Andrew Radford - Linguistics An Introduction [Second Edition]

The answer to this question is ‘yes’. Before we turn to experimental results, consider (476), which contains a covert category, namely a silent copy of the bracketed wh-phrase which paintings

(476)

[ Which paintings ] did you speak to Mary about [ which paintings ]?

In this structure, the wh-phrase originates as the complement of the preposition about, and then moves to spec-CP, leaving a trace behind in the prepositional complement position. The trace is an invisible ‘copy’ of the wh-phrase and so has the same grammatical properties as the phrase. Psycholinguists refer to the relationship between the moved wh-element and its trace as a filler-gap dependency:

the higher overt wh-phrase which paintings is regarded as the filler for the lower gap, i.e. the position occupied by the trace.

To study filler-gap dependencies experimentally, psycholinguists have used

several different techniques. One such technique is the probe-recognition task. In a study employing this task, subjects are asked to read sentences such as (476)

from a computer screen, and are then asked to determine as quickly as possible whether certain probe words (e.g. did or to) appeared in the sentence – typically the probe word is displayed by subjects pressing a button as soon as they have read the sentence on the screen, and they then press further buttons to indicate whether the word displayed occurred in the sentence or not. The result for a sentence like (476) is that reaction times (RTs) for more recent items such as to are shorter than for more distant elements such as did. In other words, subjects show a faster reaction time in recognising elements they have recently perceived (probably because they are still present in short-term memory) than for those which are further away from the end of the sentence.

This kind of recency effect can be used to investigate the role of trace copies of moved constituents. Consider the following examples:

(477) a.

John argued that Alex had seen the boys

b.

The boys argued that Alex had seen John

c.

The boys argued that Alex had seen them

d.

[Which boys] did Alex argue that he had seen [which boys]?

In all cases, the probe is the word boys, i.e. subjects have to decide as quickly as possible whether boys has occurred in the sentence they have just read (of course in an actual experiment, there will be many different sentences with many

different probes, and the order of presentation of examples will be carefully controlled). For (477b), RTs are significantly longer than they are for (477a).

This can be put down to the recency effect we have just described. Interestingly, RTs to (477c) are also significantly faster than they are to (477b), despite the fact that boys is equally distant from the end of the sentence and the appearance of the probe in both cases. However, (477c) contains them, which can be interpreted as

370

senten ces

co-referential with the boys, as a very recent item. It is plausible, therefore, to suppose that the boys in (477c) behaves as if it were in the position occupied by them, thereby giving rise to a recency effect. The most interesting result, however, is that RTs to the probe boys in (477d) are similar to those in (477c), and again significantly shorter than those in (477b). This means that there is a recency effect in (477d), too – and the only candidate for explaining this in (477d) is the trace copy of the moved wh-expression.

What this finding shows is that when subjects process wh-questions such as

(477d) and reach the position from which the wh-expression has been extracted (i.e. the position occupied by which boys in 477d), the syntactic information contained in the wh-phrase is reactivated. Otherwise, there would be no recency effect for the probe word boys. This can be accounted for straightforwardly if we suppose that movement is a copying operation, and that a silent copy of the moved wh-expression which boys remains in situ (as the complement of seen) in (477d).

The experiment shows that listeners reconstruct the relationship between a trace and its antecedent (i.e. the moved item to which it is related). (exercises 1 and 2).

Strategies of sentence processing

So far, in this section, we have shown that some notions from syntactic

theory such as constituent structure and empty categories are useful for understanding human sentence processing. This, of course, is consistent with the theory of grammar being directly interpreted as a theory of linguistic performance.

However, we shall now see that certain processing principles or strategies,

which have no place in a theory of competence, must also be operative when we process sentences. Specifically, we will look at three types of processing difficulties (involving structural ambiguities, centre-embeddings and garden-path sentences), which demonstrate that some sentences are difficult to process even though they are perfectly grammatical and do not contain any difficult words.

Structural ambiguity (see section 23) may cause processing difficulties. In fact, many of the sentences that we hear in our everyday conversations are ambiguous.

Typically, however, these ambiguities do not impede communication. Indeed, we are rarely even aware of the occurrence of an ambiguity, and we generally come up with only one interpretation for each sentence, which, in the vast majority of cases, is the correct one.

Suppose, for example, that somebody who knows the grammar of English but

who is unfamiliar with regional British culture is confronted with the following sentence:

(478)

Scotsmen like whisky more than Welshmen

This sentence has two interpretations, which can be paraphrased as (479a, b):

(479) a.

Scotsmen like whisky more than Scotsmen like Welshmen

b.

Scotsmen like whisky more than Welshmen like whisky

Sentence processing

371

The question of which interpretation is the appropriate one cannot be decided by just looking at the individual words in (478), as their meaning remains the same on both readings. The ambiguity of (478), then, must be a structural one. In other words, the grammar of English allows two different syntactic representations to be assigned to (478), each of which is associated with a different interpretation.

Hence, the difficulty of comprehending (478) results from its structural ambiguity, and since in the case of (478) there is no preferred interpretation, people typically rely on non-linguistic clues that indicate to them which interpretation is the intended one. For the case under discussion, if we equip our listener with the knowledge that whisky is the national drink of Scotland, this might be sufficient to establish a preference for the interpretation in (479b). However, this preference would not be strong and would almost certainly be overridden in a context where Scotsmen were observed fighting Welshmen (exercise 3).

To understand how the ambiguity of (478) arises, consider again (479a, b). Now assume that there exists a process of ellipsis which can erase words in the second clause that have already occurred in the first clause, but that these deleted elements remain visible to interpretation. Under these assumptions, (478) can be seen as a

‘shortened’ version of either (479a) or (479b). The two options are illustrated below, with strikethrough used to mark material which undergoes ellipsis (notice that Welshmen functions as the complement of the verb like in 480a, but is subject of like in 480b):

(480) a.

Scotsmen like whisky more than [Scotsmen like Welshmen]

b.

Scotsmen like whisky more than [Welshmen like whisky]

In other cases of structural ambiguity, we seem to strongly prefer one interpretation over the other quite independently of linguistic and non-linguistic context, and it is in connection with examples of this type that perceptual strategies become very significant. Consider the example in (481):