ELL2 Speecherrors Harley 06
ELL2 Speecherrors Harley 06
ELL2 Speecherrors Harley 06
net/publication/304041208
CITATION READS
1 8,147
1 author:
Trevor A. Harley
University of Dundee
63 PUBLICATIONS 1,744 CITATIONS
SEE PROFILE
All content following this page was uploaded by Trevor A. Harley on 01 December 2017.
1.
2.
Department of Psychology
University of Dundee
Dundee
DD1 4HN UK
3.
Natural speech is full of mismatches between intention and output. Slips of the tongue are
errors involving the sounds or words of the language, and provide a window onto the
processes of speech production. Errors can be classified according to the units of speech
(e.g. phoneme, word or phrase) and the mechanisms (e.g. exchange, substitution,
1
occurs takes place in stages, with content words and function words being accessed at
4.
Main body (2000 words; is slightly more because I also cover TOTs)
Introduction. Natural speech is far from perfect: it is replete with filled and
unfilled hesitations and errors. A speech error is a mismatch between what we intend to
say and what we actually say. This entry focuses upon a mismatch involving the sounds
of the language, or slips of the tongue, and will not cover hesitations. Speech errors made
by non-brain damaged speakers resemble the errors made by speakers with brain damage,
and the study of both sorts of errors has revealed a great deal about how humans produce
language, and about the relation between language and the brain. This entry will focus on
The units involved in slips of the tongue range from phrases, words (e.g. saying
“pass the pepper” instead of “pass the salt”), morphemes, phonemes (e.g. saying “flock of
2
bats” instead of “block of flats”), to features (e.g. saying “turn the knop” instead of
“knob”). Mechanisms include anticipations (e.g. saying “the mirst of May” instead of
“the first”), perseverations (e.g. saying “God rest re merry gentleman” instead of “God
rest ye”), exchanges (e.g. “Guess whose mind came to name?” instead of “name came to
mind”), substitutions (“Get me a fork” instead of knife), and blends (saying “chung” for
Two particular types of speech error deserve special mention. Spoonerisms are
errors where the initial consonants of words have exchanged. They are named after the
who was particularly prone to making this type of error (and indeed, who may have
1. You have wasted the whole term -> You have tasted the whole worm.
2. The Lord is a loving shepherd to his flock -> The Lord is a shoving leopard to
his flock.
Note that in both of these examples the error strings produced form words; these
errors are therefore possible instances of the lexical bias effect described below.
Malapropisms are named after Mrs Malaprop, a character in Sheridan’s play “The
Rivals”. Mrs Malaprop often used grand words incorrectly, such as saying “epitaphs” for
3
“epithets” and “reprehend” for “apprehend”. Following Fay and Cutler (1977), the term
has been used for all phonological word substitutions that inadvertently arise from
Methodology of speech error collection. There are two main techniques for
collecting speech errors. The first technique involves recorders simply noting speech
errors that they hear in everyday life. This methodology has two advantages. First, it is
naturalistic. Second, it enables a comprehensive corpus of all types of error to be built up.
The disadvantage of this technique it is that is prone to bias. Observers may be more
likely to observe certain types of error (e.g. sound-level errors might be less noticeable
than word-level errors), and they may also be less likely to remember and record certain
types of error.
participants being required to name pairs of words as rapidly as possible. The participant
makes an error on a small proportion of errors (e.g. reading BARN DOOR as “darn
bore”). The advantages of this methodology is that it is easy to collect a relatively large
number of errors of the sort the experimenter is interested in (here spoonerisms), and that
it is possible to have some element of control over the errors made. The disadvantages are
4
that the technique is more intrusive, the errors involve some element of reading, and the
recorded speech suggests that observer bias effects are not large, although they obviously
remain a concern.
to say – we form the message. Formulation involves translating the message into a
linguistic form, and has two aspects: syntactic planning and lexicalisation (the process
whereby word concepts are turned into sounds). Finally, execution involves detailed
phonetic and articulatory planning. Speech errors have informed our understanding of all
of these stages.
It is important to note that the occurrence of slips of the tongue is not random. In
particular, some sorts of errors that could occur do not (e.g. we do not observe function
words exchanging with content words). This order is the basis for explaining how and
why errors arise in terms of a model of speech production. The basic idea is that for an
error to occur, the two elements of the error (intended and actual outcome) must be
simultaneously active at the same level of processing. For example, words exchange with
words, but content words only exchange with other content words, and function words
with other function words; content words do not exchange with function words, or vice
5
versa. This finding is extraordinarily robust: in my corpus of several thousand speech
errors, there is not a single instance of a content word exchanging with a function word.
This observation supports the idea that content and function words are from
There are also very different constraints on word and sound exchange errors.
Sounds only exchange across small distances, whereas words can exchange across
phrases; words that exchange tend to come from the same syntactic class, whereas sound
exchange errors are not constrained in this way, but instead swap with words regardless
Syntactic planning. The extension of this type of argument led to the development
The capital letters indicate the primary sentence stress and the italics secondary
stress. The error left the stress unchanged, suggesting that stress is generated
independently of the particular words involved. Note also that the plural morpheme “-s”
remained where it was originally intended to be: it did not move with “maniac”, but
instead was stranded. Furthermore, the plural morpheme was realized as /z/ not /s/, so
6
that the plural ending sounds consistent with the word that actually came before it, not
with the word that was originally intended to come before it. This error is an example of
In Garrett’s model there are two major stages of syntactic planning. At the earlier
functional level, word order is not explicitly represented. The semantic content of words
is specified and assigned to roles such as subject and object. At the positional level,
words are explicitly ordered. There is a dissociation between syntactic planning and
lexical retrieval. There is a further dissociation within lexical retrieval: Garrett argued
that content and function words play very different roles in language production, with
content words selected at the functional level, whereas function words are not selected
In this model, word exchanges occur at the functional level. As only the functional
roles of the words and not their absolute positions are specified at this level, word
exchanges are constrained by syntactic category but not by the distance between the
exchanging words. Next we generate a syntactic frame for the planned sentence. The
phonological representations of content words are then accessed from the lexicon using
the semantic representation. Content words are then inserted into the syntactic planning
frame where final positions are specified to form the positional level. The function words
and other grammatical elements are then phonologically specified to give the sound-level
representation. Sound exchanges occur at this stage and, as their absolute position is now
7
Garrett’s model accounts for a great deal of the speech error evidence, but a number
of subsequent observations suggested that some aspects of it might not be correct. First, it
is not at all clear that speech production is strictly serial. There is clearly some evidence
for at least local parallel processing in word blend errors, which must be explained by
two (or more) words being simultaneously retrieved from the lexicon, as in (5). More
problematically, we find blends of phrases and sentences (e.g. 6). Furthermore, the locus
of these blends is determined phonologically, so that the two phrases cross over where
they are most similar in sound, suggesting that the alternative messages are processed in
utterance intrudes into it. The message level can intrude into lower levels of processing,
producing non-plan-internal errors (e.g. 7). These errors are often phonologically
facilitated, meaning that errors are more likely to occur if the target word and intrusion
sound alike (Harley, 1984). The names of objects or words in the outside environment
can also intrude into speech, producing environmental contamination errors (e.g. 8). Once
often than one would expect by chance (although to a lesser degree than with other
cognitive intrusions).
8
6. Utterance: I’m making the kettle on. (Target 1: I’m making some tea. + Target 2:
7. Utterance: I’ve eaten all my library books. (Target: I’ve read all my library
books. Context: The speaker reported that he was hungry and was thinking of
8. Utterance: Get out of the clark. (Target: Get out of the car. Context: The speaker
was looking at a shop-front in the background that had the name “Clark’s”
When we produce a word, processing moves from the semantic level to an intermediate
level containing lemmas, which are specified syntactically and semantically. We then
retrieve the phonological forms of these words in a stage of phonological encoding. There
Fay and Cutler (1977) observed that there were two distinct types of whole word
substitution speech error: semantic substitutions, such as (9) and (10), and form-based
9
12. historical -> hysterical
Fay and Cutler argued that the existence of two types of word substitution suggests
that the processes of word production and comprehension use the same lexicon, but in
opposite directions. Items in the lexicon are arranged phonologically for spoken word
recognition, so that words that sound similar are close together. The lexicon is accessed in
production by traversing a semantic network. Semantic errors occur when traversing the
decision tree, and phonological errors occur when the final phonological form is selected.
Although the idea that lexicalisation occurs in two stages appears to be supported by
evidence from other sources such as picture naming, tip-of-the-tongue states, studies of
people with anomia, and brain imaging, there is nevertheless some dissent (Caramazza,
choose words? Models based primarily on speech error data see speech production
primarily as an interactive process involving feedback (e.g. Dell, 1986; Dell & Reich,
1981; Harley, 1984; Stemberger, 1985) because speech errors show evidence of multiple
phonemes. In particular, lexical bias is the tendency for sound-level speech errors such as
spoonerisms to result in words rather than nonwords (e.g. “barn door” being produced as
10
“darn bore”) more often than chance would predict. Although we would expect some
sound errors to form words sometimes by chance, Dell and Reich showed that this
happens far more often than is expected by chance. This interaction between lexical and
phonological processes has been shown both for naturally occurring speech errors (Dell,
1986; Dell & Reich, 1981) and in artificially induced spoonerisms (Baars et al., 1975).
Some aphasic speakers show lexical bias in their errors (Blanken, 1998).
Similarity effects arise when the error is more similar to the target than would be
expected by chance. In mixed substitution errors the intrusion is both semantically and
phonologically related to the target (e.g. 13 and 14). We find mixed errors far more often
than would be expected by chance alone (Dell & Reich, 1981; Harley, 1984; Shallice &
McGill, 1978). We also find similar results in artificially induced speech errors (e.g.
Baars et al., 1975; Motley & Baars, 1976) and in errors arising in complex naming tasks
Similarity effects are problematical for serial models such as Fay and Cutler’s. The
basic model must be modified in one of two ways. Butterworth (1983) proposed that a
filter or editor checks the output to see that it is plausible; it is less likely to detect an
error if the word output sounds like the target should have been, or is related in meaning.
11
Such a mechanism, although it might be related to comprehension processes, is not
considered to be parsimonious.
Dell (1986), Harley (1993), and Stemberger (1985) proposed different versions of the
are slotted into frames at each level of processing. Processing units specify the syntactic,
sentence level, where items are coded for syntactic properties, through a morphological
level, to a phonological level. At each level, the most highly activated item is inserted
into the currently active slot in the frame. The flow of activation throughout the network
is time-dependent, so that the first noun in a sentence is activated before the second noun.
Feedback between the phonological and lexical levels gives rise to lexical bias and
similarity constraints.
The model gives a good account of the origin of word and sound level errors.
Several units may be active at each level of representation at any one time. If there is
sufficient random noise, an item might be substituted for another one. As items are coded
12
for syntactic category and position, the other units that are active at any one time tend to
The tip-of-the-tongue state. The failure to produce at all is the ultimate speech
where the word can take some considerable time to be produced, if it is at all. In a TOT
state, you know you know the word, but cannot access the sounds.
Brown and McNeill (1966) were the first to examine the TOT state experimentally.
(15) defines “sextant”. Brown and McNeill found that a sizeable proportion of
people are placed in a TOT state by this task. Furthermore, in a TOT, partial information
about the target word, such as the number of syllables, the initial letter or sound, and the
stress pattern, can sometimes be retrieved. People also sometimes produce near
phonological neighbours like “secant”, “sextet”, and “sexton”; these other words that
lexicalization but failure of the second stage. In support of this idea, Vigliocco, Antonini,
13
and Garrett (1997) showed that grammatical gender can be preserved in tip-of-the-tongue
states in Italian. That is, even though speakers cannot retrieve the phonological form of a
word, they can retrieve some syntactic information about it. Two theories have been
proposed to explain why the second stage of lexical access fails in TOTs. Brown and
McNeill proposed a partial activation hypothesis, which states that the target items are
inaccessible because they are only weakly represented in the system. Burke, MacKay,
Worthley, and Wade (1991) provided evidence in favour of this model from experimental
and diary studies. They argued that the retrieval deficit involves weak links between the
Woodworth (1938), states that the target item is actively suppressed by a stronger
competitor. Jones and Langford (1987) used a variant of the Brown and McNeill task
neighbour of the target word and showed that the presence of this neighbour increases the
chance of a TOT state occurring, whereas presenting a semantic neighbour does not. They
interpreted this as showing that TOTs primarily arise as a result of competition. However,
the same pattern of results is found with these materials when the blockers are not
presented, suggesting that the original results depended on the precise materials used
Harley and Bown (1998) showed that TOTs are more likely to arise on low-
frequency words that have few close phonological neighbours. This finding fits a
partial activation model of the origin of TOTs rather than an interference model. Indeed,
14
phonological neighbours appear to play a supporting rather than a blocking role in lexical
access. So far then the TOT data best support the partial activation hypothesis.
The Freudian slip. Freud argued that slips of the tongue are “a window to the
mind” in that they provide insight into the hidden, repressed desires and opinions of
noted the occurrence of slips of the tongue, and proposed that they revealed our repressed
thoughts. In one of his examples, a professor said in a lecture, “In the case of female
(experiments) …”. Not all Freudian slips need arise from a repressed sexual thought. In
another example he gives, the President of the Lower House of the Austrian Parliament
opened a meeting with “Gentlemen, I take notice that a full quorum of members is
present and herewith declare the sitting closed!” (instead of open). Freud interpreted this
as revealing the President’s true thoughts, which were that he secretly wished a
potentially troublesome meeting closed. Such errors have come to be known as Freudian
slips. Ellis (1980) reanalyzed Freud’s collection of speech errors in terms of a modern
interpretation is that most (if not all) of the examples he provides are more simply
accounted for in terms of the simple cognitive mechanisms described above. There is no
15
true Freudian slips. It is certainly the case that thoughts can intrude into overt speech, as
16. Utterance: I’ve eaten all my library books. (Target: I’ve read all my library
books. Context: The speaker was hungry and thinking of getting something to eat.)
16
5.
INSERT
Baars, B. J., Motley, M. T., & MacKay, D. G. (1975). Output editing for lexical
321-360.
Brown, R., & McNeill, D. (1966). The “tip of the tongue” phenomenon. Journal of
Burke, D. M., MacKay, D. G., Worthley, J. S., & Wade, E. (1991). On the tip of the
tongue: What causes word finding failures in young and old adults? Journal
Cutler (Ed.), Slips of the tongue and language production (pp. 73-108).
Berlin: Mouton.
17
Caramazza, A. (1997). How many levels of processing are there in lexical access?
Caramazza, A., & Miozzo, M. (1997). The relation between syntactic and
speech error data. Journal of Verbal Learning and Verbal Behavior, 20,
611-629.
Fay, D., & Cutler, A. (1977). Malapropisms and the structure of the mental lexicon.
18
Garrett, M. F. (1975). The analysis of sentence production. In G. H. Bower (Ed.),
The psychology of learning and motivation (Vol. 9, pp. 133-177). New York:
Academic Press.
Science, 8, 191-219.
291-309.
Press.
Harley, T. A., & Bown, H.. (1998). What causes a tip-of-the-tongue state? Evidence
19
Martin, N., Weisberg, R. W., & Saffran, E. M. (1989). Variables influencing the
Motley, M. T., & Baars, B. J. (1976). Semantic bias effects on the outcomes of
Shallice, T., & McGill, J. (1978). The origins of mixed errors. In J. Requin (Ed.),
Vigliocco, G., Antonini, T., & Garrett, M. F. (1997). Grammatical gender is on the
20
6.
Keywords (10-15)
level
21
7.
Pauses in speaking
8.
See figure.
No multimedia annex.
9.
Copyright permissions
None needed.
22
10.
Brief biography
carried out his PhD at the University of Cambridge on speech errors, producing a highly
interactive model of speech production where top-down influences play an important role
at all stages. His current research interests include speech errors, connectionist modelling
23
!
24
Figure 1 Garrett’s model of speech production
25
!
26