CODWOE - Comparing Dictionaries and Word Embeddings
Definition modeling for words in context, adapting SemEval-2022 Task 1 (CODWOE), which compared dictionary definitions with distributional word representations across two subtasks: definition modeling (embedding to definition) and reverse dictionary (definition to embedding). Based on Mickus et al.
Configuration Fileconfig.yaml
This Potato config reproduces the annotation task. Save it as config.yaml and run potato start config.yaml to try it.
# CODWOE - Comparing Dictionaries and Word Embeddings
# Based on Mickus et al., SemEval 2022
# Paper: https://aclanthology.org/2022.semeval-1.1/
# Dataset: https://codwoe.atilf.fr/
#
# CODWOE (SemEval-2022 Task 1) compared dictionaries and word embeddings
# through two automatic subtasks: definition modeling (generate a definition
# from a word embedding) and reverse dictionary (generate an embedding from a
# definition). Neither is directly human-annotatable, so this Potato design
# adapts the definition-modeling direction into a human writing task and
# grounds a companion judgment in the reverse-dictionary direction.
#
# This task asks annotators to (1) write a definition for a target word based
# on the context in which it appears (definition modeling), and (2) judge how
# uniquely that word could be recovered from a definition alone (reverse
# dictionary difficulty).
#
# Recoverability Labels:
# - Easily identifiable: The definition points clearly to this exact word
# - Somewhat ambiguous: The definition could fit a few related words
# - Highly ambiguous: The definition could describe many different words
annotation_task_name: "CODWOE - Comparing Dictionaries and Word Embeddings"
task_dir: "."
data_files:
- sample-data.json
item_properties:
id_key: "id"
text_key: "text"
output_annotation_dir: "annotation_output/"
output_annotation_format: "json"
port: 8000
server_name: localhost
annotation_schemes:
- annotation_type: text
name: definition
description: "Write a definition for the target word as used in this context"
- annotation_type: radio
name: recoverability
description: "If given only your definition, how uniquely could the target word be recovered?"
labels:
- "Easily identifiable"
- "Somewhat ambiguous"
- "Highly ambiguous"
keyboard_shortcuts:
"Easily identifiable": "1"
"Somewhat ambiguous": "2"
"Highly ambiguous": "3"
tooltips:
"Easily identifiable": "The definition points clearly to this exact target word"
"Somewhat ambiguous": "The definition could plausibly fit a few closely related words"
"Highly ambiguous": "The definition could describe many different words"
annotation_instructions: |
This task adapts SemEval-2022 Task 1 (CODWOE), which paired definition
modeling with a reverse-dictionary subtask. You will see a sentence with a
target word and the language of the text.
1. Read the sentence and identify the target word.
2. Write a clear, concise definition for the target word as it is used in this
context (definition modeling).
3. Judge how uniquely the target word could be recovered from your definition
alone (reverse-dictionary difficulty).
html_layout: |
<div style="padding: 15px; max-width: 800px; margin: auto;">
<div style="display: flex; gap: 12px; margin-bottom: 12px;">
<div style="background: #ecfdf5; border: 1px solid #a7f3d0; border-radius: 8px; padding: 12px; flex: 1;">
<strong style="color: #065f46;">Language:</strong>
<span style="font-size: 15px; margin-left: 8px;">{{language}}</span>
</div>
<div style="background: #fef3c7; border: 1px solid #fde68a; border-radius: 8px; padding: 12px; flex: 1;">
<strong style="color: #92400e;">Target Word:</strong>
<span style="font-size: 15px; font-weight: bold; margin-left: 8px;">{{target_word}}</span>
</div>
</div>
<div style="background: #f0f9ff; border: 1px solid #bae6fd; border-radius: 8px; padding: 16px; margin-bottom: 16px;">
<strong style="color: #0369a1;">Context:</strong>
<p style="font-size: 16px; line-height: 1.7; margin: 8px 0 0 0;">{{text}}</p>
</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": "codwoe_001",
"text": "The architect presented her vision for a sustainable bridge that would connect the two neighborhoods across the river.",
"target_word": "bridge",
"language": "English"
},
{
"id": "codwoe_002",
"text": "After years of conflict, the treaty served as a bridge between the two nations, fostering mutual understanding and cooperation.",
"target_word": "bridge",
"language": "English"
}
]
// ... 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/2022/task01-codwoe potato start config.yaml
Dataset & paper
Mickus et al., SemEval 2022
Citation (BibTeX)
@inproceedings{mickus-etal-2022-semeval,
title = "{S}emeval-2022 Task 1: {CODWOE} {--} Comparing Dictionaries and Word Embeddings",
author = "Mickus, Timothee and Van Deemter, Kees and Constant, Mathieu and Paperno, Denis",
booktitle = "Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022)",
month = jul,
year = "2022",
address = "Seattle, United States",
publisher = "Association for Computational Linguistics",
pages = "1--14",
doi = "10.18653/v1/2022.semeval-1.1",
url = "https://aclanthology.org/2022.semeval-1.1"
}Details
Annotation Types
Domain
Use Cases
Tags
Found an issue or want to improve this design?
Open an IssueRelated Designs
Hypernym Discovery
Discovery of hypernyms (broader terms) for a given target term in context, with classification of the hypernym relationship type. Based on SemEval-2018 Task 9.
BRAINTEASER - Commonsense-Defying QA
Lateral thinking and commonsense-defying question answering task requiring annotators to select answers to brain teasers that defy default commonsense assumptions and provide explanations. Based on SemEval-2024 Task 9 (BRAINTEASER).
CLARITY: Unmasking Political Question Evasions
SemEval-2026 Task 6 (CLARITY): detecting how directly a politician answers a question. Given a question-answer pair from a U.S. presidential interview, annotators classify the reply into one of three clarity levels (Clear Reply, Ambiguous, Clear Non-Reply) and, when the answer is evasive, identify the evasion technique used. Multi-part questions are first decomposed into single sub-questions so each judgment captures how well one specific inquiry is addressed.