MELT Workshop @ COLM 2025

📢 Announcements

(May 23, 2025) Please feel free to contact the organizers at: melt-workshop@googlegroups.com

Recent advances in large language models (LLMs) have led to a paradigm shift in the economy, society, and daily life, demonstrating remarkable capabilities for complex tasks, such as reasoning and multimodality. However, these innovations remain largely centered around English and are unevenly distributed across languages, cultures, and values.

That is, linguistic and cultural inclusion is lacking in language models, mainly due to an imbalance of language sources used for training. While there are more than 7,000 languages in the world, most NLP research sheds light on a handful of high-resource languages (Blasi et al., 2022; Qin et al., 2024; Zhu et al., 2024). For instance, only 7% and 17% of non-English tokens are incorporated into the pre-training data of GPT-3 (Brown et al., 2020) and BLOOM (Scao et al., 2022), respectively. Further, as we often rely on translation from English-centric datasets (Jin et al., 2024) that predominantly reflect Western cultures (Ahuja et al., 2023) to build multilingual benchmarks, these models naturally uphold social values close to North American and European cultures and neglect a wide array of languages and cultures (Held et al., 2023; AlKhamissi et al., 2024; Tao et al., 2024; Xu et al., 2024; Masoud et al., 2025).

As LLMs are increasingly deployed in diverse, region- and culture-sensitive domains worldwide, the limitations of this linguistic and cultural imbalance become more pronounced, and their implications more consequential (Yang et al., 2025; Qadri et al., 2025a; Qadri et al., 2025b). These models often exhibit significant biases, reduced performance, and a lack of cultural sensitivity when applied to underrepresented languages and diverse sociocultural settings (Foroutan et al., 2023; Liu et al., 2024; Kwok et al., 2024; Romanou et al., 2025).

Addressing these limitations is essential to ensure that LLMs provide equal access to information, resources, and services to all populations equitably, regardless of language or cultural background. Our workshop aims to meet the ultimate demand to enhance diversity and inclusion in language models. We will disseminate the state-of-the-art research on building multilingual and multicultural LLMs, fostering international cross-disciplinary collaborations, and exploring the fundamental theoretical and methodological challenges in this field.