Item — HN Reader

The really unique thing about Siri is the flexible natural language processing. You can say "Meet with Doug for dinner today", and it'll parse out the person/date/subject, realize you didn't specify a time, and ask you "OK, what time would you like to meet him at?" It's just a much more forgiving interface, which makes it more approachable for the average user.

It's actually a very hard technical problem to create software that not only parses the user's intention, but can intelligently prompt for follow-ups. Google has many of the individual pieces of technology, but making it all fit together, with a good interface, is still hard work that takes time. That's probably why it took Apple 1.5 years to go from the Siri acquisition to the announcement today.

3 comments

I agree. I think what Apple is releasing in Siri could be bigger than most of us realize. I have several android phones and don't use the voice recognition much because 1 out of every 4 or 5 tries it returns things I didn't mean. That's frustrating. The average user quits after a 20% failure rate. Siri on the iPhone 4s needs to be at 5% or less failure rate, meaning one of every 20 requests returns an error. But then again, if it returns an error at least the user can clarify and continue the conversation. I think Apple just leapfrogged Google in natural language processing and taking it to the masses.

as someone who works a lot with speech recognition: speech recognition demos differently from how it works in the field. the demoer gets to remove all of the semantic noise from the system (people who aren't Phil Schiller or their wife, choosing "Greek restaurants" instead of a phrase easily confusable with another), leaving only signal.

if it works the way it says on the box -- and remember, Apple already released Voice Control, which has a less-than-stellar reputation -- then it could be revolutionary. voice recognition often doesn't.

I agree it's a big "IF". IF Siri works as advertised, then it really could be revolutionary. I remember being pumped up by Google's demo of voice recognition, only to try it out and find it novel and cool but not dependable and accurate enough to use in all my real-life situations. And it's difficult to correct errors, or clarify requests in a conversation manner like Siri.

Another reason to hope is that Siri had 19 people when Apple acquired it and most of them are still at Apple. I would imagine that Apple scaled their team significantly. Who knows? Maybe 100 engineers working on it, since it's a cornerstone of the next generation of mobile devices (and probably coming to desktop in the near future too). But Google doesn't seem to have the same priority on natural language processing as Apple does cause more than 50% of Apple's revenue comes from the iPhone. How many engineers at Google are working on voice recognition and natural language processing? Maybe somebody here will know. Maybe max 10 engineers?

Apple also will be forced to scale Siri across multiple languages very quickly, especially if it works well. Currently they have English, French and German. But tons of people will want it, so that motivates Apple to innovate even more.

I guess we'll see very soon how good Siri on the iPhone 4s really is.

it's not an apples-to-apples (heh.) comparison, because Google builds their own speech recognition engine and Siri/Apple licenses the engine of a company called Nuance (at least, last I checked), and so their language scaling is limited by what Nuance can give them.

You're assuming that they're not able to augment/enhance the Nuance engine with their own improvements.

unless Siri has undergone a lot of changes since the last time I looked at it, there is a sharp line between speech recognition and determining user intent/question answering; it was founded as CALO (http://en.wikipedia.org/wiki/CALO), a program which didn't even do speech recognition.

speech recognition maps a space of waveforms onto a space of utterances in a language. determining user intent maps that utterance onto the space of actions.

Siri has done great things in the field of the latter; they license their technology for doing the former from another company.

How many things can it really do, though? Are those not some preprogrammed use cases, and for the rest it falls back to Wolfram Alpha, which has been around for years?

Google also has offered "special" search results for a while, such as evaluating calculations or exchange rates on the fly. I don't think what Siri does is more advanced than that. It is a nice touch to integrate it more deeply into the OS, like "get me home" - but then that is also just one button with a popular Android app (Öffi for example).

Google Calendar also does some "real text parsing" afaik.

Can't really think about a more complicated example atm. OK maybe "set up a dinner with Doug and Sandy, where Doug is a vegan and Sandy doesn't like to go farther than 2 km from her home" - can Siri parse that and pick a suitable restaurant?

I need some ideas for birthday presents for my child, can Siri help?

Have you used Android voice recognition? It's pretty good in terms of recognition. It isn't so proactive as "Go to my calendar and make me money!" but it's quite an adequate little application.

It isn't just voice recognition. It's also natural language processing. Apple just leapfrogged Google.

Google's service also does some natural language processing. You can ask it to "navigate to <restaurant>" and it will look up the restaurant address near you and start up google navigation to that address. In other cases, google's search engine does the NLP. You can ask "What is the time in Paris, France?" or "What is the weather in New York?" and it will give you the answer at the top with search results below.

They are similar services, but Apple seems to have a greater variety of intents that it recognizes.