Enabling Equal Language Access in Government with Assistive AI/ML

The Issue: Equal Language Access for Residents

There are several Federal and State laws, rules, regulations and policies that require Government agencies to provide equal language access to Limited English Proficient (LEP) residents they serve.?Two important guidelines are:

Civil Rights Act of 1964 (Federal), National Origin?

USDOL Nondiscrimination Regulations (Federal), Limited English Proficiency?

In Illinois, Government agencies are required to designate a Language Access Coordinator for their agency. Each agency is required to submit a Language Access Plan to address the primary non-English languages spoken in Illinois: Polish; Arabic; Hindi; Spanish; Chinese; Tagalog.

What is Language Access??

Language Access generally means providing equal and meaningful access to all benefits, services and programs in the LEP individual’s preferred or “native” language.?

The law requires “effective communication” with Limited English Proficiency (LEP) individuals in their primary language. This includes written communication.

There are hundreds of different languages spoken in the United States of America, making it one of the most linguistically diverse countries in the world.

The question is not if your Government agency needs to provide equitable language access, it's only a question of how your agency will provide it.

Traditional Document Language Translation

The traditional method to translate documents into various languages is to employ human experts fluent in the required languages to do the translation work. There are a number of very good language translation companies that can be contracted to do document translation work if you lack your own resources on staff. These companies typically charge fees on a per page of translation basis. The fees vary based on the number of documents, the number of pages, the formatting required, and the number of languages required.?The costs for a fee based structure can add up fast. Services of this type generally require days to weeks of turnaround time to complete.?

Automated Document Translation

Remember the original Google Translate widget launched back in 2006? At that time it was a breakthrough technology that basically did word-for-word language translation for multiple languages. It was easy to use but it wasn’t perfect. The ways in which individual words are used together to form sentences varies across languages.? With the original Google Translate technology, sometimes the translations didn’t make sense. These translations still required quality assurance by a human translator.

Modern Document Language Translation with Google Language Translation Artificial Intelligence (AI)

Over the past years, the researchers at the Google Brain Team have made incredible strides in Artificial Intelligence (AI) and Machine Learning (ML) by modeling how the human brain works using neural networks. Since 2016, Google Neural Machine Translation system (GNMT), utilizes state-of-the-art training techniques to achieve the largest improvements to date for machine translation quality.?

• A Neural Network for Machine Translation, at Production Scale

How does one know that it's working with higher translation fluency?

A BLEU score is a quality metric for Machine Translation systems that attempts to measure the correspondence between a machine translation output and a human translation. The central idea behind BLEU is that the closer a machine translation is to a professional human translation, the better it is.?

• Google Neural Network AI based translations deliver Bleu Scores from 40 - 53.????
• Google ML Translate Research: link

The research done by the Google Brain Team led to the development of the Google Cloud Language Translation API that supports translations of text and documents (HTML, DOCx, PPTx, XLSx, PDF) to over 100 different languages.??

Today you now have the ability to translate documents into different languages in multiple ways.

• Real-time (synchronous processing), online translation, for a single file. For example, translating a business document such as an enrollment form, or benefits flyer.? Online translation provides flexibility for those who have random needs on-demand.
• Batch processing, for bulk files to be translated into one or more languages. For example, translating multiple handbooks, policy manuals and forms.

Taking it a step further

Google Cloud AutoML Translation enables you to build custom translation models trained on niche glossaries of terms and lingo specific to a vocabulary.? For example legal or medical jargon.

AutoML enables developers with limited machine learning expertise to train high-quality models specific to their business needs.?Now you can quickly train custom machine learning translation models for language specific to your needs with AutoML Translation.

Example Use Cases in Government

State of Hawaii

The State of Hawaii is known as the melting pot of the world, and many residents are non-English speakers. The Government wants to make information available to people in the native languages they speak.? Manual translation of legislative documents would have cost millions of dollars.? Instead, the Google Language Translation API enables them to translate documents with speed and make them available to public.

Listen to their story: link

State of Illinois

Illinois is helping State Agencies meet Language Access requirements by leveraging Google Cloud’s Language Translation AI. An internal online one-stop portal was created for agencies to use for real-time translation of documents.? This portal can rapidly translate documents written in English into the most common alternate languages required in Illinois. Business users can simply upload the documents they want translated, then pick the target languages to translate into. The translated documents can then be saved to the location of their choosing and published where required.

The Business Impacts & ROI

Customers have seen incredible results with the Google Cloud Language Translation API.?

• Up to 80% reduction in the manual translation efforts.
• Overall translation accuracy across languages at ~92% (which is equivalent to human proficiency).?
• Significant cost savings on translation services compared to the existing approach.? From dollars per page to pennies per page.
• Significant time savings on turnaround time and effort. From 2 weeks to 2 minutes.

Human-in-the-Loop

Advances in Machine Learning & Artificial intelligence have made it possible for technology to perform tasks at incredible speed and scale.? However, the technology is still not a replacement for humans.? Rather, AI/ML technologies are “assistive” to humans, similar to the ways that glasses help with vision, and hearing aids help with hearing.?

The best outcomes when employing AI/ML technologies are realized when a human is in the loop during the development and implementation process for quality assurance.

How can I determine if Google Cloud Language Translation AI/ML right for me?

The first step in determining if any technology can be helpful is by finding out if there is a problem worth solving. Does anyone care? When it comes to Equal Language Access in Government, find out how important is it to policy makers and agency leaders.

If it seems like Equal Language Access is in fact important, then it is simply a matter of confirming the return on investment for your desired outcomes. Compare your current costs and turnaround times for document language translation to the proposed costs and turnaround times with AI/ML assisted document language translation. If the value of the difference overtime puts you in a better place, the solution could be right for you.

John Cecala - Technology Advocate for Government

Disclaimer: All views expressed in this article are my own and do not represent any official statements, opinions or positions of any entity whatsoever with which I have been or am now affiliated with.?