Producing words

Speech production: processes involved in converting an intended meaning into a spoken utterance.
1. Process falls into 3 broad areas: Conceptualization (what to say, intention), formulation (translating conceptual into linguistic form) & encoding (articulatory planning) Levelt (1989)
Speech errors: Freud- two different intentions. Repressed thoughts (1901) Freudian slip. Turn out to be rare.
1. Categories: Anticipation, Perseveration, Exchanges, Substitutions, Blends & Stranding errors
2. Ellis (1980) recognised Freud's speech errors in terms of a modern process-orientated account of speech production method of analysis, collect recording of errors and ask people what they meant to say (could produce observer bias but doesn't seem to)
  1. Two interpretations of speech errors
    1. If X and Y are involved in anticipation substitution or exchange then they must have been simultaneously available at some point in production process. Errors can tell you able scope of planning
    2. Garret (1980) we produce speech through a series of discrete levels of processing, these levels DO NOT interact with each other i.e. if x & y interact in an error they must be from the same "processing vocabulary" (x & y units at relevant level of processing)
      1. This (1976) model distinguishes between 2 major stages of syntactic planning: Functional - (word order is not yet explicitly represented. Semantic content of words is specified and assigned to syntactic roles such as subject and object). Positional - words explicitly ordered. There is a dissociation between syntactic planning and lexial retrieval
        Garrett argued that content (noun verbs semantic) and function (syntactic) words play v different roles in language production. Content words are selected at the functional level wheras functional words selected at the postional level
        Evidence for Garret- diff levels cannot exchange errors, observed that content words almost always only exchange with other content words and same for function words. Robust finding Harley found this too. Supports the idea that content and function words are from computationally distinct vocanularies
        Common exchange errors involve single words, or single sounds. Suggests there are processes fro finding and arranging words (grammatical encoding), and processes for finding and arranging sounds (phonological encoding). Likely to be separate process operating with different "processing vocabularies"
        Word exchanges usually involve the same grammatical category (content words, noun to noun). Usually between words in different phrases, constrained by syntactic factors, same level.
        Sound exchanges are often between words of diff grammatical category but sounds are phonologically similar. Usually between adjacent words, constrained by distance, same phrases.
        Errors can occur within any of the levels ie. word substitution errors occur at the functional level, sound sub at sound level
Lexical bias: speech error more likely to result in a real word than non-word. Baars et al (1975) barn door experiment
1. Evidence for monitoring process: an internal process "notices" when a non-word is going to be produced and corrects error, not as good as noticing a real word error.
Substitution errors: tend to be related to target word in either sound OR meaning. Suggest 2 process: retrieving the meaning (functional) AND retrieving the sound (sound level)
Lexicalization: process of turning a word meaning into a sequence of sounds
1. 2 stage theory: Initial stage meaning based (functional processing), a lemma is retrieved. Second phonologically based corresponding word forms (lexemes) are retrieved
  1. Lemma account: 2 layers of lexical representation- Lexical syntax must occur before phonological form. Lemmas are amodal-level of representation mediating semantics and phonology takes no account of modality
    1. Evidence for existence of 2 stage but lemmas is debatable
      1. For: Indefrey & Levelt (2004) meta analysis of brain imaging studies of word naming: lemma retrieved 150-224 ms and phonological info retrieved 250-339 ms. Different area of the brain activated in sequence as we produce words. Levelt (1990) words prime semantic neighbours early on wheras late on they prime phonological neighbours. Suggests there is an early stage when semantic candidates are activate lemma and later stage when phonological.Wheeldon & Monsell (1992) repetition priming- subjects had to speak aloud, subsewuent naming task had to name a picture. If spoken before naming times speeded up. But homophones didn't induce priming. Therefore the priming must have been mediated by meaning, not phonological form = Lemma level priming. Badecker et al (1995) Gender and anomia, patient had word finding difficulties (like TOT). Italian has grammatical gender which is often unrelated to meaning, failed to name nouns but got gender = lemma includes grammatical gender info not phon
        Orign of TOT (Brown, 1970) partial activaton-the target items are inaccessible because they are only weakly represented in the system, most data support this. Data also suggests levels of semnatic & phonological processing in lexical retrieval are distinct. The TOT state is readily explained by success of the first stage of lexicalization but failure of the second Vigliocco, Antonini & Garrett (1997) still recieve gender syntactic info not phonological
      2. Against: Butterworth (1982) formulated word retrieval explicitly in terms of a 2 stage process, semantic and phonological substitutions occur at different levels. Word substitution errors support 2 stage model but say nothing about existence of amodal, syntactically specified lemmas
        Fay & Cutler (1977) phonological and semantic word substitutions happen as a result of mistakes in different parts of the word retrieval process (independent)
    2. Evidence against 2 stage altogether
      1. Caramazz & Miozzo (1997) Sometimes TOT states allow access to partial phonological info without gender (without semantic). 2 stage model woud predict phonological info only avaliable if gender is avaiable. Lemmas are unnessary complications, if they exist (amodal) impairments involving words should not be modality specific, however patients are impaired in producing words of one grammatical class/1 modality .Moreso recent evidence to suggest grammatical gender of noun is retrieved from lexeme not lemma, this could be explained by models that don't assume a distinction between semantic and phono stage (lack a lemma)
2. Lemma (Levelt 1989): abstract words containing grammatical and semantic info but not phonological, represents a word
3. Choosing a word called lexical selection
4. Discrete or Cascade model?
  1. Discrete: lemma selection (right lemma from activates ones) is complete before 2nd stage (word form = lexeme) started
    1. Evidence: Picture-word interference (Schriefers et al (1990) picture and word heard doesn't match. Word can be semantically, phonologically related or unrelated. Can also be presented before, simultaneously or after picture. Before semantically related caused longer naming time than unrelated. Phonologically related didn't differ from unrelated. During and after phonological facillitation effect, faster naming times. Early stage of processing ONLY affected by meaning and late stage of processing ONLY affected by sound
      1. Evidence: Levelt et al (1991) name picture, lexical descion task to auditory presented IS. Named word sheep primed the semantic associate (goat) but not the phonologically related word goal. Therefore priming at strictly semantic level (lemma level) activation of goat not leaked through to phonological level
  2. Cascaded: word form selection can start before lemma selection is complete
    1. Levet used words in same category for semantic. Peterson & Savoy (1998) used stronger semantic relations-synonyms. claimed activation loses strength when it traverses through many levels)
      1. Name pictures, some trials written word presented visually on screen at variables SOA's and subject had to prepare written word. Written prime words either phonologically related to dominant form (count for coach) or secondary form (soda for sofa) or unrelated (track). Priming for target words related to dominant and secondary form ie even though couch dominant form speeded up by soda. Therefore argue that word form level is accessed before lemma has been selected. Also later priming for secondary form dimishes the system finally settles on a single word but not before phonological level is activated
        Most substitution errors similar in form OR meaning but some are mixed error (Dell & Reich (1981) Lobster instead of oyster both related in form and meaning. Hard to explain by discrete models error is phonologically related it must have happened during word form retrieval but discrete models don't allow feedback to lemma level so why are they semantically related too? In cascade model activation from wrong lemma can leak through to word form level
        Speech error likelt to occur in real word. Hard for discrete model to explain, suggests feedback from word form to lemma level
        However =Levelt 1989 proposes an output monitor to check output of speech production and correct errors. Use some parts of sentence comprehension process, the monitor might miss errors more often if they are real words (explains lexical bias)or similar to intended word, if both thats even harder that explains mixed error

Media attachments

garrett (image/jpg)

Next up

Producing words

Description

Resource summary

Media attachments

Similar

	Created by issy_hinds about 11 years ago