The Datatypes spec talks about members of the lexical space
of a given type as lexical representations of (values of)
that type.  In many sentences, however, a term is needed to
denote the strings of characters which are validated.  These
may or may not actually be lexical representations of values 
in the value space, because they may or may not be valid.

Some time ago, the editors agreed to propose to use the term
'literal' with this sense, but noted that a definition of
the term was needed.

EP-14 is a proposal to define 'literal' as 

    A sequence of zero or more characters in the Universal
    Character Set (UCS) which may or may not prove upon
    inspection to be a member of the lexical space of a given
    datatype and thus a lexical representation of a given value
    in that datatype's value space, is referred to as a
    <term>literal</term>. 

and to add (as a way of avoiding the false assumption that
'literal' is used only of unvalidated strings of uncertain
validity) the remark

    The term is used indifferently both for character sequences
    which are members of a particular lexical space and for
    those which are not.

Also, some occurrences of 'character string' used in the sense
given are changed to 'literal' in this proposal.