AdMIRe - Advancing Multimodal Idiomaticity Representation
Multimodal idiomaticity task (SemEval-2025 Task 1, Subtask A): given a potentially idiomatic nominal compound in a context sentence, annotators rank five candidate images by how well each represents the sense in which the compound is used.
Configuration Fileconfig.yaml
This Potato config reproduces the annotation task. Save it as config.yaml and run potato start config.yaml to try it.
# AdMIRe - Advancing Multimodal Idiomaticity Representation
# Based on Pickard et al., SemEval-2025 Task 1 (ACL)
# Paper: https://aclanthology.org/2025.semeval-1.330/
# Dataset: https://semeval2025-task1.github.io/
#
# Subtask A: Each item presents a potentially idiomatic nominal compound
# (NC) within a context sentence, together with five candidate images.
# The compound may be used in its idiomatic (figurative) or literal sense.
# Annotators rank the five images according to how well each represents
# the sense in which the compound is used in the given context.
#
# Sense of the compound:
# - Idiomatic: the compound is used in its figurative/idiomatic sense
# - Literal: the compound is used in its literal/compositional sense
#
# Ranking:
# - Assign each image a distinct rank from 1 (best represents the sense)
# to 5 (worst represents the sense).
annotation_task_name: "AdMIRe - Advancing Multimodal Idiomaticity Representation"
task_dir: "."
data_files:
- sample-data.json
item_properties:
id_key: "id"
text_key: "context"
output_annotation_dir: "annotation_output/"
output_annotation_format: "json"
port: 8000
server_name: localhost
annotation_schemes:
- annotation_type: radio
name: sense_used
description: "How is the highlighted nominal compound used in this sentence?"
labels:
- "Idiomatic"
- "Literal"
keyboard_shortcuts:
"Idiomatic": "i"
"Literal": "l"
tooltips:
"Idiomatic": "The compound is used in its figurative or idiomatic sense"
"Literal": "The compound is used in its literal, compositional sense"
- annotation_type: radio
name: image1_rank
description: "Rank for Image 1 (1 = best represents the sense, 5 = worst)"
labels: ["1", "2", "3", "4", "5"]
- annotation_type: radio
name: image2_rank
description: "Rank for Image 2 (1 = best represents the sense, 5 = worst)"
labels: ["1", "2", "3", "4", "5"]
- annotation_type: radio
name: image3_rank
description: "Rank for Image 3 (1 = best represents the sense, 5 = worst)"
labels: ["1", "2", "3", "4", "5"]
- annotation_type: radio
name: image4_rank
description: "Rank for Image 4 (1 = best represents the sense, 5 = worst)"
labels: ["1", "2", "3", "4", "5"]
- annotation_type: radio
name: image5_rank
description: "Rank for Image 5 (1 = best represents the sense, 5 = worst)"
labels: ["1", "2", "3", "4", "5"]
annotation_instructions: |
You will be shown a context sentence containing a highlighted nominal
compound (for example, "night owl"), together with five candidate images.
The compound may be used in its idiomatic (figurative) or its literal sense.
Your task is to:
1. Read the sentence and decide in which sense the compound is used.
2. Look at each of the five candidate images.
3. Assign each image a distinct rank from 1 to 5, where 1 is the image that
best represents the sense in which the compound is used in this context
and 5 is the image that represents it least well.
html_layout: |
<div style="padding: 15px; max-width: 800px; margin: auto;">
<div style="background: #f0f9ff; border: 1px solid #bae6fd; border-radius: 8px; padding: 16px; margin-bottom: 16px;">
<strong style="color: #0369a1;">Nominal compound:</strong>
<span style="font-size: 16px; font-weight: 600;">{{compound}}</span>
<p style="font-size: 16px; line-height: 1.7; margin: 8px 0 0 0;">{{context}}</p>
</div>
<div style="background: #fefce8; border: 1px solid #fde68a; border-radius: 8px; padding: 16px; margin-bottom: 16px;">
<strong style="color: #a16207;">Candidate images:</strong>
<ol style="font-size: 15px; line-height: 1.6; margin: 8px 0 0 0;">
<li>{{image1}}</li>
<li>{{image2}}</li>
<li>{{image3}}</li>
<li>{{image4}}</li>
<li>{{image5}}</li>
</ol>
</div>
</div>
allow_all_users: true
instances_per_annotator: 50
annotation_per_instance: 2
allow_skip: true
skip_reason_required: false
Sample Datasample-data.json
[
{
"id": "admire_001",
"compound": "night owl",
"context": "My sister is a real night owl and rarely goes to bed before 3 a.m.",
"image1": "A person working at a laptop late at night in a dimly lit room with a wall clock showing 2 a.m.",
"image2": "A nocturnal owl perched on a bare branch under a full moon.",
"image3": "A child sleeping soundly in bed in the early evening.",
"image4": "A brightly lit open-plan office bustling during the daytime.",
"image5": "A steaming cup of coffee beside an alarm clock in the morning."
},
{
"id": "admire_002",
"compound": "couch potato",
"context": "Ever since he retired, my uncle has become a couch potato who watches TV all day.",
"image1": "A person sprawled on a sofa surrounded by snacks, staring at a television.",
"image2": "A single potato resting on a kitchen counter next to a cushion.",
"image3": "A group of friends hiking up a steep mountain trail.",
"image4": "A gardener harvesting potatoes from a muddy field.",
"image5": "An empty, tidy living room with a neatly arranged couch."
}
]
// ... and 8 more itemsGet This Design
Clone or download from the repository
Quick start:
git clone https://github.com/davidjurgens/potato-showcase.git cd potato-showcase/semeval/2025/task01-admire-multimodal-idiomaticity potato start config.yaml
Dataset & paper
Pickard et al., SemEval-2025 (ACL)
Citation (BibTeX)
@inproceedings{pickard-etal-2025-semeval,
title = "{S}em{E}val-2025 Task 1: {A}d{MIR}e - Advancing Multimodal Idiomaticity Representation",
author = "Pickard, Thomas and Villavicencio, Aline and Mi, Maggie and He, Wei and Phelps, Dylan and Idiart, Marco",
booktitle = "Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval-2025)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
pages = "2597--2609"
}Details
Annotation Types
Domain
Use Cases
Tags
Found an issue or want to improve this design?
Open an IssueRelated Designs
SemEval-2022 Task 5: Multimedia Misogyny (MAMI)
MAMI is the SemEval-2022 Task 5 benchmark for detecting misogynous memes from text and image content (Fersini et al.). This Potato config reproduces both subtasks: the binary misogyny label and the four-way misogyny type.
V-WSD - Visual Word Sense Disambiguation
Visual word sense disambiguation task where annotators select the most appropriate image for a target word given its textual context. Based on SemEval-2023 Task 1 (Raganato et al.).
AfriSenti - African Language Sentiment
Sentiment analysis for tweets in African languages, classifying text as positive, negative, or neutral. Covers 14 African languages including Amharic, Hausa, Igbo, Yoruba, and Swahili. Based on SemEval-2023 Task 12 (Muhammad et al.).