Semantics are in the mind of the author, and they are in the mind of the
recipient.  They are never in data, no matter how encoded.
Moreover, the only realities that matter are the ones in our heads,
because those are the only ones upon which we can act.

Communication is possible only because we believe it is possible.  We
believe it happens only because we have believed it long enough to
accumulate sufficient context with our correspondents to confirm us in
our belief.  It's a matter of preponderance of evidence.
Disappointments -- i.e., communications failures -- are inevitable.

Parenthetically, I feel I must mention, at least in this xml-dev
context, the fact that same question was uppermost the minds of those
who insisted, many years ago, that every SGML document must have a DTD.

* Did the transmission conform?

* Did the recipient's process work correctly?

* Should the transmitter have known better than to transmit?

* Should the recipient have known better than to accept the transmission?

This takes
some rather unsettling self-training, perhaps because our brains are
designed to ignore the latter and "pay no attention to the man behind
the curtain."  The man behind the curtain is the rhetorician in the
room, and, like the Wizard of Oz, even *he* doesn't know how it works.