Does Basque have an impossible phonological opposition?

Ander Egurtzegi¹ (with Andrea García-Covelo² & Iñigo Urrestarazu-Porta³)

CNRS¹³ – IKER UMR5478¹²³, IPS-LMU², UPV/EHU³, UPPA²³

15-11-2023

Glottal fricatives

How about a nasalized glottal approximant /h̃/?

Typology

/h/ in the world’s languages

/h/ is pretty much everywhere in the world

/h~/ is not that frequent

/h̃/ in the world’s languages

Walker & Pullum: impossible segment. aerodynamic constraint.

/h̃/ in the world’s languages

Of course, Basque. Why would be speak about Basque if not?

/h/ vs. /h~ in a subgroup of lang. with /h~/

An extremely rare opposition

If /h̃/ is so rare, then languages that include both /h/ and /h̃/…

• Kwangali (kwan1273)
  
  
• ThiMbukushu (mbuk1240)
  
  

Kwangali and ThiMbukushu, two Southern Bantu languages of northern Namibia

• Seimat (seim1238)

Seimat, an Oceanic language of the Admiralty Islands

• North-Eastern Basque (basq1248)

The Basque case

Zuberoan Basque

Laryngeal approximants

• Oral /h/, potentially present in Proto-Basque (hori ‘that’)

• Nasalized /h̃/, from historical intervocalic n (Lat. (h)onore* > oh̃ore ‘honor’)

Geographic distribution:

• /h/ -> Only in continental varieties

• /h̃/ -> Only in Zuberoan & Mixean Basque

Restrictions:

• Onset only

• First 2 syllables only

• Only 1 aspirated segment in each word

Some examples of /h/

Example	Transcription	Gloss
hori	/hoɾi/	‘that (one)’
harri	/hari/	‘stone’
aho	/aho/	‘mouth’
ehun	/ehun/	‘hundred’
elhe	/elhe/	‘word’
senhar	/s̺enhar/	‘husband’

Some examples of /h̃/

Example	Transcription	Gloss
ahai	/ah̃ai̯/	‘ram’
ahuntz	/ah̃unts̻/	‘goat’
ihes	/ih̃es̺/	‘run away’
ahate	/ah̃ate/	‘duck’
dihaü	/dih̃ay/	‘money’
uhue	/uh̃ue/	‘honor’

Let’s listen to them

Oral

aihai ‘dinner’

Nasalized

ihitza ‘dew’

Diachrony

Historical origin

*n > /h̃/ /V_V

mih̃i ‘tongue’ ~ min-gain ‘upper part of the tongue’

*ardah̃o ‘wine’ ~ ardan-degi ‘winery’

*gaztah̃a ‘cheese’ ~ gaztan-bera ‘curd’

But mihi, ardo & gazta in Standard Basque

Chronology

500-800 AD
Lenition of intervocalic /n/

/VnV/ > /Vh̃V/

1000-1300 AD
Loss of laryngeals
in post-tonic last syllable
in (North) Eastern varieties

/hV.’hV.hV/ > /hV.hV.V/

Chronology

1600-1900 AD
Loss of nasality in aspirates
in most Basque varieties

/Vh̃V/ > /VhV/

Today
Nasalization of aspirates
preserved in Zuberoan Basque

ehi ‘finger’ vs. eh̃i ‘easy’

Romance languages

*n > ∅ /V_V in Romance

Galician-Portuguese

Latin	Gal-Port.	Gloss
plānum	chão	‘flat’
tenebrās	tẽevras	‘darkness’
lūnam	lũa	‘moon’
regīnam	raĩa	‘queen’
gallīnam	galĩa	‘hen’

*n > ∅ /V_V in Romance

Corsican

Chisa Corsican	Galeria Corsican	Gloss
[ˈpanɛ]	[ˈpãe]	‘bread’
[ˈpona]	[ˈpe]	‘to put’
[ˈvĩnu]	[ˈbĩu]	‘wine’
[ˈɔ̃ne]	[ɔ̃]	(suffix)

*n > ∅ /V_V in Romance

Medieval Gascon

Latin	Gascon	Gloss
Dominicus	Domeeg	(personal name)
gallīnas	garias	‘hen (pl.)’
abellanētum	aueraed	‘hazelnut grove’
camināre	camiar	‘to walk’
Castanētum	Castahied	(place name)

*n > ∅ /V_V in Romance

Sardinian

Sard. Var. A	Sard. Var. B	Sard. Var. C
[ˈpanɛ]	[ˈpãi]	-
[ˈkɛna]	[ˈtʃɛ̃a]	-
[ˈaʒina]	[ˈaʒĩa]	[ˈaʒĩʔa]
[ˈluna]	[ˈlũa]	[ˈlũʔa]

Lack of aspiration in Romance?

Scarce evidence for an aspirate outcome in Romance.

• [h] in Medieval Gascon?

- [ʔ] in Sarrabus/Isili Sardinian

Why did /h̃/ develop in Basque? / Why was it maintained?

• /h/ was likely part of the language before /h̃/ developed.

- [h̃] was likely an allophonic variant of /h/.

Phonology

Establishing the /h̃/ vs. /h/ opposition

Minimal pair/triplet:

ehi /ˈehi/ ‘finger’

vs.

ehi /ˈeh̃i/ ‘easy’

vs.

ei /ˈe.i/ ~ /ei̯/ ‘ill’

/Vh̃V/ or /ṼhṼ/?

• Zuberoan only shows contrastive nasalization in word-final stressed vowels.

• Mixean does not show contrastive nasalization in vowels.

Nasalization spread / assimilation

All vowels surrounding nasal consonants are contextually nasalized in Basque:

ona [ˈõnã] ‘the good’

ama [ˈãmã] ‘mother’

uhue [ũˈh̃ũẽ] ‘honor’

ahate [ãˈh̃ãte] ‘duck’

Nasalization spread / assimilation

Example	Transcription	Gloss
janhari	[jãnˈɦ̃ãi̯]	‘food’
sinhets	[s̺ĩɲˈɦ̃ẽts̺]	‘to believe’
nahi	[ˈnãɦ̃ĩ]	‘to want’
mehe	[ˈmẽɦ̃ẽ]	‘thin’
senhar	[ẽɦ̃ũn]	‘hundred’
lehen	[lẽɦ̃ẽn]	‘first’

Background & Research question

Previous phonetic research

It looks pretty much like this: empty

Mostly impressionistic / phonological

No research on the opposition in Basque

No research on the opposition in any other language

No nasalance-based study of [h̃]

Research is pressing

Basque is an endangered language (vulnerable according to UNESCO)

(Almost?) All Basque varieties other than the standard

Particularly the Basque varieties spoken in the Northern Basque Country in France

Research questions

Are /h/ and /h̃/ still found in Basque?

Most likely varieties:

• Mixean Basque

• Zuberoan Basque

If present, let’s describe the opposition

1st Study

Amiküze (Mixe)

Materials

15 recordings from 15 towns in the Mixean region. They were originally recorded for dialectological description (Camino 2016).

5 recordings were not used due to: - Speakers being older than 85-90 years. - Recordings being made in the 80’s (extracted from a DAT).

10 subjects. 3 females and 7 males. Age range: 80-85 years.

All recordings go well over the hour. Recorded with a portable recorder in high quality, but in a rural situation.

All audio files were manually transcribed, force-aligned using the WebMAUS application (Kisler et al. 2017) set for Basque (FR), and subsequently hand-corrected as needed.

Data

behi ‘cow’

ihizin ‘hunting’

Problem: Measuring acoustic nasality

Acoustic analysis

20 explicit measurements of vocalic nasality (Styler 2017):

• F1, F2, F3 frequency
• F1, F2, F3 bandwidth
• A1, A2, A3 amplitude
• P0, P1, P2 amplitude
• P0, P1 prominence
• A1-P0, A1-P1, A1-P2
• H1-H2, A3-P0, overall spectral center of gravity

Measurements taken in Praat and R (using wrassp; Winkelmann et al. 2017)

Additionally: MFCCs 2-12 (OpenSMILE)

All measurements taken at 5ms intervals

PCA transformation & data reduction

For each speaker, the 31-feature set was submitted to a principal components analysis (PCA) model

The number of PCs that cumulatively explained at least 80% of the total variance were retained for each speaker:

• 11-13 PCs
• Mean: 12.3, SD: 0.82

The scores for the retained PCs were used as linear predictors in speaker-specific models built to distinguish oral and nasalized vocalic contexts

Oral-nasal model training & prediction

“Unambiguously” oral and nasalized tokens were selected:

• Nasalized: 10% in NVC, 50% in NVN, and 90% in CVN
  
• Oral: 10%, 50%, and 90% in CVC

Average training data per speaker: 258 nasalized tokens, 222 oral tokens

Logistic regression models (with PC score IVs) built for each speaker

Hosmer and Lemeshow goodness of fit tests

LME with random intercepts by speaker

Pair-wise post hoc tests with Tukey alpha correction

From the oral tokens (at each time point), 80% of the data was selected for training, and 20% was retained for predicting values

Using the PC scores, response scores predicted for phone midpoints: assimilatorily nasalized [h̃], etymologically nasalized /h̃/, non-nasalized (oral) /h/, and oral vowels

Interpretations: Closer to 1 -> nasal Closer to 0 -> oral

Results

2nd Study

Larraine

Zuberoan population: 12716

Larraine: 195

Passive knowledge: 22.8%

Effective usage: 5.8%

Speakers among the oldest population

Experimental design

word reading task

Stimuli + SpeechRecorder + Nasalance device

Stimuli: wordlist

/h/	/h̃/	Nh / NVh / hVN
behi ‘cow’	ahate ‘duck’	uɲhu ‘onion’
bihotz ‘heart’	ihaute ‘carnival’	lehen ‘first, before’
ehi ‘finger’	ehi ‘easy’	nihaur ‘me, myself’

SpeechRecorder: prompting

Randomize order

French translation, to make sure they understood the target (ilitaterate in Basque)

key: no graphic difference between a priori oral and nasal segments

Nasalance device

Wooden plate

Data

Participants

self reported

5 participants

4 male : 1 female

60-70 years old

L1 Zuberoan Basque

L2 French at age 5 (school)

Recordings

aihaia ‘dinner’

desuhue ‘dishonor’

Nasalance

→

→

\[ Amplitude \]

→

→

\[ A_n \]

\[ A_o \]

↓

↓

\[ \frac{A_n}{A_n + A_o} \times 100 \]

Analysis

Bayesian generalized mixed-effects model with brms

Dependent variable: z-scored nasalance

Independent variables: aspirate category, trial

Correlated varying intercept & slope: speaker, word

Weakly informative priors

Further specs

8 chains x 10 000 iterations (5000 warm-up)

delta = 0.999

maximum tree depth = 12

Posterior distributions

Posterior distributions

Discussion

Etymologically oral aspirates ≠ Assimilated nasal aspirates ≈ Etymologically nasalized aspirates

There is an opposition between /h/ and /h̃/!!!

Discussion

Etymologically oral aspirates ≠ Assimilated nasal aspirates ≈ Etymologically nasalized aspirates

There is an opposition between /h/ and /h̃/!!!

BUT…

Not all speakers maintain the opposition to the same extent

Discussion

Etymologically oral aspirates ≠ Assimilated nasal aspirates ≈ Etymologically nasalized aspirates

There is an opposition between /h/ and /h̃/!!!

BUT…

Not all speakers maintain the opposition to the same extent

Some lexical items have lost nasalization

Discussion

Etymologically oral aspirates ≠ Assimilated nasal aspirates ≈ Etymologically nasalized aspirates

There is an opposition between /h/ and /h̃/!!!

BUT…

Not all speakers maintain the opposition to the same extent

Some lexical items have lost nasalization

There is an opposition, but it is receding

Dynamic analysis with fPCA

Work in progress

Waves have different durations

Waves with registered time

Results of fPCA

What variation does each PC capture?

Modelling PC1 scores

fit_nasalance.pc1 <- 
  brm(s1 ~ etym_asp + (1 + etym_asp | speaker),
      family = gaussian(),
      data = PCscores,
      file = "fits/fit_nasalance-pc1",
      cores = 4,
      control = list(adapt_delta = 0.99)
  )

etym_asp	emmean	lower.HPD	upper.HPD
assimilated	-0.7363393	-1.1562409	-0.2801506
nasalized	-0.4691249	-1.5250014	0.7226606
oral	0.9772860	0.4972216	1.5185178

Posterior distribution of PC1 scores

Waves reconstructed with PC1

Modelling PC2 scores

fit_nasalance.pc2 <- 
  brm(s2 ~ etym_asp + (1 + etym_asp | speaker),
      family = gaussian(),
      data = PCscores,
      file = "fits/fit_nasalance-pc2",
      cores = 4,
      control = list(adapt_delta = 0.999)
  )

etym_asp	emmean	lower.HPD	upper.HPD
assimilated	-0.3503060	-0.47628012	-0.1989825
nasalized	0.2663018	0.03689933	0.4889086
oral	0.1087322	-0.08299959	0.3058489

Posterior distribution of PC2 scores

Waves reconstructed with PC1 and PC2

Modelling PC3 scores

fit_nasalance.pc3 <-
  brm(s3 ~ etym_asp + (1 + etym_asp | speaker),
      family = gaussian(),
      data = PCscores,
      file = "fits/fit_nasalance-pc3",
      cores = 4,
      control = list(adapt_delta = 0.999)
  )

etym_asp	emmean	lower.HPD	upper.HPD
assimilated	0.08153864	-0.08457668	0.2548145
nasalized	-0.06823417	-0.26750860	0.1754418
oral	-0.00615058	-0.20369904	0.2071757

Posterior distribution of PC3 scores

Waves reconstructed with PC1, PC2 and PC3

PC1 captured the relevant variation

It’s not over!

Read our last paper

Visit my website

References

• Walker, Rachel & Pullum, Geoffrey K. 1999. Possible and Impossible Segments. Language 75(4). 764-780. doi: 10.2307/417733

Phonologic literature on Basque nasalized aspirates

• Hualde, José Ignacio. 1993. Topics in Souletin phonology. In J.I Hualde & J. Ortiz de Urbina (eds.) Generative Studies in Basque Linguistics , pp. 289-327. John Benjamins.

• Egurtzegi, Ander. 2013. Diferentes tipos de aspiración en vasco (con análisis espectrales del dialecto suletino actual) [Different kinds of aspiration in Basque (with spectral analyses of the modern Zuberoan dialect)]. In Blasco et al. (eds.), Iberia e Sardegna . Firenze: Le Monnier. 151–169.

• Egurtzegi, Ander. 2018. On the phonemic status of nasalized /h̃/ in Modern Zuberoan Basque. Linguistics 56. 1353-1367. doi: 10.1515/ling-2018-0024

• Egurtzegi, A., A. García-Covelo & I. Urrestarazu-Porta. 2023. “A nasalance-based study of the /h/ vs. /h̃/ opposition in Zuberoan Basque”. In Skarnitzl & Volín (eds.), Proceedings of ICPhS 20, 3427-3431.

• Egurtzegi, Ander & Carignan, Christopher. 2020. A typological rarity: the /h/ vs /h̃/ contrast of Mixean Basque. Labphon 17.

• Egurtzegi, Ander. 2023. /h̃/ hasperen sudurkarituaren inguran [On the nasalized aspiration /h̃/]. International Journal of Basque Linguistics and Philology 57.

Stimuli source

• Larrasquet, Jean. 1939. Le Basque de la Basse-Soule Orientale . Paris: C. Klincksieck.

Software and R packages

• Boersma, Paul & Weenink, David. 2022. Praat. Doing phonetics by computer . https://www.praat.org

• Bürkner, Paul-Christian. 2017. brms: An R Package for Bayesian Multilevel Models Using Stan. Journal of Statistical Software 80(1). 1-28. doi: 10.18637/jss.v080.i01

• Draxler, Christoph & Jänsch, Klaus. 2004. SpeechRecorder - a Universal Platform Independent Multi-Channel Audio Recording Software. In Proc. of LREC. 559-562. https://www.bas.uni-muenchen.de/Bas/software/speechrecorder/

• R Core Team. 2022. R: A language and Environment for Statistical Computing . https://www.R-project.org

• Wickhan, Hadley et al. 2019. Welcome to the tidyverse. Journal of Open Source Software 43(4). doi: 10.21105/joss.01686