LexSIC: A vocabulary task for Sicilian Dialect

Author

Arona, Besler, Cruschina, Ferin, Gyllstad, Venagli, Kupisch

Data exploration

Participants

N tested participants: 100
- 63 Sicilian speakers living in Sicily
- 22 HS of Sicilian in Germany
- 5 South Italians
- 10 North Italians

Exclusion of participants:

7 participants excluded for guessing behavior during IRT analysis for LexSIC: 156, 250, 541, 543, 566, 574, 539
3 participants excluded for guessing behavior in DIALANG ITA: 510, 537, 461
Total participant N = 90
- 56 Sicilian speakers living in Sicily
- 19 HS of Sicilian in Germany
- 5 South Italians
- 10 North Italians

Tab. 1. Sample description
Group	n	Age Mean	sd	range
Sicilian Heritage Speakers (HS)	19	28.21	10.04	14-53
Northern Italians	10	31.30	10.89	24-62
2L1 Sicilian Speakers	56	35.59	11.41	18-69
Southern Italians	5	38.00	10.99	23-52

Level of education


   0    1    2    3    4    5    6    7 
 825  600 1875  825 1500  225  825   75

Tab. 2. Participants' education levels
Group	Education Level	n
Sicilian HS	High school	8
Sicilian HS	Middle school	5
Sicilian HS	University	6
Northern Italians	High school	1
Northern Italians	University	9
Sicilian 2L1	High school	19
Sicilian 2L1	I don't know	1
Sicilian 2L1	Middle school	3
Sicilian 2L1	University	33
Southern Italians	High school	2
Southern Italians	Middle school	1
Southern Italians	University	2

Tab. 3. Participants mothers' education levels
Group	Mothers' Education Level	n
Sicilian HS	High school	1
Sicilian HS	I don't know	2
Sicilian HS	Middle school	12
Sicilian HS	No degree	2
Sicilian HS	University	2
Northern Italians	High school	4
Northern Italians	Middle school	4
Northern Italians	No degree	1
Northern Italians	University	1
Sicilian 2L1	High school	17
Sicilian 2L1	I don't know	8
Sicilian 2L1	Middle school	19
Sicilian 2L1	No degree	2
Sicilian 2L1	University	10
Southern Italians	I don't know	1
Southern Italians	Middle school	4

Tab. 4. Participants fathers' education levels
Group	Father's Education Level	n
Sicilian HS	High school	3
Sicilian HS	I don't know	2
Sicilian HS	Middle school	9
Sicilian HS	No degree	4
Sicilian HS	University	1
Northern Italians	High school	2
Northern Italians	Middle school	4
Northern Italians	No degree	2
Northern Italians	University	2
Sicilian 2L1	High school	15
Sicilian 2L1	I don't know	8
Sicilian 2L1	Middle school	20
Sicilian 2L1	No degree	2
Sicilian 2L1	University	11
Southern Italians	I don't know	1
Southern Italians	Middle school	3
Southern Italians	University	1

Self-reported Sicilian Proficiency (active vs. passive proficiency in the L1 and HS group)

Tab. 5. Active and passive Sicilian proficiency
Group	Proficiency type	M	sd
Sicilian HS	Productive skills	3.05	1.99
Sicilian HS	Receptive skills	4.05	1.70
Sicilian 2L1	Productive skills	4.39	1.60
Sicilian 2L1	Receptive skills	5.00	1.15

Use of Sicilian

Score: 1-4

LexSIC

# Formula = N(yes)[WORDS] - 2 * N(yes)[PSEUDO]
d %>% 
  group_by(participant) %>% 
  mutate(tot_real = sum(answ[cond == "REAL"]),
         tot_unreal = sum(answ[cond == "UNREAL"]),
         tot_lex = tot_real - 2*tot_unreal) %>% 
  ungroup()-> d

Tab. 7. LexSIC score by group
Group	Mean LexSIC score	sd	range
Sicilian HS	31.47	10.17	12-45
Northern Italians	19.50	8.14	3-29
Sicilian 2L1	38.41	7.55	17-50
Southern Italians	33.80	4.96	26-39

DIALANG (Ita) for the Sicilian HS and Sicilian L1 groups

Tab. 8. DIALANG (Ita) score by group
Group	Mean DIALANG_ITA score	sd	range
Sicilian HS	29.89	10.08	7-44
Sicilian 2L1	41.62	4.27	28-49

Distribution of the measures

Correlation Analysis

Following Salmela et al., 2021, we run the following correlation analyses:

Correlation between LexSIC score and age for the L1 and HS group (Spearman Correlation)
Correlation between LexSIC score and self-reported proficiency for L1 and HS group (Spearman Correlation)
Correlation between LexSIC and exposure for L1 and HS group (Spearman Correlation)
Correlation between LexSIC and DIALANG (ITA) for L1 and HS group (Spearman Correlation)

Prior to the correlation analysis, we test:

(a) whether there is a difference between the L1 and HS group in terms of LexSIC and DIALANG ITA scores (Wilcoxon rank-sum test)

(b) whether there is a difference of LexSIC score by education level in the L1 and HS score (Kruskal-Wallis test and Dunn’s multiple comparison test)

Wilcoxon rank-sum test LexSIC and DIALANG

	tot lex SIC
Predictors	Incidence Rate Ratios	CI	p
(Intercept)	29.51	26.10 – 33.37	<0.001
group [sicilian]	1.27	1.10 – 1.47	0.001
Random Effects
σ²	0.03
τ₀₀ _participant	0.07
ICC	0.73
N _participant	75
Observations	5625
Marginal R² / Conditional R²	0.097 / 0.753

Using the Wilcox test, we found a significant difference of LexSIC scores by group (p < .0001****), effect size: r = .30 (moderate)
- HS: Median = 33, IQR = 18 (Interquartile range is the difference between 3rd and 1st quartile)
- L1: Median = 41, IQR = 12
Using the Wilcox test, we found a significant difference of DIALANG scores by group (p = 0 ****), effect size: r = .57 (large)
- HS: Median = 33, IQR = 18
- L1: Median = 42, IQR = 4,25

Difference within groups as a function of educational level

Tab. 9. DIALANG (Ita) score by education level
Group	Education level	Mean DIALANG_ITA score	sd
HS	High school	30.625	11.831
HS	Middle school	28.200	8.267
HS	University	30.333	8.683
sicilian	High school	41.421	4.477
sicilian	Middle school	40.000	3.273
sicilian	University	41.939	4.236

Tab. 10. LexSIC score by education level
Group	Education level	Mean LexSIC score	sd
HS	High school	31.125	10.796
HS	Middle school	35.600	5.859
HS	University	28.500	10.978
sicilian	High school	36.737	8.139
sicilian	Middle school	38.667	5.324
sicilian	University	39.333	7.324

Kruskal-Wallis rank sum test

Within the L1 group, the difference of DIALANG scores by education level is significant (p < .001) and the pairwise comparison shows that all levels are significantly different (p < .001)
Within the L1 group, the difference of LexSIC scores by education level is significant (p < .001) and the pairwise comparison shows that both the comparison between University and High school (p < .001) and the comparison between University and Middle school are significantly different (p < .01)
Within the HS group, the difference of DIALANG scores by education level is significant (p < .001) and the pairwise comparison shows that all levels are significantly different (p < .001)
Within the HS group, the difference of LexSIC scores by education level is significant (p < .001) and the pairwise comparison shows that all levels are significantly different (p < .001)

The generalized linear model does not predict any difference between education levels within the L1 (p = 0.5) and the HS group (p = 0.4).

Correlation between LexSIC and age

L1 group: Spearman’s rank correlation .12 (p < .001) - no significant effect of age (p = .94)
In the HS group, the LexSIC score correlates with age ( Spearman .38 p<.001) - no significant effect of age (p = 0.18)

Correlations between active and receptive skills & LexSIC scores

	tot lex SIC
Predictors	Incidence Rate Ratios	CI	p
(Intercept)	29.51	26.22 – 33.21	<0.001
receptive skills	1.08	1.02 – 1.15	0.008
group [sicilian]	1.27	1.11 – 1.46	0.001
Random Effects
σ²	0.03
τ₀₀ _participant	0.07
ICC	0.71
N _participant	74
Observations	5550
Marginal R² / Conditional R²	0.155 / 0.755

Within the 2L1 group:
- Active skills:
  - Correlation (Spearman): .05 (p < 001)
  - glmer: no effect
- Receptive skills:
  - Correlation (Spearman): -.03 (p = .04)
  - glmer: no effect
- Composite measure
  - Correlation (Spearman): .02 (not significant)
  - glmer: no effect
Within the HS group:
- Active skills:
  - Correlation (Spearman): .78 (p < .001)
  - glmer: no effect
- Receptive skills:
  - Correlation (Spearman): .78 (p<.001)
  - glmer: significant effect (p <.001)
- Composite measure
  - Correlation (Spearman): .79 (p < .001)
  - glmer: significant effect (p <.001)
Within the whole sample:
- Active skills
  - Correlation (Spearman): .32 (p < .001)
  - glmer: no effect
- Receptive skills
  - Correlation (Spearman): .24 (p < .001)
  - glmer: significant effect (p <.05) [see plot below]
- Composite measure
  - Correlation (Spearman): .22 (p < .001)
  - glmer: significant effect (p <.05)

Correlation between Sicilian Use and LexSIC

	tot lex SIC
Predictors	Incidence Rate Ratios	CI	p
(Intercept)	27.03	23.31 – 31.35	<0.001
composite use family	1.01	1.00 – 1.02	0.046
group [sicilian]	1.25	1.09 – 1.44	0.002
Random Effects
σ²	0.03
τ₀₀ _participant	0.07
ICC	0.72
N _participant	74
Observations	5550
Marginal R² / Conditional R²	0.130 / 0.755

Within 2L1 group:
- cor = .13 (p < .001)
- glmer not significant
Within HS group:
- cor = .73(p < .001)
- glmer significant (p<.01)
Within whole sample:
- cor = .30 (p < .001)
- glmer not significant

Correlation between LexSIC and DIALANG ITA

	tot lex SIC
Predictors	Incidence Rate Ratios	CI	p
(Intercept)	29.51	26.34 – 33.06	<0.001
DIALANG	1.12	1.05 – 1.18	<0.001
group [sicilian]	1.27	1.11 – 1.45	<0.001
Random Effects
σ²	0.03
τ₀₀ _participant	0.06
ICC	0.69
N _participant	74
Observations	5550
Marginal R² / Conditional R²	0.200 / 0.755

Within the L1 group:
- the correlation between DIALANG ITA and LexSIC .34 (p < .001, Spearman);
Within the HS group
- the correlation between DIALANG ITA and LexSIC .57 (p < .001, Spearman)

Within the whole sample
- the correlation between DIALANG ITA and LexSIC .50 (p < .001, Spearman)

Data exploration

Data exploration

Participants

Exclusion of participants:

Level of education

Self-reported Sicilian Proficiency (active vs. passive proficiency in the L1 and HS group)

Use of Sicilian

LexSIC

DIALANG (Ita) for the Sicilian HS and Sicilian L1 groups

Distribution of the measures

Correlation Analysis

Wilcoxon rank-sum test LexSIC and DIALANG

Difference within groups as a function of educational level

Kruskal-Wallis rank sum test

Correlation between LexSIC and age

Correlations between active and receptive skills & LexSIC scores

Correlation between Sicilian Use and LexSIC

Correlation between LexSIC and DIALANG ITA

Random Forest Analysis [to be continued]