slides

We mostly discussed whether nominalization-rel should be in the MRS, for nominalized clausal complements (did not really get to clausal modifiers).

Francis: For applications such as MT, we usually need to throw away things like nominalization-rel, so perhaps better not to have it there in the first place.

Francis: On the other hand, if we don't have it there for either high or low nominalization, are we then supposed to generate both trees from the MRS? Is there no semantic difference?

Emily: That is the main question: what is the difference in the semantics, if any.

Middle nominalization: so far we haven't seen any data that would really motivate it. The example from the ERG ("The cat quickly biting the dog surprised the man") can as well be analyzed as high nominalization. In order to motivate it, we would need to see something like: FIXME (look in the recording).

Francis: if something like "The quickly cat's eating the fish" existed, i.e. so that "quickly" modifies the entire thing, including the cat, that would be motivation for FIXME (look in the recording).