mkaz.com

Powerset, natural language search platform

By marcus • Jun 30th, 2007 • Category: geek

Powerset I attended a special event at Powerset on Thursday, where they opened up what they are working and planning on. Powerset, if you hadn’t heard, is a new search start-up looking at becoming a relevant player in the search game. Their not-so secret weapon is natural language technology they licensed from Xerox PARC.

As far as technology goes, what they demoed was impressive; however search queries were over a controlled and limited set of data. The two primary examples were over Wikipedia, which is their primary test set and searches over movie data from IMDB. A question is will Powerset be able to keep high quality results when indexing less structured and less authoritative data, such as the entire internet.

Natural language search can be quite impressive, especially over known and structured data. While working at E*TRADE, I evaluated natural language search from iPhrase to power our customer service. This worked quite well; especially since we had the extra advantage to know a user’s query would be regarding financial services, account information or the like. A user searching for “how do I deposit money” would get excellent results.

One of the issues that we did come across in testing is that user’s have learned to search via keyword; thus unlikely to type sentences or phrases which the natural language engine craves. However, users are fast learners and UI clues can help solve this. iPhrase was not picked as the search vendor, but I believe the issue finally came down to price and politics and not the technology.

Now is Powerset’s natural language search going to be a destination site to compete with Google, Yahoo and the other big players, this is their current goal. However, I think the true power of their technology will work best in smaller known scopes, such as a licensed search appliance. They do plan to have open APIs and encourage all sorts of mashups; so their success may come from developers creating these specific apps on top of Powerset.

A last thing about Powerset which I really like, they have a culture of open development. They plan on launching Powerlabs in September. Powerlabs will be a community site around their product. Users will be able to submit ideas, post bugs, and play with the latest releases giving feedback on search quality. True, who wouldn’t want free QA and free product ideas from the community. Still most companies don’t have this openness and definitely not before launching the product itself.

Only time will tell, I look forward to September.

Here are some related links: Danny Sullivan posts about Natural Language Search Hype and Dan Farber has an in-depth report of the Powerset event.


One Response »

  1. interesting Marcus…I was actually a developer at iPhrase at that time (ie prior to getting purchased by ibm). Great technology, but unfortunate business strategy. They didn’t make it modular until too late. The move to a modular product suite was so that potential clients could make a smaller $ commitment.

    However, powerset is putting iphrase in the same category as Ask geeves. Saying that it was simple database queries from prepped datasets…This was NOT the case at iphrase. iPhrase did full natural language parsing via porter’s algorithm, in addition to database querying, with fallback to keyword search. Iphrase could also dynamically create relevant pages if you remember…..I’m a little offended that their coming off as a more innovative than iPhrase which did all this back in 2000. Good luck to them anyways.

    Their execution strategy along particular verticals is the same as iphrase as well.

Leave a Reply