Speech production• What happens when we speak?• One idea: a spreading activation model (Dell 1986)Speech production• How could this model be tested?– Test predictions regarding time course of production– Look at ways the model can fail and compare to humans• add noise = speech errors• damage network = patients with impairmentsSpeech errors: why?• Looking at how the model goes wrong can tell us how it is put together – What are the units of speech production? – How are units interconnected? – What is the domain of speech planning? Speech errors: how?• Problem: speech errors are very rare; we can’t easily observe them in the lab • Two solutions: – Collect overheard errors over a long time • problem: perceptual bias: are you equally likely to notice all kinds of errors? – Increase chances of making an error in an experimental situation • rapid pronunciation of difficult sequences (cf. “tongue twisters”) • Example: repeat quickly hes nev pem gek (Goldrick & Larson 2008) Types of speech error• Errors are not random; only some possible error types occurSome tendencies• Similar units tend to interact:– phonemes with other phonemes, words with other words– phonemes in same syllable position – words of same grammatical category • Errors tend to respect the rules of the language – proue tordue troue pordue more likely than– clou tordu tlou corduPhoneme similarity effect• Similarity effect extends to phonetic properties of phonemes – ...
Look at ways the model can fail and compare to humans
• add noise = speech errors
• damage network = patients with impairments
•
Speech errors:why?
Looking at how the model goes wrong can tell us how it is put together
Whatare the units of speech production?
How are units interconnected?
What is the domain of speech planning?
• •
Speech errors:how?
Problem:speecherrorsareveryrare;wecan’teasilyobserve them in the lab Two solutions: Collect overheard errors over a long time •problem: perceptual bias: are you equally likely to notice all kinds of errors? Increase chances of making an error in an experimental situation • rapid pronunciation of difficult sequences (cf. “tongue twisters) •Example:repeatquicklyhes nev pem gek
(oGlrdick&Larson2008)
•
Types of speech error
Errors are not random; only some possible error types occur
•
•
Some tendencies
Similar units tend to interact: phonemeswith other phonemes, words with other words phonemes in same syllable position words of same grammatical category
Errors tend to respect the rules of the language proue torduetroue porduemore likely than clou tordutlou cordu
•
Phoneme similarity effect
Similarity effect extends to phonetic properties of phonemes
batbadmore likely thanbatbang • [t] and [d] differ only in voicing • [t] and [ng] differ in multiple features: nasality, place, voicing
•
Lexical bias Errors tend to result in words more than non-words
ex. errors with two phonemes being swapped
•
•
expected % of resulting words: 45
observed % of resulting words: 61
“With thiswing, I theered
Modifying
the model
• •
Inner speech
How is inner speech different from audible speech? Some possibilities: meprnoivhUeisd: processing is exactly the same, except articulators aren’t moved durSpmi-ecafehsirevo: activation at phonological level is weakened or absent (because no articulation) Deep-impoverished: activation at lexical level is weakened or absent (because lexical information is not represented in short-term memory and hence in the production-perception loop)