Sticking to the Facts: Confident Decoding for Faithful Data-to-Text Generation
arXiv:1910.08684
Abstract
We address the issue of hallucination in data-to-text generation, i.e., reducing the generation of text that is unsupported by the source. We conjecture that hallucination can be caused by an encoder-decoder model generating content phrases without attending to the source; so we propose a confidence score to ensure that the model attends to the source whenever necessary, as well as a variational Bayes training framework that can learn the score from data. Experiments on the WikiBio (Lebretet al., 2016) dataset show that our approach is more faithful to the source than existing state-of-the-art approaches, according to both PARENT score (Dhingra et al., 2019) and human evaluation. We also report strong results on the WebNLG (Gardent et al., 2017) dataset.
References in corpus (14)
- Auto-Encoding Variational Bayes
- Sequence to Sequence Learning with Neural Networks
- Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
- A Neural Conversational Model
- A Deep Reinforced Model for Abstractive Summarization
- Leveraging Pre-trained Checkpoints for Sequence Generation Tasks
- Aligning Books and Movies: Towards Story-like Visual Explanations by Watching Movies and Reading Books
- Selective Encoding for Abstractive Sentence Summarization
- Variational Bayesian Inference with Stochastic Search
- Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization
- Calibration, Entropy Rates, and Memory in Language Models
- Neural Data-to-Text Generation via Jointly Learning the Segmentation and Correspondence
- Improved Natural Language Generation via Loss Truncation
- Text Generation with Exemplar-based Adaptive Decoding
Cited by in corpus (15)
- Detecting Hallucinated Content in Conditional Neural Sequence Generation
- On Hallucination and Predictive Uncertainty in Conditional Language Generation
- SynthBio: A Case Study in Human-AI Collaborative Curation of Text Datasets
- Improved Natural Language Generation via Loss Truncation
- Towards Faithfulness in Open Domain Table-to-text Generation from an Entity-centric View
- Text Generation by Learning from Demonstrations
- Logic2Text: High-Fidelity Natural Language Generation from Logical Forms
- Sometimes We Want Translationese
- Dissecting Generation Modes for Abstractive Summarization Models via Ablation and Attribution
- AgreeSum: Agreement-Oriented Multi-Document Summarization
- DYPLOC: Dynamic Planning of Content Using Mixed Language Models for Text Generation
- Logic-Consistency Text Generation from Semantic Parses
- Truth-Conditional Captioning of Time Series Data
- Neural Path Hunter: Reducing Hallucination in Dialogue Systems via Path Grounding
- Sketch and Refine: Towards Faithful and Informative Table-to-Text Generation