There's only one thing that consistently confuses people, it has nothing to do with any of the objections raised in the article, and it's being fixed to be less confusing.
Except that it was mentioned, at the very least tangentially, in the article:
what to do when, for example, someone creates an entry in an array by giving it an ordinal position that doesn't exist. Do we create an array of suitable length and fill it with bits of emptiness in order to maintain the illusion that this array is ordinal?
It may be better to think of it as a key/value table (a "dict" or "hash") that recognizes when it's being used as an array and optimizes its internal storage accordingly. It always has dict/hash operations, but indices of 1..n are stored more efficiently, and a couple functions reflect this common & algorithmically significant case: the length operator returns the n of 1..n, and "ipairs" iterates over the {1, value}..{n, value} pairs. (The un-ordered {key,value} iterator is "pairs".)
Typically, it's clear upfront whether you're using it as an array or as a table; there isn't much overlap. I understand it being confusing in the beginning, but it's really not a big problem, it's just Lua's version of "oh god, the parens/significant whitespace/etc.". (That and indexing from 1. Not a fan of that personally, but whatever, Lua is handy.)
So yes - you can use negative numbers as indices. Or strings. Or other tables, or arbitrary C pointers. Arrays are just handled better.
It has to do with "holes" in the array. If you have a Lua array like: { 1 , 2 , 3 , nil , 5 }. That "nil" causes problems. I, personally, haven't had an issue with this, but it does come up regularly on the Lua mailing list.
