LING314: Presentation Naomi Ogasawara November 4, 2004 naomi703@email.arizona.edu

Processing Missing Vowels:

Effects of Phonological and Phonotactic Knowledge in Japanese

• People have a good idea about their native language (consciously/unconsciously).

• They use knowledge of their native language when they talk (produce) / hear (perceive) utterances.

Broad Q: How does the language-specific knowledge affect production or perception of

a certain sound?

Narrow Q: How does the language-specific knowledge affect perception of voiceless

vowels in Japanese? 0. BACKGROUND: JAPANESE (JPN) PHONOLOGY/PHONOTACTICS

Terminology phonology: the study of how speech sounds are organized. phonotactics: permissible combinations of phonemes.

(1) Five vowels: [i, e, ɑ, o, ɯ,] Table 1: ENG & JPN Vowels ENG JPN F1 F2 F1 F2

i 403 2917 518 2686 e 700 2500 750 2239 a 1114 1528 965 1362 o 436 915 585 1131 u 551 1528 568 1379

1

(2) High vowels [i, ɯ] are voiceless in certain consonantal environments. • In what environments?

A: between [-voi] Cs kita ‘north’ [kji tɑ], futon [ɸɯtoɴ:], yakisoba ‘fried noodls’ [jɑkjisobɑ]

B: between a [-voi] C and a pause: okashi ‘snacks’ [okɑɕi], dasu ‘to take out’ [dɑsɯ]

(3) What is a ‘voiceless vowel’?

• devoiced vowels: no voicing, no clear traces of F1 & F2. Sounds more fricative-like.

Figure 1: Spectrograms of /hokito/ Listen

5000

500

•

500

h o kj i t o

Time (s)0 0.549252

0

0

00

deleted vowels: merged with the preceding voicel

Figure 2: Spectrograms of /mešite/ Listen

00

5000

Time (s)0 0.476213

0

0

m e e ɕ t

2

Time (s)0.54551

ess consonant, difficult to separate.

m e ɕ i t e

Time (s)0.451837

(4) Why does vowel devoicing occur? • Vowel duration: high vowels are shorter than non-high vowels

/u/ = 1.00, /i/ = 1.17, /o/ = 1.26, /e/ = 1.37, and /a/ = 1.44 (Han 1962a) Table 2: Length of Vowels

/paka/ /pika/ /puka/ /peka/ /poka/ ENG 106ms 101ms 112ms 106ms 108ms JPN 81ms 57ms 59ms 77ms 58ms

• Speech tempo: vowel devoicing occurs at normal speech tempo

(not at super-careful speech) (5) A phonotactic constraint of Japanese (inhibition of consonant clusters)

• Basically CVCV… syllables • No consonant clusters are allowed except coda nasals and geminates.

☺ /kɑt.tɑ./ katta ‘bought’ /tom.bo./ tombo ‘dragon fly’ /kɑt/ /tob.mo./

• Loanwords ☺ McDonald’s [mɑ.kɯ.do.nɑ.ſɯ.do.]

I. EXPERIMENT 1 (6) Question Is voiceless [i ] detected as quickly and as accurately as fully voiced [i] by native

Japanese listeners? (7) Conditions

A. 3 consonantal environments [-V] (devoicing): /i/ is between two voiceless obstruents. voiceless [i] is expected. [+V] (voicing): /i/ follows /ǰ/ voiced [i] is expected. [+Vnas] (voicing): /i/ follows a nasal voiced [i] is expected.

B. 2 voicing status of vowel voiceless [i] voiced [i]

3

Table 3: Experimental Conditions environment voiced voiceless

[-V]: C[-voi] ___ C[-voi] [hokito] ☺ [hoki to] [+V]: /ǰ/ ___ C[+/- voi] ☺ [tɑɟʑigɑ] [tɑɟʑigɑ] [+Vnas]: /N/ ___ C[+/- voi] ☺ [kedɑɲigɑ] [kedɑɲi gɑ] (8) Stimuli (see Appendix 1)

A. 180 experimental items (CVCiCV or CVCVCiCV). All nonsense words. B. 30 [i] fillers + 300 fillers + 10 practice items C. All words were produced by a female native speaker of the Tokyo dialect of Japanese.

(9) Procedure

A. I made two sets of word list. B. Participants were randomly assigned to one of the two sets. C. Participants were positioned in a quiet place. D. Participants were asked to listen to stimuli and hit a button on a response box as soon as

they heard /i/. E. When subjects hit a button on a response box, the next word came up after a 1000ms

pause. When subjects did not respond within 3000ms, the next word came up after a 1000ms pause.

F. Response time (RT) was recorded by a computer.

(10) Participants: 47 native speakers of the Tokyo dialect of Japanese. (11) RT Results Figure 3: Mean RTs for JPN Listeners in Phoneme Monitoring Task

618.9593.2

658.6

609.8

547

590.6

450

500

550

600

650

700

750

[-V] [+V]* [+Vnas]*

RT(ms)voiceless voiced

The symbol ‘*’ indicates that the difference was statistically significant in the environment.

4

A. [-V] (devoicing) environment: no significant difference as quickly as voiced [i].

able 4: [-V] Devoicing Environment (/i/ between voiceless consonants) voiceless [i]

voiceless [i ] was detected T voiced [i]

acoustic cues + (strong) – (weak) phonotactics:

between consonants + + )

vowel should be (facilitory) (facilitory

effects

of environment & vowel – (bad) + (good) phonology:

combinationresults no difference detection speed B. [+V] (voicing) environment: voiceless [i] was detected more slowly than voiced [i].

able 5: [+V] Voicing Environment (/i/ follows /ǰ/) voiced [i] voiceless [i]

T

acoustic cues + (strong) – (weak) phonotactics:

between consonants + + )

vowel should be (facilitory) (facilitory

effects

of environment & vowel + (good) – (bad) phonology:

combinationresults faster slower detection speed C. [+Vnas] (voicing) environment: voiceless [i] was detected more slowly than voiced [i].

Table 6: [+Vnas] Voicing Environment (/i/ follows a nasal)

voiced [i] voiceless [i] acoustic cues + (strong) – (weak) phonotactics:

need to be after a nasal – – )

a vowel doesn’t (inhibitory) (inhibitory

effects

of environment & vowel + (good) – (bad) phonology:

combinationresults faster slower detection speed

5

(12) Error Rate Results (error = the number of times when /i/ was ignored.)

Figure 4: Mean Error Rates for JPN Listeners in Phoneme Monitoring Task

13.6 9.8

43.3

14.88.5 12.2

0102030405060

[-V] [+V] [+Vnas]*

Percent Errorvoiceless voiced

The symbol ‘*’ indicates that the difference was statistically signif

. [-V] (devoicing) environment: no significant difference as accurately as voiced [i].

able 7: [-V] Devoicing Environment (/i/ between voiceless consonants) voiceless [i]

icant in the environment.

A voiceless [i ] was detected T voiced [i]

acoustic cues + (strong) – (weak) phonotactics:

between consonants + + )

vowel should be (facilitory) (facilitory

effects

of environment & vowel – (bad) + (good) phonology:

combinationresults no difference errors B. [+V] (voicing) environment: no significant difference

ed as accurately as voiced [i].

able 8: [+V] Voicing Environment (/i/ follows /ǰ/) voiced [i] voiceless [i]

voiceless [i ] was detect T

acoustic cues + (strong) – (weak) phonotactics: vowel should be between consonants

+ (facilitory) + ) (facilitoryeffects

of environment & vowel + (good) – (bad) phonology:

combinationresults no difference WHY??? errors

6

C. [+Vnas] (voicing) env ] was ignore

Table 9: [+Vnas] Voicing Environment (/i/ follows a nasal) voiced [i] voiceless [i]

ironment: voiceless [i d more than voiced [i].

acoustic cues + (strong) – (weak) phonotactics:

need to be after a nasal – – )

a vowel doesn’t (inhibitory) (inhibitory

effects

of environment & vowel + (good) – (bad) phonology:

combinationresults fewer a lot ! errors

II. EXPERIMENT 2

3) Question Is voiceless [i ] detected as quickly and as accurately as fully voiced [i] by

4) Task, Conditions, Procedures: same as Experiment 1

15) Participants: 45 native listeners of American English at U of A.

16) RT Results

igure 5: Mean RTs for AE Listeners in Phoneme Monitoring Task

(1

native listeners of American English?

(1 ( ( F

962.91001.8

911.4

817.3877.9

837.2

750800850900950

1000105011001150

[-V] [+V] [+Vnas]

RT(ms)voiceless voiced

7

(17) Error Rate Results

es for AE Listeners in Phoneme Monitoring Task Figure 6: Mean Error Rat

63.2

78.985.8

24.4 26.122.3

0

10

20

30

40

50

60

70

80

90

100

[-V]* [+V]* [+Vnas]*

Percent Errorvoiceless voiced

In all environments, voiceless [i] was detected more slowly and was ignored more than

III. EXPERIMENT 3

8) Question Are lexical words with voiceless [i] recognized as quickly and as ese

19) Conditions: same as Experiment 1 but with real words

20) Conditions: same as Experiment 1

rimental Conditions voiced voiceless1 glossary

voiced [i]. AE listeners don’t have Japanese-specific phonological and phonotactic knowledge. They relied on acoustic cues.

(1

accurately as the same words with fully voiced [i] by native Japanlisteners?

( ( Table 10: Expe

environment [-V]: C[-voi] ___ C[-voi]

‘to[aɕita] [aɕi ta] ashita

morrow’ [+V]: /ǰ/ ___ C[+/- voi] [kɯɟʑiɾa] [kɯɟʑiɾa]

‘whale’ kujira

[+Vnas]: /N/ ___ C[+/- voi] [yo ɲige] [yo ɲige] ‘to flee by night’

yonige

1 Voiceless vowels include both devoiced and deleted vowels.

8

(21) Stimuli: (see Appendix 2) All r al words.

s + 10 practice items yo dialect of Japanese.

2) Procedure: same as Experiment 1 except I asked participants to hit a button on a

3) Participants: 47 native speakers of the Tokyo dialect of Japanese.

4) RT Results

igure 7: Mean RTs for JPN Listeners in Lexical Decision Task

A. 120 experimental items. eD. 30 real word fillers + 200 non-word fillerE. All words were produced by a female native speaker of the Tok

(2response box as soon as they heard a real word.

(2 (2 F

295.9 314

414.5352.7

270.9303

150200250300350400450500550

[-V]* [+V]* [+Vnas]*

RT(ms)voiceless voiced

. [-V] (devoicing) environment: words with voiceless [i] was detected as quickly as words

able 11: [-V] Devoicing Environment (/i/ between voiceless consonants)

words with

A

with voiced [i].

T words with

voiced [i] voiceless [i] acoustic cues + (strong) – (weak) phonotactics:

between consonants + + )

vowel should be (facilitory) (facilitory

effects

phonology: combination of environment & vowel

– (bad) + (good)

results slower faster WHY??? detection speed

9

B. [+V] (voicing) environment: voiceless [i] was detected more slowly than voiced [i].

able 12: [+V] Voicing Environment (/i/ follows /ǰ/) voiced [i] voiceless [i]

T

acoustic cues + (strong) – (weak) phonotactics:

between consonants + + )

vowel should be (facilitory) (facilitory

effects

of environment & vowel + (good) – (bad) phonology:

combinationresults faster slower detection speed C. [+Vnas] (voicing) environment: voiceless [i] was detected more slowly than voiced [i].

able 13: [+Vnas] Voicing Environment (/i/ follows a nasal) oiced [i] voiceless [i]

T v

acoustic cues + (strong) – (weak) phonotactics:

need to be after a nasal – – )

a vowel doesn’t (inhibitory) (inhibitory

effects

of environment & vowel + (good) – (bad) phonology:

combinationresults faster slower detection speed

(25) Error Rates Results

es for JPN Listeners in Lexical Decision Task Figure 8: Mean Error Rat

6

19.4

42.3

13.87 8.1

05

101520253035404550

[-V]* [+V]* [+Vnas]*

Percent Errorvoiceless voiced

10

A. [-V] (devoicing) environment: words with voiceless [i ] had fewer errors than words with

able 14: [-V] Devoicing Environment (/i/ between voiceless consonants)

voiceless [i]

voiced [i].

T voiced [i]

acoustic cues + (strong) – (weak) phonotactics:

between consonants + + )

vowel should be (facilitory) (facilitory

effects

phonology: combination of environment & vowel

– (bad) + (good)

results more fewer errors B. [+V] (voicing) environment: words with voiceless [i] had more errors than words with

able 15: [+V] Voicing Environment (/i/ follows /ǰ/)

voiced [i] voiceless [i]

voiced [i].

T

acoustic cues + (strong) – (weak) phonotactics:

between consonants + + )

vowel should be (facilitory) (facilitory

effects

of environment & vowel + (good) – (bad) phonology:

combinationresults fewer more errors C. [+Vnas] (voicing) environment: words with voiceless [i] had more errors than words with

able 16: [+Vnas] Voicing Environment (/i/ follows a nasal)

voiced [i] voiceless [i]

voiced [i].

T

acoustic cues + (strong) – (weak) phonotactics:

need to be after a nasal – – )

a vowel doesn’t (inhibitory) (inhibitory

effects

of environment & vowel + (good) – (bad) phonology:

combinationresults less a lot ! errors

11

IV. CONCLUSION

• Listeners used phonotactic and phonological knowledge of their native language for

• Three factors effect differently in each experiment. [sɑŋ:kjɯ:ɸo:ɾisɯɲi ŋgɯ] ☺

SELECTED R

upoux, Emmanuel, Kazuhiko Kakehi, Yuki Hirose, Christopher Pallier & Jacques Mehler.

D Mehler. (2001). New tive

H ). Unvoicing of vowels in Japanese, in Phonetic Society of Japan (ed.)

Tsuchida, Ayako. (2001). Japanese vowel devoicing: Cases of consecutive devoicing

V y. Albany: State University of

Yuen, Chris L. – K. (2000). The perception of Japanese devoiced vowels, Proceedings of the

PPENDIX 1: Experimental Items for Experiments 1 & 2

• Devoicing (D-) Environment (30 words) hokita ita moshito meshite

• Non-Devoicing (ND-) Environment (20 words) taji yojima wajina

detecting vowel /i/.

EFERENCES

D(1999). Epenthetic vowels in Japanese: A perceptual illusion? Journal of Experimental Psychology: Human Perception and Performance 25: 1568-1578.

upoux, Emmanuel, Christopher Pallier, Kazuhiko Kakehi & Jacquesevidence for prelexical phonological processing in word recognition. Language and CogniProcesses 16: 491-505.

an, Mieko Shimizu. (1962The Study of Sounds 10: 81-99.

environments, Journal of East Asian Linguistics 10: 225-245. ance, Timothy J. (1987). An Introduction to Japanese PhonologNew York Press.

Chicago Linguistic Society 36: 531-547.

A

sekite mekito seshwachika machike nahika mahiko tekisa kachiho nehisa hokito yachite samokika toyakiko nagahita kotashike hasechito tadahika saneshita moneshito wagashite yawachike motahike nosahiko kotashite notochiko menahisa

da najida wajide tejido tajiga tajida kojiga sojime rejiza kejiso hojizo mejina yajide natajima kanojime sasajina ketajiba mokojibe narajizo merajime warejida masajido kanajida warejina tasajiba yatejima yotajino towajina

12

• Non-Devoicing-Nasal (Nas-) Environment (20 words) wan e ime tanina

PPENDIX 2: Experimental Materials for Experiment 3

• Devoicing (D-) Environment (20 words) akikan kachiku ‘domestic animals’

f a pl ce’ ’ ’

gh’ sed’ e’

voicing (ND-) Environment (20 words) ajiwau nekojita ‘too sensitive to hot food’

mensions’

’ nt person’

ean ’ ’ ’

• Non-Devoicing-Nasal (Nas-) Environment (20 words) onigiri ‘luggage’

in’

r’ nigh way w thout

id tenido saniza kenima nantanimo yonida waniba nenigo seniba hanima soniga tonize maniga hasanina yosanine kasanide naraniza hokonimo kedaniga katonino wasenida tarenido tasonibe ketanime samonima yoranide nasaniba wadonima

A

‘empty can’ hashika ‘measles’ tsukisasu ‘stub’ akisu ‘robber’ hisan ‘misery’ akita ‘the name o a ashita ‘tomorrow fukitsu ‘ill omen onshitsu ‘a green house’ koshitsu ‘a single room’ hakike ‘nausea’ fukikesu ‘to blow off’ kakikomu ‘to write down’ sekikomu ‘to couochikomu ‘to get depres tekisetsu ‘appropriate’ yakisoba ‘fried noodlsoshite ‘and then’ tekitoo ‘reasonable’

• Non-De ‘to taste’ oojisama ‘prince’

sanjigen ‘three di kajiru ‘to bite’ kejime ‘to be distinguishable’ genjitsu ‘real world’ kujira ‘whale’ kujikeru ‘to be discouraged’ tejina ‘conjuring tricks’ majime ‘serious sujigaki ‘plot’ tojiru ‘to close’ fujisan ‘Mt. Fuji’ betsujin ‘differehajiku ‘to snap’ soojiki ‘vacuum cl er najimu ‘to become familiarnejiru ‘to wrench hajimeru ‘to begin’

‘rice ball’ tanigoe ‘crossing over a valley’ tenimotsugyuuniku ‘beef’ nanimono ‘whoever’ wanigawa ‘crocodile skganimata ‘bowlegged’ inisharu ‘initial’ tanima ‘ravine’ kaniza ‘the Cancer’ onigokko ‘playing tag’ butaniku ‘pork’ uraniwa ‘back yard’ aniki ‘elder brother’ banira ‘vanilla’ hiniku ‘sarcasm’ minikui ‘ugly’ monitaa ‘monitoyonige ‘to flee by t’ kuinige ‘to run a i paying one’s bill’

13