Hi,

I looked at the various forums here, and thought that this discussion would best be had here. I hope that it's agreeable with everyone else.

The situation:

I'm a teacher at a language school. As a requirement to study at our school, we undertake level tests (English speaking and writing tests) for all incoming students. Currently this is entirely done by hand (writing) or in person (speaking) and takes quite a chunk of time, especially in times of high volume.

The solution:

Would it be possible to build a program, and 'teach it' to grade, (or at least provide an indicative grade) of a student's English proficiency? The speaking and writing tests in question are all standardized, and have rubrics matched to them. We also have some amount of tests (read:data) from past test takers that are used to train test (and standardize) graders such as myself. I imagine this would be useful to 'feed' the program information and use it to rain itself off of.

Questions:

I think it would be (much) easier to start off with the writing tests, as there are fewer variables to consider (compared to speaking).

• Where would I even start though?


• How difficult would it be to do this?

I don't have enough know-how to build the program myself but could definitely contribute to it's conceptual and developmental processes. As there isn't much budget for this (it's very much a DIY thing) I imagine it would be best to source expertise outside of NZ where labour costs are much lower, using perhaps a service such as mturk

I'm open to any other suggestions and ideas though. Currently I'm just in the information gathering stage of this project.