AI Analysis

OCR and Handwritten Letters

Optical character recognition was applied to handwritten letters in the archive. While OCR successfully identified some words and phrases, numerous transcription errors appeared due to handwriting variation, language mixing, and image quality.

Speech Transcription

Speech-to-text models were used to transcribe Burmese interviews. These models struggled with language support and accent variation, producing partial transcripts that required manual correction.

Sentiment Analysis

Sentiment analysis identified general emotional tone but often failed to capture complex emotional combinations such as grief combined with faith or resilience.

Translation Analysis

Translation workflows were used to test meaning drift across repeated Burmese ↔ English conversion passes and identify where testimony becomes unstable across model outputs.

Classification and Tagging

Visual artifacts were compared using manual tags versus AI-generated tags to evaluate category reliability and interpretive consistency.