Structure from Rank: Rank-Order Coding as a Bridge from Sequence to Structure

Xiaodan Chen, Alexandre Pitti, Mathias Quoy, Nancy F. Chen

Abstract

Understanding how structured sequence information can be represented and generalized in neural systems is key to modeling the transition from acoustic input to emergent structure. In this study, we propose a rank-order based neural network inspired by the STG-LIFG-PMC pathway, modeling the bottom-up transition from acoustic input to abstract rank representation, and the top-down generation from that representation to motor execution. Building on previous work in rank coding, we first demonstrate that this model efficiently compresses input while retaining the capacity to reconstruct full utterances from partial cues, revealing emergent structure-sensitive generation process that reflects context-general representations of sensorimotor states, which are later shaped into context-specific motor plans during speech planning. We then show that the network exhibits global-level novelty detection similar to the P3B novelty wave, replicating the global-sequence-sensitive mechanism. As a supplement, we also compare the model's behavior under local (index-level) and global (rank-level) perturbations, revealing robustness to superficial variation and sensitivity to abstract structural violation, key features associated with proto-syntactic generalization. These results suggest that rank-order coding not only serve as a compact encoding scheme but also support encoding hierarchical grammar.

Introduction

A fundamental challenge in cognitive neuroscience is understanding how biological systems transition from processing continuous acoustic streams to understanding and generating complex, hierarchical language. While modern artificial intelligence relies on massive Transformer architectures and vast datasets, the human brain achieves linguistic competence with remarkable efficiency and limited working memory. This paper explores "rank-order coding"—a mechanism where information is represented based on the relative timing and importance of neural activations—as a potential "protogrammar" that bridges the gap between simple sequential processing and complex structural understanding.

Methods and Architecture

The proposed neural network architecture is inspired by the human brain's speech processing pathways, specifically the STG-LIFG-PMC pathway. The Superior Temporal Gyrus (STG) handles initial auditory perception, while the Left Inferior Frontal Gyrus (LIFG), or Broca’s Area, acts as a sequencing hub to extract abstract structures. Finally, the Premotor Cortex (PMC) translates these abstract plans into motor actions for speech production.

The STG-LIFG-PMC pathway modeling the transition from auditory perception to motor execution via structural processing.

The model utilizes a Self-Organizing Map (SOM) for initial acoustic categorization and a Hopfield-like associative memory for pattern recall. The core innovation lies in the rank transform. Unlike standard indexing, which records specific neuron IDs, rank-order coding records the relative order of the most active neurons within a "chunk." For example, a sequence of activations [10, 42, 5] is transformed into the rank pattern [2, 1, 3]. This abstraction allows the system to focus on the underlying structural "template" rather than surface-level variations.

Hierarchical Representation

The transition from raw sound to abstract structure involves multiple layers of decomposition. By converting Mel-Frequency Cepstral Coefficients (MFCCs) from the LibriSpeech corpus into rank patterns, the model can generalize rules across different acoustic inputs.

Hierarchical decomposition of speech from raw waveform to abstract rank-order chunks, illustrating the transition from auditory perception to hierarchical structure processing.

Key Results

Efficiency and Reconstruction

The model demonstrates high compression efficiency; as sequence length increases, the variety of unique rank chunks grows significantly slower than raw index chunks. Furthermore, the network exhibits powerful reconstruction capabilities, generating coherent sequences of up to 36 indices from as few as 5 initial cue indices.

Novelty Detection and the P3B Wave

A hallmark of human linguistic processing is the P3B novelty wave, an EEG signal triggered by violations of global structural rules. The model replicated this behavior, showing a sharp spike in entropy (uncertainty) when encountering rank-level violations, even when the individual components were familiar.

Robustness to Local Perturbation

The network is robust to "local" noise (shuffling specific sound identities) but highly sensitive to "global" structural violations (changing relative order). This mirrors human perception, where we can understand speech despite accents or minor mispronunciations but immediately detect grammatical or structural errors.

Conclusion

Rank-order coding provides a biologically plausible and computationally efficient framework for representing hierarchical structures. By abstracting away from specific identities to relative patterns, the model supports proto-syntactic generalization. Future work will focus on extending this architecture to handle the complex, nested tree structures of full human syntax and integrating it with motor control systems for robotics.

Summary updated

Abstract

Understanding how structured sequence information can be represented and generalized in neural systems is key to modeling the transition from acoustic input to emergent structure. In this study, we propose a rank-order based neural network inspired by the STG-LIFG-PMC pathway, modeling the bottom-up transition from acoustic input to abstract rank representation, and the top-down generation from that representation to motor execution. Building on previous work in rank coding, we first demonstrate that this model efficiently compresses input while retaining the capacity to reconstruct full utterances from partial cues, revealing emergent structure-sensitive generation process that reflects context-general representations of sensorimotor states, which are later shaped into context-specific motor plans during speech planning. We then show that the network exhibits global-level novelty detection similar to the P3B novelty wave, replicating the global-sequence-sensitive mechanism. As a supplement, we also compare the model's behavior under local (index-level) and global (rank-level) perturbations, revealing robustness to superficial variation and sensitivity to abstract structural violation, key features associated with proto-syntactic generalization. These results suggest that rank-order coding not only serve as a compact encoding scheme but also support encoding hierarchical grammar.
BibTeX
@inproceedings{chen2026structure,
  author = {Xiaodan Chen and Alexandre Pitti and Mathias Quoy and Nancy F. Chen},
  title = {{Structure from Rank: Rank-Order Coding as a Bridge from Sequence to Structure}},
  booktitle = {Neural},
  year = {2026},
  doi = {10.1016/j.neunet.2026.108828},
}
Figure