Markus Freitag
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Markus Freitag has a total of 131 co-authors. Here are the top co-authors by number of shared publications:
- Maximilian Filsinger
Institute of Political Science, University of Bern
11 shared publications - Julian Erhardt
Institute of Political Science, University Bern, Bern, Switzerland
10 shared publications - David Grangier
Google research
7 shared publications - Kathrin Ackermann
Department of Social Sciences, University of Siegen, Siegen, Germany
6 shared publications - Steffen Wamsler
Leibniz Institute for Educational Trajectories
6 shared publications - Daniel Deutsch
ETH Zurich
5 shared publications
Latest Publications
- Open Access
- 27 May 2026
- Published by Wiley in Political Psychology
- Vol. 47 (4)
Depressive symptoms and populism: Evidence from European countries
In recent years, depression has entered the research agenda of political psychology, emerging as a meaningful psychological correlate of diverse political attitudes and behaviors. Surprisingly, however, its link to populism—the political phenomenon that has probably attracted most public and scholarly attention over the past several years—has not yet been studied. We address this notable blind spot and provide a first exploratory evaluation of the relationship between depressive symptoms and the inclination to populism. To do so, we make use of two European datasets (CRONOS‐2 and ESS 11) that include a validated, epidemiological short scale to measure depressive symptoms (CES‐D8) and allow us to consider both populist attitudes and populist voting as key individual‐level manifestations of populism. Our analyses show that individuals experiencing depressive symptoms tend to exhibit a stronger affinity for populist ideas and are more likely to support populist parties. Our study thus underlines the importance of prevention, early detection, and treatment measures targeting the widespread “black dog” of depression—also in the interest of preserving and promoting democratic health.
- 27 May 2026
- Published by Wiley in Political Psychology
- Vol. 47 (4)
Open AccessDepressive symptoms and populism: Evidence from European countries
In recent years, depression has entered the research agenda of political psychology, emerging as a meaningful psychological correlate of diverse political attitudes and behaviors. Surprisingly, however, its link to populism—the political phenomenon that has probably attracted most public and scholarly attention over the past several years—has not yet been studied. We address this notable blind spot and provide a first exploratory evaluation of the relationship between depressive symptoms and the inclination to populism. To do so, we make use of two European datasets (CRONOS‐2 and ESS 11) that include a validated, epidemiological short scale to measure depressive symptoms (CES‐D8) and allow us to consider both populist attitudes and populist voting as key individual‐level manifestations of populism. Our analyses show that individuals experiencing depressive symptoms tend to exhibit a stronger affinity for populist ideas and are more likely to support populist parties. Our study thus underlines the importance of prevention, early detection, and treatment measures targeting the widespread “black dog” of depression—also in the interest of preserving and promoting democratic health.
- Open Access
- 14 January 2026
- Published by Taylor & Francis in German Politics
- p. 1-21
Under Attack: How Social Threats Push Negative Beliefs about Scientists in Germany
- 14 January 2026
- Published by Taylor & Francis in German Politics
- p. 1-21
Open AccessUnder Attack: How Social Threats Push Negative Beliefs about Scientists in Germany
- Open Access
Deconstructing Self-Bias in LLM-generated Translation Benchmarks
As large language models (LLMs) begin to saturate existing benchmarks, automated benchmark creation using LLMs (LLM as a benchmark) has emerged as a scalable alternative to slow and costly human curation. While these generated test sets have to potential to cheaply rank models, we demonstrate a critical flaw. LLM generated benchmarks systematically favor the model that created the benchmark, they exhibit self bias on low resource languages to English translation tasks. We show three key findings on automatic benchmarking of LLMs for translation: First, this bias originates from two sources: the generated test data (LLM as a testset) and the evaluation method (LLM as an evaluator), with their combination amplifying the effect. Second, self bias in LLM as a benchmark is heavily influenced by the model's generation capabilities in the source language. For instance, we observe more pronounced bias in into English translation, where the model's generation system is developed, than in out of English translation tasks. Third, we observe that low diversity in source text is one attribution to self bias. Our results suggest that improving the diversity of these generated source texts can mitigate some of the observed self bias.
Open AccessDeconstructing Self-Bias in LLM-generated Translation Benchmarks
As large language models (LLMs) begin to saturate existing benchmarks, automated benchmark creation using LLMs (LLM as a benchmark) has emerged as a scalable alternative to slow and costly human curation. While these generated test sets have to potential to cheaply rank models, we demonstrate a critical flaw. LLM generated benchmarks systematically favor the model that created the benchmark, they exhibit self bias on low resource languages to English translation tasks. We show three key findings on automatic benchmarking of LLMs for translation: First, this bias originates from two sources: the generated test data (LLM as a testset) and the evaluation method (LLM as an evaluator), with their combination amplifying the effect. Second, self bias in LLM as a benchmark is heavily influenced by the model's generation capabilities in the source language. For instance, we observe more pronounced bias in into English translation, where the model's generation system is developed, than in out of English translation tasks. Third, we observe that low diversity in source text is one attribution to self bias. Our results suggest that improving the diversity of these generated source texts can mitigate some of the observed self bias.
- Open Access
Feeding Two Birds or Favoring One? Adequacy-Fluency Tradeoffs in Evaluation and Meta-Evaluation of Machine Translation
We investigate the tradeoff between adequacy and fluency in machine translation. We show the severity of this tradeoff at the evaluation level and analyze where popular metrics fall within it. Essentially, current metrics generally lean toward adequacy, meaning that their scores correlate more strongly with the adequacy of translations than with fluency. More importantly, we find that this tradeoff also persists at the meta-evaluation level, and that the standard WMT meta-evaluation favors adequacy-oriented metrics over fluency-oriented ones. We show that this bias is partially attributed to the composition of the systems included in the meta-evaluation datasets. To control this bias, we propose a method that synthesizes translation systems in meta-evaluation. Our findings highlight the importance of understanding this tradeoff in meta-evaluation and its impact on metric rankings.
Open AccessFeeding Two Birds or Favoring One? Adequacy-Fluency Tradeoffs in Evaluation and Meta-Evaluation of Machine Translation
We investigate the tradeoff between adequacy and fluency in machine translation. We show the severity of this tradeoff at the evaluation level and analyze where popular metrics fall within it. Essentially, current metrics generally lean toward adequacy, meaning that their scores correlate more strongly with the adequacy of translations than with fluency. More importantly, we find that this tradeoff also persists at the meta-evaluation level, and that the standard WMT meta-evaluation favors adequacy-oriented metrics over fluency-oriented ones. We show that this bias is partially attributed to the composition of the systems included in the meta-evaluation datasets. To control this bias, we propose a method that synthesizes translation systems in meta-evaluation. Our findings highlight the importance of understanding this tradeoff in meta-evaluation and its impact on metric rankings.
- Open Access
Who would take part in a pandemic preparedness cohort study? The role of vaccine-related affective polarisation: cross-sectional survey
Cohort studies are an important tool for public health research about emerging infectious diseases. Willingness to participate in research is associated with multiple factors. Little is known about the role of people’s feelings, especially in the aftermath of the COVID-19 pandemic. Affective polarisation, an affinity for people with similar attitudes to oneself and hostility toward those with opposing views, is a measure of feelings about issues, including COVID-19 vaccination. The aim of this study was to investigate factors associated with willingness to participate in a future household-based cohort study on pandemic preparedness. We did a cross-sectional online survey in the Canton of Bern, Switzerland. We invited a random sample of persons aged 18+ years from 15,000 private households. We asked about willingness to take part and collected data about demographic, social and household characteristics, and affective polarisation related to COVID-19 vaccination. We performed univariable and multivariable analyses, weighted by household size to estimate odds ratios (with 95% confidence intervals, CI). Among responders, 49.8% (95% CI 47.9-51.8, weighted proportion) were willing to take part in a cohort study. In multivariable analysis, higher educational level (adjusted odds ratio 2.48, 95% CI 1.81-3.39) and higher monthly income (1.92, 1.39-2.65) were most strongly associated with higher willingness to participate. Opposition to COVID-19 vaccination was associated with lower willingness to participate (0.53, 0.39-0.72). Affective polarisation modified the relationship between vaccination attitudes and willingness to participate. Compared with non-polarised vaccination supporters, polarised supporters were more willing to participate (marginal adjusted odds ratio 1.51, 1.05-2.16), whereas willingness to participate was lower among both non-polarised (0.53, 0.32-0.86) and affectively polarised (0.26, 0.12-0.56) vaccination opposers. Willingness to participate in a cohort study was moderate. Following the COVID-19 pandemic, addressing affective polarisation, as well as socioeconomic factors, is needed to improve participation in pandemic preparedness research.
Open AccessWho would take part in a pandemic preparedness cohort study? The role of vaccine-related affective polarisation: cross-sectional survey
Cohort studies are an important tool for public health research about emerging infectious diseases. Willingness to participate in research is associated with multiple factors. Little is known about the role of people’s feelings, especially in the aftermath of the COVID-19 pandemic. Affective polarisation, an affinity for people with similar attitudes to oneself and hostility toward those with opposing views, is a measure of feelings about issues, including COVID-19 vaccination. The aim of this study was to investigate factors associated with willingness to participate in a future household-based cohort study on pandemic preparedness. We did a cross-sectional online survey in the Canton of Bern, Switzerland. We invited a random sample of persons aged 18+ years from 15,000 private households. We asked about willingness to take part and collected data about demographic, social and household characteristics, and affective polarisation related to COVID-19 vaccination. We performed univariable and multivariable analyses, weighted by household size to estimate odds ratios (with 95% confidence intervals, CI). Among responders, 49.8% (95% CI 47.9-51.8, weighted proportion) were willing to take part in a cohort study. In multivariable analysis, higher educational level (adjusted odds ratio 2.48, 95% CI 1.81-3.39) and higher monthly income (1.92, 1.39-2.65) were most strongly associated with higher willingness to participate. Opposition to COVID-19 vaccination was associated with lower willingness to participate (0.53, 0.39-0.72). Affective polarisation modified the relationship between vaccination attitudes and willingness to participate. Compared with non-polarised vaccination supporters, polarised supporters were more willing to participate (marginal adjusted odds ratio 1.51, 1.05-2.16), whereas willingness to participate was lower among both non-polarised (0.53, 0.32-0.86) and affectively polarised (0.26, 0.12-0.56) vaccination opposers. Willingness to participate in a cohort study was moderate. Following the COVID-19 pandemic, addressing affective polarisation, as well as socioeconomic factors, is needed to improve participation in pandemic preparedness research.
- Open Access
- 25 February 2025
- Published by Springer Nature in Humanities and Social Sciences Communications
- Vol. 12 (1) , 1-12
Divided by vaccination? Evaluating the intergroup conflict between pro- and anti-vaccination groups in the post-pandemic era
The vaccination against COVID-19 has torn societies apart. Against this background we evaluate three interrelated research questions: (1) does vaccination polarize citizens even after the COVID-19 pandemic has faded; (2) do opinions about vaccination correlate with group formation and identification, and (3) do we observe opinion-based affective polarization regarding vaccination in the post-pandemic era? Based on two original surveys from Switzerland in early 2022 and late 2023, our results highlight that respondents have distinct opinions about vaccination, but that only pro-vaccination respondents have formed an opinion identity. We also observe an asymmetric affective polarization: pro-vaccination respondents show higher levels of liking toward other pro-vaccination respondents but significant dislike toward anti-vaccination respondents, while the same does not hold true for anti-vaccination respondents. Overall, affective polarization toward vaccination is less pronounced in the aftermath of the health crisis than during the COVID-19 pandemic. However, at a time when globalization is boosting the threat of pandemics, caution is warranted, as an increasing salience of vaccination could widen the divide again.
- 25 February 2025
- Published by Springer Nature in Humanities and Social Sciences Communications
- Vol. 12 (1) , 1-12
Open AccessDivided by vaccination? Evaluating the intergroup conflict between pro- and anti-vaccination groups in the post-pandemic era
The vaccination against COVID-19 has torn societies apart. Against this background we evaluate three interrelated research questions: (1) does vaccination polarize citizens even after the COVID-19 pandemic has faded; (2) do opinions about vaccination correlate with group formation and identification, and (3) do we observe opinion-based affective polarization regarding vaccination in the post-pandemic era? Based on two original surveys from Switzerland in early 2022 and late 2023, our results highlight that respondents have distinct opinions about vaccination, but that only pro-vaccination respondents have formed an opinion identity. We also observe an asymmetric affective polarization: pro-vaccination respondents show higher levels of liking toward other pro-vaccination respondents but significant dislike toward anti-vaccination respondents, while the same does not hold true for anti-vaccination respondents. Overall, affective polarization toward vaccination is less pronounced in the aftermath of the health crisis than during the COVID-19 pandemic. However, at a time when globalization is boosting the threat of pandemics, caution is warranted, as an increasing salience of vaccination could widen the divide again.
- Open Access
WMT24++: Expanding the Language Coverage of WMT24 to 55 Languages & Dialects
As large language models (LLM) become more and more capable in languages other than English, it is important to collect benchmark datasets in order to evaluate their multilingual performance, including on tasks like machine translation (MT). In this work, we extend the WMT24 dataset to cover 55 languages by collecting new human-written references and post-edits for 46 new languages and dialects in addition to post-edits of the references in 8 out of 9 languages in the original WMT24 dataset. The dataset covers four domains: literary, news, social, and speech. We benchmark a variety of MT providers and LLMs on the collected dataset using automatic metrics and find that LLMs are the best-performing MT systems in all 55 languages. These results should be confirmed using a human-based evaluation, which we leave for future work.
Open AccessWMT24++: Expanding the Language Coverage of WMT24 to 55 Languages & Dialects
As large language models (LLM) become more and more capable in languages other than English, it is important to collect benchmark datasets in order to evaluate their multilingual performance, including on tasks like machine translation (MT). In this work, we extend the WMT24 dataset to cover 55 languages by collecting new human-written references and post-edits for 46 new languages and dialects in addition to post-edits of the references in 8 out of 9 languages in the original WMT24 dataset. The dataset covers four domains: literary, news, social, and speech. We benchmark a variety of MT providers and LLMs on the collected dataset using automatic metrics and find that LLMs are the best-performing MT systems in all 55 languages. These results should be confirmed using a human-based evaluation, which we leave for future work.
- 13 February 2025
- Published by SAGE Publications in Politics
- Vol. 45 (4) , 594-615
Divided by morality? Moral foundations of affective polarisation during hard times
In this article, we examine the deeper causes of a key group conflict during the COVID-19 pandemic: attitudes towards the COVID-19 vaccination. Extending research on affective polarisation beyond partisanship, we analyse how people perceive the external groups of vaccination supporters and opponents. Furthermore, we investigate how their own moral foundations influence these perceptions. Using six samples from six European countries, our analyses reveal an affective divide between supporters and opponents of COVID-19 vaccinations. Most interestingly and contrary to conventional expectations, we find that individualising rather than binding foundations intensify affective polarisation between pro- and anti-vaccination supporters in all countries. It seems that especially in times of pandemic threat, COVID-19 vaccination supporters and opponents alike feel that their core moral foundations, in particular, fairness are being challenged by their respective out-groups. Our findings could be used to develop depolarisation interventions involving moral framing.
- 13 February 2025
- Published by SAGE Publications in Politics
- Vol. 45 (4) , 594-615
Divided by morality? Moral foundations of affective polarisation during hard times
In this article, we examine the deeper causes of a key group conflict during the COVID-19 pandemic: attitudes towards the COVID-19 vaccination. Extending research on affective polarisation beyond partisanship, we analyse how people perceive the external groups of vaccination supporters and opponents. Furthermore, we investigate how their own moral foundations influence these perceptions. Using six samples from six European countries, our analyses reveal an affective divide between supporters and opponents of COVID-19 vaccinations. Most interestingly and contrary to conventional expectations, we find that individualising rather than binding foundations intensify affective polarisation between pro- and anti-vaccination supporters in all countries. It seems that especially in times of pandemic threat, COVID-19 vaccination supporters and opponents alike feel that their core moral foundations, in particular, fairness are being challenged by their respective out-groups. Our findings could be used to develop depolarisation interventions involving moral framing.
- Open Access
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 269-285
Feeding Two Birds or Favoring One? Adequacy–Fluency Tradeoffs in Evaluation and Meta-Evaluation of Machine Translation
Behzad Shayegh, Jan-Thorsten Peter, David Vilar, Tobias Domhan, Juraj Juraska, Markus Freitag, Lili Mou. Proceedings of the Tenth Conference on Machine Translation. 2025.
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 269-285
Open AccessFeeding Two Birds or Favoring One? Adequacy–Fluency Tradeoffs in Evaluation and Meta-Evaluation of Machine Translation
Behzad Shayegh, Jan-Thorsten Peter, David Vilar, Tobias Domhan, Juraj Juraska, Markus Freitag, Lili Mou. Proceedings of the Tenth Conference on Machine Translation. 2025.
- Open Access
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 957-968
MetricX-25 and GemSpanEval: Google Translate Submissions to the WMT25 Evaluation Shared Task
Juraj Juraska, Tobias Domhan, Mara Finkelstein, Tetsuji Nakagawa, Geza Kovacs, Daniel Deutsch, Pidong Wang, Markus Freitag. Proceedings of the Tenth Conference on Machine Translation. 2025.
- 1 January 2025
- Published by Association for Computational Linguistics (ACL)
- p. 957-968
Open AccessMetricX-25 and GemSpanEval: Google Translate Submissions to the WMT25 Evaluation Shared Task
Juraj Juraska, Tobias Domhan, Mara Finkelstein, Tetsuji Nakagawa, Geza Kovacs, Daniel Deutsch, Pidong Wang, Markus Freitag. Proceedings of the Tenth Conference on Machine Translation. 2025.