timjogorman / richereventdescription Goto Github PK

We talked about the specific phenomenon of "imperative" forms. What we settled on in the call is that:

• If you have an imperative which will be likely followed, it can be treated as "determined future action" and be treated as a ACTUAL/AFTER event.
• If you have an imperative where it's in double whether anyone will follow it, you can make it AFTER/UNCERTAIN-HEDGED

This whole issue is flush with this other set of things like expression of desire or wanting, like "I asked him to leave". We're try to be consistent about treating those as HYPOTHETICAL. This might sometimes feel inconsistent, but I'm hoping that we can follow that distinction.

If you run into instances where Imperatives seem more like HYPOTHETICAL, or when the expressions of desire/ urges/ requests seem to be better as ACTUAL or UNCERTAIN, mention them here -- we want to get a sense of whether this is a good boundary here.

In THYME, we had the general tradition of "don't grab numbers", mostly as a justification for ignoring the numbers in things like "Platelets: 600".

In general, we are going to shift to explicitly allowing you to grab numbers when they stand for an ellided head that is referential. So "100s were killed" stands in for the victims, and should be coreferential with them, etc.

This brings up the awkward question - both in whether to mark it and what span to use -- for something like "$100,000". If "1,000 dollars" was referential, you would grab just "dollars", but [$]1,000 seems very unintuitive. Instead, just grab the whole number and dollar value when referential, as in [$1000].

An example brought up in the call of adjectives that are NOT eventive:
Tall and tan and young and lovely, the girl from Ipanema goes Ew, what�s that smell?�

We talked more about this distinction for pertainyms, particularly country ones like "French" and "American". The hard part is that it's quite hard to distinguish when something being mentioned is "really the entity" vs some general mention.

The current guidelines state "when the actual named entity is not being evoked itself, but merely some stereotype, origin or style of that entity, then ignore it", with the examples of things not to use being

• [Entity I ] like american [Entity music ]
• [Entity I ] like indian [Entity food ]

We still want to avoid marking those two examples as entities, but came up with a rule that is hopefully more distinguishable: if a pertainym like "american X" can be rephrased was "in the style of X", "X-style", like "american-style music" or "indian-style food", then it is NOT an entity, but if it implies more concrete relationships like "located in America", "manufactured in America", etc., then it DOES count as a mention of that country.

We've talked a great deal recently about the range of things that appear to be either an event or an entity based upon context, or that appear to have both an event component AND an entity component. We've settled upon being pretty deterministic about them, based on type.

For Messages, communications, texts, etc.:

-We are going to default to EVENT whenever you possibly can

for events representing both events and funds (like "funding")

• These too, we are defaulting to events. If I remember the conversion

For events representing laws and cases:

• We agreed to allow this to be a rare case when both events and entities could be allowed (this is testing the waters with this), for most things like a clause, a rule, a law, etc., when they both refer to the act as a thing and the act as an eventuality with a certain timeline of "being in effect" in a legal system.

For discerning the span of annotations when there are possessives, we've followed this general "treebank" rule that you don't grab possessives when separated by an 's, as in [John]'s eating of the plums, but that we DO grab the whole thing for fused possessives like "his" or "their".

Someone brought up the question of how we want to handle 's when used for plurals, as in "in the 50's". For these, we DO want to grab the whole thing, so "[50's]".

A very minor comment: at least in the LDC-released annotation, it appears that the "precondition" causal type suffix is PRECONDITION, not PRECONDITIONS (which appears in the annotation guide).

This is an archaic argument left over from when the scholastic 'the teacher failed 10 students' sense was included in this roleset. It shouldn't be possible with the existing roleset. We should be able to delete it, and any instances remaining in back annotation should be errors (or movable to fail.02)