| Sumario: | This article presents a study on the use of artificial intelligence to analyze Finnish-language newspapers published in North America between 1876 and 1923. Using GPT-4 and LLaMA 3.1, we develop and evaluate a text segmentation method on large-scale digitized historical data, and we propose a hierarchical taxonomy for register (genre) classification based on manual annotation. We present a methodology for identifying text boundaries and, separately, a manually developed register taxonomy used to classify segments. Our analysis is grounded in a manually annotated stratified corpus of 500 documents selected from 312,300 digitized pages. Results demonstrate strong segmentation accuracy, contributing both to digital humanities methodology and the historical understanding of the Finnish-American immigrant press.
|
|---|