BLAD: A Historically Contextualized, Multilingual Dataset of Bangladeshi Legal Acts (1799 to 2025)

# BLAD: A Historically Contextualized, Multilingual Dataset of Bangladeshi Legal Acts (1799–2025)
Source: [https://arxiv.org/html/2607.17111](https://arxiv.org/html/2607.17111)
Adib Sakhawat Systems and Software Lab \(SSL\) Department of Computer Science and Engineering Islamic University of Technology, Dhaka, Bangladesh sakhadib@gmail\.com

###### Abstract

We present the Bangladesh Legal Acts Dataset \(BLAD\), a curated collection of 1,484 legislative acts enacted between 1799 and 2025\. Each act is represented with its full text, structured sections and footnotes, repeal status, and metadata linking it to the governing regime, head of state, and prevailing legal framework at the time of enactment\. The corpus spans English, Bengali, and mixed\-language documents, supporting temporal and multilingual analysis of statutory law\. BLAD addresses a persistent gap in legal natural language processing \(NLP\) resources for low\-resource, civil\-law jurisdictions in South Asia\. We describe the acquisition and enrichment pipeline, report descriptive statistics over more than two centuries of legislation, and outline the research directions the corpus enables\. The dataset is publicly available under the CC BY\-SA 4\.0 license at[https://www\.kaggle\.com/datasets/sakhadib/bangladesh\-legal\-acts\-dataset](https://www.kaggle.com/datasets/sakhadib/bangladesh-legal-acts-dataset)\.

## 1Introduction

Legal natural language processing \(NLP\) has advanced rapidly in recent years, driven by the digitisation of legal corpora and by progress in transformer\-based architectures\. This growth, however, has been distributed unevenly across jurisdictions: large, well\-curated legal datasets overwhelmingly concern English\-language common\-law systems, while civil\-law frameworks and non\-English legal traditions remain comparatively underservedNiklauset al\.\([2023](https://arxiv.org/html/2607.17111#bib.bib1)\)\. The imbalance constrains the development of inclusive legal AI systems and limits cross\-jurisdictional legal research\.

Bangladesh presents an instructive case of this gap\. Its population of roughly 180 million is governed by a hybrid legal system that has accumulated across more than two centuries of colonial, post\-partition, and post\-independence rule\. Despite recent efforts to build legal datasets for neighbouring South Asian jurisdictionsK and Mishra \([2025](https://arxiv.org/html/2607.17111#bib.bib4)\), the Bangladeshi statutory corpus has remained largely inaccessible to computational researchers, owing to inconsistent digitisation, mixed\-language documentation, and the absence of structured, machine\-readable formats\.

Existing Bangladeshi legal NLP resources share three limitations\. First, they are narrow in scope: the largest English\-only corpus comprises 595 acts and 18,023 sectionsWasiet al\.\([2024](https://arxiv.org/html/2607.17111#bib.bib13)\), less than 40% of the complete statutory framework\. Second, they lack historical contextualisation, offering no systematic link between an act and the governmental period in which it was enacted, which precludes temporal legal analysis\. Third, their focus on English\-only content overlooks the bilingual character of Bangladeshi law, in which Bengali and English coexist in patterns that reflect colonial transitions and contemporary language policy\.

![Refer to caption](https://arxiv.org/html/2607.17111v1/fig/a.png)Figure 1:Number of acts per decade across the full temporal span of the corpus\.These limitations carry practical consequences\. Bangladesh’s judiciary faces a backlog of several million pending cases, which creates a concrete demand for automated legal information systems\. The absence of comprehensive, structured statutory corpora impedes the development of retrieval\-augmented generation systems, legal summarisation tools, and related applications that could broaden access to legal information\.

This paper introduces theBangladesh Legal Acts Dataset\(BLAD\), a corpus that addresses these gaps through three contributions:

Scale and coverage\.BLAD comprises 1,484 legislative acts spanning 226 years \(1799–2025\), with 35,633 sections and 14,523 footnotes, together with section\-level language identification\. This is an approximately 2\.5×\\timesincrease over the largest previously available dataset and provides substantially complete coverage of the statutory framework\.

Historical contextualisation\.Each act is enriched with metadata describing the governing regime at the time of enactment, spanning fourteen distinct government systems and forty\-two heads of state, from Company Rule \(1799–1858\) to the contemporary period\. This layer enables research on temporal legal evolution and regime\-specific legislative behaviour\.

Multilingual and structural annotation\.Every act is segmented into sections and footnotes and annotated with a detected language label \(English, Bengali, or mixed\) and a token count, providing structured supervision for multilingual and document\-structure tasks\.

## 2Related Work

Legal NLP has grown substantially over the past decade, driven by the digitisation of legal systems and by advances in transformer\-based architecturesNiklauset al\.\([2023](https://arxiv.org/html/2607.17111#bib.bib1)\)\. Recent surveys document a shift from early rule\-based approaches to large language models tailored for legal textAriai and Demartini \([2024](https://arxiv.org/html/2607.17111#bib.bib2)\)\. The distinctive properties of legal text—long documents, complex syntax, domain\-specific terminology, and a stringent requirement for accuracy and reliability—have motivated specialised methods that diverge from general\-domain NLPAriai and Demartini \([2024](https://arxiv.org/html/2607.17111#bib.bib2)\)\.

India has led regional efforts through comprehensive benchmarks\. IL\-TUR \(Indian Legal Text Understanding and Reasoning\)Joshiet al\.\([2024](https://arxiv.org/html/2607.17111#bib.bib3)\)defines eight tasks covering English and nine Indian languages\. IndicLegalQAK and Mishra \([2025](https://arxiv.org/html/2607.17111#bib.bib4)\)contributes 10,000 question–answer pairs drawn from 1,256 Indian Supreme Court judgments, addressing the need for legal information retrieval in a context where a backlog exceeding 43 million cases motivates automated processingK and Mishra \([2025](https://arxiv.org/html/2607.17111#bib.bib4)\)\.

By contrast, the South Asian legal NLP landscape has remained largely incomplete with respect to Bangladesh\. Earlier Bangladesh\-specific work was confined to preliminary efforts such as legal\-text visualisation systemsMandalet al\.\([2017](https://arxiv.org/html/2607.17111#bib.bib15)\)and narrow domain\-specific applications, rather than comprehensive coverage of the legislative framework\.

East Asian jurisdictions, particularly China and South Korea, have established leadership in large\-scale legal dataset construction and systematic benchmarkingXiaoet al\.\([2018](https://arxiv.org/html/2607.17111#bib.bib5)\)\. The Chinese AI and Law Challenge \(CAIL\) dataset, introduced in 2018, comprised 2\.6 million criminal cases from the Supreme People’s Court of China\. Subsequent Chinese benchmarks such as LexEvalLiet al\.\([2024](https://arxiv.org/html/2607.17111#bib.bib6)\)span 23 tasks and 14,150 questions across diverse legal\-reasoning abilities\. South Korea has contributed LBOX OPENHwanget al\.\([2022](https://arxiv.org/html/2607.17111#bib.bib7)\), comprising 147,000 legal precedents across tasks including judgment prediction, case classification, and summarisation; the Korean Benchmark for Legal Language understanding \(KBL\)Kimet al\.\([2024](https://arxiv.org/html/2607.17111#bib.bib8)\)extends this line with pragmatic evaluation designed for large language models\.

Standardised evaluation frameworks have been central to the field’s reproducibility\. LexGLUEChalkidiset al\.\([2022](https://arxiv.org/html/2607.17111#bib.bib10)\)established English\-language evaluation standards across seven legal tasks\. Building on it, LEXTREMENiklauset al\.\([2023](https://arxiv.org/html/2607.17111#bib.bib1)\)targets multilingual evaluation, incorporating 11 datasets across 24 languages and 17 jurisdictions\. LegalBenchGuhaet al\.\([2023](https://arxiv.org/html/2607.17111#bib.bib16)\)advances collaborative benchmark construction through 162 tasks developed jointly by legal professionals and NLP researchers\.

Recent dataset construction has emphasised systematic collection, annotation, and quality assurance, reflecting a maturation from ad hoc web scraping to structured corpus curation\. MultiLegalSBDBruggeret al\.\([2023](https://arxiv.org/html/2607.17111#bib.bib11)\)illustrates the difficulty of multilingual sentence\-boundary detection in legal preprocessing, and specialised resources demonstrate the value of expert annotation for regulatory applicationsJoshiet al\.\([2024](https://arxiv.org/html/2607.17111#bib.bib3)\)\.

Several gaps persist despite this progress\. First, most large\-scale legal datasets concern common\-law jurisdictions, particularly those using English as the primary legal languageNiklauset al\.\([2023](https://arxiv.org/html/2607.17111#bib.bib1)\)\. Second, existing multilingual datasets often lack the historical depth and systematic contextualisation required for the study of legal evolution, concentrating instead on contemporary documentsNiklauset al\.\([2024](https://arxiv.org/html/2607.17111#bib.bib12)\)\. Third, the prevalence of case\-law and judicial\-decision datasets has left statutes, regulations, and legislative documents comparatively underrepresented, despite their foundational roleNiklauset al\.\([2023](https://arxiv.org/html/2607.17111#bib.bib1)\)\.

These gaps motivate the present work\. Integrating historical government context with statutory text addresses the shortfall in temporal analysis capabilities, and constructing a comprehensive civil\-law resource addresses the underrepresentation of such systems in current multilingual benchmarks\.

## 3Dataset Description

BLAD is a comprehensive collection of legislative acts from Bangladesh spanning 1799 to 2025\. It was constructed to address the fragmented and lightly processed state of Bangladeshi legal data, which has impeded large\-scale legal NLP\. The corpus provides a structured, machine\-readable representation of the known statutory framework, enriched with contextual information about the governments and legal systems under which each act was enacted\.

### 3\.1Overview

The corpus contains 1,484 legislative acts, 35,633 sections, and 14,523 footnotes\. Acts appear in English, Bengali, and mixed\-language forms, making the corpus suitable for multilingual legal NLP\. Each act is stored in a standardised JSON record carrying its title, year, language, token count, and links to historical government and legal\-system context\.

### 3\.2Data Sources and Processing

The corpus was compiled from the official Bangladesh Laws Portal maintained by the Ministry of Law, Justice and Parliamentary Affairs\. The construction pipeline \(Figure[2](https://arxiv.org/html/2607.17111#S3.F2)\) proceeds through six stages, each addressing a specific source of noise or a specific enrichment objective\.

![Refer to caption](https://arxiv.org/html/2607.17111v1/x1.png)Figure 2:The six\-stage acquisition and enrichment pipeline used to construct BLAD\.#### Stage 1: Acquisition\.

An index crawler enumerated the act listing and, for each act, recorded its canonical source URL, act number, title, and year of enactment\. Requests were rate\-limited with a fixed delay between successive calls to avoid placing undue load on the portal, and failures were logged for retry so that individual errors did not interrupt the crawl\.

#### Stage 2: Structural parsing\.

Each act page was parsed into its constituent sections and footnotes\. Because the portal’s markup is inconsistent across acts of different eras, parsing relied on the recurring structural cues that delimit a section body and its associated footnotes, rather than on a single fixed template\. Sections and footnotes were preserved as ordered lists so that document structure is recoverable downstream\.

#### Stage 3: Normalisation and cleaning\.

Extracted text was stripped of residual markup and boilerplate, and inconsistent fields were standardised to a common schema\. Bengali numerals appearing in years and identifiers were mapped to their Western Arabic equivalents to make temporal fields sortable, and malformed or duplicated fragments were repaired or removed\.

#### Stage 4: Language identification\.

Each section was assigned a language label\. Because Bengali and Latin scripts occupy disjoint Unicode ranges, a script\-ratio heuristic distinguishes predominantly Bengali from predominantly English text, and labels a section as*mixed*when both scripts occur above a threshold\. Token counts were computed for each act using a regular\-expression tokeniser applied to the cleaned text\.

#### Stage 5: Historical and legal\-system contextualisation\.

Each act was linked to the political period in which it was enacted by matching its year against a curated lookup table of governmental periods—for example Company Rule, British Colonial Rule, the Pakistan period, and post\-independence democratic and military regimes—and to the corresponding government and head of state\. A parallel mapping records the broader legal\-system context of the period, including the governing legal framework, court structure, policing arrangements, and land\-relations regime\. This contextual layer is derived from secondary historical sources rather than from the acts themselves, and is therefore treated as an interpretive annotation rather than as primary data\.

#### Stage 6: Validation and recovery\.

Combined records were validated for structural consistency, and an automated recovery step re\-derived context for acts whose metadata was initially missing or corrupted\. Processing statistics, source URLs, and timestamps were retained with each record to support auditing and reproducibility\.

### 3\.3Dataset Features

Each act carries the following fields:

- •Act title:the title of the legislative act\.
- •Act year:the year of enactment\.
- •Language:the language\(s\) used in the act \(English, Bengali, or mixed\)\.
- •Token count:the number of tokens in the act, computed via regular\-expression tokenisation\.
- •Sections:an ordered list of sections with their content\.
- •Footnotes:an ordered list of footnotes, including amendment history where present\.
- •Repeal status:whether the act is currently repealed\.
- •Government context:the government in power at enactment, including its name, head of government, period of rule, and mode of accession\.
- •Legal\-system context:the legal framework, period, and policing arrangements in force at enactment\.
- •Processing information:the tokenisation method, source URL, and metadata\-recovery status\.

The corpus is intended as a foundational resource for research on Bangladeshi legal history, multilingual legal NLP, and legal\-technology applications in the Global South\.

## 4Dataset Statistics and Analysis

BLAD supports descriptive analysis of more than two centuries of legislative activity\. This section reports statistics on the volume of acts, legislative trends over time, the distribution of activity across government systems and leaders, and the structural complexity of individual acts\. The trends reported here are descriptive; we make no causal claims, and readers should bear in mind that archival coverage may be uneven across periods\.

### 4\.1Corpus Overview

The corpus comprises the following:

- •Total acts:1,484
- •Total sections:35,633
- •Total footnotes:14,523
- •Languages:English, Bengali, mixed
- •Time span:1799–2025
- •Government systems:14
- •Heads of state:42
- •Token count:over 2\.5 million tokens

These figures span approximately 226 years of legal history across several distinct political eras\.

### 4\.2Legislative Activity by Decade

Legislative output has increased markedly over time, with pronounced changes around major political transitions \(Figure[3](https://arxiv.org/html/2607.17111#S4.F3)\)\.

![Refer to caption](https://arxiv.org/html/2607.17111v1/x2.png)Figure 3:Distribution of legal acts by historical period\.- •1790s–1850s \(Company Rule\):fewer than ten acts per decade, establishing the foundational colonial legal framework\.
- •1860s–1950s \(British Colonial Rule\):an average of 11–82 acts per decade, indicating steady growth\.
- •1950s–1970s \(Pakistan period\):a transitional phase accounting for 72 acts \(4\.9% of the total\)\.
- •1970s \(post\-independence\):a sharp rise in activity, with 237 acts enacted\.
- •2010s:the most productive decade, with 256 acts\.
- •2020s:sustained high activity, with 205 acts to date\.

The acceleration of legislation from the 1970s onward is consistent with the growing administrative complexity of the post\-independence state\.

### 4\.3Government Systems and Legislative Productivity

Legislative output varies substantially across government systems \(Figure[4](https://arxiv.org/html/2607.17111#S4.F4)\)\.

![Refer to caption](https://arxiv.org/html/2607.17111v1/x3.png)Figure 4:Distribution of legal acts by government system\.- •Parliamentary democracy \(increasingly authoritarian\):430 acts, followed by regular parliamentary democracy with 327 acts; together these account for over half of all acts\.
- •Military rule:261 acts, concentrated on administrative reorganisation and policy implementation\.
- •British colonial rule:185 acts, primarily establishing colonial administrative frameworks\.

Democratic periods are associated with higher legislative volume than authoritarian ones in this corpus; we report this as an observed association rather than a causal effect\.

### 4\.4Leadership and Legislative Productivity

The corpus also permits analysis of output by individual leaders \(Figure[5](https://arxiv.org/html/2607.17111#S4.F5)\)\.

![Refer to caption](https://arxiv.org/html/2607.17111v1/x4.png)Figure 5:Top five heads of state by legislative activity\.- •Sheikh Hasina:the most prolific, with 477 acts \(32\.1% of the total\), reflecting extended tenure across multiple terms\.
- •Khaleda Zia:the second most prolific, with 160 acts\.
- •Sheikh Mujibur Rahman:120 acts, with the highest annual rate at roughly 40 acts per year during a brief tenure \(1972–1975\)\.
- •Military leaders\(e\.g\., Lt\. Gen\. Hussain Muhammad Ershad, General Ayub Khan\): substantial output concentrated on administrative and developmental policy\.

These figures illustrate the influence of individual leadership on the composition of the statutory corpus\.

### 4\.5Legal Complexity and Scope

Acts vary considerably in structural complexity, measured by section count:

- •The Code of Criminal Procedure \(1898\):the most extensive act, with 586 sections\.
- •The Penal Code \(1860\):583 sections\.
- •The Bangladesh Merchant Shipping Ordinance \(1983\):516 sections, indicating comparable complexity in modern legislation\.
- •Recent actssuch as the Companies Act \(1994\) and the Bangladesh Labour Act \(2006\) retain substantial complexity, with 418 and 355 sections respectively\.

Structural complexity is therefore not confined to any single period; substantial acts appear across the full temporal span\.

### 4\.6Active and Repealed Acts

Legal systems evolve in part by repealing acts that are no longer required, keeping the statutory framework current \(Figure[6](https://arxiv.org/html/2607.17111#S4.F6)\)\.

![Refer to caption](https://arxiv.org/html/2607.17111v1/x5.png)Figure 6:Active versus repealed acts\.Of the acts in the corpus, 1,280 are currently active and 203 have been repealed\.

### 4\.7Repeals and Legal Evolution

Repeal patterns offer a view of how legislation is revisited across administrations:

- •Lt\. Gen\. Hussain Muhammad Ershad:114 acts enacted during his tenure, of which 62 were subsequently repealed\.
- •Ziaur Rahman:70 acts enacted, of which 31 were subsequently repealed\.
- •Other leaders:additional acts were repealed after the enacting administration left office, consistent with ongoing legal revision\.

These patterns are consistent with the routine revision of statutory law as political leadership changes, and provide a basis for further study of legal evolution\.

## 5Use Cases

BLAD can serve as a foundation for research in multilingual NLP, legal history, and governance\. Rather than prescribing a fixed agenda, we outline several directions to indicate the corpus’s potential and to invite further work\. In each case we note the validation that responsible use would require\.

### 5\.1Legal Text Classification

With 1,484 acts spanning multiple regimes, the corpus supports experiments in automatic categorisation by subject, legal system, or period\. Careful validation remains necessary, but such models could assist document retrieval and comparative analysis\.

### 5\.2Temporal Analysis

Because the corpus spans more than two centuries, it enables exploratory study of how legal content and structure change over time—for instance the emergence of new concepts or shifts in document length—subject to caveats about archival gaps and historical context\.

### 5\.3Retrieval\-Augmented Generation \(RAG\)

The corpus can support prototype question\-answering systems that retrieve relevant passages before generating a response\. Initial work could target fact\-based queries \(e\.g\., reforms enacted during the 1970s\), with outputs evaluated against expert judgement\.

### 5\.4Historical Governance Studies

Linking acts to political timelines allows analysis of how successive administrations structured the law\. Any causal claim would require a robust methodology and additional socio\-political data\.

### 5\.5Multilingual NLP

Because statutes appear in Bengali, English, and mixed script, the corpus offers test material for tokenisation, named\-entity recognition, and domain\-specific machine translation, complementing larger multilingual resources\.

### 5\.6Fine\-tuning Legal Language Models

Fine\-tuning existing language models on the corpus may improve their handling of Bangladeshi legal prose, with initial evaluation on tasks such as clause summarisation or terminology explanation\.

### 5\.7Policy Impact Exploration

The corpus can support descriptive studies relating legislation to socio\-economic indicators\. Predictive work is feasible but would benefit from interdisciplinary collaboration and transparent error analysis\.

## Limitations

While BLAD is a historically and structurally rich resource, several limitations affect its completeness and downstream applicability\.

First, its legal scope is restricted to legislative acts and excludes case law, judicial interpretations, executive orders, and policy documents\. This restricts tasks such as judgment prediction and case matching, which require multi\-source legal corpora\.

Second, the multilingual nature of Bangladeshi legislation means the corpus contains Bengali, English, and mixed\-language documents\. Although language identification has been applied, residual errors are likely, particularly for short or code\-switched sections; transformer\-based multilingual tokenisation \(e\.g\., XLM\-R\) is a candidate for future improvement\.

Third, the contextual metadata—political regime, head of state, and government structure—is inferred from secondary sources and historical mapping rather than encoded in the acts themselves\. Such mappings may contain temporal ambiguities, especially for transitional regimes, and would benefit from expert\-validated annotation in future releases\.

Fourth, quality assurance combined rule\-based validation with manual spot\-checks, but comprehensive human verification remains pending\. Statistical audits confirmed structural consistency, yet the correctness of named\-entity mentions, token counts, and annotations would be strengthened by community review\.

Finally, the corpus does not yet include downstream benchmarking or integration with established legal NLP pipelines \(e\.g\., LexGLUEChalkidiset al\.\([2022](https://arxiv.org/html/2607.17111#bib.bib10)\)\)\. The use cases outlined above—retrieval\-augmented generation, temporal analysis, and bilingual search—are proposed directions; empirical demonstrations remain future work\.

We regard these limitations as opportunities for community\-driven improvement\. The dataset is released under CC BY\-SA 4\.0, and we welcome contributions that extend coverage, validate metadata, and improve interoperability with multilingual legal tools\.

## References

- Natural Language Processing for the Legal Domain: A Survey of Tasks, Datasets, Models, and Challenges\.arXiv\.Note:Version Number: 2External Links:[Link](https://arxiv.org/abs/2410.21306),[Document](https://dx.doi.org/10.48550/ARXIV.2410.21306)Cited by:[§2](https://arxiv.org/html/2607.17111#S2.p1.1)\.
- T\. Brugger, M\. Stürmer, and J\. Niklaus \(2023\)MultiLegalSBD: A Multilingual Legal Sentence Boundary Detection Dataset\.arXiv\.Note:Version Number: 1External Links:[Link](https://arxiv.org/abs/2305.01211),[Document](https://dx.doi.org/10.48550/ARXIV.2305.01211)Cited by:[§2](https://arxiv.org/html/2607.17111#S2.p6.1)\.
- I\. Chalkidis, A\. Jana, D\. Hartung, M\. Bommarito, I\. Androutsopoulos, D\. Katz, and N\. Aletras \(2022\)LexGLUE: A Benchmark Dataset for Legal Language Understanding in English\.InProceedings of the 60th Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\),Dublin, Ireland\.External Links:[Link](https://aclanthology.org/2022.acl-long.297),[Document](https://dx.doi.org/10.18653/v1/2022.acl-long.297)Cited by:[§2](https://arxiv.org/html/2607.17111#S2.p5.1),[Limitations](https://arxiv.org/html/2607.17111#Sx1.p6.1)\.
- N\. Guha, J\. Nyarko, D\. E\. Ho, C\. Ré, A\. Chilton, A\. Narayana, A\. Chohlas\-Wood, A\. Peters, B\. Waldon, D\. N\. Rockmore, D\. Zambrano, D\. Talisman, E\. Hoque, F\. Surani, F\. Fagan, G\. Sarfaty, G\. M\. Dickinson, H\. Porat, J\. Hegland, J\. Wu, J\. Nudell, J\. Niklaus, J\. Nay, J\. H\. Choi, K\. Tobia, M\. Hagan, M\. Ma, M\. Livermore, N\. Rasumov\-Rahe, N\. Holzenberger, N\. Kolt, P\. Henderson, S\. Rehaag, S\. Goel, S\. Gao, S\. Williams, S\. Gandhi, T\. Zur, V\. Iyer, and Z\. Li \(2023\)LegalBench: A Collaboratively Built Benchmark for Measuring Legal Reasoning in Large Language Models\.arXiv\.Note:Version Number: 1External Links:[Link](https://arxiv.org/abs/2308.11462),[Document](https://dx.doi.org/10.48550/ARXIV.2308.11462)Cited by:[§2](https://arxiv.org/html/2607.17111#S2.p5.1)\.
- W\. Hwang, D\. Lee, K\. Cho, H\. Lee, and M\. Seo \(2022\)A Multi\-Task Benchmark for Korean Legal Language Understanding and Judgement Prediction\.arXiv\.Note:Version Number: 2External Links:[Link](https://arxiv.org/abs/2206.05224),[Document](https://dx.doi.org/10.48550/ARXIV.2206.05224)Cited by:[§2](https://arxiv.org/html/2607.17111#S2.p4.1)\.
- A\. Joshi, S\. Paul, A\. Sharma, P\. Goyal, S\. Ghosh, and A\. Modi \(2024\)IL\-TUR: Benchmark for Indian Legal Text Understanding and Reasoning\.InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\),Bangkok, Thailand,pp\. 11460–11499\.External Links:[Link](https://aclanthology.org/2024.acl-long.618),[Document](https://dx.doi.org/10.18653/v1/2024.acl-long.618)Cited by:[§2](https://arxiv.org/html/2607.17111#S2.p2.1),[§2](https://arxiv.org/html/2607.17111#S2.p6.1)\.
- V\. K and A\. Mishra \(2025\)Dataset for legal question answering system in the Indian judiciary context\.Data in Brief60,pp\. 111647\(eng\)\.External Links:ISSN 2352\-3409,[Document](https://dx.doi.org/10.1016/j.dib.2025.111647)Cited by:[§1](https://arxiv.org/html/2607.17111#S1.p2.1),[§2](https://arxiv.org/html/2607.17111#S2.p2.1)\.
- Y\. Kim, Y\. R\. Choi, E\. Choi, J\. Choi, H\. J\. Park, and W\. Hwang \(2024\)Developing a Pragmatic Benchmark for Assessing Korean Legal Language Understanding in Large Language Models\.arXiv\.Note:Version Number: 1External Links:[Link](https://arxiv.org/abs/2410.08731),[Document](https://dx.doi.org/10.48550/ARXIV.2410.08731)Cited by:[§2](https://arxiv.org/html/2607.17111#S2.p4.1)\.
- H\. Li, Y\. Chen, Q\. Ai, Y\. Wu, R\. Zhang, and Y\. Liu \(2024\)LexEval: A Comprehensive Chinese Legal Benchmark for Evaluating Large Language Models\.arXiv\.Note:Version Number: 4External Links:[Link](https://arxiv.org/abs/2409.20288),[Document](https://dx.doi.org/10.48550/ARXIV.2409.20288)Cited by:[§2](https://arxiv.org/html/2607.17111#S2.p4.1)\.
- M\. K\. Mandal, P\. D\. Nath, A\. S\. Mizan, and N\. Saquib \(2017\)A visual search engine for Bangladeshi laws\.arXiv\.Note:Version Number: 1External Links:[Link](https://arxiv.org/abs/1711.05233),[Document](https://dx.doi.org/10.48550/ARXIV.1711.05233)Cited by:[§2](https://arxiv.org/html/2607.17111#S2.p3.1)\.
- J\. Niklaus, V\. Matoshi, P\. Rani, A\. Galassi, M\. Stürmer, and I\. Chalkidis \(2023\)LEXTREME: A Multi\-Lingual and Multi\-Task Benchmark for the Legal Domain\.InFindings of the Association for Computational Linguistics: EMNLP 2023,Singapore,pp\. 3016–3054\.External Links:[Link](https://aclanthology.org/2023.findings-emnlp.200),[Document](https://dx.doi.org/10.18653/v1/2023.findings-emnlp.200)Cited by:[§1](https://arxiv.org/html/2607.17111#S1.p1.1),[§2](https://arxiv.org/html/2607.17111#S2.p1.1),[§2](https://arxiv.org/html/2607.17111#S2.p5.1),[§2](https://arxiv.org/html/2607.17111#S2.p7.1)\.
- J\. Niklaus, V\. Matoshi, M\. Stürmer, I\. Chalkidis, and D\. Ho \(2024\)MultiLegalPile: A 689GB Multilingual Legal Corpus\.InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\),Bangkok, Thailand,pp\. 15077–15094\.External Links:[Link](https://aclanthology.org/2024.acl-long.805),[Document](https://dx.doi.org/10.18653/v1/2024.acl-long.805)Cited by:[§2](https://arxiv.org/html/2607.17111#S2.p7.1)\.
- A\. T\. Wasi, W\. Faisal, M\. R\. Islam, and M\. M\. Bappy \(2024\)Exploring Possibilities of AI\-Powered Legal Assistance in Bangladesh through Large Language Modeling\.arXiv\.Note:Version Number: 1External Links:[Link](https://arxiv.org/abs/2410.17210),[Document](https://dx.doi.org/10.48550/ARXIV.2410.17210)Cited by:[§1](https://arxiv.org/html/2607.17111#S1.p3.1)\.
- C\. Xiao, H\. Zhong, Z\. Guo, C\. Tu, Z\. Liu, M\. Sun, Y\. Feng, X\. Han, Z\. Hu, H\. Wang, and J\. Xu \(2018\)CAIL2018: A Large\-Scale Legal Dataset for Judgment Prediction\.arXiv\.Note:Version Number: 1External Links:[Link](https://arxiv.org/abs/1807.02478),[Document](https://dx.doi.org/10.48550/ARXIV.1807.02478)Cited by:[§2](https://arxiv.org/html/2607.17111#S2.p4.1)\.

## Appendix ADataset Schema

Each entry in BLAD is stored as a structured JSON object\. The principal fields are summarised below:

- •act\_title: full title of the act\.
- •act\_no: act number \(e\.g\., “V”\)\.
- •act\_year: year of enactment\.
- •sections: list of sections with content text\.
- •footnotes: list of footnotes with amendment history\.
- •language: detected language of the act \(Bengali, English, or mixed\)\.
- •token\_count: number of tokens after cleaning and standardisation\.
- •repealed: boolean indicating whether the act is repealed\.
- •government\_context: metadata on the governing regime at enactment\.
- •legal\_system\_context: legal framework, court structure, policing, and land relations\.
- •processing\_info: timestamps and enhancements applied during cleaning\.
- •source\_url: URL from which the act was originally retrieved\.
- •copyright\_info: attribution to the Ministry of Law\.

## Appendix BSample JSON Entry

A representative record for the act titledTHE \[\*\*\*\] WILLS AND INTESTACY REGULATION, 1799is shown in Figure[7](https://arxiv.org/html/2607.17111#A2.F7); for space, only selected fields are included\.

\{ "act\_title": "THE \[\*\*\*\] WILLS AND INTESTACY REGULATION, 1799", "act\_no": "V", "act\_year": 1799, "language": "english", "token\_count": 1277, "sections": \[ \{"section\_content": "2\. In all cases of Hindu, Mussalman \.\.\."\}, \{"section\_content": "3\. In case of a Hindu, Mussalman \.\.\."\}, \.\.\. \], "footnotes": \[ \{"footnote\_text": "1The word ’Bengal’ was omitted \.\.\."\}, \{"footnote\_text": "2The word ’Bangladesh’ was substituted \.\.\."\} \], "government\_context": \{ "govt\_system": "Company Rule \(Dual System until 1772\)", "head\_govt\_name": "Marquess of Wellesley", "period\_years": "1799\-\-1805" \}, "legal\_system\_context": \{ "period\_info": \{"period\_name": "East India Company Rule"\}, "legal\_framework": \{ "primary\_laws": \[ "Regulating Act 1773", "Cornwallis Code 1793" \], "legal\_basis": "Hybrid of British common law and Islamic/Hindu law" \} \}, "source\_url": "http://bdlaws\.minlaw\.gov\.bd/act\-print\-1315\.html" \}

Figure 7:A representative BLAD JSON record, showing selected fields forTHE \[\*\*\*\] WILLS AND INTESTACY REGULATION, 1799\.
## Appendix CDataset License and Terms of Use

The dataset is released under the Creative Commons Attribution\-ShareAlike 4\.0 International license \(CC BY\-SA 4\.0\)\. Users may share and adapt the material for any purpose, including commercial use, provided that appropriate attribution is given and that derivative works are distributed under the same license\. The underlying legal texts originate from the official Bangladesh Laws Portal and remain subject to attribution to the Ministry of Law, Justice and Parliamentary Affairs\. BLAD is an unofficial compilation intended for research; for authoritative legal reference, users should consult the official portal\.

## Appendix DExtended Charts and Tables

This appendix collects the full\-resolution charts and supplementary tables abbreviated in the main text, including per\-decade counts, per\-government\-system breakdowns, and repeal statistics by head of state\.