amrisi / amr-guidelines

244 stars 87 forks source link

Inconsistent annotation of "please" #78

Closed mgeorgescu closed 10 years ago

mgeorgescu commented 10 years ago

Currently, "please" has been rendered as :mod, or it has been dropped altogether.

Please don't judge our system on the basis of the Jackson trial!

(j / judge-01 :polarity - :mode imperative
      :ARG0 (y / you)
      :ARG1 (s / system
            :poss (w / we))
      :ARG1-of (b / base-02
            :ARG0 (t / try-02
                  :mod (p2 / person :name (n / name :op1 "Jackson"))))) 

OR

Please indicate, by noon on Saturday, Jan 8th, your picks on these 4 games:


(i / indicate-01 :mode imperative
      :ARG0 (y / you)
      :ARG1 (p2 / pick-01
            :ARG0 y
            :prep-on (g / game :quant 4
                  :mod (t / this)))
      :mod (p / please)
      :time (b / by
            :op1 (d / date-entity :month 1 :day 8
                  :weekday (s / saturday
                        :time (n / noon)))))

We noticed that there is a frame please-01 which also accommodates a "discourse usage" but as the frame is not illustrated with the :args, we were not sure if it should be used.

please.01 - "make happy"

ARG0: agent, pleaser   Cause
ARG1: patient, entity pleased   Experiencer
ARG2: instrumental 

more Example - "discourse usage"

Policies designed to encourage one type of investor over another are akin to placing a sign over the Big Board's door saying: ``Buyers welcome, sellers please go away!''
REL: please 
cbonial commented 10 years ago

There is another note about "discourse usage" in the frame that makes me think the example is only listed for cases that are essentially mistakes in the tree: "please.01: Discourse usage might also show traces in subject/object position,depending on which of the treebankers looked at that particularsentence. You may include them as targetless traces if they arepresent. This is pretty much adjectival." This is why no annotation is present in the example for PB purposes.

Given that they seem fairly simple to find, I would propose that we consider a new role for discourse items like "please," and leave it out in the meantime.

uhermjakob commented 10 years ago

Thanks for bringing this up. Might be time for a new role. Kevin and I propose a new role :polite +.

Please close the door. = Could you close the door? = Would you kindly close the door?

(c / close-01 :mode imperative :polite +
  :ARG0 (y / you)
  :ARG1 (d / door))
nschneid commented 10 years ago

Interesting idea, but I'm not sure how the boundaries would be drawn:

Is it restricted to requests? Or also to mark register?

nschneid commented 10 years ago

I think a general role for discourse markers could be useful; there are miscellaneous things that come up frequently, and it's often unclear whether to throw them out or use :mod or try something fancier. Off the top of my head:

Obviously, a proposal along these lines would require a fair bit of research and discussion.

uhermjakob commented 10 years ago

At the AMR phone meeting, we agreed to adopt ":polite +" to annotate "textual decorations" such as please and would you kindly that make a sentence more polite.

We don't use ":polite +" when something is inheritently polite (such as "You are a wonderful cook.") or more indirectly polite (such as "It was not the most accessible textbook I have read.").

However, ":polite +" is not limited to (semantic) imperatives such as "Could you please close the door?" It can also occur with (semantic) questions. For example, Could you tell me what time it is? would be annotated as What time is it? with an additional ":polite +"

And yes, Nathan is right in that we will at some point have to revisit issues such as hedges and intensifiers.