This lecture covers how a spoken sentence is produced from the formation of an idea in the speaker’s mind to the moment before it is articulated. We will discuss the processes involved and the methods by which these can be examined.
A.Why is sentence production interesting?
The storage space of the brain is finite. This means that it cannot store the infinite number of sentences that we may ever need to produce. From this it follows that we must somehow construct sentences from smaller parts or units before we are able to say them. The main issues then concern the processes by which units come to be selected and then combined in a particular order.
B.Processes of Speech Production (after Levelt 1989)
Figure 1 The Processes of Speech Production
The three main areas of speech production are:
The speaker must decide on the message to be conveyed. Very little is known about this stage. The end point is a stage at which the message itself has been decided but it has no linguistic form. It is also called the preverbal message or the message level of representation. This stage is often represented by a thought bubble.
The speaker must convert their message into a linguistic form. This stage involves
Lexicalisation – selecting the appropriate word
Syntactic planning – putting the words in the right order and adding grammatical elements.
3.Articulation / Execution
The speaker must plan the motor movements needed to convey the message.
C.Where does our evidence come from?
It’s hard to study speech production as it’s very difficult to get inside someone’s head as they plan a sentence.
II.Speech Errors (Slips of the tongue)
These are the types of errors that are relatively common in normal speech production. Errors are categorised by the mechanism and the unit involved in the error.
A unit is missed out from the intended target
1. The chimney catches fire The chimney catch fire (affix deletion)
13. I’ve read all my library books I’ve eaten all my library books (produced when the speaker was hungry)
14. Get out of the car Get out of the clark (produced when the speaker was looking at a shop called Clark’s)
We make lots of pauses while we speak. Sometimes these pauses are periods of silence (unfilled pauses) or they may contain repetitions or items such as ‘umm’ or ‘I mean’ (filled pauses).
A.Pauses before words
These pauses seem to be to do with retrieving individual words
They occur more frequently and are longer before words that are less predictable.
During such pauses people often make appropriate hand gestures that describe the word they are about to say.
Such pauses are sometimes described as a difficulty in microplanning
1.Tip-of-the-tongue state (TOT)
This state is an extreme version of a microplanning pause. The speaker knows they know what the word is (they have a ‘feeling of knowing’) and can provide semantic information about it but cannot remember the exact phonological form. Speakers may know some information about the phonological form (such as first sound or number of syllables) or produce interlopers (near phonological neighbours).
B.Pauses for sentence planning
These pauses seem to be to do with planning the syntactic and semantic content of speech.
There are fluent and hesitant phases of production.
There are more and longer pauses in the hesitant phases.
There are more of these pauses if the task is difficult or there is a high cognitive load.
These pauses are sometime described as difficulties in macroplanning.
When we speak we must put our words in a certain order and add grammatical elements to our utterance.
A.What evidence must models of syntactic planning account for?
a. The position of stress was unchanged – the primary stress and the nuclear accent remained on the final word. Suggests that prosody is generated independently of the words themselves.
b. The plural morpheme has stayed at the end of the utterance rather than moving with the stem ‘maniac’. We say the morpheme has been stranded. Suggesting that contentwords and function words are accessed and processed separately.
Another indication that content and function words are processed differently is that they show different patterns of exchange errors. Word exchanges aren’t constrained by distance whereas sound exchanges are not. Furthermore, words tend to exchange with others of the same syntactic class.
c. The plural morpheme is pronounced as /z/. This is appropriate for the phonological environment of ‘weekend’ rather than ‘maniac’. We say it shows phonological accommodation to the environment Suggests that the phonological form of function words is specified after that of content words.
B.Garrett’s model of syntactic planning
Garrett proposed a model of syntactic planning based primarily on data from speech errors like the ones above.
1.Main features of Garrett’s model
Processing is serial, that is to say that information can only flow one way.
There are two main stages, functional and positional.
Content and function words are selected at different stages.
Form abstract semantic specification and assign syntactic functions
Specify phonological forms of function words and affixes
// // // // //
2.How well does Garrett’s model account for the speech error evidence?
a)Different processing of content and function words
Content words are processed at the functional stage whereas function words are not selected until the positional stage
Words can exchange over large distances because they are retrieved before their position is established. Sounds are not specified until after the positional level which constrains the distance of their exchanges.
The tendency for words to exchange with others of the same class can be attributed to errors made when slotting the forms into the syntactic frame.
b)Late phonological specification of function words
The phonological form of function words is specified after that of content words.
c)Blends and Cognitive Intrusions
Because the model is serial and modular it can’t explain the existence of phrase blends such as 9 and cognitive intrusions such as 13 and 14.
Lexicalisation is the process of turning the semantic representation of words into the phonological specification. In Garrett’s model this isn’t really specified. We know that we retrieval the phonological forms of content words between the functional and positional levels but not exactly how this happens.
A.What speech evidence must a model of lexicalisation account for?
There are distinct types of substitution errors
10. Give me a spoon Give me a fork
Phonologically Related Substitutions (Malapropisms)
11.I think they are equivalent I think they are equivocal
12. Get me the catalogue Get me the calender
Mixed errors occur more often than would be predicted by chance
Speakers can have access to semantic information without having access to the phonological specification. I.e. they can make appropriate hand gestures during microplanning pauses and may find themselves in a tip-of-the-tongue state. How does this happen? Also, some small parts of phonological information may be available.
B.One-stage or two?
Do we go directly from the semantic representation to the phonological representation or is there an intervening level? The lemma representation has been posited as an intervening stage. A lemma is a representation containing syntactic and semantic but not phonological information.
Figure 2 One stage of lexicalisation or two?
1.How well does a two stage model explain the evidence?
Semantic substitutions come from selecting the wrong lemma, phonologically related substitutions occur when selecting the phonological representation
This model does not explain why mixed errors occur so frequently
This model doesn’t explain why word blends can occur
b)TOTs and Gestures during Hesitation
These can be explained if the lemma has been accessed but the phonological representation has not.
The model can’t explain how some but not all phonological information can be available to the speaker. This is due to the autonomous nature of the model.
VI.Modifications to models of syntactic planning and lexicalisation
Modular, autonomous models such as the ones above can explain many aspects of planning and lexicalisation. Some parts of evidence, however, support a more interactive approach where different levels of information interact and where information from outside the model can be used. Much recent work has concentrated on developing such interactive models. Such work may focus on modifying existing autonomous models or on designing completely new models.
Onset – The initial consonant or cluster of a syllable
Morpheme - The smallest unit of meaning
NuclearAccent – The final prominence giving pitch movement in an utterance
PrimaryStress – The main stress in an utterance
Prosody – Properties of duration, pitch, and loudness
ContentWord – The type of open-class words that convey most of the meaning of the utterance .
FunctionWord – The type of closed-class word that does the grammatical work of the language
References You should read the starred item and at least one of the additional chapters
*Harley, T. (1995) The Psychology of Language Cambridge CUP, 243-265. The chapter in the new edition of Harley (2001 349-376) is extensively revised especially in the treatment of lexicalisation. It’s fine to read whichever edition you can get from the library.
Fromkin, A. and Ratner, N. (1993) ‘Speech Production’ IN Gleason, J. and Ratner, N. (eds)Psycholinguistics, London: Harcourt Brace, Chapter 7.
Bock, J. and Levelt, W. (1994) ‘Language production:Grammatical encoding’. IN M.Gernsbacher (ed.), Handbook of psycholinguistics, San Diego:Academic Press, 945-984 Harris, M. and Coltheart, M.(1989) Language Processing in Children and Adults, London: Routledge, Chapter 8 Levelt, W. (1989.) Speaking: From intention to articulation Cambridge, MA: MIT Press
The first Garrett paper is also interesting but is not compulsory.
Garrett, M. (1975) The analysis of sentence production. IN G. Bower, (ed) Psychology of learning and motivation: Vol. 9. New York: Academic Press