2015-01-24 06:29:04 +00:00
|
|
|
from __future__ import unicode_literals
|
|
|
|
|
2014-12-16 11:44:43 +00:00
|
|
|
from ._state cimport State
|
2015-05-11 14:12:03 +00:00
|
|
|
from ._state cimport has_head, get_idx, get_s0, get_n0, get_left, get_right
|
2014-12-16 11:44:43 +00:00
|
|
|
from ._state cimport is_final, at_eol, pop_stack, push_stack, add_dep
|
|
|
|
from ._state cimport head_in_buffer, children_in_buffer
|
|
|
|
from ._state cimport head_in_stack, children_in_stack
|
2015-05-11 14:12:03 +00:00
|
|
|
from ._state cimport count_left_kids
|
2014-12-16 11:44:43 +00:00
|
|
|
|
2014-12-20 18:42:23 +00:00
|
|
|
from ..structs cimport TokenC
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
from .transition_system cimport do_func_t, get_cost_func_t
|
|
|
|
from .conll cimport GoldParse
|
|
|
|
|
2015-01-27 16:09:45 +00:00
|
|
|
|
2014-12-20 18:42:23 +00:00
|
|
|
DEF NON_MONOTONIC = True
|
2015-01-27 16:09:45 +00:00
|
|
|
DEF USE_BREAK = True
|
2014-12-16 11:44:43 +00:00
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef weight_t MIN_SCORE = -90000
|
2014-12-16 11:44:43 +00:00
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
# Break transition from here
|
|
|
|
# http://www.aclweb.org/anthology/P13-1074
|
2014-12-16 11:44:43 +00:00
|
|
|
cdef enum:
|
|
|
|
SHIFT
|
|
|
|
REDUCE
|
|
|
|
LEFT
|
|
|
|
RIGHT
|
2015-05-11 14:12:03 +00:00
|
|
|
|
2015-01-28 19:22:03 +00:00
|
|
|
BREAK
|
2015-05-11 14:12:03 +00:00
|
|
|
|
|
|
|
CONSTITUENT
|
|
|
|
ADJUST
|
|
|
|
|
2014-12-16 11:44:43 +00:00
|
|
|
N_MOVES
|
|
|
|
|
2015-05-11 14:12:03 +00:00
|
|
|
|
2015-03-09 11:06:01 +00:00
|
|
|
MOVE_NAMES = [None] * N_MOVES
|
|
|
|
MOVE_NAMES[SHIFT] = 'S'
|
|
|
|
MOVE_NAMES[REDUCE] = 'D'
|
|
|
|
MOVE_NAMES[LEFT] = 'L'
|
|
|
|
MOVE_NAMES[RIGHT] = 'R'
|
|
|
|
MOVE_NAMES[BREAK] = 'B'
|
2015-05-11 14:12:03 +00:00
|
|
|
MOVE_NAMES[CONSTITUENT] = 'C'
|
|
|
|
MOVE_NAMES[ADJUST] = 'A'
|
2015-03-09 11:06:01 +00:00
|
|
|
|
2015-01-27 16:18:43 +00:00
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef do_func_t[N_MOVES] do_funcs
|
|
|
|
cdef get_cost_func_t[N_MOVES] get_cost_funcs
|
2014-12-16 11:44:43 +00:00
|
|
|
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef class ArcEager(TransitionSystem):
|
2015-02-22 05:32:07 +00:00
|
|
|
@classmethod
|
|
|
|
def get_labels(cls, gold_parses):
|
2015-03-16 12:32:23 +00:00
|
|
|
move_labels = {SHIFT: {'': True}, REDUCE: {'': True}, RIGHT: {},
|
2015-05-11 14:12:03 +00:00
|
|
|
LEFT: {'ROOT': True}, BREAK: {'ROOT': True},
|
|
|
|
CONSTITUENT: {}, ADJUST: {'': True}}
|
|
|
|
for raw_text, segmented, (ids, words, tags, heads, labels, iob), ctnts in gold_parses:
|
2015-03-27 14:21:06 +00:00
|
|
|
for child, head, label in zip(ids, heads, labels):
|
2015-03-08 05:13:20 +00:00
|
|
|
if label != 'ROOT':
|
2015-03-27 14:21:06 +00:00
|
|
|
if head < child:
|
2015-03-09 05:46:22 +00:00
|
|
|
move_labels[RIGHT][label] = True
|
2015-03-27 14:21:06 +00:00
|
|
|
elif head > child:
|
2015-03-09 05:46:22 +00:00
|
|
|
move_labels[LEFT][label] = True
|
2015-05-11 14:12:03 +00:00
|
|
|
for start, end, label in ctnts:
|
|
|
|
move_labels[CONSTITUENT][label] = True
|
2015-03-09 05:46:22 +00:00
|
|
|
return move_labels
|
2015-02-22 05:32:07 +00:00
|
|
|
|
2015-03-09 11:06:01 +00:00
|
|
|
cdef int preprocess_gold(self, GoldParse gold) except -1:
|
|
|
|
for i in range(gold.length):
|
|
|
|
gold.c_heads[i] = gold.heads[i]
|
2015-03-14 15:06:35 +00:00
|
|
|
gold.c_labels[i] = self.strings[gold.labels[i]]
|
2015-05-11 14:12:03 +00:00
|
|
|
for end, brackets in gold.brackets.items():
|
|
|
|
for start, label_strs in brackets.items():
|
|
|
|
gold.c_brackets[start][end] = 1
|
|
|
|
for label_str in label_strs:
|
|
|
|
# Add the encoded label to the set
|
|
|
|
gold.brackets[end][start].add(self.strings[label_str])
|
2015-03-09 11:06:01 +00:00
|
|
|
|
|
|
|
cdef Transition lookup_transition(self, object name) except *:
|
|
|
|
if '-' in name:
|
|
|
|
move_str, label_str = name.split('-', 1)
|
|
|
|
label = self.label_ids[label_str]
|
|
|
|
else:
|
|
|
|
label = 0
|
|
|
|
move = MOVE_NAMES.index(move_str)
|
|
|
|
for i in range(self.n_moves):
|
|
|
|
if self.c[i].move == move and self.c[i].label == label:
|
|
|
|
return self.c[i]
|
|
|
|
|
2015-03-24 03:27:20 +00:00
|
|
|
def move_name(self, int move, int label):
|
|
|
|
label_str = self.strings[label]
|
|
|
|
if label_str:
|
|
|
|
return MOVE_NAMES[move] + '-' + label_str
|
|
|
|
else:
|
|
|
|
return MOVE_NAMES[move]
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef Transition init_transition(self, int clas, int move, int label) except *:
|
2015-02-22 05:32:07 +00:00
|
|
|
# TODO: Apparent Cython bug here when we try to use the Transition()
|
|
|
|
# constructor with the function pointers
|
|
|
|
cdef Transition t
|
|
|
|
t.score = 0
|
|
|
|
t.clas = clas
|
|
|
|
t.move = move
|
|
|
|
t.label = label
|
|
|
|
t.do = do_funcs[move]
|
|
|
|
t.get_cost = get_cost_funcs[move]
|
|
|
|
return t
|
2014-12-16 11:44:43 +00:00
|
|
|
|
2015-04-28 18:45:51 +00:00
|
|
|
cdef int initialize_state(self, State* state) except -1:
|
2015-03-10 17:00:23 +00:00
|
|
|
push_stack(state)
|
|
|
|
|
2015-04-28 18:45:51 +00:00
|
|
|
cdef int finalize_state(self, State* state) except -1:
|
|
|
|
cdef int root_label = self.strings['ROOT']
|
|
|
|
for i in range(state.sent_len):
|
|
|
|
if state.sent[i].head == 0 and state.sent[i].dep == 0:
|
|
|
|
state.sent[i].dep = root_label
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef Transition best_valid(self, const weight_t* scores, const State* s) except *:
|
|
|
|
cdef bint[N_MOVES] is_valid
|
|
|
|
is_valid[SHIFT] = _can_shift(s)
|
|
|
|
is_valid[REDUCE] = _can_reduce(s)
|
|
|
|
is_valid[LEFT] = _can_left(s)
|
|
|
|
is_valid[RIGHT] = _can_right(s)
|
|
|
|
is_valid[BREAK] = _can_break(s)
|
2015-05-11 14:12:03 +00:00
|
|
|
is_valid[CONSTITUENT] = _can_constituent(s)
|
|
|
|
is_valid[ADJUST] = _can_adjust(s)
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef Transition best
|
|
|
|
cdef weight_t score = MIN_SCORE
|
|
|
|
cdef int i
|
|
|
|
for i in range(self.n_moves):
|
|
|
|
if scores[i] > score and is_valid[self.c[i].move]:
|
|
|
|
best = self.c[i]
|
|
|
|
score = scores[i]
|
2015-03-08 05:13:20 +00:00
|
|
|
assert best.clas < self.n_moves
|
|
|
|
assert score > MIN_SCORE
|
2015-02-21 16:06:37 +00:00
|
|
|
# Label Shift moves with the best Right-Arc label, for non-monotonic
|
|
|
|
# actions
|
|
|
|
if best.move == SHIFT:
|
|
|
|
score = MIN_SCORE
|
|
|
|
for i in range(self.n_moves):
|
|
|
|
if self.c[i].move == RIGHT and scores[i] > score:
|
|
|
|
best.label = self.c[i].label
|
|
|
|
score = scores[i]
|
|
|
|
return best
|
2014-12-16 11:44:43 +00:00
|
|
|
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef int _do_shift(const Transition* self, State* state) except -1:
|
|
|
|
# Set the dep label, in case we need it after we reduce
|
2014-12-20 18:42:23 +00:00
|
|
|
if NON_MONOTONIC:
|
2015-03-08 05:13:20 +00:00
|
|
|
state.sent[state.i].dep = self.label
|
2015-02-21 16:06:37 +00:00
|
|
|
push_stack(state)
|
2014-12-16 11:44:43 +00:00
|
|
|
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef int _do_left(const Transition* self, State* state) except -1:
|
2015-03-27 16:39:16 +00:00
|
|
|
# Interpret left-arcs from EOL as attachment to root
|
|
|
|
if at_eol(state):
|
|
|
|
add_dep(state, state.stack[0], state.stack[0], self.label)
|
|
|
|
else:
|
|
|
|
add_dep(state, state.i, state.stack[0], self.label)
|
2015-02-21 16:06:37 +00:00
|
|
|
pop_stack(state)
|
2014-12-16 11:44:43 +00:00
|
|
|
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef int _do_right(const Transition* self, State* state) except -1:
|
|
|
|
add_dep(state, state.stack[0], state.i, self.label)
|
|
|
|
push_stack(state)
|
|
|
|
|
|
|
|
|
|
|
|
cdef int _do_reduce(const Transition* self, State* state) except -1:
|
2015-03-24 03:27:20 +00:00
|
|
|
if NON_MONOTONIC and not has_head(get_s0(state)):
|
|
|
|
add_dep(state, state.stack[-1], state.stack[0], get_s0(state).dep)
|
2015-02-21 16:06:37 +00:00
|
|
|
pop_stack(state)
|
2015-01-27 16:09:45 +00:00
|
|
|
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef int _do_break(const Transition* self, State* state) except -1:
|
|
|
|
state.sent[state.i-1].sent_end = True
|
|
|
|
while state.stack_len != 0:
|
|
|
|
if get_s0(state).head == 0:
|
2015-03-20 00:15:27 +00:00
|
|
|
get_s0(state).dep = self.label
|
2015-02-21 16:06:37 +00:00
|
|
|
state.stack -= 1
|
|
|
|
state.stack_len -= 1
|
|
|
|
if not at_eol(state):
|
|
|
|
push_stack(state)
|
|
|
|
|
|
|
|
|
2015-05-11 14:12:03 +00:00
|
|
|
cdef int _do_constituent(const Transition* self, State* state) except -1:
|
|
|
|
cdef const TokenC* s0 = get_s0(state)
|
|
|
|
if state.ctnt.head == get_idx(state, s0):
|
|
|
|
start = state.ctnt.start
|
|
|
|
else:
|
|
|
|
start = get_idx(state, s0)
|
|
|
|
state.ctnt += 1
|
|
|
|
state.ctnt.start = start
|
|
|
|
state.ctnt.end = s0.r_edge
|
|
|
|
state.ctnt.head = get_idx(state, s0)
|
|
|
|
state.ctnt.label = self.label
|
|
|
|
|
|
|
|
|
|
|
|
cdef int _do_adjust(const Transition* self, State* state) except -1:
|
|
|
|
cdef const TokenC* child
|
|
|
|
cdef const TokenC* s0 = get_s0(state)
|
|
|
|
cdef int n_left = count_left_kids(s0)
|
|
|
|
for i in range(1, n_left):
|
|
|
|
child = get_left(state, s0, i)
|
|
|
|
assert child is not NULL
|
|
|
|
if child.l_edge < state.ctnt.start:
|
|
|
|
state.ctnt.start = child.l_edge
|
|
|
|
break
|
|
|
|
else:
|
|
|
|
msg = ("Error moving bracket --- Move should be invalid if "
|
|
|
|
"no left edge to move to.")
|
|
|
|
raise Exception(msg)
|
|
|
|
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
do_funcs[SHIFT] = _do_shift
|
|
|
|
do_funcs[REDUCE] = _do_reduce
|
|
|
|
do_funcs[LEFT] = _do_left
|
|
|
|
do_funcs[RIGHT] = _do_right
|
|
|
|
do_funcs[BREAK] = _do_break
|
2015-05-11 14:12:03 +00:00
|
|
|
do_funcs[CONSTITUENT] = _do_constituent
|
|
|
|
do_funcs[ADJUST] = _do_adjust
|
2015-02-21 16:06:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
cdef int _shift_cost(const Transition* self, const State* s, GoldParse gold) except -1:
|
2015-03-08 05:13:20 +00:00
|
|
|
if not _can_shift(s):
|
|
|
|
return 9000
|
2014-12-16 11:44:43 +00:00
|
|
|
cost = 0
|
2015-02-22 05:32:07 +00:00
|
|
|
cost += head_in_stack(s, s.i, gold.c_heads)
|
|
|
|
cost += children_in_stack(s, s.i, gold.c_heads)
|
2014-12-20 18:42:23 +00:00
|
|
|
if NON_MONOTONIC:
|
2015-02-23 19:04:53 +00:00
|
|
|
cost += gold.c_heads[s.stack[0]] == s.i
|
2015-01-28 16:18:29 +00:00
|
|
|
# If we can break, and there's no cost to doing so, we should
|
2015-02-21 16:06:37 +00:00
|
|
|
if _can_break(s) and _break_cost(self, s, gold) == 0:
|
2015-01-28 16:18:29 +00:00
|
|
|
cost += 1
|
2014-12-16 11:44:43 +00:00
|
|
|
return cost
|
|
|
|
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef int _right_cost(const Transition* self, const State* s, GoldParse gold) except -1:
|
2015-03-08 05:13:20 +00:00
|
|
|
if not _can_right(s):
|
|
|
|
return 9000
|
2014-12-16 11:44:43 +00:00
|
|
|
cost = 0
|
2015-02-23 19:04:53 +00:00
|
|
|
if gold.c_heads[s.i] == s.stack[0]:
|
2015-02-22 23:18:54 +00:00
|
|
|
cost += self.label != gold.c_labels[s.i]
|
2014-12-16 11:44:43 +00:00
|
|
|
return cost
|
2015-02-22 05:32:07 +00:00
|
|
|
cost += head_in_buffer(s, s.i, gold.c_heads)
|
|
|
|
cost += children_in_stack(s, s.i, gold.c_heads)
|
|
|
|
cost += head_in_stack(s, s.i, gold.c_heads)
|
2014-12-20 18:42:23 +00:00
|
|
|
if NON_MONOTONIC:
|
2015-02-23 19:04:53 +00:00
|
|
|
cost += gold.c_heads[s.stack[0]] == s.i
|
2014-12-16 11:44:43 +00:00
|
|
|
return cost
|
|
|
|
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef int _left_cost(const Transition* self, const State* s, GoldParse gold) except -1:
|
2015-03-08 05:13:20 +00:00
|
|
|
if not _can_left(s):
|
|
|
|
return 9000
|
2014-12-16 11:44:43 +00:00
|
|
|
cost = 0
|
2015-02-23 19:04:53 +00:00
|
|
|
if gold.c_heads[s.stack[0]] == s.i:
|
|
|
|
cost += self.label != gold.c_labels[s.stack[0]]
|
2014-12-16 11:44:43 +00:00
|
|
|
return cost
|
2015-03-27 16:39:16 +00:00
|
|
|
# If we're at EOL, then the left arc will add an arc to ROOT.
|
|
|
|
elif at_eol(s):
|
|
|
|
# Are we root?
|
|
|
|
cost += gold.c_heads[s.stack[0]] != s.stack[0]
|
|
|
|
# Are we labelling correctly?
|
|
|
|
cost += self.label != gold.c_labels[s.stack[0]]
|
|
|
|
return cost
|
2014-12-16 11:44:43 +00:00
|
|
|
|
2015-02-22 05:32:07 +00:00
|
|
|
cost += head_in_buffer(s, s.stack[0], gold.c_heads)
|
|
|
|
cost += children_in_buffer(s, s.stack[0], gold.c_heads)
|
2014-12-20 18:42:23 +00:00
|
|
|
if NON_MONOTONIC and s.stack_len >= 2:
|
2015-02-23 19:04:53 +00:00
|
|
|
cost += gold.c_heads[s.stack[0]] == s.stack[-1]
|
|
|
|
cost += gold.c_heads[s.stack[0]] == s.stack[0]
|
2014-12-16 11:44:43 +00:00
|
|
|
return cost
|
|
|
|
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef int _reduce_cost(const Transition* self, const State* s, GoldParse gold) except -1:
|
2015-03-08 05:13:20 +00:00
|
|
|
if not _can_reduce(s):
|
|
|
|
return 9000
|
2014-12-20 18:42:23 +00:00
|
|
|
cdef int cost = 0
|
2015-02-22 05:32:07 +00:00
|
|
|
cost += children_in_buffer(s, s.stack[0], gold.c_heads)
|
2014-12-20 18:42:23 +00:00
|
|
|
if NON_MONOTONIC:
|
2015-02-22 05:32:07 +00:00
|
|
|
cost += head_in_buffer(s, s.stack[0], gold.c_heads)
|
2014-12-20 18:42:23 +00:00
|
|
|
return cost
|
2014-12-16 11:44:43 +00:00
|
|
|
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef int _break_cost(const Transition* self, const State* s, GoldParse gold) except -1:
|
2015-03-08 05:13:20 +00:00
|
|
|
if not _can_break(s):
|
|
|
|
return 9000
|
2015-01-28 19:22:03 +00:00
|
|
|
# When we break, we Reduce all of the words on the stack.
|
2015-01-28 16:18:29 +00:00
|
|
|
cdef int cost = 0
|
2015-01-28 16:41:58 +00:00
|
|
|
# Number of deps between S0...Sn and N0...Nn
|
|
|
|
for i in range(s.i, s.sent_len):
|
2015-02-22 05:32:07 +00:00
|
|
|
cost += children_in_stack(s, i, gold.c_heads)
|
|
|
|
cost += head_in_stack(s, i, gold.c_heads)
|
2015-01-27 16:09:45 +00:00
|
|
|
return cost
|
|
|
|
|
|
|
|
|
2015-05-11 14:12:03 +00:00
|
|
|
cdef int _constituent_cost(const Transition* self, const State* s, GoldParse gold) except -1:
|
|
|
|
if not _can_constituent(s):
|
|
|
|
return 9000
|
|
|
|
# The gold standard is indexed by end, then by start, then a set of labels
|
|
|
|
brackets = gold.brackets(get_s0(s).r_edge, {})
|
|
|
|
if not brackets:
|
|
|
|
return 2 # 2 loss for bad bracket, only 1 for good bracket bad label
|
|
|
|
# Index the current brackets in the state
|
|
|
|
existing = set()
|
|
|
|
for i in range(s.ctnt_len):
|
|
|
|
if ctnt.end == s.r_edge and ctnt.label == self.label:
|
|
|
|
existing.add(ctnt.start)
|
|
|
|
cdef int loss = 2
|
|
|
|
cdef const TokenC* child
|
|
|
|
cdef const TokenC* s0 = get_s0(s)
|
|
|
|
cdef int n_left = count_left_kids(s0)
|
|
|
|
# Iterate over the possible start positions, and check whether we have a
|
|
|
|
# (start, end, label) match to the gold tree
|
|
|
|
for i in range(1, n_left):
|
|
|
|
child = get_left(s, s0, i)
|
|
|
|
if child.l_edge in brackets and child.l_edge not in existing:
|
|
|
|
if self.label in brackets[child.l_edge]
|
|
|
|
return 0
|
|
|
|
else:
|
|
|
|
loss = 1 # If we see the start position, set loss to 1
|
|
|
|
return loss
|
|
|
|
|
|
|
|
|
|
|
|
cdef int _adjust_cost(const Transition* self, const State* s, GoldParse gold) except -1:
|
|
|
|
if not _can_adjust(s):
|
|
|
|
return 9000
|
|
|
|
# The gold standard is indexed by end, then by start, then a set of labels
|
|
|
|
gold_starts = gold.brackets(get_s0(s).r_edge, {})
|
|
|
|
# Case 1: There are 0 brackets ending at this word.
|
|
|
|
# --> Cost is sunk, but must allow brackets to begin
|
|
|
|
if not gold_starts:
|
|
|
|
return 0
|
|
|
|
# Is the top bracket correct?
|
|
|
|
gold_labels = gold_starts.get(s.ctnt.start, set())
|
|
|
|
# TODO: Case where we have a unary rule
|
|
|
|
# TODO: Case where two brackets end on this word, with top bracket starting
|
|
|
|
# before
|
|
|
|
|
|
|
|
cdef const TokenC* child
|
|
|
|
cdef const TokenC* s0 = get_s0(s)
|
|
|
|
cdef int n_left = count_left_kids(s0)
|
|
|
|
cdef int i
|
|
|
|
# Iterate over the possible start positions, and check whether we have a
|
|
|
|
# (start, end, label) match to the gold tree
|
|
|
|
for i in range(1, n_left):
|
|
|
|
child = get_left(s, s0, i)
|
|
|
|
if child.l_edge in brackets:
|
|
|
|
if self.label in brackets[child.l_edge]:
|
|
|
|
return 0
|
|
|
|
else:
|
|
|
|
loss = 1 # If we see the start position, set loss to 1
|
|
|
|
return loss
|
|
|
|
|
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
get_cost_funcs[SHIFT] = _shift_cost
|
|
|
|
get_cost_funcs[REDUCE] = _reduce_cost
|
|
|
|
get_cost_funcs[LEFT] = _left_cost
|
|
|
|
get_cost_funcs[RIGHT] = _right_cost
|
|
|
|
get_cost_funcs[BREAK] = _break_cost
|
2015-05-11 14:12:03 +00:00
|
|
|
get_cost_funcs[CONSTITUENT] = _constituent_cost
|
|
|
|
get_cost_funcs[ADJUST] = _adjust_cost
|
2014-12-16 11:44:43 +00:00
|
|
|
|
2014-12-20 18:42:23 +00:00
|
|
|
|
2015-02-21 16:06:37 +00:00
|
|
|
cdef inline bint _can_shift(const State* s) nogil:
|
|
|
|
return not at_eol(s)
|
|
|
|
|
|
|
|
|
|
|
|
cdef inline bint _can_right(const State* s) nogil:
|
|
|
|
return s.stack_len >= 1 and not at_eol(s)
|
|
|
|
|
|
|
|
|
|
|
|
cdef inline bint _can_left(const State* s) nogil:
|
|
|
|
if NON_MONOTONIC:
|
|
|
|
return s.stack_len >= 1
|
|
|
|
else:
|
|
|
|
return s.stack_len >= 1 and not has_head(get_s0(s))
|
|
|
|
|
|
|
|
|
|
|
|
cdef inline bint _can_reduce(const State* s) nogil:
|
|
|
|
if NON_MONOTONIC:
|
|
|
|
return s.stack_len >= 2
|
|
|
|
else:
|
|
|
|
return s.stack_len >= 2 and has_head(get_s0(s))
|
|
|
|
|
|
|
|
|
|
|
|
cdef inline bint _can_break(const State* s) nogil:
|
|
|
|
cdef int i
|
|
|
|
if not USE_BREAK:
|
|
|
|
return False
|
|
|
|
elif at_eol(s):
|
|
|
|
return False
|
|
|
|
else:
|
|
|
|
# If stack is disconnected, cannot break
|
|
|
|
seen_headless = False
|
|
|
|
for i in range(s.stack_len):
|
|
|
|
if s.sent[s.stack[-i]].head == 0:
|
|
|
|
if seen_headless:
|
|
|
|
return False
|
|
|
|
else:
|
|
|
|
seen_headless = True
|
|
|
|
return True
|
2015-05-11 14:12:03 +00:00
|
|
|
|
|
|
|
|
|
|
|
cdef inline bint _can_constituent(const State* s) nogil:
|
|
|
|
return s.stack_len >= 1
|
|
|
|
|
|
|
|
|
|
|
|
cdef inline bint _can_adjust(const State* s) nogil:
|
|
|
|
# Need a left child to move the bracket to
|
|
|
|
cdef const TokenC* child
|
|
|
|
cdef const TokenC* s0 = get_s0(s)
|
|
|
|
cdef int n_left = count_left_kids(s0)
|
|
|
|
cdef int i
|
|
|
|
for i in range(1, n_left):
|
|
|
|
child = get_left(s, s0, i)
|
|
|
|
if child.l_edge < s.ctnt.start:
|
|
|
|
return True
|
|
|
|
else:
|
|
|
|
return False
|