Powerset's smarts are Microsoft's gain

Larry Dignan pours lukewarm water on Microsoft's acquisition of Powerset. Talking about Google's near monopoly in the search market, Larry says:

Microsoft can reinvent search, but it’s still running up a natural Google monopoly. The analogy here is Windows: Microsoft didn’t have the best operating system on the planet. It just had the best positioned one. In search, the tables are turned in Google’s favor. I don’t see how Powerset will change that equation.

The Techmeme firehose has a party with this one with the likes of Nik Cubrilovic declaring:

Microsoft’s search engine market share is currently at 9%, far behind Google who has 61% of the market. The web division at Microsoft has been performing below expectations since its inception - and with the dominance of Google the challenges facing Microsoft are as big as any the company has ever faced in its history.

Oi! Both posts miss the point in my opinion because they ignore the fact Microsoft remains a business application provider. If Microsoft is really clever about this acquisition then consumer considerations aside, it can wrest the Google Search Appliance market out of Google's hands. In one deal that I have been involved in the last year, the first thing we did was ditch the crabby appliance. It's just too inaccurate for serious business use and I make no secret of the fact I think it's a dumb technology that is not business grade. Given that Microsoft is punting Sharepoint as hard as it can plus the fact Sharepoint in turn is being pitched as a blog/wiki holder, having solid search becomes a natural requirement.

Interestingly, my Irregular colleagues are divided on the issue. Jason Wood whose financial nouse is usually pretty darned good says this in our Google Group conversation:

No one, Microsoft included, is saying they’ve bought the future of search here. What they’re saying is they’ve acquired a quality team of engineers who are working on solving the future of search, and this will be one small step in a long road toward that. Microsoft is spending, what, 22 minutes of free cash flow on this? So by that notion, if this has ANY impact on helping build a credible if not distant #2 to Google over the next five years, how would it not be worth it in spades?

Jeff Nolan on the other hand is skeptical. He sees Powerset as an expensive proof of concept:

I don't fancy myself an expert in semantic search but I know enough to suggest that doing it in wikipedia doesn't translate into "now I can do it anywhere" so as a proof of concept I would be a little leery. Now if they have unique IP then that would be something but this area is already pretty rich in IP so not sure that is the case.

I'm happy to be proven wrong but semantic search has about as good a track record for investors as speech recognition in years past so it's a show me thing as far as I'm concerned.

Ooooh - equating to Lernout & Hauspie? I hope not.

Update: Another Irregular Bruno Haid is thoroughly dismissive:

Basically they're [Powerset] doing three things:

1.) Let Freebase extract all structured data (categories, sideboxes etc.) from wikipedia and store them in a special format. We paid an undergraduate cs/law student 6000 USD in early 2006 to do this. Freebase value comes from being able to store and query data 'semantically'. 2.) Label this data consisting of triples (WashingtonDC is CapitalOfUSA) as "factz" and present them as search results 3.) Create a simple parser that maps certain manually pre-defined query structures (when, what, where, who ... predicate, object) to search queries (mostly labor intense to get the rules right, also done by http://www.presdo.com/ and many others)

Technology bottom line:

- There is no 'understanding' happening whatsoever. - Their approach isn't really solving the problems like disambiguation or other problems mentioned in the LIVE! post. - It inherently can't scale to the web in general, because there's little to no standardized structured information out there. Yet. But even if the Yahoo SearchMonkey approach seems much more viable. - The learnings of the past 2 years don't sacrifice $100m, considering that there are around ~10 university projects on the same technology level in Europe alone. - Google is already using much more sophisticated approaches like these in their back-ends, but their are labeling and employing them as what they are: Purely statistic / heuristic methods and manual tuning labor. Very harshly put: Peter Norvig could come up with something similar to Powerset with the stuff he has on his notebook in the 10 hours it takes to cross the Atlantic.

Either MS did this to tell a good story to the markets or their text-based search team is far behind their imaging counterparts.

Harsh indeed.