{
  "id": 669062,
  "title": "Translation Alignment",
  "url": "/competitions/deep-past-initiative-machine-translation/discussion/669062",
  "author_name": "",
  "post_date": "2026-01-20T12:16:17.978189Z",
  "votes": null,
  "comment_count": 3,
  "views": 0,
  "content": "<p>Iam just starting in this competition, from my understanding of the data descritption, there is no ready data for model training? We need to extract the translations from the pdfs?</p>\n<p>Since the competition is already going for 1 month. Iam curious wheter somebody published code for a transliteration and translation alignment? I don't know want to do unnecessery work that has already been done. </p>\n<p>Thanks.</p>",
  "messages": [
    {
      "id": "3394090",
      "postDate": "01/20/2026 12:16:17",
      "content": "<p>Iam just starting in this competition, from my understanding of the data descritption, there is no ready data for model training? We need to extract the translations from the pdfs?</p>\n<p>Since the competition is already going for 1 month. Iam curious wheter somebody published code for a transliteration and translation alignment? I don't know want to do unnecessery work that has already been done. </p>\n<p>Thanks.</p>",
      "rawMarkdown": "Iam just starting in this competition, from my understanding of the data descritption, there is no ready data for model training? We need to extract the translations from the pdfs?\n\nSince the competition is already going for 1 month. Iam curious wheter somebody published code for a transliteration and translation alignment? I don't know want to do unnecessery work that has already been done. \n\nThanks.",
      "votes": null
    },
    {
      "id": "3394143",
      "postDate": "01/20/2026 14:05:42",
      "content": "<p>Not entirely true. We have around 1500 transliteration-translation pairs in <code>train.csv</code> but<a href=\"https://www.kaggle.com/competitions/deep-past-initiative-machine-translation/discussion/667483\" target=\"_blank\"> some of the texts are not aligned (~25% have either too short translations or too short transliterations)</a>.</p>\n<p>There are some texts in the <code>publications.csv</code> with translations in multiple languages, which you mentioned.</p>\n<p>There is also around 8000 transliterations in <code>published_texts.csv</code>, which at the very least could be used to learn the tokenizer.</p>",
      "rawMarkdown": "Not entirely true. We have around 1500 transliteration-translation pairs in `train.csv` but[ some of the texts are not aligned (~25% have either too short translations or too short transliterations)](https://www.kaggle.com/competitions/deep-past-initiative-machine-translation/discussion/667483).\n\nThere are some texts in the `publications.csv` with translations in multiple languages, which you mentioned.\n\nThere is also around 8000 transliterations in `published_texts.csv`, which at the very least could be used to learn the tokenizer.",
      "votes": null
    },
    {
      "id": "3394277",
      "postDate": "01/20/2026 18:59:55",
      "content": "<p>Ok, my question wasn't correctly worded. Primarily my question was wheter someone publicly aligned publications.csv with published_texts.csv.</p>",
      "rawMarkdown": "Ok, my question wasn't correctly worded. Primarily my question was wheter someone publicly aligned publications.csv with published_texts.csv.",
      "votes": null
    },
    {
      "id": "3395008",
      "postDate": "01/22/2026 04:36:19",
      "content": "<p>I don't think that has happened publicly yet, but if you would like to share your results I could help validate them publicly.</p>",
      "rawMarkdown": "I don't think that has happened publicly yet, but if you would like to share your results I could help validate them publicly.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 3394143,
      "author_name": "angantyr",
      "author_url": "",
      "post_date": "01/20/2026 14:05:42",
      "content": "<p>Not entirely true. We have around 1500 transliteration-translation pairs in <code>train.csv</code> but<a href=\"https://www.kaggle.com/competitions/deep-past-initiative-machine-translation/discussion/667483\" target=\"_blank\"> some of the texts are not aligned (~25% have either too short translations or too short transliterations)</a>.</p>\n<p>There are some texts in the <code>publications.csv</code> with translations in multiple languages, which you mentioned.</p>\n<p>There is also around 8000 transliterations in <code>published_texts.csv</code>, which at the very least could be used to learn the tokenizer.</p>",
      "votes": null,
      "replies": [
        {
          "id": 3394277,
          "author_name": "luciankucera",
          "author_url": "",
          "post_date": "01/20/2026 18:59:55",
          "content": "<p>Ok, my question wasn't correctly worded. Primarily my question was wheter someone publicly aligned publications.csv with published_texts.csv.</p>",
          "votes": null,
          "replies": [
            {
              "id": 3395008,
              "author_name": "deeppast",
              "author_url": "",
              "post_date": "01/22/2026 04:36:19",
              "content": "<p>I don't think that has happened publicly yet, but if you would like to share your results I could help validate them publicly.</p>",
              "votes": null,
              "replies": []
            }
          ]
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "3394090": "Iam just starting in this competition, from my understanding of the data descritption, there is no ready data for model training? We need to extract the translations from the pdfs?\n\nSince the competition is already going for 1 month. Iam curious wheter somebody published code for a transliteration and translation alignment? I don't know want to do unnecessery work that has already been done. \n\nThanks.",
    "3394143": "Not entirely true. We have around 1500 transliteration-translation pairs in `train.csv` but[ some of the texts are not aligned (~25% have either too short translations or too short transliterations)](https://www.kaggle.com/competitions/deep-past-initiative-machine-translation/discussion/667483).\n\nThere are some texts in the `publications.csv` with translations in multiple languages, which you mentioned.\n\nThere is also around 8000 transliterations in `published_texts.csv`, which at the very least could be used to learn the tokenizer.",
    "3394277": "Ok, my question wasn't correctly worded. Primarily my question was wheter someone publicly aligned publications.csv with published_texts.csv.",
    "3395008": "I don't think that has happened publicly yet, but if you would like to share your results I could help validate them publicly."
  },
  "source": "meta"
}