Basically, we can only speculate about why the him/himself constraint exists. Most people agree that it has its roots in abilities that are innate. But as suggested in Chapter 4, what is contentious is not that such abilities might exist, but that they might be specific to language.
Role assignment: faster than the speed ofgrammar
`Him' and 'himself' re just one kind of pronoun. There are others which have completely different properties, and are interesting because they tell us something fundamental about how roles are assigned during sentence processing. About how, in effect, we ignore grammar. The pronouns in question are, in fact, the pronouns in questions.
If Bertie tickled Cecily, one can ask `Who did Bertie tickle?'. `Who' is a pronoun whose function (in effect, its meaning) is to get you to recall, from your knowledge about Bertie (if you have any), whoever he tickled. Questions of the `who' kind are clearly and systematically related to the statements that answer them-a single statement could answer many different who- or what- questions:
Cecily hoped her second cousin would propose marriage to her.
Who did Cecily hope her second cousin would propose marriage to?
What did Cecily hope her second cousin would propose to her?
Who did Cecily hope would propose marriage to her?
What did Cecily hope?
... And so on.
Each question is related to its answer systematically. In `What did Cecily hope her second cousin would propose to her?', the `what' refers to whatever would ordinarily occur (in the active statement version) after `propose' and before `to'. It is as if, in the question, something is missing in this position-as if there is a gap between the two. We can represent this by marking the questions as follows:
Who did Cecily hope her second cousin would propose marriage to _?
What did Cecily hope her second cousin would propose - to her?
Who did Cecily hope would propose marriage to her?
What did Cecily hope _?
... And so on.
The idea that each question contains a gap helps explain how we interpret these kinds of questions. The point of asking a question is to find out who fulfilled some particular role. And to figure this out, we can look for where the gap is. We do that by finding the position in the sentence where something is missing. In `Who did she propose marriage to?' there is something missing after `to' and before `?'. After identifying the location of this gap, we can then figure out what role the thing that should be in this position would have (in this last example, the person whom the marriage was proposed to). We can then, finally, retrieve from our memory the information about who fulfilled that role.
There are other kinds of sentences that also have these gaps. Sentences with relative clauses:
The person who Cecily eventually proposed marriage to was her second cousin
They are interpreted in a similar way-in order to know which person is being referred to, you need to identify the gap, identify the role that would normally be associated with something in that position, and so figure out, in this case, that Cecily proposed marriage to this person. Hopefully you then know who this person is.
The fact that relative clauses and questions starting with `who' or `what' or `which' are very similar probably explains why `who' and `which' can start both questions and relative clauses ('The boy who ...', `The car which . . .'). Arid despite the fact that it is considered bad grammar to say `The book what I bought', people do say it!
So far, nothing has been said that is too surprising. Locating the gap is obviously an important thing to do. We do it every day, whenever someone asks a question or uses a relative clause. Sometimes, we have to wait quite a while before getting to the gap, as the second of these examples shows:
Which woman did Bertie present a wedding ring to before falling over? Which woman did Bertie present the wedding ring to his fiancee in front of
(Probably, this second question was a little difficult to process, and we shall come to why shortly.) In each case, identifying where the gap is is crucial. It tells us which woman is being referred to-whether it is the woman doing the receiving, or the woman in front of whom the whole thing took place.
But here, now, is the surprising thing. We do not wait to find out where the gap is. But if not, how else could we work out which woman was being referred to?
The verb `present' allows two kinds of object: the thing being presented (the wedding ring) and the thing, or person, it is being presented to (the woman or the fiancee). The roles associated with these two kinds of object are defined by the meaning of the verb `present'. As soon as we hear (or read) the sequence `Which woman did Bettie present . . .', we can infer, from the meaning of `present' that there are two possible roles up for grabs-the thing-that-was-presented role, and the person-that-it-was-presented-to role. In principle, therefore, we could simply take our pick. How? By selecting whichever role could most plausibly fit the woman.
But how do we know that this is what happens, and that we do not wait around until we have worked out where the gap is?
This question was addressed in a series of elegant experiments run in the late 1980s and early 1990s by a research group in Rochester, USA, headed by Michael Tanenhaus. He and his colleagues reasoned that there was a very simple way to distinguish between the two possibilities, by using two sentences that were almost identical:
Which woman did Bertie present a wedding ring to
Which horse did Bertie present a wedding ring to _?
If we start figuring out what role the woman or horse is supposed to be playing as soon as we get to the verb `present' (and without waiting to find that gap), we should realize that the woman or horse is either the thing being presented, or the recipient of whatever it was that was presented. When we get to `wedding ring', we realize this wedding ring is the thing being presented,. and that the woman or the horse must therefore be the recipient. However, a wedding ring is a plausible thing to present to a woman, but not to a horse. And in principle, we ought to notice the implausibility of this (in the horse question) as soon as we get to the wedding ring.
If we did wait until finding the gap before figuring out that the woman or the horse is the recipient of the wedding ring, we would not notice the implausibility of the horse-as-recipient until the gap itself. So all we need is a way of figuring out, as psycholinguists, when the implausibility is noticed. To cut the methodological story short, Tanenhaus and colleagues found that the implausibility is noticed when the wedding ring is found, not when the gap is found. Here is how they found it out (if you prefer, you can skip this paragraph).
There are several ways of figuring out when an implausibility is noticed. One of them is relatively simple-you sit a person in front of a computer screen and arrange things so that each time they press a button, the next word in a sentence appears on the screen in front of them. By recording how long it takes each person to press the buttons, you can get an idea of how long it took them to read each word in the sentence. Experiments such as this show consistently that if a word is implausible it takes longer to read. Another technique is technologically more exciting, and involves measuring the electrical activity produced in the upper layers of the brain. This requires the use of an electroencephalogram (EEG), a device that is routinely used in hospitals to monitor brain activity, and which is now used in a number of psycholinguistics labs. In the early 1980s, Marta Kutas and her colleagues at the University of California in San Diego demonstrated that precise and identifiable patterns of brain activity could be found in response to words which were contextually implausible. Such a pattern would be found after the word `horse' in `Bertie presented a wedding ring to his horse before falling off'. But this same pattern would not be found in `Bertie presented a sugar-lump to his horse before falling off'.
In the sentences used by Tanenhaus and his colleagues ('Which woman/horse did Bertie present a wedding ring to _?'), waiting until the gap would unambiguously tell us which role should be given t
o the woman or to the horse. But we do not wait-what we do instead is take a gamble, and assign a role to the woman or horse as soon as we have one to assign. In fact, the available evidence suggests that when we have heard or read the fragment `Which woman did Bettie present . . .', we first assume that the woman is the thing being presented (as in `Which woman did Bertie present - to his mother?'). Only subsequently do we revise this assignment when we see that the fragment continues `Which woman did Berrie present the wedding ring ...'. This explains why one of the earlier examples may have seemed a little awkward:
Which woman did Bertie present the wedding ring to his fiancee in front of _?
The reason it may seem awkward is that on encountering `the wedding ring' we immediately assume that the woman was the person that the wedding ring was presented to. But when we then find `to his fiancee', this assumption turns out to be wrong, and we have to `unassume' that the woman received the wedding ring-we have to undo that particular role assignment, leaving the woman role-less, until the end of the sentence. Of course, if we bothered to wait until finding the gap, we would not have to revise the initial assignments.
In general, what seems to happen is that as soon as we know which roles are being played out, and as soon as we know which participants are involved, we try and fit the two together. We do not wait, necessarily, until the point in the sentence at which the grammar confirms that we have made the right fit. In this sense, we ignore the grammar. But why are we so impatient?
Sentence processing is something that happens in time. And in that time, we need to integrate the information we come across with all the information we have built up so far. In the examples about Bertie, his horse, the wedding ring, and his fiancee, we could in principle wait around until our knowledge of the grammar (and hence our identification of the gaps) unambiguously told us which were the correct role assignments to make. But this would mean that certain participants would be role-less, at least for a while. Could it be that we assign roles early, without waiting around, simply to avoid a role-less vacuum? Why might we do this? Why do we want to avoid such a vacuum? And what about all the languages like Japanese and Turkish in which the verb comes at the end of the sentence? In these languages, the identity of the roles is only revealed after all the participants in the sentence have already been encountered.
In fact, these last languages are a little like German, which uses different versions of the word for `the' to indicate certain aspects of the role that the subsequent noun plays in the sentence (this was discussed in more detail in the last chapter). There is an important difference in German between `der Junge' ('the boy') and `den Junge' (also `the boy')-'der' indicates that the boy is the subject of the verb, whereas `den' indicates that it is the object of the verb (and there are finer distinctions also, regarding what kind of object). These case-marked languages therefore provide clues to the listener or reader about which participants did something, and which had something done to them, without the listener knowing what that actual something was. Even in these languages, then, the participants do not remain absolutely roleless. Perhaps there really is some underlying need to allocate each participant, as it is encountered, at least some kind of role. But this still leaves unclear why we should be in such a hurry to do this, and why we are even prepared to make preliminary role assignments which must subsequently be revised. For now, we shall just have to accept that this is the case. We shall return to this issue in Chapter 13, armed with some insights borrowed from computational models of learning.
Obeying the conventions
Much of what we now know about when specific roles are assigned during sentence processing has come from the study of sentences with gaps. But aside from their use as a kind of tool with which to study the nature and timing of role assignment, they are also interesting in their own right. We have already seen how from a single statement, many different questions can be formed, each with their gaps in a different location. But there are constraints on where the gaps can go. In the following examples, the questions marked with an asterisk are ungrammatical:
She thought Bertie's offer of marriage was long overdue.
What did she think _ was long overdue?
* What did she think Bertie's offer of _ was long overdue?
* What did she think Bertie's _ of marriage was long overdue?
* Who did she think _'s offer of marriage was long overdue?
* What was Bertie's offer of marriage long ?
If, from that first statement, we can create that first question, why can we not also create the others? In each case, it is actually quite easy to figure out what the intended interpretation is. But why are they so definitely ungrammatical? Nothing appears to be gained by preventing these questions, and yet something prevents us from interpreting them as grammatical. Similar constraints on where the gaps can go exist for the relative clause equivalents:
The marriage that she thought - was long overdue took place last week.
* The marriage that she thought Bertie's offer of_ was long overdue took place last week.
Unlike the him/himself facts that we encountered earlier, and which apparently apply to all languages, the facts about where gaps can and cannot appear change according to which language is spoken. In English we can easily ask `Who did Bertie's mother sit next to -at the wedding?'. In a language like German one has to ask instead (translated literally) `Next to whom did Bertie's mother at the wedding sit?'. The problem in German is that one cannot have a gap after a preposition, and to avoid this, one has to move the entire prepositional phrase to the front of the sentence.
So what does all of this mean? Languages contain constraints on what can and cannot be said. Some of these constraints apply across all languages, and some do not. In some respects many of these constraints appear quite arbitrary-that is, in principle we could understand some sentences that violated the constraints. So why bother with them?
There is a tension between, on the one hand, those facts which show a rigid obedience to what the conventions of the language dictate (no matter how arbitrary the conventions appear) and, on the other, those facts showing a blatant disregard for these conventions when we ignore what they have to offer and instead second-guess the correct assignment of roles to the different participants in a sentence.
The tension is, in fact, easily resolved, because we do not actually disregard the conventions when assigning roles. When we second-guess which roles to assign to which participants, we use the actual location of the gap (that is, we use the conventions within the language which tell us where that gap is) to confirm or disconfirm the original assignments. And if those assignments are disconfirmed, new assignments are made on the basis of where the gap was found. And in sentences without gaps ('Bertie loves Cecily'), we still need those conventions to tell us which participants in which locations should receive which roles. The only puzzle that needs to be explained from all this is the reason why there exist conventions which apparently serve no purpose whatsoever.
One possible solution is to assume that the seemingly arbitrary constraints on what can and cannot be done arise because in fact they reflect underlying, nonarbitrary, conventions. Here is one example of how this might work for English: the convention for interpreting questions like `Which woman did Bertie marry _?' says, in effect, `If you are trying to identify which participant is intended, you need to establish first what role it played. You can do that by looking at the verb, and seeing which roles are going, and you can then work out which role is most suitable for the participant you are interested in. Later on, you will encounter a gap in the position that will unambiguously signal which role should have been assigned to that participant, and this will either confirm or disconfirm your earlier assignment'. The important thing here is that the role that is played by the thing you are trying to identify is a role that is indicated by the verb. But not everything in a sentence can be assigned a role handed out by the verb. Here is that earlier example again:
S
he thought Bertie's offer of marriage was long overdue.
The two verbs are `thought' and `was'. The verb `thought' signals two roles-one for the person doing the thinking (the subject of the verb), and one for the thing that was thought (the object). In this case, the subject is `She', and the object is `Bertie's offer of marriage was long overdue'. The verb `was' also signals two roles-one for a subject (Bertie's offer of marriage') and one for an adjectival phrase (`long overdue'). But this means that `Bertie', `offer', `marriage' and `overdue' do not receive roles from the verbs. So the convention which allows us to interpret a question (or a relative clause, come to that) will not quite work if we try and ask something like `What did she think Bertie's offer of _ was long overdue?', because the gap is in a position (corresponding to `marriage') that does not receive a role from the verb. A similar problem arises in the relative clause construction `The marriage that she thought Bertie's offer of - was long overdue took place last week'. Once again, the gap is in a position that does not correspond to one of the roles assigned by the verb. So things are not as arbitrary as they may at first have seemed.
The Ascent of Babel: An Exploration of Language, Mind, and Understanding Page 14