"Accidental" associations through name collisions are powerful tools that can be used to link information together. This is where a global or shared namespace would be good, so that we know that when A says "[The Beatles]?
", it means the same as when B says "[The Beatles]?
". They can also lead to misleading inferences: when A says "[my mother]?
", it does not mean the same as when B says "[my mother]?
Note that this is not the general case: often, systems will need to be told that A's "Beatles, The" is the same as B's "Beatles", is the same as C's "The Beatles".
Philosophical problem: are "John, George, Paul and Ringo" identically equal to the "The Beatles"? (Consider Stuart Sutcliffe, the Silver Beatles, George Martin...) This looks like nit-picking, but once you let computers loose on datasets, silly things can happen.
Think also about the philosophical concepts of extension?
. (see http://www.swif.uniba.it/lei/foldop/foldoc.cgi?extension+-+intension