Target Phrase Classifier for Machine Translation Filtering

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Machine translation systems often introduce objectionable language into translations, either by incorrect mapping of source words, inconsistencies in training data, or human translators softening explicit content, leading to unintended offense to the original speaker.

Innovation Solution

A classifier, potentially a support vector machine, is employed to identify and remove or replace objectionable words and phrases from machine-generated translations, operating within the translation system or as a standalone service, trained using target word lists and user reports to prevent the introduction of explicit content.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If machine translation systems use translation models that map source words to destination words, then translation functionality is provided, but objectionable language may be incorrectly introduced into translations

Engineering Contradiction:

Improvetranslation functionalityVSAvoidobjectionable language introduction

Core Design Contradiction:

Adaptability or versatilityVSObject-affected harmful factors

Solution Approach 1:

A classifier is introduced as an intermediary component between the translation model and the output. The classifier receives translated text from the translation model and automatically identifies and flags objectionable language, preventing it from being presented to users. This mediator resolves the contradiction by maintaining translation functionality while filtering out harmful content through automated classification.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system implements feedback mechanisms where user reports of offensive translations are collected and used to retrain the classifier. This continuous feedback loop allows the system to learn from actual problematic cases and improve its ability to detect objectionable language, thereby preventing harmful output while preserving translation capabilities.

Inventive Principle:

Principle #23Feedback

2Quantity of substance

If training data is collected automatically from mined translations, then training data quantity is increased, but inconsistency in explicit term usage is introduced

Engineering Contradiction:

Improvetraining data quantityVSAvoidconsistency in explicit term usage

Core Design Contradiction:

Quantity of substanceVSStability of the object's composition

Solution Approach 1:

The classifier undergoes parameter changes through retraining with updated data. When inconsistencies or problematic mappings are detected in the training data, the classifier is retrained with corrected examples, changing its internal parameters to better distinguish appropriate from objectionable translations. This allows the system to maintain large training data volumes while improving consistency through iterative parameter optimization.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The system discards problematic or inconsistent training data examples that introduce objectionable language patterns, and recovers or adds corrected examples. By selectively removing harmful data and supplementing with proper examples, the training dataset maintains its quantity while improving quality and consistency of explicit term usage.

Inventive Principle:

Principle #34Discarding and recovering

3Object-affected harmful factors

If human translators soften explicit words or phrases during translation, then translations become more acceptable, but explicit content from source material is lost

Engineering Contradiction:

Improvetranslation acceptabilityVSAvoidexplicit content preservation

Core Design Contradiction:

Object-affected harmful factorsVSLoss of information

Solution Approach 1:

The classifier performs self-service by automatically detecting and flagging objectionable language without requiring manual intervention from translators. This automated system allows translators to work efficiently while the classifier independently ensures that explicit content is appropriately handled, preventing both unnecessary softening and harmful output.

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The manual process of translator judgment is supplemented and partially replaced by an automated classifier system. Instead of relying solely on human translators to judge when to soften language, the mechanical classifier system objectively identifies objectionable content, providing consistent automated decision-making that preserves explicit content when appropriate while filtering harmful material.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS10460038B2Target phrase classifier

Publication Date: 2019.10.29 META PLATFORMS INC

AI summary

Exemplary embodiments relate to detecting, removing, and/or replacing objectionable words and phrases in a machine-generated translation. A classifier identifies translations containing target words or phrases. The classifier may be applied to the output translation to remove target words and phrases from the translation, or to prevent target words and phrases from being automatically presented. Further, the classifier may be applied to a translation model to prevent the target words and phrases from appearing in the output translation. Still further, the classifier may be applied to training data so that the translation model is not trained using the target words of phrases. The classifier may remove target words or phrases only when the target words or phrases appear in the output translation but not the source language input data. The classifier may be provided as a standalone service, or may be employed in the context of a machine translation system.