Sam Shepard, who plays the legendary test pilot Chuck Yeager, with the man himself (now a retired brigadier general) and a replica of `Glamorous Glennis', the X-1 named after his wife that crashed the sound barrier.
Some wife ... And the following were headlines in US newspapers:
Retired priest may marry Springsteen Crowds rushing to see Pope trample six to death Sisters reunited after 18 years in checkout line at supermarket
And of course, these kinds of example are restricted neither to newspapers nor to the USA; church bulletins in Britain are also a rich source of grammatical innovation:
The ladies of the church have cast off clothing of every kind. They can be seen in the church basement Saturday.
For those of you who have children and don't know it, we have a nursery downstairs.
Finally, a particularly innovative offer:
Don't let worry kill you-let the church help.
As each of these examples shows, not noticing an ambiguity can have interesting consequences. With the exception of that last case, little harm is done-the consequence is that we simply interpret the sentence the wrong way. But there are other cases where the consequence is that we fail to make sense of the sentence, and we think either that it is ungrammatical, or that a word is missing:
f someone read this sentence thought it was ungrammatical because it missed an `and' between `sentence' and `thought' they would be wrong. The same person might tell the writer that he or she could not understand to get help.
These two sentences are perfectly grammatical, even if on first reading they did not seem to be. You probably interpreted the first as meaning that the person was doing the reading. But `someone read this sentence' can also be used to mean `a person who was read this sentence'-the word `read' is ambiguous between a main verb and a passive verb; a verb like `give' does not suffer this problem, and `The person given this sentence' is easily interpreted as a passive form. The reason readers consistently misinterpret that first sentence in the last paragraph is because they consistently interpret `read' as a main verb when it should be interpreted as a passive. But why?
In the second problematic sentence, the sequence `The same person might tell the writer that' is ambiguous because the word `that' can either introduce whatever it was that was told, or it can introduce some further information about the writer (e.g. `The writer that wrote this book')-'that' can signal two different roles for the following material. If the sentence had started `The same person might want the writer that', there would be no confusion, because the verb `want' is incompatible with one of the interpretations of `that', forcing it to be ruled out.
What appears to be going on in these problematic cases is that there is a consistent preference, consistent across all people, to interpret `read' or `that' in a way that is in fact inconsistent with the rest of the sentence. Again, why?
One further piece in the puzzle is that these kinds of effects can occur even in the absence of words such as `that' and `read', which have more than one meaning. Earlier, we considered the example of being reminded to read something tomorrow. Here is a similar, but now much more problematic version:
I was lent a book that I shall avidly read yesterday.
Most readers will have tried to interpret `I shall avidly read yesterday' as a single grammatical unit, and this turns out to be the wrong decision because the future tense of `I shall read' is incompatible with `yesterday'. The only way to make sense of the sentence (that is, to interpret it grammatically) is to assume instead that yesterday is when the lending must have happened. Once again, we must ask why there is such a consistent preference for one attempted reading rather than for another.
Why we cope the way we do
The reason we all show such consistent preferences in the face of ambiguity has been the source of considerable controversy since the early 1970s. This is not simply because there are different theories regarding the explanation for these phenomena (each theory having important, and different, implications for the way language understanding occurs), but because researchers have often disagreed about how to interpret the experimental tools they employ in their search for an answer (the dip-stick problem outlined in Chapter 5).
The ambiguities we have considered so far arise because at some point in the sentence, it is unclear which grammatical convention is being signalled. One school of thought maintains that there exist preferences to assume certain kinds of convention rather than certain others. In fact, the descriptions that linguists come up with for describing sentences and their meanings suggest that the less preferred interpretations in all the cases described earlier (and there are many others) all require more complex descriptions, and perhaps more complex mental processes. So perhaps ambiguities are resolved in favour of whichever grammatical convention leads to the simplest interpretation.
Another school of thought maintains that when faced with a choice of conventions, we assume that the most frequently used of those conventions is the one that is intended. Of course, it may turn out that the most frequently used conventions (or grammatical structures) are also the least complex, but the basis on which the choice is made would not be in terms of complexity itself.
The choice could even be made on the basis of which structure is most frequently associated with the individual words in the ambiguous part of the sentence. For example, `She told' can be followed by all sorts of things:
She told ...
... Sam.
... a story.
... Sam off.
...Sam a story.
... Sam about the Princess.
... Sam that it was bedtime.
Each one of these corresponds to a different structure associated with `told', and they are all used with different frequencies. When we read `She told Sam. . . ', perhaps we anticipate whichever grammatical structure is most frequently associated with the word `told'. Similarly, `read' as a main verb occurs more commonly than `read' as a passive, so perhaps we assume the main verb form just because it is the most frequently occurring.
What makes these schools of thought so interesting is the counterintuitive claim they make about the way language understanding proceeds. They both agree that ambiguities involving grammatical structure (that is, involving a decision with respect to which grammatical convention is intended) are resolved without taking into account the meaning of the words involved.
This last idea sounds absurd. Surely we do take into account meaning?
Putting things in context
Until the mid-1980s, the majority of work on these ambiguous sentences involved looking at the problems people have with these sentences in isolation-that is, without the usual context that night accompany these sentences in normal conversation. (We shall come to how this was done in a short while.) Similarly, many researchers interested in how children learn the significance of different kinds of grammatical structure studied children's understanding of different kinds of sentence, without taking into account the sorts of context that would be appropriate for those sentences. But in the early 1980s, Stephen Crain discovered at the University of California that children's abilities changed quite remarkably as soon as they were given the sentences in a context that `made sense'. For instance, it was well established that young children have problems with certain kinds of relative clause-the underlined sequence in the following example:
The elephant agirafie that bumped ay,ainst lay down and went to sleep.
(Elephants and giraffes are apparently more interesting to children than many other things.) Researchers can investigate children's ability by asking them to act out whatever the sentence means using toy models. And often, children would act out the elephant sentence assuming that it was the elephant that bumped against the giraffe, and not the other way around-they would not distinguish between that last sentence and this:
The elephant that bumped against a giraffe lay down and went to sleep.
But what Crain took into account was that, generally, relative
clauses are used so that the listener or reader can figure out which elephant is being referred to, on the assumption that there is more than one. In a variety of studies, he and his colleague Henry Hamburger found that children could act out the event described by each sentence, but only if there was more than one elephant. Children could even produce the appropriate relative clauses when shown these events. So whereas research into how children interpret isolated sentences concluded that children of a certain age could not distinguish between certain kinds of grammatical structure, research which embedded these same sentences in an appropriate context showed that children could distinguish between the different structures. The moral for research into adult sentence processing was clear: sentences should be studied in the contexts in which they might normally occur.
There followed, as a result of this research, a number of experiments into adult sentence processing that tried to find out whether some of the preferences we encountered earlier would go away if the sentences were embedded in suitable contexts. One common technique for finding this out involves seating people in front of a computer screen and measuring the movements of their eyes as they read a sentence. There are different ways of keeping track of eye movements, but perhaps the simplest involves shining an infra-red beam into the eye. Some of this is reflected back from the white of the eye (the sclera), and the more the eyes turn, the more white shows, and the greater the reflection. This can be measured using a photo-cell, which puts out a voltage in proportion to the amount of reflection. By having the person look at different parts of the screen, one can equate the voltage with position, and so figure out where the eyes are looking at each moment.
Consider the following sentence (which is a slightly modified version of one of the earlier problematic ones):
Sam told the writer that he couldn't understand to get some help from a decent editor.
Because of the preference to interpret `that' as introducing a message (and not as introducing a relative clause), the segment `to get' will be particularly difficult to interpret. Generally, a number of things can happen when reading a problematic region: either the eyes spend more time on it, or they move back to earlier parts of the sentence, or they do both, and generally this is all quite unconscious (see Chapter 11 for more about eye movements). By keeping track of what the eyes do when they get to `to get', we can figure out whether that region is problematic or not. Of course, we need some sort of baseline against which to compare our patterns so that we can be sure that, when there is a problem, the pattern is readily distinguishable from when there is no problem. This can be done by replacing the verb `told' with the verb `asked', which does not allow the word `that' to be interpreted as introducing a message:
Sam asked the writer that he couldn't understand to get some help from a decent editor.
In this case, we would not expect any problem reading `to get', as there is now only one possible interpretation for the sentence (and so there is no chance that there could be a preference to choose the wrong one).
The pattern of eye movements in the `told' case is indeed different from the pattern in the `asked' case, reflecting the difference between when there is, and when there is not, a problem with `to get'. We can now ask what will happen if the `told' case is embedded in a context which mentions that there are several writers (a bit like the several elephants in Stephen Crain's studies), and which also mentions that Sam did not understand one of them.
If Crain's work with children generalizes to adults, we should find that people have no problem reading these sentences, and that the pattern of eye movements is the same irrespective of whether, for instance, the verb `told' is used or the verb `asked' is used. In fact, when the appropriate experiments were run, this is exactly what was foundthe difficulties experienced when these sentences were read in isolation completely disappeared when they were instead embedded in a natural context.
As a result of experiments such as these, it began to look as if at least certain kinds of ambiguity -could be resolved by taking into account the way in which the alternative interpretations (that is, their meanings) fitted in with the context. The general consensus now is that in fact a whole variety of factors influence the decisions that have to be made when an ambiguity is found. The fit with context is one of these factors, but so it seems is the frequency of occurrence of the different structures associated with the ambiguous words. Accounts based on frequency originated in the early 1980s, but it was only in the early to mid-1990s, when a sufficient number of different experiments had been run, all using different sentences, that it was possible to analyse all the data together and see what factors could possibly explain the overall pattern across the different studies. A number of subsequent studies then manipulated these factors, by using, for instance, different verbs that were associated with the same alternative structures but with different frequencies. They showed that there were occasions when interpretations were chosen according to their frequency of occurrence with that particular verb.
Of course, what may be important may not be the frequency of occurrence of the grammatical structures associated with each word, but rather the frequency of occurrence of the different meanings associated with each word. One might well ask how one could tell the two apart, and this is a big problem, because the meaning of a word determines the grammatical structures in which that word can be found. This is one of those cases where two opposing theories have gradually inched closer and closer until they have become, in effect, the same. It is rather like arguing about whether a bottle of wine is half full or half empty.
Where does this leave us? One consequence of different kinds of information being taken into account when resolving ambiguities, is that if one kind of information is missing-for instance, if there is no contextual information available on which basis to interpret what is being read or heard-other kinds of information will be used instead. Many of those cases where preferences were observed when sentences were read in isolation can probably now be explained by frequency of occurrence. But there is still much controversy over exactly what kinds of information are used, and whether certain kinds of information take precedence over certain other kinds. Is the plausibility of the alternative role assignments more important than the frequency with which the alternative grammatical structures occur? And what about the fit with context? Does that take precedence over the relative frequencies? Or is precedence the wrong way to think about the interaction between the different kinds of knowledge? Do they instead each narrow down (or constrain) the range of possible interpretations, with no one kind of information more or less influential than any other kind? There are, currently, no hard and fast answers to these questions, but we shall return to them in Chapter 13, and will discuss a possible resolution of some of them.
A role for prosody
Many kinds of information may influence the decisions that must be made when ambiguities are encountered. But until now, we have ignored one potentially important source of information. We have ignored it because we have also ignored the distinction between written and spoken language. What the spoken form contains that the written lacks is, of course, prosody (intonation, rhythm, pausing, and so on-see Chapter 2 for a more complete definition). For instance, the sequence `Sam told the writer that he couldn't understand . . .' would be uttered very differently depending on whether everything after the word `that' was supposed to be something about the writer or something about the message. And this information could probably be used to disambiguate the sequence. But does this mean that it is a waste of time studying these kinds of sentence in their written form? Fortunately (for those people who study them), the answer is `no'. Many of the ambiguities we have encountered are as ambiguous in the spoken form as in the written form. Whether the waiter picks up the pizza with the extra anchovies, or whether he picks up the pizza with the oven gloves, cannot be resolved on the basis of prosodic differences alone-linguists do not always find any in these cases. Instead, this example would be resolved
with reference to whether there was just one pizza on the kitchen table, or more than one. So the principles we derive by studying sentences which might be resolved by prosodic information still apply to other sentences which cannot be resolved (or are not resolved) prosodically.
But the presence of prosodic variation in the spoken form is important in other ways, and not just in terms of being yet another source of information for disambiguating the kinds of ambiguity we have encountered so far. The extent to which the different words may be emphasized, or stressed, conveys important additional information that is not conveyed by the words themselves. In the following dialogue, the word that receives the main stress (in the spoken form) is written in capitals:
Joe: Hey-did you hear? Sam took Mary out and bought her a pizza!
Mike: You're wrong-Sam didn't buy Mary a PIZZA [he bought her something else]
The sequence in parentheses indicates the additional meaning conveyed by the stress. Each of the following could also have been uttered in response to that first `Did you hear?' sentence:
The Ascent of Babel: An Exploration of Language, Mind, and Understanding Page 12