Multilingual Chatlog Training via English Intermediary Translation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current multi-lingual chatbot systems face challenges in handling diverse languages due to the scarcity of parallel data, especially for less popular languages, which limits their effectiveness in real-world conversational scenarios.

Innovation Solution

A method that receives conversational data in multiple languages, assigns intent and language labels, and trains a machine learning model using multi-lingual and multi-intent conversational data pairing to generate a multi-lingual chatlog, enabling chatbots to handle under-resourced languages without requiring large parallel data sets.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If parallel data is collected for multiple languages, then chatbot effectiveness is improved, but data availability deteriorates for less popular languages

Engineering Contradiction:
Improvechatbot effectivenessVSAvoiddata availability
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

The patent uses English-language data as an intermediary to train chatbots for less popular languages. The system collects English conversational data, translates it to target languages, and uses this translated data to train multilingual chatbot models. This intermediary approach allows effective training without requiring large amounts of native parallel data for each language pair.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system changes the language parameter of training data by translating from English to multiple target languages. Instead of collecting native data in each language, the system transforms English data into various languages through translation, thereby creating training datasets for languages with limited available data while maintaining data quality and diversity.

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If multi-lingual data is collected, then language coverage is improved, but data quality deteriorates due to scarcity

Engineering Contradiction:
Improvelanguage coverageVSAvoiddata quality
Core Design Contradiction:
Adaptability or versatilityVSManufacturing precision

Solution Approach 1:

The patent creates a universal training approach where English-language data serves multiple functions: it is used to train chatbots for many different target languages through translation. This single source of high-quality English data is transformed into multiple language-specific training sets, ensuring consistent data quality across all language pairs while achieving broad language coverage.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system creates copies of English conversational data in multiple target languages through translation. Rather than collecting original native data for each language, the system generates translated copies that preserve the conversational patterns and quality characteristics of the source English data, thereby maintaining data quality while expanding language coverage.

Inventive Principle:
Principle #26Copying

Data Source

PatentUS11853712B2Conversational AI with multi-lingual human chatlogs
Publication Date: 2023.12.26 INTERNATIONAL BUSINESS MACHINE CORPORATION
  • US11853712B2 patent drawing
  • US11853712B2 patent drawing
  • US11853712B2 patent drawing

AI summary

A method, computer system, and computer program product for multi-lingual chatlog training are provided. The embodiment may include receiving, by a processor, a plurality of data related to conversational data in multiple languages. The embodiment may also include assigning an intent label to each conversational data. The embodiment may further include assigning a language label to each conversational data. The embodiment may also include paring the plurality of the data related to the conversational data according to the intent label and the language label. The embodiment may further include training a machine learning model using a multi-lingual and multi-intent conversational data pairing. The embodiment may also include training the machine learning model using a single language and multi-intent conversational data paring.