自然言語処理学研究室のFrederikus HUDIさん(博士後期課程1年)らが、NAACL 2025 Best Theme Paper Awardを受賞しました。(2025/5/2)
Frederikus HUDI et al. from Natural Language Processing lab received the NAACL 2025 Best Theme Paper Award at NAACL 2025.
| NAACL:Nations of the Americas Chapter of the Association for Computational Linguisticsは、北米、中南米における計算言語学と自然言語処理（NLP）研究の推進を目的とした、最高峰の国際会議です。研究者、実務家、業界の専門家が一堂に会し、最先端の研究成果を共有し、意見を交換し、協力関係を築くことを目的としています。「多文化世界におけるNLP」を特別テーマとするNAACL 2025は、2025年4月29日から5月4日まで、アメリカ合衆国ニューメキシコ州で開催されました。文化的および言語的多様性を受け入れ、支援する言語技術の開発に焦点を当てるとともに、大規模言語モデル（LLM）、多言語NLP、創発的推論能力、そして責任あるAIにおけるブレークスルーを探求しました。 |  |
| Best Theme Paper Awardは、会議の特別テーマ「多文化世界におけるNLP」に最も効果的に取り組んだ優れた研究論文を表彰します。本賞の目的は、包括的で、文化を理解し、言語的および文化的多様性を支援する言語技術の開発の分野に焦点を当て、最先端の研究を推進することです。この賞は、マイノリティの言語、モデルにおける文化的偏見、多様なグローバル文化に有益な NLP アプリケーションの作成といった課題に取り組む研究を奨励します。 | |
| The 2025 Annual Conference of the Nations of the Americas Association for Computational Linguistics (NAACL 2025) is a premier international conference dedicated to advancing research in computational linguistics and natural language processing (NLP) across North, Central, and South America. The conference was held in New Mexico, USA from April 29 to May 4. Its purpose was to bring together researchers, practitioners, and industry experts to share cutting-edge work, exchange ideas, and foster collaboration. With a special theme of "NLP in a Multicultural World," the conference aimed to highlight the development of language technologies that embrace and support cultural and linguistic diversity, alongside exploring breakthroughs in large language models (LLMs), multilingual NLP, emergent reasoning capabilities, and responsible AI. The award recognises an outstanding research paper that most effectively addresses the conference's special theme, "NLP in a Multicultural World." Its purpose is to highlight and promote cutting-edge work that focuses on developing language technologies that are inclusive, culturally aware, and supportive of linguistic and cultural diversity. This award encourages research that tackles the challenges of under-represented languages, cultural biases in models, and the creation of NLP applications beneficial to diverse global cultures. | |
- 受賞者(NAIST所属者のみ記載) Awardees (NAIST affiliates)：
Frederikus Hudi(D1), Adam Nohejl(D3), Ashmari Pramodya(M2), Maria Angelica Riera Machin(M2), Taro Watanabe(NAIST).
写真はFrederikus Hudiさん
- 受賞研究テーマ Research theme：
"WorldCuisines: A Massive-Scale Benchmark for Multilingual and Multicultural Visual Question Answering on Global Cuisines"
A Benchmark for Multilingual and Multicultural Visual Question Answering on Global Cuisines. The awarded work evaluates how Large Language Models (LLMs) recognise and describe culturally-rich culinary dishes across 30 different languages and dialects. The benchmark dataset spans over 9 language families and featuring over 1 million data points.
The paper introduces a significant new dataset designed to evaluate how Vision Language Models (VLMs) understand culture-specific knowledge, particularly in non-English and underrepresented contexts. This benchmark comprises over 1 million text-image pairs across 30 languages and dialects, focusing on global cuisines. It includes two primary tasks: predicting dish names (with no-context, contextualised, and adversarial subtasks) and identifying the regional origin of dishes. The research highlights that while VLMs perform better with correct contextual information, they still struggle with adversarial contexts and accurately predicting specific regional cuisines and languages, underscoring the need for more culturally aware and robust VLM development.
- 筆頭著者 Co-first author：
Genta Indra Winata*, Frederikus Hudi(D1)*, Patrick Amadeus Irawan*, David Anugraha*, and Rifki Afina Putri*.
- 受賞者のコメント Awardee's voice：
We are incredibly honored to receive this award as this validates our commitment to advancing culturally aware and inclusive NLP. This award encourages us to continue developing technologies that celebrate and respect global cultures, starting with the universal lens of food!
- 外部リンク Links to：
- NAACL 2025 HP：https://2025.naacl.org/