Language, Speech and Multimedia Technologies Observatory

http://www.dataversity.net/natural-language-processing-cant-beat-human-translators-yet/
01/26/2015 - 11:24

by Angela Guess Ofer Shoshan of Beta News recently wrote, “Anyone who has tried to read a Google Translated article knows that machine translations still don’t entirely get things like context, colloquial and nuance. This hysterical rendition of the hit song ‘Let It Go’ -- where the lyrics were Google Translated into another language and…

The post Natural Language Processing Can’t Beat Human Translators (Yet) appeared first on DATAVERSITY.

http://andonisagarna.blogspot.com/2015/01/googlek-irudien-oinak-automatikoki.html
01/26/2015 - 11:24

Iturria: MIT Technology Review 

Googleko ingeniariek ikasketa automatikoko algoritmo bat irudien oinak idazteko prestatu dute, itzulpen automatikorako erabiltzen dituzten teknika berberak baliatuz.

Sistema berri horrek automatikoki sortzen ditu irudien edukiak zehazki deskribatzen dituzten irudien oinak. Hori erabilgarria izango da bilatzaileetan erabiltzeko eta ikusmen-arazoak dituztenei Interneten nabigatzeko, nahiz mundu errealean ibiltzen laguntzeko.

Itzulpenak egiteko planteamendu konbentzionala hitzak banan-banan itzuliz hasten den prozesu iteratibo bat da eta gero hitzak eta esaldiak berrordenatzen dituena, itzulpena hobetzeko. Alabaina, azken urteotan, Googlek bere bilaketen datu-base erraldoia testuak beste era guztiz desberdin batez itzultzeko landu du.

Sistema horrek funtsean egiten duena da hitzak beste hitz batzuen ondoan edo hurbil zenbat bider agertzen diren zenbatu eta ondoren haiek erlazionatzen dituen bektore-espazio bat definitu. Horrela hitz bakoitza espazio horretako bektore batek ordezkatzen du eta esaldiak bektoreen konbinazioek ordezkatzen dituzte.

Horren oinarrian badago hipotesi bat: hitz jakinek erlazio berbera dutela beste hitzekin, edozein delarik ere hizkuntza. Adibidez, "errege - gizonezko + emakume = erregina" egia izango litzateke hizkuntza guztietan. Horren arabera, hizkuntzatik hizkuntzara itzultzea bektore-espazioen matematikako kontua da.

Google Translatek esaldi bakoitza bektore bihurtzen du eta gero bektore hori beste hizkuntza bateko esaldi baliokidea ekoizteko erabiltzen du.

Ikerketa horretan diharduen taldeak Google Brain deritzon proiektu bat garatu du, ikasketa sakoneko teknikak garatzeko, besteak beste, ordenagailuei giza hizketa ezagutzea irakasteko helburuarekin.

Talde horretan lan egiten duen zientzialarietako bat Oriol Vinyals kataluniarra da. Oriol Sabadellen jaio zen duela 31 urte, Telekomunikazioetako Ingeniaritza ikasi zuen Bartzelonako Universitat Politècnica de Catalunyan, formazio hori Centre de Formació Interdisciplinària Superior delakoan osatu ondoren, karrera-bukaerako proiektua AEBetako carnegie Mellon University-n egin zuen, 9 hilabeteko egonaldi bat egin zuen Berkeleyko Unibertsitatean, ikasketa automatikoan espezializatzeko, beste urtebete eman zuen San Diegon Informatikako master bat egiten eta 2009-2013 tartean doktoretza egiten, berriro ere Berkeleyn. Denbora horretan guztian, udako hilabeteetan egonaldiak egin zituen Microsoften eta Googlen, zenbait ikerlanetan parte hartzen.


Oriol Vinyals

Orain Googleko taldeak sare neuronal bat erabili du, 100.000 irudiko datu multzo bat eta irudi horien oinak aztertu eta irudien edukia nola sailkatu ikasteko. Baina irudia deskribatzen duten hitzen multzoa sortu ordez, haien algoritmoak hitzen arteko erlazioa ordezkatzen duen bektore bat sortzen du. Bektore hori gero Googleren itzulpen-algoritmoan txerta daiteke ingelesezko edo beste edozein hizkuntzatako oina ekoizteko. 

Prozedura horren eraginkortasuna probatzeko, giza ebaluatzaileak erabili zituzten, automatikoki sorturiko oinak ebaluatu zituztenak, beste prozedura automatiko batzuek eta gizakiek sorturiko oinekin konparatuz.

Proba horien emaitzek erakutsi dute sistema berriak, Googlek Neural Image Caption deritzonak, ondo funtzionatzen duela.

http://semanticweb.com/computer-neural-networks-now-effective-identifying-objects-primates_b45431
01/26/2015 - 11:24

8705693329_eb8cefeb1fDiana Depra of Tech Times reports, “For years, neuroscientists have been hard at work designing computer networks imbued with the capability to mimic visual skills that the human brain efficiently does. In a study, researchers from the Massachusetts Institute of Technology finally found a computer model called ‘deep neural networks’ that can match a primate’s brain in recognizing objects. According to James DiCarlo, Department of Brain and Cognitive Sciences head at MIT and the study’s senior author, recent network creations are based on whatever current understanding neuroscientists have when it comes to the brain’s object recognition capability.” continued…

http://semanticweb.com/know-new-skype-translator-app_b45428
01/26/2015 - 11:24

skypeJohn Brandon of Fox News reports, “Skype Translator may be the biggest advancement in real-time language translation since the babelfish on ‘Hitchhiker’s Guide to the Galaxy.’ The app, currently in a limited trial for Windows 8.1 and for English and Spanish speakers only, listens to what you say on a Skype video or audio call and then translates immediately. It can also translate the text you type during a Skype instant messaging chat. The chat works for over 40 languages beyond Spanish including German, French, Arabic and even Klingon.The technology that makes this happen involves a third-party: a robot. When you place a call, a translation engine serves as an interpreter. It literally joins the call and processes what you say.” continued…

http://semanticweb.com/baidu-research-achieves-speech-recognition-breakthrough-deep-speech_b45391
01/26/2015 - 11:24

Baidu logoSUNNYVALE, CA, Dec 18, 2014 (Marketwired via COMTEX) — Baidu Research, a division of Baidu, Inc. today announced initial results from its Deep Speech speech recognition system.

Deep Speech is a new system for speech, built with the goal of improving accuracy in noisy environments (for example, restaurants, cars and public transportation), as well as other challenging environments (highly reverberant and far-field situations).

Key to the Deep Speech approach is a well-optimized recurrent neural net (RNN) training system that uses multiple GPUs, as well as a set of novel data synthesis techniques that allowed Baidu researchers to efficiently obtain a large amount of varied data for training. continued…

http://www.unibertsitatea.net/blogak/ixa/2014/12/17/erabili-xuxen-sarean-estra-batzuekin-xuxen-eus/
01/26/2015 - 11:24

Puntueus fenomenoarekin batera xuxen.eus azaldu da.

Durangoko azokan aurkeztu zen.

Betiko zuzentzailea da, baina erabil dezakezu edonon, edozein gailurekin (mugikorra tartean) eta gainera estra batzuekin:

  • Hitzen forma hobetsiak proposatzen ditu.

    Ez da errorea baina forma hobetsi bat definitu du Euskaltzaindiak.   ;-)
  • Zeure hiztegi propioa ere erabiliko du zuzenketan nahi baduzu.  :-)
  • Interneten duten presentziaren arabera ordenatuko dizkizu hitz oker bat zuzentzeko proposamenak.
  • Elhuyar hiztegietan bilatuko dizu hitz baten gainean klik bikoitza egitean.
  • Testua entzuteko aukera ere baduzu.

xuxeneus2Webgune horretan Xuxen-ek dituen hainbat aplikaziotarako bertsioei buruzko informazioa ere baduzu:

Laister, hilabete batzuen bueltan, hobekuntza gehiago. Aupa Xuxen!

Zalantzak argitzeko, jarri harremanetan Elhuyarrekoekin

http://semanticweb.com/google-adds-dose-machine-learning-nlp-hangouts_b45353
01/26/2015 - 11:24

Google LogoStephen Shankland of CNet recently reported, “Google updated its Hangouts app on Wednesday so the multipurpose communication tool can detect when people are trying to find each other and make it easier to connect. The updated app, available for Google’s Android mobile operating system first but submitted to Apple for approval on iOS, also lets people express themselves with stickers and video filter effects, Bradley Horowitz, vice president of product at Google, said at the LeWeb conference here.”

Shankland continues, “The new app works by using natural-language processing to discover when one person has asked — in a variety of ways — where the other person is. continued…

http://semanticweb.com/will-artificially-intelligent-robots-destroy-humanity-stephen-hawking-says-maybe_b45283
01/26/2015 - 11:24

7642344074_5db123eab7Katherine Noyes of Tech News World reports, “Artificial intelligence eventually could bring about mankind’s demise, renowned physicist Stephen Hawking said in an interview published earlier this week. ‘The primitive forms of artificial intelligence we already have have proved very useful, but I think the development of full artificial intelligence could spell the end of the human race,’ Hawking told the BBC in an interview commemorating the launch of a new system designed to help him communicate.”

Noyes goes on, “Because he is almost entirely paralyzed by a motor neuron disease related to amyotrophic lateral sclerosis, Hawking relies on technology to communicate. continued…

http://semanticweb.com/quoc-le-teaching-machines-understand-emotions-deep-learning_b45263
01/26/2015 - 11:24

Google LogoDaniela Hernandez of Wired recently wrote that Quoc Le “works on the Google Brain, the search giant’s foray into ‘deep learning,’ a form of artificial intelligence that processes data in ways that mimic the human brain—at least in some ways. Le was one of the main coders behind the widely publicized first-incaration of the Google Brain, a system that taught itself to recognize cats on YouTube images, and since then, the 32-year-old Vietnam-native has been instrumental in helping to build Google systems that recognize your spoken words on Android phones and automatically tag your photos on the web, both of which are powered by deep-learning technology.” continued…

http://semanticweb.com/googles-process-translating-pictures-words_b45240
01/26/2015 - 11:24

Google LogoThe MIT Technology Review recently wrote, “Translating one language into another has always been a difficult task. But in recent years, Google has transformed this process by developing machine translation algorithms that change the nature of cross cultural communications through Google Translate. Now that company is using the same machine learning technique to translate pictures into words. The result is a system that automatically generates picture captions that accurately describe the content of images. That’s something that will be useful for search engines, for automated publishing and for helping the visually impaired navigate the web and, indeed, the wider world.” continued…

Syndicate content