Natural language multiword value in attribute enumeration/Whitesp ace in Nmtoken

Hello www-international,
 
Background/apologies
--------------------------------
 
My question ultimately is rooted in language, and thus
I figured it might be acceptable to send it to the
internationalization list; from my understanding several
people working on language-related DTDs currently
have identical questions/have to live with workarounds
like underscores instead of whitespace. I apologize
if this is not the right forum (would be happy if someone
could point me to the appropriate one.

Question
-------------

I am trying to do sth. like

<!ATTLIST myElement
        myAttribute (come over | come across) #IMPLIED>

So far did not succeed. Amongst others I tried

a) ( "come over" | "come across" )
b) ( &apos;come over&apos; | &apos;come across&apos; )
c) ( come&#32;across | come&#32over )

The error message I get from e.g. IE5 is
 
Invalid character found in ATTLIST enumeration. Line 13, Position 39 
 
<!ATTLIST fooo  associatedWord  (come across | come over)  #IMPLIED >
-----------------------------------------------------------^
 
What I am technically trying to do (my understanding) is to
have a multiword Nmtoken/whitespace within a Nmtoken. Is this simply
impossible? If not: What's the trick?
 
Thanks in advance, Christian