Jennifer Zaino
SemanticWeb.com Contributor

This summer, semantic web technology company Cognition launched SemanticMEDLINE.com, a free service that uses its SemanticNLP natural language processing technology to help health care professionals, researchers, and consumers quickly discover complex health and life science material from within the 18-million-article abstract database of health information published by the National Library of Medicine.

"When it comes to SemanticMEDLINE, the number of cases where the terminology actually used in MEDLINE content is so very ambiguous that if you were to type in a plain English phrase for what you're looking for, you'd get not only perhaps the right information, but thousands of others that have nothing to do with that," says Cognition CEO Scott Jarus. "By using our technology you can more precisely hit the target and throw out irrelevant stuff."

For example, normally a non-semantic query for "brain lobes affected by herpes encephalitis" is likely to result in volumes of data around herpes and encephalitis, without reasoning to the specific class of terms in question. At the same time, the semantic technology behind the service also ensures that individuals researching "heart attack" also can be surfaced information on myocardial infarction -- a phrase they may not have searched for because they didn't know that it was a synonym.

Cognition developed the service as an example of its semantic natural language processing technology, some 20 years in the making. It's the latest to a set of free semantic search services that also include CaseLaw, a database of federal and supreme court decisions since 1950 from Public.Resource.org, and Wiki, for semantically searching the English Wikipedia. Cognition isn't trying to be a Google killer, though. Its business model actually is to semantically enable other organizations' technologies or applications.

"So, if you have content, and you want a semantic search capability, we can help," says Jarus.

Customers include legal clients, who have applied its semantic natural language processing technologies to the ediscovery process. In fact, Cognition provides the advanced semantic search in LexisNexis Concordance. It also provides enterprise semantic search for large corporations in the biotech field, helping them organize their data semantically.
"That work over the last several years is what gave us the idea and capability of providing SemanticMEDLINE, because our semantic map--the largest English one in the world with over 500,000 word stems and 4 million semantic contexts, got augmented by those clients in the life sciences biotech fields," says Jarus.

According to Jarus, Cognition has over 99.5 percent of the common English language already accounted for. To further assist customers and grow its life sciences language base, the company sponsors a research project at the University of Texas Medical Center, which helps it continually add new terms and words into its Semantic NLP word map. Areas like genetics are adding 20,000 to 30,000 new words every year because the field is exploding, says Jarus.

What's different about Cognition's SemanticNLP compared to other semantic search technologies? "Cognition's technology is based on this very comprehensive and complete semantic map of the English language. All the other things we've added, like parsing technology to understand the structure of a sentence (nouns, verbs, nouns those verbs act on), grammar -- all are important. But if you don't have a complete comprehensive semantic map, the ability to reason within the language is severely limited."

Next up for Cognition? Possibly a semantic search along the lines of SemanticMEDLINE for patents.

"It would be interesting because patent attorneys and VC firms and entrepreneurs have a dickens of a time searching the patent and trademark database," says Jarus. "It's because they are voluminous and written in very ambiguous language, because they cross multi-disciplines and one word in one discipline could be something completely different in another one. We're looking at that to showcase our dataset."