1
0
forked from GitHub/gf-core

added experimental script for chunking in the C runtime

This commit is contained in:
kr.angelov
2012-12-03 10:07:54 +00:00
parent 2da23e9872
commit 1863e4c3d6
4 changed files with 267 additions and 8 deletions

View File

@@ -116,7 +116,8 @@ libpgf_la_SOURCES = \
bin_PROGRAMS = \
utils/pgf2yaml \
utils/pgf-print \
utils/pgf-translate
utils/pgf-translate \
utils/pgf-chunk
utils_pgf2yaml_SOURCES = utils/pgf2yaml.c
utils_pgf2yaml_LDADD = libpgf.la libgu.la
@@ -127,6 +128,9 @@ utils_pgf_print_LDADD = libpgf.la libgu.la
utils_pgf_translate_SOURCES = utils/pgf-translate.c
utils_pgf_translate_LDADD = libpgf.la libgu.la
utils_pgf_chunk_SOURCES = utils/pgf-chunk.c
utils_pgf_chunk_LDADD = libpgf.la libgu.la
AUTOMAKE_OPTIONS = foreign subdir-objects dist-bzip2
ACLOCAL_AMFLAGS = -I m4
include doxygen.am

View File

@@ -85,9 +85,7 @@ struct PgfParseState {
PgfItem* meta_item;
PgfContsMap* conts_map;
PgfGenCatMap* generated_cats;
#ifdef PGF_PARSER_DEBUG
unsigned short offset;
#endif
prob_t viterbi_prob;
@@ -1630,9 +1628,7 @@ pgf_new_parse_state(PgfParsing* ps,
state->meta_item = NULL;
state->generated_cats = gu_map_type_new(PgfGenCatMap, pool);
state->conts_map = gu_map_type_new(PgfContsMap, pool);
#ifdef PGF_PARSER_DEBUG
state->offset = next ? next->offset+1 : 0;
#endif
state->viterbi_prob = 0;
state->ps = ps;
state->ts = ts;
@@ -1884,6 +1880,102 @@ pgf_parse_result(PgfParseState* state, GuPool* pool)
return en;
}
void
pgf_parse_print_chunks(PgfParseState* state)
{
if (state->ps->completed == NULL) {
while (state->ps->completed == NULL) {
if (!pgf_parsing_proceed(state))
break;
}
if (state->ps->completed == NULL)
return;
}
GuPool* tmp_pool = gu_new_pool();
GuOut* out = gu_file_out(stdout, tmp_pool);
GuWriter* wtr = gu_new_utf8_writer(out, tmp_pool);
GuExn* err = gu_exn(NULL, type, tmp_pool);
PgfCCat* completed = state->ps->completed;
if (gu_seq_length(completed->prods) == 0)
return;
size_t n_args = 0;
size_t arg_idx = 0;
PgfCCat* ccat = NULL;
PgfProductionMeta* pmeta = NULL;
PgfProduction prod = gu_seq_get(completed->prods, PgfProduction, 0);
GuVariantInfo pi = gu_variant_open(prod);
switch (pi.tag) {
case PGF_PRODUCTION_APPLY:
n_args = 1;
arg_idx = 0;
ccat = completed;
break;
case PGF_PRODUCTION_META:
pmeta = pi.data;
n_args = gu_seq_length(pmeta->args);
arg_idx = 0;
ccat = gu_seq_index(pmeta->args, PgfPArg, arg_idx)->ccat;
break;
}
PgfParseState* next = NULL;
while (state != NULL) {
PgfParseState* tmp = state->next;
state->next = next;
next = state;
state = tmp;
}
int offset = 0;
state = next;
next = NULL;
while (state != NULL) {
if (state->ts != NULL)
{
if (ccat != NULL &&
offset == ((ccat->conts->state != NULL) ? ccat->conts->state->offset : 0)) {
PgfCCat *ccat2 = ccat;
while (ccat2->conts != NULL) {
ccat2 = ccat2->conts->ccat;
}
gu_putc('(', wtr, err);
gu_string_write(ccat2->cnccat->abscat->name, wtr, err);
gu_putc(' ', wtr, err);
}
gu_string_write(state->ts->tok, wtr, err);
offset++;
if (ccat != NULL &&
ccat ==
gu_map_get(state->generated_cats, ccat->conts, PgfCCat*)) {
gu_putc(')', wtr, err);
arg_idx++;
ccat =
(arg_idx >= n_args) ?
NULL :
gu_seq_index(pmeta->args, PgfPArg, arg_idx)->ccat;
}
gu_putc(' ', wtr, err);
}
PgfParseState* tmp = state->next;
state->next = next;
next = state;
state = tmp;
}
gu_putc('\n', wtr, err);
gu_pool_free(tmp_pool);
}
// TODO: s/CId/Cat, add the cid to Cat, make Cat the key to CncCat
PgfParseState*

View File

@@ -102,9 +102,12 @@ pgf_parse_result(PgfParseState* state, GuPool* pool);
* succesful, or ambiguously successful.
*/
PgfExpr
pgf_parse_best_result(PgfParseState* state, GuPool* pool);
// Use this procedure only on your own risk.
// It is dirty and it will probably be removed or replaced
// with something else. Currently it is here only for experimental
// purposes.
void
pgf_parse_print_chunks(PgfParseState* state);
size_t
pgf_item_lin_idx(PgfItem* item);

View File

@@ -0,0 +1,160 @@
// Don't give too much hope to this script. It is doing the wrong thing
// but let's see how far we can get with it.
#include <gu/variant.h>
#include <gu/map.h>
#include <gu/dump.h>
#include <gu/log.h>
#include <gu/enum.h>
#include <gu/file.h>
#include <pgf/pgf.h>
#include <pgf/data.h>
#include <pgf/parser.h>
#include <pgf/lexer.h>
#include <pgf/literals.h>
#include <pgf/linearize.h>
#include <pgf/expr.h>
#include <pgf/edsl.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <locale.h>
#include <time.h>
int main(int argc, char* argv[]) {
// Set the character locale, so we can produce proper output.
setlocale(LC_CTYPE, "");
// Create the pool that is used to allocate everything
GuPool* pool = gu_new_pool();
int status = EXIT_SUCCESS;
if (argc != 4) {
fprintf(stderr, "usage: %s pgf cat from_lang\n", argv[0]);
status = EXIT_FAILURE;
goto fail;
}
char* filename = argv[1];
GuString cat = gu_str_string(argv[2], pool);
GuString from_lang = gu_str_string(argv[3], pool);
FILE* infile = fopen(filename, "r");
if (infile == NULL) {
fprintf(stderr, "couldn't open %s\n", filename);
status = EXIT_FAILURE;
goto fail;
}
// Create an input stream from the input file
GuIn* in = gu_file_in(infile, pool);
// Create an exception frame that catches all errors.
GuExn* err = gu_new_exn(NULL, gu_kind(type), pool);
// Read the PGF grammar.
PgfPGF* pgf = pgf_read(in, pool, err);
// If an error occured, it shows in the exception frame
if (!gu_ok(err)) {
fprintf(stderr, "Reading PGF failed\n");
status = EXIT_FAILURE;
goto fail_read;
}
if (!pgf_load_meta_child_probs(pgf, "../../../treebanks/PennTreebank/ParseEngAbs3.probs", pool)) {
fprintf(stderr, "Loading meta child probs failed\n");
status = EXIT_FAILURE;
goto fail_read;
}
// Look up the source and destination concrete categories
PgfConcr* from_concr =
gu_map_get(pgf->concretes, &from_lang, PgfConcr*);
if (!from_concr) {
fprintf(stderr, "Unknown language\n");
status = EXIT_FAILURE;
goto fail_concr;
}
// Register a callback for the literal category Symbol
pgf_parser_add_literal(from_concr, gu_str_string("Symb", pool),
&pgf_nerc_literal_callback);
// Create an output stream for stdout
GuOut* out = gu_file_out(stdout, pool);
// Locale-encoding writers are currently unsupported
// GuWriter* wtr = gu_locale_writer(out, pool);
// Use a writer with hard-coded utf-8 encoding for now.
GuWriter* wtr = gu_new_utf8_writer(out, pool);
// We will keep the latest results in the 'ppool' and
// we will iterate over them by using 'result'.
GuPool* ppool = NULL;
// The interactive translation loop.
// XXX: This currently reads stdin directly, so it doesn't support
// encodings properly. TODO: use a locale reader for input
while (true) {
char buf[4096];
char* line = fgets(buf, sizeof(buf), stdin);
if (line == NULL) {
if (ferror(stdin)) {
fprintf(stderr, "Input error\n");
status = EXIT_FAILURE;
}
break;
} else if (strcmp(line, "") == 0) {
// End nicely on empty input
break;
}
// We create a temporary pool for translating a single
// sentence, so our memory usage doesn't increase over time.
ppool = gu_new_pool();
// Begin parsing a sentence of the specified category
PgfParseState* state =
pgf_parser_init_state(from_concr, cat, 0, ppool);
if (state == NULL) {
fprintf(stderr, "Couldn't begin parsing\n");
status = EXIT_FAILURE;
break;
}
GuReader *rdr =
gu_string_reader(gu_str_string(line, ppool), ppool);
PgfLexer *lexer =
pgf_new_lexer(rdr, ppool);
// Tokenization
GuExn* lex_err = gu_new_exn(NULL, gu_kind(type), ppool);
PgfToken tok = pgf_lexer_next_token(lexer, lex_err, ppool);
while (!gu_exn_is_raised(lex_err)) {
// feed the token to get a new parse state
state = pgf_parser_next_state(state, tok, ppool);
if (!state) {
gu_puts("Unexpected token: \"", wtr, err);
gu_string_write(tok, wtr, err);
gu_puts("\"\n", wtr, err);
goto fail_parse;
}
tok = pgf_lexer_next_token(lexer, lex_err, ppool);
}
pgf_parse_print_chunks(state);
continue;
fail_parse:
// Free all resources allocated during parsing and linearization
gu_pool_free(ppool);
ppool = NULL;
}
fail_concr:
fail_read:
fclose(infile);
fail:
gu_pool_free(pool);
return status;
}