A Hierarchical Model for Data-to-Text Generation

TL;DR

This paper introduces a hierarchical data-to-text generation model that better preserves data structure during translation, improving quality over traditional linearization methods.

Contribution

It presents a novel hierarchical approach encoding data at multiple levels, surpassing sequence linearization limitations in data-to-text tasks.

Findings

Outperforms baseline models on RotoWire dataset

Improves both qualitative and quantitative metrics

Effectively captures data structure in generated text

Abstract

Transcribing structured data into natural language descriptions has emerged as a challenging task, referred to as "data-to-text". These structures generally regroup multiple elements, as well as their attributes. Most attempts rely on translation encoder-decoder methods which linearize elements into a sequence. This however loses most of the structure contained in the data. In this work, we propose to overpass this limitation with a hierarchical model that encodes the data-structure at the element-level and the structure level. Evaluations on RotoWire show the effectiveness of our model w.r.t. qualitative and quantitative metrics.