Prompt Stylometry for On-Device Affect-Adaptive AI: A Feasibility Study in Linguistic Signal Detection and Response Steering

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Cybersecurity & Data Privacy, Internet of Things (IoT) & Connected Devices · Depth: Expert, medium

Summary

Debmalya Pal introduces Prompt Stylometry, a framework designed for detecting latent linguistic signals in user prompts to steer large language model (LLM) response generation. Unlike most LLMs optimized for generalized behavior, this framework adapts to fine-grained signals reflecting a user's affective state and cognitive style, such as analytical or indecisive reasoning. To mitigate substantial privacy risks associated with server-side processing of psychological states, the system employs a fully on-device architecture, ensuring no interaction data leaves the user's device. The study benchmarks lexicon-based, neural, and generative annotation paradigms across 600 synthetic prompts spanning 30 stylometric profiles. It evaluates affect-adaptive response steering using two small language model families under 5B parameters, demonstrating feasibility on consumer hardware while noting challenges in annotation paradigm sensitivity and cross-profile transfer.

Key takeaway

For NLP Engineers developing privacy-sensitive AI assistants, this research confirms the viability of integrating affect-adaptive capabilities directly on user devices. You should prioritize on-device architectures to detect user emotional and cognitive states from prompts without compromising privacy. Experiment with different annotation paradigms, such as lexicon-based or neural methods, to optimize signal detection. This approach can enhance response steering for small language models under 5B parameters, addressing challenges in transferability.

Key insights

Prompt Stylometry enables privacy-preserving, on-device detection of user affect and cognitive style to steer LLM responses.

Principles

• User prompts contain latent affective and cognitive-style signals.
• On-device processing is crucial for user privacy in affect detection.
• Annotation paradigms impact signal detection and response steering.

Method

The framework detects affective and cognitive-style signals from user prompts, then uses these signals to steer response generation in small language models, all within a fully on-device architecture.

In practice

• Implement on-device LLM architectures for sensitive user profiling.
• Benchmark lexicon-based, neural, and generative annotation methods.
• Evaluate affect-adaptive steering with models under 5B parameters.

Topics

• Prompt Stylometry
• On-Device AI
• Affective Computing
• Cognitive Style Detection
• Privacy-Preserving AI
• Small Language Models

Best for: Research Scientist, AI Scientist, NLP Engineer, AI Security Engineer

Related on AIssential

• PromptPrint: Behavioral Biometrics Through Natural Language Prompting in LLMs — Prompt-based identity, a new behavioral biometric, is primarily encoded in a user's surface-level lexical choices, not semantic intent.
• Explainable Detection of Depression Status Shifts from User Digital Traces — A BERT-LLM framework detects mental health shifts from digital traces, providing explainable, temporally coherent reports.
• Beating the Style Detector: Three Hours of Agentic Research on the AI-Text Arms Race — Agentic research rapidly reproduces and extends NLP studies, demonstrating LLMs' ability to evade AI-text detection.
• Stop Being Nice to Your Chatbot: The Surprising Science of Prompt Politeness — Polite prompts reduce LLM accuracy and increase computational cost, while direct or emotionally-charged prompts improve performance.
• From Feelings to Metrics: Understanding and Formalizing How Users Vibe-Test LLMs — Formalized "vibe-testing" bridges the gap between LLM benchmark scores and real-world user utility by personalizing evaluation.
• Memory Bear AI Memory Science Engine for Multimodal Affective Intelligence: A Technical Report — Affective intelligence benefits from a memory-centered approach that models emotion as an evolving, structured memory variable.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.