# Semantic Noise Matters for Neural Natural Language Generation Source: [Semantic Noise Matters for Neural Natural Language Generation]() Collection index: [llms.txt]() Data updated: 2025-12-25T14:52:07.559Z Authors: Ond\\v{r}ej Du\\v{s}ek, David M. Howcroft, Verena Rieser Published: 2019-11-10T11:24:02.000Z ## Abstract Neural natural language generation (NNLG) systems are known for their pathological outputs, i.e. generating text which is unrelated to the input specification. In this paper, we show the impact of semantic noise on state-of-the-art NNLG models which implement different semantic control mechanisms. We find that cleaned data can improve semantic correctness by up to 97%, while maintaining fluency. We also find that the most common error is omitting information, rather than hallucination. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Factual Correctness on E2E Original (test)](): Factual Correctness; E2E Original (test); Add: 0.14. - [Summarization on NYT Summarization (test)](): Summarization; NYT Summarization (test); Hallucination Rate: 7.14. - [Data-to-Text Generation on E2E Cleaned](): Data-to-text generation; E2E Cleaned; Fluency: 5.23. - [Data-to-Text Generation on E2E (test)](): Data-to-text generation; E2E (test); BLEU: 66.41. - [Data-to-text generation on Cleaned E2E (test)](): Data-to-text generation; Cleaned E2E (test); BLEU: 40.5. ## Original sources and implementation [Original paper]() [PDF]() [Implementation]()