Research

Paper

AI LLM March 20, 2026

When Contextual Inference Fails: Cancelability in Interactive Instruction Following

Authors

Natalia Bila, Kata Naszádi, Alexandra Mayn, Christof Monz

Abstract

We investigate the separation of literal interpretation from contextual inference in a collaborative block-building task where a builder must resolve underspecified instructions using contextual inferences. Building on an existing two-speaker psycholinguistic paradigm -- which contrasts a pragmatically cooperative speaker with one who is only literally reliable -- we introduce Build What I Mean (BWIM), an interactive benchmark for contextual meaning construction. In BWIM, models must resolve ambiguity by either performing a contextual inference or requesting clarification at a small communication cost. Evaluating several state-of-the-art LLMs, we find a dissociation between judgment and action: while models detect speaker unreliability in explicit confidence ratings, they fail to exploit this information to guide efficient clarification behavior. Instead, we observe suboptimal strategies, such as partner-blind over-clarification and question-averse guessing under uncertainty.

Metadata

arXiv ID: 2603.19997
Provider: ARXIV
Primary Category: cs.CL
Published: 2026-03-20
Fetched: 2026-03-23 16:54

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.19997v1</id>\n    <title>When Contextual Inference Fails: Cancelability in Interactive Instruction Following</title>\n    <updated>2026-03-20T14:46:59Z</updated>\n    <link href='https://arxiv.org/abs/2603.19997v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.19997v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>We investigate the separation of literal interpretation from contextual inference in a collaborative block-building task where a builder must resolve underspecified instructions using contextual inferences. Building on an existing two-speaker psycholinguistic paradigm -- which contrasts a pragmatically cooperative speaker with one who is only literally reliable -- we introduce Build What I Mean (BWIM), an interactive benchmark for contextual meaning construction. In BWIM, models must resolve ambiguity by either performing a contextual inference or requesting clarification at a small communication cost. Evaluating several state-of-the-art LLMs, we find a dissociation between judgment and action: while models detect speaker unreliability in explicit confidence ratings, they fail to exploit this information to guide efficient clarification behavior. Instead, we observe suboptimal strategies, such as partner-blind over-clarification and question-averse guessing under uncertainty.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-03-20T14:46:59Z</published>\n    <arxiv:primary_category term='cs.CL'/>\n    <author>\n      <name>Natalia Bila</name>\n    </author>\n    <author>\n      <name>Kata Naszádi</name>\n    </author>\n    <author>\n      <name>Alexandra Mayn</name>\n    </author>\n    <author>\n      <name>Christof Monz</name>\n    </author>\n  </entry>"
}