Durational variation in Polish fricatives provides evidence for hybrid models of phonology

Kamil Kaźmierski
Faculty of English at AMU in Poznań

ICPhS :: August 9th, 2019

kamil.kazmierski@wa.amu.edu.pl
wa.amu.edu.pl/wa/kazmierski_kamil

1 / 14

Neighborhood density vs. phonotactic probability

Spoken-word recognition (Vitevitch & Luce 1998):
- Higher neighborhood density → Slower recognition (lexical effect)
- Higher phonotactic probability → Faster recognition (sublexical effect)
- Both lexical and sublexical effects → support for hybrid phonology (Pierrehumbert 2002)
Do these effects have consequences for the acoustics of speech production?

2 / 14

Neighborhood density

No. of lexemes created by removal, addition, or substitution of a single phoneme

Low: więc /vjɛnt͡s/ 'therefore'

removal: wiec /vjɛt͡s/
addition: Ø
substitution: Ø

High: stałe /ˈstawɛ/ 'constant'

stał /staw/
stałeś /ˈstawɛɕ/, stałem /ˈstawɛm/
stare /ˈstarɛ/, stale /ˈstalɛ/, stała
/ˈstawa/, stało /ˈstawɔ/

3 / 14

Phonotactic probability

Sum of log frequencies of diphones across the corpus (Vitevitch & Luce 2004)

Low: chcą /xt͡sɔw̃/ 'they want'

xt͡sɛ, xt͡sɛʂ
Ø

High: wie /vjɛ/ 'she/he/it knows'

vjɛɕtɕ, vjast, vjɔzɛ, vjɛm, vjɛmɨ, ...
bjɛda, zjɛm, pjɛs, ...

4 / 14

Research question: Are the differential effects of phonotactic probability and neighborhood density  present in production?Motivation: Perception-production symmetry? Language specificity? Beyond lab speech?Predicition: High neighborhood density will increase and high phonotactic probability will decrease fricative durations.5 / 14

Method6 / 14

Data

Greater Poland Spoken Corpus

68 Speakers; 69,720 Word tokens

wa.amu.edu.pl/korpuswlkp

(Kaźmierski, Kul & Zydorowicz in press)

7 / 14

Data extraction

Transcripts hand-aligned at breath-group level in Praat (Boersma & Weenink 2016)

Corpus creation, management and querying with LaBB-CAT (Fromont & Hay 2012)

→ Force-aligned at word and phoneme level

→ Fricative initial (ɕ ʂ ʑ f s v x z) `C₁C₂V...`content words

8 / 14

Model architecture (N = 2,850)

Mixed-effects linear regression with lme4 (Bates et al. 2015) in (R Core Team 2019)

Response:
- Fricative duration
Predictors of theoretical interest:
- Neighborhood density, Phonotactic probability Phonological CorpusTools (Hall et al. 2018)
Control predictors:
- Average speaking rate, Rate deviation, Gender, Word duration, Frequency, Prefix, Stress
Random terms:
- (1 | Fricative), (1 + Neighborhood density + Phonotactic probability | Speaker)

9 / 14

Results10 / 14

Divergent effects of Neighb. dens. and Phon. prob.

11 / 14

Conclusions

Within the same dataset:

neighborhood density effect → Lexical level
phonotactic probability effect → Sublexical level
the presence of both → Hybrid model of phonology (Goldinger 2007; Pierrehumbert 2002; Ernestus 2014)

12 / 14

Thank you!

Neighborhood density and phonotactic probability show differential effects on duration of Polish fricatives

This research was supported by National Science Center (Poland) grant no. UMO-2017/26/D/HS2/00027

kamil.kazmierski@wa.amu.edu.pl

13 / 14

Remaining fixed effects in the model
 
    Predictor 
    B 
    p 
  


    Rate deviation 
    -0.038 
    < 0.001 
  

    Average rate 
    -0.049 
    < 0.001 
  

    Gender Male 
    -0.018 
    0.557 
  

    Word duration 
    0.920 
    < 0.001 
  

    Frequency 
    0.005 
    0.627 
  

    Prefix TRUE 
    -0.117 
    < 0.001 
  

    Stress TRUE 
    0.078 
    < 0.001 
  

14 / 14

Predictor	B	p
Rate deviation	-0.038	< 0.001
Average rate	-0.049	< 0.001
Gender Male	-0.018	0.557
Word duration	0.920	< 0.001
Frequency	0.005	0.627
Prefix TRUE	-0.117	< 0.001
Stress TRUE	0.078	< 0.001

Neighborhood density vs. phonotactic probability

Spoken-word recognition (Vitevitch & Luce 1998):

Higher neighborhood density → Slower recognition (lexical effect)
Higher phonotactic probability → Faster recognition (sublexical effect)
Both lexical and sublexical effects → support for hybrid phonology (Pierrehumbert 2002)

Do these effects have consequences for the acoustics of speech production?

2 / 14

Help

Keyboard shortcuts

↑, ←, Pg Up, k

Go to previous slide

↓, →, Pg Dn, Space, j

Go to next slide

Home

Go to first slide

End

Go to last slide

Number + Return

Go to specific slide

b / m / f

Toggle blackout / mirrored / fullscreen mode

Clone slideshow

Toggle presenter mode

Restart the presentation timer

?, h

Toggle this help

Neighborhood density vs. phonotactic probability

Neighborhood density

Low: więc /vjɛnt͡s/ 'therefore'

High: stałe /ˈstawɛ/ 'constant'

Phonotactic probability

Low: chcą /xt͡sɔw̃/ 'they want'

High: wie /vjɛ/ 'she/he/it knows'

Research question: Are the differential effects of phonotactic probability and neighborhood density present in production?

Motivation: Perception-production symmetry? Language specificity? Beyond lab speech?

Predicition: High neighborhood density will increase and high phonotactic probability will decrease fricative durations.

Method

Data

Data extraction

Transcripts hand-aligned at breath-group level in Praat (Boersma & Weenink 2016)

Corpus creation, management and querying with LaBB-CAT (Fromont & Hay 2012)

→ Force-aligned at word and phoneme level

→ Fricative initial (ɕ ʂ ʑ f s v x z) C₁C₂V...content words

Model architecture (N = 2,850)

Results

Divergent effects of Neighb. dens. and Phon. prob.

Conclusions

Thank you!

Neighborhood density and phonotactic probability show differential effects on duration of Polish fricatives

Remaining fixed effects in the model

Neighborhood density vs. phonotactic probability

Help

→ Fricative initial (ɕ ʂ ʑ f s v x z) `C₁C₂V...`content words