Datasets › SeaEval
SeaEval
SeaEval is a benchmark designed for evaluating multilingual foundation models (FMs). These large language models (LLMs) have demonstrated impressive generalizability and adaptability across various downstream tasks. The SeaEval benchmark goes beyond standard accuracy metrics and investigates how well these models understand and reason with natural language, as well as their comprehension of cultural practices, nuances, and values¹.
Here are some key aspects of the SeaEval benchmark:
-
Multilingual Context: SeaEval spans multiple languages, allowing researchers to assess model performance across diverse linguistic contexts.
-
Cultural Reasoning: In addition to traditional NLP tasks, SeaEval evaluates how well models comprehend cultural nuances and practices. This is crucial for applications in multicultural scenarios.
-
Brittleness Analysis: The benchmark explores the brittleness of foundation models in terms of semantics and multilinguality. For instance:
- Some models exhibit varied behavior when given paraphrased instructions.
- Exposure bias (e.g., positional bias, majority label bias) remains a challenge.
- Consistent responses are expected for semantically equivalent queries across languages, but most models surprisingly demonstrate inconsistency for factual, scientific, and commonsense knowledge questions.
- Multilingually-trained models have not yet achieved "balanced multilingual" capabilities.
-
Empirical Results: SeaEval provides empirical results across classic NLP tasks, reasoning, and cultural comprehension.
In summary, SeaEval serves as a launchpad for thorough investigations and evaluations of multilingual and multicultural scenarios, emphasizing the need for more generalizable semantic representations and enhanced multilingual contextualization¹.
(1) SeaEval for Multilingual Foundation Models: - arXiv.org. https://arxiv.org/html/2309.04766v2. (2) [2306.05179] M3Exam: A Multilingual, Multimodal, Multilevel Benchmark .... https://arxiv.org/abs/2306.05179. (3) SeaEval for Multilingual Foundation Models: From Cross-Lingual .... https://arxiv.org/pdf/2309.04766v1.pdf. (4) SeaEval for Multilingual Foundation Models: From Cross-Lingual .... https://arxiv.org/pdf/2309.04766v2.pdf. (5) undefined. https://github.com/SeaEval/SeaEval.
Benchmarks archive 2025-07-28
No leaderboard in the archive resolves to this dataset.
Papers archive 2025-07-28
No paper in the archive has a leaderboard row on this dataset; the archive counts 3 papers for it but never published that list.
Dataset loaders archive 2025-07-28
No loader listed in the archive.
Tasks archive 2025-07-28
No task tagged in the archive.
License archive 2025-07-28
No licence recorded in the archive. Absence here is not a statement about the dataset's terms.
Modalities archive 2025-07-28
No modality tagged.
Languages archive 2025-07-28
No language tagged.
Variants archive 2025-07-28
- SeaEval
1 variant name, as the archive lists them.
Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections