Finding Sense in Nonsense with Generated Contexts: Perspectives from Humans and Language Models

2026-06-25 · Source: Paper Index on ACL Anthology · Field: Technology & Digital — Artificial Intelligence & Machine Learning · Depth: Expert, short

Summary

The paper by Olsen and Padó (2026) investigates how humans and Large Language Models (LLMs) distinguish between anomalous and truly nonsensical sentences. They collected sensicality judgments from human raters and LLMs on sentences from five semantically deviant datasets, both with and without context. Their findings indicate that human raters perceive most sentences as merely anomalous, capable of interpretation with a supporting context, and only a small fraction as genuinely nonsensical. Furthermore, the research demonstrates that LLMs possess substantial skill in generating plausible contexts for these anomalous cases, suggesting their capability in semantic interpretation beyond simple anomaly detection. This work was presented at the 15th Joint Conference on Lexical and Computational Semantics (*SEM 2026) in San Diego, California, United States, and spans pages 98–110.

Key takeaway

For NLP Engineers developing semantic interpretation models, you should re-evaluate existing "nonsensical" datasets, as many sentences are merely anomalous and interpretable with context. Consider integrating LLMs to generate supporting contexts for ambiguous inputs, enhancing your model's ability to distinguish true nonsense from context-dependent anomalies. This approach can improve the robustness of semantic understanding systems.

Key insights

LLMs can effectively generate contexts to make anomalous sentences sensical, mirroring human judgment.

Principles

• Human raters find most "nonsensical" sentences merely anomalous.
• Context is crucial for semantic interpretation.
• LLMs excel at context generation for anomalies.

Method

Collected sensicality judgments from human raters and LLMs on sentences from five semantically deviant datasets, evaluating them context-free and with generated contexts.

In practice

• Test LLM semantic interpretation with anomalous inputs.
• Use LLMs to generate clarifying contexts.
• Re-evaluate "nonsensical" datasets with context.

Topics

• Semantic Interpretation
• Large Language Models
• Context Generation
• Nonsensical Sentences
• Anomalous Sentences
• Dataset Evaluation

Best for: Research Scientist, AI Scientist, NLP Engineer

Related on AIssential

• Your AI Probably Understands Your Marriage Joke Better Than Your SQL Parser — LLMs excel at interpreting human language's implied, non-literal meanings, unlike literal parsers.
• Revisiting the Systematicity in Negation in the Era of In-Context Learning — LLMs partially grasp negation via in-context learning, but struggle with robust scope recognition and function vector composition for it.
• Constrained Semantic Decompression in LLMs through Persian Proverb-Conditioned Story Generation — LLMs struggle with "constrained semantic decompression," failing to translate abstract cultural knowledge like proverbs into faithful narratives.
• Slang Context-based Inference Enhancement via Greedy Search-Guided Chain-of-Thought Prompting — Slang interpretation in LLMs benefits more from structured prompting than from increased model size or temperature.
• Hybrid Human-LLM Corpus Construction and LLM Evaluation for the Caused-Motion Construction — LLMs struggle with complex linguistic constructions like CMC, necessitating novel corpus creation methods.
• Probing Semantic Alignment, Lexical Invariance, and Syntactic Influence in LLM Metaphor Processing — LLMs struggle with deep metaphor comprehension, often misinterpreting concepts and relying on lexical associations over context or full syntactic understanding.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.