Repair
المؤلف:
Paul Warren
المصدر:
Introducing Psycholinguistics
الجزء والصفحة:
P75
2025-11-03
42
Repair
Repairs typically involve the interruption of an erroneous utterance. With repairs in spontaneous speech, it can often be difficult to know what the speaker would have gone on to say if they had not interrupted their utterance. For this reason, a lot of repair data come from carefully controlled speech tasks, where the speaker’s original intention is more obvious. One particularly productive source is a series of studies that has used descriptions of networks of coloured nodes, such as that shown in Figure 5.1 (Levelt, 1983).
The speaker has to describe the network so that a listener who cannot see the same network can draw it accurately based on the description. Errors arise when the speaker selects the wrong direction term (right instead of left, for example) or colour label (green for red, or light blue for dark blue). Since the experimenters know what the target node configuration is like, they have an informed idea of what the error is, even when the error is detected and corrected very rapidly, as many are.
Other similar tasks include asking speakers to describe routes around a map which might be a fictional map invented for the task, or to give route directions from one landmark to another in a locality known to the experimenter and to the participants.
One research project collected nearly a thousand self-repairs using the node network technique, providing enough data for some general pat terns to emerge (Levelt, 1983). First, three main phases were identified – interruption, editing and repair. These are illustrated in Figure 5.2 for the repair of an error that relates to the description of the network in Figure 5.1. The moment of interruption is when the speaker breaks off from their original utterance, in this case part way through a word. Taking into account the sounds produced before the interruption together with the likely words in the context of the configuration in Figure 5.1, it is probable that this word was going to be yellow. One feature of the editing phase is that the speaker may use an editing expression – in this case it is the vocalization uh. The repair is when the speaker actually makes good the damage of the error from the point of restart onwards.
As well as repairs, which involve the correction of an error in what has been said, speakers also make revisions. A revision is where what has been said is not exactly incorrect, but is in some way incomplete or not fully appropriate to the occasion hence these are often referred to as revisions of inappropriateness. If we assume that the speaker has been describing the right-hand loop of the node pattern in Figure 5.1, and has got as far as the dark blue node, then the example in 5.7 would be a revision rather than a repair. Although the interrupted utterance fails to distinguish between the two blue nodes in the figure, this is not crucial for the listener since they have added a dark blue node to their version of the figure on the basis of the earlier part of the utterance in 5.7 and are now ready to move on from that node.
Close study of repair data has shown that speakers are very efficient in carrying out repairs. It would seem that we monitor our own speech very closely and interrupt ourselves as soon as we can if we detect an error. This has been encapsulated in the main interruption rule Nooteboom, 1980, which simply states that speakers interrupt themselves immediately that they detect an error. In fact, self-interruption can be so swift that it occurs before the speaker actually utters the incorrect part of their utterance. These are known as covert repairs, and in Levelt’s (1983) study they account for 25 of interruptions. 5.8 illustrates this. We cannot tell from this example what the erroneous utterance would have been, because the speaker does not actually produce an error contrast the interruptions in Figure 5.2 and example 5.7. The evidence that suggests that a covert repair has taken place is the presence of a hesitation and the repetition of the word a.
This is not the only interpretation of the data, as the hesitation could for instance indicate an uncertainty over word selection see Chapter 3. Because of difficulties interpreting examples like 5.8, more detailed analysis of repair data tends to focus on overt repairs and revisions, i.e. where the error and repair/revision are available for scrutiny. The immediacy with which errors are detected by the speaker is shown by Levelt’s finding that just over half 51 of interruptions in overt repairs directly followed the error word Levelt, (1983). A further 18 involved interruptions within the error word itself as in the example in Figure 5.2. Closer scrutiny of the less immediate interruptions suggests that the additional delay probably arises because the error was not detected right away, rather than because the speaker wanted to complete a phrase or sentence, since even these later interruptions tend to disrupt grammatical constituents. Additional comparison of different repair types shows that interruption during a word is much more likely if the word is one that needs repair what was said was wrong than if it is one that needs revision what was said was incomplete.
The repair sequence frequently includes an editing expression, such as the vocalisation in the example in Figure 5.2. In the majority of cases, and across a wide range of languages, editing expressions are indistinguishable from the sounds found in filled pauses. Levelt noted that in his database 62 of repairs were accompanied by an editing expression, but only 28 of revisions Levelt, (1983). It has therefore been suggested that they are a signal to the listener that some dramatic change in the utterance is about to be carried out.
A number of researchers have looked more closely at different types of editing expression, and have suggested that there may be different functions associated with the different types e.g. James, (1973). A summary is given in Table 5.1.
Following the editing expression, if there is one, the speaker initiates a restart. Because errors need to be undone and repaired, when a speaker makes a repair they tend to backtrack and restart from a point before the error word. It has been argued that the first word of the repair indicates how far back the listener needs to go in backtracking. If there is an identical word before the self-interruption, then they will go back to the last instance of that word, as in the example in 5.9, where the relevant words have been underlined. If the first word of the repair differs from anything in the original, then the repair is taken to be a continuation from the last example of a word from the same syntactic category, as in 5.10, where again the relevant words are underlined.
It is claimed that the structure of a repair, and how it relates to the error, is subject to a well-formedness rule Levelt, (1989) which in essence says that the error, or the complete sentence constituent of which it would have been part, must be able to form a grammatically complete coordinated structure e.g. with a when joined with the repair. So in the case of 5.9 to the blue dot and to the red dot would be a complete coordinated structure. What counts here is the grammatical completeness rather than whether the coordinated structure makes sense. It is argued that an example like 5.11 is not likely to occur as a repair structure because should I go right and I go left is not well-formed.
In contrast to the repair of an error, revisions, involving something that is incomplete or inappropriate, often need further specification. Consequently they may involve a complete fresh start, sometimes with a completely new sentence construction. This is supported again by statistical data from Levelt’s large-scale survey of repair data 8 of repairs but 44 of revisions involved fresh starts Levelt, 1983.
Speakers often show important aspects of a repair through prosodic marking (Levelt Cutler, 1983). This usually involves the spoken emphasis of the repair word. While repairs of sound errors tend not to be prosodically marked, many 38 repairs of word errors are Cutler, 1983. One interpretation of prosodic marking is that it helps the words in the repair overwrite the error in the listener’s developing interpretation of the utterance. This is supported by the finding that prosodic marking is more likely for repairs than for revisions (Levelt & Cutler, 1983), presumably because in the former case the new text has to replace the error, rather than adding to it. Prosodic marking is also more likely when there is a high degree of contrast between error and repair terms – and presumably a higher cost involved in letting the error go uncorrected – e.g. between up and down in example 5.10.
Another interest finding that involves prosody is that an error in a word’s stress pattern is more likely to be repaired if the misplacement of stress also results in a difference in the vowels in the word – compare the vowels in the first syllable of ‘extract and ex’tract. The repair data show that 63 of lexical stress errors are corrected if they cause a change in vowel quality, but only 23 are corrected if they do not (Cutler & Clifton, 1984). This is almost certainly because misidentification of the speech sounds in a word leads to a greater risk that the word will be misidentified. On a related note, in unpublished data I have found that stress pairs where there is a difference in the vowel qualities are more likely to be recognised as different in a discrimination task than stress pairs like ‘insult and in’sult where the difference is marked- for most speakers- only by stress.
الاكثر قراءة في Linguistics fields
اخر الاخبار
اخبار العتبة العباسية المقدسة
