REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

16
REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS Sieb Nooteboom UiL OTS, Utrecht University 1

description

REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS. Sieb Nooteboom UiL OTS, Utrecht University. PHRASAL LEXICAL ITEMS (PLIs): proverbs, sayings, idioms, clichés, collocations Levelt & Meyer 2000, PLI in mental lexicon: SUPERLEMMA - PowerPoint PPT Presentation

Transcript of REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

Page 1: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

1

REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL

ITEMS

Sieb NooteboomUiL OTS, Utrecht University

Page 2: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

2

PHRASAL LEXICAL ITEMS (PLIs): proverbs, sayings, idioms, clichés, collocations

Levelt & Meyer 2000, PLI in mental lexicon:

SUPERLEMMA

lemma lemma lemma lemma lemma

SET A THIEF TO CATCH A THIEF

set a thief to catch a thief

Page 3: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

3

1) Production more automatic for PLIs than for NPs.

2) Production less closely monitored for PLIs than for NPs

PREDICTIONS:

1) In PLIs less speech errors than in NPs.

2) Speech errors less often detected and repaired in PLIs than in NPs.

Page 4: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

4

CORPUS:Utrecht corpus of speech errors:2455 speech errors in spontaneous Dutch

Here we focus on:Syntagmatic phonological speech errors

with source and target phoneme in different words,

excluding speech errors in foreign words

excluding combinations of given and family name

excludingcombinations of function words

This leaves us with635 speech errors

Page 5: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

5

Each speech error >> minimal intended error-less form, including source and target.

English example (Fromkin 1973)

Speech error: *it was such a good force because we had informants

changed intocourse because we had informants

Reason: errors varied widely in the amount of material preceding and / or following the minimal string.

Page 6: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

6

All 635 word strings presented to 3 judges. Each word string assigned to one of these categories: 1 typical (part of) PLI 2 perhaps (part of) PLI?3 typical (part of) NP

Page 7: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

7

635 speech errors:386 cases: 3 judges agreed215 cases: 2 judges agreed34 cases: 3 different judgements.

These 34 were assigned code 2, reflecting uncertainty as to lexical status.

This resulted in the following distribution:1. PLI: 92 word strings2. PLI?: 75 word strings3. NP: 468 word strings________________________________________________________________________________________

Sum total: 635 word strings

Page 8: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

8

Search Engine Count Estimates (SECEs) were assessed for all word strings, using Yahoo constrained to Dutch and to precise symbol sequences by using double quotes.

The SECEs were transformed by (1) changing all zeros to ones, and (2) by taking the 10log.

Page 9: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

9

0

1

2

3

4

5

PLI PLI? NP

aver

age

log

freq

uenc

y

categoryUnivariate one-way ANOVA:

Significant effect of category on 10log frequency (p<0.001)Post hoc analysis (Tukey’s):

Significant difference between PLI and NP (p<0.05), and between PLI? and NP (p<0.05).No significant difference between PLI and PLI? (p>0.05).PLI and PLI? Collapsed in further analysis.

Page 10: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

10

Of each of 635 speech errors we now have the following information:

a) estimated lexicality (NP vs PLI)b) speech error class (anticipation vs perseveration vs

exchange)c) repair status (repaired vs unrepaired)

This was the input for further analysis.

Page 11: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

11

Strange data: Exchanges have two chances of being detected and repaired, against only one chance for anticipations and perseverations. Why, then, is fraction repaired lower and not higher for exchanges?

0

0.2

0.4

0.6

0.8

antic persev exchan

frac

tion

repa

ired NP

PLI

Page 12: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

12

How can we explain these strange data?

1) Repaired anticipations in the corpus often of the form heft..left hemisphere.

2) In inner speech this might have been heft hemisphere, but also heft lemisphere.3) An anticipation in inner speech has only a single

erroneous form, an exchange has two erroneous forms that may trigger detection and repair. 4) Probably 2/3 of repaired anticipations in corpus really derive from half-way repaired exchanges.

Page 13: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

13

No way of knowing whether a particular repaired anticipation stems from an anticipation or an exchange.

Therefore anticipations and exchanges collapsed in further analysis as anticipations+ in further analysis.

Page 14: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

14

0

0.2

0.4

0.6

0.8

antic+ persev

frac

tion

repa

ired NP

PLI

Logistic regression:Fraction repaired higher for antic+ than for persev (p<0.003).

This can be explained by assuming a considerable contribution of repaired exchanges to antic+.Fraction repaired higher for NP than for PLI.

This confirms the main hypothesis of this paper.

Page 15: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

15

CONCLUSION

Speech errors are less often detected and repaired in Phrasal Lexical Items than in Novel Phrases.

This supports a proposal by Levelt & Meyer:

A Phrasal Lexical Item mentally represented as a Superlemma from which the constituting Sublemmas are activated in one go, making mental preparation more automatic for Phrasal Lexical Items than for Novel Phrases.

Page 16: REPAIRING PHONOLOGICAL SPEECH ERRORS IN NOVEL PHRASES AND PHRASAL LEXICAL ITEMS

16

effects coef. s.e. t p

intercept=grand mean 0.594 0.024 24.01 <0.001

novel/lexic+: +/-0.129 0.055 -2.33 < 0.020

antic+/persev: +/-0.151 0.049 -3.05 < 0.003

interaction: 0.020 0.110 0.18 < 0.860

Table 1. Estimated parameters for the best fitting binomial logistic regression model of fraction repaired using effect coding. The grand mean was used as 'intercept'. For fixed effects, regression coefficients are given, with standard errors, t values and p values. Due to the structure of the data set with no repetitions for speakers or for word sequences, there are no random effects.