Consider these sentences and their MRSs (edit: sorry the right side is clipped; you should be able to click and see the full image, or just reparse the sentences at http://delph-in.github.io/delphin-viz/demo)
- Browne merely doesn’t work.
- Browne doesn’t merely work.
These clearly have different entailments, and there are differences in the respective MRSs, but I’m confused about the first one. Can someone remind me why the ARG1 of
_mere_a_1 selects the ARG0 of
e2) and not of
e10), which it shares a label with? (The second MRS is as I would expect.)
This construction causes problems with my selection of a representative node for DMRS (
neg should be the DMRS top node, I think, and not
_mere_a_1, but currently it’s the other way around). I want to know if this is a bug in the ERG or if I should find a way to special-case this.
I may have figured it out. Is it because the index of
_work_v_1 is passed up, rather than the index of
neg, and so this is what’s available for
_mere_a_1 to select? For example, in “Browne doesn’t work”, the TOP of the MRS is the LBL of
neg but the INDEX is the ARG0 of
_work_v_1. So the
_mere_a_1 in “Browne merely doesn’t work” selects the LTOP and INDEX of the little “doesn’t work” subgraph.
Does that sound right?
Yes, that sounds right. My understanding is that we do it this way because it’s not supposed to make sense to talk about “negative events” (i.e. what the ARG0 of neg_rel putatively stands for).
Ok thanks. But it’s not just negatives, but any scopal adverb, yes? E.g., “Browne merely probably works.” has the same structure, but “Browne merely thinks he works.” (also scopal, not adverbial) does not.
I’m surprised that “merely” is treated as a non-scopal modifier, because in my mind it’s a more formal way of saying “only”, which is treated as scopal.
The strange MRS is a result of the MRS composition algebra, which makes it impossible to attach non-scopal modifiers underneath scopal modifiers, because the label of the verb is no longer available in HOOK. Similarly:
“Brown quickly doesn’t work”
Or a minimal working example:
“it heavily didn’t rain”
I don’t like either of the above sentences, but apparently this phenomenon is common in Turkish.
With the DMRS composition algebra, non-scopal modifiers can attach underneath scopal modifiers, because we can add an ARG1/EQ link to the INDEX node.
(In a way, it’s reassuring that the problem cases you’re finding are also cases with dodgy MRSs!)
I suspect you’ll find examples in any language with negation as inflection on the main verb.