INFORMATIVENESS
المؤلف:
John Field
المصدر:
Psycholinguistics
الجزء والصفحة:
P131
2025-08-31
461
INFORMATIVENESS
Book: Psycholinguistics
Author: John Field
Page: 131
The notion that, within a system of language, certain features may have greater information value than others. A feature’s information value lies in the extent to which it serves to distinguish between units of language (e.g. to distinguish one word from another). On this analysis, the vowel [ə] is not very informative in connected English speech because it is very frequent, whereas the vowel [ʊ] is informative because it is comparatively rare.
Noise, accent or speaker error often prevent a listener from perceiving all the phonological cues that are present in the signal. This raises the question of how much information is necessary for a word to be recognised. Surprisingly, even if the only cue to a word’s identity is its stress pattern, the listener can still produce an equivalence class (a group of words matching the available information) which is much smaller than the English lexicon as a whole– probably between 15 per cent and 19 per cent of the total. Several studies have used lexical statistics to examine how the candidate search space (the size of a group of possible word matches) reduces as more and more phonological information is available. Knowing the number of phonemes in an item reduces the search space to 5 per cent of the English lexicon. Adding in information as to whether the phonemes in question are vowels or consonants reduces it to about 1 per cent.
With a broad-class phonetic transcription which represents only manner of articulation, the largest group of possible word matches was calculated at about 1 per cent of an English lexicon of 20,000 words and the average group size at around 2. It was also claimed that about 32 per cent of items could be identified uniquely. However, this result was challenged on the grounds that average class size is skewed by a few very small or very large groups and that the frequency of the items needs to be considered. A formula was therefore proposed which enables the calculation of an expected class size (ECS) weighted for frequency. With the 20,000-word lexicon, ECS was 34 and uniquely identifiable words represented only 6 per cent.
An alternative way of approaching this kind of analysis (and one more in line with information theory) is in terms of how much information is needed in order to narrow the class size down to one i.e. how much information is needed to identify every word in the lexicon. Measuring the percentage of information extracted (PIE) gives very different results from measuring ECS. To give an example, the ECS for a broad transcription of the items in an English dictionary was calculated at 31.1 (0.25 per cent)– suggesting that 99.75 per cent of the original lexicon could be eliminated as not matching the information given. However, a PIE calculation indicated that the information available in the broad-class transcription represented only 79.8 per cent of what would be needed if every single word were to be uniquely identified.
See also: Probabilistic, Underspecification
Further reading: Altmann (1990)
الاكثر قراءة في Linguistics fields
اخر الاخبار
اخبار العتبة العباسية المقدسة
