Hi Andy and Co. I have just listened to this podcast. Wonderful stuff! You have raised, and looked in detail, at questions that were very much niggling in my own mind, and have given careful considered responses. As a non-BT person 'as such', I wonder how much traction this anti-AI perspective is getting in the light of the pressure to quickly produce massive quantities of translations by AI? I hope a lot of traction! Much appreciated.
Also, simply clicking on the link in the post above (Listen to the podcast episode) takes you to a Spotify page where you can click the play button to listen, even without a Spotify account.
I noticed that Andrew Case did another podcast about AI and Bible Translation on the first of January called "The Privilege of the First Draft Revisited". It is available on the same web site: https://workingfortheword.com/. For this podcast, Andrew fed the pre-publication draft of our BT Conference paper through NotebookLM (an AI agent that summarizes information orally) and asked it to create an engaging "interview" which highlights the major points of the paper. Andrew writes, "Why do this? First, because it’s kind of funny. Second, because [the paper's] message is important enough to listen to again from another angle, so that you can internalize it better. Third, because NotebookLM actually does a fantastic job at this kind of thing, and I hope it inspires more people to consider using it to generate exegetical summaries for translators to listen to in different languages of wider communication."
I feel NotebookLM did do a fantastic job, great for hearing the message from another angle and for internalizing the concepts. There are a couple of quirks, like the word "Chadian" being pronounced three different ways... isn't that a standard English word? And the AI did make up some of its own content, things that weren't actually part of the paper. But even the AI generated content was generally appropriate and very much in line with the spirit of the paper. So overall very impressive. Please note that this podcast is impressive because it was created in the high-resource language English, and not one of the low-resource for which we are translating the Bible...
Please don't miss Andrew's third reason for creating this podcast. NotebookLM has the ability to create engaging exegetical summaries in different LWCs to help translators internalize a text and prepare for their drafting. Andrew has found this very helpful in his context, and hopes to inspire others to try creative uses of AI such as this in their workflows.
Wow, that was surreal! I just listened to "The Privilege of the First Draft Revisited" in which, as Jeff mentioned, Andrew ran our BT Conference paper through NotebookLM to produce an engaging podcast with a male presenter and a female guest. First thought - Wow! We sound way younger and cooler! Second thought - Amazing! This AI process clearly knows more than we do about how podcasts typically sound. It sounds natural because it was trained on huge numbers of podcasts. Third thought - wait, there are some errors in there. Some pronunciation errors, some grammatical errors, some factual errors - and many of these errors are really subtle and hard to notice. Also some of what the AI put in this podcast was created by it, not summarized from our paper. Final thought - it would be really useful to use this AI-created oral summary to search for gems - creative ways of stating something that we could use to improve our paper.
Using AI to produce a podcast about why not to use AI - very interesting approach, Andrew! It is also closely related to the question of using a Scripture Forge draft in the translation process. There is an initial impression of naturalness, but on closer examination, errors show up. This AI-generated podcast could provide some gems to improve the wording of our paper; in a analogous way, I think a Scripture Forge draft could provide some inspiration to translation teams, giving creative new ideas on how to translate an expression. But it should be presented to the translation team AFTER they have already made their own first draft.
Remember a few key points - 1. we researched and wrote this paper then Andrew gave it to AI afterward. The thoughts and content in the resulting podcast are mostly ours. This podcast is more representative of our thoughts then if AI created a podcast from scratch, even if it were trained on our previous work. 2. We, the human authors, would be able to carefully edit this AI-produced podcast to remove inaccuracies because we have already carefully studied the material. If we didn’t know the material well, we might not find the subtle errors. 3. As Jeff mentioned, this AI-generated podcast only sounds this good because it is in English, a high-resource language. AI’s ability to summarize a paper and produce an oral summary would be far less in most minority languages.
As Andrew hoped, I am inspired to try out NotebookLM. We are holding a Prophets workshop in Chad this month. I wonder if NotebookLM could help me create a short oral summary, in French, of the history of Israel from the beginning of the Divided Kingdom to Malachi? An oral summary might be an engaging way for Chadian translators (mostly oral learners) to grasp more of Israel’s history and better understand the historical context when translating the Prophets! I’ll give it a try.
Hi Jim, thanks for those warnings. I just want to clarify, however, that I wasn’t going to ask AI to invent a history of Israel. I am feeding NotebookLM my own writings (a 2 page summary of the prophetic history of Israel - that would be the "personal sacrifice of some sort") and asking it to turn that into audio. While caution and careful checking are still clearly required, I do think there is a difference between just asking AI to come up with the information , or presenting AI with written data and asking it to shift the medium to audio.
5 comments
Sign in to join the conversation.
Hi Andy and Co. I have just listened to this podcast. Wonderful stuff! You have raised, and looked in detail, at questions that were very much niggling in my own mind, and have given careful considered responses. As a non-BT person 'as such', I wonder how much traction this anti-AI perspective is getting in the light of the pressure to quickly produce massive quantities of translations by AI? I hope a lot of traction! Much appreciated.
Also, simply clicking on the link in the post above (Listen to the podcast episode) takes you to a Spotify page where you can click the play button to listen, even without a Spotify account.
I noticed that Andrew Case did another podcast about AI and Bible Translation on the first of January called "The Privilege of the First Draft Revisited". It is available on the same web site: https://workingfortheword.com/. For this podcast, Andrew fed the pre-publication draft of our BT Conference paper through NotebookLM (an AI agent that summarizes information orally) and asked it to create an engaging "interview" which highlights the major points of the paper. Andrew writes, "Why do this? First, because it’s kind of funny. Second, because [the paper's] message is important enough to listen to again from another angle, so that you can internalize it better. Third, because NotebookLM actually does a fantastic job at this kind of thing, and I hope it inspires more people to consider using it to generate exegetical summaries for translators to listen to in different languages of wider communication."
I feel NotebookLM did do a fantastic job, great for hearing the message from another angle and for internalizing the concepts. There are a couple of quirks, like the word "Chadian" being pronounced three different ways... isn't that a standard English word? And the AI did make up some of its own content, things that weren't actually part of the paper. But even the AI generated content was generally appropriate and very much in line with the spirit of the paper. So overall very impressive. Please note that this podcast is impressive because it was created in the high-resource language English, and not one of the low-resource for which we are translating the Bible...
Please don't miss Andrew's third reason for creating this podcast. NotebookLM has the ability to create engaging exegetical summaries in different LWCs to help translators internalize a text and prepare for their drafting. Andrew has found this very helpful in his context, and hopes to inspire others to try creative uses of AI such as this in their workflows.
Wow, that was surreal! I just listened to "The Privilege of the First Draft Revisited" in which, as Jeff mentioned, Andrew ran our BT Conference paper through NotebookLM to produce an engaging podcast with a male presenter and a female guest. First thought - Wow! We sound way younger and cooler! Second thought - Amazing! This AI process clearly knows more than we do about how podcasts typically sound. It sounds natural because it was trained on huge numbers of podcasts. Third thought - wait, there are some errors in there. Some pronunciation errors, some grammatical errors, some factual errors - and many of these errors are really subtle and hard to notice. Also some of what the AI put in this podcast was created by it, not summarized from our paper. Final thought - it would be really useful to use this AI-created oral summary to search for gems - creative ways of stating something that we could use to improve our paper.
Using AI to produce a podcast about why not to use AI - very interesting approach, Andrew! It is also closely related to the question of using a Scripture Forge draft in the translation process. There is an initial impression of naturalness, but on closer examination, errors show up. This AI-generated podcast could provide some gems to improve the wording of our paper; in a analogous way, I think a Scripture Forge draft could provide some inspiration to translation teams, giving creative new ideas on how to translate an expression. But it should be presented to the translation team AFTER they have already made their own first draft.
Remember a few key points - 1. we researched and wrote this paper then Andrew gave it to AI afterward. The thoughts and content in the resulting podcast are mostly ours. This podcast is more representative of our thoughts then if AI created a podcast from scratch, even if it were trained on our previous work. 2. We, the human authors, would be able to carefully edit this AI-produced podcast to remove inaccuracies because we have already carefully studied the material. If we didn’t know the material well, we might not find the subtle errors. 3. As Jeff mentioned, this AI-generated podcast only sounds this good because it is in English, a high-resource language. AI’s ability to summarize a paper and produce an oral summary would be far less in most minority languages.
As Andrew hoped, I am inspired to try out NotebookLM. We are holding a Prophets workshop in Chad this month. I wonder if NotebookLM could help me create a short oral summary, in French, of the history of Israel from the beginning of the Divided Kingdom to Malachi? An oral summary might be an engaging way for Chadian translators (mostly oral learners) to grasp more of Israel’s history and better understand the historical context when translating the Prophets! I’ll give it a try.
Hi Jim, thanks for those warnings. I just want to clarify, however, that I wasn’t going to ask AI to invent a history of Israel. I am feeding NotebookLM my own writings (a 2 page summary of the prophetic history of Israel - that would be the "personal sacrifice of some sort") and asking it to turn that into audio. While caution and careful checking are still clearly required, I do think there is a difference between just asking AI to come up with the information , or presenting AI with written data and asking it to shift the medium to audio.