Slashdot Mirror
Is Google's Future: Star Trek?
An anonymous reader writes "ZDNet UK has an interview with Google's CTO, Craig Silverstein, and he's got some pretty cool visions: "When search grows up, it will look like Star Trek: you talk into the air ("Computer! What's the situation down on the planet?") and the computer processes your question, figures out its context, figures out what response you're looking for, searches a giant database in who-knows-how-many languages, translates/analyses/summarises all the results, and presents them back to you in a pleasant voice." Now that's the search engine I want." The NLP required for this is far off, but it sure will be cool when we get there.
Disclaimer: I did write one of the papers.
Data "never" used contractions (although in earlier episodes they either had not established this, or forgot a few times), but I don't believe the computer (or Data) had problems understanding contractions. Also, Lore (Data's evil brother, with emotions) used contractions, and this was a point in one of the Lore episodes (I believe the original, Datalore) when Lore turned Data off and was posing as him.
The preferred interpretation of NLP is "natural language processing."
Natural Language Processing or voice recognition.
I guess that there are still those amongst us that insist on trying to supplement their inadequacies by babbling in acronyms.
I've always said that if you think it's cool or leet to speak using acronyms, you should go all out and speak in hieroglyphics.
More context for that quote:
"When search grows up, it will look like Star Trek: you talk into the air ("Computer! What's the situation down on the planet?") and the computer processes your question, figures out its context, figures out what response you're looking for, searches a giant database in who-knows-how-many languages, translates/analyses/summarises all the results, and presents them back to you in a pleasant voice. I think this technology is about, oh, 300 years off. Just getting the computer to understand your question, much less the context it's being asked in, is way beyond the state of the art in computer science right now."
There are no trails. There are no trees out here.
This is not an NLP problem.
It is an AI problem.
Voice Recognition is essential in hospitals and courts all over the US.
It is a multi-million dollar a year buisness.
As for the future of speech recognition, people often make the mistake of seeing it as a replacement for all other input interfaces. This is just absurd, as you point out.
More and more products (like cell phones, car GPS, ATM's, LARAN ) use speech recognition in conjunction with other more traditional means of input/output to get the job done.
It is NOT the fastest or the most reliable means of interfacing, but it is the most NATURAL.
Acronym Finder.com has several relevant solutions:
Name Lookup Protocol
Natural Language Processing
Network-Layer Packet
Neuro-Linguistic Programming
Non-Linear Processor
Nonlinear Programming
I found the use of it sophomoric myself.
Go to Cnet and the download area. You will see the pages ranked by most requested downloads. And anyways Yahoo uses most requested links in their Categorized Links on their home page. Such as Computer>Supercomputers>Cray. While this was categorized by humans it still just as good as good as what Google does.
Hey kid, before you embarass yourself any further why don't you just go ahead and look at what PageRank actually is. Google did invent what they call "PageRank." It's name is a touch misleading.
Dacels Jewelers can't be trusted.
PageRank was named after one of its creators: Larry Page. Sure, it's a pun, but it really is named after Larry Page.
cpeterso
page rank is self explanatory.
Is it really? It's certainly beyond obvious that pages must be ranked; what's really not obvious or self-explanatory is how to rank them. CNet ranks by number of downloads, but that can't be done for web pages, because there's no system that can monitor page views. Altavista, Lycos, etc., tried to rank pages based on which ones had the closest matches to the search terms. That worked pretty will in 1995, when the web was much smaller, but soon grew to suck and was trivially easy to manipulate.
So, just how would *you* go about deciding which of the 20,000 pages that match the search terms should come up at the top of the list? And how would you do it in a tiny fraction of a second?
Please be as detailed as possible.
It's a hard problem, and Google devised (and patented) a clever, elegant and extremely effective method for figuring out which pages are most likely to be *relevant* to the searcher. Their method revolutionized the search engine industry and it was so much better that by the time Google had been operating for a year it got more traffic than all of the other search engines combined, and did it without significant advertising.
Not that Google is the end of the word in searching; but the folks at Google are a devilishly clever bunch, and good guys to talk to if you want to find out where search engines are going.
Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
I guess "far off" is relative, but take a look at GNOME Storage. It's pure theory either; look at the pretty screenshots.
NLP stands for Natural Language Parsing. NLP is feasible for very narrow domains. For example, say you want to make it easier for lock smiths to learn how to fix locks. By using syntax and grammar for questions, you can limit the number of possible combinations starting with "What" followed by an object like "tumblers." Normally there is a verb in between the two like "what is a tumbler?". Simple sentences are easy. The hard part is when you start to build complex sentences like "what is the purpose of the tumbler and how does it function within the mechanism?" This sentence can be interpreted different ways depending on the domain. The hard part is determining the relationship object/subject and mapping the valience. Valience is a term used in one school of NLP called Dependency Grammar. Valience is is meant to describe the relationship between the object and subject. Anyways, go search on google for more info.