Howler is a hypertextual, partly-collaborative Computer Assisted Language Learning system. It pervasively tracks user interaction with the system and uses this individual and group data to cluster similar learners by their behaviour. These clusters are then used to automatically generate tutor-style recommendations to resources, and to alterations in an individual's learning styles. Focusing on (preferably immersed) beginning learners of Mandarin, it presents an electronic dictionary interface to the user for basic tasks, and a reflective interface that allows them to explore the network of associations formed between items as they and other users learn.

See Howler project for more information.

• Programming languages — programming language interfaces, mobile & visual programming languages. (see Cooker project)
• Games — how players model and second-guess the worlds and user-interfaces they are given in games. (see Publications)
• Mobile — usable mobile interfaces for programming and novel mobile "side-effect" applications (for example, using a mobile with tilt sensors as a spirit-level).
• Visualisation — data visualistions, graphs/networks; demoscene

Howler is a hypertextual, partly-collaborative Computer Assisted Language Learning system and forms the basis of my PhD work.

Visual programming language targeted at beginners. The visual approach was chosen to separate the semantic and syntactic learning phases of programming. Programs are built by adding blocks which represent expressions or statements inside each other. Syntax mistakes are not possible as only valid types of blocks will be accepted by the containing block. The language is intended to be very close to non-OO Java constructs and is easily expanded via block panels.

The program contains an interpreter for the programs which allows step-by-step execution, stack traces, variable watching, and aborting program execution. The editor supports zooming in and out of the program, collapsing individual functions, and printing. A pervasive tooltip system is combined with a dynamic help system to allow local instructors to trivially augment it with class material.

The file format is XML (compressed by default) which contains checksums to detect text corruption (eg, via email) and there is a tool to assist in plagiarism detection

Coming soon: project homepage, download, GIT repo, and documentation.

Chinese/Classless Language Learning Environment. This was written as I was studying at Fudan University as a tool to help me learn. Contains a user-entered dictionary of words which allows quick, complex querying. It also has a flashcard drilling/quizing interface.

This formed the beginning of the Howler project and my PhD.

Taenarum was a distributed, collaborative, real-time programming development environment. It provided a Smalltalk-style code-browser to source code without having to have it in a particular format. It would read and parse standard source code as required so that it could be used by individual developers without having an effect on anyone else in the project (this is compared to most IDEs then, and now, which litter code with special files and file structures). Any edit actions that a user performed would be transmitted live to other Taenarum users on the same server so they could see a virtual "developing version" of the code as it happened. It had language plugins for C, Java, and Dylan, in various states of completion, CVS support, and a Java evaluation playground (via a third party interpreter).

Development slowed down as I got busy with other things and changed the way I worked. Collaborative editing was starting to be added to popular text editors which have better support than just me. Finally abandoned when I accidently left the source code in the UK when moving to live in China.

Minerva was an attempt to take high-resolution satellite images, such as those provided by the USGS, and automatically regenerate a 3D enviroment from them. The idea was that this would rapidly accelerate the first stage of content creation in open-world sandbox games which were popular at the time. Since others had better knowledge of image processing than me (ie, any), I decided to explore different approaches to identifying objects in the images.

This eventually got bogged down for three reasons:

1. An insistence on complete automation when minor user inputs could have solved significant problems. For example, the road extraction algorithm needed a start point. Large amounts of time went into finding that when a single click from the user would have done.
2. Frustration at being limited to only visible light images, since these were the only type freely available. Other wavelengths would have simplified tasks like foliage and water identification.
3. Getting side-tracked by not only starting to build an OpenGL widget set, but trying to do so without using the traditional hierarchically structured approach.

Dylan and OpenGL