# Maximizing Stylistic Control and Semantic Accuracy in NLG: Personality Variation and Discourse Contrast

## TL;DR

This paper advances neural task-oriented dialogue generation by significantly improving stylistic control and semantic accuracy through model conditioning and eliminating re-ranking, demonstrated on personality and discourse contrast benchmarks.

## Contribution

It introduces a novel approach that enhances stylistic and semantic control in neural generation, outperforming previous methods on key benchmarks.

## Key findings

- Over 15 BLEU point improvement in personality style control
- Semantic errors reduced to near zero in both tasks
- Enhanced control of discourse contrast from 0.75 to 0.81

## Abstract

Neural generation methods for task-oriented dialogue typically generate from a meaning representation that is populated using a database of domain information, such as a table of data describing a restaurant. While earlier work focused solely on the semantic fidelity of outputs, recent work has started to explore methods for controlling the style of the generated text while simultaneously achieving semantic accuracy. Here we experiment with two stylistic benchmark tasks, generating language that exhibits variation in personality, and generating discourse contrast. We report a huge performance improvement in both stylistic control and semantic accuracy over the state of the art on both of these benchmarks. We test several different models and show that putting stylistic conditioning in the decoder and eliminating the semantic re-ranker used in earlier models results in more than 15 points higher BLEU for Personality, with a reduction of semantic error to near zero. We also report an improvement from .75 to .81 in controlling contrast and a reduction in semantic error from 16% to 2%.