I guess you know a lot more about this than I do. But maybe you are overestimating the Forth-like part?

> There is no "real parser" : the parser is like any Forth parser ie one word at a time separated by blanks.

> In fact, it is an Forth-like interpreter with immediate words and parsing words that use the stack structure a lot.

Looking at an example like

    : catalan(n) n ifZero: [ 1 ] else: [ catalan(n 1-) 2 n * 1- * 2 * n 1+ / ] ;

(from http://rosettacode.org/wiki/Catalan_numbers#Oforth) it seems like, if what you say is true, then : must be a parsing word that not only reads "catalan(n)" but also somehow decomposes it in a way and sets things up such that the later call "catalan(n 1-)" is understood correctly even though "catalan", "(", and "n" are not separated by whitespace. It very much looks like there is some tokenization going on.

Also, is "ifZero:" really a separate message from "else:", i.e., or is "ifZero:" also a parsing word that looks if what follows is a well-formed expression followed by "else:"?

If you have parsing words everywhere, you have a parser. It may not be a classical parser where everything is written in one place, but it's a parser nonetheless.

I don't think there's anything wrong with implementing a Smalltalk dialect in Forth, with lots and lots of parsing words, and compiling the Smalltalk surface syntax to stack-based Forth code on the fly. That's lovely. It's just a shame to expose the underlying Forth in the surface syntax...