Theories Of Speech Production

Several theorists (e.g., Dell, 1986; Dell & O'Seaghdha, 1991; Garrett, 1976) have used evidence from speech errors to construct theories of speech production. These theories have much in common. First, it is assumed that there is a substantial amount of pre-production planning of speech. Second, most theorists agree that there is a series of processing stages in speech production, and there is even agreement that there are four processing stages. Third, it is assumed that the processes proceed from the general (the intended meaning) to the specific (the units of sound to be uttered). The processes in speech production resemble those involved in comprehension, except that the processes are in the opposite order. For example, the goal of comprehension is to understand the meaning of a message, whereas with speech production the meaning of the message is the starting point.

In view of the similarities among theories of speech production, we will consider only two theoretical approaches. First, the spreading-activation theory of Dell (1986) and Dell, Burger, and Svec (1997) will be discussed. It emphasises a psychological process (spreading activation) of general significance within language processing (see Chapter 12). Second, the theoretical approach of Levelt, Roelofs, and Meyer (1999a) is discussed.

Spreading-activation theory

Dell (1986) and Dell and O'Seaghdha (1991) put forward a spreading-activation theory. It was based on connectionist principles, and consists of four levels. The main assumptions of the theory (including descriptions of the four levels) are as follows:

• Semantic level: the meaning of what is to be said; this level is not considered in detail within the theory.

• Syntactic level: the grammatical structure of the words in the planned utterance.

• Morphological level: the morphemes (basic units of meaning or word forms) in the planned sentence.

• Phonological level: the phonemes or basic units of sound within the sentence.

• A representation is formed at each level.

• Processing during speech planning occurs the same at all four levels, and is both parallel and interactive; however, it is typically more advanced at higher levels (e.g., semantic) than at lower levels (e.g., phonological).

According to spreading-activation theory, there are categorical rules at each level. These rules are constraints on the categories of items and on the combinations of categories that are acceptable. The rules at each level define categories appropriate to that level. For example, the categorical rules at the syntactic level specify the syntactic categories of items within the sentence.

In addition to the categorical rules, there is a lexicon (dictionary) in the form of a constructionist network. It contains nodes for concepts, words, morphemes, and phonemes. When a node is activated, it sends activation to all the nodes connected to it (see Chapter 1). Finally, insertion rules select the items for inclusion in the representation at each level according to the following criterion: the most highly activated node belonging to the appropriate category is chosen. For example, if the categorical rules at the syntactic level dictate that a verb is required at a particular point within the syntactic representation, then that verb whose node is most activated will be selected. After an item has been selected, its activation level immediately reduces to zero; this prevents it from being selected repeatedly.

According to spreading-activation theory, speech errors occur because an incorrect item will sometimes have a higher level of activation than the correct item. The existence of spreading activation means that numerous nodes are all activated at the same time, and this increases the likelihood of errors being made in speech.

What kinds of errors are predicted by the theory? First, errors should belong to the appropriate category (e.g., an incorrect noun replacing the correct noun), because of the operation of the categorical rules. As expected, most errors do belong to the appropriate category (Dell, 1986).

Second, many errors should be anticipation errors, in which a word is spoken earlier in the sentence than is appropriate (e.g., "The sky is in the sky"). This happens because all of the words in the sentence tend to become activated during the planning for speech.

Third, anticipation errors should often turn into exchange errors, in which two words within a sentence are swapped (e.g., "I must write a wife to my letter"). Remember that the activation level of a selected item immediately reduces to zero. Therefore, if "wife" has been selected too early, it is unlikely to compete successfully to be selected in its correct place in the sentence. This allows a previously unselected and highly activated item such as "letter" to appear in the wrong place. Many speech errors are of the exchange variety

Fourth, anticipation and exchange errors generally involve words moving only a relatively short distance within the sentence. Those words relevant to the part of the sentence that is under current consideration will tend to be more activated than those words relevant to more distant parts of the sentence.

Fifth, it is predicted that speech errors should tend to consist of actual words or morphemes (this is known as the lexical bias effect). This effect was demonstrated by Baars, Motley, and MacKay (1975). Word pairs were presented briefly, and participants had to say both words as rapidly as possible. The error rate was twice as great when the word pair could be re-formed to create two new words (e.g., "lewd rip" can be turned into "rude lip") than when it could not (e.g. "Luke risk" turns into "ruke lisk").

Sixth, the notion that the various levels of processing interact flexibly with each other means that speech errors can be multiply determined. Dell (1986) quoted the example of someone saying "Let's stop" instead of "Let's start". The error is certainly semantic, but it could also be regarded as phonological, because the substitute word shares a common sound with the appropriate word. Detailed investigation of such wordsubstitution errors reveals that the spoken word and the intended word are more similar in sound than would be expected by chance alone (Dell & O'Seaghdha, 1991; Harley, 1984).

According to spreading-activation theory, most errors are caused by spreading activation. It might appear preferable if activation did not spread so widely through the lexicon, because then there would be fewer speech errors. However, Dell (1986) argued that widespread activation facilitates the production of novel sentences, and so prevents our utterances from becoming too stereotyped.


Spreading-activation theory makes various precise and testable predictions about the kinds of errors that should occur most often in speech production. Another strength of the theory is that its emphasis on

The relationship between overall error rate and the anticipatory proportion. The filled circles come from studies reported by Dell et al. (1997) and unfilled circles come from other studies. Adapted from Dell et al. (1997).

Stop Anxiety Attacks

Stop Anxiety Attacks

Here's How You Could End Anxiety and Panic Attacks For Good Prevent Anxiety in Your Golden Years Without Harmful Prescription Drugs. If You Give Me 15 minutes, I Will Show You a Breakthrough That Will Change The Way You Think About Anxiety and Panic Attacks Forever! If you are still suffering because your doctor can't help you, here's some great news...!

Get My Free Ebook


  • Russell White
    What is spreading activation theory of dell?
    1 year ago
  • diana
    What is traditional theory of speech production?
    1 year ago
  • christina roth
    What psychological theory is child language and speech recognition?
    5 months ago
  • mollie
    What makes speech recognition difficult psychology?
    28 days ago
    What is word production cogniitve psychology?
    26 days ago

Post a comment