Multilingual Chatbot Parallel Intent Translation for Faster Responses

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing chatbot systems struggle to accurately and efficiently translate inputs from users speaking languages other than the default language, leading to prolonged response times.

Innovation Solution

A multilingual chatbot system utilizing a dual-track approach with a translational track for literal translation and an intelligence track leveraging AI and ML to process inputs in parallel, reducing response times by internally translating and analyzing user intents.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a single translation track is used to translate requests in non-default languages, then translation accuracy can be maintained, but response time increases significantly

Engineering Contradiction:
Improvetranslation accuracyVSAvoidresponse time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The translation process is segmented into two independent tracks: a translational track for accurate translation and an intelligence track for rapid processing. Each track handles different aspects of translation independently, allowing the system to maintain accuracy while reducing overall response time by processing both tracks in parallel rather than sequentially

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The intelligence track performs a simplified form of translation that may not be fully accurate but is sufficient for understanding basic user intent. This partial translation action occurs in parallel with the complete translation in the translational track, allowing the system to respond quickly to simple queries while maintaining accuracy for complex translations

Inventive Principle:
Principle #16Partial or excessive action

2Adaptability or versatility

If a third-party translation application is used, then comprehensive language support is achieved, but response time increases by one or more orders of magnitude

Engineering Contradiction:
Improvelanguage support coverageVSAvoidresponse time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system introduces an intermediary intelligence track that processes translations internally using pre-trained models. This intermediary layer handles common translation scenarios quickly without needing to externalize to third-party applications, reserving third-party translation services only for complex cases that require comprehensive language support

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

Translation models are pre-trained and cached in advance for commonly encountered languages and phrases. When a translation request arrives, the system first checks if a pre-computed translation exists in the cache, returning it immediately if found, thereby avoiding the need to call external translation services for routine queries

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS12579370B2Multilingual chatbot
Publication Date: 2026.03.17 BANK OF AMERICA CORP
  • US12579370B2 patent drawing
  • US12579370B2 patent drawing
  • US12579370B2 patent drawing

AI summary

A method for reducing response time of an utterance received in a multilingual chatbot system may be provided. The utterance may be received in a language other than a default language with respect to the multilingual chatbot system. The method may include receiving a request utterance from a remote user device and determining that the request utterance is not in the default language. In response to the determining, the method may include transmitting the request utterance, in parallel, to both a translational track for translating the request utterance and an intelligence track for translating the request utterance. The method may include computing a first intent of the translated request utterance at the translational track. The method may include computing a second intent based on a ML request utterance. The method may include determining that the first intent matches the second intent and following the determining, generating a response utterance.