The importance of context
At FGOBC we’re making a major effort to create subtitles and lists of scripture references, topics and other information that are included in the audio of classes and sermons that we offer on this site.
Subtitles and scripture references are very important in reaching people who don’t understand English, are hearing impaired or are researching scriptures and topics.
By keeping track of this information and keeping it in sync with the audio we can accomplish so much more than just providing the audio, in English, to listeners around the world. We can use that information to bridge language barriers and expand the simple audio into a multimedia experience.
We can reach the whole world with the classes and sermons on this site.
With the assistance of current technology we can provide:
- Subtitles made directly from the audio – Current technology gives us the ability to capture the spoken words as text.
- Subtitles can be translated to other languages – Subtitles can be displayed as closed captioning right on the screen as the user listens to the audio. Current technology also gives us the ability to translate from English to other languages.
- Scripture references can be identified in the subtitles – This makes it a lot faster and easier to find scripture references, topics and other data.
- Transcripts can easily be made from subtitles – Using this we can reach people who do not have internet access. Many who are incarcerated are not allowed access to technology because it is a security risk but they are allowed to receive printed material.
- Choice of Bible translations and languages – By displaying the scripture reference, while the speaker is discussing it, the user can click on it to get more information and view it in the language and translation of their choice. With public domain or licensed translations we can display the users preferred language right on screen.
- Text in a foreign language can be ‘spoken’ by computers – Text-to-speech software in available which covers many languages. By providing this we can reach illiterate people in any of those languages. Using voice sampling the ‘spoken’ results can be made to sound like the original speaker.
- Context is everything – As we determine start and stop points for displaying scripture references, topics, etc. we are effectively identifying commentary on that reference.
- Scripture references can be searched and analyzed – It’s very possible to create an online commentary using the scripture references found in classes and sermons. We can also create compendiums of topics, parallel passages and other information gleaned from the audio and subtitles.
However, getting there isn’t easy or automatic. Some today think that artificial intelligence (AI) could do all of this. The fact is AI isn’t there yet and may never get there.
Here are just some of the limitations of technology:
- Speech-to-text technology isn’t perfect – While this is a huge time saver it’s never completely accurate so everything has to be reviewed and edited. Even the best speech-to-text programs available aren’t perfect and many charge fees (we have about 2,400 hours of classes not counting sermons).
- Speakers make mistakes and don’t always use perfect grammar – It takes a human being to fix these issues. Sometimes it requires a human spirit in communion with the Holy Spirit to understand what was really meant.
- Language translation technology isn’t perfect – Unfortunately, it’s not an exact science. There are dialects, colloquialisms, figures of speech and other issues that must be overcome. For now, to be true to the intent of the speaker, the translated text has to be edited for accuracy, clarity and localization.
- Scripture references are often spoken in pieces – It’s common for a preacher or teacher to say, ‘turn with me to the book of John’. And without giving the rest of the references he will go on a short tangent while the listeners pick up their Bible and open it. After that interlude he quotes the verse and then gives the chapter and verse, often without saying the name of the Bible book again. Then add numbers in the text (’10 virgins’, ’10 heads and 7 horns’, etc.) and it becomes almost impossible for software to get it right.
- Preachers go on tangents – I doubt the AI will ever exist that can follow a preacher’s tangents and get back to the context. And sometimes the speaker never really gets back himself. And then there are the nested tangents: tangents within tangents.
To do this right. To teach these Bible lessons in other languages around the world we have to do this in steps and get each step right:
- Run speech-to-text software on the audio – This can be done either in the free software that is easy to run with the right tools or by paying for a service to generate the text. The only real requirement is that it generates a subtitle file (usually .srt or .vtt) that includes times for each line of text and not just a text file.
- Edit the subtitles for accuracy – If the speaker misquotes or clearly uses the wrong word the correct information can be supplied in brackets, but this should be done judiciously making sure it doesn’t change the meaning. After editing is complete the English subtitles can be displayed while the audio is playing.
- Create a list of scripture references in the database – Currently this is done via Web pages on FGOBC. Most speech-to-text software seem to get the names of the Bible books correct in the output. Based on that a list of possible references is created. This list will undoubtedly contain inaccurate information (see above example). It’s up to the volunteer to determine the intent of the speaker and enter the information into the database. This information must include the entire context, start time to stop time, so that the scripture reference can display correctly and so that the text can be extracted for research and compendium creation. Topics and other information can be done in a similar way.
- Translate the edited English subtitles to other languages and edit the results – Editing the results must be done by someone familiar with the second language with an understanding of dialects and idioms. Some things get lost in translation. Names are often spelled phonetically into the second language and those must be examined. Sometime look up the story of Coca-Cola and the ‘bite the wax tadpole’ fiasco.
- Create audio in the second language using generative AI – This can only be done on edited text and should be reviewed before making it available on the site. Someday that may not be necessary, but for now it is. Consideration should be given to voice sampling of the original speaker so that the created second language audio sounds like the original speaker.
I strongly believe this is a very viable plan but will require a lot of work. It took years for David Kenady and I to capture all of the audio from cassette tapes into computer files. I expect this project will also take years to complete.
Because of all of these cautions I think the best people to do this work are the people who once sat in those classes and took notes. They have a better understanding of what was meant by the speaker and what constitutes a context.
I’ll leave the discussion of commentaries and compendiums for another post or page.