# Trainable Referring Expression Generation using Overspecification Preferences

## TL;DR

This paper introduces a REG approach that groups speakers by overspecification preferences, enabling the use of larger datasets and improving performance over personalized models.

## Contribution

It proposes a novel grouping method based on overspecification preferences to enhance trainability and performance of referring expression generation models.

## Key findings

- Grouping speakers improves REG model performance.
- The method outperforms previous personalized approaches.
- Larger training datasets are effectively utilized.

## Abstract

Referring expression generation (REG) models that use speaker-dependent information require a considerable amount of training data produced by every individual speaker, or may otherwise perform poorly. In this work we present a simple REG experiment that allows the use of larger training data sets by grouping speakers according to their overspecification preferences. Intrinsic evaluation shows that this method generally outperforms the personalised method found in previous work.