A Proactive Multi-Agent Dialogue Framework for Assessing Social Language Disorder Traits in Autism

Background

SLD is a characteristic feature of autism where individuals exhibit echoic repetition, pronoun displacement, and stereotyped media quoting. These behaviors are typically absent in spontaneous conversations but emerge only during structured clinical assessments. The latency of these behaviors means that questioning strategy selection significantly influences the diagnostic information yield.

Key Points

• Proactive Questioning Strategy: TPA uses a multi-agent dialogue framework where a doctor agent reasons about unobserved traits and selects clinically grounded strategies to generate targeted questions.
• **Patient Agent

🔗 Related Read: Autonomous AI systems test governance in physical environments

🔗 Related Read: AI Is Taking Over the Most Cursed Job in the World

**: Grounded in real ADOS-2 clinical data, this patient agent enables reproducible evaluation without actual patient participation, ensuring validity across three independent experiments.

• Performance Metrics: TPA outperforms six competitive dialogue planning baselines, achieving 82.1% SLD trait coverage and an absolute gain of +0.170 in diagnostic efficiency (AUCC: 0.628) compared to automated replays by trained clinicians.

Significance

• Proactive Strategy Selection: TPA demonstrates the potential for proactive questioning strategies to significantly improve the efficiency of automated SLD trait assessment, with direct implications for scalable AI-assisted clinical screening.
• Reproducibility and Validity: The use of a patient agent based on real clinical data ensures reproducible evaluation, enhancing the reliability of results.
• Clinical Impact: Improved diagnostic efficiency could lead to more accurate and timely identification of SLD traits, facilitating better support and intervention for individuals with autism.

TPA's proactive approach highlights the importance of dynamic strategy selection in diagnosing complex linguistic behaviors associated with SLD.