Last Translation Benchmark
- 类型：arxiv
- 标识：2609.04173
- 链接：https://arxiv.org/abs/2609.04173
- 主分类：evaluation
- 形态：benchmark
- 被引：0
- 被引来源：Semantic Scholar + OpenAlex
- S2被引：0
- OpenAlex被引：0
- 影响力被引：0
- TLDR：The Last Translation Benchmark is introduced, a collection of human-authored and peer-reviewed examples that break leading machine translation models and a new evaluation approach: each example comes with handcrafted verification rules describing concrete failure cases on that example, therefore allowing reliable and actionable future evaluation.
- OpenAlex ID：W7208726007
- OpenAlex DOI：10.48550/arxiv.2609.04173
- DOI：10.48550/arxiv.2609.04173
- DOI来源：OpenAlex
- 开放获取：green
- 开放获取链接：https://arxiv.org/pdf/2609.04173
- OpenAlex更新：2026-09-07
- 待LLM分类：否
- 标题中文：终极翻译基准（Last Translation Benchmark）
- TLDR中文：提出 The Last Translation Benchmark，一组由人工编写并经同行评审的样本，可打破领先的机器翻译模型，并提出新评估方法：每个样本附带手工编写的验证规则，描述该样本上的具体失败案例，从而支持可靠且可操作的未来评估。
- 来源文件：
- /inbox/tom/_candidates/2026-09-05-agent-rag-longcontext-candidates.json
- /inbox/tom/_candidates/2026-09-06-agent-rag-longcontext-candidates.json
- [S2 enrich]
- [OpenAlex backfill]
- /inbox/tom/_candidates/2026-09-07-agent-rag-longcontext-candidates.json