[Viki Community] We Want Your Feedback

If you have the document of the last talk with Viki, take a look at the 1st pages. You’d be surprised.

Yes, I have it. “Viki Community Call 09.27.2019”, right?
I just read again the first two pages. They actually had said that they’re thinking of machine translation for inactive projects (1 month for library projects, 1 week for on-air) and that they would advise well in advance the moderators so that they know.
They also said what they said now, that they won’t over-write our subs etc.
So there was nothing that surprised me. What were you referring to? Maybe I overlooked it, or was it on the third page?
I have to rush to work right now, I’ll be back this evening to see your reply.

1st page, yup. I don’t remember we talked about that until I took a look. I was “Whaaat?”

We’re still looking for what has been told us:

  • a better identification of inactive channels
  • massive deletion of offensive comments
  • advance notice to team members for bot/AI
  • plan to give early access to volunteers
  • giving the moderation of Disqus comments to some volunteers
  • better enforcement of guidelines and stopping channel hoarding

The other things that have been changed: they are not core problems or the main concern among many concerns volunteers have.

From what I’m reading, the intent or the purpose is not the same than what we have been told at that time, that is to say to use the bot or AI (let’s call it AI because it is not done by volunteers’ brain activity and sensibility) only as a last resource.

Feels like it is presented to us with another title like another product to deviate from the big elephant in the room, but it seems even more suspicious to me LOL I’m becoming maniac by staying here.

They are lucky to have Mariliam to talk with us, not sure it would have been so smooth.

2 Likes

No, I’m not pro automatic AI changes and decisions. I’d use it in the way I described as configured tool for the editor of that language and in a way that the AI marks segments that were changed by the English teams. So instead of implementing auto-translated subs I’d use it to compare edited English subs with OL subs and then highlight the box so the human editor can easily find and check it (and since I watched many dramas with unedited English subs I doubt that this would cause so much more editing work than now because the unedited English subs aren’t that bad at all. I didn’t have the impression they are completely wrong, it only happens that e.g. the lyrics (wording) of a song were not 100% the same for each episode but the meaning of the lyrics still was the same so it wasn’t a severe thing. On the other hand even after the English edit was done it could happen that some parts weren’t 100% correct - I had this in every project I was working on and I asked more than one native speaker to check it when I had this feeling; sometimes a sentence had a meaning that was not said in the origin dialogue but I usually didn’t message the English team because I felt that they might feel offended when I mention a overlooked line. I also had native speakers in my teams from time to time and each of them asked me if they have to translate/stick to the English subs even when it’s wrong and the origin dialogue says something different. The editing work for German subs done by native Chinese speakers didn’t need much time because I only had to do minor changes like a forgotten comma or changing a letter because of a different case).

If Viki would implement something like a software that is able to compare English with OL subs and highlights the segment fields it could make the work of the OL editors easier and then the OL teams could start translating right after the English subs are done.

In that case what @irmar described (about her work as editor/with English editing teams and creating a doc with the changes that was given to the OL editors) could be done by an AI that highlights the segments.

And then all OL viewers do not have to wait until the English edit is done and there is also no need for auto-translated subs anymore.

They’re not waiting because the German teams don’t work on it they’re waiting because the English teams do not give the GO for OL teams. With the auto-translated subs Spanish and Portuguese could watch shows with subs in their own language while all other languages still have to wait for the GO of the English team = all other languages’ viewers could have a waiting time for months (in some cases, e.g. one drama is 100% subbed in English with all episodes but only 5 episodes are allowed for OL and this situation is like that for months; it frustrates the viewers and also some volunteers because they’re bound to a project that’s not making any progress for months, sometimes the OL teams have to work/wait around a year for finishing a drama because they might get a GO for 1 episode/month or 1 episode/every second month. That might be also the reason that some OL mods and subbers are involved in ‘too many’ projects because they never know how fast it’s going).

I’m only talking about Chinese dramas that usually have ~50 episodes, sometimes even more, rarely less, maybe never? only 10.

So the delay is caused by the amount of time the English teams need for their edit and VIKI wants to solve that by using auto-translated subs (although there are enough Spanish and Portuguese human subbers).

Some posts mentioned that they’d need more editors for the English teams to solve the problem with the release for OL teams. But if that can’t be solved because you cannot summon editors from the netherworld I think instead of using auto-translations they could invest in a tool that detects segments that need to be edited (when you have a 50 episodes long Chinese drama you’ll need much time for editing anyway). So if OL were allowed to start subbing right after the English team hits 100% or let’s say 1 week later the viewers would get human made subs and the editor can do the edit later (with support of highlighted segment fields).

Yes! I really wish for improvements. This whole auto-translated subs AI investment feels quite disappointing because it is not a benefit for the volunteers.

1 Like

Bad idea.

Provide poor auto quality subs just as an interim? Viki’s reputation will go down the drain with such a pathetic solution. As a viewer, I would simply give up on the poor auto subs and can’t be bothered to watch the edited proper version later as it would take up too much time.

Have you ever watched some of the legit subscription channels in Youtube like Mango, Tencent, etc? The subs can be horrific.

6 Likes

I haven’t read everything you guys said, I admit I’m lazy because I’m late in this discussion, with dozens of messages late, so I’ll focus on this issue. Thanks Piranna! I love when you make me write monologues.

So, a few months ago, I had the idea of creating a list to boost a new dynamic in abandoned dramas, dramas without moderators. Here is the topic:Projets à reprendre en modération française : projets abandonnés, libres ou sans licence en France

Here is the updated google doc document (yes because messages are impossible to modify after two months on viki): https://docs.google.com/spreadsheets/d/1TK7xDhHK6m2LkRBP14TL2RsuK_7vPIePgyIZfjkA_lQ/edit#gid=0

First, we have (https://docs.google.com/spreadsheets/d/1TK7xDhHK6m2LkRBP14TL2RsuK_7vPIePgyIZfjkA_lQ/edit#gid=0): a list of projects abandoned by their moderator, projects without moderators with licenses in Europe or only US/Canada. So there are no unlicensed projects in US/Canada and Europe. There are also no abandoned or “break” projects from some moderators.

What is in green are the projects that have found a new moderator. What is in red are projects that have found a new moderator and have been completed. If we count, there are 17 projects that have been completed. Other dramas are still waiting to find people who will take over the drama.

Second, we have (https://docs.google.com/spreadsheets/d/1TK7xDhHK6m2LkRBP14TL2RsuK_7vPIePgyIZfjkA_lQ/edit#gid=1802398227): a list of some projects that are looking for translators because they need them. Of course, there are many more projects looking for translators, but moderators don’t always use this list.

In this list, there are 12 projects that have been completed. It is difficult to know if this made it possible to find contributors, but we can assume so.

Reality is that a tiny part of the French community got involved in this project. Some have really helped and supported this project. But if everyone isn’t contributing, it’s hard to get things done. Personally, each time I receive a message asking to translate on one of my projects where the team is complete, I answer them by giving the list of projects that are looking for translators. If all the moderators did that, it would make a difference.

Another reality is that some do not like this list, and surely think that I am just a bad person who wants to display them publicly (which is not true, since I do this for projects and viewers who are waiting. often answers in the comments, answers that will never come). “viki is a hobby”, “we come when we have the time”, it is totally true! However, the project is not owned by the moderator but by viki for viki viewers, so why not allow those who have the time to do it? That’s my philosophy. So, I decided to do it anyway, because there are already a lot of people who don’t really like me, one more or one less won’t do much for me.

Conclusion: idea is good, but it would work more if everyone supported such a project.

4 Likes

I can only talk about me, as a Portuguese volunteer I did not receive such survey. Or they know I’m not an auto-translations enthusiast so they let me out, it could be possible. :woman_shrugging:t2:

Wait, I’m remembering the last survey I took, it was about the accuracy of translations done with auto-translations, Google Translate and from volunteers. Was it that? But there was nothing asking if we were okay with bad subs.

If I were to guess, I would say that this is what they did.

2 Likes

Yes, thanks to that list, in 5 months, 17 complete projects, it makes it to 3 dramas complete per month and I participated only 1 + 16 dramas/movies have been taken over in 5 months and are being subbed right now = 33 dramas/movies in 5 months!
So if more people participated and more often, imagine how many projects could be complete in 1 month!

It is important to underline that the French community did efforts after the bot incident:

  • identify abandoned channels or channels that need help with moderating and subbing (mod and sub positions together)

  • promote it in the community and keeping it updated

  • official opening of the French subbing academy 1 month and half after the Viki bot incident! Less than 2 months! (We really fought to open this academy. To give you numbers, 9 months after the academy opened: 157 subtitlers applied, 67 have been taken care of by a sensei or have not answered, 13 graduated. A lot applied but didn’t answer later to begin their training. I think it is also because they have to wait before a sensei can take them under his wing.)

  • French mods I know (not subbers) are translating on dramas they moderate to have subtitles in 24-72 hours. They also have a job or are senseis and are editing elsewhere, making cover pages… they can do that until 1-2 a.m.!

How can the French community feel their efforts appreciated?

Note: we didn’t see intervention on Viki side for this, though it was the main concern of Viki at that time during Viki bot. We did try to find solutions on our own to satisfy Viki and viewers.

Also, let’s remind the goal of this academy is to allow new and old subbers to feel more confident on Viki, to ask questions, to be able to improve if they want to, to be assessed and have more opportunities on Viki, in teams where it can be hard to enter.

It is important to keep the community growing, and make them feel welcome and transmit this passion.

If using AI, it is like asking AI to paint, to write after analyzing Picasso or Orwell. It exists, but the beauty lies in masterpieces done by humans.

Don’t delegate something that is a passion for us. Don’t delegate hobbies, that is the answer.

I would accept a substitute only if the channel is inactive after 1 week. You can lock the epidodes in French during 1 week until it is fully 100 % subbed or 95% subbed so no one complains.

What bothers me with this idea is that volunteers who have been always at Viki’s side since its creation seem now to be relegated or overshadowed by AI. AI first, volunteers second.

Why should we intervene once there are auto-subs?
Why don’t auto-subs intervene in supplement to support us when the human intervention is not enough quick?
Why not this way instead?

Or is there a need that auto-subs are done before to be compared with the subs humans do secondly to improve AI until the surveys show that 90-95 % find AI accurate enough and can do without volunteers?

1 Like

THANK YOU so much for seeing both sides of the coin in an objective manner. Although you are not a fan of the auto-bot you understand the wants of the so called here ‘‘sub whiners.’’ of having the pleasure of seeing subs faster in the drama. I’m not taking their side either bc I’m one of those ''sub whiner myself but it’s much more bc as paying(customers) subscribers, I feel they have the right to complain.

Many here want to criticize them so much for not have any interest in quality in the subs, but the ‘‘whiners’’ don’t care about grammar or let’s say; the perfection in editing the translation, if it’s going to take a whole week before they can see any subtitles in a on air drama, no less!

Someone here said it takes 1 to 2 days for subs to be added in dramas, and that’s a lie bc to this day the earliest a drama takes to have subs in here is from 3 to 4 days(if we are lucky). If the drama airs on a weekend, it’s worst bc I never seen a single subtitle on dramas during the weekend until Monday. Like there is this one on air drama it kept saying today was uploaded but it was around 10% in the night/ it was 30% the next day again very at late night/ 60% after the 3rd.day then it stopped completely for one whole day before it was 93%. But there was so many empty segments in between I know that 60% was about 45% and the 93% was about 60 to 70% only. I’m still waiting for them to add the rest to finish that episode.

For these sub whiners and even myself; if the AI can give them/me a halfway decent sentence/subtitle, that we can halfway understand, the translation is good enough for them/me too. I think we are happy with that. I know for a fact that depending on the software name of the AI some are great, and some are very mediocre in their translations. Hopefully, they’ll get a decent one here.

I remember that on YT they had one of my favorite dramas but it had those horrible google translation subs (THIS was more than a year ago and they have improved a lot now). I really don’t care as long a at least half of the subtitle make sense, and bc of that I was able to watch/view all 16 episodes of the drama. It took more than a year before the drama was available here at this site, and although it had better quality subtitles I didn’t bothered to watch it again bc I understood the google ‘‘weird’’ subs.

I understand the volunteer’s frustration that they’re ‘‘killing’’ themselves doing all these work to provide the best subs, and I believe most whiners won’t even go back to watch the drama with better quality subtitles in them bc they’re happy with what they got from the auto generated subs. I ALSO do know that most of these ‘’ give me the subs now whiners,’’ will go back and watch the drama again with better quality sub titles. I always read their comments. ‘‘I watch with no subs and now came back to watch with the subtitles.’’

Ah ah ah… I beg to disagree here. I often can’t even understand what they are saying.

2 Likes

That sounds scary :sweat_smile:

Do you watch Chinese dramas too or only/mainly Korean? I cannot say how it’s with Korean dramas but in the rare case one I wanted to watch was available I watched it when it was still airing and without OL (but maybe it was already edited, somehow? It seems there are at least ~2-3 edits, right? So maybe the GO edit for OL is later than the one that already happened when I was watching the episode :slight_smile:

I will test that with the next/new on air drama I want to watch (watching it with subs at the moment when they’re visible :smiley:

2 Likes

@sheli

Wait, I’m remembering the last survey I took, it was about the accuracy of translations done with auto-translations, Google Translate and from volunteers. Was it that? But there was nothing asking if we were okay with bad subs.

That was definitely the survey but if that’s the case: Where they got that 50% were willing to use the auto-generated subtitles and 50% were not, when you are stating that they never ask you, if you were ok with bad subs?

Do you remember the subtitles used for AGS/G/V? Can you please give us an example if you remember them? Thanks in advance.

Here are some actual examples of pre-subs, then after the Translation Editor TE fixes the subs:

  • viki | Jan 1
    Is it anything related to play tennis?
    TE: How does meeting the stone lion at the park gate have anything to do with playing tennis?

  • viki | Jan 1
    Stone lions are rolling
    TE: Stone lions roll… Do you mean to see Sun Chao Yue? (Stone lions roll silk balls means good luck comes.)

  • viki | Jan 1
    If the we come to him
    directly it will be very
    TE: If you instantly replied “going to find Sun Chao Yue,” it would be ten points.

Blank segment for a text message.
TE: [Li Na, look at Kuaishou(video sharing app). I am on the side livestreaming.]

In my experience of working mostly with pre-subs, I found that 1/3 of the drama does not make good sense without the Translation Editor. Working with origin translators, my GEs remarked that our Viki translators have superior English.

I have watched shows with pre-subs, and I could fairly understand what was happening. Many other people could not. Honestly, I think it would be a minority who would actually benefit from having auto-generated subs. I believe it would be in Viki’s best interest to serve the majority in the best way possible.

6 Likes

It’s related to this topic we had a few months ago:

I forgot, but we have to look for Mariliam’s answer. Her answer if I remember correctly was that pre-subs in English were from the content provider.
I don’t think they are from an AI, but I am not 100 % sure of that.

Did you change your stance or change a little your perpective on pre-subs after volunteering a little more on Viki or having more exp/get more acquainted with other TEs and subbers?


To come back to OL auto-subs (I think they say English subs were safe for the moment, no auto-subs planned for them), your ex shows that it’s important that the auto-bot is activated after the English edition for this pre-subbed drama? I’m not on this drama, so I don’t know the proportion, I would just believe it’s about 1/3, the proportion you gave. What about other pre-subbed dramas?

We still don’t know the dramas where they plan to use auto-subs, but reading Mariliam’s answer again, from what I’ve understood: it’s something related to abandoned dramas? Or dramas where the English edition is not complete yet.

But it would be always wrong to subtitle for us or for the AI if the English edition is not done (at least in your case?)

So I wonder why not giving the go for other languages so they begin to translate when the English edition takes more than XXX time for example?

But it implies OL have to edit after English edition?

K drama or movies which are pre-subbed before uploading at viki are usually done by human subtitlers paid by the broadcaster or the production company. I am not aware of any which were subbed by AI. The subbers are not viki employees even though each sub will say “viki” as the subber. For example, if it is an older subtitled movie that you can see on youtube, the subs we get will often be identical.

3 Likes

@cgwm808

Do you remember the delay you gave to Viki staff subbers to come on dramas you managed to complete English subs?

I only have 1 exp on that, and was it something like 24 h after the upload on Viki, but that could be extended to … hours?
I don’t remember it well.

But I remember our last conversation where the main concern for you when you edited was to release ep to OL after xxxx time, what was that xxxx?

@piranna
On dramas which were not attractive to the volunteers was I asked that if we did not reach 90% English within 24 hours of actual upload time, the paid subbers could complete the episode. If we reached 90% within the 24 hours, the paid subbers were not to touch the episode at all. I insisted on “actual upload time” because sometimes uploads have been delayed for 24-36 hours. Even when the episodes didn’t have enough subbers to complete the episodes, I wanted volunteers to continue to have the opportunity to subtitle. I think our volunteers subbers are the best, but sometimes the timing isn’t right or, more often, the drama and the cast don’t attract enough subbers, and I recognize viki is a business site which is providing entertainment to an international audience which needs the subtitles.

1 Like

Yes, I agree with you that it’s a company with volunteers, that’s why I also agreed to have Viki staff subber intervened at that time when it was not completed fully by volunteers after 24 h, that is to say on Kdramas and on Cdramas, because there are also viewers and OL that want to sub.

If we agree on that for English, why can’t we agree on that for OL?
Can’t we do the same for OL?
That is to say, if it’s not completed fully by OL volunteers after XXX time after actual releasing by English editors, we agree that AI intervenes on an advance notice from Viki?

It is the same concept, adapted to OL? If it’s okay with English, it could be also okay with OL?


Abandoned dramas and AI:
I still think if the AI operates on abandoned dramas, what the French community did after the bot incident can inspire Viki or OL communities.
A list where Viki wants quick subs because it was abandoned or the English edition is not complete.
There would be completion by volunteers and since Portuguese and Spanish are more than French ones, the proportion of 17 dramas/movies complete in 3 months for French might be far better with Portuguese and Spanish and faster.

For me, the gain is not really good because it’s already an abandoned drama, so that one day, suddenly the AI shows subs or the volunteers do it fast because of the given pressure from AI and Viki, I mean the gained time on an abandoned drama is relative for a viewer after waiting XXX time already. Would there be a difference?

I think there is more gain because it would calm volunteers and keep peace with them + having quick subs for viewers because I think the Spanish and Portuguese teams, after this ep, would want to complete them like what we did in French.

It is better to have volunteers in their pockets than having the risk of a strike going on for Portuguese, Spanish, other languages and I think the reputation of quality is something that can’t be sacrificed for this small gain on abandoned dramas.

Just info, we need info and Viki telling us, I need quick subs on this.

When I tell my mods to complete it or to go take a look, some of them do and some of them don’t do.
It’s managing teams, the community.

I also let them the choice: subtitle yourself, recruit or just stay as subtitler because at least, I won’t be behind them with a deadline. Giving more possibilities/alternatives for them than just AI, no?

I might miss something in reading your comment, but isn’t that what is happening right now?
And you are talking about on air channels now? In the first test they will not be included.
But as far as I understood Viki the AI program relies on written subtitles.
Are you back with AI subtitles in the subtitle editor segments, so the mistakes, that will take place will have to be edited by people?
Since now the AI translations will be on another “sheet” that will not touch the subtitle editor segments?
Or do you actually mean that these AI translations should only be visible after the English translations set time is up in addition to another set time for the OLs, if the set time of the both teams is up and the subtitles are not at let’s say 90 %, then and only then should the AI translations show until the OL is done?

I am a bit confused now and might not read your comment carefully enough, so could you verify this?

How is it the same concept? It’s definitely NOT the same. Professional subbers who come to help are humans, not AI. This makes all the difference! We would probably be okay with human OL translators coming to help.

3 Likes