The answer to your first question is no. For example if you wanted to search for "What did Al Gore say?", a google wild card search for "Al Gore said wildcard" (the markup won't let me put an asterisk) would not match things like "blah, said Al Gore" or "he said blah" where he is referring to Al Gore, or "Al Gore told someone blah". For the search engine to match those results, some level of parsing is needed.
Yes, but if this is an an important news item, what he said would probably be mentioned on many web pages in many forms, including a form that matches your query.
After watching the entire video, it sounds like the system they are building fundamentally operates at a linguistic level with no attempt to really model underlying concepts and relations in a human way. A questioner asked how their approach compares to CYC (The company trying to model all human common sense knowledge with bad epistemology and contradictions), and he explained that they aren't even trying to model concepts in that way, but rather finding relations through linguistic analysis. I think the way to think about what they're doing is a very advanced pattern matching approach with definite limitations. For example, you can't ask their system to find contradictions between what someone said as it doesn't "understand" contradiction, etc. He said there would be a 5 minute tutorial on the kind of questions you can and can't ask.
I think that there is the danger, and he mentions it as well, in people having too high expectations when using NL to query their system. They apparently have exclusive license to use PARC technology finally finished in 2003, so I'm sure that there will be some interesting results. The question is will it frustrate people by its limitations so much that they move away, or will it excite people about the possibilities of a system that could actually "understand" what the concepts mean that they are expressing. The latter would be great news for Organontech.
I think they do have some modeling of underlying concepts. I think if I remember correctly they had a search where "politician" matched "mayor", so they at least have synonyms. I think they obviously would like to have understanding--it's just that NLP isn't there yet.
I agree, they do have concepts, but not in a human way. One questioner asked what their conceptual schema looks like, and the CEO said that it is basically the same as typical ontologies out there but with a few more links between concepts. We looked closely at OWL and RDF and CYC and all the other ontology languages we could find to see if it was possible to meaningfully capture human like conceptualization with these languages, and it isn't - they are extremely far from providing a framework for successfully doing it. A radically new framework is needed to make progress conceptually - which is what we are working on.
6 comments
[ 4.7 ms ] story [ 19.7 ms ] thread- For the sorts of queries he mentions in the talk, can't you just use Google's wildcard queries to get similar results?
- Can you use a more social web 2.0 approach to address difficult and/or natural language queries? If so how?
- What if you allowed the system to get back to you later with answers (possibly several hours later)? Is a social approach more feasible then?
I think that there is the danger, and he mentions it as well, in people having too high expectations when using NL to query their system. They apparently have exclusive license to use PARC technology finally finished in 2003, so I'm sure that there will be some interesting results. The question is will it frustrate people by its limitations so much that they move away, or will it excite people about the possibilities of a system that could actually "understand" what the concepts mean that they are expressing. The latter would be great news for Organontech.