Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Innovations in Neural Data-to-text Generation: A Survey

Mandar Sharma, Ajay Gogineni, Naren Ramakrishnan

TL;DR

This survey analyzes neural data-to-text generation (D2T), outlining how modern models convert structured data (MRs, graphs, tables) into natural language while navigating fidelity, coherence, and stylistic variation. It covers data representations, preprocessing, seq2seq and non-seq2seq innovations, and evolving evaluation practices, emphasizing the roles of delexicalization, linearization, graph encoders, plan-based generation, and PLM fine-tuning. Key contributions include a taxonomy of architectural and training strategies (entity/hierarchical/graph encoders, reconstruction, regularization, RL, templates), unsupervised pretraining tailored to D2T, and non-end-to-end approaches that improve interpretability and domain transfer. The paper also highlights reproducibility and fairness, advocates living benchmarks and domain-diverse datasets, and discusses future directions at the intersection of D2T with large language models, numerical reasoning, and external tools for computation and validation.

Abstract

The neural boom that has sparked natural language processing (NLP) research through the last decade has similarly led to significant innovations in data-to-text generation (DTG). This survey offers a consolidated view into the neural DTG paradigm with a structured examination of the approaches, benchmark datasets, and evaluation protocols. This survey draws boundaries separating DTG from the rest of the natural language generation (NLG) landscape, encompassing an up-to-date synthesis of the literature, and highlighting the stages of technological adoption from within and outside the greater NLG umbrella. With this holistic view, we highlight promising avenues for DTG research that not only focus on the design of linguistically capable systems but also systems that exhibit fairness and accountability.

Innovations in Neural Data-to-text Generation: A Survey

TL;DR

This survey analyzes neural data-to-text generation (D2T), outlining how modern models convert structured data (MRs, graphs, tables) into natural language while navigating fidelity, coherence, and stylistic variation. It covers data representations, preprocessing, seq2seq and non-seq2seq innovations, and evolving evaluation practices, emphasizing the roles of delexicalization, linearization, graph encoders, plan-based generation, and PLM fine-tuning. Key contributions include a taxonomy of architectural and training strategies (entity/hierarchical/graph encoders, reconstruction, regularization, RL, templates), unsupervised pretraining tailored to D2T, and non-end-to-end approaches that improve interpretability and domain transfer. The paper also highlights reproducibility and fairness, advocates living benchmarks and domain-diverse datasets, and discusses future directions at the intersection of D2T with large language models, numerical reasoning, and external tools for computation and validation.

Abstract

The neural boom that has sparked natural language processing (NLP) research through the last decade has similarly led to significant innovations in data-to-text generation (DTG). This survey offers a consolidated view into the neural DTG paradigm with a structured examination of the approaches, benchmark datasets, and evaluation protocols. This survey draws boundaries separating DTG from the rest of the natural language generation (NLG) landscape, encompassing an up-to-date synthesis of the literature, and highlighting the stages of technological adoption from within and outside the greater NLG umbrella. With this holistic view, we highlight promising avenues for DTG research that not only focus on the design of linguistically capable systems but also systems that exhibit fairness and accountability.
Paper Structure (48 sections, 8 equations, 9 figures, 4 tables)

This paper contains 48 sections, 8 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Defining Data-to-text Generation and Scope of the Survey
  3. Survey Rationale
  4. Datasets for Data-to-text Generation
  5. Meaning Representations
  6. Graph Representations
  7. Tabular Representations
  8. Data Collection & Enrichment
  9. Data-to-text Generation Fundamentals and Notations
  10. What to Say and How to Say It
  11. Hallucinations and Omissions
  12. Establishing Notation and Revisiting Seq2Seq
  13. Innovations in Data Preprocessing
  14. Delexicalization & Noise Reduction
  15. Linearization
  16. ...and 33 more sections

Figures (9)

  • Figure 1: Illustration of D2T: Narration of time-series data (COVID19 progression in the United Kingdom at the top, the Carbon-Monoxide emissions in the state of Kansas, United States at the bottom) with the LLM-based framework $T^{3}$ (T-Cube) sharma:21. This D2T framework consumes a time-series as input and generates narratives that highlight the progression and points-of-interest (regimes, trends, and peaks) in the data through LLM-generated narratives.
  • Figure 2: AMR konstas:17 and knowledge graph ribeiro:21 snapshots, representing variants of graph-based inputs to D2T systems.
  • Figure 3: Showcasing the intent of T2T, the statistics of a basketball match between the Atlanta Hawks and the Miami Heat (left) is to be translated into easily consumable narratives (right). Snapshot from the RotoWire dataset wiseman:17.
  • Figure 4: Attention-based seq2seq framework: The encoder consumes the sequential input translating it to a weighed hidden representation to be then consumed and decoded into linguistic tokens by the decoder.
  • Figure 5: Data-to-text Generation Taxonomy Corresponding to Sections in the Survey Design
  • ...and 4 more figures