Fast Albert: Smooth Swiss Search


Kentucky buzzed this month about new search system developed in Catelnau-le-Nez, France by Albert S.A. This Swiss company hired Beth Krasna as Chief Executive Officer last year. (Ms. Krasna was the CEO of Sécheron and a director of Quadrant Holding, the first listed Swiss Venture Capital company.)

Few things engender more excitement in Harrod’s Creek than a multi-lingual, natural language capable, intelligent search engine.

Search-and-retrieval in the U.S. has entered doldrums. In Europe excitement about Autonomy’s market performance engenders search opportunities. At the January 2001 Internet World in Canada, speakers and those in the audience called attention to deficiencies in existing search systems.


Albert is a “next-generation search-and-retrieval system.” It is not a single software program, nor is it a Web index such as Google. Albert is an integrated, robust suite of services that run on Linux and use MySQL to house Albert’s Knowledge Base. Built dynamically, this special data store is accessed by algorithms when a user creates a query and launches a search. Many search engines use a limited set of analytical tools to get results to the user rapidly. Some companies use equations from a maths handbook or repackage tools from third-party developers such as Thunderstone Document Retrieval and Management (Cleveland, Ohio). Some research boutiques commercialize recursive algorithms that bring Sparc Enterprise systems to their knees when a simple query is run. Enter fast Albert with its Linux OS, the Apache server, one, low-cost processor and a secret pizza sauce formulated by a multinational team working in France. (The Albert technology is available for public use from two separate demonstration sites. The French language Free site at www.free.fr and the Albert, Inc. at www.albert.com/news.htm.)

Albert blends several techniques to improve the search experience in one bundle. A typical enterprise
customer, according to Albert, may have various indexing systems. Albert ingests these systems’ data,
integrates them with Web, for-fee third party content, and standard Word and other types of documents.
The user frames a query in natural language in his or her native language and—bang—in a second or two
the relevance ranked results are displayed. If the user enters a query in German, the system displays
German documents near the top of the hit list.

From my vantage point in Kentucky, the key difference between Albert and other search-and-retrieval
systems boils down to how Albert has harnessed several technology stallions together so speed and
power are optimized. Some search-and-retrieval systems make do with a one-trick pony and maybe a
dog cart for marketing zing.

Albert adds to its NLP an innovative personalization feature. A user can enter a word or two or a
complete query, click the button, and get results in rank order. When the same user frames another
search, Albert adapts to the user’s behavior. Let’s consider a simple example. When a user enters
a term or phrase such as “Nokia product development strategy”, the Albert system takes note. The
database of information about user actions influences what Albert calls this its AlbLearner.

AlbLearner runs dynamically and independently of the other modules. It interacts with and
learns ways to make wordassociations, learn conceptual meanings, learn similarities between
documents and re-write and edit rules for that particular user. In short, Albert’s algorithms
adapt to the user’s behavior. “Users want answers,” says John R. Toedtman, Chief Executive
Officer - Americas, “not a repetitive programming demonstration.”

Albert knows languages. Now the multi-lingual capabilities of American search-and-retrieval
systems are add-ons. The result is often a jumble. Albert’s driving force—Jean-Michel
Livowsky, who holds a Ph.D. in cognitive Science and has about 100 patents—focused on
multiple languages at the outset of his development effort more than five years ago.

Albert S.A.—despite softening in the venture capital markets—attracted more than £15 million
in funding in 2000. In the course of refining its software, Albert sealed alliances with FAST
Search & Transfer (Oslo and Trondheim, Norway) and IBM Europe. Unlike most search-and-retrieval
systems that require the organization to set up, monitor, and maintain one or more software servers,
Albert’s approach is “plug in and use.” The company delivers to a client a Linux server. Once
connected to the network, the Albert system is available. If a client location requires more
processing capability, Albert places another low-cost Linux server on the network. Perhaps the
most interesting aspect of Albert’s approach is the research and software engineering teams’ commitments
to providing easy-to-use systems that make use of highly advanced search-and-retrieval techniques. Says
Albert’s founder Dr. Livowsky, “It's up to the machine to understand the man. The only thing that
counts is human language, the rest doesn't matter.” Dr. Livowsky is quick to credit his multi-national,
multilingual development team of more than 40 experts with the features of the new Albert system. How
does Albert motivate its research team to do challenging work at the frontiers of science? One part of
the answer lies in the location of the company’s main research facility. Outside of Montpelier, France, an
undistinguished office building houses carefully-selected programmers, information scientists, and engineers.

Inside the research office is a hive of activity. Albert’s R&D operation is some distance from the
placid environment captured in Frédéric Bazille’s View of the Village of Castelnau-le-Lez (1868),
on display in the Musée Fabre (Montpelier). The work style is a mix of Milpitas, California,
and the Montpelier’s Clos de l'Aube Rouge. The Albert engineers are polishing Albert’s ability
to handle Chinese, Japanese, Korean and other “double byte” languages. What else is the team
working on now? Albert’s engineers are adapting the software to cellular and PDA appliance access.
There is a voice-activated search initiative. The team is working on new data mining services.
Because Albert can build user-specific “tweaks” to algorithms, Albert staff envision recommending
associated, relevant information delivered in real time. In a matter of months, BMW and Albert
will present a version of the software integrated into a BMW automobile. What are the risks to
Albert as it begins to leverage its link with FAST, IBM, and BMW? Certainly the hotly contested
world of search-and-retrieval is a worry to any new company. However, Albert’s most serious concern
is that the supply of pizza from Le Don Camillo, Castelnau-le-lez's answer to Pizza Hut, is not
exhausted. Pizza, not coq au vin, make the Albert engineers’ intellectual engines run.

Albert S.A.
Albert-Inc. SA
13, Boulevard de la Forêt
1009 Pully - Switzerland
Phone : + 41 21 721 02 10

Stephen E. Arnold
Arnold Information Technology,
Kentucky, USA
sa@arnoldit.com


[ Top ]   [ AIT Home ]   [ Site Map ]