Zero-shot Cross-Lingual Transfer for Synthetic Data Generation in Grammatical Error Detection
Gaetan Lopez Latouche, Marc-André Carbonneau, Ben Swanson
2024-07-18
Summary
This paper discusses a new method for detecting grammatical errors in languages that lack enough human-annotated data by using synthetic data generated from other languages.
What's the problem?
Grammatical error detection (GED) systems rely on large amounts of human-annotated examples to learn how to identify mistakes. However, many languages do not have enough of these annotated datasets, making it hard to develop effective GED tools for those languages. This creates a challenge for accurately detecting grammatical errors in low-resource languages.
What's the solution?
The authors propose a method that uses multilingual pre-trained language models to generate synthetic grammatical errors in various languages. They create a two-stage training process where the GED model is first trained on this synthetic data and then fine-tuned using human-annotated data from other languages. This approach allows the model to learn from a diverse set of examples, improving its ability to detect errors without needing extensive annotated datasets for every language.
Why it matters?
This research is important because it addresses the gap in grammatical error detection for low-resource languages. By generating synthetic data, the authors provide a way to improve language tools for speakers of these languages, helping them communicate more effectively and improving language learning resources. This can lead to better educational materials and support for speakers of underrepresented languages.
Abstract
Grammatical Error Detection (GED) methods rely heavily on human annotated error corpora. However, these annotations are unavailable in many low-resource languages. In this paper, we investigate GED in this context. Leveraging the zero-shot cross-lingual transfer capabilities of multilingual pre-trained language models, we train a model using data from a diverse set of languages to generate synthetic errors in other languages. These synthetic error corpora are then used to train a GED model. Specifically we propose a two-stage fine-tuning pipeline where the GED model is first fine-tuned on multilingual synthetic data from target languages followed by fine-tuning on human-annotated GED corpora from source languages. This approach outperforms current state-of-the-art annotation-free GED methods. We also analyse the errors produced by our method and other strong baselines, finding that our approach produces errors that are more diverse and more similar to human errors.