How Google Translate Works - The Machine Learning Algorithm Explained!

Let’s take a look at how Google Translate’s Neural Network works behind the scenes! Read these references below for the best understanding of Neural Machine Translation!
SUBSCRIBE TO CODE EMPORIUM: kzbin.info...
To submit your video to CS Dojo Community, please use this link: csdojo.io/enter
REFERENCES
[1] Landmark paper of LSTM (Hochreiter et al., 1997): www.bioinf.jku.at/publication...
[2] Landmark paper of Neural Machine Translation NMT (Kalchbrenner et al., 2013): arxiv.org/abs/1306.3584
[3] Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation (Cho et al., 2014): arxiv.org/abs/1406.1078
[4] Seq to Seq learning with neural networks (Sutskever et al., 2014): arxiv.org/abs/1409.3215)
[5] The paper that introduced Bidirectional RNN : pdfs.semanticscholar.org/4b80...
[6] On the properties of NMP: Encoder-Decoder Approaches (Cho et al., 2014): arxiv.org/pdf/1409.1259.pdf Fig. 4 (a)
[7] NMT by jointly learning to align & translate (Bahdanau et al., 2016): arxiv.org/pdf/1409.0473.pdf 5.2.2
[8] Google Translate Main paper (Wu et al., 2016): ai.google/research/pubs/pub45610

[@swarnimapandey7297]

This is just the most insane video with the bestest ever content i have seen till now.. thanx a ton man!! for such an informative video.. i am saying all this because i was myself busy reading research papers these days and other contents on how actually google uses machine learning algo in translation.. and finally this video was " the cherry on the top".. keep making such videos , computer technology much easier too learn...

[@CodeEmporium]

Thanks for the kind words! I'll keep making more content back on my own channel "Code Emporium". But I'll still post some videos here too to reach more people. :)

[@Dmitriuso]

Great Job, guys! Thank you very much for such a detailed explanation!

[@davidvallejo7237]

Awesome video!! I just started studying machine learning on college, and I gotta say you were the one who made me realize I love this! Also I love languages, so it was greater!

[@juunih]

same

[@ITelefonmanI]

Thank you for including the sources at the correct time.

[@hasanabbas21]

Great explanation, I was looking for a simpler explanation of the attention model without the internal details of the lstm network itself. this one hits all the components involved in the latest in seq2seq models. Thanks

[@KrazeeKrab]

I'm writing a paper in Translation Studies about Machine Subtitle Generation. This video helped me a lot!

[@OKBYE0]

same here extremely glad

[@tempor8336]

Hello from France 🇫🇷 ! Indeed, at 7:55 the French sentence that created the algorithm looks weird

[@destinyjames6117]

This video deserves more like!! Thank you very much for making me understand this.

[@dawintch4596]

Man this is such a good video, I hope more people could actually see it.

[@subhits1903]

I am 13 now, I'm preparing myself to learn to code!

[@HarishKumar-0405]

Yeah, me too, same here. Actually I started this 1 year before, but I just spent that whole year trying to assure an ideal kind of path for me to learn to code. And then I learnt, that it can't be. There is always new things coming up in this world of coding.

[@ernestomercado8176]

@@HarishKumar-0405 yeah you need to learn 1266248727294828498 new frameworks every fucking day

[@AKOSAH-MUSIC]

Nice one there ....... thanks for this class.

[@shaileshpatil1678]

Gr8 work very nicely explained. Hats off

[@singhRajshree]

WOW dude this was amazing...i mean you should also start a channel on research/phd trends Great content and lot of hard work as can be seen

[@juunih]

GREAT video. thank you!

[@monicadesai5697]

Thank u for this amazing explanation.

[@CodeEmporium]

You are welcome! :)

[@rogendothepoet3108]

Nice tutorial, quality content

[@leliaglass1568]

thanks for this very useful video!

[@alahis]

Great content. Thanks for sharing.

[@edmarklowelldelossantos4952]

Excellent video!!! Great job!!!

[@idontknow-ms8mc]

This was really interesting 👍🏾

[@CodeEmporium]

Thanks! 🙂

[@chrisbovington9607]

Okay, I'm not sure if I understand exactly what you were going for in the first example (7:55), but you said the reverse translation broke down at "medical center" (word 19/20). If you are talking about a faithful translation of the meaning of the sentence, then it broke at "recognize" (word 12). The translation does make sense on it's own terms until word 19/20 so maybe this is what you referred to as semantics? Is that the point of this example?

[@rajatparab8116]

Nice Explanation bro

[@CodeEmporium]

Thanks homie!

[@arunadevi8724]

Nyc video about translator and I understood how it works but how to implement the LSTM-RNN using which algorithm we can implement

[@pchandu1995]

Thank you for the vid.

[@johncysamuel]

Thank you so much 👍❤️

[@morningstarcb]

how do you teach the machine about gender? for example, when I translate sentences from Russian to English, I always get the translation in the masculine form. For example: I bought a book. In Russian, the verb bought has a feminine form (if a woman is saying it) and masculine form (if a man is saying it), but google translates only in the masculine form.

[@samkitjain4883]

Thanks for this amazing video can you make a tutorial series for learning dhang please(from india)

[@Error_404.]

Morphology translation is still not there, mostly semantical syntax varies between different languages. Machine learning is not there yet but eventually accurate real time translation.

[@vikrantdhawan4062]

Sir plz explain how and what to study for crack interview

[@suryakantupadhyay1903]

What about the sentences that have drastically different grammar structures? Eg:- english and japanese

[@xtb3215]

That's exactly why Google Translate sometimes has a hard time with things like Japanese.

[@hasangur007]

Can you explain how the “gold standard” works when the algorithms analyze the target texts. And how it learns the translation solutions, what is the relationship between the training data and the way the system learns translation solutions?

[@teasugaradream]

14:44, success of both hardware (GPU o/ FPGAs) and software (AI/ML) disciplines..

[@sobitregmi31]

i am confused whether it uses one hot encoding or word2vec as input to the encoder?please kindly reply

[@abcdxx1059]

Can you make a tutorial on machine translation with tensorflow I can't find any good videos on this topic

[@CodeEmporium]

Interesting. I'll probably make programming videos in the future. Thanks for the feedback!

[@jackblundell2850]

What if you wanted to translate a completely new language? I'm interested in trying to approach this with machine learning when I learn it, I'm just not exactly sure how you would convert the unknown language into the vector

[@jackblundell2850]

Super interesting video

[@nucderpuck]

My favourite test example is the Groucho Marx quote "Time flies like an arrow; fruit flies like a banana." It is very interesting to see what Google Translate does with that when translating e.g. into German. While the first part of the sentence is translated very well, using the appropriate figurative speech equivalent used in German ("Die Zeit vergeht wie im Fllug"), but the second part is severely mistranslated. Here, "fruit" is interpreted as a noun and "flies" as a verb, rather than identifying "fruit flies" as insects. Never blindly trust a neural network...

[@ruslanmamedaliyev3912]

Also one word can have different meanings and the machine must "understand" those meanings to translate them and sentences correctly And there are idioms and other stuff that the machine has to learn too

[@deathbombs]

Looking forward and backward, which words do we focus on? Answer: none, the neural network does the work :P

[@arunadevi8724]

Do we translate only text or the text on an IMG using this networks

[@viktorpless9969]

Why isn't transformer models mentioned in this video? The related paper is from 2017 from google so I wonder if these recurrent NNs are still used at all?

[@Myselfkartik21]

can you send me the link of paper

[@Viplexify]

​@@Myselfkartik21 title is Attention is all you need

[@singhRajshree]

What about code mixed language. Does this model works on that??

[@syedraihanmahmud3136]

I learned English from google translator. Before I didn't know a simple sentence of English. I am not saying that I know English properly but, I know as much as I need to talk to others .

[@theobellash6440]

11:22 "Was signed" is actually translated by "a été" here not "été".

[@YuChiGongG]

I think he meant to say that the morpheme is the smallest unit of language, and NOT the token. Correct?

[@BunnysBro]

hello dojo i am very big fan of you.i am from india.can u teach me for my best carrier in future.please

[@watashihadesu]

7:42 - reverse translation seldom is a suitable way to check the correctness of a sentence. It does not work with humans, it will not work with machines. Or will it?

[@spadaacca]

I'm a bit confused. Why was été aligned with "was" and "signed"?

[@mohammedissam3651]

13:43 Scaled up up up it goes... dam Google translate is not accurate ! The right & best way is the first one but it costs a lot of time & effort & device performance. Great video 👍🏻 Love it 😍

[@deroytanaka]

A droite is right as in "right hand" not a right "right to vote"

[@PCHUANG-yk9pw]

I am a translator (NAATI) and I would love to see the development of AI to help us. Great progress has been made; DEEPL, for example, can do a decent job. But Google translate, to be honest, falls behind and cannot be used in professional practice. In my opinion, generated from my experience, Google translate missed a key point, which is CONTEXT. Without the awareness of the context, AI would not be able to do a good job even it knows all linguistic elements of language.

[@itorres008]

"Way back in the nineties" ha, ha, ha! 😊

[@OrangeC7]

Lol this guy is funny! ...wait a second

[@teasugaradream]

Look at the below examples: Good one: I do not think you have enough time to have someting made. >>> Je ne pense pas que vous ayez assez de temps pour faire quelque chose. Bad one: I do not think you have enough time to have anyting made. >>> Je ne pense pas que vous ayez assez de temps pour faire faire quoi que ce soit. It means you have to make the necessary fine-tuning operations on your raw sentence and this requires to be an expert. That is, the automated tools are to be used by the experts for the professional operations.

[@epicgaminglounge3840]

sir please make videos on vb.net source codes absolute beginer

[@luxtenebris764]

Another thing must be highlighted is that you in English can be understood from the phrase while French have 2 you one mean you (1 person, informal) and vous (2+ , formal ) so calling a vous by crétin IS like respect and than disrespect !!! So this ai or something like that doesn't know what to use for the phrase from the context

[@verglas.]

I like how the thumbnail is just 𝚈𝚘𝚞 𝚊𝚛𝚎 𝚊 𝚖𝚘𝚛𝚘𝚗.

[@teasugaradream]

13:26

[@pearltears8039]

FOR GET GRAMMAR! I don't get Google Translate..... Did you know if you type in ..... The word "Bless" in French it translates..."Wounded" in English If you type in...."Bless" in Dutch It translates..."injury" in English But if you type in... "Bless" in Latin.... It translates...."Love" in English SO DOES THIS MEAN IF I AM SPEAKING DUTCH AND I SAY THE WORD BLESS TO SOMEONE AM I REALLY SAYING INJURY TO THEM... OR IS IT IF I SAY THE WORD INJURY TO THEM IT MEANS BLESS TO THEM???????

[@googletranslate149]

*did you mean* : how Google translate eat the Amazon cat?

[@alisiaale565]

This youtuber looks so beautifull!

[@GamingKing-jo9py]

But we don't know what happens in the LSTM. It's still unknown.

[@CodeEmporium]

You're right. I didn't get into details of LSTM in the video. But I have videos on my channel explaining just that. Here is one with LSTMs explained. Hope this helps: kzbin.info/www/bejne/h5TMepaIra-pqa8

[@mixpsycho7564]

hello bro i want to contact you please

[@amrutha46]

i want to know the mathematics behind this, can somebody help me with it?

[@franciscolupercio8485]

We don't use cheese for tacos 🤣

[@a2sirmotivationdoses782]

Can you please make a video on text to speech?

[@FTRASARAVANAKUMARB]

Can a robotics engineer get job in IT company?

[@chethanjm2685]

ay quá

[@harrisonford4103]

I just enrolled in the institute of automation for MT, your video helped me a lot! Thank you! I understand French, the translation is just so so, but the pronunciation of French in the video is TERRIBLE! Maybe Google can do better on this.

[@credirgam4865]

It's similar to language theory

[@pearltears8039]

I don't get Google Translate..... Did you know if you type in ..... The word "Bless" in French it translates..."Wounded" in English If you type in...."Bless" in Dutch It translates..."injury" in English But if you type in... "Bless" in Latin.... It translates...."Love" in English SO DOES THIS MEAN IF I AM SPEAKING DUTCH AND I SAY THE WORD BLESS TO SOMEONE AM I REALLY SAYING INJURY TO THEM... OR IS IT IF I SAY THE WORD INJURY TO THEM IT MEANS BLESS TO THEM???????

[@thrunsguinneabottle3066]

Strange. I would have translated the last sentence as: "Vous êtes crétin".

[@gerardoa9179]

Easy, there are humans behind the algorith

[@andrewlehti801]

wait.. the answer isn't like shit?

[@chicoti3]

No matter how much google improves their algorithm, it will always chug and die if you input Japanese.

[@rowlandolanrewaju3227]

I want no more about python. How apply python in any ways

[@robertyoung6722]

Nope

[@robertabitbol6454]

You tried to explain the way the system works but you failed. Everything went downhill after :5.00. You did not explain the coder, decoder concept. It's clear you did not understand the concept yourself. This is NOT how you explain a concept: By skipping over stuff you don't understand. You would have had to take an example and explain precisely what the machine does. So no. Your explanation is not successful. However until 5:00 it was pretty good.

[@Pastor_virtual_Robson]

just a Goggle propaganda

[@haSHAH1]

huh