Constructionist approaches to language posit that all linguistic knowledge needed for language comprehension and production can be captured as a network of form-meaning mappings, called constructions. Construction Grammars (CxGs) do not distinguish between words and grammar rules, but allow for mappings between forms and meanings of arbitrary complexity and degree of abstraction. CxGs are thereby able to uniformly capture the compositional and non-compositional aspects of language use, making the theory particularly attractive to researchers in the field of Natural Language Processing (NLP). CxG theories, for example, can serve as a valuable ‘lens’ to assess and investigate the abilities of today’s large language models, which lack explicit, theoretically grounded linguistic insights. At the same time, techniques from the field of NLP are often employed for the further development and scaling of CxG theories and applications.
Prof Adele Goldberg, Professor of Psychology, Princeton University
Prof Thomas Hoffmann, Professor of English Language and Linguistics, Catholic University of Eichstätt-Ingolstadt
Prof Laura A. Michaelis, Professor of Linguistics, University of Colorado Boulder
This workshop aims to bring together researchers across theory and practice from the two complementary perspectives of Construction Grammar and NLP to explore how CxG approaches can both inform and benefit from NLP methods, with an emphasis on LLMs. Therefore, we invite original research papers from a broad spectrum of topics, including but not limited to:
Contributions to CxG linguistic theory
Formalisms for Construction Grammars
Natural Language Understanding (NLU)
Opinion pieces on the interplay between CxGs and NLP
Constructions and Language Models (Mechanistic Interpretability, BERTology, probing and evaluation of LLMs)
Resources: Constructicons and corpora annotated for Construction Grammar
Construction Grammar learning and adaptation
Applications
The 2nd CxGs+NLP workshop will be co-located with the 16th International Conference on Computational Semantics (IWCS), organized by the Heinrich Heine University (HHU) in Düsseldorf, Germany. The workshop will be a full day on 24 September 2025. Additionally, we will be hosting a community-building event in Düsseldorf on 25 September 2025, including panel discussions and breakout sessions on how to organize CxG community resources.
We are expecting the workshop to be in-person only, but are awaiting details on the possibility of a hybrid presentation option.
Jun 30th: Extended Submission deadline Jun 06: submission deadline
Aug 01: notification of acceptance, registration opens
Aug 22: camera-ready papers due
Sep 22-23: IWCS main conference
Sep 24: workshop
Sep 25: community-building event
Two types of submission are solicited: long papers and short papers. Long papers should describe original research and must not exceed 8 pages. Short papers (typically system or project descriptions, or ongoing research) must not exceed 4 pages. Acknowledgments, references, a limitations section (optional), an ethics statement (optional), and a technical appendix (optional, not subject to reviewing) do not count towards the page limit.
Accepted papers get an extra page in the camera-ready version and will be published in the conference proceedings in the ACL Anthology. Additionally, non-archival publications will be considered for acceptance into the workshop as in-person poster presentations only.
CxGs+NLP 2 papers should be formatted following the common two-column structure as used by IWCS 2021 (borrowed from ACL 2021). Please use these specific style-files or the Overleaf template.
Style files: https://iwcs2021.github.io/download/iwcs2021-templates.zip
Overleaf template: https://www.overleaf.com/latex/templates/instructions-for-iwcs-2021-proceedings/fpnsyxqqpfbw
Double submission policy: We will accept submissions that have been submitted elsewhere, but require that the authors notify us, including information on where else they are submitting and let us know if the work is accepted for publication elsewhere.
Submission site https://openreview.net/group?id=IWCS/2025/Workshop/CxGs_NLP
As reviewing will be double blind, papers must not include authors’ names and affiliations. Furthermore, self-references or links (such as github) that reveal the author’s identity, e.g., “We previously showed (Smith, 1991) …” must be avoided. Instead, use citations such as “Smith previously showed (Smith, 1991) …” Papers that do not conform to these requirements will be rejected without review. Papers should not refer, for further detail, to documents that are not available to the reviewers. For example, do not omit or redact important citation information to preserve anonymity. Instead, use third person or named reference to this work, as described above (“Smith showed” rather than “we showed”). If important citations are not available to reviewers (e.g., awaiting publication), these paper/s should be anonymised and included in the appendix. They can then be referenced from the submission without compromising anonymity. Papers may be accompanied by a resource (software and/or data) described in the paper, but these resources should also be anonymized.
Abstract:
Constructions are learned pairings of form and function, at varying levels of complexity and abstraction. They are many, varied, context-dependent and interrelated to one another. The functions of constructions can involve subtle aspects of meaning, attitude, speech act, and information structure. I will present three subtle and distinct aspects of human language that are mirrored by LLMs. Collectively, the findings suggest remarkable parallels between the representations and dynamic changes of language in both LLMs and humans, making LLMs a rich resource for clarifying the human facility with language.
Bio:
Adele Goldberg, M Taylor Pyne Professor of Psychology at Princeton, is a linguist who investigates our knowledge of language and how languages are learned and processed, in children and adults, typical and autistic populations. She is particularly interested in constructions, learned pairings of form and function. Her newest research explores representations, language change and subtle effects of constructions in LLMs.
Abstract:
Usage-based Construction Grammar (Usage-based CxG; cf. e.g. Goldberg 2019) posits that the central unit of human language are symbols – form-meaning pairings that are labelled ‘constructions’. The framework, furthermore, assumes that speakers learn constructions through usage and that input frequency as well as semantics affect how strongly constructions are entrenched as well as the degree to which constructional abstractions are stored.
Recently, several publications have noted that Large Language Models (LLMs) exhibit many properties that have parallels in central assumptions of Construction Grammar (Goldberg 2024; also see Beuls & van Eecke for a critical assessment). Moreover, several publications have investigated the constructional knowledge displayed by LLMs (inter alia, Bonial & Madabushi 2024; Weissweiler, L. et al. 2022; Weissweiler, L. et al. 2023). In this talk, I want to shift the focus to human linguistic creativity (Hoffmann 2024, 2025; Hoffmann and Turner fc.) and ask to what degree LLMs can be considered ‘creative’ (or not). Moreover, I will discuss potential avenues for future research that can benefit the field of Natural Language Processing as well as CxG as a theoretical enterprise.
Bio:
Thomas Hoffmann is Professor of English Linguistics at the Katholische Universität Eichstätt-Ingolstadt. His main research interests are usage-based Construction Grammar, synchronic syntactic variation and World Englishes. He has published articles in international journals such as Cognitive Linguistics, English World-Wide and Corpus Linguistics and Linguistic Theory. On top of that, his monograph Preposition Placement in English (2011) was published by Cambridge University Press. Currently, he is writing the textbook Construction Grammar: The Structure of English for the Cambridge Textbooks in Linguistics series.
Abstract:
Construction Grammar (CxG) links combinatoric constructions—patterns governing how signs combine—directly to interpretive and use conditions. Fillmore (2008) showed that within constructions, words can inherit the combinatory affordances of other semantic classes through coercion, an interpretive enrichment in which lexical items shift valence (e.g., verbs adopting new argument structures, nouns switching countability, adjectives alternating scalar readings). Syntax, in this view, extends lexical valence: “describing [a lexical] item is equivalent to describing the constructions in which it participates” (Fillmore 1989:41). Drawing on Igor Mel’čuk’s phraseme-based analysis (Mel’čuk 2021), Fillmore reconceived phrases not as headed categories of the VP/NP type but as complex linguistic signs, in which constructional licensing, not head-driven syntactic selection, describes conventional sign combinations. Though not spoken of, complex words are constructed too. How should complex words be analyzed? Do subparts of words have valence and semantic dependents? Are morphemes themselves signs?
Sign-Based Construction Grammar (SBCG) abandoned morphemes as signs in favor of a realization-based model, partly to avoid Hockett’s (1987) “agglutinative fraud”—the delusion that surface morphology reflects discrete units. In this talk, I propose to revive morphemes as smaller signs—fully specified daughter signs—within the Construction Grammar architecture. Constructions remain descriptions of constructs, now licensing both phrasal and morphological structures in a valence-based framework in which daughter signs select one another. By treating both derivational and inflectional morphemes as selectors without head status, this approach captures coercion phenomena, implicit/explicit multiple exponence, and licit lexical-integrity violations as products of constructional licensing. This approach bridges morphology and syntax, revealing a single combinatorial system for both enriched and head-driven semantic composition.
Bio:
Laura A. Michaelis is Professor of Linguistics at the University of Colorado Boulder and a Faculty Fellow in the Institute of Cognitive Science. She earned her PhD in Linguistics at UC Berkeley under the direction of Charles J. Fillmore. A cognitive-functional syntactician and semanticist, her research spans the tense–aspect interface, corpus syntax, syntactic innovation, lexical semantics, and the discourse–syntax interface. She is also one of the principal developers of Sign-Based Construction Grammar and a founding editor of the interdisciplinary journal Language and Cognition (CUP).
Her recent book-length publications include Syntactic Constructions of English (CUP, 2020, with Jongbok Kim), Unrealized Arguments and the Grammar of Context (CUP, 2025, with Rui Chaves and Paul Kay) and Idiomatic Expressions and Grammatical Constructions (CSLI, 2025, with Paul Kay, Dan Flickinger and the late Ivan Sag).
Professional recognitions include the title Fellow of the Linguistic Society of America (2022), the Boulder Faculty Assembly’s Excellence in Research Award (2022), and the CU Boulder Graduate School’s Outstanding Faculty Mentor Award (2022). In July 2025, she served as Charles J. Fillmore Professor at the LSA Linguistic Institute organized by the University of Oregon in Eugene, co-teaching the course “Constructions in Context” with Elaine Francis (Purdue) and giving a Forum Lecture on Fillmore’s theory of idiomaticity.
Her current research explores the syntactic dimensions of climate change discourse.
Claire Bonial is a computational linguist specializing in the murky world of event semantics. In her efforts to make this world computationally tractable, she has collaborated on and been a foundational part of several important NLP lexical resources, including PropBank, VerbNet, and Abstract Meaning Representation. A focused contribution to these projects has been her theoretical and psycholinguistic research on both the syntax and semantics of English light verb constructions (e.g., “take a walk”, “make a mistake”). Bonial received her Ph.D. in Linguistics and Cognitive Science in 2014 from the University of Colorado Boulder and began her current position in the Content Understanding Branch at the Army Research Laboratory (ARL) in 2015. Since joining ARL, she has expanded her research portfolio to include multi-modal representations of events (text and imagery/video), human-robot dialogue, and misinformation detection.
Harish's long term research goals are focused on the investigating methods of incorporating high-level cognitive capabilities into models. In the short and medium term, his research is focused on the infusion of world knowledge and common sense into pre-trained language models to improve performance on complex tasks such as multi-hop question answering, conversational agents, and social media analysis. Harish completed his PhD in Question Answering from the University of Birmingham in 2019 and began his current post as Lecturer in Artificial Intelligence at the University of Bath in 2022. In the interim, he has worked on research related to MWEs, Construction Grammar and language modelling and was the principal organiser of the SemEval 2022 Task on MWEs.
Katrien Beuls is an assistant professor in artificial intelligence at the University of Namur (Belgium). For many years, she has been a leading figure in the development of the Fluid Construction Grammar (FCG) framework. Apart from her contributions to the core of FCG, she has been involved in the application of computational construction grammars across many domains, including language tutoring, intelligent cooking assistants, and online opinion observatories. Her ongoing research is mainly concerned with agent- based models of the emergence, evolution, and acquisition of human-like languages in machines, adopting a construction grammar perspective. She has previously organised several workshops, including the 2017 AAAI Spring Symposium on Computational Construction Grammar and Natural Language Understanding and the IJCAI 2022 workshop on Semantic Techniques for Narrative-Based Understanding.
Paul Van Eecke (°1990) obtained master degrees in Artificial Intelligence (2013 – summa cum laude) and Linguistics (2012, summa cum laude) from the KU Leuven, and a PhD Degree in Computer Science (2018) from the Artificial Intelligence Laboratory of the Vrije Universiteit Brussel. From 2014 until 2017, he worked as an assistant researcher at Sony Computer Science Laboratories Paris, where he carried out research on implementing construction grammars and became one of the main developers of the Fluid Construction Grammar (FCG) system.
He is currently a tenure track research professor (Senior Research Fellow) in ‘humanlike communication in autonomous agents’ at the VUB Artificial Intelligence Laboratory. His main research interests include the emergence and evolution of language through communicative interactions, computational construction grammar and its applications, and the use of a combination of symbolic and subsymbolic AI techniques for solving advanced perception, reasoning and communication tasks.
I am a computational linguistics and natural language processing researcher at the Army Research Lab in Adelphi, Maryland. Previously, I have worked in the NERT lab, and been a member of GUCL at Georgetown. My research focuses on computational semantics and semantic parsing. My interests include deep learning, natural language understanding, and mathematical linguistics. I am particularly interested in strategies for computationally modelling word and sentence meaning in frameworks like Abstract Meaning Representation.
I am a postdoc at UT Austin Linguistics, working with Kyle Mahowald, funded by a Fellowship from the German Academic Exchange Service (DAAD).
I completed my PhD at the Center for Information and Language Processing at LMU Munich where my thesis was about Computational Approaches to Construction Grammar and Morphology. My supervisor was Hinrich Schütze. Previously, I completed my B.Sc. and M.Sc. degrees in Computational Linguistics and Computer Science at LMU, with scholarships from the German Academic Scholarship Foundation and the Max Weber Program. My M.Sc. thesis, supervised by Hinrich Schütze, was on the application of Complementary Learning Systems Theory to NLP. I spent the final year of my bachelor's degree as a visiting student at Homerton College, University of Cambridge, where I wrote my B.Sc. thesis on Character-Level RNNs under the supervision of Anna Korhonen.
Researching the challenges and risk of using large language models in Natural Language Processing (NLP), focusing on addressing these issues through the use of explainable NLP.
There has been significant progress in NLP in recent years, particularly with the development of pre-trained and large language models. These models are based on neural networks and are trained on massive amounts of text data, allowing them to learn the patterns and structure of language in a way that mimics human understanding. However, the increased use of these types of models has highlighted many challenges in NLP that still need to be addressed. Issues such as bias and lack of model interpretability are important considerations in the development and deployment of NLP models.