Learn Before
Non-LLM Methods for Problem Decomposition in Question Answering
Prior to the prevalence of Large Language Models, a common technique for problem decomposition in question answering involved using a separate, specialized neural model. This model's purpose was to generate simpler sub-questions from a complex original question, and it could produce these sub-questions either all at once in a batch or sequentially.
0
1
Tags
Ch.3 Prompting - Foundations of Large Language Models
Foundations of Large Language Models
Computing Sciences
Foundations of Large Language Models Course
Related
Example of Multi-hop Question Answering
Non-LLM Methods for Problem Decomposition in Question Answering
Example of Multi-hop Question Answering: Albert Einstein's Birth Country
An automated question-answering system is given the following query: 'Who directed the movie that won the Best Picture Oscar in the year the first person walked on the moon?' To answer this, the system must find and combine information from different sources. Which of the following sets of sub-questions correctly breaks down the original query into the logical steps needed to find the final answer?
Example of Multi-hop Question Answering Decomposition
An automated system needs to answer the complex question: 'What is the official language of the country where the inventor of the telephone was born?' Arrange the following steps into the correct logical sequence that the system must follow to find the final answer.
Decomposing a Complex Query
Learn After
Imagine a question-answering system that tackles a complex query by first using a specialized neural model to break it down into simpler sub-questions. The system is designed to generate these sub-questions sequentially, meaning it generates the first sub-question, finds its answer, and only then generates the second sub-question. Which of the following best describes a key advantage of this sequential design choice over an alternative approach that generates all sub-questions simultaneously at the beginning?
Decomposing a Complex Query
Trade-offs in Sub-Question Generation Strategies