Carlsmith (2021)
paperAuthors
Credibility Rating
Good quality. Reputable source with community review or editorial standards, but less rigorous than peer-reviewed venues.
Rating inherited from publication venue: arXiv
This paper addresses controllability in neural data-to-text generation through a Plan-then-Generate framework, relevant to AI safety concerns about controlling neural model outputs and ensuring generated text follows desired structures and constraints.
Paper Details
Metadata
Abstract
Recent developments in neural networks have led to the advance in data-to-text generation. However, the lack of ability of neural models to control the structure of generated output can be limiting in certain real-world applications. In this study, we propose a novel Plan-then-Generate (PlanGen) framework to improve the controllability of neural data-to-text models. Extensive experiments and analyses are conducted on two benchmark datasets, ToTTo and WebNLG. The results show that our model is able to control both the intra-sentence and inter-sentence structure of the generated output. Furthermore, empirical comparisons against previous state-of-the-art methods show that our model improves the generation quality as well as the output diversity as judged by human and automatic evaluations.
Summary
This paper introduces PlanGen, a Plan-then-Generate framework designed to enhance controllability in neural data-to-text generation models. The approach addresses a key limitation of existing neural models—their inability to control output structure—by separating planning from generation. Evaluated on ToTTo and WebNLG benchmarks, PlanGen demonstrates improved control over both intra-sentence and inter-sentence structure while achieving better generation quality and output diversity compared to previous state-of-the-art methods, as validated through human and automatic evaluations.
Cited by 1 page
| Page | Type | Quality |
|---|---|---|
| AI Compounding Risks Analysis Model | Analysis | 60.0 |
Cached Content Preview
# Plan-then-Generate: Controlled Data-to-Text Generation via Planning
Yixuan Su♠, David Vandyke♡ Sihui Wang♡ Yimai Fang♡ Nigel Collier♠
♠Language Technology Lab, University of Cambridge
♡Apple
{ys484,nhc30}@cam.ac.uk
{dvandyke,sihui\_wang,yimai\_fang}@apple.com Work done while the author was an intern at Apple.
###### Abstract
Recent developments in neural networks have led to the advance in data-to-text generation. However, the lack of ability of neural models to control the structure of generated output can be limiting in certain real-world applications. In this study, we propose a novel Plan-then-Generate (PlanGen) framework to improve the controllability of neural data-to-text models. Extensive experiments and analyses are conducted on two benchmark datasets, ToTTo and WebNLG. The results show that our model is able to control both the intra-sentence and inter-sentence structure of the generated output. Furthermore, empirical comparisons against previous state-of-the-art methods show that our model improves the generation quality as well as the output diversity as judged by human and automatic evaluations.
## 1 Introduction
Generating natural language from structured data Gatt and Krahmer ( [2018](https://ar5iv.labs.arxiv.org/html/2108.13740#bib.bib14 "")), i.e. data-to-text generation, is a research problem that is crucial to many downstream NLP applications. Some examples are dialogue systems Wen et al. ( [2016](https://ar5iv.labs.arxiv.org/html/2108.13740#bib.bib47 "")), restaurant assistant Novikova et al. ( [2017](https://ar5iv.labs.arxiv.org/html/2108.13740#bib.bib29 "")), and open domain question answering Chen et al. ( [2021](https://ar5iv.labs.arxiv.org/html/2108.13740#bib.bib2 "")).
To address this task, many researchers have designed sophisticated neural models based on various methods, such as soft-template Wiseman et al. ( [2018](https://ar5iv.labs.arxiv.org/html/2108.13740#bib.bib49 "")), copy mechanism Gehrmann et al. ( [2018](https://ar5iv.labs.arxiv.org/html/2108.13740#bib.bib15 "")), and pre-trained language models Kale and Rastogi ( [2020](https://ar5iv.labs.arxiv.org/html/2108.13740#bib.bib17 "")); Ribeiro et al. ( [2020](https://ar5iv.labs.arxiv.org/html/2108.13740#bib.bib38 "")). While achieving impressive results, most existing studies only focused on producing results that are close to the references. On the other hand, the controllability of such models is still under-explored, i.e. what to generate and in what order (the output structure) in their outputs cannot be explicitly controlled by the users.
We argue that the model’s ability to control the structure of its output is highly
![[Uncaptioned image]](https://ar5iv.labs.arxiv.org/html/2108.13740/assets/images/knowledge_table.png)Table 1: An Example of Knowledge Table
desirable for at least two reasons. (1) Arranging the structure of the output in a certain form enables it to have greater naturalness, as the structure of the sentence often reflects t
... (truncated, 82 KB total)64ad308db00b3ce7 | Stable ID: OTgzY2YzMm