def attributive_of_noun(sense, main_sense_of_entry):
"""
An attrib. sense of a noun is treated as the adj. equivalent
of the main sense of the entry (or of a particular sense, if referenced)
"""
# Use a particular sense if it's cross-referenced
# Check that it's an internal cross-reference to a main sense
# (hence no lemma)
xrefs = [xr for xr in sense.cross_references if
xr.refentry == sense.entry_id and xr.lemma is None]
if xrefs:
target_sense = tdb.highest_ranked(lemma=sense.lemma,
wordclass='NN',
refentry=xrefs[0].refentry,
refid=xrefs[0].refid)
if target_sense is not None and target_sense.thesclass is not None:
equiv = tdb.equivalent_class(target_sense.thesclass, 'JJ')
equiv.reason_text = 'Adjective equivalent of cross-referenced noun sense'
equiv.reason_code = 'attb'
return equiv
elif target_sense is not None:
return None
# ... otherwise, default to the main sense of the entry
if (main_sense_of_entry is not None and
main_sense_of_entry.thesclass is not None):
equiv = tdb.equivalent_class(main_sense_of_entry.thesclass, 'JJ')
equiv.reason_text = 'Adjective equivalent of main noun sense'
equiv.reason_code = 'attb'
return equiv
else:
return None
def infer_derivative(sense, main_sense_of_entry):
equiv = None
if (main_sense_of_entry is not None and
not main_sense_of_entry.is_affix() and
main_sense_of_entry.thesclass is not None):
# Take precedent from the main sense of the parent entry
equiv = tdb.equivalent_class(main_sense_of_entry.thesclass, sense.wordclass)
if equiv is not None:
equiv.reason_text = 'Parallel to "%s"' % main_sense_of_entry.lemma
equiv.reason_code = 'driv'
if equiv is None:
# Take precedent from other sibling derivatives
candidates = tdb.ranked_search(refentry=sense.entry_id,
thes_linked=True, currentOnly=True)
candidates = [c for c in candidates if c.is_derivative() and
not ' ' in c.lemma and
c.superclass() is not None]
if candidates:
j = defaultdict(int)
for c in candidates:
j[c.superclass()] += 1
parents = [p for p, num in j.items() if num == max(j.values())]
# If there's more than one possible parent, pick the one with
# the largest branch
parents.sort(key=lambda p: p.branch_size, reverse=True)
equiv = tdb.child_wordclass_branch(parents[0], sense.wordclass)
if equiv is None:
equiv = parents[0]
if equiv is not None:
equiv.reason_text = 'Parallel to %s' % ', '.join(
['"%s"' % (c.lemma,) for c in candidates if c.lemma])
equiv.reason_code = 'driv'
return equiv
def _append_derivational_classification(best_guesses, sense):
"""
If the compound is a derivative of another compound, prepend or append a
classification based on the root form
"""
base_compound_class, insert_position = compound_derivative(sense)
if base_compound_class is not None:
equiv = tdb.equivalent_class(base_compound_class, sense.wordclass)
if (equiv is not None and
equiv.wordclass is not None and
not any([b.target.is_descendant_of(equiv) for b in best_guesses])):
# Create a new derivative-based guess object
# Give it a dummy score, calculated to be higher or lower
# than the set of existing guesses
if not best_guesses:
score = 1
elif insert_position == 'first':
score = best_guesses[0].target_score * 1.1
else:
score = best_guesses[-1].target_score * 0.9
new_guess = BestGuess(None, equiv, score, 'derivative')
# ... and prepend/append it to the best_guesses list
if insert_position == 'first':
best_guesses.insert(0, new_guess)
elif insert_position == 'last':
best_guesses.append(new_guess)
return best_guesses
def infer_from_etyma(sense, subjectFilter=False):
etymon, target_instance = (None, None)
if len(sense.etyma) == 1 and sense.etyma[0][0] == sense.lemma:
etymon = sense.etyma[0]
elif (len(sense.etyma) == 2 and
re.search(r'^[a-zA-Z]+$', sense.etyma[0][0]) and
re.search(r'^-[a-z]+$', sense.etyma[1][0])):
suffix = sense.etyma[1][0]
if (deriv_tester.is_neutral_suffix(suffix) or
(suffix in ('-ist', '-ian', '-ful') and sense.wordclass == 'JJ')):
etymon = sense.etyma[0]
if etymon is not None:
# First try to find the exact sense, in case the etymon points to a
# specific sense - see e.g. lam n./3
target_instance = tdb.highest_ranked(lemma=etymon[0],
refentry=etymon[1],
refid=etymon[2],
exact_sense=True)
# ...but if the etymon just points to an entry in general, find that
# entry's main sense
if target_instance is None and not subjectFilter:
target_instance = main_sense_finder.main_sense(lemma=etymon[0],
refentry=etymon[1])
elif target_instance is None and subjectFilter:
main_sense = tdb.highest_ranked(lemma=etymon[0],
refentry=etymon[1],
subjects=sense.subjects)
if main_sense is not None and main_sense.entry_size < 100:
target_instance = main_sense
if target_instance is not None:
# Check if the target is also referenced in the sense's
# cross-references (in case a particular sense is pointed to,
# as in 'nocturning').
for xr in sense.cross_references:
if (xr.lemma == target_instance.lemma and
xr.refentry == target_instance.refentry):
specific_target = tdb.highest_ranked(lemma=xr.lemma,
refentry=xr.refentry,
refid=xr.refid,
exact_sense=True)
if specific_target is not None:
target_instance = specific_target
break
if target_instance is not None and target_instance.thesclass is not None:
if target_instance.wordclass == sense.wordclass:
match = target_instance.thesclass.wordclass_parent()
else:
match = tdb.equivalent_class(target_instance.thesclass, sense.wordclass)
if match is not None:
match.reason_code = 'etym'
match.reason_text = 'Analogy with "%s" in etymology' % etymon[0]
return match
else:
return None
def superordinate_adjective_state(self, sense):
if sense.superordinate is not None:
m = re.search(r'^state of being ([a-z-]+)-JJ$', sense.superordinate)
if m is None:
m = re.search(r'^being ([a-z-]+)-JJ state$', sense.superordinate)
if m is not None:
adjective = m.group(1)
target_sense = MAIN_SENSE_FINDER.main_sense(lemma=adjective,
wordclass='JJ')
if (target_sense is not None and
target_sense.thesclass is not None):
return tdb.equivalent_class(target_sense.thesclass, 'NN')
return None
def superordinate_lookup(sense, **kwargs):
match = superordinate_manager.superordinate_lookup(sense, **kwargs)
if match is not None and match.wordclass is not None:
if sense.wordclass == 'JJ':
match = tdb.equivalent_class(match, 'JJ')
match.reason_code = 'adeq'
match.reason_text = 'Adjective equivalent of "%s"' % sense.superordinate
else:
match.reason_code = 'supe'
match.reason_text = 'Classification of superordinate "%s" ("%s")'\
% (sense.superordinate, sense.superordinate_full)
#if sense.wordclass == 'VB':
# print('\n----------------------------------------')
# print(trace_sense(sense))
# print(trace_class(match))
return match
else:
return None
def infer_from_neighbouring_wordclass(sense):
match = None
if sense.wordclass in ('JJ', 'VB'):
opposite_class = 'NN'
elif sense.wordclass in ('NN', 'RB'):
opposite_class = 'JJ'
else:
opposite_class = None
if opposite_class is not None:
opposite = tdb.highest_ranked(lemma=sense.lemma,
refentry=sense.entry_id,
wordclass=opposite_class)
if opposite is not None and opposite.thesclass is not None:
match = tdb.equivalent_class(opposite.thesclass, sense.wordclass)
if match is not None:
match.reason_code = 'nbor'
match.reason_text = 'Inferred from neighbouring %s branch' % opposite_class
return match
def cf_cross_reference(sense):
"""
If the current sense has 'cf'-type cross-reference, the classification
of the target sense is taken as a good guide to the classification
of the current sense.
If the target sense and the current sense are in the same wordclass and
have the same end word-ending, we assume that the current sense should
go in the target sense's class (similar to how 'equals'-type xrefs are
treated.
Otherwise, we put the current sense at the top of the equivalent
wordclass-level branch.
"""
xr = sense.cf_crossreference()
# Nb don't specify a wordclass here, since the target wordclass may well
# be different from the current sense's wordclass
target_senses, sense_count = tdb.cross_reference_target(lemma=xr.lemma,
refentry=xr.refentry, refid=xr.refid)
# Don't attempt if the target is too ambiguous (too many possible senses)
if target_senses and sense_count <= 3:
high_scores = [t for t in target_senses
if t.rating() == target_senses[0].rating()]
high_scores.sort(key=lambda i: i.node_size(), reverse=True)
if high_scores[0].thesclass is not None:
target = high_scores[0]
if (target.wordclass == sense.wordclass and
target.lemma[-2:] == sense.lemma[-2:]):
match = target.thesclass
elif target.wordclass == sense.wordclass:
match = target.wordclass_parent()
else:
match = tdb.equivalent_class(target.thesclass, sense.wordclass)
if match is not None:
match.reason_text = 'Analogy with target of cf-type xref ("%s")' % target.lemma
match.reason_code = 'cfxr'
return match
return None
def find_branch_from_superordinate(self, sense):
"""Classify by finding the main or only sense of the superordinate
"""
if (sense.wordclass not in ('NN', 'JJ') or
not sense.superordinate or
len(sense.superordinate) < 3 or
sense.superordinate in GENERICS):
return None
target_sense = None
# If the superordinate is (more or less) single-sense, we assume that
# sense to be the correct one
candidates = tdb.ranked_search(
lemma=sense.superordinate,
wordclass='NN',
current_only=True)
if candidates and tdb.distinct_senses(candidates) <= 2:
target_sense = candidates[0]
# Otherwise, narrow by Bayes classification
if target_sense is None and sense.bayes.confidence() >= 8:
target_sense = tdb.highest_ranked(
lemma=sense.superordinate,
wordclass='NN',
branches=sense.bayes_based_classifications,
current_only=True)
# Otherwise, narrow by branches based on subject labels
if target_sense is None and sense.label_based_classifications:
target_sense = tdb.highest_ranked(
lemma=sense.superordinate,
wordclass='NN',
branches=sense.label_based_classifications,
current_only=True)
# Otherwise, narrow by branches based on cross-references
if target_sense is None and sense.xref_branches:
target_sense = tdb.highest_ranked(
lemma=sense.superordinate,
wordclass='NN',
branches=sense.xref_branches,
current_only=True)
# Last gasp: If the gloss consists more or less *only* of the
# superordinate (e.g. 'an abbey'), then it should be adequate to
# just use the main sense of the superordinate, even if it's
# multi-sense.
# But don't risk this is there are cross-references or subject
# labels which might suggest a more specific use
if (target_sense is None and not sense.subjects and
not sense.xref_branches and sense.gloss is not None):
g = re.sub(r'^(a|an|the) ', '', sense.gloss.lower())
if g == sense.superordinate:
target_sense = MAIN_SENSE_FINDER.main_sense(
lemma=sense.superordinate, wordclass='NN')
# Otherwise, narrow by Bayes classification
if target_sense is None and sense.bayes.is_usable():
target_sense = tdb.highest_ranked(
lemma=sense.superordinate,
wordclass='NN',
branches=sense.bayes_based_classifications,
current_only=True)
if target_sense is not None and target_sense.thesclass is not None:
match = target_sense.thesclass
if sense.wordclass == 'JJ':
match = tdb.equivalent_class(match, 'JJ')
return match
else:
return None
def formal_compound_analysis(sense, entry_main_sense):
"""
Figure out a likely thesaurus class based on the form of a
two-part compound lemma.
This is the main way of classifying undefined compounds, and
can be used to support other methods for defined compounds.
Returns a CompoundAnalysisResult object
"""
# Bug out if this is not a workable compound
if (sense.last_element() is None or
sense.last_element() in STOPWORDS or
sense.first_element() in STOPWORDS or
sense.wordclass not in WORDCLASSES):
return formal_compound_null_result()
# Initialize the CompoundAnalysisResult object which will be returned
output = CompoundAnalysisResult(lemma=sense.lemma,
refentry=sense.entry_id,
refid=sense.node_id)
output.is_undefined = sense.is_undefined()
#=====================================
# Build the core tables for the result object
#=====================================
# Insert consensus of different Bayes evaluations
if sense.is_possibly_parasynthetic():
bayes_modes = ('main', 'bias_high',)
else:
bayes_modes = ('main', 'bias_low',)
output.bayes_consensus = compute_bayes_consensus(sense, bayes_modes)[0:10]
# Insert ranked senses for the second word
output.ranked_senses = rank_senses_for_last_element(sense)[0:10]
# Likely thesaurus branches for the first and last elements,
# derived from the index of compound elements
word1_index = retrieve_from_compound_index(sense.first_element(),
'first')
word2_index = retrieve_from_compound_index(sense.last_element(),
sense.wordclass)
if (word1_index is not None and
word1_index.count >= 5 and
word2_index is not None and
word2_index.count >= 5):
p = word2_index.combined_probabilities(word1_index)
elif word2_index is not None and word2_index.count >= 5:
p = word2_index.combined_probabilities(None)
else:
p = []
output.index_consensus = p[0:10]
if word2_index is not None:
output.index_count = word2_index.count
# Establish the main sense of the first and last words
word1_main_sense = entry_main_sense
word2_main_sense = MAIN_SENSE_FINDER.main_sense(
lemma=sense.last_element(),
wordclass=sense.wordclass
)
#=====================================
# Handlers for special cases
#=====================================
# Special handling of 'doctor-like', 'cat-wise', etc. - we just return
# the wordclass branch equivalent to the first element
if ((sense.last_element() in SIMILATIVE and
sense.wordclass in ('JJ', 'RB')) or
(sense.last_element() in ('piece', 'part') and
sense.wordclass == 'NN')):
if (word1_main_sense is not None and
word1_main_sense.thesclass is not None):
t = tdb.equivalent_class(word1_main_sense.thesclass,
sense.wordclass)
if t is not None:
output.forced_result = t.wordclass_parent() or t
output.forced_result = t
return output
# Special handling of 'pro-...' and 'anti-...'- we just return
# the wordclass branch equivalent to the first element
if (sense.first_element() in ('pro', 'anti', 'un') and
sense.wordclass in ('JJ', 'NN')):
if (word2_main_sense is not None and
word2_main_sense.thesclass is not None):
t = tdb.equivalent_class(word2_main_sense.thesclass,
sense.wordclass)
if t is not None:
output.forced_result = t.wordclass_parent() or t
return output
# Special handling of '...-shaped', '...-coloured', etc.
if sense.last_element() in PARASYN_ENDINGS:
output.forced_result = PARASYN_ENDINGS[sense.last_element()]
return output
# Special handling of cases where the Second element is single-sense
# (but don't risk this with upper-case forms or possible plurals)
if (sense.wordclass == 'NN' and
len(output.ranked_senses) == 1 and
len(output.ranked_senses[0].classes) == 1 and
output.ranked_senses[0].classes[0] is not None and
sense.last_element().islower() and
not re.search(r'[^s]s$', sense.last_element())):
output.forced_result = output.ranked_senses[0].classes[0]
return output
#=====================================
# Handler for regular cases (compute best guesses)
#=====================================
output.best_guesses = compute_best_guesses(sense, output)
return output