Stefan L. Frank
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Stefan L. Frank has a total of 202 co-authors. Here are the top co-authors by number of shared publications:
Josef Pfeilschifter
Pharmazentrum frankfurt/ZAFES, Klinikum der Johann Wolfgang Goethe-Universität, Theodor-Stern-Kai 7, D-60590 Frankfurt am Main, Germany
21 shared publicationsHeiko Kämpfer
Pharmazentrum frankfurt/ZAFES, Klinikum der Johann Wolfgang Goethe-Universität, Frankfurt am Main, Germany
9 shared publicationsItamar Goren
Pharmazentrum frankfurt/ZAFES, Klinikum der Johann Wolfgang Goethe-Universität, Frankfurt am Main, Germany
9 shared publicationsAntal van Den Bosch
Max Planck Institute for Psycholinguistics, Nijmegen, Netherlands
7 shared publications- Gabriella Vigliocco
Department of Experimental Psychology, University College London
6 shared publications Peter de Swart
6 shared publications
Latest Publications
- Open Access
- 21 August 2026
- Published by SAGE Publications in International Journal of Bilingualism
Do Not Use Inferential Statistics to Evaluate Quality of Matching
Aims and Objectives: Quantitative research in bilingualism often involves matching between groups on one or more potentially confounding variables. A common approach to assessing the quality of matching is to apply inferential statistics to the variable(s). I argue and demonstrate that this approach is theoretically and practically problematic. Approach: I use simulated data from fictitious studies to demonstrate the problems that can arise from the use of inferential statistics for evaluating matching quality. I also analyse statistical practices for group matching in published articles from two bilingualism journals. Data and Analysis: The simulated data are analysed by t -tests, but this is merely for ease of illustration, as any inferential statistical analysis leads to the same problem. The statistical practices in published articles are analysed with logistic regression. Conclusions: There is no purely quantitative, objective alternative to inferential statistics. Matching quality can only be evaluated by a comparison of the sample distributions, informed by expert knowledge about the research topic. Originality: Although I am not the first to identify the issue, it has not yet been laid out for an audience of researchers studying bilingualism. Also, earlier attempts to improve the evaluation of matching quality proposed objective quantities as an alternative, which I argue is not viable. Significance/Implications: If researchers no longer take the outcome of an inferential statistical test as a measure of matching quality and instead present rich, descriptive statistics and embrace the subjective nature of the issue of matching, this will contribute to improving the reliability of published findings.
- 21 August 2026
- Published by SAGE Publications in International Journal of Bilingualism
Open AccessDo Not Use Inferential Statistics to Evaluate Quality of Matching
Aims and Objectives: Quantitative research in bilingualism often involves matching between groups on one or more potentially confounding variables. A common approach to assessing the quality of matching is to apply inferential statistics to the variable(s). I argue and demonstrate that this approach is theoretically and practically problematic. Approach: I use simulated data from fictitious studies to demonstrate the problems that can arise from the use of inferential statistics for evaluating matching quality. I also analyse statistical practices for group matching in published articles from two bilingualism journals. Data and Analysis: The simulated data are analysed by t -tests, but this is merely for ease of illustration, as any inferential statistical analysis leads to the same problem. The statistical practices in published articles are analysed with logistic regression. Conclusions: There is no purely quantitative, objective alternative to inferential statistics. Matching quality can only be evaluated by a comparison of the sample distributions, informed by expert knowledge about the research topic. Originality: Although I am not the first to identify the issue, it has not yet been laid out for an audience of researchers studying bilingualism. Also, earlier attempts to improve the evaluation of matching quality proposed objective quantities as an alternative, which I argue is not viable. Significance/Implications: If researchers no longer take the outcome of an inferential statistical test as a measure of matching quality and instead present rich, descriptive statistics and embrace the subjective nature of the issue of matching, this will contribute to improving the reliability of published findings.
- Open Access
- 1 July 2026
- Published by Center for Open Science in PsyArXiv
Neural Language Models’ sensitivity to long-distance dependencies and their island constraints is not universal: a comparison between English and Dutch humans and models
A much-debated question in linguistics is whether learning language, specifically grammar, requires a language-specific learning capacity or can be achieved from input alone through domain-general mechanisms. The current study adds to this debate by testing whether Neural language models (NLMs), which learn solely from input and their non-linguistic inductive biases, without any built-in linguistic representations, can model the wh-island constraint. We improve on previous research by testing both NLMs and human participants in the same experimental set-up and directly comparing their results, and doing this cross-linguistically in both English and Dutch. Our results show that the NLMs can successfully model long-distance filler-gap dependencies and the wh-island constraint thereon in English but not in Dutch, opposing the arguments in favor of learning from input alone through domain-general mechanisms.
- 1 July 2026
- Published by Center for Open Science in PsyArXiv
Open AccessNeural Language Models’ sensitivity to long-distance dependencies and their island constraints is not universal: a comparison between English and Dutch humans and models
A much-debated question in linguistics is whether learning language, specifically grammar, requires a language-specific learning capacity or can be achieved from input alone through domain-general mechanisms. The current study adds to this debate by testing whether Neural language models (NLMs), which learn solely from input and their non-linguistic inductive biases, without any built-in linguistic representations, can model the wh-island constraint. We improve on previous research by testing both NLMs and human participants in the same experimental set-up and directly comparing their results, and doing this cross-linguistically in both English and Dutch. Our results show that the NLMs can successfully model long-distance filler-gap dependencies and the wh-island constraint thereon in English but not in Dutch, opposing the arguments in favor of learning from input alone through domain-general mechanisms.
- Open Access
- 30 April 2026
- Published by Center for Open Science in PsyArXiv
Why Produce Garden-Path Sentences? Evidence from Corpus Data across Registers
Garden-path sentences induce processing difficulty by requiring readers to revise an initial syntactic analysis (e.g., The policeman saw the lights were off). While garden-path comprehension has been widely studied, far less is known about why such sentences are produced in natural language. We thus address this gap by analyzing large-scale written-text corpora in English and Japanese across five registers. English Noun/Sentence garden-path constructions, arising from optional complementizer omission, were frequent in blogs and newspaper, suggesting that their emphasis on communicative efficiency promotes their use. Rather, Japanese Main Clause/Relative Clause garden-path constructions, which arise from grammatical constraints, were most frequent in legal texts, reflecting the register's tendency to favor complexity within a clause. Legal texts also exhibit the longest ambiguous regions across constructions, which can increase reanalysis cost. These findings suggest that grammatical properties and register-specific writing styles jointly shape the production of garden-path sentences in real-world language.
- 30 April 2026
- Published by Center for Open Science in PsyArXiv
Open AccessWhy Produce Garden-Path Sentences? Evidence from Corpus Data across Registers
Garden-path sentences induce processing difficulty by requiring readers to revise an initial syntactic analysis (e.g., The policeman saw the lights were off). While garden-path comprehension has been widely studied, far less is known about why such sentences are produced in natural language. We thus address this gap by analyzing large-scale written-text corpora in English and Japanese across five registers. English Noun/Sentence garden-path constructions, arising from optional complementizer omission, were frequent in blogs and newspaper, suggesting that their emphasis on communicative efficiency promotes their use. Rather, Japanese Main Clause/Relative Clause garden-path constructions, which arise from grammatical constraints, were most frequent in legal texts, reflecting the register's tendency to favor complexity within a clause. Legal texts also exhibit the longest ambiguous regions across constructions, which can increase reanalysis cost. These findings suggest that grammatical properties and register-specific writing styles jointly shape the production of garden-path sentences in real-world language.
- Open Access
- 15 April 2026
- Published by Center for Open Science in PsyArXiv
Do not use inferential statistics to evaluate quality of matching
Quantitative research in bilingualism often involves between-group comparisons in the presence of on one or more potentially confounding variables. There is no risk of confounding if the potential confounders are matched between groups. But when is a variable sufficiently matched? A common approach to assessing the quality of matching is to apply inferential statistics to the variable(s). In this paper, I argue that this approach is theoretically and practically problematic, and that there is no purely quantitative, objective alternative. Matching quality can only be evaluated by a comparison of the sample distributions, informed by expert knowledge about the research topic.
- 15 April 2026
- Published by Center for Open Science in PsyArXiv
Open AccessDo not use inferential statistics to evaluate quality of matching
Quantitative research in bilingualism often involves between-group comparisons in the presence of on one or more potentially confounding variables. There is no risk of confounding if the potential confounders are matched between groups. But when is a variable sufficiently matched? A common approach to assessing the quality of matching is to apply inferential statistics to the variable(s). In this paper, I argue that this approach is theoretically and practically problematic, and that there is no purely quantitative, objective alternative. Matching quality can only be evaluated by a comparison of the sample distributions, informed by expert knowledge about the research topic.
- Open Access
- 3 September 2025
- Published by Center for Open Science in PsyArXiv
Connectionist Models of Second Language Acquisition and Processing
Connectionist models treat language processing as the flow of activation through a network of simple units and the connections between them. Language learning is modeled as the adjustment of the strengths of these connections. Such models have led to important insights into second language acquisition and processing. This chapter introduces the fundamentals of connectionist models in cognitive science and reviews the most influential word- and sentence-level models of bilingualism.
- 3 September 2025
- Published by Center for Open Science in PsyArXiv
Open AccessConnectionist Models of Second Language Acquisition and Processing
Connectionist models treat language processing as the flow of activation through a network of simple units and the connections between them. Language learning is modeled as the adjustment of the strengths of these connections. Such models have led to important insights into second language acquisition and processing. This chapter introduces the fundamentals of connectionist models in cognitive science and reviews the most influential word- and sentence-level models of bilingualism.
- Open Access
- 25 May 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-11
MultiplEYE: Creating a multilingual eye-tracking-while-reading corpus
- 25 May 2025
- Published by Association for Computing Machinery (ACM)
- p. 1-11
Open AccessMultiplEYE: Creating a multilingual eye-tracking-while-reading corpus
- Open Access
- 10 March 2025
- Published by Center for Open Science in PsyArXiv
BLiMP-NL: A corpus of Dutch minimal pairs and acceptability judgements for language model evaluation
We present a corpus of 8400 Dutch sentence pairs, intended primarily for the grammatical evaluation of language models. Each pair consists of a grammatical sentence and a minimally different ungrammatical sentence. The corpus covers 84 paradigms, classified into 22 syntactic phenomena. Ten sentence pairs of each paradigm were created by hand, while the remaining ninety were generated semi-automatically and manually validated afterwards. Nine of the ten hand-crafted sentences of each paradigm are rated for acceptability by at least 30 participants each, and for the same 9 sentences reading times are recorded per word, through self-paced reading. Here, we report on the construction of the dataset, the measured acceptability ratings and reading times, as well as the extent to which a variety of language models can be used to predict both the ground-truth grammaticality and human acceptability ratings.
- 10 March 2025
- Published by Center for Open Science in PsyArXiv
Open AccessBLiMP-NL: A corpus of Dutch minimal pairs and acceptability judgements for language model evaluation
We present a corpus of 8400 Dutch sentence pairs, intended primarily for the grammatical evaluation of language models. Each pair consists of a grammatical sentence and a minimally different ungrammatical sentence. The corpus covers 84 paradigms, classified into 22 syntactic phenomena. Ten sentence pairs of each paradigm were created by hand, while the remaining ninety were generated semi-automatically and manually validated afterwards. Nine of the ten hand-crafted sentences of each paradigm are rated for acceptability by at least 30 participants each, and for the same 9 sentences reading times are recorded per word, through self-paced reading. Here, we report on the construction of the dataset, the measured acceptability ratings and reading times, as well as the extent to which a variety of language models can be used to predict both the ground-truth grammaticality and human acceptability ratings.
- Open Access
- 4 January 2025
- Published by California Digital Library (CDL) in Glossa Psycholinguistics
- Vol. 4 (1)
Revisiting processing complexity of nested and cross-serial dependencies
In two web-based experiments, we compare comprehension difficulty between Dutch and German sentences with clusters of two or three verbs. In Dutch, such sentences involve crossing dependencies, whereas these dependencies are nested in German. Replicating the seminal finding of Bach et al. (1986), we find that the crossing (Dutch) structure is easier to comprehend than the nested (German) structure, although we find a different pattern of results in terms of where this difficulty manifests. The results are in line with predictions from the Dependency Locality Theory.
- 4 January 2025
- Published by California Digital Library (CDL) in Glossa Psycholinguistics
- Vol. 4 (1)
Open AccessRevisiting processing complexity of nested and cross-serial dependencies
In two web-based experiments, we compare comprehension difficulty between Dutch and German sentences with clusters of two or three verbs. In Dutch, such sentences involve crossing dependencies, whereas these dependencies are nested in German. Replicating the seminal finding of Bach et al. (1986), we find that the crossing (Dutch) structure is easier to comprehend than the nested (German) structure, although we find a different pattern of results in terms of where this difficulty manifests. The results are in line with predictions from the Dependency Locality Theory.
- Open Access
- 1 October 2024
- Published by John Benjamins Publishing Company in Linguistic Approaches to Bilingualism
- Vol. 15 (4) , 571-599
The Bilingual Dual-path model
Experimental research has yielded many important psycholinguistic findings in bi-/multilingualism. However, cognitive computational models of sentence processing were limited to the single-language case until recently. In this methodological review, we discuss cognitive modelling work that uses the Bilingual Dual-path model to simulate experimental research on bilingual sentence production, processing, and development. As a detailed example of such work, we then report on new simulations conducted with the model. Finally, we suggest directions for future cognitive modelling research using this model.
- 1 October 2024
- Published by John Benjamins Publishing Company in Linguistic Approaches to Bilingualism
- Vol. 15 (4) , 571-599
Open AccessThe Bilingual Dual-path model
Experimental research has yielded many important psycholinguistic findings in bi-/multilingualism. However, cognitive computational models of sentence processing were limited to the single-language case until recently. In this methodological review, we discuss cognitive modelling work that uses the Bilingual Dual-path model to simulate experimental research on bilingual sentence production, processing, and development. As a detailed example of such work, we then report on new simulations conducted with the model. Finally, we suggest directions for future cognitive modelling research using this model.