Prior works formulate the extraction of event-specific arguments as a span extraction problem, where event arguments are explicit — i.e. assumed to be contiguous spans of text in a document. In this study, we revisit this definition of Event Extraction (EE) by introducing two key argument types that cannot be modeled by existing EE frameworks. First, implicit arguments are event arguments which are not explicitly mentioned in the text, but can be inferred through context. Second, scattered arguments are event arguments that are composed of information scattered throughout the text. These two argument types are crucial to elicit the full breadth of information required for proper event modeling.To support the extraction of explicit, implicit, and scattered arguments, we develop a novel dataset, DiscourseEE, which includes 7,464 argument annotations from online health discourse. Notably, 51.2% of the arguments are implicit, and 17.4% are scattered, making DiscourseEE a unique corpus for complex event extraction. Additionally, we formulate argument extraction as a text generation problem to facilitate the extraction of complex argument types. We provide a comprehensive evaluation of state-of-the-art models and highlight critical open challenges in generative event extraction. Our data and codebase are available at https://omar-sharif03.github.io/DiscourseEE.

Internet memes have gained immense traction as a medium for individuals to convey emotions, thoughts, and perspectives on social media. While memes often serve as sources of humor and entertainment, they can also propagate offensive, incendiary, or harmful content, deliberately targeting specific individuals or communities. Identifying such memes is challenging because of their satirical and cryptic characteristics. Most contemporary research on memes’ detrimental facets is skewed towards high-resource languages, often sidelining the unique challenges tied to low-resource languages, such as Bengali. To facilitate this research in low-resource languages, this paper presents a novel dataset MIMOSA (MultIMOdal aggreSsion dAtaset) in Bengali. MIMOSA encompasses 4,848 annotated memes across five aggression target categories: Political, Gender, Religious, Others, and non-aggressive. We also propose MAF (Multimodal Attentive Fusion), a simple yet effective approach that uses multimodal context to detect the aggression targets. MAF captures the selective modality-specific features of the input meme and jointly evaluates them with individual modality features. Experiments on MIMOSA exhibit that the proposed method outperforms several state-of-the-art rivaling approaches. Our code and data are available at https://github.com/shawlyahsan/Bengali-Aggression-Memes.

Multimodal hateful content detection is a challenging task that requires complex reasoning across visual and textual modalities. Therefore, creating a meaningful multimodal representation that effectively captures the interplay between visual and textual features through intermediate fusion is critical. Conventional fusion techniques are unable to attend to the modality-specific features effectively. Moreover, most studies exclusively concentrated on English and overlooked other low-resource languages. This paper proposes a context-aware attention framework for multimodal hateful content detection and assesses it for both English and non-English languages. The proposed approach incorporates an attention layer to meaningfully align the visual and textual features. This alignment enables selective focus on modality-specific features before fusing them. We evaluate the proposed approach on two benchmark hateful meme datasets, viz. MUTE (Bengali code-mixed) and MultiOFF (English). Evaluation results demonstrate our proposed approach’s effectiveness with F1-scores of 69.7% and 70.3% for the MUTE and MultiOFF datasets. The scores show approximately 2.5% and 3.2% performance improvement over the state-of-the-art systems on these datasets. Our implementation is available at https://github.com/eftekhar-hossain/Bengali-Hateful-Memes.

Internet memes have become a powerful means for individuals to express emotions, thoughts, and perspectives on social media. While often considered as a source of humor and entertainment, memes can also disseminate hateful content targeting individuals or communities. Most existing research focuses on the negative aspects of memes in high-resource languages, overlooking the distinctive challenges associated with low-resource languages like Bengali (also known as Bangla). Furthermore, while previous work on Bengali memes has focused on detecting hateful memes, there has been no work on detecting their targeted entities. To bridge this gap and facilitate research in this arena, we introduce a novel multimodal dataset for Bengali, BHM (Bengali Hateful Memes). The dataset consists of 7,148 memes with Bengali as well as code-mixed captions, tailored for two tasks: (i) detecting hateful memes, and (ii) detecting the social entities they target (i.e., Individual, Organization, Community, and Society). To solve these tasks, we propose DORA (Dual cO-attention fRAmework), a multimodal deep neural network that systematically extracts the significant modality features from the memes and jointly evaluates them with the modality-specific features to understand the context better. Our experiments show that DORA is generalizable on other low-resource hateful meme datasets and outperforms several state-of-the-art rivaling baselines.

Stance detection on social media is challenging for Large Language Models (LLMs), as emerging slang and colloquial language in online conversations often contain deeply implicit stance labels. Chain-of-Thought (COT) prompting has recently been shown to improve performance on stance detection tasks — alleviating some of these issues. However, COT prompting still struggles with implicit stance identification. This challenge arises because many samples are initially challenging to comprehend before a model becomes familiar with the slang and evolving knowledge related to different topics, all of which need to be acquired through the training data. In this study, we address this problem by introducing COT Embeddings which improve COT performance on stance detection tasks by embedding COT reasonings and integrating them into a traditional RoBERTa-based stance detection pipeline. Our analysis demonstrates that 1) text encoders can leverage COT reasonings with minor errors or hallucinations that would otherwise distort the COT output label. 2) Text encoders can overlook misleading COT reasoning when a sample’s prediction heavily depends on domain-specific patterns. Our model achieves SOTA performance on multiple stance detection datasets collected from social media.

Amidst the sharp rise in the evaluation of large language models (LLMs) on various tasks, we find that semantic textual similarity (STS) has been under-explored. In this study, we show that STS can be cast as a text generation problem while maintaining strong performance on multiple STS benchmarks. Additionally, we show generative LLMs significantly outperform existing encoder-based STS models when characterizing the semantic similarity between two texts with complex semantic relationships dependent on world knowledge. We validate this claim by evaluating both generative LLMs and existing encoder-based STS models on three newly-collected STS challenge sets which require world knowledge in the domains of Health, Politics, and Sports. All newly-collected data is sourced from social media content posted after May 2023 to ensure the performance of closed-source models like ChatGPT cannot be credited to memorization. Our results show that, on average, generative LLMs outperform the best encoder-only baselines by an average of 22.3% on STS tasks requiring world knowledge. Our results suggest generative language models with STS-specific prompting strategies achieve state-of-the-art performance in complex, domain-specific STS tasks.

Recently, detection and categorization of undesired (e. g., aggressive, abusive, offensive, hate) content from online platforms has grabbed the attention of researchers because of its detrimental impact on society. Several attempts have been made to mitigate the usage and propagation of such content. However, most past studies were conducted primarily for English, where low-resource languages like Bengali remained out of the focus. Therefore, to facilitate research in this arena, this paper introduces a novel multilabel Bengali dataset (named M-BAD) containing 15650 texts to detect aggressive texts and their targets. Each text of M-BAD went through rigorous two-level annotations. At the primary level, each text is labelled as either aggressive or non-aggressive. In the secondary level, the aggressive texts have been further annotated into five fine-grained target classes: religion, politics, verbal, gender and race. Baseline experiments are carried out with different machine learning (ML), deep learning (DL) and transformer models, where Bangla-BERT acquired the highest weighted f₁-score in both detection (0.92) and target identification (0.83) tasks. Error analysis of the models exhibits the difficulty to identify context-dependent aggression, and this work argues that further research is required to address these issues.

Posting and sharing memes have become a powerful expedient of expressing opinions on social media in recent days. Analysis of sentiment from memes has gained much attention to researchers due to its substantial implications in various domains like finance and politics. Past studies on sentiment analysis of memes have primarily been conducted in English, where low-resource languages gain little or no attention. However, due to the proliferation of social media usage in recent years, sentiment analysis of memes is also a crucial research issue in low resource languages. The scarcity of benchmark datasets is a significant barrier to performing multimodal sentiment analysis research in resource-constrained languages like Bengali. This paper presents a novel multimodal dataset (named MemoSen) for Bengali containing 4417 memes with three annotated labels positive, negative, and neutral. A detailed annotation guideline is provided to facilitate further resource development in this domain. Additionally, a set of experiments are carried out on MemoSen by constructing twelve unimodal (i.e., visual, textual) and ten multimodal (image+text) models. The evaluation exhibits that the integration of multimodal information significantly improves (about 1.2%) the meme sentiment classification compared to the unimodal counterparts and thus elucidate the novel aspects of multimodality.

The exponential surge of social media has enabled information propagation at an unprecedented rate. However, it also led to the generation of a vast amount of malign content, such as hateful memes. To eradicate the detrimental impact of this content, over the last few years hateful memes detection problem has grabbed the attention of researchers. However, most past studies were conducted primarily for English memes, while memes on resource constraint languages (i.e., Bengali) are under-studied. Moreover, current research considers memes with a caption written in monolingual (either English or Bengali) form. However, memes might have code-mixed captions (English+Bangla), and the existing models can not provide accurate inference in such cases. Therefore, to facilitate research in this arena, this paper introduces a multimodal hate speech dataset (named MUTE) consisting of 4158 memes having Bengali and code-mixed captions. A detailed annotation guideline is provided to aid the dataset creation in other resource constraint languages. Additionally, extensive experiments have been carried out on MUTE, considering the only visual, only textual, and both modalities. The result demonstrates that joint evaluation of visual and textual features significantly improves (≈ 3%) the hateful memes classification compared to the unimodal evaluation.

With the substantial rise of internet usage, social media has become a powerful communication medium to convey information, opinions, and feelings on various issues. Recently, memes have become a popular way of sharing information on social media. Usually, memes are visuals with text incorporated into them and quickly disseminate hatred and offensive content. Detecting or classifying memes is challenging due to their region-specific interpretation and multimodal nature. This work presents a meme classification technique in Tamil developed by the CUET NLP team under the shared task (DravidianLangTech-ACL2022). Several computational models have been investigated to perform the classification task. This work also explored visual and textual features using VGG16, ResNet50, VGG19, CNN and CNN+LSTM models. Multimodal features are extracted by combining image (VGG16) and text (CNN, LSTM+CNN) characteristics. Results demonstrate that the textual strategy with CNN+LSTM achieved the highest weighted f₁-score (0.52) and recall (0.57). Moreover, the CNN-Text+VGG16 outperformed the other models concerning the multimodal memes detection by achieving the highest f₁-score of 0.49, but the LSTM+CNN model allowed the team to achieve 4^th place in the shared task.

With the proliferation of internet usage, a massive growth of consumer-generated content on social media has been witnessed in recent years that provide people’s opinions on diverse issues. Through social media, users can convey their emotions and thoughts in distinctive forms such as text, image, audio, video, and emoji, which leads to the advancement of the multimodality of the content users on social networking sites. This paper presents a technique for classifying multimodal sentiment using the text modality into five categories: highly positive, positive, neutral, negative, and highly negative categories. A shared task was organized to develop models that can identify the sentiments expressed by the videos of movie reviewers in both Malayalam and Tamil languages. This work applied several machine learning techniques (LR, DT, MNB, SVM) and deep learning (BiLSTM, CNN+BiLSTM) to accomplish the task. Results demonstrate that the proposed model with the decision tree (DT) outperformed the other methods and won the competition by acquiring the highest macro f₁-score of 0.24.

Recently, emotion analysis has gained increased attention by NLP researchers due to its various applications in opinion mining, e-commerce, comprehensive search, healthcare, personalized recommendations and online education. Developing an intelligent emotion analysis model is challenging in resource-constrained languages like Tamil. Therefore a shared task is organized to identify the underlying emotion of a given comment expressed in the Tamil language. The paper presents our approach to classifying the textual emotion in Tamil into 11 classes: ambiguous, anger, anticipation, disgust, fear, joy, love, neutral, sadness, surprise and trust. We investigated various machine learning (LR, DT, MNB, SVM), deep learning (CNN, LSTM, BiLSTM) and transformer-based models (Multilingual-BERT, XLM-R). Results reveal that the XLM-R model outdoes all other models by acquiring the highest macro f₁-score (0.33).

With the widespread usage of social media and effortless internet access, millions of posts and comments are generated every minute. Unfortunately, with this substantial rise, the usage of abusive language has increased significantly in these mediums. This proliferation leads to many hazards such as cyber-bullying, vulgarity, online harassment and abuse. Therefore, it becomes a crucial issue to detect and mitigate the usage of abusive language. This work presents our system developed as part of the shared task to detect the abusive language in Tamil. We employed three machine learning (LR, DT, SVM), two deep learning (CNN+BiLSTM, CNN+BiLSTM with FastText) and a transformer-based model (Indic-BERT). The experimental results show that Logistic regression (LR) and CNN+BiLSTM models outperformed the others. Both Logistic Regression (LR) and CNN+BiLSTM with FastText achieved the weighted F₁-score of 0.39. However, LR obtained a higher recall value (0.44) than CNN+BiLSTM (0.36). This leads us to stand the 2^nd rank in the shared task competition.

Although research on emotion classification has significantly progressed in high-resource languages, it is still infancy for resource-constrained languages like Bengali. However, unavailability of necessary language processing tools and deficiency of benchmark corpora makes the emotion classification task in Bengali more challenging and complicated. This work proposes a transformer-based technique to classify the Bengali text into one of the six basic emotions: anger, fear, disgust, sadness, joy, and surprise. A Bengali emotion corpus consists of 6243 texts is developed for the classification task. Experimentation carried out using various machine learning (LR, RF, MNB, SVM), deep neural networks (CNN, BiLSTM, CNN+BiLSTM) and transformer (Bangla-BERT, m-BERT, XLM-R) based approaches. Experimental outcomes indicate that XLM-R outdoes all other techniques by achieving the highest weighted f_1-score of 69.73% on the test data.

In recent years, several systems have been developed to regulate the spread of negativity and eliminate aggressive, offensive or abusive contents from the online platforms. Nevertheless, a limited number of researches carried out to identify positive, encouraging and supportive contents. In this work, our goal is to identify whether a social media post/comment contains hope speech or not. We propose three distinct models to identify hope speech in English, Tamil and Malayalam language to serve this purpose. To attain this goal, we employed various machine learning (SVM, LR, ensemble), deep learning (CNN+BiLSTM) and transformer (m-BERT, Indic-BERT, XLNet, XLM-R) based methods. Results indicate that XLM-R outdoes all other techniques by gaining a weighted f_1-score of 0.93, 0.60 and 0.85 respectively for English, Tamil and Malayalam language. Our team has achieved 1st, 2nd and 1st rank in these three tasks respectively.

The increasing accessibility of the internet facilitated social media usage and encouraged individuals to express their opinions liberally. Nevertheless, it also creates a place for content polluters to disseminate offensive posts or contents. Most of such offensive posts are written in a cross-lingual manner and can easily evade the online surveillance systems. This paper presents an automated system that can identify offensive text from multilingual code-mixed data. In the task, datasets provided in three languages including Tamil, Malayalam and Kannada code-mixed with English where participants are asked to implement separate models for each language. To accomplish the tasks, we employed two machine learning techniques (LR, SVM), three deep learning (LSTM, LSTM+Attention) techniques and three transformers (m-BERT, Indic-BERT, XLM-R) based methods. Results show that XLM-R outperforms other techniques in Tamil and Malayalam languages while m-BERT achieves the highest score in the Kannada language. The proposed models gained weighted f_1 score of 0.76 (for Tamil), 0.93 (for Malayalam ), and 0.71 (for Kannada) with a rank of 3rd, 5th and 4th respectively.

In the past few years, the meme has become a new way of communication on the Internet. As memes are in images forms with embedded text, it can quickly spread hate, offence and violence. Classifying memes are very challenging because of their multimodal nature and region-specific interpretation. A shared task is organized to develop models that can identify trolls from multimodal social media memes. This work presents a computational model that we developed as part of our participation in the task. Training data comes in two forms: an image with embedded Tamil code-mixed text and an associated caption. We investigated the visual and textual features using CNN, VGG16, Inception, m-BERT, XLM-R, XLNet algorithms. Multimodal features are extracted by combining image (CNN, ResNet50, Inception) and text (Bi-LSTM) features via early fusion approach. Results indicate that the textual approach with XLNet achieved the highest weighted f_1-score of 0.58, which enable our model to secure 3rd rank in this task.

This paper illustrates the details description of technical text classification system and its results that developed as a part of participation in the shared task TechDofication 2020. The shared task consists of two sub-tasks: (i) first task identify the coarse-grained technical domain of given text in a specified language and (ii) the second task classify a text of computer science domain into fine-grained sub-domains. A classification system (called ‘TechTexC’) is developed to perform the classification task using three techniques: convolution neural network (CNN), bidirectional long short term memory (BiLSTM) network, and combined CNN with BiLSTM. Results show that CNN with BiLSTM model outperforms the other techniques concerning task-1 of sub-tasks (a, b, c and g) and task-2a. This combined model obtained f1 scores of 82.63 (sub-task a), 81.95 (sub-task b), 82.39 (sub-task c), 84.37 (sub-task g), and 67.44 (task-2a) on the development dataset. Moreover, in the case of test set, the combined CNN with BiLSTM approach achieved that higher accuracy for the subtasks 1a (70.76%), 1b (79.97%), 1c (65.45%), 1g (49.23%) and 2a (70.14%).