Neural Machine Translation Method Based on Emotional Semantics Enhanced Encoding and Decoding
To address the problem that current neural machine translation models rely solely on parallel corpus training and cannot fully tap into deep linguistic knowledge,we propose a neural machine translation method based on emotional semantic enhancement coding and decoding to improve the model's ability to understand deep linguistic information by introducing additional emotional semantics.Firstly,word2vec technology is used to obtain word embeddings for all words in the corpus,which are then input into a fusion model for training.This fusion model combines mechanisms based on GRU and document embedding to obtain emotional semantic representations at the word and document levels.Secondly,in the emotional fusion stage,a weighted formula is used to integrate the emotional semantic repre-sentations at the word and document levels organically,forming a more comprehensive emotional semantic representation.Finally,this representation is added bitwise with contextual semantic representations to fully introduce emotional information,and passed as input to the encoder and decoder of the machine translation model.Experiments on multiple benchmark datasets show that compared to traditional Transformer models,the proposed method significantly improves performance on the IWSLT dataset,with BLEU values increasing by 1.3 to 1.62.It also achieves good performance on the WMT dataset,confirming the effectiveness of integrating emotional semantics in machine translation.
emotional semanticsenhanced encoding and decodingneural machine translationTransformerparallel corpus
国家科技期刊平台
NSTL
万方数据
