Constructional approaches to signed languages are becoming increasingly popular within sign language linguistics. Current approaches, however, focus primarily on theoretical description, while formalization and computational implementation remain largely unexplored. This paper provides an initial step towards addressing this gap by studying and operationalizing the core mechanisms required for representing and processing manual signed forms using computational construction grammar. These include a phonetic representation of individual manual signs and a formal representation of the complex temporal synchronization patterns between them. The implemented mechanisms are integrated into Fluid Construction Grammar and are available as a module within the Babel software library. Through an interactive web demonstration, we illustrate how this module lays the groundwork for future computational exploration of constructions that bidirectionally map between signed forms and their meanings.

This paper investigates false positive constructions: grammatical structures which an LLM hallucinates as distinct constructions but which human introspection does not support. Both a behavioural probing task using contextual embeddings and a meta-linguistic probing task using prompts are included, allowing us to distinguish between implicit and explicit linguistic knowledge. Both methods reveal that models do indeed hallucinate constructions. We then simulate hypothesis testing to determine what would have happened if a linguist had falsely hypothesized that these hallucinated constructions do exist. The high accuracy obtained shows that such false hypotheses would have been overwhelmingly confirmed. This suggests that construction probing methods suffer from a confirmation bias and raises the issue of what unknown and incorrect syntactic knowledge these models also possess.

This paper applies Sentence-Bert embeddings to the analysis of three competing constructions in Canadian English: be perfect, predicate adjective and have perfect. Samples are drawn from a Canadian news media database. Constructional exemplars are vectorized and mean-pooled to create constructional centroids, from which top-ranked exemplars and cross-construction similarities are calculated. Clause type distribution and definiteness marking are also examined. The embeddings-based analysis is cross-validated by a traditional quantitative study, and both lines of inquiry converge on the following tendencies: (1) prevalence of embedded – and particularly adverbial – clauses in the be perfect and predicate adjective constructions, (2) prevalence of matrix clauses in the have perfect, (3) prevalence of definiteness marking in the direct object of the be perfect, and (4) greater statistical similarities between be perfects and predicate adjectives. These findings support the argument that be perfects function as topic-marking constructions within a usage-based framework.

Large language models excel at statistical pattern recognition but may lack explicit understanding of constructional form-meaning correspondences that characterize human grammatical competence. This paper presents Construction-Aware LoRA (CA-LoRA), a parameter-efficient fine-tuning method that incorporates constructional templates through specialized loss functions and targeted parameter updates. We focus on five major English construction types: ditransitive, caused-motion, resultative, way-construction, and conative. Evaluation on BLiMP, CoLA, and SyntaxGym shows selective improvements: frequent patterns like ditransitive and caused-motion show improvements of approximately 3.5 percentage points, while semi-productive constructions show minimal benefits (1.2 points). Overall performance improves by 1.8% on BLiMP and 1.6% on SyntaxGym, while maintaining competitive performance on general NLP tasks. Our approach requires only 1.72% of trainable parameters and reduces training time by 67% compared to full fine-tuning. This work demonstrates that explicit constructional knowledge can be selectively integrated into neural language models, with effectiveness dependent on construction frequency and structural regularity.

Argument structure constructions (ASCs) offer a theoretically grounded lens for analyzing second language (L2) proficiency, yet scalable and systematic tools for measuring their usage remain limited. This paper introduces the ASC analyzer, a publicly available Python package designed to address this gap. The analyzer automatically tags ASCs and computes 50 indices that capture diversity, proportion, frequency, and ASC-verb lemma association strength. To demonstrate its utility, we conduct both bivariate and multivariate analyses that examine the relationship between ASC-based indices and L2 writing scores.

Is the framework of Universal Dependencies (UD) compatible with findings from linguistic typology about constructions in the world’s languages? To address this question, we need to systematically review how UD represents these constructions, and how it handles the range of morphosyntactic variation attested across languages. In this paper, we present the results of such a review focusing on verbal predication constructions. We find that, although UD can represent all major constructions in this area, the guidelines are not completely coherent with respect to the criteria for core argument relations and not completely systematic in the definition of subtypes for nonbasic voice constructions. To improve the overall coherence of the guidelines, we propose a number of revisions for future versions of UD.

Linguistic evaluations of how well LMs generalize to produce or understand novel text often implicitly take for granted that natural languages are generated by symbolic rules. Grammaticality is thought to be determined by whether sentences obey such rules. Interpretation is believed to be compositionally generated by syntactic rules operating on meaningful words. Semantic parsing is intended to map sentences into formal logic. Failures of LMs to obey strict rules have been taken to reveal that LMs do not produce or understand language like humans. Here we suggest that LMs’ failures to obey symbolic rules may be a feature rather than a bug, because natural languages are not based on rules. New utterances are produced and understood by a combination of flexible, interrelated, and context-dependent constructions. We encourage researchers to reimagine appropriate benchmarks and analyses that acknowledge the rich, flexible generalizations that comprise natural languages.

Linguistic theories and models of natural language can be divided into two categories, depending on whether they represent and process linguistic information numerically or symbolically. Numerical representations, such as the embeddings that are at the core of today’s large language models, have the advantage of being learnable from textual data, and of being robust and highly scalable. Symbolic representations, like the ones that are commonly used to formalise construction grammar theories, have the advantage of being compositional and interpretable, and of supporting sound logic reasoning. While both approaches build on very different mathematical frameworks, there is no reason to believe that they are incompatible. In the present paper, we explore how numerical, in casu distributional, representations of linguistic forms, constructional slots and grammatical categories can be integrated in a computational construction grammar framework, with the goal of reaping the benefits of both symbolic and numerical methods.

Constructionist approaches to language posit that all linguistic knowledge is captured in constructions. These constructions pair form and meaning at varying levels of abstraction, ranging from purely substantive to fully abstract and are all acquired through situated communicative interactions. In this paper we provide computational support for these foundational principles. We present a model that enables an artificial learner agent to acquire a construction grammar directly from its sensory experience. The grammar is built from the ground up, i.e. without a given lexicon, predefined categories or ontology and covers a range of constructions, spanning from purely substantive to partially schematic. Our approach integrates two previously separate but related experiments, allowing the learner to incrementally build a linguistic inventory that solves a question-answering task in a synthetic environment. These findings demonstrate that linguistic knowledge at different levels can be mechanistically acquired from experience.

The empirically established null subject (NS) stage, lasting until about 4 years of age, involves frequent omission of subjects by children. Orfitelli and Hyams (2012) observe that young English speakers often confuse imperative NS utterances with declarative ones due to performance influences, promoting a temporary null subject grammar. We propose a new computational parameter to measure this misinterpretation and incorporate it into a simulated model of obligatory subject grammar learning. Using a modified version of the Variational Learner (Yang, 2012) which works for superset-subset languages, our simulations support Orfitelli and Hyams’ hypothesis. More generally, this study outlines a framework for integrating computational models in the study of grammatical acquisition alongside other key developmental factors.

Generic statements like *A dog has four legs* are central to encode general knowledge. Yet their form–meaning mapping remains elusive. Some predicates sound natural with indefinite singulars (*a*-generics), while others require the definite article (*the*-generics) or the bare plural (bare-plural generics). For instance, why do we say *The computer revolutionized education* but not *A computer revolutionized education*? We propose a construction-based account explaining why not all generic statements are created equal. Prior accounts invoke semantic notions like kind-reference, stage-levelness, or accidental generalization, but offer no unified explanation. This paper introduces a new explanatory dimension: predicate collectivity level, i.e. whether the predicate applies to each member of a group or to the whole group as a unit (without necessarily applying to each of its members individually). Using two preregistered acceptability experiments we show that *a*-generics, unlike *the*-generics and bare-plural generics, are dispreferred with collective predicates. The findings offer a functionally motivated, empirically supported account of morphosyntactic variation in genericity, providing a new entry point for Construction Grammar.

Constructional approaches to language have evolved from rigid tree-based representations to framing constructions as flexible, multidimensional pairings of form and function. However, it remains unclear how to formalize this conceptual shift in ways that are both computationally scalable and scientifically insightful. This paper proposes dynamic tensegrity – a term derived from “tensile integrity” – as a novel architecture metaphor for modelling linguistic form. It argues that linguistic structure emerges from dynamically evolving networks of constraint-based tensions rather than fixed hierarchies. The paper explores the theoretical consequences of this view, supplemented with a proof-of-concept implementation in Fluid Construction Grammar, demonstrating how a tensegrity-inspired approach can support robustness and adaptivity in language processing.

This paper proposes a formal framework based on Tree Adjoining Grammar (TAG) that aims to incorporate central tenets of Construction Grammar while integrating mechanisms from a psycholinguistically motivated variant of TAG. Central ideas are (i) to give TAG-inspired tree representation to various constructions including schematic constructions like argument structure constructions, (ii) to link schematic constructions that are extensions of each other within a network of constructions, (iii) to make the derivation proceed incrementally, (iv) to allow the prediction of upcoming constructions during derivation and (v) to introduce the incremental extension of schematic constructions to larger ones via extension trees in a usage-based manner. The final point is the major novel contribution, which can be conceptualized as the on-the-fly traversal of the inheritance links in the network of constructions. Moreover, we present first experiments towards a parser implementation. We report preliminary results of extracting constructions from the Penn Treebank and automatically identifying constructions to be added during incremental parsing, based on a generative language model (GPT-2).

Chinese verb-resultative complement construction (VRCC), constitute a distinctive syntactic-semantic pattern in Chinese that integrates agent-patient dynamics with real-world state changes; yet widely used benchmarks such as CLiMP and ZhoBLiMP provide few minimal-pair probes tailored to these constructions. We introduce ZhVrcMP, a 1,204 pair dataset spanning two paradigms: resultative complement presence versus absence, and verb–complement order. The examples are drawn from Modern Chinese and are annotated for linguistic validity. Using mean log probability scoring, we evaluate Zh-Pythia models (14M-1.4B) and Mistral-7B-Instruct-v0.3. Larger Zh-Pythia models perform strongly, especially on the order paradigm, reaching 89.87% accuracy. Mistral-7B-Instruct-v0.3 shows lower perplexity yet overall weaker accuracy, underscoring the remaining difficulty of modeling constructional semantics in Chinese.

The usage-based constructionist (UCx) approach to language posits that language comprises a network of learned form-meaning pairings (constructions) whose use is largely determined by their meanings or functions, requiring them to be graded and probabilistic. This study investigates whether the internal representations in Large Language Models (LLMs) reflect the proposed function-infused gradience. We analyze representations of the English Double Object (DO) and Prepositional Object (PO) constructions in Pythia-1.4B, using a dataset of 5000 sentence pairs systematically varied by human-rated preference strength for DO or PO. Geometric analyses show that the separability between the two constructions’ representations, as measured by energy distance or Jensen-Shannon divergence, is systematically modulated by gradient preference strength, which depends on lexical and functional properties of sentences. That is, more prototypical exemplars of each construction occupy more distinct regions in activation space, compared to sentences that could have equally well have occured in either construction. These results provide evidence that LLMs learn rich, meaning-infused, graded representations of constructions and offer support for geometric measures for representations in LLMs.

This paper introduces a usage-based framework that models argument structure annotation as nearest-neighbor classification over verb–argument structure (VAS) embeddings. Instead of parsing sentences separately, the model aligns new tokens with previously observed constructions in an embedding space derived from semi-automatic corpus annotations. Pilot studies show that cosine similarity captures both form and meaning, that nearest-neighbor classification generalizes to dative alternation verbs, and that accuracy in locative alternation depends on the corpus source of exemplars. These results suggest that analogical classification is shaped by both structural similarity and corpus alignment, highlighting key considerations for scalable, construction-based annotation of new sentence inputs.

Sequence to Sequence models struggle at compositionality and systematic generalisation even while they excel at many other tasks.We attribute this limitation to their failure to internalise constructions—conventionalised form–meaning pairings that license productive recombination. Building on these insights, we introduce an unsupervised procedure for mining pseudo-constructions: variable-slot templates automatically extracted from training data. When applied to the SCAN dataset, ourmethod yields large gains out-of-distribution splits: accuracy rises to 47.8% on ADD JUMP and to 20.3% on AROUND RIGHT without any architectural changes or additional supervision. The model also attains competitive performance with ≤ 40% of the original training data, demonstrating strong data efficiency. Our findings highlight the promise of construction-aware preprocessing as an alternative to heavy architectural or training-regime interventions.

We present CoGS-NLI, a Natural Language Inference (NLI) evaluation benchmark testing understanding of English phrasal constructions drawn from the Construction Grammar Schematicity (CoGS) corpus. This dataset of 1,500 NLI triples facilitates assessment of constructional understanding in a downstream inference task. We present an evaluation benchmark based on the performance of two language models, where we vary the number and kinds of examples given in the prompt, with and without chain-of-thought prompting. The best-performing model and prompt combination achieves a strong overall accuracy of .94 when provided in-context learning examples with the target phrasal constructions, whereas providing additional general NLI examples hurts performance. This evidences the value of resources explicitly capturing the semantics of phrasal constructions, while our qualitative analysis suggests caveats in assuming this performance indicates a deep understanding of constructional semantics.

We probe large language models’ ability to learn deep form-meaning mappings as defined by construction grammars. We introduce the ConTest-NLI benchmark of 80k sentences covering eight English constructions from highly lexicalized to highly schematic. Our pipeline generates diverse synthetic NLI triples via templating and the application of a model-in-the loop filter. This provides aspects of human validation to ensure challenge and label reliability. Zero-shot tests on leading LLMs reveal a 24% drop in accuracy between naturalistic (88%) and adversarial data (64%), with schematic patterns proving hardest. Fine-tuning on a subset of ConTest-NLI yields up to 9% improvement, yet our results highlight persistent abstraction gaps in current LLMs and offer a scalable framework for evaluating construction informed learning.

In this paper, we apply the lens of Construction Grammar to provide linguistically-grounded evidence for the recently introduced view of LLMs that moves beyond the “stochastic parrot” and “emergent Artificial General Intelligence” extremes. We provide further evidence, this time rooted in linguistic theory, that the capabilities of LLMs are best explained by a process of context-directed extrapolation from their training priors. This mechanism, guided by in-context examples in base models or the prompt in instruction-tuned models, clarifies how LLM performance can exceed stochastic parroting without achieving the scalable, general-purpose reasoning seen in humans. Construction Grammar is uniquely suited to this investigation, as it provides a precise framework for testing the boundary between true generalization and sophisticated pattern-matching on novel linguistic tasks. The ramifications of this framework explaining LLM performance are three-fold: first, there is explanatory power providing insights into seemingly idiosyncratic LLM weaknesses and strengths; second, there are empowering methods for LLM users to improve performance of smaller models in post-training; third, there is a need to shift LLM evaluation paradigms so that LLMs are assessed relative to the prevalence of relevant priors in training data, and Construction Grammar provides a framework to create such evaluation data.

This paper presents a pilot study of metaphors of motion in African American literary language (AALL) in two sub-corpora of novels published in 1920-1925 and 1926-1930. It assesses the effectiveness of Dunn’s (2024) unsupervised learning approach to computational construction grammar (c2xg) as a basis for searching for constructional metaphors, a purpose beyond its original design as a grammar-learning tool. This method is chosen for its statistical orientation and employed without pre-trained models to minimize bias towards standard language; its output is also used to choose a target search term. Focusing on the verbal phrase ‘come to’, the study analyzes argument-structure constructions that instantiate conceptual metaphors, most prominently experiencer-as-theme (e.g., ‘he came to know’) and experiencer-as-goal (e.g., ‘thoughts came to her’). The evaluation compares c2xg coverage against a manually annotated set of metaphors and examines the uniformity of metaphor types extracted. Results show that c2xg captures 52% and 63% of metaphoric constructions in the two sub-corpora, with variation in coverage and uniformity depending on the ambiguity of the construct. The study underscores the value of combining computational and manual analysis to obtain outcomes that are both informative and ethically aware when studying marginalized varieties of English.