Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MST5 -- Multilingual Question Answering over Knowledge Graphs

Nikit Srivastava, Mengshi Ma, Daniel Vollmers, Hamada Zahera, Diego Moussallem, Axel-Cyrille Ngonga Ngomo

TL;DR

KGQA remains English-centric, and MST5 addresses multilingual KGQA by using a single pretrained multilingual transformer (mT5 MT5) augmented with linguistic context and entity information to generate SPARQL queries end-to-end. Formally, given a natural language query $Q$ and auxiliary data $A$, MST5 seeks $\hat{S} = \arg\max_{S'} P(S'\mid Q, A; \theta)$ and minimizes the loss $\mathcal{L}(\theta) = -\log P(S\mid Q, A; \theta)$. The approach preprocesses SPARQL targets, concatenates features, and leverages attention to jointly model $Q$ and $A$. On QALD-9-Plus (updated) and QALD-10, MST5 variants outperform baselines (e.g., DeepPavlov-2023) across languages, with added coverage for Chinese and Japanese and open-source code facilitating replication; limitations include third-party tool dependencies and lower performance on some low-resource languages, pointing to multi-task learning as future work.

Abstract

Knowledge Graph Question Answering (KGQA) simplifies querying vast amounts of knowledge stored in a graph-based model using natural language. However, the research has largely concentrated on English, putting non-English speakers at a disadvantage. Meanwhile, existing multilingual KGQA systems face challenges in achieving performance comparable to English systems, highlighting the difficulty of generating SPARQL queries from diverse languages. In this research, we propose a simplified approach to enhance multilingual KGQA systems by incorporating linguistic context and entity information directly into the processing pipeline of a language model. Unlike existing methods that rely on separate encoders for integrating auxiliary information, our strategy leverages a single, pretrained multilingual transformer-based language model to manage both the primary input and the auxiliary data. Our methodology significantly improves the language model's ability to accurately convert a natural language query into a relevant SPARQL query. It demonstrates promising results on the most recent QALD datasets, namely QALD-9-Plus and QALD-10. Furthermore, we introduce and evaluate our approach on Chinese and Japanese, thereby expanding the language diversity of the existing datasets.

MST5 -- Multilingual Question Answering over Knowledge Graphs

TL;DR

KGQA remains English-centric, and MST5 addresses multilingual KGQA by using a single pretrained multilingual transformer (mT5 MT5) augmented with linguistic context and entity information to generate SPARQL queries end-to-end. Formally, given a natural language query and auxiliary data , MST5 seeks and minimizes the loss . The approach preprocesses SPARQL targets, concatenates features, and leverages attention to jointly model and . On QALD-9-Plus (updated) and QALD-10, MST5 variants outperform baselines (e.g., DeepPavlov-2023) across languages, with added coverage for Chinese and Japanese and open-source code facilitating replication; limitations include third-party tool dependencies and lower performance on some low-resource languages, pointing to multi-task learning as future work.

Abstract

Knowledge Graph Question Answering (KGQA) simplifies querying vast amounts of knowledge stored in a graph-based model using natural language. However, the research has largely concentrated on English, putting non-English speakers at a disadvantage. Meanwhile, existing multilingual KGQA systems face challenges in achieving performance comparable to English systems, highlighting the difficulty of generating SPARQL queries from diverse languages. In this research, we propose a simplified approach to enhance multilingual KGQA systems by incorporating linguistic context and entity information directly into the processing pipeline of a language model. Unlike existing methods that rely on separate encoders for integrating auxiliary information, our strategy leverages a single, pretrained multilingual transformer-based language model to manage both the primary input and the auxiliary data. Our methodology significantly improves the language model's ability to accurately convert a natural language query into a relevant SPARQL query. It demonstrates promising results on the most recent QALD datasets, namely QALD-9-Plus and QALD-10. Furthermore, we introduce and evaluate our approach on Chinese and Japanese, thereby expanding the language diversity of the existing datasets.
Paper Structure (22 sections, 2 equations, 2 figures, 5 tables)

This paper contains 22 sections, 2 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. LM-based SPARQL query generation
  4. Multilingual KGQA
  5. MST5 Approach
  6. Problem Definition
  7. MST5 Architecture
  8. Named Entity Recognition and Linking
  9. Linguistic feature extraction
  10. Preprocessing the target SPARQL-queries
  11. Creating Input Data for the MST5 Model
  12. Experimental Setup
  13. Training and Evaluation Datasets
  14. Model Training Details
  15. Ablation Study
  16. ...and 7 more sections

Figures (2)

  • Figure 1: An overview of the MST5 approach (from left to right). First, linguistic context and entity information is extracted from the input natural language question. Then, the extracted information is concatenated with the input before being passed on to the language model. The language model generates the resulting SPARQL query.
  • Figure 2: An example of linguistic context including dependency parsing (red) and POS-tags (green) alongisde the disambiguated entity for the text: Who are the grandchildren of Bruce Lee?