"With 1 follower I must be AWESOME : P". Exploring the role of irony markers in irony recognition

Debanjan Ghosh, Smaranda Muresan

Abstract

Conversations in social media often contain the use of irony or sarcasm, where users say the opposite of what they truly mean. Irony markers are meta-communicative clues that inform the reader that an utterance is ironic. We propose a thorough analysis of theoretically grounded irony markers across two social media platforms: Twitter and Reddit. Classification and frequency analysis show that for Twitter, typographic markers such as emoticons and emojis are the most discriminative markers to recognize ironic utterances, while for Reddit, morphological markers (e.g., interjections, tag questions) are the most discriminative.

Introduction: The Challenge of Irony Detection

Detecting irony and sarcasm is a notorious "white whale" in Natural Language Processing (NLP). While humans often use context and tone to spot a sarcastic remark, machines struggle because the literal meaning of an utterance is often the opposite of the intended meaning.

Most previous research has focused on building high-performing models without a strong theoretical foundation. This paper bridges the gap between linguistics and NLP by focusing on irony markers. According to linguistic theory (Attardo, 2000), these are optional clues that "alert the reader to the fact that a sentence is ironical." Unlike "irony factors," which are essential to the irony itself, markers can be removed without destroying the irony, but they make it much easier to recognize. The authors seek to understand which markers are most effective and how their usage varies across different social media environments.

Methodology

The authors framed the problem as a binary classification task (ironic vs. non-ironic) using a Support Vector Machine (SVM) with a linear kernel. They engineered features based on three categories of irony markers:

  1. Tropes: Figurative language including metaphors, hyperboles (intensifiers), and rhetorical questions.
  2. Morpho-syntactic (MS) markers: Structural clues such as exclamation marks, tag questions (e.g., "isn't it?"), and interjections (e.g., "wow," "yay").
  3. Typographic markers: Visual cues including capitalization, quotation marks, hashtags, emoticons, and emojis.

To evaluate the importance of these features, they performed ablation tests, systematically removing one category of markers at a time to observe the impact on Precision (P), Recall (R), and F1 scores.

Datasets

The study utilized two large-scale, self-labeled datasets to capture platform-specific nuances:

  • Twitter: 350,000 tweets. Ironic tweets were identified by hashtags like #irony or #sarcasm. Non-ironic tweets were collected using sentiment hashtags like #happy or #sad.
  • Reddit: 50,000 instances where users explicitly mark sarcasm with the /s tag. The authors selected replies at least two sentences long to ensure sufficient context.
Examples of ironic utterances and their platform-specific markers (#ironic on Twitter and /s on Reddit).

Key Insights: A Stylistic Divide

The most striking finding is the "stylistic divide" between platforms. Twitter is a visual medium where emojis and emoticons are primary discriminators. In contrast, Reddit users—who typically write longer posts—rely on linguistic complexity, such as metaphors and tag questions.

The "Sentiment Flip" in Emojis

On Twitter, ironic tweets often feature emojis with negative sentiments (angry faces, "unamused" faces) to signal the irony of a seemingly positive statement. Conversely, non-ironic tweets are more likely to use positive emojis (hearts, wedding symbols).

Top features for ironic (I) and non-ironic (NI) tweets on Twitter, highlighting the use of negative-sentiment emojis to signal irony.

Hyperbole and Contextual Intensity

In Twitter, removing tropes (specifically hyperboles) caused a massive 15% drop in Recall. This suggests that while hyperbole doesn't occur in every ironic tweet, it is a very strong signal when present. Furthermore, irony markers are significantly more frequent in controversial subreddits like Politics and Religion compared to Technology or Sports. Users in heated debates seem to "over-mark" their sarcasm to ensure their point is not misinterpreted.

Key Results

  • Twitter Performance: Typographic markers were the most discriminative. Removing them significantly lowered the Precision for the irony category.
  • Reddit Performance: Morpho-syntactic markers (like tag questions and interjections) were the most impactful. Removing them had a much larger negative impact on the F1 score than removing typographic markers.
  • Frequency Analysis: Emojis and emoticons are nearly non-existent in the Reddit irony dataset compared to Twitter. Meanwhile, Rhetorical Questions (RQs) and exclamations are far more common on Reddit.

Conclusion and Future Work

The authors acknowledge that their feature set is not exhaustive. Future work should explore more complex markers such as ironic echo (mocking repetition), understatements, and repetition. Additionally, while self-labeling provides a large dataset, it may introduce a "reporting bias," capturing only irony that the user intended to be obvious. Subtle, unmarked irony remains a significant challenge for future NLP research.

Summary updated

Abstract

Conversations in social media often contain the use of irony or sarcasm, when the users say the opposite of what they really mean. Irony markers are the meta-communicative clues that inform the reader that an utterance is ironic. We propose a thorough analysis of theoretically grounded irony markers in two social media platforms: $Twitter$ and $Reddit$. Classification and frequency analysis show that for $Twitter$, typographic markers such as emoticons and emojis are the most discriminative markers to recognize ironic utterances, while for $Reddit$ the morphological markers (e.g., interjections, tag questions) are the most discriminative.
BibTeX
@misc{ghosh2018with,
  author = {Debanjan Ghosh and Smaranda Muresan},
  title = {{"With 1 follower I must be AWESOME : P". Exploring the role of irony markers in irony recognition}},
  year = {2018},
  eprint = {1804.05253},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1804.05253},
}
Figure