[erg] Missing article in a sentence causes a problem in parsing?

[<]

Dan

thank you for those details. i will try the robust grammar on our sample sentences and let you know what happens. 

but i understand your point about robustness vs accuracy and i wonder if there is a halfway position between the addition of a specific noncount entry for "male" ( and similar) and the use of the robust grammar to reduce the loss of accuracy.  in this example domain, only some limited telegraphic forms occur, such as a missing article, and i might like to just add a specific rule for "determiner phrases without a determiner", as it were, which would perhaps minimise the accuracy loss. i think i will take it as an exercise to try and do that!

this leads to a more general thought. in our project, one of the aspirations is to provide the user with the capability to add such rules themselves. i know some of my colleagues feel this is crazy, and you may do so too, and i appreciate that the ERG has taken decades to develop using deep linguistic knowledge. however the case in hand, missing determiners,  is i think a phenomenon that most english speakers could understand, and it is not impossible to imagine that they could describe the structure given a language and set if concepts to do so. or am i being totally naive? perhaps the problem is easy in isolation, but is made difficult because such a description has to be seamlessly added to an already complex set of rules? again maybe this is a good exercise for me to try

best

David

Sent from my iPad

On 16 Sep 2013, at 23:16, Dan Flickinger <danf at stanford.edu> wrote:

> Hi David -
> 
> You can invoke a robust version of the ERG which will allow for some common grammatical errors, including missing articles of singular count nouns as in your example, but you won't be surprised to hear that this more robust grammar introduces a lot of additional ambiguity in parsing, so the likelihood rises of an unwanted analysis being ranked most probable.  Still, you might experiment with this version, and see if the trade-off in robustness vs. accuracy is acceptable for your use.  (The references to `educ' in the version name etc. are because the primary use of this version is in an online English grammar teaching course.)
> 
> To run the robust version of the grammar with the PET parser, you can compile it as follows (in a terminal window):
>  flop -t educ
> (Don't be alarmed at the many warning messages produced by the grammar compiler; they are intentional redefinitions of types, rules, and lexical entries from the standard grammar.)
> Then invoke `cheap' in the usual way, but using `educ.grm' instead of `english.grm'
>  cheap -t -cm educ.grm
> You can do this either with the stable "1212" version of the ERG, or with `trunk' if you feel adventurous.
> 
> To use the robust version with the LKB, bring up a Lisp image with the LKB, and then before loading the grammar, set a global variable which tells the LKB to make alternate choices when loading, mostly in loading additional files found in the `educ' subdirectory of the grammar directory:
>  (push :educ *features*)
> Then load the grammar as usual (either "1212" as here, or the `trunk' version):
>  (read-script-file-aux "~/logon/lingo/erg/lkb/script")
> 
> If you experiment with it, I'll be glad to get your reactions on how well it might serve your purposes.
> 
> Best,
> 
> Dan
> 
> 
> 
> 
> ----- Original Message -----
> From: "Emily M. Bender" <ebender at uw.edu>
> To: "David Mott" <mottdh at googlemail.com>, erg at delph-in.net
> Sent: Monday, September 16, 2013 12:36:45 PM
> Subject: Re: [erg] Missing article in a sentence causes a problem in parsing?
> 
> [Keeping this on-list]
> 
> Hi David,
> 
> Dan has some "mal-rules", which are used in the grammar checking
> application among other things, and which surely include a rule for
> missing determiners.  I don't know the fine details of how to enable them
> (and what to do about parse selection in case they are enabled).
> 
> Emily
> 
> 
> On Mon, Sep 16, 2013 at 11:31 AM, David Mott <mottdh at googlemail.com> wrote:
>> Emily
>> 
>> Thanks. In this context "male" is intended to be a count noun, its just that
>> the writer got sloppy and omitted the determiner.
>> 
>> I will have a go, but I wonder if there is a deeper question here, of how to
>> handle "sloppy" style?.
>> 
>> David
>> 
>> 
>> On 16 September 2013 18:26, Emily M. Bender <ebender at uw.edu> wrote:
>>> 
>>> Hi David,
>>> 
>>> The problem isn't the "adjectival nature" but that the
>>> lexical entry for the noun "male" treats it as a count
>>> noun, i.e. one which requires a determiner when used in
>>> the singular.  So yes, a new entry would fix this.
>>> 
>>> Emily
>>> 
>>> 
>>> On Fri, Sep 13, 2013 at 9:39 AM, David Mott <mottdh at googlemail.com> wrote:
>>>> Hi
>>>> 
>>>> I have noticed the following behaviour of the ERG in different variants
>>>> of a
>>>> sentence:
>>>> 
>>>> call from a male to a female   ( parses)
>>>> call from man to a female      ( parses)
>>>> 
>>>> call from male to a female     ( *** does not parse)
>>>> 
>>>> 
>>>> Is this because of the adjectival nature of "male"? Do I need to make a
>>>> new
>>>> lexical entry for "male"?
>>>> 
>>>> thanks
>>> 
>>> 
>>> 
>>> --
>>> Emily M. Bender
>>> Associate Professor
>>> Department of Linguistics
>>> Check out CLMS on facebook! http://www.facebook.com/uwclma
>> 
>> 
> 
> 
> 
> -- 
> Emily M. Bender
> Associate Professor
> Department of Linguistics
> Check out CLMS on facebook! http://www.facebook.com/uwclma