{"paper":{"title":"Prompt Segmentation and Annotation Optimisation: Controlling LLM Behaviour via Optimised Segment-Level Annotations","license":"http://creativecommons.org/licenses/by-sa/4.0/","headline":"Optimised segment-level annotations on decomposed prompts improve LLM responses while preserving the original to avoid degradation.","cross_cats":[],"primary_cat":"cs.AI","authors_text":"Anjin Liu, Anna Leontjeva, Coco Wu, Devika Prasad, Henry Xiao, Luiz Pizzato, Luke Gerschwitz, Tong Li","submitted_at":"2026-05-14T08:33:47Z","abstract_excerpt":"Prompt engineering is crucial for effective interaction with generative artificial intelligence systems, yet existing optimisation methods often operate over an unstructured and vast prompt space, leading to high computational costs and potential distortions of the original intent. We introduce Prompt Segmentation and Annotation Optimisation (PSAO), a structured prompt optimisation framework designed to improve prompt optimisation controllability and efficiency. PSAO decomposes a prompt into interpretable segments (e.g., sentences) and augments each with human-readable annotations (e.g., {not "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"optimised segment-level annotations can lead to improved LLM responses, with the original prompt retained as a candidate in the optimisation space to prevent performance degradation. Empirical evaluations indicate that PSAO benefits from annotations in terms of improved reasoning accuracy and self-consistency.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That human-readable annotations such as {important} or {not important} can reliably guide LLMs in allocating focus and clarifying confusion during response generation without distorting the original intent.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"PSAO decomposes prompts into annotated segments to improve LLM reasoning accuracy and self-consistency as a proof-of-concept framework.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Optimised segment-level annotations on decomposed prompts improve LLM responses while preserving the original to avoid degradation.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"b4632f3d2c75554a9c9363def596aa7e7318d6f9554f99c835d084162ec32f87"},"source":{"id":"2605.14561","kind":"arxiv","version":1},"verdict":{"id":"6e960847-3ff6-4d80-8b85-deb6521c6a79","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T01:25:14.198673Z","strongest_claim":"optimised segment-level annotations can lead to improved LLM responses, with the original prompt retained as a candidate in the optimisation space to prevent performance degradation. Empirical evaluations indicate that PSAO benefits from annotations in terms of improved reasoning accuracy and self-consistency.","one_line_summary":"PSAO decomposes prompts into annotated segments to improve LLM reasoning accuracy and self-consistency as a proof-of-concept framework.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That human-readable annotations such as {important} or {not important} can reliably guide LLMs in allocating focus and clarifying confusion during response generation without distorting the original intent.","pith_extraction_headline":"Optimised segment-level annotations on decomposed prompts improve LLM responses while preserving the original to avoid degradation."},"references":{"count":37,"sample":[{"doi":"","year":2025,"title":"GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning","work_id":"40b60d06-dc1c-4799-b75d-ff1eca653049","ref_index":1,"cited_arxiv_id":"2507.19457","is_internal_anchor":true},{"doi":"","year":2024,"title":"Xcoop: Explainable prompt learning for computer-aided diagnosis via concept-guided context optimization","work_id":"522057a1-f9a0-471b-9fe3-e9a3808696d8","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2023,"title":"Evoprompting: Language models for code-level neu- ral architecture search","work_id":"49995824-ec19-40f9-bc0e-6c230c88b5b0","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Instructzero: Effi- cient instruction optimization for black-box large language models","work_id":"472340d6-ef90-45f8-8269-d54b058c514b","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2021,"title":"Training Verifiers to Solve Math Word Problems","work_id":"acab1aa8-b4d6-40e0-a3ee-25341701dca2","ref_index":5,"cited_arxiv_id":"2110.14168","is_internal_anchor":true}],"resolved_work":37,"snapshot_sha256":"7dd494948a478276f155ad5a8f22d16dd71ef22e3fcd62dd29600c0b8ce08d79","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"a8e46f5e5fb0f0e8e5f454655a9a847e7eae3009d0077b40573bc19b62896442"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}