- to stress that the task of language technology research efforts should be to build knowledge representations of some sort, preferably as explicit as possible;
- even in an unsupervised research effort, hypotheses must be stated;
- for research purposes the explanatory level of the resulting knowledge representation must be non-trivial;
- for engineering purposes, the knowledge representation must be designed answer to stated requirements.
Quite often, these various goals are conflated. My talk was a bit grumpy, but I still really like unsupervised methods.