---
title: "LEX-EC: A Lexical Evidence-Channel Audit Framework for Zero-Shot LLM Personality Classification in Black-Box Settings"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2607.24435v1"
permalink: "https://feed.astralor.com/items/itm_47a9ccb7f48421d571fb42a2"
published: "2026-07-27T13:45:12.000Z"
score: 70
channel: "tech"
tags: ["LEX-EC","Personality Classification","Large Language Models","Lexical Ablation","Black-Box Interpretability"]
---

# LEX-EC: A Lexical Evidence-Channel Audit Framework for Zero-Shot LLM Personality Classification in Black-Box Settings

## AI commentary

> This paper offers a novel audit framework for black-box interpretability in LLM personality classification, especially notable for revealing how different text genres significantly affect trait association signals, making it highly valuable for researchers working on model transparency and fairness.

## At a glance

The paper introduces LEX-EC, a reusable black-box audit framework for zero-shot LLM personality classification interpretability. Combining prevalence and agreement diagnostics with controlled lexical ablation, it distinguishes marginal-distribution effects from trait-associated signal. Experiments reveal sharply different profiles across text genres: free-form essays show broadest but weak signal; in graduate introductions, Extraversion association weakens after masking; single Facebook statuses yield little stable evidence. Masking topical/demographic content weakens some associations while function words, affective terms, and cognitive-style vocabulary remain detectable. The framework validates a novel application of lexical methods to black-box interpretability in personality labeling.

## Key points

- Proposes LEX-EC, a black-box audit framework combining prevalence, agreement diagnostics, and controlled lexical ablation.
- Distinguishes marginal-distribution effects from trait-associated signals, revealing varying signal strengths across text genres.
- After masking topical and demographic content, function words, affective terms, and cognitive-style vocabulary still yield detectable associations.
- Validates a novel application of lexical methods to black-box interpretability, enhancing model transparency.

## Full text

Large language models may easily assign personality labels from text, but model interpretability remains an open problem. To address this gap, we introduce LEX-EC, a reusable black-box audit framework combining prevalence and agreement diagnostics with controlled lexical ablation to distinguish marginal-distribution effects from trait-associated signal recoverable under restricted evidence. Using this framework, we illustrate how various text genres may exhibit sharply different profiles: free-form essay text contains the broadest, but still weak, signal; in graduate student introductions, an observable Extraversion association weakened after masking; and single Facebook statuses yield little stable evidence even in a trait-balanced sample, indicating a possible lower bound of content or length. Masking topical and demographic content weakened some associations while leaving others detectable from function words, affective terms, and cognitive-style vocabulary. Linguistic prompting shifted model self-explanations but did not eliminate topical content. LEX-EC jointly evaluates classification prevalence, item-level association, chance-corrected agreement, persistence under lexical restriction, and prompt sensitivity in model-generated explanations. Across datasets, models, and prompts, LEX-EC characterizes how trait associations may vary with available lexical evidence, introducing a novel application of lexical methods to black-box interpretability in personality labeling.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_47a9ccb7f48421d571fb42a2