{ "id": 664518, "title": "Clarification: Gap Format in Test Set ( vs ) - Did I understand correctly?", "url": "/competitions/deep-past-initiative-machine-translation/discussion/664518", "author_name": "", "post_date": "2025-12-25T22:38:38.070285700Z", "votes": 19, "comment_count": 11, "views": 0, "content": "

Hi,

\n

I've been analyzing the data formats and wanted to confirm my understanding about the gap notation in the hidden test set.

\n

Based on the discussion about formatting, here's what I understood:

\n

Gap Format Examples:

\n

Example 1 - Single gap:\nOriginal: x-4 GÍN AN.NA\nNormalized: -4 GÍN AN.NA

\n

Example 2 - Multiple gaps:\nOriginal: {large break} … x … na-áb-sú-en₆ ša ḫu-bu-li-…\nNormalized: na-áb-sú-en₆ ša ḫu-bu-li-

\n

My Observations from the Data:

\n
    \n
  1. train.csv: No gaps found (clean parallel data)
    2. \n
  2. published_texts.csv: Has two columns:
      \n
    • transliteration_orig: Original format with …, x, {large break}
      • \n
    • transliteration: Normalized format with <gap>, <big_gap>
    3. \n
\n

My Questions:

\n
    \n

  1. Is the hidden test set using the normalized format (<gap>, <big_gap>) in the transliteration column, similar to published_texts.csv ?

    2. \n

  2. For the ground truth translations, are gaps represented as ... or some other notation?

    3. \n

  3. Subscript digits (like ₄, ₅, ₆) - are these converted to normal digits (4, 5, 6) in the test set as well?

    4. \n
\n

Understanding these formatting details would help ensure our models are trained on data that matches the test set distribution.

\n

Thanks in advance for any clarification!

", "messages": [ { "id": "3381917", "postDate": "12/25/2025 22:38:38", "content": "

Hi,

\n

I've been analyzing the data formats and wanted to confirm my understanding about the gap notation in the hidden test set.

\n

Based on the discussion about formatting, here's what I understood:

\n

Gap Format Examples:

\n

Example 1 - Single gap:\nOriginal: x-4 GÍN AN.NA\nNormalized: -4 GÍN AN.NA

\n

Example 2 - Multiple gaps:\nOriginal: {large break} … x … na-áb-sú-en₆ ša ḫu-bu-li-…\nNormalized: na-áb-sú-en₆ ša ḫu-bu-li-

\n

My Observations from the Data:

\n
    \n
  1. train.csv: No gaps found (clean parallel data)
    2. \n
  2. published_texts.csv: Has two columns:
      \n
    • transliteration_orig: Original format with …, x, {large break}
      • \n
    • transliteration: Normalized format with <gap>, <big_gap>
    3. \n
\n

My Questions:

\n
    \n

  1. Is the hidden test set using the normalized format (<gap>, <big_gap>) in the transliteration column, similar to published_texts.csv ?

    2. \n

  2. For the ground truth translations, are gaps represented as ... or some other notation?

    3. \n

  3. Subscript digits (like ₄, ₅, ₆) - are these converted to normal digits (4, 5, 6) in the test set as well?

    4. \n
\n

Understanding these formatting details would help ensure our models are trained on data that matches the test set distribution.

\n

Thanks in advance for any clarification!

", "rawMarkdown": "Hi,\n\nI've been analyzing the data formats and wanted to confirm my understanding about the gap notation in the hidden test set.\n\nBased on the discussion about formatting, here's what I understood:\n\n\n### Gap Format Examples:\n**Example 1 - Single gap:**\nOriginal: x-4 GÍN AN.NA\nNormalized: -4 GÍN AN.NA\n\n**Example 2 - Multiple gaps:**\nOriginal: {large break} … x … na-áb-sú-en₆ ša ḫu-bu-li-…\nNormalized: na-áb-sú-en₆ ša ḫu-bu-li-\n\n### My Observations from the Data:\n\n1. **train.csv**: No gaps found (clean parallel data)\n2. **published_texts.csv**: Has two columns:\n - `transliteration_orig`: Original format with `…`, `x`, `{large break}`\n - `transliteration`: Normalized format with ``, ``\n\n### My Questions:\n\n1. **Is the hidden test set using the normalized format** (``, ``) in the transliteration column, similar to `published_texts.csv` ?\n\n2. **For the ground truth translations**, are gaps represented as `...` or some other notation?\n\n3. **Subscript digits** (like ₄, ₅, ₆) - are these converted to normal digits (4, 5, 6) in the test set as well?\n\nUnderstanding these formatting details would help ensure our models are trained on data that matches the test set distribution.\n\nThanks in advance for any clarification!", "votes": null }, { "id": "3382647", "postDate": "12/28/2025 07:03:52", "content": "

@deeppast Pease clarify on this

", "rawMarkdown": "deeppast Pease clarify on this", "votes": null }, { "id": "3382901", "postDate": "12/28/2025 19:16:30", "content": "

Thanks for your questions. Regarding the hidden test set:

\n
    \n
  1. Transliterations use the normalized format (< gap >, < big_gap >) for (x, [x], …, […])
    2. \n
  2. Translations also have < gap >, < big_gap > for (x, [x], …, […]) — up to two < gap > in a row, along with up to four < big_gap > combination (but no more)
    3. \n
  3. Subscript digits are converted to normal digits, except for syllables with A₂ E₂ I₂ U₂ = Á É Í Ú and A₃ E₃ I₃ U₃ = À È Ì Ù, which are represented by diacritic marks over the vowels (see overview for more details).
    4. \n
\n

Also, there are fractions (½) and not decimals (.5). I realize this poses an issue and I'm working on a supplemental dataset which includes the original integers and fractions.

", "rawMarkdown": "Thanks for your questions. Regarding the hidden test set: \n1. Transliterations use the normalized format (< gap >, < big_gap >) for (x, [x], ..., [...])\n2. Translations also have < gap >, < big_gap > for (x, [x], ..., [...]) — up to two < gap > in a row, along with up to four < big_gap > combination (but no more)\n3. Subscript digits are converted to normal digits, except for syllables with A₂ E₂ I₂ U₂ = Á É Í Ú and A₃ E₃ I₃ U₃ = À È Ì Ù, which are represented by diacritic marks over the vowels (see overview for more details).\n\nAlso, there are fractions (½) and not decimals (.5). I realize this poses an issue and I'm working on a supplemental dataset which includes the original integers and fractions.", "votes": null }, { "id": "3382919", "postDate": "12/28/2025 20:17:37", "content": "

All guides are referring to < gap > < big_gap > conversions. What is this (,) conversion about? Can you clarify?

", "rawMarkdown": "All guides are referring to < gap > < big_gap > conversions. What is this (,) conversion about? Can you clarify?", "votes": null }, { "id": "3382933", "postDate": "12/28/2025 21:05:51", "content": "

So my subscript conversion was wrong:

\n
SUBSCRIPT_MAP = str.maketrans({\n    \"₀\": \"0\", \"₁\": \"1\", \"₂\": \"2\", \"₃\": \"3\", \"₄\": \"4\",\n    \"₅\": \"5\", \"₆\": \"6\", \"₇\": \"7\", \"₈\": \"8\", \"₉\": \"9\",\n    \"ₓ\": \"x\"\n})\n# Normalize Unicode subscripts → ASCII\ntext = text.translate(SUBSCRIPT_MAP)\n
\n

It should be limited to:

\n
SUBSCRIPT_MAP = str.maketrans({\n    \"₀\": \"0\", \"₁\": \"1\", \"₄\": \"4\",\n    \"₅\": \"5\", \"₆\": \"6\", \"₇\": \"7\", \"₈\": \"8\", \"₉\": \"9\",\n    \"ₓ\": \"x\"\n})\n# Normalize Unicode subscripts → ASCII\ntext = text.translate(SUBSCRIPT_MAP)\n
\n

Transliteration normalization is not so easy.

\n
\n

Also, there are fractions (½) and not decimals (.5).

\n
\n

@deeppast Do you mean there are no decimal at all in the expected translation? Because we've a lot of 1.8333, 2.6666 …\nAs you've reported also here? \nhttps://www.kaggle.com/competitions/deep-past-initiative-machine-translation/discussion/664795#3382916

\n

For example: 0faa50f7-b86c-466c-a8ab-3a6f48fcb00a

\n
\n

I paid 4 shekels for a scarf of -weave, 1.3333 shekel for a wagonload of reed; I paid 1.6666 shekel for ; I paid 0.3333 mina of silver for an Abarnian textile, 1 / 6 shekel for a piece of dried meat, 1 / 4 shekel for onions, x shekels for x bags of wheat … … ;I paid … … … … … … shekels for ?, … … … … for a … … … … ; I paid 0.6666 shekel for firewood; 0.6666 shekels I paid for chaff, for a double load I paid 0.5 shekel; 1 / 4 shekel for the blacksmith. I supplied 95 drinks to .. at 52.5

\n
\n

Indeed, even after some sentence alignment (around 4500 sentences from 1561 document) before training, the CV / LB is so huge for me (CV=45, LB=29). I need to understand the gap.

", "rawMarkdown": "So my subscript conversion was wrong:\n\n```python\nSUBSCRIPT_MAP = str.maketrans({\n \"₀\": \"0\", \"₁\": \"1\", \"₂\": \"2\", \"₃\": \"3\", \"₄\": \"4\",\n \"₅\": \"5\", \"₆\": \"6\", \"₇\": \"7\", \"₈\": \"8\", \"₉\": \"9\",\n \"ₓ\": \"x\"\n})\n# Normalize Unicode subscripts → ASCII\ntext = text.translate(SUBSCRIPT_MAP)\n```\n\nIt should be limited to:\n```python\n\nSUBSCRIPT_MAP = str.maketrans({\n \"₀\": \"0\", \"₁\": \"1\", \"₄\": \"4\",\n \"₅\": \"5\", \"₆\": \"6\", \"₇\": \"7\", \"₈\": \"8\", \"₉\": \"9\",\n \"ₓ\": \"x\"\n})\n# Normalize Unicode subscripts → ASCII\ntext = text.translate(SUBSCRIPT_MAP)\n```\n\nTransliteration normalization is not so easy.\n\n\n>Also, there are fractions (½) and not decimals (.5). \n\n@deeppast Do you mean there are no decimal at all in the expected translation? Because we've a lot of 1.8333, 2.6666 ...\nAs you've reported also here? \nhttps://www.kaggle.com/competitions/deep-past-initiative-machine-translation/discussion/664795#3382916\n\nFor example: 0faa50f7-b86c-466c-a8ab-3a6f48fcb00a\n>I paid 4 shekels for a scarf of -weave, 1.3333 shekel for a wagonload of reed; I paid 1.6666 shekel for ; I paid 0.3333 mina of silver for an Abarnian textile, 1 / 6 shekel for a piece of dried meat, 1 / 4 shekel for onions, x shekels for x bags of wheat ... ... ;I paid ... ... ... ... ... ... shekels for ?, ... ... ... ... for a ... ... ... ... ; I paid 0.6666 shekel for firewood; 0.6666 shekels I paid for chaff, for a double load I paid 0.5 shekel; 1 / 4 shekel for the blacksmith. I supplied 95 drinks to .. at 52.5 \n\nIndeed, even after some sentence alignment (around 4500 sentences from 1561 document) before training, the CV / LB is so huge for me (CV=45, LB=29). I need to understand the gap.", "votes": null }, { "id": "3382934", "postDate": "12/28/2025 21:09:19", "content": "

Same, cv 45-48, lb 28-31. There definitely is a formatting confusion

", "rawMarkdown": "Same, cv 45-48, lb 28-31. There definitely is a formatting confusion", "votes": null }, { "id": "3382936", "postDate": "12/28/2025 21:23:02", "content": "

I guess (,) is (< gap >, < big_gap >) without spaces

", "rawMarkdown": "I guess (,) is (< gap >, < big_gap >) without spaces", "votes": null }, { "id": "3383007", "postDate": "12/29/2025 05:51:33", "content": "

I converted all instances of x [x] … etc. in the translation column of the final test predictions into the corresponding and tags, and my score rose to 30.3.

", "rawMarkdown": "I converted all instances of x [x] ... etc. in the translation column of the final test predictions into the corresponding and tags, and my score rose to 30.3.", "votes": null }, { "id": "3383015", "postDate": "12/29/2025 06:11:22", "content": "

Here CV is GM of Chrf++ and BLEU or just CHrf++/Chrf

", "rawMarkdown": "Here CV is GM of Chrf++ and BLEU or just CHrf++/Chrf", "votes": null }, { "id": "3383043", "postDate": "12/29/2025 08:10:20", "content": "

Yes, Geometric Mean

", "rawMarkdown": "Yes, Geometric Mean", "votes": null }, { "id": "3393646", "postDate": "01/19/2026 13:52:19", "content": "

How would you convert KIŠIB xxxxx ša 21 x KÙ.BABBAR ?\nI think it should be KIŠIB < big_gap > ša 21 < gap > KÙ.BABBAR but as per published_texts, it is KIŠIB < big_gap > < gap > ša 21 < gap > .\nTo me it looks like any sequence of x's like xxxxx gets converted to < big_gap > < gap > although I would expect it to be just < big_gap >

", "rawMarkdown": "How would you convert `KIŠIB xxxxx ša 21 x KÙ.BABBAR` ?\nI think it should be `KIŠIB < big_gap > ša 21 < gap > KÙ.BABBAR` but as per published_texts, it is `KIŠIB < big_gap > < gap > ša 21 < gap >` .\nTo me it looks like any sequence of x's like `xxxxx` gets converted to `< big_gap > < gap >` although I would expect it to be just `< big_gap >`", "votes": null }, { "id": "3394032", "postDate": "01/20/2026 10:11:48", "content": "

Yeah I am still confused on gap , big-gap merging and my LB scores are not giving me a clear signal of which might be right. \nAs of now I am processing xxxx -> <big_gap>

", "rawMarkdown": "Yeah I am still confused on gap , big-gap merging and my LB scores are not giving me a clear signal of which might be right. \nAs of now I am processing xxxx -> ``", "votes": null } ], "comments": [ { "id": 3382647, "author_name": "hrithik22122", "author_url": "", "post_date": "12/28/2025 07:03:52", "content": "

@deeppast Pease clarify on this

", "votes": null, "replies": [] }, { "id": 3382901, "author_name": "deeppast", "author_url": "", "post_date": "12/28/2025 19:16:30", "content": "

Thanks for your questions. Regarding the hidden test set:

\n
    \n
  1. Transliterations use the normalized format (< gap >, < big_gap >) for (x, [x], …, […])
    2. \n
  2. Translations also have < gap >, < big_gap > for (x, [x], …, […]) — up to two < gap > in a row, along with up to four < big_gap > combination (but no more)
    3. \n
  3. Subscript digits are converted to normal digits, except for syllables with A₂ E₂ I₂ U₂ = Á É Í Ú and A₃ E₃ I₃ U₃ = À È Ì Ù, which are represented by diacritic marks over the vowels (see overview for more details).
    4. \n
\n

Also, there are fractions (½) and not decimals (.5). I realize this poses an issue and I'm working on a supplemental dataset which includes the original integers and fractions.

", "votes": null, "replies": [ { "id": 3382919, "author_name": "nlztrk", "author_url": "", "post_date": "12/28/2025 20:17:37", "content": "

All guides are referring to < gap > < big_gap > conversions. What is this (,) conversion about? Can you clarify?

", "votes": null, "replies": [ { "id": 3382936, "author_name": "dunglt", "author_url": "", "post_date": "12/28/2025 21:23:02", "content": "

I guess (,) is (< gap >, < big_gap >) without spaces

", "votes": null, "replies": [] }, { "id": 3383007, "author_name": "qifeihhh666", "author_url": "", "post_date": "12/29/2025 05:51:33", "content": "

I converted all instances of x [x] … etc. in the translation column of the final test predictions into the corresponding and tags, and my score rose to 30.3.

", "votes": null, "replies": [] } ] }, { "id": 3382933, "author_name": "mpware", "author_url": "", "post_date": "12/28/2025 21:05:51", "content": "

So my subscript conversion was wrong:

\n
SUBSCRIPT_MAP = str.maketrans({\n    \"₀\": \"0\", \"₁\": \"1\", \"₂\": \"2\", \"₃\": \"3\", \"₄\": \"4\",\n    \"₅\": \"5\", \"₆\": \"6\", \"₇\": \"7\", \"₈\": \"8\", \"₉\": \"9\",\n    \"ₓ\": \"x\"\n})\n# Normalize Unicode subscripts → ASCII\ntext = text.translate(SUBSCRIPT_MAP)\n
\n

It should be limited to:

\n
SUBSCRIPT_MAP = str.maketrans({\n    \"₀\": \"0\", \"₁\": \"1\", \"₄\": \"4\",\n    \"₅\": \"5\", \"₆\": \"6\", \"₇\": \"7\", \"₈\": \"8\", \"₉\": \"9\",\n    \"ₓ\": \"x\"\n})\n# Normalize Unicode subscripts → ASCII\ntext = text.translate(SUBSCRIPT_MAP)\n
\n

Transliteration normalization is not so easy.

\n
\n

Also, there are fractions (½) and not decimals (.5).

\n
\n

@deeppast Do you mean there are no decimal at all in the expected translation? Because we've a lot of 1.8333, 2.6666 …\nAs you've reported also here? \nhttps://www.kaggle.com/competitions/deep-past-initiative-machine-translation/discussion/664795#3382916

\n

For example: 0faa50f7-b86c-466c-a8ab-3a6f48fcb00a

\n
\n

I paid 4 shekels for a scarf of -weave, 1.3333 shekel for a wagonload of reed; I paid 1.6666 shekel for ; I paid 0.3333 mina of silver for an Abarnian textile, 1 / 6 shekel for a piece of dried meat, 1 / 4 shekel for onions, x shekels for x bags of wheat … … ;I paid … … … … … … shekels for ?, … … … … for a … … … … ; I paid 0.6666 shekel for firewood; 0.6666 shekels I paid for chaff, for a double load I paid 0.5 shekel; 1 / 4 shekel for the blacksmith. I supplied 95 drinks to .. at 52.5

\n
\n

Indeed, even after some sentence alignment (around 4500 sentences from 1561 document) before training, the CV / LB is so huge for me (CV=45, LB=29). I need to understand the gap.

", "votes": null, "replies": [ { "id": 3382934, "author_name": "nlztrk", "author_url": "", "post_date": "12/28/2025 21:09:19", "content": "

Same, cv 45-48, lb 28-31. There definitely is a formatting confusion

", "votes": null, "replies": [ { "id": 3383015, "author_name": "hrithik22122", "author_url": "", "post_date": "12/29/2025 06:11:22", "content": "

Here CV is GM of Chrf++ and BLEU or just CHrf++/Chrf

", "votes": null, "replies": [ { "id": 3383043, "author_name": "mpware", "author_url": "", "post_date": "12/29/2025 08:10:20", "content": "

Yes, Geometric Mean

", "votes": null, "replies": [] } ] } ] } ] }, { "id": 3393646, "author_name": "einherjer", "author_url": "", "post_date": "01/19/2026 13:52:19", "content": "

How would you convert KIŠIB xxxxx ša 21 x KÙ.BABBAR ?\nI think it should be KIŠIB < big_gap > ša 21 < gap > KÙ.BABBAR but as per published_texts, it is KIŠIB < big_gap > < gap > ša 21 < gap > .\nTo me it looks like any sequence of x's like xxxxx gets converted to < big_gap > < gap > although I would expect it to be just < big_gap >

", "votes": null, "replies": [ { "id": 3394032, "author_name": "hrithik22122", "author_url": "", "post_date": "01/20/2026 10:11:48", "content": "

Yeah I am still confused on gap , big-gap merging and my LB scores are not giving me a clear signal of which might be right. \nAs of now I am processing xxxx -> <big_gap>

", "votes": null, "replies": [] } ] } ] } ], "raw_markdown_by_id": { "3381917": "Hi,\n\nI've been analyzing the data formats and wanted to confirm my understanding about the gap notation in the hidden test set.\n\nBased on the discussion about formatting, here's what I understood:\n\n\n### Gap Format Examples:\n**Example 1 - Single gap:**\nOriginal: x-4 GÍN AN.NA\nNormalized: -4 GÍN AN.NA\n\n**Example 2 - Multiple gaps:**\nOriginal: {large break} … x … na-áb-sú-en₆ ša ḫu-bu-li-…\nNormalized: na-áb-sú-en₆ ša ḫu-bu-li-\n\n### My Observations from the Data:\n\n1. **train.csv**: No gaps found (clean parallel data)\n2. **published_texts.csv**: Has two columns:\n - `transliteration_orig`: Original format with `…`, `x`, `{large break}`\n - `transliteration`: Normalized format with ``, ``\n\n### My Questions:\n\n1. **Is the hidden test set using the normalized format** (``, ``) in the transliteration column, similar to `published_texts.csv` ?\n\n2. **For the ground truth translations**, are gaps represented as `...` or some other notation?\n\n3. **Subscript digits** (like ₄, ₅, ₆) - are these converted to normal digits (4, 5, 6) in the test set as well?\n\nUnderstanding these formatting details would help ensure our models are trained on data that matches the test set distribution.\n\nThanks in advance for any clarification!", "3382647": "deeppast Pease clarify on this", "3382901": "Thanks for your questions. Regarding the hidden test set: \n1. Transliterations use the normalized format (< gap >, < big_gap >) for (x, [x], ..., [...])\n2. Translations also have < gap >, < big_gap > for (x, [x], ..., [...]) — up to two < gap > in a row, along with up to four < big_gap > combination (but no more)\n3. Subscript digits are converted to normal digits, except for syllables with A₂ E₂ I₂ U₂ = Á É Í Ú and A₃ E₃ I₃ U₃ = À È Ì Ù, which are represented by diacritic marks over the vowels (see overview for more details).\n\nAlso, there are fractions (½) and not decimals (.5). I realize this poses an issue and I'm working on a supplemental dataset which includes the original integers and fractions.", "3382919": "All guides are referring to < gap > < big_gap > conversions. What is this (,) conversion about? Can you clarify?", "3382933": "So my subscript conversion was wrong:\n\n```python\nSUBSCRIPT_MAP = str.maketrans({\n \"₀\": \"0\", \"₁\": \"1\", \"₂\": \"2\", \"₃\": \"3\", \"₄\": \"4\",\n \"₅\": \"5\", \"₆\": \"6\", \"₇\": \"7\", \"₈\": \"8\", \"₉\": \"9\",\n \"ₓ\": \"x\"\n})\n# Normalize Unicode subscripts → ASCII\ntext = text.translate(SUBSCRIPT_MAP)\n```\n\nIt should be limited to:\n```python\n\nSUBSCRIPT_MAP = str.maketrans({\n \"₀\": \"0\", \"₁\": \"1\", \"₄\": \"4\",\n \"₅\": \"5\", \"₆\": \"6\", \"₇\": \"7\", \"₈\": \"8\", \"₉\": \"9\",\n \"ₓ\": \"x\"\n})\n# Normalize Unicode subscripts → ASCII\ntext = text.translate(SUBSCRIPT_MAP)\n```\n\nTransliteration normalization is not so easy.\n\n\n>Also, there are fractions (½) and not decimals (.5). \n\n@deeppast Do you mean there are no decimal at all in the expected translation? Because we've a lot of 1.8333, 2.6666 ...\nAs you've reported also here? \nhttps://www.kaggle.com/competitions/deep-past-initiative-machine-translation/discussion/664795#3382916\n\nFor example: 0faa50f7-b86c-466c-a8ab-3a6f48fcb00a\n>I paid 4 shekels for a scarf of -weave, 1.3333 shekel for a wagonload of reed; I paid 1.6666 shekel for ; I paid 0.3333 mina of silver for an Abarnian textile, 1 / 6 shekel for a piece of dried meat, 1 / 4 shekel for onions, x shekels for x bags of wheat ... ... ;I paid ... ... ... ... ... ... shekels for ?, ... ... ... ... for a ... ... ... ... ; I paid 0.6666 shekel for firewood; 0.6666 shekels I paid for chaff, for a double load I paid 0.5 shekel; 1 / 4 shekel for the blacksmith. I supplied 95 drinks to .. at 52.5 \n\nIndeed, even after some sentence alignment (around 4500 sentences from 1561 document) before training, the CV / LB is so huge for me (CV=45, LB=29). I need to understand the gap.", "3382934": "Same, cv 45-48, lb 28-31. There definitely is a formatting confusion", "3382936": "I guess (,) is (< gap >, < big_gap >) without spaces", "3383007": "I converted all instances of x [x] ... etc. in the translation column of the final test predictions into the corresponding and tags, and my score rose to 30.3.", "3383015": "Here CV is GM of Chrf++ and BLEU or just CHrf++/Chrf", "3383043": "Yes, Geometric Mean", "3393646": "How would you convert `KIŠIB xxxxx ša 21 x KÙ.BABBAR` ?\nI think it should be `KIŠIB < big_gap > ša 21 < gap > KÙ.BABBAR` but as per published_texts, it is `KIŠIB < big_gap > < gap > ša 21 < gap >` .\nTo me it looks like any sequence of x's like `xxxxx` gets converted to `< big_gap > < gap >` although I would expect it to be just `< big_gap >`", "3394032": "Yeah I am still confused on gap , big-gap merging and my LB scores are not giving me a clear signal of which might be right. \nAs of now I am processing xxxx -> ``" }, "source": "meta" }