BERTGEN: Multi-task Generation through BERT

TL;DR

BERTGEN is a multi-task, decoder-only generative model that combines multimodal and multilingual pretraining to excel in image captioning, translation, and zero-shot language generation, outperforming many baselines.

Contribution

It introduces BERTGEN, a novel model that fuses VL-BERT and M-BERT for multi-task generation, demonstrating effective transfer and zero-shot capabilities.

Findings

Outperforms strong baselines across tasks

Shows competitive zero-shot language generation

Benefits significantly from multi-task training

Abstract

We present BERTGEN, a novel generative, decoder-only model which extends BERT by fusing multimodal and multilingual pretrained models VL-BERT and M-BERT, respectively. BERTGEN is auto-regressively trained for language generation tasks, namely image captioning, machine translation and multimodal machine translation, under a multitask setting. With a comprehensive set of evaluations, we show that BERTGEN outperforms many strong baselines across the tasks explored. We also show BERTGEN's ability for zero-shot language generation, where it exhibits competitive performance to supervised counterparts. Finally, we conduct ablation studies which demonstrate that BERTGEN substantially benefits from multi-tasking and effectively transfers relevant inductive biases from the pre-trained models.