Summary of the presentation of Gary S. Dell’s "Spreading-Activation Theory of Retrieval in Sentence Production"

Essay , 2012 , 7 Pages , Grade: 1,7

English Language and Literature Studies - Linguistics

In Gary S. Dell’s article ”A Spreading-Activation Theory of Retrieval in Sentence Production” which was published in Volume 93 of the Psychological Review in 1986, he analyses the production of language which is “the least studied and the least understood” by psycholinguistics. The former research concerning the production of language is not sufficient in Dell’s opinion. Dell’s theory tries to make predictions that can be fitted to models.

Table of Contents

1. Introduction

2. Differences and similarities to M. F. Garrett’s model

4. Types of errors

5. Factors that make errors more likely

6. Spreading Activation and Sound Errors

7. Conclusion

8. Reference

1. Introduction

In Gary S. Dell’s article A Spreading-Activation Theory of Retrieval in Sentence Production which was published in Volume 93 of the Psychological Review in 1986, he analyses the production of language which is “the least studied and the least understood” (p. 283) by psycholinguistics. The former research concerning the production of language is not sufficient in Dell’s opinion. Dell’s theory tries to make predictions that can be fitted to models. The basis of his model is M. F. Garrett’s model of sentence production, which Dell develops further. The principal assumption of his model is that there are three steps on the way from meaning to sound:

1. The syntactic encoding: words must be chosen and arranged according to the grammatical rules.
2. The morphological encoding: words must be specified in terms of their constituent morphemes.
3. Phonological encoding: words must be spelled out in terms of their sound.

Dell focuses on this phonological encoding to underline his spreading activation theory of retrieval. This theory is based on data given by speech errors, often called “Freudian slips”.

In our presentation of Gary S. Dell’s theory we concentrated on his analysis of how speech errors are created and what kinds of errors can occur. We explained how his model works. The basis of his model is the theory that all four representations (semantic, syntactic, morphological, phonological) are in a “hierarchical structure” (p. 288). These four representations are connected by nodes, which interact with one another in any direction, “top-down connections” and “bottom-up connections” (p. 288). There are no inhibitory connections.

2. Differences and similarities to M. F. Garrett’s model

As he is particularly concerned with M. F. Garrett’s model of sentence production, which says that “it is not a theory of why a speaker says what is said but of how it is said.” (p. 283). The assumption that there is always a bi-directional connection between the different levels, the so-called Spread of Activation, is the main difference to Garrett’s model, which says that there is no connection at all. So in Dell’s model the levels must not be viewed distinctively and there is no differentiation between open and closed classes. The model is interactive: a network of nodes. As Dell’s model is an advancement of Garrett’s model there is also a similarity like the fact that both models account for the lexical bias effect which is the tendency for people when making speech errors to use true words which they know rather than inventing non-words.

3. Gar1y S. Dell’s model

In his chapter “Speech Errors in the Theory” Dell shows all possible connections between the different nodes that can create speech errors. To underline his theory of sentence production Dell uses the sentence “Some swimmers sink” (p. 291).

When a word is selected, all nodes that represent the word are activated. Slips occur when nodes within the same level are competing for activation. For this sentence he offers some possible errors among which is the sentence: “Sim swimmers sink”, a phoneme anticipation, which he explains in detail in a model that we used in our presentation.

The model shows three levels of nodes. The upper level is the syntactical level, the middle level is the morphological level, and at the bottom is the phonological level. Slips occur when a “wrong” node (error node) has a higher activation level than the “right” node (target node). This is usually done by context effects.

In his model the syllable /sÙm/ is encoded at the phonological level and the node for the vowel /I/ is activated by “swim” and “sink” which are both in the planning process. So the “wrong” node is the /I/ from the words “swim” and “sink”. The node for /I/ will replace /Ù/ and an error occurs. The right sentence “Some swimmers sink” becomes “Sim swimmers sink”. Other possible errors are e.g.: “Swum simmers sink” (phoneme shift), “Some simmers sink” (phoneme deletion) and “Some swimmer drown” (word substitution), (p. 291).

4. Types of errors

There are misordering or contextual errors (the intruding item comes from the speaker’s intention) and there are non-contextual errors (the intruding word comes from outside), (p. 292).

1. Contextual errors: The main reason for these errors is that “upcoming items receive activation from nodes in the higher level representation” (p. 292). The sentence “Some swummers sink” belongs to the error type phoneme perseveration. The sentence “Sim swummers sink” belongs to the error type phoneme exchange and the sentence “Some swinkers sink” to the error type cluster anticipation.
2. Non-contextual errors: They occur when a node outside the intended utterance has a higher activation level, e. g. “Some swimmer drown” is a so-called substitution error and belongs to this type of errors.

5. Factors that make errors more likely

1. There are output biases where meaningful words and expressions are created (p. 292). To this group belong lexical biases (actual words or morphemes are created) and semantic biases (wrong words are semantically related to other words in the environment).
2. Another factor is similarity: Interacting items in an error tend to be similar and the immediate environment of the interacting items tends to be identical.
3. Speaking rate is also an important factor. At fast speaking rates there is not enough time for the recall of the correct words.
4. Furthermore, distance is another factor for the production of errors. Misordered sounds and morphemes are likely to move to nearby positions.

6. Spreading Activation and Sound Errors

Output biases:

In this section of his article Dell mentions three output biases.

1. Lexical bias or syllable bias:

This is a bias for errors to create existing syllables over phonologically legal but non-occurring syllables, e.g. // as in spatula.

2. Frequency biases:

Initial /t/ is more frequent and has more connections from its node to syllable nodes than initial /f/. This has consequences for its activation level. “Frequent sounds get more activation from activated syllables than do in frequent sounds” (p. 301).

3. Contingent frequency:

Just as the frequency bias concerns a single unit here Dell refers to combinations of units that are frequent. As an example he mentions that /kæ/ occurs more often than /kU/.

The repeated-phoneme effect:

Furthermore, the section contains the repeated-phoneme effect. Dell explains that effect by saying that “Repeated sounds tend to induce the misordering of sound around them” (p. 301). He gives the example of “had pat” which leads to the exchange error “pat had” or the anticipation error “pat pad” or the perseveration error “hat had”. A repeated-phoneme effect is also the error that changes “hat pet” into “pat het”. Here, sounds that are not neighbours of the repeated sound are misordered. “This repeated-phoneme effect is the consequence of the structure of the network” (p. 301).

The speaking-rate effect:

1. Lexical bias and speaking-rate:

If by increasing the speaking rate the available time is reduced, there is less time for constructive spreading. So the wrong sounds are more activated than the correct sounds. Dell differentiates between lexical bias and speaking-rate on the one hand and repeated-phoneme effect and speaking-rate on the other hand.

Dell presents the word “kitten” and explains how it can become “ditten” or “mitten” (p. 302). If the speech rate is slow, the activation is given time to spread and existing words occur, e.g. “mitten”. If the speech rate is at a higher level, which means that there is less time and, as there is no morpheme node for “ditten”, the activation of non-existing words is more likely, e.g. “ditten”.

2. Repeated-phoneme effect and speaking rate: The repeated phoneme effect is stronger at slower speaking rates.

3. Error types and speaking rate:

For this type of error, Dell gives two examples: an exchange error like “pancake” and “canpake” and an anticipation error “pancake” and “cancake” (p. 302). Here, the speaking rate affects the proportion of exchanges and anticipations.

7. Conclusion

Gary S. Dell’s model is generally accepted by modern linguistics, as it incorporates other models, especially M. F. Garrett’s model. But he comes to different conclusions. Some error phenomena are proved by experiments, but as he admits himself, some have not been tested empirically and are predictions (p. 302) from his own theory.

8. Reference

Dell, G.S. 1986. A Spreading-Activation Theory of Retrieval in Sentence Production. Psychological Review, 93, 283–321.

[...]

• College: University of Frankfurt (Main) (Institut für England- und Amerikastudien)
• Course: Language Production
• Grade: 1,7
• Author: Judith Leitermann (Author)
• Publication Year: 2012
• Pages: 7
• Catalog Number: V338037
• ISBN (eBook): 9783668274433
• Language: English
• Tags: summary, gary, dell’s, spreading-activation, theory, sentence, production