The field of natural language processing is witnessing significant developments in multimodal and social media discourse analysis. Researchers are exploring innovative methods to detect scientific discourse on social media, including the use of ensemble models and few-shot prompting of large language models. Another area of focus is the development of benchmarks to evaluate the multilingual capabilities of large multimodal models, with a emphasis on fairness and bias assessment. Noteworthy papers in this area include:

• LinguaMark, a benchmark for evaluating state-of-the-art large multimodal models on a multilingual Visual Question Answering task, which reveals that closed-source models generally achieve the highest overall performance.
• The DS@GT team's work on ensemble methods for detection of scientific discourse on social media, which achieved a macro-averaged F1 score of 0.8611.