hakia Participating in LangTech 2008

I recently returned from Rome, where I represented hakia at the LangTech2008 conference. Hans Uszkoreit of the DFKI had invited several search scientists to join a panel on “Human Language Technology in tomorrow’s search application.” The night before the panel, conference participants could take a tour of the impressive Castel St.Angelo, where we enjoyed the view over the city and a great buffet dinner, during which I had a lively conversation with Prof. Uszkoreit and Dr. Lenz.

For the panel the next day, Hans Uszkoreit started us off with a set of questions on the promises and limitations of NLP in search. Next, Geoff Zweig of Microsoft focused on their research on voice search with summarization improvement; Mario Lenz, CTO of empolis (part of avarto, part of Bertelsmann), the leader and coordinator of the EU Theseus project, gave a general overview of NLP support and its limitations; Tom Hofmann, Director of Engineering of Google’s research team in Zurich presented recent advances into NLP support for their search engine; finally, I talked about how all that was yesterday and how real semantics will mean relevance and how statistics will continue to mean staying below the threshold. Most questions during the following half hour of audience questions and panelist discussions were to Tom Hofmann and me, to him how they do what they do, and to me if what we claim we’ll do is really possible.

I’m safely back in our New York office now, with the usual travel-induced cold. But I’m truly glad to have had the opportunity to meet and talk with these colleagues and other participants of the conference and present our approach to the audience at LangTech 2008. I was particularly happy to have also found other researchers who think that statistical and formal methods alone will continue to not deliver acceptable applications in NLP. I discussed this in detail over lunch with Claude Roux of Xerox, Grenoble. Very telling was a remark by a member of the audience after a paper on the semantic web by Christopher Welty of IBM. The remark cited Hubert Dreyfus’s criticism of the claims of progress in standard statistical artificial intelligence: “…the first man to climb a tree could claim tangible progress toward reaching the moon. Rather than climbing blindly, it’s better to look where one is going.” (Dreyfus, Hubert L. 1992. What Computers Still Can’t Do: A Critique of Artificial Reason. Cambridge, MA: MIT Press: page 100.). We know where we want to go with search and we know we need nothing less than real semantics, Ontological Semantics, to get there. Others are still climbing the statistical trees.

        

This entry was posted on Friday, March 7th, 2008 at 3:40 am and is filed under News, Technology. You can follow any responses to this entry through the RSS 2.0 feed. You can leave a response, or trackback from your own site.