When implementing the WIG scanner you may have to take into account the fact that WIG has multiple so-called lexical scopes. You already encountered lexical scopes when implementing multi-line comments for JOOS: inside comments, there exists a different set of tokens/keywords than outside comments. WIG has even more such scopes. Consider the following little WIG fragment:

service {
  const html Compliment = <html> <body>
    This is a <[fin]> great service, man!
  </body> </html>;

  const html Pledge = <html> <body>
    What is your name?
    <input name=name type="text" size=20>
  </body> </html>;

  string name; //name is an id here, although it is a keyword inside HTML tags
               //inside HTML text, it's considered plain text

  session Contribute() {

In this snippet I identified the following lexical scopes:

WIG syntax: Here, stuff like service, const, html and so on are keywords. "name" is not a keyword.
HTML syntax:

is entered when <html> is scanned and left when </html> is scanned
unlike in WIG syntax, service, const etc. are no keywords
> and < have different meaning than in WIG syntax (although the scanner may not necessarily have to distinguish those)

HTML Tags: Here input, name etc. should be keywords so that the parser can recognize them specially.
Holes: only allow for identifiers - any identifiers in fact, including those that would be keywords in other scopes, e.g. <[html]> is valid
HTML right-hand side values: It may be useful to have another scope here so that e.g. name is not recognized as a keyword.

Can you think of other lexical scopes? What about HTML comments? Do those exist in the benchmarks? If so, can HTML comments be nested?

You can extend a Flex scanner with lexical scopes using so-called start conditions. You prefix a regular expression with <c> to denote that it should only be scanned when being in state c. You switch to a state c by calling BEGIN(c) in the scanner's action.

SableCC supports a similar mechanism using so-called states (see pages 35 ff).