Jonathan Falk did a double-take, and quite rightly, when he saw this opening sentence in a recent article by Megan McArdle in the Business section of The Atlantic:
Oddly enough, the New York Times health blog has an item on performance reviews, which suggests that they're probably a bad idea.
Unh? They're saying that the mere fact of the New York Times health blog having an item on performance reviews makes performance reviews ipso facto a bad idea? Could they possibly think that?
Finally the penny dropped, and he realized he was supposed to take the relative clause as restrictive. Under the intended sense, what suggests performance reviews are a bad idea is not the fact of the New York Times health blog having published the item; it is the content of the item.
What has gone wrong with McArdle's writing here? Could the initial misunderstanding be some kind of vindication of the purported that/which rule so beloved of the Fowler brothers?
No, it isn't. But the comma after the word reviews was an intelligibility-wrecking error. The sentence clearly doesn't express its intended meaning in a way that the grammar of Standard English allows.
In the terminology of The Cambridge Grammar of the English Language, McArdle's relative clause is supposed to be an integrated one, integrated into the structure of the noun phrase whose noun it modifies. Integrated relative clauses are not set off with commas. The whole noun phrase, if properly punctuated, would be an item on performance reviews which suggests that they're probably a bad idea.
By putting in her comma, McArdle forced the relative clause to be read (by anyone who follows the usual constraints of written Standard English) as nonrestrictive (or supplementary, in CGEL's terms), and that kind of relative clause it is naturally read as having a clause as the constituent it is semantically associated with — what CGEL calls its anchor. (In She insulted me, which made me angry, the part after the comma is a supplementary relative clause, and its anchor is the clause before the comma. Semantically, it says that what made me angry was the event constituted by her insulting me. Arnold Zwicky has proposed calling this a summative supplementary relative. They have been deprecated by some prescriptive grammarians, a point that I return to at the end.)
If McArdle wanted to have a sort of mental pause before the clause about performance reviews being a bad idea, so that the following phrase came as an afterthought, it would have been better to use coordination and the pronoun it instead of a relative clause:
If not, and she intended a relative clause of the integrated type, she should have left the comma out:
If she wanted to make sure, redundantly, that no one would be tempted to read the relative clause as supplementary and thus nonrestrictive (though they shouldn't, if there's no comma), she could have used the option of introducing it with that, since this virtually never introduces supplementary relative clauses any more; so she could have written this:
(I don't like the double occurrence of that very much, but it's entirely grammatical, and many copy editors would insist that this is the version to pick.)
Another way to accomplish the same thing, since the relative clause happens to be of the kind where the relativized element is the subject, would have been to use a nonfinite clause with a gerund-participial verb as modifier instead of a relative clause:
That's probably the neatest and tightest version, in my opinion. But any of the above suggestions would be all right. What she wrote is not. It's actually ungrammatical. Not because of the which (integrated restrictive relatives allow either which or that at the writer's discretion, though American copy editors insist on needlessly changing every restrictive which to that wherever they can get away with it), but because of the comma.
What do we learn from this?
First, expert speakers and writers make mistakes. When Language Log pours scorn on the promoters of stupid rules that do not reflect the syntactic reality of Standard English and never did (the that/which rule, the mythical ban on split infinitives, the bar against singular antecedents for they, and so on), the point being made is not that there are no rules and you can do as you like when you write. Rather, there are facts about what the correct rules are, and it is possible (if sometimes rather tricky) to figure out what the correct rules say. Speakers and writers who unintentionally deviate from what the correct set of rule specifies have made grammatical mistakes. Their lifelong experience in using the language does give them real expertise, but it does not give them infallibility.
Second, the principle that supplementary relative clauses are flanked by commas and integrated ones aren't is really important. It can make the difference between saying what you meant and saying something totally insane that you didn't intend.
Third, newspapers and magazines still need copy editors. It is just possible that The Atlantic actually employed one on McArdle's post, and perhaps saw the which and popped in a comma before it to comply with the fictive that/which rule, but it seems unlikely. It was probably just that McArdle herself didn't realize that by making her relative clause an afterthought with that comma she had unintentionally made it extremely likely that it would be understood as a supplementary relative, and the main clause would be the obvious anchor.
And finally, a fourth point: you can't base good prescription on bad description. (Recall Zwicky's warning against "unexamined grammatical dogma that's been transformed into folk linguistics (and bad advice)".) There is a prejudice against what Zwicky calls summative supplementary relative clauses (like She insulted me, which made me angry). Some prescriptive grammarians (Merriam-Webster's Dictionary of English Usage, under which, cites Lurie, Barnard, Bernstein, and Copperud as examples) want to say that it is bad to construct sentences with summative relatives. But that simply doesn't enable us to understand what went on when Jonathan Falk (like me, and quite possibly you too) did that double-take on reading the McArdle example.
The misreading occurs because the syntax of English does, unquestionably, allow summative supplementary relative clauses as a fully natural and familiar construction. It's not you that can be accused of a grammatical error because of your misreading: you kept your part of the bargain. You (if you're like Jonathan and me) interpreted the relative clause as supplementary because of that pesky incorrect comma, and since it didn't immediately follow a suitable noun as anchor, you read it as summative. A grammarian who tells you that you shouldn't have because the summative relative is "vague" and therefore bad is a grammarian who's a damn fool. The summative relative is in most cases amply clear and precise enough, fit to be seen in the most excellent prose. But McArdle didn't intend to write one.
[Or did she? After writing this post, I was persuaded by an alert reader to change my mind. See this newer post for the recantation and further discussion.