TechFest event looks to the future of translation for text-to-911

There is an increasingly pressing need for PSAPs to be able to translate text-to-911 messages sent in languages other than English


By IJIS Institute

The introduction of texting to 911 emergency call centers from mobile phones opened up a new way to contact public safety officials, but public safety answering points (PSAPs) face a real problem if the texts are in a foreign language. Few PSAPs have access to a human translator and instead often deploy machine translation tools. Machine translation software automates the process of translating text from one language to another and can translate text without human editing.

With the growing adoption of text-to-911 services, industry leaders see an increasingly pressing need to be able to translate text-to-911 messages sent in languages other than English. This need is urgent, as more than 60 million people in the United States speak a language other than English in their homes, while some 28 million people are identified as having limited English proficiency.

A dispatcher in Roswell, Ga., works with a variety of screens while handling a 911 call. The Roswell call center is one of the few in the United States that accepts text messages. (AP Photo/Lisa Marie Pane)
A dispatcher in Roswell, Ga., works with a variety of screens while handling a 911 call. The Roswell call center is one of the few in the United States that accepts text messages. (AP Photo/Lisa Marie Pane)

In 2018, according to voluntary reporting to the Federal Communications Commission, it was estimated that 23% of PSAPs in the United States accepted text messages sent to 911. There is no national program at the helm of the implementation of text-to-911 services. In fact, the widely used message, “Call if you can, text if you can’t,” conveys the preference for voice calls over texts.

However, the adoption of text-to-911 services is expected to grow through a local patchwork approach. There are compelling use cases for text-to-911, such as situations of domestic abuse, home invasion, school shootings or other incidents in which callers fear being discovered if they speak.

“Texts sent to 911 are still a very low proportion of overall emergency requests,” said Michael Alagna, director of technology at the IJIS Institute, a nonprofit alliance working to promote and enable technology in the public sector. “But just as many emergency call centers have voice translation services for calls, we also need to be able to offer translation for texts sent to 911 to best serve our communities.”

Toward this end, the IJIS Institute, the Department of Homeland Security Science and Technology Directorate, and Google hosted a text-to-911 TechFest event at the Google campus in Kirkland, Washington.

Public safety leaders, technologists and government representatives attended TechFest to discuss how text-to-911 is being used across the country and the need for wider implementation, public education about the services and more technology tools to support translation services for text-to-911.

TechFest was part of a broader effort by DHS S&T to work with the project team to develop, pilot and test a solution for nationally relevant translation capability for PSAPs to obtain accurate text-to-911 information from LEP users.

“When DHS S&T and IJIS began the project in February 2016, less than 3% of the nation’s 6,000 PSAPs, also known as 911 call centers, had implemented text-to-911,” said Denis Gusty, DHS S&T program manager. “Since then, not only has the number of PSAPs using the platform increased to 30%, but federal, state and local laws have required call centers to ensure that the platform is available to the limited English proficient (LEP) population. Currently, almost 28 million people across the United States are identified as LEP and need to be accommodated as more PSAPs implement the technology in their respective communities. We anticipate the result of this joint project will be a national standard for implementing text-to-911 to LEP populations, as well as operational, business and training protocols that will ensure consistent national implementation.”

Current State of Translation Services for Emergency Calls

Many PSAPs nationwide have contracts with language service providers to offer interpretation and translation services for emergency phone calls made by people with LEP.

However, as discussed among attendees at the TechFest, no representatives from the PSAPs present were using existing agreements with language service providers to translate text communications.

In many locations, public safety officials have found that soon after launching text-to-911 services, 911 call centers began receiving texts in non-English languages. By some estimates, as many as 2 million texts per year require language assistance, but there is currently no standard model for translating texts to 911. Many jurisdictions have not implemented any tools to translate text messages.

The Emergency Communications Center in Arlington County, Virginia, began accepting texts to 911 in February 2016 but has seen much slower uptake than anticipated. Over a little more than three years, the center had received around 400 texts. Dave Mulholland, administrator of the Arlington County, Virginia, Emergency Communications Center, estimates that less than 15% of those texts were legitimate public safety concerns.

In Arlington County, an estimated 140 languages are spoken over only 26 square miles. The Emergency Communications Center does use a language interpretation service for voice calls but does not currently have a method for translating texts sent in languages other than English.

At the TechFest event, the roundtable discussion among attendees confirmed that many PSAPs do not translate texts, or if they are currently engaging in responding to texts in languages other than English, dispatchers rely on machine translation tools. Public safety leaders are keenly interested in understanding the efficacy of machine translation software particularly in calls where life and safety are at risk.

Efficacy of Machine Translation

In PSAPs that are using machine translation tools, these systems are understood to be "best-effort" delivery systems, meaning the provider does not guarantee the results. There is an underlying assumption that these "best-effort" systems are better than no effort at all to accept texts to 911 in non-English languages.

A commissioned analysis of machine translation services for public safety uses found there were many instances where Google Translate produced high-quality translations, with acceptable translations over 50% of the time. But misspellings, colloquial terminology, diacritical marks and text shorthand can all interfere with a successful translation.

In cases where life and safety are in danger, there are concerns with using machine translation services, which may miss key elements of language interpretation. In contrast, human translators can see if the intended meaning of a message is conveyed comprehensively and can add important context missed by machine translation tools.

Ecosystem of Solution Providers

Representatives from technology companies that offer machine translation services attended the TechFest event, including staff from Google, Agent511, INdigital and RapidDeploy. In recent years, machine translation has grown in sophistication and some have suggested that human translators could become obsolete. Google Translate is one of the most used machine services: It currently translates over 100 billion words a day and offers translation between English and more than 100 other languages.

The experts concluded that given the serious nature of 911 emergency communications, PSAPs should understand the accuracy of machine translation as a means of enabling communication via text message between the LEP public and public safety officials. A successful translation does more than just translate the message in isolation. It conveys the intended meaning and considers additional contextual information.

While human translators can provide this as they do with 911 voice calls, at present, there is no established business case for human-assisted translation of text-to-911. Without this, PSAPs will have to rely on the efforts of technology firms to improve machine translation quality.

James Kinney, Chief Innovation Officer at INdigital, sees achieving cost-effective, human language translation or high-quality machine translation as overlapping challenges: “I think we still stand behind machine translation as better than nothing, but it is also far from stellar yet. Getting to that middle ground is where the concern lies. What is it going to ultimately cost us: money or quality? At this point, it seems like we have to pick one and there is no happy, truly safe medium yet.”

The Future of Text-to-911 Services

At the national level, without federal program guidance or a clear funding stream to support text-to-911, industry leaders anticipate slow but steady uptake and implementation by jurisdiction, which creates challenges for adoption, standardization and affordability.

Leaders from public safety emergency call centers, industry technologists and language service providers will continue to collaborate to meet the need for an affordable, public-safety-grade solution for translating texts sent to 911.

“Translation is critical to the growth and inclusiveness of text-to-911 in serving all communities,” said Jay Malin, founder and managing director at Agent511. “At a minimum, an automated best-effort response is better than no solution in expediting response and dispatch of emergency services.”

TechFest highlighted that further research and development is needed to develop a national standard for text-to-911 translation for LEP populations. Another long-range goal is to have emergency response technology providers use machine learning to get computers to improve based on public safety use cases. By feeding data and information in the form of observations and real-world interactions, machine translation will continue to improve.

Join the discussion

Copyright © 2019 firerescue1.com. All rights reserved.