generative-semantic-diff / llm_query /span_marking /configs /config.span_marking.v0.6.json
jvamvas's picture
Deploy generative semantic diff demo (Azure gpt-5.6-terra)
1a212f3 verified
Raw
History Blame Contribute Delete
1.63 kB
{
"prompt_template": "Identify the spans in Input A that are semantically dissimilar to Input B. Surround each dissimilar span with the special markers {{ }}. Leave the rest of Input A unchanged. Do not rewrite, translate, or paraphrase Input A — only add markers.\nNote that the two inputs might be in different languages. It is important to preserve the language of Input A – mark spans in Input A only; do not translate.\n\nHere are examples:\n\nExample 1:\nInput A (to be marked):\nCeltic defeated Everton 1–0 in the final, on 10 June 1938, with a goal from Johnny Crum in extra time.\n\nInput B:\nCeltic gewann das Endspiel durch ein Tor von Johnny Crum mit 1:0.\n\nMarked Input A:\nCeltic defeated {{Everton}} 1–0 in the final, {{on 10 June 1938}}, with a goal from Johnny Crum {{in extra time}}.\n\nExample 2:\nInput A (to be marked):\nC'est la quatrième plus grande ville de Slovaquie avec une population de 81 114 habitants en 2015.\n\nInput B:\nIt is the fourth-largest city of the province, with a population of 118,450.\n\nMarked Input A:\nC'est la quatrième plus grande ville de {{Slovaquie}} avec une population de {{81 114}} habitants {{en 2015}}.\n\nExample 3:\nInput A (to be marked):\nSi Elvis peut le faire, je peux le faire.\n\nInput B:\nIf Elvis can do it, I can do it.\n\nMarked Input A:\npass\n\nNow mark the following:\n\nInput A (to be marked):\n{text_a}\n\nInput B:\n{text_b}\n\nIf the texts are already completely equivalent to each other (no dissimilar spans), just respond with a single word: \"pass\". Otherwise, simply respond with Input A with dissimilar spans surrounded by {{ }}."
}