Reproduce “Question Generation from SQL Queries Improves Neural Semantic Parsing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Daya Guo, Yibo Sun, Duyu Tang, Nan Duan, Jian Yin, Hong Chi, James Cao, Peng Chen and 1 more
Researchers at Sun Yat-sen University and Microsoft Research developed a method to improve neural semantic parsing by augmenting training data with synthetically generated natural language questions from SQL queries. This approach achieved comparable performance with 30% of supervised data and established a new state-of-the-art on WikiSQL with 63.7% logical form accuracy and 75.5% execution accuracy.
Key insights
- Semantic parser accuracy exhibits a logarithmic relationship with the amount of training data, showing diminishing returns as data volume increases.
- The copying mechanism in the question generation model is essential for replicating specific terms from SQL queries, while latent variables enhance the diversity of generated natural language questions.
- Generated data particularly benefits the semantic parsing of complex SQL queries, especially those with multiple WHERE conditions, which are often under-represented in human-annotated datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.