We present our submission to the HIPE-2026 shared task on multilingual person-place relation classification in historical documents, using the FI CODE approach. This task requires the prediction of both a ternary at relation and a binary isAt relation for sampled person-location pairs in OCR-processed historical documents written in German, English, and French. Three systems designed for efficiency are compared: a distance heuristic over entity mention positions with no training, an encoder-based ITER relation classifier, and an in-context learning system based on a small large language model. Our experiments demonstrate that, while simple positional cues provide a competitive baseline, encoder- and prompt-LLM-based systems offer additional advantages. In the HIPE-2026 efficiency ranking, we achieved first place for French texts and second place for English texts using distance-based relation classification.     «

We present our submission to the HIPE-2026 shared task on multilingual person-place relation classification in historical documents, using the FI CODE approach. This task requires the prediction of both a ternary at relation and a binary isAt relation for sampled person-location pairs in OCR-processed historical documents written in German, English, and French. Three systems designed for efficiency are compared: a distance heuristic over entity mention positions with no training, an encoder-base...     »