Keynote presentation at the Deakin University Faculty of Arts and Education HDR Summer School, Geelong, 24 February 2017.

Cite as: Tim Sherratt, ‘The Practice of Play’, presented at Deakin University Faculty of Arts and Education HDR Summer School, Geelong, 24 February 2017, online at <https://doi.org/10.6084/m9.figshare.4696258>.

I’m a historian. But in the past decade the nature of my research has changed quite profoundly. Instead of heading off to the archives, taking lots of notes, and writing up a book or an article, I now make things. Generally these things are online, and open to the public. I make things for people to use, to explore, to play, and to ponder.

I started down this track before I realised there was a name for what I do – practice-led research. The things that I make even have their own acronym – they’re NTROs, or Non Traditional Research Outputs.

But practice-led research is not just about making things. New knowledge is generated through cycles of creation and reflection. My aim in making is not to follow a blueprint, or check off a list of requirements, but to end up asking ‘What is this thing?’, ‘What does it do?’, ‘How does it do it?’.

In the past, I’ve tended to talk about my research practice as playing with data. I think there’s an important argument to be made for the role of play in research, particularly when confronted with large cultural datasets. But ‘play’ doesn’t quite capture what I do, nor does it look very convincing in a research proposal. So what do I really do?

Let’s play a game.

Headline Roulette is a very simple game. Presented with the title of a digitised newspaper article drawn at random from Trove’s collection of more than 200 million you are challenged to guess the year in which the article was published. Sounds easy, but you only get ten guesses. It’s sort of like a cross between hangman and The Price is Right.

Despite its simplicity, I’ve known it to unleash the competitive instincts of a workshop full of historians. But for me, Headline Roulette is important because it provides an example of what becomes possible once we make cultural heritage collections available online. Our interactions are no longer limited to conventional modes of viewing or reading – we can play, and we can build.

I made the first version of Headline Roulette back in 2010. It was a game, but it was also an argument about access and possibilities.

Perhaps we should first take a step back. Who’s used Trove?

Trove is a fundamental part of Australia’s research infrastructure – and not just for those of us in the humanities or social sciences. Trove is a lot more than digitised newspapers, but access to more than 150 years worth of digitised newspapers has profoundly changed historical practice.

I say this not just because you have been spared the pain and suffering wrought by microfilm readers upon a generation of historians, but because the meaning of access itself has changed. Headline Roulette is just one simple and silly example of how once cultural heritage resources are in digital form we can use them differently. We can see them differently.

Imagine your search in Trove’s newspapers zone returns 10,000 or 100,000 results. How do you make sense of that? How do you get an understanding of the whole, when all you see is page after page of search results?

QueryPic extracts data from Trove to visualise your search as a single chart – showing you the number of articles per year that match your query. You can even compare the occurrence of particular words or phrases.

But that’s only the beginning, because once you think about web resources as data rather than just another type of publication you can aggregate and analyse – you can look for big, dramatic pictures as well as tiny, fragile fragments.

Trove Harvester is tool that delivers historical newspaper articles in bulk – thousands, even millions of articles saved to your computer for offline exploration.

What might you do with a million newspaper articles?

Research using digital resources like Trove is not constrained to the window of your web browser. You can ask new types of questions.

But back in 2010–11 when I created the first versions of Headline Roulette, QueryPic and the Trove Harvester there was no easy way of getting data out of Trove. The thing is, web pages are good for delivering data to human beings, but not so good for computers. Computers are actually pretty dumb, and you need to be quite explicit in packaging up data for them. Nowadays Trove has a thing called an API (an Application Programming Interface) which delivers data in a carefully structured format that even computers can understand. You can use APIs to harvest data, or to build new tools or interfaces. APIs are cool.

Without an API, the first versions of my tools had to turn human-readable web pages into computer-readable data – a process known as screen scraping. They were, therefore, not only useful or interesting applications in their own right, they were arguments about why things like APIs matter. Why web pages aren’t enough. Why researchers need access to data.

These are arguments we’re still making. Next week I’m heading to a workshop in California where we’ll be discussing how libraries and other cultural institutions can deliver their data in ways that support new forms of research.

But we don’t have to wait. By screen scraping web pages, by reverse engineering online databases, we can continue to develop the argument for access by extracting, sharing, and using data.

What could you do with 70gb of digitised surveillance files from the Australian Security Intelligence Organisation (ASIO)? I’f you’d like a copy I have them here on a USB drive.

Don’t worry – we’re not about to be raided by the security services. These are all files that have been carefully examined and released to the public through the National Archives of Australia. You can find them by searching the Archives’ online database – RecordSearch.

Who’s used RecordSearch? It’s not the most friendly system, but the collection it documents, and the metadata it provides, is rich and wondrous. I’ve spent a lot of time trying to get useful data out of RecordSearch – not just ASIO files, also records documenting the administration of the White Australia Policy, as well as higher-level data aimed at building my understanding of how the Archives, and its descriptive systems, actually work.

It is painful and frustrating work. But, I would argue, it is research. Terms like ‘data mining’ and ‘text mining’ fly around all the time, making it seem as if the the accumulation of data is a mechanical process – as if we’re just digging it up. But the practice of screen scraping, or of liberating data from any cultural heritage source, is not simply extractive – it’s iterative and interpretative. It’s a process through which you begin to understand how the data is organised, what its limits and assumptions are, what its history is. What it means. We’re not just taking things out, we’re putting them back.

Frederick Gibbs and Trevor Owens argue that historical data need not be deployed solely as statistical evidence. ‘It can also help’, they suggest, ‘with discovering and framing research questions’ – questions, not answers; interpretation not calculation. Gibbs and Owens describe an ‘iterative interaction with data as part of the hermeneutic process’.

For me, RecordSearch is like an archaeological site. Excavating data from it involves digging through layers of technology, institutional history, and descriptive practice to try and understand why we have what we have.

Those of you undertaking projects using the collections of the National Archives will almost certainly come across the process of ‘access examination’. Under the Archives Act, government records more than twenty years old are expected to be opened to the public. However, the act also defines a number of exceptions to this rule – for example, records that endanger national security or infringe an individual’s privacy can be completely, or partially, withheld from scrutiny. The process of assessing records against this set of exemptions is called ‘access examination’.

The vast majority of records are opened without problem – they are, after all, more than 20 years old. But a significant number are not. While you can’t use these records, RecordSearch does provide some information about them. So I decided to see what we couldn’t see.

In January 2016 I fired up my screen scraper and harvested details of all the files in RecordSearch that have the access status of ‘closed’ – there were 14,370 of these files that had been through the process of access examination and withheld from public view. I then created my own interface that lets you explore this data from a variety of angles – such as the reasons why files were closed, when decisions were made about them, how old they are, and which government agencies created them.

It is perhaps the most frustrating search interface ever devised, given that you’re not allowed to see any of the files you find.

Those of you currently planning research projects might be interested to know where most of these files come from. It’s not defence or the intelligence agencies, but what is now the Department of Foreign Affairs and Trade (DFAT) – in January 2016, there were 1,747 closed files from just one DFAT series. But if you dig deeper you see that most of these files aren’t withheld for one of the reasons defined by the Archives Act, they are described as ‘closed pending advice’. The National Archives is still waiting to hear back from DFAT about them. Using my interface you can see that there were 54 files in this series where the Archives has been waiting for more than five years. So if you’re embarking on a project using the National Archives, make sure you get your access examination requests in early. Just in case.

My aim in extracting and sharing this data is to better understand access examination itself as a historical process. It’s work that enables us to ask different types of questions, but it also makes a change in the process itself. My interface is public, offering a critical commentary on the ‘official’ system. As a result of my research, the Archives has made changes to the way it describes closed files. It’s both research and intervention, history and hack.

‘Hack’ has a number of definitions, both positive and negative. Mark Olsen describes the ‘hacker ethos’ as:

‘a way of feeling your way forward through trial and error, up to and perhaps beyond the limits of your expertise, in order to make something, perhaps even something new. It is provisional, sometimes ludic, and involves a willingness to transgress boundaries, to practice where you don’t belong… Whether eloquent or a kludge, a hack gets things done.’1

Olsen explores what hacking means in the context of the humanities, arguing not only that hacking has a legitimate place in humanities practice, but that the humanities itself needs to be hacked to foster the development of new skills and literacies.

At this point you’re probably thinking, ‘But I don’t do any of this wacky digital stuff, what has this got to do with me?’

Who’s heard of filter bubbles, or search personalisation? Who’s read one of the many reports recently about the way computer algorithms are shaping our online experience? Olsen argues for a humanities practice that equips us to wrestle with complex techno-social systems.

And we’re not just talking about Google.

Last year Matthew Reidsma published an analysis of algorithmic bias in library discovery systems. He hacked a common commercial library product to show some of the biases underlying its recommendations system. The interfaces we use to access information are never neutral. The databases we search are products of selection and exclusion. Hacking enables us to interact with these systems as critics, and not just consumers.

Using the Trove API you can create a chart showing the number of digitised newspaper articles available per year from 1803 onwards. If you do this, you’ll notice two significant features. First, there is a dramatic drop-off in the number of articles after 1954. This is the ‘copyright cliff of death’. Few things are certain in our overly-complex copyright system, but 1954 provides a practical cut-off point. History stops in 1954.

You’ll also notice a substantial peak in the number of articles around 1914. Why might this be? Did something significant happen in 1914?

In fact, it’s all about money. In the lead up to the centenary of WWI it was decided to focus limited digitisation resources on newspapers from the WWI period. It was a perfectly reasonable decision, but the consequences are effectively invisible to any user of the web interface. You don’t know what you’re searching.

The power of Google encourages us to put a lot of faith in search interfaces. We trust that they will just work. And if we can’t find what we’re looking for, we often assume that it doesn’t exist.

Hansard, the recorded proceedings of the Australian parliament from 1901 can be searched using the ParlInfo database on the Australian Parliament House website. Perhaps you’ve used it – it’s a wonderfully rich resource. Powering the search results are a series of well-structured XML files, one for each sitting day, that identify individual debates and speeches.

Last year I reverse-engineered ParlInfo and harvested all those XML files. I thought they’d provide a great dataset for exploring changes in political speech, and so I created a repository containing all the files for the House of Representatives and the Senate from 1901 to 1980. Feel free to download and play.

But in the process of harvesting the files I noticed that some of the XML files were empty. After a bit more analysis I realised that about 100 sitting days were missing – they didn’t show up in search results on ParlInfo.

The ‘missing’ days were concentrated in the Senate between 1910 and 1920. So anyone relying on ParlInfo to research the WWI period would have missed significant amounts of content. This ‘black hole’ was effectively invisible to any user of the web interface. It was only though hacking that its shape and extent was revealed.

Fortunately staff at the Parliamentary Library have investigated and fixed the problem. But it’s a good example of why we should, as researchers, start from the assumption that search interfaces lie. Processes of selection and description shape the ‘reality’ of online collections. We then explore them through complex technological systems that appear comprehensive, even when they are not. You can’t find what’s not there. Online collections hide as much as they reveal.

Of course this is true of all historical sources. We are trained to analyse both context and content, to make judgements about authenticity and accuracy. These same skills need to applied to digital resources, to data. Indeed, Gibbs and Owen argue that ‘historians must treat data as text, which needs to be approached from multiple points of view and as openly as possible’. But how do we find multiple points of view when interfaces construct our experiences and limit our perspectives. How do we open data to new possibilities? How do we see data differently?

No doubt you’ve been encouraged to find a way of expressing your research questions succinctly, in a way that communicates with a non-specialist audience – yes, I mean the dreaded elevator pitch. You’re not the only one.

I’ve landed back in academia after a number of years working in cultural heritage institutions, and pursuing my own research interests with the support of the international digital humanities community.

Believe me when I say, Twitter changed my life. There I was, hacking away on cultural heritage data without any real assistance or encouragement, when I discovered, via Twitter, that there were people out there like me. Many of these people are now my friends, and I’ve been lucky to travel around the world to meet and work with them.

But coming back to academia I’ve found that my collection of projects, tools, experiments, and obsessions was not quite enough – my research needs a ‘narrative’.

So, like you, I’ve had to think about why I do what I do. What motivates my research? What matters?

For me it comes back to the nature of this thing we call ‘access’. Cultural heritage organisations talk about ‘access’ all the time, particularly in relation to online collections. But what does it actually mean? I want to overturn our assumptions about access – exploring it not as a process of opening things up, but as a system of controls and limits. It’s not a state of being, it’s a struggle for meaning and power.

My methodology, and I think I can call it that, is the multiplication of contexts. Context is, of course, critical to cultural heritage collections – it enables us to locate them within history and culture, to analyse their authenticity, to mobilise their value as evidence. But the descriptive systems we use to manage and explore collections represent only a privileged subset of possible contexts.

Now I’m still figuring this out, but I think what my work does is that it removes collections from these highly-controlled systems and lets them loose in a variety of new contexts. This allows unexpected features, or new uses, to emerge – we see them differently, and in that moment, the nature of access shifts, however slightly. It’s those moments I’m trying to catch and observe.

If you’ve ever tried to use Hansard through the ParlInfo database you’ll realise that it’s just really difficult to read. You’re presented with a series of nested fragments, so it’s hard to get a sense of the context and flow of the day’s proceedings. Having downloaded all those XML files, I thought I’d have a go at presenting Hansard in a form that privileged reading over search.

So I created Historic Hansard – dedicated to lovers of political speech. It does nothing very fancy, but I think it does it pretty well.

In the end, however, Hansard is still just text. What’s lost in the documentation process is the performance – the theatre of parliament. But not completely. As well as formal speeches, many interjections have been recorded and preserved.

A few weeks ago I extracted all those interjections from 1901 to 1980, about a million of them, and saved them to a new database. As I fiddled with different presentation methods, I started to see them as something akin to tweets – quick, pithy, and pointed. What would happen, I wondered, if we reimagined interjections from a century ago in an age of social media.

Like many of my projects, this whatever it is took me a couple of days to build. No research grants were harmed in its creation, no committees were needlessly formed. This is not because I’m a whizz-bang coder – I’m certainly not. It has to do with the nature of this work – it’s rapid, experimental, and sometimes even ephemeral. I don’t design websites, I make interventions – things that are not only of the world, but in the world. They do something.

Stephen Ramsay explores the hermeneutical possibilities of screwing around with technology and texts. The ‘screwmeneutical imperative’ he suggests is based on the fact that:

’a writerly, anarchic text… is more useful than the readerly, institutional text. Useful and practical not in spite of its anarchic nature, but as a natural consequence of the speed and scale that inhere in all anarchic systems.

Digital technologies give us the opportunity to play with scale and speed. We can manipulate millions of newspaper articles, and we can build a new version of Hansard in a weekend. But this shift also applies to the way we communicate. Instead of waiting months or years for an article to appear in print, we can post it on a blog, or in a digital repository. It is fundamental to the work that I do that it is shared, it is public by default – not just the results, but the code, the data, the process, and yes the licensing. Access is not just what we take, it’s what we do.

The multiplication of contexts has some interesting precedents as a research methodology. In the literary world the Oulipo movement sought to play with the contraints of composition. Lisa Samuels and Jerome McGann suggested that the deliberate misreading of a text, what they termed ‘deformance’, could yield critical insights. More recently, Mark Sample has argued for a ‘deformed humanities’ where we learn about things by breaking them.

In history we have the counterfactual – a creative reimagining of a past that never was, aimed at revealing perspectives and possibilities too quickly closed and forgotten. As Sean Scalmer argues, ‘counterfactuals are fun’:

‘Conventions can be disregarded, or even mocked. Worlds might be remade, the tyrannical overthrown, and the humble elevated. New orders can be imagined.’2

But counterfactuals are not fiction. They work best when they sail close to an accepted version of the past; when they play with the constraints of documentary evidence rather than just ignore them. Just because an approach is playful, it doesn’t mean that there are no rules. As Ian Bogost has recently argued, the fun of play is ‘not doing what we want, but doing what we can with what is given’.3 Play is an investigation of limits.

While some of the ASIO files held by the National Archives are closed to the public, most are ‘open with exception’. This means that sensitive parts of the files have been removed. Whole pages can be withheld, or sections of text blacked out – a process known as redaction.

A redaction is, by definition, an absence of information, and yet the frequency, density, and placement of redactions across a large collection of documents could conceivably tell us something interesting. So last year I wrote a kludgy computer vision script that found and extracted redactions from digitised ASIO files. I now have a collection of 250,000 redactions which I’ve shared on Figshare – grab a copy now!

I’m continuing to explore the possibilities of these redactions as data points. But there was also something visually interesting about the redactions, particularly when they were assembled on masse.

Here you can browse all 250,000 redactions. But that’s not all, you can also use them as entry points to the documents they were intended to obscure.

Contexts here have been reversed, the files have been turned inside out – the limits remain, indeed the scale of redaction is emphasised, and yet within these limits, perhaps even because of these limits, we can experience the files quite differently. We are no longer simply the subjects of state surveillance, we can reverse the gaze, inspect the process, and ask new questions.

The manipulation of contexts is not mere invention. The limits of access offer both meaning and rules. We have skin in this game, its outcomes matter, what is at stake is our ability to see, and be seen, within the cultural record. Access changes who we can imagine ourselves to be.

In the first half of the twentieth century, if you were deemed not to be ‘white’ and wanted to travel overseas from your home here in Australia, you had to carry special documents. Without them, you’d probably be stopped from returning – from coming home. This was ‘extreme vetting’ White Australia style.

Many thousands of these documents are now held by the National Archives of Australia. In 2011, I used my screen scraper to harvest about 12,000 images like this from RecordSearch. I then ran them through a facial detection script and created The Real Face of White Australia.

There are about 7,000 faces in this seemingly endless scrolling wall. And that’s from just a small sample of the White Australia records. It’s powerful, compelling and discomfiting. But the power comes not from any technical magic, but from the faces themselves – from what we feel when meet their gaze. Once again the records have been turned inside out – instead of seeing files, metadata, or a list of search results, we see the people inside.

Play can be serious. It can make you feel things you don’t expect. It can challenge your certainties and take you to the limits of what you know.

That sounds a lot like research to me.

1. M. J. Olson, ‘Hacking the humanities: Twenty-first-Century literacies and the “becoming other”of the humanities’, in Eleonora Belfiore and Anna Upchurch (eds), Humanities in the Twenty-first Century: Beyond utility and markets, Palgrave Macmillan, 2013, pp. 237–250.
2. Sean Scalmer, ‘Introduction’, in Stuart Macintyre and Sean Scalmer (eds), What if? Australian history as it might have been, Melbourne University Press, Melbourne, 2006, pp. 1–11.
3. Ian Bogost, Play Anything: The Pleasure of Limits, the Uses of Boredom, and the Secret of Games,Basic Books, New York, 2016, p. 236.

A few years ago it seemed fashionable for cultural institutions to create a ‘My [insert institution name]’ space on their website where visitors could create their own online exhibits from collection materials. It bothered me at the time because it seemed to be a case of creating silos within silos. What could people do with their collections once they’d assembled them?

I was reminded of this recently as I undertook my Christmas-break mini project to think about Trove integration with Zotero. Some years ago I created a Zotero translator for the Trove newspaper zone, but I’d much rather we just exposed metadata within Trove pages that Zotero (and other tools like Hypothes.is and Mendeley) could pick up without any special code. More about that soon…

However. embedded metadata only addresses part of the problem — there’s also questions around tools and workflows. Trove includes a number of simple tools that enable users to annotate and organise resources — tags, comments, and lists. Tags and comments… well you know what they are. Lists are just collections of resources, and like tags, they can be public or private.

They may not be terribly exciting tools, but they are very heavily used. More than 2 million tags have been added by Trove users, but it’s lists that have shown the most growth in recent times. There are currently more than 47,000 lists and 30,000 of those are public. That’s a pretty impressive exercise in collection building. What are the lists about? A few months ago I harvested the titles of all public lists and threw them into Voyant for a quick word frequency check.

Given what we know about Trove users, it wasn’t surprising to see that many of the lists related to family history, but there are a few wonderful oddities buried in there as well. I love to be surprised by the passions of Trove users.

I suspect these tools are popular because they’re simple and open-ended. There are few constraints on what you can use for a tag or add to a list. Following some threads about game design recently I came upon a discussion of ‘underspecified’ tools — ‘the use of which you can never fully predict’. By underspecifying we leave open possibilities for innovation, experimentation and play. It seems like a pretty good design approach for the cultural heritage sector.

But wait a minute, you might be wondering, what sort of Trove Manager magic did I have to weave in order to extract those thousands of list titles? None, none at all. You could do exactly the same thing.

I’ve been talking a lot in recent months about Trove as a platform rather than a website — something to build on. One of our main construction tools is, of course, the Trove API. I suppose a good cultural heritage API is also underspecified — focused enough to be useful, but fuzzy enough to encourage a bit of screwing around. What you may not know about the Trove API is that as well as giving you access to around 300 million resources, it lets you access user comments, tags and lists.

I’m looking forward to researchers using the API to explore the various modes of meaning-making that occur around resources in Trove. But right now one thing it offers is portability — the collections people make can be moved. And that brings us back to Zotero.

Why should a Trove user have to decide up front whether they want to use Zotero or create a Trove list? Pursuing Europeana’s exciting vision of putting collections in our workflows we need to recognise that workflows change, projects grow, and new uses emerge. We should support and encourage this by making it as easy as possible for people to move their stuff around.

So of course I had to build something.

My Christmas project has resulted in some Python code that lets you export a Trove list or tag to a Zotero collection — API to API. Again it’s a simple idea, but I think it opens up some interesting possibilities for things like collaborative tagging projects — with a few lines of code hundreds of tagged items could be saved to Zotero for further organisation or annotation.

Along the way I ended up starting a more general Trove-Python library — it’s very incomplete, but it might be useful to someone. It’s all on GitHub — shared, as usual, not because I think the code is very good, but because I think it’s really important to share examples of what’s possible. Hopefully someone will find a slender spark of inspiration in my crappy code and build something better. Needless to say, this isn’t an official Trove product.

So what do you do if you want to export a list or tag?

First of all get yourself Python 2.7 and set up a virtualenv where you can play without messing anything up. Then install my library…

git clone https://github.com/Trove-Toolshed/trove-python.git
cd trove-python
python setup.py install

You’ll also need to install PyZotero. Once that’s done you can fire up Python and export a list from the command line like this…

from pyzotero import zotero
from trove_python.trove_core import trove
from trove_python.trove_zotero import export

zotero_api = zotero.Zotero('[Your Zotero user id]', 'user', '[Your Zotero API key]')
trove_api = trove.Trove('[Your Trove API key]')

export.export_list(list_id='[Your Trovelist id]', zotero_api=zotero_api, trove_api=trove_api)

Obviously you’ll also need to get yourself a Trove API key, and a Zotero key for your user account.

Exporting items with a particular tag is just as easy…

from pyzotero import zotero
from trove_python.trove_core import trove
from trove_python.trove_zotero import export

zotero_api = zotero.Zotero('[Your Zotero user id]', 'user', '[Your Zotero API key]')
trove_api = trove.Trove(['Your Trove API key'])
exporter = export.TagExporter(trove_api, zotero_api)

exporter.export('[Your tag]')

What do you end up with? Here’s my test list on Trove and the resulting Zotero collection. Here’s a set of resources tagged with ‘inigo’, and here’s the collection I created from them. You’ll notice that I added a few little extras, like attaching pdf copies where they’re available.

Sorry, no GUIs, no support, and not much documentation. Just a bit of rough code and some ideas to play with.

This is the full version of a paper I presented at the National Digital Forum, 30 November 2011.

In 1901, one of the first acts of the Commonwealth of Australia was to create a system of exclusion and control designed to keep the newly-formed nation ‘white’. But White Australia was always a myth. As well as the Indigenous population, there were already many thousands of people classified as ‘non-white‘ living in Australia — most were Chinese, but there were also Japanese, Indians, Syrians and Indonesians.

Here are some of them…

The administration of what became known as the White Australia Policy created a huge volume of records, much of which is still preserved within the National Archives of Australia. These photographs are attached to certificates that non-white residents needed to get back into the country if they decided to travel overseas. There are thousands upon thousands of these certificates in the Archives. Thousands of certificates representing thousands of lives — all monitored and controlled.

But is is too easy to see these people as the powerless victims of a repressive system. There were many acts of resistance. Some argued against the need to be identified ‘just like a criminal’. Others exercised control over their representation, submitting formal studio portraits instead of mug shots.

Most commonly and most powerfully, people resisted the policy simply by going ahead and living rich and productive lives.

My partner, Kate Bagnall, is helping to rewrite Australian-Chinese history by overthrowing the stereotype of the culturally isolated Chinese man living a lonely, meagre existence surrounded by gambling and opium dens. By mining the available records, by reading against the grain of contemporary reports and by working with family historians, Kate is documenting their intimate lives — their wives, their lovers, their families and descendants — the sorts of relationships that sent a shudder through the edifice of White Australia. Power can be reclaimed in many subtle and subversive ways.

‘The real face of White Australia’ is an experiment. It uses facial detection to technology to find and extract the photographs from digital copies of the original certificates made available through the National Archives of Australia’s collection database. The photographs you see here come from just one series, ST84/1. There’s no API to the collection so I reverse-engineered the web interface to create a script that would harvest the item metadata and download copies of all the digitised images. There are 2,756 files in this series. On the day I harvested the metadata, 347 of those files had been digitised, comprising 12,502 images. It took a few hours, but I just ran my script and soon I had a copy of all of this in my local database.

Then came the exciting part. Using a facial detection script I found through Google and an open source computer vision library, I started experimenting with ways of extracting the photos. After a few tweaks I had something that worked pretty well, so I pointed my aging laptop at the 12,502 images and watched anxiously as the CPU temperature rose and rose. It took a few emergency cooling measures, but the laptop survived and I had a folder containing 11,170 cropped images. About a third of these weren’t actually faces, but it was easy to manually remove the false positives, leaving 7,247 photos.

These photos. These people.

With my database fully primed and loaded it was just a matter of creating a simple web interface using Django for the backend and Isotope (a jQuery plugin) at the front. Both are open source projects. All together, from idea to interface, it took a bit more than a weekend to create, and most of that was waiting for the harvesting and facial detection scripts to complete. It would be silly to say it was easy, but I would say that it wasn’t hard.

What we ended up with was a new way of seeing and understanding the records — not as the remnants of bureaucratic processes, but as windows onto the lives of people. All the faces are linked to copies of the original certificates and back to the collection database of the National Archives. So this is also a finding aid. A finding aid that brings the people to the front.

According to Margaret Hedstrom the archival interface ‘is a site where power is negotiated and exercised’.1 Whether in a reading room or online, finding aids or collection databases are ‘neither neutral nor transparent’, but the product of ‘conscious design decisions’. We would like to think that this interface gives some power back to the people within the records. Their photographs challenge us to do something, to think something, to feel something. We cannot escape their discomfiting gaze.

But this interface represents another subtle shift in power. We could create it without any explicit assistance or involvement by the National Archives itself. Simply by putting part of the collection online, they provided us with the opportunity to develop a resource that both extends and critiques the existing collection database. Interfaces to cultural heritage collections are no longer controlled solely by cultural heritage institutions.

It’s these two aspects of the power of interfaces that I want to focus on today.

There are a growing number of examples where the records created by repressive or discriminatory regimes have, in Eric Ketelaar’s words, ‘become instruments of empowerment and liberation, salvation and freedom’.2 Nazi records of assets confiscated during the Holocaust have been used to inform processes of restitution and reparation. Government records have helped members of Australia’s Stolen Generations trace family members. Descendants of inmates incarcerated by American colonial authorities in what was the world’s largest leprosy colony in the Philippines, have embraced the administrative record as an affirmation of their own heritage and survival.3 Records can find new meanings. Power can be reclaimed.

Technology can help. Tim Hitchcock has described how something as simple as keyword searching can turn archives on their heads. Recordkeeping systems tend to reflect the structures and power relations of the organisations that create them. The ‘hierarchical and institutional nature of most archives’, Hitchcock argues, ‘contains an ideological component which is sucked in with every dust-filled breath’.4 But digitisation and keyword searching free us from having to follow the well-worn paths of institutional power. We can find people and follow their lives against the flow of bureaucratic convenience. We can gain a wholly new perspective on the workings of society. ‘What changes’, Hitchcock asks, ‘when we examine the world through the collected fragments of knowledge that we can recover about a single person, reorganised as a biographical narrative, rather than as part of an archival system?’5

Projects such as Unknown no longer may help us answer that question.

It’s aiming to extract the names and biographical details of slaves from the 8 million manuscript documents held by the Virginia Historical Society. The documents include court records, receipts, wills and inventories. Here is a page from the ‘Inventory of Negroes at Berry Plain Plantation, King George County, Virginia’ for 1855, listing names, occupations and valuations.

Tim Hitchcock is one of the directors of London Lives a project that similarly seeks to find the people in 240,000 manuscript pages documenting the lives of plebeian Londoners in the 17th century.

More than three million names have already been extracted from the records of courts, workhouses, hospitals and other institutions. Work is continuing to link these names together, to merge these various shards of identity and piece together the experiences of London’s poorest inhabitants.

Remember me from the US Holocaust Memorial Museum is working with photographs taken by relief agencies in the aftermath of World War Two. The photographs are of displaced children who survived the Holocaust but were separated from families. What happened to them? The project is seeking public help to identify and trace the children.

These are all projects about finding people. Finding the oppressed, the vulnerable, the displaced, the marginalized and the poor and giving them their place in history. This is what Kate and I hope to do with Invisible Australians, the broader project of which our faces experiment is part.

‘Invisible Australians’ aims to extract more than just photographs. We want to record and aggregate the biographical data contained within the records of the White Australia Policy — to extract the data and rebuild identities.

But we want to do more, we want to link these identities up with with other records, with the research of family and local historians, with cemetery registers and family trees, with newspaper articles and databases we don’t even know about yet. We want to find people, families and communities.

It’s ridiculously ambitious and totally unfunded. But it is possible.

The most exciting part of online technology is the power it gives to people to pursue their passions. As with the faces, we don’t need the help of the National Archives. We need the records to be digitized, but that’s happening anyway and we can afford to be patient. Most of the tools we need already exist, and are free. In the past 12 months, for example, there have been a number of open source tools released for crowd-sourced transcription of manuscript records.

People with passions, people with dreams, people who are just annoyed and impatient, don’t have to wait for cultural institutions to create exactly what they need. They can take what’s on offer and change it.

Interfaces can be modified. It is amazingly easy to write a script that will change the way a web page looks and behaves in your browser. I was frustrated by the standard interface to digitized files in the National Archives of Australia’s Recordsearch database — so I changed it.

Not only did make it look a bit nicer, I added new functions. My script lets you print a whole file or a range of pages and display the entire contents of the file on a pretty cool 3d wall.

I’ve shared this script, and a few other Recordsearch enhancements. Anyone can install them with a click and use them.

Interfaces are sites of power and we can claim some of that power for ourselves. Online technologies not only free us from the having to brave the physical intimidation of the reading room, they free us up to engage with the records in new ways. The archivist-on-duty would probably not be pleased if I pulled out some scissors and started snipping photos out of certificates. Or if I pulled a file apart and pasted it’s contents on the wall. But online we are free to experiment.

The power of cultural heritage organisations is perhaps expressed most forcefully in their ability to control the arrangement and description of their collections. ‘Every representation, every model of description, is biased’, note Verne Harris and Wendy Duff, ‘because it reflects a particular world-view and is constructed to meet specific purposes’.6 Archives, libraries and museums are already starting to share this power, by allowing tagging, or seeking public assistance with description through crowd sourcing projects. But most of the these activities still happen within spaces created and curated by the institutions themselves. Our cathedrals of culture might be opening their doors and inviting the public to participate in their ceremonies, but that doesn’t make them bazaars. The architecture stills speaks of authority.

In any case, people already have a space where they can explore and enrich collections — it’s called the internet.

It would be great to see cultural institutions doing more to watch, understand and support what people are doing with collections in their own spaces — following them as they pursue their passions, rather than thinking of ways to motivate them.

A quick example… You might have heard of Zotero, it’s an open source project that lets you capture, annotate and organize your research materials.

One cool thing about Zotero is that you can build and contribute little screen scrapers, called translators, that let Zotero extract structured data from any old collection database. You might not be surprised to learn that I’ve created a translator for Recordsearch. Another cool thing about Zotero is that you can share the stuff that you collect in public groups.

Put those two cool things together and what do you have? Well to me they spell out user generated finding aids — parallel collection databases created by researchers simply pursuing their own passions.

Linked Open Data greatly increases opportunities for collection description to leak into the wider web. If objects and documents are identified with a unique URL, then anyone can can make and publish statements about them in machine-readable form. These statements can then be aggregated and explored. Initiatives such as the Open Annotation Collaboration will hasten the development of these shared descriptive and interpretative layers around our cultural collections.

And of course all this descriptive and interpretative work can be harvested back to enhance existing collection databases. We could start doing it now — though I will spare you today my rant about the possibilities of mining footnotes.

As well as exploring the possibilities of user-generated content, cultural institutions are starting to open up their collection data for re-use. APIs are great (though Linked Open Data is better), and New Zealand is lucky to have an organisation like DigitalNZ which just gets it. People can and will make cool things with your stuff.

But again, we don’t have to wait for everything to be delivered in a convenient, machine-readable form. If it’s on the web anybody can scrape, harvest and experiment.

You probably all know about the National Library of Australia’s newspaper digitisation project — it’s building a magnificent resource. But I wanted to do more than just find articles. I wanted to explore and analyze their content on a large scale. So I built a screen scraper to extract structured data from search results, and then used the scraper to  power a series of tools. I have a harvester that lets you download an entire results set — hundreds or thousands of articles — with metadata neatly packaged for further analysis.

Or what about a script that graphs the occurrence of search terms over time, and allows you to ask questions like When did the Great War become the First World War?.

In the end I got a bit carried away and built my own public API to the Trove newspaper database.

I think it’s important to note that the tools I developed were guided by the types of questions I wanted to ask. While we should welcome APIs and celebrate their possibilities, we should also remain critical. APIs are interfaces, they too embed power relations. Every API has an argument. What questions do they let us ask? What questions do they prevent us from asking?

Even as we move from the age of lumbering, slow-witted data silos into the rapidly-evolving realms of Linked Open Data, we have to constantly question the models we make of the world. Ontologies and vocabularies are culturally determined and historically specific. Yes, they too are interfaces, complete with their own distributions of power and authority. But we can revisit and change them. And we can relate our new models to our old models, capturing complex, long-term shifts in the way we think about the world. That’s incredibly exciting.

All of this hacking, harvesting, questioning, enriching and meaning-making makes me think about the possibilities of grassroots leadership. Online technologies enable people to take cultural institutions into unexpected realms. They can build their own interfaces, ask their own questions, determine their own needs — they can point the way instead of simply waiting to be served.

You might wonder what the National Library of Australia thinks of my various scrapers and harvesters. I can’t speak for them, but I can say that they’ve awarded me a fellowship to explore further the possibilities of text-mining in their newspaper database.

The idea of grassroots leadership brings me back to the title of this talk — ‘It’s all about the stuff’. It seems to me that we tend to model the interactions between cultural institutions and the public as transactions. The public are ‘clients’, ‘patrons’, ‘users’ or ‘visitors’. But the sorts of things I’ve been talking about today give us a chance to put the collections themselves squarely at the centre of our thoughts and actions. Instead of concentrating on the relationship between the institution and the public, we can can focus on the relationship we both have with the collections.

It’s all about the stuff.

It’s all about the respect and responsibility we both have for our collections.

It’s all about the respect and responsibility we both have for people like this.

1. Margaret Hedstrom, ‘Archives, Memory, and Interfaces with the Past’, Archival Science, vol. 2, 2002, p. 22.
2. Eric Ketelaar, ‘Archival Temples, Archival Prisons: Modes of Power and Protection’, Archival Science, vol. 2, 2002, p. 229.
3. Ricardo L Punzalan, ‘Archives and the Segregation of People with Leprosy in the Philippines’, presented at ICHORA-4, University of Western Australia, 2008, pp. 93-102.
4. Tim Hitchcock, ‘Digital Searching and the Re-formulation of Historical Knowledge’, in Mark Greengrass and Lorna Hughes (eds), The Virtual Representation of the Past, Farnham, UK: Ashgate, 2008, p.87.
5. Tim Hitchcock, p.90.
6. Wendy M. Duff and Verne Harris, ‘Stories and Names: Archival Description as Narrating Records and Constructing Meanings’, Archival Science, vol. 2, 2002, p. 275.

For a project that I’m working on at the National Museum of Australia, I’ve started collecting various sources of date-identified data. Most recently I had a go at extracting historical population data from the Australian Bureau of Statistics.

The data can all be downloaded as .xls files, but they’re not simple, flat spreadsheets – they’re data cubes. As the name suggests, data cubes are organised along a number of dimensions. In the case of the population data it’s year, state and gender.

This means that you can’t just export the data to CSV and suck it into your database – first you’ve got to flatten the cube. No doubt there are other ways to do this, but I just wrote a simple python script. It uses xlrd to read from the spreadsheet, does a bit or reorganisation, then writes the output to a CSV file. The code, for what it’s worth, is available at Bitbucket.

Once I had the CSV file I just imported it into MySQL and used Django and Piston to build a basic API. So if you want to know the population of NSW in 1856, you just go to:

http://wraggelabs.com/api/json/population/nsw/1856/

The number of infant deaths in Tasmania in 1932:

http://wraggelabs.com/api/json/infantdeaths/tas/1932/

The number of female births in Australia in 1959:

http://wraggelabs.com/api/json/births/australia/females/1959/

I’m sure you get the picture. You can change the ‘json’ to ‘xml’ if you’d like another flavour of data.

With an API delivering JSON you can start playing around with all sorts of fun AJAX-y stuff. To demonstrate I built a simple population browser using JQuery. Just drag the slider!