[erg] issues / questions re the latest head of the ERG

[Dan Flickinger]

Hi Paul -

Thanks for reminding me that I have not yet put together a summary of the relevant changes to the ERG from version "1212" to the new stable version "1214".  I will figure out where best to put the information on the wiki, but in the meantime, below is a list of the changes that might affect users of the grammar, a couple of which you have already run into.

Cheers,

 Dan

-------------------------------------
Changes in the ERG from 1212 to 1214:

1. Attribute name change

I changed the name of the attribute PRONTYPE to PT for layout reasons, since the other attribute names on ref-inds were already shorter, and this renaming just lets us align the attribute value pairs more easily. 

2. Values of PT renamed

For similar motivations, the following three type names for possible values of PT also changed:
  std_pron  => std
  zero_pron => zero
  not_pron => notpro

3. Abstract type name changed:

In aiming for more consistency in the naming conventions of abstract predicates, I renamed 
quant_or_wh_rel => abstract_q_rel 
so that searches for quantifier relations could consistently look for the "_q_" pattern.

4. Rule name changes

On the syntax side, I renamed several rules in striving for more consistency in naming conventions:

n-nh_v-cpd_c => n-v_j-cpd_c
n-nh_j-t-cpd_c => n-j_j-t-cpd_c
n-nh_j-cpd_c => n-j_j-cpd_c
j-n_n-ed_c => j_n-ed_c
hdn_num-seq_c => n-n_num-seq_c
mnp_deg_prd_c => mnp_deg-prd_c

j_enough_wc_dlr => j_enough-wc_dlr


5. Linguistic analysis change

After struggling for a year with an unsatisfactory analysis of gapping constructions as in "We gave Kim a cat and Abrams a dog", I have retreated back to the drawing board, and removed that analysis in the 1214 version.  This is a minor regression in apparent linguistic coverage, but the removal of that machinery for gapping results in a significant gain in parsing efficiency, and leaves the door open to a more defensible and scalable analysis down the road.





----- Original Message -----
From: "Paul Haley" <paul at haleyai.com>
To: erg at delph-in.net
Sent: Friday, November 21, 2014 2:07:04 PM
Subject: Re: [erg] issues / questions re the latest head of the ERG

Might it be that redwoods.mem contains references to this feature while 
the ERG does not define it?

Also, the ERG semantics pages on Delphin's site are quite helpful. I 
found them by looking at recent changes.  Great work by the author.  
Haven't found anything comprehensive at the argument feature level, 
unfortunately.

Best,
Paul

On 11/18/2014 2:39 PM, Paul Haley wrote:
> Greetings,
>
> It appears that some semantic attributes of features of arguments of 
> predications have changed in MRS, such as PRONTYPE becoming PT and one 
> of its values changing from STD_PRON to STD.  Looking at semi.vpm and 
> (R)MRS discussions on DELPH-IN, it appears that more has changed, but 
> I cannot find a good summary or discussion of changes at the feature 
> level, in particular.  Is there a reference or some brief summary of 
> what has changed in the ERG with respect to MRS over the last year?
>
> Thank you,
> Paul
>
>
> P.S. In compiling the ERG (with a year old flop), I receive a number 
> of messages such as follows.  The only references I can find to the 
> type are commented out in constructions.tdl and ctype.tdl.  The result 
> seems to work fine, but any insights or suggestions would also be 
> appreciated.
>
>
> WARN    : Unknown type/instance `n-hdn_ttl-cpd_c' in feature #636307
> WARN    : Unknown type/instance `n-hdn_ttl-cpd_c' in feature #636308
> WARN    : Unknown type/instance `n-hdn_ttl-cpd_c' in feature #636309
> WARN    : Unknown type/instance `n-hdn_ttl-cpd_c' in feature #636959
> WARN    : Unknown type/instance `n-hdn_ttl-cpd_c' in feature #637573
> WARN    : Unknown type/instance `n-hdn_ttl-cpd_c' in feature #637575
> WARN    : Unknown type/instance `n-hdn_ttl-cpd_c' in feature #637578
>