Understanding Model Robustness to User-generated Noisy Texts
This addresses robustness issues in NLP models for real-world applications with user-generated noisy texts, but it is incremental as it builds on existing noise generation methods.
The paper tackles the problem of deep neural models' sensitivity to naturally occurring noise like spelling errors by modeling errors statistically from grammatical-error-correction corpora, and it evaluates state-of-the-art NLP systems' robustness across multiple languages and tasks, showing performance drops and comparing mitigation approaches.
Sensitivity of deep-neural models to input noise is known to be a challenging problem. In NLP, model performance often deteriorates with naturally occurring noise, such as spelling errors. To mitigate this issue, models may leverage artificially noised data. However, the amount and type of generated noise has so far been determined arbitrarily. We therefore propose to model the errors statistically from grammatical-error-correction corpora. We present a thorough evaluation of several state-of-the-art NLP systems' robustness in multiple languages, with tasks including morpho-syntactic analysis, named entity recognition, neural machine translation, a subset of the GLUE benchmark and reading comprehension. We also compare two approaches to address the performance drop: a) training the NLP models with noised data generated by our framework; and b) reducing the input noise with external system for natural language correction. The code is released at https://github.com/ufal/kazitext.