Role of Feedback

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

The Role of

Feedback
in

Speech Production

by

Christina May

Paper submitted in partial fulfillment of the

requirements for a graded credit (8 ECTS points) for the course

“Psycholinguistics”

in Winter Term 2007/2008

Submission Date: 19 March 2008

Approved by: Prof. Dr. Jürgen Handke

Philipps-University Marburg
1 Introduction 2

Contents

1 Introduction .....................................................................................................3

2 The Production of Natural Speech................................................................. 4


2.1 Levelt’s Model of Speech Production ...............................................................4

2.2 Garrett’s Model of Sentence Processing ...........................................................5

3 System Errors – Slips of the Tongue.............................................................. 8


3.1 Garret’s Analysis of Speech Errors ................................................................... 8

3.1.1 Speech Error Patterns 8

3.1.2 Localization of Speech Errors 10

3.2 Levelt’s Categorization of Speech Errors........................................................ 11

4 Feedback Mechanisms in Speech Production............................................. 14


4.1 Feedback Loops...............................................................................................14

4.2 The Importance of Feedback Processes........................................................... 16

5 Conclusion ......................................................................................................18

6 References ...................................................................................................... 19

7 Confirmation of Authorship .........................................................................20


1 Introduction 3

1 Introduction

“The inner workings of a highly complex system are often revealed by the way in
which the system break down.”
Dell 1986

Speech errors have attracted linguists’ attention ever since: Slips of the tongue,
demonstrating the complexity of natural speech, provide an insight into the highly
complex system of natural language which has been studied extensively during
the past two centuries. However, most stages of language production remain
hidden.

On the base of different speech errors psycholinguists tempted to discover the


mechanisms underlying speech production. Using large corpora of spontaneous
speech samples scientist like Merril F. Garret or Wilhelm J. Levelt developed
models of language processing clarifying the compound procedures. Moreover it
was possible to determine mechanisms responsible for system errors or slips of
the tongue. Studying their characteristics linguists were looking for so called
monitors able to avoid those errors. Actually, the existence of those feedback
loops reducing the number of uttered speech errors down to 50 percent seems to
be proven today (POSTMA 2000: 2).

In this paper, various feedback mechanisms monitoring natural speech


production shall be depicted. Their function as well as the importance of the
monitoring features will attract special interest.

Illustrating different linguistic models, assumptions about the units involved


in natural speech production will be explained. In this context, attention will be
drawn on speech errors initiating different feedback mechanisms. Error categories
as well as their origin in the language production system shall be introduced.
Moreover, error detection via specialized feedback loops as well as its importance
for fluent speech acts will be demonstrated.
2 The Production of Natural Speech 4

2 The Production of Natural Speech

Although most speech comprehension processes were revealed, linguists only


know little about the speech act itself. Based on the analysis of spontaneous
speech and modern neurological examinations, various stages of speech
production could be defined.

To get an insight into those complex mechanisms two models of language


production shall be introduced.

2.1 Levelt’s Model of Speech Production

The speech act – a complex system “involving a number of cognitive, linguistic


and motoric processes” – starts with the construction of the message (POSTMA
2000: 98).

Fig. 1: A blueprint for the speaker; adapted from LEVELT 1989: 9


2 The Production of Natural Speech 5

The Conceptualizer defining the utterance’s content and communicative


intention produces a preverbal message which is transferred to the Formulator
(LEVELT 1989: 9).

The Formulator translating the preverbal message into a phonetic string


involves different processing levels: While encoding a message grammatically,
the utterances lemmas, i.e. phonologically unspecified lexical items, have to be
retrieved. The activation of a lemma’s grammatical and syntactic role initiates the
generation of a syntactic frame: lemmas are arranged according to their
grammatical and syntactic role. At this stage of sentence processing, the
Grammatical Encoder has produced the message’s surface structure (LEVELT
1983: 48).

Before the surface structure is ready to be uttered by the articulators, it has to


be encoded phonologically. The message’s lexical form, the information about an
item’s internal composition stored in the mental lexicon has to be retrieved. Apart
from the lemma’s phonological and morphological restrictions, information about
its stress is activated (LEVELT 1989: 12).

The output of the Phonological Encoder, the articulatory plan, can now be
passed to the Articulator. Since human articulators cannot produce different
phonetic plans which reaching the articulators simultaneously, some of the plans
have to be stored temporarily in an Articulatory Buffer (ibid. 12). Efferent signal
are sent to the articulators which produce audible or overt speech. Consisting of
phonetic strings overt speech is perceived by the interlocutors’ speech
comprehension system which allows them to react on the received message.

2.2 Garrett’s Model of Sentence Processing

Linguist Merril F. Garrett’s assumptions about the production of natural speech


are based on the analysis of spontaneous speech errors. Every utterance is
conceptualized in the so called Message Source. In addition to the message’s
content, its purpose and the knowledge about previously uttered messages
influence the Message Source’s output.
2 The Production of Natural Speech 6

This is transferred to the Production System. Passing different levels of


speech production the message is translated into efferent signals necessary for
articulatory processes. While the message’s lemmas have to be retrieved, a
functional frame builder is arranging clause type units in which the lemmas can be
arranged according to their functional syntactic roles (GARRET 1975: 176).
Subsequent to this functional level representation, a phonologically specified
string of the message is produced.

Fig. 2: Garret’s model of sentence processing (GARRETT 1975: 176)

Another subcomponent belonging to the production system is the positional


level of representation. A syntactic frame of the message has to be processed.
Furthermore, morphological phrases are produced in order to inflect the
utterance’s lemmas.

At the next stage – the sound level of representation – the message has to be
translated into a phonological string the articulators will be able to utter.
Furthermore the utterance will be marked referring its word stress and intonation.
2 The Production of Natural Speech 7

Now the message is ready to be uttered. Efferent neurons transfer signals to


the articulators, all part of the articulatory system. These will turn the abstract in
overt speech audible for interlocutors and the speaker himself (LEVELT 1981: 49).
3 System Errors – Slips of the Tongue 8

3 System Errors – Slips of the Tongue

Classifying slips of the tongue concerning their pattern as well as their


linguistic units, Merril F. Garret gained an insight into speech production
mechanisms. Based on the assumptions that

(1) “elements of a sentence interact in an error (e.g. exchange position), they


must be elements of the same hypothesized processing vocabulary” and
(2) “the structural constraints for a given error type must be of a single
processing type” (GARRETT 1980: 183),
he succeeded in localizing different stages of speech production responsible for
certain error patterns.

In the following chapter various categories of speech errors as well as the


levels causing these slips shall be depicted. Besides Garret’s analysis, Willem J.
M. Levelt’s considerations about speech errors will be illustrated.

3.1 Garret’s Analysis of Speech Errors

Based on the analysis of Victoria Fromkin and other linguists, Garrett developed
the following categorization of speech errors. This work laid the foundations for
his consideration about speech production like explained above.

3.1.1 Speech Error Patterns

Consulting different dictionaries, one might find the following definitions of the
term error: “something produced by mistake” but also “an act involving an
(un)intended deviation from truth or accuracy” are listed in connection with this
term [INT 1]. Those actions can be incorrect concerning some external criterion
reflecting “a generally agreed upon idea of well-formedness, successful
performance, and/or error-free output” (POSTMA 2000: 100). Transferring this
concept to speech production, these external criteria are “based upon the linguistic
rules” (ibid. 101).
3 System Errors – Slips of the Tongue 9

Categorized into different patterns, speech errors have examples in nearly


every linguistic type, i.e. sound, morpheme, word etc. Analysing the MIT corpora,
Merril F. Garrett found the following categories:

a) Additions

An extra unit inserted in an already complete utterance is categorized as addition.


The redundant unit causes content, syntactic, as well as grammar errors (GARRETT
1975: 138):

(1) “I don’t see any many paddocks around here.”


intended: either any or many
(2) “Positively and negatively remarked”
intended: positively and negatively marked” (ROELOFS 1998: 666)

b) Deletion

As deletion Garrett defines the omission of an utterance’s unit. Like the addition,
this error pattern can affect every kind of linguistic unit (GARRETT 1975: 138):

(3) “I’ll just get up and muttered _intelligibly.”


intended: unintelligibly

c) Substitution

The replacement of an utterance’s unit by another linguistic unit is defined as a so


called substitution.

(4) “At low speed it’s too light.”


intended: too heavy (ibid. 138)

d) Shifts

Linguistic units like sounds or morphemes are slipt up form a word and added to
another lemma in the message.

(5) “She’ll be ready in case she decide_ to hits this.”


intended: she decides to hit this (ibid. 138)
3 System Errors – Slips of the Tongue 10

e) Exchanges

If units in an utterance change positions this error is called exchange. If single


sound units exchange, the well-formedness at the sound level is preserved (ibid.
1980: 184).

(6) a. “Fancy getting your model renosed.”


intended: nose remodelled (ibid. 1975: 138)
b. “I have to fill up the gas with car.”
intended: car with gas (ibid. 155)

f) Fusion

Two (semantically related) units are blended. In most of the cases the units
affected by the fusion can be recovered (ibid. 1980: 185f.).

(7) “At the end of todays lection…”


intended: lecture or lesson

3.1.2 Localization of Speech Errors

Depending on the lexical unit affected as well as the type of the error, Garret
defines different loci producing the slip. So he seems to be able to define the
functional level of representation as origin for substitution, word or phrase
exchanges, as well as fusions. The fact that all of the incorrectly chosen or
blended lemmas take the right syntactic roles and undergo all of the necessary
morphological and phonetic processes supports Garrett’s assumption (ibid. 176).
Furthermore this stage of speech production seems to be responsible for deletion
of whole words. Trying to select their lemmas in the mental lexicon, the word
cannot be retrieved and therefore do not occur in the uttered message (ibid. 1980:
183f.).

In contrast to slips referring to whole words, errors affecting single sounds or


morphemes tend to occur on the positional level. On this stage of production,
sound elements as well as morphemes defining the lemmas syntactic and
grammatical roles are selected and inserted in the syntactic frame. In some
occasions, the speaker might choose the wrong positional slot and produce shifts
as well as form or sound exchanges. Moreover the positional level is producing
3 System Errors – Slips of the Tongue 11

word shifts if the right lemma has been picked before on the functional level (ibid.
199).

More severe sound errors like simple or complex sound deletions are caused
on the sound level of representation. Furthermore Garrett tends to define this stage
as origin for word or sentence stress errors (ibid.1975: 176).

Another type of slip, Garrett included into his model, are so called tongue
twisters. These errors occur, if wrong signals are sent to the articulators or if they
just fail to pronounce an utterance because of its complex (phonetic) structure e.g.
(ibid. 176).

3.2 Levelt’s Categorization of Speech Errors

On the basis of Merril Garrett’s analysis of speech errors, Willem Levelt


attempted to define different types of errors a speaker is able to monitor.
Categorizing seven groups of error, Levelt affirms the Cutler’s thesis about he
possibility of monitoring every little piece of speech. However the following
categories should be sufficient for the following considerations of monitoring
processes:

a) Conceptual errors

Conceptual errors originate in the first stage of speech production, the


Conceptualizer. They concern the message’s content or intention like it can be
seen in the following examples:

(8) a. “Tell me, uh what – d’you need a hot sauce?”


b. “We start in the middle with – in the middle of the paper with a blue
disk.”

While in the first example the error affected utterance’s intention, the speaker of
sample two rethought his message with respect of the content. Although there is
some sign of hesitation, the speakers do not show any over sign of repair (Levelt
1989: 460).
3 System Errors – Slips of the Tongue 12

b) Slips concerning social standards

This error category is summing up slips referring to the register used to express
the speaker’s mind. Impolite or informal formulations in an official talk (e.g. an
job interview) belong to this group of errors. Defective conceptualizing processes
seem to e responsible for those slips.

c) Lexical errors

Lexical errors, caused by incorrect lemma selection during grammatical encoding,


are probably the most frequent caught slips of the tongue. Substitutions,
exchanges or blending effects belong to this error category.

(9) a. “Left to pink – er, straight to pink…”


b. “Well, let me right it back – er, down, so that…”

d) Syntactic and morphological slips

Besides lexical errors, other slips are due to incorrect grammatical encoding.
Inserting lemmas into wrong slots causes deviant syntax. The retrieval of
erroneous morphemes might lead to morphological slips like it can be seen in the
examples below (ibid. 462):

(10) a. “What things are this kid – er, is this kid going to say incorrectly?”
b. “Why it is – why is that nobody makes a decent toilet seat?”

e) Sound errors

Another category classified by Levelt is that of the so called sound errors. The
phonetic plan processed during phonological encoding can contain segmental or
supra-segmental sound shift, deletion, additions and many other patterns (cf.
GARRETT 1975: 138):

(11) a. “A unut – unit from the yellow dot.”


b. “… from my prOsodic – prosOdic colleagues…” (LEVELT 1989: 462)

Comparing these assumptions to Garrett’s error analysis, there are no striking


differences. In contrast to Garrett, dividing the slips into different patterns, Levelt
3 System Errors – Slips of the Tongue 13

concentrates on the different linguistic units which can be affected by errors.


However, both categorizations showing up many parallels will serve as base for
the definition as well as localization of different feedback loops in the following
chapter.
4 Feedback Mechanisms in Speech Production 14

4 Feedback Mechanisms in Speech


Production

Also if we just know little about the mechanisms involved in the production of
natural speech, there seems to be no doubt about the existence of some monitoring
systems. These feedback processes actuate so called editing mechanisms
correcting inaccurate utterances. The amount of speech errors can so be reduced
to a total number of about 50 percent (POSTMA 2000: 2).

In this chapter the main functions and types of feedback mechanisms shall be
illustrated. Furthermore the stages of speech production profiting from monitoring
mechanisms will be explored. In addition, the important role of feedback loops in
speech production shall be depicted.

4.1 Feedback Loops

Illustrating the different types of feedback which monitor natural speech


production, it might be advisable to follow the different levels of sentence
procession.

The first stage of the speech act and therefore the first possible error source is
the level of conceptualizing. A so called conceptual loop can monitor conceptual
as well as appropriateness errors originating on this stage (ibid. 102). Detecting an
error, the speaker has to rethink his message what might take some time. In most
of the cases, the speech act has to be interrupted what might cause some
hesitations or even pauses (ibid. 104).

Following Garrett’s model of sentence processing, the output of the functional


level of representation can be monitored with respect to its lemma selection.
Lemmas deviating from the utterances concept are substituted by new units
matching the utterances intention and content (ibid. 104).
4 Feedback Mechanisms in Speech Production 15

Fig. 3: Feedback Mechanisms in Speech Production; adapted from POSTMA 2000: 99

This feedback loop can be localized in Levelt’s model as well. The lexicality
monitor is situated in the formulator, to be precise on the stage of lexical selection
(ibid. 99). In this way exchanges, substitutions, blends, and deletions can be
detected and immediately edited without even uttering the erroneous message.

The next monitor the preverbal message has to pass is the syntactic
construction loop. This feedback loop whose position is assumed to be situated on
the positional and syntactic frame generation level (LEVELT 1989: 275)
respectively is monitoring the lemmas’ as well as the single morphemes’ positions
in the message (POSTMA 2000: 107). All kind of syntactic deadlock as well as
shift of words, morphemes, or single sounds can be prevented.
4 Feedback Mechanisms in Speech Production 16

Now the message is ready to be translated into a phonological string, the so


called phonetic plan or parsed speech. Before this plan is transferred to the
articulator, the speaker’s comprehension system is able to parse the internal
speech. Detecting a slip the speaker has to rethink his message and conceptualize
a new, correct message. In most cases, these editing processes are done – without
even being noticed – within about 150-200 ms. Since the articulators need about
250 ms to utter a sentence it leaves enough time to correct the erroneous message
before it is accessible for other interlocutors (LEVELT 1989: 469).

Once an incorrect message should have been uttered, auditory feedback is


able to detect the slip. Using the speech comprehension system, self-produced
speech as well as messages uttered by an interlocutor can be monitored by the
speaker’s conceptual loop. Correcting the unit affected by the slip, the edited
message can be uttered again (POSTMA 2000: 101). Levelt, acting on the
assumption that – apart from the speech comprehension system – no additional
monitoring mechanisms are involved in the detection of slips of the tongue,
established the perceptual loop theory as his hypothesis is called (LEVELT 1989:
469).

Apart from auditory feedback – part of the external monitoring mechanisms –


tactile feedback can monitor overt speech. Similar to error detection via efferent
or muscle spindle feedback loops tactile receptors inform the speaker about
obstacles constraining the articulators’ movements. Receiving fed-back signals,
the speaker is able to ‘calibrate’ his articulators without interrupting the speech act
(POSTMA 2000: 110f.).

4.2 The Importance of Feedback Processes

Considering the variety of complex feedback loops involved in speech production


mechanisms one might ask for their relevance in the natural speech act.

Probably the most obvious purpose of monitoring acts is to improve the


quality of a speaker’s utterances. Detecting a huge number of errors and initiating
special editing mechanisms, feedback loops can reduce the amount of erroneous
4 Feedback Mechanisms in Speech Production 17

articulations down to a number of at least 50 percent (POSTMA 2000: 98). This can
improve the interlocutors’ comprehension what might benefit mutual
understanding.

Moreover the fluency of speech might profit from the different feedback
mechanisms: detecting and correcting an error before it is even uttered, the
speaker does not have to interrupt his stream of speech. Since most feedback
loops are activated unconsciously, they do not constrain the speaker in his
conversational intentions at all. The speech act remains homogenous, an
apparently effortless activity (INDEFREY 2007: 547).

Thinking about tactile or efferent monitors, feedback enables the speaker to


continue the stream of speech in disadvantageous external conditions. Dental
plates, food, or other things blocking the normal speech act can easily be
compensated (BORDEN 2003: 127).

The most fascinating effect feedback mechanisms might cause is their role in
speech control. Feedback deprivation can cause high damage in the speech control
system like as in the case of deafness. While deaf-born people show extreme
deficits in language production, people suffering deafness as a consequence of
diseases or accidents do not seem to be limited in their quality of speech
production. Therefore it seems that auditory feedback holds a tuning function, i.e.
it calibrates the articulators to improve their movements. This training seems to be
essential for the whole process of speech articulation (GARMAN 1990: 225).

These examples, exemplifying the important role of feedback mechanisms in


speech processing activities, give a short insight into this complex and fascinating
topic. Since it would go beyond the scope of this paper, no more detail about
monitors as well as editing mechanisms will be added at this point. To learn more
about the importance of feedback, the interested reader can consult POSTMA’S
declarations.
5 Conclusion 18

5 Conclusion

In this paper two facts should have become apparent: Although speaking
seems to be “a homogenous and largely effortless activity” (INDEFREY 2007: 547),
the production of speech is a highly complex system vulnerable to a number of
different errors. These slips can affect all kinds of linguistic units and be of
numerous patterns.

Anyhow, there seems to be a kind of detection mechanism, a monitor,


controlling the different stages of speech production. Due to these feedback loops
slips can easily be identified. Once detected, language production mechanisms
initiate editing units reducing the immense number of uttered faults.

To cut it short, feedback mechanisms enable fluent and in most cases correct
speech. Moreover they seem to be responsible for proper articulation like
demonstrated on the basis of deaf people’ speech skills. Hence these loops have
immense influence on our speaking acts. Although their work is as sufficient we
not even notice their existence when producing speech. They are one of the many
little things which make the fascinating process of human speech and therefore the
expression of our deepest feelings and thoughts possible.
6 References 19

6 References

Borden, Gloria J. et al. 2003. Speech Science Primer: physiology, acoustics and
perception of speech. Philadelphia: Lippincott Williams & Wilkins.

Dell, Gary S. 1986. “A spreading-activation theory of retrieval in sentence production.”


In Psychological Review. 93.

Garman, Michael. 1990. Psycholinguistics. Cambridge: University Press.

Garrett, Merril F. 1975. The Analysis of Sentence Production. In Bower, Gordon H. The
Psychology of Learning and Motivation. Stanford: Academic Press. pp. 133- 177.

Garrett, Merril F. 1980. Sentence Processing. In Butterworth, B. language Production,


Volume 1: Speech and Talk. Cambridge: Academic Press. pp. 178-220.

Indefrey, Peter. 2007. Brain-imaging studies of language production. In Gaskell, M.


Gareth. The Oxford Handbook of Psycholinguistics. pp. 545-574.

Levelt, Willem J. M. 1983. “Monitoring and Self-Repair.” In Cognition. 14. pp. 41-104.

Levelt, Willem J. M. 1989. Speaking. From Intention to Articulation. Massachusetts:


MIT Press.

Postma, Albert. 2000. “Detection of errors during speech production: a review of speech
monitoring models.” In Cognition. 77. pp. 97-130.

Roelofs, Ardi. 1998. Syllabification in Speech Production: Evaluation of WEAVER. In


Altmann, Gerry T. M. 1998. Cognitive Models of Speech Production. York:
Psychology Press. pp. 657-694.

Internet Sources

[INT1] Merriam-Webster’s Online Dictionary: http://www.merriam-


webster.com/dictionary/error, accessed 13 March 2008
7 Confirmation of Authorship 20

7 Confirmation of Authorship

The Role of Feedback in Speech Production

I hereby formally declare that the work submitted is entirely my own and
does not involve any additional human assistance. I also confirm that it has not
been submitted for credit before, neither as a whole nor in part and neither by
myself nor by any other person. All quotations and paraphrases but also
information and ideas that have been taken from sources used are cited
appropriately with the corresponding bibliographical references provided. The
same is true of all drawings, sketches, pictures and the like that appear in the text,
as well as of all Internet resources used.

Violation of these terms will result in failure of the seminar (PS, SE, etc.) and
no credits will be awarded. I am aware that plagiarism is serious academic
misconduct which can lead to further sanctions on reoccurrence.

You might also like