Actual words were easier for the telegraph operator to send and receive so they were charged less than a scramble of letters or nonsense. Words were charged individually, but could be up to 10 letters long. It makes sense to avoid the shortest words to get some error correction, "— ." = "T E" could be misheard as "—." = "N".

The international regulations limited the cheaper rate messages to only certain languages, of which one was Latin, so the codebooks used that to avoid confusion with modern languages.

(See my other comment and read the introduction in the linked book.)

Presumably has built in error-correction. You don't want to go from t56 "Cholera, family dead, stock market crash" to t66 "Buy farmland, rains predicted good" in one typo.

Can imagine this approach being less error-prone for laymen, in the same way that passphrases (word word word word word) can be friendlier than passwords (word12345678).

Because the telegram company charges more for non words.

Good point to link to what3words.

Have a look, it’s a geographic encoding system.