Deflanderization for Game Dialogue: Balancing Character Authenticity with Task Execution in LLM-based NPCs

TL;DR

Deflanderization for Game Dialogue investigates balancing character authenticity with task execution in LLM-based NPCs. It proposes a triad of methods—Deflanderization prompting, retrieval-augmented grounding, and LoRA/SFT finetuning—and evaluates them in the CPDC 2025 competition across API and GPU tracks. API results show lightweight prompting can improve task fidelity and reduce over-embellishment, while GPU experiments demonstrate that large-scale finetuning with memory-augmented retrieval yields stronger overall performance. The work highlights the trade-offs between persona-consistency and functional precision and points to hybrid approaches for jointly optimizing dialogue quality and action reliability in fantasy RPG NPCs.

Abstract

The emergence of large language models (LLMs) has opened new opportunities for creating dynamic non-player characters (NPCs) in gaming environments, enabling both functional task execution and persona-consistent dialogue generation. In this paper, we (Tu_Character_lab) report our participation in the Commonsense Persona-Grounded Dialogue Challenge (CPDC) 2025 Round 2, which evaluates agents across three tracks: task-oriented dialogue, context-aware dialogue, and their integration. Our approach combines two complementary strategies: (i) lightweight prompting techniques in the API track, including a Deflanderization prompting method to suppress excessive role-play and improve task fidelity, and (ii) fine-tuned large models in the GPU track, leveraging Qwen3-14B with supervisedfinetuning (SFT) and Low-Rank Adaptation(LoRA). Our best submissions ranked 2nd on Task 1, 2nd on Task 3 (API track), and 4th on Task 3 (GPU track).

Figures (7)

• Figure 1: Examples of player–NPC interactions using LLM-based agents in the CPDC 2025 competition, Top panel: Early Summer 7 PM, clear night at the Weapon Shop, showing an example of user-NPC interaction in Task 1 (function generation). Bottom panel: Late Winter 2 PM, rainy conditions at the Quest Reception Desk, showing an example of dialogue generation in Task 2.
• Figure 2: Main pipeline for the API Track task 3. The prompting stages are Step1 and Step4 and generataion stages are in Step2 and Step5
• Figure 3: Age-gender of characters in Task 2_train.json, the diagram shown that balanced NPC characters(20 merchant and 20 guild receptionist) most NPC are women with the younger age than men.
• Figure 4: Date-time distribution in Task 2_train.json, most of event occur after 1 pm and there are only quest reception place event in winter season.
• Figure 5: Guild NPC Response Return Value Ratios in Task 1_train.json (Green = return; Red = no return)