On Tue, Feb 28, 2012 at 18:13, Ahmed Hassan ahmed@linuxism.com wrote:
You mentioned some good points about such as "Have a fixed number of template sentences", but you almost didn't mention how to implement your proposal.
We didn't elaborate at length because we thought the implementation details of that weren't important for understanding the idea at this stage. We will of course provide a demo implementation (probably in Ruby or Perl) once we have the dictionaries compiled. I don't know the Privoxy source at all, so would need help in translating it to work there, but that can be done afterwards.
Compiling the dictionaries is a significant amount of work (cf our requirements list, which eg RFC1751 & RFC2289 utterly fail). That's why we're stopping for input now. It's also important to have the dictionaries compiled before we can generate a full set of templates; there are some complicated linguistic interconstraints that make just generating one without the other a bad idea.
Testing can be done in various ways, e.g. just asking subjects to remember (both for recognition and input) a random hash-phrase for 1h, 1d, or 1w, before full deployment.
- Sai