Just taking a wild guess: It appears you're using MSFT Office to generate the pages. (MicrosoftOfficeWebServer: 5.0_Pub) In that case, the character encoding is probably windows-1252. That encoding uses values that are invalid on the web, e.g. hyphens, single and double "curly" quotes. My guess is that Office escapes those invalid characters as valid ASCII/ANSI entities. I didn't spot any of those invalid characters (they're not invalid within their declared set) in the body of the page.
Blowing my whole theory out of the water is the page's http-equiv charset utf-8. :shrug: So, there's the SWAG.