X-Blocks: Linguistic Building Blocks of Natural Language Explanations for Automated Vehicles

TL;DR

X-Blocks (eXplanation Blocks) is introduced, a hierarchical analytical framework that identifies the linguistic building blocks of natural language explanations for AVs at three levels: context, syntax, and lexicon, and provides evidence-based linguistic design principles for generating scenario-aware explanations that support transparency, user trust, and cognitive accessibility in automated driving systems.

Abstract

Natural language explanations play a critical role in establishing trust and acceptance of automated vehicles (AVs), yet existing approaches lack systematic frameworks for analysing how humans linguistically construct driving rationales across diverse scenarios. This paper introduces X-Blocks (eXplanation Blocks), a hierarchical analytical framework that identifies the linguistic building blocks of natural language explanations for AVs at three levels: context, syntax, and lexicon. At the context level, we propose RACE (Reasoning-Aligned Classification of Explanations), a multi-LLM ensemble framework that combines Chain-of-Thought reasoning with self-consistency mechanisms to robustly classify explanations into 32 scenario-aware categories. Applied to human-authored explanations from the Berkeley DeepDrive-X dataset, RACE achieves 91.45 percent accuracy and a Cohens kappa of 0.91 against cases with human annotator agreement, indicating near-human reliability for context classification. At the lexical level, log-odds analysis with informative Dirichlet priors reveals context-specific vocabulary patterns that distinguish driving scenarios. At the syntactic level, dependency parsing and template extraction show that explanations draw from a limited repertoire of reusable grammar families, with systematic variation in predicate types and causal constructions across contexts. The X-Blocks framework is dataset-agnostic and task-independent, offering broad applicability to other automated driving datasets and safety-critical domains. Overall, our findings provide evidence-based linguistic design principles for generating scenario-aware explanations that support transparency, user trust, and cognitive accessibility in automated driving systems.

Figures (13)

• Figure 1: (a) Key factors affecting the effectiveness of explanations in AVs: Insights from user-centered studies yousefizadeh2025psylingxav and (b) Building blocks of language vajjala2020practical.
• Figure 2: X-Blocks Level 1: The proposed RACE framework, (a) a CoT-SC-based classification pipeline for context scenario labelling of explanations in driving and (b) an evaluation pipeline for assessing reliability and human-model agreement.
• Figure 3: X-Blocks Level 2: Extracting syntactic building blocks of driving explanations. Raw explanations are parsed using dependency parsing to obtain interpretable syntactic cues (Stage 1), augmented with coverage-oriented detection of causal and purpose constructions (Stage 2), combined into discrete syntactic signatures representing grammar families (Stage 3), and abstracted into reusable slot-based templates that preserve high-level structural patterns (Stage 4).
• Figure 4: X-Blocks Level 3: Pipeline for extracting morpheme- and lexeme-level building blocks of driving explanations. Raw explanations are normalised via lemmatisation and filtering (Stage 1), scored for contextual distinctiveness using log-odds with an informative Dirichlet prior (Stage 2), and mapped back to representative surface forms for human interpretability (Stage 3).
• Figure 5: Distribution of driving-context labels in the annotated BDD-X explanation corpus. Counts are shown on a logarithmic scale, revealing a highly imbalanced, long-tailed distribution across scenario contexts.