Natural Language Processing Not Suited for Web Searches
August 9, 2011
There’s a new cowboy in town and he’s shaking up the search engine industry. The article, Real Language Q&A: The Next Generation of Search?, on Search Engine Journal, explores the practicality of Oren Etzioni’s recommendations for search engines in his new paper, titled, Search Needs a Shake Up, published in Nature.
According to Etzioni, current search engines have not kept up with the time. The reliance they have on old algorithms with results displayed as a list that can run into the millions is no longer practical. As the article explains,
“In Etzioni’s view, the next generation of search would abandon the “blue link” structure in favor of directly answering the questions of users. “Moving up the information food chain requires a search engine that can interpret a user’s question, extract facts from all the information on the web, and select an appropriate answer,” he states. The tricky part, though, is in finding the answer. With so many ambiguities, it’s difficult to see how most questions could be answered by a search site.”
Conveniently, Etzioni offers his own University of Washington’s Reverb program as a step in the right direction. Reverb relies on Natural Language Processing (NLP) which is an interesting direction for search engines, but depends entirely on the reliability of the user’s question.
In a world of Etzioni’s search engine, the functionally illiterate would never receive an accurate search result because the search engine would never recognize, “who be prez bamas baby mama?”
While it would be a lovely world to live in if life was as well-spoken as Jeopardy and Watson could answer our every question quickly and precisely, that is not the case and never will be. NLP works well with voice searches and should stay there. Though Etzioni poses some interesting questions and points out the while elephant in the search engine room, the answer is not as simple as NLP. At least not yet.
Catherine Lamsfuss, August 9, 2011
Sponsored by Pandia.com, publishers of The New Landscape of Enterprise Search
Comments
One Response to “Natural Language Processing Not Suited for Web Searches”
Hello Stephen,
have you tried searching for “who be prez bamas baby mama?”? Google actually does deliver good results in my opinion (if you do not use quotation marks for your query). The first hit is an article in the Huffington Post that basically answers the question. All ten hits I got are directly relevant to the question. (Bing, unfortunately, does not do quite as well for this query)
I think this whole question-answering-topic is an interesting debate. I am not sure if it would really help if you could ask questions; in my experience, web search is an iterative process, and I find it hard if not impossible to formulate a direct question for most of my searches.
What do you think?
Best regards,
Florian