Neural Data-to-Text Generation via Jointly Learning the Segmentation and Correspondence
arXiv:2005.01096
Abstract
The neural attention model has achieved great success in data-to-text generation tasks. Though usually excelling at producing fluent text, it suffers from the problem of information missing, repetition and "hallucination". Due to the black-box nature of the neural attention architecture, avoiding these problems in a systematic way is non-trivial. To address this concern, we propose to explicitly segment target text into fragment units and align them with their data correspondences. The segmentation and correspondence are jointly learned as latent variables without any human annotations. We further impose a soft statistical constraint to regularize the segmental granularity. The resulting architecture maintains the same expressive power as neural attention models, while being able to generate fully interpretable outputs with several times less computational cost. On both E2E and WebNLG benchmarks, we show the proposed model consistently outperforms its neural attention counterparts.
Accepted at ACL 2020
References in corpus (9)
- PyTorch: An Imperative Style, High-Performance Deep Learning Library
- Evaluating the State-of-the-Art of End-to-End Natural Language Generation: The E2E NLG Challenge
- Structured Attention Networks
- Latent Alignment and Variational Attention
- A Tutorial on Deep Latent Variable Models of Natural Language
- Unsupervised Pidgin Text Generation By Pivoting English Data and Self-Training
- Improving Multi-turn Dialogue Modelling with Utterance ReWriter
- NEXUS Network: Connecting the Preceding and the Following in Dialogue Generation
- Select and Attend: Towards Controllable Content Selection in Text Generation
Cited by in corpus (8)
- Sticking to the Facts: Confident Decoding for Faithful Data-to-Text Generation
- Fork or Fail: Cycle-Consistent Training with Many-to-One Mappings
- Towards Faithfulness in Open Domain Table-to-text Generation from an Entity-centric View
- Safe Handover in Mixed-Initiative Control for Cyber-Physical Systems
- Learning Fine-grained Fact-Article Correspondence in Legal Cases
- DART: A Lightweight Quality-Suggestive Data-to-Text Annotation Tool
- Sequence-to-Sequence Learning with Latent Neural Grammars
- DocuBot : Generating financial reports using natural language interactions