Machine Translation Style Vectors for Consistent Paragraph Tone
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine translation systems often produce translated sentences with unnatural tones due to inconsistent translation styles within a paragraph.
Innovation Solution
An electronic apparatus and method that identifies style information, including degree of honorific, formality, colloquialism, and persuasiveness, to generate style vectors for each sentence and apply consistent translation styles across sentences, using a processor to maintain the tone and style of the original text.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine translation is performed sentence by sentence using conventional methods, then translation speed is improved, but translation style consistency deteriorates
Solution Approach 1:
The system performs preliminary action by extracting style information from the source text before translation begins. Style vectors are generated and stored in advance, representing the desired translation style. During translation, these pre-extracted style vectors guide the translation process, ensuring consistency without slowing down sentence-by-sentence translation.
Solution Approach 2:
The system implements feedback by continuously comparing the translation style of each sentence against the reference style vector extracted from the source text. The style vector acts as a feedback mechanism that guides adjustments in translation style, ensuring that translated sentences maintain consistency with the original text's tone and style while allowing rapid translation processing.
2Stability of the object's composition
If translation style is standardized across all sentences, then translation style consistency is improved, but natural tone expression deteriorates
Solution Approach 1:
The system applies local quality by extracting style information specifically from sentences that best represent the source text's style (e.g., sentences with complete stylistic features). These localized style extractions are then applied selectively to guide translations, allowing different parts of the text to maintain their specific stylistic characteristics rather than forcing uniformity across all sentences.
Solution Approach 2:
The system implements dynamics by using style vectors that can adapt to different sentences. Rather than applying a rigid standardized style, the style vector dynamically guides each translation based on the source sentence's characteristics, allowing the translation style to naturally vary while maintaining overall consistency with the source text's tone.
3Adaptability or versatility
If style information is extracted from each individual sentence, then natural tone is improved, but translation style consistency deteriorates
Solution Approach 1:
The system merges style information by combining style vectors from multiple source sentences to create a comprehensive reference style vector. This aggregated style representation captures the overall stylistic characteristics of the source text more accurately than any single sentence could provide, while still allowing natural tone variation in translations.
Solution Approach 2:
The style vector serves multiple functions: it guides translation style consistency across all sentences, adapts to different sentence types, and maintains natural tone expression. This universal style representation works across various contexts and sentence structures, providing consistent guidance while allowing natural variation.
Data Source
AI summary
An electronic apparatus includes a memory configured to store instructions, and a processor configured to execute the instructions to receive a text of a first language, generate, based on the text of the first language, style information indicating a translation style to be applied to the text of the first language and machine-translate the text of the first language into text of a second language based on the generated style information.