IJIS Institute Studies Reliable Text-to-9-1-1 Translation Techniques for PSAPs
Thursday, May 14, 2020 | Comments

As public-safety answering points (PSAPs) begin implementing text-to-9-1-1 capabilities, they run into the challenge of answering texts written in languages besides English.

Currently, somewhere between 20% and 40% of the nation’s 6,500 PSAPs have text-to-9-1-1 capabilities, said Michael Alagna, director of technology for the IJIS Institute, during a webinar hosted by the National 911 Program.

In the U.S., about 60 million people speak another language in addition to English at home, and data on text-to-9-1-1 shows that soon after a PSAP implements text to 9-1-1, it begins receiving texts for assistance in other languages. Translating those texts in other languages can prove quite difficult for PSAPs. Most PSAPs take one of two approaches or a combination of both in translating those texts. Machine translation offers the benefits of being instantly available, but the reliability of the translation does not always meet public-safety requirements.

The IJIS Institute worked with the Department of Homeland Security (DHS) Science & Technology Directorate (S&T) to analyze automated machine translation of texts from Spanish to English, Alagna said. Each message received a BLEU score, which ranks the translation quality and assigns a score, with 1 being a perfect match and 0 being a perfect mismatch.

The average BLEU score for the translations was a 0.45, meaning generally high-quality translations. Additionally, human linguists deemed about 56% of the machine translations acceptable, Alagna said.

One of the largest issues causing machine translation failures was that the computer could not interpret the text properly because of typos, misspellings, lack of punctuation and what Alagna described as “text speech,” or informal speech used regularly in texting.

Additionally, while the machine translation is capable of translating a message correctly, it will not always convey the full meaning of that message because of a lack of context, he said. Most machine translation applications are commercial applications, so they don’t have the needed public-safety context built into them to ensure that the dispatcher receives a full idea of the situation.

Instead of machine translation or in addition to it, PSAPs will also use in-person or over-the-phone translations by humans to respond to the texts. However, this also brings challenges. If a PSAP staff member has knowledge of a specific language, that person will often answer the text, Alagna said. However, that person will not be working every time a text comes through, and there are so many different languages spoken in the U.S. that it’s unlikely a PSAP will have a staff member who speaks all spoken languages in the U.S, he said.

Many PSAPs also use over-the-phone translation services that offer translation of more than 200 different languages, but the voice nature of those services proves difficult for translation of text messages. The over-the-phone market is big enough that PSAPs can benefit from sharing the costs of those services. However, there is a lack of a business case for human-assisted interpretation of 9-1-1 text messages, meaning that such a translation method is too costly for PSAPs to adopt.

Because of the current challenges to human translation of the messages, many PSAPs will have to depend on machine translations because of the low investment costs for the technology, said Alagna.

Together with the DHS S&T and Google, the IJIS Institute hosted a techfest that simulated a PSAP using machine translation with an integrated option that allows a dispatcher to conference in a human translator to help coach the dispatcher through a machine-translated text.

Some version of a similar solution is likely the best option to help solve the translation issue, but the lack of a business case for the human and machine-based translation makes such a solution difficult right now, Alagna said.

To allow uninterrupted service to those texting to 9-1-1 as more PSAPs implement the technology, some work needs to be done to make both translation options viable, he said. First, public safety should work, in partnership with industry, to develop a business case for including human translation in any text translation solution.

At the same time, public safety and industry must also work on machine translation that takes public-safety situations into account. Many machine translation applications use machine learning and artificial intelligence (AI) to learn and improve translation, but because most translation software is commercial, that machine learning isn’t geared toward public-safety situations, limiting effectiveness in a public-safety solution. To help public safety, machine translation applications that use machine learning and AI to better learn public-safety context and implement that in their translations are needed, Alagna said.

The IJIS Institute initially planned to trial a hybrid solution in real PSAPs, but the initial phase of the project ended, and IJIS did not receive another round of funding to continue it, said Alagna.

Would you like to comment on this story? Find our comments system below.



 
 
Post a comment
Name: *
Email: *
Title: *
Comment: *
 

Comments

No Comments Submitted Yet

Be the first by using the form above to submit a comment!

Site Navigation

Close