How Skype Used AI to Build Its New Language Translator(wired.com)
wired.com
How Skype Used AI to Build Its New Language Translator
http://www.wired.com/2014/12/skype-used-ai-build-amazing-new-language-translator/
4 comments
I would be really surprised if this kind of software could be used accurately in any way. Especially on the vocal recognition part. As any kind of neural network technology, vocal recognition needs a lot of training input to work properly. As the web is primarily a text-based medium, this kind of technology works very well in languages where the spoken language is close to the written one (that's why it works quite well in English for example) but it's going to be really hard for languages where the spoken & written forms have taken a completely separated route.
"Pakistan may ban Skype in Karachi, citing use by terrorists" - The Washington Post
"Facebook, Gmail, Skype face Russia ban under 'anti-terror' data snoop law" - ZDNet
"Terrorists move to Skype, frustrate eavesdroppers" - The Times of India
"Pakistan's Sindh Province To Ban Skype, Viber For 3 Months of Terrorist Usage" - TechCrunch
"Talk Like a Terrorist: Use Skype" - United Liberty
"U.S. Jihadis being recruited by ISIS via Skype" - NY Daily News
Among the first languages to get speech support: Arabic, Russian, Chinese. (A full list is here: https://translator.skype.com/#/home)
Honestly, I don't know how widely speech-to-speech will be used (how many people are trying to Skype with someone who exclusively speaks a language they don't?). But I do know how speech understanding will bolster Skype, one of the major PRISM programs. Foreign Intelligence Surveillance will gain a lot of speed and drastically reduce costs by automating most foreign intelligence collection.
"Facebook, Gmail, Skype face Russia ban under 'anti-terror' data snoop law" - ZDNet
"Terrorists move to Skype, frustrate eavesdroppers" - The Times of India
"Pakistan's Sindh Province To Ban Skype, Viber For 3 Months of Terrorist Usage" - TechCrunch
"Talk Like a Terrorist: Use Skype" - United Liberty
"U.S. Jihadis being recruited by ISIS via Skype" - NY Daily News
Among the first languages to get speech support: Arabic, Russian, Chinese. (A full list is here: https://translator.skype.com/#/home)
Honestly, I don't know how widely speech-to-speech will be used (how many people are trying to Skype with someone who exclusively speaks a language they don't?). But I do know how speech understanding will bolster Skype, one of the major PRISM programs. Foreign Intelligence Surveillance will gain a lot of speed and drastically reduce costs by automating most foreign intelligence collection.
That's brilliant deduction there. Among the languages to get speech support are also German, Italian and Japanese, so the NSA also still has one eye on the old axis foes as well. Then Spanish and Portuguese. wait a minute - isn't this list remarkably similar to a list of the most widely spoken languages in the world?
Cantonese, Italian, Korean stand out as smaller languages. Hindi isn't present despite being the fourth largest language in the world. Bengali, Punjabi, Javanese, Wu, Malay, Telugu, and Vietnamese are all more popular than Korean, French and Marathi, Tamil, Urdu, Persian and Turkish still more popular than Italian, Cantonese.
I believe there is a win-win here. Of course Microsoft wants a feature that applies as much as possible to its market (which appears from the list of languages not to include Eastern languages outside of Chinese and specifically excludes all of India). The US government wants as much foreign intelligence as it can get, so it makes sense to focus on more popular languages.
Do you believe that speech understanding will not be incorporated into the Skype PRISM program?
I believe there is a win-win here. Of course Microsoft wants a feature that applies as much as possible to its market (which appears from the list of languages not to include Eastern languages outside of Chinese and specifically excludes all of India). The US government wants as much foreign intelligence as it can get, so it makes sense to focus on more popular languages.
Do you believe that speech understanding will not be incorporated into the Skype PRISM program?
"Eastern languages outside of China" - except for Japanese and Korean, which were listed. India has the benefit of widespread English use across all its language communities. The lack of a widely spoken language like Farsi on the list argues AGAINST your claim that this list is targeted to cultures of interest to the NSA.
I'm just not sure I see any relevance to Prism at all here. Does Skype have to offer instantaneous translation in order to permit multilingual transcription of calls? No - if PRISM has that capability, it has that capability, independently of translation being provided to users.
I guess you can argue that if people are using the translation service, they will maybe speak slower, and repeat and reiterate what they are saying until it is communicated clearly by the translator, so the quality of the transcript may be higher - that's maybe valid, but obviously only applies to conversations that are actually mediated by the translator. A skype conversation in Arabic becomes no easier to transcribe by virtue of the existence of this service. Creating this service, in fact, potentially reveals the extent of their speech-to-text capability - unless they have a greater capability behind the scenes, in which case - again, what does this have to do with PRISM? And if you approach this with a paranoid mindset, surely in theory this technology makes it EASIER for non-English-speaking bad actors in countries of interest like Syria to communicate and recruit vulnerable English-speakers in the US, which strikes me as counter to the interests of homeland security.
A more positive view of this would be that anything which permits easier communication across language barriers facilitates global understanding, trade, co-operation and cultural exchange, and that long term that is the only thing which is going to end the cultural clash which drives terrorism and global conflict. So in some ways, it would be nice if this was part of a long-term US foreign policy initiative aimed at making the world a more harmonious place.... but I think it's more likely that it's just a way for Microsoft to expand the market of their product.
I'm just not sure I see any relevance to Prism at all here. Does Skype have to offer instantaneous translation in order to permit multilingual transcription of calls? No - if PRISM has that capability, it has that capability, independently of translation being provided to users.
I guess you can argue that if people are using the translation service, they will maybe speak slower, and repeat and reiterate what they are saying until it is communicated clearly by the translator, so the quality of the transcript may be higher - that's maybe valid, but obviously only applies to conversations that are actually mediated by the translator. A skype conversation in Arabic becomes no easier to transcribe by virtue of the existence of this service. Creating this service, in fact, potentially reveals the extent of their speech-to-text capability - unless they have a greater capability behind the scenes, in which case - again, what does this have to do with PRISM? And if you approach this with a paranoid mindset, surely in theory this technology makes it EASIER for non-English-speaking bad actors in countries of interest like Syria to communicate and recruit vulnerable English-speakers in the US, which strikes me as counter to the interests of homeland security.
A more positive view of this would be that anything which permits easier communication across language barriers facilitates global understanding, trade, co-operation and cultural exchange, and that long term that is the only thing which is going to end the cultural clash which drives terrorism and global conflict. So in some ways, it would be nice if this was part of a long-term US foreign policy initiative aimed at making the world a more harmonious place.... but I think it's more likely that it's just a way for Microsoft to expand the market of their product.
"Privacy is not dead: Microsoft lawyer prepares to take on US government" http://www.theguardian.com/technology/2014/dec/14/privacy-is...
Related?
HN Discussion: https://news.ycombinator.com/item?id=8750751
Related?
HN Discussion: https://news.ycombinator.com/item?id=8750751
Undoubtedly related in some way, but it's hard to tell how much and in which direction. Microsoft is a huge organization and for the most part (from what we've seen from their behavior and from Snowden documents) they are for the most part on board with the US using the tech sector and their technology to gain intelligence and decision advantages overseas.
Essentially Microsoft wants the best of BOTH worlds. It wants to comply with government requests and it wants to win on the technological front.
Microsoft does NOT want to lose customers and their unique place in the market, however. It's going to be a very careful balancing act for them to both comply with US government requests and to maintain a veil of innocence to both domestic and foreign customers. So far the damage to their reputation could have been a lot worse and their PR has more than proven its worth.
I feel Microsoft is fighting on both fronts.
Essentially Microsoft wants the best of BOTH worlds. It wants to comply with government requests and it wants to win on the technological front.
Microsoft does NOT want to lose customers and their unique place in the market, however. It's going to be a very careful balancing act for them to both comply with US government requests and to maintain a veil of innocence to both domestic and foreign customers. So far the damage to their reputation could have been a lot worse and their PR has more than proven its worth.
I feel Microsoft is fighting on both fronts.
Maybe they should use AI to block the non-stop spammers I seem to get on the service that drained my account of credit.