changes take place. Reloading and old version without change even if the syntax is correct will not take place in the final result on the web app. Maybe it is included somewhere in the documentation but if not, I think it is important to include it somewhere in the LL docs.

I am not sure I fully understood this part. Is it about the changes not being deployed? In theory, the deployment happens once an hour, so there is a potentially one hour delay in the deployment (unless of course, something goes wrong for example). This is because the whole site is statically built, it's not based on a "live backend" or database. (Though there is the idea of building each course separately, or even build just one specific page after a change, or even to have a live backend in addition to the static build to reduce such delays to trivially low)

this should indeed be documented.

btw if you want to keep track of exactly when the last deployment has happened, you can always look at this page: https://github.com/LibreLingo/LibreLingo/actions

[kantord]

there is not really a LibreLingo database and manually adding files could be tricky. it would require manually adding the audios to the course, and then also adding them to a metadata file. I mean, anytime the data is changed, you would automatically be forced to also change the audio and such, and would receive no warnings for these problems and such 🤔 but I guess, it's technically possible with many many asterisks, and it might almost be easier to just do it properly and source the audios from tatoeba for example.

Another advantage of that approach is that you could start by contributing your audio files to tatoeba, which would mean that people who need these audio files can already use them, even though they are not in librelingo itself yet. Then once we implement connecting your course with the tatoeba audio, it would quickly appear in librelingo itlsef

[Bouaziz-aitd]

I understand the dependency on having files versed to some LibreLingo folder. As I explained in the comment from last night, I downloaded the common voice single word data set and I found that only a few words (literally numbers from 0 to 9) have been included with many voice types.
I was trying to find a patch in the interim as the Tatoeba solution might take a long time before it comes to fruition (sorry I am not a champion in this field to work on it; it is just a matter of competence :(. ) Since these courses at this level do not need a huge dataset, I thought we can implement a temporary solution and get it replaced transparently when the Tatoeba solution is up and running. Maybe, I am just looking through my pink glasses.
One thing certain is that my course is not top notch without the voice because the pronunciation is key.
Any thoughts??

[kantord]

Maybe there is a way to have an intermediate solution, although I think that even the intermediate solution would need some kind of technical help, so I 'm not sure exactly how much time could be saved. We could look into this in more detail.

it is important to fix the audios as it is kinda stopping your end users from being able to access the course, unless they have at least a minimal pre-existing knowledge of Kabyle pronunciation. On the other hand at least while the audios are not there, you can change any of the phrases or words without having to re-record audios, so maybe you can take it as a period of trying to make some parts of the course stable before recording the audios 😅 Also maybe you can focus on the parts of the course that are useful to somehow who already knows basic pronunciation. That could be for example a person who grew up hearing some Kabyle but cannot speak it as it was not their main language. Those learners might be able to start playing with your course even before the audio system is working.

I hope that helps a bit, I wish I could just tell you all of the audio system will just work by tomorrow but that's not realistic 😅

[Bouaziz-aitd]

I understand the it is not easy but I am sure there should be an interim solution. I started talking around me if anyone has any smart idea. One of my IT relatives informed me that Glosbe has an API and she will have a quick look when she has time.

[Bouaziz-aitd]

@kantord as I informed above an IT person has taken a look at the question of common voice and she came to the same idea you had pushing a list of words/sentences into common voice and recuperate them from there using API. I invited her to come into this discussion thread in order that you maybe have to interject and comment on the route take (it will be more efficient). We will see if she will join or just help from the backstage. She pointed me out a repo from common voice on GitHub here: https://github.com/common-voice/common-voice/tree/main/server/data/kab.
If this approach suits LibreLingo programing structure then she will investigate on how to post a dictionary into Common voice and how the API can be used to fetch the voices.
Does this make sense?

[kantord]

At this time, I understand that the course dictionary is populated through the skills. Is there any documentation on the structure of the this dictionary? I would like to understand how it is built in order that I work on the skills more effectively to built the database that a learner needs to integrate through the learning process. Is the use of past learned vocabulary used randomly in the course? Any guidance to understand the flow structure?

If I remember correctly, the only documentation we have regarding the mini-dictionary it's located here: https://librelingo.app/docs/courses/skill.html basically there is an explanation regarding that the mini dictionary is for

[Bouaziz-aitd]

That's correct. Although I think we should add some specifics to better explain how this is saved in the global dictionary.
I think this is key since it helps the course developer to optimize the entries in the mini-dict. As a lot of words will come back in the long run, knowing which ones are already taught in the previous skills will save me time in not repeating them again. I hope this makes sense?

[kantord]

One thing I have noticed through the use of the web app is that when I wanted to cancel my learning session, LL platform does not bring me to the main module's menu. It pushes me to the test page. I don't know if this is done intentionally?

No, this is for sure a bug or a configuration issue 😮 could you please report it? https://github.com/LibreLingo/LibreLingo/issues/new

[Bouaziz-aitd]

This is the screen I get after I cancel the learning session:

[kantord]

ok, this one should be fixed now

[Bouaziz-aitd]

Yes it works now.

[kantord]

One last thing, I have mentioned earlier is about the the English text that is used in the questions on the web app. Where should we switch that to French?

So actually, this is kinda a new feature, basically we need to translate all pages. Although, the home page is already translated, so there is an infrastructure built for this, so all we should need to do is add a configuration somewhere that will force the UI to use the translation to be in French for this course, plus make sure all relevant text has it's own translation key. I will do at least an initial MR for this when I have a moment. For now I created an issue for it: #2796

[Bouaziz-aitd]

If you would like me to work on the French translation. I can help.

[kantord]

Sorry about this long email and I always appreciate your guidance and huge effort you deploy to making this project happen.

Don't worry, all your questions and feedback are extremely useful and I also thank you for your contribution by creating a new course and testing this software that is admittedly not the most stable and optimized piece of software yet.

If you'd like to help, it would be also extremely useful if you helped by editing/extending existing documentation, since time has proved that I'm not able to write the best documentation as I don't have the same perspective as the users have. The documentation files are located here: https://github.com/LibreLingo/LibreLingo/tree/main/docs

[Bouaziz-aitd]

@kantord, here are some screen shots of an error I am not able to understand. You can see the quoted sentence that I believe is correctly written in the sentence along with its declaration in the mini-dict. Note that it is also specified in the French part of the mini-dict. You may see the error is pointing to line 125 which is the Phrase.
Maybe you can see anything I do not see that causes the issue?

[MignonBelongie]

Hello! I'm just trying to figure this stuff out myself, so I did a little searching through various yaml files. I think that based on https://github.com/LibreLingo/LibreLingo/blob/main/apps/tools/tests/fixtures/e2/basics/skills/hello.yaml
that line 125 should be:
- Phrase: "{tizwal uṛumi} d tizeggaɣin"
and line 184 should be:
- tizwal uṛumi: fraises
Unfortunately, I'm not set up yet to test this myself, but I thought it might be useful to bring this file to your attention.

(To avoid confusion, I should point out that this file does actually have a deliberate error, namely the undefined word "uncompersick". See also https://github.com/LibreLingo/LibreLingo/blob/main/apps/tools/tests/fixtures/e1/basics/skills/hello.yaml, which doesn't have that error, but has the (deliberate) error that "Excuse me" is not defined.)

[Bouaziz-aitd]

Thank you @MignonBelongie for you input. I actually managed to go around it and decided to use a hyphen to link the two words and make it one word. I though used the the "{}" in the Mini dictionary like I described it and it worked in other instance. My suspicion is that there is bug somewhere in the programming.
I just used your suggestion and I got the below error message suggesting that the composed word in the Min-dict at line 184 was not recognized.

[Bouaziz-aitd]

Hi All, I made an extensive test to better understand how things work in the Mini-dict and I came to the conclusion that when you add a sentence as an explanation you may omit the curls and quotes. The sentence (string of words) will be considered during the exercises and will be displayed. One should though define every word in the sentence individually. My example below:

• please: s'il vous plait
  in the case please will show the sentence as a hint "s'il vous plait". To allow this to work "s'il", "vous", and "plait" should be defined individually.
  Alternatively if you want the whole sentence to be defined as one word you should then use the curls and quotes like defined in LibreLingo documentation.
  I hope this helps.

[kantord]

ok I think I understand now, I mean, I guess the quotes are required simply to make it clear in YAML syntax that it's a string and not an object, because "{}" can also mean an object.

I think this could be a little bit of a problem, maybe we should consider using something other than {} for the same purpose in order to avoid having to put the entire phrase into quotes because of it, since this is not intuitive

(although this is another problem that would be solved if we had a GUI, since then it would be automatically serialized as it should)

[Bouaziz-aitd]

@kantord I managed to complete my course based on the template you provided from "ES from EN". I did and adaptation of the skills as Kabyle does not have the same structure as Spanish when it comes to "Estar and Ser" verbs among others.
I am still missing the voice which is key in learning the language especially for the sonority of some of the letter that are not common. I downloaded the single work package from Common Voice and I realized that the only words that are included are about YES, NO and numbers 0 through 9. I am still wondering how Glosbe managed to parse the sentences to single words since they appear on their online dictionary.
I noticed that there is a set of MP3 files here: https://github.com/LibreLingo/LibreLingo/tree/main/apps/web/static/voice but I am not sure how they are called upon from the web app. I went through the Spanish course but I was not able to detect the sequence the pronounced phrases are chosen,
If you tell me how to identify the phrases and words to be selected in the skills, I will manage to record those and add them to LibreLingo voice files set.

When I was testing the Kab course I noticed that when you go through a skill it is marked as complete even if you go over it another time. In my opinion, we should make the skill appear in progress when you over it again. This will mark it as kind of under study, In such a way that the user can engage in multiple skills in parallel and the would know where they stand in the progress. I am not sure if this is clear?

Hopefully will be able to make the voice work so I am planning to add junior section tailored to kids. That section should base the learning very much on pictures and voice through short words or sentences. I would like for example to introduce the alphabet with words accompanied with voice. I did not yet make my head around this but I am still thinking about the best way to include it. Since you know better than me the structure behind LibreLingo platform, I turn it to you hoping to get an option for this, please!

I will be sharing the app among a limited group for now just to get some feedback and make corrections if any.

Lastly, you may have noticed, I edited the the skill documentation and slightly amended a bit the text related the above "{}" question. I am not sure those additions are acceptable.

Thanks for your help!

[kantord]

I did and adaptation of the skills as Kabyle does not have the same structure as Spanish when it comes to "Estar and Ser" verbs among others.

That is ok, feel free to completely restructure the course. The structure of the course, especially in terms of the grammar has to be adapted to the languages, both the language that your learners speak (French in this case) and also the language you are teaching. Basically the things to consider is, how is this language typically taught (if there are some typical structures available, if not you can take for example wikipedia's grammar explanation as inspiration https://en.wikipedia.org/wiki/Kabyle_grammar). Another thing to consider in structuring your course is useful/important vocabulary that your learners can use in a lot of sentences in diverse ways. For that, word frequency lists should be helpful, a list of the most commonly used 1000 words for example. If you cannot find that, it can also be an option to use a list for French, and translate the words, which is not the perfect approach but it could be a starting point.

I am still wondering how Glosbe managed to parse the sentences to single words since they appear on their online dictionary.

it could be worth asking them, maybe they have a different source

I noticed that there is a set of MP3 files here: https://github.com/LibreLingo/LibreLingo/tree/main/apps/web/static/voice but I am not sure how they are called upon from the web app. I went through the Spanish course but I was not able to detect the sequence the pronounced phrases are chosen,

If you tell me how to identify the phrases and words to be selected in the skills, I will manage to record those and add them to LibreLingo voice files set.

Well what is happening with the audios is that right now they are fetched from a TTS engine (Amazon Polly) and the connection between the audio and the exercises is established automatically using a hash. This process is not very ideal for adding audio files manually and I'm not sure that it could ever be ideal to maintain the audio files so manually, like manually uploading them and associating them with the exercises, as that would lead to frequent errors. So this process has to be automated for sure, at least I don't have an idea how to make it completely manual while also avoiding frequent errors. When I say frequent errors I mean things like associating the wrong audio file with the wrong phrase when a phrase or word is edited after the fact.

When I was testing the Kab course I noticed that when you go through a skill it is marked as complete even if you go over it another time. In my opinion, we should make the skill appear in progress when you over it again. This will mark it as kind of under study, In such a way that the user can engage in multiple skills in parallel and the would know where they stand in the progress. I am not sure if this is clear?

It's not really clear to me, could you show me some examples with screenshots? Or I'm not sure what it means to "go over it again" 🤔

[kantord]

regarding your changes in the documentation, I have added some comments in the pull request

[Bouaziz-aitd]

Thanks for the information:

• Grammar and language learning. I have plenty of reference documentation and certainly Wikipedia seems to be of good quality information.
• I think Glosbe seems to offer free API that allows developers to access their database of multilingual translations, including audio pronunciations for many words. We need to figure out how the voice can be used. To be continued...
• Concerning the the issue (a nice to have, I would say) related to the app I am not able to replicate that because I went through all the module at least one time. This is not urgent.

[Bouaziz-aitd]

@kantord, I came accross a very strange case with the word "l'eau". I have two distinct sentences in the skill that use the composed word "l'eau" (see the attached snapshot).
"eau" is defined in the Mini-dict in both languages as follows:

• eau: aman
• aman: eau

From the attached sentences work fine with the above definitions.
The verry strange thing is that when I add the "l'" for "eau" in the sentence on line 77 to get "l'eau" the compiler throws an error saying that the "l" is not defined in the dictionary.

Any thoughts?

[kantord]

hmmm, that is weird, I think that the error should be that "l'eau" is not in the dictionary 🤔 I think this must be a bug, let's create a new issue for it, and I will try to fix it by first adding a test case to make sure I can reproduce it

[Bouaziz-aitd]

@kantord, I am at the point where I will distribute the link to the course for select people for feedback and any corrections.
I have added the welcome text below in the main course .md file and in the basics entry .md file. Please review and let me know if you are ok with the credits reference.

Bienvenue sur LibreLingo, la plateforme gratuite pour l'apprentissage des langues! Ansuf yis-wen ɣer LibreLingo; allal n ulmad n tutlayt Taqbaylit!

Cours de Kabyle destinés aux francophones. Ce cours est bâti sur l'apprentissage à répétition de notions de vocabulaire et d'expressions en usage courants par les locuteurs de la langue.

Les notions grammaticales sont développées à travers les exercices présentés dans le cours.

Le cours contient plusieurs modules lesquels contiennent des compétences (Skills) dont le contenu est construit de manière à permettre un apprentissage progressif.

Il est hautement suggéré de reprendre les compétences plusieures fois jusqu'à bien assimiler les notions enseignées.

Le cours est aussi destiné aux locuteurs de langue kabyle qui n'ont pas eu l'opportunité de pratiquer l'écrit de la langue. Ce cours est une opportunité de s'initier à l'écriture du Kabyle en utilisant la transcription en caractère latins.

Dda Lmulud At Mammer yenna-d: "Win iran Taqbaylit yaru-tt!"

La plateforme LibreLingon est développée par Daniel Kantor et al. Le cours de Kabyle est adapté et monté par Bouaziz Ait Driss et al.

Un grand merci à tous les collaborateurs et collaboratrices. Pour tous commentaires veuillez contacter l'équipe de développement à : [email protected]

[kantord]

of course, I think this is fine, but I would also include a link to the original repo, where the whole project can be seen including the full list of contributors to the main project. https://github.com/LibreLingo/LibreLingo

also maybe it's a good idea to include a link to the parts of the documentation that you personally found useful (if any) so that contributors have a place to start with) https://librelingo.app/docs/courses/index.html

[Bouaziz-aitd]

Added, thank you for the input.

[Bouaziz-aitd]

@kantord, I would like to move forward with enriching the kab course by adding skills (mainly basics for junior or kids) and modules. The basics_jr will be mainly focused on pictures recognition with more animals, objects and simple actions. I am not sure if I should just add files for skills in the folder directly in the repository? And I do not know how to create a new module.
Your help will be highly appreciated.

[Bouaziz-aitd]

@kantord, I generated an ID from UUID and included it in the Test_JR skill. I still do not see the inclusion of the new skill despite a successful compilation and push.
I think this will need your intervention please. Can you help?
Thanks

[Bouaziz-aitd]

Hi @kantord, I went through the actions and I realize that the job "Deployment to Production" is failing since last week. I assume this is the action that updates the course on the web app?
The last successful action is 1166 from last week. I suspect that is why I do not see any changes on the Kab course despite the additions and changes I was making.
Can you please have a quick look to figure out the issue?
Thank you

[Bouaziz-aitd]

@kantord, this is to let you know that I figured out that the issue came from the fact that the new folder is not recognized. I removed it from the course.yaml and included a new skill in an existing module and it worked. Now I am trying to create a new module folder with a name not containing any cap letters as I suspect it might the issue. I will let you know if this is indeed the problem.

[Bouaziz-aitd]

My attempt to creating a module folder has failed. Creating a new skill in an existing module is finally working but I failed in creating a new module with a new skill. I checked the failure message and it seems the path to the skill file is not recognized (see below). I will appreciate if you gave me any tip to solving this issue.

[Bouaziz-aitd]

Hi Daniel,
I don't know if you did anything from your side but now it works. The new folder is showing up now on the web app.
Regards.

[Bouaziz-aitd]

@kantord is there a way to see the trafic on a given course? I am anticipating on the moment I will share the course more widely.

[Bouaziz-aitd]

I was told that using google analytics can help monitor trafic on a static website without collecting any personal data. We still need to announce that the web site is measuring trafic.
An alternative might be to integrate a simple counter in the app program if possible where any user may hit a button to start like pushing for a token while this action could be caught by a counter and saved somewhere.
Does this make sense?

[kantord]

I was told that using google analytics can help monitor trafic on a static website without collecting any personal data. We still need to announce that the web site is measuring trafic.

Do you have some resources about this? Would it be GDPR compliant? What I have read that even making requests to a different IP address can be a problem, as IP addresses count as personal data by GDPR 🤔 For example even Google Fonts was deemed illegal https://meetanshi.com/blog/google-fonts-gdpr-compliant/ 🤔

An alternative might be to integrate a simple counter in the app program if possible where any user may hit a button to start like pushing for a token while this action could be caught by a counter and saved somewhere.

well, this would need to be stored somewhere though, so in the end we still need some kind of software for this.

[RymRabehi]

Hi @kantord and @Bouaziz-aitd

I was told that using google analytics can help monitor trafic on a static website without collecting any personal data. We still need to announce that the web site is measuring trafic.

Do you have some resources about this? Would it be GDPR compliant? What I have read that even making requests to a different IP address can be a problem, as IP addresses count as personal data by GDPR thinking For example even Google Fonts was deemed illegal https://meetanshi.com/blog/google-fonts-gdpr-compliant/ thinking

It's possible, but we will need to anonymize the data through a proxy. Here is an article about this https://www.cnil.fr/en/google-analytics-and-data-transfers-how-make-your-analytics-tool-compliant-gdpr

An alternative might be to integrate a simple counter in the app program if possible where any user may hit a button to start like pushing for a token while this action could be caught by a counter and saved somewhere.

well, this would need to be stored somewhere though, so in the end we still need some kind of software for this.

@kantord may I ask you how is hosted https://librelingo.app/ ?

An alternative to Google Analytics is Matmoto Analytics
It's a french tool approved by the CNIL (National Commission on Informatics and Liberty), the data privacy regulator in France.
The code is open-source https://github.com/matomo-org/, it's free if self-hosted (it will obviously cost at least a server and some time to maintain), the cloud solution cost 19€/month.

There is also another solution which is GoAccess if you use an ngnix webserver.
It's also open-source and IP addresses can be anonymized using --anonymize-level option allinurl/goaccess#2282

++
Rym R.

[Bouaziz-aitd]

Thank you @RymRabehi for the valuable information.
I don't know if I oversimplify things here but knowing that our objective is only about computing the trafic through a course then we could only advertise that we do not keep any user data and flush any collected information but keep the counter on for cumulative counts.
Does this makes sense?

[kantord]

@kantord may I ask you how is hosted https://librelingo.app/ ?

It is at the moment hosted on GitHub pages. But it would be nice to migrate to a similar static site provider that makes GDPR compliance very clear

[Bouaziz-aitd]

Hi Daniel,
I am now up an running with my course development. I still have a few work to do but I anticipate some of the steps ahead. I did pictures for learning the the alphabet and I intend to post them on unsplash platform. Did you have any contingency for adding pictures in the list LibreLingo is using? Can I then add the files into https://github.com/LibreLingo/LibreLingo/tree/main/apps/web/static/images ?
This will take a couple of week since I have a restriction of 10 images to be added to unsplash.
I am coming back again (sorry about that) on the voice. In the interim, I thought I can point the user to some media such as a short video on Youtube for example where I can teach the sounds for the alphabet. So my question is : is there a way to include a live link to a video residing outside LibreLingo ? What do you think?
Thank you

[kantord]

I am now up an running with my course development. I still have a few work to do but I anticipate some of the steps ahead. I did pictures for learning the the alphabet and I intend to post them on unsplash platform. Did you have any contingency for adding pictures in the list LibreLingo is using? Can I then add the files into https://github.com/LibreLingo/LibreLingo/tree/main/apps/web/static/images ?

Are pictures on unsplash needed for learning the alphabet? I think that if there are exercises needed for learning the alphabet it should be a separate type of challenge, a new type of thing added to the YAML file format. The current way pictures are added is specifically optimized for learning words, and the plan is to keep it optimized that way. I think that there should be a separate "Script" or "Alphabet" or "Characters" section in the YAML files then.

In any case, the way pictures can be added to the static folder is indeed if you upload the pictures to unsplash (or choose an existing one on the same site). Then you can link them in this file: https://github.com/LibreLingo/LibreLingo/blob/main/docs/image_attributions.csv and I can run a script to automatically fetch the images in the repo which makes them available (automatically generates the images in the static folder)

[RymRabehi]

Hi @kantord

I took a look to the audio feature and how it can be implemented.

For common voice, the api seems not available. My guess is that we need to download the dataset, store it somewhere (s3 bucket, drive,...) and then embed each url in the code.
Some websites, using common voice that I found:
https://datasets-server.huggingface.co/assets/common_voice/--/kab/train/0/audio/audio.mp3
https://glosbe.com/fb_aud/mp3/Im2765885_audio-by-166701.mp3

Also, while doing some research on this I found this issue #714 where a discussion about this topic started.
The idea of having a static url in the yaml file sounds great to me. It will be easy to have diverse audio sources?
What do you think about it?

++
Rym R.

[Bouaziz-aitd]

Interesting to see that Glosbe has an easy way to reading the audio.
Tatoeba can be read easily as well: https://audio.tatoeba.org/sentences/kab/6718401.mp3

[kantord]

Keep in mind that there is the danger of linking the wrong audio file. Imagine that you link an audio file, and then change the sentence itself. Now the wrong pronunciation will be showed in the app which will be basically a bug and major blocker for the users.

However with platforms like tatoeba that also have the text transcription for the audio, it's possible to avoid that issue as there can be an automated check that does not allow you to link the wrong audio to a sentence. Another problem is licensing which also needs to be verified as a lot of users might upload "stolen" audio. So I think that diverse audio sources is a good idea, but still it needs to be limited to platforms that respect copyright and have transcription, and specifically allow this usage.

[kantord]

And regarding common voice I think that it would be basically required to set up some kind of API based on common voice. I think that common voice meets all the criteria in terms of licensing and having transcriptions, but setting up an API to access the audio files in common voice I think it's a separate bigger project and I think that this should not be part of the LibreLingo project, rather, someone should create a solution for this and we should use that solution in LibreLingo, as I think that the magnitude of that task exceeds the complexity of what we can maintain in LibreLingo itself.

If you know someone who is currently working on such a service or already providing it, or has some open source tool that can do it, let me know!

[Bouaziz-aitd]

Hi Daniel,
I hope you are doing well. Despite my absence, I did not stop scratching my head to find any feasible solution for the audio issue. I finally got two options I tested on a standalone basis in python environment, I would like to propose.

In summary I tested two options i) using Tatoeba database and ii) using the MMS TTS model (as you may know this is a Meta initiative).

i) Tatoeba dataset
I worked with Tatoeba team and started contributing in creating a set of sentences that may be of interest to my course. It is important to understand (and I did not know that before I contributed) that it is not permitted to upload single word sentences unless the one word sentence makes sense such as "Hello!". Any sentence needs to have punctuation. In any way, sentences can be saved there and work with them through and API call from python. I did several tests on existing sentences that have audio included and my own sentences and it works perfect! It is important to register the contributions with a descent to allow Libre Lingo to use those without any breach of the license. Since I am working for promoting my language so I decided to include those free of any license so we can use them without any issue.
In my testing I tried to go around the single word aspect. I wrote a python script that parse the words from an audio sentence. The result is not perfect and not consistent because it depends a lot on the silence length between the words. To ease a bit the work, I tried to record sentences with the first word followed with a little silence. This may work but it is extremely important the the narrator puts the silence long enough to parse it with success. The downside of it is that sometimes it is difficult to have a natural audio sentence with a descent silences after the first word. I started thinking about having kind of exclamation tone after the first to mark the gap with the following words of the audio sentence. This can be worked out latter if we decided to go this way. The easiest alternative is to avoid one word dictation in Libre Lingo at least for the beginning, until we proof test the solution.
The advantage of using Tatoeba would be good because the audio is natural (human dictation).
In anticipation I started mounting a register of those sentences with their registration ID in the dataset. ID number is used to read the audio form Tatoeba. For Libre Lingo implementation, I was thinking of a mapping matrix where the sentences are for a given course are included in an excel file with the corresponding IDs. At the time the user hits the speak button, the corresponding sentence is searched in the list and the corresponding ID picked up used to call Tatoeba. Maybe you have better ideas but at least we are safe having an option here.

ii) Massively Multilingual Speech (MMS TTS) of Meta. This project has targeted more than 4000 languages around the world. they make a good deal of work for some languages but not all. At this time they have trained with the equivalent of 32 hours of data more that a thousand language among them Kabyle. The latter has been trained for text to speech conversion and for speech recognition. I focused on the TTS. They have a repo on Github here: https://github.com/facebookresearch/fairseq/blob/main/examples/mms/tts/tutorial/MMS_TTS_Inference_Colab.ipynb
I was curious and I looked at some of the languages that are in Libre Lingo such as Occitan and Basque. The former is not trained yet but the latter is.
I tried first the TTS on Google Colab for kab and the model produced descent audio in wav format ( I copied a script that has been posted for all languages that I found shared on the web).
I wanted to proof the concept by including this model in my python script outside Colab and I managed to make it work. At this time I can write a sentence and let the program create the wav file that I can read. The result is not perfect but it works in general with good precision at about 95% of the cases. i noticed that the composed sentences perform better that single words.
It should be noted that this solution requires to download the model which is a zipped file that should be unzipped to some location but it is done one time when the model is installed.

In conclusion, we have at least two options we work around. I did what I could till here and I do not know how we can move this to bring it to fruition.

What do you think?
Please tell if I should duplicate this comment in the open issue for the audios #714?
Sorry about the long comment. Many thanks

[Bouaziz-aitd]

Hi Daniel,
Any thoughts about the above?
Have a great day.

[Bouaziz-aitd]

Hi Daniel,
I do not want to appear too pushy but I am wondering if you had a chance to read my messages?