A Declarative-Procedural Perspective on Expert Routing in Bilingual Mixture-of-Experts Language Models

📅 2026-08-15
📈 Citations: 0
Influential: 0
📄 PDF
🤖 AI Summary
This study investigates whether bilingual Mixture-of-Experts (MoE) routing exhibits linguistic structure by adopting a declarative-procedural framework. Utilizing probing techniques and mutual information to quantify expert specialization, the research integrates cognitive theory into interpretability analysis to reveal emergent language organization patterns under non-curriculum training and the suppressive effects of curriculum learning on monolingual dominance. Results demonstrate that mixed training yields stronger aggregated specialization compared to sequential approaches, with the non-curriculum model achieving a mutual information score of 0.2599. Conversely, curriculum training facilitates stable language-balanced routing. These findings offer novel insights into the internal mechanisms of multilingual models, highlighting the interplay between training strategies and linguistic representation in MoE architectures.
📝 Abstract
We investigate whether Mixture-of-Experts (MoE) language models develop linguistically structured expert routing during bilingual language acquisition. Inspired by the Declarative-Procedural framework, we analyze lexical, grammatical, and syntactic processing in a decoder-only English-German MoE Transformer trained under sequential language exposure. We construct a probe-based validation set and extract token-level routing distributions to quantify category-dependent specialisation using mutual information, routing entropy, and Jensen-Shannon distance. The curriculum-trained model exhibits a peak mutual information of 0.1148 at layer 5, indicating category-dependent differences in routing distributions across linguistic categories. Surprisingly, a no-curriculum baseline trained on mixed English-German data shows stronger aggregate specialisation, reaching a peak mutual information of 0.2599 at the same layer. These results suggest that interpretable linguistic organization emerges within MoE routing patterns even without sequential language exposure. A replication at a second training seed shows that the no-curriculum condition's specialisation concentrates on a single language whose identity is seed-dependent, whereas the curriculum consistently yields a stable, language-balanced routing profile; rather than uniformly increasing specialisation, staged bilingual exposure reduces single-language dominance. The official Github repository: https://github.com/Amrit828/DP-Theory-MOE-Interpretability-Research
Problem

Research questions and friction points this paper is trying to address.

Mixture-of-Experts
Bilingual Language Acquisition
Expert Routing
Declarative-Procedural Framework
Innovation

Methods, ideas, or system contributions that make the work stand out.

Mixture-of-Experts
Declarative-Procedural Framework
Expert Routing
Bilingual Language Acquisition
Interpretability
A
Amrit Gopinath
Sri Sivasubramaniya Nadar College of Engineering, Chennai, India
R
Raghul
Sri Sivasubramaniya Nadar College of Engineering, Chennai, India
D
Durairaj Thenmozhi
Shiv Nadar University Chennai, India