Chapter 8 Speech Production

Notice that speech errors are organized mistakes.

1) Elements that interact tend to come from similar linguistic environments.

word-initial sounds are exchanged with other word-initial sounds, etc.

2) Elements that interact tend to be similar to each other

consonants with consonants

3) If mistakes produce nonsense words, they tend to follow the sound pattern (phonological rules) of the language.

4) Affixes and stress stay in their initial position in the sentence.

Drink is the curse of the working class. ->

Work is the curse of the drinking class.

5) We make phonological accomodations for errors

a language acquisition project --->

an anguage lacquisition project

6) Lexical bias (we have a bias toward producing real words)

7) We avoid producing taboo words.

Conclusion:

The speech production system operates in terms of linguistic units--features (sub-phonemic), phonemes/sound level, word level, free morphemes, bound morphemes, etc.

Why do we make these organized errors?

The Freudian View: We are thinking about more things than we are talking about. Errors reveal problems in organization at the conceptual level of production.

How would a psycholinguist model speech production?

An Example of the Fromkin Serial Model:

It's a nice day for a ball game --> It's a nice ball for a day game.

Stage 1) Identify the meaning:

day nice enough for playing ball

Stage 2) Select a syntactic structure

Stage 3) intonation contour generated decide which syllables will have stress and which won't.

NICE DAY BALL GAME

Stage 4) Insert content words into the syntactic structure

Stage 5) Add the function morphemes

Stage 6) Phonetic segments according to phonological rules of the language

Why separate the stages like this? Word exchanges tend to occur across clauses, while sound exchanges tend to occur across at most two words.

Four levels of nodes in permanent memory:

semantic

syntactic

morphological

phonological

The nodes at all levels are connected, so as some get activated, they activate other closely connected ones. There is forward and backward activation.

What the parallel model has that the serial one doesn't: a way to explain why we tend to produce real words : activation is only for what is in the lexicon, so presumably you don't store nonwords in your lexicon.