{
  "id": 666498,
  "title": "First competition",
  "url": "/competitions/deep-past-initiative-machine-translation/discussion/666498",
  "author_name": "",
  "post_date": "2026-01-07T13:00:46.502299600Z",
  "votes": 2,
  "comment_count": 2,
  "views": 0,
  "content": "<p>Hello, there. I hope everyone is doing well. I'm curious to understand how to implement ORACC (Open Richly Annotated Cuneiform Corpus), RINAP (Royal Inscriptions of the Neo-Assyrian Period) and/or \nHugging Face / Kaggle. What is the proper protocol? </p>",
  "messages": [
    {
      "id": "3387671",
      "postDate": "01/07/2026 13:00:46",
      "content": "<p>Hello, there. I hope everyone is doing well. I'm curious to understand how to implement ORACC (Open Richly Annotated Cuneiform Corpus), RINAP (Royal Inscriptions of the Neo-Assyrian Period) and/or \nHugging Face / Kaggle. What is the proper protocol? </p>",
      "rawMarkdown": "Hello, there. I hope everyone is doing well. I'm curious to understand how to implement ORACC (Open Richly Annotated Cuneiform Corpus), RINAP (Royal Inscriptions of the Neo-Assyrian Period) and/or \nHugging Face / Kaggle. What is the proper protocol?",
      "votes": null
    },
    {
      "id": "3387893",
      "postDate": "01/07/2026 18:51:12",
      "content": "<p>Hello, </p>\n<p>It is important to note that ORACC is a project based database, and is not as comprehensive as the CDLI. It is also important to know that there are no Old Assyrian texts in ORACC, outside of a couple royal inscriptions, which are of a very different genre and style.</p>\n<p>The good thing about ORACC is that the texts are lemmatized, which means they have been linked to an internal dictionary / lexicon and thus the data is cleaner than what we encounter in the Old Assyrian datasets presented in this challenge. If you choose to use the ORACC data and the different projects in there, you will want to pay close attention to the formatting differences, outlined in the Overview Additional Characters &amp; Formats (you may encounter), which indicates the format of the challenge as \"Character\", the format of the text in the \"CDLI\", and the format of the text in \"ORACC\".</p>\n<p>A couple years ago I acquired all the ORACC data and placed it in this Zenodo repository, labeled finialdf.csv\n<a href=\"https://zenodo.org/records/10794626/files/factgrid_df.csv?download=1\" target=\"_blank\">https://zenodo.org/records/10794626/files/factgrid_df.csv?download=1</a></p>\n<p><strong>The header fields are as follows:</strong></p>\n<ul>\n<li>0 = index</li>\n<li>lang = language-dialect</li>\n<li>form = transliteration form</li>\n<li>id_word = unique word id</li>\n<li>label = published line label of the tablet</li>\n<li>id_text = unique text id (also corresponding to CDLI id, when it begins with P######)</li>\n<li>delim = blank</li>\n<li>gdl = for ORACC's internal use (redundant fields in an array)</li>\n<li>pos = part of speech tag</li>\n<li>cf = citation form, a normalization of the transliteration form</li>\n<li>gw = guide word, a basic English gloss</li>\n<li>sense = English sense of the word, somewhat more verbose</li>\n<li>norm = standard normalization of the transliteration form (often identical to cf)</li>\n<li>epos = effective part of speech, i.e. part-of-speech labels, but null values for breaks and numbers</li>\n<li>headform = the lexeme used for lemmatization with Oracc's internal dictionaries (only for Sumerian)</li>\n<li>contrefs = word_ids for when the same word occurs multiple times in the same text</li>\n<li>norm0 = the lexeme used for lemmatization with Oracc's internal dictionaries (usually the same as headform, primarily used for Sumerian)</li>\n<li>base = the base of the form (without morphological affixes, only for Sumerian)</li>\n<li>morph = the morphological affixes when present (only for Sumerian)</li>\n<li>stem = a sparsely attested label for the Akkadian verbal stem (e.g. G, D, N, Ntn, Š), for more on this see an Akkadian grammar (e.g., Huehnergard 2011 <a href=\"https://www.academia.edu/234695/2011_A_Grammar_of_Akkadian_3rd_edition_\" target=\"_blank\">https://www.academia.edu/234695/2011_A_Grammar_of_Akkadian_3rd_edition_</a>)</li>\n<li>cont = morphological decomposition of the affix (only for Sumerian)</li>\n<li>syntax_ub-after = rare instances of '-.' syntax for numbers (decimals) and Sumerian logograms (not comprehensive)</li>\n<li>morph2 = blank</li>\n<li>aform = blank</li>\n</ul>",
      "rawMarkdown": "Hello, \n\nIt is important to note that ORACC is a project based database, and is not as comprehensive as the CDLI. It is also important to know that there are no Old Assyrian texts in ORACC, outside of a couple royal inscriptions, which are of a very different genre and style.\n\nThe good thing about ORACC is that the texts are lemmatized, which means they have been linked to an internal dictionary / lexicon and thus the data is cleaner than what we encounter in the Old Assyrian datasets presented in this challenge. If you choose to use the ORACC data and the different projects in there, you will want to pay close attention to the formatting differences, outlined in the Overview Additional Characters & Formats (you may encounter), which indicates the format of the challenge as \"Character\", the format of the text in the \"CDLI\", and the format of the text in \"ORACC\".\n\nA couple years ago I acquired all the ORACC data and placed it in this Zenodo repository, labeled finialdf.csv\nhttps://zenodo.org/records/10794626/files/factgrid_df.csv?download=1\n\n**The header fields are as follows:**\n- 0 = index\n- lang = language-dialect\n- form = transliteration form\n- id_word = unique word id\n- label = published line label of the tablet\n- id_text = unique text id (also corresponding to CDLI id, when it begins with P######)\n- delim = blank\n- gdl = for ORACC's internal use (redundant fields in an array)\n- pos = part of speech tag\n- cf = citation form, a normalization of the transliteration form\n- gw = guide word, a basic English gloss\n- sense = English sense of the word, somewhat more verbose\n- norm = standard normalization of the transliteration form (often identical to cf)\n- epos = effective part of speech, i.e. part-of-speech labels, but null values for breaks and numbers\n- headform = the lexeme used for lemmatization with Oracc's internal dictionaries (only for Sumerian)\n- contrefs = word_ids for when the same word occurs multiple times in the same text\n- norm0 = the lexeme used for lemmatization with Oracc's internal dictionaries (usually the same as headform, primarily used for Sumerian)\n- base = the base of the form (without morphological affixes, only for Sumerian)\n- morph = the morphological affixes when present (only for Sumerian)\n- stem = a sparsely attested label for the Akkadian verbal stem (e.g. G, D, N, Ntn, Š), for more on this see an Akkadian grammar (e.g., Huehnergard 2011 https://www.academia.edu/234695/2011_A_Grammar_of_Akkadian_3rd_edition_)\n-  cont = morphological decomposition of the affix (only for Sumerian)\n- syntax_ub-after = rare instances of '-.' syntax for numbers (decimals) and Sumerian logograms (not comprehensive)\n- morph2 = blank\n- aform = blank",
      "votes": null
    },
    {
      "id": "3415176",
      "postDate": "02/28/2026 10:41:01",
      "content": "<p>Hello,</p>\n<p>To implement resources such as <strong>ORACC (Open Richly Annotated Cuneiform Corpus)</strong> and <strong>RINAP (Royal Inscriptions of the Neo-Assyrian Period)</strong> together with platforms like <strong>Hugging Face</strong> or <strong>Kaggle</strong>, the general protocol usually involves a few key steps:</p>\n<ol>\n<li><p><strong>Data access and licensing</strong>\nFirst, verify the usage terms and licenses of ORACC and RINAP datasets to ensure that redistribution and machine-learning use are allowed.</p></li>\n<li><p><strong>Data extraction and formatting</strong>\nDownload the corpus data (often available in XML, JSON, or structured text formats) and convert it into a clean, consistent format suitable for analysis or model training.</p></li>\n<li><p><strong>Preprocessing and annotation handling</strong>\nPreserve linguistic annotations (translations, transliterations, metadata, etc.) while preparing the data. This may include normalization, tokenization, and cleaning depending on the research goal.</p></li>\n<li><p><strong>Integration with ML platforms</strong></p>\n<ul>\n<li><strong>Hugging Face:</strong> convert the dataset into a compatible dataset format (e.g., using the <code>datasets</code> library) for NLP experimentation or model training.</li>\n<li><strong>Kaggle:</strong> structure the dataset into tabular or text files (CSV/JSON) and provide clear documentation so it can be used in notebooks or competitions.</li></ul></li>\n<li><p><strong>Documentation and reproducibility</strong>\nClearly document preprocessing steps, scripts, and data sources so that results can be reproduced and validated by others.</p></li>\n</ol>",
      "rawMarkdown": "Hello,\n\nTo implement resources such as **ORACC (Open Richly Annotated Cuneiform Corpus)** and **RINAP (Royal Inscriptions of the Neo-Assyrian Period)** together with platforms like **Hugging Face** or **Kaggle**, the general protocol usually involves a few key steps:\n\n1. **Data access and licensing**\n   First, verify the usage terms and licenses of ORACC and RINAP datasets to ensure that redistribution and machine-learning use are allowed.\n\n2. **Data extraction and formatting**\n   Download the corpus data (often available in XML, JSON, or structured text formats) and convert it into a clean, consistent format suitable for analysis or model training.\n\n3. **Preprocessing and annotation handling**\n   Preserve linguistic annotations (translations, transliterations, metadata, etc.) while preparing the data. This may include normalization, tokenization, and cleaning depending on the research goal.\n\n4. **Integration with ML platforms**\n\n   * **Hugging Face:** convert the dataset into a compatible dataset format (e.g., using the `datasets` library) for NLP experimentation or model training.\n   * **Kaggle:** structure the dataset into tabular or text files (CSV/JSON) and provide clear documentation so it can be used in notebooks or competitions.\n\n5. **Documentation and reproducibility**\n   Clearly document preprocessing steps, scripts, and data sources so that results can be reproduced and validated by others.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3387893,
      "author_name": "deeppast",
      "author_url": "",
      "post_date": "01/07/2026 18:51:12",
      "content": "<p>Hello, </p>\n<p>It is important to note that ORACC is a project based database, and is not as comprehensive as the CDLI. It is also important to know that there are no Old Assyrian texts in ORACC, outside of a couple royal inscriptions, which are of a very different genre and style.</p>\n<p>The good thing about ORACC is that the texts are lemmatized, which means they have been linked to an internal dictionary / lexicon and thus the data is cleaner than what we encounter in the Old Assyrian datasets presented in this challenge. If you choose to use the ORACC data and the different projects in there, you will want to pay close attention to the formatting differences, outlined in the Overview Additional Characters &amp; Formats (you may encounter), which indicates the format of the challenge as \"Character\", the format of the text in the \"CDLI\", and the format of the text in \"ORACC\".</p>\n<p>A couple years ago I acquired all the ORACC data and placed it in this Zenodo repository, labeled finialdf.csv\n<a href=\"https://zenodo.org/records/10794626/files/factgrid_df.csv?download=1\" target=\"_blank\">https://zenodo.org/records/10794626/files/factgrid_df.csv?download=1</a></p>\n<p><strong>The header fields are as follows:</strong></p>\n<ul>\n<li>0 = index</li>\n<li>lang = language-dialect</li>\n<li>form = transliteration form</li>\n<li>id_word = unique word id</li>\n<li>label = published line label of the tablet</li>\n<li>id_text = unique text id (also corresponding to CDLI id, when it begins with P######)</li>\n<li>delim = blank</li>\n<li>gdl = for ORACC's internal use (redundant fields in an array)</li>\n<li>pos = part of speech tag</li>\n<li>cf = citation form, a normalization of the transliteration form</li>\n<li>gw = guide word, a basic English gloss</li>\n<li>sense = English sense of the word, somewhat more verbose</li>\n<li>norm = standard normalization of the transliteration form (often identical to cf)</li>\n<li>epos = effective part of speech, i.e. part-of-speech labels, but null values for breaks and numbers</li>\n<li>headform = the lexeme used for lemmatization with Oracc's internal dictionaries (only for Sumerian)</li>\n<li>contrefs = word_ids for when the same word occurs multiple times in the same text</li>\n<li>norm0 = the lexeme used for lemmatization with Oracc's internal dictionaries (usually the same as headform, primarily used for Sumerian)</li>\n<li>base = the base of the form (without morphological affixes, only for Sumerian)</li>\n<li>morph = the morphological affixes when present (only for Sumerian)</li>\n<li>stem = a sparsely attested label for the Akkadian verbal stem (e.g. G, D, N, Ntn, Š), for more on this see an Akkadian grammar (e.g., Huehnergard 2011 <a href=\"https://www.academia.edu/234695/2011_A_Grammar_of_Akkadian_3rd_edition_\" target=\"_blank\">https://www.academia.edu/234695/2011_A_Grammar_of_Akkadian_3rd_edition_</a>)</li>\n<li>cont = morphological decomposition of the affix (only for Sumerian)</li>\n<li>syntax_ub-after = rare instances of '-.' syntax for numbers (decimals) and Sumerian logograms (not comprehensive)</li>\n<li>morph2 = blank</li>\n<li>aform = blank</li>\n</ul>",
      "votes": null,
      "replies": []
    },
    {
      "id": 3415176,
      "author_name": "ayoubelmarhraoui",
      "author_url": "",
      "post_date": "02/28/2026 10:41:01",
      "content": "<p>Hello,</p>\n<p>To implement resources such as <strong>ORACC (Open Richly Annotated Cuneiform Corpus)</strong> and <strong>RINAP (Royal Inscriptions of the Neo-Assyrian Period)</strong> together with platforms like <strong>Hugging Face</strong> or <strong>Kaggle</strong>, the general protocol usually involves a few key steps:</p>\n<ol>\n<li><p><strong>Data access and licensing</strong>\nFirst, verify the usage terms and licenses of ORACC and RINAP datasets to ensure that redistribution and machine-learning use are allowed.</p></li>\n<li><p><strong>Data extraction and formatting</strong>\nDownload the corpus data (often available in XML, JSON, or structured text formats) and convert it into a clean, consistent format suitable for analysis or model training.</p></li>\n<li><p><strong>Preprocessing and annotation handling</strong>\nPreserve linguistic annotations (translations, transliterations, metadata, etc.) while preparing the data. This may include normalization, tokenization, and cleaning depending on the research goal.</p></li>\n<li><p><strong>Integration with ML platforms</strong></p>\n<ul>\n<li><strong>Hugging Face:</strong> convert the dataset into a compatible dataset format (e.g., using the <code>datasets</code> library) for NLP experimentation or model training.</li>\n<li><strong>Kaggle:</strong> structure the dataset into tabular or text files (CSV/JSON) and provide clear documentation so it can be used in notebooks or competitions.</li></ul></li>\n<li><p><strong>Documentation and reproducibility</strong>\nClearly document preprocessing steps, scripts, and data sources so that results can be reproduced and validated by others.</p></li>\n</ol>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "3387671": "Hello, there. I hope everyone is doing well. I'm curious to understand how to implement ORACC (Open Richly Annotated Cuneiform Corpus), RINAP (Royal Inscriptions of the Neo-Assyrian Period) and/or \nHugging Face / Kaggle. What is the proper protocol?",
    "3387893": "Hello, \n\nIt is important to note that ORACC is a project based database, and is not as comprehensive as the CDLI. It is also important to know that there are no Old Assyrian texts in ORACC, outside of a couple royal inscriptions, which are of a very different genre and style.\n\nThe good thing about ORACC is that the texts are lemmatized, which means they have been linked to an internal dictionary / lexicon and thus the data is cleaner than what we encounter in the Old Assyrian datasets presented in this challenge. If you choose to use the ORACC data and the different projects in there, you will want to pay close attention to the formatting differences, outlined in the Overview Additional Characters & Formats (you may encounter), which indicates the format of the challenge as \"Character\", the format of the text in the \"CDLI\", and the format of the text in \"ORACC\".\n\nA couple years ago I acquired all the ORACC data and placed it in this Zenodo repository, labeled finialdf.csv\nhttps://zenodo.org/records/10794626/files/factgrid_df.csv?download=1\n\n**The header fields are as follows:**\n- 0 = index\n- lang = language-dialect\n- form = transliteration form\n- id_word = unique word id\n- label = published line label of the tablet\n- id_text = unique text id (also corresponding to CDLI id, when it begins with P######)\n- delim = blank\n- gdl = for ORACC's internal use (redundant fields in an array)\n- pos = part of speech tag\n- cf = citation form, a normalization of the transliteration form\n- gw = guide word, a basic English gloss\n- sense = English sense of the word, somewhat more verbose\n- norm = standard normalization of the transliteration form (often identical to cf)\n- epos = effective part of speech, i.e. part-of-speech labels, but null values for breaks and numbers\n- headform = the lexeme used for lemmatization with Oracc's internal dictionaries (only for Sumerian)\n- contrefs = word_ids for when the same word occurs multiple times in the same text\n- norm0 = the lexeme used for lemmatization with Oracc's internal dictionaries (usually the same as headform, primarily used for Sumerian)\n- base = the base of the form (without morphological affixes, only for Sumerian)\n- morph = the morphological affixes when present (only for Sumerian)\n- stem = a sparsely attested label for the Akkadian verbal stem (e.g. G, D, N, Ntn, Š), for more on this see an Akkadian grammar (e.g., Huehnergard 2011 https://www.academia.edu/234695/2011_A_Grammar_of_Akkadian_3rd_edition_)\n-  cont = morphological decomposition of the affix (only for Sumerian)\n- syntax_ub-after = rare instances of '-.' syntax for numbers (decimals) and Sumerian logograms (not comprehensive)\n- morph2 = blank\n- aform = blank",
    "3415176": "Hello,\n\nTo implement resources such as **ORACC (Open Richly Annotated Cuneiform Corpus)** and **RINAP (Royal Inscriptions of the Neo-Assyrian Period)** together with platforms like **Hugging Face** or **Kaggle**, the general protocol usually involves a few key steps:\n\n1. **Data access and licensing**\n   First, verify the usage terms and licenses of ORACC and RINAP datasets to ensure that redistribution and machine-learning use are allowed.\n\n2. **Data extraction and formatting**\n   Download the corpus data (often available in XML, JSON, or structured text formats) and convert it into a clean, consistent format suitable for analysis or model training.\n\n3. **Preprocessing and annotation handling**\n   Preserve linguistic annotations (translations, transliterations, metadata, etc.) while preparing the data. This may include normalization, tokenization, and cleaning depending on the research goal.\n\n4. **Integration with ML platforms**\n\n   * **Hugging Face:** convert the dataset into a compatible dataset format (e.g., using the `datasets` library) for NLP experimentation or model training.\n   * **Kaggle:** structure the dataset into tabular or text files (CSV/JSON) and provide clear documentation so it can be used in notebooks or competitions.\n\n5. **Documentation and reproducibility**\n   Clearly document preprocessing steps, scripts, and data sources so that results can be reproduced and validated by others."
  },
  "source": "meta"
}