Speech errors and lexical selection
Our second source of data relating to lexical selection is provided by speech errors. Several large corpora of speech errors now exist, some of which are available online. What makes speech errors particularly interesting and useful for psycholinguistics is that they are not random. Instead they display patterns. By looking at what errors occur and possibly those that do not, we can figure out the mechanisms that would have allowed the errors. We could think, for instance, about what linguistic units must be available for the following errors to occur.
In the case of 3.5, the speaker has reached a stage in the sentence production process at which the words sea and s have been selected from the mental lexicon, together with relevant information about their grammatical category. These words are in a queue, waiting to be inserted into an utterance frame. Under this kind of slots-and-fillers or scan-copier approach Shattuck-Hufnagel, 1979, the choice of the wrong fillers words for the waiting slots positions in the sentence frame can result in an exchange of words, as in 3.5. At another level, we can see the same kind of process applying in 3.6, the difference being that in this case the slots and the fillers are individual speech sounds, rather than complete words. We will return to sound errors like this in Chapter 4.
Before we go any further, it is important to note that the speech errors we are interested in here are the slips that normal competent language users produce as part of everyday speech. That is, we are not looking at clinically impaired language output, although that is also revealing about the so-called normal’ processes of language production. Nor are we studying the errors that first or second language learners make during the learning process, although again such errors are interesting, but on the whole for different purposes.
Similarly, it is important to distinguish between the causes and mechanisms of speech errors. The causes of speech errors are manifold. Tiredness, distraction, drunkenness, the fear of public speaking – all of these can result in an increase in the count of errors in our speech. What is interesting for psycholinguists when they investigate the process of language production is not the probable cause of such errors, but what the nature of the errors might tell us about the production system. Freudian slips’ are of course of interest to both psycholinguists and psychoanalysts and undoubtedly to comedians. Primarily, though, the psycholinguist is interested in the fact that there are slots for words in sentences and slots for sounds or letters in words, and that the wrong words, sounds or letters can end up in the available slots. This might make the psycholinguistic analysis of speech errors seem rather bland, but our interest is in the mechanics of errors, which appear to be universal reflecting aspects of the language production system, while the causes vary with the speaker and with the situation.
Some comments are also required on the reliability of error data. Most speech error corpora have been collected over many years by researchers observing errors in lectures, conferences, cafe conversations etc. But we need to be certain that these are errors in production rather than in perception Chapter 7 includes discussion of slips of the ear’ and of what they can tell us about speech perception processes. Recordings can help here, but are not always available. We also need to be confident in the accuracy of the descriptions not only of the error but also of the intended utterance. Often, the intention is clear from the context or from some sort of self-correction on the part of the speaker once they realise they have slipped. Sometimes, the error is made by the researchers themselves, and they will have a good idea of what their intention had been. But often the intended utterance will be unclear, in which case we need to be careful to avoid any over-interpretation. Chapter 4 illustrates how researchers have also worked with experimentally induced errors, which allows greater control and reliability in reporting, but may result in atypical errors.
The examples in Table 3.1 illustrate a range of spontaneously occurring word errors, organised into a number of different types. The different types reflect different suggested stages of the production process. Examples 3.7 and 3.8 involve errors during the selection of words from the mental lexicon, while examples 3.9–3.11 involve a later stage in the process, when the words that have been selected are positioned in the utterance.
Substitutions and blends
On the whole, substitutions and blends, i.e. errors of mis-selection, involve words that are semantically related, that is words which have a clear meaning relationship to one another. This is not surprising – we select words to express a concept that we want to talk about see Figure 3.1, and concepts are likely to be linked to a number of related words. What is interesting is the difference in the nature of the relationship in these two types. Substitutions, like 3.2 and 3.7 above and 3.14 below, are quite likely to involve antonyms, words that are some kind of opposite of one another.
On the other hand, blends more frequently involve words that are synonyms or near-synonyms, i.e. words that have very similar meanings. Example 3.8 above illustrates this, as do the examples in 3.15.
It has been suggested that the differing nature of the meaning relationships involved in substitutions and blends reflects different stages of the production process at which the errors occur. To understand this, we must distinguish between relationships at the conceptual level and relationships at the lemma level.
Concept-level relationships involve pre-linguistic abstract ideas. Let us assume that during speech production there is sometimes ambivalence as to which of two closely related ideas best represents the speaker’s intention. This has been referred to as a situation of alternative plans’. In such a situation, the related concepts activate their lemmas at the same time (e.g. (baggage and luggage) in 3.3. If the speaker is unable to resolve the competition between the alternative plans, then these activated lemmas may both be inserted into the same slot, and the lexemes linked to these lemmas become blended at the level of phonological processing to produce (buggage).
By contrast, lemma-level relationships tend to be associative. That is, they arise through the associations that words have with one another. Researchers have collected data on native speakers’ word associations, as we will see below. Some associations involve collocations i.e. between words that tend to co-occur in the same phrase or sentence, such as bread and butter, but many involve some kind of antonymy or oppositeness.
Unlike the near-synonyms involved in blends, words in an associative relationship differ crucially in some core aspect of their meaning. Interestingly, associated words are hardly ever involved in blends, but they do feature quite regularly in substitutions. In substitutions, then, the intended concept activates its lemma, and activation flows through the associative links between lemmas, so that an associate of the initially accessed lemma is also activated, and the wrong lexeme is inserted into the utterance see Figure 3.3.
We now need to explain why the associate becomes available before the target lemma and replaces it in the utterance. One possible factor is the relative frequency of the lemmas involved. If an associated lemma is more frequent than the target lemma, then it might become available before the target. Frequency effects are abundant in psycholinguistics, and are discussed in more detail from the perspective of word recognition in Chapters 8 and 9. The results in this respect seem to be rather mixed, with frequency being a factor in some studies, but not reliably so. Indeed, one study found a stronger effect of imageability, a semantic effect that has independently been shown to influence ease of access to words in production. Substituting words were generally more imageable than substituted words Harley & Macandre, 2001.
It is also possible that the replacement of a target by an associate comes about because the target lexeme is for some reason unavailable, even if only momentarily. The result is similar to the longer-lasting effects found in the speech of some patients with brain injury e.g. after strokes, who will often produce words associated with an intended word, often without realising that they have produced the wrong word.
The discussion above has suggested that blends and substitutions differ in the type of semantic relationship that exists between the words involved synonyms are blended, while substitutions involve antonyms or other types of associative relationship. This oversimplifies the situation. For instance, there are errors where substitutions involve near-synonyms, including the rather more complicated example in 3.16, where it would appear that a footwear’ concept has led to lemmas for both shoes and boots being available, and shoes has substituted for o se s early in the sentence.
In addition, we have no easy way of determining that substitutions do not also involve synonyms, simply because substitution of one word by another word with a similar meaning is unlikely to be noticed. However, the fact that blends rarely involve anything other than synonyms, while substitutions clearly involve a range of associative relationships, suggests that these two classes of error involve different levels or types of processing.
Malapropisms
A further common observation about blends is that they often involve words that have similar sounds, as was illustrated by the examples in 3.8 and 3.15. These are not the only word-selection errors that involve formal sound-based relationships between words. Consider for instance the errors known as malapropisms Fay & Cutler, 1977 . These are errors where the word produced is similar to the intended word in its sound shape, but not necessarily in its meaning, as illustrated in 3.17.
Malapropisms are named after the character Mrs Malaprop in Sheridan’s play The Rivals 1775. This character made somewhat ridiculous substitutions of words that were similar in form, such as the example from the play in 3.18.
The examples from Sheridan’s play are obviously constructed, as are countless others that have been placed by scriptwriters into the mouths of actors over the centuries. However, naturally occurring malapropisms also exist, although they are relatively rare compared with other error types. Some errors infamously produced by former United States President George W. Bush have been described as malapropisms as well as being called Bushisms. Two of these are listed in 3.19 and 3.20. Many of Bush’s examples may of course have their origins in misreading, rather than being spontaneous speech errors.
If lexical selection is meaning-based i.e. based on the conceptual representation of the message, how do form-based word-selection errors occur One explanation suggests that there are links from the sounds in the target word to other words that contain the same sounds. As the form of the target is retrieved, this activates its component sounds. If activation flows back from these sounds to the other words containing the same sounds, as shown by the dashed arrows in Figure 3.4, then a similar sounding word may become activated. If the activation is strong enough, then this competing word may replace the target in the actual utterance produced. The activation of the substituting word may be higher because it is a more frequent word, as has been demonstrated in an analysis of malapropisms Vitevitch, 1997. It may even be increased by activation from a conceptual aspect of the context. In this particular case it could be argued that the concept of the word hostile is an active one in the context of the sentence in 3.20, and that this has also activated the hostile lemma and the /hostile/ lexeme. The same process operates in the case of many blends, where the words involved frequently also overlap both in their sounds and in their meaning.
Serial and interactive models of lexicalization
Notice that Figure 3.4 differs from the simpler model of lexicalisation in Figure 3.1, in that it allows information between the levels to flow in more than one direction. These two figures are rather simplified forms of two theories of speech production. In one type, reflected in Figure 3.1, it is argued that the speaker has access to one word at a time following a rather discrete and unidirectional flow of information between levels.
Models of this type are variously referred to as symbolic or serial search models. In another type, more like Figure 3.4 , information spreads by way of activation from units at one level down to multiple units at the next level, but then also back up to the higher-level units. These are connectionist models, including interactive activation IA models. Chapters 8 and 9 also consider such model types, from the point-of-view of lexical access during word recognition, and explore these models in more detail than is given here.
As well as the positive or facilitative activation illustrated in Figure 3.4, IA models typically also include the notion of inhibition through competitive links, usually between units at the same level. Such links help to explain the speed with which successful word candidates become available – the more active a candidate word is, the more strongly it inhibits its competitors. For example, a high-frequency word will quickly dominate other candidate words both because it starts off being more active and also because, all else being equal, it will inhibit its competitors and become even more dominant.
In addition to evidence from speech errors, there is also experimental evidence which shows that both meaning- and form-based relation ships are important in lexical selection, and which also supports the notion of inhibitory links between words, at least at some levels. For example, Levelt et al 1999 review results from picture/word interference tasks. These are tasks where participants have to name objects in a series of pictures, but at around the time they are about to start producing the name, they either hear or see superimposed on the picture a word related to the object name. If this distractor word is related in meaning to the object name e.g. GOAT if the picture is of a sheep, then it slows the naming response. If the distractor is phono logically related e.g. SHEET for sheep, then it speeds up the naming response.
الاكثر قراءة في Linguistics fields
اخر الاخبار
اخبار العتبة العباسية المقدسة
