Yesterday I mentioned Wason and Reich's 30-year-old paper on sentences like No head injury is too trivial to be ignored ("A Verbal Illusion", The Quarterly Journal of Experimental Psychology 31(4):591-97, 1979), and promised to sketch their argument. Here's their basic analysis, in their own words:
There would appear to be at least three sources of potential difficulty in the target sentence. First, its syntax is permeated with negativity. Clark (personal communication) has pointed out that it probably contains four negatives, and this number might be expected simply to overload the channel capacity of the individual and render the sentence incomprehensible. In particular, there are intuitive grounds for supposing that the final negative (the verb ignored) presents the greatest obstacle to the correct interpretation.
Second, the construction, too x to be y-ed, is associated with a relation between x and y such that the more x the less y . Thus, it makes sense to argue that the more small (smaller) a missile, the less it should be banned, the more convincing an alibi the less it should be doubted, the more serious a head injury the less it should be ignored. But No head injury is too trivial to be ignored, presupposes the more trivial a head injury the less one should ignore it. Thus it implies that only trivial head injuries require treatment and serious ones may be ignored. Hence a more consistent version might be: No head injury is too serious to be ignored. However, this sentence is still not easily understood. In fact, like the target sentence, it differs from the missile-banned sentence and the head injury-noticed sentence in yet another way.
This third factor may best be illustrated by considering the injunction that is conveyed by each sentence. No missile is too small to be banned and No head injury is too trivial to be noticed may be construed to mean Ban all missiles and Treat all head injuries respectively. Both these injunctions are consistent with current views on missiles and head injuries. However, the injunction, Ignore all head injuries, which follows from the sentence, No head injury is too serious to be ignored, is unusual if not actually incredible. Although it is possible to construct a sensible context for such a sentence, it is not easy to do so. The sentence, as it stands, is inconsistent with most of our beliefs and opinions about head injuries and their treatment. For this reason it may be difficult to understand, and may be termed a non-pragmatic sentence. In contrast, information that is consistent with our opinions and beliefs about the world, such as the missile sentence, may be termed pragmatic.
A number of letters from psychologists and linguists, as well as face to face encounters, showed that the target sentence is almost invariably misconstrued. Individuals were shown the target printed on one side of a card and the sentence, No head injury is too trivial to be noticed, printed on the other side. Not only was there a pronounced tendency to misconstrue the target in the expected way, but it was also often claimed that the other sentence was in some way ill-formed. Rational attempts to correct this error frequently seemed to generate some affect on the part of the respondents.
Another way to express their theory, I think, is that certain kinds of sentences are so difficult to process that we often (unconsciously) abandon compositional semantics and jump directly to an interpretation that puts the pieces together in a sensible way. They focus on negation as the difficulty, but their "target sentence" has two other common sources of strain, namely a scalar predicate ("too trivial") and an implicit (deontic) modal.
They ran a couple of pilot experiments to verify that people (presumably, in this case, UCL undergraduates) have trouble with sentences of this kind. For example, when presented with the sentence No WUG is too DAX to be ZONGED, six out of ten subjects felt that this implies WUGS should be ZONGED (which is correct, according to Wason and Reich), while four out of ten concluded that WUGS should not be ZONGED.
Having demonstrated that misconstrual of such sentences is "a robust phenomenon", they designed an experiment to "[compare] the understanding of sentences which expressed pragmatic information and sentences which expressed non-pragmatic information", where by pragmatic they mean something like "common-sensical".
Their "pragmatic sentences" were:
1. No missile is too small to be banned.
2. No government is too secure to be overthrown.
3. No dictatorship is too benevolent to be condemned.
4. No weather forecast is too plausible to be mistrusted.
And their "non-pragmatic sentences" were
1. No error is too gross to be overlooked.
2. No message is too urgent to be ignored.
3. No film is too good to be missed.
4. No book is too interesting to be put down.
Each subject was given one sentence of each type to paraphrase, based on these printed instructions:
I am going to give you a sentence which may express a point of view on a particular topic. You are to disregard your point of view, or the point of view you think you ought to hold on the issue, and merely put the sentence into your own words as it stands.
There were 64 subjects (16 pairs X 2 orders X 2 replications).
The results are consistent with their hypothesis (numbers in each cell are correct/incorrect):
|No. 1||No. 2||No. 3||No. 4||Total|
|Pragmatic sentences||16/0||16/0||11/5||8/8||51/13 (80%)|
|Non-pragmatic sentences||3/13||4/12||9/7||11/5||27/37 (42%)|
[Their choice of sentences is open to question, it seems to me. For example, bookish children are often told that they should put down what they're reading, no matter how interesting it is, and go out to play. (At least, this would have been true back in the childhood of 1979 undergraduates, when children still went out to play.) So the correct reading of No book is too interesting to be put down is far from "non-pragmatic" — as their results suggest. And I suspect that (especially where British weather forecasts are concerned) many people retain some traces of the old meaning of plausible as "having a false appearance of reason or veracity; specious", which would move No weather forecast is too plausible to be mistrusted into the non-pragmatic category.]
Wason and Reich conclude with some observations about the peculiar difficulty of the original example, No head injury is too trivial to be ignored:
There may be still another reason why the non-pragmatic sentences were less resistant to a correct construal than our original target sentence … [T]he target presupposes that the more trivial a head injury the less we should ignore it, and it enjoins us to ignore all head injuries. But by merely dropping the negative on the verb, both these implausibilities are simultaneously removed: No head injury is too trivial to be noticed. Hence if the processor fails to handle this negative verb correctly, he reaches an apparently satisfactory interpretation. But suppose that the target were to be presented so that it is implausible in just one of these ways. No head injury is too serious to be ignored, has a plausible presupposition (the more serious a head injury the less it should be ignored), but an implausible injunction (ignore all head injuries). Conversely, No head injury is too serious to be noticed, has an implausible presupposition (the more serious a head injury the less it should be noticed), but a plausible injunction (notice all head injuries).
As a result, they argue, it's paradoxically easier to see the problem with these other sentences, precisely because their problems are less pervasive.
They also connect their target example's peculiar difficulty ("When the correct interpretation was explained it was often adamantly rejected …, as if the informants literally could not see an alternative view. […] [This] is analogous to the susceptibility to the Ames Illusion which, unlike most illusions, also does not fluctuate. Like the Ames Illusion the sentence appears to be normal and regular, and this tempts us into a plausible but wrong interpretation.
This case also helps to bring out the reasons why I feel that the terms "prescriptive" and "descriptive" are misleading. If Wason and Reich are right, then the intuitions of the overwhelming majority of native English speakers are wrong. But few if any "descriptive" linguists would take that as a knock-down argument against Wason and Reich's analysis.
A final note: I haven't found any work on analogies to this illusion in other languages. The details would generally need to be different, but the basic ingredients should all be there. Does anyone know of any cross-linguistic work on such things?
[Update — there seems to be some confusion, perhaps also on my part, about exactly what W & R's terminology means. So let me try to explain, diagrammatically, what I think they were getting at.
Behind the phrase No missile is too small to be banned there is a background assumption that we can diagram something like this:
|← less small||-||more small →|
In other words, on the scale of missile smallness, we presuppose that there is a threshold T such that missiles smaller than T (i.e. farther to the right in the diagram) are (or should be) not banned, while missiles less small than T are (or should be) banned.
Against this background, we assert that T is in fact so far along the scale of smallness that there are no missiles "too small to be banned", i.e. on the not-banned side of T:
|← less small||-||more small →|
It follows that all missiles should be banned.
Now consider No error is too gross to be overlooked. The presupposition is exactly analogous:
|← less gross||-||more gross→|
That is, there's a threshold T on the scale of error grossness, such that errors less gross than T are (or should be) overlooked, and errors more gross than T are not overlooked. But now the analogous assertion can be diagrammed as
|← less gross||-||more gross →|
This means that T is so far along the scale of grossness that there are no errors below T on this scale, and therefore all errors are (or should be) overlooked. This assertion is (in W&R's terminology) non-pragmatic, in the sense that it's contrary to common sense.
If we now examine the original "target" sentence No head injury is too trivial to be ignored, we see that there are problems at both stages. The presupposition
|← less trivial||-||more trivial →|
tells us that there is a threshold T on the scale of head-injury triviality, such that injuries that are less trivial than T are (or should be) ignored, while injuries that are more trivial than T are (or should be) not ignored. This is already backwards — simply by virtue of the meaning of the words, the scale should work in the opposite direction.
And the assertion is equally problematic:
|← less trivial||-||more trivial →|
This tells us that T is so far along the scale of triviality that there are no head injuries below T on this scale (i.e. less trivial than the threshold), and therefore all head injuries are (or should be) ignored.
W&R wanted to make their test a fair one, so they crafted their "non-pragmatic" sentences to have sensible presuppositions with counter-sensical assertions. But (as they point out) the original target sentence No head injury is too trivial to be ignored is bass-ackwards in both respects. This undoubtedly makes it harder to see the problem, since one interpretive flip changes both, and leaves the interpreter with an all-around sensible sentence. In their "non-pragmatic" sentences, the analogous flip fixes one problem at the expense of creating another one.]