AI for Translation in Criminal Investigations

Summary:

A European law enforcement agency utilizes AI to translate large amounts of evidence.

Client: 

The European Union Agency for Law Enforcement Cooperation (Europol) is the law enforcement body of the EU, tasked with supporting national law enforcement authorities across EU member states. Its mission is to enhance safety in Europe by providing assistance in crime prevention, investigation, and operational coordination among member countries.

Problem Statement: 

The law enforcement agency was tasked with handling large volumes of evidence written in low-resource foreign languages, which lacked sufficient digital tools and data for efficient translation. The agency faced challenges due to the strict timelines for investigations and evidence admissibility, along with a limited number of linguists available to translate the material. These constraints jeopardized their ability to prosecute criminals effectively. To overcome this, they needed a flexible and efficient solution to translate vast amounts of content in low-resource languages within tight deadlines.

Results: 

• Achieved translation speeds of up to 150,000 words per minute, benchmarked across four different languages. 
• Increased translation throughput during peak workloads, ensuring that no end users faced delays and no mission was left unsupported due to resource limitations.
• Ability to process large volumes of evidence much faster, ensuring that critical evidence was translated and reviewed within the required timeframes. 
• Achieved a 30X increase in character throughput, along with a 5X increase in model size, improving both latency and translation quality. 

AI Solution Overview:

When a European law enforcement agency faced the challenge of translating vast amounts of content in low-resource languages under tight deadlines, they turned to LILT’s generative AI platform. Powered by large language models, and using NVIDIA GPUs and NVIDIA NeMo, LILT’s platform enabled faster, large-scale translation of time-sensitive material. This solution helped prevent resource limitations from hindering the agency’s ability to combat crime.

LILT’s platform offers a versatile workflow that empowers non-linguists to autonomously triage documents with machine translation (MT), identifying high-value content and sending bulk translations via API. Professional linguists then focus on translating key evidentiary documents using LILT’s predictive CAT tool. This workflow has been deployed in major operations, contributing to hundreds of criminal arrests and the confiscation of illegal drugs and weapons.

By leveraging NVIDIA A100 Tensor Core GPUs for accelerated model training and NVIDIA T4 Tensor Core GPUs for model inference, LILT achieved a 30X increase in character throughput, along with a 5X increase in model size, improving both latency and translation quality. LILT’s adaptive machine learning models continue to evolve as linguists provide feedback, enabling the platform to keep up with changing colloquial language and diverse content sources such as social media.

NVIDIA NeMo, an end-to-end cloud-native framework, further supports the scalability and efficiency of LILT’s generative AI models, offering an enterprise-ready, cost-effective solution for rapid AI adoption.

References: 

1. AI-Powered Language Translation in Criminal Investigations. https://www.nvidia.com/en-us/case-studies/lilt/
2. Law Enforcement Agency Leverages LILT in Criminal Investigations. https://lilt.com/customer-stories/euro-law-enforcement

Industry: Legal Services

Vendor:  LILT, NVIDIA and AWS

Client: The European Union Agency for Law Enforcement Cooperation (Europol)

Publication Date: 2024