@deeppast as per the competition overview we should remove word dividers like : OR ..\nDoes that mean we should change KÙ.BABBAR to KÙBABBAR ? In other words, are the . in ITU.KAM, KÙ.BABBAR etc word dividers (that should be removed)?
\n", "messages": [ { "id": "3386035", "postDate": "01/04/2026 12:49:21", "content": "Remove (modern scribal notations):
\n: OR . (word divider)
\n
@deeppast as per the competition overview we should remove word dividers like : OR ..\nDoes that mean we should change KÙ.BABBAR to KÙBABBAR ? In other words, are the . in ITU.KAM, KÙ.BABBAR etc word dividers (that should be removed)?
\n", "rawMarkdown": "deeppast as per the competition overview we should remove word dividers like `:` OR `.`.\nDoes that mean we should change KÙ.BABBAR to KÙBABBAR ? In other words, are the `.` in ITU.KAM, KÙ.BABBAR etc word dividers (that should be removed)? \n\n>Remove (modern scribal notations):\n\n>: OR . (word divider)", "votes": null }, { "id": "3387244", "postDate": "01/06/2026 15:57:21", "content": "Remove (modern scribal notations):
\n: OR . (word divider)
\n
No, those word dividers are only found together with white spaces on both sides: ' : ' OR ' . ' \nThe logograms you listed were not changed from their original format, but it is good to know that scholars use both 'KÙ.BABBAR' and 'KÙ.B.' / 'KÙ.B' in their transliterations.
", "rawMarkdown": "No, those word dividers are only found together with white spaces on both sides: ' : ' OR ' . ' \nThe logograms you listed were not changed from their original format, but it is good to know that scholars use both 'KÙ.BABBAR' and 'KÙ.B.' / 'KÙ.B' in their transliterations.", "votes": null }, { "id": "3387821", "postDate": "01/07/2026 17:09:13", "content": "Is somewhere everything we need to know about train / test differences and how the actual metrics are calculated maybe with examples in one place?\nIt is discouraging that things are scattered through the discussions, and I can't even let an LLM deep research read all the comments to collect these information because the kaggle platform does not make it possible.
", "rawMarkdown": "Is somewhere everything we need to know about train / test differences and how the actual metrics are calculated maybe with examples in one place?\nIt is discouraging that things are scattered through the discussions, and I can't even let an LLM deep research read all the comments to collect these information because the kaggle platform does not make it possible.", "votes": null }, { "id": "3387834", "postDate": "01/07/2026 17:30:26", "content": "That's a good suggestion. As yet there is no such document, but I can work on that.
", "rawMarkdown": "That's a good suggestion. As yet there is no such document, but I can work on that.", "votes": null }, { "id": "3388435", "postDate": "01/08/2026 20:18:03", "content": "Hi @nofreewill,
\nYou can find a brief description of the evaluation metric on the Evaluation page. The actual implementation we use for scoring is in this notebook: DPI BLEU / chrF++. The Evaluation page links to papers describing BLEU and chrF++, or you could refer to the sacrebleu documentation and source code.
\n\nRemove (modern scribal notations):
\n\n
\n- ! (certain reading)
\n- ? (questionable reading)
\n
@deeppast how to distinguish between an \"!\" as punctuation vs one that is a scribal notation? Same question for the question mark \"?\"\nThe train.csv looks mostly clean with some exceptions like \"The which Mati? and Tibula ..\" and there are cases where one has \"(?)\" which is possibly a scribal notation(?)
BTW it isn't entirely clear to me if this section on formatting suggestions (Formatting Suggestions for Transliterations and Translations) is meant for train.csv or only/mainly for the publications.csv?
I am aware of the fact that published_texts.csv has cleaned up the transliterations as per the formatting suggestions but I am asking about the translations.
No, those word dividers are only found together with white spaces on both sides: ' : ' OR ' . ' \nThe logograms you listed were not changed from their original format, but it is good to know that scholars use both 'KÙ.BABBAR' and 'KÙ.B.' / 'KÙ.B' in their transliterations.
", "votes": null, "replies": [ { "id": 3387821, "author_name": "nofreewill", "author_url": "", "post_date": "01/07/2026 17:09:13", "content": "Is somewhere everything we need to know about train / test differences and how the actual metrics are calculated maybe with examples in one place?\nIt is discouraging that things are scattered through the discussions, and I can't even let an LLM deep research read all the comments to collect these information because the kaggle platform does not make it possible.
", "votes": null, "replies": [ { "id": 3387834, "author_name": "deeppast", "author_url": "", "post_date": "01/07/2026 17:30:26", "content": "That's a good suggestion. As yet there is no such document, but I can work on that.
", "votes": null, "replies": [] }, { "id": 3388435, "author_name": "ryanholbrook", "author_url": "", "post_date": "01/08/2026 20:18:03", "content": "Hi @nofreewill,
\nYou can find a brief description of the evaluation metric on the Evaluation page. The actual implementation we use for scoring is in this notebook: DPI BLEU / chrF++. The Evaluation page links to papers describing BLEU and chrF++, or you could refer to the sacrebleu documentation and source code.
\n\nRemove (modern scribal notations):
\n\n
\n- ! (certain reading)
\n- ? (questionable reading)
\n
@deeppast how to distinguish between an \"!\" as punctuation vs one that is a scribal notation? Same question for the question mark \"?\"\nThe train.csv looks mostly clean with some exceptions like \"The which Mati? and Tibula ..\" and there are cases where one has \"(?)\" which is possibly a scribal notation(?)
BTW it isn't entirely clear to me if this section on formatting suggestions (Formatting Suggestions for Transliterations and Translations) is meant for train.csv or only/mainly for the publications.csv?
I am aware of the fact that published_texts.csv has cleaned up the transliterations as per the formatting suggestions but I am asking about the translations.