Multimodal-USElecDeb60To16: Audio-Enhanced Political Debates for Argument Mining

Multimodal USElecDeb60To16 provides audio features and synthetic speech for U.S. presidential debates from 1960 to 2016. The dataset was created to augment pre-trained language models for argumentation mining, as described in a 2023 EACL Findings paper. It offers a version with large audio files and a lighter version without them.

Use Cases

• [Argumentation mining] based on [transcribed political debate text]
• [Multimodal model training] based on [paired audio features and text]
• [Synthetic speech analysis] based on [included artificial voices]
• [Political discourse study] based on [debates spanning 1960-2016]

Strengths

• Multimodal design explicitly includes audio feature embeddings and artificial voices.
• Dataset is associated with peer-reviewed research published at EACL 2023.
• Offers a version choice to manage file size, indicating consideration for user needs.

Limitations

• Key metadata like row count, column names, and specific size are unavailable from the sources.
• The dataset's freshness is unclear as the last update date is not provided.
• Relies on inference for the 1960-2016 time range based on the dataset name; the exact coverage is not explicitly stated.

Provenance

Source

Created by Rafael Mestre, affiliated with the University of Southampton.

Collection Method

Likely compiled from recordings and transcripts of U.S. presidential debates, with extracted audio features and generated synthetic voices.

Time Range

1960-2016 (inferred from dataset name 'USElecDeb60To16')

Geography

United States (inferred from 'US' in dataset name and focus on presidential debates)