metadata
dataset_info:
features:
- name: user
dtype: string
- name: label
dtype: string
- name: timestamp
dtype: string
- name: source
dtype: string
- name: subreddit
dtype: string
- name: id_original
dtype: string
- name: text
dtype: string
- name: parent_id_original
dtype: string
- name: parent_text
dtype: string
- name: Language_instance
dtype: string
- name: Language_variety
dtype: string
- name: Age
dtype: string
- name: Sex
dtype: string
- name: Ethnicity simplified
dtype: string
- name: Country of birth
dtype: string
- name: Country of residence
dtype: string
- name: Nationality
dtype: string
- name: Language_annotator
dtype: string
- name: Student status
dtype: string
- name: Employment status
dtype: string
splits:
- name: train
num_bytes: 7299373
num_examples: 14172
download_size: 1038853
dataset_size: 7299373
EPIC_Irony
- paper: EPIC: Multi-Perspective Annotation of a Corpus of Irony at ACL 2023
Key features:
- EPIC (English Perspectivist Irony Corpus) is an annotated corpus for irony analysis based on data perspectivism principles.
- The corpus contains social media conversations in five regional varieties of English, annotated by contributors from corresponding countries.
- The dataset explores the perspectives of annotators, taking into account their origin, age, and gender.
- Perspective-aware models were created to validate EPIC, and these proved more effective and confident in identifying irony than non-perspectivist models.
- The models showcase variation in irony perception across different demographic groups.
- EPIC serves as a valuable resource for training perspective-aware models for irony detection.
Metadata in Creative Language Toolkit (CLTK)
- CL Type: Irony
- Task Type: detection
- Size: 14k
- Created time: 2023