The following quote regarding the use of \+ in DCGs
comes from this email by Richard O'Keefe:

    | Date: Thu, 5 Jun 2014 17:40:07 +1200
    | Subject: Re: [PROLOG-STANDARD] DCGs
    | From: ok@cs.otago.ac.nz

BEGIN QUOTE

    The original definition of DCGs made some notable
    omissions.  It turns out that there is good reason.
    Waving my hands vaguely in the air,
    if G is a "nice" grammar,
       NT is a nonterminal of G
       S0 is a ground sequence of tokens, and
       S is a suffix of S0
    then we would like or hope it do be the case that
       G |- phrase(NT, S0, S)
    if and only if
       S0 \ S is in the language of NT with respect to G.

    Now suppose we have a language in which a singular
    noun phrase can be a proper noun NOT followed by the
    word "maa".  (This is based on a real language.)
    So we have a rule
        np(s) --> proper_noun, \+[maa].
        proper_noun --> [hone]. % amongst others

    So we expect that
        phrase(np(s), [hone,maa], [maa]).
    After all, [hone,maa]\[maa] = [hone]...

    But this will FAIL because  \+[maa] looks ahead,
    in a completely unconstrained way, into part of
    the input that isn't supposed to be there.

    Worse still,
        phrase(np(s), S0, [])
    will *never* be able to generate a singular noun phrase
    using this rule.

    We can look at this another way.  The translation of
    negated bodies offered later in the document produces
    a use of negation-as-failure which is pretty much
    guaranteed not to be sound.

    You might want to object, "but since \+ is definable
    in terms of (_ -> _ ; _), if-then-else is subject to
    the same problem.  So why not reject that too?"  And
    the answer is that I *do* reject if->then;else in a
    grammar rule *except* where the test that drives it is
    {pure Prolog} not involving any terminals or non-terminals.

    I would very strongly urge that \+Body and (Body->_;_) and
    (Body->_) where Body contains any terminal or nonterminal
    should be rejected by the standard.  I would go so far as
    to say "let it be a type error" that *must* be reported.
    If someone wants to offer the general case as an exceptionally
    confusing extension, we can't stop them, but the blood should
    not be on *our* hands.

    What about cuts, then?   The answer to that is that Prolog
    programmers are *taught* that cuts are potentially dangerous
    and can destroy reversible execution and other logical
    properties.  if->then;else and \+negation are regarded as
    (at least somewhat) safer alternatives.

END QUOTE