re.findall() returns all non-overlapping matches. We need to use a positive lookahead on the third part of the pattern to stop it from consuming the non-character:

  re.findall(r"[^a-zA-Z]([a-zA-Z]+)(?=[^a-zA-Z])", ..)