Language Log

Soon before

The story begins with an American Dialect Society posting by Alison Murie on 5/18/05.  Murie found the soon before in "...security official said in an interview soon before the transfer of sovereignty that..." (Seymour Hersh, Chain of Command, pp. 355-6) to be very odd, adding that she would have expected shortly before here.  Others agreed, though I myself found no problem with soon before.

Google web searches quickly turned up a huge disparity between (infrequent) soon before and (very frequent) soon after, an unsurprising difference given OED2's definition of soon as "within a short time (after a particular point of time specified or implied)", which has afterness as well as shortness in it (plus a reference point).  Those (like me) who accept soon before are innovators who have extended the meaning of soon by dropping the afterness component.

Then it turned out that the disparity between soon before and soon after essentially disappears when we look at how soon before and how soon after.  From the Google figures and preliminary judgment collection, it appears that there are three varieties: one without soon before, one with soon before only when modified by how, and one with soon before generally.  I'll speculate about how the intermediate variety might have arisen.

Finally, later discussion on ADS-L suggested that acceptability judgments on soon before might be like acceptability judgments on positive anymore (as in Gas is expensive anymore), in that these judgments are sometimes unreliable.  I'll argue that the two situations aren't parallel, and that judgments on positive anymore aren't chaotic or generally unreliable.


The examples in question are ones like like following:


In what follows, I'll assume that types (2) and (4) are generally acceptable; the meaning of soon is entirely compatible with the meaning of after.  It's types (1) and (3) that we're interested in.

Now, the raw Google web hits:

"soon before" -how	31,900	"how soon before"	20,300	"shortly before"	2,960,000
"soon after" -how	2,670,000	"how soon after"	86,400	"shortly after"	11,400,000
after/before ratio	83.70	after/before ratio	4.25	after/before ratio	3.85

"very soon before"	3,900
"very soon after"	187,000
after/before ratio	47.95

In the left column of the first table, we see the gross disparity between soon before and soon after when soon isn't modified by how, a disparity reproduced in the second table for modification by very.  By themselves, these figures suggest a general disfavoring of soon before, which is entirely consonant with its being an innovative combination.  Still, the numbers for soon before aren't tiny; my variety is well represented.

The center column of the first table has the surprise: under modification of soon by how, the disparity between before and after essentially disappears, falling almost to the level of shortly before vs. shortly after, where after is favored over before, though not hugely.  It looks like there are rather a lot of people who use how soon before, but little or no soon before otherwise.

This impression is borne out by preliminary (and still unsystematic) collections of judgments.  So far most informants fall clearly into three types: full innovators, those who accept soon before generally, in examples (1) and (3); partial innovators, those who accept how soon before (in (3)) but not otherwise (as in (1)); and conservatives, who reject soon before, in both (1) and (3).

Where do the partial innovators come from?  Their comments on examples like (3) are telling.  How else would you say it, they ask?  For soon before in (1), these informants offer paraphrases with other adverbs denoting short duration: shortly before, just before, right before.  But these adverbs either resist modification by how (shortly: ?how shortly before/after) or reject it entirely (just and right: *how just/right before/after).  There are adverbs that are fine modified by how -- long and much, as in how much/long before/after -- but these lack the semantic component of short duration.  Short of recasting the question thoroughly, there's no way to package short duration into a how question.  That is, how soon before fills an expressive gap.  Even if you won't go all the way to (1), you might be willing to go as far as (3).  Yes, this is all highly speculative.

In further discussion on ADS-L (5/19/05), Ron Butters suggested that soon before might be like positive anymore, in that informants' judgments are unreliable, not always in accord with their practice.  But the two situations aren't parallel: a great many positive anymore speakers have been confronted with criticism or correction from others -- the feature has even made it into some usage manuals, as a regional variant to be avoided in formal writing -- while soon before seems to have escaped notice.  Nothing confounds acceptability judgments quite so much as explicit regulation, so that it's scarcely a surprise that some people who use positive anymore claim not to.  (Lots of people who use restrictive relative which -- E. B. White and Jacques Barzun, for example -- claim not to, after all.)  For soon before, there is no explicit regulation and no reason to treat informant judgments as any more suspect than informant judgments on other unregulated features.

In any case, informant judgments on positive anymore aren't simply a morass.  Some people don't use positive anymore and report, accurately, that they don't.  Some people use positive anymore and have escaped explicit regulation or failed to attend to it, and they report, accurately, that they use it (and where they use it).  Alas, some people are unreliable judges.  But that's no reason to throw everybody out; our task is to figure out who's who.

There is one way in which soon before probably is like positive anymore: it's not simply a matter of having the feature or not having it.  Instead, the feature is allowed or prohibited, favored or disfavored, in certain contexts, and the details of these distributions differ from speaker to speaker.  That's the way variation works.

zwicky at-sign csli period stanford period edu

Posted by Arnold Zwicky at May 30, 2005 08:11 PM