[erg] Logon missing MRS for

Dan Flickinger danf at stanford.edu
Wed Apr 1 19:27:35 CEST 2015

Ah, I picked the wrong variable to focus on in my reply, so I confused both of us, and failed to see your point, which is that you rightly expected an imperative analysis where the ARG1 is an implicit "you".  Upon inspection, I found that there is a bug in the grammar that has been preventing imperatives with right-node raising.  I had noticed a while ago that in ordinary imperatives of coordinated verb phrases, we were getting two analyses, as in |Buy books and read them!|, where one analysis first conjoins the two VPS and then applies the imperative rule, while the other first made each VP an imperative, and then conjoined the two sentences.  I wrongly thought this to be an unmotivated ambiguity, so I restricted the imperative rule to block conjoined VPs, thus only allowing [S-coord [S-imp [VP buy books]] [and [S-imp [VP read them]]].  But in right-node-raising, there is only one possible analysis, namely the one where we first construct the VP [[buy and read] bpoks] and then apply the imperative rule.  So it seems that we have to live with the ambiguity for ordinary conjoined imperatives.  I have repaired the imperative rule, and will check in the changes to `trunk' in the next few days.

Thanks, Paul, for the bug report, even if it took me a while to catch on!


----- Original Message -----
From: "Paul Haley" <paul at haleyai.com>
To: "Dan Flickinger" <danf at stanford.edu>
Sent: Tuesday, March 31, 2015 4:53:09 PM
Subject: Re: [erg] Logon missing MRS for

Yes, Dan, thanks.  We do add such quantifiers.  I was simply surprised 
not to have the implicit "you" argument for such ubiquitous imperative 
forms.  Actually, I am confident that I have seen it in other (simpler?) 
parses, which probably led to my mistaken expectation.

If you can spare a minute, I'd be interested in better understanding 
"unknown_rel".  It seems related to this missing 2nd person pronoun 
(from my perspective).


On 3/31/2015 5:56 PM, Dan Flickinger wrote:
> Those unexpressed argument "i" variables will show up a lot in MRSs, wherever a predication has an optional argument position that does not correspond to an overt expression in a sentence.  I make them of type "i" rather than "x" so we can preserve the expectation that all x's are bound by a quantifier, while avoiding the need for gratuitous quantifiers whenever an argument slot does not get filled.
> You might well argue that in many instances it would be plausible to infer the existence of an entity to fill such an argument position, such that an existential quantifier would seem well-motivated, but we don't want to make the grammar uniformly supply such an existential quantifier for every uninstantiated argument variable: for a relational noun like "solution" we would get a strange MRS for "That isn't a solution" in a setting where we mean "That isn't a solution to any problem" or equivalently "That is a solution to no problem".  The grammar can't add any information about an unexpressed argument, so we leave it underspecified in the MRS.  You might of course for a given application be able to supply additional constraints on some such arguments, in which case you could enrich the parser's output MRS to add a quantifier for that argument variable as a postprocessing operation on the MRS.
>   Dan
> ----- Original Message -----
> From: "Paul Haley" <paul at haleyai.com>
> To: "Dan Flickinger" <danf at stanford.edu>
> Cc: "erg" <erg at delph-in.net>
> Sent: Tuesday, March 31, 2015 12:23:49 PM
> Subject: Re: [erg] Logon missing MRS for
> Ooops, thank you very much (and regrets).
> I went on-line to see why there were no quantifiers in the MRS I was
> getting, but I see now that the only x has one.  But note the
> co-occurrence of i7 without quantification, which was the interesting
> case (at least it broke my code).  I was expecting an implicit 2nd
> person pronoun and its quantification.
> Thank you,
> Paul
> On 3/31/2015 3:18 PM, Dan Flickinger wrote:
>> Paul, your screen shot shows that you did not click the box asking for MRS output.  If you click that MRS box and resubmit the input, you should see the MRSs for each of the three analyses.
>>    Dan
>> ----- Original Message -----
>> From: "Paul Haley" <paul at haleyai.com>
>> To: "erg" <erg at delph-in.net>
>> Sent: Tuesday, March 31, 2015 12:11:44 PM
>> Subject: [erg] Logon missing MRS for
>> Parsing the following sentence produces a VP parse without any MRS. Is
>> that expected?
>> Generate and compare multiple solutions!
>> Thanks,
>> Paul

More information about the erg mailing list