Hrm, first I'd better point out I don't agree with the definition of Chaos being an extension of Evil and Law being an extension of Good. Chaos and law are thier own scale which combine with Good and Evil to create a more specific personality aspect, adherance to rules and stuff like that.. but thats just my point of view Anywho..
If you want to use the -1000 to 1000 scale for Chaos -> Law then fine, but remeber doing that in the present code makes "Good" and "Evil" less potent as statements. Someone who was Evil is really evil, yet under your system they are just sort just over quasi-evil.
If you want Chaos to represent the present Evil state then do as you say, Chaos == -1000 and Law == 1000, or what I suggest is you extend the scale a bit, say -1500 is Chaos and 1500 is Law and leave the present Evil/Good at -1000/1000. This way you don't have to change any references to the 1000 mark in either direction, just extend for additional checks, which should mean less to change (all the good/evil stuff stays where it is), and just additions.
Where barbus_007 mentions about flags, there will be fun when dealing with antichaotic and antilawful, as "evil" is termed as "not good and not neutral" thus you have a particular range of alignment. Introduce Chaos and you have antievil flags meaning "not good, not lawful, not neutral, not chaotic" (unless you want to include chaotic in the "evil" definition"
which means more range checks. So you have to not only add your anti flags if you want them, but check the existing for range checks on the alignment.
Other than various output routines to display words rather than numbers when it comes to alignment, I can't think of much else that barbus_007 hasn't already mentioned