Does the hidden test follow the suggested formattings you provided? (like < gap > for ground truth translations, transforming subscript to normal digits for transliterations etc.)
", "messages": [ { "id": 3381525, "postDate": "2025-12-24T20:24:46.970Z", "content": "Does the hidden test follow the suggested formattings you provided? (like < gap > for ground truth translations, transforming subscript to normal digits for transliterations etc.)
", "rawMarkdown": "Does the hidden test follow the suggested formattings you provided? (like < gap > for ground truth translations, transforming subscript to normal digits for transliterations etc.)", "votes": 24 }, { "id": 3381529, "postDate": "2025-12-24T20:40:49.880Z", "content": "yes, the hidden test follows these formats < gap > for one signs missing (e.g., x or [x]) and < big_gap > for multiple signs missing (e.g. [x x] […] x x x … …) — Note that I added spaces around the pointy brackets and the words 'gap' and 'big_gap' here, because otherwise they are not displayed in this viewer.
", "rawMarkdown": "yes, the hidden test follows these formats < gap > for one signs missing (e.g., x or [x]) and < big_gap > for multiple signs missing (e.g. [x x] [...] x x x ... ...) — Note that I added spaces around the pointy brackets and the words 'gap' and 'big_gap' here, because otherwise they are not displayed in this viewer.", "votes": 5, "replies": [ { "id": 3381573, "postDate": "2025-12-24T23:07:46.853Z", "content": "This knowledge might bump my score a bit since I trained a model that outputs … time to find out!
\nEdit: nope!?
", "rawMarkdown": "This knowledge might bump my score a bit since I trained a model that outputsThanks! Another one:
\nShould we expect {tug₂} or () in the given test set? The train has it as (TÚG) like a third form, which transformation should I do? Some parts of the legend has the raw text on the left side and the transformed states on the right, but the last part of the legend seems the opposite. I'm really confused.
", "rawMarkdown": "Thanks! Another one:\n\nShould we expect {tug₂} or (TÚG) in the given test set? The train has it as (TÚG) like a third form, which transformation should I do? Some parts of the legend has the raw text on the left side and the transformed states on the right, but the last part of the legend seems the opposite. I'm really confused.", "votes": 4, "replies": [ { "id": 3381794, "postDate": "2025-12-25T14:45:16.687Z", "content": "Same for (d) and {d}
We can normalize it what ever the enclosing characters if we don't get an answer.
\nI'm surprised to learn that gap / big_gap is already available in test set (I don't see it in the data description).
", "rawMarkdown": "Same for `(d)` and `{d} `\n\nSee: https://www.kaggle.com/competitions/deep-past-initiative-machine-translation/discussion/664177#3381788\n\nWe can normalize it what ever the enclosing characters if we don't get an answer.\n\nI'm surprised to learn that gap / big_gap is already available in test set (I don't see it in the data description).", "votes": 2 }, { "id": 3381799, "postDate": "2025-12-25T14:57:34.253Z", "content": "gap, big gap conversion just give me a big jump (~+6) today… We need way more clearer definitions about the test set. A bunch of example translations formatted like the test set would be a good reference.
", "rawMarkdown": "gap, big gap conversion just give me a big jump (~+6) today... We need way more clearer definitions about the test set. A bunch of example translations formatted like the test set would be a good reference.", "votes": 5 }, { "id": 3381807, "postDate": "2025-12-25T15:17:49.680Z", "content": "Anil, are you formatting just the transliteration? I trained and predicted with the replacement and got worse performance.. My ideas are clearly right but the implementation is definitely wrong 😭 This is my first text-based competition so it is what it is… Also, if you join train.csv with published_texts.csv you can get transliterations with brackets, < big gap >, etc. @mpware
", "rawMarkdown": "Anil, are you formatting just the transliteration? I trained and predicted with the replacement and got worse performance.. My ideas are clearly right but the implementation is definitely wrong 😭 This is my first text-based competition so it is what it is… Also, if you join train.csv with published_texts.csv you can get transliterations with brackets, < big gap >, etc. @mpware ", "votes": 2 }, { "id": 3381808, "postDate": "2025-12-25T15:21:09.207Z", "content": "yep, just transliteration
", "rawMarkdown": "yep, just transliteration", "votes": 1 }, { "id": 3381818, "postDate": "2025-12-25T16:06:41.557Z", "content": "lol，I've been through the same thing.
", "rawMarkdown": "lol，I've been through the same thing." }, { "id": 3381858, "postDate": "2025-12-25T18:20:06.300Z", "content": "You must not be using byt5, right? I’m going to try some other stuff once I’m on my home pc… no gpu hours on kaggle and I’m traveling for the holidays.
", "rawMarkdown": "You must not be using byt5, right? I’m going to try some other stuff once I’m on my home pc… no gpu hours on kaggle and I’m traveling for the holidays. " }, { "id": 3382099, "postDate": "2025-12-26T13:59:50.767Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3382100, "postDate": "2025-12-26T14:00:00.757Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3382101, "postDate": "2025-12-26T14:00:28.343Z", "content": "@jackvd Did you use the published_texts data after merging? Did the training process show any improvement?
", "rawMarkdown": "@jackvd Did you use the published_texts data after merging? Did the training process show any improvement?" }, { "id": 3382103, "postDate": "2025-12-26T14:00:55.517Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3382165, "postDate": "2025-12-26T17:43:22.060Z", "content": "@qianyuu No, I'm working on my preprocessing while going through all the training pairs. There's a lot of weirdness that I'm planning on making a discussion post about if Anil or someone else doesn't beat me to it.. it's important to know if it exists in the hidden test and without any clarification on that we're just throwing darts a bit
", "rawMarkdown": "@qianyuu No, I'm working on my preprocessing while going through all the training pairs. There's a lot of weirdness that I'm planning on making a discussion post about if Anil or someone else doesn't beat me to it.. it's important to know if it exists in the hidden test and without any clarification on that we're just throwing darts a bit" }, { "id": 3382926, "postDate": "2025-12-28T20:35:57.890Z", "content": "Expect TÚG, it is rarely in brackets in the Old Assyrian texts. It is usually written by itself TÚG or in plural form TÚG.HI.A
", "rawMarkdown": "Expect TÚG, it is rarely in brackets in the Old Assyrian texts. It is usually written by itself TÚG or in plural form TÚG.HI.A", "votes": 2 } ] } ] }, { "id": 3383030, "postDate": "2025-12-29T07:14:50.263Z", "content": "Subscript to normal digits for transliterations? @jackvd
", "rawMarkdown": "Subscript to normal digits for transliterations? @jackvd " }, { "id": 3382574, "postDate": "2025-12-27T21:53:07.697Z", "content": "learnt from this question, thanks!
", "rawMarkdown": "learnt from this question, thanks!" } ], "comments": [ { "id": 3381529, "author_name": "Adam Anderson", "author_url": "", "post_date": "2025-12-24T20:40:49.880000", "content": "yes, the hidden test follows these formats < gap > for one signs missing (e.g., x or [x]) and < big_gap > for multiple signs missing (e.g. [x x] […] x x x … …) — Note that I added spaces around the pointy brackets and the words 'gap' and 'big_gap' here, because otherwise they are not displayed in this viewer.
", "votes": 5, "replies": [ { "id": 3381573, "author_name": "Jack", "author_url": "", "post_date": "2025-12-24T23:07:46.853000", "content": "This knowledge might bump my score a bit since I trained a model that outputs … time to find out!
\nEdit: nope!?
", "votes": 0, "replies": [] }, { "id": 3381675, "author_name": "Anil Ozturk", "author_url": "", "post_date": "2025-12-25T07:42:07.647000", "content": "Thanks! Another one:
\nShould we expect {tug₂} or () in the given test set? The train has it as (TÚG) like a third form, which transformation should I do? Some parts of the legend has the raw text on the left side and the transformed states on the right, but the last part of the legend seems the opposite. I'm really confused.
", "votes": 4, "replies": [ { "id": 3381794, "author_name": "MPWARE", "author_url": "", "post_date": "2025-12-25T14:45:16.687000", "content": "Same for (d) and {d}
We can normalize it what ever the enclosing characters if we don't get an answer.
\nI'm surprised to learn that gap / big_gap is already available in test set (I don't see it in the data description).
", "votes": 2, "replies": [] }, { "id": 3381799, "author_name": "Anil Ozturk", "author_url": "", "post_date": "2025-12-25T14:57:34.253000", "content": "gap, big gap conversion just give me a big jump (~+6) today… We need way more clearer definitions about the test set. A bunch of example translations formatted like the test set would be a good reference.
", "votes": 5, "replies": [] }, { "id": 3381807, "author_name": "Jack", "author_url": "", "post_date": "2025-12-25T15:17:49.680000", "content": "Anil, are you formatting just the transliteration? I trained and predicted with the replacement and got worse performance.. My ideas are clearly right but the implementation is definitely wrong 😭 This is my first text-based competition so it is what it is… Also, if you join train.csv with published_texts.csv you can get transliterations with brackets, < big gap >, etc. @mpware
", "votes": 2, "replies": [] }, { "id": 3381808, "author_name": "Anil Ozturk", "author_url": "", "post_date": "2025-12-25T15:21:09.207000", "content": "yep, just transliteration
", "votes": 1, "replies": [] }, { "id": 3381818, "author_name": "耶✌", "author_url": "", "post_date": "2025-12-25T16:06:41.557000", "content": "lol，I've been through the same thing.
", "votes": 0, "replies": [] }, { "id": 3381858, "author_name": "Jack", "author_url": "", "post_date": "2025-12-25T18:20:06.300000", "content": "You must not be using byt5, right? I’m going to try some other stuff once I’m on my home pc… no gpu hours on kaggle and I’m traveling for the holidays.
", "votes": 0, "replies": [] }, { "id": 3382099, "author_name": "", "author_url": "", "post_date": "2025-12-26T13:59:50.767000", "content": "", "votes": 0, "replies": [] }, { "id": 3382100, "author_name": "", "author_url": "", "post_date": "2025-12-26T14:00:00.757000", "content": "", "votes": 0, "replies": [] }, { "id": 3382101, "author_name": "QianYuu", "author_url": "", "post_date": "2025-12-26T14:00:28.343000", "content": "@jackvd Did you use the published_texts data after merging? Did the training process show any improvement?
", "votes": 0, "replies": [] }, { "id": 3382103, "author_name": "", "author_url": "", "post_date": "2025-12-26T14:00:55.517000", "content": "", "votes": 0, "replies": [] }, { "id": 3382165, "author_name": "Jack", "author_url": "", "post_date": "2025-12-26T17:43:22.060000", "content": "@qianyuu No, I'm working on my preprocessing while going through all the training pairs. There's a lot of weirdness that I'm planning on making a discussion post about if Anil or someone else doesn't beat me to it.. it's important to know if it exists in the hidden test and without any clarification on that we're just throwing darts a bit
", "votes": 0, "replies": [] }, { "id": 3382926, "author_name": "Adam Anderson", "author_url": "", "post_date": "2025-12-28T20:35:57.890000", "content": "Expect TÚG, it is rarely in brackets in the Old Assyrian texts. It is usually written by itself TÚG or in plural form TÚG.HI.A
", "votes": 2, "replies": [] } ] } ] }, { "id": 3383030, "author_name": "Navneet", "author_url": "", "post_date": "2025-12-29T07:14:50.263000", "content": "Subscript to normal digits for transliterations? @jackvd
", "votes": 0, "replies": [] }, { "id": 3382574, "author_name": "Irakoze Ntawigenga Kelly", "author_url": "", "post_date": "2025-12-27T21:53:07.697000", "content": "learnt from this question, thanks!
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "3381525": "Does the hidden test follow the suggested formattings you provided? (like < gap > for ground truth translations, transforming subscript to normal digits for transliterations etc.)", "3381529": "yes, the hidden test follows these formats < gap > for one signs missing (e.g., x or [x]) and < big_gap > for multiple signs missing (e.g. [x x] [...] x x x ... ...) — Note that I added spaces around the pointy brackets and the words 'gap' and 'big_gap' here, because otherwise they are not displayed in this viewer.", "3383030": "Subscript to normal digits for transliterations? @jackvd ", "3382574": "learnt from this question, thanks!" } }