Spoken word recognition Flashcards
Are there boundaries between phonemes?
No! E.g. /p/ /b/ – there is no definite boundary, the two phonemes slowly ‘merge’ into one. We are sensitive to the contrast though, hence why we can usually tell what one the person is saying. We are initially sensitive to all contrasts, but eventually specialise to just being sensitive to those in our native language.
What is the cohort model?
As soon as a word starts being pronounced, you start to recognise it. In the cohort model, you have the first cohort, which is comprised of all the words possible which have been activated by the first phoneme. or so. When the next bit of the word is added (e.g. ‘st’ –> ‘star’), the second cohort becomes smaller, because words drop out of it if their activation lessens. This continues til the end of the word, where there should only be one word left in the cohort (the most activated one) which is selected.
What is the segmentation problem?
There are no gaps in natural speech like there are in written sentences.
How do we overcome the segmentation problem?
We use acoustic cues (e.g. vowels tend to lengthen at the end of words). Also, segmentation follows recognition: first we recognise, then we segment into words. Many words (60%, mainly high frequency, short words) are only unique at offset.
How do we segment?
Metric Segmentation Strategies: in English 1st syllable is usually stressed. Can (sometimes!) assume that if a syllable is stressed, it’s the onset of a new word.
Also, we might use the possible word constraint principle: segment natural speech so that ALL phonemes are accounted for by words.
