January 22, 2004

Deny, disprove, refute

Here's some more on the verb refute, which Geoff Nunberg and I have written about here recently. It's more than any sane person would care to know about refute, actually, but I hope I can convince you that some more general issues emerge.

Originally, I cited a case where a college newspaper reporter wrote that "[i]n his weekly radio address last Saturday, Bush refuted that the law sets unreasonable standards." This differs from standard usage in two ways: first, the context suggests that the writer means only to say that Bush denied something, not that he presented arguments to prove it wrong; second, refute is here used with a "sentential complement" rather a "noun phrase object", i.e. refuted that such-and-such is so rather refuted such-and-such. Geoff assured us that the American Heritage Dictionary's august usage panel has this problem in its sights..

In my earlier post, I did a bit of googling, and concluded that refute with sentential complement is a pretty rare bird: "refute/refutes/refuted the" is about 115 times commoner than "refute/refutes/refuted that the". By comparison, claim/claims/claimed the is only about 1.17 times commoner than claim/claims/claimed that the, a difference in ratios of two orders of magnitude. Philip Resnik pointed out that Alta Vista (unlike Google) allows both start and end times for searches, making it easier to track changes over time, so I redid the same searches on Altavista for three two-year periods starting at the beginning of 1998. The results were pretty stable, or at least don't show any consistent trend (except for the volume of indexed material to grow!):

 
1/1/98-12/31/99
1/1/00-12/31/01
1/1/2002-12/31/03
refute/
refutes/
refuted that the
35
107
601
refute/
refutes/
refute the
2,776
8,916
45,762
RATIO
120
129
116

So if the sentential-complement use is a change in progress, it's not happening fast enough to be seen at this temporal resolution. Of course, this is what we expect if it's happening on the time scale of a generation or two. But another possibility is that this is just a sporadic difference in how learners of English generalize from the examples they hear and read.

Now, this particular usage question is of little real interest in itself. However, it exemplifies a range of interesting questions about how and why ideolects vary, and how individual variation is related to large-scale change across space, time and communicative connectivity. The technology of networked computing might have been designed by a cosmic sociolinguist specifically for the purpose of instrumenting these patterns in increasingly minute detail, and over the next few decades, we'll learn a lot about how this works.

Some other issues seem to be lurking here. In general, intrinsically negative verbs having to do with attitudes towards propositions seem to occur with nominal objects much more often than with "that S" complements. In some cases, the preference is pretty much categorical: for me it seems impossible to say "I refuse that S" or "I reject that S". In other cases, it's more of a statistical preference. By comparison, intrinsically positive verbs of similar sorts seem much more likely to permit "that S" complements, and to use them more often when they are possible. The table below glosses over a multitude of problems, but may be enough to support the plausibility of the generalization:

  ___ the ___ that the Ratio
doubt
1,140,000
681,000
1.7
deny
796,000
115,000
6.9
disbelieve
15,400
979
16.2
disprove
76,800
2,310
33.2
distrust
37,500
742
50.5
refute
122,000
1,050
116.2
misunderstand
72,200
524
137.8
ignore
1,880,000
8,710
215.8
disavow
12,800
34
376.5
reject
901,000
2,160
417.1
negate
140,000
276
507.2
refuse
249,000
393
633.6
contradict
201,000
314
640.1
falsify
26,400
23
1147.8
 
conclude
351,000
762,000
0.5
conjecture
30,900
43,100
0.7
assert
175,000
193,000
0.9
believe
4,070,000
3,330,000
1.2
prove
1,190,000
647,000
1.8
verify
2,110,000
1,010,000
2.1
affirm
338,000
112,000
3.0
confirm
2,090,000
647,000
3.2
reckon
53,200
16,400
3.2
demonstrate
2,280,000
699,000
3.3
acknowledge
1,170,000
318,000
3.7
know
11,000,000
2,650,000
4.2
recognize
2,840,000
529,000
5.2
understand
7,570,000
950,000
8.0
establish
2,630,000
225,000
11.7
trust
2,700,000
120,000
22.5
validate
610,000
27,000
22.6

So "refute that S" is fighting a larger-scale battle, it seems. Is this apparent pattern real? Is it part of a larger one? What's the cause, if any?

Here's another issue. At least in principle, the choice of syntactic frame for refute -- sentential complement vs. nominal object -- is independent of the choice of meaning -- "prove to be wrong" vs. "deny". Some of the sentential complement uses clearly mean "prove to be wrong", e.g. this statement from a document on the EPA's web site:

The On-site Verification test should be able to be completed within an hour, and is used in the
field to verify or refute that the waste is behaving as predicted from the Characterization and
Compliance testing.

and similarly, this sentence from a dissertation on "Vertical Integration in Commercial Fisheries":

Speculation aside, however, there is no empirical evidence to confirm or refute that the use of quota management actually leads to increased vertical coordination.

Other uses of the same syntactic pattern clearly mean nothing more than "deny" or "dismiss":

Pavkovic did not admit - but neither did he refute - that the Battallion was filled by members and supporters of Momir Bulatovic's Party.

I have no doubt that this woman was the sneak who called security and hotel management and I absolutely refute that the shout "Look out, here come the Indians! Circle the wagons!" was made.

But are the two variables really uncorrelated? Since deny is one of the most sentential of the negative verbs in the list above, whereas disprove is about five time more nominal, and falsify is the most nominal of all, might we not expect that that refute would be more likely to become sentential for people who use it to mean deny, as opposed to people who retain the standard meaning of falsify? I didn't tally things up to check on this, but as I was poking around, I got the impression that it's not (very) true.

Finally, here's another kind of evidence about usage standards, brought to my attention by David Nash.

According to John Quiggin, the New York Times ran an AP wire story under the headline "Powell Refutes Report saying U.S. Overstated Iraq Threat", where "[t]he body of the article makes it clear that Powell said he disagreed but produced nothing that would prove the report false." However, within hours of his post, the headline had been changed to read "... dismisses ..." This kind of correction of on-line news stories is very common, though the changes are more often to add or fix information than to fix language. Still, if you automatically tracked the changes and aligned the versions, it might be possible to see how often various usages are corrected -- a sort of automated editorial usage panel.

 

Posted by Mark Liberman at January 22, 2004 08:24 AM