AmsterdamHawkins

Ann: A lot of people working in similar areas, so I would start from asking what we would get from Delph-in resources compared to what's been done; separate this from how Hawkins' proposal compares to others.

David: Delph-in provides scale.

Ann: People have done similar things on the PTB, and more recently UD. Would suggest looking at Gibson, Futrell, Levy... (probably others as I'm not working in this area.)

Emily: Something Delph-in can provide is both parsing and generation.

---

Glenn: How do you combine the different sums in MaOP?

David: Just summed (not weighted)

Eric: Could this be a way to rank analyses without a treebank?

David: Before doing that, want to see if it correlates with reality.

---

Emily: May need to choose corpus more carefully, e.g. spoken corpus; connect to Olga's LLM vs. human, Petter's incremental trees.

---

Guy: Could we formalise this in terms of dependencies instead of constituents?

Dan: DMRS could provide the flatter structure.

Emily: Could compare with UD and what Gibson et al. are doing.

David: Articles also treated as identifying an NP.

Guy: Then DMRS doesn't directly match his analysis for determiners.

Emily: Reminds me of graph-walking for scope/focus of negation.

Francis: This sounds like more work than flattening the tree.

---

Emily: This analysis seems to rely on word boundaries.

David: Yes, but he doesn't discuss this.

Emily: What about Romance prepositions and determiners squished together?

David: I haven't seen a tree for that.

Dan: What about coordination?

David: Right, he would make a strong claim about that.

---

Guy: MiD seems interesting, but MaOP seems out-of-date, e.g. very clear that semantics and discourse context affect garden paths.

David: Also not probabilistic.

Ann: Would need to look at psycholinguistic experiments.