Pausing and predictability
المؤلف:
Paul Warren
المصدر:
Introducing Psycholinguistics
الجزء والصفحة:
P40
2025-11-01
55
Pausing and predictability
A central claim in the pausing research outlined in Chapter 2 is that pause patterns vary across speech tasks because these tasks require different amounts and types of planning. A specific instance of this general claim concerns the process of retrieving words from the mental lexicon. Speakers start off with a general abstract idea of what they want to say and they need to find linguistic expression for this, including finding the appropriate words. But not all words are equal – there are some words that are used more frequently than others; and there are some words that fit a particular topic or context better than others. As a result of such frequency and predictability effects, we would expect to find that some words are easier to retrieve from the mental lexicon than others. If lexical retrieval is hard, it may take longer, and may be more likely to result in a pause before the difficult word. Conversely, it has long been known that ease of access to words is associated with greater fluency in speech Beattie & Butterworth, 1979; Lounsbury, 1954.
Studies of large corpora of pause data have shown that – in general – pauses are more likely and longer before content words than before function words. Recall that the closed’ set of function words is much more limited than the open’ set of content words. The finding of more pausing before content words might therefore reflect some additional time needed to select the appropriate word from a larger set of possibilities. There are alternative explanations for this overall effect. One is that function words sit in a separate part of the mental lexicon with faster access. Another is that function words become available at a different stage of the production process, when the grammatical sentence frame is constructed.
To get an experimental measure of the likelihood of different content words in an utterance, researchers have used various sentence completion techniques. In what is known as Shannon’s ‘Guessing game’ (Shannon, 1951), an incomplete transcript of an utterance is presented to respondents. Initially, they see the beginning portion of the utterance, and have to guess the next word. They carry on making suggestions for each word until they get it right, up to a specified time limit e.g. one minute. They then go on to guess the next word, and so on. The lower the ratio of correct guesses to incorrect guesses for any particular word, the lower the likelihood of that word in the given context. A variant on this procedure starts at the end of the utterance and works backwards. Another, also known as the Cloze task (Taylor, 1953), deletes a word or words from a transcript and asks participants to fill in the blanks. One measure taken from such tasks is known as transitional probability. For example, consider the sentence fragments in 3.4. Native speakers familiar with English idioms will find o highly predictable in the first sentence. There will be a relatively limited set of words that would be likely in response to the second fragment, rather more for the third, and still more for the fourth fragment. Therefore, the transitional probability for o in sentence a. will be higher than that for, say, short bread in sentence c.
Sentences have been taken from transcripts of actual spontaneous speech and used in the types of completion task outlined above. The patterns of transitional probabilities for the content words in these sentences have then been compared with the likelihood and duration of pauses before each of these words. A long-attested effect is that pauses are both more likely and longer when the word is unpredictable (Lounsbury, 1954). But early research also noted a strong relationship of pausing with a likely hood measure derived from the right-to-left, backward variant of the guessing game, and an even stronger relationship with likelihood derived from a combination of forwards and backwards guessing (Goldman-Eisler, 1961). These findings show that predictability and lexical selection is based not just on what is the most likely next word in a linear string of words. Rather, it depends also on a more hierarchical structure, with aspects of what we want to go on to say influencing our current word choices.
Note though that words which are unlikely in a given context such as opera ticket after his in-sentence b. in 3.4 are also words which have high information value, i.e. they contribute a lot to the meaning. If pauses are also used to highlight less predictable words and make them easier for listeners to make out, then pausing at points of low transitional probability may be a strategy by speakers to help their listeners. For a general discussion of predictability and speech prosody, see Turk, 2010.
الاكثر قراءة في Linguistics fields
اخر الاخبار
اخبار العتبة العباسية المقدسة
