What question did this study set out to answer?

To develop a lexicon-free, prompt-based evaluation framework for measuring intersectional demographic stereotypes and representational harms in large language model outputs.

May 29, 2023Open Access

Marked Personas: Using Natural Language Prompts to Measure Stereotypes in Language Models

View Full Paper

Ask AI

Bookmark

To develop a lexicon-free, prompt-based evaluation framework for measuring intersectional demographic stereotypes and representational harms in large language model outputs.
Operationalized the sociolinguistic concept of markedness to contrast generated personas of target demographic groups against unmarked default groups.
Prompted GPT-3.5 and GPT-4 to produce natural language descriptions of marked and unmarked demographics without relying on predefined lexicons or manual labels.
Identified statistically distinguishing vocabulary across groups and compared LLM stereotype rates against human-written portrayals.
Portrayals generated by GPT-3.5 and GPT-4 contained higher rates of racial stereotypes than human-authored portrayals under identical prompts.
Distinguishing terms for non-white and non-male groups demonstrated systematic othering, exoticization, and intersectional tropes, including tropicalism and hypersexualization.

Discussion

No takes yet. Share an insight, caveat, or question.

Implication

Prompt-based evaluation reveals elevated rates of racial stereotyping and exoticization in large language models, highlighting representational risks in downstream text generation.

Cite This Study

Cheng et al. (2023) studied this question.

synapsesocial.com/papers/6a153b91cb0379474a8208fe https://doi.org/10.48550/arxiv.2305.18189

View Full Paper

Ask AI

Bookmark

Implication

Prompt-based evaluation reveals elevated rates of racial stereotyping and exoticization in large language models, highlighting representational risks in downstream text generation.

Marked Personas: Using Natural Language Prompts to Measure Stereotypes in Language Models

Key Points

Discussion

Implication

Cite This Study

Marked Personas: Using Natural Language Prompts to Measure Stereotypes in Language Models

Key Points

Discussion

Implication

Cite This Study

Also Consider

Also Consider