2.04 AI-assisted processing of heritage texts

The “content refinery”: Maximising the value of each story

This module’s framework begins with a crucial asset: a written transcript of the oral history interview. Once you have a clean, accurate text transcript, an AI assistant can become a powerful partner in what can be thought of as a “content refinery” process. A single, rich story can be efficiently processed and transformed into multiple valuable assets for different audiences and purposes, dramatically increasing the value and reach of the original collected heritage. This “refinery” framework – from raw text to summary, to keywords, to translated and adapted versions – demonstrates how AI can help you maximise the impact and accessibility of every single story you collect, turning one interview into a multitude of valuable educational assets. Do note that other approaches to transcriptions are possible. For instance, you may keep your own notes while conducting the interview and then use an AI tool to transcribe the interview. This initial AI transcription may not be perfect. However, by leveraging your notes, you can quickly identify interesting stories within the text. You must then manually review these selected sections for accuracy.

A quick note on the AI tools that will be mentioned in this course

Many of the specialised creative platforms we will explore operate on a “freemium” model. This means they offer excellent free tiers for you to experiment with, but may require a subscription for advanced features, higher-quality exports, or extensive use. We encourage you to explore the free options to learn the concepts. We highly suggest to explore the capabilities of Google Gemini, especially through the Google AI Studio, that offers a substantial daily usage in text, image, sound, and video creation.

A note on transcription

For the purposes of our demonstration, we will assume the audio recordings from Module 2 have been carefully transcribed. The process of transcription – converting spoken words into written text – requires great care. In a project dealing with diverse local dialects and the nuanced speech patterns of seniors, manual transcription by a human who understands the local context is often the best way to ensure accuracy.

However, this is extremely time-consuming.

AI-powered transcription services represent an accelerated alternative. Tools like Sonix, Rev, and Descript can rapidly convert audio to text.

Sonix is noted for its high accuracy (up to 99% on clear audio) and advanced analysis features like sentiment analysis and topic detection.

Rev offers a hybrid model with both AI and human transcription services, with its AI boasting over 90% accuracy.

Descript integrates transcription with powerful audio and video editing features, making it ideal for creating polished content.

Even with the best AI, however, a transcript of a dialect-rich oral history will almost always require significant human review and correction to catch errors, properly identify speakers, and ensure cultural nuances are not lost. Also note that almost all of the tools in this category are freemium. In other words, they offer limited capabilities or limited amount of audio recordings you can transcribe for free. The rising multimodal AI models such as previously mentioned Google Gemini models can be accessed for free, with some limitation on daily usage, through the Google AI studio.

AI demonstration: Summarisation

Long interview transcripts are dense and difficult to work with. Summarisation is often the first and most critical step in making the content manageable. AI tools, powered by advanced Natural Language Processing (NLP), are exceptionally good at distilling a large body of text to its core ideas.

For academic or research contexts, specialised tools like Scholarcy can provide structured summaries with highlighted findings.

Prompt:

Here is a 500-word excerpt from an oral history interview about traditional fishing on the river Drava. Summarise the main points in a short, clear paragraph. It is very important to preserve the key cultural details and the narrator’s personal perspective.

Example AI response:

The narrator fondly recalls how his grandfather taught him to read the river’s subtle signs – such as the colour of the water and the specific behaviour of birds – to know precisely where the fish would be. He describes this as a form of traditional ecological knowledge passed down through generations, not from books, but from direct experience. He also recounts the community’s collective effort to build and maintain intricate wooden weirs, a specific fishing technique that was central not only to the village’s economy but also to its social life before the 1960s. Ultimately, the narrator expresses a profound sense of loss, noting that while fishing is now mostly a hobby, the deep, communal connection to the river as a source of livelihood, identity, and shared culture has significantly faded over time.

AI demonstration: Keyword extraction

Once you have a summary, extracting keywords or subject tags makes the story easily discoverable within a digital archive. This process is fundamental to building any searchable database.

Tools like Writesonic’s Keyword Extractor can assist in identifying relevant terms from a body of text.

Prompt:

From the following summary of a heritage story, extract 5-7 relevant keywords that would be useful for archival tagging. Include important themes, locations, specific practices, and significant events mentioned.

Example AI response:

Drava River, Traditional Fishing, Weirs, Intergenerational Knowledge, Community Life, Ecological Change, Oral History.

AI demonstration: Translation

Translation makes your heritage collection accessible to a global audience, including international researchers and diaspora communities, which is essential for a cross-border project like HER[AI]TAGE.

Prompt:

Translate the following short passage from a Croatian heritage story into clear, natural-sounding English.

Passage: “Naši stari su znali točno kada će riba gristi. Gledali su u mjesec i vodostaj, to je bilo njihovo znanje, nije bilo zapisano u knjigama.”

Example AI response:

Original: Naši stari su znali točno kada će riba gristi. Gledali su u mjesec i vodostaj, to je bilo njihovo znanje, nije bilo zapisano u knjigama.

Translation: “Our elders knew exactly when the fish would bite. They looked at the moon and the water level; that was their knowledge, it was not written in books.”

AI demonstration: Plain-language rewrite

A single story can serve many different audiences. An AI assistant can help you adapt the language, style, and complexity of a text for various contexts, making the heritage you have collected more inclusive and effective.

Prompt:

Rewrite the following academic-style passage for a younger audience (ages 10-12) for a museum exhibit. The goal is to make it easy to understand and engaging.

Passage: “The intricate socio-economic dynamics of the riverside community were intrinsically linked to the seasonal migratory patterns of the ichthyofauna.”

Example AI response:

Transformed text: “The way people in the village lived and worked together was all connected to the fish. When the fish travelled up or down the river at different times of the year, it changed everything for the community.”