[initial submission below, fuller position paper to follow]
At HSBC we are using 'Natural Rule Language', (an evolution of the CliX Language - see www.systemwire.com) together with FpML (see fpml.org).
I would say that NRL is a horizontal DSL (a near English syntax for the use of business analysts) whilst FpML is an industry standard domain specific ontology. Using them together creates a specific language that business analysts can use to express rules about Derivatives products, and which business end users can read and understand.
This is all very good - as far as it goes....
The snag is that there are hundreds of products out there processing trades - and hardly any of them uses a rules based approach. Where they do they have a proprietary implementation. So, although it would be quite possible to generate java directly from NRL, the problem that we are faced with is that we would need to tailor the generated code to suit particular software implementations.
It seems to me that DSLs suffer from the same impediment as MDA initiatives (at least in the following respect) in that they assume a standardised execution environment for the output code. Unless we are in the (fortunate?) position of being able to write everything from scratch, I can't see that DSLs or MDA are going to make any real impression on the 'software crisis'.