Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper investigates the impact of schema key wording on the performance of LLMs in structured generation tasks under constrained decoding. It reveals that varying the linguistic formulation of schema keys, without altering prompts or model parameters, significantly affects model performance. The study demonstrates that different model families exhibit varying sensitivities to schema-level versus prompt-level instructions, with Qwen models favoring schema instructions and LLaMA models relying more on prompts.
LLMs can be subtly steered during constrained decoding by simply rewording schema keys, revealing a previously overlooked instruction channel.
Constrained decoding has been widely adopted for structured generation with large language models (LLMs), ensuring that outputs satisfy predefined formats such as JSON and XML. However, existing approaches largely treat schemas as purely structural constraints and overlook the possibility that their linguistic formulation may affect model behavior. In this work, we study how instruction placement influences model performance in structured generation and show that merely changing the wording of schema keys, without modifying the prompt or model parameters, can significantly alter model performance under constrained decoding. Based on this observation, we propose to reinterpret structured generation as a multi-channel instruction problem, where instructions can be conveyed explicitly through prompts and implicitly through schema keys during decoding. To the best of our knowledge, this is the first work to systematically study how schema key formulation acts as an implicit instruction channel and affects model performance under constrained decoding. Experiments on multiple mathematical reasoning benchmarks show that different model families exhibit distinct sensitivities to these instruction channels: Qwen models consistently benefit from schema-level instructions, while LLaMA models rely more heavily on prompt-level guidance. We further observe non-additive interaction effects between instruction channels, showing that combining multiple channels does not always lead to further improvement. These findings suggest that schema design not only determines output structure, but also carries instruction signals, offering a new perspective on structured generation in LLMs.