引导式测验 · 硬 等级
共指解析测验
测试提及身份、代词照应、Winograd 推理、下游错误、跨度排名、语言线索、实体链接和偏差评估。
问题 1 的 8
When grouping Maria and she in Maria said she was tired, what identity relationship does coreference resolution investigate?
预览每个问题8 问题
当您参加测验时,就会出现答案和解释。
- When grouping Maria and she in Maria said she was tired, what identity relationship does coreference resolution investigate?
- Whether the two mentions refer to the same entity
- Whether the passage must be translated first
- Whether both words have the same spelling
- Whether the sentence has exactly two punctuation marks
- Linking a pronoun to an earlier expression for the same referent is a familiar example of which relation?
- Document-level sentiment classification
- 图像分割
- Pronominal anaphora
- Alphabetical token sorting
- 为什么“奖杯无法放入手提箱,因为它太大”这句话是经典的硬箱(Winograd 模式)?
- 它包含一个拼写错误
- 它根本没有代词
- 它是用被动语态写的
- 决定“它”指的是什么需要现实世界的知识,而不仅仅是语法
- 为什么共指消解对于总结和问答等任务很重要?
- 它使文本字体变大
- 错误地理解单词所指的实体会导致错误或混乱的输出
- 它可以加快计算机硬盘的速度
- 只对诗歌有用
- How do influential span-ranking coreference models construct mention links?
- By requiring all names to be alphabetically adjacent
- By deleting every pronoun before analyzing identity
- By scoring candidate mentions and potential earlier antecedents
- By assigning every span the same database record
- Which information can provide cues for a coreference model without guaranteeing a correct link?
- Mention distance, grammatical agreement, and contextual information
- Text color alone
- The speed at which the user typed
- A requirement that occupation stereotypes determine gender
- What extra step distinguishes entity linking from grouping coreferent mentions inside a document?
- Counting how many pronouns appear without resolving them
- Associating a mention with a disambiguated external knowledge-base entry
- Changing every mention to uppercase
- Requiring every mention to be a different entity
- What does WinoBias examine through occupation-based, Winograd-style pronoun examples?
- The screen resolution used to display occupations
- A guarantee that every occupation has one gender
- The file size of every trained model
- Differences between pro-stereotypical and anti-stereotypical coreference performance