We can push buttons, we can touch screens, but the future will be the ability to tell computers what we want them to do with our voice.
I hope this future will come in my lifetime as much as the next guy, though so far every time I try speech-to-text and the likes of speech-command software I use it optimistically for the first few days, then give it up.
Context aware software with speech recognition might be about to change all of that. And maybe it's going to take another garage startup (think Apple) to make it all fall into place.
But 8 hours of work seems like too small an effort if you're wanting to compete with Siri.
Essentially all they have is a glorified search engine that accepts speech instead of text and gives you a summary of one of the results. This isn't Siri, or anything close.
I agree, except I think there are places for voice, and places for touch. Things that would take multiple steps to accomplish, e.g. "remind me to take out the trash when I get home," is an awesome application for voice, but for choosing from a menu of a bunch of items, I want to just select the item.
My wife got a 4S, and from the few days we've been playing with it, we agree that it's sort of still in the uncanny valley of AI. It's too hard to tell what's going to work and what's not, esp. for the technically lay.
It's amazing, though, for Apple to have opened everyone's eyes (ears?) once again wrt to what direction the near future will take.
And the more I think of it, the less I am convinced. For instance, the hole Chinese-speaking world cannot conveniently use speech to communicate with a machine, because of Chinese language structure. If Siri do not understand an English name I am telling it, I can always "spell" it. Not in Chinese. You can't conveniently "spell" a Chinese character, and even if you could (using wubi method), you would still need a visual feedback to confirm the correct character has been chosen.
Nowadays, any technology that can't scale well to other languages than English has a profound problem, I suppose.
Every tv-show with any amount of tech that I watched when I was a kid in the 90s featured a computer that talked and could be talked to. Plenty of people hoping this will become a reality, even if only for languages using letters (ie. not Chinese).
Any complex task that is somehow generic can easily be done with voice control and makes the UI needed for simpler tasks to be more concise and less bloated.
For instance setting the alarm clock. You could scroll through your apps looking for the clock app. Open it (ie. wait for it to load), then navigate to the alarm clock tab, create a new alarm, set the correct time and hit save. (on my nokia phone this is actually faster than saying saying the word "alarm", but my ipod touch is slow as hell)
On a computer with a keyboard a lot of this can be sped up because I can search for the application from the keyboard, I can tab through fields and use shortcuts to do things (like ctrl-n for new). But my mum can never remember any shortcuts, and she can't type very fast. And if you think that the new generation is any better...then have a look at anyone studying in the human sciences department...most of them have trouble enough using word.
My point is, if something doesn't require perfect speech recognition, is a complex yet generic task...then voice control seems like a very good solution.
I agree that this is likely missing some pretty important pieces.
Due to all the recent talk about Apple and Siri I decided to look into Siri the startup a little more closely. That's the company that Apple bought to bring you Siri the feature. I was curious how much of this technology was created by the startup and how much Apple contributed. While I didn't find the answer to that question, I did discover something even more startling.
It turns out that Siri the company was born out of one of the largest computer science research programs in U.S. history. Or so claims the press release [1]. Looking at the the research project page [2], it appears to involve at least seven major universities, a lot of government funding, and several private corporate collaborators. While some of the pieces may be open, the license for the project itself only allows for non-commercial use.
It's also fantastically complex and combines many state of the art techniques. While it's certainly possible for a company to reproduce much of the implementation from the open pieces, it is a lot more than 8 hours of work.
All said, Siri is an amazing accomplishment, and more than anything, a great example of why science funding is important.
In all seriousness, they probably should have called it something other than Iris. I'm not a fan of Android these days (since my Galaxy S died on me at a really bad time), but this name will (most likely) fuel the fire of Android being a total rip of iPhone.
Of course, I might just be reading too far into the name thing. Their humor is pretty obvious.
Not going into details, but yes, I was/am pretty mad at Google and Samsung. Seven months of use followed by a total loss of data left me with a negative attitude.
Fortunately for the Android brand, the bulk of the phone market doesn't read Techcrunch and will never even notice the existence of this hack, however cute its name is.
The same goes for all the cheap-and-cheery Siri-ish hacks on the iOS App Store, of which I'm sure there are several.
"SpeakToIt" seems a lot like Siri. It seems a lot of companies were already doing this. It's just now that Apple pushed it to its new iPhone that people are really starting to get interested in them.
SpeaktoIt seems to have more personality based off responses it gives. It is not as serious as Siri. Though Siri does have its Easter eggs too though less so then speak to it. Both are equally useful and enjoyable and have their strengths and weaknesses. Siri is built into iPhone and it's speech is superior to speak to it.
Context aware software can be built only when rich repositories of background knowledge are available. They don't come cheap. Either you need hordes of linguists and domain experts or you get garbage. One can try using OpenCyc, FrameNet, WordNet or UMLS but it ain't easy. You might as well spent the time on writing Q/A templates for your bot. Hurdles, hurdles, hurdles.
Nonetheless, Watson and Apple showed that there is something shiny on the other side of the rainbow.
"Suddenly, I got the urge to do something similar for Android. Since we have been working on NLP and Machine learning for over an year now, I had a crazy belief that I could pull this off. " (from http://blog.dexetra.com/a-day-when-siri-inspired-us-to-creat...)
It's cool that they banged up a cool app in a very short time but to think that this in anyway can compete with probably hundreds of man-years of research that went into CALO (what became Siri) and, of course, with Apple's polish is insane (just check these funny Siri comebacks to see this: http://pocketnow.com/iphone/funny-things-siri-says-screensho...).
Pushing half-baked apps as competitors to those from Apple with similar functionality will only push more people to iPhones.
in just 8 hours, they built the app…having already spent more than a year on natural language processing and machine learning. in 8 eight hours they have a working but unpolished simple app that plugs into a lot of research. does anyone else find the sensationalism absurd? "just 8 hours"
and a question for HN—what is this fascination with techcrunch? if there is a story out on the interwebs and techcrunch has covered it, it's usually at the top. is it because they do venture capital pieces? is it because green is (was) the color of money? is it because michael arrington held some sort of strange fascination by tech culture at large that has remained now that he's been ousted?
Android has a great voice recognition technology.It works like a dream. Also it's pretty easy to integrate voice-to-text into your applications and again it's very easy to launch apps/do google searchs/open a map/etc in Android with Intents.
I've been playing with command voices and writing emails etc since maybe more than a year ago and if you have a little training in NLP (or even if you don't) you realize it's pretty easy to get your Android device to return wikipedia data in a pretty layout or get a few simple "natural language"-like commands working.
But what's amazing in Apple's system is not sending an email by saying "email to...". The really cool and difficult part is how the handle paraphrasing and uncertainty. Paraphrasing ("how's the weather today?" x “What is the weather like today?”) and ambiguity resolution ("What's up next" -> shows your next appointments) are still very difficult problems in NLP and even though research has improved lots[1][2] it's still totally non-trivial. The way they got it to work in the phone is in a total different level from what those guys working 8 hours achieved.
Reminds me of this blog post, "How to clone Delicious in 48 hours"[1]. It's cool that these people took their existing work and put a mockup of Siri's UI over it, but getting it to work smoothly and be useful is going to take a lot longer than 8 hours.
Besides, Android doesn't even need this, right? I thought it already had a Siri equivalent, and Siri represents Apple yet again playing catchup. Right?
On the other hand, Siri has supposedly one of the largest engineering teams at Apple. If a small shop can beat that team at their own game, that would be pretty neat to see. But, I am not holding my breath.
26 comments
[ 2.6 ms ] story [ 73.4 ms ] threadI hope this future will come in my lifetime as much as the next guy, though so far every time I try speech-to-text and the likes of speech-command software I use it optimistically for the first few days, then give it up.
Context aware software with speech recognition might be about to change all of that. And maybe it's going to take another garage startup (think Apple) to make it all fall into place.
But 8 hours of work seems like too small an effort if you're wanting to compete with Siri.
Essentially all they have is a glorified search engine that accepts speech instead of text and gives you a summary of one of the results. This isn't Siri, or anything close.
My wife got a 4S, and from the few days we've been playing with it, we agree that it's sort of still in the uncanny valley of AI. It's too hard to tell what's going to work and what's not, esp. for the technically lay.
It's amazing, though, for Apple to have opened everyone's eyes (ears?) once again wrt to what direction the near future will take.
It is hard to tell. I am not convinced that the near future is in natural language and voice control. I stated why here http://news.ycombinator.com/item?id=3119528.
And the more I think of it, the less I am convinced. For instance, the hole Chinese-speaking world cannot conveniently use speech to communicate with a machine, because of Chinese language structure. If Siri do not understand an English name I am telling it, I can always "spell" it. Not in Chinese. You can't conveniently "spell" a Chinese character, and even if you could (using wubi method), you would still need a visual feedback to confirm the correct character has been chosen.
Nowadays, any technology that can't scale well to other languages than English has a profound problem, I suppose.
Any complex task that is somehow generic can easily be done with voice control and makes the UI needed for simpler tasks to be more concise and less bloated.
For instance setting the alarm clock. You could scroll through your apps looking for the clock app. Open it (ie. wait for it to load), then navigate to the alarm clock tab, create a new alarm, set the correct time and hit save. (on my nokia phone this is actually faster than saying saying the word "alarm", but my ipod touch is slow as hell)
On a computer with a keyboard a lot of this can be sped up because I can search for the application from the keyboard, I can tab through fields and use shortcuts to do things (like ctrl-n for new). But my mum can never remember any shortcuts, and she can't type very fast. And if you think that the new generation is any better...then have a look at anyone studying in the human sciences department...most of them have trouble enough using word.
My point is, if something doesn't require perfect speech recognition, is a complex yet generic task...then voice control seems like a very good solution.
Due to all the recent talk about Apple and Siri I decided to look into Siri the startup a little more closely. That's the company that Apple bought to bring you Siri the feature. I was curious how much of this technology was created by the startup and how much Apple contributed. While I didn't find the answer to that question, I did discover something even more startling.
It turns out that Siri the company was born out of one of the largest computer science research programs in U.S. history. Or so claims the press release [1]. Looking at the the research project page [2], it appears to involve at least seven major universities, a lot of government funding, and several private corporate collaborators. While some of the pieces may be open, the license for the project itself only allows for non-commercial use.
It's also fantastically complex and combines many state of the art techniques. While it's certainly possible for a company to reproduce much of the implementation from the open pieces, it is a lot more than 8 hours of work.
All said, Siri is an amazing accomplishment, and more than anything, a great example of why science funding is important.
[1] http://www.sri.com/news/releases/020510.html
[2] https://pal.sri.com/Plone/framework
Of course, I might just be reading too far into the name thing. Their humor is pretty obvious.
The same goes for all the cheap-and-cheery Siri-ish hacks on the iOS App Store, of which I'm sure there are several.
For stupid people, yes. Who cares.
http://www.youtube.com/watch?v=myE498nyfGw
Nonetheless, Watson and Apple showed that there is something shiny on the other side of the rainbow.
"Suddenly, I got the urge to do something similar for Android. Since we have been working on NLP and Machine learning for over an year now, I had a crazy belief that I could pull this off. " (from http://blog.dexetra.com/a-day-when-siri-inspired-us-to-creat...)
It's cool that they banged up a cool app in a very short time but to think that this in anyway can compete with probably hundreds of man-years of research that went into CALO (what became Siri) and, of course, with Apple's polish is insane (just check these funny Siri comebacks to see this: http://pocketnow.com/iphone/funny-things-siri-says-screensho...).
Pushing half-baked apps as competitors to those from Apple with similar functionality will only push more people to iPhones.
and a question for HN—what is this fascination with techcrunch? if there is a story out on the interwebs and techcrunch has covered it, it's usually at the top. is it because they do venture capital pieces? is it because green is (was) the color of money? is it because michael arrington held some sort of strange fascination by tech culture at large that has remained now that he's been ousted?
i humbly await the downvotes and replies.
thanks
hacker n00b
But what's amazing in Apple's system is not sending an email by saying "email to...". The really cool and difficult part is how the handle paraphrasing and uncertainty. Paraphrasing ("how's the weather today?" x “What is the weather like today?”) and ambiguity resolution ("What's up next" -> shows your next appointments) are still very difficult problems in NLP and even though research has improved lots[1][2] it's still totally non-trivial. The way they got it to work in the phone is in a total different level from what those guys working 8 hours achieved.
[1] http://www.nist.gov/tac/2011/RTE/index.html
[2] http://arxiv.org/abs/0912.3747
Besides, Android doesn't even need this, right? I thought it already had a Siri equivalent, and Siri represents Apple yet again playing catchup. Right?
On the other hand, Siri has supposedly one of the largest engineering teams at Apple. If a small shop can beat that team at their own game, that would be pretty neat to see. But, I am not holding my breath.
[1] http://notes.torrez.org/2010/12/learn-to-program-in-24-hours...