This blog post is by Dr Adi Keinan-Schoonbaert, Digital Curator for Asian and African Collections, British Library. She's on Mastodon as @[email protected] and Bluesky as @adi-keinan.bsky.social

The Wikisource 2025 Conference, held in the lush setting of Bali, Indonesia between 14-16 February 2025, brought together a global community of Wikimedians, heritage enthusiasts, and open knowledge advocates. Organised by a coalition of Wikisource contributors, Wikimedia Foundation and Wikimedia Indonesia, the conference served as a dynamic space to discuss the evolving role of Wikisource, explore new technologies, and strengthen collaborations with libraries, cultural institutions, and other global stakeholders.

Wikisource Conference 2025 participants. Photo by Memora Productions for Wikimedia Indonesia.

The conference, themed “Wikisource: Transform & Preserve Heritage Digitally,”  featured a rich programme of keynote talks, long presentations, lightning talks, and informal meet-ups. Central themes included governance, technological advancements, community engagement, and the challenge of scaling Wikisource as a set of collaborative, multilingual platforms. We also enjoyed a couple of fantastic cultural events, celebrating the centuries-old, unique heritage of Bali!

Keynotes and Indonesian Partnerships

Following a kick-off session on the state of Wikisource community and technology, several Indonesian partners shared insights into their work on heritage, preservation, and digital accessibility. Dr Munawar Holil (Kang Mumu) highlighted the efforts of Manassa (the Indonesian Manuscript Society) to safeguard over 121,000 manuscripts, the majority of which remain undigitised, with key collections located in Bali, Jakarta, and Aceh. Challenges include limited public awareness, sacred perceptions requiring ceremonial handling, and structural gaps in institutional training.

Dr Cokorda Rai Adi Paramartha from Udayana University addressed the linguistic diversity of Indonesia – home to 780 languages and 40 scripts, only eight (!) of which are in Unicode – and stressed the importance of developing digital tools like a Balinese keyboard to engage the younger generation. Both speakers underscored the role of community collaboration and technological innovation in making manuscripts more accessible and relevant in the digital age.

Dr Munawar Holil (left), Dr Cokorda Rai Adi Paramartha (right) and session moderator Ivonne Kristiani (WMF; centre).

I had the honour – and the absolute pleasure! – of being invited as one of the keynote speakers for this conference. In my talk I explored collaborations between the British Library and Wikisource, focusing on engaging local communities, raising awareness of library collections, facilitating access to digitised books and manuscripts, and enhancing them with accurate transcriptions.

We have previously collaborated with Bengali communities on two competitions to proofread 19^th century Bengali books digitised as part of the Two Centuries of Indian Print project. More recently, the Library partnered with the Wikisource Loves Manuscripts (WiLMa) project, sharing Javanese manuscripts digitised through the Yogyakarta Digitisation Project. I’ve highlighted past and present work with Transkribus undertaken to develop Machine Learning training models aimed at automating transcriptions in various languages, encouraging further collaborations that could benefit communities worldwide, and highlighting the potential of such partnerships in expanding access to digitised heritage.

Dr Adi Keinan-Schoonbaert delivering a keynote address at the conference. Photo by Memora Productions for Wikimedia Indonesia.

Another keynote was delivered by Andy Stauder from the READ-COOP. After introducing the cooperative and Transkribus, Andy talked about a key component of their approach – CCR – which stands for Clean, Controllable, and Reliable data coupled with information extraction (NER), powered by end-to-end ATR (automated text recognition) models. This approach is essential for both training and processing with large language models (LLMs). The future may move beyond pre-training to embrace active learning, fine-tuning, retrieval-augmented generation (RAG), dynamic prompt engineering, and reinforcement learning, with an aim to generate linked knowledge—such as integration with Wikidata IDs. Community collaboration remains central, as seen in projects like the digitisation of Indonesian palm-leaf manuscripts using Transkribus.

Andy Stauder (READ-COOP) talking about collaboration around the Indonesian palm-leaf manuscripts digitisation

Cassie Chan (Google APAC Search Partnerships) gave a third keynote on Google's role in digitising and curating cultural and literary heritage, aligning with Wikisource’s mission of providing free access to source texts. Projects like Google Books aim to make out-of-copyright works discoverable online, while Google Arts & Culture showcases curated collections such as the Timbuktu Manuscripts, aiding preservation and accessibility. These efforts support Wikimedia goals by offering valuable, context-rich resources for contributors. Additionally, Google's use of AI for cultural exploration – through tools like Poem Postcards and Art Selfie – demonstrates innovative approaches to engaging with global heritage.

Spotlight on Key Themes and Takeaways

The conference featured so many interesting talks and discussions, providing insights into projects, sharing knowledge, and encouraging collaborations. I’ll mention here just a few themes and some key takeaways, from my perspective as someone working with heritage collections, communities, and technology.

Starting with the latter, a major focus was on Optical Character Recognition (OCR) improvements. Enhanced OCR capabilities on Wikisource platforms not only improve text accuracy but also encourage more volunteers to engage in text correction. Implementing Google OCR, Tesseract, and more recently – Transkribus – are driving increased participation, as volunteers enjoy refining text accuracy. Among other speakers, User:Darafsh, Chairman of the Iranian Wikimedians User Group, mentioned the importance of teaching how to use Wikisource and OCR, and the development of Persian OCR at the University of Hamburg. Other talks relating to technology covered the introduction of new extensions, widgets, and mobile apps, highlighting the push to make Wikisource more user-friendly and scalable.

Nicolas Vigneron showcasing the languages for which Google OCR was implemented on Wikisource

Some discussions explored the potential of WiLMa (Wikisource Loves Manuscripts) as a model for coordinating across stakeholders, ensuring the consistency of tools, and fostering engagement with cultural institutions. For example, Irvin Tomas and Maffeth Opiana talked about WiLMa Philippines. This project launched in June 2024 as the first WiLMa project outside of Indonesia, focusing on transcribing and proofreading Central Bikol texts through activities like monthly proofread-a-thons, a 12-hour transcribe-a-thon, and training sessions at universities.

Another interesting topic was that of Wikidata and Metadata. The integration of structured metadata remains a key area of development, enabling better searchability and linking across digital archives. Bodhisattwa Mandal (West Bengal Wikimedians User Group) talked about Wikisource content including both descriptive metadata and unstructured text. While most data isn’t yet stored in a structured format, using Wikidata enables easier updates, avoids redundancy, and improves search, queries, and visualisation. There are tools that support metadata enrichment, annotation, and cataloguing, and a forthcoming mobile app will allow Wikidata-based book searches. Annotating text with Wikidata items enhances discoverability and link content more effectively across Wikimedia projects.

Working for the British Library, I (naturally!) picked up on a few collaborative projects between Wikisource and public or national libraries. One talk was about a digitisation project for traditional Korean texts, a three-year collaboration with Wikimedia Korea and the National Library of Korea, successfully revitalising the Korean Wikisource community by increasing participation and engaging volunteers through events and partnerships.

Another project built a Wikisource community in Uganda by training university students, particularly from library information studies, alongside existing volunteers. Through practical sessions, collaborative tasks, and support from institutions like the National Library of Uganda and Wikimedia contributors, participants developed digital literacy and archival skills.

Nanteza Divine Gabriella giving a talk on ‘Training Wikisource 101’ and building a Wikisource community in Uganda

A third Wikisource and libraries talk was about a Wikisource to public library pipeline project, which started initially in a public library in Hokitika, New Zealand. This pipeline enables scanned public domain books to be transcribed on Wikisource and then made available as lendable eBooks via the Libby app, using OverDrive's Local Content feature. With strong librarian involvement, a clear workflow, and support from a small grant, the project has successfully bridged Wikisource and library systems to increase accessibility and customise reading experiences for library users.

The final session of the conference focused on shaping a future roadmap for Wikisource through community-driven conversation, strategic planning, and partnership development. Discussions emphasised the need for clearer vision, sustainable collaborations with technology and cultural institutions, improved tools and infrastructure, and greater outreach to grow both readership and contributor communities. Key takeaways included aligning with partners’ goals, investing in editor growth, leveraging government language initiatives, and developing innovative workflows. A strong call was made to prioritise people over platforms and to ensure Wikisource remains a meaningful and inclusive space for engaging with knowledge and heritage.

Looking Ahead

The Wikisource 2025 Conference reaffirmed the platform’s importance in the digital knowledge ecosystem. However, sustaining momentum requires ongoing advocacy, technological refinement, and deeper institutional partnerships. Whether through digitising new materials or leveraging already-digitised collections, there is a clear hunger for openly accessible public domain texts.

As the community moves forward, a focus on governance, technology, and strategic partnerships will be essential in shaping the future of Wikisource. The atmosphere was so positive and there was so much enthusiasm and willingness to collaborate – see this fantastic video available via Wikimedia Commons, which successfully captures the sentiment. I’m sure we’re going to see a lot more coming from Wikisource communities in the future!

Digital Curator Mia Ridge writes...

While the British Library works to recover from the October 2023 cyber-attack, we're putting some information from our currently inaccessible website into an easily readable and shareable format. This blog post is based on a page captured by the Wayback Machine in September 2023.

Crowdsourcing at the British Library

Example of a crowdsourcing task

For the British Library, crowdsourcing is an engaging form of online volunteering supported by digital tools that manage tasks such as transcription, classification and geolocation that make our collections more discoverable.

The British Library has run several popular crowdsourcing projects in the past, including the Georeferencer, for geolocating historical maps, and In the Spotlight, for transcribing important information about historical playbills. We also integrated crowdsourcing activities into our flagship AI / data science project, Living with Machines.

• Agents of Enslavement uses 18th/19th century newspapers to research slavery in Barbados and create a database of enslaved people.
• Living with Machines, which is mostly based on research questions around nineteenth century newspapers

Crowdsourcing Projects at the British Library

• Living with Machines (2019-2023) created innovative crowdsourced tasks, including tasks that asked the public to closely read historical newspaper articles to determine how specific words were used.
• Agents of Enslavement (2021-2022) used 18th/19th century newspapers to research slavery in Barbados and create a database of enslaved people.
• In the Spotlight (2017-2021) was a crowdsourcing project from the British Library that aimed to make digitised historical playbills more discoverable, while also encouraging people to closely engage with this otherwise less accessible collection of ephemera.
• Canadian wildlife: notes from the field (2021), a project where volunteers transcribed handwritten field notes that accompany recordings of a wildlife collection within the sound archive.
• Convert a Card (2015) was a series of crowdsourcing projects aimed to convert scanned catalogue cards in Asian and African languages into electronic records. The project template can be found and used on GitHub.
• Georeferencer (2012 - present) enabled volunteers to create geospatial data from digitised versions of print maps by adding control points to the old and modern maps.
• Pin-a-Tale (2012) asked people to map literary texts to British places.

Research Projects

The Living with Machines project included a large component of crowdsourcing research through practice, led by Digital Curator Mia Ridge.

Mia was also the Principle Investigator on the AHRC-funded Collective Wisdom project, which worked with a large group of co-authors to produce a book, The Collective Wisdom Handbook: perspectives on crowdsourcing in cultural heritage, through two 'book sprints' in 2021:

This book is written for crowdsourcing practitioners who work in cultural institutions, as well as those who wish to gain experience with crowdsourcing. It provides both practical tips, grounded in lessons often learned the hard way, and inspiration from research across a range of disciplines. Case studies and perspectives based on our experience are woven throughout the book, complemented by information drawn from research literature and practice within the field.

More Information

Our crowdsourcing projects were designed to produce data that can be used in discovery systems (such as online catalogues and our item viewer) through enjoyable tasks that give volunteers an opportunity to explore digitised collections.

Each project involves teams across the Library to supply digitised images for crowdsourcing and ensure that the results are processed and ingested into various systems. Enhancing metadata through crowdsourcing is considered in the British Library's Collection Metadata Strategy.

We previously posted on twitter @LibCrowds and currently post occasionally on Mastodon https://glammr.us/@libcrowds and via our newsletter.

Past editions of our newsletter are available online.

This blog post is by Dr Adi Keinan-Schoonbaert, Digital Curator for Asian and African Collections, British Library. She's on Mastodon as @[email protected]. 

The British Library’s vast Southeast Asian collection includes manuscripts, periodicals and printed books in the languages of the countries of maritime Southeast Asia, including Indonesia, Malaysia, Singapore, Brunei, the Philippines and East Timor, as well as on the mainland, from Thailand, Laos, Cambodia, Myanmar (Burma) and Vietnam.

The display of literary manuscripts from Southeast Asia outside of the Asian and African Studies Reading Room in St Pancras (photo by Adi Keinan-Schoonbaert)

Several languages and scripts from the mainland were the focus of recent development work commissioned by the Library and done on the script conversion platform Aksharamukha. These include Shan, Khmer, Khuen, and northern Thai and Lao Dhamma (Dhamma, or Tham, meaning ‘scripture’, is the script that several languages are written in).

These and other Southeast Asian languages and scripts pose multiple challenges to us and our users. Collection items in languages using non-romanised scripts are mainly catalogued (and therefore searched by users) using romanised text. For some language groups, users need to search the catalogue by typing in the transliteration of title and/or author using the Library of Congress (LoC) romanisation rules.

Items’ metadata text converted using the LoC romanisation scheme is often unintuitive, and therefore poses a barrier for users, hindering discovery and access to our collections via the online catalogues. In addition, curatorial and acquisition staff spend a significant amount of time manually converting scripts, a slow process which is prone to errors. Other libraries worldwide holding Southeast Asian collections and using the LoC romanisation scheme face the same issues.

Excerpt from the Library of Congress romanisation scheme for Khmer

Having faced these issues with Burmese language, last year we commissioned development work to the open-access platform Aksharamukha, which enables the conversion between various scripts, supporting 121 scripts and 21 romanisation methods. Vinodh Rajan, Aksharamukha’s developer, perfectly combines language and writing systems knowledge with computer science and coding skills. He added the LoC romanisation system to the platform’s Burmese script transliteration functionality (read about this in my previous post).

The results were outstanding – readers could copy and paste transliterated text into the Library's catalogue search box to check if we have items of interest. This has also greatly enhanced cataloguing and acquisition processes by enabling the creation of acquisition records and minimal records. In addition, our Metadata team updated all of our Burmese catalogue records (ca. 20,000) to include Burmese script, alongside transliteration (side note: these updated records are still unavailable to our readers due to the cyber-attack on the Library last year, but they will become accessible in the future).

The time was ripe to expand our collaboration with Vinodh and Aksharamukha. Maria Kekki, Curator for Burmese Collections, has been hosting this past year a Chevening Fellow from Myanmar, Myo Thant Linn. Myo was tasked with cataloguing manuscripts and printed books in Shan and Khuen – but found the romanisation aspect of this work to be very challenging to do manually. In order to facilitate Myo’s work and maximise the benefit from his fellowship, we needed to have a LoC romanisation functionality available. Aksharamukha was the right place for this – this free, open source, online tool is available to our curators, cataloguers, acquisition staff, and metadata team to use.

Former Chevening Fellow Myo Thant Linn reciting from a Shan manuscript in the Asian and African Studies Reading Room, September 2024 (photo by Jana Igunma)

In addition to Maria and Myo’s requirements, Jana Igunma, Ginsburg Curator for Thai, Lao and Cambodian Collections, noted that adding Khmer to Aksharamukha would be immensely helpful for cataloguing our Khmer backlog and assist with new acquisitions. Northern Thai and Lao Dhamma scripts would be mostly useful to catalogue new acquisitions for print material, and add original scripts to manuscript records. The automation of LoC transliteration could be very cost-effective, by saving many cataloguing, acquisitions and metadata team’s hours. Khmer is a great example – it has the most extensive alphabet in the world (74 letters), and its romanisation is extremely complicated and time consuming!

First three leaves with text in a long format palm leaf bundle (សាស្ត្រាស្លឹករឹត/sāstrā slẏk rẏt) containing part of the Buddhist cosmology (សាស្ត្រាត្រៃភូមិ/Sāstrā Traibhūmi) in Khmer script, 18th or 19th century. Acquired by the British Museum from Edwards Goutier, Paris, on 6 December 1895. British Library, Or 5003, ff. 9-11

It was required, therefore, to enhance Aksharamukha’s script conversion functionality with these additional scripts. This could generally be done by referring to existing LoC conversion tables, while taking into account any permutations of diacritics or character variations. However, it definitely has not been as simple as that!

For example, the presence of diacritics instigated a discussion between internal and external colleagues on the use of precomposed vs. decomposed formats in Unicode, when romanising original script. LoC systems use two types of coding schemata, MARC 21 and MARC 8. The former allows for precomposed diacritic characters, and the latter does not – it allows for decomposed format. In order to enable both these schemata, Vinodh included both MARC 8 and MARC 21 as input and output formats in the conversion functionality.

Another component, implemented for Burmese in the previous development round, but also needed for Khmer and Shan transliterations, is word spacing. Vinodh implemented word separation in this round as well – although this would always remain something that the cataloguer would need to check and adjust. Note that this is not enabled by default – you would have to select it (under ‘input’ – see image below).

Screenshot from Aksharamukha, showcasing Khmer word segmentation option

It is heartening to know that enhancing Aksharamukha has been making a difference. Internally, Myo had been a keen user of the Shan romanisation functionality (though Khuen romanisation is still work-in-progress); and Jana has been using the Khmer transliteration too. Jana found it particularly useful to use Aksharamukha’s option to upload a photo of the title page, which is then automatically OCRed and romanised. This saved precious time otherwise spent on typing Khmer!

It should be mentioned that, when it comes to cataloguing Khmer language books at the British Library, both original Khmer script and romanised metadata are being included in catalogue records. Aksharamukha helps to speed up the process of cataloguing and eliminates typing errors. However, capitalisation and in some instances word separation and final consonants need to be adjusted manually by the cataloguer. Therefore, it is necessary that the cataloguer has a good knowledge of the language.

On the left: photo of a title page of a Khmer language textbook for Grade 9, recently acquired by the British Library; on the right: conversion of original Khmer text from the title page into LoC romanisation standard using Aksharamukha

The conversion tool for Tham (Lanna) and Tham (Lao) works best for texts in Pali language, according to its LoC romanisation table. If Aksharamukha is used for works in northern Thai language in Tham (Lanna) script, or Lao language in Tham (Lao) script, cataloguer intervention is always required as there is no LoC romanisation standard for northern Thai and Lao languages in Tham scripts. Such publications are rare, and an interim solution that has been adopted by various libraries is to convert Tham scripts to modern Thai or Lao scripts, and then to romanise them according to the LoC romanisation standards for these languages.

Other libraries have been enjoying the benefits of the new developments to Aksharamukha. Conversations with colleagues from the Library of Congress revealed that present and past commissioned developments on Aksharamukha had a positive impact on their operations. LoC has been developing a transliteration tool called ScriptShifter. Aksharamukha’s Burmese and Khmer functionalities are already integrated into this tool, which can convert over ninety non-Latin scripts into Latin script following the LoC/ALA guidelines. The British Library funding Aksharamukha to make several Southeast Asian languages and scripts available in LoC romanisation has already been useful!

If you have feedback or encounter any bugs, please feel free to raise an issue on GitHub. And, if you’re interested in other scripts romanised using LoC schemas, Aksharamukha has a complete list of the ones that it supports. Happy conversions!

This is a repeated and updated blog post by Dr Adi Keinan-Schoonbaert, Digital Curator for Asian and African Collections. She shares some background information on how a new post advertised for a Digital Curator for OCR/HTR will help the Library streamline post-digitisation work to make its collections even more accessible to users. Our previous run of this recruitment was curtailed due to the cyber-attack on the Library - but we are now ready to restart the process!

We’ve been digitising our collections for about three decades, opening up access to incredibly diverse and rich collections, for our users to study and enjoy. However, it is important that we further support discovery and digital research by unlocking the huge potential in automatically transcribing our collections.

We’ve done some work over the years towards making our collection items available in machine-readable format, in order to enable full-text search and analysis. Optical Character Recognition (OCR) technology has been around for a while, and there are several large-scale projects that produced OCRed text alongside digitised images – such as the Microsoft Books project. Until recently, Western languages print collections have been the main focus, especially newspaper collections. A flagship collaboration with the Alan Turing Institute, the Living with Machines project, applied OCR technology to UK newspapers, designing and implementing new methods in data science and artificial intelligence, and analysing these materials at scale.

OCR of Bengali books using Transkribus, Two Centuries of Indian Print Project

Machine Learning technologies have been dealing increasingly well with both modern and historical collections, whether printed, typewritten or handwritten. Taking a broader perspective on Library collections, we have been exploring opportunities with non-Western collections too. Library staff have been engaging closely with the exploration of OCR and Handwritten Text Recognition (HTR) systems for English, Bangla, Arabic, Urdu and Chinese. Digital Curators Tom Derrick, Nora McGregor and Adi Keinan-Schoonbaert have teamed up with PRImA Research Lab and the Alan Turing Institute to run four competitions in 2017-2019, inviting providers of text recognition methods to try them out on our historical material.

We have been working with Transkribus as well – for example, Alex Hailey, Curator for Modern Archives and Manuscripts, used the software to automatically transcribe 19^th century botanical records from the India Office Records. A digital humanities work strand led by former colleague Tom Derrick saw the OCR of most of our digitised collection of Bengali printed texts, digitised as part of the Two Centuries of Indian Print project. More recently Transkribus has been used to extract text from catalogue cards in a project called Convert-a-Card, as well as from Incunabula print catalogues.

An example of a catalogue card in Transkribus, showing segmentation and transcription

We've also collaborated with Colin Brisson from the READ_Chinese project on Chinese HTR, working with eScriptorium to enhance binarisation, segmentation and transcription models using manuscripts that were digitised as part of the International Dunhuang Programme. You can read more about this work in this brilliant blog post by Peter Smith, who's done a PhD placement with us last year.

The British Library is now looking for someone to join us to further improve the access and usability of our digital collections, by integrating a standardised OCR and HTR production process into our existing workflows, in line with industry best practice.

For more information and to apply please visit the ad for Digital Curator for OCR/HTR on the British Library recruitment site. Applications close on Sunday 21 July 2024. Please pay close attention to questions asked in the application process. Any questions? Drop us a line at [email protected].

Good luck!

This blog post is by Peter Smith, DPhil Student at the Faculty of Asian and Middle Eastern Studies, University of Oxford

Introduction

The study of writing and literature has been transformed by the mass transcription of printed materials, aided significantly by the use of Optical Character Recognition (OCR). This has enabled textual analysis through a growing array of digital techniques, ranging from simple word searches in a text to linguistic analysis of large corpora – the possibilities are yet to be fully explored. However, printed materials are only one expression of the written word and tend to be more representative of certain types of writing. These may be shaped by efforts to standardise spelling or character variants, they may use more formal or literary styles of language, and they are often edited and polished with great care. They will never reveal the great, messy diversity of features that occur in writings produced by the human hand. What of the personal letters and documents, poems and essays scribbled on paper with no intention of distribution; the unpublished drafts of a major literary work; or manuscript editions of various classics that, before the use of print, were the sole means of preserving ancient writings and handing them onto future generations? These are also a rich resource for exploring past lives and events or expressions of literary culture.

The study of handwritten materials is not new but, until recently, the possibilities for analysing them using digital tools have been quite limited. With the advent of Handwritten Text Recognition (HTR) the picture is starting to change. HTR applications such as Transkribus and eScriptorium are capable of learning to transcribe a broad range of scripts in multiple languages. As the potential of these platforms develops, large collections of manuscripts can be automatically transcribed and consequently explored using digital tools. Institutions such as the British Library are doing much to encourage this process and improve accessibility of the transcribed works for academic research and the general interest of the public. My recent role in an HTR project at the Library represents one small step in this process and here I hope to provide a glimpse behind-the-scenes, a look at some of the challenges of developing HTR.

As a PhD student exploring classical Chinese texts, I was delighted to find a placement at the British Library working on HTR of historical Chinese manuscripts. This project proceeded under the guidance of my British Library supervisors Dr Adi Keinan-Schoonbaert and Mélodie Doumy. I was also provided with support and expertise from outside of the Library: Colin Brisson is part of a group working on Chinese Historical documents Automatic Transcription (CHAT). They have already gathered and developed preliminary models for processing handwritten Chinese with the open source HTR application eScriptorium. I worked with Colin to train the software further using materials from the British Library. These were drawn entirely from the fabulous collection of manuscripts from Dunhuang, China, which date back to the Tang dynasty (618–907 CE) and beyond. Examples of these can be seen below, along with reference numbers for each item, and the originals can be viewed on the new website of the International Dunhuang Programme. Some of these texts were written with great care in standard Chinese scripts and are very well preserved. Others are much more messy: cursive scripts, irregular layouts, character corrections, and margin notes are all common features of handwritten work. The writing materials themselves may be stained, torn, or eaten by animals, resulting in missing or illegible text. All these issues have the potential to mislead the ‘intelligence’ of a machine. To overcome such challenges the software requires data – multiple examples of the diverse elements it might encounter and instruction as to how they should be understood.

The challenges encountered in my work on HTR can be examined in three broad categories, reflecting three steps in the HTR process of eScriptorium: image binarisation, layout segmentation, and text recognition.

Image binarisation

The first task in processing an image is to reduce its complexity, to remove any information that is not relevant to the output required. One way of doing this is image binarisation, taking a colour image and using an algorithm to strip it of hue and brightness values so that only black and white pixels remain. This was achieved using a binarisation model developed by Colin Brisson and his partners. My role in this stage was to observe the results of the process and identify strengths and weaknesses in the current model. These break down into three different categories: capturing details, stained or discoloured paper, and colour and density of ink.

1. Capturing details

In the process of distinguishing the brushstrokes of characters from other random marks on the paper, it is perhaps inevitable that some thin or faint lines – occurring as a feature of the hand written text or through deterioration over time – might be lost during binarisation. Typically the binarisation model does very well in picking them out, as seen in figure 1:

Fig 1. Good retention of thin lines (S.3011, recto image 23)

While problems with faint strokes are understandable, it was surprising to find that loss of detail was also an issue in somewhat thicker lines. I wasn’t able to determine the cause of this but it occurred in more than one image. See figures 2 and 3:

Fig 2. Loss of detail in thick lines (S.3011, recto image 23)

Fig 3. Loss of detail in thick lines (S.3011, recto image 23)

2. Stained and discoloured paper

Where paper has darkened over time, the contrast between ink and background is diminished and during binarisation some writing may be entirely removed along with the dark colours of the paper. Although I encountered this occasionally, unless the background was really dark the binarisation model did well. One notable success is its ability to remove the dark colours of partially stained sections. This can be seen in figure 4, where a dark stain is removed while a good amount of detail is retained in the written characters.

Fig 4. Good retention of character detail on heavily stained paper (S.2200, recto image 6)

3. Colour and density of ink

The majority of manuscripts are written in black ink, ideal for creating good contrast with most background colourations. In some places however, text may be written with less concentrated ink, resulting in greyer tones that are not so easy to distinguish from the paper. The binarisation model can identify these correctly but sometimes it fails to distinguish them from the other random markings and colour variations that can be found in the paper of ancient manuscripts. Of particular interest is the use of red ink, which is often indicative of later annotations in the margins or between lines, or used for the addition of punctuation. The current binarisation model will sometimes ignore red ink if it is very faint but in most cases it identifies it very well. In one impressive example, shown in figure 5, it identified the red text while removing larger red marks used to highlight other characters written in black ink, demonstrating an ability to distinguish between semantic and less significant information.

Fig 5. Effective retention of red characters and removal of large red marks (S.2200, recto image 7)

In summary, the examples above show that the current binarisation model is already very effective at eliminating unwanted background colours and stains while preserving most of the important character detail. Its response to red ink illustrates a capacity for nuanced analysis. It does not treat every red pixel in the same way, but determines whether to keep it or remove it according to the context. There is clearly room for further training and refinement of the model but it already produces materials that are quite suitable for the next stages of the HTR process.

Layout segmentation

Segmentation defines the different regions of a digitised manuscript and the type of content they contain, either text or image. Lines are drawn around blocks of text to establish a text region and for many manuscripts there is just one per image. Anything outside of the marked regions will just be ignored by the software. On occasion, additional regions might be used to distinguish writings in the margins of the manuscript. Finally, within each text region the lines of text must also be clearly marked. Having established the location of the lines, they can be assigned a particular type. In this project the options include ‘default’, ‘double line’, and ‘other’ – the purpose of these will be explored below.

All of this work can be automated in eScriptorium using a segmentation model. However, when it comes to analysing Chinese manuscripts, this model was the least developed component in the eScriptorium HTR process and much of our work focused on developing its capabilities. My task was to run binarised images through the model and then manually correct any errors. Figure 6 shows the eScriptorium interface and the initial results produced by the segmentation model. Vertical sections of text are marked with a purple line and the endings of each section are indicated with a horizontal pink line.

Fig 6. Initial results of the segmentation model section showing multiple errors. The text is the Zhuangzi Commentary by Guo Xiang (S.1603)

This example shows that the segmentation model is very good at positioning a line in the centre of a vertical column of text. Frequently, however, single lines of text are marked as a sequence of separate lines while other lines of text are completely ignored. The correct output, achieved through manual segmentation, is shown in figure 7. Every line is marked from beginning to end with no omissions or inappropriate breaks.

Fig 7. Results of manual segmentation showing the text region (the blue rectangle) and the single and double lines of text (S.1603)

Once the lines of a text are marked, line masks can be generated automatically, defining the area of text around each line. Masks are needed to show the transcription model (discussed below) exactly where it should look when attempting to match images on the page to digital characters. The example in figure 8 shows that the results of the masking process are almost perfect, encompassing every Chinese character without overlapping other lines.

Fig 8. Line masks outline the area of text associated with each line (S.1603)

The main challenge with developing a good segmentation model is that manuscripts in the Dunhuang collection have so much variation in layout. Large and small characters mix together in different ways and the distribution of lines and characters can vary considerably. When selecting material for this project I picked a range of standard layouts. This provided some degree of variation but also contained enough repetition for the training to be effective. For example, the manuscript shown above in figures 6–8 combines a classical text written in large characters interspersed with double lines of commentary in smaller writing, in this case it is the Zhuangzi Commentary by Guo Xiang. The large text is assigned the ‘default’ line type while the smaller lines of commentary are marked as ‘double-line’ text. There is also an ‘other’ line type which can be applied to anything that isn’t part of the main text – margin notes are one example. Line types do not affect how characters are transcribed but they can be used to determine how different sections of text relate to each other and how they are assembled and formatted in the final output files.

Fig 9. A section from the Lotus Sūtra with a text region, lines of prose, and lines of verse clearly marked (Or8210/S.1338)

Figures 8 and 9, above, represent standard layouts used in the writing of a text but manuscripts contain many elements that are more random. Of these, inter-line annotations are a good example. They are typically added by a later hand, offering comments on a particular character or line of text. Annotations might be as short as a single character (figure 10) or could be a much longer comment squeezed in between the lines of text (figure 11). In such cases these additions can be distinguished from the main text by being labelled with the ‘other’ line type.

Fig 10. Single character annotation in S.3011, recto image 14 (left) and a longer annotation in S.5556, recto image 4 (right)

Fig 11. A comment in red ink inserted between two lines of text (S.2200, recto image 5)

Other occasional features include corrections to the text. These might be made by the original scribe or by a later hand. In such cases one character may be blotted out and a replacement added to the side, as seen in figure 12. For the reader, these should be understood as part of the text itself but for the segmentation model they appear similar or identical to annotations. For the purpose of segmentation training any irregular features like this are identified using the ‘other’ line type.

Fig 12. Character correction in S.3011, recto image 23.

As the examples above show, segmentation presents many challenges. Even the standard features of common layouts offer a degree of variation and in some manuscripts irregularities abound. However, work done on this project has now been used for further training of the segmentation model and reports are promising. The model appears capable of learning quickly, even from relatively small data sets. As the process improves, time spent using and training the model offers increasing returns. Even if some errors remain, manual correction is always possible and segmented images can pass through to the final stage of text recognition.

Text recognition

Although transcription is the ultimate aim of this process it consumed less of my time on the project so I will keep this section relatively brief. Fortunately, this is another stage where the available model works very well. It had previously been trained on other print and manuscript collections so a well-established vocabulary set was in place, capable of recognising many of the characters found in historical writings. Dealing with handwritten text is inevitably a greater challenge for a transcription model but my selection of manuscripts included several carefully written texts. I felt there was a good chance of success and was very keen to give it a go, hoping I might end up with some usable transcriptions of these works. Once the transcription model had been run I inspected the first page using eScriptorium’s correction interface as illustrated in figure 13.

Fig 13. Comparison of image and transcription in eScriptorium’s correction interface.

The interface presents a single line from the scanned image alongside the digitally transcribed text, allowing me to check each character and amend any errors. I quickly scanned the first few lines hoping I would find something other than random symbols – I was not disappointed! The results weren’t perfect of course but one or two lines actually came through with no errors at all and generally the character error rate seems very low. After careful correction of the errors that remained and some additional work on the reading order of the lines, I was able to export one complete manuscript transcription bringing the whole process to a satisfying conclusion.

Final thoughts

Naturally there is still some work to be done. All the models would benefit from further refinement and the segmentation model in particular will require training on a broader range of layouts before it can handle the great diversity of the Dunhuang collection. Hopefully future projects will allow more of these manuscripts to be used in the training of eScriptorium so that a robust HTR process can be established. I look forward to further developments and, for now, am very grateful for the chance I’ve had to work alongside my fabulous colleagues at the British Library and play some small role in this work.

This blog post is by Dr Adi Keinan-Schoonbaert, Digital Curator for Asian and African Collections, British Library. She's on Mastodon as @[email protected].

Last week I came back from my very first ICDAR conference, inspired and energised for things to come! The International Conference on Document Analysis and Recognition (ICDAR) is the main international event for scientists and practitioners involved in document analysis and recognition. Its 17^th edition was held in San José, California, 21-26 August 2023.

ICDAR 2023 featured a three-day conference, including several competitions to challenge the field, as well as post-conference workshops and tutorials. All conference papers were made available as conference proceedings with Springer. 155 submissions were selected for inclusion into the scientific programme of ICDAR 2023, out of which 55 were delivered as oral presentations, and 100 as posters. The conference also teamed up with the International Journal of Document Analysis and Recognition (IJDAR) for a special journal track. 13 papers were accepted and published in a special issue entitled “Advanced Topics of Document Analysis and Recognition,” and were included as oral presentations in the conference programme. Do have a look at the programme booklet for more information!

ICDAR 2023 Logo

Each conference day included a thought-provoking keynote talk. The first one, by Marti Hearst, Professor and Interim Dean of the UC Berkeley School of Information, was entitled “A First Look at LLMs Applied to Scientific Documents.” I learned about three platforms using Natural Language Processing (NLP) methods on PDF documents: ScholarPhi, Paper Plain, and SCIM. These projects help people read academic scientific publications, for example by enabling definitions for mathematical notations, or generating glossary for nonce words (e.g. acronyms, symbols, jargon terms); make medical research more accessible by enabling simplified summaries and Q&A; and classifying key passages in papers to enable quick and intelligent paper skimming.

The second keynote talk, “Enabling the Document Experiences of the Future,” was by Vlad Morariu, Senior Research Scientist at Adobe Research. Vlad addressed the need for human-document interaction, and took us through some future document experiences: PDF re-flows for mobile devices, documents read themselves, and conversational functionalities such as asking questions and receiving answers. Enabling this type of ultra-responsive documents is reliant on methods such as structural element detection, page layout understanding, and semantic connections.

The third and final keynote talk was by Seiichi Uchida, Distinguished Professor and Senior Vice President, Kyushu University, Japan. In his talk, “What Are Letters?,” Seiichi took us through the four main functions of letters and text: message (transmission of verbalised info), label (disambiguation of objects and environments), design (give a nonverbal info, such as impression), and code (readability under various noises and deformations). He provoked us to contemplate how our lives were affected by texts around us, and how could we analyse the correlation between our behaviour and the texts that we read.

Prof Seiichi Uchida giving his keynote talk on “What Are Letters?”

When it came to papers submitted for review by the conference committee, the most prominent topic represented in those submissions was handwriting recognition, with a growing number of papers specifically tackling historical documents. Other submission topics included Graphics Recognition, Natural Language Processing for Documents (D-NLP), Applications (including for medical, legal, and business documents), and other types of Document Analysis and Recognition topics (DAR).

Screenshot of a slide showing the main submission topics for ICDAR 2023

Some of the papers that I attended tackled Named Entity Recognition (NER) evaluation methods and genealogical information extraction; papers dealing with Document Understanding, e.g. identifying the internal structure of documents, and understanding the relations between different entities; papers on Text and Document Recognition, such as looking into a model for multilingual OCR; and papers looking into Graphics, especially the recognition of table structure and content, as well as extracting data from structure diagrammes, for example in financial documents, or flowchart recognition. Papers on Handwritten Text Recognition (HTR) dealt with methods for Writer Retrieval, i.e. identifying documents likely written by specific authors, the creation of generic models, text line detection, and more.

The conference included two poster sessions, featuring an incredibly rich array of poster presentations, as well as doctoral consortia. One of my favourite posters was presented by Mirjam Cuper, Data Scientist at the National Library of the Netherlands (KB), entitled “Unraveling confidence: examining confidence scores as proxy for OCR quality.” Together with colleagues Corine van Dongen and Tineke Koster, she looked into confidence scores provided by OCR engines, which indicate the level of certainty in which a word or character were accurately recognised. However, other factors are at play when measuring OCR quality – you can watch a ‘teaser’ video for this poster.

Conference participants at one of the poster sessions

As mentioned, the conference was followed by three days of tutorials and workshops. I enjoyed the tutorial on Computational Analysis of Historical Documents, co-led by Dr Isabelle Marthot-Santaniello (University of Bale, Switzerland) and Dr Hussein Adnan Mohammed (University of Hamburg, Germany). Presentations focused on the unique challenges, difficulties, and opportunities inherent to working with different types of historical documents. The distinct difficulties posed by historical handwritten manuscripts and ancient artifacts necessitate an interdisciplinary strategy and the utilisation of state-of-the-art technologies – and this fusion leads to the emergence of exciting and novel advancements in this area. The presentations were interwoven with great questions and a rich discussion, indicative of the audience’s enthusiasm. This tutorial was appropriately followed by a workshop dedicated to Computational Palaeography (IWCP).

I especially looked forward to the next day’s workshop, which was the 7^th edition of Historical Document Imaging and Processing (HIP’23). It was all about making documents accessible in digital libraries, looking at methods addressing OCR/HTR of historical documents, information extraction, writer identification, script transliteration, virtual reconstruction, and so much more. This day-long workshop featured papers in four sessions: HTR and Multi-Modal Methods, Classics, Segmentation & Layout Analysis, and Language Technologies & Classification. One of my favourite presentations was by Prof Apostolos Antonacopoulos, talking about his work with Christian Clausner and Stefan Pletschacher on “NAME – A Rich XML Format for Named Entity and Relation Tagging.” Their NAME XML tackles the need to represent named entities in rich and complex scenarios. Tags could be overlapping and nested, character-precise, multi-part, and possibly with non-consecutive words or tokens. This flexible and extensible format addresses the relationships between entities, makes them interoperable, usable alongside other information (images and other formats), and possible to validate.

Prof Apostolos Antonacopoulos talking about “NAME – A Rich XML Format for Named Entity and Relation Tagging”

I’ve greatly enjoyed the conference and its wonderful community, meeting old colleagues and making new friends. Until next time!

Digitised archives and library collections are rich resources for creative practitioners, including video game makers, who can bring history to life in new ways with immersive storytelling. A wonderful example of this is Astrologaster by Nyamyam, an interactive comedy set in Elizabethan London, based on the manuscripts of medical astrologer Simon Forman, which is currently showcased in the British Library’s Digital Storytelling exhibition.

On Friday 15^th September we are delighted to host an event to celebrate the making and the music of Astrologaster. Featuring game designer Jennifer Schneidereit in conversation with historian Lauren Kassell discussing how they created the game. Followed by a vocal quartet who will sing madrigal songs from the soundtrack composed by Andrea Boccadoro. Each character in the game has their own Renaissance style theme song with witty lyrics written by Katharine Neil. This set has never before been performed live, so we can’t wait to hear these songs at the Library and we would love for you to join us, click here to book. We've had the title song, which you can play below, as an earworm for the last few months!

Simon Forman was a self-taught doctor and astrologer who claimed to have cured himself of the plague in 1592. Despite being unlicensed and scorned by the Royal College of Physicians he established a practice in London where he analysed the stars to diagnose and solve his querents’ personal, professional and medical problems. Forman documented his life and work in detail, leaving a vast quantity of papers to his protégé Richard Napier, whose archive was subsequently acquired by Elias Ashmole for the Ashmolean Museum at the University of Oxford. In the nineteenth century this collection transferred to the Bodleian Library, where Forman’s manuscripts can still be consulted today.

Screen capture image of the Casebooks digital edition showing ‘CASE5148’.

Lauren Kassell, Michael Hawkins, Robert Ralley, John Young, Joanne Edge, Janet Yvonne Martin-Portugues, and Natalie Kaoukji (eds.), ‘CASE5148’, The casebooks of Simon Forman and Richard Napier, 1596–1634: a digital edition, https://casebooks.lib.cam.ac.uk/cases/CASE5148, accessed 1 September 2023.

Funded by the Wellcome Trust, the Casebooks Project led by Professor Lauren Kassell at the University of Cambridge, spent over a decade researching, digitising, documenting and transcribing these records. Producing The casebooks of Simon Forman and Richard Napier, 1596–1634: a digital edition published by Cambridge Digital Library in May 2019. Transforming the archive into a rich searchable online resource, with transcriptions and editorial insights about the astrologers’ records, alongside digitised images of the manuscripts.

In 2014 Nyamyam’s co-founder and creative director Jennifer Schneidereit saw Lauren present her research on Simon Forman’s casebooks, and became fascinated by this ambitious astrologer. Convinced that Forman and his patients’ stories would make an engaging game with astrology as a gameplay device, she reached out to Lauren to invite her to be a consultant on the project. Fortunately Lauren responded positively and arranged for the Casebooks Project to formally collaborate with Nyamyam to mine Forman’s patient records for information and inspiration to create the characters and narrative in the Astrologaster game.  

Still image of a playthrough video demonstrating how to play Astrologaster made by Florence Smith Nicholls for the Digital Storytelling exhibition

At the British Library we are interested in collecting and curating interactive digital narratives as part of our ongoing emerging formats research. One method we are investigating is the acquisition and creation of contextual information, such as recording playthrough videos. In the Digital Storytelling exhibition you can watch three gameplay recordings, including one demonstrating how to play Astrologaster. These were made by Florence Smith Nicholls, a game AI PhD researcher based at Queen Mary University of London, using facilities at the City Interaction Lab within the Centre for Human-Computer Interaction Design at City, University of London. Beyond the exhibition, these recordings will hopefully benefit researchers in the future, providing valuable documentation on the original ‘look and feel’ of an interactive digital narrative, in addition to instructions on use whenever a format has become obsolete.

The Digital Storytelling exhibition is open until the 15th October 2023 at the British Library, displaying 11 narratives that demonstrate the evolving field of interactive writing. We hope you can join us for upcoming related events, including the Astrologaster performance on Friday 15^th September, and an epic Steampunk Late on Friday 13th October. We are planning this Late with Clockwork Watch, Blockworks and Lancaster University's Litcraft initiative, so watch this blog for more information on this event soon.

This blog post was originally published on Wikimedia’s community blog, Diff, by Satdeep Gill (WMF) and Dr Adi Keinan-Schoonbaert (Digital Curator for Asian and African Collections, British Library)

The British Library has joined hands with the Wikimedia Foundation to support the Wikisource Loves Manuscripts (WiLMa) project, sharing 76 Javanese manuscripts, including what is probably the largest Javanese manuscript in the world, digitised as part of the Yogyakarta Digitisation Project. The manuscripts, which are now held in the British Library, were taken from the Kraton (Palace) of Yogyakarta following a British attack in June 1812. The British Library’s digitisation project was funded by Mr. S P Lohia and included conservation, photography, quality assurance and publication on the Library’s Digitised Manuscripts website, and the presentation of complete sets of digital images to the Governor of Yogyakarta Sri Sultan Hamengkubuwono X, the National Library of Indonesia, and the Library and Archives Board of Yogyakarta.

3D model of Menak Amir Hamza (British Library Add MS 12309), probably the largest Javanese manuscript in the world

For the WiLMa project, the scanned images, representing more than 30,000 pages, were merged into pdfs and uploaded to Wikimedia Commons by Ilham Nurwansah, Wikimedian-in-Residence at PPIM and User:Bennylin from the Indonesian community. The manuscripts are now available on Wikimedia Commons in the Category:British Library manuscripts from Yogyakarta Digitisation Project.

“Never before has a library of Javanese manuscripts of such importance been made available to the internet, especially for easy access to the almost 100 million Javanese people worldwide.”

User:Bennylin said about the British Library donation

As a global movement, Wikimedia is able to connect the Library with communities of origin, who can use the digitised manuscripts to revitalise their language online. As such, we have a history of collaboration with the Wikimedia community, hosting Wikimedians-in-Residence and working with the Wikisource community. In 2021, we collaborated with the West Bengal Wikimedians User Group to organise two Wikisource competitions (in Spring and Autumn). Forty rare Bengali books, digitised as a part of the Two Centuries of Indian Print project, were made available on Wikisource. The Bengali Wikisource community has corrected more than 5,000 pages of text, which were OCRed as part of the project.

“As part of our global engagement with Wikimedia communities, we were thrilled to engage in a partnership with the Bengali Wikisource community for the proofreading of rare and unique books digitised as part of the Two Centuries of Indian Print project. We extend our gratitude towards the community’s unwavering commitment and the enthusiasm of its members, which have greatly enhanced the accessibility of these historic gems for readers and researchers.”

Dr Adi Keinan-Schoonbaert, Digital Curator, British Library

The developing Javanese Wikisource community has already started using the newly digitised Javanese manuscripts in their project, and has plans ranging from transliteration and translation, to recording the content being sung, as originally intended. (Recording of Ki Sujarwo Joko Prehatin, singing (menembang) the texts of Javanese manuscripts, at the British Library, 12 March 2019; recording by Mariska Adamson).

Screenshot of a Javanese manuscript being used for training an HTR model using Transkribus

The Library’s collaboration with the Javanese community started earlier this year, when the Wikisource community included three manuscripts from the Library’s Henry D. Ginsburg Legacy Digitisation Projects in the list of focus texts for a Wikisource competition. Parts of these three long manuscripts were proofread by the community during the competition and now they are being used to create a Handwritten Text Recognition (HTR) model for the Javanese script using Transkribus, as part of our ongoing WiLMa initiative.

Stay tuned for further updates about WiLMa Learning Partners Network!