Speech act conditionals

[nschneid]

Inspired by a recent XKCD comic:

This is called a speech act conditional: the "if X, Y" means something like "if X, then it is relevant to the conversation that Y".

How should we treat this in AMR? I don't see a guideline in the AMR Dictionary. Clearly :condition is suboptimal for such sentences as:

[1]   cjconsensus   DF-199-192823-675_6015.9   (snt. 9 in workset  dfa-wset-11, last updated on Tue Jun 9, 2015)
Here's a place where there are some livestreams if you want to check it out.

(p / place 
      :location-of (s2 / stream-03 
            :quant (s / some) 
            :mod (l / live)) 
      :location (h / here) 
      :condition (w / want-01 
            :ARG0 (y / you) 
            :ARG1 (c / check-out-05 
                  :ARG0 y 
                  :ARG1 p))) 

[kevincrawfordknight]

My name is Bill if you need me.

[timjogorman]

Could I suggest something like "have-pragmatic-condition-91"? That would be in line with PDTB, which has a separate relation type of "Pragmatic Conditional" (a Contingency subtype) for this.

have-pragmatic-condition-91 arg1: Event which is true arg2: situation in which the information arg1 is relevant or needed arg3: entity to whom it is relevant (not sure if we want this one)

PDTB examples encompass both pure relevance conditionals like " If you are thirsty, there’s beer in the fridge" and what they call "implicit assertions" like “If you want to keep the crime rates high,O’Connor is your man.”

[uhermjakob]

We might also consider expanding the existing have-condition-91

have-condition-91 arg1: main event, true if condition in arg2 is met arg2: condition arg3: context info, true regardless of condition, but particularly relevant if condition in arg2 is met

Pragmatic conditions would then typically use only arg2 and arg3 with arg1 being implied.

[nschneid]

Decision at today's meeting: adopt a new frame, have-relevance-91, which would cover pragmatically relevant information in the case of speech act conditionals.

have-relevance-91 arg1: relevant information arg2: relevant to what (if omitted, assumed to be the present discourse)

We take speech act conditionals to introduce a piece of information possibly relevant to the current discourse—it is the relevance of the state of affairs (to the discourse) that is conditional, not the truth of the proposition. The speaker doesn't wish to presume that it is relevant, and is effectively hedging. Thus, for:

I'll be in your city tomorrow if you want to hang out (with me).

The following represents a logical condition reading:

(b / be-located-at-91
    :ARG1 (i / i)
    :ARG2 (c / city :poss (y / you))
    :condition (w / want-01
        :ARG0 y
        :ARG1 (h / hang-out-06 :ARG1 y :ARG2 i)))

Contrast with the speech act conditional reading:

(b / be-located-at-91
    :ARG1 (i / i)
    :ARG2 (c / city :poss (y / you))
    :ARG1-of (h2 / have-relevance-91
        :condition (w / want-01
            :ARG0 y
            :ARG1 (h / hang-out-06 :ARG1 y :ARG2 i))))

This says that being-in-your-city is relevant on the condition that you-want-to-hang-out. (Paraphrase: "If you want to hang out, then IT IS WORTHWHILE TO KNOW THAT I'll be in your city".) What is it relevant to (implicit ARG2 of have-relevance-91)? Roughly, "the current conversation", but we do not attempt to formalize this in the AMR.

We can also exploit have-relevance-91 for a proposition's contextual background presented as something less than a cause (e.g., "Given the weather, we should stay indoors."). See #136.

[kevincrawfordknight]

have-relevance-91

Need to create this frame.

[uhermjakob]

Added new frame have-relevance-91 to AMR Editor. Added new entry to AMR Dictionary: http://www.isi.edu/~ulf/amr/lib/amr-dict.html#have-relevance-91

[uhermjakob]

Having second thoughts about have-relevance-91 that seems to be a duplicate of previously existing relevant-01 with the exact same argument structure. I propose to just use relevant-01 for conditional relevance sentences such as "My name is Bill if you need me."

(n / name-01
  :ARG1 (i / i)
  :ARG2 (p / person :name (n2 / name :op1 "Bill"))
  :ARG1-of (r / relevant-01
             :ARG2 (y / you)
             :condition (n3 / need-01
                          :ARG0 y
                          :ARG1 i)))

[nschneid]

It seems like it would be odd to paraphrase with the adjective "relevant", though. Might be worth distinguishing metalinguistic relevance vs. explicit discussion of relevance.

On Jan 10, 2017 5:42 PM, "Ulf Hermjakob" notifications@github.com wrote:

Having second thoughts about have-relevance-91 that seems to be a duplicate of previously existing relevant-01 with the exact same argument structure. I propose to just use relevant-01 for conditional relevance sentences such as "My name is Bill if you need me."

(n / name-01 :ARG1 (i / i) :ARG2 (p / person :name (n2 / name :op1 "Bill")) :ARG1-of (r / relevant-01 :ARG2 (y / you) :condition (n3 / need-01 :ARG0 y :ARG1 i)))

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/amrisi/amr-guidelines/issues/177#issuecomment-271721690, or mute the thread https://github.com/notifications/unsubscribe-auth/AA8Ir_2UZ9BlT-Newin1hsMGNmtgpw7lks5rRAlTgaJpZM4HtNIT .

[nschneid]

Decision: use relevant-01. @ulfulf will update documentation.