!7 static Map> theSet; static new LinkedHashSet allObjects; static long changes; static new MultiMap> groupingsByA; static new MultiMap subClasses; static S groupings = [[ + = + = + = = + = + = + = ]]; sclass Updatable { void update {} void setField(S field, O value) { if (eq(get(this, field), value)) ret; set(this, field, value); change(); } } sclass Expectation { S ifClass; Runnable action; *() {} *(S *ifClass, Runnable *action) {} } sclass Word extends Updatable { S text; // or null if unknown new LinkedHashSet prev; new LinkedHashSet next; L constituents; // if group new L expectations; new L fulfilledExpectations; new TreeSet classes; int classesConvertedToTraits; new LinkedHashSet groups; // I am part of new L traits; void update { // Add direct word classes if (text != null) for (S c : reverseLookupInMapToSets(theSet, text)) addClass(c); // Make text for group if (isGroup() && text == null) { L l = collect(constituents, 'text); if (!anyNull(l)) setField(text := joinWithSpace(l)); } // Process expectations for (Expectation e : cloneList(expectations)) { //print("Checking expected class " + e.ifClass); if (classes.contains(e.ifClass)) { moveElementFromCollectionToCollection(e, expectations, fulfilledExpectations); change(); callF(e.action); } } if (l(classes) > classesConvertedToTraits) { for (fS c : dropFirst(classesConvertedToTraits, classes)) addTraitsForClass(c); classesConvertedToTraits = l(classes); } for (Trait t : iterateListConcurrently(traits)) t.update(); } bool isGroup() { ret constituents != null; } bool addClass(S c) { if (!classes.add(c)) false; change(); long n; do { n = changes; // optimizable for (S d : iterateListConcurrently(asList(classes))) for (S e : subClasses.get(d)) if (classes.add(e)) change(); } while (changes != n); true; } void addExpectation(Expectation e) { //print("addExpectation " + e); expectations.add(e); change(); } void addTraitsForClass(S c) { for (PairS p : groupingsByA.get(c)) addTrait(LinkWithTo(p.a, p.b)); } void addTrait(Trait t) { set(t, w := this); traits.add(t); } } static Word makeGroup(Word a, Word b, S newClass) { //print("makeGroup " + a.text + " / " + b.text); L list = ll(a, b); // look for existing group for (Word g : a.groups) if (eq(g.constituents, list)) { if (g.addClass(newClass)) print("Added class " + newClass + " to existing group: " + a.text + " + " + b.text); ret g; } // new group print("Making group " + newClass + " " + a.text + " + " + b.text); print(" prev=" + sfu(collect(a.prev, 'text))); print(" next=" + sfu(collect(b.next, 'text))); new Word g; allObjects.add(g); g.addClass(newClass); g.constituents = list; for (Word w : list) w.groups.add(g); g.prev.addAll(a.prev); g.next.addAll(b.next); for (Word prev : a.prev) prev.next.add(g); for (Word next : b.next) next.prev.add(g); ret g; } sclass Trait extends Updatable { Word w; } sclass LinkWithTo extends Trait { S linkWith, linkTo; // classes int expectationsSentToNext; *() {} *(S *linkWith, S *linkTo) {} void update { if (l(w.next) > expectationsSentToNext) { for (final Word next : dropFirst(expectationsSentToNext, w.next)) next.addExpectation(Expectation(linkWith, r { makeGroup(w, next, linkTo) })); expectationsSentToNext = l(w.next); } } } p-exp { fS sentence = "In the movies Dracula always wears a cape"; L rawWords = printStruct(words(sentence)); theSet = ai_englishWordCategoriesWithElements(); parseGroupings(); new L words; for (S w : rawWords) words.add(nu(Word, text := w)); for (int i = 0; i < l(words)-1; i++) linkWords(words.get(i), words.get(i+1)); //printStruct(first(words)); addAll(allObjects, words); long lastChanges; do { lastChanges = changes; print(n2(changes, "change")); for (Updatable w : cloneList(allObjects)) w.update(); } while (lastChanges != changes); //for (Word w : words) printStruct(cloneForPrinting(w)); for (Word w : words) print(" " + textAndClasses(w)); print(); L groups = [Word w : instancesOf(Word, allObjects) | w.constituents != null]; print(); print(n2(groups, "group")); for (Word g : groups) { print("Group: " + textAndClasses(g)); print(" " + grouped(g)); } assertNempty([Word g : groups | eq(g.text, sentence) && g.classes.contains("")]); print("OK"); } sS textAndClasses(Word w) { ret w.text + " (" + joinWithComma(w.classes) + ")"; } svoid linkWords(Word a, Word b) { a.next.add(b); b.prev.add(a); } static O cloneForPrinting(Word w) { ret cloneWithoutFields(w, 'prev, 'next, 'constituents); } svoid change() { ++changes; } svoid parseGroupings() { for (S s : tlft(groupings)) { L tok = javaTokWithAngleBracketsC(s); if (l(tok) == 5) groupingsByA.put(tok.get(0), pair(tok.get(2), tok.get(4))); else if (l(tok) == 3) subClasses.put(tok.get(0), tok.get(2)); } } sS grouped(Word g) { if (g.constituents == null) ret g.text; new L l; for (Word w : g.constituents) l.add(curlyBraceIfMultipleTokens(grouped(w))); ret joinWithSpace(l); }