RealizationDescriptions.py 45.9 KB
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141
import datetime
import os

from collections import Counter, defaultdict
from itertools import chain

from shellvalier.settings import BASE_DIR, DEBUG

from meanings.models import LexicalUnit, Synset
from semantics.models import SemanticRole, RoleAttribute

from entries.phrase_descriptions.utils import get_form
from entries.phrase_descriptions.polish_strings import TO
from entries.phrase_descriptions.descriptions import make_phraseologisms

from importer.Phrase import Case, Preposition, Modification, Words, LexPhrase, Fixed, NP, LexNP, LexNumP, PrepNP, LexPrepNP, LexPrepGerP, AdjP, LexAdjP, LexPrepAdjP, PActP, LexPActP
from importer.RealizationDescriptionUtils import *

def get_prefs_list(argument):
    return sorted(
        map(str, argument.predefined.all())
    ) + sorted(
        map(str, argument.synsets.all())
    ) + sorted(
        map(str, argument.relations.all())
    )

LOCATION_ROLES = {'Location', 'Path'}

def select_predefined(predefs):
    if len(predefs) == 1:
        return predefs[0]
    return 'ALL'
    # TODO inne heurystyki?
    raise RealisationDescriptionError('couldn’t choose predef lemma: {}'.format('/'.join(predefs)))

def select_predefined_for_xp(predefs, role):
    if predefs == ['ISTOTY']:
        return 'ISTOTY'
    return 'ALL'
    # TODO heurystyki?
    raise RealisationDescriptionError('couldn’t choose predef lemma for XP: {}'.format('/'.join(predefs)))

def get_predefined_lemma(argument, xp=False):
    predefined = argument.predefined.all()
    if not predefined:
        return None
    predefs = sorted(p.key for p in predefined)
    role = argument.role.role.role
    if role not in LOCATION_ROLES and {'LUDZIE', 'PODMIOTY'}.intersection(predefs):
        return ['LUDZIE']
    if xp:
        return [select_predefined_for_xp(predefs, role)]
    else:
        return [select_predefined(predefs)]

def get_hyponyms(synset, seen=None, tab=' '):
    if seen is None:
        seen = set()
    hyponyms = set()
    for hypo in synset.hyponyms.all():
        if hypo not in seen:
            seen.add(hypo)
            hyponyms.add(hypo)
            hyponyms.update(get_hyponyms(hypo, seen, tab=tab + '  '))
    return hyponyms

# for benchmarking
BENCH3 = defaultdict(list)

# precalculated for the largest ones
HYPONYM_CACHE = {
    # sklep-1
    4747 : 46,
    # obiekt budowlany-1
    53426 : 590,
    # konstrukcja-1
    7218 : 614,
    # cecha człowieka-1
    36347 : 676,
    # aberracja-1 nieprawidłowość-1 zaburzenie-2 zakłócenie-3
    4127 : 700,
    # znak-1
    7416 : 732,
    # coś na ząb-1 jedzenie-2 pokarm-1 pożywienie-3 żywność-1
    10738 : 766,
    # materiał-1 tworzywo-1
    1612 : 879,
    # jednostka miary-1 jednostka-4 miano-2 miara-3
    1161 : 881,
    # związek chemiczny-1 związek-1
    19589 : 882,
    # zjawisko naturalne-1
    5351 : 901,
    # dzieło-2 praca-6
    7469 : 927,
    # część-1
    462 : 957,
    # cecha czynności-1 cecha działania-1
    5953 : 1033,
    # część-3
    104936 : 1056,
    # cecha fizyczna-1
    5464 : 1056,
    # wypowiedź-1
    3998 : 1062,
    # proces-1
    54253 : 1103,
    # ciąg wydarzeń-1 ciąg zdarzeń-1
    47401 : 1107,
    # grupa-2 zespół ludzi-1 zespół-2
    7653 : 1176,
    # człowiek charakteryzowany ze względu na kwalifikacje-1
    6779 : 1188,
    # substancja chemiczna-1
    5233 : 1206,
    # przyrząd-1
    7425 : 1260,
    # ilość-1
    1078 : 1427,
    # grupa ludzi-1 grupa-5 ludzie-1
    7702 : 1510,
    # kategoria-3 pojęcie-2
    8170 : 1522,
    # urządzenie-5
    7446 : 1524,
    # historia-3 wydarzenie-1 wypadek-3 zdarzenie-2
    6526 : 1533,
    # grupa istot-1
    103330 : 1585,
    # miejsce-1
    4750 : 1632,
    # stan-1
    3243 : 1761,
    # narzędzie-1
    7610 : 1800,
    # roślina-1
    4603 : 1928,
    # artefakt-1 twór-5 wytwór-2
    2605 : 2029,
    # człowiek ze względu na swoje zajęcie-1
    6797 : 2184,
    # nazwa człowieka uwzględniająca jego cechy-1 nosiciel cechy-1
    6778 : 2308,
    # płód-3 wytwór umysłu-1
    8137 : 2599,
    # człowiek ze względu na relacje społeczne-1
    6775 : 2642,
    # fenomen-1 zjawisko-1
    5371 : 2674,
    # środek-1
    28294 : 2793,
    # człowiek, który coś robi-1
    241977 : 2828,
    # substancja-1
    5236 : 2871,
    # zwierzę-1
    5621 : 2966,
    # materia-3
    247979 : 2970,
    # spowodowanie-1 sprawienie-1
    102579 : 4255,
    # atrybut-1 cecha-1 przymiot-1 własność-2 właściwość-1
    323 : 4579,
    # grupa-4 zbiór-1
    1282 : 4587,
    # uczynienie-1 zrobienie-1
    102576 : 4851,
    # całość-1 ogół-1
    2129 : 5668,
    # człowiek-1 istota ludzka-1 jednostka-2 osoba-1
    6047 : 6151,
    # osoba-4
    28688 : 6170,
    # wytwór-1
    2903 : 7230,
    # efekt-1 rezultat-1 skutek-1 wynik-1
    5195 : 7915,
    # przedmiot-1
    2646 : 7552,
    # istota żywa-1 stworzenie-5 twór-1
    6045 : 8448,
    # istota-1
    1027 : 8536,
    # czynność-1
    10765 : 8653,
    # rzecz-4
    103156 : 9480,
    # egzemplarz-1 indywiduum-1 jednostka-3 organizm-1 osobnik-2
    6731 : 10609,
    # obiekt-2
    234224 : 21435,
}

def select_synsets(synsets):
    by_num_hyponyms = defaultdict(set)
    for synset in synsets:
        sid = synset.id
        if sid not in HYPONYM_CACHE:
            #-------
            t1 = datetime.datetime.now()
            #-------
            hyponyms = get_hyponyms(synset)
            HYPONYM_CACHE[sid] = len(hyponyms)
            #-------
            t2 = datetime.datetime.now()
            # deciseconds :)
            d = round((t2 - t1).total_seconds() * 10)
            if DEBUG:
                BENCH3[d].append((HYPONYM_CACHE[sid], sid, synset))
            # ----
        N = HYPONYM_CACHE[sid]
        by_num_hyponyms[N].add(synset)
    M = max(by_num_hyponyms.keys())
    return list(by_num_hyponyms[M])

FREQ = Counter()
with open(os.path.join(BASE_DIR, 'data/freq/sgjp-freq-23032021.tab')) as f:
    for l in f:
        lemma, pos, freq = l.strip('\n').split('\t')
        if pos not in ('adj', 'subst'):
            continue
        freq = int(freq)
        if freq < 10:
            continue
        # this is inaccurate, but conflate multiple occurrences
        FREQ[lemma] += freq

def rank_units(units, ranker):
    buckets = defaultdict(set)
    for unit in units:
        buckets[ranker(unit)].add(unit)
    ranked = dict()
    for rank, (n, unts) in enumerate(sorted(buckets.items())):
        for unit in unts:
            ranked[unit] = rank
    return ranked

meaning_no_ranker = lambda unit: int(unit.sense)
# TODO lepiej mniej znaczeń (bardziej specyficzne -> precyzyjniejsze?) czy więcej (częstsze -> bardziej zrozumiałe?)
num_meanings_ranker = lambda unit: LexicalUnit.objects.filter(base=unit.base).count()
# w ten sposób nadajemy też najniższy priorytet wielowyrazowym, jeśli istnieje 1-wyrazowa notowana na liście frek.
freq_ranker = lambda unit: -FREQ.get(unit.base, 0)
words_ranker = lambda unit: len(unit.base.split())


# różnice przejrzane oczami na próbce dla:
# [meaning_no_ranker, freq_ranker, num_meanings_ranker]
# [freq_ranker, meaning_no_ranker, num_meanings_ranker] -> [freq_ranker, num_meanings_ranker, meaning_no_ranker] -> takie same wyniki na próbce, TODO sugestia Eli: druga opcja brzmi intuicyjniej
# [num_meanings_ranker, meaning_no_ranker, freq_ranker]
# [meaning_no_ranker, num_meanings_ranker, freq_ranker]

def select_units(units, rankers=[freq_ranker, num_meanings_ranker, meaning_no_ranker, words_ranker]):
    units = [unit for unit in units if (unit.base, unit.sense) not in UNIT_KILL_LIST]
    unit2rank = defaultdict(lambda: [0 for i in range(len(rankers))])
    for i, ranker in enumerate(rankers):
        for unit, rank in rank_units(units, ranker).items():
            unit2rank[unit][i] = rank
    by_rank = defaultdict(set)
    for unit, rank in unit2rank.items():
        by_rank[tuple(rank)].add(unit)
    #for rank, units in sorted(by_rank.items()):
    #    print('        ***', rank, units)
    return sorted(by_rank.items())[0][1]

LEMMA_CACHE = dict()

#returns [lemmata], is_predef
def get_synsets_lemma(argument, pos):
    synsets = argument.synsets.filter(lexical_units__pos=pos).distinct()
    synsets = [(Synset.objects.get(id=SYNSET_MAP[s.id]) if s.id in SYNSET_MAP else s) for s in synsets if s.id not in SYNSET_KILL_LIST]
    if not synsets:
        return None
    key = tuple(sorted(map(str, synsets)))
    if key in LEMMA_CACHE:
        return LEMMA_CACHE[key]
    synsets = synsets if len(synsets) == 1 else select_synsets(synsets)
    for synset in synsets:
        if synset.id in SYNSET2LEMMA:
            return [SYNSET2LEMMA[synset.id]], True
    units = list(chain.from_iterable(synset.lexical_units.all() for synset in synsets))
    units = [units[0]] if len(units) == 1 else select_units(units)
    ret = (sorted(unit.base for unit in units), False)
    if ret[0] == ['cecha czynności', 'cecha działania']:
        return (['cecha'], False)
    LEMMA_CACHE[key] = ret
    return ret

# for benchmarking
BENCH2 = defaultdict(list)

def get_argument_lemma(argument, xp=False):
    t1 = datetime.datetime.now()
    ret = get_argument_lemma2(argument, xp=xp)
    t2 = datetime.datetime.now()
    # deciseconds :)
    d = round((t2 - t1).total_seconds() * 10)
    if DEBUG:
        BENCH2[d].append((argument.predefined.all(), argument.synsets.all(), ret))
    return ret

def get_argument_lemma2(argument, xp=False):
    lemma = get_predefined_lemma(argument, xp=xp)
    if lemma:
        return lemma, True
    lemma = get_synsets_lemma(argument, 'noun')
    if lemma:
        # get_synsets_lemma returns [lemmata], is_predef
        return lemma
    lemma = get_synsets_lemma(argument, 'adj')
    if lemma:
        return lemma
    # TODO!!! np. akuratność
    return ['ALL'], True
    lemma = get_relations_lemma(argument)
    assert(lemma)
    return lemma, False

# nie powinny występować razem:
#   * LUDZIE + PODMIOTY
#   * MIEJSCE + OTOCZENIE + POŁOŻENIE

def process_lemma(lemma, phrase_type):
    mod = NATR
    if lemma in PREDEF2LEMMA:
        lemma, gend, num, pos, mod = PREDEF2LEMMA[lemma].get(phrase_type, PREDEF2LEMMA[lemma]['_'])
        return lemma, gend, num, pos, mod
    
    if ' ' in lemma:
        # eg. ‹środki pieniężne›
        words = lemma.split(' ')
        tags = []
        for i, word in enumerate(words):
            tags.append(sorted(get_simplified_tags(word)))
        if len(words) == 2 and 'subst:nom' in tags[0] and 'adj' in tags[1]:
            # np. ‹środki pieniężne›
            # ‹napój wyskokowy› -> ‹napój› również impt,
            # ‹stan psychiczny› -> ‹psychiczny› również subst,
            lemma = words[0]
            mod = make_adjp_mod(words[1])
            mod._order = 'post'
        elif len(words) == 2 and 'subst:nom' in tags[1] and 'adj' in tags[0]:
            # np. ‹zły uczynek›
            lemma = words[1]
            mod = make_adjp_mod(words[0])
        elif len(words) == 2 and 'subst:nom' in tags[0] and 'pact' in tags[1]:
            # np. ‹pojazd latający›
            lemma = words[0]
            mod = make_pactp_mod(words[1])
            mod._order = 'post'
        elif len(words) == 2 and 'subst:nom' in tags[0] and 'subst:gen' in tags[1]:
            # np. ‹dziedzina wiedzy›
            lemma = words[0]
            mod = make_npgen_mod(words[1])
        elif len(words) == 2 and 'subst:nom' in tags[0] and 'ger:gen' in tags[1]:
            # np. ‹język programowania›
            lemma = words[0]
            # nie mamy lexgerp, więc używamy fixed
            mod = make_fixed_mod(words[1])
            mod._order = 'post'
        elif len(words) == 3 and 'subst:nom' in tags[0] and 'prep:gen' in tags[1] and 'subst:gen' in tags[2]:
            # np. ‹maszyna do szycia›
            lemma = words[0]
            mod = make_prepnp_mod(words[2], words[1], 'gen')
        else:
            raise RealisationDescriptionError('couldn’t parse lemma: {} {}'.format(lemma, tags))
    
    if lemma == 'lata':
        return 'rok', 'm3', 'pl', 'subst', mod
    if lemma in GERUNDS:
        return lemma, 'n', 'sg', 'subst', mod
    
    subst_sg_interps = get_interps(lemma, lemma=lemma, tag_constraints=['subst', 'sg', 'nom'])
    if subst_sg_interps:
        return lemma, get_gender(subst_sg_interps), 'sg', 'subst', mod
    subst_pl_interps = get_interps(lemma, lemma=lemma, tag_constraints=['subst', 'pl', 'nom'])
    if subst_pl_interps:
        # lemat „mnogi” notowany w Morfeuszu jako plurale tantum, np. ‹środki›
        return lemma, get_gender(subst_pl_interps), 'pl', 'subst', mod
    pt_interps = get_interps(lemma, tag_constraints=['subst', 'pl', 'nom'])
    if pt_interps: 
        # lemat „mnogi” nie notowany w Morfeuszu, jako plurale tantum, np. ‹pieniądze›
        lemmata = set(lemma for lemma, tag in pt_interps)
        if len(lemmata) == 1:
            return lemmata.pop(), get_gender(pt_interps), 'pl', 'subst', mod
    if get_interps(lemma, lemma=lemma, tag_constraints=['adj', 'sg', 'nom', 'm1']):
        # przymiotnik
        return lemma, None, 'sg', 'adj', mod
    ger_interps = get_interps(lemma, tag_constraints=['ger', 'sg', 'nom'])
    if ger_interps:
        # gerundium
        lemmata = set(lemma for lemma, tag in ger_interps)
        if len(lemmata) == 1:
            return lemmata.pop(), 'n', 'sg', 'ger', mod
    
    raise RealisationDescriptionError('couldn’t process lemma: {} {}'.format(lemma, get_interps(lemma)))
    
    '''
    # TODO rodzaj w zależności od hiperonimów?
    if lemma == 'członek':
        return lemma, 'sg', 'subst', mod
    try:
        get_form(lemma, ['subst', 'sg', 'nom'])
        return lemma, 'sg', 'subst', mod
    except:
        pass
    try:
        # lemat „mnogi” notowany w Morfeuszu jako plurale tantum, np. ‹środki›
        get_form(lemma, ['subst', 'pl', 'nom'])
        return lemma, 'pl', 'subst', mod
    except:
        pass
    try:
        # przymiotnik
        get_form(lemma, ['adj', 'sg', 'nom', 'm1'])
        return lemma, 'sg', 'adj', mod
    except:
        # lemat „mnogi” nie notowany w Morfeuszu, jako plurale tantum, np. ‹pieniądze›
        subst_pl_nom_lemmata = set(interp[2][1].split(':')[0] for interp in morfeusz.analyse(lemma) if interp[2][2].startswith('subst:pl:nom'))
        if len(subst_pl_nom_lemmata) == 1:
            return subst_pl_nom_lemmata.pop(), 'pl', 'subst', mod
        print('============', lemma)
        print('============', subst_pl_nom_lemmata)
        raise
    '''

PREP_2GRAMS = Counter()
with open(os.path.join(BASE_DIR, 'data/freq/2grams_prep_nkjp')) as f:
    for l in f:
        digram, freq = l.strip('\n').split('\t')
        freq = int(freq)
        PREP_2GRAMS[digram] = freq

XP2PREPNP = {
    'abl'   : (('z', 'gen'),),
    # do domu / na basen
    'adl'   : (('do', 'gen'), ('na', 'acc'),),
    # w mieście, na wsi, u Janka
    'locat' : (('w', 'loc'), ('na', 'loc',), ('u', 'gen'),),
    'perl'  : (('przez', 'acc'),),
    'temp'  : (('podczas', 'gen'),),
    'dur'   : (('przez', 'acc'),),
}

def xp2prepnp(advcat, lemma, num):
    if advcat in XP2PREPNP:
        preps = XP2PREPNP[advcat]
        if len(preps) == 1:
            return preps[0]
        else:
            ranked = []
            for prep, case in preps:
                form = get_form(lemma, ['subst', num, case])[0]
                digram = '{} {}'.format(prep, form)
                ranked.append((-PREP_2GRAMS[digram], (prep, case)))
            return sorted(ranked)[0][1]
    else:
        return None, None

XP2COMPREPNP = {
     'caus'  : 'z powodu',
     # TODO: ożywione: dla ..., nieożywione: w celu ...
     'dest'  : 'w celu',
     'instr' : 'za pomocą',
}

def generate_phrases(function, negativity, phrase, lemma, is_predef, head_gender, controller=None, controller_grammar=None):
    phrase_type = phrase._name
    dummy_id = None
    
    if is_predef and phrase_type == 'xp' and not phrase._category._limitations:
        advcat = phrase._category._value
        # np. „komuś podobało się gdzieś”
        return [PREDEFXP[advcat][lemma]], 'n', 'sg'
    
    distrp = False
    processed_lemma, gend, num, pos, mod = process_lemma(lemma, phrase_type)
    if phrase_type in ('adjp', 'prepadjp') and pos != 'adj':
        # np. aborcja - Manner - lek - adjp(agr)/xp(instr) -> ‹jakaś aborcja›
        processed_lemma, gend, pos, mod = 'jakiś', None, 'adj', NATR
    if phrase_type == 'nonch':
        phrase_type = 'np'
        phrase = NP(Case('nom'), dummy_id)
        # bo nonch może być realizowana wyłącznie przez ‹coś› itp.
        processed_lemma, gend, pos, mod = 'coś', 'n', 'subst', NATR
        # i przetwarzanie dalej jako np
    if phrase_type == 'distrp':
        # ‘po jabłku’ byłoby OK, ale np. ‘po pieniądzach’ brMzmi idiotycznie, więc
        # robimy np(gen) i potem dokleimy ‘po ileś’ (czegoś)
        distrp = True
        phrase_type = 'np'
        phrase = NP(Case('gen'), dummy_id)
        # i przetwarzanie dalej jako np
    
    print('PHRASE TYPE:', phrase_type, 'LEMMA:', processed_lemma, 'MODIFICATION:', mod, 'FUNCTION:', function)
    words = Words('concat', 'xor', [processed_lemma])
    
    # TODO
    if phrase_type in ('cp', 'ncp', 'prepncp'):
        cptype = phrase._type._value
        assert(cptype in ('int', 'rel') or not phrase._type._realisations)
        phr = None
        if cptype == 'int':
            if phrase._type._realisations:
                phr = '/'.join(phrase._type._realisations) + ' …'
            else:
                phr = 'kto/co/czy/… robi/się dzieje/…'
        elif cptype == 'rel':
            if phrase._type._realisations:
                phr = '/'.join(phrase._type._realisations) + ' …'
            else:
                raise RealisationDescriptionError('rel phrase without realisations: {}'.format(phrase))
        elif cptype == 'żeby2':
            comp = 'że' if negativity != 'neg' else 'żeby'
            phr = 'że coś się stało'
        elif cptype in ('żeby', 'jakoby', 'jakby',):
            phr = '{} coś się stało'.format(cptype)
        elif cptype in ('że', 'bo', 'gdy', 'jak', 'jeśli', 'kiedy',):
            phr = '{} coś się dzieje'.format(cptype)
        elif cptype in ('aż', 'zanim',):
            phr = '{} coś się stanie'.format(cptype)
        else:
            print(phrase)
            1 / 0
        if phrase_type == 'cp':
            return [phr], 'n', 'sg'
        if phrase_type == 'ncp':
            return ['{}, {}'.format(TO[phrase._case._value], phr)], 'n', 'sg'
        if phrase_type == 'prepncp':
            return ['{} {}, {}'.format(phrase._prep._value, TO[phrase._prep._case._value], phr)], 'n', 'sg'
    if phrase_type == 'or':
        # TODO? absurd „coś się dzieje”? absurd: coś się dzieje?
        return ['„coś się dzieje”'], 'n', 'sg'
    if phrase_type in ('refl', 'recip'):
        # TODO?
        return ['się'], None, None
    if phrase_type == 'advp':
        # TODO!
        if pos == 'adj':
            return [adj2adv(processed_lemma)], None, None
        # dla nie-przymiotników i tak nic nie wymyślimy
        return ['jakoś'], None, None
    if phrase_type == 'infp':
        # TODO?
        return ['coś robić' if negativity != 'neg' else 'czegoś robić'], 'n', 'sg'
    if phrase_type == 'E':
        # TODO?
        return ['∅'], 'n', 'sg'
    
    if pos == 'adj' and phrase_type not in ('possp', 'adjp', 'prepadjp',):
        # TODO? np. aktualizacja - Manner - automatyczny - xp(instr)
        # TODO źle się generuje dla chlastać, ale tam Instrument ma pref. przymiotnikową ‹ostry›, powinno być raczej ‹ostrze›
        phrase_type = 'adjp'
        phrase = AdjP(Case('agr'), dummy_id)
        # i przetwarzanie dalej jako adjp
    
    if phrase_type == 'possp' and processed_lemma == 'czyjś':
        return [get_form(processed_lemma, ['sg', 'nom', head_gender, 'pos'])[0]], None, None
    if phrase_type == 'comprepnp':
        # TODO wielowyrazowe! ‹abonament w wysokości środków pieniężnych›
        # TODO może ładniej by było „w czyjejś sprawie”, „na czyjąś rzecz”, ale
        # to trochę trudniejsze
        return make_comprepnp(phrase._prep._value, words, num, mod), None, None
        #return ['{} {}'.format(phrase._prep._value, get_form(lemma, [num, 'gen'])[0])]
    
    lex_phrases = []
    phrases = []
    
    if phrase_type == 'np':
        # gerundium; TODO? lista wyjątków jeśli więcej
        if (processed_lemma, function, phrase._case._value) == ('przyrządzanie', 'subj', 'str'):
            return ['przyrządzanie'], 'n', 'sg'
        if (processed_lemma, function, phrase._case._value) == ('szarpnięcie', None, 'inst'):
            return ['szarpnięciem'], 'n', 'sg'
        lex_phrases.append(LexNP(phrase, num, words, mod, dummy_id))
    if phrase_type == 'possp':
        np = NP(Case('gen'), dummy_id)
        lex_phrases.append(LexNP(np, num, words, mod, dummy_id))
    if phrase_type == 'prepnp':
        # gerundium; TODO? lista wyjątków jeśli więcej
        if (processed_lemma, phrase._prep._case._value, phrase._prep._value) == ('przyrządzanie', 'gen', 'do'):
            return ['do przyrządzania'], None, None
        if phrase._prep._value in ('między', 'pomiędzy', 'wśród', 'pośród') and processed_lemma not in ('ktoś', 'coś'):
            num = 'pl'
        if pos == 'subst':
            lex_phrases.append(LexPrepNP(phrase, num, words, mod, dummy_id))
        if pos == 'ger':
            lex_phrases.append(LexPrepGerP(phrase, num, 'aff', words, '', mod, dummy_id))
    if phrase_type == 'adjp':
        # TODO! gender & control
        lex_phrases.append(LexAdjP(phrase, 'sg', head_gender if head_gender else 'm1', 'pos', words, mod, dummy_id))
    if phrase_type == 'prepadjp':
        lex_phrases.append(LexPrepAdjP(phrase, 'sg', 'm1', 'pos', words, mod, dummy_id))
    if phrase_type == 'compar':
        lex_phrases.append(make_compar(phrase, words, num, mod, controller))
    if phrase_type == 'xp':
        if phrase._category._limitations:
            for realisation in phrase._category._limitations:
                phrs, g, n = generate_phrases(function, negativity, realisation, lemma, is_predef, head_gender)
                for phr in phrs:
                    if phr not in phrases:
                        phrases.append(phr)
            return phrases, 'n', 'sg'
        else:
            advcat = phrase._category._value
            if advcat == 'mod':
                phrase2 = NP(Case('inst'), dummy_id)
                lex_phrases.append(LexNP(phrase2, num, words, mod, dummy_id))
            prep, case = xp2prepnp(advcat, processed_lemma, num)
            if prep:
                phrase2 = PrepNP(Preposition(prep, Case(case)), dummy_id)
                lex_phrases.append(LexPrepNP(phrase2, num, words, mod, dummy_id))
            if advcat in XP2COMPREPNP:
                if pos == 'subst':
                    comprep = XP2COMPREPNP[advcat]
                    return make_comprepnp(comprep, words, num, mod), None, None
                if pos == 'ger':
                    assert(mod == NATR)
                    return ['{} {}'.format(comprep, get_form(processed_lemma, ['ger', num, 'gen', head_gender])[0])], 'n', 'sg'
    
    for lex_phrase in lex_phrases:
        for phr in make_phraseologisms(lex_phrase, function, negativity, controller=controller, controller_grammar=controller_grammar):
            if phr not in phrases:
                # TODO? porządna lista wyjątków, jeśli będzie więcej
                if phr == 'na członek rodziny':
                    phr = 'na członka rodziny'
                if distrp:
                    # po iluś facetów/po ileś dziewczyn/kotów...
                    phr = 'po {} {}'.format('iluś' if gend == 'm1' else 'ileś', phr)
                phrases.append(phr)
    
    assert(phrases)
    return phrases, gend if phrase_type == 'np' else None, num if phrase_type == 'np' else None

def get_lex_gender_number(phrase):
    if isinstance(phrase, LexNP):
        number = phrase._number
        # take the first lemma since first expansion is taken for whole meaning description
        lemma = phrase._words._lemmas[0]
        if lemma == 'siebie':
            gender = 'm1'
        elif lemma == 'łupień':
            gender = 'm2'
        else:
            interps = get_interps(lemma, lemma=lemma, tag_constraints=['subst', 'nom'])
            gender = get_gender(interps)
        return gender, number if number != '_' else 'sg'
        '''
        genders = list()
        for lemma in phrase._words._lemmas:
            if lemma == 'siebie':
                genders.append('m1')
            elif lemma == 'łupień':
                genders.append('m2')
            else:
                interps = get_interps(lemma, lemma=lemma, tag_constraints=['subst', 'nom'])
                genders.append(get_gender(interps))
        return genders[0], number if number != '_' else 'sg'
        '''
    if isinstance(phrase, LexNumP):
        # take the first lemma since first expansion is taken for whole meaning description
        lemma = phrase._words._lemmas[0]
        interps = get_interps(lemma, lemma=lemma, tag_constraints=['subst', 'nom'])
        gender = get_gender(interps)
        lemma = phrase._nums._lemmas[0]
        recs = set()
        if lemma == '2':
            recs.add('congr')
        else:
            for interp in get_interps(lemma, lemma=lemma, tag_constraints=['num', 'nom']):
                recs.add(interp[1].split(':')[-1])
        assert(len(recs) == 1)
        rec = recs.pop()
        if rec == 'rec':
            # wiele/pięciu/trzydzieści osiem kotów/facetów/kobiet przyszło
            return 'n', 'sg'
        else:
            # trzy kobiety/koty przyszły/trzej faceci przyszli
            return gender, 'pl'
    return None, None

PHRASE_CACHE = dict()

PHRASE_SEP = ' / '

# for benchmarking
BENCH = defaultdict(list)

def get_phrase_description(subentry, argument, position, phrase, controller_grammar=None):
    t1 = datetime.datetime.now()
    ret = get_phrase_description2(subentry, argument, position, phrase, controller_grammar=controller_grammar)
    t2 = datetime.datetime.now()
    # deciseconds :)
    d = round((t2 - t1).total_seconds() * 10)
    if DEBUG:
        BENCH[d].append((subentry.entry.name, argument.role.role.role, ret[0]))
    return ret

# subentry, argument: DB model objects
# schema, phrase: importer objects
def get_phrase_description2(subentry, argument, position, phrase, controller_grammar=None):
    print()
    print(argument)
    print(phrase)
    gender, number = None, None
    function = position._function._value if position._function else None
    control = None
    if position._control:
        #assert(len(position._control) == 1)
        #control = position._control[0]._function
        ee = [c._function for c in position._control if c._function.endswith('controllee')]
        er = [c._function for c in position._control if c._function.endswith('controller')]
        assert(len(ee) <= 1)
        assert(len(er) <= 1)
        # e.g. ‹uznać› — controllee and pred_controller on the same position, take controllee
        if ee:
            control = ee[0]
        else:
            control = er[0]
    negativity = subentry.negativity.name if subentry.negativity else '_'
    head_lemma, head_gender = subentry.entry.name, None
    
    controller, controller_features, controller_function = None, None, None
    if control and control.endswith('controllee'):
        controller = position._schema.getController(control)
        try:
            controller_features = controller_grammar[controller]
        except KeyError:
            raise RealisationDescriptionError('couldn’t determine grammar features for {}: {} {}'.format(' '.join(map(str, argument.frame.lexical_units.all())), control, phrase))
        controller_function = controller._function._value if controller._function else None
    
    if subentry.entry.pos.tag == 'noun':
        interps = get_interps(head_lemma, lemma=head_lemma, tag_constraints=['subst', 'nom'])
        head_gender = get_gender(interps)
    
    # TODO
    # TODO gender, number
    # TODO (‹jakieś›) oko * (‹jakieś›) oczy *błyszczy* z powodu substancji
    if isinstance(phrase, LexPhrase) or isinstance(phrase, Fixed):
        phrs = []
        # TODO to powinny być tylko brakujące [...] w lex(cp)
        try:
            for phr in make_phraseologisms(phrase, function, negativity, controller=controller, controller_grammar=controller_features):
                if phr not in phrs:
                    phrs.append(phr)
        except:
            phrs.append('!!!???')
        gender, number = get_lex_gender_number(phrase)
        return PHRASE_SEP.join(phrs), gender, number
    lemmata, is_predef = get_argument_lemma(argument, xp=(phrase._name == 'xp' and not phrase._category._limitations))
    if len(lemmata) != 1:
        raise RealisationDescriptionError('couldn’t choose single lemma: {}'.format('/'.join(lemmata)))
    phrases = []
    # TODO since there’s one lemma, drop the loop
    for lemma in lemmata:
        key = (function, negativity, str(phrase), lemma, str(head_gender), control, controller_features, controller_function)
        if key in PHRASE_CACHE:
            lemma_phrases, gender, number = PHRASE_CACHE[key]
        else:
            lemma_phrases, gender, number = generate_phrases(function, negativity, phrase, lemma, is_predef, head_gender, controller=controller, controller_grammar=controller_features)
            PHRASE_CACHE[key] = (lemma_phrases, gender, number)
        phrases += lemma_phrases
    return PHRASE_SEP.join(phrases), gender, number

def get_only_value(d):
    return list(d.values())[0]

PRIORITY, ATTR, SUBPRIORITY = 'priority', 'attr', 'subpriority'
LOW_PRIORITY = 200
CP_PRIO = {
    'żeby'   : 0, # że
    'kiedy'  : 0, # gdy, jak
    'żeby2'  : 1, # jak
    'że'     : 2, # jak
    # prefer phrases introduced by complementisers where present
    'int' : LOW_PRIORITY + 1,
}
PHRASE_PRIORITY = {
    'xp' : {
        PRIORITY : 10,
        ATTR : lambda phrase: phrase._category._value,
        SUBPRIORITY : {
            'adl'   : 0, # nawigacja xp(adl)/xp(locat)
            'locat' : 1, # powycierać xp(abl)/xp(locat)
            'caus'  : 2, # ucierpieć xp(caus)/xp(temp)
        },
    },
    'np' : {
        PRIORITY : 20,
        ATTR : lambda phrase: phrase._case._value,
        SUBPRIORITY : {
            'str' : 0,
        },
    },
    'prepnp' : {
        PRIORITY : 22,
        ATTR : lambda phrase: (phrase._prep._value, phrase._prep._case._value),
        SUBPRIORITY : {
            ('do', 'gen')      : 0, # adekwatny do/dla; kolejka do/za
            ('za', 'inst')     : 1, # agitować za/przeciw
            ('o', 'acc')       : 1, # apel o/przeciw
            ('w', 'acc')       : 1, # całować w/po
            ('w', 'loc')       : 1, # defilada w/na pojeździe
            ('między', 'inst') : 2, # debata między/z/wśród
            ('o', 'loc')       : 2, # debata o/wokół/nad
            ('wobec', 'gen')   : 2, # dług wobec/względem, konsekwentny wobec/dla
            ('dla', 'gen')     : 3, # certyfikat dla/za
            ('z', 'gen')       : 2, # dochód z/za/od
            ('o', 'acc')       : 3, # kampania o/za
            ('pod', 'inst')    : 4, # kruszyć się pod/od
            ('o', 'loc')       : 4, # książka o czymś/z czegoś
            ('po', 'loc')      : 5, # odlatywać od/po
            ('od', 'gen')      : 6, # podatek od/za
            ('przeciw', 'dat') : 7, # przestępstwo z/przeciw
            ('na', 'loc')      : 7, # skoncentrować się na/nad
            ('za', 'acc')      : 7, # zabulić na/za
            ('z', 'acc')       : LOW_PRIORITY + 1, # mandat – błąd w danych, jest tam też za:acc
        },
    },
    'comprepnp' : {
        PRIORITY : 24,
        ATTR : lambda phrase: phrase._prep._value,
        SUBPRIORITY : {
            'w sprawie'   : 0, # w kwestii
            'w zakresie'  : 0, # dyletant w zakresie/w kwestii
            'w kwestii'   : 1, # dyskrecja co do/w kwestii
            'z dziedziny' : 1, # referat w dziedzinie/z dziedziny
        },
    },
    'cp' : {
        PRIORITY : 30,
        ATTR : lambda phrase: phrase._type._value,
        SUBPRIORITY : CP_PRIO,
    },
    'ncp' : {
        PRIORITY : 32,
        ATTR : lambda phrase: phrase._type._value,
        SUBPRIORITY : CP_PRIO,
    },
    'prepncp' : {
        PRIORITY : 34,
        ATTR : lambda phrase: phrase._type._value,
        SUBPRIORITY : CP_PRIO,
    },
}

def get_phrase_priority(phrase):
    lex = False
    if isinstance(phrase, LexPhrase):
        lex = True
        phrase = phrase._lex_phrase()
    phrase_type = phrase._name
    if phrase_type == 'xp' and phrase._category._limitations:
        # TODO? heurystyka: bierzemy pierwszą
        phrase, phrase_type = phrase._category._limitations[0], phrase._category._limitations[0]._name
    if phrase_type not in PHRASE_PRIORITY:
        return (LOW_PRIORITY, LOW_PRIORITY)
    attr = PHRASE_PRIORITY[phrase_type][ATTR](phrase)
    # lower the priority by 1 for lexes, eg. dostępność prepnp(dla, gen)/lex(prepnp(‹dla kieszeni›))
    return (PHRASE_PRIORITY[phrase_type][PRIORITY] + (1 if lex else 0), PHRASE_PRIORITY[phrase_type][SUBPRIORITY].get(attr, LOW_PRIORITY))

# position: importer object
# phrase_descriptions: dict
#    key: phrase importer object
#    value: (description, gender, number)
# result: phrase description to use in the realisation description
def select_phrase_description(position, phrase_descriptions):
    #print(type(position))
    #print(phrase_descriptions)
    if len(phrase_descriptions) == 1:
        desc = get_only_value(phrase_descriptions)
        assert(desc[0] != '???')
        return desc
    by_priority = defaultdict(set)
    for p, d in phrase_descriptions.items():
        by_priority[get_phrase_priority(p)].add((p, d))
    min_priority_phrases = by_priority[min(by_priority.keys())]
    if len(min_priority_phrases) == 1:
        p, desc = min_priority_phrases.pop()
        assert (desc[0] != '???')
        return desc
    else:
        # all are lex phrases
        assert(all(isinstance(p, LexPhrase) for p, d in min_priority_phrases))
        # all have the same grammatical type
        assert(len(set(str(p._lex_phrase()) for p, d in min_priority_phrases)) == 1)
        # heuristic: return first lexicographically
        return sorted(min_priority_phrases, key=lambda x: x[1][0])[0][1]
        #raise RealisationDescriptionError('couldn’t select phrase description: {}'.format(' * '.join(desc[0] for desc in phrase_descriptions.values())))


FUNCTION_RANK = {
    'subj' : 0,
    'head' : 0,
    'obj'  : 2,
    None   : 4,
}

def is_np(phrase, case):
    if phrase._name != 'np':
        return False
    if isinstance(phrase, LexPhrase):
        return phrase._np._case._value == case
    else:
        return phrase._case._value == case

# TODO: possp na początku tylko, jeśli jest przymiotnikowe
def get_argument_realisation_priority(ar, entry_pos):
    position = ar._position
    function = position._function._value if position._function else None
    # first rank by subj or possp, obj, rest
    rank1 = FUNCTION_RANK[function]
    phrase_types = set(phrase._name for phrase in position._phrases)
    if (phrase_types == {'adjp'} and entry_pos == 'noun') or phrase_types == {'possp'}:
        # jakieś COŚ, ale UCZYNIĆ kogoś jakimś
        rank1 = 0
        # np(dat) after verb ‹ktoś daje komuś coś›
    if [p for p in ar._position._phrases if is_np(p, 'dat')]:
        rank1 = 1
    # np(str) without function (TODO? error in data, e.g. chwytać ustami *powietrze* – should be obj?)
    if function is None and [p for p in ar._position._phrases if is_np(p, 'str')]:
        rank1 = 3
    # clauses at the end
    if {'cp', 'ncp', 'prepncp'}.issuperset(phrase_types):
        rank1 = 5
    # then rank by phrase type: refl/recip, then nominal, then rest
    rank2 = 2
    if {'refl', 'recip'}.intersection(phrase_types):
        rank2 = 0
    elif 'np' in phrase_types:
        rank2 = 1
    # finally rank by semantic argument priority
    sem_role = ar._argument._semantic_role
    role_prio = SemanticRole.objects.get(role=sem_role._value).priority
    attribute_prio = RoleAttribute.objects.get(attribute=sem_role._attribute).priority if sem_role._attribute else 0
    rank3 = (role_prio, attribute_prio)
    return [rank1, rank2, rank3]

# jeśli nie ma nic na początku, a jest np(dat), to przesuwamy na początek
def rerank(ars):
    print(ars)
    before, after, np_dat = [], [], []
    for rank, fallback, ar in ars:
        if rank[0] == 0:
            before.append((rank, fallback, ar))
        elif [p for p in ar._position._phrases if is_np(p, 'dat')]:
            np_dat.append((rank, fallback, ar))
        else:
            after.append((rank, fallback, ar))
    if before:
        return ars
    else:
        #assert(len(np_dat) <= 1) #TODO? hasło: daleki
        return [([0] + rank[1:], fallback, ar) for rank, fallback, ar in np_dat] + after

# for multi-position Lemma arguments, e.g. dostać się z deszczu pod rynnę

FALLBACK = {
    'z deszczu' : 1,
    'pod rynnę' : 2,
    'od ściany' : 1,
    'do ściany' : 2,
    'żywcem'   : 1,
    'ze skóry' : 2,
    'pięknym'    : 1,
    'za nadobne' : 2,
    'od Annasza'  : 1,
    'do Kajfasza' : 2,
    'z (brudnymi) buciorami / z (swoimi) buciorami / z (brudnymi swoimi) buciorami / z (brudnymi) butami / z (swoimi) butami / z (brudnymi swoimi) butami' : 1,
    'do łóżka / do łóżek'                                                                                                                                  : 2,
    'samego'            : 1,
    'w (‹jakieś›) ręce' : 2,
    'z (‹jakiejś›) radości / z (‹jakiegoś›) szczęścia' : 1,
    'pod sufit'                                        : 2,
    'z jednej skrajności' : 1,
    'w drugą'             : 2,
    'ze skrajności' : 1,
    'w skrajność'   : 2,
    'z motyką'  : 1,
    'na słońce' : 2,
    'z nogi'  : 1,
    'na nogę' : 2,
    'z pustego' : 1,
    'w próżne'  : 2,
    'z (‹jakiejś›) klasy'  : 1,
    'do (‹jakiejś›) klasy' : 2,
    'z (‹jakiegoś›) kwiatka' : 1,
    'na (‹jakiś›) kwiatek'   : 2,
    'w dno'    : 1,
    'od spodu' : 2,
    'po rozum' : 1,
    'do głowy' : 2,
    'z pazurami / z pięściami' : 1,
    'do oczu'                  : 2,
    'na ziemię' : 1,
    'z obłoków' : 2,
    'prosto' : 1,
    'w (‹jakieś›) serce / w (‹jakieś›) serca' : 2,
    'z rąk'  : 1,
    'do rąk' : 2,
    'z ręki'  : 1,
    'do ręki' : 2,
    'o pomstę' : 1,
    'do nieba' : 2,
    'ze zbiornika' : 1,
    'do zbiornika' : 2,
}

def fallback(description):
    return FALLBACK.get(description, 0)

WINIEN = ('powinien', 'winien',)

# realisation: importer object
# subentry: DB model object
# TODO wszystkie lex-y chyba powinny wejść do tej reprezentacji,
# np. ktoś babrze ‹sobie› ‹rączki›: ‹sobie› nie jest powiązane z argumentem...
def get_realisation_description(realisation, subentry, aspect):
    entry = subentry.entry
    ars = [(get_argument_realisation_priority(ar, entry.pos.tag), fallback(ar._description), ar) for ar in realisation._argument_realizations]
    print([(p1, p2, ar._description) for p1, p2, ar in ars])
    try:
        ars = sorted(ars)
    except:
        raise RealisationDescriptionError('couldn’t order argument realisations: {}'.format(' * '.join('{}{} {}'.format(ar._argument._semantic_role._value, ar._argument._semantic_role._attribute, ar._description) for ar in realisation._argument_realizations)))
    if entry.pos.tag == 'verb':
        # dla innych nie przesuwamy np(dat): bliski *komuś*
        ars = rerank(ars)
    before = [('<b>{}</b>' if ar._argument._semantic_role._value == 'Lemma' else '{}').format(ar._description.split(PHRASE_SEP)[0]) for rank, fallback, ar in ars if rank[0] == 0]
    after = [('<b>{}</b>' if ar._argument._semantic_role._value == 'Lemma' else '{}').format(ar._description.split(PHRASE_SEP)[0]) for rank, fallback, ar in ars if rank[0] > 0]
    subj_ars = [ar for ar in realisation._argument_realizations if ar._position._function and ar._position._function._value == 'subj']
    if len(subj_ars) > 1:
        raise RealisationDescriptionError('> 1 subject argument realisations: {}'.format(' * '.join('{}{} {}'.format(ar._argument._semantic_role._value, ar._argument._semantic_role._attribute, ar._description) for ar in subj_ars)))
    subj_ar = subj_ars[0] if subj_ars else None
    head_ars = [ar for ar in realisation._argument_realizations if ar._position._function and ar._position._function._value == 'head']
    if len(head_ars) > 1:
        raise RealisationDescriptionError('> 1 head argument realisations: {}'.format(' * '.join('{}{} {}'.format(ar._argument._semantic_role._value, ar._argument._semantic_role._attribute, ar._description) for ar in head_ars)))
    head_ar = head_ars[0] if head_ars else None
    entry_form = entry.name
    if entry.name == 'naleźć':
        #TODO błąd w słowniku
        aspect = 'perf'
    if entry.pos.tag == 'adj' and head_ar:
        entry_form = get_form(entry.name, ['adj', head_ar._number, 'nom', head_ar._gender, 'pos'])[0]
    elif entry.name == 'bootować':
        # nienotowane w Morfeuszu
        entry_form = 'bootuje'
    elif entry.name == 'wtyczkować':
        # nienotowane w Morfeuszu
        entry_form = 'wtyczkuje'
    elif entry.pos.tag == 'verb':
        assert(aspect)
        entry_base = entry.name
        if entry_base == 'doprząc':
            entry_base = 'doprzęgnąć'
        if aspect == '_':
            # eg. aresztować
            aspect = 'imperf'
        try:
            subj_num = subj_ar._number if subj_ar else 'sg'
            if subj_ar and (aspect == 'perf' or entry_base in WINIEN):
                # potrzebne tylko dla dokonanych (zrobił/a/o) i winien/na
                if subj_ar._gender:
                    subj_gend = subj_ar._gender
                else:
                    raise RealisationDescriptionError('couldn’t determine subject’s gender: {} {} {}'.format(subj_ar, subj_ar._position._phrases, subj_ar._argument))
            else:
                # no subject: ‹jestem kotem — olśniło kogoś›
                subj_gend = 'n'
            if entry_base in WINIEN:
                entry_form = get_form(entry_base, ['winien', subj_num, subj_gend, 'imperf'])[0]
            elif aspect == 'imperf':
                # niedokonane: fin (cz. teraźnieszy)
                # TODO? lista wyjątków, jeśli będzie więcej
                if entry_base == 'sparować' and subj_num == 'sg':
                    # bokser sparuje — imperf nienotowane w Morfeuszu
                    entry_form = 'sparuje'
                else:
                    print(entry_base, ['fin', subj_num, 'ter', 'imperf'])
                    entry_form = get_form(entry_base, ['fin', subj_num, 'ter', 'imperf'])[0]
            else:
                # dokonane: praet (cz. przeszły)
                # TODO? lista wyjątków, jeśli będzie więcej
                if entry_base == 'nasuwać' and (subj_num, subj_gend) == ('sg', 'm1'):
                    # „Nasuwał się mebli przy odnawianiu mieszkania.” — perf nienotowane w Morfeuszu
                    entry_form = 'nasuwał'
                elif entry_base == 'wybzykać' and (subj_num, subj_gend) == ('sg', 'm1'):
                    # nienotowane w Morfeuszu
                    entry_form = 'wybzykał'
                elif entry_base == 'wytuszować' and (subj_num, subj_gend) == ('sg', 'm1'):
                    # nienotowane w Morfeuszu
                    entry_form = 'wytuszował'
                elif entry_base == 'zależeć' and (subj_num, subj_gend) == ('sg', 'm2'):
                    # nienotowane w Morfeuszu
                    entry_form = 'zależał'
                elif entry_base == 'zemdlić' and (subj_num, subj_gend) == ('sg', 'f'):
                    # formy inne niż „zemdliło” nienotowane w Morfeuszu
                    entry_form = 'zemdliła'
                else:
                    entry_form = get_form(entry_base, ['praet', subj_num, subj_gend, 'perf', ['nagl', '']])[0]
        except:
            entry_form = get_form(entry_base, ['pred'])[0]
        if entry.name == 'napaść' and {'wal_69620-mng', 'wal_80242-mng', 'wal_174604-mng', 'wal_174605-mng', 'wal_174603-mng', 'wal_174606-mng'}.issuperset(realisation._frame._meanings):
            # znaczenie ‹napaść (się) jedzeniem›
            entry_form = entry_form.replace('dł', 'sł')
        if entry.name == 'oblec' and {'wal_85605-mng', 'wal_85615-mng'}.issuperset(realisation._frame._meanings):
            # znaczenie ‹oblec twierdzę›
            entry_form = entry_form.replace('kł', 'gł')
        if entry.name == 'odpaść' and {'wal_68230-mng', 'wal_68225-mng', 'wal_79689-mng'}.issuperset(realisation._frame._meanings):
            # znaczenie ‹odpaść (się) jedzeniem›
            entry_form = entry_form.replace('dł', 'sł')
        if entry.name == 'podpaść' and {'wal_86356-mng', 'wal_86350-mng', 'wal_174582-mng', 'wal_174584-mng', 'wal_174585-mng', 'wal_174586-mng'}.issuperset(realisation._frame._meanings):
            # znaczenie ‹podpaść (się) jedzeniem›
            entry_form = entry_form.replace('dł', 'sł')
        if entry.name == 'popaść' and {'wal_174529-mng', 'wal_174530-mng'}.issuperset(realisation._frame._meanings):
            # znaczenie ‹popaść (się) jedzeniem›
            entry_form = entry_form.replace('dł', 'sł')
    
    if subentry.negativity and subentry.negativity.name == 'neg':
        entry_form = 'nie ' + entry_form
    if subentry.inherent_sie.name == 'true':
        entry_form += ' się'
    elements = before + ['<b>{}</b>'.format(entry_form)] + after
    
    if entry_form[0] > 'z':
        #-------
        for t in sorted(BENCH3.keys()):
            if t > 4:
                print('    ************', t, len(BENCH3[t]), BENCH3[t][:10])
                for n, sid, synset in BENCH3[t]:
                    print('    ************', synset)
                    print('    ************', sid, ':', n)
        #for t in sorted(BENCH2.keys()):
        #    if t > 4:
        #        print('    ********', t, len(BENCH2[t]), BENCH2[t][:10])
        #for t in sorted(BENCH.keys()):
        #    if t > 4:
        #        print('    ****', t, len(BENCH[t]), BENCH[t][:10])
        #-------
    
    return ' '.join(elements)