Speech World Model: Causal State-Action Planning with Explicit Reasoning for Speech

arXiv 25.12 2025 31.7 method

TLDR

Proposes a modular speech world model with causal graph for explicit reasoning and interpretability under partial supervision.

Reasoning

Strengths include a novel modular causal graph approach inspired by cognitive science, enabling counterfactual interventions and interpretability. Weaknesses are the lack of empirical results or real-world evaluation in the abstract, and reliance on future open-sourcing without evidence of performance.

Read-first score

Read-first score 31.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 19.

Recency 5%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 15%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=analysis,experiment,result

Reproducibility 15%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code,open source

Topical relevance 25%

27.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Citation impact 15%

10.2

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=1

Citation velocity 10%

8.3

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.50

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 243.

Keyword Scores

world model

world dynamics prediction

generative world model

world simulator

interactive world model

video world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Explicit reasoning over speech states and actions with modular and transparent decisions
Factorization of speech understanding into four modules communicating through a causal graph
Cognitive state search space guided by posterior traces
Instruction-tuned language model for causal analysis and user-facing response
Counterfactual interventions and interpretability under partial supervision
First graph-based modular speech model for explicit reasoning

Methodology

The system adopts a world model perspective, learning forward dynamics over latent states. It factorizes speech understanding into four modules that interact via a causal graph, establishing a cognitive state search space. An instruction-tuned language model uses posterior traces from this space to generate causal analysis and responses.

Key Results

The abstract does not report quantitative experimental results. It claims the model enables counterfactual interventions and interpretability under partial supervision, and that it is the first graph-based modular speech model for explicit reasoning.

Resources

Paper