Andrew Radford - Linguistics An Introduction [Second Edition]

If the representations of sentences are not retrieved from a memory store, this means that they are constructed on-line (in a step-by-step fashion) in accordance with syntactic principles or rules. It follows that sentence comprehension involves segmenting the sentence into relevant processing units and constructing a syntactic representation for the sentence (the technical term for this is parsing).

But how do we go about processing sentences? According to one view (which

is favoured by many psychologists), speakers/listeners rely on parsing and production strategies that have nothing much to do with the units and operations that linguists employ in their syntactic analyses of sentences. On this view, the detailed tree structures we have been associating with sentences throughout this part of the book bear no relationship to the procedures native speakers employ when parsing.

Alternatively, it has been suggested that such structures do play an important role in sentence processing, to an extent to be determined by psycholinguistic research.

Proponents of this alternative view claim that when producing or comprehending a sentence, we make use of essentially the same processing units and operations as are used in linguistic analysis, such as constituents, tree structures and movement operations. If this is correct, it means, for example, that listeners segment sentences into VPs, TPs, CPs, etc., and that linguistically complex sentences are more difficult to comprehend than simple ones. In other words, the more complex the 366

Sentence processing

367

syntactic derivation (in terms of the operations it involves), the more difficult the sentence is to process. This view came to be known as the Derivational Theory of Complexity (DTC), and many psycholinguists have explored the extent to which the DTC actually holds. When this research began, in the late 1960s, the idea that a generative grammar could provide not just a theory of syntactic knowledge

(competence), but at the same time a theory of syntactic processing (a central aspect of performance) was adopted with considerable enthusiasm. Subsequently, however, these rather naïve ideas have been abandoned, and more complex

questions are now being asked. In what follows, we will look at two sets of

experimental results which suggest that the syntactic constructs theoretical linguists have postulated are in fact used by normal listeners when they process sentences. Positive results of this kind do not, of course, constitute a comprehensive theory of sentence perception. They do, however, indicate that a grammar, as we have understood this notion throughout this book, will be a central component of such a theory.

Click studies

The purpose of click studies is to determine whether listeners segment

sentences to which they are listening into units similar to those postulated in syntactic theory, namely phrases and clauses. In this type of experiment, sentences such as (474) are recorded, and superimposed on each sentence is a ‘click’ or

‘beep’, i.e. a short acoustic signal, which may be located at any one of a number of different places within the sentence.

(474)

The man [who nobody likes] is leaving soon

Immediately after hearing the sentence (including the superimposed ‘click’), subjects are given a written copy of it and are asked to indicate the point in the sentence at which they perceived the click. In sentences like (474), the bracketed clause is a relative clause, in which the relative pronoun who ‘relates to’ the preceding expression the man (see section 18, p. 253). The possible locations of the click for subjects hearing this sentence are indicated by + in (475):

(475) a.

The + man [who nobody likes] is leaving soon

b.

The man [who + nobody likes] is leaving soon

c.

The man + [who nobody likes] is leaving soon

In (475a), the click occurs before the relative clause boundary, in (475b), it occurs after this boundary, and in (475c), it is located exactly at the boundary. Subjects hear a range of sentences of this (and other) structural types with the position of the click systematically varied.

The basic finding in such studies is that subjects misplace clicks towards or into major clause boundaries. An early click, which in the stimulus is objectively located immediately before the noun man in (475a), is reported as occurring

368

senten ces

towards or at the clause boundary (i.e. in the word man or between man and who).

Similarly, a late click located after the clause boundary in (475b) is reported as occurring earlier, again towards or at the clause boundary. By contrast, clicks objectively located at the clause boundary are accurately perceived as having occurred in this position. Similar results have been obtained with respect to the second clause boundary position in (474), i.e. between likes and is, and using a variety of different clause types.

Click experiments are deliberately constructed in such a way as to overstretch a subject’s processing capacity. The task is extremely demanding as it involves two processing tasks to be undertaken simultaneously, the comprehension of the sentences (which can be tested by asking subjects questions) and the location of the clicks. The idea is that because of the demands of the task, the experiment produces errors in click location, and this is in fact what happens. What is most interesting here is the types of errors that the subjects make, which are not random.

Firstly, of the three possibilities, click misplacements tend not to occur for (475c) and other sentences, where the click is located at the clause boundary. By contrast, errors are common in the ‘early’ and ‘late’ conditions of respectively (475a, b).

Secondly, click mislocations tend to go into the clause boundary. These results suggest two things, namely (i) that the placement errors reflect the way the stimulus sentences are segmented into structural units, and (ii) that the clause is the major sentence-processing unit. Using the same technique with different

stimuli has yielded evidence for perceptual segmentation at constituent boundaries within clauses, too, specifically for a constituent boundary before VP, but these clause-internal boundaries give rise to a weaker effect than do major clause boundaries such as that in (474).

Finally, it is important to be clear that ‘common sense’ does not provide an explanation of these findings. For instance, it might be thought that there is a clear

‘acoustic gap’ between man and who in (474) and that it is this superficial aspect of the signal which is ‘attracting’ clicks. But this is not so: acoustic analysis of stimuli used in these experiments indicates that there is no such ‘acoustic gap’ – the speech signal is continuous – and reinforces the conclusion that subjects are relying on a linguistic segmentation of the input signal in their perception of the sentence.

Processing empty categories

As we have seen in section 20, syntactic theory postulates a range of so-called empty categories, phonetically null place-holders that occupy specific phrase-structure positions. Among these are the trace copies left behind by syntactic movement, discussed at some length in section 21. In fact, it is more accurate to refer to such objects as covert categories, since – if the theory is correct – they are not in fact empty of syntactic information. For example, PRO has the categorial status of a D and traces, by virtue of being silent copies of moved items, retain the syntactic characteristics (as a DP-trace, V-trace, etc.) of those items. Is there any evidence

Sentence processing

369

from psycholinguistic experiments which independently confirms that empty categories are involved in the processing of sentence structure?