Suicide, Police Knowledge, and AI-Assisted Research in Wartime and Postwar Shanghai, 1939–1948

On May 3, 1945, Zhou Zifa jumped into a well near his home and died. His son reported that Zhou, a small book seller (書攤) at the entrance of their alley on Qimen Road (332祁門路), had complained of economic difficulties since the New Year, but that day his father had only told him he was going to the public latrines. He had no suspicion of anything until one of his neighbors came to share the news that his father had been found in the well. Nothing in the police file tells us why Zhou chose to end his life (SMA R44-2-5520).

Suicide left a substantial documentary trace in modern Shanghai. A sudden death activated the police, hospitals, coroners, relatives, neighbors, and other witnesses. Their statements and reports might reconstruct a room, a rooftop, a well, a bottle of poison, the position of a body, or the movements that preceded death. Yet this administrative abundance did not necessarily answer the question that mattered most to survivors and that most readily attracts the historian: why did a person decide to die?

This essay explores that question through a curated collection of suicide-related dossiers in the Shanghai Municipal Archives. It asks what circumstances surrounded these deaths and how police officers, relatives, and witnesses explained—or failed to explain—the acts they investigated. The corpus is not a representative sample from which to calculate the incidence of suicide in Shanghai. This is a dataset collected in the course of my work on Scythe and the City (Henriot 2016). The corpus is uneven in date, provenance, and documentary depth. Its strongest concentration lies in 1944 and 1945, with a smaller postwar cluster in 1948. It therefore offers a focused view of suicide during the late occupation and the unsettled immediate postwar years, filtered through the surviving practices of police and judicial record keeping.

The purpose is deliberately modest. This is not a comprehensive social history of suicide in Republican China and does not attempt a full historiographical review. Three works nevertheless provide important points of orientation. Bryna Goodman’s The Suicide of Miss Xi reconstructs the extraordinary public, legal, and political afterlife of a young woman’s suicide in 1922 Shanghai (Goodman 2021). It shows how one death could generate rival accounts of gender, economic conduct, citizenship, and responsibility. Hou Yanxing’s study of female suicide in Shanghai between 1927 and 1937 places women at the center of a sustained analysis of the social problem before the period examined here (Hou Yanxing 2008). Emily Baum’s work on the invention of “madness” in modern China supplies a necessary warning against translating unstable contemporary terms such as 神經病 directly into present-day psychiatric diagnoses (Baum 2018). The dossiers studied below are more routine and usually far less public than the celebrated case examined by Goodman. Precisely for that reason, they reveal the ordinary evidentiary limits of police knowledge.

A second purpose makes the exploration unusual. The dossiers were read and converted into structured case metadata with the assistance of ChatGPT 5.6. This was an experiment in using a multimodal language model to process mixed-format, mostly handwritten Chinese archives at corpus scale. The experiment did not abolish illegibility or uncertainty. It made their distribution visible. Some fields, notably year and method, could be extracted with relative consistency; others, especially addresses and motives, remained partial, contradictory, or simply unknown. The historical argument and the methodological experience thus converge: the archive records the effort to make suicide intelligible, but it also preserves the limits of that effort.

1. From archival dossiers to structured evidence

The starting collection contains fifty-eight catalogue entries. Eight are aggregate statistical files rather than individual case records: seven police statistical returns covering parts of 1942 and 1943, and one undated municipal compilation. Removing these produces a working set of fifty individual or suicide-related dossiers. Because one file concerns the deaths of two women, the dossiers represent at least fifty-one persons. The records come chiefly from two Shanghai Municipal Archives series. Thirty-two files belong to R44-2 and cluster in 1944–45, under the wartime municipal police administration. Fifteen belong to Q186-2 and are predominantly postwar Nationalist municipal records, concentrated in 1948. Single files come from R4-1 and R43-2. This bimodal shape reflects a change in the institutions that created and preserved the records as much as a difference in chronology. The collection cannot therefore be treated as a continuous, uniformly sampled series extending from 1939 to 1948.

A typical dossier is composite. Police investigation reports sit alongside witness interrogations, procuratorial or court documents, hospital records, and medico-legal examinations. Occasionally, investigators preserved suicide notes or farewell letters. These layers make it possible to reconstruct physical circumstances with considerable precision. They also establish whose voices dominate the file: investigators and survivors rather than the deceased. Even where personal writing survives, it was selected, copied, translated, or embedded within an official inquiry.

The fifty retained dossiers total 1,120 pages. The median file contains seventeen pages, and the mean 22. One exceptional 169-page criminal case concerning alleged instigation of suicide pulls that average upward; without it, the mean falls to 19 pages. The forty-one-page investigation of the deaths of Raissa Zartsine and Anna Lyssai (SMA R44-2-5804) and the thirty-eight-page Zhuang Yupeng dossier (SMA R44-2-5691) are also unusually substantial. Most routine inquests, however, are compact. Unequal dossier length matters because investigative depth is itself a source of informational inequality: the cases that generated more documents naturally offer more opportunities to recover context.

Five files resist straightforward inclusion in an ordinary-suicide corpus. One concerns a drowning that police ultimately classified as accidental. Another is principally a criminal prosecution for alleged instigation of suicide. A third records a poisoning for which suicidal intent remains uncertain. A fourth began as an inquiry into suicide or homicide but eventually supported suicide. The fifth contains two deaths in a single dossier. These cases have been retained for the descriptive profile because they belong to the archival collection under study, but they are flagged rather than silently normalized. For the analysis of motive, the accidental drowning and instigation case are excluded, leaving forty-eight suicide or possible-suicide dossiers.

Reading handwritten Chinese archives with ChatGPT 5.6

The practical challenge was to move from hundreds of photographed or scanned pages to case records that could be compared. ChatGPT 5.6 assisted with reading handwritten and typewritten Chinese documents, identifying names, dates, institutions, places, and document types, drafting case-level abstracts, and extracting metadata into a structured workbook. Work proceeded iteratively rather than as a single act of transcription. A name or age read on one page could be checked against an interrogation record elsewhere in the file; a provisional classification could be revised after a medico-legal report; conflicting dates or addresses could be retained explicitly rather than collapsed into one apparently certain value. This is an important distinction. The model was not used as an autonomous archivist, nor were its first readings treated as authoritative. It functioned as a research assistant within a repeated cycle of reading, extraction, comparison, correction, and audit. The complete dataset was subsequently examined for internal consistency. The audit separated original archival descriptions from derived analytical variables and avoided inferring sex from a Chinese personal name when the documentation did not make sex explicit.

The results were uneven in historically revealing ways. Year and physical method are comparatively solid. Sex is explicit in forty-one of fifty dossiers. A preferred numeric age could be assigned to only thirty individuals. Precise event locations are weakest: approximately two-thirds of the files contain addresses that are uncertain, partial, or difficult to decipher. The present data therefore support a chronological and methodological profile, but not reliable street-level mapping. AI assistance accelerated the movement between individual pages and the corpus as a whole; it did not make every variable equally knowable.

AI USE STATEMENT ChatGPT 5.6 was used to assist with the reading of digitized archival documents, including handwritten and typewritten Chinese records, and with the extraction, organization, and auditing of case-level metadata. It also assisted in producing preliminary abstracts, identifying recurring terms, and comparing information across dossiers. All interpretations, classifications, and quantitative results were reviewed by the author. Uncertain readings were retained as uncertain rather than silently resolved, and personal names, ages, addresses, motives, and case classifications were not inferred when the surviving documents did not provide sufficient evidence. The model had no independent access to the archival collections beyond the documents supplied by the author.

The ability to retain uncertainty proved more valuable than any promise of automatic completion. When the sources did not support an address, a motive, or even a secure classification of death, the appropriate result was not a plausible reconstruction but an explicit unknown. That discipline becomes central to the historical interpretation that follows.

2. Fifty dossiers: a profile of suicide in wartime and postwar Shanghai

The chronology of the corpus is sharply concentrated. Seven dossiers date from 1939, two from 1943, fifteen from 1944, seventeen from 1945, and eight from 1948; one cannot be dated at all. Thus thirty-two of the fifty files (64 percent) belong to 1944–45. This distribution describes surviving and selected records, not the incidence of suicide in the city. Without comparison to external municipal statistics and a fuller account of archival survival, the wartime concentration cannot be read as evidence that suicide itself peaked in those years.

Year	Dossiers
1939	7
1943	2
1944	15
1945	17
1948	8
Uncertain	1
Total	50

Where sex is explicit, the distribution is almost even: twenty-one female and twenty male dossiers. Nine remain unknown. At the person level, the double-suicide dossier adds one woman, producing twenty-two identified women among at least fifty-one persons. The unknown category is essential. Closer reading of the original scans would likely narrow it.

Age survives much less consistently. A numeric age is available in thirty dossiers. Within that subset, the mean is 38.4, the median thirty-eight, and the range sixteen to fifty-eight. Nine of the thirty individuals fall between forty and forty-nine, the largest decade group. These figures characterize only the known-age subset. They also conceal two informative exceptions: Pan Yaohua is described simply as being in her early twenties, while Anna Lyssai is listed as approximately eighty-five in an English police report and eighty in a later Chinese form.

Suicide method is far better documented. Fifteen dossiers concern jumping or falling from a height and fourteen poisoning. Together they account for 58 percent of the corpus. Seven involve drowning or immersion, six a blade or self-inflicted wound, four hanging, and one a firearm; the remaining files belong to exceptional or unspecified classifications. The drowning total includes the file eventually judged accidental, and the double-suicide dossier securely establishes hanging for the daughter while leaving the mother’s method less clear.

Method	Dossiers	Share
Jumping/fall	15	30%
Poisoning	14	28%
Drowning/immersion	7	14%
Blade/wound	6	12%
Hanging	4	8%
Firearm	1	2%
Instigation case	1	2%
Other/unspecified	2	4%

Poisoning opens a particularly material window onto urban life. The files mention sleeping pills, arsenic, opium, Lysol or 來蘇水, 黃鏹水, and substances that investigators could not identify securely. Yet the comparative solidity of method should not be mistaken for explanatory power. An inquest could determine what entered a person’s body or from which structure someone fell while leaving the circumstances of decision entirely opaque. The archive is most confident at precisely the point where the historian’s causal question begins.

3. Explaining the inexplicable: illness, precarity, and personal crisis

The analytical corpus contains forty-eight individual suicide or possible-suicide dossiers after excluding the accidental drowning and the criminal-instigation case. A meaningful contextual or explanatory factor can be identified in only eighteen, or 37.5 percent. In the remaining thirty (62.5 percent) the abstracts and comments do not support a confident explanation. This majority is not just an inconvenience produced by incomplete metadata. In many cases, the surviving investigation itself did not establish a reason. The categories used below overlap. Ten dossiers contain health-related circumstances, seven material insecurity, and four marital or intimate strain. Two record prior suicidal behavior or ideation, and one includes a threatened loss of housing. Adding these totals would count some people more than once, but that overlap is analytically productive. The files more often suggest clusters of vulnerability and precipitating events than a single cause.

Illness and physical suffering

Health is the largest identifiable cluster. Seven cases involve chronic or serious physical illness, and six use language of nervous or mental disturbance; several belong to both groups. In one of the clearest formulations, Chen Maoyin had been ill for a long time without recovery (久病不愈) and was said to have developed (厭世之念) thoughts or feelings of weariness with life (SMA R44-2-3944). The wording draws a causal line more directly than most files, but it remains the language of an official reconstruction rather than transparent access to inner experience.

Other cases describe physical suffering without producing a single decisive explanation. Wang Xiunian suffered from advanced pulmonary tuberculosis, and testimony linked her prolonged failure to recover with her decision to jump (SMA R44-2-4818). Xu Fukang had lived with paralysis or hemiplegia for roughly three years despite treatment (SMA R44-2-5631). Nikolay Voronchihin also suffered from serious tuberculosis and had poor prospects of recovery. Immediately before attempting suicide, however, he lost a wallet containing his savings. His coworker considered the disease more important than the financial loss while acknowledging both. The file thereby preserves not one motive but a contemporary debate over relative causal weight (SMA R44-2-5768).

Six dossiers employ expressions such as 神經病 (nervous disturbance) or 神經不正常 (mental confusion). These terms cannot safely be converted into retrospective clinical diagnoses. Their meaning depended on medical, popular, familial, and administrative usage. Mrs. Li-Zhang had reportedly suffered from 神經病 for about three years (SMA R44-2-4309). Lin Kangshi’s file heading directly connected her suicide to the same condition (SMA R44-2-5600). Mrs. Zhang-Chen’s husband likewise described her as suffering from 神經病 and reported more than ten earlier suicidal episodes from which she had been rescued (SMA R44-2-5621).

Zhang’s case makes the temporal problem especially clear. Witnesses knew of a long history of suicidal behavior, but they could not explain why the final attempt occurred on that day or why it proved fatal. The archive documents chronic vulnerability without recovering the immediate precipitant. A tentative gender contrast also appears: four of the six files using nervous or mental terminology concern women, whereas documented physical illness appears in five male and two female cases. With numbers this small, the pattern is a question for further investigation, not a demographic conclusion.

Material insecurity

Seven dossiers highlight unemployment, difficulty making a living, indebtedness, dismissal, loss of savings, precarious residence, or threatened displacement. These circumstances do not appear as an abstract condition called poverty. They enter the record as concrete disruptions in the organization of daily life. Chen Gushi’s husband had recently become unemployed. The household faced severe subsistence difficulties, and the couple quarreled shortly before she jumped (SMA R44-2-4246). Xu Nian, formerly a carpenter, was also unemployed and described as experiencing life difficulties (SMA R44-2-5464). Neither dossier permits economic hardship to be isolated from the rest of the person’s life, but each shows how unemployment entered the explanatory vocabulary of witnesses and police.

Xu Qinzhang’s circumstances were still more entangled. Testimony described gambling, visits to prostitutes, debts, deteriorating relations with his employer, and dismissal (SMA R44-2-5788). A label such as “economic motive” would blur the line between the interaction of money, work, reputation, and social relations. Voronchihin’s lost savings present a sudden financial shock superimposed on tuberculosis (SMA R44-2-5768). The deaths of Raissa Zartsine and Anna Lyssai raise the problem of housing insecurity: before their deaths, Zartsine had been informed that their flat would be treated as vacant from 1 July. The notice makes threatened displacement highly relevant, but it does not prove that eviction alone caused the double suicide (SMA R44-2-5804).

These dossiers are most persuasive when read as sequences of pressure. A livelihood could become precarious over months or years; a wallet could disappear, an employer could dismiss a worker, or a housing notice could introduce a sudden crisis. The administrative file often sought an intelligible proximate explanation, but the surviving evidence repeatedly points beyond a single event.

Marriage, relationships, and cumulative pressure

Four dossiers provide evidence of marital strain. Three concern women, and three involve poisoning, although neither distribution can support a general claim. Tan Chenshi’s file directly connects her poisoning to 與夫口角, a quarrel with her husband (SMA R44-2-4421). Chen Gushi’s case also combines household economic distress with a recent marital conflict. Zhuang Shenshi’s partner or husband was frequently absent and himself connected her unhappiness to his inability to spend time with her. Here the relevant circumstance was not an overt quarrel but possible emotional isolation (SMA R44-2-4246). Zheng Xinchen is especially revealing because two farewell letters survive. His separation from his wife and precarious circumstances form part of the documentary context, yet his wife explicitly stated that she did not know why he had killed himself (SMA R44-2-4439). Documentary abundance, however, does not provide all the answers. A note can confirm an intentional departure, address survivors, distribute responsibilities, or record a final wish without explaining why death became preferable to continued life.

Only three dossiers explicitly preserve suicide or farewell writing: those of Yu Side, Zheng Xinchen, and Voronchihin, roughly 6 percent of the analytical corpus. Yu’s handwritten note does not establish a definitive motive in the present metadata. Zheng’s letters communicate with his wife and concern future arrangements but do not explain why he chose to die. Voronchihin’s note confirms that his act was intentional, while the account of illness and financial loss comes mainly from his coworker. Evidence of suicidal intent, in short, is not equivalent to evidence of motive. Cumulative pressure is therefore more faithful to the stories in the  files than an illusory taxonomy. Chronic vulnerability included tuberculosis, paralysis, recurring illness, nervous disturbance, repeated suicidal behavior, unemployment, debt, and relational instability. Immediate precipitating events included a quarrel, the loss of savings, dismissal, an accumulation of debts, or notice affecting housing.

The strongest aggregate finding remains the thirty dossiers for which no certain contextual explanation can be recovered. Some investigations assembled witnesses and circumstances but ended without an account of motive. Others survive in abstracts too thin to sustain interpretation. Silence therefore has more than one origin: the deceased may have concealed distress; relatives or acquaintances may genuinely not have known; witnesses may have withheld information; investigators may not have asked, believed, or recorded; and later extraction may preserve the physical acts more fully than the testimony surrounding them. The chronological distribution of identifiable factors demonstrates the need for caution. Seventeen of the eighteen dossiers containing such evidence date from 1944–45. Only one dates from 1948, and none from the 1939 or 1943 groups. It would be tempting to convert this into a wartime transformation in the causes of suicide. The more plausible explanation at this stage is documentary asymmetry: earlier and postwar abstracts and comments are generally thinner. Before any chronological causal claim can be made, those dossiers must be reread systematically.

4. What the police knew and what remains unknowable

The dossiers are far stronger on how people died than on why. Police and medico-legal investigators could examine injuries, identify substances, measure rooms and buildings, reconstruct movements, question witnesses, and decide whether a death appeared accidental, suicidal, or possibly homicidal. Such work transformed a body and a scene into an administratively classified event. Subjective intention belonged to a different evidentiary order. Even the motives that appear most clearly were socially constructed after death. They generally came from spouses, parents, relatives, neighbors, employers, coworkers, hospital personnel, and investigators. Recurrent phrases (久病不愈, 厭世之念, 神經病, 生活困難, 失業, 與夫口角) are therefore double evidence. They refer to illness, hardship, unemployment, and conflict, but they also reveal the categories through which survivors and officials rendered a death comprehensible. This does not mean that every explanation was arbitrary. Tuberculosis, paralysis, lost savings, dismissal, and marital quarrels were real circumstances attested in the files. It means that the step from circumstance to motive was interpretive even for contemporaries. Witnesses weighed competing facts, and police reports compressed them into concise causal formulations. Where no such formulation emerged, the investigation preserved opacity rather than knowledge.

The AI-assisted workflow sharpened rather than resolved this source-critical problem. At its best, it enabled rapid comparison across many files and made recurrent terms and omissions visible. It also created a danger familiar to all structured-data projects: an empty field invites completion. Refusing that invitation was methodologically essential. A plausible motive generated from surrounding circumstances would be precisely the kind of speculation that these records cannot support. The 62.5 percent without a secure explanation must remain a substantive result. The present exploration consequently offers no epidemiology of Shanghai suicide and no general ranking of causes. It identifies a body of wartime and postwar dossiers in which illness, material insecurity, intimate conflict, and prior suicidal behavior sometimes surrounded death. More consistently, it documents the work performed by families, neighbors, medical personnel, and the police to transform a disturbing act into an intelligible event.

Returning to Zhou Zifa’s death reveals both the density and the limits of the police archive. Its final pages reconstruct almost everything surrounding the recovery of his body. Zhou, a fifty-four-year-old native of Yancheng who made his living from a small bookstall, left home on the afternoon of 2 May after saying that he was going to a public toilet on Jiangning Road. When he failed to return, his family did not immediately know what had happened. The next morning, an unnamed woman from the same alley reported that his body was in the well. His two sons pulled him out, laid him beside the well, and notified the neighborhood head. Police officers questioned the family, inspected the body, recorded its dimensions and physical condition, and found no fatal external injury. Their conclusion was that Zhou had entered the water alive and died by drowning.

The dossier did not lack explanations. Zhou’s son reported that business at his father’s small bookstall had become depressed, that earning a living had grown extremely difficult, and that Zhou had expressed thoughts of 厭世 since the beginning of the year. The police summary, however, placed greater emphasis on another history: Zhou was said to suffer from 神經病, had previously attempted to kill himself by entering a river, and had once been treated at Pudong Hospital. Material hardship, expressed despair, previous suicidal behavior, and an administratively defined nervous disorder were thus assembled around the same death. They may have been connected, but the file does not show precisely how.

What these final pages provide, and what they fail to provide, can stand for the larger collection. They establish the identity of the dead man, the condition of his body, the absence of evident violence, the route by which the death became known, and the administrative reasoning through which it was classified as suicide. They even preserve words attributed to Zhou before his death. Yet they do not preserve his own account of what those words meant or why, on that particular afternoon, he did not return. The archive offers not an absence of explanation but several explanations produced by different observers and subsequently compressed into an official conclusion. These files preserve a material history of death with remarkable density. They also preserve the incompleteness of explanation. That opacity is not external to the history of suicide in Shanghai, nor is it simply a technical defect to be repaired by better extraction. It is one of the central historical facts the archive allows us to recover.

References

Baum, Emily. The Invention of Madness: State, Society, and the Insane in Modern China. Chicago: University of Chicago Press, 2018.

Goodman, Bryna. The Suicide of Miss Xi: Democracy and Disenchantment in the Chinese Republic. Cambridge, MA: Harvard University Asia Center, 2021.

Hou Yanxing 侯艷興. Shanghai nüxing zisha wenti yanjiu (1927–1937) 上海女性自殺問題研究（1927–1937), Shanghai: Shanghai cishu chubanshe 上海辭書出版社, 2008.

Archival materials

Shanghai Municipal Archives (SMA), chiefly R44-2 and Q186-2 series, with individual dossiers from R4-1 and R43-2. All available on BN-Asie [query “suicide” to pull the files]

---

The text only may be used under licence Creative Commons Attribution Non Commercial 4.0 International. All other elements (illustrations, imported files) are “All rights reserved”, unless otherwise stated.

---

---