Overview of Memotion 3: Sentiment & Emotion Analysis of Codemixed Hinglish - Abstract & Introduction

21 Feb 2024

This paper is available on arxiv under CC 4.0 license.


(1) Shreyash Mishra has an equal contribution from IIIT Sri City, India;

(2) S Suryavardan has an equal contribution from IIIT Sri City, India;

(3) Megha Chakraborty, University of South Carolina, USA;

(4) Parth Patwa, UCLA, USA;

(5) Anku Rani, University of South Carolina, USA;

(6) Aman Chadha, work does not relate to a position at Amazon from Stanford University, USA, or Amazon AI, USA;

(7) Aishwarya Reganti, CMU, USA;

(8) Amitava Das, University of South Carolina, USA;

(9) Amit Sheth, University of South Carolina, USA;

(10) Manoj Chinnakotla, Microsoft, USA;

(11) Asif Ekbal, IIT Patna, India;

(12) Srijan Kumar, Georgia Tech, USA.

Abstract & Introduction

Related Work

Task Details

Participating systems


Conclusion and Future Work and References


Analyzing memes on the internet has emerged as a crucial endeavor due to the impact this multi modal form of content wields in shaping online discourse. Memes have become a powerful tool for expressing emotions and sentiments, possibly even spreading hate and misinformation, through humor and sarcasm. In this paper, we present the overview of the Memotion 3 shared task, as part of the DeFactify 2 workshop at AAAI-23. The task released an annotated dataset of Hindi-English code-mixed memes based on their Sentiment (Task A), Emotion (Task B), and Emotion intensity (Task C). Each of these is defined as an individual task and the participants are ranked separately for each task. Over 50 teams registered for the shared task and 5 made final submissions to the test set of the Memotion 3 dataset. CLIP, BERT modifications, ViT etc. were the most popular models among the participants along with approaches such as Student-Teacher model, Fusion, and Ensembling. The best final F1 score for Task A is 34.41, Task B is 79.77 and Task C is 59.82.

Keywords Memes, codemixed, multimodal, Hindi-English

1. Introduction

The term meme is derived from the Ancient Greek word mimema, meaning imitated thing, which comes from the verb mimeisthai, meaning to mimic. The term was coined by Richard Dawkins, a British evolutionary biologist, in his book The Selfish Gene [1]. Dawkins used the concept of a meme to explain the spread of cultural phenomena and ideas, drawing parallels between memes and genes. Dawkins provided examples of memes such as melodies, catchphrases, fashion, and arch-building technology in his book. Interestingly, the word meme is a self-describing term, also known as autological, meaning that it is a meme itself.

The reach of online content is vast and has the capacity to reach a massive viewership. Memes propagate through multiple mediums including social media, messaging, apps, and emails. An article on memes quotes "When you plant a fertile meme in my mind, you literally, parasitize my brain" [2]. This quote indicates the impact of memes on a viewer’s mind. It spreads to others who find it motivating, offensive, amusing, or relatable in some ways [3]. Memes are used to convey an opinion. When a person comes across a meme that they find relatable, they often share it with others who they think would find relatable. Different people interpret memes differently which means what can be offensive for one person might not be for other. This difference in interpretation is a result of cultural, gender, and demographic differences.

The memes can also be used to spread offensive and hateful content [4]. Identifying and halting the dissemination of hateful memes is an arduous undertaking for both human beings and AI models, as it requires comprehending the intricate subtleties and socio-political contexts that form their interpretation. The deficiencies of current hate speech moderation techniques highlight the pressing need to enhance the effectiveness of automated hate speech detection. Given their subtlety and multi-modal nature, memes pose a more challenging issue to address compared to only text.

In this paper, we present the findings of the Memotion 3 shared task, where participants were provided with an annotated dataset of 10k memes [5], and were tasked with detecting the sentiment, emotion and emotion intensity of the meme. Unlike the previous iteration of memotion [6, 7] which provided English memes, current iteration studies codemixed HindiEnglish (Hinglish) memes.

The paper is organized as follows: we describe the related work and the task in section 2 and 3 respectively; Section 4 contains the details of the dataset we collected for Memotion analysis: Memotion 3.0; followed by a brief description of baseline models and their results in 5. We conclude and mention some possible future works in section 6.