let's standardize!

[yhamoudi]

A first way to produce a standardized tree from triples.

treeTranslation.py contains the algo. I think some code could be factorized (but i have errors whenever i try something)

If you want to try it:

• run demo6.py (in /demo folder)
• enter your question

If someone is comfortable with the filling of requesthandler.py (and the other parts to communicate with external modules), please do it :)

[progval]

you shouldn't have to edit ppp_nlp_classical/__init__.py, this is the file used to expose stuff to external software, and only the app is enough.

[progval]

And the communication protocol has been changing a lot yesterday and today, so I suggest we wait a day or two before starting working on the implementation of requesthandler. All I need to implement it is a method that takes a string as input and returns:

• (mandatory) a tree, as defined in the data model
• (optional) measures

[Ezibenroc]

Valentin -> We have modified __init__.py to do our unit tests.

[progval]

I tried to type four different questions, and all of them get exit: question word has child (please, report your sentence on http://goo.gl/EkgO5l) as a reply. I am using it wrong?

[Ezibenroc]

What questions? What script?

[progval]

demo6.py, and the questions I wrote on the pad:

Who am I?
Who are you?
What is my birth date?
What is the birth date of the first president of the United States?

[yhamoudi]

What branch did you used? All of these questions work on triple_standardize (but is seems there is a bug on non_question_word_questions branch)

[Ezibenroc]

I think that you did not launch the server with the option for the copula. Use this to launch the server:

CORENLP="stanford-corenlp-full-2014-08-27" CORENLP_OPTIONS="-parse.flags \" -makeCopulaHead\"" python3 -m corenlp

(eventually replacing the option CORENLP by the appropriate directory)

[progval]

I used triple_standardize.

[Ezibenroc]

With the above command line to launch the server, the four questions are ok for me.

[progval]

It works for How old are you?, though

[progval]

oh I see, I forgot the copula stuff

[progval]

That format looks quite good, very nice!

However, I am not sure I understand this output:

progval@ganymede:~/etudes/ens/ppp/nlp-classical(triple_standardize)$ echo "Who are you?" | python3 demo/demo6.py 
{
    "object": {
        "type": "missing"
    },
    "subject": {
        "object": {
            "value": "you",
            "type": "resource"
        },
        "subject": {
            "type": "missing"
        },
        "type": "triple",
        "predicate": {
            "value": "are",
            "type": "resource"
        }
    },
    "type": "triple",
    "predicate": {
        "value": "identity",
        "type": "resource"
    }
}

Should the whole first-level subject be {"type": "resource", "value": "you"}?

[Ezibenroc]

Yassine: for the question Who are you?, the tree before simplification has a node are as ROOT child. I think that we should add the following as question words: Who is, Who are, Who am, Who were, Who was...

Other solution: do not launch the server with copula option, and merge the copula dependency.

[yhamoudi]

I don't understand the problem (2 last comments). You want another triples for Who are you ?

[progval]

Yes, I think this one would make more sens:

{
    "object": {
        "type": "missing"
    },
    "subject": {
        "value": "you",
        "type": "resource"
    },
    "type": "triple",
    "predicate": {
        "value": "identity",
        "type": "resource"
    }
}

[yhamoudi]

No, I think it's relevant. The second level will identify you (Valentin) and the first one will provide a description/identity of you.

In the case of Who questions it can seem a bit too difficult. But if you replace Who by Where it performs just as you want : ((?,are,you), location, ?)

The current algorithm just removes the question word, product some triples, and then add an extra triple depending on the question word (see https://github.com/ProjetPP/PPP-NLP-classical/issues/14).

But perhaps it could be interessant to add question words such as Who are in order to delete a level.

[progval]

Oh, I see

[Ezibenroc]

We should merge if you are ok (and do a new branch for the part about the requestHandler).

[yhamoudi]

as you want, i've nothing to add on this branch