Per the docs:

Under natural sorting, strings are splitted at word and number boundaries, and the resulting substrings are compared as follows:

* numeric substrings are compared numerically
* alphabetic substrings are compared lexically
* numeric substrings come always before alphabetic substrings