ZusammenQA: Data Augmentation with Specialized Models for Cross-lingual Open-retrieval Question Answering System

Abstract

This paper introduces our proposed system for the MIA Shared Task on Cross-lingual Openretrieval Question Answering (COQA). In this challenging scenario, given an input question the system has to gather evidence documents from a multilingual pool and generate from them an answer in the language of the question. We devised several approaches combining different model variants for three main components: Data Augmentation, Passage Retrieval, and Answer Generation. For passage retrieval, we evaluated the monolingual BM25 ranker against the ensemble of re-rankers based on multilingual pretrained language models (PLMs) and also variants of the shared task baseline, re-training it from scratch using a recently introduced contrastive loss that maintains a strong gradient signal throughout training by means of mixed negative samples. For answer generation, we focused on languageand domain-specialization by means of continued language model (LM) pretraining of existing multilingual encoders. Additionally, for both passage retrieval and answer generation, we augmented the training data provided by the task organizers with automatically generated question-answer pairs created from Wikipedia passages to mitigate the issue of data scarcity, particularly for the low-resource languages for which no training data were provided. Our results show that language-and domain-specialization as well as data augmentation help, especially for low-resource languages.

MIA 2022 Shared Task: Evaluating Cross-lingual Open-Retrieval Question Answering for 16 Diverse Languages

“…ZusammenQA. Hung et al (2022) follow the retrieve-then-read system, but with the expansion of several components, along with training methods and data augmentation. Their retriever ensembles supervised models (mDPR and mDPR with a MixCSE loss; along with unsupervised sparse (Oracle BM-25) and unsupervised dense models (DISTIL, LaBSE, MiniLM, MPNet).…”

“…Texttron and Team Utah combine both BM25 and mDPR, while ZusammenQA explore a diverse set of unsupervised and supervised retrieval approaches including BM25 and LaBSE (Feng et al, 2022). Team Utah shows that combining BM25 with mDPR helps, while ZusammenQA shows that only using BM25 gives significantly lower scores than the original baseline (Hung et al, 2022), as BM25 does not have cross-lingual phrase matching capabilities. Texttron iteratively trained their dense retriever, mining increasingly hard negative examples using BM25 and query translation, filtered using simple heuristics.…”