Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PONTE: Personalized Orchestration for Natural Language Trustworthy Explanations

Vittoria Vineis, Matteo Silvestri, Lorenzo Antonelli, Filippo Betello, Gabriele Tolomei

Abstract

Explainable Artificial Intelligence (XAI) seeks to enhance the transparency and accountability of machine learning systems, yet most methods follow a one-size-fits-all paradigm that neglects user differences in expertise, goals, and cognitive needs. Although Large Language Models can translate technical explanations into natural language, they introduce challenges related to faithfulness and hallucinations. To address these challenges, we present PONTE (Personalized Orchestration for Natural language Trustworthy Explanations), a human-in-the-loop framework for adaptive and reliable XAI narratives. PONTE models personalization as a closed-loop validation and adaptation process rather than prompt engineering. It combines: (i) a low-dimensional preference model capturing stylistic requirements; (ii) a preference-conditioned generator grounded in structured XAI artifacts; and (iii) verification modules enforcing numerical faithfulness, informational completeness, and stylistic alignment, optionally supported by retrieval-grounded argumentation. User feedback iteratively updates the preference state, enabling quick personalization. Automatic and human evaluations across healthcare and finance domains show that the verification-refinement loop substantially improves completeness and stylistic alignment over validation-free generation. Human studies further confirm strong agreement between intended preference vectors and perceived style, robustness to generation stochasticity, and consistently positive quality assessments.

PONTE: Personalized Orchestration for Natural Language Trustworthy Explanations

Abstract

Explainable Artificial Intelligence (XAI) seeks to enhance the transparency and accountability of machine learning systems, yet most methods follow a one-size-fits-all paradigm that neglects user differences in expertise, goals, and cognitive needs. Although Large Language Models can translate technical explanations into natural language, they introduce challenges related to faithfulness and hallucinations. To address these challenges, we present PONTE (Personalized Orchestration for Natural language Trustworthy Explanations), a human-in-the-loop framework for adaptive and reliable XAI narratives. PONTE models personalization as a closed-loop validation and adaptation process rather than prompt engineering. It combines: (i) a low-dimensional preference model capturing stylistic requirements; (ii) a preference-conditioned generator grounded in structured XAI artifacts; and (iii) verification modules enforcing numerical faithfulness, informational completeness, and stylistic alignment, optionally supported by retrieval-grounded argumentation. User feedback iteratively updates the preference state, enabling quick personalization. Automatic and human evaluations across healthcare and finance domains show that the verification-refinement loop substantially improves completeness and stylistic alignment over validation-free generation. Human studies further confirm strong agreement between intended preference vectors and perceived style, robustness to generation stochasticity, and consistently positive quality assessments.
Paper Structure (12 sections, 1 equation, 2 figures, 2 tables)

This paper contains 12 sections, 1 equation, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background and Prior Work
  3. Proposed Solution
  4. Empirical Validation
  5. Experimental Setup
  6. Evaluation Setup
  7. Automatic Evaluation.
  8. Human evaluation.
  9. Main Results
  10. Automatic evaluation.
  11. Human evaluation.
  12. Conclusion and Future Work

Figures (2)

  • Figure 1: Visual overview of PONTE.
  • Figure 2: Proportion of respondents (N=17) selecting Agree or Strongly Agree for three evaluation items: Linguistic Realization Quality ("The narrative is clear, logically structured, and easy to understand"), Actionable Insights ("The narrative helps me understand which changes would be necessary to obtain a different outcome") and Overall Satisfaction ("The narrative is satisfactory and appropriate for understanding the model’s decision'"), disaggregated by use-case persona and personal field of expertise.