Comp models of reading (SM89 and PMSP) Flashcards
What does SM89 stand for and nicknamed?
its creators, Seidenberg and Mcclelland 89. Nicknamed the ‘triangle model’ for 3 components; orthographic, semantic and phonological
Explain units of SM89
uses back-propogation methods of input (visual) -hidden- output (phonological).
uses 400 ortho, 200 hidden and 460 phono units
what is a ‘wickelfeature’
Used in SM89 phonemes encoded using grapheme and phoneme pattern of activation; bad as disperse graphemes of spelling-sound regularities
How does SM89 use feedback?
with semantics but not from ortho to phono as no necessarily linked (esp in english)
How was the SM89 trained?
with novel word presentation of 3000 words; to repeatedly computer error score (actual desired outcomes)
evidence FOR SM89?
good for spelling as humans do (Slower for exceptions)
lexical processing involves information and statistical regularities
good for representing phonoloigcal dyslexia (farah 96) - impairment at phonoligcal level.
evidence AGAINST SM89?
Coltheart 93: doesnt account for reading aloud, visual judgements, and represents surface D badly.
Besner 90: only reads non-words as a human 70% of time
significantly less word input than humans and makes lexical decision errors
How did Forster criticise computational accounts?
Just because output is correct doesn’t mean this is actually how humans do it necessarily.
What is the revised connectionist account know as ?
PMSP after its authors losely from 97-97 (plaut, mclelland, seidenberg and patterson)
why is PMSP considered a better model?
it is much better at pronouncing non words and at lexical decision tasks, has more realistic input and output
how are constraints dealt with in the PMSP
phonological representations based on phonotactic constraints of language (phonemes that are legal together)
orthographic reps are based on graphotactic constraints (letters that are legal together)
Features of the SM89?
depends on frequency of exposure; positive response for ‘friends’ and negative for ‘enemies’; non linear response to letter strings
two pathways: direct (ortho-sem OS) or indirect (ortho-phon-sem OPS/ phonoloigcal mediation)
- modulate by skil and word frequency
Can PMSP model dyslexia?
ys good for surface dyslexia (patterson, seidenberg and mcclelland 89) by impairing OS and OPS to simulate semantic deficits
Surface D is when the semantic pathway is damaged leaving only the isolated phonological pathway
Plaut 97
individual differences in reading exception ability are explained by the devision of labour betwen sem and phon pathways
what about deep dyslexia?
Neither model can explain, so Hinton and shallice 91 created a new one that uses back-prop to associate pronunciation with representations of meaning (i.e. paralexia and visual errors)