There is a Definicon cross-assembler but it is an MSDOS executable binary so I can't run it on my Linux PC nor can I compile it for Linux because there is no source-code.

There is C source code for one in Dr. Dobbs' Journal. The OCR text is not good because all the OCR tools I can use for free try to automatically recognise text in columns, with no way to specify single-column operation. Which mangles the order of source code scans.

Several days of tedious salvage work later, I've got the 32000 cross-assembler to the point where it compiles, but it crashes with a memory segmentation fault. Probably trying to do something that used to be okay on MSDOS but not on a modern OS.

I don't have enough spare time to get it working, but there are many capable coders out there, and I am sure one of them could get it running fairly quickly.

I notice that most of the code is doing lexical analysis, splitting the source into lines and words, turning words into symbols and numbers. 

I have just started learning Python, and it has vast libraries for doing this kind of work. If I were to write one from scratch, I would start with Python to make the job easier and quicker.