Automatic Detection of Machine Generated Text: A Critical Survey

Understand

Text generative models (TGMs) excel in producing text that matches the style of human language reasonably well.

• Such TGMs can be misused by adversaries, e.g., by automatically generating fake news and fake product reviews that can look authentic and fool humans.
• Detectors that can distinguish text generated by TGM from human written text play a vital role in mitigating such misuse of TGMs.
• Recently, there has been a flurry of works from both natural language processing (NLP) and machine learning (ML) communities to build accurate detectors for English.