Contact and activation
المؤلف:
Paul Warren
المصدر:
Introducing Psycholinguistics
الجزء والصفحة:
P126
2025-11-05
26
Contact and activation
After the pre-lexical analysis of the speech input, the next stage of the word recognition process is the contact stage, involving a mapping from the output of pre-lexical analysis onto forms stored in the mental lexicon. Such mapping from the input to the lexicon is an example of bottom-up processing. Many models of word recognition argue that this initial con tact is based solely on automatic bottom-up processing, while others also claim a role here for top-down processing, e.g. the use of context to pre select words from a particular area of meaning.
Simultaneous contact with more than one stored word is a characteris tic of parallel models of lexical processing. The alternative is a class of models that can be characterised as serial search models, where one word is considered at a time, such as Forster’s Search Model 9Forster, 1976), discussed in Chapter 9. Support for the notion of parallel processing of all lexical entries that match the input comes from studies using a cross modal priming technique. One study (Zwitserlood, 1989) presented visual targets after either a test prime such as captain or a control prime such as justice. The test primes all had partner words which started with the same sequence of phonemes (e.g. in this case capital), but which were never heard in the experiment. The test and control prime were each combined with three types of target. One (ship) was related to the test prime (captain), the second (money) was related to the partner word (capital), and the third was related to neither the test prime nor its partner. The study was actually done in Dutch, but the examples are similar.
The experimental measure of interest is whether the targets show facilitation, i.e. faster and more accurate responses, after test primes compared with after control primes, and whether this varies depending on how much of a prime has been heard. In an Early condition, the targets were presented part way through the prime, i.e. before the whole prime word had been heard. For a test prime, this was a point that included all the material which the word shares with its partner word – for captain / capital this would be at the end of the /p/. In the Late condition the targets were shown at the end of the prime. Facilitation was found in the Early condition for targets related either to the test prime or to the partner word. So in our example both money and ship would be responded to more rapidly and with fewer errors after the initial portion of captain than after the initial portion of the control word justice. The control target (lamp) showed no such facilitation. In the Late condition, when all of the prime word has been heard, only ship, related to captain, shows facilitation. The earlier facilitation shown for money has now disappeared.
The Cohort Model
The above result shows that during the early stages of the processing of the spoken word captain, both this word and the phonetically similar word capital have been contacted, and words related to both of these are primed. In other words, the initial contact is with multiple words matching the input, i.e. there is parallel lexical processing. Note that further experiments with rhyming word pairs shows that it is the overlap of the initial portion of the word that is important, and not just any portion of the word. That is, while bee is primed by honey, it is not primed by money, which has most of the same sounds as honey, but does not start with the same sound. It makes good sense for the beginning sounds of words to carry primary responsibility for making contact with words in the mental lexicon, since we hear these parts of the word first. The set of words contacted in this way has been referred to as a word-initial cohort.
The word-initial cohort is a key feature of the parallel and primarily bottom-up approach to spoken word recognition known as the Cohort model (Marslen-Wilson, 1987; Marslen-Wilson & Welsh, 1978). In this model, once the initial sounds of a word have been heard, all words in the listener’s mental lexicon that have the same initial sequence of sounds will be contacted. Once initial contact has been made with the cohort members, the lexical entries for these words become activated, making available further information about these words that is stored in the lexicon as discussed in more detail below.
Zwitserlood’s experiment described above also shows that as the input continues, further bottom-up information concerning the next sounds becomes available, so that some of the initially contacted words no longer match this input, and fall out of contention. This is sometimes referred to using a directional metaphor from reading at least as applies to the reading of languages such as English, i.e. that there is left-to-right processing of spoken language. By the end of the spoken word, it is clear that the word is captain and not capital, and so only ship is primed.
In early versions of the Cohort model, it was assumed that a word was either available or not available, depending on whether the sounds matched the input and on whether the lexical information about that word matched the developing utterance context, as we will see later. However, it soon became widely acknowledged that lexical entries in the mental lexicon are not just on’ or off’, but that they assume levels of activation (Marslen-Wilson, 1987). The more input there is that supports a particular lexical item, the higher the level of activation for that entry. So the activation levels of both captain and capital increase as the first few sounds /k æ p/ are heard, but once the sound pattern of the input diverges from that expected for the word capital, then the activation of this word falls, while that for captain continues to rise. This is illustrated in Figure 8.1, which shows what might happen to the activation levels of captain, captive and capital, as examples of the word-initial cohort of captain, as more of the word captain is heard.
One of the advantages of an activation-based model rather than an all or-nothing model of recognition is that it can more easily allow for partial matches and for recovery from mispronunciation or perhaps from unexpected pronunciations. So, for example, if a speaker says shtudent’ for student’ (an increasingly common pronunciation in New Zealand English), the initial sh’ sound partially activates the similar /s/ and the subsequent input /tju .../ continues to increase the activation of words such as student’, which we might suppose is represented in the lexicon as / stjudənt /.
الاكثر قراءة في Linguistics fields
اخر الاخبار
اخبار العتبة العباسية المقدسة
