Hearing and Speech Sciences

You are here: /Home / Newman / HESP 300 / Book Chapter 7 Review

Chapter Seven
Speech production

Outline

Introduction

From concept to expression

Sources of data for models of speech production

  • Speech Errors
  • Disfluencies

Issues in speech production

  • The Units of Speech Production
  • Word Selection and Placement Errors
  • Morphemes and Speech Errors
  • Lexical Search and Pausal Phenomena
  • Grammatical Rules
  • The Phrase as a Planning Unit
  • How Far Ahead Do We Plan?

What speech error data suggest about the process of speech production?

  • Speech Is Planned in Advance
  • the Lexicon Is Organized Both Semantically and Phonologically
  • Morphologically Complex Words Are Assembled
  • Affixes and Functors Behave Differently than Content Words in Slips of the Tongue
  • Speech Errors Reflect Rule Knowledge

Speech production processing models

  • The Utterance Generator Model of Speech Production
  • The Garrett Model
  • Levelt's Model
  • Dell's Model

Summary

Key Concepts From the Book

I. The study of speech production attempts to explain how people turn a mental concept into a spoken utterance. Because of the difficulty of constructing experiments to reveal the complex steps in this process, psycholinguists rely primarily upon analysis of speech errors and speech disfluencies. (Page 310)

II. Speech communication may be viewed as a chain of events linking the speaker's brain with the listener's brain. We know a great deal about the physiological, articulatory and acoustic aspects of speech production, but we are still far from understanding the processes by which speakers put the message they wish to convey into linguistic form. (Page 310)

III. Although philosophers have long speculated about the language of thought, how concepts are represented in the brain is still a mystery. Numerous alternative views have been posited, with little empirical evidence to support one any more than the rest. (Pages 311-312)

IV. Speech errors or slips of the tongue, (such as Spoonerisms) occur regularly in normal conversation. These errors provide indirect evidence for the units, stages and cognitive computations involved in speech production. Study of speech produced in seminars, classes, and business meetings reveals that 98% of utterances are grammatical. The remaining 2%, which is ungrammatical, provides insight into the production process. (Pages 312-313)

V. Disfluencies, such as filler words (creating filled pauses) hesitations, repetitions, and false starts also provide insight into speech production. When describing pictures, hesitations (unfilled pauses) occur roughly every 5 words. When conversing naturally, hesitations and fillers occur every 7-8 words. (Page 313)

VI. Speech is produced by stringing together, arranging and rearranging a finite number of stored items. One of the major issues in the study of speech production is to determine just what these stored units are. Analysis of speech errors has show that utterances are composed of discrete units of varying sizes and classes. (Page 313-315)

VII. Speech error collections show many examples of phonological errors in which single phonemic-sized elements are anticipated (sounds which will come later on appear earlier than intended) or persevere (sounds produced early in the utterance reappear in an incorrect location later on.) Other errors include phonemes that are deleted, added, or exchanged. (Page 315)

VIII. Phonetic segments as units of speech appear to be organized as vowels, consonants, and consonant clusters. (Page 315)

IX. The most elementary unit of speech production is the quark, which is composed of the phonetic features (or distinctive features) of the utterance. Phonological features errors are common. Vowel features may exchange or become confused with other vowel features and consonant features with other consonant features, but consonant features and vowel features do not interfere with each other. (Page 316)

X. Syllables also constitute units in the phonetic programming system. Initial syllables interact with other initial syllables, medials with medials and finals with finals. (Page 316)

XI. Word and phrasal stress may also be processing units. Studies of speech errors in tonal languages show that tone can become disordered.(Page 317-318)

XII. Word selection and placement errors are also an object of study. Disfluencies and hesitations occur more before content words than before function words, suggesting that the speaker has not yet arrived at his lexical target. (Page 318)

XIII. Morphological units serve as production building blocks. Stem morphemes, derivational morphemes and inflectional morphemes undergo rearrangement in different sorts of errors. The rules of inflectional and derivational morphology surface in speech errors. (Page 319-320)

XIV. Even larger linguistic structures, such as phrases, can serve as speech units. Sentence constituents (noun phrases, verb phrases, prepositional phrases) are marked off as units when we speak. Noun phrases and verb phrases may be exchanged in speech errors, but the grammatical structure of the sentence is not disordered. (Page 320)

XV. Other support for the phrase being a separate speech unit is that in self corrections and retracings, speakers tend to go back to the beginning of the constituent containing the error. (Page 320-321).

XVI. Pauses are likely to occur at clause boundaries, suggesting that speakers may need the pause time to encode the following clause. Syntactically more complex speech tends to contain more hesitations and filled pauses, suggesting that pauses reflect active sentence-planning efforts. (Page 321)

XVII. Because from 79%-85% of exchange errors occur in the same clause, researchers believe that speech is mostly planned in clausal units. However, the 15%-21% of errors found outside of clausal boundaries show that speech planning must include larger segments of discourse. (Page 321-322)

XVIII. Theoretical, explanatory models of speech production must account for the hierarchical nature of speech production. A viable model of production must posit all and only the necessary stages of production, and predict the form of the utterance representation at that level. (Page 322-323)

XIX. Substitution errors and blends suggest that words are organized both semantically and phonologically. Erroneous selection occurs only after the appropriate class of word has been selected -- nouns substitute for nouns, verbs for verbs, etc. The words involved are often phonologically similar.(Page 323)

XX. Some lexical selection errors seem to fall into the category of Freudian slips, the result of unconscious competing plans or nonlinguistic interference, either internally or externally introduced. (Page 324)

XXI. Some word substitutions occur that appear to be influenced by previous words in the string -- they are syntagmatically conditioned. In these cases, the speaker may be selecting lexical entries listed in close proximity to the intended item. (Page 324)

XXII. Derivational errors in morphologically complex words appear to occur before lexical insertion. Morphological rules of word formation therefore appear to be actively engaged during speech production and morphologically complex items are compiled, even if they are stored as wholes. (Page 324-325)

XXIII. Affix morphemes and minor sentence elements are often moved or shifted in the sentence, whereas major category stems and words are more likely to be involved in exchange errors. (Page 326)

XXIV. Speech errors reflect rule knowledge. The productive use of such rules are shown in utterances where a regular rule has been applied to an irregular form.(Page 326-327)

XXV. The Utterance Generator model proposed by Fromkin (1971) is a top-down generator that distinguishes six stages at which different representations of the utterance occur. The stages are:

A. the generation of the meaning to be conveyed

B. the mapping of the meaning onto a syntactic structure

C. the generation of the intonation contours of the utterance

D. the selection of words from the lexicon

E. phonological specification

F. the generation of the motor commands for speech

(Page 328-331)

XXVI. The Garrett model of speech production (1975) makes some of the implicit aspects of the Fromkin model explicit and provides a major framework for further work in the field. This model distinguishes between a conceptual level, a functional level and a positional level. (Page 331)

XXVII. Levelt's model (1989) says message generation is initiated by the conceptualization of the utterance. First, intention is conceived, creating the preverbal message which is then fed to the formulator. The formulator includes both a grammatical encoder (which retrieves lexical items) and a phonological encoder. The articulator then executes the phonetic plan by sending messages to the neuromuscular system. Finally, the speech comprehension system checks the output for errors. (Pages 331-334)

XXVIII. A lemma contains an item's meaning as well as its syntactic properties.(Page 333)

XXIX. Dell's model uses the concept of spreading activation to account for speech production. In this model, words and rules are organized into a network with connections between units based on semantic and phonological relatedness. Activation is assumed to be bidirectional: it is possible to have interactions between semantic and phonological representations. (Page 337)

XXX. Recently retrieved lexical items can prime the speaker to make a slip based on the phonetic bias paradigm. (Page 338)

Print This Page
Department of Hearing & Speech Sciences
0100 Lefrak Hall, College Park, MD 20742
Tel: 301-405-4214
Fax: 301-314-2023