Support for users of CCG software closed 7-27-20

[Sourajit Basak <sourajit.basak AT gmail.com>]

hi,
As an alternate, I was looking at the shallow parser (chunker) developed at http://cogcomp.cs.illinois.edu/page/software_view/13. If we use that we need to convert the POS tagged input to column format for which there is a perl script.

Given an input "Why choose Ruby on Rails ?", the chunker outputs [ADVP Why ] [VP choose ] [NP Ruby ] [PP on ] [NP Rails ] ? However, the same cannot be used by the perl script

C:\Projects\learning>perl chunks_to_cols.pl pos_tagged
Error: input doesn't match prescribed format. at chunks_to_cols.pl line 169, <CHUNK> line 1.

However, if I convert it to [(ADVP Why) (VP choose) (NP Ruby) (PP on) (NP Rails) ()] it works.
Note that in order to process the last word, I had to add a blank ().

It will not be difficult to do this conversion programatically, just curious to know if anyone has faced this problem.

This does not seem to be look like a full fledged FDG parser o/p because it lacks the role mappings. What is your opinion on using it ?


O       0       0       B-(ADVP          Why      x       0       0
O       0       1       I-(ADVP VP      choose  x       0       0
O       0       2       I-(ADVP NP      Ruby     x       0       0
O       0       3       I-(ADVP PP      on         x       0       0
O       0       4       I-(ADVP NP      Rails     x       0       0

rgds,
Sourajit Basak

On Tue, Dec 28, 2010 at 9:01 PM, Mark Sammons <mssammon AT illinois.edu> wrote:

Hi, Sourajit.

Fex uses a non-standard representation for input of dependency paths etc.
I don't know of any parsers that provide output that can be used without
modification.

Fex uses sensors in its Sensors.cpp file to define feature generation functions:
to extend it to accept your output, you'd need to write sensors that process
the relevant column info.  Fex reads input as whitespace-separated columns,
so when you generate the column format you need to keep that in mind.

Regards,

Mark

---- Original message ----
>Date: Tue, 28 Dec 2010 12:53:54 +0530
>From: Sourajit Basak <sourajit.basak AT gmail.com>
>Subject: [Illinois-ml-nlp-users] Which FDG Parsers are compatible with Fex ?
>To: illinois-ml-nlp-users AT cs.uiuc.edu
>
>   I am working on an experimental chatbot and planning
>   to use FEX.
>
>   Can I use the output from Connexor's FDG Parser
>   verbatim as the corpus to Fex ? The FDG parser demo
>   can be found online at
>   http://www.connexor.eu/technology/machinese/demo/
>   Here is the output from "A chatbot is a software
>   program that can talk to human beings"
>

>    #   Text       Baseform   Syntactic   Syntax and
>                              relation    morphology

>    1  A         a         det:>2     @DN> %>N DET
>                                          SG  

>                                          @SUBJ %NH
>    2  chatbot   chatbot   subj:>3    Heur N NOM SG

>                                           
>    3  is        be        main:>0    @+FMAINV %VA
>                                          V PRES SG3  
>    4  a         a         det:>6     @DN> %>N DET
>                                          SG  
>    5  software  software  attr:>6    @A> %>N N NOM
>                                          SG  
>    6  program   program   comp:>3    @PCOMPL-S %NH
>                                          N NOM SG  
>    7  that      that      subj:>8    @SUBJ %NH
>                                          <Rel> PRON  
>    8  can       can       v-ch:>9    @+FAUXV %AUX
>                                          V AUXMOD  
>    9  talk      talk      mod:>6     @-FMAINV %VA
>                                          V INF  
>   10  to        to        ha:>9      @ADVL %EH
>                                          PREP  
>   11  human     human     attr:>12   @A> %>N A ABS
>                                           
>   12  beings    being     pcomp:>10  @<P %NH N NOM
>                                          PL  
>   13  .         .  
>   14  <s>       <s>  
>
>   If not, what FDG parsers are usable with Fex. I am
>   looking for the ones that do not require any manual
>   modification of the output.
>
>   rgds,
>   Sourajit Basak
>   India

>________________
>_______________________________________________
>illinois-ml-nlp-users mailing list
>illinois-ml-nlp-users AT cs.uiuc.edu
>http://lists.cs.uiuc.edu/mailman/listinfo/illinois-ml-nlp-users