Retrieval as Generation: A Unified Framework with Self-Triggered Information Planning

We revisit retrieval-augmented generation (RAG) by embedding retrieval control directly into generation. Instead of treating retrieval as an external intervention, we express retrieval decisions within token-level decoding, enabling end-to-end coordination without additional controllers or classifiers. Under the paradigm of Retrieval as Generation, we propose $\mathbf{\text{GRIP}}$ ($\mathbf{\text{G}}$eneration-guided $\mathbf{\text{R}}$etrieval with $\mathbf{\text{I}}$nformation $\mathbf{\text{P}}$lanning), a unified framework in which the model regulates retrieval behavior through control-token emission. Central to GRIP is $\mathit{\text{Self-Triggered}}\mathit{\text{Information}}\mathit{\text{Planning}}$, which allows the model to decide when to retrieve, how to reformulate queries, and when to terminate, all within a single autoregressive trajectory. This design tightly couples retrieval and reasoning and supports dynamic multi-step inference with on-the-fly evidence integration. To supervise these behaviors, we construct a structured training set covering answerable, partially answerable, and multi-hop queries, each aligned with specific token patterns. Experiments on five QA benchmarks show that GRIP surpasses strong RAG baselines and is competitive with GPT-4o while using substantially fewer parameters.