Discourse processing suffers from data sparsity, especially for dialogues. As
a result, we explore approaches to build discourse structures for dialogues,
based on attention matrices from Pre-trained Language Models (PLMs). We
investigate multiple tasks for fine-tuning and show that the dialogue-tailored
Sentence Ordering task performs best. To locate and exploit discourse
information in PLMs, we propose an unsupervised and a semi-supervised method.
Our proposals achieve encouraging results on the STAC corpus, with F1 scores of
57.2 and 59.3 for unsupervised and semi-supervised methods, respectively. When
restricted to projective trees, our scores improved to 63.3 and 68.1.

本文研究基于 Pre-trained Language Models (PLMs) 的注意力矩阵构建对话结构的方法，提出了一种无监督和半监督的方法来定位和利用 PLMs 中的语篇信息，并在 STAC 语料库上获得了令人鼓舞的结果。

基于预训练和微调语言模型的对话中的话语结构提取

Discourse Structure Extraction from Pre-Trained and Fine-Tuned Language Models in Dialogues

Earnings conference calls are significant information events for volatility
forecasting, which is essential for financial risk management and asset
pricing. Although some recent volatility forecasting models have utilized the
textual content of conference calls, the dialogue structures of conference
calls and company relationships are almost ignored in extant literature. To
bridge this gap, we propose a new model called Temporal Virtual Graph Neural
Network (TVGNN) for volatility forecasting by jointly modeling conference call
dialogues and company networks. Our model differs from existing models in
several important ways. First, we propose to exploit more dialogue structures
by encoding position, utterance, speaker role, and Q\&A segments. Second, we
propose to encode the market states for volatility forecasting by extending the
Gated Recurrent Units (GRU). Third, we propose a new method for constructing
temporal company networks in which the messages can only flow from temporally
preceding to successive nodes, and extend the Graph Attention Networks (GAT)
for modeling company relationships. We collect conference call transcripts of
S\&P500 companies from 2008 to 2019, and construct a dataset of conference call
dialogues with additional information on dialogue structures and company
networks. Empirical results on our dataset demonstrate the superiority of our
model over competitive baselines for volatility forecasting. We also conduct
supplementary analyses to examine the effectiveness of our model's key
components and interpretability.

该研究提出了一种新模型 ——TVGNN，通过共同建模会议通话对话和公司网络，对波动率进行预测，取得了较好的效果。