def compare_same_duration_ordinal(dataset, narrative, rel_diff=0.1):
questions = [
'Was the <O1> [sound event,sound] [roughly,approximately] as <D> as the <O2> [sound event,sound]?', # noqa: E501
'Was the <O1> and <O2> [sound events,sounds] [roughly,approximately] as <D>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> [sound event,sound] and the <O2> [sound event,sound], were they [roughly,approximately] as <D>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> [sound event,sound] and the <O2> [sound event,sound], did they [roughly,approximately] have the same duration?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> and <O2> [sound events,sounds], were they [roughly,approximately] as <D>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> and <O2> [sound events,sounds], did they [roughly,approximately] have the same duration?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number_1, ordinal_1 = sample_number(len(lst_events))
duration = sample_duration()
number_2, ordinal_2 = sample_second_number(len(lst_events), number_1)
assert number_1 != number_2, 'Question (compare_same_duration_ordinal) illposed.'
question = question.replace('<O1>', ordinal_1) # insert ordinal
question = question.replace('<D>', duration) # insert duration
question = question.replace('<O2>', ordinal_2) # insert ordinal
question = sanitize_question(question) # correct grammar
lst_duration = get_lst_durations(narrative)
e_1_duration = lst_duration[number_1 - 1]
e_2_duration = lst_duration[number_2 - 1]
rel_duration_diff = compute_rel_diff(np.array(e_1_duration),
np.array(e_2_duration))
# Assert a good margin in relative duration
assert np.sum(np.logical_and(rel_duration_diff > rel_diff,
rel_duration_diff < (2 * rel_diff))) <= 0, \
'Question (compare_same_duration_ordinal) illposed.'
answer = 'yes' if rel_duration_diff <= rel_diff else 'no'
return question, answer
def what_was_duration_relative_ordinal(dataset, narrative, rel_diff=0.1):
questions = [
'What was the <AD> sound <RO> the <O> sound?',
'What was the <AD> sound <RO> [hearing,listening to] the <O> sound?',
'What was the <AD> sound <RO> the <O> sound was heard?',
]
question = str(np.random.choice(questions)) # sample question
duration = sample_absolute_duration()
preposition = sample_preposition()
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<AD>', duration) # insert duration
question = question.replace('<RO>', preposition) # insert preposition
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
lst_durations = get_lst_durations(narrative)
event_idx = (number - 1)
answer = None
if 'before' in question:
if (event_idx - 1) < 0:
answer = 'nothing'
else:
lst_events_e = lst_events[:event_idx]
lst_events_d = lst_durations[:event_idx]
elif 'after' in question:
if (event_idx + 1) >= len(lst_events):
answer = 'nothing'
else:
lst_events_e = lst_events[(event_idx + 1):]
lst_events_d = lst_durations[(event_idx + 1):]
else:
assert False, \
'Preposition illdefined in Question (what_was_duration_relative_ordinal).'
if answer is None:
assert len(lst_events_e) > 0, \
'Question (what_was_duration_relative_ordinal) illposed.'
if 'long' in question:
est = np.argmax(lst_events_d)
elif 'short' in question:
est = np.argmin(lst_events_d)
else:
assert False, \
'Duration illdefined in Question (what_was_duration_relative_ordinal).'
# Assert a good margin in relative duration
evt_duration = lst_events_d[est]
x_durations = [j for i, j in enumerate(lst_events_d) if i != est]
rel_duration_diff = compute_rel_diff(np.array(x_durations),
np.array(evt_duration))
assert np.sum(rel_duration_diff < rel_diff) <= 0, \
'Question (what_was_duration_relative_ordinal) illposed.'
e = lst_events_e[est]
answer = (str(np.random.choice(dataset['sources'][e])) + ' ' +
str(np.random.choice(dataset['actions'][e])))
return question, answer
def how_many_ordinal(dataset, narrative, _):
questions = ['How many times did you [hear,listen to] a sound that [sounded,seemed] like the <O> [sound event,sound]?', # noqa: E501
'What is the number of times did you [hear,listen to] a sound that [sounded,seemed] like the <O> [sound event,sound]?', # noqa: E501
'[Hearing,Listening to] the <O> [sound event,sound], how many sounds were [the same, similar]?', # noqa: E501
'[Hearing,Listening to] the <O> [sound event,sound], what is the number of sounds that were [the same, similar]?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
event = lst_events[number - 1]
answer = numbers_to_words(lst_events.count(event) - 1) # -1 for base event
return question, answer
def what_was(dataset, narrative, _):
questions = [
'What was the <O> sound you [heard,listened to]?',
'What was the <O> sound?',
'What did the <O> sound [sound,seem] like?',
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
event = lst_events[number - 1]
answer = (str(np.random.choice(dataset['sources'][event])) + ' ' +
str(np.random.choice(dataset['actions'][event])))
return question, answer
def was_there_similar_ordinal(dataset, narrative, _):
questions = [
'Were there any similar sounds to the <O> sound?',
'Were there any sounds that were similar to the <O> sound?',
'Was there at least a sound similar to the <O> sound?',
'Was there at least a sound that was similar to the <O> sound?', # noqa: E501
'Was there at least [one,a single] sound similar to the <O> sound?',
'Was there at least [one,a single] sound that was similar to the <O> sound?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
event = lst_events[number - 1]
answer = 'yes' if lst_events.count(event) > 1 else 'no' # 1 for reference
return question, answer
def compare_ordinal(dataset, narrative, _):
questions = [
'Was the <O1> [sound event,sound] [the same as,similar to] the <O2> [sound event,sound]?', # noqa: E501
'Was the <O1> [sound event,sound] and <O2> [sound event,sound] [the same,similar]?', # noqa: E501
'Were the <O1> and <O2> [sound events,sounds] [the same,similar]?',
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number_1, ordinal_1 = sample_number(len(lst_events))
number_2, ordinal_2 = sample_second_number(len(lst_events), number_1)
assert number_1 != number_2, 'Question (compare_ordinal) illposed.'
question = question.replace('<O1>', ordinal_1) # insert ordinal
question = question.replace('<O2>', ordinal_2) # insert ordinal
question = sanitize_question(question) # correct grammar
answer = 'yes' if lst_events[number_1 - 1] == lst_events[number_2 - 1] \
else 'no'
return question, answer
def how_many_sounds_relative_ordinal(dataset, narrative, _):
questions = ['How many [sound events,sounds] after the <O> [sound event,sound] were there?', # noqa: E501
'How many [sound events,sounds] after the <O> [sound event,sound] [did,could] you [hear,listen to]?', # noqa: E501
'How many [sound events,sounds] after the <O> [sound event,sound] have you [heard,listened to]?', # noqa: E501
'What is the number of [sound events,sounds] after the <O> [sound event,sound]?', # noqa: E501
'What is the number of [sound events,sounds] after the <O> [sound event,sound] [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of [sound events,sounds] after the <O> [sound event,sound] have you [heard,listened to]?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
assert number < (len(lst_events) - 1), \
'Question (how_many_sounds_relative_ordinal) illposed.'
lst_events_e = lst_events[number:]
answer = numbers_to_words(len(lst_events_e))
return question, answer
def compare_duration_ordinal(dataset, narrative, rel_diff=0.1):
questions = [
'Was the <O1> [sound event,sound] <RD> than the <O2> [sound event,sound]?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> [sound event,sound] and the <O2> [sound event,sound], was the former <RD>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O1> and <O2> [sound events,sounds], was the former <RD>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O2> [sound event,sound] and the <O1> [sound event,sound], was the latter <RD>?', # noqa: E501
'[Comparing,Listening to,Hearing] the <O2> and <O1> [sound events,sounds], was the latter <RD>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number_1, ordinal_1 = sample_number(len(lst_events))
rel_duration = sample_rel_duration()
number_2, ordinal_2 = sample_second_number(len(lst_events), number_1)
assert number_1 != number_2, 'Question (compare_duration_ordinal) illposed.'
question = question.replace('<O1>', ordinal_1) # insert ordinal
question = question.replace('<RD>', rel_duration) # insert duration
question = question.replace('<O2>', ordinal_2) # insert ordinal
question = sanitize_question(question) # correct grammar
lst_duration = get_lst_durations(narrative)
e_1_duration = lst_duration[number_1 - 1]
e_2_duration = lst_duration[number_2 - 1]
# Assert a good margin in relative duration
rel_duration_diff = compute_rel_diff(np.array(e_1_duration),
np.array(e_2_duration))
assert np.sum(rel_duration_diff < rel_diff) <= 0, \
'Question (compare_duration_ordinal) illposed.'
if 'short' in question:
answer = 'yes' if e_1_duration < e_2_duration else 'no'
elif 'long' in question:
answer = 'yes' if e_1_duration > e_2_duration else 'no'
else:
assert False, 'Duration illdefined in Question (compare_duration_ordinal).'
return question, answer
def how_many_sounds_duration_ordinal(dataset, narrative, rel_diff=0.1):
questions = ['How many [sound events,sounds] [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'How many [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'How many [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound [did,could] you [hear,listen to]?', # noqa: E501
'How many [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound have you heard?', # noqa: E501
'How many [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound?', # noqa: E501
'How many [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound [did,could] you [hear,listen to]?', # noqa: E501
'How many [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound have you heard?', # noqa: E501
'What is the number of [sound events,sounds] [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'What is the number of [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'What is the number of [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of [sound events,sounds] that are [roughly,approximately] as <D> as the <O> sound have you heard?', # noqa: E501
'What is the number of [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound?', # noqa: E501
'What is the number of [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of [sound events,sounds] that have [roughly,approximately] the same duration as the <O> sound have you heard?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
duration = sample_duration() # sample duration
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<D>', duration) # insert duration
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
lst_durations = get_lst_durations(narrative)
evt_duration = lst_durations[number - 1]
x_durations = [j for i, j in enumerate(lst_durations) if i != (number - 1)]
rel_durations_diff = compute_rel_diff(np.array(x_durations),
np.array(evt_duration))
# Assert a good margin in relative loudness
assert np.sum(np.logical_and(rel_durations_diff > rel_diff,
rel_durations_diff < (2 * rel_diff))) <= 0, \
'Question (how_many_sounds_duration_ordinal) illposed.'
answer = numbers_to_words(np.sum(rel_durations_diff <= rel_diff))
return question, answer
def compare_ordinal_event(dataset, narrative, _):
questions = [
'Was the <O> [sound event,sound] [a,an] <S> <A>?', # noqa: E501
'Did the <O> [sound event,sound] [sound,seem] like [a,an] <S> <A>?', # noqa: E501
'[Listening to,Hearing] the <O> [sound event,sound], was it [a,an] <S> <A>?', # noqa: E501
'[Listening to,Hearing] the <O> [sound event,sound], did it [sound,seem] like [a,an] <S> <A>?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
event = str(np.random.choice(dataset['events'])) # sample event
source = str(np.random.choice(dataset['sources'][event])) # sample source
action = str(np.random.choice(dataset['actions'][event])) # sample action
question = question.replace('<O>', ordinal) # insert ordinal
question = question.replace('<S>', source) # insert source
question = question.replace('<A>', action) # insert action
question = sanitize_question(question) # correct grammar
answer = 'yes' if lst_events[number - 1] == event else 'no'
return question, answer
def was_there_similar_duration_ordinal(dataset, narrative, rel_diff=0.1):
questions = [
'Were there any sounds [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'Were there any sounds that were [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'Were there any sounds that were [roughly,approximately] the same duration as the <O> sound?', # noqa: E501
'Was there any sound [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'Was there any sound that was [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'Was there any sound that was [roughly,approximately] the same duration as the <O> sound?', # noqa: E501
'Was there at least a sound [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'Was there at least a sound that was [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'Was there at least a sound that was [roughly,approximately] the same duration as the <O> sound?', # noqa: E501
'Was there at least [one,a single] sound that was [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'Was there at least [one,a single] sound [roughly,approximately] as <D> as the <O> sound?', # noqa: E501
'Was there at least [one,a single] sound that was [roughly,approximately] the same duration as the <O> sound?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
duration = sample_duration() # sample duration
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<D>', duration) # insert duration
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
lst_durations = get_lst_durations(narrative)
evt_duration = lst_durations[number - 1]
x_durations = [j for i, j in enumerate(lst_durations) if i != (number - 1)]
rel_durations_diff = compute_rel_diff(np.array(x_durations),
np.array(evt_duration))
# Assert a good margin in relative duration
assert np.sum(np.logical_and(rel_durations_diff > rel_diff,
rel_durations_diff < (2 * rel_diff))) <= 0, \
'Question (was_there_similar_duration_ordinal) illposed.'
answer = 'yes' if np.sum(rel_durations_diff <= rel_diff) >= 1 else 'no'
return question, answer
def how_many_event_relative_ordinal(dataset, narrative, _):
questions = ['How many <S>s <A> <RO> the <O> [sound event,sound] were there?',
'How many <S>s <A> <RO> the <O> [sound event,sound] [did,could] you [hear,listen to]?', # noqa: E501
'How many <S>s <A> <RO> the <O> [sound event,sound] have you [heard,listened to]?', # noqa: E501
'What is the number of <S>s <A> <RO> the <O> [sound event,sound]?',
'What is the number of <S>s <A> <RO> the <O> [sound event,sound] [did,could] you [hear,listen to]?', # noqa: E501
'What is the number of <S>s <A> <RO> the <O> [sound event,sound] have you [heard,listened to]?', # noqa: E501
]
question = str(np.random.choice(questions)) # sample question
event = str(np.random.choice(dataset['events'])) # sample event
source = str(np.random.choice(dataset['sources'][event]))
action = str(np.random.choice(dataset['actions'][event]))
preposition = sample_preposition()
lst_events = get_lst_events(narrative)
number, ordinal = sample_number(len(lst_events))
question = question.replace('<S>', source) # insert source
question = question.replace('<A>', action) # insert action
question = question.replace('<RO>', preposition) # insert preposition
question = question.replace('<O>', ordinal) # insert ordinal
question = sanitize_question(question) # correct grammar
if 'before' in question:
assert number > 1, 'Question (how_many_event_relative_ordinal) illposed.'
lst_events_e = lst_events[:(number - 1)]
elif 'after' in question:
assert number < (len(lst_events) - 1), \
'Question (how_many_event_relative_ordinal) illposed.'
lst_events_e = lst_events[number:]
else:
assert False, \
'Relative preposition illdefined in Question (how_many_event_relative_ordinal).' # noqa: E501
answer = numbers_to_words(lst_events_e.count(event))
return question, answer