/* Assumes dc is set according to type of speaker. (Generation would normally
* be set the same way: generate in American English when talking to an
* American...)
* todo: Add FT_ADDRESS (useful only for French, to reduce probability of
* singular "vous" interpretation when "tu" is expected)
* todo: If a parse is accepted using a noncongruent lexical item, consider
* updating the speaker model.
*/
Float LexEntryToObjScore(LexEntryToObj *leo, Discourse *dc)
{
Float score;
if (leo == NULL) return(SCORE_MAX);
switch (FeatureGet(leo->features, FT_FREQ)) {
/* todoSCORE */
case F_FREQUENT:
score = 1.0;
break;
case F_INFREQUENT:
score = 0.2;
break;
default:
score = 0.9;
break;
}
score = ScoreCombine(score,
StyleScore(FeatureGet(leo->features, FT_STYLE),
DC(dc).style));
score = ScoreCombine(score,
DialectScore(FeatureGet(leo->features, FT_DIALECT),
DC(dc).dialect));
return(score);
}
Float Syn_ParseFilterWY_W(PNode *w, PNode *y, int lang)
{
Float score;
if (Syn_Parse_IsOnlyRelativeW(w, lang)) return(0.0);
if (Syn_ParseRightmostIsPronoun(w)) return(0.0);
if (Syn_Parse_IsAdvShiftableToVP(w, y, lang)) return(0.0);
if (!XBarSatisfiesArgRestrictions(w, W_MAX_NPS, W_MAX_PPS-1, W_MAX_ARGS-1)) {
return(0.0);
}
if (LexEntryConceptIsAncestor(N("prep-of"), PNodeLeftmostLexEntry(y))) {
/* See Hindle and Rooth (1993, p. 109)
* "default preference with of for noun attachment" todoSCORE
*/
score = 0.8;
} else {
score = SCORE_MAX;
}
if (!PNodeEndGroupingPunc(w)) {
return(ScoreCombine(score, ScoreCount(PNodeNumberOfWords(w), 8, 20)));
/* todoSCORE */
}
return(score);
}
/* Returns how many matching pns were found <r_cnt>, and how many of those
* would be spliced out <r_spliced>.
*/
void TA_TaggerPrune2(PNode *pn, PNode *prev, int dosplice, size_t pos,
int firstchar, Channel *ch, char *word, char *tag,
/* RESULTS */ PNode **r_pn, PNode **r_prev,
int *r_cnt, int *r_spliced)
{
int cnt, spliced;
Float score;
cnt = spliced = 0;
for (; pn && pn->lowerb <= pos; pn = pn->next) {
/*
Dbg(DBGGEN, DBGHYPER, "%ld %ld\n", pn->lowerb, pos);
*/
if (pn->lowerb == pos &&
pn->type == PNTYPE_LEXITEM &&
pn->lexitem &&
pn->lexitem->word &&
(!PNodeIsPhrase(pn))) {
if (firstchar != pn->lexitem->word[0]) {
if (!dosplice) {
Dbg(DBGGEN, DBGBAD, "TA_TaggerPrune1 firstchar mismatch <%c> <%s>",
(char)firstchar, pn->lexitem->word);
}
prev = pn;
} else if (0.0 ==
(score = PennTreebankTTCompatNBest(tag,
pn->lexitem->features))) {
/* Splice out. */
cnt++;
spliced++;
if (dosplice) {
Dbg(DBGGEN, DBGDETAIL, "tag <%s> rules out <%s>.<%s>",
tag, pn->lexitem->word, pn->lexitem->features);
if (prev) {
prev->next = pn->next;
if (prev->next == NULL) ch->pnf->last = prev;
/* todoFREE */
} else {
ch->pnf->first = pn->next;
if (ch->pnf->first == NULL) ch->pnf->last = NULL;
}
}
} else {
cnt++;
if (dosplice && pn->score < 1.0) {
pn->score = ScoreCombine(pn->score, score);
Dbg(DBGGEN, DBGDETAIL,
"nonbest tag <%s> modifies score of <%s>.<%s> to %d",
tag, pn->lexitem->word, pn->lexitem->features, pn->score);
}
prev = pn;
}
} else {
prev = pn;
}
}
if (r_prev) *r_prev = prev;
if (r_pn) *r_pn = pn;
if (r_cnt) *r_cnt = cnt;
if (r_spliced) *r_spliced = spliced;
}
