Translation and Transcription for Investigative Data

Overview

Development stage

DeployedSource: c) Deployed – The use case is being actively authorized or utilized to support the functions or mission of an agency.

High-impact designation

Presumed high-impact, determined notSource: b) Presumed high-impact, but determined not high impact

Impact justification

This use case provides more efficient data processing for HSI personnel. The AI output may be used to produce investigative insights in the form of data, information leads or connections that HSI personnel can use to inform investigations, but the output itself is data preparation and organization so HSI personnel can produce those leads when combining the AI output with the personnel’s expertise and other relevant investigative data and information. Personnel may use these insights for law enforcement purposes in ongoing investigations with existing targets to assist in activities such as producing risk assessments about individuals or identifying criminal suspects; however, all insights are reviewed and validated by both personnel and supervisors before being included in any official case management system, and do not serve as a principal basis for law enforcement action or decision. Any enforcement decisions related to these insights are outcomes of the full Federal Investigation Process involving verifying any insights as evidence (including validating AI-translated material by a certified interpreter), presentation to a U.S. Attorney’s Office and potentially a District Court judge, decision to prosecute, judicial review, and trial and sentencing.

Start date

2024-02-01 (day precision)Source: 2024-02-01T00:00:00

Withheld from public reporting

Not reported

Mission

Topic area

Law Enforcement

Operational functions

Translation and transcription Derived by keyword rules; see methodology

Problem the AI is intended to solve

This use case intends to solve the problem of the time-consuming process of translating and transcribing data for investigative purposes.

Expected benefits

HSI investigators often encounter data from various sources, including legal and administrative processes, enforcement actions, and open-source materials, in languages other than English. To unlock the value of this data, it must be translated into English before further analysis can be conducted. The Translation and Transcription Service leverages neural machine translation (NMT) models for text translation and automatic speech recognition (ASR) and deep neural network (DNN) models with normalization for voice-to-text transcription. This innovative approach enables users to quickly triage large datasets and identify key information relevant to investigations. Any data deemed critical for court proceedings is then submitted to certified human translators for final review, ensuring that government resources are allocated efficiently and only used for necessary translations and transcriptions.

System outputs

The Translation and Transcription Service leverages neural machine translation (NMT) models for text translation and automatic speech recognition (ASR) and deep neural network (DNN) models with normalization for voice-to-text transcription.

Technology

AI classification

Natural language processingSource: Natural Language Processing: AI that processes, interprets, and shares information in human language.

System name(s)

Repository for Analytics in a Virtualized Environment,

Custom-developed code

Yes

Public source code

Not reported

Sourcing

How it was built

Contract and in-houseSource: c) Developed with both contracting and in-house resources

Vendor (as reported)

Booz Allen

Vendors (standardized)

BABooz Allen Hamilton

A vendor is the supplier named by the agency. It does not identify the underlying model or AI technology.

Data and privacy

Involves PII

Yes

Privacy Impact Assessment

www.dhs.gov ↗

Authorization to Operate

Yes

Demographic features

Not reported

Training and evaluation data

AI models within the use case are not trained on agency data. Open-source models (i) Whisper is trained on 680K hours of multilingual and multitask supervised data collected from the web, (ii) No Language Left Behind (NLLB) is trained on a combination of publicly available datasets (additional information available in Section 5 of Meta’s NLLB whitepaper: https://research.facebook.com/file/585831413174038/No-Language-Left-Behind--Scaling-Human-Centered-Machine-Translation.pdf).

Federal Data Catalog

Not reported

Governance

0 of 8 minimum-practice questions answered. A blank answer means the agency reported nothing; it does not mean the practice is absent.

Potential impacts and how they were identified

Not reported