File size: 65,560 Bytes
6fa4bc9 |
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 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 |
{
"paper_id": "I05-1014",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:26:15.475638Z"
},
"title": "Chunking Using Conditional Random Fields in Korean Texts",
"authors": [
{
"first": "Yong-Hun",
"middle": [],
"last": "Lee",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "POSTECH and AITrc",
"location": {
"addrLine": "San 31, Hyoja-dong, Nam-gu",
"postCode": "790-784",
"settlement": "Pohang",
"country": "R. of Korea"
}
},
"email": ""
},
{
"first": "Mi-Young",
"middle": [],
"last": "Kim",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "POSTECH and AITrc",
"location": {
"addrLine": "San 31, Hyoja-dong, Nam-gu",
"postCode": "790-784",
"settlement": "Pohang",
"country": "R. of Korea"
}
},
"email": ""
},
{
"first": "Jong-Hyeok",
"middle": [],
"last": "Lee",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "POSTECH and AITrc",
"location": {
"addrLine": "San 31, Hyoja-dong, Nam-gu",
"postCode": "790-784",
"settlement": "Pohang",
"country": "R. of Korea"
}
},
"email": "jhlee@postech.ac.kr"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "We present a method of chunking in Korean texts using conditional random fields (CRFs), a recently introduced probabilistic model for labeling and segmenting sequence of data. In agglutinative languages such as Korean and Japanese, a rule-based chunking method is predominantly used for its simplicity and efficiency. A hybrid of a rule-based and machine learning method was also proposed to handle exceptional cases of the rules. In this paper, we present how CRFs can be applied to the task of chunking in Korean texts. Experiments using the STEP 2000 dataset show that the proposed method significantly improves the performance as well as outperforms previous systems.",
"pdf_parse": {
"paper_id": "I05-1014",
"_pdf_hash": "",
"abstract": [
{
"text": "We present a method of chunking in Korean texts using conditional random fields (CRFs), a recently introduced probabilistic model for labeling and segmenting sequence of data. In agglutinative languages such as Korean and Japanese, a rule-based chunking method is predominantly used for its simplicity and efficiency. A hybrid of a rule-based and machine learning method was also proposed to handle exceptional cases of the rules. In this paper, we present how CRFs can be applied to the task of chunking in Korean texts. Experiments using the STEP 2000 dataset show that the proposed method significantly improves the performance as well as outperforms previous systems.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Text chunking is a process to identify non-recursive cores of various phrase types without conducting deep parsing of text [3] . Abney first proposed it as an intermediate step toward full parsing [1] . Since Ramshaw and Marcus approached NP chunking using a machine learning method, many researchers have used various machine learning techniques [2, 4, 5, 6, 10, 11, 13, 14] . The chunking task was extended to the CoNLL-2000 shared task with standard datasets and evaluation metrics, which is now a standard evaluation task for text chunking [3] .",
"cite_spans": [
{
"start": 123,
"end": 126,
"text": "[3]",
"ref_id": "BIBREF2"
},
{
"start": 197,
"end": 200,
"text": "[1]",
"ref_id": "BIBREF0"
},
{
"start": 347,
"end": 350,
"text": "[2,",
"ref_id": "BIBREF1"
},
{
"start": 351,
"end": 353,
"text": "4,",
"ref_id": "BIBREF3"
},
{
"start": 354,
"end": 356,
"text": "5,",
"ref_id": "BIBREF4"
},
{
"start": 357,
"end": 359,
"text": "6,",
"ref_id": "BIBREF5"
},
{
"start": 360,
"end": 363,
"text": "10,",
"ref_id": "BIBREF9"
},
{
"start": 364,
"end": 367,
"text": "11,",
"ref_id": "BIBREF10"
},
{
"start": 368,
"end": 371,
"text": "13,",
"ref_id": "BIBREF12"
},
{
"start": 372,
"end": 375,
"text": "14]",
"ref_id": "BIBREF13"
},
{
"start": 544,
"end": 547,
"text": "[3]",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Most previous works with relatively high performance in English used machine learning methods for chunking [4, 13] . Machine learning methods are mainly divided into the generative approach and conditional approach. The generative approach relies on generative probabilistic models that assign a joint probability p(X,Y) of paired input sequence and label sequence, X and Y respectively. It provides straightforward understanding of underlying distribution. However, this approach is intractable in most domains without strong independence assumptions that each input element is independent from the other elements in input sequence, and is also difficult to use multiple interacting features and long-range dependencies between input elements. The conditional approach views the chunking task as a sequence of classification problems, and defines a conditional probability p(Y|X) over label sequence given input sequence. A number of conditional models recently have been developed for use. They showed better performance than generative models as they can handle many arbitrary and overlapping features of input sequence [12] .",
"cite_spans": [
{
"start": 107,
"end": 110,
"text": "[4,",
"ref_id": "BIBREF3"
},
{
"start": 111,
"end": 114,
"text": "13]",
"ref_id": "BIBREF12"
},
{
"start": 1123,
"end": 1127,
"text": "[12]",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "A number of methods are applied to chunking in Korean texts. Unlike English, a rule-based chunking method [7, 8] is predominantly used in Korean because of its well-developed function words, which contain information such as grammatical relation, case, tense, modal, etc. Chunking in Korean texts with only simple heuristic rules obtained through observation on the text shows a good performance similar to other machine learning methods [6] . Park et al. proposed a hybrid of rule-based and machine learning method to handle exceptional cases of the rules, to improve the performance of chunking in Korean texts [5, 6] .",
"cite_spans": [
{
"start": 106,
"end": 109,
"text": "[7,",
"ref_id": "BIBREF6"
},
{
"start": 110,
"end": 112,
"text": "8]",
"ref_id": "BIBREF7"
},
{
"start": 438,
"end": 441,
"text": "[6]",
"ref_id": "BIBREF5"
},
{
"start": 613,
"end": 616,
"text": "[5,",
"ref_id": "BIBREF4"
},
{
"start": 617,
"end": 619,
"text": "6]",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper, we present how CRFs, a recently introduced probabilistic model for labeling and segmenting sequence of data [12] , can be applied to the task of chunking in Korean texts. CRFs are undirected graphical models trained to maximize conditional probabilities of label sequence given input sequence. It takes advantage of generative and conditional models. CRFs can include many correlated, overlapping features, and they are trained discriminatively like conditional model. Since CRFs have single exponential model for the conditional probability of entire label sequence given input sequence, they also guarantee to obtain globally optimal label sequence. CRFs successfully have been applied in many NLP problems such as part-of-speech tagging [12] , text chunking [13, 15] and table extraction from government reports [19] .",
"cite_spans": [
{
"start": 123,
"end": 127,
"text": "[12]",
"ref_id": "BIBREF11"
},
{
"start": 755,
"end": 759,
"text": "[12]",
"ref_id": "BIBREF11"
},
{
"start": 776,
"end": 780,
"text": "[13,",
"ref_id": "BIBREF12"
},
{
"start": 781,
"end": 784,
"text": "15]",
"ref_id": "BIBREF14"
},
{
"start": 830,
"end": 834,
"text": "[19]",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The rest of this paper is organized as follows. Section 2 gives a simple introduction to CRFs. Section 3 explains how CRFs is applied to the task of chunking in Korean texts. Finally, we present experimental results and draw conclusions.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Conditional Random Fields (CRFs) are conditional probabilistic sequence models first introduced by Lefferty et al [12] . CRFs are undirected graphical models, which can be used to define the joint probability distribution over label sequence given the entire input sequence to be labeled, rather than being directed graphical models such as Maximum Entropy Markov Models (MEMMs) [11] . It relaxes the strong independence assumption of Hidden Markov Models (HMMs), as well as resolves the label bias problem exhibited by MEMMs and other non-generative directed graphical models such as discriminative Markov models [12] .",
"cite_spans": [
{
"start": 114,
"end": 118,
"text": "[12]",
"ref_id": "BIBREF11"
},
{
"start": 379,
"end": 383,
"text": "[11]",
"ref_id": "BIBREF10"
},
{
"start": 614,
"end": 618,
"text": "[12]",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conditional Random Fields",
"sec_num": "2"
},
{
"text": "CRFs may be viewed as an undirected graphical model globally conditioned on input sequence [14] . Let X=x 1 x 2 x 3 \u2026x n be an input sequence and Y=y 1 y 2 y 3 \u2026y n a label sequence. In the chunking task, X is associated with a sequence of words and Y is associated with a sequence of chunk types. If we assume that the structure of a graph forms a simple first-order chain, as illustrated in Figure 1 , CRFs define the conditional probability of a label sequence Y given an input sequence X by the Hammersley-Clifford theorem [16] as follows:",
"cite_spans": [
{
"start": 91,
"end": 95,
"text": "[14]",
"ref_id": "BIBREF13"
},
{
"start": 527,
"end": 531,
"text": "[16]",
"ref_id": "BIBREF15"
}
],
"ref_spans": [
{
"start": 393,
"end": 401,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Fundamentals of CRFs",
"sec_num": "2.1"
},
{
"text": "\u239f \u23a0 \u239e \u239c \u239d \u239b = \u2211\u2211 \u2212 i k i i k k i X y y f X Z X Y p ) , , , ( exp ) ( 1 ) | ( 1 \u03bb (1)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Fundamentals of CRFs",
"sec_num": "2.1"
},
{
"text": "where Z(X) is a normalization factor; f k (y i-1 , y i , X, i) is a feature function at positions i and i-1 in the label sequence; k \u03bb is a weight associated with feature k f . ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Fundamentals of CRFs",
"sec_num": "2.1"
},
{
"text": "+ = \u2211\u2211 \u2211\u2211 \u2212 i k i k k i k i i k k i X y s i X y y t X Z X Y p ) , , ( ) , , , ( exp ) ( 1 ) | ( 1 \u00b5 \u03bb (2) where t k (y i-1 , y i , X, i)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Fundamentals of CRFs",
"sec_num": "2.1"
},
{
"text": "is a transition feature function of the entire input sequence and the labels at positions i and i-1 in the label sequence; s k (y i , X, i) is a state feature function of the label at position i and the observed input sequence; and k \u03bb and k \u00b5 are parameters to be estimated from training data. The parameters k \u03bb and k \u00b5 play similar roles to the transition and emission probabilities in HMMs [12] . Therefore, the most probable label sequence for input sequence X is Y* which maximizes a posterior probability.",
"cite_spans": [
{
"start": 394,
"end": 398,
"text": "[12]",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Fundamentals of CRFs",
"sec_num": "2.1"
},
{
"text": ") | ( max arg * X Y P Y Y \u03bb = (3)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Fundamentals of CRFs",
"sec_num": "2.1"
},
{
"text": "We can find Y* with dynamic programming using the Viterbi algorithm.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Fundamentals of CRFs",
"sec_num": "2.1"
},
{
"text": "Assuming the training data {(X (n) , Y (n) )} are independently and identically distributed, the product of Equation 1 over all training sequences is a likelihood function of the parameter \u03bb . Maximum likelihood training chooses parameter values such that the log-likelihood is maximized [10] . For CRFs, the log-likelihood ) (\u03bb L is given by",
"cite_spans": [
{
"start": 31,
"end": 34,
"text": "(n)",
"ref_id": null
},
{
"start": 288,
"end": 292,
"text": "[10]",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter Estimation for CRFs",
"sec_num": "2.2"
},
{
"text": "\u2211 \u2211\u2211 \u2211 \u23a5 \u23a6 \u23a4 \u23a2 \u23a3 \u23a1 \u2212 = = \u2212 n n i k n n i n i k k n n n X Z i X y y f X Y P L ) ( log ) , , , ( ) | ( log ) ( ) ( ) ( ) ( ) ( 1 ) ( ) ( \u03bb \u03bb \u03bb (4)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter Estimation for CRFs",
"sec_num": "2.2"
},
{
"text": "It is not possible to analytically determine the parameter values that maximize the log-likelihood. Instead, maximum likelihood parameters must be identified using an iterative technique such as iterative scaling [12] or gradient-based methods [13, 14] .",
"cite_spans": [
{
"start": 213,
"end": 217,
"text": "[12]",
"ref_id": "BIBREF11"
},
{
"start": 244,
"end": 248,
"text": "[13,",
"ref_id": "BIBREF12"
},
{
"start": 249,
"end": 252,
"text": "14]",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter Estimation for CRFs",
"sec_num": "2.2"
},
{
"text": "Lafferty et al. proposed two iterative scaling algorithms to find parameters for CRFs. However, these methods converge into a global maximum very slowly. To overcome this problem of slow convergence, several researchers adopted modern optimization algorithms such as the conjugate-gradient method or the limited-memory BFGS(L-BFGS) method [17] .",
"cite_spans": [
{
"start": 339,
"end": 343,
"text": "[17]",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Parameter Estimation for CRFs",
"sec_num": "2.2"
},
{
"text": "We now describe how CRFs are applied to the task of chunking in Korean texts. Firstly, we explore characteristics and chunk types of Korean. Then we explain the features for the model of chunking in Korean texts using CRFs. The ultimate goal of a chunker is to output appropriate chunk tags of a sequence of words with part-ofspeech tags.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Chunking Using Conditional Random Fields in Korean Texts",
"sec_num": "3"
},
{
"text": "Korean is an agglutinative language, in which a word unit (called an eojeol) is a composition of a content word and function word(s). Function words -postpositions and endings -give much information such as grammatical relation, case, tense, modal, etc. Well-developed function words in Korean help with chunking, especially NP and VP chunking. For example, when the part-of-speech of current word is one of determiner, pronoun and noun, the following seven rules for NP chunking in Table 1 can find most NP chunks in text, with about 89% accuracy [6] . For this reason, boundaries of chunks are easily found in Korean, compared to other languages such as English or Chinese. This is why a rule-based chunking method is predominantly used. However, with sophisticated rules, the rule-based chunking method has limitations when handling exceptional cases. Park et al. proposed a hybrid of the rule-based and the machine learning method to resolve this problem [5, 6] . Many recent machine learning techniques can capture hidden characteristics for classification. Despite its simplicity and efficiency, the rule-based method has recently been outdone by the machine learning method in various classification problems.",
"cite_spans": [
{
"start": 548,
"end": 551,
"text": "[6]",
"ref_id": "BIBREF5"
},
{
"start": 959,
"end": 962,
"text": "[5,",
"ref_id": "BIBREF4"
},
{
"start": 963,
"end": 965,
"text": "6]",
"ref_id": "BIBREF5"
}
],
"ref_spans": [
{
"start": 483,
"end": 490,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "Characteristics of Korean",
"sec_num": "3.1"
},
{
"text": "Abney was the first to use the term 'chunk' to represent a non-recursive core of an intra-clausal constituent, extending from the beginning of constituent to its head. In Korean, there are four basic phrases: noun phrase (NP), verb phrase (VP), adverb phrase (ADVP), and independent phrase (IP) [6] . As function words such as postposition or ending are well-developed, the number of chunk types is small compared to other languages such as English or Chinese. Table 2 lists the Korean chunk types, a simple explanation and examples of each chunk type. Like the CoNLL-2000 dataset, we use three types of chunk border tags, indicating whether a word is outside a chunk (O), starts a chunk (B), or continues a chunk (I). Each chunk type XP has two border tags: B-XP and I-XP. XP should be one of NP, VP, ADVP and IP. There exist nine chunk tags in Korean.",
"cite_spans": [
{
"start": 295,
"end": 298,
"text": "[6]",
"ref_id": "BIBREF5"
}
],
"ref_spans": [
{
"start": 461,
"end": 468,
"text": "Table 2",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Chunk Types of Korean",
"sec_num": "3.2"
},
{
"text": "One advantage of CRFs is that they can use many arbitrary, overlapping features. So we take advantage of all context information of a current word. We use words, partof-speech tags of context and combinations of part-of-speech tags to determine the chunk tag of the current word,. The window size of context is 5; from left two words to right two words. Table 3 summarizes the feature set for chunking in Korean texts. Table 3 . Feature set for the chunking in Korean texts",
"cite_spans": [],
"ref_spans": [
{
"start": 354,
"end": 361,
"text": "Table 3",
"ref_id": null
},
{
"start": 419,
"end": 426,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Feature Set of CRFs",
"sec_num": "3.3"
},
{
"text": "POS tag Bi-gram of tags Tri-gram of tags w i-2 = w w i-1 = w w i = w w i+1 = w w i+2 = w",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Word",
"sec_num": null
},
{
"text": "For evaluation of our proposed method, we use the STEP 2000 Korean chunking dataset (STEP 2000 dataset) 1 , which is converted from the parsed KAIST Corpus [9] . The STEP 2000 dataset consists of 12,092 sentences. We divide this corpus into training data and test data. Training data has 10,883 sentences and test data has 1,209 sentences, 90% and 10% respectively. Table 4 summarizes characteristics of the STEP 2000 dataset. Figure 2 shows an example sentence of the STEP 2000 dataset and its format is equal to that of CoNLL-2000 dataset. Each line is composed of a word, its part-of-speech (POS) tag and a chunk tag.",
"cite_spans": [
{
"start": 156,
"end": 159,
"text": "[9]",
"ref_id": "BIBREF8"
}
],
"ref_spans": [
{
"start": 366,
"end": 373,
"text": "Table 4",
"ref_id": "TABREF2"
},
{
"start": 427,
"end": 435,
"text": "Figure 2",
"ref_id": null
}
],
"eq_spans": [],
"section": "Data Preparation",
"sec_num": "4.1"
},
{
"text": "The standard evaluation metrics for chunking performance are precision, recall and Fscore (F ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation Metric",
"sec_num": "4.2"
},
{
"text": "Experiments were performed with C++ implementation of CRFs (FlexCRFs) on Linux with 2.4 GHz Pentium IV dual processors and 2.0Gbyte of main memory [18] . We use L-BFGS to train the parameters and use a Gaussian prior regularization in order to avoid overfitting [20] . Table 5 , the performances of most chunk type are 96~100%, very high performance. However, the performance of NP chunk type is lowest, 94.27% because the border of NP chunk type is very ambiguous in case of consecutive nouns. Using more features such as previous chunk tag should be able to improve the performance of NP chunk type. [6] . We add the experimental results of the chunking methods using HMMs-bigram and CRFs. In Table 6 , F-score of chunking using CRFs in Korean texts is 97.19%, the highest performance of all. It significantly outperforms all others, including machine learning methods, rule-based methods and hybrid methods. It is because CRFs have a global optimum solution hence overcoming the label bias problem. They also can use many arbitrary, overlapping features. Figure 3 shows the performance curve on the same test set in terms of the precision, recall and F-score with respect to the size of training data. In this figure, we can see that the performance slowly increases in proportion to the size of training data. In the experiment, we can see that CRFs can help improve the performance of chunking in Korean texts. CRFs have many promising properties except for the slow convergence speed compared to other models. In the next experiment, we have tried to analyze the importance of each feature and to make an additional experiment with various window sizes and any other useful features.",
"cite_spans": [
{
"start": 147,
"end": 151,
"text": "[18]",
"ref_id": "BIBREF17"
},
{
"start": 262,
"end": 266,
"text": "[20]",
"ref_id": "BIBREF19"
},
{
"start": 602,
"end": 605,
"text": "[6]",
"ref_id": "BIBREF5"
}
],
"ref_spans": [
{
"start": 269,
"end": 276,
"text": "Table 5",
"ref_id": "TABREF3"
},
{
"start": 695,
"end": 702,
"text": "Table 6",
"ref_id": "TABREF4"
},
{
"start": 1058,
"end": 1066,
"text": "Figure 3",
"ref_id": "FIGREF2"
}
],
"eq_spans": [],
"section": "Experimental Results",
"sec_num": "4.3"
},
{
"text": "In this paper, we proposed a chunking method for Korean texts using CRFs. We observed that the proposed method outperforms other approaches. Experiments on the STEP 2000 dataset showed that the proposed method yields an F-score of 95.36%. This performance is 2.82% higher than that of SVMs and 1.15% higher than that of the hybrid method. CRFs use a number of correlated features and overcome the label bias problem. We obtained a very high performance using only small features. Thus, if we use more features such as semantic information or collocation, we can obtain a better performance.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "From the experiment, we know that the proposed method using CRFs can significantly improve the performance of chunking in Korean texts. CRFs are a good framework for labeling an input sequence. In our future work, we will investigate how CRFs can be applied to other NLP problems: parsing, semantic analysis and spam filtering. Finally, we hope that this work can contribute to the body of research in this field.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "t i-2 = t t i-1 = t t i = t t i+1 = t t i+2 = t t i-2 = t', t i-1 = t t i-1 = t', t i = t t i = t', t i+1 = t t i+1 = t',t i+2 = t t i-2 = t\", t i-1 = t', t i = t t i-1 = t\", t i = t', t i+1 = t t i = t\", t i+1 = t', t i+2 = t4 ExperimentsIn this section, we present experimental results of chunking using CRFs in Korean texts and compare the performance with previous systems of Park et al[6]. To make a fare comparison, we use the same dataset as Park et al[6].",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "STEP is an abbreviation of Software Technology Enhancement Program. We download this dataset from http://bi.snu.ac.kr/~sbpark/Step2000. If you want to know the part-of-speech tags used in the STEP 2000 dataset, you can reference KAIST tagset[9].",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Performances of all methods except HMMs and CRFs are cited from the experiment of Park et al[6]. They also use the STEP 2000 dataset and similar feature set. Therefore, the comparison of performance is reasonable.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "This work was supported by the KOSEF through the Advanced Information Technology Research Center (AITrc) and by the BK21 Project.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Parsing by chunks",
"authors": [
{
"first": "S",
"middle": [],
"last": "Abney",
"suffix": ""
}
],
"year": 1991,
"venue": "Principlebased Parsing",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "S. Abney: Parsing by chunks. In R. Berwick, S. Abney, and C. Tenny, editors, Principle- based Parsing. Kluwer Academic Publishers (1991).",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Text chunking using transformation-based learning",
"authors": [
{
"first": "L",
"middle": [
"A"
],
"last": "Ramashaw",
"suffix": ""
},
{
"first": "M",
"middle": [
"P"
],
"last": "Marcus",
"suffix": ""
}
],
"year": 1995,
"venue": "Proceedings of the Thired ACL Workshop on Very Large Corpora",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "L. A. Ramashaw and M. P. Marcus: Text chunking using transformation-based learning. Proceedings of the Thired ACL Workshop on Very Large Corpora (1995).",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Introduction to the CoNLL-2000 shared task: Chunking. Proceedings of CoNLL-2000",
"authors": [
{
"first": "E",
"middle": [
"F"
],
"last": "Tjong Kim Sang",
"suffix": ""
},
{
"first": "S",
"middle": [],
"last": "Buchholz",
"suffix": ""
}
],
"year": 2000,
"venue": "",
"volume": "",
"issue": "",
"pages": "127--132",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "E. F. Tjong Kim Sang and S. Buchholz: Introduction to the CoNLL-2000 shared task: Chunking. Proceedings of CoNLL-2000 (2000) 127-132.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Chunking with support vector machines. Proceedings of NAACL2001, ACL",
"authors": [
{
"first": "T",
"middle": [],
"last": "Kudo",
"suffix": ""
},
{
"first": "Y",
"middle": [],
"last": "Matsumoto",
"suffix": ""
}
],
"year": 2001,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "T. Kudo and Y. Matsumoto: Chunking with support vector machines. Proceedings of NAACL2001, ACL (2001).",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Combining a Rule-based Method and a k-NN for Chunking Korean Text",
"authors": [
{
"first": "S.-B",
"middle": [],
"last": "Park",
"suffix": ""
},
{
"first": "B.-T",
"middle": [],
"last": "Zhang",
"suffix": ""
}
],
"year": 2001,
"venue": "Proceedings of the 19 th International Conference on Computer Processing of Oriental Languages",
"volume": "",
"issue": "",
"pages": "225--230",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Park, S.-B. and Zhang, B.-T.: Combining a Rule-based Method and a k-NN for Chunking Korean Text. Proceedings of the 19 th International Conference on Computer Processing of Oriental Languages (2001) 225-230.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Text Chunking by Combining Hand-Crafted Rules and Memory-Based Learning",
"authors": [
{
"first": "S.-B",
"middle": [],
"last": "Park",
"suffix": ""
},
{
"first": "B.-T",
"middle": [],
"last": "Zhang",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of the 41 st Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "497--504",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Park, S.-B. and Zhang, B.-T.: Text Chunking by Combining Hand-Crafted Rules and Memory-Based Learning. Proceedings of the 41 st Annual Meeting of the Association for Computational Linguistics (2003) 497-504.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Maximally Efficient Syntactic Parsing with Minimal Resources",
"authors": [
{
"first": "H.-P",
"middle": [],
"last": "Shin",
"suffix": ""
}
],
"year": 1999,
"venue": "Proceedings of the Conference on Hangul and Korean Language Information Processing",
"volume": "",
"issue": "",
"pages": "242--244",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "H.-P. Shin: Maximally Efficient Syntactic Parsing with Minimal Resources. Proceedings of the Conference on Hangul and Korean Language Information Processing (1999) 242-244.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Dependency Parsing by Chunks. Proceedings of the 27 th KISS Spring Conference",
"authors": [
{
"first": "M.-Y",
"middle": [],
"last": "Kim",
"suffix": ""
},
{
"first": "S.-J",
"middle": [],
"last": "Kang",
"suffix": ""
},
{
"first": "J.-H",
"middle": [],
"last": "Lee",
"suffix": ""
}
],
"year": 1999,
"venue": "",
"volume": "",
"issue": "",
"pages": "327--329",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "M.-Y. Kim, S.-J. Kang and J.-H. Lee: Dependency Parsing by Chunks. Proceedings of the 27 th KISS Spring Conference (1999) 327-329.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Study on KAIST Corpus",
"authors": [
{
"first": "J.-T",
"middle": [],
"last": "Yoon",
"suffix": ""
},
{
"first": "K.-S",
"middle": [],
"last": "Choi",
"suffix": ""
}
],
"year": 1999,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J.-T. Yoon and K.-S. Choi: Study on KAIST Corpus, CS-TR-99-139, KAIST CS (1999).",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "A maximum entropy approach to natural language processing",
"authors": [
{
"first": "A",
"middle": [
"L"
],
"last": "Berger",
"suffix": ""
},
{
"first": "S",
"middle": [
"A"
],
"last": "Della Pietra",
"suffix": ""
},
{
"first": "V",
"middle": [
"J"
],
"last": "Della Pietra",
"suffix": ""
}
],
"year": 1996,
"venue": "Computational Linguistics",
"volume": "22",
"issue": "1",
"pages": "39--71",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "A. L. Berger, S. A. Della Pietra and V. J. Della Pietra: A maximum entropy approach to natural language processing. Computational Linguistics, 22(1) (1996) 39-71.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Maximum entropy Markov models for information extraction and segmentation",
"authors": [
{
"first": "Andrew",
"middle": [],
"last": "Mccallum",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Freitag",
"suffix": ""
},
{
"first": "F",
"middle": [],
"last": "Pereira",
"suffix": ""
}
],
"year": 2000,
"venue": "Proceedings of International Conference on Machine Learning",
"volume": "",
"issue": "",
"pages": "591--598",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Andrew McCallum, D. Freitag and F. Pereira: Maximum entropy Markov models for in- formation extraction and segmentation. Proceedings of International Conference on Ma- chine Learning , Stanford, California (2000) 591-598.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data. Proceedings of the 18 th International Conference on Machine Learning",
"authors": [
{
"first": "John",
"middle": [],
"last": "Lafferty",
"suffix": ""
},
{
"first": "Andrew",
"middle": [],
"last": "Mccallum",
"suffix": ""
},
{
"first": "Fernando",
"middle": [],
"last": "Pereira",
"suffix": ""
}
],
"year": 2001,
"venue": "",
"volume": "",
"issue": "",
"pages": "282--289",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "John Lafferty, Andrew McCallum and Fernando Pereira: Conditional Random Fields: Probabilistic Models for Segmenting and Labeling Sequence Data. Proceedings of the 18 th International Conference on Machine Learning (2001) 282-289.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Shallow Parsing with Conditional Random Fields. Proceedings of Human Language Technology-NAACL",
"authors": [
{
"first": "Fei",
"middle": [],
"last": "Sha",
"suffix": ""
},
{
"first": "Fernando",
"middle": [],
"last": "Pereira",
"suffix": ""
}
],
"year": 2003,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Fei Sha and Fernando Pereira: Shallow Parsing with Conditional Random Fields. Proceed- ings of Human Language Technology-NAACL, Edmonton, Canada (2003).",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Efficient Training of Conditional Random Fields",
"authors": [
{
"first": "Hanna",
"middle": [],
"last": "Wallach",
"suffix": ""
}
],
"year": 2002,
"venue": "Thesis. Master of Science School of Cognitive Science",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hanna Wallach: Efficient Training of Conditional Random Fields. Thesis. Master of Sci- ence School of Cognitive Science, Division of Informatics. University of Edinburgh (2002).",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Applying Conditional Random Fields to Chinese Shallow Parsing",
"authors": [
{
"first": "Yongmei",
"middle": [],
"last": "Tan",
"suffix": ""
},
{
"first": "Tianshun",
"middle": [],
"last": "Yao",
"suffix": ""
},
{
"first": "Qing",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Jingbo",
"middle": [],
"last": "Zhu",
"suffix": ""
}
],
"year": 2005,
"venue": "The 6 th International Conference on Intelligent Text Processing and Computational Linguistics (CICLing-2005)",
"volume": "3406",
"issue": "",
"pages": "167--176",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yongmei Tan, Tianshun Yao, Qing Chen and Jingbo Zhu: Applying Conditional Random Fields to Chinese Shallow Parsing. The 6 th International Conference on Intelligent Text Processing and Computational Linguistics (CICLing-2005) . LNCS, Vol.3406, Springer, Mexico City, Mexico (2005) 167-176.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Markov fields on finite graphs and lattices. Unpublished manuscript",
"authors": [
{
"first": "J",
"middle": [],
"last": "Hammersley",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Clifford",
"suffix": ""
}
],
"year": 1971,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Hammersley and P. Clifford. Markov fields on finite graphs and lattices. Unpublished manuscript (1971).",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "On the limited memory bfgs method for large-scale optimization",
"authors": [
{
"first": "D",
"middle": [
"C"
],
"last": "Liu",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "",
"suffix": ""
}
],
"year": 1989,
"venue": "Mathematic Programming",
"volume": "45",
"issue": "",
"pages": "503--528",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "D. C. Liu and J. Nocedal: On the limited memory bfgs method for large-scale optimiza- tion. Mathematic Programming, 45 (1989) 503-528.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "FlexCRFs: A Flexible Conditional Random Fields Toolkit",
"authors": [
{
"first": "Minh",
"middle": [
"Le"
],
"last": "Hieu Xuan Phan",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Nguyen",
"suffix": ""
}
],
"year": 2004,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hieu Xuan Phan and Minh Le Nguyen: FlexCRFs: A Flexible Conditional Random Fields Toolkit. http:://www.jaist.ac.jp/~hieuxuan/flexcrfs/flexcrfs.html (2004).",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Table extraction using conditional random fields",
"authors": [
{
"first": "D",
"middle": [],
"last": "Pinto",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Mccallum",
"suffix": ""
},
{
"first": "X",
"middle": [],
"last": "Wei",
"suffix": ""
},
{
"first": "W",
"middle": [
"B"
],
"last": "Croft",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of the ACM SIGIR",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "D. Pinto, A. McCallum, X. Wei and W. B. Croft: Table extraction using conditional ran- dom fields. Proceedings of the ACM SIGIR (2003).",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "A Gaussian prior for smoothing maximum entropy models",
"authors": [
{
"first": "S",
"middle": [
"F"
],
"last": "Chen",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Rosenfeld",
"suffix": ""
}
],
"year": 1999,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "S. F. Chen and R. Rosenfeld: A Gaussian prior for smoothing maximum entropy models. Technical Report CMU-CS-99-108, Carnegie Mellon University (1999).",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"type_str": "figure",
"uris": null,
"num": null,
"text": "Graphical structure of chain-structured CRFs Equitation 1, the general form of a graph structure for modeling sequential data, can be expanded to Equation 2,"
},
"FIGREF1": {
"type_str": "figure",
"uris": null,
"num": null,
"text": ". F-score is used for comparisons with other reported results. Each equation is defined as follows."
},
"FIGREF2": {
"type_str": "figure",
"uris": null,
"num": null,
"text": "The performance curve respect to the size of training data"
},
"TABREF0": {
"html": null,
"type_str": "table",
"num": null,
"text": "Rules for NP chunking in Korean texts",
"content": "<table><tr><td>No</td><td>Previous eojeol</td><td>Chunk tag of current word</td></tr><tr><td>1</td><td>determiner</td><td>I-NP</td></tr><tr><td>2</td><td>pronoun</td><td>I-NP</td></tr><tr><td>3</td><td>noun</td><td>I-NP</td></tr><tr><td>4</td><td>noun + possessive postposition</td><td>I-NP</td></tr><tr><td>5</td><td>noun + relative postfix</td><td>I-NP</td></tr><tr><td>6</td><td>adjective + relative ending</td><td>I-NP</td></tr><tr><td>7</td><td>others</td><td>B-NP</td></tr></table>"
},
"TABREF1": {
"html": null,
"type_str": "table",
"num": null,
"text": "The Korean chunk types",
"content": "<table><tr><td colspan=\"2\">No Category Explanation</td><td>Example</td><td/><td/><td/></tr><tr><td>1 NP</td><td>Noun Phrase</td><td colspan=\"2\">[NP\uc800 ([the beautiful woman] [look]) \uc544 \ub984 \ub2e4 \uc6b4 \uc5ec \uc778 ] [\ubcf4\uc138\uc694]. \uc744</td><td/><td/></tr><tr><td>2 VP</td><td>Verb Phrase</td><td colspan=\"4\">[\uc9c0\ubd95\uc774] [\ubabd\ub545] [VP\ub0b4\ub824\uc549\uc544 ([the roof] [completely] [has fallen in]) \uc788 ]. \ub2e4</td></tr><tr><td>3 ADVP</td><td>Adverb Phrase</td><td>[\uc0c8\uac00] [ADVP ([a bird] [very high] [is flying]) \ub9e4 \uc6b0 \ub192 \uc774 ] [\ub0a0\uace0</td><td>\uc788</td><td>\ub2e4</td><td>].</td></tr><tr><td>4 IP</td><td>Independent Phrase</td><td colspan=\"2\">[IP ([wow] [this] [very] [is delicious]) ], [\uc774\uac70] [\uc815\ub9d0] [\ub9db\uc788\ub2e4]. \uc640</td><td/><td/></tr></table>"
},
"TABREF2": {
"html": null,
"type_str": "table",
"num": null,
"text": "Simple statistics on the STEP 2000 dataset",
"content": "<table><tr><td/><td/><td colspan=\"2\">Information</td><td>Value</td></tr><tr><td/><td/><td>POS tags</td><td/><td>52</td></tr><tr><td/><td/><td>Words</td><td/><td>321,328</td></tr><tr><td/><td/><td>Sentences</td><td/><td>12,092</td></tr><tr><td/><td/><td>Chunk tags</td><td/><td>9</td></tr><tr><td/><td/><td>Chunks</td><td/><td>112,658</td></tr><tr><td>\uadf8 \uc758 \ucc45 \uc740 \ud30c \ub418 \uc5c8 \ub2e4 .</td><td>\uae30</td><td>npp jcm ncn jxt ncpa xsv ep ef sf</td><td>B-NP I-NP I-NP I-NP B-VP I-VP I-VP I-VP O</td><td>his postposition: possessive book postposition: topic destructed be pre-final ending : past ending : declarative</td></tr></table>"
},
"TABREF3": {
"html": null,
"type_str": "table",
"num": null,
"text": "The performance of proposed method",
"content": "<table><tr><td>Chunk tag</td><td>Precision</td><td>Recall</td><td>F-score</td></tr><tr><td>NP</td><td>94.23</td><td>94.30</td><td>94.27</td></tr><tr><td>VP</td><td>96.71</td><td>96.28</td><td>96.49</td></tr><tr><td>ADVP</td><td>96.90</td><td>97.02</td><td>96.96</td></tr><tr><td>IP</td><td>99.53</td><td>99.07</td><td>99.30</td></tr><tr><td>All</td><td>95.42</td><td>95.31</td><td>95.36</td></tr><tr><td colspan=\"3\">Total number of CRF features is 83,264. As shown in</td><td/></tr></table>"
},
"TABREF4": {
"html": null,
"type_str": "table",
"num": null,
"text": "The experimental results of various chunking methods2",
"content": "<table><tr><td/><td>HMMs</td><td>DT</td><td>MBL</td><td>Rule</td><td colspan=\"3\">SVMs Hybrid CRFs</td></tr><tr><td>Precision</td><td>73.75</td><td>92.29</td><td>91.41</td><td>91.28</td><td>93.63</td><td>94.47</td><td>95.42</td></tr><tr><td>Recall</td><td>76.06</td><td>90.45</td><td>91.43</td><td>92.47</td><td>91.48</td><td>93.96</td><td>95.31</td></tr><tr><td>F-score</td><td>74.89</td><td>91.36</td><td>91.38</td><td>91.87</td><td>92.54</td><td>94.21</td><td>95.36</td></tr><tr><td colspan=\"7\">Park et al. reported the performance of various chunking methods</td><td/></tr></table>"
}
}
}
} |