Synthetic Data Augmentation for Zero-Shot Cross-Lingual Question Answering

By Arij Riabi, Thomas Scialom, Rachel Keraron, Benoît Sagot, Djamé Seddah, Jacopo Staiano. 2020

62 citations recorded by the source corpus.

Coupled with the availability of large scale datasets, deep learning architectures have enabled rapid progress on the Question Answering task. However, most of those datasets are in English, and the performances of state-of-the-art multilingual models are significantly lower when evaluated on non-English data. Due to high data collection costs, it is not realistic to obtain annotated data for each language one desires to support. We propose a method to improve the Cross-lingual Question Answering performance without requiring additional annotated data, leveraging Question Generation models to produce synthetic samples in a cross-lingual fashion. We show that the proposed method allows to significantly outperform the baselines trained on English data only. We report a new state-of-the-art on four datasets: MLQA, XQuAD, SQuAD-it and PIAF (fr).

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Information Systems, Natural Language Processing, Question Answering & Reading Comprehension

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Luke S. Zettlemoyer
2. Hannaneh Hajishirzi
3. Mohit Bansal
4. Percy Liang
5. Noah A. Smith
6. Jianfeng Gao
7. Caiming Xiong
8. Graham Neubig

In Computer Vision and Pattern Recognition

1. Hannaneh Hajishirzi
2. Sebastian Ruder
3. Mirella Lapata
4. Thomas Scialom
5. Akari Asai
6. Dipanjan Das
7. Kuzman Ganchev
8. Tuhin Chakrabarty

In Information Systems

1. Akari Asai
2. Dipanjan Das
3. Mirella Lapata
4. Joshua Maynez
5. Smaranda Muresan
6. Hannaneh Hajishirzi
7. Sebastian Ruder
8. Thomas Scialom

In Natural Language Processing

1. Hannaneh Hajishirzi
2. Luke S. Zettlemoyer
3. Mohit Bansal
4. Thomas Scialom
5. Caiming Xiong
6. Mirella Lapata
7. Akari Asai
8. Richard Socher

In Question Answering & Reading Comprehension

1. Dipanjan Das
2. Mirella Lapata
3. Hannaneh Hajishirzi
4. Thomas Scialom

Citations in the corpus

• Data Augmentation Approaches in Natural Language Processing: A Survey (2021, unknown): “Pretrained [44], [30], [29], [32], [94], [28], [81], [127], [128] [21], [30], [83] [21], [129]”
• QAmeleon: Multilingual QA with Only 5 Examples (2022, methodology): “Other approaches (Riabi et al., 2021: Shakeri et al., 2021; Kramchaninova and Defauw, 2022) focus on zero-shot transfer, i.e., a multilingual model trained on QA data generation on SQ U AD (and optionally automatically translated SQ U AD data) is applied to other languages.”
• Fine-tuned Language Models are Continual Learners (2022, background): “In the context of Continual Learning, we also suspect that using only a comparison of the log-likelihood of respective classes would not reflect the actual model’s memory, since the decoders are known to suffer from catastrophic forgetting more than the encoders (Riabi et al., 2021).”
• One Question Answering Model for Many Languages with Cross-lingual Dense Passage Retrieval (2021, background): “Several studies augment multilingual machine reading comprehension training data by generating new question-answer pairs from randomly sampled non-English Wikipedia paragraphs (Riabi et al., 2021; Shakeri et al., 2020).”
• Continual-T0: Progressively Instructing 50+ Tasks to Language Models Without Forgetting (2022, background): “In the context of Continual Learning, we also suspect that using only a comparison of the log-likelihood of respective classes would not reflect the actual model’s memory, since the decoders are known to suffer from catastrophic forgetting more than the encoders (Riabi et al., 2021).”