基于注意力机制的手语语序转换方法

A word order conversion method based attention mechanism for sign language

文章考虑听障人士与健全人士在语法和语言结构上的差异,设计一种基于注意力机制的手语语序转换器,实现手语语序到书面表达的转换。语序转换器在编码阶段使用双向长短期记忆网络(long short-term memory,LSTM)提取手语语序特征,解码阶段使用一维卷积提取编码器隐藏状态的特征,并利用注意力机制避免了长距离的依赖问题,从而得到书面表达。实验结果表明,语序转换器准确率最高为92.64%。

Considering the differences in grammar and language structure between hearing-impaired and able-bodied people,this paper designs a sign language order converter based on attention mechanism to realize the conversion of sign language order to written expression.In the encoding stage,the word order converter uses bidirectional long short-term memory(LSTM)to extract the features of the sign language order,and in the decoding stage,it uses one-dimensional convolution to extract the features of the hidden state of the encoder.In addition,the attention mechanism is used to avoid the long-distance dependency problem,so as to obtain the written expression.The results show that the highest accuracy of the word order converter is 92.64%.