Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Paying Attention to Deflections: Mining Pragmatic Nuances for Whataboutism Detection in Online Discourse

Khiem Phi, Noushin Salek Faramarzi, Chenlu Wang, Ritwik Banerjee

TL;DR

This work introduces new datasets from Twitter and YouTube, revealing overlaps as well as distinctions between whataboutism, propaganda, and the tu quoque fallacy, and distinguishes the `what about' lexical construct from whataboutism.

Abstract

Whataboutism, a potent tool for disrupting narratives and sowing distrust, remains under-explored in quantitative NLP research. Moreover, past work has not distinguished its use as a strategy for misinformation and propaganda from its use as a tool for pragmatic and semantic framing. We introduce new datasets from Twitter and YouTube, revealing overlaps as well as distinctions between whataboutism, propaganda, and the tu quoque fallacy. Furthermore, drawing on recent work in linguistic semantics, we differentiate the `what about' lexical construct from whataboutism. Our experiments bring to light unique challenges in its accurate detection, prompting the introduction of a novel method using attention weights for negative sample mining. We report significant improvements of 4% and 10% over previous state-of-the-art methods in our Twitter and YouTube collections, respectively.

Paying Attention to Deflections: Mining Pragmatic Nuances for Whataboutism Detection in Online Discourse

TL;DR

This work introduces new datasets from Twitter and YouTube, revealing overlaps as well as distinctions between whataboutism, propaganda, and the tu quoque fallacy, and distinguishes the `what about' lexical construct from whataboutism.

Abstract

Whataboutism, a potent tool for disrupting narratives and sowing distrust, remains under-explored in quantitative NLP research. Moreover, past work has not distinguished its use as a strategy for misinformation and propaganda from its use as a tool for pragmatic and semantic framing. We introduce new datasets from Twitter and YouTube, revealing overlaps as well as distinctions between whataboutism, propaganda, and the tu quoque fallacy. Furthermore, drawing on recent work in linguistic semantics, we differentiate the `what about' lexical construct from whataboutism. Our experiments bring to light unique challenges in its accurate detection, prompting the introduction of a novel method using attention weights for negative sample mining. We report significant improvements of 4% and 10% over previous state-of-the-art methods in our Twitter and YouTube collections, respectively.
Paper Structure (45 sections, 2 theorems, 5 equations, 6 figures, 4 tables)

This paper contains 45 sections, 2 theorems, 5 equations, 6 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Pragmatic flexibility
  3. "What about" $\centernot\Leftrightarrow$ whataboutism:
  4. The Datasets
  5. A thematic scrutiny
  6. Discourse coherence:
  7. Perceptions, framing, and Tu Quoque:
  8. Sarcasm and irony:
  9. Experiments
  10. Baselines: (Habitual) Transfer learning
  11. (Conventional) Semantic similarity
  12. Mining negatives with attention
  13. Results and analysis
  14. Baseline models:
  15. Semantic similarity models:
  16. ...and 30 more sections

Key Result

Theorem 1

Let $X_1, \ldots, X_n$ be independent random variables with $\mathbb{E}\left[X_i\right] = 0$ and $\vert X_i \vert \leq 1$ for all $i$. Let $X = \sum_{i \in [n]} X_i$, and let $\sigma^2$ denote the variance of $X_i$. Then,

Figures (6)

  • Figure 1: What about in YouTube comments: implicit use (top) to discredit the source and redirect the topic, and explicit use (bottom) as an attempt toward reasonable argumentation instead of propaganda.
  • Figure 2: Syntactically and semantically similar (or even identical) responses may exhibit extreme pragmatic flexibility: being instances of whataboutism [w] or not [nw] depending on the discursive context. Furthermore, the latter category may include valid argumentation tools such as suggestion or challenge, instead of propaganda.
  • Figure 3: MINA (Mining Negatives with Attention) employs attention weights in the final layer of the Transformer encoder as a measure of pragmatic contrast. The complete architecture is shown here in situ.
  • Figure 4: t-SNE visualization of the target class (w; blue), amid everything else (nw; red) with (a) transfer-learning with pretrained language models, and (b) mining negative samples using mina. The latter demonstrates better separability of the target class.
  • Figure 5: Labeled data distribution, spanning six divisive sociopolitical topics for the collection of YouTube comments, and eight such topics for the collection of Twitter responses. Each topic is partitioned into three sets (training, validation, and test), each with comments labeled as whataboutism (w) or not (nw).
  • ...and 1 more figures

Theorems & Definitions (5)

  • Theorem 1: Bernstein's Inequality
  • Definition 1
  • Definition 2
  • Theorem 2
  • proof