# The unread oracle bone characters > Oracle bone script is the earliest Chinese writing, cut into ox shoulder blades and turtle shells at the Shang capital Yinxu (Anyang) between about 1250… Source: https://mysteriesdept.com/mysteries/oracle-bone-undeciphered Publisher: Mysteries Department Licence: CC BY 4.0 (credit Mysteries Department and link the page) Updated: 2026-10-31T15:26:58+00:00 --- **MYS** 0DEDM4BQVECR·Mystery · script # The unread oracle bone characters **0** [Follow][Ask a question](https://mysteriesdept.com/questions/key-questions?mystery=oracle-bone-undeciphered) The unread oracle bone characters ![Photograph of an inscribed Shang ox scapula fragment, Heji 1, in the National Museum of China](https://mysteriesdept.com/media/folio/oracle-bone-undeciphered-lead) Gary Lee Todd · CC0 1.0 Where Yinxu, the last Shang capital, at Xiaotun near Anyang, Henan; a few hundred Western Zhou pieces from Zhouyuan, Shaanxi⁠[[1]] When c. 1250 to 1041 BC, from the reign of Wu Ding to the fall of the Shang⁠[[6]] Corpus About 150,000 inscribed pieces, about 90 percent of them fragments⁠[[13]] Standard edition Jiaguwen heji, 13 volumes, 41,956 pieces, 1978 to 1982; supplement 1999⁠[[22]] Sign inventory About 4,500 distinct graphs counted; between a third and a half read with agreement⁠[[10]] Language Early Old Chinese, the ancestor of the Chinese writing system⁠[[7]] Holders National Library of China (over 35,000), Academia Sinica (about 20,000), British Library (about 500) and about 14 countries in all⁠[[23]] Bounty 100,000 yuan per accepted new reading, National Museum of Chinese Writing, 2016 onward; three awards by 2025⁠[[10]] Status Unexplained Questions 0 of 7 answered Places [All 6](https://mysteriesdept.com/places?mystery=oracle-bone-undeciphered) [**British Library, London** library](https://mysteriesdept.com/records/PLC0DEDJ57D6PD2/british-library-london)[**National Library of China, Beijing** library](https://mysteriesdept.com/records/PLC0DEDJ4MP2HBA/national-library-of-china-beijing)[**National Museum of China, Beijing** museum](https://mysteriesdept.com/records/PLC0DEDJ5QZSV6B/national-museum-of-china-beijing)[**National Museum of Chinese Writing, Anyang** museum](https://mysteriesdept.com/records/PLC0DEDJ42NMZ3X/national-museum-of-chinese-writing-anyang) Concepts [All](https://mysteriesdept.com/concepts) [Decipherment](https://mysteriesdept.com/concepts/CPT0DECW01ZWFP5/decipherment)[Held-out test](https://mysteriesdept.com/concepts/CPT0DECW06RF522/held-out-test)[Null model](https://mysteriesdept.com/concepts/CPT0DECW06JED24/null-model)[Radiocarbon dating](https://mysteriesdept.com/concepts/CPT0DECW013KR34/radiocarbon-dating) Catalog key**MYS** 0DEDM4BQVECR Oracle bone inscriptions are the records of divination at the court of the late Shang kings. A diviner applied heat to a prepared ox scapula or turtle plastron, read the cracks, and the result was cut into the bone beside them: the date and the diviner's name, the question put, sometimes the king's forecast and the outcome⁠[[1]]. Farmers around Xiaotun village near Anyang had been digging the bones up for years before [Wang Yirong](https://mysteriesdept.com/records/FIG0DEDJKZ9S54Q/wang-yirong) recognised the marks as ancient writing in 1899⁠[[1]]. [Luo Zhenyu](https://mysteriesdept.com/records/FIG0DEDJM7VWF81/luo-zhenyu) traced the source to Anyang and in 1914 read several hundred graphs by comparing them with bronze inscriptions and the small seal forms recorded in the Shuowen jiezi dictionary of AD 100⁠[[2]]. [Wang Guowei](https://mysteriesdept.com/records/FIG0DEDJMCDJZ2K/wang-guowei) then showed in 1917 that the ancestors who receive sacrifice in the texts are the Shang kings of Sima Qian's king list, with corrections, which fixed the bones as Shang royal records⁠[[3]]. Scientific excavation began in October 1928 under [Dong Zuobin](https://mysteriesdept.com/records/FIG0DEDJMJFQ1RB/dong-zuobin) and Li Ji, and in 1933 Dong published ten criteria that sort the inscriptions into five periods tied to reigns⁠[[4]]⁠[[5]]. [Radiocarbon dating](https://mysteriesdept.com/concepts/CPT0DECW013KR34/radiocarbon-dating) of king-attributed bones in 2021 put Wu Ding's reign at 1254 to 1197 BC and the end of the sequence at about 1041 BC⁠[[6]]. The script is therefore read in the sense that matters for history. Its language is an early stage of Chinese, its graphs are the ancestors of modern characters, and a competent reader can translate most inscriptions⁠[[7]]⁠[[1]]. What is not read is a large part of the sign inventory. Counts differ with how variants are merged, but about 4,500 distinct graphs are usually cited, of which between 1,160 and 2,200 are read with general agreement depending on the standard applied⁠[[8]]⁠[[9]]⁠[[10]]. One recent tally gives 3,763 catalogued graphs, 1,273 read, 378 disputed and 2,112 unread⁠[[11]]. Most of the unread graphs occur only a few times, and many are proper names of people, places, clans and sacrificial victims that left no descendant in later Chinese writing⁠[[11]]. Each of them is a small gap in an otherwise legible text, and together they are the reason a proportion of every inscription cannot be translated with confidence. Reading a new graph means tracing a chain of forms from the bone through Western Zhou bronze inscriptions, Warring States scripts and small seal to a known character, and then showing that the proposed word fits every context in which the graph appears⁠[[7]]. Since 2016 the [National Museum of Chinese Writing](https://mysteriesdept.com/records/PLC0DEDJ42NMZ3X/national-museum-of-chinese-writing-anyang) in Anyang has offered 100,000 yuan for each reading that passes its expert review and 50,000 for settling a disputed one; [Jiang Yubin](https://mysteriesdept.com/records/FIG0DEDJNZD8Y0R/jiang-yubin) received the first award in 2018 for the graph chun, and a second batch of awards followed in 2023⁠[[12]]⁠[[13]]⁠[[14]]. Three awards had been made by 2025⁠[[10]]. In parallel, the Anyang Normal University laboratory under Liu Yongge opened the Yinqi Wenyuan platform in 2019, image datasets such as OBC306, HUST-OBC and EVOBC were released, and a conditional diffusion model that generates a candidate modern glyph from a bone graph won the ACL 2024 best paper award⁠[[15]]⁠[[16]]⁠[[17]]⁠[[18]]. Specialists judge the generated hints of limited use so far, and a 2025 benchmark found that the best multimodal models decipher at about the level of an untrained person⁠[[11]]⁠[[19]]. The corpus itself is also unfinished. About 90 percent of the pieces are fragments, and joining them recovers context that single fragments lack⁠[[20]]. Rejoining by hand has produced several thousand joins over a century; deep-learning systems since 2020 propose candidate joins with high recall on [held-out](https://mysteriesdept.com/concepts/CPT0DECW06RF522/held-out-test) data, and tools built with Tencent have found about 500 duplicate rubbings in the published record⁠[[21]]⁠[[10]]. The standard edition is the Jiaguwen heji, 41,956 rubbings in 13 volumes published between 1978 and 1982 under [Guo Moruo](https://mysteriesdept.com/records/FIG0DEDJMT6DBC9/guo-moruo) and Hu Houxuan, with a 1999 supplement of 13,450 pieces⁠[[22]]. The largest holdings are the [National Library of China](https://mysteriesdept.com/records/PLC0DEDJ4MP2HBA/national-library-of-china-beijing) with upwards of 35,000 pieces, Academia Sinica in Taipei with about 20,000 from the 1928 to 1937 excavations, and the [British Library](https://mysteriesdept.com/records/PLC0DEDJ57D6PD2/british-library-london) with about 500⁠[[23]]⁠[[24]]. Of the 7 questions on this site, 0 are answered. ### Key questionsset by the editors [**Can computational methods produce readings of unread graphs that palaeographers accept?** ](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQFMCXW4D/can-computational-methods-produce-readings-of-unread-graphs)No answer ahead yet [**How many distinct graphs does the oracle-bone script have once scribal variants are merged?** ](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQEXW42BX/how-many-distinct-graphs-does-the-oracle-bone-script-have)No answer ahead yet [**How many of the 150,000 fragments can still be rejoined, and how much context does rejoining recover for rare graphs?** ](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQG2AZ12C/how-many-of-the-150-000-fragments-can-still-be-rejoined-and)No answer ahead yet [**What are the roughly 3,000 unread oracle-bone graphs, and how many of them can ever be read?** ](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQEESA0GT/what-are-the-roughly-3-000-unread-oracle-bone-graphs-and-how)No answer ahead yet [**What standard of proof should count as an accepted reading of an oracle-bone graph?** ](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQG872N9M/what-standard-of-proof-should-count-as-an-accepted-reading)No answer ahead yet [All 7 questions ](https://mysteriesdept.com/questions/key-questions?mystery=oracle-bone-undeciphered) ### Research sessions1 so far · last 31 Oct `S-001`[The unread oracle bone characters seeded: corpus, readings and open data](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/sessions/SES0DEDVWXKJN4C/session-1-the-unread-oracle-bone-characters-seeded-corpus) 31 Oct **11** evidence**7** questions raised**7** answers proposedNo question changed standing [All 1 sessions ](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/sessions) ## The object An oracle bone is a prepared ox scapula or turtle plastron with rows of hollows drilled into its back. A diviner applied a heated point to a hollow so that a crack formed on the front, read the crack, and the record was cut beside it with a knife: a preface giving the day in the sixty-day cycle and the diviner's name, the charge put to the ancestors or spirits, sometimes the king's reading of the crack, and sometimes what happened⁠[[1]]. Charges concern sacrifice, harvest, rain, war, hunting, illness, childbirth and the ten-day period ahead. Pieces were stored in pits at Xiaotun; pit YH127, found in 1936, held over 17,000 inscribed pieces, and the 1991 Huayuanzhuang East cache held 561 inscribed pieces from a prince's household⁠[[24]]⁠[[1]]. About 150,000 inscribed pieces are known, held in China and about fourteen other countries; roughly 90 percent are fragments⁠[[13]]⁠[[20]]. The National Library of China holds upwards of 35,000, Academia Sinica about 20,000 from the 1928 to 1937 excavations, and the British Library about 500 under shelfmark Or. 7694⁠[[23]]⁠[[24]]. The standard edition is the Jiaguwen heji (1978 to 1982), 41,956 rubbings and drawings in 13 volumes arranged by Dong Zuobin's periods, with a 1999 supplement of 13,450 pieces⁠[[22]]. Later excavated groups have their own volumes. Digital access is split between a subscription database of over 143,000 inscriptions from Zhonghua Book Company, the free Academia Sinica rubbings database, the Anyang Normal University platform Yinqi Wenyuan, the CHANT transcriptions at the Chinese University of Hong Kong, and Xiaoxuetang's graph-by-graph forms⁠[[25]]⁠[[26]]⁠[[10]]⁠[[27]]⁠[[28]]. ### Timeline **Timeline** Squeezed where little happened. Each mark is a published claim, by how it stands:AcceptedMost likelyStill possibleDisputedRuled out ## What is known The language is early Chinese and the script is the direct ancestor of modern Chinese characters. Its graphs work as later Chinese graphs do: some are pictures of what they mean, some are borrowed for a word that sounds alike, and most combine a sound element with a meaning element⁠[[7]]. Text runs in columns, usually top to bottom, and paired charges on one bone are often cut in mirror image on left and right⁠[[1]]. Numerals, the sixty-day cycle, the names of the kings and their consorts, the diviners, the sacrifices and the common verbs are all read, so most inscriptions can be translated in full and the rest in large part⁠[[1]]⁠[[29]]. The chronology is fixed on three legs. Wang Guowei's 1917 matching of sacrificial recipients to the Shiji king list established that the archive runs from Wu Ding to Di Xin⁠[[3]]. Dong Zuobin's ten criteria of 1933, since refined into diviner groups by Li Xueqin, [Qiu Xigui](https://mysteriesdept.com/records/FIG0DEDJNQ4CN2R/qiu-xigui) and Lin Yun, assign most pieces to a reign or pair of reigns⁠[[4]]⁠[[5]]. Radiocarbon dating of 48 king-attributed bones in 2021 gave 1254 to 1197 BC for Wu Ding and about 1041 BC for the end, and placed the disputed Li diviner group in the Wu Ding to Zu Geng reigns rather than later⁠[[6]]. What is known about the unread graphs is mostly negative. Counts of distinct graphs range from about 3,800 to about 4,500 depending on how scribal variants are merged, and the share read ranges from a third to a half depending on whether disputed readings count⁠[[8]]⁠[[9]]⁠[[10]]⁠[[11]]. Most unread graphs are rare. Context often shows what kind of word a graph is, a place after the verb for hunting, a person after a verb of sacrifice, a clan sign on a bronze, without showing which word⁠[[11]]⁠[[7]]. ## What has been tested The method that has produced every accepted reading since 1914 is the chain of forms. A graph on bone is matched to a bronze form of the Western Zhou, then to Warring States and small seal forms, and finally to a character whose sound and meaning are recorded; the reading is then checked against every context in which the graph occurs⁠[[2]]⁠[[7]]. Jiang Yubin's 2018 reading of chun, meaning stirring or unrest, followed this route through bronze inscriptions and was accepted by the Anyang museum's reviewers; Zhou Zhongbing's reading of tai, accepted in 2023, took two years of collection and drafting⁠[[13]]⁠[[14]]. The bounty has produced three awards in nine years, which is the measure of how hard the remaining graphs are⁠[[10]]. Computation has been tested on three tasks. For recognition of known graphs, datasets such as OBC306 (309,551 scanned images, 306 classes) and HUST-OBC (140,053 images; 1,588 read classes and 9,411 unread classes) support classifiers that reach high accuracy on read graphs⁠[[15]]⁠[[16]]⁠[[9]]. For [decipherment](https://mysteriesdept.com/concepts/CPT0DECW01ZWFP5/decipherment), EVOBC provides 229,170 images along six stages from bone to clerical script so that a model can learn how forms change; a [baseline](https://mysteriesdept.com/concepts/CPT0DECW06JED24/null-model) predicts the modern character of a held-out read graph with 16.7 percent top-1 and 55.8 percent top-20 accuracy⁠[[17]]. The OBSD diffusion model generates a candidate modern glyph directly from a bone graph image and was judged the best paper at ACL 2024, but its output for unread graphs is a visual hint rather than a reading, and a 2026 review finds such hints of limited use to palaeographers⁠[[18]]⁠[[11]]. OBI-Bench tested 23 large multimodal models on 5,523 images and found the best of them at about the level of untrained people on deciphering and below them on fine perception⁠[[19]]. For rejoining, Zhang and colleagues' 2020 system reached 98.39 percent top-10 accuracy on held-out joins, the Microsoft and Capital Normal University system Diviner matches fragment outlines across publications, and the Tencent and Anyang Normal University platform has found about 500 duplicate rubbings⁠[[21]]⁠[[20]]⁠[[10]]. These recover context rather than readings, but context is what the chain-of-forms method needs. ## The readings Five positions are live about the unread graphs, and they are not exclusive. The first is that most are proper names: people, places, clans, victims and spirits whose graphs were never used for a common word and so left no descendant to be traced. On this view they can be identified as names and dated but never read as words, and the undeciphered count will never fall far⁠[[11]]. The second is that many are common words that died before the bronze age of writing, so that no later form exists; these can be read only from internal evidence, by showing from every context what the word must mean, which the museum's reviewers have so far accepted only when a form chain also exists⁠[[7]]⁠[[14]]. The third is that a large share of the unread classes are variants, miscuts or damaged forms of graphs already read. HUST-OBC lists 9,411 unread classes against 1,588 read ones, far more than the 2,000 to 3,000 unread graphs scholars cite, which shows how much depends on merging variants⁠[[16]]⁠[[9]]. If so, the true problem is smaller than the headline figures and a careful variant study would shrink it. The fourth is that many unread graphs do have later descendants obscured by changes of form, and that systematic chains such as EVOBC's six stages, together with the slow stream of bounty awards, will keep yielding readings⁠[[17]]⁠[[10]]. The fifth is that generative and multimodal models will propose readings that experts then confirm; no accepted reading has yet come this way, and the benchmarks place current models at the level of untrained people⁠[[18]]⁠[[19]]⁠[[11]]. All five share a dependence on context, and context is what fragments lack. The rejoining work is therefore relevant to every position⁠[[21]]. ### Published claims **Published claims and their status** | Year | Who | Claimed | Status | Reason | |---|---|---|---|---| | 1899 | Wang Yirong⁠[1] | The marks on the bones sold by Anyang dealers are an ancient form of Chinese writing. | Accepted | Every later reading confirmed it, and the graphs are the ancestors of modern characters⁠[7]. | | 1914 | Luo Zhenyu⁠[2] | Several hundred graphs can be read by comparison with bronze inscriptions and Shuowen seal forms, including the names of the Shang kings. | Accepted | The comparative method he used is the one behind every accepted reading since⁠[7]. | | 1917 | Wang Guowei⁠[3] | The ancestors who receive sacrifice in the inscriptions are the Shang kings of the Shiji, whose list is largely correct. | Accepted | The match fixed the archive as Shang royal records and has held through a century of new finds⁠[1]. | | 1933 | Dong Zuobin⁠[4] | Ten criteria sort the inscriptions into five periods tied to Shang reigns. | Most likely | The scheme stands in outline but diviner groups can overlap reigns, and his 1949 placing of the Li group was overturned by radiocarbon dates⁠[5]⁠[6]. | | 1978 | Guo Moruo and Hu Houxuan⁠[22] | The 41,956 pieces of the Jiaguwen heji give a usable corpus arranged by period. | Accepted | It remains the standard edition that every database and dataset is keyed to⁠[25]. | | 1978 | David Keightley⁠[1] | A divination record has a fixed structure of preface, charge, prognostication and verification, which lets most inscriptions be read in full. | Accepted | The structure is the basis of every translation handbook since⁠[29]. | | 1982 | Li Xueqin, Qiu Xigui and Lin Yun⁠[5] | Dating should proceed by diviner group rather than by Dong's five periods, and the Li group belongs with Wu Ding and Zu Geng. | Most likely | The 2021 radiocarbon series placed the Li group in late Wu Ding to Zu Geng as they argued⁠[6]. | | 1988 | Qiu Xigui⁠[7] | A graph is read by tracing its form through later scripts to a known character and checking the word against every context. | Accepted | Both bounty awards so far followed this route and were accepted on it⁠[14]. | | 2018 | Jiang Yubin⁠[13] | A frequent unread graph is chun, meaning stirring or unrest, in both bone and bronze texts. | Accepted | Accepted by the museum's expert review as the first 100,000 yuan award⁠[8]. | | 2020 | Chongsheng Zhang and colleagues⁠[21] | A deep network can rank candidate fragment joins with 98.39 percent top-10 accuracy on held-out pairs. | Accepted | The accuracy is on known joins; how many new joins it has produced in the corpus is not reported⁠[21]. | | 2021 | Kexin Liu and colleagues⁠[6] | Radiocarbon places Wu Ding at 1254 to 1197 BC and the end of the archive at about 1041 BC. | Accepted | A Bayesian model of 48 king-attributed bones agrees with the historical sequence and resolves the Li group⁠[6]. | | 2024 | Haisu Guan and colleagues⁠[18] | A conditional diffusion model can generate the modern glyph of an oracle-bone graph and so hint at readings for unread ones. | Still possible | It won ACL 2024 best paper on read graphs, but no unread graph has been read from its hints and reviewers judge them of limited use⁠[11]. | | 2024 | Pengjie Wang and colleagues⁠[16] | HUST-OBC's 9,411 unread classes against 1,588 read ones measure the size of the problem. | Still possible | The unread count is three to four times the figure palaeographers cite, which points to unmerged variants rather than distinct words⁠[9]. | | 2025 | Zijian Chen and colleagues⁠[19] | Large multimodal models decipher oracle-bone graphs at about the level of untrained people and trail them on fine perception. | Accepted | Measured on 5,523 images and five tasks across 23 models⁠[19]. | | 2026 | Yiran Rex Ma⁠[11] | Of 3,763 catalogued graphs 2,112 are unread, and most of these are proper nouns such as personal and place names. | Still possible | It is the consensus view among palaeographers but no published slot-by-slot classification of the unread graphs supports the proportion⁠[11]. | ## What would settle it The central question is not whether the script can be read but how many of the unread graphs can ever be read, and each position predicts something testable. A full variant study of the unread classes in HUST-OBC and Yinqi Wenyuan, merging allographs by position, period and diviner group, would show how many distinct unread words there are; if the count falls well below 2,000 the variant position gains⁠[[16]]⁠[[10]]. A classification of every unread graph by syntactic slot, using the CHANT transcriptions, would show what share can only be names; if it is most of them, the names position is confirmed and the bounty's ceiling is known⁠[[27]]⁠[[11]]. For the computational position the test is a held-out one. A model trained on graphs read before a cut-off date should be scored on the graphs accepted since 2018 by the Anyang reviewers, with the generated glyph or ranked candidates compared with the accepted reading; EVOBC's own baseline of 16.7 percent top-1 on read graphs is the number to beat, and a model that places an accepted new reading in its top 20 on a graph it never saw would be the first real evidence⁠[[17]]⁠[[18]]⁠[[14]]. Any proposed reading must then do what Jiang Yubin's did: fit every attested context and connect to a later form or a documented word⁠[[13]]⁠[[7]]. New finds remain possible. Yinxu is still excavated, the 1991 Huayuanzhuang cache and the Zhouyuan bones of the 1970s and 2003 were unexpected, and more rejoins among the 150,000 fragments would supply the contexts that rare graphs need⁠[[24]]⁠[[21]]. ## Questions 0 of 7 questions are answered. **Key questions** [All 7](https://mysteriesdept.com/questions/key-questions?mystery=oracle-bone-undeciphered) **0** [Can computational methods produce readings of unread graphs that palaeographers accept?](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQFMCXW4D/can-computational-methods-produce-readings-of-unread-graphs) 0 of 1 sub-questions answered·0 No evidence yet **0** [How many distinct graphs does the oracle-bone script have once scribal variants are merged?](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQEXW42BX/how-many-distinct-graphs-does-the-oracle-bone-script-have) 0 of 0 sub-questions answered·0 No evidence yet **0** [How many of the 150,000 fragments can still be rejoined, and how much context does rejoining recover for rare graphs?](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQG2AZ12C/how-many-of-the-150-000-fragments-can-still-be-rejoined-and) 0 of 0 sub-questions answered·0 Open: no answer ahead yet **0** [What are the roughly 3,000 unread oracle-bone graphs, and how many of them can ever be read?](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQEESA0GT/what-are-the-roughly-3-000-unread-oracle-bone-graphs-and-how) 0 of 1 sub-questions answered·0 Open: 6 answers, none ahead **0** [What standard of proof should count as an accepted reading of an oracle-bone graph?](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQG872N9M/what-standard-of-proof-should-count-as-an-accepted-reading) 0 of 0 sub-questions answered·0 No evidence yet ## Evidence 10 findings [What are the roughly 3,000 unread oracle-bone graphs, and how many of them can ever be read?](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQEESA0GT/what-are-the-roughly-3-000-unread-oracle-bone-graphs-and-how)9 findings [Most unread graphs are names without descendants](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/answers/ANS0DEDS19358ER/most-unread-graphs-are-names-without-descendants)Still possible +[A 2026 survey reports 2,112 of 3,763 catalogued graphs unread and states that most are proper nouns such as personal and place names.](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTF0NZRNA/a-2026-survey-reports-2-112-of-3-763-catalogued-graphs)Suggestive +[Three accepted readings in nine years of a 100,000 yuan bounty show that the remaining graphs resist the form-chain method that read the rest.](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTFGAHS19/three-accepted-readings-in-nine-years-of-a-100-000-yuan)Suggestive [Lost words with no later form](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/answers/ANS0DEDS1VM7WTQ/lost-words-with-no-later-form)Still possible +[The structure of Chinese graphs means a word that fell out of use leaves a graph with no phonetic or semantic anchor in later writing.](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTG099HTB/the-structure-of-chinese-graphs-means-a-word-that-fell-out)Suggestive [Unmerged variants inflate the unread count](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/answers/ANS0DEDS2J0FVE4/unmerged-variants-inflate-the-unread-count)Still possible +[HUST-OBC catalogues 9,411 unread image classes against 1,588 read ones, while palaeographers count 2,000 to 3,000 unread graphs, a gap of three to four times.](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTG4YX83Y/hust-obc-catalogues-9-411-unread-image-classes-against-1-588)Supported −[Even the expert-merged tallies leave 2,112 unread graphs of 3,763, so variant merging alone cannot remove the problem.](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTGF070DA/even-the-expert-merged-tallies-leave-2-112-unread-graphs-of)Suggestive [Descendants exist and form chains will find them](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/answers/ANS0DEDS3C777CK/descendants-exist-and-form-chains-will-find-them)Still possible +[Jiang Yubin's 2018 reading of chun and Zhou Zhongbing's 2023 reading of tai were both reached by form chains through bronze inscriptions and accepted by the Anyang museum's reviewers.](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTGX7E8FK/jiang-yubins-2018-reading-of-chun-and-zhou-zhongbings-2023)Supported −[The rate of accepted readings has been about one every three years since 2016, which at the present pace would take centuries for 2,000 graphs.](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTHBFGPBG/the-rate-of-accepted-readings-has-been-about-one-every-three)Suggestive [Models will propose readings experts confirm](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/answers/ANS0DEDS3NDW5K3/models-will-propose-readings-experts-confirm)Still possible +[OBSD generates plausible modern glyphs for read oracle-bone graphs and was judged best paper at ACL 2024.](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTHKXFKA9/obsd-generates-plausible-modern-glyphs-for-read-oracle-bone)Suggestive −[OBI-Bench finds the best of 23 large multimodal models deciphering at about the level of untrained people, and EVOBC's evolution-chain baseline scores 16.7 percent top-1 on held-out read graphs.](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTJ2A3HW3/obi-bench-finds-the-best-of-23-large-multimodal-models)Supported [How many of the 150,000 fragments can still be rejoined, and how much context does rejoining recover for rare graphs?](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/questions/QST0DEDQG2AZ12C/how-many-of-the-150-000-fragments-can-still-be-rejoined-and)1 finding [Automated rejoining recovers context at scale](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/answers/ANS0DEDS4CYCRZY/automated-rejoining-recovers-context-at-scale)Still possible +[A 2020 deep-learning system ranked the correct join in its top 10 for 98.39 percent of held-out fragment pairs, and platform tools have since found about 500 duplicate rubbings.](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/evidence/EVD0DEDTM293ZBW/a-2020-deep-learning-system-ranked-the-correct-join-in-its)Supported ## Artifacts The films, photos, papers and objects this mystery rests on, each with its own key and its distance from the original. **Artifacts** [See all 6](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/artifacts) Photo1 step from object [Heji 1: Bin-group scapula of King Wu Ding's reign, National Museum of China](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/artifacts/ART0DEDNMFA8PPV/heji-1-bin-group-scapula-of-king-wu-dings-reign-national) Gary Lee Todd · CC0 1.0 Photo1 step from object [The British Library oracle bones (Or. 7694, Couling-Chalfant and Hopkins collections)](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/artifacts/ART0DEDNMQ5N18C/the-british-library-oracle-bones-or-7694-couling-chalfant) British Library · CC0 1.0 Dataset2 steps from object [Pit YH127 archive, Xiaotun North (Academia Sinica)](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/artifacts/ART0DEDNMYN2V62/pit-yh127-archive-xiaotun-north-academia-sinica) Institute of History and Philology, Academia Sinica · Academia Sinica terms of use; free with registration Dataset2 steps from object [The Jiaguwen heji corpus (41,956 rubbings, 1978 to 1982)](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/artifacts/ART0DEDNN6XKVTM/the-jiaguwen-heji-corpus-41-956-rubbings-1978-to-1982) Zhonghua Book Company, Gulian, Chen Nianfu · subscription, copyright Zhonghua Book Company [See all 6 artifacts](https://mysteriesdept.com/mysteries/oracle-bone-undeciphered/artifacts) ## See also [ **Proto-cuneiform: the archaic texts of Uruk** The oldest writing, 3350 to 3000 BC: numbers and goods are read, the language is not. ](https://mysteriesdept.com/mysteries/proto-cuneiform) [ **Etruscan: the readable language that cannot be fully translated** An alphabet everyone can read, 13,000 inscriptions, about 250 understood words, and four long texts that still will not translate. ](https://mysteriesdept.com/mysteries/etruscan-language) [ **Cretan Hieroglyphic** The first script of Minoan Crete, 1900 to 1600 BC, on seals and clay, with no word read. ](https://mysteriesdept.com/mysteries/cretan-hieroglyphic) [ **Eteocretan and Eteocypriot** Two Iron Age languages, from Crete and Cyprus, written in scripts we can read and languages we cannot. ](https://mysteriesdept.com/mysteries/eteocretan-eteocypriot) ## References 1. 1.[The standard English handbook on the oracle-bone inscriptions: how a divination record is structured, how the corpus was found and published, Dong Zuobin's periodization and the collections. University of California Press, Berkeley.](https://mysteriesdept.com/sources/SRC0DEDFXAMZWYF/sources-of-shang-history-the-oracle-bone-inscriptions-of)cited 6 times here**SRC** 0DEDFXAMZWYF 1. 2.[The first large systematic set of graph identifications, reading several hundred graphs by comparison with bronze and Shuowen seal forms and identifying the Shang royal names. Library record used for the bibliographic data.](https://mysteriesdept.com/sources/SRC0DEDFYKS8DWF/yinxu-shuqi-kaoshi-explanations-of-the-yin-ruins)cited 4 times here**SRC** 0DEDFYKS8DWF 1. 3.[Identified 31 Shang ancestors and kings as sacrifice recipients in the inscriptions and showed that Sima Qian's king list in the Shiji was largely correct, with corrections. This is the check that fixed the inscriptions as Shang royal records. Pointer page used for the bibliographic data.](https://mysteriesdept.com/sources/SRC0DEDGA29RA0K/yin-buci-zhong-suojian-xiangong-xianwang-kao-the-predynastic)cited 3 times here**SRC** 0DEDGA29RA0K 1. 4.[Set out ten criteria (genealogy, royal titles, diviner names, graph forms, calligraphy and others) that divide the inscriptions into five periods tied to Shang kings; written 1932, published 1933. Pointer page used for the bibliographic data.](https://mysteriesdept.com/sources/SRC0DEDGA9AN4WS/jiaguwen-duandai-yanjiu-li-examples-of-research-on-the)cited 4 times here**SRC** 0DEDGA9AN4WS 1. 5.[Reviews Dong Zuobin's five-period scheme and the diviner-group revisions by Li Xueqin, Qiu Xigui and Lin Yun after the 1973 Xiaotun South finds. Early China 8, pp. 1-13.](https://mysteriesdept.com/sources/SRC0DEDFXVSSASC/recent-approaches-to-oracle-bone-periodization-a-review)[doi](https://doi.org/10.1017/S0362502800005411)cited 3 times here**SRC** 0DEDFXVSSASC 1. 6.[First systematic AMS radiocarbon series on king-attributed oracle bones: Wu Ding 1254-1197 BC, end of Shang c. 1041 BC, Li group placed in Wu Ding to Zu Geng. Radiocarbon 63(1), pp. 155-175.](https://mysteriesdept.com/sources/SRC0DEDFY3TDWMF/radiocarbon-dating-of-oracle-bones-of-late-shang-period-in)[doi](https://doi.org/10.1017/RDC.2020.90)cited 4 times here**SRC** 0DEDFY3TDWMF 1. 7.[The standard account of how Chinese graphs are built (pictographs, phonetic loans, semantic-phonetic compounds) and of the method for identifying an oracle-bone graph by tracing it through bronze and seal forms. Early China Special Monograph Series 4, Berkeley; Chinese original 1988.](https://mysteriesdept.com/sources/SRC0DEDFXKZ1AH3/chinese-writing-wenzixue-gaiyao-translated-by-gilbert-l)cited 5 times here**SRC** 0DEDFXKZ1AH3 1. 8.[Reports Chen Jian and Xie Mingwen honoured by the National Museum of Chinese Writing in January 2024, Jiang Yubin's 2018 award, and Jiang's estimate that about 160,000 pieces exist, about 80,000 are published, and about 1,160 graphs are confirmed of a dictionary of 3,500 to 4,000.](https://mysteriesdept.com/sources/SRC0DEDH8230WQM/fudan-scholars-honoured-for-oracle-bone-interpretations)cited 4 times here**SRC** 0DEDH8230WQM 1. 9.[Survey of oracle character recognition with a table of datasets (Oracle-20K, OBC306, HWOBC, HUST-OBC, EVOBC, Oracle-MNIST, OBI125) and their repositories; states about 4,500 classes found and about 2,200 mapped to modern characters.](https://mysteriesdept.com/sources/SRC0DEDGBCYE989/a-comprehensive-survey-of-oracle-character-recognition)[doi](https://doi.org/10.48550/arXiv.2411.11354)cited 2 times here**SRC** 0DEDGBCYE989 1. 10.[China Daily report on the 2025 Anyang AI decipherment competition (Henan heritage administration, Anyang government, Anyang Normal University, Tencent), Liu Yongge's laboratory, the Yinqi Wenyuan platform launched 2019, about 500 duplicate rubbings found, and three bounty recipients to date; gives over 4,500 graphs with more than 3,000 unread.](https://mysteriesdept.com/sources/SRC0DEDH7SYE3A5/race-on-to-decipher-oracle-bone-characters-with-ai)cited 8 times here**SRC** 0DEDH7SYE3A5 1. 11.[Position paper mapping digital resources and methods for oracle bone, bronze and bamboo scripts; gives the count 3,763 catalogued graphs, 1,273 deciphered, 378 debated, 2,112 undeciphered, and judges generative hints such as OBSD of limited use to experts.](https://mysteriesdept.com/sources/SRC0DEDGQBEJ1T5/towards-computational-chinese-paleography)[doi](https://doi.org/10.48550/arXiv.2601.06753)cited 5 times here**SRC** 0DEDGQBEJ1T5 1. 12.[Reports the National Museum of Chinese Writing's bounty of 100,000 yuan per accepted decipherment (50,000 for a disputed one resolved), with the counts then current: about 160,000 pieces, 4,000 graphs, 1,600 read.](https://mysteriesdept.com/sources/SRC0DEDGTMEFXG2/chinese-museum-offers-big-cash-reward-for-identifying-oracle)cited 3 times here**SRC** 0DEDGTMEFXG2 1. 13.[Reports Jiang Yubin's 2018 first prize (100,000 yuan) for reading the graph chun, over 150,000 pieces excavated, about 4,400 graphs with one third read, and National Museums Scotland's 1,700 bones.](https://mysteriesdept.com/sources/SRC0DEDGTZ7HEWQ/a-vocation-with-character-beijing-review-on-the-120th)cited 4 times here**SRC** 0DEDGTZ7HEWQ 1. 14.[Reports the museum's second batch of accepted decipherments (2023): Zhou Zhongbing's first prize for reading the graph tai, Li Chuntao's second prize, and the count of about 4,500 graphs with one third read.](https://mysteriesdept.com/sources/SRC0DEDGV9Z1RMK/amazing-oracle-bone-scripts-how-hard-is-deciphering-oracle)cited 3 times here**SRC** 0DEDGV9Z1RMK 1. 15.[ICDAR 2019: 309,551 scanned character images in 306 classes cropped from rubbings, distributed by the Anyang Normal University lab. The largest scanned-glyph recognition set.](https://mysteriesdept.com/sources/SRC0DEDGSD7N23J/obc306-a-large-scale-oracle-bone-character-recognition)cited 2 times here**SRC** 0DEDGSD7N23J 1. 16.[Describes HUST-OBC: 140,053 images, 1,588 deciphered classes (77,064 images) and 9,411 undeciphered classes (62,989 images), built from two scanned books, GuoXueDaShi, Yinqi Wenyuan and HWOBC; dataset CC BY-NC 4.0 on Figshare. Scientific Data 11.](https://mysteriesdept.com/sources/SRC0DEDGAFWQ9AQ/an-open-dataset-for-oracle-bone-character-recognition-and)[doi](https://doi.org/10.1038/s41597-024-03807-x)cited 5 times here**SRC** 0DEDGAFWQ9AQ 1. 17.[EVOBC: 229,170 images in 13,714 categories across six stages (oracle bone, bronze, seal, Spring and Autumn, Warring States, clerical); decipherment baseline top-1 16.7%, top-20 55.8%. Used for the evolution-chain approach.](https://mysteriesdept.com/sources/SRC0DEDGB3QH6XA/an-open-dataset-for-the-evolution-of-oracle-bone-characters)[doi](https://doi.org/10.48550/arXiv.2401.12467)cited 4 times here**SRC** 0DEDGB3QH6XA 1. 18.[OBSD: a conditional diffusion model that generates a candidate modern glyph from an oracle-bone graph image, with a zero-shot refinement stage; ACL 2024 best paper. Used for the computational wave and its limits.](https://mysteriesdept.com/sources/SRC0DEDGARH43R9/deciphering-oracle-bone-language-with-diffusion-models)[doi](https://doi.org/10.18653/v1/2024.acl-long.831)cited 4 times here**SRC** 0DEDGARH43R9 1. 19.[Benchmark of 5,523 images and five tasks (recognition, rejoining, classification, retrieval, deciphering) on 23 large multimodal models; the best models match untrained humans on deciphering and trail them on fine perception. ICLR 2025.](https://mysteriesdept.com/sources/SRC0DEDGRVJARQR/obi-bench-can-lmms-aid-in-study-of-ancient-script-on-oracle)[doi](https://doi.org/10.48550/arXiv.2412.01175)cited 3 times here**SRC** 0DEDGRVJARQR 1. 20.[Describes the Diviner fragment-matching system built by Microsoft Research Asia with Capital Normal University's oracle bone centre, the Zhuiyu lianzhu rejoin database at Fudan, and the estimate that about 90% of the 160,000 pieces are fragments.](https://mysteriesdept.com/sources/SRC0DEDH8D7P6XC/the-shang-dynasty-has-an-unexpected-ally-ai)cited 2 times here**SRC** 0DEDH8D7P6XC 1. 21.[IJCAI 2020 demo: OB-Rejoin dataset and algorithm (98.39% top-10 on held-out joins), OracleBone-8000 character-level annotations, 80.9% recognition accuracy. Used for the fragment-rejoining line of work.](https://mysteriesdept.com/sources/SRC0DEDGQZJF41K/ai-powered-oracle-bone-inscriptions-recognition-and)[doi](https://doi.org/10.24963/ijcai.2020/779)cited 3 times here**SRC** 0DEDGQZJF41K 1. 22.[The standard corpus edition: 41,956 rubbings and drawings in 13 volumes, Zhonghua Shuju, 1978-1982, arranged by Dong Zuobin's periods; supplement (Heji bubian) 1999 with 13,450 further pieces. Pointer page used for the bibliographic data.](https://mysteriesdept.com/sources/SRC0DEDFYD5ZFHF/jiaguwen-heji-collected-oracle-bone-inscriptions-13-volumes)cited 5 times here**SRC** 0DEDFYD5ZFHF 1. 23.[States that the British Library holds about 500 oracle bones (shelfmark Or. 7694), digitised in 2015, and that the National Library of China holds upwards of 35,000.](https://mysteriesdept.com/sources/SRC0DEDHMYWT69H/oracle-bones-a-conservation-collaboration-with-the-national)cited 2 times here**SRC** 0DEDHMYWT69H 1. 24.[Used only for secondary counts not found in the primary sources consulted: Academia Sinica's about 20,000 excavated pieces, pit YH127 (1936, over 17,000 pieces), Xiaotun South 1973 (4,612 inscribed pieces), Huayuanzhuang East 1991 (561 inscribed pieces), and the Zhouyuan Western Zhou bones.](https://mysteriesdept.com/sources/SRC0DEDHPK7J06Z/oracle-bone-wikipedia-article-used-as-a-pointer-to)cited 5 times here**SRC** 0DEDHPK7J06Z 1. 25.[Reviews the Zhonghua Book Company's subscription database of over 143,000 inscriptions (obid.ancientbooks.cn) and compares it with the free Academia Sinica rubbings database and CHANT.](https://mysteriesdept.com/sources/SRC0DEDH914JK07/cracking-the-oracle-bones-the-yinxu-oracle-bone-inscription)cited 3 times here**SRC** 0DEDH914JK07 1. 26.[Database of the Institute's own oracle-bone rubbings (over 40,000 held, 21,556 digitised records searchable), built 2004-2009 under Yuan Kuo-hua and maintained by the Fu Ssu-nien Library; free with registration.](https://mysteriesdept.com/sources/SRC0DEDH9CH9E84/oracle-bones-rubbings-database-institute-of-history-and)cited 3 times here**SRC** 0DEDH9CH9E84 1. 27.[Describes the CHANT oracle-bone component: transcriptions of inscriptions from seven major collections into standard characters, over a million words with character frequencies, keyed to Yinxu jiagu keci leizuan; searchable by character or phrase.](https://mysteriesdept.com/sources/SRC0DEDHB1Z5YYS/a-computerized-database-of-oracular-inscriptions-on-tortoise)cited 2 times here**SRC** 0DEDHB1Z5YYS 1. 28.[Free database of character forms by stage (oracle bone, bronze, small seal, clerical) and reconstructed pronunciations (Old Chinese, Middle Chinese, Guangyun, Jiyun); the usual reference for tracing a graph across scripts.](https://mysteriesdept.com/sources/SRC0DEDH9ZPFN6V/xiaoxuetang-chinese-character-form-and-sound-databases)cited 2 times here**SRC** 0DEDH9ZPFN6V 1. 29.[Annotated English readings of selected inscriptions including Heji 1 (the king orders the zhongren to work the fields); Springer and Shanghai People's Publishing House, ISBN 978-981-15-6213-6.](https://mysteriesdept.com/sources/SRC0DEDHP4CSQ3R/reading-of-shang-inscriptions)[doi](https://doi.org/10.1007/978-981-15-6214-3)cited 3 times here**SRC** 0DEDHP4CSQ3R Last edited 31 Oct 2026