int sentence_parse(Sentence sent, Parse_Options opts)
{
int rc;
sent->num_valid_linkages = 0;
/* If the sentence has not yet been split, do so now.
* This is for backwards compatibility, for existing programs
* that do not explicitly call the splitter.
*/
if (0 == sent->length)
{
rc = sentence_split(sent, opts);
if (rc) return -1;
}
else
{
/* During a panic parse, we enter here a second time, with leftover
* garbage. Free it. We really should make the code that is panicking
* do this free, but right now, they have no API for it, so we do it
* as a favor. XXX FIXME someday. */
free_sentence_disjuncts(sent);
}
/* Check for bad sentence length */
if (MAX_SENTENCE <= sent->length)
{
prt_error("Error: sentence too long, contains more than %d words\n",
MAX_SENTENCE);
return -2;
}
resources_reset(opts->resources);
/* Expressions were set up during the tokenize stage.
* Prune them, and then parse.
*/
expression_prune(sent, opts);
print_time(opts, "Finished expression pruning");
if (opts->use_sat_solver)
{
sat_parse(sent, opts);
}
else
{
classic_parse(sent, opts);
}
print_time(opts, "Finished parse");
if ((verbosity > 0) &&
(PARSE_NUM_OVERFLOW < sent->num_linkages_found))
{
prt_error("Warning: Combinatorial explosion! nulls=%zu cnt=%d\n"
"Consider retrying the parse with the max allowed disjunct cost set lower.\n"
"At the command line, use !cost-max\n",
sent->null_count, sent->num_linkages_found);
}
return sent->num_valid_linkages;
}
int sentence_parse(Sentence sent, Parse_Options opts)
{
int rc;
verbosity = opts->verbosity;
debug = opts->debug;
test = opts->test;
sent->num_valid_linkages = 0;
/* If the sentence has not yet been split, do so now.
* This is for backwards compatibility, for existing programs
* that do not explicitly call the splitter.
*/
if (0 == sent->length)
{
rc = sentence_split(sent, opts);
if (rc) return -1;
}
/* Check for bad sentence length */
if (MAX_SENTENCE <= sent->length)
{
prt_error("Error: sentence too long, contains more than %d words\n",
MAX_SENTENCE);
return -2;
}
/* Initialize/free any leftover garbage */
free_sentence_disjuncts(sent); /* Is this really needed ??? */
resources_reset_space(opts->resources);
if (resources_exhausted(opts->resources))
return 0;
/* Expressions were previously set up during the tokenize stage. */
expression_prune(sent);
print_time(opts, "Finished expression pruning");
if (opts->use_sat_solver)
{
sat_parse(sent, opts);
}
else
{
chart_parse(sent, opts);
}
print_time(opts, "Finished parse");
if ((verbosity > 0) &&
(PARSE_NUM_OVERFLOW < sent->num_linkages_found))
{
prt_error("WARNING: Combinatorial explosion! nulls=%zu cnt=%d\n"
"Consider retrying the parse with the max allowed disjunct cost set lower.\n"
"At the command line, use !cost-max\n",
sent->null_count, sent->num_linkages_found);
}
return sent->num_valid_linkages;
}
static void free_sentence_words(Sentence sent)
{
for (WordIdx i = 0; i < sent->length; i++)
{
free_X_nodes(sent->word[i].x);
free(sent->word[i].alternatives);
}
free_sentence_disjuncts(sent);
free((void *) sent->word);
sent->word = NULL;
}
void sentence_delete(Sentence sent) {
/*free_andlists(sent); */
free_sentence_disjuncts(sent);
free_sentence_expressions(sent);
string_set_delete(sent->string_set);
free_parse_set(sent);
free_post_processing(sent);
post_process_close_sentence(sent->dict->postprocessor);
free_lookup_list();
free_deletable(sent);
free_effective_dist(sent);
xfree(sent->is_conjunction, sizeof(char)*sent->length);
xfree((char *) sent, sizeof(struct Sentence_s));
}
int sentence_parse(Sentence sent, Parse_Options opts) {
int nl;
verbosity = opts->verbosity;
free_sentence_disjuncts(sent);
resources_reset_space(opts->resources);
if (resources_exhausted(opts->resources)) {
sent->num_valid_linkages = 0;
return 0;
}
expression_prune(sent);
print_time(opts, "Finished expression pruning");
prepare_to_parse(sent, opts);
init_fast_matcher(sent);
init_table(sent);
/* A parse set may have been already been built for this sentence,
if it was previously parsed. If so we free it up before building another. */
free_parse_set(sent);
init_x_table(sent);
for (nl = opts->min_null_count;
(nl<=opts->max_null_count) && (!resources_exhausted(opts->resources)); ++nl) {
sent->null_count = nl;
sent->num_linkages_found = parse(sent, sent->null_count, opts);
print_time(opts, "Counted parses");
post_process_linkages(sent, opts);
if (sent->num_valid_linkages > 0) break;
}
free_table(sent);
free_fast_matcher(sent);
print_time(opts, "Finished parse");
return sent->num_valid_linkages;
}
/**
* classic_parse() -- parse the given sentence.
* Perform parsing, using the original link-grammar parsing algorithm
* given in the original link-grammar papers.
*
* Do the parse with the minimum number of null-links within the range
* specified by opts->min_null_count and opts->max_null_count.
*
* To that end, call do_parse() with an increasing null_count, from
* opts->min_null_count up to (including) opts->max_null_count, until a
* parse is found.
*
* A note about the disjuncts save/restore that is done here:
* To increase the parsing speed, before invoking do_parse(),
* pp_and_power_prune() is invoked to remove connectors which have no
* possibility to connect. It includes a significant optimization when
* null_count==0 that makes a more aggressive removal, but this
* optimization is not appropriate when null_count>0.
*
* So in case this optimization has been done and a complete parse (i.e.
* a parse when null_count==0) is not found, we are left with sentence
* disjuncts which are not appropriate to continue do_parse() tries with
* null_count>0. To solve that, we need to restore the original
* disjuncts of the sentence and call pp_and_power_prune() once again.
*/
void classic_parse(Sentence sent, Parse_Options opts)
{
fast_matcher_t * mchxt = NULL;
count_context_t * ctxt = NULL;
bool pp_and_power_prune_done = false;
Disjunct **disjuncts_copy = NULL;
bool is_null_count_0 = (0 == opts->min_null_count);
int max_null_count = MIN((int)sent->length, opts->max_null_count);
/* Build lists of disjuncts */
prepare_to_parse(sent, opts);
if (resources_exhausted(opts->resources)) return;
if (is_null_count_0 && (0 < max_null_count))
{
/* Save the disjuncts in case we need to parse with null_count>0. */
disjuncts_copy = alloca(sent->length * sizeof(Disjunct *));
for (size_t i = 0; i < sent->length; i++)
disjuncts_copy[i] = disjuncts_dup(sent->word[i].d);
}
for (int nl = opts->min_null_count; nl <= max_null_count; nl++)
{
Count_bin hist;
s64 total;
if (!pp_and_power_prune_done)
{
if (0 != nl)
{
pp_and_power_prune_done = true;
if (is_null_count_0)
opts->min_null_count = 1; /* Don't optimize for null_count==0. */
/* We are parsing now with null_count>0, when previously we
* parsed with null_count==0. Restore the save disjuncts. */
if (NULL != disjuncts_copy)
{
free_sentence_disjuncts(sent);
for (size_t i = 0; i < sent->length; i++)
sent->word[i].d = disjuncts_copy[i];
disjuncts_copy = NULL;
}
}
pp_and_power_prune(sent, opts);
if (is_null_count_0) opts->min_null_count = 0;
if (resources_exhausted(opts->resources)) break;
free_count_context(ctxt, sent);
free_fast_matcher(sent, mchxt);
pack_sentence(sent);
ctxt = alloc_count_context(sent);
mchxt = alloc_fast_matcher(sent);
print_time(opts, "Initialized fast matcher");
}
if (resources_exhausted(opts->resources)) break;
free_linkages(sent);
sent->null_count = nl;
hist = do_parse(sent, mchxt, ctxt, sent->null_count, opts);
total = hist_total(&hist);
lgdebug(D_PARSE, "Info: Total count with %zu null links: %lld\n",
sent->null_count, total);
/* total is 64-bit, num_linkages_found is 32-bit. Clamp */
total = (total > INT_MAX) ? INT_MAX : total;
total = (total < 0) ? INT_MAX : total;
sent->num_linkages_found = (int) total;
print_time(opts, "Counted parses");
extractor_t * pex = extractor_new(sent->length, sent->rand_state);
bool ovfl = setup_linkages(sent, pex, mchxt, ctxt, opts);
process_linkages(sent, pex, ovfl, opts);
free_extractor(pex);
post_process_lkgs(sent, opts);
if (sent->num_valid_linkages > 0) break;
if ((0 == nl) && (0 < max_null_count) && verbosity > 0)
prt_error("No complete linkages found.\n");
/* If we are here, then no valid linkages were found.
* If there was a parse overflow, give up now. */
if (PARSE_NUM_OVERFLOW < total) break;
//if (sent->num_linkages_found > 0 && nl>0) printf("NUM_LINKAGES_FOUND %d\n", sent->num_linkages_found);
}
sort_linkages(sent, opts);
if (NULL != disjuncts_copy)
{
for (size_t i = 0; i < sent->length; i++)
free_disjuncts(disjuncts_copy[i]);
}
free_count_context(ctxt, sent);
free_fast_matcher(sent, mchxt);
}
