[developers] "Quite" problematic: MRS -> EDS conversion

Michael Wayne Goodman goodmami at uw.edu
Thu Jul 26 22:07:59 CEST 2018


Thanks for explaining, Guy. I misunderstood in your previous email: simply
removing loc_nonsp does not help me resolve whether _in_p_temp or _hire_v_1
is the representative EP, but it certainly helps if the latter is the ARG1
of the former. Unfortunately no output of the ERG (niether 1214 nor trunk)
gives such an MRS. I'm hoping Dan or Emily (when they have a chance to
catch up with this thread) can shed some light onto whether the analysis we
have is correct or if the (more intuitive) one you propose is what we
should get.

Note that the cleft form is similar to "It was a day in January that Browne
was hired", which has a stronger case for the loc_nonsp. But this might
help explain why it's used in the original sentence (without "a day"),
e.g., in case the "in" is hidden under that "was" and for technical reasons
cannot be linked up with "hired".

On Thu, Jul 26, 2018 at 7:59 AM, Guy Emerson <gete2 at cam.ac.uk> wrote:

> "Browne was hired in January"
> - in the 1214 ERG, the ARG1 of _in_p_temp is the ARG0 of _hire_v_1.
> - we can therefore choose _hire_v_1 as the representative EP
>
> "it was in January that Browne was hired"
> - we need to add something to mark the cleft construction (presumably now
> in ICONS)
> - otherwise, nothing else needs to change in the semantics
> - if nothing else changes, we can choose _hire_v_1 as the representative
> EP, for the same reason as above
> - in the 1214 ERG, there is a loc_nonsp that mediates between _in_p_temp
> and _hire_v_1, but I think this is a bug
>
> As explained in that wiki page, loc_nonsp is used for implicit locatives,
> but there's nothing implicit here, because we have an explicit "in".  For
> example, compare:
> "Browne was hired"
> "it was Browne that was hired"
> - in the 1214 ERG, the only difference between the two is the
> _be_v_itcleft (which presumably is going to move to ICONS), and I think
> this is correct
> - whatever difference we have here should also be the only difference
> between "Browne was hired in January" and "it was in January that Browne
> was hired"
>
>
> Am 25.07.2018 20:19 schrieb "Michael Wayne Goodman" <goodmami at uw.edu>:
>
> Forgot to CC the list...
>
>
> On Wed, Jul 25, 2018 at 10:53 AM, Michael Wayne Goodman <goodmami at uw.edu>
> wrote:
>
>> On Wed, Jul 25, 2018 at 2:21 AM, Guy Emerson <gete2 at cam.ac.uk> wrote:
>>
>>> For "It was in January that Browne was hired", is the MRS correct?  I
>>> was surprised to see both _in_p_temp and loc_nonsp mediating between
>>> mofy(Jan) and _hire_v_1.  Is there a reason to have both?  It would seem
>>> more consistent to have just _in_p_temp, which takes the intrinsic variable
>>> of _hire_v_1 as its ARG1.  This would match the MRS for "Brown was hired in
>>> January".  Adding the it-cleft seems to have the side effect of adding
>>> loc_nonsp.
>>>
>>
>> Hmm, good point (this is relevant, for those following along:
>> http://moin.delph-in.net/ErgSemantics/ImplicitLocatives). I think this
>> version of the ERG is simply unable to link up the ARG1 of _in_p_temp to
>> the index of _hire_v_1. Notice that the ARG1 of _in_p_temp is
>> underspecified, so the loc_nonsp is required, then, to establish the
>> relationship. Other analyses that do not have loc_nonsp don't actually get
>> us a _in_p_temp that hooks up with _hire_v_1; rather, the "in January" is
>> modifying the "was" (i.e., not an it-cleft analysis). Also, if I remove
>> the loc_nonsp and change the ARG1 of _in_p_temp to select the ARG0 of
>> _hire_v_1, the MRS no longer generates. I'm not sure if the original
>> selected analysis is an underspecification of some valid ambiguity or a
>> technical compromise to deal with limitations in our rules of composition.
>>
>>
>>> The reason this is relevant is that, without the loc_nonsp, there would
>>> be no need to look at TENSE.
>>>
>>
>> I'm not following here. What would change in the graph to make _hire_v_1
>> stand out as the representative EP?
>>
>>
>>> 2018-07-24 23:18 GMT+01:00 Michael Wayne Goodman <goodmami at uw.edu>:
>>>
>>>> Answering one of my own questions; see below...
>>>>
>>>> On Fri, Jul 20, 2018 at 10:49 PM, Michael Wayne Goodman <
>>>> goodmami at uw.edu> wrote:
>>>>>
>>>>>
>>>>> If anyone has read this far, I have a question: both the ERG and Jacy
>>>>> have a [ TENSE untensed ] property, but is this a more widespread
>>>>> convention? I'm reluctant to rely on it because I want to avoid
>>>>> parametrizing my semantic conversion functions for grammar-specific values.
>>>>>
>>>>
>>>> I surveyed some grammars, and noted that the following use [ TENSE
>>>> untensed ]: ERG, Jacy, gg, SRG
>>>>
>>>> The following do not: NorSource, Semitic Grammar (HeGram), BURGER, HaG,
>>>> KRG, INDRA, Zhong
>>>>
>>>> Also note:
>>>> * HaG has [ TAM untensed ] exported in the VPM, but not [ TENSE
>>>> untensed ]
>>>> * Zhong does not use the TENSE property at all
>>>>
>>>> So I think it's safe to say that it's *not* a widespread convention
>>>> among medium-sized or larger grammars.
>>>>
>>>> One alternative to this specific tense property is the pos field of
>>>> predicates, which is part of MRS and not grammar-defined. The idea is that
>>>> predicates of certain pos values are more likely to modify others, such as
>>>> verbs modifying nouns ("sleeping dog"), adpositionals modifying verbs ("ran
>>>> quickly", "ran in the park"), degree modifiers on adpositionals ("ran very
>>>> quickly", "the cat was very much in the bag"), etc. Abstract predicates
>>>> (which do not formally have a pos) would come last, I suppose. But I'm not
>>>> certain that such a gradation is not language-specific, and there are
>>>> probably counter-examples.
>>>>
>>>
>>>
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/developers/attachments/20180726/64d13001/attachment-0001.html>


More information about the developers mailing list