I am at @aaclmeeting.bsky.social in Mumbai to present our paper "The Learning Dynamics of Subword Segmentation for Morphologically Diverse Languages". I'm giving a talk tomorrow at 2pm in Room 21, in the MC-OP2 session on Multilingualism and Cross-Lingual NLP.

If a language model could dynamically optimise subword tokenisation, how would its subwords evolve during training? In our new paper we study the learning dynamics of subword segmentation: arxiv.org/pdf/2511.09197

🌍Introducing BabyBabelLM: A Multilingual Benchmark of Developmentally Plausible Training Data! LLMs learn from vastly more data than humans ever experience. BabyLM challenges this paradigm by focusing on developmentally plausible data We extend this effort to 45 new languages!

𝐃𝐨 𝐲𝐨𝐮 𝐫𝐞𝐚𝐥𝐥𝐲 𝐰𝐚𝐧𝐭 𝐭𝐨 𝐬𝐞𝐞 𝐰𝐡𝐚𝐭 𝐦𝐮𝐥𝐭𝐢𝐥𝐢𝐧𝐠𝐮𝐚𝐥 𝐞𝐟𝐟𝐨𝐫𝐭 𝐥𝐨𝐨𝐤𝐬 𝐥𝐢𝐤𝐞? 🇨🇳🇮🇩🇸🇪 Here’s the proof! 𝐁𝐚𝐛𝐲𝐁𝐚𝐛𝐞𝐥𝐋𝐌 is the first Multilingual Benchmark of Developmentally Plausible Training Data available for 45 languages to the NLP community 🎉 arxiv.org/abs/2510.10159

Today our poster will be up at @loreslm.bsky.social Poster Session #2 (2-3pm local time Abu Dhabi). It's also available online at Whova: whova.com/portal/webap...

Our paper "BabyLMs for isiXhosa: Data-Efficient Language Modelling in a Low-Resource Context" will be presented at The First Workshop on Language Models for Low-Resource Languages at #COLING2025 in Abu Dhabi. Paper: arxiv.org/pdf/2501.03855