panoptes (solution for fb.ai/babi 20 QA tasks)

See also: http://jamesknighton.com/2015/babi/

1. Mind

Organization

panoptes/mind/
├── idea
│   ├── base.py
│   ├── clause.py
    def __init__(self, status=Status.ACTUAL, purpose=Purpose.INFO,
                 is_intense=False, verb=None, adverbs=None, rel2xxx=None):
│   ├── comparative.py
    def __init__(self, polarity, adjective, than_x):

│   ├── direction.py
│   ├── __init__.py
│   ├── noun.py
    def __init__(self, query=None, name=None, gender=None,
                 is_animate=None, attributes=None, kind=None,
                 rel2xxx=None, carrying=None):
│   ├── reverb.py
│   └── time.py
├── __init__.py
├── know
│   ├── cause_and_effect.yaml
│   ├── cause_effect.py
│   ├── graph.py
│   ├── __init__.py
│   ├── location.py
│   ├── memory.py
│   └── user.py
├── mind.py
└── verb
    ├── base.py
    ├── be.py
    ├── carry.py
    ├── fear.py
    ├── fit.py
    ├── give.py
    ├── go.py
    ├── __init__.py
    └── manager.py

Flow

        checkpoint = self.memory.make_checkpoint()

        from_xx = list(map(self.user_mgr.get, from_uids))
        to_xx = list(map(self.user_mgr.get, to_uids))
        x = self.memory.decode_dsen(dsen, from_xx, to_xx)

        if x is None:
            self.memory.load_checkpoint(checkpoint)
            return None

        c = self.memory.ideas[x]
        r = self.semantics_mgr.handle(c)

        if not r:
            self.memory.load_checkpoint(checkpoint)
            return None

        return r

2. Language

Recognize

Text >> parse >> Parse >> recog >> S-Structure >> unfront >> D-Structure >> resolve >> Memory

Generate

Text << contract << Tokens << render << S-Structure << front << D-Structure << refer << Memory

3. Python example ("headless" mode)

Setup

from contextlib import contextmanager
import json
from time import sleep
from tqdm import trange
import yaml

from panoptes.ling.english import English
from panoptes.ling.glue.idiolect import Idiolect
from panoptes.ling.tree.arg_loader import ArgLoader
from panoptes.ling.tree.deep.sentence import DeepSentence
from panoptes.ling.verb.verb import DeepVerb

@contextmanager
def step(name):
    indent = ' ' * 4
    print()
    print(indent + name)
    print(indent + '-' * len(name))
    print()
    yield
    sleep(1.337)

dump = lambda x: json.dumps(x, indent=4, sort_keys=True)
dump = lambda x: yaml.safe_dump(x) 

loader = ArgLoader()
english = English()

Get the verb

conf = {
    'lemma': 'be',
    'polarity': {
        'tf': True,
        'is_contrary': False,
    },
    'tense': 'PRESENT',
    'aspect': {
        'is_perf': False,
        'is_prog': False,
    },
    'modality': {
        'flavor': 'IMPERATIVE',
        'is_cond': False,
    },
    'verb_form': 'FINITE',
    'is_pro_verb': False,
}
be = DeepVerb.load(conf)
print(dump(be))

Get the implied subject

conf = {
    'type': 'PersonalPronoun',
    'declension': 'YALL',
    'ppcase': 'SUBJECT',
}
yall = loader.load(conf)
print(dump(yall))

Get a locative verb argument

conf = {
    'type': 'DeepCommonNoun',
    'possessor': None,
    'selector': {
        'correlative': 'PROX',
        'n_min': 'SING',
        'n_max': 'SING',
        'of_n_min': 'SING',
        'of_n_max': 'SING',
    },
    'number': None,
    'attributes': [],
    'noun': 'place',
    'rels_nargs': [],
}
here = loader.load(conf)
print(dump(here))

Get a temporal verb argument

conf = {
    'type': 'DeepCommonNoun',
    'possessor': None,
    'selector': {
        'correlative': 'PROX',
        'n_min': 'SING',
        'n_max': 'SING',
        'of_n_min': 'SING',
        'of_n_max': 'SING',
    },
    'number': None,
    'attributes': [],
    'noun': 'time',
    'rels_nargs': [],
}
now = loader.load(conf)
print(dump(now))

Now connect them together into a content clause

conf = {
    'type': 'DeepContentClause',
    'status': 'ACTUAL',
    'purpose': 'INFO',
    'is_intense': False,
    'verb': be.dump(),
    'adverbs': [],
    'rels_vargs': [
        ('AGENT', yall.dump()),
        ('PLACE', here.dump()),
        ('TIME', now.dump()),
    ],
    'subj_index': 0,
}
clause = loader.load(conf)
print(dump(clause))

That clause is our sentence

conf = {
    'type': 'DeepSentence',
    'root': conf,
}
sentence = DeepSentence.load(conf, loader)
print(dump(sentence))

Configure how it will be said

conf = {
    'archaic_pro_adverbs': False,
    'contractions': True,
    'pedantic_plurals': False,
    'stranding': True,
    'split_infinitive': True,
    'subjunctive_were': True,
    'whom': False,
}
idiolect = Idiolect(**conf)
dump(idiolect.__dict__)

Generate

text = english.say(sentence, idiolect)
print('$', text)  # "Be here now."