amrisi / amr-guidelines

250 stars 87 forks source link

Precondition #136

Open timjogorman opened 9 years ago

timjogorman commented 9 years ago

Discussion of Nathan's post #132 brought up the idea of whether it's appealing to have a role for "contributing factors that are weaker than causes".

RED has just such a relation, which we call "precondition" (many discourse relation inventories call the same thing "enablement"). Our current definition in the guidelines is:

"we annotate "X PRECONDITIONS Y" if, according to the writer, had the particular EVENT X not happened, the particular EVENT Y would not have been able to happen. Put differently, although a precondition EVENT is not sufficient to cause another EVENT, its occurrence is necessary in order for the other EVENT to have happened."

Some prototypical examples from the guidelines being:

I think that often these kinds of relations would get no marking at all in AMR, or merely temporal. But for the cases like Nathan mentioned with phrases like "in light of" or "given that", where we are marking an explicit causal relationship, it might make sense to consider something like this. The "in light of" example: Yisrael Beiteinu Knesset member and Deputy Interior Minister Faina Kirshenbaum asked party chairman Avigdor Lieberman not to seek re-election in light of the police investigation into her alleged role in the party's graft scandal.

What do people think?

(The paper Martha mention on causation and RED is http://acl2014.org/acl2014/W14-29/pdf/W14-2903.pdf)

nschneid commented 9 years ago

Hmm. I'm wondering if there's an even weaker version than necessary-condition. What about:

Does this imply that she would have decided differently if not for the apple? I'm not sure it does; to me it just indicates that the apple helped inform the decision, weighing in in favor of what was ultimately decided. If it had weighed in against what was decided, it would have been linked with something else:

I wonder if "given that" is marking something less like enablement and more like facilitation/increased-likelihood.

I like this example from the FrameNet Book:

The because is linking an indirect cause: what is implied is that somebody put the lavender in the fridge in order to fulfill the instructions on the packet, and that is how the lavender came to be in the fridge. But again, I'm not sure this is framed as a necessary condition.

nschneid commented 9 years ago

Some other examples that might be facilitation rather than enablement:

I assume that this would instead be causation:

nschneid commented 9 years ago

Kira: not just facilitation—also an element of contextualization (in the discourse)

nschneid commented 9 years ago

The name :given might capture both facilitation and contextualization.

nschneid commented 9 years ago

To emphasize: We are interested not in deeply expressing all possible causal(ish) relations between events in the text; we are talking about cases where such a relationship is marked linguistically ("given", "in light of", "whereas", some absolute phrases, etc.) and annotators might be hesitant to use cause-01.

nschneid commented 8 years ago

177 introduced the frame have-relevance-91:

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

We exploit this here for a proposition that serves as contextual background for anther proposition:

Given that she had an apple for breakfast, she decided to have a banana for lunch.

(d / decide-01
    :ARG0 (s / she)
    :ARG1 [to have a banana for lunch]
    :ARG2-of (h / have-relevance-91
        :ARG1 [she had an apple for breakfast]))

Likewise:

With temperatures in the low 70s, it was a perfect day for a walk. = It was a perfect day for a walk: after all, temperatures were in the low 70s.

(p / perfect-02
    :ARG1 (d / day)
    :ARG2 (w / walk-01)
    :ARG2-of (h / have-relevance-91
        :ARG1 [temperatures were in the low 70s]))
nschneid commented 8 years ago

Suggestion: :given could be an editor shortcut for :ARG2-of have-relevance-91. (I'm wary of :relevant because it doesn't quite capture the asymmetry between the thing that is relevant and the thing it's relevant to.)