Grok Learned To Transcribe Jonathan Fisher’s Shorthand

I’ve been teaching Grok 4.6 how to transcribe Jonathan Fisher’s shorthand into English. The results have been promising.

Well, sort of.

I uploaded every resource I had access to, including Raoul Smith’s works on Fisher and course materials from the transcription workshop presented at the Jonathan Fisher House.

Next, I uploaded images of Fisher’s Sermon #3 and our human transcription of the sermon. It took me about 40 minutes to upload and discuss the training materials with Grok. After uploading some new sermon images, Grok was able to decipher and transcribe the pages in a fraction of a second. See the images below.

Next, I pointed it to additional training materials and asked it to transcribe a later sermon written in Fisher’s mature shorthand, which consists almost entirely of symbols. Grok struggled. It offered a general analysis but not a word-for-word transcription. It also flagged the image resolution, a fair complaint. Using the consumer version of Grok, I am limited by file size, so later sermons would need to be uploaded at original resolution and likely enhanced in Photoshop or a similar program.

My conclusion after roughly an hour of testing: a dedicated AI agent with access to all available training materials, university-grade computing, and someone to organize and upload high-resolution images of Fisher’s shorthand documents could decipher and transcribe Fisher’s entire extant corpus in about a week or, perhaps, two.

That would be a genuine service to scholars of New England religion, society, politics, economics, and culture in the Early Republic and the Second Great Awakening. Imagine this: a nineteenth-century minister’s private script might be opened to scholars within days instead of years with the help of an AI agent, under the direction of an enthusiastic scholar. That’s 2,900+ sermons along with other personal notes, journals, inventories, and sundry items. All waiting to be read after 200 years.