Methodological implications of single-response declarative vs complex annotative language use on the example of Ukrainian refugees in Poland

Methodological implications of single-response declarative vs complex annotative language use on the example of Ukrainian refugees in Poland

1. Methodological implications of single-response declarative vs complex annotative language use on the example of Ukrainian refugees in Poland Andrzej Jarynowski 1,2, Karolina Czopek 1, Andrea Palmini 2, Alexander Semenov 3, Vitaly Belik 2, Michał B. Paradowski 1 1) University of Warsaw  2) Freie Universität Berlin 3) University of Florida

2. HPAI COVID-19 PARASITES C ontact networks from vision (tracking) , RF I D from vision, bluetooth Human/Human/ minks interactions from vision (tracking) , RF I D - - UKR/PL (Meseals, TB) Spatiotemporal patterns (covariates) Wild birds, climate, poultry density , Weather seasonality Pigs density, human mobility , Production seasonality. trade registries Adherence to measures, excess mortality, causal models Exposure to environment (prediction) Weather seasonality Satellite images Spatical custering of Berlin Early detection ( sensors ) Mortality, eggs production, movements (vision), coughing (sound) Mortality, movements (vision) Coughing (sound analys is ) eggs production, movements, (time series) - biomonitoring Risk Maps Infoveill a nce - - Bug of words (demand and supply of information - discussion and search on AM Discussion of group of interests - Infodemiology Adherence to biosecurity, social tensions (NLP) Adherence to biosecurity, social tensions (NLP) Antisanitarian attitude - Perception of AM Comparing discussion in Poland and Germany - Stigma (M-pox, HIV) Antimicrobial USE ASF Ecological disasters Travel assosiated Diseases

3. 3 Epidemic reproduction rate for Tuberculosis in Lower Silesia

4. 4

5. Triage: Refugees health (Infermedica and Aidmed) Quick and cheap screening fot. Kamil Krasoń

6. Forced migration and language learning TL learning is crucial in the process of integration (Morice et al., 2021; Sharifan et al., 2021; Abu-Khalil et al., 2019) , but it might be inhibited by war trauma (Finn, 2010; Iversen et al., 2014) Education (and language learning) facilitate developing resilience in refugees and minimizing the effects of trauma (Ameen & Cinkara, 2018; Cinkara, 2017; Bouchane, 2016; Sleijpen et al., 2016) A Jarynowski PAN metrics, 16.05.2024 6

7. Learning Polish as a second language by Ukrainians Studies focusing on Polish-Ukrainian interference errors (Czapla, 2020; Krawczuk, 2020; Maliszewski, 2020; Górska, 2015; Kowalewski, 2015) and Ukrainian students in Polish primary schools (Jędryka, 2015, 2021) Intercultural pedagogy (Gębal, 2018) Little research on the sociolinguistic context of using L2 Polish with L1 Ukrainian (Levchuk, 2016) A Jarynowski PAN metrics, 16.05.2024 7

8. Ego-network questionnaire (Paradowski et al. 2021a,b; 2022; ) A Jarynowski PAN metrics, 16.05.2024 8

9. Participants N = 251 Ukrainian refugee students of L2 PL residing in a large city in the Western Poland 130 attending a 30-hr course over 10 weeks with 3 meetings per week 92 attending a 30-hr course over 15 weeks twice a week 29 attending a 60-hr course over 4 weeks with three 5-hour meetings/week Levels A0 – C1 85% ♀ (men mainly teenegers, elderly or handicups i.e. PTSD) ; mean age 34.5 (SD 12.67) On average, resident in Poland for 148 days (SD 209) 16% planned to stay in Poland forever; others planned to leave within 1-120 mo 82% claimed L1 UKR, whereas 17% RUS At the same time, 20% noted that RUS was their second L1, and 30% indicated it was one of the languages they were speaking at home. Motivation to learn PL was generally high (4.50, SD .83, range 1-5). The main reasons for learning TL included living in Poland (55%), working (28%) and studying (7%)

10. A Jarynowski PAN metrics, 16.05.2024 10 Participants’ origins Primarily from eastern UA (Donbas), south, north and centre; Kyiv (17%), Kharkiv (12%), Dnipro (6%), Odessa (4%), Zaporizhia (4%) The further east the origins, the higher the proportion of declared use of Russian

11. Measuring language distance A Jarynowski PAN metrics, 16.05.2024 11 Тищенко (2000:267)

12. Measuring language distance A Jarynowski PAN metrics, 16.05.2024 12 Wichmann (2023)

13. Language Transfer and Interference in L3 Polish Acquisition Effects of language transfer for Ukrainian refugee students: Language transfer influences Polish acquisition by incorporating Ukrainian/Russian syntax patterns, and common vocabulary (in writing/speaking). Negative impact of interference on language learning: Interference leads to errors in grammar and pronunciation, hindering effective communication. Strategies to overcome interference in language learning: Teaching awareness of interference, practicing correct usage, and providing feedback are essential.

14. L1 UA vs L1 RU There is no statistically significant difference in terms of progress in PL between participants with declared L1 UA and L1 RU (with a sample >1 k we might have been able to observe a slight edge for L1 UA, but the effect size would have been small) A Jarynowski PAN metrics, 16.05.2024 14

15. L1 UA vs L1 RU There is no statistically significant difference in terms of progress in PL between participants with declared L1 UA and L1 RU (with a sample >1 k we might have been able to observe a slight edge for L1 UA, but the effect size would have been small) but L1 RU speakers showed: concealment of L1 RU use A Jarynowski PAN metrics, 16.05.2024 15

16. 16

17. 17

18. Avarege of % of general interaction and % summaric time by context 18

19. PAN metrics, 16.05.2024 A Jarynowski 19 Concealment of L1 RU use: 62% of users of Russian in the private sphere declared Ukrainian as their L1 (5% vice versa)

20. L1 UA vs L1 RU There is no statistically significant difference in terms of progress in PL between participants with declared L1 UA and L1 RU (with a sample >1 k we might have been able to observe a slight edge for L1 UA, but the effect size would have been small) but L1 RU speakers showed: concealment of L1 RU use lower centrality in the peer contact networks A Jarynowski PAN metrics, 16.05.2024 20

21. Network of sample group with declared use of Ukrainian and Russian PAN metrics, 16.05.2024 A Jarynowski 21 Weighted degree centrality is higher among students declaring Ukrainian as their L1, while L1 Russian speakers are at the network periphery suggesting linguistic segregation with hints of marginalisation

22. L1 UA vs L1 RU There is no statistically significant difference in terms of progress in PL between participants with declared L1 UA and L1 RU (with a sample >1 k we might have been able to observe a slight edge for L1 UA, but the effect size would have been small) but L1 RU speakers showed: concealment of L1 RU use lower centrality in the peer contact networks less interaction with native speakers of PL , thus a shallower level of immersion in society higher affective filter, reflected both in more perceived barriers in contacts with locals, and a lower degree of motivation A Jarynowski PAN metrics, 16.05.2024 22

23. S ocial desirability bias prevalence of self-report bias in questionnaires According to linguistic, Silesian is a dialect of Polish, not a language sensu stricto . However, it my be perceved as language sensu lato as a social constrain. It would be extremely difficult to count L1 and L2 users of Silesian, as it could be confounded with political declaration (self-declaration in the population census is meaningless). A Jarynowski PAN metrics, 16.05.2024 23

24. Etic and emic Etic (objective) and emic are terms often used in anthropology and linguistics to describe different perspectives or approaches to studying cultures and languages. 24 Context Etic Perspective Emic Perspective Vocabulary use Quantitative assessments of language proficiency or literacy rates across populations using standardized tests or surveys. Exploring how speakers within a specific linguistic community perceive, their proficency in Speaking/Reading Language L1 Use Objectively observing and analyzing how individuals use their native language in real-life contexts, such as tracking language use in various situations or studying language shift and maintenance over time. Exploring individuals' subjective experiences, perceptions, and attitudes toward their native language, considering factors such as language proficiency, cultural identity, and personal connections to the language.

25. Proximal Sociolinguistic Context • What constitutes proximal sociolinguistic context • Influence on language acquisition 25

26. Missing data imputation Given the high volume of missing values, we implemented an imputation process with the missing values treated as missing at random, and using a predictive mean matching technique for the numerical variables, logistic regression imputation for the binary variables, polytomous regression imputation for the categorical variables with more than two categories. After the missing values have been imputed, a random forest process was applied to understand the importance of the different predictors on the outcome variables (self-perceived as well as objective improvement, across subskills and global). A Jarynowski PAN metrics, 16.05.2024 26

27. The first 20 most important predictors for improvement in reading A Jarynowski PAN metrics, 16.05.2024 27

28. The first 20 most important predictors for improvement in writing A Jarynowski PAN metrics, 16.05.2024 28

29. The first 20 most important predictors for improvement in grammar A Jarynowski PAN metrics, 16.05.2024 29

30. Predictor(s) of overall improvement A Jarynowski PAN metrics, 16.05.2024 30 estimate SE p intercept 6.065 2.056 .00995** Communication with a partner in L1 −0.065 0.047 .181 Communication with a partner in Russian −0.002 0.025 .950 Communication with friends in Russian  −0.059 0.022 .01731*  Communication with family in Russian  0.005 0.022 .838

31. In terms of network predictors weighted degree centrality is a significant predictor of progress , both overall : *** p ~0; ** p < .001 Residual SE: .8734 on 221 DF (130 observations deleted due to missingness) BUT: multiple R ²: .05245, adj. R ²: .0353 F (4,221) = 3.058, p = .01765 A Jarynowski PAN metrics, 16.05.2024 31 estimate SE p (intercept) 4.24116 .19128 < 2 e-16 *** weighted degree centrality 0.00686 .00217 .00182** days in country 0.00038 .00036 .29079 sum FL 0.04680 .07095 .51025 % UA in total 0.14770 .18808 .43313

32. In terms of network predictors and to a lesser extent in reading : *** p ~0; ** p < .001 Residual SE: 1.054 on 218 DF (133 observations deleted due to missingness) Multiple R ²: .05016, adj. R ²: .03273 F (4,218) = 2.878, p = .02369 A Jarynowski PAN metrics, 16.05.2024 32 Estimate SE p (intercept) 4.84088 .23276 <2 e-16 *** weighted degree centrality 0.00807 .00263 .00238** days in country 0.00048 .00043 .26714 sum FL −0.06246 .08618 .46940 % UA in total −0.05294 .23040 .81847

33. Length of residence in Poland is a significant negative predictor of objective progress *** p ~0 Residual SE: 14.43 on 21 DF (9 observations deleted due to missingness) Multiple R ²: .5359, adj. R ²: .4475 F (4,21) = 6.063, p = .002091 A Jarynowski PAN metrics, 16.05.2024 33 Estimate SE t p (intercept) 69.42848 12.05729 5.758 1.03 e-05 *** weighted degree centrality −0.16758 0.18088 −0.926 .36474 days in country −0.41717 0.08988 −4.641 .00014*** sum FL 1.62940 3.47309 0.469 .64380 % UA in total 11.93291 10.33820 1.154 .26136

34. Conceivable explanations: respondents may be underestimating the proportion of time when they actually switch (back) into their L 1 , whether unbeknownst or to provide the ‘expected’ answer and project a positive image of themselves seeing faces associated with a particular culture triggers easier processing in the congruent language ( Hartsuiker & Declerck 2013) opportunities to fall back on the shared common denominator mean that calques and code-switching will be easier and less time-consuming ways out of communication breakdowns than attempts at repair in the TL same language background may lead to interference errors being unnoticed, perpetuated, spread and stabilised increased amounts of time spent with same L 1 -speakers naturally diminish the exposure to speakers hailing from other linguistic backgrounds, who may be providing more varied hence enriching input A Jarynowski PAN metrics, 16.05.2024 34

35. Longitudinally We observe two time regimes for subjectively perceived improvement, with a phase transition point around day ~85 of immersion in the country, the former whereof we provisionally label Latinisation. During this period, we observe highest increase in TL proficiency in reading, and to some extent pronunciation and cultural/pragmatic competence, but not so much in grammar or writing, potentially due to interference effects. After the phase transition, there is no longer a detectable pattern regarding differential progress across the different subskills. A Jarynowski PAN metrics, 16.05.2024 35

36. A Jarynowski PAN metrics, 16.05.2024 36 Longitudinally

37. The regime breaking point was identified using an optimization research task (maximising t -statistics of mean difference between grammar and reading improvement). Then, we transform improvements to detrend and normalise the signal (subtracting the mean of improvement in a given dimension for the whole observation time, and dividing the difference by the same mean). Finally, we apply paired t -tests for the mean difference between improvement in grammar and reading (after transformation) separately for both regimes, and non-paired t -tests to compare the same detrended dimension of improvement across the phases. Pedagogical implications: At least in the initial stages, instructional materials should not be L1-agnostic A Jarynowski PAN metrics, 16.05.2024 37 Detrended grammar and reading Initial phase Later phase Mean difference (grammar vs reading) −0.200*** ( p = .009) 0.012 ( p = .496) Latinisation vs standard phase

38. Thank you very much for listening! We are open to collaboration! References: https://peerlang.ils.uw.edu.pl/publications/ Paradowski, M.B., Jarynowski, A., Czopek, K. & Jelińska, M. (2021). Peer interactions and second language learning: The contributions of Social Network Analysis in Study Abroad vs At-Home environments. In: R. Mitchell & H. Tyne (Eds.), Language, Mobility and Study Abroad in the Contemporary European Context (pp. 99–116). New York: Routledge. 10.4324/9781003087953-8 Paradowski, M.B., Jarynowski, A., Jelińska, M. & Czopek, K. (2021). Out-of-class peer interactions matter for second language acquisition during short-term overseas sojourns: The contributions of Social Network Analysis. Language Teaching , 54 (1), 139–143. 10.1017/S0261444820000580 Paradowski, M.B., Ochab, J.K., Cierpich-Kozieł, A. & Chen, C.-C. (2022) How output outweighs input and interlocutors matter for Study-Abroad SLA: Novel insights from Social Network Analysis of learner interactions. The Modern Language Journal , 106 (4), 694–725. 10.1111/modl.12811 Best of MLJ award 💥 Paradowski, M.B., Czuba, M. & Bródka, P. (2024, forthc.) Peer interaction dynamics and L2 learning trajectories during study abroad: A longitudinal investigation using dynamic computational social network analysis. Language Learning . https://uw.academia.edu/paradowski/ A Jarynowski PAN metrics, 16.05.2024 38

39. A Jarynowski PAN metrics, 16.05.2024 39 Comfort in communication with different groups of interlocutors The refugee students are most satisfied with their TL communication with neighbours and volunteers, less so in service encounters and the workplace, and the least in the administrative sphere

40. Highest proportion of Polish language use in text messages A Jarynowski PAN metrics, 16.05.2024 40

41. Conceivable explanations: respondents may be underestimating the proportion of time when they actually switch (back) into their L 1 , whether unbeknownst or to provide the ‘expected’ answer and project a positive image of themselves seeing faces associated with a particular culture triggers easier processing in the congruent language (Hartsuiker & Declerck 2013) opportunities to fall back on the shared common denominator mean that calques and code-switching will be easier and less time-consuming ways out of communication breakdowns than attempts at repair in the TL same language background may lead to interference errors being unnoticed, perpetuated, spread and stabilised increased amounts of time spent with same L 1 -speakers naturally diminish the exposure to speakers hailing from other linguistic backgrounds, who may be providing more varied hence enriching input A Jarynowski PAN metrics, 16.05.2024 41

42. Conclusions and pedagogical implications By identifying social behaviours which can positively or negatively impact learners’ language attainment one can hope to accelerate students’ progress by looking at ways to enhance and encourage the beneficial ones while containing and discouraging the detrimental ones. Results may help teachers choose the optimal forms of in- and out-of-class activities ( cf. e.g. Chi 2009) , and help students themselves raise their awareness of both more and less straightforward relations between their interactions within a group and the effects of learning. A Jarynowski PAN metrics, 16.05.2024 42

43. Limitations and areas for future research prevalence of self-report bias in questionnaires; the Dunning-Kruger effect (Kruger & Dunning 1999; Ehrlinger & Dunning 2003) the classroom is not the sole place of interactions and exposure to language ( Lindgren & Muñoz 2013; out-of-class behaviour exhibits markedly different dynamics ; Ricca 2012 ) try to systematically tease apart whether native and other TL-interlocutors make a difference for language development ( Dewey, Ring, Gardner & Belnap (2013) did not identify a significant difference between the benefits of interacting with native-speaking peers—local Arab students—vs. extended network interactions—shopkeepers, custodians, etc.) A Jarynowski PAN metrics, 16.05.2024 43

44. Limitations and areas for future research common issue of a ceiling effect: students with high levels of proficiency at the commencement of the measurement naturally have less room to show their progress than those who start out lower – not merely because there are fewer areas for them to catch up in (and the linguistic subtleties are harder to come by in the input), but also because progression along higher levels of the CEFR scale requires incomparably more time A Jarynowski PAN metrics, 16.05.2024 44