XunEndong（荀恩东）

Professor XunEndong

Speaker introduction

Professor Xun Endong, a Ph.D. supervisor, serves as the director of the Advanced Innovation Center for Language Resources at Beijing Language and Culture University, and the dean of the Institute of Language Intelligence. He received his Ph.D. from Harbin Institute of Technology. His research fields include natural language processing and language education technology, with a focus on Chinese semantic analysis and international intelligent Chinese education. He has published five academic monographs and over 70 papers. He presided over the development of the BCC Corpus, which is the largest online corpus both domestically and internationally. He also led the development of the International Intelligent Chinese Teaching Platform. He proposed the Yihe Graph theory and its construction method for Chinese semantic analysis.

Abstract

Large language models, when serving as natural language interfaces and applied to solve specific tasks in real-world scenarios, encounter challenges essentially stemming from their inability to generate, with high accuracy, interface structures that bridge the gap between natural language and application tasks. The target structure to be generated in this context corresponds to the pragmatic aspects of natural language. This report delves into various natural language objects and their corresponding linguistic structures from different perspectives, elucidating the relationships among them.

It distinguishes between explicit and implicit manifestations of linguistic structures and discusses, from a knowledge-based perspective, the data requirements, models, and fusion methods necessary for computing linguistic structures across different aspects. By exploring the complexities and nuances within natural language, the report aims to provide insights into how large language models can be enhanced to more effectively interact with and adapt to real-world applications, overcoming the challenges posed by the generation of precise interface structures for natural language to application tasks.

荀恩东教授

主讲人介绍

荀恩东 教授，博士生导师，北京语言大学语言资源高精尖创新中心主任，语言智能研究院院长，博士毕业于哈尔滨工业大学。研究领域为自然语言处理和语言教育技术，重点研究方向为中文语义分析和国际中文智慧教育。发表5部学术专著和70余篇论文；主持开发的BCC语料库，是国内外最大的在线语料库；主持开发国际中文智慧教学平台；提出了用于汉语语义分析的意合图理论和构建方法。

摘要

大语言模型作为自然语言接口，落地应用解决场景任务时遇到挑战，本质是大语言模型无法高正确率地从自然语言生成与应用任务对接的接口结构，这时目标要生成的结构即为自然语言语用侧面的结构。本报告论述不同自然语言对象、不同侧面的语言结构，说明之间的关系。讨论显示和隐式两种样态的语言结构，从知识角度讨论不同侧面的语言结构计算所需要的数据，模型和融合方法。