/robowaifu/ - DIY Robot Wives

Advancing robotics to a point where anime catgrill meidos in tiny miniskirts are a reality.

Build Back Better

More updates on the way. -r

Max message length: 6144

Drag files to upload or
click here to select them

Maximum 5 files / Maximum size: 20.00 MB

More

(used to delete files and postings)


Have a nice day, Anon!


Open file (32.03 KB 400x400 FXCY9fGv.jpg)
THE LANGUAGE PROBLEM Robowaifu Technician 11/20/2020 (Fri) 13:22:39 No.6937
'Sup anons? I am here to remind you guys something important, TO DO YOUR RESEARCH IN MULTIPLE LANGUAGES. Our mutual language is English however that is not enough. We need people who can speak those 3 important languages: Japanese, Chinese, Russian. I've been learning japanese for 4 months and with the help of dictionary I am able to understand basic stuff. Here is the point, there is a whole another world out there. 1) Chinese: Chinese people work under such hard circumstances and put really much effort into their jobs. Nearly none of the projects are translated into English since Google is banned at China. However there are a lot of great stuff there, I mean like even Microsoft runs their virtual woman project there. Since it is too hard for me to learn I generally use DeepL (best translator out there) and Baidu (Chinese search engine) and read latest researches and projects there. I wish I knew chinese well, in that case I would be able to find not-so-popular webpages and grab more information on topics. 2) Japanese: Even though good Japanese projects gets translated into English most of the research there gets translated only when the projects is ready to publish and sometimes they are too hard to find. I try to read as much AI papers in Japanese as I can. Scientists there do great stuff. I've seen a lot of robowomen projects there. You can also find some 3d printing projects for anime girls. Really worth looking. 3) Russian: Russian is the least important one in my opinion. But a professor of mine graduated from a university there and he has a lot of academic books that aren't translated to English. You would be amazed at how much work they have on stuff such as algorithm theory, artificial intelligence, computer science. Most of the stuff there are focused on "science" part of CS. So they are theory-weighted. So right now we need people who can speek Japanese and Chinese (Korean would be good as well, but there isn't that much of research there tbh) Using DeepL is enough to understand most of the pages but only a person with fluent Chinese/Japanese would be able to find goldmines deep there. I am pretty sure that there are hundreds of Chinese people working on robowaifu related projects that we are not aware of. Same applies for Japanese people, but since Chinese people are in a much worse situation it becomes really hard to find those people. Anyone has some recommendations? I wish I had time and skills to learn all those languages but I can only afford to learn one and I am going with Japanese since I have a dream of moving there in the future. We need to brainstorm on this issue.
For example with a little search I was able to find some interesting news. Link: https://news.mydrivers.com/1/459/459199.htm I really enjoyed reading this link, here is a quote from the link "Another reason is that the main force behind the development of intelligent robots is men, so female robots are more attractive to them than male robots and bring a bit of fun to a boring job - who wouldn't want to create the perfect woman themselves?" Another link: https://baijiahao.baidu.com/s?id=1612937377753302098&wfr=spider&for=pc "Artificially Intelligent Beautiful Robots May Become Human Companions" :^) Another link (from the website above): https://baijiahao.baidu.com/s?id=1628957393261869647&wfr=spider&for=pc Apperently University of Science and Technology of China is working on robowomen, I better check their website and papers if there exist any. When I have time I will also look at some Japanese websites to see what they are up to. If I find anything worth looking, I will share them in this thread.
>>6937 >Microsoft runs their virtual woman project sauce?
If I understand the gist of your position OP, it's that we should all learn one of these languages fluently enough to do academic research in those languages natively. Correct? While I admire your personal determination and skill in learning Japanese, it's not something I myself can tackle in a big way. If I ever move there I'll just have to pick it up organically as I go, and I certainly won't be likely to manage skill high enough to study technical research in the language. However, it's definitely true that men all over the world, particularly from the areas where the four languages you mention are spoken (whether multi-lingually or not). The drive of men to deal with the failures of most modern women is basically universal, as our welcome thread points out. So there's is plenty of room for connections and cross-pollination there, absolutely. This is a global phenomenon. That's a good thing. From my angle of trying to actually implement something with robowaifu speech and languages, I'm somewhat interested in the generic characteristics of language itself. Kind of a meta-language-language if you will. Fortunately for me, there has been plenty of research done in English heh in this area, and I hope to apply some of it to my software development efforts. I realize my goals aren't quite the same as yours for multi-languages, but we can both agree that languages other than just English are important.
>>6947 There you go anon. These are the links I saved, maybe if you search web you can find more information. http://www.msxiaoice.com/ https://archive.is/QMXze https://github.com/Iamnvincible/msxiaoicemsg https://archive.is/BSYla >>6948 >fluently enough to do academic research in those languages natively. Correct? Yes, I know this not an easy task but there is a whole another world there. There are people working on dolls, there are people working on NLP and creating virutal anime girls, there are people working on creating realistic 3D anime waifus... But the problem is, they don't speak English. They are alike us, but their research is in Japanese and they can't speak English. I was pretty shocked when I first started to read researches in Japanese. I am not only interested in creating robowaifus, I also work on cybernetics, cyborg technologies, learning models, theory of algorithms, deep and machine learning, functional analysis, humanistic psychology, Systems theory, neurology... You got my point. I recently realised that even though we think English is the largest language and it is the language of science, this is not true. I borrowed around 6~ books about CS from my proffessor written in Russian that are not translated to English (he told me so, I did not bother checking if they were translated after he bought them). There are a lot of great information inside them. I could create at least 7 academic papers out of those books (at least I imagine so, because there are some concepts there that are not well known). The same applies for Japanese, I found a lot of research and projects about cybernetics that aren't translated to English. I soon will start to lurk around Japanese/Chinese forums to see what I can find regarding robowaifus. Chinese is too hard to learn, I really had a hell of a time to learn the basic grammar. I think you can manage to search what you are looking for after learning basic grammar and using translators which would take around 1-2 months with studying only a little per day. It will be interesting to see what other people around the world are doing. I have a contest idea, let's see who will be the first one to create a functional robowaifu. Us or them.
>>6949 Sure enough. Thanks. They plainly have the Microsoft logo on the original page. They seem to have some misplaced that recently haha. https://web.archive.org/web/20140626001247/http://www.msxiaoice.com/
>>6949 >books about CS from my proffessor written in Russian that are not translated to English (he told me so, I did not bother checking if they were translated after he bought them). There are a lot of great information inside them. I recall Lex Friedman on his podcast talking about this. They seem to have some approaches of their own. Don't remember which guest, though. So yes, I wouldn't underestimate Russian. However, it's a good idea to keep that in mind, but only some of us here will be able to pull it off to learn another language on top of everything else.
English imageboards are very international, I think we already have a significant amount of people who know a language other than English. Russians and Japanese are likely the most common easterners in English imageboards.
Anon, I've been keeping tabs on Japanese projects on Twitter, the robowaifu guys there love to retweet anything interesting. I notice most of them use Raspberry Pi, Nvidia Jetson Nano, M5Stack, and the multitudes of Shenzhen arduino clones and microcontrollers. What they mainly got different is they got their own seemingly Japan-exclusive servos or for some reason are able to afford Dynamixels. As for language processing, if we are training chatbots they will be dependent on the specific language unless they just all make cute noises like Chii -- well our mascot is Chii so a cute noise generator waifu will be more universally appealing without the need for translations.
>>6975 We miss you no longer reporting on updates from those guys. That really added some interesting flavor and inspiration to /robowaifu/.
My Japanese is no where near good enough to read research papers. I just opened one and got stuck on NLP, kek. >はじめに従来,自然言語処理 (shizen gengo shori) >Introduction: Natural language processing >>6975 >if we are training chatbots they will be dependent on the specific language Not necessarily, it's possible to create a chatbot that can speak in multiple languages and translate between them. However, with current methods those languages have to be targeted at from the beginning. I've been researching ways to create a language model that can handle text in any language and automatically segment it across different languages. So far I've created one that can operate on UTF-8 text, a way to encode and decode variable length sequences, and a way to structure sentences in a hierarchy of compressed sequences. It works on toy examples but I haven't had much success yet training on large datasets. I'm thinking it might be better to create some sort of neural byte pair encoding so that whenever the model sees a repeated pattern it automatically encodes it into a token. I think this ability to compress information and improve the compression will be essential to creating high-quality conversational AIs. There's a Japanese roboticist on Twitter looking for a decent Japanese chatbot for his robot, Kibo-chan. He's pretty friendly and I plan to show him my project once it's ready. I think it would be a great starting point for English and Japanese robowaifuists to come together, share their knowledge and collaborate. They're leagues ahead of us in robotics but the West is leagues ahead in AI. If we can combine our strengths it'll be amazing. And of course we could also reach out to Russian, Korean and Chinese developers too.
>>7267 >I think it would be a great starting point for English and Japanese robowaifuists to come together Not that anon, but that would be absolutely wonderful. Kibo-chan-Anon knows a thing or two about character-animating servo based armatures obviously. I would absolutely love to hear her speaking with some kawaii synthesized voice as well.
Open file (23.09 KB 480x360 hamtaro-FIGHTO.jpg)
>>7267 >My Japanese is no where near good enough to read research papers yet. FTFY, Anon. Gambate!
Realize this is an old post, but here are some useful resources for Japanese https://itazuraneko.neocities.org/
>>11389 >is an old post Heh, don't worry Anon. This isn't a typical IB in that sense. That is, there's no such thing as 'necrobumping' here (or any complaints about it either). If you have something to add, by all means put it in the correct thread.
Open file (476.56 KB 1100x600 example2.png)
Sometimes PDFs don't copy and paste text correctly because researchers upload scanned documents and whatever OCR they used on it sucks. For a long time I've been using Google Keep which has a great multilingual OCR feature but I'm looking for a simpler open-source solution so I don't have to copy pages and pages of paragraphs. So far I've found these two that support Asian languages: https://github.com/PaddlePaddle/PaddleOCR https://github.com/JaidedAI/EasyOCR It would be great to have a tool one day that automates PDF OCR and prepares it into a document for translation on DeepL. A lot of the time I just ignore research in other languages because it's such a hassle to read.
>>11491 I think I understand the need you're describing Anon. Having no experience with what's being depicted, I'm confused by the provided image however. Any chance you can break down what's being represented there for the uninitiate?
>>11492 It's optical character recognition. It's outputting the bounding box coordinates of text in the image, the predicted text, and the confidence level.
>>11494 Ah, I suspected that might be the case, seemed to make sense. I presume the Asian characters would be sent through some kind of translation software afterwards?

Report/Delete/Moderation Forms
Delete
Report