The Communication Bottleneck: round-trip study of tree-structured expression serialization in language models

KEY POINTS

1. The paper proposes a round-trip protocol to measure how well tree-structured arithmetic expressions survive serialization into natural-language word problems and back.
2. Evaluating 16 models pairwise, the authors find the channel is lossy and asymmetric (swapping generator and extractor can change accuracy by up to 60.4 points), the best generator–extractor pair achieves 92.9% round-trip correctness, at least 73.6% of failures originate in generation, structure (operator count, depth, right-branching) drives difficulty more than model family, and ~3,600 targeted fine-tuning examples substantially improve open-weight models (raising them above untrained Gemini-3.1-Pro and helping in disjoint-domain cases).
3. This quantifies a concrete, trainable limitation—serialization of hierarchical structure into text—and pinpoints generation-phase and tree shape as primary bottlenecks, guiding where to improve model communication and fine-tuning.