19 papers · ranked by Valyu relevance
Nathan Klinedinst
Human languages use complex, structured signals whose meanings are compositional. Recent empirical research has claimed to demonstrate compositionality in bird and primate communication (Berthet et al. [2]; Engesser et al. [4]; Girard-Buttoz et al. [6]; Leroux et al. [8]; Suzuki et al. [21]). While the compositionality…
Guangyao Qi, Olga Dal Monte, Siqi Fan, Steve W. C. Chang
Social gaze underpins primate communication, yet the neural principles enabling its flexibility remain unknown. Each social gaze can be deconstructed into three primitives: gaze content, social state, and gaze duration. To reduce dimensionality and facilitate generalization, the brain needs to represent these…
Ziyao Xu, Cong Wang, Houfeng Wang
Compositional generalization tests are often used to estimate the compositionality of LLMs. However, such tests have the following limitations: (1) they only focus on the output results without considering LLMs' understanding of sample compositionality, resulting in explainability defects; (2) they rely on dataset…
Dat H. Do, Rushi Shah, Duc V. Le, Dianbo Liu
Compositionality is believed to be the foundation for generalization, enabling models to reuse meaningful primitives in novel combinations. Yet, models trained with standard gradient-based optimization rarely, and often only weakly, exhibit compositional internal structure, and it remains unclear how or why such…
Joshua B. Tan, Isabella F. Orlando, Jungwoo Kim, Christopher J. Cueva + 6 more
Human cognition depends on the ability to flexibly recombine existing knowledge in new ways. Although this capacity for compositionality has traditionally been attributed to cortical networks, its broader neural basis remains unclear. Here, we combined dimensionality reduction of task-based fMRI with recurrent neural…
Fausto Carcassi
Formal semantics has shown that sentence meanings arise by recursively composing lexical meanings, yet much of the literature on semantic universals models either lexicons with fixed signal structures or holistic composition without interpretable lexical parts. We introduce a framework that integrates this fundamental…
Fabio De Ponte, Eloise Gaines-White, Conor Houghton, Seth Bullock
The iterated learning model was introduced to investigate language evolution: the way in which the characteristic properties of human languages have been shaped, at least partly, by repeated transmission from one language user to another. The key finding is that language compositionality can arise spontaneously as a…
Devon Jarvis, Richard Klein, Benjamin Rosman, Andrew M. Saxe
Title: Significance Children rapidly acquire an ability for language during early development. One theory, called iterated learning, posits that language evolves over generations to become more structured. This structure can then be exploited by learners through systematic generalization, where past experiences are…
Louis Pezon, Alexander van Meegen
Flexible cognition utilizes reusable components to enable rapid adaptation of behavior to different contexts or tasks. Analysis of artificial neural networks trained on multiple tasks suggested that this compositionality is supported by dynamical structures which are shared and re-used across tasks. However, the nature…
Mahnoor Shahid, Hannes Rothe
Compositional generalization remains a foundational weakness of modern neural networks, limiting their robustness and applicability in domains requiring out-of-distribution reasoning. A central, yet unverified, assumption in neuro-symbolic AI is that compositional reasoning will emerge as a byproduct of successful…
Anne Reboul, Nicolas Claidière, Isabelle Dautriche, Joël Fagot + 1 more
This study investigates whether baboons are capable of semantic compositionality, specifically, whether they can apply compositional rules to new situations (generalization). In language, semantic compositionality is linked to productivity, the generalization of a rule to new combinations. Across four experiments…
Giulia Palazzolo
Is syntax an evolutionary novelty in the human lineage? This question, along with the question of how human syntax evolved, is highly debated in the field of language evolution. In this paper, I reconstruct two prominent frameworks for studying the evolution of human syntax, which I call “unbounded hierarchy” (Bolhuis…
Aldo Battista, Camillo Padoa-Schioppa, Xiao-Jing Wang
Title: SUMMARY Value-guided decisions are a cornerstone of cognition, yet the underlying circuit-level mechanisms remain elusive. We used reinforcement learning to train recurrent neural network models endowed with Dale’s law on a battery of economic choice tasks, which revealed a two-stage computational framework.…
Sowmya Manojna Narasimha, Jingya Huang, Ram Dyuthi Sristi, Vikash Gilja + 1 more
Recent brain-computer interfaces (BCIs) have achieved state-of-the-art performance in decoding behavior from neural activity. These models are typically trained on a con-strained set of behaviors, which limits their ability to generalize to real-world settings where behavior is variable, complex, and context-dependent.…
Chris Jenkins, Emma Raimundo Schulz, Filip Miletić, Sabine Schulte im Walde
We explore the phenomenon of semantic change of German and English noun compounds, with the objective of investigating and modeling gradual changes of meanings and degrees of compositionality in the past and over time. To do so, we introduce the Compositionality Trend Prediction task, which is evaluated against a novel…
Yuma Osako, Aineias Arango, Toshitake Asabuki
Animals flexibly combine learned behaviors into novel actions without practicing their combinations, yet the computational mechanisms that enable independently acquired computations to be expressed in parallel remain unclear. Here we show that feedback geometry during learning determines whether recurrent dynamics can…
Authors not listed
Self-driving laboratories (SDLs) promise accelerated scientific discovery and product development by closing the loop between robotic execution and AI/ML-driven decision making. In practice, however, SDL orchestration remains fragmented; workflows are typically encoded as laboratory-specific scripts or bespoke…
Authors not listed
The Polytope Formalism provides a rigorous and unifying mathematical framework for representing all possible molecular configurations and their interrelationships. Extending its application from stereoisomerism to molecular constitution reveals that both arise from a common structural foundation linking discrete and…
Authors not listed
Machine learning is increasingly used to predict reaction properties such as barrier heights, reaction energies, rates, or yields, as well as the underlying molecular geometries, including transition state structures. While such predictions have the potential to provide mechanistic insight for high-impact applications…