Not Just Farming: How an LLM Helped Me Study Saskatchewan’s Smallest Settlements

Jessica Jack with Chloë Farr and Jacob Polay

This post is part of a series on AI and Collaboration.

Good AI-assisted historical research isn’t a shortcut around other people’s labour. It’s a way of putting that labour into conversation. In comparison to the extractive practices for which some LLMs are used, some historians have used LLMs to be a medium of research collaboration.

What does that look like? Take one finding from my research into Saskatchewan’s settlement history. For decades, the province’s 429 settlements as of 1921 were sorted by population size. Larger often meant more important.

Sort those same settlements by what they did and the picture changes. I classified them by economic function rather than population. Farm villages, the smallest settlements tracked in the census, had been long assumed to be purely agricultural. However, tracking economic function showed 79% of the classifiable workforce held jobs that had nothing to do with farming. They were grain buyers, bank clerks, blacksmiths, and station agents. Whole local economies built around moving goods and money were hiding inside these smaller places. Settlements dismissed due to their population size were, in many cases, running a meaningful share of the province’s day-to-day commercial life.

That finding didn’t come from one dataset. It came from me putting several separate sources into conversation, each built by different people for different purposes. Census enumerators recorded, and archivists later digitized, the occupation of nearly every working person in Saskatchewan’s 1921 census: roughly 760,000 individual records, built for a government count. A group of university researchers then cleaned, digitized, and mapped those census results and made them available for others to use. A separate research project had compiled grain elevator licensing records, meant to track industry activity, not classify communities. Government historians had preserved records of every post office opened and closed across the province. Railway historians and GIS specialists had digitized rail lines station by station. Finally, hundreds of volunteers, local historical societies, and archivists across Saskatchewan contributed to over 400 community-written local history books, supplying what the official records alone couldn’t: what people in a place actually said the place was for.

Each of these datasets was built to answer a narrower question, by people who never expected their work to end up joined to others. Building a settlement typology out of them is where an LLM came in. I did not use it to generate the findings, but it facilitated my work in merging disparate community sources into one. It helped by converting them into spreadsheets, importing them into a relational database (Neo4j), filtering this database for relevant information, standardizing the data against chronological change, and then effectively and accurately querying the results to answer new questions asked of old material. A census enumerator’s tally sheet, a grain elevator inspector’s license log, and a local history society’s book were never built to talk to each other. Now they can.

The result reframes Saskatchewan settlement around its division of labour. Cities, regional hubs, service towns, farm villages, grain hamlets, and elevator sidings each did structurally different economic work, whatever their population. That distinction was invisible in any single source. It only became visible once the people who built and published those sources, separately, for separate reasons, some of them a century apart, were brought into the same conversation.

LLMs as collaborative tools in historical research can be an aid to keeping people’s labour in conversation. My local history compilation alone drew on more than 600 named contributors and institutions: authors, archivists, and local historical societies. None of them set out to build a settlement typology. Yet when I used an LLM to help me combine them with census records, industry licensing data, postal archives, and railway data built by entirely separate teams, they became something entirely new.

LLMs ran through the rest of this series, piece by piece, doing the same kind of work. They helped to transcribe documents, making hard-to-read historical material legible at a broader, more accessible scale. They facilitated naming and tagging historically relevant entities so they could be more easily accessed by researchers. They kept a knowledge graph’s sources attached to the graph, rather than letting them dissolve into a single unaccountable answer, enabling a researcher to query over a million pages. These are different methods with the same shape: the LLMs enabled us to expand our work beyond the traditional boundaries imposed by time, budget, and capability.

Talkie is what that same tool looks like without the collaboration. It extracts historical and archival labour and returns one closed answer, with no path back to the people whose work it was trained on and no attempt to keep their labour in conversation. When LLMs are used well, they can avoid this knowledge extraction. They can keep the people behind that work visible, connected, and in dialogue.

The settlement typology above exists because an LLM did exactly that kind of connecting work. It matched historical data points that meant different things depending on who recorded them. It reconciled categories built decades apart. It held enough separate pieces of other people’s labour at once that a census enumerator, a grain elevator inspector, a postal clerk, and hundreds of local history volunteers ended up inside the same analysis, despite never working together and never having any reason to expect their work would meet. An LLM was a collaborator that helped to join the efforts of all of those contributors in a timely and efficient way.

We invite other historians and GLAM professionals who have used LLMs in their research to contribute further to this series, and hope readers take from this series the importance of knowledge collaboration. Whether the people whose labour built the archive get treated as collaborators, or as raw material to be mined once and discarded, is the choice we have tried to demonstrate throughout this series. It is one we hope you will consider in your own work too. An LLM can extract as easily as it can connect. These are the ethical roads that diverge in a wood, but it is the user who decides which road is taken. Less scrupulous actors follow each other down the well-traveled roads. We have chosen to take the one less traveled by, and we hope it makes all the difference.

Editor’s note: If you wish to publish a piece on your own technological research tools or respond to any aspect of this series, please write to activehistoryinfo@gmail.com.

Jessica Jack is a PhD student in History at the University of Saskatchewan, developing applications for Large Language Models in historical research. They are doing so through studying settler land use in late 19th century and early 20th century Saskatchewan.

Chloë Farr is a researcher working at the intersection of artificial intelligence, archives, and digital humanities. Working out of the Open Science Lab at TIB – Leibniz Information Centre for Science and Technology, her research focuses on large-scale text recognition and analysis of historical documents, including newspapers, maps, and archival records. Learn more about Farr’s work on GitHub.

Jacob Polay is a PhD student in History at the University of Saskatchewan, studying the roles Large Language Models have in the historical method. His current research involves creating an information retrieval pipeline using artificial intelligence tools to unlock the early modern archive at scale.