{
  "id": 132008,
  "title": "[Native Speaker please help] Component Combinations?",
  "url": "/competitions/bengaliai-cv19/discussion/132008",
  "author_name": "",
  "post_date": "2020-02-23T11:33:09.557462200Z",
  "votes": 1,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Well, it is no news now that there exists a consistent .0008-.001 gap between CV and LB. I assume that this is because some triplets (grapheme+vowel+conso) do not appear in the training set.</p>\n\n<p>Here is the problem that I need help with: would a component appear different visually if it is combined with different counterparts? e.g. would grapheme root 1 appear different when with vowel #3+conso #2 than with vowel #4+conso #1? </p>\n\n<p>Assuming that components do appear differently, how are we supposed to generalize to the test set???</p>",
  "messages": [
    {
      "id": "754311",
      "postDate": "02/23/2020 11:33:09",
      "content": "<p>Well, it is no news now that there exists a consistent .0008-.001 gap between CV and LB. I assume that this is because some triplets (grapheme+vowel+conso) do not appear in the training set.</p>\n\n<p>Here is the problem that I need help with: would a component appear different visually if it is combined with different counterparts? e.g. would grapheme root 1 appear different when with vowel #3+conso #2 than with vowel #4+conso #1? </p>\n\n<p>Assuming that components do appear differently, how are we supposed to generalize to the test set???</p>",
      "rawMarkdown": "Well, it is no news now that there exists a consistent .0008-.001 gap between CV and LB. I assume that this is because some triplets (grapheme+vowel+conso) do not appear in the training set.\n\nHere is the problem that I need help with: would a component appear different visually if it is combined with different counterparts? e.g. would grapheme root 1 appear different when with vowel #3+conso #2 than with vowel #4+conso #1? \n\nAssuming that components do appear differently, how are we supposed to generalize to the test set???",
      "votes": null
    },
    {
      "id": "760723",
      "postDate": "03/01/2020 16:50:18",
      "content": "<p>ক্ট্রো, ক্টে, ক্ট্র, ক্টা -  like this? same grapheme_root with different counterparts. </p>",
      "rawMarkdown": "ক্ট্রো, ক্টে, ক্ট্র, ক্টা -  like this? same grapheme_root with different counterparts.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 760723,
      "author_name": "ipythonx",
      "author_url": "",
      "post_date": "03/01/2020 16:50:18",
      "content": "<p>ক্ট্রো, ক্টে, ক্ট্র, ক্টা -  like this? same grapheme_root with different counterparts. </p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "754311": "Well, it is no news now that there exists a consistent .0008-.001 gap between CV and LB. I assume that this is because some triplets (grapheme+vowel+conso) do not appear in the training set.\n\nHere is the problem that I need help with: would a component appear different visually if it is combined with different counterparts? e.g. would grapheme root 1 appear different when with vowel #3+conso #2 than with vowel #4+conso #1? \n\nAssuming that components do appear differently, how are we supposed to generalize to the test set???",
    "760723": "ক্ট্রো, ক্টে, ক্ট্র, ক্টা -  like this? same grapheme_root with different counterparts."
  },
  "source": "meta"
}