Neural Translation Decoder for Past and Future Source Content

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing neural machine translation systems face challenges in effectively modeling untranslated and translated source content during the training process, leading to difficulties in decoder performance and translation accuracy.

Innovation Solution

The proposed method introduces additional hidden layers in the decoder to model past and future translation vectors, allowing for the separation of untranslated and translated source content, thereby simplifying the training process and improving translation quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If traditional neural machine translation systems are used, then the system structure is simple, but the decoder training complexity increases and translation accuracy decreases

Engineering Contradiction:

Improvetranslation accuracyVSAvoiddecoder training complexity

Core Design Contradiction:

Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent segments the source content into untranslated and translated parts, and separately models their translation vectors. This segmentation allows the decoder to focus on specific translation tasks, reducing training complexity while improving translation accuracy through targeted vector modeling.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent extracts and separately models the translation vectors for untranslated and translated source content. By taking out these specific vector components and handling them independently, the system simplifies the overall decoder training process while enhancing translation precision.

Inventive Principle:

Principle #2Taking out (Extraction)

2Adaptability or versatility

If traditional neural machine translation systems are used, then the model structure is straightforward, but the ability to handle untranslated and translated content separately deteriorates

Engineering Contradiction:

Improveability to handle untranslated and translated contentVSAvoidmodel structure complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent divides the translation modeling into separate components for untranslated and translated source content. This segmentation enables the system to adaptively handle different content types with specialized vector representations, improving versatility without requiring a completely new model architecture.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent creates a multi-functional translation model that can simultaneously handle untranslated and translated content through separate vector representations. This universal approach allows a single model structure to perform multiple translation tasks effectively.

Inventive Principle:

Principle #6Universality (Multi-functionality)

Data Source

PatentUS12493756B2Translation method, target information determining method, related apparatus, and storage medium

Publication Date: 2025.12.09 TENCENT TECHNOLOGY (SHENZHEN) CO LTD

AI summary

A translation method is provided, including: encoding to-be-processed text information to obtain a source vector representation sequence, the to-be-processed text information belonging to a first language; obtaining a source context vector corresponding to a first instance according to the source vector representation sequence, the source context vector indicating to-be-processed source content in the to-be-processed text information at the first instance; determining a translation vector according to the source vector representation sequence and the source context vector; and decoding the translation vector and the source context vector, to obtain target information of the first instance, the target information belonging to a second language.