🔥 Popular | Latest

Computers, Google, and Omg: Exploiting Similarities among Languages for Machine Translation lya Sutskever Google Inc Mountain Vicw 1lyasu@google.com Ouoc V. Le Google Inc. Mountain View qv1@google.com Tomas Mikolov Google Inc Mountain View tmikolov@google.com Our stady found that it is posaible to infer missing d ted repeeseats Abstract dictionaly shrases We achieve arning hat ep arsiection betwe ad coests e two eet cach language simple saep angr d dictionary o lo doveksam e build monolingal modcls ing large amounts of test wang lincr peoiec- ingw d mapping he m nal b the sest sime, we ca e seen in the tion between the lan by aing on large water dae any wo sheep vecton by projecting dsribad Ium the source lang he vector in the lagage space Once urpet lang a on arget bwen nd leans ervages Despie enbod in s we ouput the mot similar e ca r ra we learned Continuous recently peoposed 2O13) These models leare The epresentations of langua the disirituted Skip gram e Bar of Words CBOW NaKelnde ngas o lon tables for ay y (Mikolov et al using a simple neural aers to predict the Becan of ts simp t amount of esl d ip-gram and CHOW I Introduction Statiotical machine ramlatione ysiems have been odels can mentation can learn a moded our par Nliom of words in bours vears and became very cly developed uliztion to illtrate kIn Figure Figre 1 ves Comments 1.7K Add a public comment... jellymantanki 1 second ago 1:18 WATER SHEEP OMG ITI The vid is called why computers suck at translation if you want to know

The vid is called why computers suck at translation if you want to know

Save
Smoking, Good, and Jokes: SPORTCHAF NOROR Anyone else notice the similarities between Jack and Chandler? They both enjoy smoking but hide if from their wife’s, make really good jokes, and rock sweaters

Anyone else notice the similarities between Jack and Chandler? They both enjoy smoking but hide if from their wife’s, make really good jokes...

Save
Animals, Computers, and Google: Exploiting Similarities among Languages for Machine Translation Ilya Sutskever Google Inc. Mountain View Quoc V. Le Google Inc. Mountain View qv1@google.com omas Mikolov Google Inc. Mountain View ilyasu@google . com tmikolov@google.com Our study found that it is possible to infer missing dictionary entries using dis of words and phrase lincar pro Abstract ibuted representations achieve this by learning Dictionaries and phrase tables are the en statistical maciine hd that can between vector spaces language. The method co resent sems. This paper d ecnerating and automate aries and phras d phrase eual models of steps. First, we build m languages using large amo a small bilingual tion between translat al corpora tion from the source blain the vector i ts of text. Next, we use ary to learn a lincar projec languages. At the test time. mehod can translale sructures hydraulic entries by lenal data and mappdta n lanuages from s tio of words tuses distribul ning between word that has been seer projecting its ector representa- ram o the target space and leams uees Despite its simp we cu surprisingly S for transla- language space target language space, we output the most similar word vector method translation. achieve almot Enelish and Spanish. entations of languages are learned Continuous distributed Skip- ently proposed The repr bod makes little assur the days of "hydraulic ram" being translated |as "water sheep" languages, so it can fine dictionaries language are pretty much in the Introduction our parallelized implementation can lean in hours aiems have been I became very successfut aMems rely on dictionaries h efforts to generate atistical machine trais 1:16/5:04or years d from bill gives simple visualiza how our method works ors for numb Figur In Figure 1. we and animals in En- that these Why Computers Suck At Translation 1,104,604 views 26K.לן 253 SHARE ESAVE Exploiting Similarities among Languages for Machine Translation lya Sutskever Google Quoc V, Le Google Mountain View qv1@google.com Tomas Mikolov Google Inc. Mountain View tmikolov@google.com View Moun ilyasu@google.com Our study found that it is possible to infer missing distributed representat carning Abstract dictionary entries hrases We achieve this by of words that rep- ojection between ve consists Dictionaries and phrase tables are the basis two f modern statistic sent cach language. The simple steps. First, we languages usdictionary dockkes a method that the process of gener Out nding dictionanes a word and phrase d monolingual models of amounts of text. Next, we u learn a line a small erween the languages. At th ranslate any word that has lingual corpora tion from the method can tra language structure water annine be laree monolingual da hilingual data. its vector representa- the sheep tween lang represemation of s leas a lincar mapp simplicity. e language space n the langu t language space, we output the most similar wond vector as the translath The repres ed Skip-gram using CBOW) models rec earn word Ba by (Mikolov et al., 2013a). These n spaces of languages risingly effective me t90% precisione languages are learned Conti and Spanish. tion of words betwee umption about the proposed languag is and translation ables fot any lanquage pairs tations using a simple neural network archi- ecture that aims to predi past. the Skin f its simpli Because Introduction don a large a models can l implementation can Icarn ine transzn ems have been p m ens in hours Figure 1 gives simp works. In Figure 1, we acni 1:20 5:04 vears and became very su tionaries and systems rely on forts to generate for numbers and animals Why Computers Suck At Translation CC 1,104,604 views 26K 253 SHARE ESAVE Tom Scott tried to warn us. We didn't listen.

Tom Scott tried to warn us. We didn't listen.

Save
Google, Target, and Dictionary: Exploiting Similarities among Languages for Machine Translation lya Sutskever Google Inc. Mountain View ilyasu@google.com Quoe V. Le Google Inc. View Tomas Mikolov Google Inc Moun View Mountain V qv1@google.com tmikolov@google.com infer missing Our study found that it is possibl nary entries using distribsted representations words and phrases. We achieve this by learning lincar projection bet resent each language. simple steps. First. languages using a small bilings Abstract Dictionaries and phrase tables are the basis f modern statistica vector spaces that rep- The method consists of two chine translation sys- develops a method that the process of generati tems. This build monolingual models ftext. Next. dictionary to learn a line tablies. Our word and phrase language structures based opolingual data and mappin data. distributed representation of words. nding dictionaries and phra method can translate entries by lear amounts projec- tion between the languages. At the test translat ime, we can hydraulic on large tween languages from small the mono- word that has been corpora by projecting its maun tion from the source languag ctor representa- the target the lincar mappin veen vector nd learns ram espite its simplicity spaces of languag nely effective: we e space. Once we obtain the vector langua language space, we output the most similar our method is achieve ta precision@5 for words between English vector as the translatio The representations using the distribute Bag-of-Words (CBOW) models recently propos by (Mikolov et al., 2013a). These models learn word representatio tecture th Весаи models can be trained on a large a parallelized implementat Snanish languages are learned Skip-gram or Continuous This method makes little tion about the sed to extend and re- languages, so it can fine dictionari language translation tables for any k archi- using a simple neural netv aims to predict the neighbors of a word. its simplicity, the Skip and CBOW Introduction nount of text data: can learn a model inhours Exploiting Similarities among Languages for Machine Translation lya Sutskever Google Inc. Quoc V. Le Google In Mountain View Tomas Mikolov Google Inc. Mountain View Mountain View ilyasu@google.com qv1@google.com tmikolov@google.com Our study found that it is possible to infer missing using distributed representations learning Abstract dictionary entri of words and phrases. We achieve this a linear naries and phrase tables are the basis modern statistical machine tems. This paper deveeenerating and ex automate the and phrase tabies. tending ces that rep- I consists rojection between vector dation sys- two resent each language. The meth simple steps. First, we build monoingual models of languages using large amoun small bilingual dictionary method that can of text. Next, we use n translate missing word an phrase yto learn a linear projec- tures based ries by learning lang nd mapping be At the test time, we tion between the lang translate any word that has been seen in the mono- lingual corpora tion from th on large monoling small bilingual d tween lang ed representation of v water projecting its vector representa- sheep It us leams a linear mapping snaces of languages. effective: we can source language space . Once we obtain target language space, we output vector as the translation. The representations of the target vector vector in the most similar simplicity language our method is 90% nrecision@5 for tran anish. achieve ds between English and nguages are learned ip-gram or Continuous models recently proposed about the ethod makes little as extend and re using the distributed Bag-of-Words (CBC by (Mikolov et al, 2013a). These models learn representations tecture that Because of its simplicity, the Skip-gram and CBOW languages, so it can fine dictionaries and translation tables language pairs. rd ng a simple neural network archi- s to predict the neighbors of a word. alaree amount of text data: WATER SHEEP?! Looks like 'Water sheep" had some mysterious connections with other youtubers way before pewds(3 years to be exact)

WATER SHEEP?! Looks like 'Water sheep" had some mysterious connections with other youtubers way before pewds(3 years to be exact)

Save
Animals, Computers, and Google: Exploiting Similarities among Languages for Machine Translation Ilya Sutskever Google Inc. Mountain View Quoc V. Le Tomas Mikolov Google In Mountain View tmikolov@google.com Inc. Goog View Mount ilyasu@google.com qv1@google.com possible to infer missing stributed representations We achieve this by learning Our study found that it dictionary entries using of words and phrases linear project Abstract Dictionaries and phrase tables are the bas between vector spaces language. The method consis statistical machine sts of two tems. This paper develop erating and ex- automate the p and phrase tables. tending di resent models of simple steps. First, we build mono lar text. Next, we use ages using large amounts small bilingual dictionary tion between the l has been seen in the mon translate any word lingual corpora tion from the source language space n translate missing word entries by learning language learn a lincar projec- uctures hased and mapping be tween languages from small bilingual data representation of linear mapping betwe water large monolin senta- projecting its vector rep sheep target the language Once we obtain the vector target vector as the translation. The representations of languages are learned snaces of languages. Desp simplicity effective: we can our method is sur recision@5 for t ost similar space, we output the mos raisia achieve woods hetween English and his method makes little assu d and re- about the -gram or Continuous using the distributed R of Words (CBOW) models recently propose the days of "hydraulic ram" being translated as "water sheep" translati fine dictio language pairs These models learn word 1 Introduction are pretty much in the Cu machine translation systems have been SIweessful in on dictionaries and h efforts to generate our parallelizeu from billions of words in hours. 1:19/5:04cloned for years and becan These systems rely ate visualization to illustrate Figure 1 give why and how our method works. In Why Computers Suck At Translation ectors for numbers and animals in En- Следующее АВТОВОСПРОИЗВЕДЕНИЕ П W A T E R S H E E P

W A T E R S H E E P

Save
Animals, Cars, and Google: Exploiting Similarities among Languages for Machine Translation Ilya Sutskever Quoc V. Le Tomas Mikolov Google Inc. Mountain Vie Google Inc. ountain View Google Inc. Mountain View 1lyasu@google.com qv1@google.com tmikolov@google.com infer missing Our study found that it is possible dictionary entries using distributed representations of words and phrases. We achieve this by learning a lincar projection between vector spaces that rep resent cach languag simple steps. ! languag l bilingual dictionary to leart Abstract Dictionaries and phrase tables are the basis machine tratslation sys . The method consists e dels of we build monolingual of modern sta puper develops a method ems omate the process of pe cnding diction word and method can trans catries ree monolingsal data and and ex- brase ubles. Our ext, we ase g large amounts of text. lincar projec- a smal hydraulic ning language siructures hd tion between the languages. At the ansdate any word that has bo time, we can ping be- seen in the mono- its vector representa- on larp languages from small bilingual data Hodistributed repres and learns a incas lingual corpora by pro from the source language space t language space. Once w target language sp word vector as the translation The rep using Bag-of words CBOW) models r of words eping between vector simpli the obtain the vector ram the ages Despite spaces issprisingly etior trala chieve almost 90% pr h and Spanish we output the most similar atations of languages are learned Continuous tien of words berwe This method he wed Csi rle assumption about the disributed Skip-gram or the days of "hydraulic ram" being translated ently proposed et al, 2013a) These language pai ecture th as "water sheep" are pretty much in the 1 Introduction Statistical mac ccme very sC developed for ycars. These systems rely on dictionaries anu h efforts to generate Figure 1 CS S why and how our method works e a P mbers and animals in En- Exploiting Similarities among Languages for Machine Translation lya Sutskever Google Inc. Mountain View lyasu@google.com Quoc V. Le Google Inc. Mountain View qv1@google.com Tomas Mikolov Google Inc. Mountain View tmikolov@google.com Our study found that it is possible to infer minsing entations Abstract nries using distributed re dictionary learning and phrases We spaces that reP lincar projection bet ncthod consists of two Dictionaries and p e the basis ranslation sys moders stals ccs a method thal resent cach language. simple steps. Fint languages using large amounts of test. Next. small tween the languages. At the build monolingual models of e use d es- e the process dig dictionaries and p geaera ahies Our word and phrae nethod can anguage strctures cotn lingal dala igal data. projec- gual dictionary to learn a l time, we can water seen in the mono ndate any word that has beo lingual corpora by projectin tion from the so language sp E obtain the vector arget langua word ween languages from we dstri its vector representa age space to the target sheep crion of words lece mapping between ve city oflective we can space, we output the most similar r as the translation languages Despin method is surg s or ra achive a English and The representations of languages are learned using the distributed S ds recently peoposed Baof Words(CBOW bo the Coetinuous eages so c ne dictionary auage pai the days of "hydraulic ram" being translated am a hesmedcls cars as "water sheep" are pretty much in the I Introduction Satistical machine ranlabo developed for years and tme These systems rely uaful in dictionaries aIN fforts to generate on bi Figue v y and how our orks 16 orn for numbers and anima these Help, hydraulic ram is haunting me

Help, hydraulic ram is haunting me

Save
Anime, Parents, and Relationships: l 43%02:15 "I'm not talking to that creep until he quits writing these messed up novels," Fushimi's sister told Anime Maru. "My brother has always been kind of... off and I've tried to tolerate it as best as I can until now. During the whole Oreimo thing he kept insisting that fiction and reality were separate and that it didn't at all reflect any of the his views. In hindsight I probably should have just started avoiding him back then, but now with this Eromanga Sensei anime he's really gone too far." Eromanga Sensei can be considered by some to be spiritual sequel to the Oreimo. The show started a airing this season with many similarities to Fushimi's prior work, mainly being that both of the shows have a focus on sibling relationships. "My sister is just being tsundere" Tsukasa Fushimi stated when questioned by our correspondent. "Any similarities are simply a coincidence. For my stories I am just trying to convey the strongest bonds and purest relationships to my readers. To say this is just a projection of some sort of hidden fantasy or something is ridiculous." "Let me guess, he made them unrelated by blood or something like that right?" Fushimi's sister later added. "The parents probably inexplicably gone are for some reason too. I can see what that weirdo is trying to do. And the little sister draws lewd light novel covers for a living? Give me a break. He's not even trying to be subtle anymore just look at the damn title!" Tsukasa Fushimi later announced his next project, Creator of Eromanga Sensei and other incestuous anime is literally being called a creep by his sister, and he is simply the most perfect example of neckbeard there has ever been. This whole article is hilarious.
Save
Save
The Real, World, and Real World: A The creative similarities between the real world and the other world (Coraline 2009)

The creative similarities between the real world and the other world (Coraline 2009)

Save
Save
Minecraft, Ps4, and Bar: + Typed watersheep into the seed bar in Minecraft ps4. Anyone else see similarities between this and pewds’ place?

Typed watersheep into the seed bar in Minecraft ps4. Anyone else see similarities between this and pewds’ place?

Save
Future, Life, and Party: unDeRSTANDIng The SpecTRum Introducad by Archie! Langиage can be confusing For me. It takes me longer than the average person to process conversations. And although I am good at making conversation, it can take me longer than normal to respond. But, neurotypical people Pind language confusing too. And it can lead to some people mispercieving who I am. That is why I would like to explain what is meant by 'spectrum when we talk about the 'autistic spectrum'. Sometimes when people think of this word, theu think of the autism spectrum as being like this: Ala Very autistic autistic A very linear looking 'spectrum' which gives the impression that From being ittle autistic to very autistic' Hm, How can you be little autistic'? It's that vaaue language that I always Find conFusing You're only a little autistic, Archie The problem with thinking of the spectrum in this way, is that a perception оf an autistic still don't understand, can you be less vague? Hи you're able to have person also becomes linear ation pretty normal! You're not severely autistic Not Very autistic autistic And so you see, someone thinks you're na 'low end' of this spectrum this often happens: Mch NOIE Archie uou сan handle all of this ROUTENE you're not that Autistic. tiqht ing oNT Converasonce TPON'T FIDGET BE HORE CT LIKE EVE Ah Ahh- FAoiBUT CONCE ME IN OX MORE HARDEPAea INFORMAI NORMAL W RWISE AND LOok DOING IT PEOPL A heina so over dramatic, aet over it! Woah, uow're more autistie than I thougkt. GE And if you're seen as being on the high end of this spectrum- Im aonne re-label yow on this spectrum. have a job, just to vknou sate It can lead to some people labeling you ae doing anything at all. Very utistic autistic The truth is though, someone who is neurodiverse in some areas of their brain, will also be no different to your average person in other areas of their brain. Motor ki1l5 Language recutive frction you see, the autistic spectrum looks something more like this. Some traits create The soectrum consists diFFie lie of many dipferent shem do'suReI brain (hence being diagnosed) ir uage Lan processes information. arecutive ftion But also many traits everu day life. Each person with autism in dieferent areas of the soectrum The aneas where theu don't have a trait will Function no dipferently to a neurotupical brain, but may be aPfected by circumstances. In example, I am good at making sory averload in laud and crowded spaces, which then makes conversation very hard For me And so, another autistic person might be very happy in loud crowds, but Find conversation hard in general. you could sau I'm iust a real 'party animal'! Hou can see witk this spectrum than that not every autistic person kas 'savent skills" verbally migkt etill udetend te saying, but just need a dipPerent way to communicate such as sian language It shows how nat ever autistic person acts the same way, and we are all сараble of varying weaknesses Sometimes, if someone is diagnosed as being on the spectrum', and informs әnother person of this, it's so that they can get some understanding and respect For the things they are unable to do. But, it is alsO so that they can cooperate with the world around them- so that they can be the best in the things they can do. I hope that in the Future, people will better understand the term spectrum', and continue to respect the differences and similarities we all share in how we experience the world. Stitch It ec yotd d Sensoc
Save
Computers, Funny, and Ally: THIS KI.... READ THE SECOND PARAGRAPH, AND THEN CHECK OUT HIS GRADE. 4- Nic Job MIDTERM EXAM PART 2 (70 points) Select one of the following qeestions Circle your doice and wrte a 34 pagr cohert, well nferenced essay supporting your desis, Use drect quotes wbes appeopriate and make aure yos soe pecifc eamples YOUMAY USE A COPY OF ANY OF THE TEXTS EIT NO LAPTOP COMPUTERS OPTION A: Many of Shakespeare's works contain similar elements, themes, and plot points Compare and contrast these similarities for two of the assigned plays. Cite specific examples as generalized answers will not receive full credit. OPTION B: (Note: A Midsumimer Night's Dream is not acceptable for this question) For one of the assigned tragedies, explore the feasibility of it occurring in a modern setting. Would advances in technology, communication, and other factors overcome the negative, yet timeless, concepts of human nature displayed in that play? エn t。 Rom eo and Juliet then is of rash aned hastly made decisns. There s an under lyn Motal of prvdence lustteel by t Snabolling effets choices, ltinatly colminatag Suicide monets before Toliet aakes, md a l king is fe befte fnding ot she had done ingngIf Common Sease hod been afplied at preuous es these trag ic chegs mey have never metetializnd Ls all henesty, I on already bered with this tpie. iar less inteeshing than I had heped mdI Wnt to finish this essayIn fairly sufe yor den t ally tead these, sae Tn just gea4 toPot enoush wotds don t make it seem like te alot while I kill time wana heol Sone wrds that rhyme vith Mime (hah minesate funny) ChiMe, lime, fwdude you kow uhat has lne in ? Sprite, its iKe lemon lime. could tally go for one of these abot ew, but tSert Mist het st At the Sae, I+ tres too hald to be sprite but * jst cort Pll it off エ+ shae, セ be tsef, and stap tojias to mtas ore も ather Sosh Ets of Shake spentes most fameus plays Ofhelloand feaccutin hene of the consnces Romco's fa lis h in feally den't time? Crme, dnt
Save
Apparently, Crime, and Dating: writing-prompt-s A dating service where matching is based on people's search history exists. You're a serial killer. You go on a date with a writer. endreams-s Serial Killer: metaphorically, if you were to kill someone, how would you do it? Writer: Air shot between the toes, it'll look like a heart attack. Serial Killer who is obviously in love already: *sucks in a breath* ok fangoddess817 Writer: how long would it take to die if you were to potentially stab someone in the guts Serial killer: anywhere from 2 to 30 minutes Writer, already bringing a ring out: *shaking* thanks infinityonthot A++ addition tetsuskitten Writer: *shows the serial killer the murder scene they're writing* babe, i'm not sure if this would actually work? Serial killer: *kisses writer on the forehead and leaves, comes back later, a suspicious scent of blood coming off them* it works baby, you're doing great tigerliliesandcherryblossoms I LOVE THIS vmohlere Oh no, murder comedy is my jam laziestofthedreamers I love this, I love all of this, but quick question, does the author know? Like are they aware that their significant other is a serial killer or do they just think that they have a morbid sense of humor? It'd be even funnier if the author had no fucking clue, like how Aurthur Conan Doyle was apparently stupidly gullible, and on top of it they're a horror or crime novelist. Like the serial killer works at a butcher shop or something so it's completely normal for them to come home smelling like blood, here, from a long day at work. no murders going sirey. Just my darling coming back home on no Now fast forward a bit and the author has managed to get their first book published, with loving support from the serial killer who helped them fine tune all the murder scenes, and it's a big hit. Enough a detective with the local police department has noticed some disturbing similarities to several active cases, including details that were never released to the press. Obviously he brings this up to his superior and convinces him that there's something to the theory, but it's all circumstantial right now. He stakes out the author's home and is super convinced that the author is the murderer, but they don't seem to do anything??? Like they literally are at the house all day, that's it. Most they do is leave for groceries so that So you get this dynamic of the serial killer mining the author for creative murder schemes, the author being lovingly encouraged by the serial killer, and finally the detective who is just so sure that the author is the killer and that if he sticks it out long enough he'll FINALLY have proof. annieutimagines Plot twist, The serial killer and detective use to go out so it gets sub what personal. "You need to stop seeing them. I think they are a serial killer." Serial killer breaths in. "Look-" Please write this but if the writer is guilty (or any innocent person) I will have to block you
Save