CALMA: A Process for Deriving Context-aligned Axes for Language Model Alignment

TL;DR

CALMA is a participatory methodology that derives context-specific axes for evaluating and aligning language models, addressing the limitations of broad, Western-centric benchmarks by incorporating diverse community priorities.

Contribution

The paper introduces CALMA, a novel grounded approach for eliciting context-relevant axes for language model evaluation and alignment, emphasizing participatory and open-ended processes.

Findings

CALMA surfaced community-specific priorities absent from standard benchmarks.

Evaluation practices based on CALMA are more open-ended and use-case-driven.

The methodology enhances pluralistic and transparent alignment pipelines.

Abstract

Datasets play a central role in AI governance by enabling both evaluation (measuring capabilities) and alignment (enforcing values) along axes such as helpfulness, harmlessness, toxicity, quality, and more. However, most alignment and evaluation datasets depend on researcher-defined or developer-defined axes curated from non-representative samples. As a result, developers typically benchmark models against broad (often Western-centric) values that overlook the varied contexts of their real-world deployment. Consequently, models trained on such proxies can fail to meet the needs and expectations of diverse user communities within these deployment contexts. To bridge this gap, we introduce CALMA (Context-aligned Axes for Language Model Alignment), a grounded, participatory methodology for eliciting context-relevant axes for evaluation and alignment. In a pilot with two distinct communities, CALMA surfaced novel priorities that are absent from standard benchmarks. Our findings demonstrate the value of evaluation practices based on open-ended and use-case-driven processes. Our work advances the development of pluralistic, transparent, and context-sensitive alignment pipelines.