ElecDeb60to20 Dataset

In this repository is available the dataset named ElecDeb60to20 that is split in different annotation types:

• Fallacy Annotation
• Argument Component Annotations
• Argument Relationship Annotations

In addition, it is possible to explore the debates from the "speech turn" point of view in the folder data/debates.

This dataset is a collection of televised debates of the US presidential campaign debates from 1960 to 2020 [1], [2], [3].

The guidelines used to annotate the fallacy are provided here, whereas for the argumentative component and relationship are here.

Fallacy

In the pragma-dialectical theory of argumentation [Eemeren and Grootendorst, 1992; Eemeren, 2010], fallacies are defined as “derailments of strategic manoeuvring”, meaning speech acts that violate the rules of a rational argumentative discussion for assumed persuasive gains.

Focus on six category of main fallacies and three of them are further divided into sub-categories:

1. Ad Hominem
   1. General
   2. Bias ad Hominem
   3. Tu quoque
   4. Name-calling, Labeling
2. Appeal to Emotion
   1. Appeal to Fear
   2. Appeal to Pity
   3. Loaded Language
   4. Flag Waving
3. Appeal to Authority
   1. Without evidence
   2. False authority
   3. Popular Opinion
4. False Cause
5. Slippery Slope
6. Slogans

Argumentative Component

The components are split into:

• Claims
• Premises

Claims

Being them the ultimate goal of an argument, in the context of political debates, claims can be a policy advocated by a party or a candidate to be undertaken which needs to be justified in order to be accepted by the audience.

Premises

Premises are assertions made by the debaters for supporting their claims (i.e., reasons or justifications).

Argumentative Relationship

The relationships are split into:

• Support
• Attack

Support

Links two components from a supporting argument component to a supported argument component.

Attack

Holds when one argument component is in contradiction with another argument component