Sequence-Based Lock Unlock Classifier for Software Localization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine translation systems for software localization face challenges such as functional breaks and over-translation due to the inability to accurately identify and lock specific strings or substrings that should not be translated, leading to cumbersome, costly, and error-prone manual processes or incomplete deterministic approaches.
Innovation Solution
A system that tokenizes software input strings, applies a context-dependent lock/unlock classifier to identify tokens that should be locked for translation, converts the string with locking indicators, and then submits it to a machine translation system, ensuring locked tokens remain unchanged in the translation result.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If manual processes are used to identify strings to lock from translation, then translation accuracy can be maintained, but the process becomes very cumbersome, time consuming, and costly
Solution Approach 1:
The patent replaces manual mechanical processes with an automated machine learning system. A classifier trained on software resources automatically identifies strings that should be locked from translation, substituting human manual identification with an automated computational system that processes software resources efficiently without manual intervention
Solution Approach 2:
The system enables self-service by allowing the machine learning classifier to autonomously identify and lock strings without requiring manual human intervention. The classifier learns from training data and independently determines which strings should remain untranslated, making the system self-sufficient for the localization process
2Manufacturing precision
If deterministic approach is used to lock strings, then some translation accuracy is maintained, but it is very difficult to list all issues exhaustively and cannot cover all possible cases
Solution Approach 1:
The patent transitions from static deterministic rules to a dynamic machine learning system. The classifier adapts its behavior based on learned patterns from training data, allowing it to dynamically identify strings that should be locked across diverse software contexts and domains, covering cases that predetermined rules cannot anticipate
Solution Approach 2:
The system changes the approach from fixed deterministic parameters to learned probabilistic parameters. The machine learning model learns optimal thresholds and patterns from training data, adjusting its decision parameters based on the specific characteristics of different software resources, thereby improving both accuracy and adaptability
3Productivity
If entire software resource is submitted for translation without locking, then translation completeness is achieved, but over-translation occurs where certain parts that should have been locked are actually translated
Solution Approach 1:
The patent applies preliminary action by training the machine learning classifier beforehand on software resources with known translation requirements. This pre-training enables the system to proactively identify and lock strings that should not be translated before the actual translation process begins, preventing over-translation errors while maintaining completeness
Data Source
AI summary
A software input string is received and tokenized into a sequence of tokens. The sequence of tokens is applied to a trained sequence-dependent lock/unlock classifier so that each of the tokens is classified as a token that should be locked, or remain unlocked, for subsequent translation. The software input string is converted to a converted string, in which the locked tokens are identified and the converted string is submitted for machine translation. A machine translation result is received and converted so that the locked tokens are replaced in the machine translation result, to obtain a translated software string.