Recognition and context effects
المؤلف:
Paul Warren
المصدر:
Introducing Psycholinguistics
الجزء والصفحة:
P130
2025-11-06
31
Recognition and context effects
In many models of word recognition, including the Cohort model, lexical access is only possible after a word has been activated on the basis of the input speech signal, i.e. on the basis of bottom-up information. This seems to run counter to the intuitive notion that there is pre-selection of words based on information from the contexts in which they are uttered. That is, we often find ourselves able to complete or continue someone else’s sentence for them. This suggests that we can predict a word on the basis of top-down or contextual information.
However, there is little evidence that contextual information leads to the pre-selection of candidate words during the normal rapid word recognition process. For instance, when sentence contexts such as 8.1 are used in cross-modal priming experiments and a related visual target word s is presented immediately before participants hear the actual prime word (captain), then there is no facilitation of ship, relative to a control condition. But if the target word is presented later in the prime word e.g. after /kæ p/, then there is facilitation. For discussion of the efficacy of using the cross-modal priming technique with sentence primes, see (Nicol, Swinney, Love & Hald, 2006). It seems, then, that bottom-up information the phonetic input has priority over contextual information in the initial stages of making contact with stored lexical forms.
It is argued that the experience of being able to finish someone’s sentence for them reflects processes that are slower than the normal processes of word recognition. That is, just as words can be primed by other words related in meaning, so the interpretation of a sentence context can activate a field of meaning and can cause a set of words to become available, just as they are during language production. However, this process is not as fast as bottom-up activation.
A further argument against pre-selection is the sheer quantity of words that would be activated in many contexts see examples discussed under Predictability in Chapter 3.
Although context might not pre-select words, it is involved in efficient word recognition, as follows. Words are activated on the basis of bottom-up information; lexical access makes available lexical i.e. grammatical and semantic information associated with these words; this lexical information is assessed against the developing interpretation of the sentence, resulting in changes in the activation levels of the activated words. The rapid integration of contextual and lexical information means that words can be recognised earlier in sentence contexts than in isolation. That is, the recognition point of a word can be earlier than its uniqueness point, if contextual information rules out other candidate words before this point. An oft-cited example is the word trespass. In isolation this word’s recognition point is the same as its uniqueness point, at the /p/. It is at this point that other members of the word-initial cohort, such as ess and trestle, can be eliminated. If this word is heard in a context such as that in 8.2, then its recognition point is earlier, at the /s/, since ess and es le do not fit the context and so the activation levels of their lexical representations will have dropped away.
Not all contexts are the same, and some may have a greater constraining influence of this type than others. This is revealed in word monitoring experiments using the same target words in different spoken sentence contexts (Marslen-Wilson, Brown & Tyler, 1988). In one example, participants were told to listen out for the target word guitar and to press the response button as soon as they heard this word. Depending on which participant group they were assigned to, they would have heard the word in one of the four contexts in 8.3--8.6
Participants’ response times RTs were measured from the onset of the word guitar. The results showed that across sentences 8.3 to 8.6 response times got increasingly longer. This reflects the changing contextual constraints across these sentences, as determined by the changes in the verb. In 8.3, a guitar is something that very plausibly can be a e mean RT was 241 msec, measured from the onset of the word being monitored for, and averaged across a set of examples with the properties of 8.3. In 8.4, while a guitar can be buried, this is a somewhat unusual thing to do mean RT was 268 msec. In 8.5, although the verb can have an object noun phrase the thing that is consumed by drinking, this noun phrase usually denotes a liquid, and guitar are not liquid mean RT was 291 msec. Finally, in 8.6, sleep is a verb that does not take an object mean RT was 320 msec. An additional finding was that word monitoring times tended to decrease the later a word is in a sentence, as long as that sentence is grammatically well-formed. This shows that as we hear more words in a sentence, we are able to develop a more complete interpretation of the context sentence, which becomes more constraining on the possible words that can occur.
A neuroimaging study (Kuperberg e al , 2000) also used sentences such as those in 8.3--8.6. Participants had to listen to the entire sentence and say whether it made sense. The study found that all of the violation conditions 8.4--8.6 showed increased activation in a brain area which has also been associated in brain-damaged patients with comprehension difficulties. In addition, separate areas of increased activity were found for the different types of violation. It is conjectured that the area showing activation increases for all three violation types is the area that is responsible for integration of information, which is of course important in the context effects on recognition that we have discussed in this section.
الاكثر قراءة في Linguistics fields
اخر الاخبار
اخبار العتبة العباسية المقدسة
