Text generation

Text generation encompasses systems that produce natural language output, from rule-based templates to neural language models (GPT-2, GPT-3). Key challenges include evaluating quality and detecting when generated text is indistinguishable from human-written content.

Key papers

• Adelani et al. (2019) — Generating Sentiment-Preserving Fake Online Reviews: Demonstrates a two-step approach to generate high-quality fake product reviews using fine-tuned GPT-2 with BERT-based validation to preserve sentiment; comprehensive evaluation of human and automated detection methods.
• Clark et al. (2021) — All That's 'Human' Is Not Gold: Studies human ability to evaluate generated text across three domains (stories, news, recipes) using GPT2 and GPT3 models.

Related topics

• Neural text generation — neural approaches to text generation
• Machine-generated text detection — methods for identifying generated text
• NLP evaluation — evaluation methods for generated text