Research from the School of Informatics has been recognised with two ACL 2026 Test-of-Time Paper Awards, honouring work published a decade ago that continues to shape natural language processing and artificial intelligence today. The awards recognise pioneering contributions to both modern language models and responsible AI, with honours going to Professor Alexandra Birch, Professor Rico Sennrich, Dr Barry Haddow and Dr Zeerak Talat, alongside their collaborators. Researchers from the School of Informatics have received two 2026 ACL Test-of-Time Paper Awards, recognising research that has had a lasting and significant influence on the fields of computational linguistics and natural language processing (NLP).Presented at the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) in San Diego, the awards honour papers published in 2016 whose impact continues to be felt across research and industry today. Foundations of modern language models One of the awards was presented to Professor Alexandra Birch, Professor Rico Sennrich and Dr Barry Haddow, all from the School of Informatics, for their paper:"Neural Machine Translation of Rare Words with Subword Units" (ACL 2016)The paper addressed a major challenge in early neural machine translation systems: how to deal with rare or previously unseen words.The researchers introduced an approach based on Byte Pair Encoding (BPE), adapting a compression algorithm to split words into smaller subword units. This enabled neural systems to move beyond fixed vocabularies and handle open-vocabulary translation more effectively.What began as a solution to a machine translation problem has since become one of the foundational technologies underpinning modern AI systems.Today, subword tokenisation methods derived from this work are used in many of the world's most widely deployed large language models and multilingual AI systems.Reflecting on the award, Professor Birch noted that the original work was developed to solve a specific problem in machine translation but ultimately proved influential far beyond its initial scope. Tokenization, something that felt like plumbing, turned out to be one of the key components of LLMs. Every model used, from the first GPT to whatever shipped this week, still starts by doing some version of what we described in 2016. Professor Alexandra Birch Chair of Multilingual Natural Language Processing in the School of Informatics The paper's influence extends well beyond machine translation, contributing to the development of multilingual language technologies and large-scale generative AI systems used across academia and industry. Link to paper Lasting impact on responsible AI and content moderation A second Test-of-Time Award was presented to Dr Zeerak Talat, Chancellor's Fellow in Responsible Machine Learning and AI at the School of Informatics, and Professor Dirk Hovy (University of Milan), for their paper:"Hateful Symbols or Hateful People? Predictive Features for Hate Speech Detection on Twitter" (NAACL Student Research Workshop, 2016)Published when the field of online content moderation was still in its infancy, the paper helped introduce hate speech detection and online harms research to the Natural Language Processing (NLP) community.The research combined linguistic and contextual features to examine how harmful content can be identified online, challenging the idea that individual keywords alone are sufficient for detecting hate speech.Over the past decade, the paper has become highly influential in research on content moderation, online harms, platform governance and responsible AI.Reflecting on the award, Dr Talat highlighted both its personal significance and the broader message it sends about the value of student-led research. What brings me joy here, beyond the award itself obviously, is the signal it also sends: early work by students can have a large impact on our field and can be rewarding in so many ways. Dr Zeerak Talat Chancellor's Fellow in Responsible Machine Learning and AI at the School of Informatics The paper was originally published while Dr Talat was a student researcher and remains one of the foundational works in the study of computational approaches to online harms. Link to paper Research with enduring influence The ACL Test-of-Time Awards recognise papers whose influence extends well beyond their original publication, highlighting research that continues to shape both academic inquiry and real-world technologies many years later.The recognition of these two papers demonstrates the breadth of Informatics research impact—from foundational methods that underpin today's large language models, to early work that helped establish responsible AI and content moderation as major areas of study. Related links ACL Test-of-Time Awards 2026 Neural Machine Translation of Rare Words with Subword Units Hateful Symbols or Hateful People? Predictive Features for Hate Speech Detection on Twitter Research Blog: 10 years of Byte-Pair Encoding Tokenization This article was published on Wednesday 15 July 2026
