nschneid
commented
5 years ago It seems like a slippery slope to start using unanalyzable for a) instances of fully productive patterns, and b) fully compositional phrases that use normal syntax but happen to convey a date/time ("this Tuesday at 4 late in the afternoon"—is "afternoon" part of a time expression, or just a descriptor? should the date and time be considered 1 unit or two? etc.).
Maybe the principle of the foundational layer is that it covers the abstract semantics of what IS expressed as predicate-argument or modifier-center relations, and expressions where we cannot designate a semantic head should have multiple centers:
arrive_P [on_R November_C 5_C , 2019_C]_T arrive on the 5th of November in 2019 — normal elaborator/relator structures [St. Louis UNA]_C , MO_C , USA_C 10_C [Main Street UNA]_C [Apt_C 207_E]_C
There's a question of what sorts of grouping to use: I put "Apt 207" as C-E because it's the 207th apartment/the apartment numbered 207, so "apartment" is arguably the semantic head. But when different parts of a "locator" expression provide a location at different scales, as with date/month/year, street/house number/apartment number, and city/state/country, it's harder to say that any of these components is the head of the others.
omriabnd
dotdv
Ideally I'd like to ideally make all definite time expressions unanalyzalbe. So dates, days of the week, time of the day etc. The reason is that they have their own unique syntax, which I don't want to model (it's not part of the foundational layer and frankly it's not that interesting). The problem is with cases like: the 3rd and 5th of November, where the time expression syntax interacts with the other types of syntax. I think in these cases, we can still say it's unanalyzable, even though it isn't, just so we won't get into this thorny bush. What do you think?