一种端到端的藏文La格浅层语义分析  

An End-to-End Shallow Semantic Analysis of Tibetan La Case

作　　者：班玛宝 色差甲 才让加[1,2,3,4,5] 张瑞 柔特 BAN Mabao;SE Chajia;CAI Rangjia;ZHANG Rui;ROU Te(College of Computer Science and Technology,Qinghai Normal University,Qinghai,Xi'ning 810016,China;The State Key Laboratory of Tibetan Intelligent Information Processing and Application,Qinghai,Xi'ning 810008,China;Tibetan Information Processing and Machine Translation Key Laboratory of Qinghai Province,Qinghai,Xi'ning 810008,China;Key Laboratory of Tibetan Information Processing,Ministry of Education,Qinghai,Xining 810008,China;Tibetan Information Processing Engineering Technology and Research Centerof Qinghai Province,Qinghai,Xi'ning 810008,China)

机构地区：[1]青海师范大学计算机学院,青海西宁810016 [2]省部共建藏语智能信息处理及应用国家重点实验室,青海西宁810008 [3]青海省藏文信息处理与机器翻译重点实验室,青海西宁810008 [4]藏文信息处理教育部重点实验室,青海西宁810008 [5]青海省藏文信息处理工程技术研究中心,青海西宁810008

出　　处：《中文信息学报》2023年第2期62-70,共9页Journal of Chinese Information Processing

基　　金：国家自然科学基金(61866032,619660316,62206146);青海省重点研发项目(2022-GX-104)。

摘　　要：该文通过详细分析La格的用法及特征,在研制La格(■)浅层语义标记规范的基础上,提出一种端到端的长短时记忆神经网络藏文La格浅层语义分析方法,该方法首先借鉴LSTM的设计思路,通过在LSTM的垂直方向上装置一个新颖的“门控高速连接”机制(Gated high-speed connection mechanism,GM),学习了输入句子的时序语义特征。GM包含对单元内部输入和输出的线性连接,使信息可以通畅地在不同层之间传播;然后使用Softmax计算每一时刻语义标签的局部归一化分布,以供输出层进行约束解码;最后使用维特比算法进行解码时通过强制执行该文设定的BIO和La格浅层语义标注约束,规范了输出语义标签之间的结构关系。虽然这个模型比较简单,不需要输入任何额外特征,但取得了理想的实验结果,在测试集上的藏文La格浅层语义分析准确率达到90.59%。Based on the detailed analysis of the usage and characteristics of La case()and the designed shallow semantic annotation standard of La,this paper proposes an end-to-end LSTM based shallow semantic analysis method for Tibetan La lattice.Firstly,this method proposes a novel“gated high-speed connection”mechanism(GM)in the vertical direction of LSTM to allow the input information spread into between different layers.Then,softmax is used to calculate the local normalized distribution of semantic tags at each time.Finally,when decoding with Viterbi algorithm,the structural relationship between output semantic tags is realized by enforcing the constraints of BIO and La case set in this paper.Experimental results demonstate an accuracy of 90.59%for shallow semantic analysis of Tibetan La case.

关 键 词：自然语言处理 La格 浅层语义分析 门控高速连接 约束解码 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]