The efgect of word frequency and position-in-utterance in Mandarin - - PowerPoint PPT Presentation

▶

the efgect of word frequency and position in utterance in

The efgect of word frequency and position-in-utterance in Mandarin - - PowerPoint PPT Presentation

Dec 28, 2023 369 likes •656 views

. . . . . . . . . . . . . . The efgect of word frequency and position-in-utterance in Mandarin speech errors A connectionist model of speech production I-Ping WAN and Marc ALLASSONNIERE-TANG Graduate Institute of Linguistics, NCCU

slide-1

SLIDE 1

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

The efgect of word frequency and position-in-utterance in Mandarin speech errors

A connectionist model of speech production I-Ping WAN and Marc ALLASSONNIERE-TANG

Graduate Institute of Linguistics, NCCU Lab Dynamics of Language, U-Lyon2

CLSW 2020

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 1 / 11 CLSW 2020

slide-2

SLIDE 2

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Research questions

How speech processing happens (connectionist view)

When I want to say ‘cat’ (Dell et al, 1999)

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 2 / 11 CLSW 2020

slide-3

SLIDE 3

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Research questions

How to test this model?

Speech errors do not occur arbitrarily e.g., I warn you » I want you Efgects from semantics and phonology Words that are used frequently are easily activated PRED 1: lower freq » higher probability of speech error Given-before-new principle PRED 2: sentence fjnal » higher probability of speech error

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 3 / 11 CLSW 2020

slide-4

SLIDE 4

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Research questions

How to test this model?

Speech errors do not occur arbitrarily e.g., I warn you » I want you Efgects from semantics and phonology Words that are used frequently are easily activated PRED 1: lower freq » higher probability of speech error Given-before-new principle PRED 2: sentence fjnal » higher probability of speech error

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 3 / 11 CLSW 2020

slide-5

SLIDE 5

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Research questions

How to test this model?

Speech errors do not occur arbitrarily e.g., I warn you » I want you Efgects from semantics and phonology Words that are used frequently are easily activated PRED 1: lower freq » higher probability of speech error Given-before-new principle PRED 2: sentence fjnal » higher probability of speech error

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 3 / 11 CLSW 2020

slide-6

SLIDE 6

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Research questions

How to test this model?

Speech errors do not occur arbitrarily e.g., I warn you » I want you Efgects from semantics and phonology Words that are used frequently are easily activated PRED 1: lower freq » higher probability of speech error Given-before-new principle PRED 2: sentence fjnal » higher probability of speech error

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 3 / 11 CLSW 2020

slide-7

SLIDE 7

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Research questions

How to test this model?

Speech errors do not occur arbitrarily e.g., I warn you » I want you Efgects from semantics and phonology Words that are used frequently are easily activated PRED 1: lower freq » higher probability of speech error Given-before-new principle PRED 2: sentence fjnal » higher probability of speech error

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 3 / 11 CLSW 2020

slide-8

SLIDE 8

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Research questions

How to test this model?

Speech errors do not occur arbitrarily e.g., I warn you » I want you Efgects from semantics and phonology Words that are used frequently are easily activated PRED 1: lower freq » higher probability of speech error Given-before-new principle PRED 2: sentence fjnal » higher probability of speech error

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 3 / 11 CLSW 2020

slide-9

SLIDE 9

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Research questions

How to test this model?

Speech errors do not occur arbitrarily e.g., I warn you » I want you Efgects from semantics and phonology Words that are used frequently are easily activated PRED 1: lower freq » higher probability of speech error Given-before-new principle PRED 2: sentence fjnal » higher probability of speech error

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 3 / 11 CLSW 2020

slide-10

SLIDE 10

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Materials

How to test those predictions?

Database on speech errors

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 4 / 11 CLSW 2020

slide-11

SLIDE 11

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Materials

How to test those predictions?

Database on speech errors

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 4 / 11 CLSW 2020

slide-12

SLIDE 12

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Materials

How to test those predictions?

Database on speech errors

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 4 / 11 CLSW 2020

slide-13

SLIDE 13

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Materials

How to test those predictions?

382 speech errors extracted with sentence (3022 words) Annotated with word frequency and position-in-utterance Word frequency from the Academia Sinica corpus Position-in-utterance = normalized distance to sentence start

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 5 / 11 CLSW 2020

slide-14

SLIDE 14

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Materials

How to test those predictions?

382 speech errors extracted with sentence (3022 words) Annotated with word frequency and position-in-utterance Word frequency from the Academia Sinica corpus Position-in-utterance = normalized distance to sentence start

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 5 / 11 CLSW 2020

slide-15

SLIDE 15

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Materials

How to test those predictions?

382 speech errors extracted with sentence (3022 words) Annotated with word frequency and position-in-utterance Word frequency from the Academia Sinica corpus Position-in-utterance = normalized distance to sentence start

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 5 / 11 CLSW 2020

slide-16

SLIDE 16

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Materials

How to test those predictions?

382 speech errors extracted with sentence (3022 words) Annotated with word frequency and position-in-utterance Word frequency from the Academia Sinica corpus Position-in-utterance = normalized distance to sentence start

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 5 / 11 CLSW 2020

slide-17

SLIDE 17

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Materials

Overview of data

PCA visualization

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 6 / 11 CLSW 2020

slide-18

SLIDE 18

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Analysis

Generalized Linear Mixed Models

Higher frequency » less likely to be a speech error Far from sentence start » more likely to be a speech error Supports the two predictions

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 7 / 11 CLSW 2020

slide-19

SLIDE 19

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Analysis

Generalized Linear Mixed Models

Higher frequency » less likely to be a speech error Far from sentence start » more likely to be a speech error Supports the two predictions

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 7 / 11 CLSW 2020

slide-20

SLIDE 20

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Analysis

Generalized Linear Mixed Models

Higher frequency » less likely to be a speech error Far from sentence start » more likely to be a speech error Supports the two predictions

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 7 / 11 CLSW 2020

slide-21

SLIDE 21

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Analysis

Binary recursive partitioning tree

Higher frequency » less likely to be a speech error Far from sentence start » more likely to be a speech error Supports the two predictions

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 8 / 11 CLSW 2020

slide-22

SLIDE 22

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Analysis

Binary recursive partitioning tree

Accuracy: The question is complex since errors are rare Better than with only frequency or position-in-utterance (5% up) Precision and recall: Need better identifjcation of speech errors

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 9 / 11 CLSW 2020

slide-23

SLIDE 23

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Analysis

Binary recursive partitioning tree

Accuracy: The question is complex since errors are rare Better than with only frequency or position-in-utterance (5% up) Precision and recall: Need better identifjcation of speech errors

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 9 / 11 CLSW 2020

slide-24

SLIDE 24

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Analysis

Binary recursive partitioning tree

Accuracy: The question is complex since errors are rare Better than with only frequency or position-in-utterance (5% up) Precision and recall: Need better identifjcation of speech errors

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 9 / 11 CLSW 2020

slide-25

SLIDE 25

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Analysis

Take-away message

Word frequency and position-in-utterance help to predict speech errors But more variables are needed

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 10 / 11 CLSW 2020

slide-26

SLIDE 26

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Analysis

Take-away message

Word frequency and position-in-utterance help to predict speech errors But more variables are needed

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 10 / 11 CLSW 2020

slide-27

SLIDE 27

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Analysis

Take-away message

Word frequency and position-in-utterance help to predict speech errors But more variables are needed Dell et al (1999)

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 10 / 11 CLSW 2020

slide-28

SLIDE 28

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Comments or questions?

Special thanks to The organizers of the CLSW 2020 政治大學語音暨心理語言綜合實驗室資工顧問李謙吾博士師範大學甯俐馨教授實驗室助理：曾子伊, 林志安, 李哲柔, 許惟翔, 楊璧綺

I-P, Wan and M. A-Tang (NCCU,DDL) The efgect of word frequency and position-in-utterance in Mandarin speech errors 11 / 11 CLSW 2020