That/Which

[Arnold Zwicky]

various people are citing style sheets with the second part of the
  WHICH, not THAT, for nonrestrictives;
  THAT, not WHICH, for restrictives
rule (in some version or another), but the original question was about
the origin of this rule, the second part of which is not grounded in
the practice of careful writers, british or american.

MWDEU, discussing fowler 1907, hypothesizes that the reasoning went as
follows: "if _that_ was being confined to introducing restrictive
clauses, might it not be useful (as well as symmetrical) to confine
_which_ to nonrestrictive clauses?"  this is/was a proposal to
*change* the practices of most (and most of the best) writers - to
change it to put relative WHICH and THAT into complementary
distribution with one another.

copy editors with whom i've contended on this matter almost invariably
cite the complementary distribution justification, in the form:
there should be only one correct choice in any given context.  in
combination with a desire for rules that are as simple as possible
(none of this stuff about euphony, prosodic heaviness, parallelism,
avoidance of repetition, shades of (in)formality, etc.), you're
driven to prescribing only THAT for restrictives.

now i think the idea that there's only one correct choice in any
given context is preposterous.  but for some folks this is a moral
principle.

faced with data about the actual practices of good writers, many copy
editors, in my experience, simply reply that that shows these
writers aren't good enough; nobody's perfect, everybody makes
mistakes.  i mean, the rule's there in the style book, it's easy
to apply, and it makes sense.

arnold (zwicky at csli.stanford.edu)