Generative Pretrained Structured Transformers: Unsupervised Syntactic Language Models at Scale
Abstract
A syntactic language model (SLM) incrementally generates a sentence with its syntactic tree in a left-to-right manner. We present Generative Pretrained Structured Transformers (GPST), an unsupervised SLM at scale capable of being pre-trained from scratch on raw texts with high parallelism. GPST circumvents the limitations of previous SLMs such as relying on gold trees and sequential training. It consists of two components, a usual SLM supervised by a uni-directional language modeling loss, and an additional composition model, which induces syntactic parse trees and computes constituent representations, supervised by a bi-directional language modeling loss. We propose a representation surrogate to enable joint parallel training of the two models in a hard-EM fashion. We pre-train GPST on OpenWebText, a corpus with billion tokens, and demonstrate the superiority of GPST over GPT-2 with a comparable size in numerous tasks covering both language understanding and language generation. Meanwhile, GPST also significantly outperforms existing unsupervised SLMs on left-to-right grammar induction, while holding a substantial acceleration on training.
Cite
@article{arxiv.2403.08293,
title = {Generative Pretrained Structured Transformers: Unsupervised Syntactic Language Models at Scale},
author = {Xiang Hu and Pengyu Ji and Qingyang Zhu and Wei Wu and Kewei Tu},
journal= {arXiv preprint arXiv:2403.08293},
year = {2024}
}
Comments
accepted by ACL 2024