Datasets:

jonc
/

cei-benchmark

language:
  - en
license: cc-by-4.0
size_categories:
  - n<1K
task_categories:
  - text-classification
task_ids:
  - emotion-classification
tags:
  - pragmatic-reasoning
  - theory-of-mind
  - emotion-inference
  - indirect-speech
  - benchmark
  - multi-annotator
  - plutchik-emotions
  - vad-dimensions
dataset_info:
  features:
    - name: id
      dtype: int64
    - name: subtype
      dtype: string
    - name: context
      dtype: string
    - name: speaker
      dtype: string
    - name: listener
      dtype: string
    - name: utterance
      dtype: string
    - name: power_relation
      dtype: string
    - name: social_context
      dtype: string
    - name: gold_standard
      dtype: string
  splits:
    - name: train
      num_examples: 210
    - name: validation
      num_examples: 45
    - name: test
      num_examples: 45

CEI: A Benchmark for Evaluating Pragmatic Reasoning in Language Models

Dataset Description

CEI (Contextual Emotional Inference) is a benchmark of 300 expert-authored scenarios for evaluating how well language models interpret pragmatically complex utterances in social contexts. Each scenario presents a communicative exchange involving indirect speech (sarcasm, mixed signals, strategic politeness, passive aggression, or deflection) where the speaker's literal words diverge from their actual emotional state.

Paper: CEI: A Benchmark for Evaluating Pragmatic Reasoning in Language Models (DMLR 2026)
Repository: https://github.com/jon-chun/cei-tom-dataset-base
Zenodo: https://doi.org/10.5281/zenodo.18528706
License: CC-BY-4.0 (data), MIT (code)

Dataset Structure

Scenarios

300 scenarios across 5 pragmatic subtypes (60 each)
3 independent annotations per scenario (900 total)
Predefined splits: train (210), validation (45), test (45), stratified by subtype and power relation

Pragmatic Subtypes

Subtype	Description	Fleiss' kappa
Sarcasm/Irony	Speaker says the opposite of what they mean	0.25
Passive Aggression	Hostility expressed through superficial compliance	0.22
Strategic Politeness	Polite language masking negative intent	0.20
Mixed Signals	Contradictory verbal and contextual cues	0.16
Deflection/Misdirection	Speaker redirects to avoid revealing feelings	0.06

Labels

Primary emotion: One of Plutchik's 8 basic emotions (joy, trust, fear, surprise, sadness, disgust, anger, anticipation)
VAD ratings: Valence, Arousal, Dominance on 7-point scales mapped to [-1.0, +1.0]
Confidence: Annotator self-reported confidence
Gold standard: Majority vote with expert adjudication

Power Relations

Peer (72%), High-to-Low authority (20%), Low-to-High authority (7%)

Key Statistics

Inter-annotator agreement: Overall kappa = 0.21 (fair), ranging from 0.06 (deflection) to 0.25 (sarcasm)
Human accuracy (vs. gold): 61% mean, 14.3% unanimous, 31.3% three-way split
Best LLM baseline: 25.7% accuracy (Phi-4, zero-shot) vs. 54% human majority agreement
Random baseline: 12.5% (8-class)

Intended Uses

Benchmarking LLM pragmatic reasoning capabilities
Diagnosing model failure modes on indirect speech subtypes
Research on emotion inference, social AI, Theory of Mind
Soft-label training using per-annotator distributions

Limitations

All scenarios are expert-authored (not naturalistic)
English only
15 undergraduate annotators from a single institution
Small scale (300 scenarios) optimized for annotation quality over quantity

Citation

@article{chun2026cei,
  title={CEI: A Benchmark for Evaluating Pragmatic Reasoning in Language Models},
  author={Chun, Jon and Sussman, Hannah and Mangine, Adrian and Kocaman, Murathan and Sidorko, Kirill and Koirala, Abhigya and McCloud, Andre and Eisenbeis, Gwen and Akanwe, Wisdom and Gassama, Moustapha and Gonzalez Chirinos, Eliezer and Enright, Anne-Duncan and Dunson, Peter and Ng, Tiffanie and von Rosenstiel, Anna and Idowu, Godwin},
  journal={Journal of Data-centric Machine Learning Research (DMLR)},
  year={2026}
}