Continuous Interpretive Steering for Scalar Diversity

TL;DR

This paper introduces Continuous Interpretive Steering (CIS) and GraSD, a dataset for probing graded pragmatic interpretation in large language models, revealing that models encode scalar diversity sensitivity in their representations.

Contribution

The study presents a novel method, CIS, for systematically probing graded pragmatic interpretation in LLMs, supported by a new dataset GraSD that encodes scalar diversity.

Findings

Graded activation steering yields interpretive shifts aligned with scalar diversity.

Uniform activation increases pragmatic interpretations but collapses item-level variation.

Models encode scalar diversity sensitivity in their representation space.

Abstract

Pragmatic inference is inherently graded. Different lexical items give rise to pragmatic enrichment to different degrees. Scalar implicature exemplifies this property through scalar diversity, where implicature strength varies across scalar items. However, evaluations of pragmatic inference in large language models (LLMs) often rely on prompt-based manipulations. Beyond prompt-level effects, this study introduces Continuous Interpretive Steering (CIS), a method that probes graded pragmatic interpretation by treating activation-level steering strength as a continuous experimental variable. To support this analysis, this study introduces a new dataset, GraSD, which encodes graded scalar diversity. Experiments on four LLMs show that uniform activation steering increases pragmatic interpretations globally but collapses item-level variation, whereas graded activation steering yields differentiated interpretive shifts aligned with scalar diversity grades. It indicates that graded sensitivity is encoded in the representation space and can be systematically recovered through controlled intervention. Together, CIS and GraSD provide a principled framework for evaluating graded pragmatic sensitivity in LLMs.