Unified Question Answering in Slovene

TL;DR

This paper adapts the UnifiedQA question-answering approach to Slovene using SloT5 and mT5 models, demonstrating that a general model can effectively handle multiple question formats and improve performance through cross-lingual transfer.

Contribution

It introduces a unified question-answering model for Slovene, leveraging existing datasets and cross-lingual transfer, achieving state-of-the-art results for the language.

Findings

The unified model performs comparably to specialized models across formats.

Cross-lingual transfer from English improves Slovene question answering.

Performance in Slovene still lags behind English but is significantly improved.

Abstract

Question answering is one of the most challenging tasks in language understanding. Most approaches are developed for English, while less-resourced languages are much less researched. We adapt a successful English question-answering approach, called UnifiedQA, to the less-resourced Slovene language. Our adaptation uses the encoder-decoder transformer SloT5 and mT5 models to handle four question-answering formats: yes/no, multiple-choice, abstractive, and extractive. We use existing Slovene adaptations of four datasets, and machine translate the MCTest dataset. We show that a general model can answer questions in different formats at least as well as specialized models. The results are further improved using cross-lingual transfer from English. While we produce state-of-the-art results for Slovene, the performance still lags behind English.