File size: 64,566 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 |
{
"paper_id": "2019",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T07:29:35.815919Z"
},
"title": "Towards Handling Verb Phrase Ellipsis in English-Hindi Machine Translation",
"authors": [
{
"first": "Niyati",
"middle": [],
"last": "Bafna",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Ashoka University",
"location": {}
},
"email": ""
},
{
"first": "Dipti",
"middle": [],
"last": "Misra",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Indian Institute of Technology",
"location": {
"settlement": "Hyderabad"
}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "English-Hindi machine translation systems have difficulty interpreting verb phrase ellipsis (VPE) in English, and commit errors in translating sentences with VPE. We present a solution and theoretical backing for the treatment of English VPE, with the specific scope of enabling English-Hindi MT, based on an understanding of the syntactical phenomenon of verb-stranding verb phrase ellipsis in Hindi (VVPE). We implement a rule-based system to perform the following sub-tasks: 1) Verb ellipsis identification in the English source sentence, 2) Elided verb phrase head identification 3) Identification of verb segment which needs to be induced at the site of ellipsis 4) Modify input sentence; i.e. resolving VPE and inducing the required verb segment. This system is tested in two parts. It obtains 94.83 percent precision and 83.04 percent recall on subtask (1), tested on 3900 sentences from the BNC corpus (Leech, 1992). This is competitive with state-of-the-art results. We measure accuracy of subtasks (2) and (3) together, and obtain a 91 percent accuracy on 200 sentences taken from the WSJ corpus(Paul and Baker, 1992). Finally, in order to indicate the relevance of ellipsis handling to MT, we carried out a manual analysis of the MT outputs of 100 sentences after passing it through our system. We set up a basic metric (1-5) for this evaluation, where 5 indicates drastic improvement, and obtained an average of 3.55.",
"pdf_parse": {
"paper_id": "2019",
"_pdf_hash": "",
"abstract": [
{
"text": "English-Hindi machine translation systems have difficulty interpreting verb phrase ellipsis (VPE) in English, and commit errors in translating sentences with VPE. We present a solution and theoretical backing for the treatment of English VPE, with the specific scope of enabling English-Hindi MT, based on an understanding of the syntactical phenomenon of verb-stranding verb phrase ellipsis in Hindi (VVPE). We implement a rule-based system to perform the following sub-tasks: 1) Verb ellipsis identification in the English source sentence, 2) Elided verb phrase head identification 3) Identification of verb segment which needs to be induced at the site of ellipsis 4) Modify input sentence; i.e. resolving VPE and inducing the required verb segment. This system is tested in two parts. It obtains 94.83 percent precision and 83.04 percent recall on subtask (1), tested on 3900 sentences from the BNC corpus (Leech, 1992). This is competitive with state-of-the-art results. We measure accuracy of subtasks (2) and (3) together, and obtain a 91 percent accuracy on 200 sentences taken from the WSJ corpus(Paul and Baker, 1992). Finally, in order to indicate the relevance of ellipsis handling to MT, we carried out a manual analysis of the MT outputs of 100 sentences after passing it through our system. We set up a basic metric (1-5) for this evaluation, where 5 indicates drastic improvement, and obtained an average of 3.55.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Verb phrase ellipsis is a particularly frequent form of ellipsis, both in speech and in text. English VPE is the elimination of a non-finite verb phrase, introduced by an auxiliary or the particle 'to' (Kenyon-Dean et al., 2016) .",
"cite_spans": [
{
"start": 202,
"end": 228,
"text": "(Kenyon-Dean et al., 2016)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We observe that state-of-the-art MT systems often cannot correctly interpret and translate sentences with VPE. For example, (elliptical phrase and site of ellipsis are in bold and italics respectively):",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Ram cooked the food quickly, but Shyam did not \u2192 In our initial analysis, we find that Google Translate could translate only 6/50 VPE sentences, taken from the WSJ corpus, correctly from English to Hindi. This motivated us to identify and resolve VPE in English and give this modified sentence to the machine translation system to check whether it improves the quality of translation. We used our MT system Anusaaraka for this purpose. 1 We present a rule-based approach that performs three sub-tasks in order to solve this problem: 1) ellipsis identification 2) identification of the antecedent head verb 3) addition of necessary auxiliaries and/or complements to the head verb. Finally, it modifies the input sentence by inducing the identified verb segment at the site of ellipsis. We tailor this algorithm for the purpose of English-Hindi MT, and therefore provide a special treatment to compound verbs (including phrasal verbs), serial verbs, and verb complements in English.",
"cite_spans": [
{
"start": 436,
"end": 437,
"text": "1",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Our solution is based on an understanding of verb phrase ellipsis in Hindi. Hindi does not exhibit VPE in the same manner as English; however, it exhibits a phenomenon called verb-stranding verb phrase ellipsis (VVPE) (Manetta, 2018) . This means that a verbal phrase, including objects and other arguments, may be completed elided, stranding the head verb, which then appears at the site of ellipsis.",
"cite_spans": [
{
"start": 218,
"end": 233,
"text": "(Manetta, 2018)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We propose that English VPE can be transferred into Hindi VVPE in order to improve MT results on the modified sentence. For this, we claim that it is enough to identify the head verb antecedent in the English sentence, and perform an analysis to support the claim.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Finally, we want to see that MT systems indeed perform better on statements with ellipsis after our treatment. Since we want to test the performance of the MT on a very specific facet: i.e. the elliptical clause, we do not utilize standard evaluation metrics but set up and define our own scale, and perform a manual analysis of 100 sentences. This evaluation dataset is meant to be merely indicative of the benefits of ellipsis handling for MT. The results are explained below.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "There have been several previous works addressing the problem of antecedent head resolution for VPE in English. Cheung et al adapt the Margin-Infused-Relaxed Algorithm (MIRA) for target detection and antecedent resolution and obtain an accuracy of 65 percent (Kenyon-Dean et al., 2016 (Nielsen, 2005 ) (Hardt, 1992) . Liu et al, 2016 experiment with various joint modelling techniques, and obtain a recall of 83.46 percent for boundary identification (Liu et al., 2016) . Earlier works include Daniel Hardt's linguistically motivated rule-based system, that eliminates impossible antecedents by looking at be-do conflicts, contained antecedents and assigned scores based on co-reference of the noun subjects and clausal relationships, such as 'as' constructions. (Hardt, 1992) The necessity for tools to deal with VPE is widely recognized in literature, for the purposes of information extraction, finding event co-occurrence, etc (Kenyon-Dean et al., 2016) . In the context of MT, a possible solution is to identify the antecedent, or the source verb phrase, and induce it at the site of ellipsis to gain a legitimate, simplified sentence. This solution functions by reiterating the antecedent at the site of ellipsis. By breaking the link between the ellipsis and the antecedent, we give the MT two independent clauses, which it can translate without error.",
"cite_spans": [
{
"start": 259,
"end": 284,
"text": "(Kenyon-Dean et al., 2016",
"ref_id": "BIBREF2"
},
{
"start": 285,
"end": 299,
"text": "(Nielsen, 2005",
"ref_id": "BIBREF6"
},
{
"start": 302,
"end": 315,
"text": "(Hardt, 1992)",
"ref_id": "BIBREF1"
},
{
"start": 451,
"end": 469,
"text": "(Liu et al., 2016)",
"ref_id": "BIBREF4"
},
{
"start": 763,
"end": 776,
"text": "(Hardt, 1992)",
"ref_id": "BIBREF1"
},
{
"start": 931,
"end": 957,
"text": "(Kenyon-Dean et al., 2016)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background and Our Contribution",
"sec_num": "2"
},
{
"text": "Indeed, the previous works listed above are aimed at identifying the antecedent verb phrase, which includes compulsorily a head verb, and optionally its arguments and adverbials. While it is important to pick up arguments and adverbials of the head verb for the purpose of comprehension, there are some problems that it introduces: namely, it must often disambiguate by context and therefore can be a great source of error, and in the context of MT, it might make the output sentence clumsy and unnatural.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background and Our Contribution",
"sec_num": "2"
},
{
"text": "Since we are looking at antecedent head resolution from a particular angle i.e. transforming the English sentence containing VPE in order to align with Hindi VVPE and therefore help the MT, our problem does not require us to find the antecedent boundary of the head verb, while we do provide an additional treatment of certain verb constructions in English according to the manner in which they would be translated into Hindi. As far as we are aware, this is the first attempt to tailor English VPE resolution in the context of English-Hindi MT.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background and Our Contribution",
"sec_num": "2"
},
{
"text": "Verb phrase ellipsis in English is introduced by an auxiliary. We consider five classes of el-lipsis depending upon the auxiliary at the site of ellipsis: 1) to_be, 2) to_have, 3) to_do, 4) modals, and 5) to_particle ellipsis. Cheung et al include a sixth class: the do-so anaphora, while acknowledging that modals and Do-X anaphora are not technically auxiliaries (Kenyon-Dean et al., 2016) . We have chosen to identify ellipsis introduced by modals, however, because 1. the behaviour of the former is identical with ellipsis by true auxiliaries, 2. Likewise, we observe that it poses a problem to state-of-the-art MT English-Hindi systems. We do not identify Do-X anaphora in this system, however, since this is simply a pronominalization of the antecedent rather than eliding. These have a different treatment than VPE -for example, they may be directly pronominalized in Hindi as well, rather than transferred into VVPE. Indeed, MT systems are able to do this:",
"cite_spans": [
{
"start": 365,
"end": 391,
"text": "(Kenyon-Dean et al., 2016)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Ellipsis in English",
"sec_num": "3"
},
{
"text": "Although Mr. Azoff won't produce films at first, it is possible that he could do so later, the sources said \u2192 ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Ellipsis in English",
"sec_num": "3"
},
{
"text": "We know that state-of-the-art machines cannot interpret VPE. (Voita et al., 2019) . Translation to Hindi from English requires a prediction of the elided English verb. An elided VP consists of a head verb, optionally along with its object arguments and adverbials. For example, it is clear that the adverbial is interpreted as part of the VP in the following sentence:",
"cite_spans": [
{
"start": 61,
"end": 81,
"text": "(Voita et al., 2019)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Antecedent Resolution in the Context of MT",
"sec_num": "4"
},
{
"text": "I could walk quickly, at that age, across traffic-filled roads, but now I cannot. (created example) One solution to eliminate errors due to VPE, is to identify the antecedent and induce it at the site of ellipsis entirely, as a preliminary step before translation, thus eliminating the ellipsis entirely. However, this naive approach has the following issues:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Antecedent Resolution in the Context of MT",
"sec_num": "4"
},
{
"text": "1. Making the decision of whether to import a particular adjunct is complicated, and might be governed by semantic context, and (when verbal), by emphasis.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Antecedent Resolution in the Context of MT",
"sec_num": "4"
},
{
"text": "2. Reiterating the entire verb phrase at the site of ellipsis may sound clumsy and unnatural, and make for a worse translation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Antecedent Resolution in the Context of MT",
"sec_num": "4"
},
{
"text": "Identifying the boundaries of the elided verb phrase is a much harder task than identifying the antecedent head verb. We note, for example, the consistent drop in Exact Match accuracy in Neilson's study of Hardt's algorithm over different corpora (Nielsen, 2005) -sometimes dipping as much as 30 percent lower. The matter is not as simple as picking up the entire verb sub-tree -always importing all arguments and adjuncts is not permissible, since the correct interpretation often depends on surrounding knowledge. For example, in Ram would have liked to eat out with you on Sunday afternoon, but he can't. (created example)",
"cite_spans": [
{
"start": 247,
"end": 262,
"text": "(Nielsen, 2005)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Antecedent Resolution in the Context of MT",
"sec_num": "4"
},
{
"text": "There are more than one interpretations of what Ram can't do: eat out, eat out with you, or eat out with you on Sunday afternoon. A native speaker selects one of these in context. If we are seeking to eliminate ellipsis with our system, we must select one of these interpretations to paste at the site of ellipsis, or the whole verb phrase. We may also not adopt intermediate policies such as importing noun objects but not adjuncts, because this risks placing undue emphasis on certain arguments, and may result in an incorrect semantics.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Antecedent Resolution in the Context of MT",
"sec_num": "4"
},
{
"text": "The second problem with this strategy is that it may render sentences, after pasting, clumsy, unnatural or nonsensical, as illustrated, respectively, by the following example, taken from the WSJ corpus (Bos and Spenader, 2011 ).",
"cite_spans": [
{
"start": 202,
"end": 225,
"text": "(Bos and Spenader, 2011",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Antecedent Resolution in the Context of MT",
"sec_num": "4"
},
{
"text": "The Volokhs were afraid that they'd end up like a friend of theirs who'd applied for a visa and waited for 10 years, having been demoted from his profession of theoretical mathematician to shipping clerk. They didn't (end up like\u2026shipping clerk)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Antecedent Resolution in the Context of MT",
"sec_num": "4"
},
{
"text": "Verb-stranding verb phrase ellipsis is the phenomenon wherein a verb is stranded at a site of ellipsis, and its internal arguments are elided. Raj also read (the paper twice). (Manetta, 2018) Here, 'paRha' provides access to the internal arguments i.e. the direct object and the adverb via VVPE. Manetta supports her analysis theoretically and by a survey across native speakers. This provides us an intuition for a solution for our larger problem: that is, instead of disambiguating the antecedent boundaries to resolve English VPE as simplification for English-Hindi MT, we may provide the Hindi clause containing ellipsis only the head verb, which has access to the internal arguments of the antecedent. If we induce the head verb at the site of ellipsis in the English before translating, now, then we create a valid, syntactical Hindi sentence with all the interpretations of the original sentence intact. We illustrate with an example: Resolving the ellipsis in (b) we get an English sentence with 'eat' at the previous site of ellipsis, that means that Ram is incapable of eating. However, (c) in Hindi, that similarly has 'eat' at previous site of English VPE, is a perfectly acceptable translation of the original sentence, exhibiting VVPE. In (c) the stranded verb 'kha' (eat) has access to the internal arguments 'aapke saath', (with you), 'Sunday dopahar' (Sunday afternoon), etc., and therefore it carries all the interpretations of the original. This is the core idea of the system.",
"cite_spans": [
{
"start": 176,
"end": 191,
"text": "(Manetta, 2018)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Hindi VVPE",
"sec_num": "5"
},
{
"text": "a (Original",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hindi VVPE",
"sec_num": "5"
},
{
"text": "While we do not require to identify the boundaries of the head verb, we do need to identify all the components of the verb. This may be required in several cases, such as phrasal verbs, idioms, or serial verbs.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dealing with Multi-Word Verbs in English",
"sec_num": "6"
},
{
"text": "These can be categorized into phrasal verbs and prepositional verbs. The former is a class of verbs that consists of a head verb and a preposition, like \"take over\", \"get around\" or \"sink in\". Since phrasal verbs are opaque, we require to pick up its preposition to maintain the correct sense in which it appears. For example, in She hasn't got over her old failures as yet, although she should. (created example)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Compound Verbs",
"sec_num": "6.1"
},
{
"text": "We want the stranded segment to be \"get over\", not simply \"get\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Compound Verbs",
"sec_num": "6.1"
},
{
"text": "Prepositional verbs are also verbs followed by a preposition e.g \"stare at\", \"care for\", but they are different from phrasal verbs in certain ways. Their meaning is derived primarily from the head verb. We can see that in the first case, \"stare at\", it would admissible to strand simply \"stare\", although in the subsequent case, the sense of 'care' changes without the preposition. It is necessary, therefore, to pick up the prepositional verb as a unit.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Compound Verbs",
"sec_num": "6.1"
},
{
"text": "In general, when the head verb has verb complement arguments, it is not necessary to pick them up, because the stranded head verb in Hindi will have access to them. For example, (b) is a legitimate translation of (a), with the stranded verb \"wants\". However, this treatment assumes that the sequence of complements in the head verb lexically translate into a sequence of simple verbs in Hindi. This might not be the case. The reason we cannot strand is because 'padega' (has: obligation aspect) is an auxiliary of 'jaana' (to go) in the Hindi sentence, not the head verb of the compound verb \"jaana padega\" (has to go). The head here is \"jaana\", and therefore it must be induced in the site of ellipsis. Therefore, with the English verb \"has\", among others, we must also pick up its complements to induce at the site of ellipsis. To generalize, we require the complement cE of a head verb hE, when hE + cE results in a VV or verb-light verb complex in Hindi.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Verbal Complements",
"sec_num": "6.2"
},
{
"text": "Examples of serial verbs in English are a I'll go see if she's okay b Why don't you run get a taxi?",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Serial Verbs",
"sec_num": "6.3"
},
{
"text": "These can be treated as verb-complement series, as the Stanford Universal Dependency Framework treats them. Hindi also treats the verb complex as a head verb followed by complements. Taking the first example, it is acceptable to strand \"jaati\" (go) from this sentence, in: ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Serial Verbs",
"sec_num": "6.3"
},
{
"text": "We are using a dependency tree of the input sentence for all tasks.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Algorithm to Identify Ellipsis and Head Verb of Antecedent",
"sec_num": "7"
},
{
"text": "We assume that each word w in the input sentence that belongs to our five classes, is a site of ellipsis. Then we go through a process of elimination. We have a different set of rules for each class.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Identification: Rules and Results",
"sec_num": "7.1"
},
{
"text": "Some of the basic criteria for elimination include:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Identification: Rules and Results",
"sec_num": "7.1"
},
{
"text": "1. w is a copula (for to_be) 2. w is an auxiliary child of a verb (for to_be, to_do, to_have, modals)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Identification: Rules and Results",
"sec_num": "7.1"
},
{
"text": "3. w has a direct object noun child (for to_be, to_do, to_have, modals)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Identification: Rules and Results",
"sec_num": "7.1"
},
{
"text": "We perform part 1 of our two-part testing at this stage. These rules, tested on 3900 sentences from the BNC corpus (Leech, 1992), give us a precision of 94.83 percent and a recall of 83.04 percent.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "w is not an xcomp child (for to_particle)",
"sec_num": "4."
},
{
"text": "If we find an ellipsis, then we perform 2 subtasks:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Antecedent Head Resolution: Algorithm",
"sec_num": "7.2"
},
{
"text": "1. Find the head verb 2. Supplement the head verb",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Antecedent Head Resolution: Algorithm",
"sec_num": "7.2"
},
{
"text": "We collect all the verbs in the input, eliminating according to the constraints on to_be/non-to_be ellipsis discussed in Section 3. We use a score-based approach, as does Hardt (Hardt, 1992) .",
"cite_spans": [
{
"start": 177,
"end": 190,
"text": "(Hardt, 1992)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Finding the Head Verb",
"sec_num": "7.2.1"
},
{
"text": "Here are the features that we look at, for verb v as a candidate for ellipsis e: 1. Positive scores for noun subjects matching in number, negative score for noun subjects not matching in 'passivity', positive score if both noun subjects are proper, positive score for identical noun subjects.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Finding the Head Verb",
"sec_num": "7.2.1"
},
{
"text": "2. Negative score if e belongs to the complement clause child of v. For example, in: He told me that he had passed the exam, and then he told me that John hadn't. \"told\" gets ruled out as the clause \"John hadn't\" is a complement child of \"told\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Finding the Head Verb",
"sec_num": "7.2.1"
},
{
"text": "3. Negative score if v belongs to the complement clause child of e. 4. Positive score if v has an auxiliary in the same class as e. For example: Sita could walk while texting, but now she can't.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Finding the Head Verb",
"sec_num": "7.2.1"
},
{
"text": "5. Finally, we assign a positive score to the first verb that we obtain by backtracking up the dependency tree from e, if it is contained in our candidate verbs. This gives us the correct antecedent head several times, even when it is far away. It captures clausal relationships between antecedent verb and e, such as as\u2026as, \u2026than, which Hardt also mentions in their scoring algorithm.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Finding the Head Verb",
"sec_num": "7.2.1"
},
{
"text": "6. We then evaluate the scores. If there is a clash, we choose the closest verb, advantaging forward ellipsis to backward ellipsis, as the latter is widely acknowledged to be much rarer than the former.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Finding the Head Verb",
"sec_num": "7.2.1"
},
{
"text": "As we said earlier, the head verb might need to be supplemented before it is 'stranded'. For example, in:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Supplement Main Verb",
"sec_num": "7.2.2"
},
{
"text": "Both banks have been battered, as other Arizona banks have, by falling real estate prices.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Supplement Main Verb",
"sec_num": "7.2.2"
},
{
"text": "The above algorithm will identify \"battered\" as the head verb; however, we do need to supplement it with the auxiliary \"been\", to create a grammatical verb after inducing.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Supplement Main Verb",
"sec_num": "7.2.2"
},
{
"text": "We perform three sub-tasks in supplementing the main verb v:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Supplement Main Verb",
"sec_num": "7.2.2"
},
{
"text": "1. Add auxiliaries: we add any auxiliaries of v, after skipping the first if any that belongs to the same class as e. For example, we skip the auxiliary \"have\" in the above example.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Supplement Main Verb",
"sec_num": "7.2.2"
},
{
"text": "2. Add particles: here, we check whether v is part of a phrasal verb/prepositional verb, and add the preposition(s) if so. The dependency tree marks particle dependants of the verb: however, since it doesn't always do so, we maintain and import a list of common phrasal verbs/prepositional verbs for reference.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Supplement Main Verb",
"sec_num": "7.2.2"
},
{
"text": "3. Add verb complements. Similarly, we maintain and import a list of verbs of which verb complements, if any, we need to pick up, since they result in nonstrandable verbs in isolation in Hindi: \"let\", \"have\". These also include verbs which may not always give a correct lexical translation in isolation, e.g. \"feel\", \"seem\", for which it is safer to also induce the complements.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Supplement Main Verb",
"sec_num": "7.2.2"
},
{
"text": "The lists above can always be augmented, of course. Currently, our system does not deal with idioms, but one solution that we suggest is to maintain a list of frequently occurring idioms and induce them as a whole.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Supplement Main Verb",
"sec_num": "7.2.2"
},
{
"text": "In part 2 of our testing, we tested this system on 200 sentences from the WSJ corpus and got an accuracy of 91 percent on antecedent head resolution. Here are some errors, and their analysis:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "Mr. Dinkins also has failed to allay Jewish voters' fears about his association with the Rev. Jesse Jackson, despite the fact that few local non-Jewish politicians have been as vocal for Jewish causes in the past 20 years as Mr. Dinkins has.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "Here, the algorithm identifies \"failed\" instead of \"been\". It awards \"failed\" for common noun subject, common auxiliary and being the first verb upon backtracking, whereas \"been\" is only awarded for common auxiliary.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "But Sony also says in its filing that the Warner contract \"doesn't require that Guber and Peters take any affirmative steps to produce motion pictures; it simply rewards them when they do and prohibits them from producing for another entertainment company.\"",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "Here the algorithm identifies \"rewards\" as the source verb instead of \"take\". It awards \"take\" for noun subject number, \"require\" for common auxiliary, and \"rewards\" by backtracking. Finally, it resolves the tie by choosing \"rewards\" which is the closest. Now they know who you mean and you know who you mean -but no one else does.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "The algorithm gives \"mean\" instead of \"know\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "\"mean\" gets awarded for noun subject number, and \"know\" is awarded for backtracking; however, \"mean\" wins the tie since it is the closest.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "There are errors introduced by the POS tagger; for example, in:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "A good half-hour into breakfast at the Palmer House, Mr. O'Brien looks up from his plate after Mr. Straszheim says something about people who believe interest rates are about to nosedive -\"I'm one of them who hope they will, with 6 billion in debt on the books. \"nosedive\" is not recognized as a verb, similar to, in another examples, \"program-trade\", and \"move\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "There are errors introduced by the parser:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "The text by Patrick O'Connor is a tough read, but the pictures make her magnetism clear and help explain why Ernest Hemmingway called Baker, \"The most sensational women anybody ever saw -or ever will.\"",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "Here, the algorithm wrongly identifies \"called\" as the antecedent main verb, instead of \"saw\". The clause \"or ever will\" is labelled a conjugate dependent of the noun Baker, instead of a conjunct of the verb \"saw\". If this had been so, \"saw\" would have got scores for a common subject (\"anybody\") and being the verb obtained upon backtracking.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "\"A lot of people think I will give away the store, but I can assure you I will not,\" he says.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "The algorithm identifies \"assure\". This would be avoided if the dependency marked \"I will not\" as a complement clause of \"assure\" -however the dependency misses this relation. If \"assure\" was given a penalty on this grounds, the next highest candidate is indeed the correct one: \"give\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results and Error Analysis for Antecedent Head Resolution",
"sec_num": "8"
},
{
"text": "We now show that inducing the head elliptical verb makes the input sentence easier for the MT system to comprehend. We perform a manual analysis of 100 sentences with ellipsis, taken from the WSJ corpus: we create a \"before\" and \"after\" translation pair for each, and compare to identify improvements. This was done by two fluent speakers of Hindi and English. We define a scale (1-5) to quantify this improvement:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "5: Improvement from incoherent to perfect translation of ellipsis, and surroundings 4: The meaning is fairly clearer than it was in the original 3: The translation is as good or as bad as it originally was 2: The meaning is fairly more obscure than it originally was 1: The sentence is rendered completely incoherent from an original good translation",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "We add some flags to further nuance this scale: we also mark the translations for fluidity, i.e. if the translation while rendered better is still not fluent (f) or if the translation while not making the meaning clearer is rendered more fluent (F), and for the overall meaning of the entire sentence -i.e. beyond the clauses of the antecedent and the ellipsis. These markers, however, are only for exceptional cases, since most of the translations we got, both \"before\" and \"after\" were not fluent.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "The average score over 100 sentences was found to be 3.55, with 18 cases of correct nonfluent sentences, and 5 cases of incorrect sentences with especial improvement in fluency.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "These are the large-scale sources of lack of improvement that we found: 1. Sentences that the MT cannot translate overall due to other complexities such as nested clauses, etc., for which its output is close to gibberish, show little to no improvement with addition of the elliptical verb. These sentences passed through a system that can handle them, sometimes Google Translate, almost always show high improvement with the addition of the verb. We had about 35 such sentences, all marked 3, sometimes marked for improvement or deterioration in fluency.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "2. Most sentences of the type \"as did\", as in 'X ate apples as did Y' fail to show any improvement and often show a deterioration in fluency. This is because the processed sentence is rendered ungrammatical and perhaps more incomprehensible.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "Again, Google Translate often shows improvement on these sentences from \"before\" to \"after\".",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "An example of the first instance is:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "(1) American Enterprise Institute scholar Norman Ornstein in the Oct. 21 TV Guide on \"What TV News Doesn't Report About Congressand Should\"",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "The system induces the elliptical verb complex \"report\". Both Anusaaraka and Google output incorrect translations for the original sentence, although Google shows errors only due to the ellipsis. Therefore, it is able to improve its translation after we induce the elliptical verb:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "(2) amerikee (Score: 3f, for deterioration in fluency)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "The original gives a similarly incorrect output, though not quite as ungrammatical. This output, even after the elliptical verb has been added, conveys little to no meaning in Hindi.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "Here are some micro-level sources of nonimprovement:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "1. When an antecedent verb is being used idiomatically, the MT system may interpret it literally when stranded in the elliptical clause, even if it catches the correct sense in the antecedent clause, possible because the former construction is more unusual.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "2. In general, the VVPE construction fails if the system makes two different lexical (Score: 4)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "We note here that these figures are dependent upon how well the base translation system can translate the original. We performed the same analysis on samples from this dataset with Google Translation and got consistently better results per each batch of 10, and an average of 3.7. This indicative exercise is intended to give an idea of why targeted VPE handling for specific language pairs holds significance in bettering MT results.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation of Effect on MT outputs",
"sec_num": "9"
},
{
"text": "The concept and the system that we have introduced above, for handling ellipsis in a targeted manner to improve English-Hindi MT, are still in their nascent stages. There are three primary entry points for future work: firstly, the conceptual negotiation of the phenomenon in English and Hindi. We have decided, as we explain, only to induce the main verb. While this gives satisfactory results most of the times, it also fails in some cases: it might help, for example, to make a decision to induce object arguments in these cases. Secondly, the identification of VPE in English, and the antecedent resolution. We are already dealing with complex verbs, verbal complements etc. in a certain manner, but this treatment invites further and more rigorous work, both in terms in nuance with the treatment, and how exhaustive we are with the lists that we have drawn up, introducing, for example, treatment of idioms. Thirdly, the application of our system as over the input, before it is passed through the MT. There are certain problems that may be solved by pipelining this ellipsis handling differently into the MT system: we leave this to future investigation.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Future Work",
"sec_num": "10"
},
{
"text": "https://ltrc.iiit.ac.in/Anusaaraka/anu_ home.html",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "interpretations in the antecedent and elliptical clause. This may be for different reasons: e.g. on transitive verbs, since they appear in the elliptical verb without their objects, which is unnatural. The MT system will possibly attempt to catch an intransitive sense of the verb in such a situation. However, this is only in certain few cases of such verbs.An example of both of the above is:During the takeover, Mr. Hahn said he would put his account up for review if WPP's bid were successful, but he didn't.The system induces the elliptical verb complex \"put up\". Both Anusaaraka and Google output incorrect translations for the original sentence, although Google interprets the idiomatic meaning of \"put up\" correctly. Therefore, it is able to improve its translation after we induce the elliptical verb:Anusaaraka makes error type 1 (literal interpretation of verb): 4 (Score: 3) Google makes error type 2 (different lexical interpretation). In this case, it is minor, because the meaning of the sentence is restored from the original. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "annex",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "An annotated corpus for the analysis of vp ellipsis",
"authors": [
{
"first": "J",
"middle": [],
"last": "Bos",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Spenader",
"suffix": ""
}
],
"year": 2011,
"venue": "Language Resources and Evaluation",
"volume": "45",
"issue": "4",
"pages": "463--494",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "J. Bos and J. Spenader. An annotated corpus for the analysis of vp ellipsis. Language Resources and Evaluation, 45(4):463-494, 2011.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "An algorithm for vp ellipsis",
"authors": [
{
"first": "D",
"middle": [],
"last": "Hardt",
"suffix": ""
}
],
"year": 1992,
"venue": "Proceedings of the 30th annual meeting on Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "9--14",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "D. Hardt. An algorithm for vp ellipsis. In Proceed- ings of the 30th annual meeting on Association for Computational Linguistics, pages 9-14. As- sociation for Computational Linguistics, 1992.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Verb phrase ellipsis resolution using discriminative and margin-infused algorithms",
"authors": [
{
"first": "K",
"middle": [],
"last": "Kenyon-Dean",
"suffix": ""
},
{
"first": "J",
"middle": [
"C K"
],
"last": "Cheung",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Precup",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "1734--1743",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "K. Kenyon-Dean, J. C. K. Cheung, and D. Pre- cup. Verb phrase ellipsis resolution using dis- criminative and margin-infused algorithms. In Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 1734-1743, 2016.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Leech. 100 million words of english: the british national corpus (bnc)",
"authors": [
{
"first": "G",
"middle": [
"N"
],
"last": "",
"suffix": ""
}
],
"year": 1992,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "G. N. Leech. 100 million words of english: the british national corpus (bnc). 1992.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Exploring the steps of verb phrase ellipsis",
"authors": [
{
"first": "Z",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "E",
"middle": [
"G"
],
"last": "Pellicer",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Gillick",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the Workshop on Coreference Resolution Beyond OntoNotes (CORBON 2016)",
"volume": "",
"issue": "",
"pages": "32--40",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Z. Liu, E. G. Pellicer, and D. Gillick. Exploring the steps of verb phrase ellipsis. In Proceedings of the Workshop on Coreference Resolution Be- yond OntoNotes (CORBON 2016), pages 32-40, 2016.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Verb-phrase ellipsis and complex predicates in hindi-urdu",
"authors": [
{
"first": "E",
"middle": [],
"last": "Manetta",
"suffix": ""
}
],
"year": 2018,
"venue": "Natural Language & Linguistic Theory",
"volume": "",
"issue": "",
"pages": "1--39",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "E. Manetta. Verb-phrase ellipsis and complex predicates in hindi-urdu. Natural Language & Linguistic Theory, pages 1-39, 2018.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "A corpus-based study of Verb Phrase Ellipsis Identification and Resolution. PhD thesis",
"authors": [
{
"first": "L",
"middle": [
"A"
],
"last": "Nielsen",
"suffix": ""
}
],
"year": 2005,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "L. A. Nielsen. A corpus-based study of Verb Phrase Ellipsis Identification and Resolution. PhD the- sis, King's College London, 2005.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "The design for the wall street journal-based csr corpus",
"authors": [
{
"first": "D",
"middle": [
"B"
],
"last": "Paul",
"suffix": ""
},
{
"first": "J",
"middle": [
"M"
],
"last": "Baker",
"suffix": ""
}
],
"year": 1992,
"venue": "Proceedings of the workshop on Speech and Natural Language",
"volume": "",
"issue": "",
"pages": "357--362",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "D. B. Paul and J. M. Baker. The design for the wall street journal-based csr corpus. In Proceed- ings of the workshop on Speech and Natural Lan- guage, pages 357-362. Association for Computa- tional Linguistics, 1992.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "When a good translation is wrong in context: Contextaware machine translation improves on deixis, ellipsis, and lexical cohesion",
"authors": [
{
"first": "E",
"middle": [],
"last": "Voita",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Sennrich",
"suffix": ""
},
{
"first": "I",
"middle": [],
"last": "Titov",
"suffix": ""
}
],
"year": 2019,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"arXiv": [
"arXiv:1905.05979"
]
},
"num": null,
"urls": [],
"raw_text": "E. Voita, R. Sennrich, and I. Titov. When a good translation is wrong in context: Context- aware machine translation improves on deixis, ellipsis, and lexical cohesion. arXiv preprint arXiv:1905.05979, 2019.",
"links": null
}
},
"ref_entries": {}
}
} |