hah, I want to say something naughty but I'm afraid of starting a religious conflict. Also the author is just regurgitating what they heard/read.
However I would like to report that in the 70's and 80's I converted millions of lines of FORTRAN code into APL, mostly models from academics. Typically the APL implementation, including the test suite, was 1/100 to 1/1000 the size of the original. From this I assumed the actual information density of FORTRAN was very low. Of course, the ratio of bugs to LOC is roughly the same across languages, all other things being roughly equal. Often my test suite would discover severe bugs in the original unnoticed by the original authors because they didn't have grant money to eradicate bugs, haha.
I never used APL but saw it in Byte magazine, but I did use FORTH.
These are very different languages. APL was tight from its symbolic structure, as I recall, while FORTH was tight due to incredibly recursive code that could be interpreted (including inline assembly language) or compiled.
It was technically possible to have FORTH be smaller than any compiled program because no compiler would do microrecursion to eliminate redundancy, while the FORTH programmer did this by nature and need.
Those were some good days. I remember being in high school and having saved up $150 for an amazing FORTH environment in the 1980s.