How Can AI Help Read Old Newspapers?
Updated: Sep 17
When Old Print Gets in the Way
You are no doubt familiar with the problem—you come across an interesting article in a historical newspaper, but it’s not only printed in old type and orthography, it’s also written in historical language with archaic expressions. It’s not easy but AI can help. I use AI tools in my own research every day to get help with various tasks.
This post was inspired by a question in a Facebook group, where my suggestion to use AI for this problem was met with skepticism. Therefore, I decided to demonstrate that AI (artificial intelligence) can indeed be a helpful assistant.
Where to Find Digitized Latvian Newspapers
The website periodika.lv automatically provides electronic versions of the printed texts of old newspapers, but often this version isn’t much easier to read—the words are still in old orthography, and OCR errors are common. Below is an example.
But I've learned to use AI to make this process easier. I'll walk you through it step by step:
Step 1: Copy the Electronic Text
In this example, I used a speech delivered at the opening of the beekeeping society “Drava” on August 9, 1898. The transcribed text is full of errors—for instance, “w” is most often read by computer as “m.” If the entire text had to be corrected manually, it would require a great deal of time and patience. Moreover, it is not easy to immediately “get the hang of it” and correctly read, for example, “ee” as “ie,” or to know when “s” should be understood as “z,” and so on.
Step 2: Run It Through ChatGPT
Other AI tools could certainly help as well, but at the moment I’m working with this paid ChatGPT-4. I ask it to transcribe the text into modern orthography without changing the words or sentence structure.
Of course, one could also ask AI to rewrite the text in modern language—allowing changes to words and sentence structure—but then it becomes harder to spot mistakes or compare with the original.
My prompt for AI in this case was: "Here is a transcript of a historical newspaper text. The computer recognized the letters, but a large part of them inaccurately. Write the text in modern orthography, but do not change the structure of the words or sentences." In my example I worked in Latvian but you can test with an English prompt, too but you have to point out that your
Step 3: Check the Result
Once I have the first version, I compare it with the original. AI does make mistakes. At this stage, it’s not really possible for an English speaker to verify the transformed text, but you can assume that most of it will be accurate
Step 4: Put the Text to Use
Once the text has been reviewed and corrected, I can ask AI to rewrite it in modern language, translate it, or summarize it—depending on what’s needed in the specific case.
A Note for English Speakers
For English speakers, I wouldn’t skip step 3 anyway, because AI might not translate the text correctly at first. However, once it has been converted into modern orthography, the translation is usually much more accurate
This is awesome Agnese! Thanks for the tip!