RSS .92| RSS 2.0| ATOM 0.3
  • Home
  • About
  • Team

    Connecting archives with linked geodata – Part I

    October 22nd, 2010

    This is the first half of the talk I gave at FOSS4G 2010 covering the Chalice project and the Unlock services. Part ii to follow shortly….

    My starting talk title, written in a rush, was “Georeferencing archives with Linked Open Geodata” – too many geos; though perhaps they cancel one another out, and just leave *stuff*.

    In one sense this talk is just about place-name text mining. Haven’t we seen all this before? Didn’t Schuyler talk about Gutenkarte (extracting place-names from classical texts and exploring them using a map) in like, 2005, at OSGIS before it was FOSS4G? Didn’t Metacarta build a multi-million business on this stuff and succeed in getting bought out by Nokia? Didn’t Yahoo! do good-enough gazetteer search and place-name text mining with Placemaker? Weren’t *you*, Jo, talking about Linked Data models of place-names and relations between them in 2003? If you’re still talking about this, why do you still expect anyone to listen?

    What’s different now? One word: recursion. Another word: potentiality. Two more words: more people.

    Before i get too distracted, i want to talk about a couple of specific projects that i’m organising.

    One of them is called Chalice, which stands for Connecting Historical Authorities with Linked Data, Contexts, and Entities. Chalice is a text-mining project, using a pipeline of Natural Language Processing and data munging techniques to take some semi-structured text and turn the core of it into data that can be linked to other data.

    The target is a beautiful production called the English Place Name Survey. This is a definitive-as-possible guide to place-names in England, their origins, the names by which things were known, going back through a thousand years of documentary evidence, reflecting at least 1500 years of the movement of people and things around the geography of England. There are 82 volumes of the English Place Name Survey, which started in 1925, and is still being written (and once its finished, new generations of editors will go back to the beginning, and fill in more missing pieces).

    Place-name scholars amaze me. Just by looking at words and thinking about breaking down their meanings, place-name scholars can tell you about drainage patterns, changes in the order of political society, why people were doing what they were doing, where. The evidence contained in place-names helps us cross the gap between the archaeological and the digital.

    So we’re text mining EPNS and publishing the core (the place-name, the date of the source from which the name comes, a reference to the source, references to earlier and later names for “the same place”). But why? Partly because the subject matter, the *stuff*, is so very fascinating. Partly to make other, future historic text mining projects much more successful, to get a better yield of data from text, using the one to make more sense of the other. Partly just to make links to other *stuff*.

    In newer volumes the “major names”, i.e. the contemporary names (or the last documented name for places that have become forgotten) have neat grid references, point-based, thus they come geocoded. The earliest works have no such helpful metadata. But we have the technology; we can infer it. Place-name text mining, as my collaborators at the Language Technology Group in the School of Informatics in Edinburgh would have it, is a two-phase process. First phase is “geo-tagging”, the extraction of the place-names themselves; using techniques that are either rule-based (“glorified regular expressions”) or machine-learning based (“neural networks” for pattern cognition, like spam filters, that need a decent volume of training data).

    Second phase is “geo-resolution”; given a set of place-names and relations between them, figuring out where they are. The assumption is that places cluster together in space similarly as they do in words, and on the whole that works out better than other assumptions. As far as i can see, the state of the research art in Geographic Information Retrieval is still fairly limited to point-based data, projections onto a Cartesian plane. This is partly about data availability, in the sense of access to data (lots of research projects use geonames data for its global coverage, open license, and linked data connectivity). It’s partly about data availability in the sense of access to thinking. Place-name gazetteers look point-based, because the place-name on a flat map begins at a point on a cartesian plane. (So many place-name gazetteers are derived visually from the location of strings of text on maps; they are for searching maps, not for searching *stuff*)

    So next steps seem to involve

    • dissolving the difference between narrative, and data-driven, representations of the same thing
    • inferring things from mereological relations (containment-by, containment-of) rather than sequential or planar relationsOn the former – data are documents, documents are data.

    On the latter, this helps explain why i am still talking about this, because it’s still all about access to data. Amazing things, that i barely expected to see so quickly, have happened since i started along this path 8 years ago. We now have a significant amount of UK national mapping data available on properly open terms, enough to do 90% of things. OpenStreetmap is complete enough to base serious commercial activity on; Mapquest is investing itself in supporting and exploiting OSM. Ordnance Survey Open Data combines to add a lot of as yet hardly tapped potential…

    Read more, if you like, in Connecting archives with linked geodata – Part II which covers the use of and plans for the Unlock service hosted at the EDINA data centre in Edinburgh.

    Chalice poster from AHM 2010

    October 22nd, 2010

    Chalice had a poster presentation at All Hands Meeting in Cardiff, the poster session was an evening over drinks in the National Museum of Wales, and all very pleasant.

    Chalice poster

    View the poster on scribd and download if from there if you like, be aware the full size version is rather large.

    I’ve found the poster very useful; projected it instead of presentation slides while I talked at FOSS4G and at the Place-Names workshop in Nottingham on September 3rd.

    Quality of text correction analysis from CDDA

    October 21st, 2010

    The following post is by Elaine Yeates, project manager at the Centre for Data Digitisation and Analysis in Belfast. Elaine and her team have been responsible for taking scans of a selection of volumes of the English Place Name Survey and turning them into corrected OCR’d text, for later text mining to extract the data structures and republish them as Linked Data.

    “I’ve worked up some figures based on an average character count from Cheshire, Buckinghamshire, Cambridgeshire and Derbyshire.

    We had two levels of quality control:

    1st QA Spelling and Font:- On completion of the OCR process and based on 40 pages averaging 4000 characters per page the error rate was 346 character errors (average per page 8.65) = 0.22

    1st QA Unicode:- On completion of the OCR process and based on 40 pages averaging 4000 characters per page the error rate was 235 character errors (average per page 5.87)= 0.14.

    TOTAL Error Rate 0.36
    2nd QA – Encompasses all of 1st QA and based on 40 pages averaging 4000 characters per page the error rate was 18 character errors (average per page 0.45) = 0.01.

    Through the pilot we indentified that there are quite a few Unicodes unique to this material. CDDA developed an in-house online Unicode database for analysts, they can view, update the capture file and raise new codes when found. I think for a more substantial project we might direct our QA process through an online audit system, where we could identify issues with material, OCR of same, macro’s and the 1st and 2nd stages of quality control.

    We are pleased with these figures and it looks encouraging for a larger scaled project.”

    Elaine also wrote in response to some feedback on markup error rates from Claire Grover on behalf of the Language Technology Group:

    ‘Thanks for these. Our QA team our primarily looking for spelling errors, from your list the few issues seem to be bold, spaces and small caps.

    Of course when tagging, especially automated, you’re looking for certain patterns, however moving forward I feel this error rate is very encouraging and it helps our QA team to know what patterns might be searchable for future capture.

    Looking at your issues so far, on part Part IV (5 issues e-mailed) and a total word count of 132,357 (an error rate of 0.00003).”

    I am happy to have these numbers, as one can observe consistency of quality over iterations, as means are found to work with more volumes of EPNS.

    Musings on the first Chalice Scrum

    October 18th, 2010

    For a while i’ve been hearing enthusiastic noises about how Scrum development practise can focus productivity and improve morale; and been agitating within EDINA to try it out. So Chalice became the guinea-pig first project for a “Rapid Application Development” team; we did three weeks between September 20th and October 7th. In the rest of this post I’ll talk about what happened, what seemed to work, and what seemed soggy.

    What happened?

    • We worked as a team 4 days a week, Monday-Thursday, with Fridays either to pick up pieces or to do support and maintenance work for other projects.
    • Each morning we met at 9:45 for 15 minutes to review what had happened the day before, what would happen the next day
    • Each item of work-in-progress went on a post-it note in our meeting room
    • The team was of 4+1 people – four software developers, with a database engineer consulting and sanity checking
    • We had three deliverables –
          a data store and data loading tools
          a RESTful API to query the data
          a user interface to visualise the data as a graph and map

    In essence, this was it. We slacked on the full Scrum methodology in several ways:

    • No estimates.

    Why no estimates? The positive reason: this sprint was mostly about code re-use and concept re-design; we weren’t building much from scratch. The data model design, and API to query bounding boxes in time and space, were plundered and evolved from Unlock. The code for visualising queries (and the basis for annotating results) was lifted from Addressing History. So we were working with mostly known quantities.

    • No product owner

    This was mostly an oversight; going into the process without much preparation time. I put myself in the “Scrum master” role by instinct, whereas other project managers might be more comfortable playing “product owner”. With hindsight, it would have been great to have a team member from a different institution (the user-facing folk at CeRch) or our JISC project officer, visit for a day and play product owner.

    What seemed to work?

    The “time-boxed” meeting (every morning for 15 minutes at 9:45) seemed to work very well. It helped keep the team focused and communicating. I was surprised that team members actually wanted to talk for longer, and broke up into smaller groups to discuss specific issues.

    The team got to share knowledge on fundamentals, that should be re-useful across many other projects and services – for example, the optimum use of Hibernate to move objects around in Java decoupled from the original XML sources and the database implementation.

    Emphasis on code re-use meant we could put together a lot of stuff in a compressed amount of time.

    Where did things go soggy?

    From this point we get into some collective soul-searching, in the hope that it’s helpful to others for future planning.

    The start and end were both a bit halting – so out of 12 days available, for only 7 or 8 of those were we actually “on”. The start went a bit awkwardly because:

        We didn’t have the full team available ’til day 3 – holidays scheduled before the Scrum was planned
        It wasn’t clear to other project managers that the team were exclusively working on something else; so a couple of team members were yanked off to do support work before we could clearly establish our rules (e.g. “you’ll get yours later”).

    We could address the first problem through more upfront public planning. If the Scrum approach seems to work out and EDINA sticks with it for other projects and services, then a schedule of intense development periods can be published with a horizon of up to 6 months – team members know which times to avoid – and we can be careful about not clashing with school holidays.

    We could address the second problem by broadcasting more, internally to the organisation, about what’s being worked on and why. Other project managers will hopefully feel happier with arrangements once they’ve had a chance to work with the team. It is a sudden adjustment in development practise, where the norm has been one or two people full-time for a longish stretch on one service or project.

    The end went a bit awkwardly because:

      I didn’t pin down a definite end date – I wasn’t sure if we’d need two or three weeks to get enough-done, and my own dates for the third week were uncertain
      Non-movable requirements for other project work came up right at the end, partly as a by-product of this

    The first problem meant we didn’t really build to a crescendo, but rather turned up at the beginning of week 3, looked at how much of the post-it-note map we still had to cover. Then we lost a team member, and the last couple of days turned into a fest of testing and documentation. This was great in the sense that one cannot underestimate the importance of tests and documentation. This was less great in that the momentum somewhat trickled away.

    On the basis of this, I imagine that we should:

    • Schedule up-front more, making sure that everyone involved has several months advance notice of upcoming sprints
    • Possibly leave more time than the one review week between sprints on different projects
    • Wait until everyone, or almost everyone, is available, rather than make a halting start with 2 or 3 people

    We were operating in a bit of a vacuum as to end-user requirements, and we also had somewhat shifting data (changing in format and quality during the sprint). This was another scheduling fail for me – in an ideal world we would have waited another month, seen some in-depth use case interviews from CeRch and had a larger and more stable collection of data from LTG. But when the chance to kick off the Scrum process within the larger EDINA team came up so quickly, I just couldn’t postpone it.

    We plan a follow-up sprint, with the intense development time between November 15th and 25th. The focuses here will be

    • adding annotation / correction to the user interface and API (the seeds already existing in the current codebase)
    • adding the ability to drop in custom map layers

    Everything we built at EDINA during the sprint is in Chalice’s subversion repository on Sourceforge – which I’m rather happy with.