Text translation method, device and equipment and computer readable storage medium

Abstract

The embodiment of the invention provides a text translation method, device and equipment and a computer readable storage medium, and relates to the technical field of artificial intelligence. The method comprises the steps of calling an encoder to sequentially perform mapping processing on to-be-translated words in a to-be-translated text to obtain implicit state representation of the to-be-translated text, and performing prediction according to the implicit state representation of the to-be-translated text to obtain a candidate word set; in the process of calling a decoder to decode and translate the hidden state representation, for each to-be-translated word, obtaining an intermediate vector representation corresponding to a translated word before the to-be-translated word; predicting future word vector representation of the to-be-translated word according to the intermediate vector representation; and calling a decoder to predict a translation corresponding to the to-be-translated text in the candidate word set based on the word vector representation and the future word vector representation of each to-be-translated word in sequence. According to the method and the device, the predicted global information and local future information can be fully utilized to assist the translation process, so the text translation efficiency is effectively improved.

Description

technical field [0001] The embodiment of the present application relates to the technical field of the Internet, and relates to but not limited to a text translation method, device, equipment and computer-readable storage medium. Background technique [0002] In the traditional machine translation model, for a given source input (that is, the text to be translated), the machine translation model usually generates words in the machine translation one by one from left to right. This process is called "automatic translation". regression modelling. In autoregressive modeling, each decoding step can only see segments of the translation that have been generated in the past (i.e., prefixes of the full translation), but not segments of the translation that have not yet been generated (i.e., the suffix of the full translation). Obviously, this makes the information that the machine translation model can obtain during decoding limited by the translated local context. [0003] In rel...

AI Technical Summary

Problems solved by technology

[0004] However, in related technologies, whether it is for the utilization of the global context or the utilization of the local future context, there are problems of low efficiency, and the prediction results are not used to assist the translation process, and the model improvement is limited.

Images