license: cc-by-4.0
language:
- en
task_categories:
- text-generation
- visual-question-answering
pretty_name: imprecision-bench
tags:
- pragmatics
- computational-linguistics
- multimodal
- time-reference
- benchmarking
- LLM-evaluation
imprecision-bench
A multimodal benchmark for evaluating whether LLMs calibrate linguistic precision to pragmatic context, paired with 475 human productions and a peer-reviewed RSA baseline (r² ≈ 0.97).
This dataset accompanies the paper:
Modeling (Im)precision in Context Roland Mühlenbernd, Stephanie Solt Linguistics Vanguard, 2022 [Paper] · [Source Data] · [Companion Repo]
Notebook
notebook.ipynb — guided walkthrough: data loading, sample evaluation (1 row per condition), clock-reading accuracy, pragmatic shift analysis, and Wasserstein distance against the human baseline.
 |
Interactive (Google account required) |
 |
Interactive (no account needed; slower start) |
 |
Read-only rendered view |
Overview
When a witness tells a police officer "It happened at 8:31" rather than "It happened around 8:30," the choice of precision level is not arbitrary — it reflects the interlocutor's needs, the communicative context, and the speaker's pragmatic judgment. This benchmark tests whether large language models and vision-language models make the same kind of context-sensitive precision adjustments that humans do.
The dataset contains 475 human time-reference productions across 24 conditions (12 clock states × 2 pragmatic contexts: police witness statement vs. neighbor at a party). Each item includes the clock image used as stimulus, a textual clock description for text-only LLM evaluation, annotated motive labels (12 categories, multi-label), and motive-explanation free text from a follow-up elicitation task.
The benchmark supports two evaluation tasks:
- Task 1 — Production: given a clock + scenario, produce a time expression to complete
"It happened ___." - Task 2 — Motive elicitation: given the production + context, explain why that wording was chosen.
Dataset
| Property | Value |
|---|---|
| Size | 475 items, 312 unique speakers |
| Conditions | 12 clock states × 2 contexts (police / neighbor) |
| Language | English |
| Modalities | Clock images (PNG) + textual clock descriptions |
| Annotation | Multi-label motive categories (12 classes); free-text motive explanations |
| Source | Mühlenbernd & Solt (2022), figshare DOI 10.6084/m9.figshare.21629531 |
| License | CC BY 4.0 |
Clock states
11 precise times (8:25–8:35, one minute apart) plus one range stimulus (8:26–8:34, shown as a yellow wedge on the clock face). The precise times span ±5 minutes around the canonical half-hour, allowing analysis of how offset magnitude interacts with rounding behavior.
Pragmatic contexts
- Police: formal witness statement; officer establishing a detailed event timeline.
- Neighbor: casual party conversation; neighbor curious about what happened.
Human productions show a reliable cross-context shift: police context elicits more precise time expressions; neighbor context elicits more rounding and approximation.
Quick Start
from datasets import load_dataset
ds = load_dataset("RolandM/imprecision-bench", split="train")
# Inspect a row
row = ds[0]
print(row["prompt"]) # Task 1 scenario text
print(row["clock_description"])# Text clock stimulus
print(row["production"]) # Human reference answer
print(row["motive_labels"]) # Annotated motive categories
Data Format
Each row represents one human production. Key columns:
| Column | Type | Description |
|---|---|---|
item_id |
string | Unique item identifier |
subject_id |
int | Anonymized participant ID |
context |
ClassLabel | police or neighbor |
target_time |
string | Canonical time (e.g. "8:30", "8:26-8:34") |
stimulus_type |
ClassLabel | precise or range |
signed_offset |
int8 | Minutes offset from 8:30 (null for range) |
abs_offset |
int8 | Absolute offset in minutes (null for range) |
production |
string | Human time expression (fills "It happened ___.") |
production_code |
int8 | Precision code for the production |
approximator |
ClassLabel | Lexical approximator used (e.g. around, about, none) |
motive_labels |
Sequence(ClassLabel) | Multi-label motive annotation (12 categories) |
motive_text |
string | Free-text motive explanation from follow-up task |
clock_image |
Image | PNG clock image (432 × 429 px) |
clock_description |
string | Textual clock description (with Clock description: tag) |
prompt |
string | Task 1 scenario text (unified across modalities) |
prompt_motive |
string | Task 2 context-embedded motive elicitation prompt |
Motive label categories
Precision, Accuracy, Info lack, Misinfo, Safe, H needs, Context, S ease, H ease, Habit, Sound, Other
Approximator values
none, around, about, just before/after, approximately, ish, nearly, roughly, round about
Evaluation
Task 1 — Production (multimodal)
import anthropic, base64
from datasets import load_dataset
ds = load_dataset("RolandM/imprecision-bench", split="train")
client = anthropic.Anthropic()
row = ds[0]
img_bytes = row["clock_image"].tobytes() # PIL Image
img_b64 = base64.b64encode(img_bytes).decode()
response = client.messages.create(
model="claude-opus-4-6",
max_tokens=50,
messages=[{
"role": "user",
"content": [
{"type": "image",
"source": {"type": "base64",
"media_type": "image/png",
"data": img_b64}},
{"type": "text", "text": row["prompt"]},
],
}],
)
print(response.content[0].text)
Task 1 — Production (text-description)
content = row["clock_description"] + "\n\n" + row["prompt"]
response = client.messages.create(
model="claude-opus-4-6",
max_tokens=50,
messages=[{"role": "user", "content": content}],
)
print(response.content[0].text)
Task 2 — Motive elicitation
response = client.messages.create(
model="claude-opus-4-6",
max_tokens=200,
messages=[{"role": "user", "content": row["prompt_motive"]}],
)
print(response.content[0].text)
Suggested metrics
- Production task: distribution distance from human productions (Wasserstein distance over precision codes); cross-context shift detection (does the model produce more rounding in neighbor vs. police context?); approximator usage rate.
- Motive task: label match against
motive_labels(multi-label F1); qualitative analysis of free-text explanations. - ESR / CDS: effect-size ratio and calibration distance score, as defined in Mühlenbernd (2026, CMCL).
Human baseline (RSA model)
A Rational Speech Act (RSA) speaker model fit to this dataset achieves r² ≈ 0.97 against human production distributions, providing a strong peer-reviewed reference baseline. See the companion repository for the model implementation.
Baseline Results
Results from three vision-language models evaluated on the full dataset (n = 475), using evaluate.py. All numbers are from Task 1 (production task); Task 2 (motive elicitation) is not yet scored and open for contribution.
Each response is classified as precise (conveys the exact target time), rounded (conveys a canonical approximation — 8:25, 8:30, or 8:35 — but not the exact time), or other (wrong, vague, or non-answer). The same classifier is applied to human productions for a direct apples-to-apples comparison.
Finding 1 — Modality gap
Models read textual clock descriptions far more accurately than clock images. Accuracy = fraction of responses where the exact target time is conveyed.
| Model | Image accuracy | Text accuracy |
|---|---|---|
| GPT-4o mini | 2.7% | 35.8% |
| Claude Haiku 4.5 | 0.2% | 38.1% |
| Gemini 2.5 Flash | 26.9% | 54.3% |
GPT-4o mini and Claude Haiku essentially fail to read clock images (≤3%), yet manage 35–38% on textual descriptions. Gemini is substantially better on the image task but still well below its own text performance. Open challenge: reliable analog-clock reading for current VLMs.
Finding 2 — Under-rounding
On off-round targets (8:26–8:34), models almost never choose to round to a canonical approximation, whereas humans do so frequently. Human-style rounding means saying "8:30" for a target of 8:28 — deliberately trading precision for simplicity. This is distinct from hedging language on an exact time (e.g., "around 8:30" when the target is 8:30).
| Source | Police rounding | Neighbor rounding |
|---|---|---|
| GPT-4o mini | 0.9% | 4.7% |
| Claude Haiku 4.5 | 0.0% | 0.0% |
| Gemini 2.5 Flash | 0.0% | 8.5% |
| Human baseline | 23.5% | 43.4% |
Models round at 5–25× lower rates than humans on off-round targets. Claude Haiku never rounds at all. Gemini is the closest to human-like: it does produce rounded responses in the neighbor context, though still far below the human rate. This under-rounding means models operate in a fundamentally different region of the precision-rounding space than humans do — they default to precision (or produce misreadings) where humans routinely choose approximation. Open challenge: models that use canonical approximations as a natural production strategy, not just as a fallback.
Finding 3 — Pragmatic shift
Humans produce more precise time expressions in the formal police context than in the casual neighbor context. Do models replicate this shift? (Text task, n = 475.)
"Δ cond" is the conditional shift — precise rate within responses that are either precise or rounded (i.e., excluding misreadings). This is the cleaner signal: it asks whether a model that can produce a time expression at all adjusts its precision level to context.
| Source | Police precise | Neighbor precise | Δ precise | Δ cond | WD |
|---|---|---|---|---|---|
| GPT-4o mini | 32.5% | 38.9% | −6.5% | −5.3% | 0.120 |
| Claude Haiku 4.5 | 31.2% | 37.7% | −6.5% | −10.7% | 0.066 |
| Gemini 2.5 Flash | 55.0% | 53.7% | +1.3% | +4.7% | 0.044 |
| Human baseline | 62.3% | 53.3% | +9.1% | +9.9% | 0.097 |
WD = Wasserstein distance between police and neighbor distributions (3-way coding: 0=precise, 1=rounded, 2=other). Higher = more context-sensitive.
GPT-4o mini and Claude Haiku show an inverted shift — they are less precise in the formal police context. Gemini moves in the right direction but at less than half the human magnitude. Open challenge: context-sensitive precision calibration at human magnitude.
Finding 4 — Off-round subset
Restricting to targets where rounding vs. precision is unambiguous (8:26–8:29, 8:31–8:34, n = 244) — the methodology of Mühlenbernd & Solt (2022) — the human rounding signal strengthens considerably. Δ cond = conditional precise shift (police − neighbor); p-value from one-tailed Mann-Whitney U (H₁: neighbor rounds more).
| Source | Police prec/rnd | Neighbor prec/rnd | Δ cond | p |
|---|---|---|---|---|
| GPT-4o mini | 24.3% / 0.9% | 30.2% / 4.7% | +9.9% | 0.039 * |
| Claude Haiku 4.5 | 21.7% / 0.0% | 26.4% / 0.0% | 0.0% | 1.000 |
| Gemini 2.5 Flash | 51.3% / 0.0% | 50.4% / 8.5% | +14.5% | 0.001 *** |
| Human baseline | 70.4% / 23.5% | 51.2% / 43.4% | +20.9% | **0.001 ***** |
Gemini replicates the human direction significantly (p = 0.001). GPT-4o mini shows a weak but significant effect (p = 0.039). Claude Haiku produces zero rounding responses on off-round targets in both contexts — it defaults to "other" (wrong or vague times), making pragmatic shift detection impossible.
Prompts
Task 1 — Police context, precise stimulus
[clock stimulus here — image or description]
One morning when you leave your house, you witness an automobile accident
in your street. You look at your watch when it happens. Later that day you
are invited to the police station to give a formal witness statement about
the accident. The police officer is trying to establish a detailed timeline
of the event. He asks you: "What time did the accident happen?" You remember
that it happened at the time shown on the clock as given above.
How would you answer in this situation? (Fill the blank)
"It happened ___."
For the range stimulus, "at the time" is replaced by "in the time range." For the neighbor context, the police station passage is replaced by a party-at-a-neighbor's-house framing; "He asks" becomes "She asks."
Task 2 — Police context, precise stimulus (example)
One morning when you leave your house, you witness an automobile accident
in your street. You look at your watch when it happens. Later, you gave a
formal witness statement at a police station. The officer, trying to
establish a detailed timeline, asked you what time the accident happened.
You knew that the accident happened at 8:31, and your answer was
"It happened just after half past eight". Why did you choose to answer
this way?
Caveats and Limitations
- Human baseline is image-only. Participants saw clock images; they did not see textual descriptions. Comparisons between text-description LLM outputs and human productions are meaningful but involve a modality translation step that should be flagged in analyses.
- Description format is a benchmark design choice. One canonical textual description format is fixed as part of the benchmark specification. Results may differ with alternative description strategies.
- Original stimulus wording. The original M&S 2022 experiment used "the clock on the left" (referring to GUI layout). This dataset uses "the clock as given above," which is layout-agnostic but a minor deviation from the source wording.
- Gendered interlocutors. The original stimuli used "He asks" for the police officer and "She asks" for the neighbor. These gender assignments are preserved here as faithful to the source experiment. Researchers should be aware of potential gender-stereotyping effects.
- Task 2 prompt is context-embedded. The original follow-up task gave participants only the minimal prompt ("In your task, you knew that…"). This dataset's
prompt_motiveembeds the police/neighbor context so single-turn LLM evaluation has access to the pragmatic framing. This is a deliberate design choice for LLM eval; the source wording is preserved in the paper. - 5 rows have empty
motive_text. Participants declined to respond on those items.prompt_motiveis still valid; human-baseline comparisons for Task 2 simply have no ground-truth for those rows. - English only. All productions are in English by English-speaking participants.
Citation
If you use this benchmark, please cite the original paper:
@article{muehlenbernd2022imprecision,
title = {Modeling (im)precision in context},
author = {M{\"u}hlenbernd, Roland and Solt, Stephanie},
journal = {Linguistics Vanguard},
year = {2022},
doi = {10.1515/lingvan-2022-0035}
}
Please also cite the source data:
@misc{muehlenbernd2022imprecision_data,
title = {Modeling (im)precision in context — supplementary data},
author = {M{\"u}hlenbernd, Roland and Solt, Stephanie},
year = {2022},
doi = {10.6084/m9.figshare.21629531}
}
License
This dataset is released under CC BY 4.0, consistent with the source data license on figshare. You are free to share and adapt the material for any purpose, provided appropriate credit is given.
Related Resources
- Companion repo (RSA model + analysis notebook): muehlenbernd/imprecision-in-context
- LLM social calibration paper (ESR/CDS metrics): muehlenbernd/llm-social-calibration · arXiv 2604.02512