>Anything not in the document charset should be ignored (try parsing
>with sgmls and see what the SGML parsing model produces).
I think an error should be indicated in some manner. Perhaps we
should decide on a glyph image to be displayed. Isn't there something
like this in the Unicode standard?