What if.... we could automatically test our textual instructions?

A common problem in any complex computational environments are stale instructions. Software systems change but their related instructions are left unmodified. This problem is made further complicated in Debian by the large number of human languages supported by the project.

Branavan et al. (2009) describe an automated way to translate human, textual instructions into executable actions. [1]

Their approach involves reinforcement learning with no human intervention, by probing alternative interpretations of the instructional text in a virtual machine. The computer learns which different actions (such as "choose third option from the menu") correspond to different English words in the instructions. Their system also learns how to re-order instructions (as text presentation doesn't necessarily imply execution order, e.g. "click done after selecting /dev/usb"). They are also making their code available (targeted towards a MS Windows VM). [2]

In this talk, I will sketch their approach, show a prototype of a Debian-adapted system and seek out collaborators to put this technology to use in different parts of the project, especially in d-i and i18n.

[1] Reinforcement Learning for Mapping Instructions to Actions ACL 2009 [Best paper award] S.R.K. Branavan, Harr Chen, Luke Zettlemoyer, Regina Barzilay. <a href="http://people.csail.mit.edu/branavan/papers/acl2009.pdf">http://people.csail.mit.edu/branavan/papers/acl2009.pdf</a>

[2] <a href="http://groups.csail.mit.edu/rbg/code/rl/">http://groups.csail.mit.edu/rbg/code/rl/</a>