Challenge: supervised learning is a key component of offensive language detection, but there is little attention given to the quality of annotated data.
Approach: They propose to examine the level of agreement among annotators while selecting data to create offensive language datasets, a task involving a high level of subjectivity.
Outcome: The proposed datasets show that annotators' agreement has a strong effect on classifiers performance and robustness.

Similar Papers

Why Don’t You Do It Right? Analysing Annotators’ Disagreement in Subjective Tasks (2023.eacl-main)

Copied to clipboard

Challenge: Disagreement can reflect different aspects of linguistic annotation, from annotators’ subjectivity to sloppiness or lack of context to interpret a text.
Approach: They propose a taxonomy of possible reasons leading to annotators' disagreement in subjective tasks and manually label part of a Twitter dataset for offensive language detection in english following this taxonomies.
Outcome: The proposed taxonomy of disagreements in linguistic datasets can be used to assess how accurate tweets belonging to different disagreement categories can be classified as offensive or not.
D3CODE: Disentangling Disagreements in Data across Cultures on Offensiveness Detection and Evaluation (2024.emnlp-main)

Copied to clipboard

Challenge: Recent studies on annotator subjectivity focus on Western contexts and only document differences across age, gender, or racial groups.
Approach: They propose a large-scale cross-cultural dataset of parallel annotations for offensive language in over 4.5K English sentences annotated by a pool of more than 4k annotators from 21 countries.
Outcome: The proposed dataset captures annotators’ moral values along six moral foundations: care, equality, proportionality, authority, loyalty, and purity.
Subjective Crowd Disagreements for Subjective Data: Uncovering Meaningful CrowdOpinion with Population-level Learning (2023.acl-long)

Copied to clipboard

Challenge: Annotator disagreements are resolved before learning takes place, but researchers question the performance of a system when annotators disagree.
Approach: They propose a method that uses language features and label distributions to pool similar items into larger labels.
Outcome: The proposed method is based on five publicly available datasets with varying levels of disagreements on social media and in the wild using a dataset from Facebook.
Is LLM an Overconfident Judge? Unveiling the Capabilities of LLMs in Detecting Offensive Language with Annotation Disagreement (2025.findings-acl)

Copied to clipboard

Challenge: Existing studies focus on evaluating large language models' ability to handle disagreement cases.
Approach: They evaluate the performance of large language models in detecting offensive language at varying levels of agreement.
Outcome: The proposed model improves detection accuracy and model alignment with human judgment by using disagreement samples in training.
You Are What You Annotate: Towards Better Models through Annotator Representations (2023.findings-emnlp)

Copied to clipboard

Challenge: Annotator disagreement is ubiquitous in natural language processing tasks.
Approach: They propose to model annotators' idiosyncrasies and account for their idioms by creating representations for each annotator and their annotations.
Outcome: The proposed model improves on an existing dataset with eight annotators with inherent disagreements while increasing model size by 1%.
Would you describe a leopard as yellow? Evaluating crowd-annotations with justified and informative disagreement (2020.coling-main)

Copied to clipboard

Challenge: Existing evaluation methods rely on agreement between annotators, which implies a single correct interpretation.
Approach: They propose an agreement-independent quality metric based on answer-coherence to evaluate on expected disagreement.
Outcome: The proposed model shows that agreement is the most important indicator of quality in semantic annotation tasks.
When the Majority is Wrong: Modeling Annotator Disagreement for Subjective Tasks (2023.emnlp-main)

Copied to clipboard

Challenge: a number of studies have questioned assumptions of majority vote aggregated labels.
Approach: They construct a model that predicts individual annotator ratings on potentially offensive text and combines this information with the predicted target group of the text to predict the ratings of target group members.
Outcome: The proposed model raises performance over baseline by 22% and 33% at predicting variance among annotators.
Which Examples Should be Multiply Annotated? Active Learning When Annotators May Disagree (2023.findings-acl)

Copied to clipboard

Challenge: Disagreement in annotations is natural for humans, depending on background, identity, positionality . many active learning approaches focus on examples where model entropy and annotator entropicy are the most different.
Approach: They propose an active learning approach that focuses annotations on examples where model entropy and annotator entropic are the most different.
Outcome: The proposed approach reduces the number of annotations required by 24% on average across datasets.
The Promises and Pitfalls of LLM Annotations in Dataset Labeling: a Case Study on Media Bias Detection (2025.findings-naacl)

Copied to clipboard

Challenge: Recent research suggests using Large Language Models (LLMs) to automate the annotation process, reducing these costs while maintaining data quality.
Approach: They propose to use Large Language Models to automate annotation process and train classifiers on large datasets.
Outcome: The proposed model outperforms all of the annotator LLMs on two media bias benchmark datasets (BABE and BASIL) while maintaining data quality.
Improving Adversarial Data Collection by Supporting Annotators: Lessons from GAHD, a German Hate Speech Dataset (2024.naacl-long)

Copied to clipboard

Challenge: Hate speech detection models are only as good as the data they are trained on, but adversarial datasets are slow and costly . data sourced from social media suffer from systematic gaps and biases, leading to unreliable models with simplistic decision boundaries.
Approach: They propose a German Adversarial Hate speech Dataset comprising 11k examples . they explore new strategies for supporting annotators and provide manual analysis of disagreements for each strategy .
Outcome: The proposed dataset is challenging even for state-of-the-art hate speech detection models and it significantly improves model robustness.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations