Debug ressources

Sample of headwords extractions

To get a sample of 100 headword extractions into a pickle file, use:

./headwordextractor.py -s 100 sample_file

Each sample contains three lines which are:

• the real argument
• the closest matching node of the syntactic tree of this sentence
• the headword

Dump module

The dump module is a script which facilitate the visualisation of the impact of a code modification or of an option on the results.

The –dump option allows the user to dump the state of the algorithm after the frame-matching and after the probability model instead of displaying statistics, and the dumper module can display the differences between two dump files when called directly.

By default, only the differences that have an impact on the performance statistics are displayed (ie, slots that have a good state like one_role_good in one dump and a bad state like no_role in another). To display every difference, use the --all option of dumper.py.

For instance, to see the exact differences between the slot and the slot-class probability models:

./main.py --model=slot_class --dump test1
./main.py --model=slot --dump test2
./dumper.py test1 test2

Outputs:

Differences after frame matching:
Good -> bad:
Bad -> good:


Differences after probability model:
Good -> bad:


Sentence: The stimulus package will jumpstart the economy and make sure it
does n't slide into recession again .
Predicate: slide
Argument: into recession
Frame structure: ['NP', 'V', 'into', 'NP']
Frame name: Motion
Role list: {'Location'}
Correct role: Goal (FrameNet) -> {'Location'} (VerbNet)
Status: one_role_good

Sentence: The stimulus package will jumpstart the economy and make sure it
does n't slide into recession again .
Predicate: slide
Argument: into recession
Frame structure: ['NP', 'V', 'into', 'NP']
Frame name: Motion
Role list: {'Result'}
Correct role: Goal (FrameNet) -> {'Location'} (VerbNet)
Status: one_role_bad



Sentence: The stimulus package will jumpstart the economy and make sure it
does n't slide into recession again .
Predicate: slide
Argument: into recession
Frame structure: ['NP', 'V', 'into', 'NP']
Frame name: Motion
Role list: {'Location'}
Correct role: Goal (FrameNet) -> {'Location'} (VerbNet)
Status: one_role_good

Sentence: The stimulus package will jumpstart the economy and make sure it
does n't slide into recession again .
Predicate: slide
Argument: into recession
Frame structure: ['NP', 'V', 'into', 'NP']
Frame name: Motion
Role list: {'Result'}
Correct role: Goal (FrameNet) -> {'Location'} (VerbNet)
Status: one_role_bad

(...)

Bad -> good:


Sentence: But Jamaica is not simply turning blindly into a small version of
its bigger brother .
Predicate: turn
Argument: into a small version of its bigger brother
Frame structure: ['NP', 'V', 'into', 'NP']
Frame name: Undergo_change
Role list: {'Location'}
Correct role: Final_category (FrameNet) -> {'Result'} (VerbNet)
Status: one_role_bad

Sentence: But Jamaica is not simply turning blindly into a small version of
its bigger brother .
Predicate: turn
Argument: into a small version of its bigger brother
Frame structure: ['NP', 'V', 'into', 'NP']
Frame name: Undergo_change
Role list: {'Result'}
Correct role: Final_category (FrameNet) -> {'Result'} (VerbNet)
Status: one_role_good



Sentence: Right , yeah , I heard about that on the news , yeah .
Predicate: hear
Argument: about that
Frame structure: ['NP', 'V', 'about', 'NP']
Frame name: Perception_experience
Role list: {'Theme'}
Correct role: Phenomenon (FrameNet) -> {'Stimulus'} (VerbNet)
Status: one_role_bad

Sentence: Right , yeah , I heard about that on the news , yeah .
Predicate: hear
Argument: about that
Frame structure: ['NP', 'V', 'about', 'NP']
Frame name: Perception_experience
Role list: {'Stimulus'}
Correct role: Phenomenon (FrameNet) -> {'Stimulus'} (VerbNet)
Status: one_role_good