Integrating PropS proposition extraction

[gabrielStanovsky]

@kleinay This is a first pull request adding the ability to get the Json objects we discussed. You can see a usage example in the "main" of the PropS wrapper, and also an example output json.

Please skim these changes, and feel free to ask to change anything that you think can be done better, or ask for clarifications where needed.

Let's discuss these changes here, and decide together when we want to merge them upstream.

Thanks!

[kleinay]

regarding the "Template" atribute of a predicate: I wonder whether there is a way to facilitate the usage of such template (replacing signs in a string demands a bit proccessing).

[gabrielStanovsky]

Changes in c789353729557bda13fe0fb5e31c6467b8c4625b:

• Template - placeholders for predicates and arguments appear with curly braces, so you can replace by using format with dictionary:

  "{A1} forced to land in {A3}".format(**{"A1" : "Plane", "A3" : "Moscow"})

• Added the features you requested for the head predicate. These will appear as:

  'P1': {"Bare predicate": "was forced",
                     "Head": {"Lemma": "force",
                              "POS": "VBN",
                              "Surface": "forced"},
                     "Template": "{A1} was forced {P2}"}

• You can test the single sentence extraction by giving it an input file with raw sentences and an output file where to output a corresponding list of extractions (json format). Try:

  python ./props_wrapper.py --in=../../../examples/raw_sentences --out=../../../examples/raw_sentences.json

Additionally, in 22e4849d4d2b4d7294baa69438b3610b71eb5d05 I added list of word indices for both predicates and entities, this usually comes in handy at some point :) So now, everything that includes surface words should be a tuple of words and indices:

{'Entities': {'A1': ('The Syrian plane', (0, 1, 2)), 'A2': ('Moscow', (8,))},
 'Predicates': {'P1': {'Bare predicate': ('was forced', (3, 4)),
                       'Head': {'Lemma': u'force',
                                'POS': 'VBN',
                                'Surface': ('forced', 4)},
                       'Template': '{A1} was forced {P2}'},
                'P2': {'Bare predicate': ('to land in', (5, 6, 7)),
                       'Head': {'Lemma': 'land',
                                'POS': 'VB',
                                'Surface': ('land', 6)},
                       'Template': '{A1} to land in {A2}'}},
 'Sentence': 'The Syrian plane was forced to land in Moscow .'}

[kleinay]

I have another change request for the Predicate data - please add a "Arguments" attribute, containing the symbols in the template. e.g. for predicate with template '{A1} to land in {A2}', add "Arguments":['A1', 'A2'] to its information. should be easy.

[gabrielStanovsky]

Done, please review and approve here: https://github.com/vered1986/OKR/pull/5