��������Ƶ

Please register if you want to attend.

Time: Thursday 26 March (15:00-17:00)

Location: s.KS 203

Abstract

A much-debated question in linguistics is whether learning language requires alanguage-specific learning capacity or can arise from input alone. Neurallanguage models (NLMs) can greatly influence this debate as they learn solelyfrom input and their inductive biases, without built-in linguisticrepresentations. Recent work has examined how NLM’s handle various grammaticalphenomena, including syntactic island constraints, which block filler-gapdependency formation in structures like wh-phrases.

(1)   *What_i do you wonder [wh

whether John bought _i]?

Although island violations like (1) rarely occur in language input andNLMs cannot fall back on built-in linguistic knowledge, studies show that theycan model these constraints successfully. However, these studies typically onlyassume a correspondence between model and human behavior, and focus almostexclusively on English. This study addresses these gaps by directly comparingNLMs’ and humans’ behavior on wh-islands in English and Dutch.

We introduce two key improvements. First, five modelsand >70 human participants were presented with the same sentencesmanipulated for presence of island, gap, and filler. By comparingmodel-assigned probabilities with human acceptability judgments, we testwhether NLMs represent a wh-island sensitivity in a human-like way. Second,we extend this approach cross-linguistically, contrasting English (SVO) andDutch (SOV), two related languages differing in word order. Our resultsreplicate previous findings that NLMs show a wh-island sensitivitycomparable to English participants. However, the same pattern does notgeneralize to Dutch: while Dutch participants showed a strong wh-islandsensitivity similar to English participants and models, Dutch NLMs did not. Thesefindings suggest that NLMs’ apparent success on English does notstraightforwardly extend across languages. Cross-linguistic evidence istherefore crucial before NLMs can be claimed to bear on the human capacity forgrammar learning. 

Bio

Michelle Suijkerbuijk completed a Bachelor in Linguistics and a Research Master in Linguistics and Communication sciences at the Radboud University in Nijmegen, before starting as a PhD candidate in 2022 at the Centre for Language Studies at the same university. Her PhD project is a continuation of her Master’s thesis and is called Deep Learning Across Languages. She is supervised by dr. Stefan Frank and dr. Peter de Swart and is part of the Grammar and Cognition research group. In her PhD project, she investigates whether computational models can model a specific grammatical phenomenon (namely the syntactic island constraint) comparable to human speakers across different languages with different word orders and varying morphological complexity (e.g. Dutch and Turkish).

Please register  if you wish to attend.

Time: Tuesday 16/12/2025, 13:00-15:00

Location: S.M.107

Abstract 

In individuals with language disorders, studies consistently report increased disfluencies – both in children with Developmental Language Disorder (DLD) and in adults with aphasia (Befi-Lopes et al., 2014; Saharoui & Martínez-Ferreiro, 2021). However, if there are studies on typically developing children (Bellifemine, 2024) and adults (Özkan et al., 2023), little is known about how these populations manage communicative breakdowns multimodally to repair their speech. We examine the way adults with non-fluent aphasia and children with expressive DLD, compared to  healthy controls, organize self-repair sequences through gestures and pauses. Participants included 22  French-speaking children with expressive DLD (aged 7-10) and 22 age- and gender-matched controls, filmed during a narrative task; and seven adults with non-fluent aphasia (aged 34–80) with age-matched controls, filmed during spontaneous conversation. Self-repairs (lexical, syntactic, phonological), gestures (referential, pragmatic), and pauses (filled, silent) were annotated. Each self-repair was analyzed in relation to its gestural and temporal construction. Results show that typically developing children produced more non-referential beat gestures during selfrepairs, whereas children with DLD relied primarily on referential gestures. Adults produced both gesture types equally. No difference in pauses was observed between the two groups of children, whereas adults with aphasia produced longer pauses than healthy controls. Four multimodal patterns emerged, defined by the coordination of pauses and gesture types. These patterns were shared by children with and without DLD but differentiated adults with and without aphasia. Thus, while disfluencies such as pauses and self-repairs are not pathological markers in children, they may serve as such in adults.

References 

Befi-Lopes, D. M., Cáceres-Assenço, A. M., Marques, S. F., & Vieira, M. (2014). School-age children with specific language impairment produce more speech disfluencies than their peers. CoDAS, 26(6), 439–443. Bellifemine, C. (2024). Non-referential beat gestures and their function of self-repair in children's narratives. Journal of Child Language Acquisition and Development-JCLAD, 12(1), 963-987.Özkan, E. E., Healey, P. G. T., Gurion, T., Hough, J., & Jamone, L. (2023). Speakers Raise Their Hands and Head During Self-Repairs in Dyadic Conversations. IEEE Transactions on Cognitive and Developmental Systems, 15(4), 1993-2003.Sahraoui, H. & Martínez-Ferreiro, S. (2021). Disfluency patterns in the connected discourse of individuals with agrammatic aphasia, Virtual Meeting: 9th Novi Sad Workshop on Psycholinguistic, Neurolinguistic and Clinical Linguistic Research, 30th October 2021.

Bio

Doctor of Language Sciences (Université Sorbonne Nouvelle, Paris), I have been an Associate Professor at Université de Lorraine (Metz) since 2024, within the UR 3476 CREM Research Lab. My PhD thesis focused on the relationship between syntactic complexity in discourse, gesture use in children with and without Developmental Language Disorder (DLD), and the influence of activity type.Over the years, my research interests have broadened. They now center on interaction analysis, syntax, reference management, and disfluencies in both typical and atypical language, in children and adults—always approached from a multimodal perspective.I have worked with children with neurodevelopmental disorders during my Master’s and PhD studies, and contributed to data collection from adults with aphasia for the AADI project (Aphasie, Analyse du Discours en Interaction). I am a member of the RaProChe project (Recherche-Action Professionnel·les-Chercheur·es), which aims to observe childcare professionals’ child-directed speech to facilitate language acquisition. I am also a member of the DinLang project (Dîners familiaux, Langage et multimodalité), which studies multimodal, multi-party interactions between adults and children during family meals.I have supervised Master’s theses on multimodal and verbal behaviors in children with Autism Spectrum Disorder (ASD) and adults with Aphasia, Schizophrenia, and Alzheimer’s Disease (AD). I currently co-supervise two PhD theses on discourse coherence in schizophrenia and non-verbal training in AD.

Time: Friday 07/11/2025 at 14h:00 

Location: S.M. 104

Abstract

​Over the past few decades, cognitive-functional approaches have transformed how we think about the primitive units of language. Gone are the rigid syntax trees and mechanical slot-and-filler templates of earlier theories; in their place stand constructions: flexible, multidimensional pairings of form and function. This shift has opened up new research avenues for understanding language as a meaning-driven, dynamic, and evolving system. At the same time, the spectacular success of contemporary language models has shown that data-driven learning can capture many of the surface regularities with astonishing fluency. Yet this very success has also sharpened an old question: how can we formalize the richness of linguistic structure in ways that are both scientifically insightful and computationally scalable?

In this presentation, I will argue that part of the answer lies in embracing population thinking and complex systems perspectives. Rather than treating constructions as static entities in the grammar, we can view them as members of evolving populations that interact, compete, and cooperate within a linguistic ecosystem. Through these locally situated interactions, higher-order structures emerge – patterns that cannot be reduced to the sum of their parts. Drawing on insights from Construction Grammar, cognitive science, and agent-based modelling, I will show how this ecological view helps us rethink functional diversity, network organization, and the elusive “strata” of linguistic structure as emergent layers rather than fixed hierarchies.

Bio

Remi van Trijp is a researcher at the Sony Computer Laboratories in Paris, where he uses agent-based experiments and computational modelling to explore the intersections of linguistics, complex systems science, human-centric artificial intelligence, and language philosophy. He is one of the chief architects of Fluid Construction Grammar, an open-source platform for exploring constructional language processing as well as the emergence and evolution of language. Delighted to return to his alma mater – where he earned both his Master’s and doctoral degree – he looks forward to exchanging ideas with fellow researchers and the next generation of curious minds.

Click to register.

Time - location

Thursday 03/07/2025 - Room S.R.012

Schedule:

11:30 - 12:15: presentation

12:15 - 12:30: coffee break

12:30 - 13:15: discussion

Abstract 

In this talk, I will give an overview of work that shows how different aspects of bilingual syntax can be modeled as error-based implicit learning. I will first introduce the Dual-path model (Chang et al., 2006) and the Bilingual Dual-path model (Tsoukala et al., 2021) and I will give a short overview of the previous work that has been done with these models (Khoe & Frank, 2024). I will then discuss in more depth the simulated experiments that I have conducted on cross-language structural priming using the Bilingual Dual-path model (Khoe et al., 2023). The results of these simulations increase support for implicit learning as an account for structural priming by extending the account from within-language to crosslanguage structural priming. Next, I will present simulated ERP experiments on P600/N400 effects in L2 learners in response to syntactic violations (Verwijmeren et al., 2023). Then, I will give an overview of ongoing work on how proficiency might modulate the strength of cross-language structural priming in the model. Finally, I will present a study on how codeswitching increases cross-language structural priming in the Bilingual Dual-path model and in Spanish-English bilinguals.

References 

- Chang, F., Dell, G. S., & Bock, K. (2006). Becoming syntactic. Psychological Review, 113(2), 234. https://doi.org/10.1037/0033-295X.113.2.234 

- Khoe, Y. H., Tsoukala, C., Kootstra, G.J., Frank, S. (2023). Is structural priming between different languages a learning effect? Modelling priming as error-driven implicit learning, Language, Cognition and Neuroscience, 38 (pp. 537-557), https://doi.org/10.1080/23273798.2021.1998563 

- Khoe, Y.H., Frank, S.L. (2024). The Bilingual Dual-path model: Simulating bilingual production, comprehension, and development, Linguistic Approaches to Bilingualism, https://doi.org/10.1080/23273798.2021.1998563 

- Tsoukala, C., Broersma, M., Van Den Bosch, A., & Frank, S. L. (2021). Simulating codeswitching using a neural network model of bilingual sentence production. Computational Brain & Behavior, 4, 87–100. https://doi.org/10.1007/s42113-020-00088-6 

- Verwijmeren, S., Frank, S.L., Fitz, H. & Khoe, Y.H. (2023). A neural network simulation of event-related potentials in response to syntactic violations in second-language learning, Proceedings of the 21st International Conference on Cognitive Modelling, preprint: http://www.stefanfrank.info/pubs/ICCM_paper_Stephan.pdf

Bio

I recently started as a postdoc in the School of Psychology at the University of Birmingham. My research there focuses on implicit learning mechanisms for syntactic production in healthy older adults. It is part of a project that is funded by a Leverhulme Trust Research Project Grant, supervised by Katrien Segaert, Linda Wheeldon (University of Agder, Norway) and Evelien Heyselaar (Radboud University, Netherlands). Until I defend my thesis, I am also still a PhD candidate at the Centre for Language Studies (CLS) at Radboud University in the Netherlands. My PhD research focused on understanding how people learn, process and produce (multiple) language(s) by using traditional experimental techniques paired with computational cognitive models. The PhD is supervised by Stefan Frank, Gerrit Jan Kootstra and Rob Schoonen.