language - word recognition

Question 1

learning objectives lecture 3

Answer 1

-understand the uniquness point of a word
-understand the basics of a cohort model
-access , selection and integration stages
-the distinction between early and late versions of the model
-understand that there are other models of spoken word recognition
-template matching ,TRACE etc

Question 2

what happens after phoneme recognition? after we recognise specific sounds

Answer 2

-theres a number of models on this-one is cohort model

Question 3

early models of word recognition
-template matching model

Answer 3

template matching models
-targets stored as templates (a store in your brain with all the words you know)
-when you get an incoming speech stream what your doing is comparing the words in that incoming speech stream to the template of words you know, once you find a match that’s how identify a word
-a template is an exact description of the sound or words being searched for
-identification occurs when a match is found

Question 4

problems with template matching approach

Answer 4

-far too much variation in speech eg dialect,basic pitch,speech rate
-variation in how words are prnounced - far too muvh variation in speech eg dialect, basic, speech rate, youd have to store a lot of differnt version of words in this template

Question 5

early models of word recognition
analysis by synthesis

Answer 5

-when you listen to speech you recognise it by generating what you would have to do to pronounce that speech yourself
-when we hear speech we produce or synthesise a succession of speech sounds until we match what we hear
-largely discredited as unclear how this might work

Question 6

problems with analysis by synthesis models

Answer 6

-although there is evidence motor areas in the brain are activated when we perceive speech, whether we recognise speech this way is another question
-in children comprehension skills always ahead of production skills (kids can understand more words than they can produce eg)(from a developmental perspective it not clear how that would work)- if you were using the ways that you synthesis language to understand language why would it be that comprehension is ahead of production
-people who have trouble producing speech paraphasia, but can still understand speech, so makes it unclear how this model would work

Question 7

key debate in word recognition today
top down vs bottom down processing
definitions

Answer 7

top down processing - processing that involves knowledge coming from higher levels , eg,context, experience , expectations ,predictions

bottom up processing
processing that is data driven i.e directly influenced by environmental stimuli, so stimulus triggers a sequence of actions that then results in word recognition

Question 8

key debate in word recognition- top down versus bottom up processing

Answer 8

debate
-for example to what extent does sentence context help us recognise a word
-do we make predictions based on the context of a sentence to help us recognise a word or is it just the incoming speech stream that activates the process of word recognition

Question 9

cohort model
-developed by
-key terms cohort and uniqueness

Answer 9

-developed by William Marslen Wilson and his colleagues 1970s onwards
key terms
cohort- the set of candidates that a heard word could possibly be
uniqueness point: the point at which a word can be distinguished uniquely from other similar words
early and later versions of the cohort model differ slightly

Question 10

uniquness point

Answer 10

the more you hear of a word being pronounced the fewer the candidates there are for what that word might be
/c/ canoe, cocoon,karate ,confess,confetti,cook
/cuh/-cocoon,karate,confess,confetti
/conf/ -confess, confetti
/confess/ - confess

Question 11

demonstration
late uniqueness words and early uniqueness words

Answer 11

late uniqueness eg
-ballot,palace,fillet,locust,relic,
early uniqueness eg
-breakfast, dungeon,biscuit,onion,sergeant,trumpet
-you hear the exact same info for these words in the demonstration, it sounds like you hear more with the early uniq words
-late uniq words, at the point that they were cut off in the demonstration, there are still many words they could have been
-early uniq words, by the time you get the cut off point there are no other words it could be/not many competitors

Question 12

cohort model
access stage
selection stage
integration stage

Answer 12

access stage
-perceptual representation used to generate a candidate set of words it might hear eghear k sound the cohort is developed (set of words it could be )
selection stage
-one item is chosen from this set ,it does this when the uniqueness point of the word is reched and theres no more competitors
integration stage
-semantic and syntactic properties of word are accessed, what the meaning of the word is etc

first two stages are prelexical (before word recogniton happens)
last stage is postlexical (after word recognition happens)

Question 13

earlier versions of cohort model

Answer 13

-sensory information (i.e the sounds of the word) and context information (eg the rest of the sentence ) are used at the same time
-an element of top down processing
-we only listen to the whole word when there’s no useful sentence information

Question 13

late and early uniqueness words
explain in terms of cohort model

Answer 13

late uniqueness words
ballot - ballad,balacalava,balnace,ballerina,ballast
locust-local,locomotion,locum,locust
so these words have many competitors so a large cohort which means they are stuck at the access stage

early uniqueness words
breakfast
dungeon
no competitors at the cut off point
can proceed to selection and integration stages

Question 13

cohort model early version
-activation of words is all or nothing

Answer 13

-words are either in the cohort or out the cohort (no in between)
-activation of words is all or nothing
-once a word is eliminated from the cohort it cannot be reintroduced
-this means that if the beginning of a word is mispronounced, the cohort model cannot recognise it
-eg intented word is pleasent but speaker says bleasent, cohort model wont recognise it as due to the b sound pleasent is already eliminated

Question 14

cohort model-a later version
-what does the later version say about context

Answer 14

context does not influence the cohort
-it cannot be used to eliminate members of the cohort before the uniqueness point
-you cant use context you just need to use the uniqueness point and thats all

Question 15

what does the later version of the cohort model say about activation of words

Answer 15

-elimination of candidates not all or none
-the cohort can retain words with similar initial phonemes
-eg bleasent vs pleasant
-this means that if the model mis hears the first sound of a word , it can recover

Question 16

evidence for the cohort model
mispronunciation task

Answer 16

-early work on the cohort model involved tasks in which people listened for mispronunciations
-eg participants listen to speech where the sound is distorted eg boot changed to po ot

-turns out-participants are more sensitive to changes at the beginning of words than at the end - we do not pay attention equally to all parts of a word

this is consistent with two aspects of the cohort model
-that we not always listen to the whole of the word if we can identify it
-that word recognition proceeds from the beginning of the word

Question 17

cohort model evidence
semantic priming

Answer 17

-uses a lexical decision task
-word shows up on a screen and you have to press one button if its a word and another button if its a non word
-lexical decision tasks are really good at getting at how difficult processing an individual word is
-semantic priming-it is easier to recognise a word if you have just een or heard a word thats related in meaning
eg captain will help you recognise ship

Question 18

cohort model evidence
semantic priming
will a fragment of a word respond the same?

Answer 18

-captain will prime responses to ship
-‘capt’ is almost as good a prime for ship as ‘captain’
-‘tain’ is not a prime for ship
again suggests that the start of the word is important in generating this cohort,
-suggests that you start to access words as soon as you hear them
-you do not access words unless they begin with the sounds you hear

Question 19

cohort model evidence
-visual world paradigm

Answer 19

-you have a screen infront of you and a head mounted eye tracker and it monitors which parts of the screen your looking at
-the logic is that the place your looking at on the screen is where your attention is focused
-in this case you would hear something like click on the cloud (in the headphones) and then you can monitor where people are looking as soon as they start to hear this word
-as soon as they hear cl- they start looking at the clown or the cloud so as soon as the word starts they’ve ruled out any other pictures

Question 20

cohort model - early verses late

Answer 20

early version
-words are either in or out of the cohort
-context can influence the cohort

later version
-cohort can contain words with similar initial phonemes
-context does not influence the cohort

Question 21

evidence for the later version of the cohort model
phoneme detection study

Answer 21

-in our out assumption
frauenfelder et al 2001 french study
-phoneme detection study where they had to listen out for various sounds
-they were interested in looking at what happened when you distorted the intial phonemes of words
eg focabulaire instead of vocabulaire
-listeners were disrupted by the mismatch, but still recognised the relevant word

Question 22

later version of the cohort model - evidence
-why take out the role of context?

Answer 22

-freidrich and kotz (2007)
-in this study the presented sentences like ‘to light up the dark she needed her can___ and then presented candle,lantern,candy,number
-ERP (electrodes on scalp)
evidence suggests that candy was activated
-erp component called p250 is an index of word access-tells you if someone has activated a word

Question 23

evaluate the cohort model of word recognition

Answer 23

-good strong evidence base
-intutitivley makes sense

-debate on how plausible the model is

Question 24

cohort evaluation
-problem with context

Answer 24

-wberer and crocker 2012
-sometimes context does seem to matter
-it depends on the stimuli that you use in your experiment
-they used the german equivalent of ‘the women irons the_____’
-then they would hear bl
-speakers looked more to a picture of a blouse (german bluse) than a flower (german bloom)
-but they also looked more at the flower than dissimilar words

Question 25

other problems with cohort model

Answer 25

-there isnt really a role for word meaning in this whole word recognition process
de emphasises the role of word meaning
-word meaning seems to influence word recognition (especially how concrete a word is) concrete words refer to physical objects, these stand in contrast to abstract words - a word that is a thing but its not tangible like eg idea
-people are faster to respond to concrete words than abstract words

-how does the cohort model know when words begin?
-cohort model is good at recognising words in isolation
-heavily relies on knowing when words start without having any explicit mechanisms for finding the starts of words
-great for isolated words but what about sentences

Question 26

other models of word recognition
TRACE

Answer 26

-connectionist model
-units that can be activated
-involves computer simulations with many simple processing units
-can train it to do a ceryain tasks
-theyve been trained on word recognition

assumes that top down and bottom up processes interact flexibly
-all sources of information used at the same time (unlike cohort)
-emphasis on the role of context on word recognition
-in contrast with the later version of the cohort model, which has a bottom up emphasis