Smartphones are everywhere and it seems like there are apps for everything. I log into my business accounts platform using an authorisation app. I listen to the Bible using a Bible app and I rely on a calendar app to remind me when to do things. Might it be that apps could allow churches to break the language barrier?
And the Answer is…
It’s complicated. Since I wrote an entire book about machines doing interpreting, I won’t bore you with the details. It is useful however to understand how these apps work since that knowledge will help us make better decisions.
Broadly speaking, there are two ways to build a speech translation app. The first is call the cascade model.
In the cascade model, what you say is turned into text by a voice recognition engine. This text is then fed into a machine translation engine. What comes out of that is then sent to a speech synthesiser to read aloud.
This means that what people hear at the end has come from written text. There is no place for intonation, emotion, speed or even accent data in this model. What you say is turned into text, that text is run through machine translation and whatever comes out the other end is what you get. If you would like to hear what that might sound like with the sample of voice synthesis from Microsoft Word.
This isn’t actually that bad but it isn’t going to stir an emotional reaction in anyone. Even the best cascade model interpreting apps will fail to produce output with any noticeable emotion.
Nowadays, a small number of providers are testing an alternative model, called the end-to-end model. That model goes straight from speech in one language to speech in another, without ever turning anything into text. The point of doing things that way is precisely to keep emotion and intonation. A sample of what an end-to-end model can sound like can be found on the Google AI blog. There are no apps that currently offer that model to the public.
While end-to-end models can keep emotion and they are billed as keeping voice characteristics, what they lack is knowing what works in context. They have no idea about politeness or what is acceptable in different cultures.
The Limits of Speech Translation in Church
Where all forms of speech translation fail is that, unless you pay out the money to train a system yourself, they will be trained on large samples of language, which probably doesn’t reflect how language is used in your church. To this day, Google Translate tends to produce French which is more in line with Roman Catholic terminology than protestant terminology. Apps certainly won’t produce terms that line up perfectly with the ones you are used to. Just think about key words like eucharist/Lord’s supper/communion and priest/minister/pastor and you can imagine how using the wrong term could lead to discomfort.
Speech translation is also thrown off by background noise, accents, unclear speech, and slang words. It doesn’t do a great job of reflecting excitement, sorrow, or anger.
It is a useful tool but it has limits.
Using Speech Translation Apps Wisely
None of this means that churches should ignore the potential of speech translation. For everyday conversations, times when simply relaying information is what is important, and for initial contact, speech translation apps can be useful. It is impossible to ignore the fact that they do break down some language barriers and allow people to communicate, even if they don’t share a language.
If the situation is right for automatic speech translation, do everything you can to minimise background noise, speak in complete sentences and eliminate any local or slang language. Do also monitor how quickly you are speaking and avoid interrupting the other person. Also work at adding in questions to check they have understood, such as “do you know what you need to do next?” or “do you remember what we just spoke about?” Asking questions to check for specific pieces of information is much more effective than asking “do you understand?”
But any time something has theological importance, any time that getting the right terminology matters, any time something is read from a Bible translation, and any time how something is said matters as much as what was said, then speech translation apps simply cannot deliver what is needed. In those circumstances, churches need to look at other solutions.
There are lots of possible solutions, which will be covered in an upcoming guide and are covered in detail in the book, Multilingual Church.