Great question. On second thoughts, is there much to say except "Combine the parts you like from existing languages and the new things you want"? And, read stuff about how existing languages were designed.
There's a family tree of languages here: http://www.digibarn.com/collections/posters/tongues/tongues....
The buck stops somewhere, the history is not that long - if you trace all the ancestors and key influencers of a single language, that's generally 10 at most; and if you trace another language, half of them will overlap.
You can't really skip that step anyway, it's not like you can make an informed decision about a particular syntax feature if you haven't taken a serious look on how that feature works or doesn't work in a dozen languages; and the more obscure languages tend to be very useful in this regard because they explore a larger variety of features and their combinations. Getting a short overview of syntax of at least some 25 languages would be a reasonable prerequisite before designing another one, and that overview would still be the easy and straightforward part.
This actually sounds quite interesting. You're starting from a good language (Smalltalk) and adding a feature absent from it but widely regarded as desirable (strong static typing).
> and clean up its syntax a bit (it's already pretty clean, just a few things could be improved)
If you go for type inference instead of having types explicitly declared, you might be able to avoid both run-time type checks and changes to Smalltalk's syntax.
> essentially a faster Smalltalk.
Bryce Kampjes (http://www.kampjes.demon.co.uk/index.html) was working on compiling Smalltalk byte code into machine code to speed it up, but that didn't involve changes to the language, so I think run-time type checks would still have been necessary. Making static type checks would result in a further speed-up.
Good luck.
Thanks!
Report can be found at https://gist.github.com/ndrewxie/54d3299c998569c27e281d2320b...