对话式人工智能测验

当您参加测验时，就会出现答案和解释。

1. 在经典的面向任务的对话式 AI 管道中，自然语言理解 (NLU) 组件的作用是什么？
   • 将文本回复转换为语音音频
   • 识别用户的意图并提取称为槽的关键细节
   • 将对话永久存储在数据库中
   • 训练语音识别器
2. What does a structured dialogue-state tracker maintain in the pipeline described here?
   • Relevant slot values and other task information across turns
   • Only the final audio waveform
   • A mandatory permanent record inside every model weight
   • Only an alphabetical word count
3. 为什么旧的基于规则或严格的基于意图的聊天机器人经常被描述为“脆弱”？
   • 他们用了太多的电
   • 它们只能在智能音箱上运行
   • 当用户以意想不到的方式提出请求时，他们会陷入困境
   • 他们生成的答案总是太长
4. What role does natural language generation serve in a modular dialogue pipeline?
   • Only transcribing incoming audio
   • Only predicting a hardware temperature
   • Only storing a destination slot without wording
   • Expressing a selected response or action result in words
5. How can earlier chat turns affect a model response when they are included in its context?
   • They supply information the current generation can condition on
   • They necessarily become permanent new model weights
   • They guarantee unlimited recall of omitted messages
   • They erase the current user request by definition
6. 对话式人工智能面临的主要挑战是什么？
   • 在屏幕上显示文本
   • 向菜单添加更多按钮
   • 自信地避免错误答案并知道何时将其交给人工处理
   • 计算用户输入的单词数
7. Which pipeline role selects whether to ask for a missing slot or take the next allowed task action?
   • Only the waveform decoder
   • Only a document filename sorter
   • Only a visual theme selector
   • The dialogue policy
8. Which component converts a spoken user request into text before a text-based dialogue pipeline processes it?
   • Only natural language generation of the reply
   • 自动语音识别
   • Only a database backup
   • Only a model-weight quantizer