Fillers and gaps
The discussion of anaphora above looked at the relationships between different parts of the discourse, and pointed out that the interpretation of text or speech often requires the comprehender to make links between anaphors and their antecedents. A further type of linkage that needs to be made, more usually in this case within sentences, is between elements moved to the front of a sentence and the locations in the sentence structure from which they have been moved (Clifton & Frazier, 1989). The fronted elements are known as fillers and the locations they have been moved from are known variously as traces or gaps. Because the traces or gaps have no phonological content, i.e. are not pronounced, they are also referred to as empty categories.
The most obvious fillers are wh-words or phrases such as who in the question sentence in (12.42), or relative pronouns as in examples (12.43) - (12.44). To understand sentence (12.42) appropriately, the comprehender needs to work out that this word represents the direct object of introduce, i.e. the question is asking for the identity of the direct object of this verb.
To see why the comprehension of such sentences is an interesting issue, recall the principles of sentence processing discussed in Chapter 10 in connection with garden path sentences. The model introduced there as a starting point for our discussion argued that a single syntactic analysis is computed, on a word-by-word basis, just as long as the syntactic category of each word can be determined unambiguously. This analysis follows phrase structure rules i.e. the rules that linguists might use in drawing well-formed syntactic trees, which establish the grammatical relations between words. Grammatical relations in English are largely determined by the position a word has in a structure and so the sentences Dogs chase cats and Cats chase dogs are different, even though they contain the same words.
Building well-formed syntactic structures has the potential to become problematic when constituents have been moved from their typical positions, as in (12.42). To start with, the moved constituent, which in the case of (12.42) is the filler who, has to be held in memory while a place is found in the sentence to which it can be linked. It has been shown that keeping words in memory is more costly if there is a longer distance between the filler and its position in the sentence (Gibson, 1998; Warren & Gibson, 2002). Thus, object relative clauses, as in (12.43), impose a greater load on working memory than subject relatives such as (12.44). In object relatives the filler represents a constituent in the position of object to the verb in the relative clause, and therefore after that verb, whereas in subject relatives the moved element comes from before that verb.
Processing of filler–gap dependencies is also made complicated, as we will see through further examples below, by the fact that it is not always easy to determine where the filler belongs’. This is because the gap is not usually marked in any way.
Once the gap has been identified, it is assumed that the processor has to retrieve the filler from memory and integrate it with the relevant parts of the sentence in the case of the sentence in (12.42) this is with the verb (introduce).
It has been argued that wh-question sentences such as (12.42) result from movement operations applied to an underlying sentence such as (12.45), and that although as a consequence these two sentences have different word orders, the argument structure of the verb introduce remains the same. That is, introduce has a subject (John in both sentences), an indirect object (Mary), and an object (which is Bill in (12.45), but is of course unknown and the purpose of the question in (12.42)).
So for the reader or listener to interpret such sentences, the wh-word has to be identified as filling the appropriate argument slot in the sentence. The challenge for theories of sentence processing is to explain how this is achieved. Following on from the argument that wh-question sentences result from the movement of a sentence constituent from its underlying grammatical position in a corresponding declarative sentence (John introduced X to Mary), it is claimed that sentences like (12.42) have a gap in that grammatical position, as indicated by the underscore in (12.46). An alternative representation is to show that there is a residue or trace of that moved element, as shown by the letter t for trace with a subscript in (12.47). Of course, a wh-word at the beginning of a sentence may have been moved from other positions in the underlying structure, depending on what the question is. Other possibilities are shown in (12.48) and (12.49).
The three main tasks that have to be accomplished in interpreting these filler–gap constructions, then, are first to identify the fillers, second to identify the gaps, and third to associate the fillers to their gaps. Most studies of filler–gap dependencies have involved wh-words, in which case the first task, finding the filler, is straightforward. But there are cases where it is not so obvious. The interpretation of the sentence in (12.50) clearly involves understanding that Susan is the subject of Marry, and so it has been argued that there is an underlying gap before the infinitive verb to marry in this sentence. Despite the additional complications thrown up by examples such as this, we will focus here on those filler–gap sentences where the filler is more readily obvious, in the form of a wh-word.
Active filler hypothesis Once a filler has been identified, the next task is to locate the gap from which it might have come. Under the acti e filler hypothesis (Clifton & Frazier, 1989), if there is a filler that is still active, i.e. has not been associated with a particular gap, then locating the gap is a priority. If there is only one possible gap, then this is straightforward, and cases like (12.46) (12.49) above, as well as (12.51) and (12.52) below, are relatively easy to work out.
But placing active fillers into gaps is not always easy. For one, there are often pseudogaps. For instance, (12.53) starts off looking like (12.51), and so there might be a gap after, just as there is in (12.51). However, there is a noun phrase after the verb that could also be the object of that verb. This does in fact turn out to be the case, and so the potential gap after buy is not a real one – the filler should instead be interpreted as the object of the preposition with. If the comprehension system initially tries to link the filler to the pseudo gap after buy, then we have an example of a garden path, i.e. an attachment that turns out to be the wrong analysis.
Another issue is that there may be multiple gaps in a sentence. Take for instance the sentence in (12.54). In addition to the pseudo gap after buy, there are two further gaps here. One is for the subject of decorate (which is filled by John), and the other is for the object of decorate (filled by what). So, the sentence in (12.54) means something like John bought the paint in order that John could decorate what? (12.55) is even more complex, with three actual gaps – the object of buy (what), the subject of paint (John), and the object of with (what). The first and third gaps ultimately need to be filled by the same element – the what (so the sentence means something like John bought what so that John could paint the porch with it (=what)?).
In addition, there are ambiguous sentences involving filler–gap dependencies, such as the sentences which we met in Chapter 2 in connection with wanna-contraction. These are where the verb has transitivity ambiguity, i.e. can be used either with or without an object, as is the case with leave in (12.56). The possible interpretations of this sentence are illustrated in (12.57) and (12.58). In the former, leave is used intransitively, and the question is about who has to leave. In the latter, leave is transitive, and the question concerns who Max is going to quit.
Since wanna-contraction is more likely in the case of (12.58) than (12.57), it would appear that at least some aspect of filler–gap dependencies remains relevant to the final stages of speech production.
Processing strategies for matching fillers to gaps
It has been argued that the processing of filler–gap sentences follows processing strategies. However, as Clifton and Frazier (1989) explain, different and often conflicting strategies have been proposed as the best or most appropriate account. A few of these will be briefly introduced here as illustrations of the complexity involved.
The first is the strategy of gap as first resort( Clifton & Frazier, 1989: 281). This stipulates that the processing system the comprehender will postulate a gap at the earliest possible position. Since the wh-phrases we are considering here are noun phrases, this earliest possible gap will be at the first position where an NP could occur. A corollary of this strategy would have to be that if this first gap turns out to not in fact be a gap but rather a pseudogap, see above then the next gap position will be considered, and so on. It should be clear that this will lead to plenty of gap postulations in (12.59), with the potential gap after ask being cancelled when Meg is read, that after persuade cancelled when Jill is read, and so on. This does not seem to be a very effective strategy in cases like this. But then again, this is a rather unusual though possible sentence.
At the other extreme we find the strategy of gap as last resort (Clifton & Frazier, 1989: 282). In this case the approach is for the processor to delay postulating a gap until it is forced to. Now, though, this wrongly predicts that (12.60) (with kill used intransitively) will be no more difficult to process than (12.61). In both cases under gap as last resort there is no requirement to postulate a gap until the end of the sentence, which turns out to be correct for each sentence. Intuitively, though, (12.60) seems to be more difficult to process, and this appears to be because of the assumption of a gap after kill, i.e. that kill is being used transitively, which then has to be abandoned.
An approach that may be able to find the middle ground between first resort and last resort is one that expects gap postulation to follow lexical expectations. Verbs differ from one another in their expectations regarding transitivity, as we saw in Chapter 11. This can in part reflect the frequency with which they are used in different syntactic constructions. For example, read occurs very frequently with a following object, but although al can take an object it rarely does so. sing this information, the processor may take a gamble, when encountering read with no apparent object, that there is a gap after read with which the filler should be linked. With walk, it makes a different bet, that al has no object. Fodor (1978, 1989) shows that a lexical expectation account leads, correctly, to the pre diction that (12.62) is easier to process than (12.63), and that (12.64) is more difficult to process than (12.65), even though (12.62) and (12.64) have the same structure and (12.63) and (12.65) have the same structure.
Along similar lines, ask prefers an object, but the verb race does not. In (12.66) and (12.67), therefore, there is a strong potential gap after asked. The filler in (12.66) is plausible here reporters do ask witnesses, but that in (12.67) is not reporters do not ask churches. Tanenhaus et al. (1985) found that the sentence in (12.66) is accepted as sensible more often and more rapidly than that in (12.67).
In the pair in (12.68) and (12.69), since race does not prefer an object, a gap after raced is not so tempting. Correspondingly, Tanenhaus et al. found no difference between these two sentences, i.e. the processor does not try to fill a gap after raced, neither for the sentence in which it is plausible that the wh-phrase could be the object of race (12.68), nor for the one in which it is implausible (12.69).
In some cases, the successful interpretation of a sentence requires the processor to deal with more than one filler and more than one gap. A strategy that has been proposed for dealing with this situation is the most recent filler hypothesis (Frazier, Clifton & Randall, 1983). In its simplest form, this says that where there are two gaps, the processor should assign the most recent of two possible fillers to the first gap, and the more distant to the second. This is illustrated in (12.70) and (12.71). Note that in both cases the more distant first fillers are deliberately chosen to be ones that could – on semantic grounds – be linked to the wrong’ early gap (e.g. who could be the filler for the gap marked in (12.70), as in who did you want to make a potholder?), but the preferred interpretations are the ones indicated by the subscripts.
However, Boland et al. (1990) show that this hypothesis does not apply in cases such as (12.72), which can be contrasted with the superficially similar sentence in (12.73). The difference arises because promise and persuade expect different underlying sentence structures – the subject of promise is also the subject of the subordinate clause (i.e. John promised Mary that John would read the book), while it is the object of persuade that is the subject of the subordinate clause (John persuaded Mary that Mary would read the book).
A final consideration in this section concerns the efficiency with which comprehenders fill gaps. This is done very quickly, and Tanenhaus e al 1985 found that the anomaly is detected just as early in (12.75) as it is in (12.74). In (12.74) the snake is object of force, but it is also coindexed with the subject position of to hop. In an anomaly detection task, listeners are quick to detect that there is something wrong with the meaning of this sentence as soon as they have recognised the word hop. The anomaly is that snakes do not hop. Now consider (12.75). With this sentence the process is more complicated. It is argued that the subject position of hop has to be linked to the object position of force, which in turn has to be linked to the filler phrase which snake. However, the anomaly with hop is detected just as quickly here as it is in (12.74).
الاكثر قراءة في Linguistics fields
اخر الاخبار
اخبار العتبة العباسية المقدسة
