Disclosure: I work at the NY Genome Center.
The reason the language processing is important is because this application of Watson ingests journal articles to "learn" that tumors with mutation X affect pathway P and cross reference that with other articles that might say drug D is effective for regulating pathway P.
This creates a knowledge base watson can then use to apply to a sequenced tumor that has mutations that may include X and can suggest the possible effectiveness of drug D.
The Watson team has done a fair amount of work over the past few years to get Watson to "understand" the language and terms used in medical journal articles to enable it to extract information from the plain text articles.
One of the main wins of Watson isn't so much the query front end, but the ability to use unstructured text to develop networks of related assertions.