powersetlogo

Powerset has raised $12.5-million (U.S.) in a first round of venture capital hoping to do Google one better in the search engine business.

The company is looking to implement natural language search, which so far has been attempted but not implemented.

Natural language search means that you can pose a query just like asking a question: “Who was the fourteenth president of the United States?” The search engine should be able to pull out the meaning and return the proper search results.

As you can imagine, this gets very complicated given the convoluted syntax of the English language (and we won’t even get into internationalization issues). “List all U.S. Senators who are not Democrats” can easily stymie a regular search engine, which sees “U.S.”, “Senators”, and “Democrats”, but misses the crucial “not.”

Reaction has been mixed. Danny Sullivan of Search Engine Watch reposts a 10-year-old article called “Hello Natural Language Search, My Old Over-Hyped Search Friend” which basically argues that we currently don’t search using natural language; rather, we type one or two words like ‘britney spears’ or ‘peak oil.’ As a result, search engines can parse results and group them into logical clusters…but search engines like Clusty are already doing that, and it’s never quite taken off.

Barney Pell at Powerset responds by arguing that there’s much richness to be found in stopwords (words like if, and, or by, which search engines typically ignore), and that we’re going to relearn how to formulate queries in natural language - because the search results (as powered by Powerset) will be so much more accurate.

There are some problems with this view.

First, user behavior is notoriously difficult to change. Google succeeded because it didn’t require users to change their behavior - it just did a better job than the competing search engines of the day with the same query strings.

Second, natural language just hasn’t made a lot of strides in shipping software. We’re still stuck with Microsoft’s help technology, which isn’t bad, but clearly isn’t at a level of sophistication that would be required for a large domain set. It’s also clear that Microsoft’s natural language help search has been nudged along with manual editing of results, which would be impossible to do on a large scale.

It may seem like an argument against innovation, but I would be more likely to believe that full-blown natural language search was around the corner if we’d seen incremental improvements over the years. Even as far back as the mid-80s, the Q&A database was touting natural-language queries.

And finally, while I believe that limited natural language search is possible, the complexities of the English language combined with the frequent spelling and grammatical errors we see in search queries are going to sorely try any such search engine. The first time a user types in a complicated query and Powerset gets it wrong, the user will go somewhere else…like Google.

Still, I can’t wait to see how Powerset tackles the problem. If nothing else, it’ll be an incremental approach on the way to a solution for better queries…and better search results.

Tags: ,