Missing sense of "leave"

[goodmami]

When I parse something like "Kim left." I get a _leave_v_1 with ARG2 having an i variable for the dropped argument (filled by, e.g., "the office" in "Kim left the office"). This seems fine. However the SEM-I entry has this for _leave_v_1:

  _leave_v_1 : ARG0 e, ARG1 i, ARG2 i, [ ARG3 h ].

As far as I can tell, sentences that can fill the ARG3 use the leave-in-a-state sense of "leave" instead of the depart sense. Compare:

(1) Kim left the room quickly. (without ARG3) (2) Kim left the room speechless. (with ARG3)

I know we don't distinguish predicates solely by sense, but it seems that sentences like (1) cannot have an ARG3 while those like (2) must have one. I think these should be different predicates.

The other predicates for "leave" are not better candidates for these as they just combine with other words (and make lots of ambiguity):

  _leave_v_behind : ARG0 e, ARG1 i, ARG2 i.
  _leave_v_home : ARG0 e, ARG1 i, ARG2 i.
  _leave_v_i : ARG0 e, ARG1 i, ARG2 i.
  _leave_v_in : ARG0 e, ARG1 i, ARG2 i.
  _leave_v_off : ARG0 e, ARG1 i, [ ARG2 i ].
  _leave_v_on : ARG0 e, ARG1 i, ARG2 i.
  _leave_v_open : ARG0 e, ARG1 i, ARG2 i.
  _leave_v_out : ARG0 e, ARG1 i, ARG2 i.
  _leave_v_over : ARG0 e, ARG1 i, ARG2 i.
  _leave_v_to : ARG0 e, ARG1 i, ARG2 u, ARG3 h.
  _leave_v_transfer : ARG0 e, ARG1 i, ARG2 i, ARG3 i.

I think this problem may be more general than just "leave" but I don't have a good way of detecting them yet. I came across this example by chance. However it is in line with a more systemic issue of our predicates not actually having fixed arity.

[guyemerson]

It took me a while to make sense of this, but I agree. I initially misunderstood because (2) is ambiguous between:

(2a) Kim left the room; Kim was speechless. (2b) Kim caused the room of people to be speechless.

The two readings are distinguished in the ERG: (2a) uses a subord rel; (2b) uses ARG3.

I agree there should be a separate predicate because (2a) implies Kim left the room, but (2b) does not.

[guyemerson]

I don't think this impacts on the more general issue of predicates having a fixed arity, which I'm not convinced we need to do anyway.

[goodmami]

Yes, sorry, I didn't think about the (2a) reading. I had "Kim left the office flooded" (with water, not emotion), but used (2) because it made a nice minimal pair with (1).

the more general issue of predicates having a fixed arity, which I'm not convinced we need to do anyway.

I thought we make a big deal about fixed arity being required for MRS to be resolvable to LF, as mentioned here: http://moin.delph-in.net/RmrsFaq . And I'm finding it a useful property when trying to use a SEM-I to make sense of synthesized MRSs (e.g., from transfer), or in reading/writing the indexed-mrs format.

There's also this workshop on it: http://moin.delph-in.net/TheAbbey/Chrysalis2014Arity (I'm linking this but I don't have time right now to read it carefully; will try to return to this on Monday).

[guyemerson]

Fixed arity is needed for a Davidsonian approach, but it's not needed in a neo-Davidsonian approach (which looks like RMRS in that predicates are unary relations and ARG relations are separate binary relations).

I think fixed arity is a plausible goal, but I'm not convinced it's a necessary one.

[goodmami]

which looks like RMRS in that predicates are unary relations and ARG relations are separate binary relations

Yes, although as I recall RMRS is still Davidsonian on the technicality that arguments aren't linked via event variables but by anchors and that adjuncts have their own events, etc. But I think that's a pretty gray area.

I think fixed arity is a plausible goal, but I'm not convinced it's a necessary one.

I had been under the assumption that instances of variable arity in our outputs were due to grammar bugs. Maybe at the summit we can discuss the issue further, or see if we can formalize what the Chrysalis2014 workshop introduced as UDMRS so as to distinguish it from "formal MRS".

[fcbond]

But aside from the question of fixed arity, I think these should probably be different predicates as the senses are so different.

On Fri, May 10, 2019 at 10:30 PM Michael Wayne Goodman < notifications@github.com> wrote:

which looks like RMRS in that predicates are unary relations and ARG relations are separate binary relations

Yes, although as I recall RMRS is still Davidsonian on the technicality that arguments aren't linked via event variables but by anchors and that adjuncts have their own events, etc. But I think that's a pretty gray area.

I think fixed arity is a plausible goal, but I'm not convinced it's a necessary one.

I had been under the assumption that instances of variable arity in our outputs were due to grammar bugs. Maybe at the summit we can discuss the issue further, or see if we can formalize what the Chrysalis2014 workshop introduced as UDMRS so as to distinguish it from "formal MRS".

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/delph-in/erg/issues/15#issuecomment-491309492, or mute the thread https://github.com/notifications/unsubscribe-auth/AAIPZRSCP2EF63D53A6SP5DPUWBG7ANCNFSM4HMA2OKQ .

-- Francis Bond http://www3.ntu.edu.sg/home/fcbond/ Division of Linguistics and Multilingual Studies Nanyang Technological University

[danflick]

I have now assigned distinct senses to the two-argument vs. three-argument "leave" entries, but the more general issue is still open about a general methodology for deciding when we should assign distinct predicate names for sufficiently distinct senses.

[danflick]

We might revisit this issue once I have added coarse-grained WordNet senses to ERG lexical entries, but for now I'll close this issue.