量化四个世纪奥斯曼法庭登记册中的文化心理构念

arXiv cs.CL 论文

摘要

本文提出了一个计算文本分析框架,用于量化1513—1911年间54,670份奥斯曼土耳其语法庭记录中的文化心理构念(个人主义、集体主义、荣誉),并介绍了经过适配的句子转换器模型(OTA-BERTurk),以及一个公开发布、配有交互式网页界面的评分流程。

arXiv:2611.24981v1 Announce Type: new Abstract: Populations vary along cultural dimensions such as individualism, collectivism, and honor. But how can these dimensions be studied in historical societies for which conventional psychological data are unavailable? Here, we introduce a computational text-analytic framework for quantifying psychological constructs in 54,670 Ottoman Turkish court records from 21 Istanbul courts spanning 1513-1911 CE. We develop sentence-embedding models adapted to Ottoman legal language and extend a theory-driven, text-analytic pipeline to capture cultural variation along the dimensions of individualism, collectivism, and honor in this historical corpus. We evaluate our models through multiple methods: document retrieval, semantic textual similarity, questionnaire-item classification, and validation against expert annotations. Our adapted sentence transformer (OTA-BERTurk) achieved a Recall@1 of .843 on the test set for document retrieval, a Mean Reciprocal Rank (MRR) of .806 for semantic textual similarity (meaning the best answer usually appeared as a top result), and classified questionnaire items with 89.58\% accuracy. Document-level performance varied across constructs and methods. Qualitative analysis further showed that high scores reflected historically specific legal and institutional patterns. We publicly release the models, construct representations, scoring pipeline, as well as an interactive web interface to support transparent, reproducible historical-psychological analyses of Ottoman Turkish court records spanning approximately four centuries. These computational tools enable the study of psychological variation across centuries of cultural and institutional change in a diverse empire.
查看原文
查看缓存全文

缓存时间: 2026/11/20 09:04

# Quantifying Cultural-Psychological Constructs in Four Centuries of Ottoman Court Registers
Source: [https://arxiv.org/html/2611.24981](https://arxiv.org/html/2611.24981)
Historical Psychology in the Ottoman World\\authorsnames\[1, 2, 3\]Sila Mutaf, Enes Yılandiloğlu, Mohammad Atari\\authorsaffiliationsDepartment of Psychology, University of Michigan, Department of Digital Humanities, University of Helsinki, Department of Psychological and Brain Sciences, University of Massachusetts Amherst\\authornoteSila Mutaf and Enes Yılandiloğlu contributed equally to this work and share first authorship\. Acknowledgments: We thank Mina Ozkaraduman and Ahmet Ziya Boz for their help with annotating documents\. This publication was partly supported through the grant \#63867 from the John Templeton Foundation to Mohammad Atari\. \\addORCIDlinkSila Mutaf0009\-0000\-9213\-031X \\addORCIDlinkEnes Yılandiloğlu0009\-0009\-2764\-8011 \\addORCIDlinkMohammad Atari0000\-0002\-4358\-7783\\leftheaderMutaf et al\.

###### Abstract

Populations vary along cultural dimensions such as individualism, collectivism, and honor\. But how can these dimensions be studied in historical societies for which conventional psychological data are unavailable? Here, we introduce a computational text\-analytic framework for quantifying psychological constructs in 54,670 Ottoman Turkish court records from 21 Istanbul courts spanning 1513–1911 CE\. We develop sentence\-embedding models adapted to Ottoman legal language and extend a theory\-driven, text\-analytic pipeline to capture cultural variation along the dimensions of individualism, collectivism, and honor in this historical corpus\. We evaluate our models through multiple methods: document retrieval, semantic textual similarity, questionnaire\-item classification, and validation against expert annotations\. Our adapted sentence transformer \(OTA\-BERTurk\) achieved a Recall@1 of \.843 on the test set for document retrieval, a Mean Reciprocal Rank \(MRR\) of \.806 for semantic textual similarity \(meaning the best answer usually appeared as a top result\), and classified questionnaire items with 89\.58% accuracy\. Document\-level performance varied across constructs and methods\. Qualitative analysis further showed that high scores reflected historically specific legal and institutional patterns\. We publicly release the models, construct representations, scoring pipeline, as well as an interactive web interface to support transparent, reproducible historical\-psychological analyses of Ottoman Turkish court records spanning approximately four centuries\. These computational tools enable the study of psychological variation across centuries of cultural and institutional change in a diverse empire\.

###### keywords

Historical Psychology, Digital Humanities, Natural Language Processing, Cultural Evolution, Ottoman History

Our psychology leaves a unique trace in our language\([Pennebaker et al\., 2003](https://arxiv.org/html/2611.24981#bib.bib21)\)\. When language is preserved in written form, it can thereafter serve as an indirect archive of the minds and cultures that produced it\([Iliev et al\., 2015](https://arxiv.org/html/2611.24981#bib.bib44)\)\. Because written records emerge from particular cultural and institutional settings, recurring patterns can reveal how societies collectively made sense of the world, categorized people, organized social relationships, and evaluated behavior\([Boyd, 2017](https://arxiv.org/html/2611.24981#bib.bib59)\)\. Contemporary computational techniques allow us to excavate these “cognitive fossils” from historical texts – incomplete yet informative proxies for psychology preserved in textual records\([Atari and Henrich, 2023](https://arxiv.org/html/2611.24981#bib.bib1);[Baumard et al\., 2024](https://arxiv.org/html/2611.24981#bib.bib5)\)\.

Here, we study three foundational cultural dimensions — individualism, collectivism, and honor — in Ottoman court registers\. These court records provide an important source for studying social and cultural life in the Ottoman Empire, which spanned more than six centuries \(1299–1922 CE\) and encompassed three continents\. Although historians have studied these materials extensively\([Aykan and Ergene, 2020](https://arxiv.org/html/2611.24981#bib.bib6);[Ekinci, 2005](https://arxiv.org/html/2611.24981#bib.bib7), e\.g\.,\), their potential for large\-scale computational analysis is largely untapped\([Akça, 2025](https://arxiv.org/html/2611.24981#bib.bib10);[Coşgel et al\., 2025](https://arxiv.org/html/2611.24981#bib.bib18)\)\. The corpus analyzed here comprises 54,670 Latin\-script transcribed records drawn from 100 Istanbul qadi registers published through the Istanbul Qadi Registers project\([Türkiye Diyanet Vakfı İslâm Araştırmaları Merkezi, 2023](https://arxiv.org/html/2611.24981#bib.bib39)\)\. Individualism and collectivism capture the relative emphasis placed on personal agency, uniqueness, and self\-expression, on the one hand, and relational obligation, group harmony, and interdependence, on the other\([Triandis et al\., 1988](https://arxiv.org/html/2611.24981#bib.bib25);[Kitayama and Salvador, 2024](https://arxiv.org/html/2611.24981#bib.bib45)\)\. These orientations have been widely used to explain cultural variation in self\-construal, emotion, cognition, and social behavior\([Markus and Kitayama, 1991](https://arxiv.org/html/2611.24981#bib.bib26);[Oyserman et al\., 2002](https://arxiv.org/html/2611.24981#bib.bib27)\)\. Honor constitutes a related but distinct cultural logic organized around personal and family reputation, social respect, and responses to insults or moral transgressions\. It is especially relevant to the Turkish cultural context\([Uskul and Cross, 2019](https://arxiv.org/html/2611.24981#bib.bib23);[Uskul et al\., 2023](https://arxiv.org/html/2611.24981#bib.bib46)\)and has been proposed as an additional dimension of cultural variation that complements individualism and collectivism\([Oyserman, 2017](https://arxiv.org/html/2611.24981#bib.bib24)\)\. Together, these three constructs provide a theoretical framework for quantifying cultural\-psychological expression in Ottoman legal texts\.

Importantly, however, these psychological constructs may not transfer unchanged in historical and cultural contexts\. Although individualism and collectivism were developed to explain cross\-cultural variation, their dominant definitions and measurement instruments emerged largely from contemporary academic psychology in the U\.S\.\([Talhelm et al\., 2026](https://arxiv.org/html/2611.24981#bib.bib61)\)and may carry assumptions about personhood, agency, and group membership that do not map neatly onto Ottoman society\([Berry, 1969](https://arxiv.org/html/2611.24981#bib.bib53), see\)\. Contemporary formulations of honor likewise cannot be assumed to capture its historical meanings\. This problem reflects a broader issue in psychological science\([Medin et al\., 2017](https://arxiv.org/html/2611.24981#bib.bib62);[Rad et al\., 2018](https://arxiv.org/html/2611.24981#bib.bib63)\), whose empirical and theoretical foundations remain disproportionately based on Western, educated, industrialized, rich, and democratic \(WEIRD\) populations\([Henrich et al\., 2010](https://arxiv.org/html/2611.24981#bib.bib56)\)\. Emerging historical approaches in psychology address this issue by treating psychological variation as an outcome of processes unfolding over time and investigating how changing cultural, ecological, and institutional environments shape human psychology\([Atari et al\., 2025](https://arxiv.org/html/2611.24981#bib.bib22)\)\. Making Ottoman Turkish materials accessible allows researchers to examine the cultural evolution of psychological traits in an understudied, non\-WEIRD context and facilitates theoretical refinement\.

### 0\.1From Historical Corpora to Psychological Measurement

The expansion of digitized archives has enabled researchers to investigate historical variation in cultural\-psychological patterns during periods for which surveys and experiments are unavailable\. Early studies used word frequencies in large book collections to trace cultural trends, including changes in collective memory, the pursuit of fame, censorship, and historical epidemiology\([Michel et al\., 2011](https://arxiv.org/html/2611.24981#bib.bib30)\)\. This shift from individual texts to aggregate patterns parallels the practice ofdistant readingin literary scholarship, which often uses computational methods to identify recurring patterns and long\-term trends across large collections\([Moretti, 2013](https://arxiv.org/html/2611.24981#bib.bib32)\)\. Word\-counting analyses were also used to test specific hypotheses about historical change, such as the vocabulary associated with social obligation, authority, individual choice, and materialism in relation to urbanization\([Greenfield, 2013](https://arxiv.org/html/2611.24981#bib.bib31)\), the increasing prominence of individualistic language\([Twenge et al\., 2012](https://arxiv.org/html/2611.24981#bib.bib50);[Fan et al\., 2025](https://arxiv.org/html/2611.24981#bib.bib83)\), changes in emotional expression\([Acerbi et al\., 2013](https://arxiv.org/html/2611.24981#bib.bib51);[Masui and Miyamoto, 2025](https://arxiv.org/html/2611.24981#bib.bib81)\), and the declining cultural salience of moral and virtue\-related language\([Kesebir and Kesebir, 2012](https://arxiv.org/html/2611.24981#bib.bib52);[Wheeler et al\., 2019](https://arxiv.org/html/2611.24981#bib.bib82)\)\.

These dictionary\-based approaches, however, remain limited to the frequencies of individual terms or predefined groups of words and therefore provide limited information about how meaning varies across contexts and periods\([Almelhem et al\., 2026](https://arxiv.org/html/2611.24981#bib.bib60)\)\. Diachronic word embeddings addressed this limitation by representing words through their distributional relationships and comparing those representations over time\([Hamilton et al\., 2016](https://arxiv.org/html/2611.24981#bib.bib33)\)\. Historians subsequently evaluated word embeddings as tools for studying conceptual and semantic change\([Wevers and Koolen, 2020](https://arxiv.org/html/2611.24981#bib.bib36)\)\. In computational social science, word embedding models have been used to examine the changing cultural meanings of social class\([Kozlowski et al\., 2019](https://arxiv.org/html/2611.24981#bib.bib35)\), gender and ethnic stereotypes\([Garg et al\., 2018](https://arxiv.org/html/2611.24981#bib.bib34)\), and representations of social groups\([Charlesworth et al\., 2022](https://arxiv.org/html/2611.24981#bib.bib47)\)\. In parallel, methods from Natural Language Processing have been used to quantify psychological constructs, such as personality traits\([Park et al\., 2015](https://arxiv.org/html/2611.24981#bib.bib48)\), group identity\([Ashokkumar and Pennebaker, 2022](https://arxiv.org/html/2611.24981#bib.bib84)\), depression\([Eichstaedt et al\., 2018](https://arxiv.org/html/2611.24981#bib.bib49)\), and moral values\([Kennedy et al\., 2021](https://arxiv.org/html/2611.24981#bib.bib58)\)from social\-media posts\.

Applying contextual embedding methods to historical corpora nevertheless presents methodological challenges\. Their reliability depends on the quality of the training corpus, their adaptation to the historical language and genre, and the availability of appropriate validation data\([Manjavacas and Fonteyn, 2022](https://arxiv.org/html/2611.24981#bib.bib85)\)\. Historical corpora also introduce challenges around transcription errors, uneven corpus composition, and limited annotated data\([Pettit, 2016](https://arxiv.org/html/2611.24981#bib.bib37);[Martins and Baumard, 2022](https://arxiv.org/html/2611.24981#bib.bib38)\)\. These limitations are especially important for low\-resource languages\. Thus, reliable historical\-psychological text analysis requires adaptation to the language and corpus, together with validation against expert judgments or external historical evidence\([Baumard et al\., 2024](https://arxiv.org/html/2611.24981#bib.bib5);[Feuerriegel et al\., 2025](https://arxiv.org/html/2611.24981#bib.bib4);[Atari and Henrich, 2023](https://arxiv.org/html/2611.24981#bib.bib1);[Aiello and Simeone, 2019](https://arxiv.org/html/2611.24981#bib.bib65);[Kennedy et al\., 2022](https://arxiv.org/html/2611.24981#bib.bib66)\)\.

Recent work has taken steps toward addressing some of these challenges by combining psychometric theory with contextual language modeling\.[Atari et al\. \(2023\)](https://arxiv.org/html/2611.24981#bib.bib2)introduced Contextualized Construct Representation \(CCR\), which uses sentence transformers to represent validated psychometric items \(or face\-valid descriptions of a psychological construct\) and target texts in an embedding space\([Reimers and Gurevych, 2019](https://arxiv.org/html/2611.24981#bib.bib3)\)\. Then, relevance of the construct, or “loading,” is estimated by calculating the semantic similarity between the text’s and the scale items’ vectorized representations\.[Chen et al\. \(2024\)](https://arxiv.org/html/2611.24981#bib.bib9)adapted CCR to Classical Chinese and developed an indirect supervised contrastive\-learning procedure for fine\-tuning sentence transformers on the Chinese Historical Psychology Corpus, comprising 21,539 paragraphs from 667 historical works\. The authors fine\-tuned a Classical Chinese model on two historical corpora\. By representing passages holistically, this approach can identify construct\-relevant meaning that may not be expressed through individual words\. Applying CCR to Ottoman court records, hence, requires adapting sentence representations to a specific legal language, determining how contemporary constructs are represented in that language, and benchmarking the resulting scores against expert judgments\.

### 0\.2Ottoman Court Registers as a Psychological Archive

Ottoman court registers were maintained by a judge \(qadı\) or their deputy \(naib\) and recorded by court scribes \(kātib\)\. They document legal relations involving people from different social positions in the imperial center and provinces and illuminate family, social, economic, administrative, and legal life\([Uğur, 2026](https://arxiv.org/html/2611.24981#bib.bib8);[Aykan and Ergene, 2020](https://arxiv.org/html/2611.24981#bib.bib6)\)\. Their contents include family disputes, marriage, homicide, inheritance divisions, contracts, settlements, guardianship, manumission, taxation, price regulation, guild disputes, endowments, appointments, and administrative orders\.

The registers provide broad access to Ottoman legal and social discourse, but constitute an institutionally mediated rather than fully representative record of Ottoman social life; their contents reflect specific patterns of jurisdiction, court use, and documentary practice\. Because scribes generally recorded only case summaries, they provide limited detail about proceedings or the actual enforcement of judgments\([Ekinci, 2005](https://arxiv.org/html/2611.24981#bib.bib7)\)\. Despite these limitations, their breadth has made them important sources for historians, legal scholars, and economists\([Jennings, 1978](https://arxiv.org/html/2611.24981#bib.bib28);[Cosgel and Ergene, 2013](https://arxiv.org/html/2611.24981#bib.bib54)\)\. Scholars have edited, transcribed, and digitized court registers from across the Ottoman Empire, including collections from Istanbul\([Yılmaz, 2010c](https://arxiv.org/html/2611.24981#bib.bib13)\), Diyarbakır\([İzgöer, 2013](https://arxiv.org/html/2611.24981#bib.bib14)\), Jerusalem\([Rabaia, 2015](https://arxiv.org/html/2611.24981#bib.bib16)\), and Crimea\([Yaşa, 2025](https://arxiv.org/html/2611.24981#bib.bib15)\)\. These collections have supported research on Islamic legal institutions, slavery and social status, gender and property, legal agency, and family and household organization\([Baldwin, 2017](https://arxiv.org/html/2611.24981#bib.bib12);[Wagner et al\., 2024](https://arxiv.org/html/2611.24981#bib.bib11);[Peirce, 2003](https://arxiv.org/html/2611.24981#bib.bib40);[Agmon, 2004](https://arxiv.org/html/2611.24981#bib.bib41);[Doumani, 2017](https://arxiv.org/html/2611.24981#bib.bib43)\)\.

The registers’ content makes them especially useful for examining social identity and shared cultural beliefs because Ottoman legal documents frequently identify people through relational markers such as parentage, spouse, neighborhood, occupation, religious community, household, or patronage ties\. Their coverage across courts and historical periods creates opportunities for examining variation in legal and cultural\-psychological language at a scale that would be difficult to address through close reading alone\.

### 0\.3The Present Research

We develop and evaluate a computational framework for measuring cultural\-psychological constructs in Ottoman court records\. First, we domain\-adapt sentence\-transformer and long\-context embedding models to Ottoman legal language and evaluate the resulting representations through document retrieval, embedding\-space analysis, semantic textual similarity \(STS\)\([Cer et al\., 2017](https://arxiv.org/html/2611.24981#bib.bib67)\), and questionnaire\-item classification \(QIC\)\([Chen et al\., 2024](https://arxiv.org/html/2611.24981#bib.bib9)\)\. Second, we compare three approaches to representing psychological constructs: translated questionnaire items, proxy sentences that elaborate semantically aligned and opposed meanings, and expert\-authored definitions\. In the court records, constructs such as individualism, collectivism, and honor are rarely named directly; instead, they may be expressed through relationships and actions\. Representing each construct through multiple proxy sentences allows the models to identify different semantic expressions even when the records do not contain the modern psychological vocabulary\. Third, we rely on these representations to generate document\-level scores for individualism, collectivism, and honor, and validate these scores against manual annotations by experts in Ottoman history\. Moreover, we compare CCR\-based measurement with zero\-shot and few\-shot classifications produced by open\-weight Large Language Models \(LLMs\)\([Abdurahman et al\., 2024](https://arxiv.org/html/2611.24981#bib.bib68)\)\. Finally, we qualitatively examine the passages with the highest scores to determine whether the retrieved evidence corresponds to the intended constructs and how those concepts are expressed through legal and institutional terminology\. We publicly release the adapted models, construct representations, scoring pipeline, and an open\-source web interface for other researchers to test their hypotheses and extend our computational approach\.

## 1Method

### 1\.1Corpus and Preprocessing

#### 1\.1\.1Data source

The corpus used in this study comprises 100 digitized volumes of Istanbul court registers, known variously asqadi sicilleriorşer‘iyye sicilleri, with Latin\-script transcriptions, put together through a collaboration among the Istanbul Metropolitan Municipality, the Centre for Islamic Studies, and Istanbul Medipol University\([Türkiye Diyanet Vakfı İslâm Araştırmaları Merkezi, 2023](https://arxiv.org/html/2611.24981#bib.bib39)\)\. The whole dataset comprises 54,670 documents totaling 11,497,610 words\. The dataset spans from 1513 to 1911 CE, recorded by 21 different courts located in Istanbul\. The registers were written in the Perso\-Arabic script and drew on Ottoman Turkish, although their linguistic composition varied across courts, periods, and document types\([Akça, 2025](https://arxiv.org/html/2611.24981#bib.bib10)\)\. The dataset was transcribed to Latin script, digitized, and made publicly accessible by[Türkiye Diyanet Vakfı İslâm Araştırmaları Merkezi \(2023\)](https://arxiv.org/html/2611.24981#bib.bib39)\. We utilize the data provided by[Yüksek \(2024\)](https://arxiv.org/html/2611.24981#bib.bib17)\. Each record includes the court title, the year of the issue \(both Hijri and Gregorian dates\), the title, a summary, the text of the case, and the ID information\. The summaries in our data were manually prepared by domain experts during the process of transcription into Latin script\. These one\-sentence summaries have the names of the main actors and the main act or appeal \(see Supplementary Materials for examples of these summaries\)\.

#### 1\.1\.2Preprocessing and Exclusions

The original dataset contained 54,670 records\. For document\-level scoring, we removed 56 records that contained a summary but no main text, leaving 54,614 records\. We applied stricter criteria when constructing the model\-training subset, excluding 133 records from the original dataset: 116 with fewer than 20 text characters, including the 56 records with no main text; 10 with fewer than 20 characters after markup tags were removed; one with an alphabetic\-character ratio below \.30; and six with summaries shorter than six characters\. These exclusions left 54,537 training\-eligible records, of which 52,360 had unique summaries \(96\.0%\)\. The dataset initially contained 24 court\-title labels, including alternative names or spellings for the same courts\. We consolidated three pairs:Eyüp MahkemesiandEyüb Mahkemesi \(Havass\-ı Refia\);Anadolu Sadâreti MahkemesiandAnadolu Sadareti Mahkemesi; andRumeli Sadâreti MahkemesiandRumeli Sadareti Mahkemesi\. This standardization resulted in 21 unique courts\.

#### 1\.1\.3Corpus Temporal Coverage

The courts are not equally represented over time \(see Figure[1](https://arxiv.org/html/2611.24981#S1.F1)\)\. While Bab, Istanbul, Galata, and Üsküdar persisted over decades, other courts appear only for a small period\. Across 54,614 documents with text, there are 11,497,610 words, corresponding to an average of 210\.5 words per document\. The number of documents and words varies notably across courts, as shown in Figure[1](https://arxiv.org/html/2611.24981#S1.F1)\.

Figure 1:Temporal Coverage and Corpus Size Across Ottoman Courts![Refer to caption](https://arxiv.org/html/2611.24981v1/court_distribution.png)

### 1\.2Representing Cultural\-Psychological Constructs

We focused on three cultural\-psychological constructs: individualism, collectivism, and honor\.

#### 1\.2\.1Questionnaire Items

Questionnaire items are central to CCR because they provide theoretically grounded semantic anchors for locating a construct within the embedding space\([Atari et al\., 2023](https://arxiv.org/html/2611.24981#bib.bib2);[Teitelbaum and Simchon, 2025](https://arxiv.org/html/2611.24981#bib.bib86)\)\. CCR embeds the questionnaire items and target documents in a shared semantic space and estimates document relevance from their similarity to the resulting construct representation\. The content of the questionnaire items therefore determines what the model recognizes as linguistic evidence of a construct\. Because validated measures of these constructs are unavailable in Ottoman Turkish, we manually translated each questionnaire item into Latin\-transcribed Ottoman Turkish\.

##### Individualism and Collectivism

We represented individualism and collectivism using the 22\-item measure developed by[Oyserman \(1993\)](https://arxiv.org/html/2611.24981#bib.bib20)\. The measure contains separate sets of items assessing individualistic and collectivistic orientations\.

##### Honor

We represented honor using 25 items drawn from[Rodriguez Mosquera et al\. \(2002\)](https://arxiv.org/html/2611.24981#bib.bib19)\. The measure includes items assessing moral integrity, family honor, feminine honor, and masculine honor, with one item representing both feminine and family honor\.

#### 1\.2\.2Proxy Sentences and Expert Definitions

We expanded each item into five aligned and five opposed sentences designed to express semantically aligned and opposed meanings\([Teitelbaum and Simchon, 2025](https://arxiv.org/html/2611.24981#bib.bib86), see\)\. We also prepared a single\-paragraph expert definition for each construct\. Illustrative examples of the questionnaire items, Ottoman Turkish adaptations, aligned and opposed proxy sentences, and expert\-authored definitions are provided in Appendix[B](https://arxiv.org/html/2611.24981#A2)\(Table[7](https://arxiv.org/html/2611.24981#A2.T7)\)\. The complete materials are provided in the Supplementary Materials\.

### 1\.3Expert\-Annotated Evaluation Set

#### 1\.3\.1Sampling Procedure

We constructed an expert\-annotated evaluation set by sampling 20 documents from each court label represented in the metadata\. We sampled records within courts and stratified the selection by decade, excluding documents used in an earlier pilot sample of five records per court\. Because two labels for the Eyüp court —Eyüp MahkemesiandEyüb Mahkemesi \(Havass\-ı Refia\)— were sampled separately before court names were consolidated, Eyüp contributed 40 documents, whereas each other court contributed 20\. Therefore, this procedure resulted in 440 documents\.

#### 1\.3\.2Annotation and Label Aggregation

Three annotators, including a co\-first author, independently coded each document for individualism, collectivism, and honor\. Annotators followed a detailed codebook containing construct definitions and positive and negative examples drawn from the corpus\. They assigned each document a separate label for each construct: present \(11\), absent \(00\), or uncertain \(−1\-1\)\. All annotators held bachelor’s degrees in history or Turkish philology and had relevant knowledge of Ottoman Turkish and training in Ottoman history\. We combined the annotations into a binary document\-level label through majority vote\. We treated uncertain labels as missing and calculated the majority over the remaining valid labels\. For a given construct, we excluded documents when the two remaining valid labels disagreed\. This procedure yielded honor labels for 435 documents \(31 present and 404 absent\), individualism labels for 440 documents \(368 present and 72 absent\), and collectivism labels for 438 documents \(84 present and 354 absent\)\. The complete annotation codebook is provided in the Supplementary Materials\.

#### 1\.3\.3Inter\-Annotator Agreement

Based on prior work on psychological annotation\([Zewail et al\., 2026](https://arxiv.org/html/2611.24981#bib.bib89)\), we evaluated inter\-annotator agreement using Gwet’s AC1, Krippendorff’sα\\alpha, Fleiss’κ\\kappa, PABAK, and percentage agreement\. We treated Gwet’s AC1 as the primary coefficient because the prevalence of positive labels differed substantially across constructs, particularly for honor\([Gwet, 2008](https://arxiv.org/html/2611.24981#bib.bib87);[Krippendorff, 2004](https://arxiv.org/html/2611.24981#bib.bib88)\)\.

### 1\.4Model Development

#### 1\.4\.1Data Splits and Evaluation Separation

After removing two duplicate summary–text pairs, the model\-development corpus contained 54,535 records\. We held out the 440 expert\-annotated documents from all model training and checkpoint selection\. We divided the remaining 54,095 records into approximately 80% training, 10% validation, and 10% test sets using a fixed random seed of 42\. We performed the split at the summary level, ensuring that records sharing the same summary were assigned to the same subset\. We did not stratify the data by court or decade\. The resulting subsets contained 43,108 training records, 5,715 validation records, and 5,272 test records\. We selected checkpoints using the validation set and conducted document\-retrieval and embedding\-space evaluations on the held\-out test set, using the same records and sampled pairs for every model\.

#### 1\.4\.2Fine\-tuning Overview

The training\-pair procedure used in earlier historical applications of CCR identifies related documents through the similarity of broad thematic titles\([Chen et al\., 2024](https://arxiv.org/html/2611.24981#bib.bib9)\)\. The Ottoman court records instead contain one\-sentence summaries with personal names and specific legal acts, which can produce high similarity between otherwise unrelated cases\. We therefore paired each summary directly with its corresponding court record\.

We adapted OTA\-XLM\-RoBERTa\([Yılandıloğlu, 2025](https://arxiv.org/html/2611.24981#bib.bib57)\)and BERTurk\([Schweter, 2020](https://arxiv.org/html/2611.24981#bib.bib55)\)through two training stages\. First, we used unsupervised SimCSE\([Gao et al\., 2021](https://arxiv.org/html/2611.24981#bib.bib29)\)to obtain sentence\-level representations from the masked\-language encoders\. Second, we trained the models on summary–record pairs using Multiple Negatives Ranking Loss \(MNRL\), treating other records in the batch as negatives\. We adapted the long\-context EmbeddingGemma\-2 model using the same summary–record objective but without the preliminary SimCSE stage because it was already trained as a sentence\- and document\-embedding model\. The Supplementary Materials provide the complete training procedures, hyperparameters, prompt formats, and implementation settings\.

### 1\.5Document\-Level Construct Scoring

#### 1\.5\.1Pooling Strategies

We calculated cosine similarity between each construct representation and the Ottoman court records using three pooling strategies\. The whole\-document approach encoded the available text of each record as a single vector:

Scoredoc⁡\(D,q\)=cos⁡\(𝐞⁡\(D\),𝐞⁡\(q\)\),\\operatorname\{Score\}\_\{\\mathrm\{doc\}\}\(D,q\)=\\cos\\left\(\\mathbf\{e\}\(D\),\\mathbf\{e\}\(q\)\\right\),\(1\)
whereDDdenotes a court record andqqdenotes a questionnaire item, proxy sentence, or definition sentence\. Because a construct may be expressed in only part of a record, we also divided each record into sentences\. Maximum pooling retained the highest sentence\-level similarity:

Scoremax⁡\(D,q\)=maxsi∈D⁡cos⁡\(𝐞⁡\(si\),𝐞⁡\(q\)\),\\operatorname\{Score\}\_\{\\mathrm\{max\}\}\(D,q\)=\\max\_\{s\_\{i\}\\in D\}\\cos\\left\(\\mathbf\{e\}\(s\_\{i\}\),\\mathbf\{e\}\(q\)\\right\),\(2\)
whereas mean pooling averaged similarity across all sentences:

Scoremean⁡\(D,q\)=1N​∑i=1Ncos⁡\(𝐞⁡\(si\),𝐞⁡\(q\)\),\\operatorname\{Score\}\_\{\\mathrm\{mean\}\}\(D,q\)=\\frac\{1\}\{N\}\\sum\_\{i=1\}^\{N\}\\cos\\left\(\\mathbf\{e\}\(s\_\{i\}\),\\mathbf\{e\}\(q\)\\right\),\(3\)
whereNNis the number of sentences inDD\. Maximum sequence lengths and sentence\-segmentation procedures are reported in the Supplementary Materials\.

#### 1\.5\.2Construct Representations

We represented each construct in four ways\.Itemsused only the translated questionnaire items\.Alignedcombined each original item with five semantically aligned proxy sentences\.Contrastadditionally incorporated five opposed proxy sentences for each item\.Definitionused the sentences of the expert\-authored construct definition\.

For a set of construct queriesQQ, we averaged the query\-level scores:

Score⁡\(D,Q\)=1\|Q\|​∑q∈QScore⁡\(D,q\),\\operatorname\{Score\}\(D,Q\)=\\frac\{1\}\{\|Q\|\}\\sum\_\{q\\in Q\}\\operatorname\{Score\}\(D,q\),\(4\)
whereScore⁡\(D,q\)\\operatorname\{Score\}\(D,q\)was calculated using one of the three pooling strategies described above\. For contrast scoring, we subtracted the mean similarity to the opposed query set from the mean similarity to the aligned query set:

Scorecontrast⁡\(D,C\)=Score⁡\(D,QC\+\)−Score⁡\(D,QC−\),\\operatorname\{Score\}\_\{\\mathrm\{contrast\}\}\(D,C\)=\\operatorname\{Score\}\(D,Q\_\{C\}^\{\+\}\)\-\\operatorname\{Score\}\(D,Q\_\{C\}^\{\-\}\),\(5\)
whereQC\+Q\_\{C\}^\{\+\}andQC−Q\_\{C\}^\{\-\}contain the aligned and opposed representations of constructCC, respectively\.

Honor was represented by 25 questionnaire items, whereas individualism and collectivism were each represented by 11 items\. Each item was expanded into five aligned and five opposed proxy sentences\. The Supplementary Materials provide the original items, Ottoman Turkish adaptations, proxy sentences, and expert\-authored definitions\.

### 1\.6Comparison Methods

#### 1\.6\.1Embedding Models

We evaluated the domain\-adapted models alongside off\-the\-shelf masked\-language, sentence\-embedding, and long\-context embedding models\. The complete list of models, identifiers, and implementation settings is provided in Appendix[A](https://arxiv.org/html/2611.24981#A1)\(Table[6](https://arxiv.org/html/2611.24981#A1.T6)\)\.

#### 1\.6\.2LLM Benchmark

We evaluated seven open\-weight LLMs as automated classifiers of the 440 expert\-coded documents\. Each model classified the presence \(11\) or absence \(00\) of honor, individualism, and collectivism under both zero\-shot and few\-shot prompting, resulting in six classifications per document\. We used greedy decoding without sampling and disabled model reasoning\. We compared the predictions with the expert\-derived document labels\. The Supplementary Materials provide model identifiers, prompt templates, quantization and inference settings, API\-access details, output constraints, and post\-processing procedures\.

### 1\.7Evaluation Tasks

We evaluated the models at two levels\. First, representation\-level evaluations assessed whether the embeddings captured general semantic relationships in Ottoman legal language, independently of the expert construct labels\. These evaluations included document retrieval, questionnaire\-item classification, and semantic textual similarity\. Second, document\-level validation assessed whether the embedding\-based CCR scores and LLM classifications corresponded to expert annotations of honor, individualism, and collectivism\.

#### 1\.7\.1Document Retrieval

We evaluated summary–record alignment on the held\-out test split\. Each case summary served as a query, and the model ranked the 5,272 test records according to cosine similarity\. A record was considered relevant when it corresponded to the query summary\. When multiple records shared the same summary, all corresponding records were treated as relevant\. We used the same queries, candidate records, and relevance judgments for every model\. We report Recall@1, Recall@5, Recall@10, MRR@10, and NDCG@10\. Recall@kkmeasures the proportion of all records associated with a query summary that appear among the firstkkretrieved records, averaged across queries\. MRR@10 reflects the rank of the first relevant record, whereas NDCG@10 evaluates the ranking of all relevant records within the first ten positions\.

#### 1\.7\.2Questionnaire\-Item Classification

The QIC task evaluates how effectively sentence embeddings capture the boundaries between psychological concepts\. Using an SVM classifier on item\-level vector representations of the questionnaire, this evaluation demonstrates the capability of domain\-adapted sentence transformer models to reliably distinguish psychological constructs\. More specifically, we analyze how well a given model classifies the items in the questionnaires\. We used each questionnaire item and its five aligned proxy sentences as separate samples \(282 samples from 47 items: 150 honor, 66 individualism, 66 collectivism\)\. Items were classified with a support vector machine \(RBF kernel,C=1\.0C=1\.0,γ=scale\\gamma=\\text\{scale\}\) with balanced class weights, using fixed hyperparameters without additional tuning\. We used 10\-fold cross\-validation grouped by questionnaire item \(GroupKFold\), thus an item and its proxy sentences never appear in both the training and the test folds\. Reported values are the means across the 10 folds, and the values after±\\pmare the standard deviations across folds\. All items in the honor scale from[Rodriguez Mosquera et al\. \(2002\)](https://arxiv.org/html/2611.24981#bib.bib19)are labeled as ‘honor,’ while items from[Oyserman \(1993\)](https://arxiv.org/html/2611.24981#bib.bib20)are labeled as either ‘individualism’ or ‘collectivism\.’ For the EmbeddingGemma models, the items and proxy sentences were encoded with theDocumentprompt\.

#### 1\.7\.3Semantic Textual Similarity

In addition to QIC, we implemented a pairwise STS evaluation task to assess how the models distinguish semantically related from unrelated sentence pairs in Ottoman Turkish legal texts\. First, we prepared 100 semantically similar within\-topic pairs in Ottoman Turkish on legal topics such as debt, land dispute, and embezzlement\. Ottoman Turkish sentences cover a wide range of styles, from heavy legal language to plain prose that stays close to modern Turkish\. As none of these sentences were in the training data of the models, this dataset serves as novel, unseen data for evaluation\. The data can be accessed through Supplementary Materials\.

Then we constructed 19,900 pairwise combinations from the 200 sentences, consisting of 100 semantically similar within\-topic pairs \(positive class\) and 19,800 cross\-topic dissimilar pairs \(negative class\)\. Cosine similarity scores were extracted for each pair\. We report two kinds of metrics\. First, retrieval metrics: each sentence is used as a query against its matched sentence, and vice versa, and we report the Mean Reciprocal Rank \(MRR\) and Recall@1 with 95% bootstrap confidence intervals over topics\. Second, ROC\-AUC: the cross\-register AUC uses the 100 positives and only the negatives, and the all\-pairs AUC uses all 19,900 pairs\. We also report the area under the precision–recall curve \(PR\-AUC\), because 100 positives are outnumbered by 19,800 negatives, and ROC\-AUC can look high even when precision at a usable threshold is low\. For the EmbeddingGemma models, both sides of a pair \(and of a questionnaire\-item pair in QIC\) were encoded with theDocumentprompt, because both sides are the same kind of object\. Other prompts are reported in the Supplementary Materials\. We compare models with paired bootstrap differences over topics\.

#### 1\.7\.4Comparative Document\-Level Validation

We evaluated two approaches to document\-level construct measurement against the expert\-derived labels\. The embedding\-based approach produced continuous CCR scores for honor, individualism, and collectivism\. The zero\-shot and few\-shot LLM approaches produced binary predictions indicating whether each construct was present \(11\) or absent \(00\)\. All methods were evaluated on the same expert\-annotated documents, which were excluded from model training and checkpoint selection\.

For the continuous embedding scores, we report ROC\-AUC with 95% confidence intervals obtained through 1,000\-iteration bootstrap resampling\. We additionally report average precision because the prevalence of positive labels differed substantially across constructs: 7% for honor, 84% for individualism, and 19% for collectivism\. Pearson’srrand Spearman’sρ\\rhoare reported in the Supplementary Materials\.

For the binary LLM predictions, we report balanced accuracy, F1, precision, sensitivity, and specificity\. For a hard binary predictor, ROC\-AUC is numerically equivalent to balanced accuracy\. Nevertheless, continuous embedding scores provide more detailed ranking information than binary LLM outputs, so direct differences between the two approaches should be interpreted cautiously\. We conducted all analyses separately for honor, individualism, and collectivism\.

### 1\.8Qualitative Examination of High\-Scoring Records

To interpret what the document\-level construct scores captured, we ranked the records separately by their individualism, collectivism, and honor scores, calculated using OTA\-BERTurk with contrast scoring and whole\-document embeddings, and qualitatively examined the top\-10 scoring records per construct\. For each record, we considered the document type, the actors and relationships represented, and the passages most closely associated with the construct\. The upper tail of the score distribution differed substantially across constructs\. For example, only 35 individualism records but 208 honor records scored more than three standard deviations above their respective corpus means\. We therefore used rank rather than a standardized\-score threshold to define the primary qualitative sample\. For honor, we additionally included records that are outside the top\-10 sample that nevertheless received extreme scores \(i\.e\., higher than three SDs above the mean\)\.

### 1\.9Transparency and Openness

We report how the corpus and expert\-annotated evaluation sample were constructed, all data exclusions, all construct representations, and all evaluation procedures used in this study\. The original Latin\-script transcriptions of the Ottoman court records are accessible through the Istanbul Qadi Registers project at[https://kadisicilleri\.org](https://kadisicilleri.org/)\. Due to copyright restrictions, we do not redistribute the complete record texts\. The domain\-adapted models are available at https://zenodo\.org/records/24282664\. The analysis and model\-training code, scoring pipeline, preprocessing procedures, and code used to reproduce the tables and figures are available at https://zenodo\.org/records/24282664\. The translated questionnaire items, aligned and opposed proxy sentences, expert\-authored construct definitions, annotation codebook, LLM prompts, expert annotations, and evaluation materials are available in the Supplementary Materials and at https://zenodo\.org/records/24282664\. The interactive web interface is available at https://koras0ff\.github\.io/ottoman\-court\-explorer/\. Of note, neither the study design nor the analyses were preregistered\.

### 1\.10Use of Generative Artificial Intelligence in Manuscript Preparation

During manuscript preparation, the authors used ChatGPT\([OpenAI, 2026](https://arxiv.org/html/2611.24981#bib.bib79)\)and Claude\([Anthropic, 2026](https://arxiv.org/html/2611.24981#bib.bib80)\)to assist with language editing andLaTeXformatting\. The authors reviewed and revised all AI\-assisted material and independently verified the accuracy of factual claims and citations\. This use was limited to manuscript preparation and was distinct from the open\-weight LLM benchmark described above\. The tool was not used to generate or modify the study data or expert annotations, conduct statistical analyses, or determine the reported results\.

## 2Results

### 2\.1Model Performance Evaluation

#### 2\.1\.1Document Retrieval

Table[1](https://arxiv.org/html/2611.24981#S2.T1)presents the document\-retrieval results on the held\-out model test set\. The SimCSE stage alone did not substantially improve retrieval: NDCG@10 remained \.005 for the XLM\-RoBERTa model and \.024 for the BERTurk model\. Performance improved substantially after summary–record training, with NDCG@10 reaching \.889 for OTA\-XLM\-RoBERTa, \.898 for OTA\-BERTurk, and \.955 for OTA\-EmbeddingGemma\-2\. Among the domain\-adapted models, OTA\-EmbeddingGemma\-2 achieved the strongest overall document\-retrieval performance\.

Table 1:Document Retrieval Performance on the Test SetNote\.Bold values indicate the highest performance in each column\. NDCG@10 = normalized discounted cumulative gain at 10; MRR@10 = mean reciprocal rank at 10; R@1, R@5, and R@10 = recall at ranks 1, 5, and 10, respectively\.

#### 2\.1\.2Questionnaire\-Item Classification

We evaluated model performance across a 3\-construct classification scheme \(individualism, collectivism, and honor\)\. Table[2](https://arxiv.org/html/2611.24981#S2.T2)summarizes accuracy, macro F1, and weighted F1 for the masked language models and our domain\-adapted models\. The performance of off\-the\-shelf embedding models is reported in the Supplementary Materials\. Among our models, OTA\-BERTurk performed best \(89\.58%89\.58\\%accuracy\), outperforming its base model BERTurk \(85\.08%85\.08\\%\)\. Several larger general\-purpose embedding models were higher, such as Octen\-Embed 8B \(93\.50%93\.50\\%\)\. OTA\-XLM\-RoBERTa \(80\.17%80\.17\\%\) was far above the XLM\-RoBERTa masked language models \(47\.17%47\.17\\%\), which score at about the share of the largest class \(honor, 53%\)\. Fine\-tuning did not improve EmbeddingGemma\-2: OTA\-EmbeddingGemma\-2 scored85\.25%85\.25\\%with theDocumentprompt, against87\.00%87\.00\\%for the base model, and it performed worse under every prompt \(Supplementary Materials\)\. A plausible reason is that the summary–text objective teaches the model to match short summaries to long records, which is a different task than classifying short questionnaire items\.

Table 2:Model performance on the QIC task\.Note\.Bold values indicate the best result within each model family\. Accuracy is the percentage of samples assigned to the correct class\. Macro F1 is the unweighted mean of the class\-specific F1 scores\. Weighted F1 averages the class\-specific F1 scores according to class size\.

Table 3:Performance on the Semantic Textual Similarity TaskNote\.Bold values indicate the best result within each model family\. MRR = mean reciprocal rank; R@1 = recall at 1; Cross\-reg\. AUC = cross\-register area under the receiver operating characteristic curve; AUC \(all\) = area under the receiver operating characteristic curve calculated across all pairs; PR\-AUC = area under the precision–recall curve\.

#### 2\.1\.3Semantic Textual Similarity

Performance on the semantic textual similarity task varied considerably across model families \(Table[3](https://arxiv.org/html/2611.24981#S2.T3)\)\. The two largest general\-purpose embedding models were nearly perfect: Qwen3\-Embed 8B \(MRR = \.968, cross\-register AUC = \.999\) and Octen\-Embed 8B \(MRR = \.965, cross\-register AUC = \.998\)\. As their intervals overlap, we do not rank them\. Among our domain\-adapted models, OTA\-BERTurk was the strongest \(MRR = \.806 \[\.740, \.866\]\), followed by OTA\-EmbeddingGemma\-2 \(MRR = \.762\) and OTA\-XLM\-RoBERTa \(MRR = \.762\)\. Paired bootstrap differences \(MRR\) among the three adapted models were not significant \(OTA\-BERTurk versus OTA\-EmbeddingGemma\-2:Δ\\DeltaMRR=\.043=\.043\[−\.022,\.104\]\[\-\.022,\.104\]; versus OTA\-XLM\-RoBERTa:\.044\.044\[−\.006,\.095\]\[\-\.006,\.095\]\)\. Thus, STS scores do not clearly separate them alone\. Domain adaptation clearly helped each base model as EmbeddingGemma\-2 went from an MRR of \.607 to \.762 \(cross\-register AUC from \.897 to \.961\), and the OTA\-XLM\-RoBERTa masked language model went from \.351 to \.762\. The choice of prompt made no significant difference for OTA\-EmbeddingGemma\-2 \(Supplementary Materials\)\. Precision stays modest\. Yet, the PR\-AUC is \.521 for OTA\-BERTurk and \.334 for OTA\-EmbeddingGemma\-2, even though their ROC\-AUCs are above \.90\. In Table[3](https://arxiv.org/html/2611.24981#S2.T3), the best value within each model category is marked in bold; the off\-the\-shelf embedding models are reported in the Supplementary Materials\.

### 2\.2Expert\-Annotation Reliability

Agreement varied considerably across the three constructs \(Table[4](https://arxiv.org/html/2611.24981#S2.T4)\)\. For honor, annotations reached 86% observed agreement, and Gwet’s AC1 was 0\.89, indicating substantial agreement\. The low Krippendorffα\\alpha\(0\.37\) and Fleissκ\\kappa\(0\.40\) for honor can be explained by the skew in the annotations rather than genuine disagreement\. Individualism is the most reliable construct, with high agreement across every metric \(AC1 = 0\.89,α\\alpha= 0\.71,κ\\kappa= 0\.71, PABAK = 0\.83; 88% raw agreement\)\. Agreement estimates for collectivism varied across metrics \(AC1 = 0\.53;α\\alpha= 0\.17;κ\\kappa= 0\.17\), with raw agreement of 55%\. This divergence may partly reflect substantial differences in annotators’ use of the positive label: annotator 1 assigned it to 205 of 440 items, compared with 29 and 73 items for annotators 2 and 3, respectively\.

Table 4:Inter\-Annotator Agreement Scores- •Note\.AC1 = Gwet’s first\-order agreement coefficient;α\\alpha= Krippendorff’s alpha;κ\\kappa= Fleiss’ kappa; PABAK = prevalence\-adjusted bias\-adjusted kappa\. Values in brackets are 95% confidence intervals\. % Agreement indicates percentage agreement among the annotators\.

### 2\.3Document\-Level Validation Against Expert Annotations

Table[5](https://arxiv.org/html/2611.24981#S2.T5)shows the ROC\-AUC of the MLM encoders and our domain\-adapted models under one fixed configuration \(the contrast score with whole\-document embeddings\) and the balanced accuracy of the four LLMs discussed below under zero\-shot and few\-shot prompting; the off\-the\-shelf embedding models and the remaining LLMs are in the Supplementary Materials\. No single approach consistently outperformed the others across all constructs\.

Table 5:Document\-Level Validation Against Expert Annotations- •Note\.Values for MLM encoders and domain\-adapted embedding models are ROC\-AUCs with 95% bootstrap confidence intervals\. Results for off\-the\-shelf embedding models and the remaining LLMs are reported in the Supplementary Materials\.

#### 2\.3\.1Individualism

Detecting individualism was challenging for our domain\-adapted embedding models, whereas few\-shot LLMs performed better\. Among the MLM encoders, BERTurk achieved the highest ROC\-AUC \(\.712 \[\.639, \.783\]\), while Jina Embeddings v5 Small was the strongest off\-the\-shelf embedding model \(\.710 \[\.636, \.781\]; see Supplementary Materials\)\. Among our models, OTA\-EmbeddingGemma\-2 performed best, although its ROC\-AUC remained close to chance \(\.556 \[\.475, \.637\]\); OTA\-XLM\-RoBERTa performed below chance \(\.424\)\. LLM performances varied and notably depended on the prompting strategy\. Zero\-shot balanced accuracy ranged from \.496 to \.560, whereas few\-shot Qwen 3\.8 27B reached \.833, followed by Qwen 3\.5 9B \(\.778\) and Gemma 4\-E4B\-it \(\.768\)\.

#### 2\.3\.2Collectivism

For collectivism, our models fared well\. OTA\-XLM\-RoBERTa had the highest AUC of all embedding models \(\.727 \[\.663, \.793\]\), although its interval overlaps with that of the best off\-the\-shelf model, Octen\-Embed 8B \(\.666 \[\.600, \.730\]; Supplementary Materials\)\. OTA\-BERTurk \(\.664\) was on par with it, but OTA\-EmbeddingGemma\-2 stayed close to chance \(\.529\)\. BERTurk was the best MLM encoder \(\.627 \[\.558, \.694\]\)\. The strongest LLM was zero\-shot Qwen 3\.8 27B \(balanced accuracy \.740\), slightly above our best AUC, although the two metrics cannot be compared directly\. Collectivism also has the least reliable labels of the three constructs, so these scores should be interpreted with caution, and tested in the future with larger labeled data\.

#### 2\.3\.3Honor

Honor is the construct where the embedding models did best\. The best off\-the\-shelf model, Qwen3\-Embed 8B, scored \.800 \[\.708, \.890\] \(Supplementary Materials\)\. The MLM encoders were behind: OTA XLM\-RoBERTa was the best of them \(\.697 \[\.599, \.790\]\)\. Our three domain\-adapted models have similar scores, OTA\-EmbeddingGemma\-2 on top \(\.697 \[\.587, \.796\]\), then OTA\-XLM\-RoBERTa \(\.687\) and OTA\-BERTurk \(\.654\)\. That puts the best off\-the\-shelf model about \.10 above our best one\. The intervals overlap, though, and with only 31 positive documents we cannot separate the groups with high confidence\. The LLMs did worse than the embedding models on honor: zero\-shot DeepSeek\-V4\-Flash reached a balanced accuracy of \.677, and zero\-shot Qwen 3\.8 27B was slightly less accurate \(\.669\)\.

### 2\.4Qualitative Examination of High\-Scoring Records

#### 2\.4\.1Individualism

The 10 highest\-scoring individualism records were all related to manumission \(see Table[8](https://arxiv.org/html/2611.24981#A3.T8)\)\. These documents identify enslaved persons through their names, origins, physical characteristics, and legal status before formally recording their transition to freedom\. In one case, Mehmed Çelebi declares that he has freed Ruhısâni, an enslaved woman of Russian origin, for the sake of God and divine approval; the court subsequently recognizes her as free, like other freeborn adult women\([Yılmaz, 2010b](https://arxiv.org/html/2611.24981#bib.bib72), judgment no\. 573, fol\. 85b, entry 3, p\. 428\)\. Another record documents, through an authorized representative and witnesses, the manumission of Şaban, an enslaved man of European origin\([Yılmaz, 2011](https://arxiv.org/html/2611.24981#bib.bib73), judgment no\. 32, original\-text no\. 17\-3, p\. 51\)\. Importantly, each deed formally reclassifies a named person from property to a free legal subject\. The model therefore appears to associate individualism primarily with legal individuation through emancipation, which is the formal recognition of a person as possessing a legal status separate from that of a former owner\.

#### 2\.4\.2Collectivism

The highest\-scoring collectivism records were dominated by religious endowment deeds \(see Table[9](https://arxiv.org/html/2611.24981#A3.T9)\)\. In one of the highest\-scoring examples, Musa Efendi dedicates a residence first to Kadızâde Şeyh Mehmed Efendi, then to his own descendants, and ultimately to a member of the Naqshbandi order or a poor Muslim\([Yılmaz, 2019a](https://arxiv.org/html/2611.24981#bib.bib70), judgment no\. 94, fol\. 17a, entry 1, p\. 120\)\. Successive occupants must administer, maintain, and repair the property\. When Musa Efendi attempts to revoke the endowment, the trustee refuses and the court declares it binding\. The model thus appears to associate collectivism with the transformation of individual wealth into a collective institution that links family, religious community, and future beneficiaries through shared responsibilities and intergenerational obligations\.

#### 2\.4\.3Honor

The highest\-scoring honor records were dominated by manumission deeds and other disputes over freedom and legal status \(see Table[10](https://arxiv.org/html/2611.24981#A3.T10)\)\. In one case, Fatıma Hanım formally frees Sertâbe, an enslaved woman of Russian origin, and the court records that she has joined the ranks of freeborn women and that her former owner retains no claim over her beyond the legally recognized bond of patronage\([Yılmaz, 2019b](https://arxiv.org/html/2611.24981#bib.bib69), judgment no\. 160, fols\. 33a–33b, entry 4, p\. 176\)\. Other records that are high\-scoring \(i\.e\., more than 3 SDs above the mean\) link honor more directly to the public evaluation of conduct\. A record, which was not on the top 10\-highest scoring cases, but was still 3\.47 SDs above the mean, describes a group identified by the court as Roma who admitted drinking wine and were accused of insulting janissaries and directing improper remarks toward Muslim families\. Community members further portrayed their behavior as habitual and socially disruptive, and these allegations were entered into the register\([Yılmaz, 2010a](https://arxiv.org/html/2611.24981#bib.bib71), judgment no\. 372, fol\. 35a, entry 4, p\. 250\)\. The model therefore appears to associate honor both with the formal recognition of social and legal standing and with communal judgments of morally appropriate conduct\. This interpretation is consistent with accounts of Ottoman honor as a relational process shaped through publicly negotiated reputation\([Peirce, 2014](https://arxiv.org/html/2611.24981#bib.bib42)\)\. However, this interpretation warrants caution because the highest\-scoring records were dominated by manumission deeds\. The pattern may reflect the formulaic language of legal status, freedom, and pious motivation more than the broader historical connotations of honor\.

### 2\.5Interactive Corpus Exploration Using the Web Interface

As a publicly available research resource, we release an open\-access web interface to analyze 54,670 Ottoman court records to quantify other psychological constructs\. The outputs in the interface are calculated using the OTA\-BERTurk model with contrastive scoring across the full document, the same approach used to generate results in Table[5](https://arxiv.org/html/2611.24981#S2.T5)\. The interface provides the outputs yielded by the CCR method \(an example shown in Figure[2](https://arxiv.org/html/2611.24981#S2.F2)\)\. The interface aggregates vector similarity scores for each psychological construct, providing a visualization of semantic trends across decades and court districts\. Users can view the outputs generated by both the base and fine\-tuned embedding models\. Additionally, users can select which constructs they want to view in the visualizations tab\.

Figure 2:An Example of Cultural\-Psychological Construct Loadings, Displayed on the Web Interface![Refer to caption](https://arxiv.org/html/2611.24981v1/interface_scores.png)The interface presents both high\-level statistical visualizations and court records themselves\. This allows researchers to inspect standardized construct scores over time\. Users can compare constructs across decades to identify historical variation \(see Figure[3](https://arxiv.org/html/2611.24981#S2.F3)\)\. We use embedding\-based scores in the interface rather than LLM annotation because the interface covers all 54,670 records\. An LLM would need six prompts per record \(three constructs times two prompt types\) over more than 50,000 records, which is slow and expensive, and raises reproducibility concerns\([McLoughlin et al\., 2026](https://arxiv.org/html/2611.24981#bib.bib90)\)\. Embedding the records once and scoring them by cosine similarity is fast and reproducible, since readers can inspect which sentences drive a score up or down\.

Figure 3:An Example of Psychological Variation Over Time, Displayed on the Web Interface\.![Refer to caption](https://arxiv.org/html/2611.24981v1/change.png)

## 3Discussion

We develop a domain\-adapted computational framework for analyzing Ottoman court records and demonstrate its application with examples to historical\-psychological measurement\. Following the idea that written records can serve as indirect archives of the psychological orientations of the populations that produced them\([Atari and Henrich, 2023](https://arxiv.org/html/2611.24981#bib.bib1)\), we treat the Ottoman court registers as a site where individual and collective identity can be examined empirically\. The framework identifies and compares the linguistic and institutional forms through which cultural\-psychological constructs, in this case individualism, collectivism, and honor, were expressed in the legal discourse of the historical Ottoman state\.

Across 54,670 records from 100 court\-register volumes, we adapted sentence\-embedding models to Ottoman legal language and evaluated the resulting representations across several tasks\. Domain adaptation generally improved summary–record alignment, document retrieval, and semantic textual similarity, although the levels of improvement varied across models and tasks, and larger off\-the\-shelf models sometimes performed better\. Document\-level performance depended jointly on the construct, model family, construct representation, and pooling strategy: no single model or scoring approach performed best across the constructs of interest\.

The results for individualism and collectivism revealed different strengths and limitations\. For individualism, the domain\-adapted embedding models performed weakly against the expert labels\. For collectivism, OTA\-XLM\-RoBERTa achieved the strongest performance among the embedding models\. The qualitative results for high\-scoring individualism and collectivism documents were conceptually coherent\. High\-scoring individualism records were consistently manumission records that show legal individuation, whereas high\-scoring collectivism records centered on endowments that dedicated private property to religious and communal purposes, creating lasting systems of shared responsibility and intergenerational obligation\. This systematic retrieval suggests that the pipeline identified historically situated textual patterns aligned with the intended constructs, even though quantitative performance varied across models and operationalizations\. Close reading of high\-scoring records further shows why historical interpretation must accompany computational scoring\.

Honor produced a comparatively complex pattern across the quantitative and qualitative evaluations\. At the document level, the embedding models distinguished honor more successfully than the other constructs: Qwen3\-Embed 8B achieved a ROC\-AUC of \.800, while the strongest domain\-adapted model, OTA\-EmbeddingGemma\-2, reached \.697\. The open\-weight LLMs performed less strongly, although their balanced accuracies are not directly comparable with the ROC\-AUCs calculated from continuous embedding scores\. These results indicate that the construct representations captured a signal corresponding to the expert honor labels\. Close reading, however, showed that the highest\-scoring records were dominated by manumission deeds, suggesting that the models may have responded strongly to their language of publicly recognized status, freedom, pious motivation, and moral self\-presentation\. Records more directly concerning reputation and communal evaluations of conduct also received high scores but appeared less frequently at the extreme upper end of the distribution\.

Future research should evaluate honor within documentary genres and develop construct representations grounded more directly in historically situated concepts\. We used the Honor Scale developed by[Rodriguez Mosquera et al\. \(2002\)](https://arxiv.org/html/2611.24981#bib.bib19), a well\-established cross\-cultural measure that has also been adapted and validated in Turkish samples\([Ceylan\-Batur and Sakallı Uğurlu, 2019](https://arxiv.org/html/2611.24981#bib.bib74)\)\. Nevertheless, honor is not a unitary concept, and the scale’s dimensions may not cover all the meanings expressed in Ottoman legal discourse\. In the Ottoman state, honor was commonly employed by terms such asırzandnamusand operated through relational processes of public reputation even when it was not named directly\([Peirce, 2014](https://arxiv.org/html/2611.24981#bib.bib42)\)\. Contemporary related terms do not necessarily denote the same phenomenon: research on Turkish usage distinguishes amongnamus,şeref, andonur, associating them with different aspects of sexual and familial reputation, public standing, and dignity\([Dilmaç, 2014](https://arxiv.org/html/2611.24981#bib.bib75)\)\. These meanings also changed historically; for example,ırzacquired specifically gendered and legal meanings through nineteenth\-century Ottoman penal codification\([Tuğ Onaran, 2022](https://arxiv.org/html/2611.24981#bib.bib76)\)\. Future work could therefore construct separate representations of these concepts that are period\-sensitive\.

This result supports treating contemporary constructs as analytical starting points rather than historically invariant categories\. Repeated retrieval of a narrower, historically coherent pattern may reveal legal or institutional configuration that aligns with one component of the contemporary construct, or how part of that contemporary construct was organized differently in the given discourse\. At the same time, interpretive coherence alone does not establish validity\. The proposed proxy\-sentence method is most useful when it remains traceable, meaning researchers can inspect the sentences that define a construct, examine the passages that drive a document’s score, compare alternative scales or expert definitions, and revise the representation when it retrieves an unintended legal or linguistic pattern\. This iterative movement is precisely what the framework is designed to support\.

### 3\.1Limitations

Several limitations should be acknowledged\. First, the registers are institutionally mediated documents written by court scribes, not transparent reports of private beliefs, everyday speech, or population\-level prevalence\. Their contents reflect who used the courts, which disputes entered their jurisdiction, how cases were summarized, and which documentary formulas scribes employed\. Construct scores should consequently be interpreted as properties of recorded legal discourse, not direct estimates of the psychological traits of Ottoman populations\.

Second, the expert benchmark is small relative to the corpus, and agreement varies markedly by construct\. High agreement for individualism supports firmer conclusions, whereas weaker agreement for collectivism makes model\-human correspondence more difficult to interpret\. Honor presents the additional problem of strong prevalence imbalance, which can yield high raw agreement and Gwet’s AC1 alongside much lower kappa\-type coefficients\. Honor has only 31 positive documents, so its intervals are wide, and the cross\-document analysis \(131 positive pairs\) and the STS set \(100 pairs\) are small\. Our comparison between the sentence transformers \(256 tokens\) and EmbeddingGemma\-2 \(2,048 tokens\) also mixes model and context length\. Finally, pooling choices can reverse a construct’s ranking, thus document\-level scores depend on the configuration\.

Third, as we have discussed before, the constructs examined here are not culturally or historically neutral\. Although individualism and collectivism were developed to explain cross\-cultural variation, their dominant definitions and measurement instruments emerged largely from contemporary academic psychology and may carry assumptions that do not transfer to Ottoman society\([Berry, 1969](https://arxiv.org/html/2611.24981#bib.bib53)\)\. As discussed, honor may have greater contextual relevance given its importance in Turkish cultural psychology, but contemporary formulations of honor likewise cannot be assumed to capture its historical meanings\. Moreover, the questionnaire items, proxy sentences, and expert definitions used to represent these constructs shape what the models can recognize as relevant evidence\. Construct scores should therefore be treated as properties of the specific discourse rather than direct measurements of psychological traits\. Therefore close\-reading should complement quantitative scores when possible\.

### 3\.2Future Directions

The study of historical texts opens up new possible questions to be asked and answered by social science researchers, and especially for cultural psychologists who are interested in cultural change\. Previous research has used historical literature to trace changes in romantic love\([Baumard et al\., 2022](https://arxiv.org/html/2611.24981#bib.bib77)\), prosociality\([Martins and Baumard, 2020](https://arxiv.org/html/2611.24981#bib.bib78)\), contemporary psychological variation and historical information about Church exposure and kinship institutions to explain contemporary psychological variation in individualism, conformity and impersonal prosociality\([Schulz et al\., 2019](https://arxiv.org/html/2611.24981#bib.bib64)\)\. Similarly, our framework opens new avenues for examining how cultural\-psychological patterns varied across time and geography – in this case, across four centuries and regions on three continents over which the Ottoman Empire exercised authority\. Temporal analyses can investigate whether such cultural\-psychological constructs became more or less prominent during periods such as imperial expansion or political decline, and look at how specific administrative reforms, economic crises, or certain demographic changes have affected the cultural\-psychological constructs reflected in historical writings\. Comparisons with contemporaneous archives from other societies could then examine whether such patterns were specific to the Ottoman context or reflected broader developments occurring across states during the same historical period\. Such analyses can also generate historically grounded hypotheses about the origins of contemporary cultural\-psychological patterns, which can then be tested using data from living populations\.

Our framework also supports questions about demographic diversity\. The 21 courts represented in the corpus include courts serving Muslim\-majority and non\-Muslim\-majority neighborhoods \(see Supplementary Materials, Table S1, for the complete list\), as well as courts with specialized institutional jurisdictions\. Across these different social settings, the courts formed part of a shared judicial apparatus through which the Ottoman state exerted its authority\. Researchers can therefore ask whether courts used different linguistic patterns when recording the claims of different communities or represented agency and moral responsibility differently across social settings\. Such comparisons may reveal the diverse moral worlds and forms of social life that operated in the historical neighborhoods of the Ottoman Empire\.

Much can be done to strengthen and refine the present framework and improve its results\. Future work should expand and adjudicate the expert\-coded sample, report construct\-specific error analyses, validate scores against independent historical classifications, and test robustness across construct definitions and pooling strategies\. Diachronic and cross\-court analyses should use balanced samples or multilevel models and should be paired with close readings of the records that drive apparent differences\. The web interface can support this work by allowing researchers to move from aggregate patterns to the underlying passages\. Used in this way, computational scores can accompany historical interpretation by providing a reproducible method for locating, comparing, and refining hypotheses about culturally and historically situated forms of cultural\-psychological constructs\.

## 4Conclusion

Ultimately, this application to Ottoman court registers broadens the borders of both psychology and history\. We introduced a reproducible computational framework that combines the construction of a 54,670 record corpus, the domain adaptation of language models to Ottoman legal discourse which was evaluated through semantic textual similarity and questionnaire\-item classification, and document\-level validation against expert annotations\. To represent psychological constructs, we compared different opperationalizations of the same construct, namely questionnaire items, proxy sentences, and expert\-authored definitions\. We complemented these quantitative evaluations with close readings of high\-scoring records to identify the textual and institutional patterns underlying the model outputs\. By making the corpus, adapted models, construct representations, scoring pipeline, and web interface publicly available, we provide a reusable methodological infrastructure for historical\-psychological research in Ottoman Empire\. The framework extends psychological inquiry beyond living respondents and modern survey instruments while complementing close historical reading with corpus\-scale analysis\. In doing so, it expands the temporal, cultural, and methodological scope of social\-scientific research and creates opportunities to investigate how culturally meaningful representations of agency, obligation, status, and reputation varied across courts and historical periods\.

## References

- S\. Abdurahman, M\. Atari, F\. Karimi\-Malekabadi, M\. J\. Xue, J\. Trager, P\. S\. Park, P\. Golazizian, A\. Omrani, and M\. DehghaniPerils and opportunities in using large language models in psychological research\.PNAS nexus3\(7\),pp\. pgae245\.Cited by:[§0\.3](https://arxiv.org/html/2611.24981#S0.SS3.p1.1)\.
- Acerbiet al\.\(2013\)A\. Acerbi, V\. Lampos, P\. Garnett, and R\. A\. BentleyThe expression of emotions in 20th century books\.PLOS ONE8\(3\),pp\. e59030\.External Links:[Document](https://dx.doi.org/10.1371/journal.pone.0059030),[Link](https://doi.org/10.1371/journal.pone.0059030)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p1.1)\.
- Agmon \(2004\)I\. AgmonWomen’s history and ottoman sharia court records: shifting perspectives in social history\.2\(2\),pp\. 172–209\.External Links:[Document](https://dx.doi.org/10.1163/1569208041514680)Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1)\.
- Aiello and Simeone \(2019\)K\. D\. Aiello and M\. SimeoneTriangulation of history using textual data\.Isis110\(3\),pp\. 522–537\.Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p3.1)\.
- Akça \(2025\)S\. AkçaTopic modelling in the ottoman kadi registers\.InNew Horizons in Artificial Intelligence in Libraries,E\. Balnaves, L\. Bultrini, A\. Cox, and R\. Uzwyshyn \(Eds\.\),pp\. 213–226\.External Links:ISBN 9783111336435,[Document](https://dx.doi.org/10.1515/9783111336435-016)Cited by:[§1\.1\.1](https://arxiv.org/html/2611.24981#S1.SS1.SSS1.p1.1),[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Almelhemet al\.\(2026\)A\. Almelhem, M\. Iyigun, A\. Kennedy, and J\. RubinEnlightenment ideals and belief in progress in the run\-up to the industrial revolution: a textual analysis\.The Quarterly Journal of Economics141\(1\),pp\. 263–314\.Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p2.1)\.
- Anthropic \(2026\)AnthropicClaude\.Note:Large language modelExternal Links:[Link](https://claude.ai/)Cited by:[§1\.10](https://arxiv.org/html/2611.24981#S1.SS10.p1.1)\.
- Ashokkumar and Pennebaker \(2022\)A\. Ashokkumar and J\. W\. PennebakerTracking group identity through natural language within groups\.PNAS nexus1\(2\),pp\. pgac022\.Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p2.1)\.
- Atariet al\.\(2025\)M\. Atari, J\. Henrich, and J\. SchulzThe chronospatial revolution in psychology\.Nature Human Behaviour9\(7\),pp\. 1319–1327\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p4.1)\.
- Atari and Henrich \(2023\)M\. Atari and J\. HenrichHistorical psychology\.32\(2\),pp\. 176–183\.External Links:[Document](https://dx.doi.org/10.1177/09637214221149737)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p3.1),[§3](https://arxiv.org/html/2611.24981#S3.p1.1),[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p2.1)\.
- Atariet al\.\(2023\)M\. Atari, A\. Omrani, and M\. DehghaniContextualized construct representation: leveraging psychometric scales to advance theory\-driven text analysis\.PsyArXiv\.External Links:[Document](https://dx.doi.org/10.31234/osf.io/m93pd)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p4.1),[§1\.2\.1](https://arxiv.org/html/2611.24981#S1.SS2.SSS1.p1.1)\.
- Aykan and Ergene \(2020\)Y\. Aykan and B\. ErgeneShari’a courts in the ottoman empire before the tanzimat\.The Medieval History Journal22\(2\),pp\. 203–228\.Note:Original work published 2019External Links:[Document](https://dx.doi.org/10.1177/0971945819897437),[Link](https://doi.org/10.1177/0971945819897437)Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p1.1),[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Baldwin \(2017\)J\. E\. BaldwinIslamic law and empire in ottoman cairo\.Edinburgh University Press,Edinburgh\.External Links:ISBN 9781474403061,[Link](http://www.jstor.org/stable/10.3366/j.ctt1g052kk)Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1)\.
- Baumardet al\.\(2022\)N\. Baumard, E\. Huillery, A\. Hyafil, and L\. SafraThe cultural evolution of love in literary history\.Nature Human Behaviour6\(4\),pp\. 506–522\.External Links:[Document](https://dx.doi.org/10.1038/s41562-022-01292-z)Cited by:[§3\.2](https://arxiv.org/html/2611.24981#S3.SS2.p1.1)\.
- Baumardet al\.\(2024\)N\. Baumard, L\. Safra, M\. Martins, and C\. ChevallierCognitive fossils: using cultural artifacts to reconstruct psychological changes throughout history\.Trends in Cognitive Sciences28\(2\),pp\. 172–186\.Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p3.1),[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p2.1)\.
- Berry \(1969\)J\. W\. BerryOn cross\-cultural comparability\.International Journal of Psychology4\(2\),pp\. 119–128\.External Links:[Document](https://dx.doi.org/10.1080/00207596908247261),[Link](https://doi.org/10.1080/00207596908247261)Cited by:[§3\.1](https://arxiv.org/html/2611.24981#S3.SS1.p3.1),[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p4.1)\.
- Boyd \(2017\)R\. L\. BoydPsychological text analysis in the digital humanities\.InData analytics in digital humanities,pp\. 161–189\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p2.1)\.
- Ceret al\.\(2017\)D\. Cer, M\. Diab, E\. Agirre, I\. Lopez\-Gazpio, and L\. SpeciaSemEval\-2017 task 1: semantic textual similarity multilingual and crosslingual focused evaluation\.InProceedings of the 11th International Workshop on Semantic Evaluation \(SemEval\-2017\),S\. Bethard, M\. Carpuat, M\. Apidianaki, S\. M\. Mohammad, D\. Cer, and D\. Jurgens \(Eds\.\),Vancouver, Canada,pp\. 1–14\.External Links:[Document](https://dx.doi.org/10.18653/v1/S17-2001)Cited by:[§0\.3](https://arxiv.org/html/2611.24981#S0.SS3.p1.1)\.
- Ceylan\-Batur and Sakallı Uğurlu \(2019\)S\. Ceylan\-Batur and N\. Sakallı UğurluOnur Ölçeği: geçerlik ve güvenirlik Çalışması\.Nesne Psikoloji Dergisi7\(15\),pp\. 236–253\.External Links:[Document](https://dx.doi.org/10.7816/nesne-07-15-06)Cited by:[§3](https://arxiv.org/html/2611.24981#S3.p5.1)\.
- Charlesworthet al\.\(2022\)T\. E\. S\. Charlesworth, A\. Caliskan, and M\. R\. BanajiHistorical representations of social groups across 200 years of word embeddings from Google Books\.Proceedings of the National Academy of Sciences119\(28\),pp\. e2121798119\.External Links:[Document](https://dx.doi.org/10.1073/pnas.2121798119)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p2.1)\.
- Chenet al\.\(2024\)Y\. Chen, S\. Li, Y\. Li, and M\. AtariSurveying the dead minds: historical\-psychological text analysis with contextualized construct representation \(ccr\) for classical chinese\.InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing,pp\. 2597–2615\.External Links:[Link](http://dx.doi.org/10.18653/v1/2024.emnlp-main.151),[Document](https://dx.doi.org/10.18653/v1/2024.emnlp-main.151)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p4.1),[§0\.3](https://arxiv.org/html/2611.24981#S0.SS3.p1.1),[§1\.4\.2](https://arxiv.org/html/2611.24981#S1.SS4.SSS2.p1.1)\.
- Cosgel and Ergene \(2013\)M\. M\. Cosgel and B\. A\. ErgeneThe selection bias in court records: settlement and trial in eighteenth century kastamonu\.Working papersTechnical Report2013\-12,University of Connecticut, Department of Economics\.Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1)\.
- Coşgelet al\.\(2025\)M\. M\. Coşgel, J\. Espín\-Sánchez, and E\. ÖzerIntergenerational mobility in Ottoman Istanbul: evidence from court records\.Yale University, Economic Growth Center,New Haven, CT\.Note:EGC Discussion Paper No\. 1116External Links:[Link](https://hdl.handle.net/10419/324809)Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Dilmaç \(2014\)J\. A\. Dilmaç“Our” honor and “their” honor: the case of honor killings in turkey\.InGendered Perspectives on Conflict and Violence: Part B,M\. T\. Segal and V\. Demos \(Eds\.\),Advances in Gender Research, Vol\.18B,pp\. 251–274\.External Links:[Document](https://dx.doi.org/10.1108/S1529-21262014000018B014)Cited by:[§3](https://arxiv.org/html/2611.24981#S3.p5.1)\.
- Doumani \(2017\)B\. B\. DoumaniFamily life in the ottoman mediterranean: a social history\.Cambridge University Press\.External Links:[Document](https://dx.doi.org/10.1017/9780511989605),ISBN 9781107154544Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1)\.
- Eichstaedtet al\.\(2018\)J\. C\. Eichstaedt, R\. J\. Smith, R\. M\. Merchant, L\. H\. Ungar, P\. Crutchley, D\. Preoţiuc\-Pietro, D\. A\. Asch, and H\. A\. SchwartzFacebook language predicts depression in medical records\.Proceedings of the National Academy of Sciences115\(44\),pp\. 11203–11208\.External Links:[Document](https://dx.doi.org/10.1073/pnas.1802331115)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p2.1)\.
- Ekinci \(2005\)E\. B\. EkinciOsmanlı devleti’nde mahkemeler ve kadılık müessesesi literatürü\.Türkiye Araştırmaları Literatür Dergisi3\(5\),pp\. 417–439\.Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1),[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Fanet al\.\(2025\)Z\. Fan, V\. Thouzeau, C\. de Dampierre, C\. Chevallier, and N\. BaumardName uniqueness and the rise of individualism in the western hemisphere \(1500\-2000\)\.Current Research in Ecological and Social Psychology,pp\. 100235\.Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p1.1)\.
- Feuerriegelet al\.\(2025\)S\. Feuerriegel, A\. Maarouf, D\. Bär,et al\.Using natural language processing to analyse text data in behavioural science\.Nature Reviews Psychology4\(2\),pp\. 96–111\.External Links:[Document](https://dx.doi.org/10.1038/s44159-024-00392-z),[Link](https://doi.org/10.1038/s44159-024-00392-z)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p3.1)\.
- Gaoet al\.\(2021\)T\. Gao, X\. Yao, and D\. ChenSimCSE: simple contrastive learning of sentence embeddings\.InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing,M\. Moens, X\. Huang, L\. Specia, and S\. W\. Yih \(Eds\.\),Online and Punta Cana, Dominican Republic,pp\. 6894–6910\.External Links:[Link](https://aclanthology.org/2021.emnlp-main.552/),[Document](https://dx.doi.org/10.18653/v1/2021.emnlp-main.552)Cited by:[§1\.4\.2](https://arxiv.org/html/2611.24981#S1.SS4.SSS2.p2.1)\.
- Garget al\.\(2018\)N\. Garg, L\. Schiebinger, D\. Jurafsky, and J\. ZouWord embeddings quantify 100 years of gender and ethnic stereotypes\.115\(16\),pp\. E3635–E3644\.External Links:[Document](https://dx.doi.org/10.1073/pnas.1720347115)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p2.1)\.
- Greenfield \(2013\)P\. M\. GreenfieldThe changing psychology of culture from 1800 through 2000\.24\(9\),pp\. 1722–1731\.External Links:[Document](https://dx.doi.org/10.1177/0956797613479387)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p1.1)\.
- Gwet \(2008\)K\. L\. GwetComputing inter\-rater reliability and its variance in the presence of high agreement\.British Journal of Mathematical and Statistical Psychology61\(1\),pp\. 29–48\.Cited by:[§1\.3\.3](https://arxiv.org/html/2611.24981#S1.SS3.SSS3.p1.1)\.
- Hamiltonet al\.\(2016\)W\. L\. Hamilton, J\. Leskovec, and D\. JurafskyDiachronic word embeddings reveal statistical laws of semantic change\.InProceedings of the 54th Annual Meeting of the Association for Computational Linguistics,Vol\.1,pp\. 1489–1501\.External Links:[Document](https://dx.doi.org/10.18653/v1/P16-1141),[Link](https://aclanthology.org/P16-1141/)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p2.1)\.
- Henrichet al\.\(2010\)J\. Henrich, S\. J\. Heine, and A\. NorenzayanThe weirdest people in the world?\.Behavioral and Brain Sciences33\(2\-3\),pp\. 61–83\.External Links:[Document](https://dx.doi.org/10.1017/S0140525X0999152)Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p4.1)\.
- Ilievet al\.\(2015\)R\. Iliev, M\. Dehghani, and E\. SagiAutomated text analysis in psychology: methods, applications, and future developments\.Language and cognition7\(2\),pp\. 265–290\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p2.1)\.
- İzgöer \(2013\)A\. Z\. İzgöerDiyarbekir Şeriyye sicilleri Âmid mahkemesi \(3715, 3789, 3828 numaralı defterler\)\.Yayın No: 24,Dicle Üniversitesi İlahiyat Fakültesi Yayınları,Diyarbakır\.External Links:ISBN 978\-975\-7635\-32\-1Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1)\.
- Jennings \(1978\)R\. C\. JenningsZimmis \(non\-muslims\) in early 17th century ottoman judicial records\.Journal of the Economic and Social History of the Orient21\(1\),pp\. 225–293\.External Links:[Document](https://dx.doi.org/10.1163/156852078X00143)Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1)\.
- Kennedyet al\.\(2022\)B\. Kennedy, A\. Ashokkumar, R\. L\. Boyd, and M\. DehghaniText analysis for psychology: methods, principles, and practices\.InHandbook of Language Analysis in Psychology,M\. Dehghani and R\. L\. Boyd \(Eds\.\),pp\. 3–62\.External Links:ISBN 9781462548439Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p3.1)\.
- Kennedyet al\.\(2021\)B\. Kennedy, M\. Atari, A\. M\. Davani, J\. Hoover, A\. Omrani, J\. Graham, and M\. DehghaniMoral concerns are differentially observable in language\.Cognition212,pp\. 104696\.Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p2.1)\.
- Kesebir and Kesebir \(2012\)P\. Kesebir and S\. KesebirThe cultural salience of moral character and virtue declined in twentieth\-century america\.The Journal of Positive Psychology7\(6\),pp\. 471–480\.External Links:[Document](https://dx.doi.org/10.1080/17439760.2012.715182),[Link](https://doi.org/10.1080/17439760.2012.715182)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p1.1)\.
- Kitayama and Salvador \(2024\)S\. Kitayama and C\. E\. SalvadorCultural psychology: beyond east and west\.Annual review of psychology75\(1\),pp\. 495–526\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Kozlowskiet al\.\(2019\)A\. C\. Kozlowski, M\. Taddy, and J\. A\. EvansThe geometry of culture: analyzing the meanings of class through word embeddings\.84\(5\),pp\. 905–949\.External Links:[Document](https://dx.doi.org/10.1177/0003122419877135)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p2.1)\.
- Krippendorff \(2004\)K\. KrippendorffReliability in content analysis: some common misconceptions and recommendations\.Human communication research30\(3\),pp\. 411–433\.Cited by:[§1\.3\.3](https://arxiv.org/html/2611.24981#S1.SS3.SSS3.p1.1)\.
- Manjavacas and Fonteyn \(2022\)E\. Manjavacas and L\. FonteynAdapting vs\. pre\-training language models for historical languages\.Journal of Data Mining & Digital Humanities\(Digital humanities in languages\)\.Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p3.1)\.
- Markus and Kitayama \(1991\)H\. R\. Markus and S\. KitayamaCulture and the self: implications for cognition, emotion, and motivation\.\.Psychological Review98\(2\),pp\. 224\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Martins and Baumard \(2020\)M\. d\. J\. D\. Martins and N\. BaumardThe rise of prosociality in fiction preceded democratic revolutions in early modern europe\.Proceedings of the National Academy of Sciences117\(46\),pp\. 28684–28691\.External Links:[Document](https://dx.doi.org/10.1073/pnas.2009571117)Cited by:[§3\.2](https://arxiv.org/html/2611.24981#S3.SS2.p1.1)\.
- Martins and Baumard \(2022\)M\. d\. J\. D\. Martins and N\. BaumardHow to develop reliable instruments to measure the cultural evolution of preferences and feelings in history?\.13\.External Links:[Document](https://dx.doi.org/10.3389/fpsyg.2022.1034058)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p3.1)\.
- Masui and Miyamoto \(2025\)H\. Masui and Y\. MiyamotoEmotions in japanese song lyrics over 50 years: trajectory over time and the impact of economic hardship and disasters\.Current Research in Ecological and Social Psychology8,pp\. 100218\.Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p1.1)\.
- McLoughlinet al\.\(2026\)K\. L\. McLoughlin, A\. Palmer, J\. N\. Matias, and M\. CrockettProprietary llms impede scientific transparency and reproducibility\.Nature Reviews Psychology,pp\. 1–2\.Cited by:[§2\.5](https://arxiv.org/html/2611.24981#S2.SS5.p2.1)\.
- Medinet al\.\(2017\)D\. Medin, B\. Ojalehto, A\. Marin, and M\. BangSystems of \(non\-\) diversity\.Nature human behaviour1\(5\),pp\. 0088\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p4.1)\.
- Michelet al\.\(2011\)J\. Michel, Y\. K\. Shen, A\. Presser Aiden, A\. Veres, M\. K\. Gray, Google Books Team, J\. P\. Pickett, D\. Hoiberg, D\. Clancy, P\. Norvig, J\. Orwant, S\. Pinker, M\. A\. Nowak, and E\. Lieberman AidenQuantitative analysis of culture using millions of digitized books\.331\(6014\),pp\. 176–182\.External Links:[Document](https://dx.doi.org/10.1126/science.1199644)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p1.1)\.
- Moretti \(2013\)F\. MorettiDistant reading\.Verso\.Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p1.1)\.
- OpenAI \(2026\)OpenAIChatGPT\.Note:Large language modelExternal Links:[Link](https://chatgpt.com/)Cited by:[§1\.10](https://arxiv.org/html/2611.24981#S1.SS10.p1.1)\.
- Oysermanet al\.\(2002\)D\. Oyserman, H\. M\. Coon, and M\. KemmelmeierRethinking individualism and collectivism: evaluation of theoretical assumptions and meta\-analyses\.\.Psychological bulletin128\(1\),pp\. 3\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Oyserman \(1993\)D\. OysermanThe lens of personhood: viewing the self and others in a multicultural society\.Journal of Personality and Social Psychology65\(5\),pp\. 993–1009\.External Links:[Document](https://dx.doi.org/10.1037/0022-3514.65.5.993)Cited by:[§1\.2\.1](https://arxiv.org/html/2611.24981#S1.SS2.SSS1.Px1.p1.1),[§1\.7\.2](https://arxiv.org/html/2611.24981#S1.SS7.SSS2.p1.1)\.
- Oyserman \(2017\)D\. OysermanCulture three ways: culture and subcultures within countries\.Annual review of psychology68,pp\. 435–463\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Parket al\.\(2015\)G\. Park, H\. A\. Schwartz, J\. C\. Eichstaedt, M\. L\. Kern, M\. Kosinski, D\. J\. Stillwell, L\. H\. Ungar, and M\. E\. P\. SeligmanAutomatic personality assessment through social media language\.Journal of Personality and Social Psychology108\(6\),pp\. 934–952\.External Links:[Document](https://dx.doi.org/10.1037/pspp0000020)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p2.1)\.
- Peirce \(2003\)L\. PeirceMorality tales: law and gender in the ottoman court of aintab\.University of California Press\.External Links:[Document](https://dx.doi.org/10.1525/california/9780520228900.001.0001),ISBN 9780520228900Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1)\.
- Peirce \(2014\)L\. PeirceHonor, reputation, and reciprocity\.European Journal of Turkish Studies\(18\)\.External Links:[Document](https://dx.doi.org/10.4000/ejts.4850)Cited by:[§2\.4\.3](https://arxiv.org/html/2611.24981#S2.SS4.SSS3.p1.1),[§3](https://arxiv.org/html/2611.24981#S3.p5.1)\.
- Pennebakeret al\.\(2003\)J\. W\. Pennebaker, M\. R\. Mehl, and K\. G\. NiederhofferPsychological aspects of natural language use: our words, our selves\.Annual review of psychology54\(1\),pp\. 547–577\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p2.1)\.
- Pettit \(2016\)M\. PettitHistorical time in the age of big data: cultural psychology, historical change, and the Google Books Ngram Viewer\.19\(2\),pp\. 141–153\.External Links:[Document](https://dx.doi.org/10.1037/hop0000023)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p3.1)\.
- Rabaia \(2015\)I\. RabaiaKudüs Şer’iyye sicilleri, sicil no\. 167\.IRCICA,İstanbul\.External Links:ISBN 9789290632894Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1)\.
- Radet al\.\(2018\)M\. S\. Rad, A\. J\. Martingano, and J\. GingesToward a psychology of homo sapiens: making psychological science more representative of the human population\.Proceedings of the National Academy of Sciences115\(45\),pp\. 11401–11405\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p4.1)\.
- Reimers and Gurevych \(2019\)N\. Reimers and I\. GurevychSentence\-BERT: sentence embeddings using siamese BERT\-networks\.InProceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing \(EMNLP\-IJCNLP\),Hong Kong, China,pp\. 3982–3992\.External Links:[Link](https://aclanthology.org/D19-1410),[Document](https://dx.doi.org/10.18653/v1/D19-1410)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p4.1)\.
- Rodriguez Mosqueraet al\.\(2002\)P\. M\. Rodriguez Mosquera, A\. S\. R\. Manstead, and A\. H\. FischerThe role of honour concerns in emotional reactions to offences\.Cognition and Emotion16\(1\),pp\. 143–163\.External Links:[Document](https://dx.doi.org/10.1080/0269993014300016),[Link](https://doi.org/10.1080/0269993014300016)Cited by:[§1\.2\.1](https://arxiv.org/html/2611.24981#S1.SS2.SSS1.Px2.p1.1),[§1\.7\.2](https://arxiv.org/html/2611.24981#S1.SS7.SSS2.p1.1),[§3](https://arxiv.org/html/2611.24981#S3.p5.1)\.
- Schulzet al\.\(2019\)J\. F\. Schulz, D\. Bahrami\-Rad, J\. P\. Beauchamp, and J\. HenrichThe church, intensive kinship, and global psychological variation\.Science366\(6466\),pp\. eaau5141\.Cited by:[§3\.2](https://arxiv.org/html/2611.24981#S3.SS2.p1.1)\.
- Schweter \(2020\)BERTurk \- bert models for turkishExternal Links:[Document](https://dx.doi.org/10.5281/zenodo.3770924),[Link](https://doi.org/10.5281/zenodo.3770924)Cited by:[§1\.4\.2](https://arxiv.org/html/2611.24981#S1.SS4.SSS2.p2.1)\.
- Talhelmet al\.\(2026\)T\. Talhelm, L\. Wei, R\. Sun, D\. Medvedev, Á\. San Martin, M\. Helmy, A\. Samekin, A\. Zaragoza Scherman, and A\. S\. EnglishDefining collectivism as duties in relationships rather than generalized warmth improves predictive validity in 100 cultures\.Nature Human Behaviour,pp\. 1–13\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p4.1)\.
- Teitelbaum and Simchon \(2025\)L\. Teitelbaum and A\. SimchonNeural text embeddings in psychological research: a guide with examples in r\.\.Psychological Methods\.Cited by:[§1\.2\.1](https://arxiv.org/html/2611.24981#S1.SS2.SSS1.p1.1),[§1\.2\.2](https://arxiv.org/html/2611.24981#S1.SS2.SSS2.p1.1)\.
- Triandiset al\.\(1988\)H\. C\. Triandis, R\. Bontempo, M\. J\. Villareal, M\. Asai, and N\. LuccaIndividualism and collectivism: cross\-cultural perspectives on self\-ingroup relationships\.\.Journal of personality and Social Psychology54\(2\),pp\. 323\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Tuğ Onaran \(2022\)B\. Tuğ Onaran1858 ceza kanunname\-i hümayunu’nda irz ve namus kavramlarının kodifikasyonu ve toplumsal cinsiyet\.Akdeniz Kadın Çalışmaları ve Toplumsal Cinsiyet Dergisi5\(2\),pp\. 209–239\.Cited by:[§3](https://arxiv.org/html/2611.24981#S3.p5.1)\.
- Türkiye Diyanet Vakfı İslâm Araştırmaları Merkezi \(2023\)Türkiye Diyanet Vakfı İslâm Araştırmaları MerkeziİStanbul kadı sicilleri\(Website\)İSAM\.External Links:[Link](https://www.kadisicilleri.org/)Cited by:[§1\.1\.1](https://arxiv.org/html/2611.24981#S1.SS1.SSS1.p1.1),[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Twengeet al\.\(2012\)J\. M\. Twenge, W\. K\. Campbell, and B\. GentileIncreases in individualistic words and phrases in american books, 1960–2008\.PLOS ONE7\(7\),pp\. e40181\.External Links:[Document](https://dx.doi.org/10.1371/journal.pone.0040181),[Link](https://doi.org/10.1371/journal.pone.0040181)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p1.1)\.
- Uğur \(2026\)Y\. UğurŞer’iyye sicilleri\.InTDV İslâm Ansiklopedisi,External Links:[Link](https://islamansiklopedisi.org.tr/seriyye-sicilleri)Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p1.1)\.
- Uskul and Cross \(2019\)A\. K\. Uskul and S\. E\. CrossThe social and cultural psychology of honour: what have we learned from researching honour in turkey?\.European Review of Social Psychology30\(1\),pp\. 39–73\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Uskulet al\.\(2023\)A\. K\. Uskul, A\. Kirchner\-Häusler, V\. L\. Vignoles, R\. Rodriguez\-Bailón, V\. A\. Castillo, S\. E\. Cross, M\. G\. Yalçın, C\. Harb, S\. Husnu, K\. Ishii,et al\.Neither eastern nor western: patterns of independence and interdependence in mediterranean societies\.\.Journal of Personality and Social Psychology125\(3\),pp\. 471\.Cited by:[Quantifying Cultural\-Psychological Constructs in Four Centuries of Ottoman Court Registers](https://arxiv.org/html/2611.24981#p3.1)\.
- Wagneret al\.\(2024\)V\. Wagner, Z\. Y\. Gökçe, Z\. D\. Abacı, and T\. AllahverdiyevaTranscending boundaries: rethinking slave agency in the ottoman empire through religious conversion practices\.Working PaperTechnical Report14,Bonn Center for Dependency and Slavery Studies \(BCDSS\)\.External Links:[Link](https://hdl.handle.net/20.500.11811/11527),[Document](https://dx.doi.org/10.48565/bonndoc-279)Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1)\.
- Wevers and Koolen \(2020\)M\. Wevers and M\. KoolenDigital begriffsgeschichte: tracing semantic change using word embeddings\.53\(4\),pp\. 226–243\.External Links:[Document](https://dx.doi.org/10.1080/01615440.2020.1760157)Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p2.1)\.
- Wheeleret al\.\(2019\)M\. A\. Wheeler, M\. J\. McGrath, and N\. HaslamTwentieth century morality: the rise and fall of moral concepts from 1900 to 2007\.PLoS one14\(2\),pp\. e0212267\.Cited by:[§0\.1](https://arxiv.org/html/2611.24981#S0.SS1.p1.1)\.
- Yaşa \(2025\)F\. YaşaC\. Kafadar and G\. A\. Tekin \(Eds\.\)Tracing the path of justice: the crimean qadi court and its records, manuscript division of the russian national library, fond 917, crimean court records, vol\. 22 \(1087\-1089/1677\-1678\)\.Sources of Oriental Languages and Literatures,Harvard University Press,Cambridge, MA\.Note:Analysis, Text, and FacsimileCited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1)\.
- Yüksek \(2024\)Sufism\-and\-Urban\-ReligiosityNote:[https://github\.com/yuksekyusuf/Sufism\-and\-Urban\-Religiosity](https://github.com/yuksekyusuf/Sufism-and-Urban-Religiosity)Accessed: February 19, 2026Cited by:[§1\.1\.1](https://arxiv.org/html/2611.24981#S1.SS1.SSS1.p1.1)\.
- Yılandıloğlu \(2025\)E\. YılandıloğluXLM\-roberta base for ottoman turkish court records\.Hugging Face\.Note: urlhttps://huggingface\.co/enesyila/ota\-roberta\-base\-kadisicilleriCited by:[§1\.4\.2](https://arxiv.org/html/2611.24981#S1.SS4.SSS2.p2.1)\.
- C\. Yılmaz \(Ed\.\) \(2010a\)C\. Yılmaz \(Ed\.\)İStanbul kadı sicilleri 10: Üsküdar mahkemesi 84 numaralı sicil \(h\. 999–1000 / m\. 1590–1591\)\.Türkiye Diyanet Vakfı İslâm Araştırmaları Merkezi\.Cited by:[§2\.4\.3](https://arxiv.org/html/2611.24981#S2.SS4.SSS3.p1.1)\.
- C\. Yılmaz \(Ed\.\) \(2010b\)C\. Yılmaz \(Ed\.\)İStanbul kadı sicilleri 35: galata mahkemesi 20 numaralı sicil \(h\. 1005–1007 / m\. 1596–1599\)\.Türkiye Diyanet Vakfı İslâm Araştırmaları Merkezi\.Cited by:[§2\.4\.1](https://arxiv.org/html/2611.24981#S2.SS4.SSS1.p1.1)\.
- C\. Yılmaz \(Ed\.\) \(2010c\)C\. Yılmaz \(Ed\.\)İstanbul kadı sicilleri Üsküdar mahkemesi 2 numaralı sicil \(h\. 924\-927 / m\. 1518\-1521\)\.İSAM Yayınları, Kadı Sicilleri Dizisi: 2,Türkiye Diyanet Vakfı İslâm Araştırmaları Merkezi \(İSAM\),İstanbul\.External Links:ISBN 978\-605\-5586\-03\-4Cited by:[§0\.2](https://arxiv.org/html/2611.24981#S0.SS2.p2.1)\.
- C\. Yılmaz \(Ed\.\) \(2011\)C\. Yılmaz \(Ed\.\)İStanbul kadı sicilleri 32: galata mahkemesi 5 numaralı sicil \(h\. 983–984 / m\. 1575–1576\)\.Türkiye Diyanet Vakfı İslâm Araştırmaları Merkezi\.Cited by:[§2\.4\.1](https://arxiv.org/html/2611.24981#S2.SS4.SSS1.p1.1)\.
- C\. Yılmaz \(Ed\.\) \(2019a\)C\. Yılmaz \(Ed\.\)İStanbul kadı sicilleri 46: rumeli sadâreti mahkemesi 40 numaralı sicil \(h\. 1033–1034 / m\. 1623–1624\)\.İstanbul Büyükşehir Belediyesi Kültür A\.Ş\.\.Cited by:[§2\.4\.2](https://arxiv.org/html/2611.24981#S2.SS4.SSS2.p1.1)\.
- C\. Yılmaz \(Ed\.\) \(2019b\)C\. Yılmaz \(Ed\.\)İStanbul kadı sicilleri 53: bab mahkemesi 11 numaralı sicil \(h\. 1081 / m\. 1670–1671\)\.İstanbul Büyükşehir Belediyesi Kültür A\.Ş\.\.Cited by:[§2\.4\.3](https://arxiv.org/html/2611.24981#S2.SS4.SSS3.p1.1)\.
- Zewailet al\.\(2026\)A\. Zewail, A\. Setia, R\. Mohammadsadegh, F\. Şeker, A\. Hajian, H\. J\. Sosa, S\. R\. Gavireddy, L\. Morhayim, and M\. AtariThe cultural and moral expressions in language \(CAMEL\) corpus\.Behavior Research Methods\.Cited by:[§1\.3\.3](https://arxiv.org/html/2611.24981#S1.SS3.SSS3.p1.1)\.

## Appendix AModels Evaluated

Table 6:Embedding Models Evaluated
## Appendix BExamples of Construct Representations

Table 7:Illustrative Examples of Construct RepresentationsNote\.The examples illustrate the structure of the construct representations\. Document\-level scoring used the complete sets of questionnaire items, aligned proxies, opposed proxies, and expert\-definition sentences\. Complete materials are provided in the Supplementary Materials\.

## Appendix CRecords Discussed in the Qualitative Examination

Table 8:Highest\-Scoring Individualism Records Examined QualitativelyNote\.SD above mean indicates the number of standard deviations by which a record’s individualism score exceeded the corpus\-wide mean\. All 10 records concern manumission\. The published texts are subject to copyright restrictions\.

Table 9:Highest\-Scoring Collectivism Records Examined QualitativelyNote\.SD above mean indicates the number of standard deviations by which a record’s collectivism score exceeded the corpus\-wide mean\. All 10 records concern religious endowments, including endowments of money, land, residences, and other property\. The published texts are subject to copyright restrictions\.

Table 10:Highest\-Scoring Honor Records Examined QualitativelyNote\.SD above mean indicates the number of standard deviations by which a record’s honor score exceeded the corpus\-wide mean\. Records are ordered from highest to lowest honor score\. The published texts are subject to copyright restrictions\.

相似文章