{ "id": 667480, "title": "Floating Point Number Questions", "url": "/competitions/deep-past-initiative-machine-translation/discussion/667480", "author_name": "justinlamlamlam", "post_date": "2026-01-12T21:02:00.190000", "votes": 18, "comment_count": 18, "views": 0, "content": "

Hi @deeppast . \nI have a few questions. Any insights on them would be super helpful.\nI have found that there are a quite few long floating point numbers, which is likely due to floating point errors. To give a few examples

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
oare_idtransliterationtranslation
672697fdae3-abd6-4de6-89ac-03796381be510.33333 ma-na 1.3333300000000001 GÍN KÙ.BABBAR ṣa-ru-pá-am i li-bi šé-ep-IŠTAR 1 GÍN iš-tí e-wa-ri-mu-ša0.3333 mina 1.3333 shekel of refined silver is in Šēp-Ištar's possession.
1301d1bc99fc-3f4b-476d-8c28-f072aa2800833.6666600000000003 ma-na 2 GÍN AN.NA i li-bi₄ a-šur-SIPA IGI a-šur-dan IGI ir-ma-a-šur3 minas 42 shekels of tin owed by Aššur-rē'ī. In the presence of Aššur-dān, of Ir'am-Aššūr.
1458ef2c3a21-2451-436b-b1a3-16b8574ff27a1.3333300000000001 ma-na KÙ.BABBAR a-mu-ru-um i li-bi₄ a-da-da x um … DUMU šu-… 0.33333 ma-na KÙ.BABBAR ta-… ša a-lá-ḫi-im iš-tù ḫa-mu-uš-tim ša kur-ub-IŠTAR DINGIR ša-pá-tám i-li-ik KÙ.BABBAR il₅-qé1.3333 mina of checked silver is owed by Adada.
9509676573d-7242-4ee0-a393-fea152d64de9xx GÍN KÙ.BABBAR a-na du-li i-na ú-lá-ma a-dí-in um-ma šu-ut-ma ku-ta-nam ú-šé-ba-lá-kum 1.3333300000000001 ma-na KÙ.BABBAR KI ḫa-lu-a x xI gave Duli in Ulama [x] shekels of silver. He said: I will send you one -textile.\" 1.3333 mina of silver is with halua …\"
1339d93fadca-089c-4135-8228-bc755873b2bf10.66666 GÍN a-šùr-ma-lik i-lá-qé 0.66666 ma-na 0.83333 GÍN kàr-wa-a 11 GÍN iṣ-mì-id-DINGIR 0.5 ma-na 3.6666600000000003 GÍN šu-ku-bu-um 15.33333 GÍN bu-zu-ta-a10.6666 shekels Aššur-mālik will take, 40.83333 shekels Karwaya, 11 shekels Iṣmid-Ilum, 33.6666 shekels Šu-Kūbum, 15.3333 shekels Buztāya
\n

There are many more containing such floating point numbers. I could find 79 rows with 9 consecutive 0s with this code.

\n
df = pd.read_csv(\"/kaggle/input/deep-past-initiative-machine-translation/train.csv\")\ndf = df[df['transliteration'].str.contains('000000000', na=False)]\n
\n

Will such numbers appear in the test set? \nHow should we translate such numbers? Sometimes these numbers are translated to fractions, and sometimes they are translated to numbers with inconsistent decimal places.

", "messages": [ { "id": 3390252, "postDate": "2026-01-12T21:02:00.190Z", "content": "

Hi @deeppast . \nI have a few questions. Any insights on them would be super helpful.\nI have found that there are a quite few long floating point numbers, which is likely due to floating point errors. To give a few examples

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
oare_idtransliterationtranslation
672697fdae3-abd6-4de6-89ac-03796381be510.33333 ma-na 1.3333300000000001 GÍN KÙ.BABBAR ṣa-ru-pá-am i li-bi šé-ep-IŠTAR 1 GÍN iš-tí e-wa-ri-mu-ša0.3333 mina 1.3333 shekel of refined silver is in Šēp-Ištar's possession.
1301d1bc99fc-3f4b-476d-8c28-f072aa2800833.6666600000000003 ma-na 2 GÍN AN.NA i li-bi₄ a-šur-SIPA IGI a-šur-dan IGI ir-ma-a-šur3 minas 42 shekels of tin owed by Aššur-rē'ī. In the presence of Aššur-dān, of Ir'am-Aššūr.
1458ef2c3a21-2451-436b-b1a3-16b8574ff27a1.3333300000000001 ma-na KÙ.BABBAR a-mu-ru-um i li-bi₄ a-da-da x um … DUMU šu-… 0.33333 ma-na KÙ.BABBAR ta-… ša a-lá-ḫi-im iš-tù ḫa-mu-uš-tim ša kur-ub-IŠTAR DINGIR ša-pá-tám i-li-ik KÙ.BABBAR il₅-qé1.3333 mina of checked silver is owed by Adada.
9509676573d-7242-4ee0-a393-fea152d64de9xx GÍN KÙ.BABBAR a-na du-li i-na ú-lá-ma a-dí-in um-ma šu-ut-ma ku-ta-nam ú-šé-ba-lá-kum 1.3333300000000001 ma-na KÙ.BABBAR KI ḫa-lu-a x xI gave Duli in Ulama [x] shekels of silver. He said: I will send you one -textile.\" 1.3333 mina of silver is with halua …\"
1339d93fadca-089c-4135-8228-bc755873b2bf10.66666 GÍN a-šùr-ma-lik i-lá-qé 0.66666 ma-na 0.83333 GÍN kàr-wa-a 11 GÍN iṣ-mì-id-DINGIR 0.5 ma-na 3.6666600000000003 GÍN šu-ku-bu-um 15.33333 GÍN bu-zu-ta-a10.6666 shekels Aššur-mālik will take, 40.83333 shekels Karwaya, 11 shekels Iṣmid-Ilum, 33.6666 shekels Šu-Kūbum, 15.3333 shekels Buztāya
\n

There are many more containing such floating point numbers. I could find 79 rows with 9 consecutive 0s with this code.

\n
df = pd.read_csv(\"/kaggle/input/deep-past-initiative-machine-translation/train.csv\")\ndf = df[df['transliteration'].str.contains('000000000', na=False)]\n
\n

Will such numbers appear in the test set? \nHow should we translate such numbers? Sometimes these numbers are translated to fractions, and sometimes they are translated to numbers with inconsistent decimal places.

", "rawMarkdown": "Hi @deeppast . \nI have a few questions. Any insights on them would be super helpful.\nI have found that there are a quite few long floating point numbers, which is likely due to floating point errors. To give a few examples\n\n| | oare_id | transliteration | translation |\n|-----:|:-------------------------------------|:--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|:------------------------------------------------------------------------------------------------------------------------------------------|\n| 672 | 697fdae3-abd6-4de6-89ac-03796381be51 | 0.33333 ma-na 1.3333300000000001 GÍN KÙ.BABBAR ṣa-ru-pá-am i li-bi šé-ep-IŠTAR 1 GÍN iš-tí e-wa-ri-mu-ša | 0.3333 mina 1.3333 shekel of refined silver is in Šēp-Ištar's possession. |\n| 1301 | d1bc99fc-3f4b-476d-8c28-f072aa280083 | 3.6666600000000003 ma-na 2 GÍN AN.NA i li-bi₄ a-šur-SIPA IGI a-šur-dan IGI ir-ma-a-šur | 3 minas 42 shekels of tin owed by Aššur-rē'ī. In the presence of Aššur-dān, of Ir'am-Aššūr. |\n| 1458 | ef2c3a21-2451-436b-b1a3-16b8574ff27a | 1.3333300000000001 ma-na KÙ.BABBAR a-mu-ru-um i li-bi₄ a-da-da x um … DUMU šu-… 0.33333 ma-na KÙ.BABBAR ta-… ša a-lá-ḫi-im iš-tù ḫa-mu-uš-tim ša kur-ub-IŠTAR DINGIR ša-pá-tám i-li-ik KÙ.BABBAR il₅-qé | 1.3333 mina of checked silver is owed by Adada. |\n| 950 | 9676573d-7242-4ee0-a393-fea152d64de9 | xx GÍN KÙ.BABBAR a-na du-li i-na ú-lá-ma a-dí-in um-ma šu-ut-ma ku-ta-nam ú-šé-ba-lá-kum 1.3333300000000001 ma-na KÙ.BABBAR KI ḫa-lu-a x x | I gave Duli in Ulama [x] shekels of silver. He said: I will send you one -textile.\" 1.3333 mina of silver is with halua ...\" |\n| 1339 | d93fadca-089c-4135-8228-bc755873b2bf | 10.66666 GÍN a-šùr-ma-lik i-lá-qé 0.66666 ma-na 0.83333 GÍN kàr-wa-a 11 GÍN iṣ-mì-id-DINGIR 0.5 ma-na 3.6666600000000003 GÍN šu-ku-bu-um 15.33333 GÍN bu-zu-ta-a | 10.6666 shekels Aššur-mālik will take, 40.83333 shekels Karwaya, 11 shekels Iṣmid-Ilum, 33.6666 shekels Šu-Kūbum, 15.3333 shekels Buztāya |\n\n\nThere are many more containing such floating point numbers. I could find 79 rows with 9 consecutive 0s with this code.\n```\ndf = pd.read_csv(\"/kaggle/input/deep-past-initiative-machine-translation/train.csv\")\ndf = df[df['transliteration'].str.contains('000000000', na=False)]\n```\nWill such numbers appear in the test set? \nHow should we translate such numbers? Sometimes these numbers are translated to fractions, and sometimes they are translated to numbers with inconsistent decimal places. \n", "votes": 18 }, { "id": 3390651, "postDate": "2026-01-13T15:55:15.177Z", "content": "

No, there are no floats in the test set. I would encourage you to convert the decimals to fractions. Their system is base-60 so these fractions are predictable, these are the only ones you'll encounter:

\n\n

If you consider that 1 mina = 60 shekels, then you get the following conversion into shekels:

\n

% of mina (ma-na) in shekels:

\n", "rawMarkdown": "No, there are no floats in the test set. I would encourage you to convert the decimals to fractions. Their system is base-60 so these fractions are predictable, these are the only ones you'll encounter:\n\n- ½\t= 0.5\n- ¼\t= 0.25\n- ⅓\t= 0.333...\n- ⅔\t= 0.666...\n- ⅙\t= 0.166...\n- ⅚\t= 0.833...\n\nIf you consider that 1 mina = 60 shekels, then you get the following conversion into shekels:\n\n% of mina (ma-na) in shekels:\n- ½\t= 0.5\t\t= 30 shekels\n- ¼\t= 0.25\t\t= 15 shekels\n- ⅓\t= 0.333...\t= 20 shekels\n- ⅔\t= 0.666...\t= 40 shekels\n- ⅙\t= 0.166...\t\t= 10 shekels\n- ⅚\t= 0.833...\t= 50 shekels", "votes": 7, "replies": [ { "id": 3390711, "postDate": "2026-01-13T20:26:00.577Z", "content": "

Hi, what's the relationship between grains (ŠE) and mina and shekels?

", "rawMarkdown": "Hi, what's the relationship between grains (ŠE) and mina and shekels?", "replies": [ { "id": 3390722, "postDate": "2026-01-13T21:04:05.837Z", "content": "

180 ŠE equals one GIN (shekel); sixty shekels equal one (MA.NA) mina; 60 mina equal one talent (GU2)

\n

See also:

\n

https://shs.hal.science/halshs-03923497/file/12-Michel-CulturesComputationQuantification.pdf

", "rawMarkdown": "180 ŠE equals one GIN (shekel); sixty shekels equal one (MA.NA) mina; 60 mina equal one talent (GU2)\n\nSee also:\n\nhttps://shs.hal.science/halshs-03923497/file/12-Michel-CulturesComputationQuantification.pdf", "votes": 5 }, { "id": 3390724, "postDate": "2026-01-13T21:07:01.893Z", "content": "

Thanks! quite difficult to decipher the units when translators are doing the math in their heads…

", "rawMarkdown": "Thanks! quite difficult to decipher the units when translators are doing the math in their heads..." }, { "id": 3390732, "postDate": "2026-01-13T21:34:12.567Z", "content": "

You're welcome and I know. We translators also often make mistakes because we mess up addition/subtraction of fractions in the sexagesimal system as we jump units –– e.g. 3 5/6 mana equals 180+50 shekels. Now add 3/4 mana 2 shekels 15 ŠE.

\n

When I torture my students I give them the following, authentic school test frpm around 1730 BCE:

\n

\"I found a stone I did not weigh it. I added 1/7 to its weight. I added 1/11. I weighed the stone - it was 1 mina. How much did it weigh before?\"

\n

I always promise an A grade for the entire class if they can solve it within 60 seconds. For over a decade there have been no takers.

", "rawMarkdown": "You're welcome and I know. We translators also often make mistakes because we mess up addition/subtraction of fractions in the sexagesimal system as we jump units –– e.g. 3 5/6 mana equals 180+50 shekels. Now add 3/4 mana 2 shekels 15 ŠE. \n\nWhen I torture my students I give them the following, authentic school test frpm around 1730 BCE: \n\n\"*I found a stone I did not weigh it. I added 1/7 to its weight. I added 1/11. I weighed the stone - it was 1 mina. How much did it weigh before?*\" \n\nI always promise an A grade for the entire class if they can solve it within 60 seconds. For over a decade there have been no takers.", "votes": 4 }, { "id": 3390745, "postDate": "2026-01-13T22:12:36.140Z", "content": "

If I may ask another number question, I saw a lot of examples of this (ITU.KAM ṣí-ip-im) - (Month VII), which I assume here ṣí-ip-im means VII somehow, is there any reference material on this conversion? So I can understand what each of these corresponds to (I, II, III, IV, V, etc.)

", "rawMarkdown": "If I may ask another number question, I saw a lot of examples of this (ITU.KAM ṣí-ip-im) - (Month VII), which I assume here ṣí-ip-im means VII somehow, is there any reference material on this conversion? So I can understand what each of these corresponds to (I, II, III, IV, V, etc.)", "votes": 2 }, { "id": 3390747, "postDate": "2026-01-13T22:22:40.100Z", "content": "

The Roman numeral refers to the month # -- so, \"ṣip'um\" is month seven -- meaning that you could translate \"ITU.KAM ṣí-ip-im\" as \"month seven\" or \"the month of Ṣip'um\" or similar.

\n

For the entire calendar, see:

\n

https://cdli.ox.ac.uk/wiki/doku.php?id=the_old_assyrian_calendar

", "rawMarkdown": "The Roman numeral refers to the month # -- so, \"ṣip'um\" is month seven -- meaning that you could translate \"ITU.KAM ṣí-ip-im\" as \"month seven\" or \"the month of Ṣip'um\" or similar.\n\nFor the entire calendar, see:\n\nhttps://cdli.ox.ac.uk/wiki/doku.php?id=the_old_assyrian_calendar", "votes": 3 }, { "id": 3390784, "postDate": "2026-01-14T00:07:58.823Z", "content": "

This is an important question, and it should be addressed in terms of the test data as well. The train file doesn't have a lot of month names present, however the published_texts.csv does contain all of these month names. The first month is not attested in the test set, it is rarely mentioned, as discussed in chronological studies, such as Stratford (2017) A Year of Vengance.

\n

The roman numerals (Roman) were converted to integers for the test set, so it would be wise to know that when Month X is mentioned in translation, it should be converted to Month 10 (Month). You can use the following table for the conversions, and I also gathered a number of the written forms (forms) for these months as well as their aliases in translations (AKA).

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
MonthRomanMNformsAKA
Month 2IIIša-sarratimša sá-ra-tim
Month 3IIIKenātimke-na-timša kēnātim
Month 4IVMahur-ilīMa-hu-ur-DINGIR; ma-ḫu-ur-ì-lí
Month 5VAbšarraniáb-ša-ra-ni; áb ša-ra-ni; áb-ša-ra-nuab šarrāni; abšarrani
Month 6VIHuburHu-bu-ur
Month 7VIIṢip'umṣí-ip-imṣipum
Month 8VIIIQarrātumqá-ra-a-tí; qá-ra-a-tim
Month 9IXKanwartakán-bar-ta; Kà-an-ma-ar-taKanmarta
Month 10XTe’inātumté-i-na-tim
Month 11XIKuzallumku-zal-li; ku-zal-lu
Month 12XIIAllanātuma-lá-na-tum; a-lá-na-tim
", "rawMarkdown": "This is an important question, and it should be addressed in terms of the test data as well. The train file doesn't have a lot of month names present, however the published_texts.csv does contain all of these month names. The first month is not attested in the test set, it is rarely mentioned, as discussed in chronological studies, such as Stratford (2017) A Year of Vengance.\n\nThe roman numerals (Roman) were converted to integers for the test set, so it would be wise to know that when Month X is mentioned in translation, it should be converted to Month 10 (Month). You can use the following table for the conversions, and I also gathered a number of the written forms (forms) for these months as well as their aliases in translations (AKA).\n\n| Month | Roman | MN | forms | AKA |\n|--------|--------|-----|-------|------|\n|Month 2|\tIII |\tša-sarratim |\tša sá-ra-tim|\t|\n|Month 3|\tIII |\tKenātim |\tke-na-tim |\tša kēnātim|\n|Month 4|\tIV |\tMahur-ilī |\tMa-hu-ur-DINGIR; ma-ḫu-ur-ì-lí |\t|\n|Month 5|\tV |\tAbšarrani |\táb-ša-ra-ni; áb ša-ra-ni; áb-ša-ra-nu |\tab šarrāni; abšarrani |\n|Month 6|\tVI |\tHubur |\tHu-bu-ur\t| |\n|Month 7|\tVII |\tṢip'um |\tṣí-ip-im |\tṣipum|\n|Month 8|\tVIII |\tQarrātum |\tqá-ra-a-tí; qá-ra-a-tim |\t|\n|Month 9|\tIX |\tKanwarta |\tkán-bar-ta; Kà-an-ma-ar-ta |\tKanmarta |\n|Month 10|\tX |\tTe’inātum |\tté-i-na-tim |\t|\n|Month 11|\tXI |\tKuzallum |\tku-zal-li; ku-zal-lu |\t|\n|Month 12|\tXII |\tAllanātum |\ta-lá-na-tum; a-lá-na-tim |\t|", "votes": 8 }, { "id": 3390796, "postDate": "2026-01-14T00:40:46.290Z", "content": "

Thanks. I believe this is also useful for others.

", "rawMarkdown": "Thanks. I believe this is also useful for others.", "votes": 2 }, { "id": 3390982, "postDate": "2026-01-14T08:36:52.840Z", "rawMarkdown": "", "isDeleted": true }, { "id": 3390983, "postDate": "2026-01-14T08:40:25.490Z", "content": "
\n

When I torture my students I give them the following, authentic school test frpm around 1730 BCE:

\n

\"I found a stone I did not weigh it. I added 1/7 to its weight. I added 1/11. I weighed the stone - it was 1 mina. How much did it weigh before?\"

\n
\n

The answer to be given in mina, shekels, grains or everything properly divided into all three?

", "rawMarkdown": "> When I torture my students I give them the following, authentic school test frpm around 1730 BCE: \n> \n> \"*I found a stone I did not weigh it. I added 1/7 to its weight. I added 1/11. I weighed the stone - it was 1 mina. How much did it weigh before?*\" \n\nThe answer to be given in mina, shekels, grains or everything properly divided into all three?" }, { "id": 3394513, "postDate": "2026-01-21T07:16:55.250Z", "content": "

There are also cases like \"ITU.1.KAM na-ar-ma-ak a-šur\".

\n

Without the next word \"Sarrātim\" or \"Kēnātim\" this could be month 2 or 3. How should we deal with this case?

", "rawMarkdown": "There are also cases like \"ITU.1.KAM na-ar-ma-ak a-šur\".\n\nWithout the next word \"Sarrātim\" or \"Kēnātim\" this could be month 2 or 3. How should we deal with this case?" }, { "id": 3394544, "postDate": "2026-01-21T08:52:42.977Z", "content": "

@deeppast

\n

Transliteration : ITU.1.KAM áb-ša-ra-ni li-mu-um šu-sú-in DUMU ba-bi-lim

\n

Which would be the correct translation according to test set formatting

\n
 Month Ab-šarrāni, in the eponymy Šu-Suen son of Babilum.\n Month Abšarrani, in the eponymy Šu-Suen son of Babilum.\n Month V, in the eponymy of Šu-Suen, son of Babilum.\n Month 5, in the eponymy of Šu-Suen, son of Babilum.\n
", "rawMarkdown": "@deeppast \n\nTransliteration : ITU.1.KAM áb-ša-ra-ni li-mu-um šu-sú-in DUMU ba-bi-lim\n\nWhich would be the correct translation according to test set formatting \n```\n Month Ab-šarrāni, in the eponymy Šu-Suen son of Babilum.\n Month Abšarrani, in the eponymy Šu-Suen son of Babilum.\n Month V, in the eponymy of Šu-Suen, son of Babilum.\n Month 5, in the eponymy of Šu-Suen, son of Babilum.\n```", "votes": 3 }, { "id": 3394805, "postDate": "2026-01-21T17:25:14.817Z", "content": "

my guess is the last one

", "rawMarkdown": "my guess is the last one" }, { "id": 3394811, "postDate": "2026-01-21T17:36:17.643Z", "content": "

Thinking the same , but just to be sure

", "rawMarkdown": "Thinking the same , but just to be sure " }, { "id": 3394822, "postDate": "2026-01-21T18:01:20.733Z", "content": "

What can be confirmed so far is that there is no third option.

", "rawMarkdown": "What can be confirmed so far is that there is no third option." }, { "id": 3394905, "postDate": "2026-01-21T20:59:48.347Z", "content": "

Exactly, it definitely won't be Month V. The translators were not consistent with the name normalization (hence the onomasticon I provided), and they weren't consistent with the conversion of month names to month numbers (hence my month name to number chart). In general, it's more common to see the month name rather than the number.

", "rawMarkdown": "Exactly, it definitely won't be Month V. The translators were not consistent with the name normalization (hence the onomasticon I provided), and they weren't consistent with the conversion of month names to month numbers (hence my month name to number chart). In general, it's more common to see the month name rather than the number.", "votes": 1 } ] }, { "id": 3390737, "postDate": "2026-01-13T21:45:47.577Z", "content": "

Thank you for the response! This is very helpful.

", "rawMarkdown": "Thank you for the response! This is very helpful. " } ] } ], "comments": [ { "id": 3390651, "author_name": "Adam Anderson", "author_url": "", "post_date": "2026-01-13T15:55:15.177000", "content": "

No, there are no floats in the test set. I would encourage you to convert the decimals to fractions. Their system is base-60 so these fractions are predictable, these are the only ones you'll encounter:

\n\n

If you consider that 1 mina = 60 shekels, then you get the following conversion into shekels:

\n

% of mina (ma-na) in shekels:

\n", "votes": 7, "replies": [ { "id": 3390711, "author_name": "hongan", "author_url": "", "post_date": "2026-01-13T20:26:00.577000", "content": "

Hi, what's the relationship between grains (ŠE) and mina and shekels?

", "votes": 0, "replies": [ { "id": 3390722, "author_name": "Cackler", "author_url": "", "post_date": "2026-01-13T21:04:05.837000", "content": "

180 ŠE equals one GIN (shekel); sixty shekels equal one (MA.NA) mina; 60 mina equal one talent (GU2)

\n

See also:

\n

https://shs.hal.science/halshs-03923497/file/12-Michel-CulturesComputationQuantification.pdf

", "votes": 5, "replies": [] }, { "id": 3390724, "author_name": "hongan", "author_url": "", "post_date": "2026-01-13T21:07:01.893000", "content": "

Thanks! quite difficult to decipher the units when translators are doing the math in their heads…

", "votes": 0, "replies": [] }, { "id": 3390732, "author_name": "Cackler", "author_url": "", "post_date": "2026-01-13T21:34:12.567000", "content": "

You're welcome and I know. We translators also often make mistakes because we mess up addition/subtraction of fractions in the sexagesimal system as we jump units –– e.g. 3 5/6 mana equals 180+50 shekels. Now add 3/4 mana 2 shekels 15 ŠE.

\n

When I torture my students I give them the following, authentic school test frpm around 1730 BCE:

\n

\"I found a stone I did not weigh it. I added 1/7 to its weight. I added 1/11. I weighed the stone - it was 1 mina. How much did it weigh before?\"

\n

I always promise an A grade for the entire class if they can solve it within 60 seconds. For over a decade there have been no takers.

", "votes": 4, "replies": [] }, { "id": 3390745, "author_name": "hongan", "author_url": "", "post_date": "2026-01-13T22:12:36.140000", "content": "

If I may ask another number question, I saw a lot of examples of this (ITU.KAM ṣí-ip-im) - (Month VII), which I assume here ṣí-ip-im means VII somehow, is there any reference material on this conversion? So I can understand what each of these corresponds to (I, II, III, IV, V, etc.)

", "votes": 2, "replies": [] }, { "id": 3390747, "author_name": "Cackler", "author_url": "", "post_date": "2026-01-13T22:22:40.100000", "content": "

The Roman numeral refers to the month # -- so, \"ṣip'um\" is month seven -- meaning that you could translate \"ITU.KAM ṣí-ip-im\" as \"month seven\" or \"the month of Ṣip'um\" or similar.

\n

For the entire calendar, see:

\n

https://cdli.ox.ac.uk/wiki/doku.php?id=the_old_assyrian_calendar

", "votes": 3, "replies": [] }, { "id": 3390784, "author_name": "Adam Anderson", "author_url": "", "post_date": "2026-01-14T00:07:58.823000", "content": "

This is an important question, and it should be addressed in terms of the test data as well. The train file doesn't have a lot of month names present, however the published_texts.csv does contain all of these month names. The first month is not attested in the test set, it is rarely mentioned, as discussed in chronological studies, such as Stratford (2017) A Year of Vengance.

\n

The roman numerals (Roman) were converted to integers for the test set, so it would be wise to know that when Month X is mentioned in translation, it should be converted to Month 10 (Month). You can use the following table for the conversions, and I also gathered a number of the written forms (forms) for these months as well as their aliases in translations (AKA).

\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n
MonthRomanMNformsAKA
Month 2IIIša-sarratimša sá-ra-tim
Month 3IIIKenātimke-na-timša kēnātim
Month 4IVMahur-ilīMa-hu-ur-DINGIR; ma-ḫu-ur-ì-lí
Month 5VAbšarraniáb-ša-ra-ni; áb ša-ra-ni; áb-ša-ra-nuab šarrāni; abšarrani
Month 6VIHuburHu-bu-ur
Month 7VIIṢip'umṣí-ip-imṣipum
Month 8VIIIQarrātumqá-ra-a-tí; qá-ra-a-tim
Month 9IXKanwartakán-bar-ta; Kà-an-ma-ar-taKanmarta
Month 10XTe’inātumté-i-na-tim
Month 11XIKuzallumku-zal-li; ku-zal-lu
Month 12XIIAllanātuma-lá-na-tum; a-lá-na-tim
", "votes": 8, "replies": [] }, { "id": 3390796, "author_name": "hongan", "author_url": "", "post_date": "2026-01-14T00:40:46.290000", "content": "

Thanks. I believe this is also useful for others.

", "votes": 2, "replies": [] }, { "id": 3390982, "author_name": "", "author_url": "", "post_date": "2026-01-14T08:36:52.840000", "content": "", "votes": 0, "replies": [] }, { "id": 3390983, "author_name": "Angantyr", "author_url": "", "post_date": "2026-01-14T08:40:25.490000", "content": "
\n

When I torture my students I give them the following, authentic school test frpm around 1730 BCE:

\n

\"I found a stone I did not weigh it. I added 1/7 to its weight. I added 1/11. I weighed the stone - it was 1 mina. How much did it weigh before?\"

\n
\n

The answer to be given in mina, shekels, grains or everything properly divided into all three?

", "votes": 0, "replies": [] }, { "id": 3394513, "author_name": "Tomorin", "author_url": "", "post_date": "2026-01-21T07:16:55.250000", "content": "

There are also cases like \"ITU.1.KAM na-ar-ma-ak a-šur\".

\n

Without the next word \"Sarrātim\" or \"Kēnātim\" this could be month 2 or 3. How should we deal with this case?

", "votes": 0, "replies": [] }, { "id": 3394544, "author_name": "Hrithik Reddy", "author_url": "", "post_date": "2026-01-21T08:52:42.977000", "content": "

@deeppast

\n

Transliteration : ITU.1.KAM áb-ša-ra-ni li-mu-um šu-sú-in DUMU ba-bi-lim

\n

Which would be the correct translation according to test set formatting

\n
 Month Ab-šarrāni, in the eponymy Šu-Suen son of Babilum.\n Month Abšarrani, in the eponymy Šu-Suen son of Babilum.\n Month V, in the eponymy of Šu-Suen, son of Babilum.\n Month 5, in the eponymy of Šu-Suen, son of Babilum.\n
", "votes": 3, "replies": [] }, { "id": 3394805, "author_name": "hongan", "author_url": "", "post_date": "2026-01-21T17:25:14.817000", "content": "

my guess is the last one

", "votes": 0, "replies": [] }, { "id": 3394811, "author_name": "Hrithik Reddy", "author_url": "", "post_date": "2026-01-21T17:36:17.643000", "content": "

Thinking the same , but just to be sure

", "votes": 0, "replies": [] }, { "id": 3394822, "author_name": "耶✌", "author_url": "", "post_date": "2026-01-21T18:01:20.733000", "content": "

What can be confirmed so far is that there is no third option.

", "votes": 0, "replies": [] }, { "id": 3394905, "author_name": "Adam Anderson", "author_url": "", "post_date": "2026-01-21T20:59:48.347000", "content": "

Exactly, it definitely won't be Month V. The translators were not consistent with the name normalization (hence the onomasticon I provided), and they weren't consistent with the conversion of month names to month numbers (hence my month name to number chart). In general, it's more common to see the month name rather than the number.

", "votes": 1, "replies": [] } ] }, { "id": 3390737, "author_name": "justinlamlamlam", "author_url": "", "post_date": "2026-01-13T21:45:47.577000", "content": "

Thank you for the response! This is very helpful.

", "votes": 0, "replies": [] } ] } ], "raw_markdown_by_id": { "3390252": "Hi @deeppast . \nI have a few questions. Any insights on them would be super helpful.\nI have found that there are a quite few long floating point numbers, which is likely due to floating point errors. To give a few examples\n\n| | oare_id | transliteration | translation |\n|-----:|:-------------------------------------|:--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|:------------------------------------------------------------------------------------------------------------------------------------------|\n| 672 | 697fdae3-abd6-4de6-89ac-03796381be51 | 0.33333 ma-na 1.3333300000000001 GÍN KÙ.BABBAR ṣa-ru-pá-am i li-bi šé-ep-IŠTAR 1 GÍN iš-tí e-wa-ri-mu-ša | 0.3333 mina 1.3333 shekel of refined silver is in Šēp-Ištar's possession. |\n| 1301 | d1bc99fc-3f4b-476d-8c28-f072aa280083 | 3.6666600000000003 ma-na 2 GÍN AN.NA i li-bi₄ a-šur-SIPA IGI a-šur-dan IGI ir-ma-a-šur | 3 minas 42 shekels of tin owed by Aššur-rē'ī. In the presence of Aššur-dān, of Ir'am-Aššūr. |\n| 1458 | ef2c3a21-2451-436b-b1a3-16b8574ff27a | 1.3333300000000001 ma-na KÙ.BABBAR a-mu-ru-um i li-bi₄ a-da-da x um … DUMU šu-… 0.33333 ma-na KÙ.BABBAR ta-… ša a-lá-ḫi-im iš-tù ḫa-mu-uš-tim ša kur-ub-IŠTAR DINGIR ša-pá-tám i-li-ik KÙ.BABBAR il₅-qé | 1.3333 mina of checked silver is owed by Adada. |\n| 950 | 9676573d-7242-4ee0-a393-fea152d64de9 | xx GÍN KÙ.BABBAR a-na du-li i-na ú-lá-ma a-dí-in um-ma šu-ut-ma ku-ta-nam ú-šé-ba-lá-kum 1.3333300000000001 ma-na KÙ.BABBAR KI ḫa-lu-a x x | I gave Duli in Ulama [x] shekels of silver. He said: I will send you one -textile.\" 1.3333 mina of silver is with halua ...\" |\n| 1339 | d93fadca-089c-4135-8228-bc755873b2bf | 10.66666 GÍN a-šùr-ma-lik i-lá-qé 0.66666 ma-na 0.83333 GÍN kàr-wa-a 11 GÍN iṣ-mì-id-DINGIR 0.5 ma-na 3.6666600000000003 GÍN šu-ku-bu-um 15.33333 GÍN bu-zu-ta-a | 10.6666 shekels Aššur-mālik will take, 40.83333 shekels Karwaya, 11 shekels Iṣmid-Ilum, 33.6666 shekels Šu-Kūbum, 15.3333 shekels Buztāya |\n\n\nThere are many more containing such floating point numbers. I could find 79 rows with 9 consecutive 0s with this code.\n```\ndf = pd.read_csv(\"/kaggle/input/deep-past-initiative-machine-translation/train.csv\")\ndf = df[df['transliteration'].str.contains('000000000', na=False)]\n```\nWill such numbers appear in the test set? \nHow should we translate such numbers? Sometimes these numbers are translated to fractions, and sometimes they are translated to numbers with inconsistent decimal places. \n", "3390651": "No, there are no floats in the test set. I would encourage you to convert the decimals to fractions. Their system is base-60 so these fractions are predictable, these are the only ones you'll encounter:\n\n- ½\t= 0.5\n- ¼\t= 0.25\n- ⅓\t= 0.333...\n- ⅔\t= 0.666...\n- ⅙\t= 0.166...\n- ⅚\t= 0.833...\n\nIf you consider that 1 mina = 60 shekels, then you get the following conversion into shekels:\n\n% of mina (ma-na) in shekels:\n- ½\t= 0.5\t\t= 30 shekels\n- ¼\t= 0.25\t\t= 15 shekels\n- ⅓\t= 0.333...\t= 20 shekels\n- ⅔\t= 0.666...\t= 40 shekels\n- ⅙\t= 0.166...\t\t= 10 shekels\n- ⅚\t= 0.833...\t= 50 shekels" } }