Results

Next?

Lessons Learned

Web as lab works.

Mechanical Turk as recruiter is cheap and fast.

Routines managing sessions and data work.

No-answer button should move.

Disambiguating representations can contain new ambiguities.

Results nonobvious and mysterious.

Main Study

Worth doing.

Don't revise what works.

Future Work

Introduce context.

Define objective correctness (versus consistency and agreement).

Systematically vary verb arguments and cause-effect order.

Escalate task from selection to formulation.

Multilingualize.

Valid XHTML 1.1!