# Semantic Parsing for Conversational Question Answering over Knowledge Graphs

**Authors:** Laura Perez-Beltrachini, Parag Jain, Emilio Monti, Mirella Lapata  
**arXiv:** [2301.12217](https://arxiv.org/abs/2301.12217)  
**Date:** 2023-01-28  
**Summary page:** https://www.summarizepaper.com/en/paper/2301.12217v1/

## Key points

- The paper explores the development of semantic parsers that can understand natural language questions in a conversation with a user and translate them into formal queries over definitions in a general-purpose knowledge graph (KG) with vast vocabularies.
- The KG covers thousands of concept names and relations, as well as millions of entities.
- The authors develop a dataset where user questions are annotated with Sparql parses, and system answers correspond to execution results.
- The paper presents two different semantic parsing approaches and highlights the challenges involved in the task, including dealing with large vocabularies, modeling conversation context, predicting queries with multiple entities, and generalizing to new questions at test time.
- The authors hope that their dataset will serve as a useful testbed for the development of conversational semantic parsers.
- The authors release their dataset and models at https://github.com/EdinburghNLP/SPICE.

## Summary

The paper titled "Semantic Parsing for Conversational Question Answering over Knowledge Graphs" explores the development of semantic parsers that can understand natural language questions in a conversation with a user and translate them into formal queries over definitions in a general-purpose knowledge graph (KG) with vast vocabularies. The KG covers thousands of concept names and relations, as well as millions of entities. To achieve this goal, the authors develop a dataset where user questions are annotated with Sparql parses, and system answers correspond to execution results. The paper presents two different semantic parsing approaches and highlights the challenges involved in the task. These challenges include dealing with large vocabularies, modeling conversation context, predicting queries with multiple entities, and generalizing to new questions at test time. The authors hope that their dataset will serve as a useful testbed for the development of conversational semantic parsers. The authors release their dataset and models at https://github.com/EdinburghNLP/SPICE. Overall, this paper provides valuable insights into developing conversational semantic parsers that can understand natural language questions in conversations with users and ground them to formal queries over definitions in a general-purpose knowledge graph. The authors are Laura Perez-Beltrachini, Parag Jain, Emilio Monti, and Mirella Lapata.