Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

NASP-T: A Fuzzy Neuro-Symbolic Transformer for Logic-Constrained Aviation Safety Report Classification

Fadi Al Machot, Fidaa Al Machot

TL;DR

NASP-T addresses multi-label aviation safety report classification by enforcing domain logic through a neuro-symbolic pipeline that combines ASP-based rules with a transformer. It introduces rule-based data augmentation and a differentiable fuzzy ASP regularizer, validated by Clingo, to produce predictions that are both accurate and logically consistent. The approach yields substantial reductions in rule violations on the ASRS dataset while maintaining competitive F1 scores, marking a step toward trustworthy, interpretable safety-critical NLP. The methodology generalizes to other high-stakes domains by uniting symbolic reasoning with scalable neural models $L_{total} = L_{BCE} + L_{fuzzy}$, and paves the way for deeper integration of logic into end-to-end learning.

Abstract

Deep transformer models excel at multi-label text classification but often violate domain logic that experts consider essential, an issue of particular concern in safety-critical applications. We propose a hybrid neuro-symbolic framework that integrates Answer Set Programming (ASP) with transformer-based learning on the Aviation Safety Reporting System (ASRS) corpus. Domain knowledge is formalized as weighted ASP rules and validated using the Clingo solver. These rules are incorporated in two complementary ways: (i) as rule-based data augmentation, generating logically consistent synthetic samples that improve label diversity and coverage; and (ii) as a fuzzy-logic regularizer, enforcing rule satisfaction in a differentiable form during fine-tuning. This design preserves the interpretability of symbolic reasoning while leveraging the scalability of deep neural architectures. We further tune per-class thresholds and report both standard classification metrics and logic-consistency rates. Compared to a strong Binary Cross-Entropy (BCE) baseline, our approach improves micro- and macro-F1 scores and achieves up to an 86% reduction in rule violations on the ASRS test set. To the best of our knowledge, this constitutes the first large-scale neuro-symbolic application to ASRS reports that unifies ASP-based reasoning, rule-driven augmentation, and differentiable transformer training for trustworthy, safety-critical NLP.

NASP-T: A Fuzzy Neuro-Symbolic Transformer for Logic-Constrained Aviation Safety Report Classification

TL;DR

NASP-T addresses multi-label aviation safety report classification by enforcing domain logic through a neuro-symbolic pipeline that combines ASP-based rules with a transformer. It introduces rule-based data augmentation and a differentiable fuzzy ASP regularizer, validated by Clingo, to produce predictions that are both accurate and logically consistent. The approach yields substantial reductions in rule violations on the ASRS dataset while maintaining competitive F1 scores, marking a step toward trustworthy, interpretable safety-critical NLP. The methodology generalizes to other high-stakes domains by uniting symbolic reasoning with scalable neural models , and paves the way for deeper integration of logic into end-to-end learning.

Abstract

Deep transformer models excel at multi-label text classification but often violate domain logic that experts consider essential, an issue of particular concern in safety-critical applications. We propose a hybrid neuro-symbolic framework that integrates Answer Set Programming (ASP) with transformer-based learning on the Aviation Safety Reporting System (ASRS) corpus. Domain knowledge is formalized as weighted ASP rules and validated using the Clingo solver. These rules are incorporated in two complementary ways: (i) as rule-based data augmentation, generating logically consistent synthetic samples that improve label diversity and coverage; and (ii) as a fuzzy-logic regularizer, enforcing rule satisfaction in a differentiable form during fine-tuning. This design preserves the interpretability of symbolic reasoning while leveraging the scalability of deep neural architectures. We further tune per-class thresholds and report both standard classification metrics and logic-consistency rates. Compared to a strong Binary Cross-Entropy (BCE) baseline, our approach improves micro- and macro-F1 scores and achieves up to an 86% reduction in rule violations on the ASRS test set. To the best of our knowledge, this constitutes the first large-scale neuro-symbolic application to ASRS reports that unifies ASP-based reasoning, rule-driven augmentation, and differentiable transformer training for trustworthy, safety-critical NLP.

Paper Structure

This paper contains 18 sections, 12 equations, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. Domain-specific Language Models
  4. Neuro-Symbolic Learning and Constraint Integration
  5. Dataset and Rule Generation
  6. Methodology
  7. Problem Setup
  8. Baseline Loss: Binary Cross-Entropy (BCE)
  9. Data Augmentation Using Logical Rules
  10. Fuzzy Rule Regularization
  11. Relation to Symbolic Clingo Reasoning
  12. Total Objective Function
  13. Rule Violation Metrics
  14. Total Rule Violations.
  15. Violation Rate.
  16. ...and 3 more sections

Figures (1)

  • Figure 1: Overview of the proposed Neuro-Symbolic ASP-Constrained Transformer (NASP-T) framework. The pipeline integrates transformer-based text encoding with domain knowledge expressed as soft Answer Set Programming (ASP) rules. These rules are first used for logic-driven data augmentation, enriching the training set with rule-consistent examples, and later as a fuzzy ASP regularizer that penalizes logical inconsistencies during optimization. The BCE loss ensures data-driven learning, while the fuzzy ASP term enforces symbolic coherence. Together, they yield predictions that are both statistically accurate and logically consistent—crucial for safety-critical NLP tasks.