In addition to victor hugo's answer it is possible to get all known character references legal in an XSLT file, like this:
<!DOCTYPE stylesheet [
<!ENTITY % w3centities-f PUBLIC "-//W3C//ENTITIES Combined Set//EN//XML"
"http://www.w3.org/2003/entities/2007/w3centities-f.ent">
%w3centities-f;
]>
...
<xsl:text>& –</xsl:text>
There is also certain difference in the result of this approach as compared to <xsl:text disable-output-escaping="yes"> one. The latter is going to produce string literals like for all kinds of output, even for <xsl:output method="text">, and this may happen to be different from what you might wish... On the contrary, getting entities defined for XSLT template via <!DOCTYPE ... <!ENTITY ... will always produce output consistent with your xsl:output settings.
And when including all character references, it may be wise to use a local entity resolver to keep the XSLT engine from fetching character entity definitions from the Internet. JAXP or explicit Xalan-J users may need a patch for Xalan-J to use the resolver correctly. See my blog XSLT, entities, Java, Xalan... for patch download and comments.