Neural Text Generation from Rich Semantic Representations

TL;DR

This paper introduces neural models that generate high-quality English text from rich semantic structures called MRS, outperforming previous methods and demonstrating the effectiveness of MRS in structured semantic to text generation.

Contribution

It presents a novel neural approach to generate text from MRS, a detailed semantic representation, and shows improved performance using a grammar-based parser for training data augmentation.

Findings

Sequence-to-sequence model achieves BLEU 66.11 on gold data.

Using a grammar-based parser increases BLEU to 77.17 with silver data.

MRS-based representations effectively support structured semantics and natural language generation.

Abstract

We propose neural models to generate high-quality text from structured representations based on Minimal Recursion Semantics (MRS). MRS is a rich semantic representation that encodes more precise semantic detail than other representations such as Abstract Meaning Representation (AMR). We show that a sequence-to-sequence model that maps a linearization of Dependency MRS, a graph-based representation of MRS, to English text can achieve a BLEU score of 66.11 when trained on gold data. The performance can be improved further using a high-precision, broad coverage grammar-based parser to generate a large silver training corpus, achieving a final BLEU score of 77.17 on the full test set, and 83.37 on the subset of test data most closely matching the silver data domain. Our results suggest that MRS-based representations are a good choice for applications that need both structured semantics and the ability to produce natural language text as output.