Abstraction versus exemplars 1
Normalization Word is stripped of irrelevant info and stored as an abstract, normalized entity 2
Normalization Experiment 1 People heard words spoken by 1 or 10 people When the words were presented with noise there were fewer correct identifications with 10 speakers voices 3
Normalization Experiment 2 People repeated the word they heard They were slower to repeat with 10 voices than with one 4
Normalization If different voices were irrelevant, they all were normalized you would not expect differences between hearing one or 10 voices This sounds like evidence against normalization Authors concluded that a normalization process caused this, and words are stored as abstract normalized entities 5
Normalization If they found no effect of # of voices they’d also conclude that normalization occurs 6
Normalization If they found no effect of # of voices they’d also conclude that normalization occurs There is no evidence that would be a problem for their theory. Not falsifiable 7
Exemplar storage Evidence for storage of the irrelevant People remember where word was on page People remember the person, voice, context a word was said in People remember the font or case a word was presented in 8
Exemplar storage People heard 150 words Later (5 minutes, day week) they heard 300 words and decided if the words were new or old. Some words presented in same voice others in different voice There were 2, 6, 10 different voices 9
Exemplar storage 10
Exemplar model Mental representations of phonology are built up through experience, addition of exemplars 11
Abstract model Words stored in phonemic form, then pass through rules which give it its phonetic detail 12
Abstract model In formal theories the mental lexicon (storage) is separate from processing In formal theories words stored with small inventory of phonemes Phonemes are identical in all words Fact that rules can apply to new words is “evidence” for existence of rule 13
Abstract model In formal theories sound changes apply to all words with same phonemes So, all words must be comprised of same units 14
Exemplar dynamics In exemplar theory words stored in surface form The mental lexicon (UR) and processing (rules) are not separate but interrelated A word or phone or morpheme is a category 15
Exemplar dynamics Members of the category are stored in a cloud of related exemplars More similar exemplar are stored “closer” to each other (or with stronger connections) 16
Exemplar dynamics Exemplar are related/belong to same cloud/category because they are perceived to be the same thing A form can belong to different categories e.g. shoot stored under different meanings stored with relationships to frustration stored with voice of person who says it a lot 17
Exemplar dynamics Can you store everything you've ever said/heard? Possibilities Older memories decay Many exemplars are so similar they fuse into the same thing 18
Exemplar dynamics Exemplar: a memory trace of an experience 19
Prototype Theory Prototype: Most central member of category The one new exemplars are compared with to determine what category it belongs to
Exemplar Theory No real single prototype, Instead Densely populated part of cloud Many exemplars High frequency exemplar Lots of connections between exemplar
Exemplar Theory No real single prototype, Instead Densely populated part of cloud Many exemplars High frequency exemplar Lots of connections between exemplar
Exemplar Theory There is competition between clouds of exemplars “the probability for a label is influenced both by the activation of exemplars having that label, and by competition from other labels having exemplars in the same area of the cognitive map.”
Exemplar Theory “In perception, the encoded phonetic character of an incoming stimulus locates it in the parameter space. Activation of exemplars in the neighborhood is passed upwards to the labels, with the most probable label winning in competition with alternatives. Production proceeds in the opposite direction.”
Exemplar Theory “The likelihood that a particular exemplar will be selected is proportionate to its strength. Production is taken to be sensitive to strength in exactly the same way that perceptual classification is.”
Frequency effects Schwa deletion Occurs more in high frequency words Every, family 26
Frequency effects Schwa deletion Occurs more in high frequency words Every, family In mid-frequency words it gives syllabic /r/ Memory, salary 27
Frequency effects Schwa deletion Occurs more in high frequency words Every, family In mid-frequency words it gives syllabic /r/ Memory, salary In low frequency words is gives schwa plus /r/ Mammary, artillery 28
Frequency effects Final /t, d/ deletion Occurs more in high frequency words And, went 29
Frequency effects Final /t, d/ deletion Occurs more in high frequency words And, went Abstract rule models can’t explain frequency effects 30
Exemplar theory Phonological patterns arise from language use/experience Words stored with phonetic details This explains frequency effects 31
Exemplar theory All exemplars are categorized As same sound, word, meaning Exemplars have different strengths Depending on how recently they were activated 32
Exemplar theory Frequency effect happen because Frequent categories have more members Frequent categories have more highly activated exemplars 33
Exemplar theory Prototype No need to calculate and store prototype 34
Exemplar theory Exemplar cloud for two categories 35
Exemplar theory Consider vowel categorization as example Exemplars of two vowels. Height is strength. Distance is phonetic similarity Exemplar to be categorized is * 36
Exemplar theory In perception,most activated exemplars determine the category New exemplar is added to cloud 37
Exemplar theory In perception,most activated exemplars determine the category New exemplar is added to cloud In production a random exemplar from labeled cloud is chosen and the stored motor program for that exemplar is used Stronger exemplars are more likely to be chosen 38
Exemplar theory Modeling lenition Spirantization Not hitting full closure Voicing VCV V voicing encroaches on C Degemination Shortening of duration of C 39
Exemplar theory Example k > x Occasional instances of x produced Over time more instances of x are produced Distributions shifts toward x 40
Exemplar theory If exemplars are words, high frequency words have more exemplars hence more that are lenited They are farther along in historical change 41
Exemplar theory Model will eventually spread out too far How do you prevent that? 42
Exemplar theory Model will eventually spread out too far How do you prevent that? Instead of choosing single exemplar as model of production, randomly choose a space in the exemplar cloud All close and highly activated exemplars influence production 43
Exemplar theory Model will eventually spread out too far How do you prevent that? Instead of choosing single exemplar as model of production, randomly choose a space in the exemplar cloud All close and highly activated exemplars influence production 44
Exemplar theory Regression toward the mean eliminates wide flat distribution 45
Exemplar theory Hearing exemplars of people with same speech causes entrenchment Hearing people with novel speech causes shift in own speech 46
Exemplar theory Hearing exemplars of people with same speech causes entrenchment Hearing people with novel speech causes shift in own speech Older people have more exemplar of older speech variant than young people which is why they don’t adopt newer patterns 47
Exemplar theory Merger: voiced and voiceless stops word finally It’s hard to maintain voicing Phonetic trend is toward devoicing There are more voiceless than voiced stops 48
Exemplar theory 49