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

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

hi Mark,
I did go through these tutorials which helped my understanding of a feature space. But I think those do not contain illustrative examples of using FeX.

I am actually extracting feature space for question recognition. I am particularly interested in  -
2) What will be the feature rule in FeX for POS tags around the target ?
3) What will be the feature for non overlapping phrases around the target ?

If I am allowed, I am willing to enhance the tutorial after I am conversant with FeX.

rgds,
Sourajit

On Fri, Mar 18, 2011 at 9:28 AM, Mark Sammons <mssammon AT illinois.edu> wrote:

Hi, Sourajit.

You might find the fex/snow tutorial helpful:

http://cogcomp.cs.illinois.edu/page/tutorial

-- what you are trying to do looks like the approach taken in the context-sensitive
spell checker.  You specify a set of target words, and only those words
generate examples.  However, instead of specifying the words in the script file,
you use the "-t" option and name the target words. Other applications are also
covered in the tutorial.

Regards,

Mark

---- Original message ----
>Date: Tue, 15 Mar 2011 13:58:28 +0530
>From: Sourajit Basak <sourajit.basak AT gmail.com>
>Subject: query on FeX language
>To: Mark Sammons <mssammon AT illinois.edu>, illinois-ml-nlp-users AT cs.uiuc.edu
>
>   hello,
>
>   I am trying to extract features as mentioned in the
>   following examples. The attached file contains the
>   sentence (question) and its POS annotated
>   representation. I have the following queries.
>
>   1)  Feature: One word each around the target word
>   Is this rule correct ? -1 loc: w(x=Java) [-1,1]; -1
>   loc: w(x=XML-RPC) [-1,1] I do not see the lexicon
>   and example populated; guess that the rule is not
>   fired. This fired "4 loc: w [-1, -1]" whereas this
>   "Java loc: w [-1, 1]" failed.
>
>   2) What will be the feature rule for POS tags around
>   the target ?
>
>   3) What will be the feature for non overlapping
>   phrases around the target ?
>
>   I am afraid I didn't fully comprehend the FeX
>   manual. A brief explanation for the above will be
>   extremely helpful.
>
>   rgds,
>   Sourajit

>________________
>Preprocessing steps:-
>#1. Example: How to call an XML-RPC method from Java code ?
>
>#2. POS tagger and chunker output
>$$java -Xmx512m edu.illinois.cs.cogcomp.lbj.chunk.ChunksAndPOSTags
<plain_text_file>
>POS Tagged and chunked: [ADVP (WRB How) ] [VP (TO to) (VB call) ] [NP (DT an)
(NNP XML-RPC) (NN method) ] [PP (IN from) ] [NP (NNP Java) (NNP code) ] (. ?)
>
>#3. Columnar output. Used as the actual corpus.
>$$  perl chunks_to_cols.pl <pos_tagged_question>
>
>O       0       0       B-ADVP  WRB     How     x       0       0
>O       0       1       B-VP    TO      to      x       0       0
>O       0       2       I-VP    VB      call    x       0       0
>B-NP    0       3       B-NP    DT      an      x       0       0
>I-NP    0       4       I-NP    NNP     XML-RPC x       0       0
>O       0       5       B-PP    IN      from    x       0       0
>B-NP    0       6       B-NP    NNP     Java    x       0       0
>I-NP    0       7       I-NP    NNP     Code    x       0       0
>O       0       8       O       .       ?       x       0       0