id sid tid token lemma pos bracis-28376 1 1 evaluating evaluate VERB bracis-28376 1 2 recent recent ADJ bracis-28376 1 3 legal legal ADJ bracis-28376 1 4 rhetorical rhetorical ADJ bracis-28376 1 5 role role NOUN bracis-28376 1 6 labeling labeling NOUN bracis-28376 1 7 approaches approach NOUN bracis-28376 1 8 supported support VERB bracis-28376 1 9 by by ADP bracis-28376 1 10     SPACE bracis-28376 1 11 transformer transformer NOUN bracis-28376 1 12 encoders encoder NOUN bracis-28376 1 13 | | ADV bracis-28376 1 14 springer springer NOUN bracis-28376 1 15 nature nature PROPN bracis-28376 1 16 link link PROPN bracis-28376 1 17 ( ( PUNCT bracis-28376 1 18 formerly formerly ADV bracis-28376 1 19 springerlink springerlink NOUN bracis-28376 1 20 ) ) PUNCT bracis-28376 1 21 skip skip VERB bracis-28376 1 22 to to ADP bracis-28376 1 23 main main ADJ bracis-28376 1 24 content content NOUN bracis-28376 1 25 advertisement advertisement NOUN bracis-28376 1 26 log log NOUN bracis-28376 1 27 in in ADP bracis-28376 1 28 menu menu NOUN bracis-28376 1 29 find find VERB bracis-28376 1 30 a a DET bracis-28376 1 31 journal journal NOUN bracis-28376 1 32 publish publish VERB bracis-28376 1 33 with with ADP bracis-28376 1 34 us we PRON bracis-28376 1 35 track track VERB bracis-28376 1 36 your your PRON bracis-28376 1 37 research research NOUN bracis-28376 1 38 search search NOUN bracis-28376 1 39 cart cart NOUN bracis-28376 1 40 home home NOUN bracis-28376 1 41 intelligent intelligent ADJ bracis-28376 1 42 systems system NOUN bracis-28376 1 43 conference conference NOUN bracis-28376 1 44 paper paper NOUN bracis-28376 1 45 evaluating evaluate VERB bracis-28376 1 46 recent recent ADJ bracis-28376 1 47 legal legal ADJ bracis-28376 1 48 rhetorical rhetorical ADJ bracis-28376 1 49 role role NOUN bracis-28376 1 50 labeling labeling NOUN bracis-28376 1 51 approaches approach NOUN bracis-28376 1 52 supported support VERB bracis-28376 1 53 by by ADP bracis-28376 1 54     SPACE bracis-28376 1 55 transformer transformer NOUN bracis-28376 1 56 encoders encoder NOUN bracis-28376 1 57 conference conference NOUN bracis-28376 1 58 paper paper NOUN bracis-28376 1 59 first first ADV bracis-28376 1 60 online online ADV bracis-28376 1 61 : : PUNCT bracis-28376 1 62 12 12 NUM bracis-28376 1 63 october october NOUN bracis-28376 1 64 2023 2023 NUM bracis-28376 1 65 pp pp ADP bracis-28376 1 66 18–32 18–32 NUM bracis-28376 1 67 cite cite VERB bracis-28376 1 68 this this DET bracis-28376 1 69 conference conference NOUN bracis-28376 1 70 paper paper NOUN bracis-28376 1 71 access access NOUN bracis-28376 1 72 provided provide VERB bracis-28376 1 73 by by ADP bracis-28376 1 74 university university PROPN bracis-28376 1 75 of of ADP bracis-28376 1 76 notre notre PROPN bracis-28376 1 77 dame dame PROPN bracis-28376 1 78 hesburgh hesburgh PROPN bracis-28376 1 79 library library PROPN bracis-28376 1 80 download download PROPN bracis-28376 1 81 book book NOUN bracis-28376 1 82 pdf pdf PROPN bracis-28376 1 83 download download NOUN bracis-28376 1 84 book book NOUN bracis-28376 1 85 epub epub PROPN bracis-28376 1 86 intelligent intelligent ADJ bracis-28376 1 87 systems system NOUN bracis-28376 1 88 ( ( PUNCT bracis-28376 1 89 bracis bracis NOUN bracis-28376 1 90 2023 2023 NUM bracis-28376 1 91 ) ) PUNCT bracis-28376 1 92 evaluating evaluate VERB bracis-28376 1 93 recent recent ADJ bracis-28376 1 94 legal legal ADJ bracis-28376 1 95 rhetorical rhetorical ADJ bracis-28376 1 96 role role NOUN bracis-28376 1 97 labeling labeling NOUN bracis-28376 1 98 approaches approach NOUN bracis-28376 1 99 supported support VERB bracis-28376 1 100 by by ADP bracis-28376 1 101     SPACE bracis-28376 1 102 transformer transformer NOUN bracis-28376 1 103 encoders encoder NOUN bracis-28376 1 104 download download NOUN bracis-28376 1 105 book book NOUN bracis-28376 1 106 pdf pdf PROPN bracis-28376 1 107 download download NOUN bracis-28376 1 108 book book PROPN bracis-28376 1 109 epub epub PROPN bracis-28376 1 110 alexandre alexandre PROPN bracis-28376 1 111 gomes gomes PROPN bracis-28376 1 112 de de PROPN bracis-28376 1 113 lima lima PROPN bracis-28376 1 114     SPACE bracis-28376 1 115 orcid orcid NOUN bracis-28376 1 116 : : PUNCT bracis-28376 1 117 orcid.org/0000-0001-7251-00089,10 orcid.org/0000-0001-7251-00089,10 PROPN bracis-28376 1 118 , , PUNCT bracis-28376 1 119 josé josé PROPN bracis-28376 1 120 g. g. PROPN bracis-28376 1 121 moreno moreno PROPN bracis-28376 1 122     SPACE bracis-28376 1 123 orcid orcid PROPN bracis-28376 1 124 : : PUNCT bracis-28376 1 125 orcid.org/0000-0002-8852-579711 orcid.org/0000-0002-8852-579711 NOUN bracis-28376 1 126 , , PUNCT bracis-28376 1 127 taoufiq taoufiq NOUN bracis-28376 1 128 dkaki11 dkaki11 PROPN bracis-28376 1 129 , , PUNCT bracis-28376 1 130 eduardo eduardo PROPN bracis-28376 1 131 henrique henrique PROPN bracis-28376 1 132 da da PROPN bracis-28376 1 133 s. s. PROPN bracis-28376 1 134 aranha aranha AUX bracis-28376 1 135     SPACE bracis-28376 1 136 orcid orcid NOUN bracis-28376 1 137 : : PUNCT bracis-28376 1 138 orcid.org/0000-0002-8446-638x10 orcid.org/0000-0002-8446-638x10 NOUN bracis-28376 1 139 & & CCONJ bracis-28376 1 140 … … PUNCT bracis-28376 1 141 mohand mohand NOUN bracis-28376 1 142 boughanem11 boughanem11 NOUN bracis-28376 1 143     SPACE bracis-28376 1 144 show show VERB bracis-28376 1 145 authors author NOUN bracis-28376 1 146 part part NOUN bracis-28376 1 147 of of ADP bracis-28376 1 148 the the DET bracis-28376 1 149 book book NOUN bracis-28376 1 150 series series NOUN bracis-28376 1 151 : : PUNCT bracis-28376 1 152 lecture lecture NOUN bracis-28376 1 153 notes note NOUN bracis-28376 1 154 in in ADP bracis-28376 1 155 computer computer NOUN bracis-28376 1 156 science science NOUN bracis-28376 1 157 ( ( PUNCT bracis-28376 1 158 ( ( PUNCT bracis-28376 1 159 lnai lnai ADJ bracis-28376 1 160 , , PUNCT bracis-28376 1 161 volume volume NOUN bracis-28376 1 162 14196 14196 NUM bracis-28376 1 163 ) ) PUNCT bracis-28376 1 164 ) ) PUNCT bracis-28376 1 165 included include VERB bracis-28376 1 166 in in ADP bracis-28376 1 167 the the DET bracis-28376 1 168 following follow VERB bracis-28376 1 169 conference conference NOUN bracis-28376 1 170 series series NOUN bracis-28376 1 171 : : PUNCT bracis-28376 1 172 brazilian brazilian ADJ bracis-28376 1 173 conference conference NOUN bracis-28376 1 174 on on ADP bracis-28376 1 175 intelligent intelligent ADJ bracis-28376 1 176 systems system NOUN bracis-28376 1 177 510 510 NUM bracis-28376 1 178 accesses access NOUN bracis-28376 1 179 1 1 NUM bracis-28376 1 180 citation citation NOUN bracis-28376 1 181 abstract abstract ADJ bracis-28376 1 182 pre pre ADJ bracis-28376 1 183 - - ADJ bracis-28376 1 184 trained train VERB bracis-28376 1 185 transformer transformer NOUN bracis-28376 1 186 models model NOUN bracis-28376 1 187 have have AUX bracis-28376 1 188 been be AUX bracis-28376 1 189 used use VERB bracis-28376 1 190 to to PART bracis-28376 1 191 improve improve VERB bracis-28376 1 192 the the DET bracis-28376 1 193 results result NOUN bracis-28376 1 194 of of ADP bracis-28376 1 195 several several ADJ bracis-28376 1 196 nlp nlp NOUN bracis-28376 1 197 tasks task NOUN bracis-28376 1 198 , , PUNCT bracis-28376 1 199 which which PRON bracis-28376 1 200 includes include VERB bracis-28376 1 201 the the DET bracis-28376 1 202 legal legal ADJ bracis-28376 1 203 rhetorical rhetorical ADJ bracis-28376 1 204 role role NOUN bracis-28376 1 205 labeling labeling NOUN bracis-28376 1 206 ( ( PUNCT bracis-28376 1 207 legal legal ADJ bracis-28376 1 208 rrl rrl PROPN bracis-28376 1 209 ) ) PUNCT bracis-28376 1 210 one one NUM bracis-28376 1 211 . . PUNCT bracis-28376 2 1 this this DET bracis-28376 2 2 task task NOUN bracis-28376 2 3 assigns assign VERB bracis-28376 2 4 semantic semantic ADJ bracis-28376 2 5 functions function NOUN bracis-28376 2 6 , , PUNCT bracis-28376 2 7 such such ADJ bracis-28376 2 8 as as ADP bracis-28376 2 9 fact fact NOUN bracis-28376 2 10 and and CCONJ bracis-28376 2 11 argument argument NOUN bracis-28376 2 12 , , PUNCT bracis-28376 2 13 to to ADP bracis-28376 2 14 sentences sentence NOUN bracis-28376 2 15 from from ADP bracis-28376 2 16 judgment judgment NOUN bracis-28376 2 17 documents document NOUN bracis-28376 2 18 . . PUNCT bracis-28376 3 1 several several ADJ bracis-28376 3 2 legal legal ADJ bracis-28376 3 3 rrl rrl NOUN bracis-28376 3 4 works work VERB bracis-28376 3 5 exploit exploit VERB bracis-28376 3 6 pre pre ADJ bracis-28376 3 7 - - ADJ bracis-28376 3 8 trained train VERB bracis-28376 3 9 transformers transformer NOUN bracis-28376 3 10 to to PART bracis-28376 3 11 encode encode VERB bracis-28376 3 12 sentences sentence NOUN bracis-28376 3 13 but but CCONJ bracis-28376 3 14 only only ADV bracis-28376 3 15 a a DET bracis-28376 3 16 few few ADJ bracis-28376 3 17 employ employ NOUN bracis-28376 3 18 approaches approach NOUN bracis-28376 3 19 other other ADJ bracis-28376 3 20 than than ADP bracis-28376 3 21 fine fine ADV bracis-28376 3 22 - - PUNCT bracis-28376 3 23 tuning tuning NOUN bracis-28376 3 24 to to PART bracis-28376 3 25 improve improve VERB bracis-28376 3 26 the the DET bracis-28376 3 27 performance performance NOUN bracis-28376 3 28 of of ADP bracis-28376 3 29 models model NOUN bracis-28376 3 30 . . PUNCT bracis-28376 4 1 in in ADP bracis-28376 4 2 this this DET bracis-28376 4 3 work work NOUN bracis-28376 4 4 , , PUNCT bracis-28376 4 5 we we PRON bracis-28376 4 6 implement implement VERB bracis-28376 4 7 three three NUM bracis-28376 4 8 of of ADP bracis-28376 4 9 such such ADJ bracis-28376 4 10 approaches approach NOUN bracis-28376 4 11 and and CCONJ bracis-28376 4 12 evaluate evaluate VERB bracis-28376 4 13 them they PRON bracis-28376 4 14 over over ADP bracis-28376 4 15 the the DET bracis-28376 4 16 same same ADJ bracis-28376 4 17 datasets dataset NOUN bracis-28376 4 18 to to PART bracis-28376 4 19 achieve achieve VERB bracis-28376 4 20 a a DET bracis-28376 4 21 better well ADJ bracis-28376 4 22 perception perception NOUN bracis-28376 4 23 of of ADP bracis-28376 4 24 their their PRON bracis-28376 4 25 impacts impact NOUN bracis-28376 4 26 . . PUNCT bracis-28376 5 1 in in ADP bracis-28376 5 2 our our PRON bracis-28376 5 3 experiments experiment NOUN bracis-28376 5 4 , , PUNCT bracis-28376 5 5 approaches approach NOUN bracis-28376 5 6 based base VERB bracis-28376 5 7 on on ADP bracis-28376 5 8 data datum NOUN bracis-28376 5 9 augmentation augmentation NOUN bracis-28376 5 10 and and CCONJ bracis-28376 5 11 positional positional ADJ bracis-28376 5 12 encoders encoder NOUN bracis-28376 5 13 do do AUX bracis-28376 5 14 not not PART bracis-28376 5 15 provide provide VERB bracis-28376 5 16 performance performance NOUN bracis-28376 5 17 gains gain NOUN bracis-28376 5 18 to to ADP bracis-28376 5 19 our our PRON bracis-28376 5 20 models model NOUN bracis-28376 5 21 . . PUNCT bracis-28376 6 1 conversely conversely ADV bracis-28376 6 2 , , PUNCT bracis-28376 6 3 the the DET bracis-28376 6 4 models model NOUN bracis-28376 6 5 based base VERB bracis-28376 6 6 on on ADP bracis-28376 6 7 the the DET bracis-28376 6 8 dfcsc dfcsc NOUN bracis-28376 6 9 approach approach NOUN bracis-28376 6 10 overcome overcome VERB bracis-28376 6 11 the the DET bracis-28376 6 12 appropriate appropriate ADJ bracis-28376 6 13 baselines baseline NOUN bracis-28376 6 14 , , PUNCT bracis-28376 6 15 and and CCONJ bracis-28376 6 16 they they PRON bracis-28376 6 17 do do VERB bracis-28376 6 18 remarkably remarkably ADV bracis-28376 6 19 well well ADV bracis-28376 6 20 as as ADP bracis-28376 6 21 the the DET bracis-28376 6 22 lowest low ADJ bracis-28376 6 23 and and CCONJ bracis-28376 6 24 highest high ADJ bracis-28376 6 25 improvements improvement NOUN bracis-28376 6 26 respectively respectively ADV bracis-28376 6 27 are be AUX bracis-28376 6 28 5.9 5.9 NUM bracis-28376 6 29 % % NOUN bracis-28376 6 30 and and CCONJ bracis-28376 6 31 10.4 10.4 NUM bracis-28376 6 32 % % NOUN bracis-28376 6 33 . . PUNCT bracis-28376 7 1 this this DET bracis-28376 7 2 work work NOUN bracis-28376 7 3 was be AUX bracis-28376 7 4 supported support VERB bracis-28376 7 5 by by ADP bracis-28376 7 6 npad npad PROPN bracis-28376 7 7 / / SYM bracis-28376 7 8 ufrn ufrn PROPN bracis-28376 7 9 , , PUNCT bracis-28376 7 10 by by ADP bracis-28376 7 11 the the DET bracis-28376 7 12 coordenação coordenação PROPN bracis-28376 7 13 de de PROPN bracis-28376 7 14 aperfeiçoamento aperfeiçoamento PROPN bracis-28376 7 15 de de X bracis-28376 7 16 pessoal pessoal PROPN bracis-28376 7 17 de de PROPN bracis-28376 7 18 nível nível PROPN bracis-28376 7 19 superior superior PROPN bracis-28376 7 20 brasil brasil PROPN bracis-28376 7 21 ( ( PUNCT bracis-28376 7 22 capes cape NOUN bracis-28376 7 23 , , PUNCT bracis-28376 7 24 finance finance NOUN bracis-28376 7 25 code code NOUN bracis-28376 7 26 001 001 NUM bracis-28376 7 27 ) ) PUNCT bracis-28376 7 28 and and CCONJ bracis-28376 7 29 by by ADP bracis-28376 7 30 the the DET bracis-28376 7 31 lawbot lawbot ADJ bracis-28376 7 32 project project NOUN bracis-28376 7 33 ( ( PUNCT bracis-28376 7 34 anr-20 anr-20 NOUN bracis-28376 7 35 - - PUNCT bracis-28376 7 36 ce38 ce38 PROPN bracis-28376 7 37 - - PUNCT bracis-28376 7 38 0013 0013 NUM bracis-28376 7 39 ) ) PUNCT bracis-28376 7 40 , , PUNCT bracis-28376 7 41 granted grant VERB bracis-28376 7 42 by by ADP bracis-28376 7 43 anr anr PROPN bracis-28376 7 44 the the DET bracis-28376 7 45 french french PROPN bracis-28376 7 46 agence agence PROPN bracis-28376 7 47 nationale nationale PROPN bracis-28376 7 48 de de X bracis-28376 7 49 la la X bracis-28376 7 50 recherche recherche PROPN bracis-28376 7 51 . . PUNCT bracis-28376 8 1 access access NOUN bracis-28376 8 2 provided provide VERB bracis-28376 8 3 by by ADP bracis-28376 8 4 university university PROPN bracis-28376 8 5 of of ADP bracis-28376 8 6 notre notre PROPN bracis-28376 8 7 dame dame PROPN bracis-28376 8 8 hesburgh hesburgh PROPN bracis-28376 8 9 library library PROPN bracis-28376 8 10 . . PUNCT bracis-28376 9 1 download download PROPN bracis-28376 9 2 conference conference NOUN bracis-28376 9 3 paper paper NOUN bracis-28376 9 4 pdf pdf NOUN bracis-28376 9 5 similar similar ADJ bracis-28376 9 6 content content NOUN bracis-28376 9 7 being be AUX bracis-28376 9 8 viewed view VERB bracis-28376 9 9 by by ADP bracis-28376 9 10 others other NOUN bracis-28376 9 11 deeprhole deeprhole PROPN bracis-28376 9 12 : : PUNCT bracis-28376 9 13 deep deep ADJ bracis-28376 9 14 learning learning NOUN bracis-28376 9 15 for for ADP bracis-28376 9 16 rhetorical rhetorical ADJ bracis-28376 9 17 role role NOUN bracis-28376 9 18 labeling labeling NOUN bracis-28376 9 19 of of ADP bracis-28376 9 20 sentences sentence NOUN bracis-28376 9 21 in in ADP bracis-28376 9 22 legal legal ADJ bracis-28376 9 23 case case NOUN bracis-28376 9 24 documents document NOUN bracis-28376 9 25 article article NOUN bracis-28376 9 26 13 13 NUM bracis-28376 9 27 november november PROPN bracis-28376 9 28 2021 2021 NUM bracis-28376 9 29 marro marro X bracis-28376 9 30 : : PUNCT bracis-28376 9 31 multi multi ADJ bracis-28376 9 32 - - ADJ bracis-28376 9 33 headed headed ADJ bracis-28376 9 34 attention attention NOUN bracis-28376 9 35 for for ADP bracis-28376 9 36 rhetorical rhetorical ADJ bracis-28376 9 37 role role NOUN bracis-28376 9 38 labeling labeling NOUN bracis-28376 9 39 in in ADP bracis-28376 9 40 legal legal ADJ bracis-28376 9 41 documents document NOUN bracis-28376 9 42 article article NOUN bracis-28376 9 43 18 18 NUM bracis-28376 9 44 march march NOUN bracis-28376 9 45 2025 2025 NUM bracis-28376 9 46 rhetorical rhetorical ADJ bracis-28376 9 47 role role NOUN bracis-28376 9 48 prediction prediction NOUN bracis-28376 9 49 in in ADP bracis-28376 9 50     SPACE bracis-28376 9 51 legal legal ADJ bracis-28376 9 52 documents document NOUN bracis-28376 9 53 : : PUNCT bracis-28376 9 54 a a DET bracis-28376 9 55 comparison comparison NOUN bracis-28376 9 56 of of ADP bracis-28376 9 57     SPACE bracis-28376 9 58 machine machine NOUN bracis-28376 9 59 learning learn VERB bracis-28376 9 60 algorithms algorithm NOUN bracis-28376 9 61 chapter chapter NOUN bracis-28376 9 62 © © PROPN bracis-28376 9 63 2025 2025 NUM bracis-28376 9 64 explore explore VERB bracis-28376 9 65 related relate VERB bracis-28376 9 66 subjects subject NOUN bracis-28376 9 67 discover discover VERB bracis-28376 9 68 the the DET bracis-28376 9 69 latest late ADJ bracis-28376 9 70 articles article NOUN bracis-28376 9 71 , , PUNCT bracis-28376 9 72 books book NOUN bracis-28376 9 73 and and CCONJ bracis-28376 9 74 news news NOUN bracis-28376 9 75 in in ADP bracis-28376 9 76 related related ADJ bracis-28376 9 77 subjects subject NOUN bracis-28376 9 78 , , PUNCT bracis-28376 9 79 suggested suggest VERB bracis-28376 9 80 using use VERB bracis-28376 9 81 machine machine NOUN bracis-28376 9 82 learning learning NOUN bracis-28376 9 83 . . PUNCT bracis-28376 10 1 computational computational ADJ bracis-28376 10 2 linguistics linguistics PROPN bracis-28376 10 3 corpus corpus PROPN bracis-28376 10 4 linguistics linguistics PROPN bracis-28376 10 5 literature literature PROPN bracis-28376 10 6 and and CCONJ bracis-28376 10 7 law law NOUN bracis-28376 10 8 natural natural ADJ bracis-28376 10 9 language language NOUN bracis-28376 10 10 processing processing NOUN bracis-28376 10 11 ( ( PUNCT bracis-28376 10 12 nlp nlp NOUN bracis-28376 10 13 ) ) PUNCT bracis-28376 10 14 research research NOUN bracis-28376 10 15 methods method NOUN bracis-28376 10 16 in in ADP bracis-28376 10 17 language language NOUN bracis-28376 10 18 and and CCONJ bracis-28376 10 19 linguistics linguistics PROPN bracis-28376 10 20 speech speech NOUN bracis-28376 10 21 act act PROPN bracis-28376 10 22 theory theory NOUN bracis-28376 10 23 1 1 NUM bracis-28376 10 24 introduction introduction NOUN bracis-28376 10 25 one one NUM bracis-28376 10 26 of of ADP bracis-28376 10 27 the the DET bracis-28376 10 28 main main ADJ bracis-28376 10 29 goals goal NOUN bracis-28376 10 30 of of ADP bracis-28376 10 31 legal legal ADJ bracis-28376 10 32 ai ai NOUN bracis-28376 10 33 , , PUNCT bracis-28376 10 34 the the DET bracis-28376 10 35 application application NOUN bracis-28376 10 36 of of ADP bracis-28376 10 37 artificial artificial ADJ bracis-28376 10 38 intelligence intelligence NOUN bracis-28376 10 39 ( ( PUNCT bracis-28376 10 40 ai ai NOUN bracis-28376 10 41 ) ) PUNCT bracis-28376 10 42 in in ADP bracis-28376 10 43 the the DET bracis-28376 10 44 legal legal ADJ bracis-28376 10 45 domain domain NOUN bracis-28376 10 46 , , PUNCT bracis-28376 10 47 is be AUX bracis-28376 10 48 task task NOUN bracis-28376 10 49 automation automation NOUN bracis-28376 10 50 , , PUNCT bracis-28376 10 51 which which PRON bracis-28376 10 52 increases increase VERB bracis-28376 10 53 the the DET bracis-28376 10 54 productivity productivity NOUN bracis-28376 10 55 of of ADP bracis-28376 10 56 legal legal ADJ bracis-28376 10 57 professionals professional NOUN bracis-28376 10 58 and and CCONJ bracis-28376 10 59 makes make VERB bracis-28376 10 60 the the DET bracis-28376 10 61 law law NOUN bracis-28376 10 62 more more ADV bracis-28376 10 63 accessible accessible ADJ bracis-28376 10 64 . . PUNCT bracis-28376 11 1 several several ADJ bracis-28376 11 2 works work NOUN bracis-28376 11 3 related relate VERB bracis-28376 11 4 to to ADP bracis-28376 11 5 legal legal ADJ bracis-28376 11 6 ai ai NOUN bracis-28376 11 7 are be AUX bracis-28376 11 8 about about ADP bracis-28376 11 9 tasks task NOUN bracis-28376 11 10 concerning concern VERB bracis-28376 11 11 judgments judgment NOUN bracis-28376 11 12 such such ADJ bracis-28376 11 13 as as ADP bracis-28376 11 14 judgment judgment NOUN bracis-28376 11 15 prediction prediction NOUN bracis-28376 11 16 [ [ X bracis-28376 11 17 14 14 NUM bracis-28376 11 18 , , PUNCT bracis-28376 11 19 24 24 NUM bracis-28376 11 20 ] ] PUNCT bracis-28376 11 21 , , PUNCT bracis-28376 11 22 and and CCONJ bracis-28376 11 23 legal legal ADJ bracis-28376 11 24 summarization summarization NOUN bracis-28376 11 25 [ [ X bracis-28376 11 26 7 7 NUM bracis-28376 11 27 , , PUNCT bracis-28376 11 28 9 9 NUM bracis-28376 11 29 ] ] PUNCT bracis-28376 11 30 , , PUNCT bracis-28376 11 31 which which PRON bracis-28376 11 32 rely rely VERB bracis-28376 11 33 on on ADP bracis-28376 11 34 judgment judgment NOUN bracis-28376 11 35 documents document NOUN bracis-28376 11 36 as as ADP bracis-28376 11 37 the the DET bracis-28376 11 38 main main ADJ bracis-28376 11 39 source source NOUN bracis-28376 11 40 of of ADP bracis-28376 11 41 data datum NOUN bracis-28376 11 42 . . PUNCT bracis-28376 12 1 a a DET bracis-28376 12 2 judgment judgment NOUN bracis-28376 12 3 document document NOUN bracis-28376 12 4 is be AUX bracis-28376 12 5 a a DET bracis-28376 12 6 record record NOUN bracis-28376 12 7 of of ADP bracis-28376 12 8 a a DET bracis-28376 12 9 lawsuit lawsuit NOUN bracis-28376 12 10 authored author VERB bracis-28376 12 11 by by ADP bracis-28376 12 12 the the DET bracis-28376 12 13 court court NOUN bracis-28376 12 14 that that PRON bracis-28376 12 15 sets set VERB bracis-28376 12 16 forth forth ADP bracis-28376 12 17 its its PRON bracis-28376 12 18 decision decision NOUN bracis-28376 12 19 about about ADP bracis-28376 12 20 the the DET bracis-28376 12 21 case case NOUN bracis-28376 12 22 . . PUNCT bracis-28376 13 1 the the DET bracis-28376 13 2 document document NOUN bracis-28376 13 3 includes include VERB bracis-28376 13 4 such such ADJ bracis-28376 13 5 as as ADP bracis-28376 13 6 statements statement NOUN bracis-28376 13 7 of of ADP bracis-28376 13 8 the the DET bracis-28376 13 9 facts fact NOUN bracis-28376 13 10 , , PUNCT bracis-28376 13 11 analysis analysis NOUN bracis-28376 13 12 of of ADP bracis-28376 13 13 relevant relevant ADJ bracis-28376 13 14 law law NOUN bracis-28376 13 15 , , PUNCT bracis-28376 13 16 rights right NOUN bracis-28376 13 17 and and CCONJ bracis-28376 13 18 liabilities liability NOUN bracis-28376 13 19 of of ADP bracis-28376 13 20 the the DET bracis-28376 13 21 parties party NOUN bracis-28376 13 22 , , PUNCT bracis-28376 13 23 and and CCONJ bracis-28376 13 24 rulings ruling NOUN bracis-28376 13 25 , , PUNCT bracis-28376 13 26 among among ADP bracis-28376 13 27 others other NOUN bracis-28376 13 28 . . PUNCT bracis-28376 14 1 specific specific ADJ bracis-28376 14 2 information information NOUN bracis-28376 14 3 is be AUX bracis-28376 14 4 required require VERB bracis-28376 14 5 or or CCONJ bracis-28376 14 6 can can AUX bracis-28376 14 7 be be AUX bracis-28376 14 8 exploited exploit VERB bracis-28376 14 9 to to PART bracis-28376 14 10 boost boost VERB bracis-28376 14 11 legal legal ADJ bracis-28376 14 12 ai ai NOUN bracis-28376 14 13 tasks task NOUN bracis-28376 14 14 : : PUNCT bracis-28376 14 15 facts fact NOUN bracis-28376 14 16 for for ADP bracis-28376 14 17 fact fact NOUN bracis-28376 14 18 - - PUNCT bracis-28376 14 19 based base VERB bracis-28376 14 20 search search NOUN bracis-28376 14 21 ; ; PUNCT bracis-28376 14 22 facts fact NOUN bracis-28376 14 23 and and CCONJ bracis-28376 14 24 violated violate VERB bracis-28376 14 25 statutes statute NOUN bracis-28376 14 26 for for ADP bracis-28376 14 27 judgment judgment NOUN bracis-28376 14 28 prediction prediction NOUN bracis-28376 14 29 ; ; PUNCT bracis-28376 14 30 facts fact NOUN bracis-28376 14 31 , , PUNCT bracis-28376 14 32 arguments argument NOUN bracis-28376 14 33 , , PUNCT bracis-28376 14 34 and and CCONJ bracis-28376 14 35 judicial judicial ADJ bracis-28376 14 36 reasoning reasoning NOUN bracis-28376 14 37 for for ADP bracis-28376 14 38 legal legal ADJ bracis-28376 14 39 summarization summarization NOUN bracis-28376 14 40 . . PUNCT bracis-28376 15 1 the the DET bracis-28376 15 2 legal legal ADJ bracis-28376 15 3 rhetoric rhetoric NOUN bracis-28376 15 4 role role NOUN bracis-28376 15 5 labeling labeling NOUN bracis-28376 15 6 ( ( PUNCT bracis-28376 15 7 legal legal ADJ bracis-28376 15 8 rrl rrl PROPN bracis-28376 15 9 ) ) PUNCT bracis-28376 15 10 task task NOUN bracis-28376 15 11 assigns assign NOUN bracis-28376 15 12 semantic semantic ADJ bracis-28376 15 13 functions function NOUN bracis-28376 15 14 to to ADP bracis-28376 15 15 sentences sentence NOUN bracis-28376 15 16 in in ADP bracis-28376 15 17 judgment judgment NOUN bracis-28376 15 18 documents document NOUN bracis-28376 15 19 ( ( PUNCT bracis-28376 15 20 e.g. e.g. ADV bracis-28376 15 21 , , PUNCT bracis-28376 15 22 fact fact NOUN bracis-28376 15 23 , , PUNCT bracis-28376 15 24 argument argument NOUN bracis-28376 15 25 , , PUNCT bracis-28376 15 26 ruling ruling NOUN bracis-28376 15 27 ) ) PUNCT bracis-28376 15 28 . . PUNCT bracis-28376 16 1 thus thus ADV bracis-28376 16 2 , , PUNCT bracis-28376 16 3 legal legal ADJ bracis-28376 16 4 rrl rrl NOUN bracis-28376 16 5 is be AUX bracis-28376 16 6 a a DET bracis-28376 16 7 way way NOUN bracis-28376 16 8 of of ADP bracis-28376 16 9 finding find VERB bracis-28376 16 10 information information NOUN bracis-28376 16 11 that that PRON bracis-28376 16 12 is be AUX bracis-28376 16 13 useful useful ADJ bracis-28376 16 14 in in ADP bracis-28376 16 15 itself itself PRON bracis-28376 16 16 as as ADV bracis-28376 16 17 well well ADV bracis-28376 16 18 as as ADP bracis-28376 16 19 for for ADP bracis-28376 16 20 downstream downstream ADJ bracis-28376 16 21 tasks task NOUN bracis-28376 16 22 . . PUNCT bracis-28376 17 1 the the DET bracis-28376 17 2 legal legal ADJ bracis-28376 17 3 rrl rrl NOUN bracis-28376 17 4 task task NOUN bracis-28376 17 5 is be AUX bracis-28376 17 6 not not PART bracis-28376 17 7 trivial trivial ADJ bracis-28376 17 8 , , PUNCT bracis-28376 17 9 even even ADV bracis-28376 17 10 for for ADP bracis-28376 17 11 humans human NOUN bracis-28376 17 12 . . PUNCT bracis-28376 18 1 one one NUM bracis-28376 18 2 of of ADP bracis-28376 18 3 the the DET bracis-28376 18 4 limiting limit VERB bracis-28376 18 5 factors factor NOUN bracis-28376 18 6 is be AUX bracis-28376 18 7 the the DET bracis-28376 18 8 heterogeneity heterogeneity NOUN bracis-28376 18 9 in in ADP bracis-28376 18 10 the the DET bracis-28376 18 11 structure structure NOUN bracis-28376 18 12 of of ADP bracis-28376 18 13 documents document NOUN bracis-28376 18 14 . . PUNCT bracis-28376 19 1 it it PRON bracis-28376 19 2 is be AUX bracis-28376 19 3 common common ADJ bracis-28376 19 4 for for SCONJ bracis-28376 19 5 a a DET bracis-28376 19 6 lawsuit lawsuit NOUN bracis-28376 19 7 to to PART bracis-28376 19 8 be be AUX bracis-28376 19 9 judged judge VERB bracis-28376 19 10 by by ADP bracis-28376 19 11 different different ADJ bracis-28376 19 12 courts court NOUN bracis-28376 19 13 ( ( PUNCT bracis-28376 19 14 e.g. e.g. ADV bracis-28376 19 15 , , PUNCT bracis-28376 19 16 lower low ADJ bracis-28376 19 17 courts court NOUN bracis-28376 19 18 , , PUNCT bracis-28376 19 19 supreme supreme PROPN bracis-28376 19 20 court court PROPN bracis-28376 19 21 ) ) PUNCT bracis-28376 19 22 , , PUNCT bracis-28376 19 23 which which PRON bracis-28376 19 24 may may AUX bracis-28376 19 25 adopt adopt VERB bracis-28376 19 26 different different ADJ bracis-28376 19 27 standards standard NOUN bracis-28376 19 28 or or CCONJ bracis-28376 19 29 guidelines guideline NOUN bracis-28376 19 30 to to PART bracis-28376 19 31 write write VERB bracis-28376 19 32 judgment judgment NOUN bracis-28376 19 33 documents document NOUN bracis-28376 19 34 . . PUNCT bracis-28376 20 1 another another DET bracis-28376 20 2 issue issue NOUN bracis-28376 20 3 is be AUX bracis-28376 20 4 the the DET bracis-28376 20 5 subjectivity subjectivity NOUN bracis-28376 20 6 related relate VERB bracis-28376 20 7 to to ADP bracis-28376 20 8 the the DET bracis-28376 20 9 interpretation interpretation NOUN bracis-28376 20 10 of of ADP bracis-28376 20 11 different different ADJ bracis-28376 20 12 rhetorical rhetorical ADJ bracis-28376 20 13 roles role NOUN bracis-28376 20 14     SPACE bracis-28376 20 15 [ [ X bracis-28376 20 16 3 3 NUM bracis-28376 20 17 , , PUNCT bracis-28376 20 18 15 15 NUM bracis-28376 20 19 ] ] PUNCT bracis-28376 20 20 . . PUNCT bracis-28376 21 1 inter inter ADJ bracis-28376 21 2 - - ADJ bracis-28376 21 3 annotator annotator ADJ bracis-28376 21 4 agreements agreement NOUN bracis-28376 21 5 reported report VERB bracis-28376 21 6 from from ADP bracis-28376 21 7 works work NOUN bracis-28376 21 8 about about ADP bracis-28376 21 9 the the DET bracis-28376 21 10 development development NOUN bracis-28376 21 11 of of ADP bracis-28376 21 12 legal legal ADJ bracis-28376 21 13 rrl rrl NOUN bracis-28376 21 14 datasets dataset NOUN bracis-28376 21 15 are be AUX bracis-28376 21 16 good good ADJ bracis-28376 21 17 evidence evidence NOUN bracis-28376 21 18 of of ADP bracis-28376 21 19 such such ADJ bracis-28376 21 20 subjectivity subjectivity NOUN bracis-28376 21 21 . . PUNCT bracis-28376 22 1 in in ADP bracis-28376 22 2 [ [ X bracis-28376 22 3 15 15 NUM bracis-28376 22 4 ] ] PUNCT bracis-28376 22 5 , , PUNCT bracis-28376 22 6 the the DET bracis-28376 22 7 authors author NOUN bracis-28376 22 8 report report VERB bracis-28376 22 9 a a DET bracis-28376 22 10 low low ADJ bracis-28376 22 11 inter inter ADJ bracis-28376 22 12 - - NOUN bracis-28376 22 13 agreement agreement ADJ bracis-28376 22 14 score score NOUN bracis-28376 22 15 ( ( PUNCT bracis-28376 22 16 kappa kappa ADJ bracis-28376 22 17 index index NOUN bracis-28376 22 18 is be AUX bracis-28376 22 19 0.65 0.65 NUM bracis-28376 22 20 ) ) PUNCT bracis-28376 22 21 among among ADP bracis-28376 22 22 three three NUM bracis-28376 22 23 annotators annotator NOUN bracis-28376 22 24 on on ADP bracis-28376 22 25 judgments judgment NOUN bracis-28376 22 26 from from ADP bracis-28376 22 27 the the DET bracis-28376 22 28 income income NOUN bracis-28376 22 29 tax tax NOUN bracis-28376 22 30 domain domain NOUN bracis-28376 22 31 over over ADP bracis-28376 22 32 1,600 1,600 NUM bracis-28376 22 33 sentences sentence NOUN bracis-28376 22 34 . . PUNCT bracis-28376 23 1 similarly similarly ADV bracis-28376 23 2 , , PUNCT bracis-28376 23 3 in in ADP bracis-28376 23 4 [ [ X bracis-28376 23 5 8 8 NUM bracis-28376 23 6 ] ] X bracis-28376 23 7 a a DET bracis-28376 23 8 low low ADJ bracis-28376 23 9 inter inter ADJ bracis-28376 23 10 - - NOUN bracis-28376 23 11 agreement agreement ADJ bracis-28376 23 12 score score NOUN bracis-28376 23 13 ( ( PUNCT bracis-28376 23 14 kappa kappa ADJ bracis-28376 23 15 index index NOUN bracis-28376 23 16 is be AUX bracis-28376 23 17 0.59 0.59 NUM bracis-28376 23 18 ) ) PUNCT bracis-28376 23 19 is be AUX bracis-28376 23 20 reported report VERB bracis-28376 23 21 for for ADP bracis-28376 23 22 40,305 40,305 NUM bracis-28376 23 23 sentences sentence NOUN bracis-28376 23 24 and and CCONJ bracis-28376 23 25 30 30 NUM bracis-28376 23 26 annotators annotator NOUN bracis-28376 23 27 . . PUNCT bracis-28376 24 1 legal legal ADJ bracis-28376 24 2 rrl rrl PROPN bracis-28376 24 3 automation automation NOUN bracis-28376 24 4 can can AUX bracis-28376 24 5 be be AUX bracis-28376 24 6 implemented implement VERB bracis-28376 24 7 as as ADP bracis-28376 24 8 a a DET bracis-28376 24 9 sentence sentence NOUN bracis-28376 24 10 classification classification NOUN bracis-28376 24 11 task task NOUN bracis-28376 24 12 , , PUNCT bracis-28376 24 13 where where SCONJ bracis-28376 24 14 sentences sentence NOUN bracis-28376 24 15 are be AUX bracis-28376 24 16 fed feed VERB bracis-28376 24 17 into into ADP bracis-28376 24 18 a a DET bracis-28376 24 19 machine machine NOUN bracis-28376 24 20 learning learn VERB bracis-28376 24 21 model model NOUN bracis-28376 24 22 that that PRON bracis-28376 24 23 assigns assign VERB bracis-28376 24 24 a a DET bracis-28376 24 25 label label NOUN bracis-28376 24 26 to to ADP bracis-28376 24 27 each each DET bracis-28376 24 28 sentence sentence NOUN bracis-28376 24 29 . . PUNCT bracis-28376 25 1 the the DET bracis-28376 25 2 model model NOUN bracis-28376 25 3 consists consist VERB bracis-28376 25 4 of of ADP bracis-28376 25 5 two two NUM bracis-28376 25 6 core core NOUN bracis-28376 25 7 components component NOUN bracis-28376 25 8 , , PUNCT bracis-28376 25 9 a a DET bracis-28376 25 10 sentence sentence NOUN bracis-28376 25 11 encoder encoder NOUN bracis-28376 25 12 and and CCONJ bracis-28376 25 13 a a DET bracis-28376 25 14 classifier classifier NOUN bracis-28376 25 15 . . PUNCT bracis-28376 26 1 the the DET bracis-28376 26 2 sentence sentence NOUN bracis-28376 26 3 encoder encoder NOUN bracis-28376 26 4 is be AUX bracis-28376 26 5 responsible responsible ADJ bracis-28376 26 6 to to PART bracis-28376 26 7 produce produce VERB bracis-28376 26 8 numeric numeric ADJ bracis-28376 26 9 representations representation NOUN bracis-28376 26 10 ( ( PUNCT bracis-28376 26 11 i.e. i.e. X bracis-28376 26 12 , , PUNCT bracis-28376 26 13 vectors vector NOUN bracis-28376 26 14 ) ) PUNCT bracis-28376 26 15 , , PUNCT bracis-28376 26 16 commonly commonly ADV bracis-28376 26 17 known know VERB bracis-28376 26 18 as as ADP bracis-28376 26 19 sentence sentence NOUN bracis-28376 26 20 embeddings embedding NOUN bracis-28376 26 21 , , PUNCT bracis-28376 26 22 of of ADP bracis-28376 26 23 text text NOUN bracis-28376 26 24 sentences sentence NOUN bracis-28376 26 25 while while SCONJ bracis-28376 26 26 the the DET bracis-28376 26 27 classifier classifier NOUN bracis-28376 26 28 assigns assign VERB bracis-28376 26 29 labels label NOUN bracis-28376 26 30 to to ADP bracis-28376 26 31 each each DET bracis-28376 26 32 representation representation NOUN bracis-28376 26 33 as as SCONJ bracis-28376 26 34 illustrated illustrate VERB bracis-28376 26 35 in in ADP bracis-28376 26 36 fig fig NOUN bracis-28376 26 37 . . PUNCT bracis-28376 26 38     SPACE bracis-28376 27 1 1 1 NUM bracis-28376 27 2 . . PUNCT bracis-28376 27 3 fig fig NOUN bracis-28376 27 4 . . PUNCT bracis-28376 28 1 1 1 X bracis-28376 28 2 . . X bracis-28376 28 3 sentence sentence NOUN bracis-28376 28 4 classification classification NOUN bracis-28376 28 5 task task NOUN bracis-28376 28 6 . . PUNCT bracis-28376 29 1 full full ADJ bracis-28376 29 2 size size NOUN bracis-28376 29 3 image image NOUN bracis-28376 29 4 the the DET bracis-28376 29 5 representation representation NOUN bracis-28376 29 6 vectors vector NOUN bracis-28376 29 7 directly directly ADV bracis-28376 29 8 impact impact VERB bracis-28376 29 9 the the DET bracis-28376 29 10 performance performance NOUN bracis-28376 29 11 of of ADP bracis-28376 29 12 the the DET bracis-28376 29 13 model model NOUN bracis-28376 29 14 since since SCONJ bracis-28376 29 15 richer rich ADJ bracis-28376 29 16 representations representation NOUN bracis-28376 29 17 lead lead VERB bracis-28376 29 18 to to ADP bracis-28376 29 19 more more ADV bracis-28376 29 20 precise precise ADJ bracis-28376 29 21 classifications classification NOUN bracis-28376 29 22     SPACE bracis-28376 30 1 [ [ X bracis-28376 30 2 21 21 NUM bracis-28376 30 3 ] ] PUNCT bracis-28376 30 4 . . PUNCT bracis-28376 31 1 the the DET bracis-28376 31 2 exploitation exploitation NOUN bracis-28376 31 3 of of ADP bracis-28376 31 4 pre pre ADJ bracis-28376 31 5 - - ADJ bracis-28376 31 6 trained train VERB bracis-28376 31 7 transformer transformer NOUN bracis-28376 31 8 models model NOUN bracis-28376 31 9     SPACE bracis-28376 32 1 [ [ X bracis-28376 32 2 20 20 NUM bracis-28376 32 3 ] ] PUNCT bracis-28376 32 4 is be AUX bracis-28376 32 5 the the DET bracis-28376 32 6 current current ADJ bracis-28376 32 7 state state NOUN bracis-28376 32 8 - - PUNCT bracis-28376 32 9 of of ADP bracis-28376 32 10 - - PUNCT bracis-28376 32 11 the the DET bracis-28376 32 12 - - PUNCT bracis-28376 32 13 art art NOUN bracis-28376 32 14 in in ADP bracis-28376 32 15 many many ADJ bracis-28376 32 16 nlp nlp ADJ bracis-28376 32 17 tasks task NOUN bracis-28376 32 18 and and CCONJ bracis-28376 32 19 this this PRON bracis-28376 32 20 includes include VERB bracis-28376 32 21 the the DET bracis-28376 32 22 sentence sentence NOUN bracis-28376 32 23 encoding encode VERB bracis-28376 32 24 one one NUM bracis-28376 32 25 . . PUNCT bracis-28376 33 1 several several ADJ bracis-28376 33 2 pre pre ADJ bracis-28376 33 3 - - ADJ bracis-28376 33 4 trained train VERB bracis-28376 33 5 transformer transformer NOUN bracis-28376 33 6 models model NOUN bracis-28376 33 7 are be AUX bracis-28376 33 8 publicly publicly ADV bracis-28376 33 9 available available ADJ bracis-28376 33 10 , , PUNCT bracis-28376 33 11 such such ADJ bracis-28376 33 12 as as ADP bracis-28376 33 13 bert bert PROPN bracis-28376 33 14     SPACE bracis-28376 34 1 [ [ X bracis-28376 34 2 5 5 NUM bracis-28376 34 3 ] ] PUNCT bracis-28376 34 4 and and CCONJ bracis-28376 34 5 roberta roberta PROPN bracis-28376 34 6     SPACE bracis-28376 35 1 [ [ X bracis-28376 35 2 13 13 NUM bracis-28376 35 3 ] ] PUNCT bracis-28376 35 4 . . PUNCT bracis-28376 36 1 when when SCONJ bracis-28376 36 2 generating generate VERB bracis-28376 36 3 the the DET bracis-28376 36 4 vector vector NOUN bracis-28376 36 5 representation representation NOUN bracis-28376 36 6 of of ADP bracis-28376 36 7 a a DET bracis-28376 36 8 word word NOUN bracis-28376 36 9 w w NOUN bracis-28376 36 10 in in ADP bracis-28376 36 11 a a DET bracis-28376 36 12 sentence sentence NOUN bracis-28376 36 13 , , PUNCT bracis-28376 36 14 a a DET bracis-28376 36 15 transformer transformer NOUN bracis-28376 36 16 model model NOUN bracis-28376 36 17 takes take VERB bracis-28376 36 18 w w NOUN bracis-28376 36 19 and and CCONJ bracis-28376 36 20 its its PRON bracis-28376 36 21 neighbor neighbor NOUN bracis-28376 36 22 words word NOUN bracis-28376 36 23 into into ADP bracis-28376 36 24 account account NOUN bracis-28376 36 25 . . PUNCT bracis-28376 37 1 therefore therefore ADV bracis-28376 37 2 , , PUNCT bracis-28376 37 3 it it PRON bracis-28376 37 4 yields yield VERB bracis-28376 37 5 different different ADJ bracis-28376 37 6 vector vector NOUN bracis-28376 37 7 representations representation NOUN bracis-28376 37 8 for for ADP bracis-28376 37 9 w w NOUN bracis-28376 37 10 when when SCONJ bracis-28376 37 11 it it PRON bracis-28376 37 12 occurs occur VERB bracis-28376 37 13 in in ADP bracis-28376 37 14 different different ADJ bracis-28376 37 15 sentences sentence NOUN bracis-28376 37 16 or or CCONJ bracis-28376 37 17 different different ADJ bracis-28376 37 18 positions position NOUN bracis-28376 37 19 of of ADP bracis-28376 37 20 the the DET bracis-28376 37 21 same same ADJ bracis-28376 37 22 sentence sentence NOUN bracis-28376 37 23 . . PUNCT bracis-28376 38 1 this this DET bracis-28376 38 2 context context NOUN bracis-28376 38 3 exploitation exploitation NOUN bracis-28376 38 4 ability ability NOUN bracis-28376 38 5 allows allow VERB bracis-28376 38 6 the the DET bracis-28376 38 7 model model NOUN bracis-28376 38 8 to to PART bracis-28376 38 9 yield yield VERB bracis-28376 38 10 rich rich ADJ bracis-28376 38 11 text text NOUN bracis-28376 38 12 representations representation NOUN bracis-28376 38 13 . . PUNCT bracis-28376 39 1 fine fine ADJ bracis-28376 39 2 - - PUNCT bracis-28376 39 3 tuning tuning NOUN bracis-28376 39 4 is be AUX bracis-28376 39 5 a a DET bracis-28376 39 6 common common ADJ bracis-28376 39 7 procedure procedure NOUN bracis-28376 39 8 for for ADP bracis-28376 39 9 exploiting exploit VERB bracis-28376 39 10 pre pre ADJ bracis-28376 39 11 - - ADJ bracis-28376 39 12 trained trained ADJ bracis-28376 39 13 transformers transformer NOUN bracis-28376 39 14 that that PRON bracis-28376 39 15 adjusts adjust VERB bracis-28376 39 16 the the DET bracis-28376 39 17 weights weight NOUN bracis-28376 39 18 of of ADP bracis-28376 39 19 a a DET bracis-28376 39 20 pre pre ADJ bracis-28376 39 21 - - ADJ bracis-28376 39 22 trained train VERB bracis-28376 39 23 model model NOUN bracis-28376 39 24 for for ADP bracis-28376 39 25 a a DET bracis-28376 39 26 specific specific ADJ bracis-28376 39 27 task task NOUN bracis-28376 39 28 and and CCONJ bracis-28376 39 29 dataset dataset NOUN bracis-28376 39 30 . . PUNCT bracis-28376 40 1 this this PRON bracis-28376 40 2 is be AUX bracis-28376 40 3 a a DET bracis-28376 40 4 general general ADJ bracis-28376 40 5 procedure procedure NOUN bracis-28376 40 6 and and CCONJ bracis-28376 40 7 it it PRON bracis-28376 40 8 can can AUX bracis-28376 40 9 be be AUX bracis-28376 40 10 employed employ VERB bracis-28376 40 11 to to PART bracis-28376 40 12 tackle tackle VERB bracis-28376 40 13 the the DET bracis-28376 40 14 legal legal ADJ bracis-28376 40 15 rrl rrl NOUN bracis-28376 40 16 task task NOUN bracis-28376 40 17     SPACE bracis-28376 41 1 [ [ X bracis-28376 41 2 1 1 NUM bracis-28376 41 3 , , PUNCT bracis-28376 41 4 3 3 NUM bracis-28376 41 5 , , PUNCT bracis-28376 41 6 8 8 NUM bracis-28376 41 7 , , PUNCT bracis-28376 41 8 18 18 NUM bracis-28376 41 9 ] ] PUNCT bracis-28376 41 10 . . PUNCT bracis-28376 42 1 some some DET bracis-28376 42 2 recent recent ADJ bracis-28376 42 3 works work NOUN bracis-28376 42 4 [ [ X bracis-28376 42 5 10,11,12 10,11,12 NUM bracis-28376 42 6 , , PUNCT bracis-28376 42 7 15 15 NUM bracis-28376 42 8 ] ] PUNCT bracis-28376 42 9 propose propose VERB bracis-28376 42 10 additional additional ADJ bracis-28376 42 11 procedures procedure NOUN bracis-28376 42 12 as as ADP bracis-28376 42 13 a a DET bracis-28376 42 14 way way NOUN bracis-28376 42 15 of of ADP bracis-28376 42 16 improving improve VERB bracis-28376 42 17 the the DET bracis-28376 42 18 performance performance NOUN bracis-28376 42 19 of of ADP bracis-28376 42 20 models model NOUN bracis-28376 42 21 handling handle VERB bracis-28376 42 22 the the DET bracis-28376 42 23 legal legal ADJ bracis-28376 42 24 rrl rrl NOUN bracis-28376 42 25 task task NOUN bracis-28376 42 26 but but CCONJ bracis-28376 42 27 , , PUNCT bracis-28376 42 28 despite despite SCONJ bracis-28376 42 29 their their PRON bracis-28376 42 30 potential potential NOUN bracis-28376 42 31 , , PUNCT bracis-28376 42 32 such such ADJ bracis-28376 42 33 works work NOUN bracis-28376 42 34 employ employ VERB bracis-28376 42 35 different different ADJ bracis-28376 42 36 datasets dataset NOUN bracis-28376 42 37 what what PRON bracis-28376 42 38 hinders hinder VERB bracis-28376 42 39 a a DET bracis-28376 42 40 fair fair ADJ bracis-28376 42 41 comparison comparison NOUN bracis-28376 42 42 . . PUNCT bracis-28376 43 1 in in ADP bracis-28376 43 2 this this DET bracis-28376 43 3 context context NOUN bracis-28376 43 4 , , PUNCT bracis-28376 43 5 the the DET bracis-28376 43 6 goal goal NOUN bracis-28376 43 7 of of ADP bracis-28376 43 8 this this DET bracis-28376 43 9 work work NOUN bracis-28376 43 10 is be AUX bracis-28376 43 11 to to PART bracis-28376 43 12 perform perform VERB bracis-28376 43 13 an an DET bracis-28376 43 14 equitable equitable ADJ bracis-28376 43 15 comparison comparison NOUN bracis-28376 43 16 among among ADP bracis-28376 43 17 approaches approach NOUN bracis-28376 43 18 proposed propose VERB bracis-28376 43 19 in in ADP bracis-28376 43 20 recent recent ADJ bracis-28376 43 21 works work NOUN bracis-28376 43 22     SPACE bracis-28376 44 1 [ [ X bracis-28376 44 2 10,11,12 10,11,12 NUM bracis-28376 44 3 , , PUNCT bracis-28376 44 4 15 15 NUM bracis-28376 44 5 ] ] PUNCT bracis-28376 44 6 . . PUNCT bracis-28376 45 1 to to PART bracis-28376 45 2 achieve achieve VERB bracis-28376 45 3 this this PRON bracis-28376 45 4 , , PUNCT bracis-28376 45 5 we we PRON bracis-28376 45 6 implement implement VERB bracis-28376 45 7 models model NOUN bracis-28376 45 8 based base VERB bracis-28376 45 9 on on ADP bracis-28376 45 10 three three NUM bracis-28376 45 11 of of ADP bracis-28376 45 12 such such ADJ bracis-28376 45 13 approaches approach NOUN bracis-28376 45 14 and and CCONJ bracis-28376 45 15 evaluate evaluate VERB bracis-28376 45 16 them they PRON bracis-28376 45 17 over over ADP bracis-28376 45 18 two two NUM bracis-28376 45 19 datasets dataset NOUN bracis-28376 45 20 . . PUNCT bracis-28376 46 1 only only ADV bracis-28376 46 2 the the DET bracis-28376 46 3 dfcsc dfcsc NOUN bracis-28376 46 4 - - PUNCT bracis-28376 46 5 based base VERB bracis-28376 46 6 models model NOUN bracis-28376 46 7 overcome overcome VERB bracis-28376 46 8 the the DET bracis-28376 46 9 appropriate appropriate ADJ bracis-28376 46 10 baselines baseline NOUN bracis-28376 46 11 , , PUNCT bracis-28376 46 12 and and CCONJ bracis-28376 46 13 they they PRON bracis-28376 46 14 do do VERB bracis-28376 46 15 so so ADV bracis-28376 46 16 remarkably remarkably ADV bracis-28376 46 17 well well ADV bracis-28376 46 18 : : PUNCT bracis-28376 46 19 the the DET bracis-28376 46 20 lowest low ADJ bracis-28376 46 21 and and CCONJ bracis-28376 46 22 highest high ADJ bracis-28376 46 23 improvements improvement NOUN bracis-28376 46 24 respectively respectively ADV bracis-28376 46 25 are be AUX bracis-28376 46 26 5.9 5.9 NUM bracis-28376 46 27 % % NOUN bracis-28376 46 28 and and CCONJ bracis-28376 46 29 10.4 10.4 NUM bracis-28376 46 30 % % NOUN bracis-28376 46 31 . . PUNCT bracis-28376 47 1 2 2 NUM bracis-28376 47 2 recent recent ADJ bracis-28376 47 3 approaches approach NOUN bracis-28376 47 4 to to PART bracis-28376 47 5     SPACE bracis-28376 47 6 handle handle VERB bracis-28376 47 7 legal legal ADJ bracis-28376 47 8 rrl rrl NOUN bracis-28376 47 9 pre pre ADJ bracis-28376 47 10 - - ADJ bracis-28376 47 11 trained train VERB bracis-28376 47 12 transformer transformer NOUN bracis-28376 47 13 models model NOUN bracis-28376 47 14 have have AUX bracis-28376 47 15 achieved achieve VERB bracis-28376 47 16 remarkable remarkable ADJ bracis-28376 47 17 results result NOUN bracis-28376 47 18 in in ADP bracis-28376 47 19 many many ADJ bracis-28376 47 20 nlp nlp NOUN bracis-28376 47 21 tasks task NOUN bracis-28376 47 22 . . PUNCT bracis-28376 48 1 thus thus ADV bracis-28376 48 2 , , PUNCT bracis-28376 48 3 it it PRON bracis-28376 48 4 is be AUX bracis-28376 48 5 not not PART bracis-28376 48 6 a a DET bracis-28376 48 7 surprise surprise NOUN bracis-28376 48 8 the the DET bracis-28376 48 9 publication publication NOUN bracis-28376 48 10 of of ADP bracis-28376 48 11 legal legal ADJ bracis-28376 48 12 rrl rrl NOUN bracis-28376 48 13 works work NOUN bracis-28376 48 14 that that PRON bracis-28376 48 15 rely rely VERB bracis-28376 48 16 on on ADP bracis-28376 48 17 such such ADJ bracis-28376 48 18 models model NOUN bracis-28376 48 19     SPACE bracis-28376 49 1 [ [ X bracis-28376 49 2 1 1 NUM bracis-28376 49 3 , , PUNCT bracis-28376 49 4 3 3 NUM bracis-28376 49 5 , , PUNCT bracis-28376 49 6 8 8 NUM bracis-28376 49 7 , , PUNCT bracis-28376 49 8 18 18 NUM bracis-28376 49 9 ] ] PUNCT bracis-28376 49 10 . . PUNCT bracis-28376 50 1 fine fine ADJ bracis-28376 50 2 - - PUNCT bracis-28376 50 3 tuning tuning NOUN bracis-28376 50 4 is be AUX bracis-28376 50 5 a a DET bracis-28376 50 6 common common ADJ bracis-28376 50 7 training training NOUN bracis-28376 50 8 procedure procedure NOUN bracis-28376 50 9 for for ADP bracis-28376 50 10 one one NUM bracis-28376 50 11 working work VERB bracis-28376 50 12 with with ADP bracis-28376 50 13 pre pre ADJ bracis-28376 50 14 - - ADJ bracis-28376 50 15 trained train VERB bracis-28376 50 16 models model NOUN bracis-28376 50 17 , , PUNCT bracis-28376 50 18 whose whose DET bracis-28376 50 19 goal goal NOUN bracis-28376 50 20 is be AUX bracis-28376 50 21 the the DET bracis-28376 50 22 adjustment adjustment NOUN bracis-28376 50 23 of of ADP bracis-28376 50 24 the the DET bracis-28376 50 25 model model NOUN bracis-28376 50 26 weights weight VERB bracis-28376 50 27 to to ADP bracis-28376 50 28 the the DET bracis-28376 50 29 task task NOUN bracis-28376 50 30 at at ADP bracis-28376 50 31 hand hand NOUN bracis-28376 50 32 . . PUNCT bracis-28376 51 1 a a DET bracis-28376 51 2 pre pre ADJ bracis-28376 51 3 - - ADJ bracis-28376 51 4 trained train VERB bracis-28376 51 5 transformer transformer NOUN bracis-28376 51 6 model model NOUN bracis-28376 51 7 can can AUX bracis-28376 51 8 be be AUX bracis-28376 51 9 employed employ VERB bracis-28376 51 10 out out ADP bracis-28376 51 11 - - PUNCT bracis-28376 51 12 of of ADP bracis-28376 51 13 - - PUNCT bracis-28376 51 14 the the DET bracis-28376 51 15 - - PUNCT bracis-28376 51 16 box box NOUN bracis-28376 51 17 ( ( PUNCT bracis-28376 51 18 i.e. i.e. X bracis-28376 51 19 , , PUNCT bracis-28376 51 20 without without ADP bracis-28376 51 21 fine fine ADV bracis-28376 51 22 - - PUNCT bracis-28376 51 23 tuning tuning NOUN bracis-28376 51 24 ) ) PUNCT bracis-28376 51 25 but but CCONJ bracis-28376 51 26 since since SCONJ bracis-28376 51 27 the the DET bracis-28376 51 28 task task NOUN bracis-28376 51 29 at at ADP bracis-28376 51 30 hand hand NOUN bracis-28376 51 31 is be AUX bracis-28376 51 32 mostly mostly ADV bracis-28376 51 33 different different ADJ bracis-28376 51 34 from from ADP bracis-28376 51 35 the the DET bracis-28376 51 36 task task NOUN bracis-28376 51 37 used use VERB bracis-28376 51 38 in in ADP bracis-28376 51 39 its its PRON bracis-28376 51 40 pre pre NOUN bracis-28376 51 41 - - NOUN bracis-28376 51 42 training training NOUN bracis-28376 51 43 , , PUNCT bracis-28376 51 44 the the DET bracis-28376 51 45 fine fine ADV bracis-28376 51 46 - - PUNCT bracis-28376 51 47 tuning tuning NOUN bracis-28376 51 48 improves improve VERB bracis-28376 51 49 the the DET bracis-28376 51 50 performance performance NOUN bracis-28376 51 51 of of ADP bracis-28376 51 52 the the DET bracis-28376 51 53 model model NOUN bracis-28376 51 54 . . PUNCT bracis-28376 52 1 other other ADJ bracis-28376 52 2 training training NOUN bracis-28376 52 3 procedures procedure NOUN bracis-28376 52 4 can can AUX bracis-28376 52 5 be be AUX bracis-28376 52 6 employed employ VERB bracis-28376 52 7 together together ADV bracis-28376 52 8 with with ADP bracis-28376 52 9 the the DET bracis-28376 52 10 fine fine ADV bracis-28376 52 11 - - PUNCT bracis-28376 52 12 tuning tune VERB bracis-28376 52 13 one one NUM bracis-28376 52 14 as as ADP bracis-28376 52 15 a a DET bracis-28376 52 16 way way NOUN bracis-28376 52 17 of of ADP bracis-28376 52 18 improving improve VERB bracis-28376 52 19 the the DET bracis-28376 52 20 performance performance NOUN bracis-28376 52 21 of of ADP bracis-28376 52 22 a a DET bracis-28376 52 23 model model NOUN bracis-28376 52 24 . . PUNCT bracis-28376 53 1 regarding regard VERB bracis-28376 53 2 the the DET bracis-28376 53 3 legal legal ADJ bracis-28376 53 4 rrl rrl NOUN bracis-28376 53 5 task task NOUN bracis-28376 53 6 , , PUNCT bracis-28376 53 7 there there PRON bracis-28376 53 8 are be VERB bracis-28376 53 9 a a DET bracis-28376 53 10 few few ADJ bracis-28376 53 11 recent recent ADJ bracis-28376 53 12 works work NOUN bracis-28376 53 13 that that PRON bracis-28376 53 14 follow follow VERB bracis-28376 53 15 this this DET bracis-28376 53 16 strand strand NOUN bracis-28376 53 17 . . PUNCT bracis-28376 54 1 mixup mixup NOUN bracis-28376 54 2     SPACE bracis-28376 55 1 [ [ X bracis-28376 55 2 23 23 NUM bracis-28376 55 3 ] ] PUNCT bracis-28376 55 4 , , PUNCT bracis-28376 55 5 a a DET bracis-28376 55 6 data data NOUN bracis-28376 55 7 augmentation augmentation NOUN bracis-28376 55 8 technique technique NOUN bracis-28376 55 9 , , PUNCT bracis-28376 55 10 is be AUX bracis-28376 55 11 exploited exploit VERB bracis-28376 55 12 in in ADP bracis-28376 55 13     SPACE bracis-28376 55 14 [ [ X bracis-28376 55 15 10 10 NUM bracis-28376 55 16 ] ] PUNCT bracis-28376 55 17 to to PART bracis-28376 55 18 augment augment VERB bracis-28376 55 19 the the DET bracis-28376 55 20 training training NOUN bracis-28376 55 21 set set VERB bracis-28376 55 22 and and CCONJ bracis-28376 55 23 hence hence ADV bracis-28376 55 24 improve improve VERB bracis-28376 55 25 model model NOUN bracis-28376 55 26 performance performance NOUN bracis-28376 55 27 . . PUNCT bracis-28376 56 1 mixup mixup NOUN bracis-28376 56 2 relies rely VERB bracis-28376 56 3 on on ADP bracis-28376 56 4 the the DET bracis-28376 56 5 weighted weighted ADJ bracis-28376 56 6 interpolation interpolation NOUN bracis-28376 56 7 of of ADP bracis-28376 56 8 two two NUM bracis-28376 56 9 inputs input NOUN bracis-28376 56 10 to to PART bracis-28376 56 11 generate generate VERB bracis-28376 56 12 a a DET bracis-28376 56 13 synthetic synthetic ADJ bracis-28376 56 14 vector vector NOUN bracis-28376 56 15 as as SCONJ bracis-28376 56 16 defined define VERB bracis-28376 56 17 by by ADP bracis-28376 56 18 the the DET bracis-28376 56 19 following follow VERB bracis-28376 56 20 equations equation NOUN bracis-28376 56 21 : : PUNCT bracis-28376 56 22 $ $ SYM bracis-28376 56 23 $ $ SYM bracis-28376 56 24 \begin{aligned \begin{aligne VERB bracis-28376 56 25 } } PUNCT bracis-28376 56 26 \mathbf \mathbf PROPN bracis-28376 56 27 { { PUNCT bracis-28376 56 28 \tilde{x \tilde{x NOUN bracis-28376 56 29 } } PUNCT bracis-28376 56 30 } } PUNCT bracis-28376 56 31 & & CCONJ bracis-28376 56 32 = = NOUN bracis-28376 56 33 \lambda \lambda PROPN bracis-28376 56 34 \textbf{x}_{i}+(1-\lambda \textbf{x}_{i}+(1-\lambda PROPN bracis-28376 56 35 ) ) PUNCT bracis-28376 56 36 \textbf{x}_{j \textbf{x}_{j PROPN bracis-28376 56 37 } } PUNCT bracis-28376 56 38 , , PUNCT bracis-28376 56 39 { { PUNCT bracis-28376 56 40 } } PUNCT bracis-28376 56 41 & & CCONJ bracis-28376 56 42 { { PUNCT bracis-28376 56 43 } } PUNCT bracis-28376 56 44 \textbf{x}_{i \textbf{x}_{i PROPN bracis-28376 56 45 } } PUNCT bracis-28376 56 46 , , PUNCT bracis-28376 56 47 \textbf{x}_{j}\in \textbf{x}_{j}\in ADV bracis-28376 56 48 \mathcal \mathcal PROPN bracis-28376 56 49 { { PUNCT bracis-28376 56 50 r}^e r}^e NOUN bracis-28376 56 51 \\ \\ NOUN bracis-28376 56 52 \mathbf \mathbf PROPN bracis-28376 56 53 { { PUNCT bracis-28376 56 54 \tilde{y \tilde{y NOUN bracis-28376 56 55 } } PUNCT bracis-28376 56 56 } } PUNCT bracis-28376 56 57 & & CCONJ bracis-28376 56 58 = = NOUN bracis-28376 56 59 \lambda \lambda PROPN bracis-28376 56 60 \textbf{y}_{i}+(1-\lambda \textbf{y}_{i}+(1-\lambda PROPN bracis-28376 56 61 ) ) PUNCT bracis-28376 56 62 \textbf{y}_{j \textbf{y}_{j PROPN bracis-28376 56 63 } } PUNCT bracis-28376 56 64 , , PUNCT bracis-28376 56 65 { { PUNCT bracis-28376 56 66 } } PUNCT bracis-28376 56 67 & & CCONJ bracis-28376 56 68 { { PUNCT bracis-28376 56 69 } } PUNCT bracis-28376 56 70 \textbf{y}_{i \textbf{y}_{i PROPN bracis-28376 56 71 } } PUNCT bracis-28376 56 72 , , PUNCT bracis-28376 56 73 \textbf{y}_{j \textbf{y}_{j PROPN bracis-28376 56 74 } } PUNCT bracis-28376 56 75 \in \in PROPN bracis-28376 56 76 \mathcal \mathcal PROPN bracis-28376 56 77 { { PUNCT bracis-28376 56 78 r}^{|l| r}^{|l| NOUN bracis-28376 56 79 } } PUNCT bracis-28376 56 80 , , PUNCT bracis-28376 56 81 \end{aligned}$$ \end{aligned}$$ ADP bracis-28376 56 82 where where SCONJ bracis-28376 56 83 \(\textbf{x}_{i}\ \(\textbf{x}_{i}\ NOUN bracis-28376 56 84 ) ) PUNCT bracis-28376 56 85 and and CCONJ bracis-28376 56 86 \(\textbf{x}_{j}\ \(\textbf{x}_{j}\ NOUN bracis-28376 56 87 ) ) PUNCT bracis-28376 56 88 are be AUX bracis-28376 56 89 sentence sentence NOUN bracis-28376 56 90 embedding embed VERB bracis-28376 56 91 vectors vector NOUN bracis-28376 56 92 , , PUNCT bracis-28376 56 93 e e X bracis-28376 56 94 is be AUX bracis-28376 56 95 the the DET bracis-28376 56 96 embedding embed VERB bracis-28376 56 97 dimension dimension NOUN bracis-28376 56 98 , , PUNCT bracis-28376 56 99 \(\textbf{y}_{i}\ \(\textbf{y}_{i}\ PROPN bracis-28376 56 100 ) ) PUNCT bracis-28376 56 101 and and CCONJ bracis-28376 56 102 \(\textbf{y}_{j}\ \(\textbf{y}_{j}\ PROPN bracis-28376 56 103 ) ) PUNCT bracis-28376 56 104 are be AUX bracis-28376 56 105 one one NUM bracis-28376 56 106 - - PUNCT bracis-28376 56 107 hot hot ADJ bracis-28376 56 108 label label NOUN bracis-28376 56 109 encodings encoding NOUN bracis-28376 56 110 , , PUNCT bracis-28376 56 111 and and CCONJ bracis-28376 56 112 |l| |l| NOUN bracis-28376 56 113 is be AUX bracis-28376 56 114 the the DET bracis-28376 56 115 number number NOUN bracis-28376 56 116 of of ADP bracis-28376 56 117 labels label NOUN bracis-28376 56 118 or or CCONJ bracis-28376 56 119 classes class NOUN bracis-28376 56 120 . . PUNCT bracis-28376 57 1 \((\textbf{x}_{i \((\textbf{x}_{i NOUN bracis-28376 57 2 } } PUNCT bracis-28376 57 3 , , PUNCT bracis-28376 57 4 \textbf{y}_{i})\ \textbf{y}_{i})\ PROPN bracis-28376 57 5 ) ) PUNCT bracis-28376 57 6 and and CCONJ bracis-28376 57 7 \((\textbf{x}_{j \((\textbf{x}_{j NOUN bracis-28376 57 8 } } PUNCT bracis-28376 57 9 , , PUNCT bracis-28376 57 10 \textbf{y}_{j})\ \textbf{y}_{j})\ X bracis-28376 57 11 ) ) PUNCT bracis-28376 57 12 are be AUX bracis-28376 57 13 two two NUM bracis-28376 57 14 examples example NOUN bracis-28376 57 15 drawn draw VERB bracis-28376 57 16 at at ADP bracis-28376 57 17 random random ADJ bracis-28376 57 18 from from ADP bracis-28376 57 19 the the DET bracis-28376 57 20 training training NOUN bracis-28376 57 21 data datum NOUN bracis-28376 57 22 and and CCONJ bracis-28376 57 23 \(\lambda \(\lambda NOUN bracis-28376 57 24 \ \ NOUN bracis-28376 57 25 ) ) PUNCT bracis-28376 57 26 is be AUX bracis-28376 57 27 a a DET bracis-28376 57 28 value value NOUN bracis-28376 57 29 in in ADP bracis-28376 57 30 the the DET bracis-28376 57 31 interval interval NOUN bracis-28376 58 1 [ [ X bracis-28376 58 2 0 0 NUM bracis-28376 58 3 , , PUNCT bracis-28376 58 4     SPACE bracis-28376 58 5 1 1 NUM bracis-28376 58 6 ] ] PUNCT bracis-28376 58 7 drawn draw VERB bracis-28376 58 8 at at ADP bracis-28376 58 9 random random ADJ bracis-28376 58 10 from from ADP bracis-28376 58 11 a a DET bracis-28376 58 12 beta\((\alpha beta\((\alpha PROPN bracis-28376 58 13 , , PUNCT bracis-28376 58 14 \alpha \alpha ADJ bracis-28376 58 15 ) ) PUNCT bracis-28376 58 16 \ \ ADJ bracis-28376 58 17 ) ) PUNCT bracis-28376 58 18 distribution distribution NOUN bracis-28376 58 19 , , PUNCT bracis-28376 58 20 where where SCONJ bracis-28376 58 21 \(\alpha \(\alpha ADV bracis-28376 58 22 > > X bracis-28376 58 23 0\ 0\ PROPN bracis-28376 58 24 ) ) PUNCT bracis-28376 58 25 . . PUNCT bracis-28376 59 1 in in ADP bracis-28376 59 2 their their PRON bracis-28376 59 3 setup setup NOUN bracis-28376 59 4     SPACE bracis-28376 59 5 [ [ X bracis-28376 59 6 10 10 NUM bracis-28376 59 7 ] ] PUNCT bracis-28376 59 8 , , PUNCT bracis-28376 59 9 the the DET bracis-28376 59 10 authors author NOUN bracis-28376 59 11 exploit exploit VERB bracis-28376 59 12 sentence sentence NOUN bracis-28376 59 13 bert bert NOUN bracis-28376 59 14     SPACE bracis-28376 60 1 [ [ X bracis-28376 60 2 17 17 NUM bracis-28376 60 3 ] ] PUNCT bracis-28376 60 4 as as ADP bracis-28376 60 5 an an DET bracis-28376 60 6 encoder encoder NOUN bracis-28376 60 7 without without ADP bracis-28376 60 8 fine fine ADV bracis-28376 60 9 - - PUNCT bracis-28376 60 10 tuning tuning NOUN bracis-28376 60 11 . . PUNCT bracis-28376 61 1 the the DET bracis-28376 61 2 exploited exploit VERB bracis-28376 61 3 dataset dataset NOUN bracis-28376 61 4 comprises comprise VERB bracis-28376 61 5 60 60 NUM bracis-28376 61 6 judgments judgment NOUN bracis-28376 61 7 , , PUNCT bracis-28376 61 8 10,024 10,024 NUM bracis-28376 61 9 sentences sentence NOUN bracis-28376 61 10 , , PUNCT bracis-28376 61 11 and and CCONJ bracis-28376 61 12 seven seven NUM bracis-28376 61 13 rhetorical rhetorical ADJ bracis-28376 61 14 roles role NOUN bracis-28376 61 15 . . PUNCT bracis-28376 62 1 as as ADP bracis-28376 62 2 a a DET bracis-28376 62 3 result result NOUN bracis-28376 62 4 , , PUNCT bracis-28376 62 5 the the DET bracis-28376 62 6 mixup mixup NOUN bracis-28376 62 7 - - PUNCT bracis-28376 62 8 based base VERB bracis-28376 62 9 models model NOUN bracis-28376 62 10 overcome overcome VERB bracis-28376 62 11 the the DET bracis-28376 62 12 respective respective ADJ bracis-28376 62 13 baselines baseline NOUN bracis-28376 62 14 in in ADP bracis-28376 62 15 terms term NOUN bracis-28376 62 16 of of ADP bracis-28376 62 17 f1 f1 NOUN bracis-28376 62 18 score score NOUN bracis-28376 62 19 . . PUNCT bracis-28376 63 1 for for ADP bracis-28376 63 2 example example NOUN bracis-28376 63 3 , , PUNCT bracis-28376 63 4 a a DET bracis-28376 63 5 linear linear ADJ bracis-28376 63 6 classifier classifier NOUN bracis-28376 63 7 achieves achieve VERB bracis-28376 63 8 a a DET bracis-28376 63 9 gain gain NOUN bracis-28376 63 10 of of ADP bracis-28376 63 11 4.3 4.3 NUM bracis-28376 63 12 % % NOUN bracis-28376 63 13 in in ADP bracis-28376 63 14 the the DET bracis-28376 63 15 worst bad ADJ bracis-28376 63 16 case case NOUN bracis-28376 63 17 and and CCONJ bracis-28376 63 18 6.8 6.8 NUM bracis-28376 63 19 % % NOUN bracis-28376 63 20 in in ADP bracis-28376 63 21 the the DET bracis-28376 63 22 best good ADJ bracis-28376 63 23 case case NOUN bracis-28376 63 24 . . PUNCT bracis-28376 64 1 another another DET bracis-28376 64 2 work work NOUN bracis-28376 64 3     SPACE bracis-28376 65 1 [ [ X bracis-28376 65 2 12 12 NUM bracis-28376 65 3 ] ] PUNCT bracis-28376 65 4 leverages leverage VERB bracis-28376 65 5 positional positional ADJ bracis-28376 65 6 information information NOUN bracis-28376 65 7 in in ADP bracis-28376 65 8 addition addition NOUN bracis-28376 65 9 to to ADP bracis-28376 65 10 encoded encode VERB bracis-28376 65 11 sentences sentence NOUN bracis-28376 65 12 . . PUNCT bracis-28376 66 1 each each DET bracis-28376 66 2 sentence sentence NOUN bracis-28376 66 3 is be AUX bracis-28376 66 4 represented represent VERB bracis-28376 66 5 by by ADP bracis-28376 66 6 a a DET bracis-28376 66 7 sentence sentence NOUN bracis-28376 66 8 embedding embed VERB bracis-28376 66 9 and and CCONJ bracis-28376 66 10 a a DET bracis-28376 66 11 positional positional ADJ bracis-28376 66 12 embedding embedding NOUN bracis-28376 66 13 , , PUNCT bracis-28376 66 14 the the DET bracis-28376 66 15 latter latter ADJ bracis-28376 66 16 indicating indicate VERB bracis-28376 66 17 the the DET bracis-28376 66 18 position position NOUN bracis-28376 66 19 a a DET bracis-28376 66 20 sentence sentence NOUN bracis-28376 66 21 occupies occupie NOUN bracis-28376 66 22 in in ADP bracis-28376 66 23 its its PRON bracis-28376 66 24 source source NOUN bracis-28376 66 25 document document NOUN bracis-28376 66 26 . . PUNCT bracis-28376 67 1 positional positional ADJ bracis-28376 67 2 embeddings embedding NOUN bracis-28376 67 3 are be AUX bracis-28376 67 4 computed compute VERB bracis-28376 67 5 by by ADP bracis-28376 67 6 the the DET bracis-28376 67 7 same same ADJ bracis-28376 67 8 sinusoidal sinusoidal NOUN bracis-28376 67 9 positional positional ADJ bracis-28376 67 10 encoding encoding NOUN bracis-28376 67 11 method method NOUN bracis-28376 67 12 exploited exploit VERB bracis-28376 67 13 in in ADP bracis-28376 67 14 the the DET bracis-28376 67 15 original original ADJ bracis-28376 67 16 transformer transformer NOUN bracis-28376 67 17 architecture architecture NOUN bracis-28376 67 18     SPACE bracis-28376 68 1 [ [ X bracis-28376 68 2 20 20 NUM bracis-28376 68 3 ] ] NOUN bracis-28376 68 4 : : PUNCT bracis-28376 68 5 $ $ SYM bracis-28376 68 6 $ $ SYM bracis-28376 68 7 \begin{aligned \begin{aligne VERB bracis-28376 68 8 } } PUNCT bracis-28376 68 9 { { PUNCT bracis-28376 68 10 \begin{matrix \begin{matrix NOUN bracis-28376 68 11 } } PUNCT bracis-28376 68 12 pe_{(pos,2k pe_{(pos,2k NOUN bracis-28376 68 13 ) ) PUNCT bracis-28376 68 14 } } PUNCT bracis-28376 68 15 & & CCONJ bracis-28376 68 16 { { PUNCT bracis-28376 68 17 } } PUNCT bracis-28376 68 18 = = SYM bracis-28376 68 19 \sin \sin NOUN bracis-28376 68 20 ( ( PUNCT bracis-28376 68 21 pos pos NOUN bracis-28376 68 22 / / SYM bracis-28376 68 23 m^{2k m^{2k PROPN bracis-28376 68 24 / / SYM bracis-28376 68 25 e e NOUN bracis-28376 68 26 } } PUNCT bracis-28376 68 27 ) ) PUNCT bracis-28376 68 28 \\ \\ NOUN bracis-28376 68 29 pe_{(pos,2k+1 pe_{(pos,2k+1 NOUN bracis-28376 68 30 ) ) PUNCT bracis-28376 68 31 } } PUNCT bracis-28376 68 32 & & CCONJ bracis-28376 68 33 { { PUNCT bracis-28376 68 34 } } PUNCT bracis-28376 68 35 = = SYM bracis-28376 68 36 \cos \cos PROPN bracis-28376 68 37 ( ( PUNCT bracis-28376 68 38 pos pos NOUN bracis-28376 68 39 / / SYM bracis-28376 68 40 m^{2k m^{2k PROPN bracis-28376 68 41 / / SYM bracis-28376 68 42 e e NOUN bracis-28376 68 43 } } PUNCT bracis-28376 68 44 ) ) PUNCT bracis-28376 68 45 \end{matrix \end{matrix NOUN bracis-28376 68 46 } } PUNCT bracis-28376 68 47 } } PUNCT bracis-28376 68 48 \end{aligned}$$ \end{aligned}$$ ADP bracis-28376 68 49 where where SCONJ bracis-28376 68 50 pos pos NOUN bracis-28376 68 51 is be AUX bracis-28376 68 52 an an DET bracis-28376 68 53 integer integer NOUN bracis-28376 68 54 indicating indicate VERB bracis-28376 68 55 the the DET bracis-28376 68 56 position position NOUN bracis-28376 68 57 of of ADP bracis-28376 68 58 a a DET bracis-28376 68 59 sentence sentence NOUN bracis-28376 68 60 , , PUNCT bracis-28376 68 61 k k PROPN bracis-28376 68 62 is be AUX bracis-28376 68 63 a a DET bracis-28376 68 64 dimension dimension NOUN bracis-28376 68 65 of of ADP bracis-28376 68 66 a a DET bracis-28376 68 67 \(\textbf{pe}\ \(\textbf{pe}\ NOUN bracis-28376 68 68 ) ) PUNCT bracis-28376 68 69 vector vector NOUN bracis-28376 68 70 where where SCONJ bracis-28376 68 71 \(0 \(0 PROPN bracis-28376 68 72 \le \le PROPN bracis-28376 69 1 k k X bracis-28376 69 2 < < X bracis-28376 69 3 ( ( PUNCT bracis-28376 69 4 e-1)/2\ e-1)/2\ PROPN bracis-28376 69 5 ) ) PUNCT bracis-28376 69 6 , , PUNCT bracis-28376 69 7 e e X bracis-28376 69 8 is be AUX bracis-28376 69 9 the the DET bracis-28376 69 10 embedding embed VERB bracis-28376 69 11 dimension dimension NOUN bracis-28376 69 12 , , PUNCT bracis-28376 69 13 and and CCONJ bracis-28376 69 14 m m PROPN bracis-28376 69 15 is be AUX bracis-28376 69 16 the the DET bracis-28376 69 17 max max PROPN bracis-28376 69 18 number number NOUN bracis-28376 69 19 of of ADP bracis-28376 69 20 sentences sentence NOUN bracis-28376 69 21 in in ADP bracis-28376 69 22 a a DET bracis-28376 69 23 document document NOUN bracis-28376 69 24 . . PUNCT bracis-28376 70 1 the the DET bracis-28376 70 2 sentence sentence NOUN bracis-28376 70 3 and and CCONJ bracis-28376 70 4 positional positional ADJ bracis-28376 70 5 embeddings embedding NOUN bracis-28376 70 6 must must AUX bracis-28376 70 7 be be AUX bracis-28376 70 8 combined combine VERB bracis-28376 70 9 before before ADP bracis-28376 70 10 being be AUX bracis-28376 70 11 fed feed VERB bracis-28376 70 12 to to ADP bracis-28376 70 13 the the DET bracis-28376 70 14 models model NOUN bracis-28376 70 15 ’ ' PUNCT bracis-28376 70 16 classifier classifier NOUN bracis-28376 70 17 and and CCONJ bracis-28376 70 18 to to PART bracis-28376 70 19 do do VERB bracis-28376 70 20 this this PRON bracis-28376 70 21 the the DET bracis-28376 70 22 authors author NOUN bracis-28376 70 23 employ employ VERB bracis-28376 70 24 vector vector NOUN bracis-28376 70 25 sum sum NOUN bracis-28376 70 26 and and CCONJ bracis-28376 70 27 vector vector NOUN bracis-28376 70 28 concatenation concatenation NOUN bracis-28376 70 29 . . PUNCT bracis-28376 71 1 in in ADP bracis-28376 71 2 their their PRON bracis-28376 71 3 setup setup NOUN bracis-28376 71 4     SPACE bracis-28376 71 5 [ [ X bracis-28376 71 6 12 12 NUM bracis-28376 71 7 ] ] PUNCT bracis-28376 71 8 , , PUNCT bracis-28376 71 9 they they PRON bracis-28376 71 10 experiment experiment VERB bracis-28376 71 11 with with ADP bracis-28376 71 12 four four NUM bracis-28376 71 13 pre pre ADJ bracis-28376 71 14 - - ADJ bracis-28376 71 15 trained train VERB bracis-28376 71 16 transformers transformer NOUN bracis-28376 71 17 as as ADP bracis-28376 71 18 sentence sentence NOUN bracis-28376 71 19 encoders encoder NOUN bracis-28376 71 20 . . PUNCT bracis-28376 72 1 the the DET bracis-28376 72 2 exploited exploit VERB bracis-28376 72 3 dataset dataset NOUN bracis-28376 72 4 comprises comprise VERB bracis-28376 72 5 369 369 NUM bracis-28376 72 6 judgments judgment NOUN bracis-28376 72 7 , , PUNCT bracis-28376 72 8 54,244 54,244 NUM bracis-28376 72 9 sentences sentence NOUN bracis-28376 72 10 , , PUNCT bracis-28376 72 11 and and CCONJ bracis-28376 72 12 two two NUM bracis-28376 72 13 rhetorical rhetorical ADJ bracis-28376 72 14 roles role NOUN bracis-28376 72 15 . . PUNCT bracis-28376 73 1 the the DET bracis-28376 73 2 authors author NOUN bracis-28376 73 3 report report VERB bracis-28376 73 4 gains gain NOUN bracis-28376 73 5 from from ADP bracis-28376 73 6 3.3 3.3 NUM bracis-28376 73 7 % % NOUN bracis-28376 73 8 to to ADP bracis-28376 73 9 12.6 12.6 NUM bracis-28376 73 10 % % NOUN bracis-28376 73 11 achieved achieve VERB bracis-28376 73 12 by by ADP bracis-28376 73 13 fine fine ADV bracis-28376 73 14 - - PUNCT bracis-28376 73 15 tuned tune VERB bracis-28376 73 16 models model NOUN bracis-28376 73 17 . . PUNCT bracis-28376 74 1 the the DET bracis-28376 74 2 approaches approach NOUN bracis-28376 74 3 presented present VERB bracis-28376 74 4 in in ADP bracis-28376 74 5 [ [ X bracis-28376 74 6 10 10 NUM bracis-28376 74 7 , , PUNCT bracis-28376 74 8 12 12 NUM bracis-28376 74 9 ] ] PUNCT bracis-28376 74 10 rely rely VERB bracis-28376 74 11 on on ADP bracis-28376 74 12 the the DET bracis-28376 74 13 single single ADJ bracis-28376 74 14 sentence sentence NOUN bracis-28376 74 15 classification classification NOUN bracis-28376 74 16 paradigm paradigm NOUN bracis-28376 74 17 : : PUNCT bracis-28376 74 18 a a DET bracis-28376 74 19 sentence sentence NOUN bracis-28376 74 20 is be AUX bracis-28376 74 21 encoded encode VERB bracis-28376 74 22 in in ADP bracis-28376 74 23 isolation isolation NOUN bracis-28376 74 24 and and CCONJ bracis-28376 74 25 so so ADV bracis-28376 74 26 its its PRON bracis-28376 74 27 context context NOUN bracis-28376 74 28 ( ( PUNCT bracis-28376 74 29 neighbor neighbor NOUN bracis-28376 74 30 sentences sentence NOUN bracis-28376 74 31 ) ) PUNCT bracis-28376 74 32 is be AUX bracis-28376 74 33 ignored ignore VERB bracis-28376 74 34 . . PUNCT bracis-28376 75 1 although although SCONJ bracis-28376 75 2 , , PUNCT bracis-28376 75 3 transformers transformer NOUN bracis-28376 75 4 are be AUX bracis-28376 75 5 notable notable ADJ bracis-28376 75 6 for for ADP bracis-28376 75 7 exploiting exploit VERB bracis-28376 75 8 context context NOUN bracis-28376 75 9 to to PART bracis-28376 75 10 encode encode VERB bracis-28376 75 11 sentences sentence NOUN bracis-28376 75 12 and and CCONJ bracis-28376 75 13 thus thus ADV bracis-28376 75 14 the the DET bracis-28376 75 15 work work NOUN bracis-28376 75 16 of of ADP bracis-28376 75 17 [ [ X bracis-28376 75 18 11 11 NUM bracis-28376 75 19 ] ] PUNCT bracis-28376 75 20 leverages leverage VERB bracis-28376 75 21 this this DET bracis-28376 75 22 ability ability NOUN bracis-28376 75 23 to to PART bracis-28376 75 24 produce produce VERB bracis-28376 75 25 richer rich ADJ bracis-28376 75 26 sentence sentence NOUN bracis-28376 75 27 embeddings embedding NOUN bracis-28376 75 28 as as ADP bracis-28376 75 29 a a DET bracis-28376 75 30 way way NOUN bracis-28376 75 31 to to PART bracis-28376 75 32 achieve achieve VERB bracis-28376 75 33 better well ADJ bracis-28376 75 34 results result NOUN bracis-28376 75 35 . . PUNCT bracis-28376 76 1 the the DET bracis-28376 76 2 encoder encoder NOUN bracis-28376 76 3 is be AUX bracis-28376 76 4 fed feed VERB bracis-28376 76 5 with with ADP bracis-28376 76 6 chunks chunk NOUN bracis-28376 76 7 of of ADP bracis-28376 76 8 sentences sentence NOUN bracis-28376 76 9 and and CCONJ bracis-28376 76 10 a a DET bracis-28376 76 11 special special ADJ bracis-28376 76 12 character character NOUN bracis-28376 76 13 is be AUX bracis-28376 76 14 employed employ VERB bracis-28376 76 15 to to PART bracis-28376 76 16 signal signal VERB bracis-28376 76 17 sentences sentence NOUN bracis-28376 76 18 ’ ' PUNCT bracis-28376 76 19 limits limit NOUN bracis-28376 76 20 . . PUNCT bracis-28376 77 1 the the DET bracis-28376 77 2 authors author NOUN bracis-28376 77 3 propose propose VERB bracis-28376 77 4 a a DET bracis-28376 77 5 chunk chunk ADJ bracis-28376 77 6 layout layout NOUN bracis-28376 77 7 and and CCONJ bracis-28376 77 8 an an DET bracis-28376 77 9 assembly assembly NOUN bracis-28376 77 10 procedure procedure NOUN bracis-28376 77 11 , , PUNCT bracis-28376 77 12 named name VERB bracis-28376 77 13 dfcsc dfcsc NOUN bracis-28376 77 14 ( ( PUNCT bracis-28376 77 15 dynamic dynamic ADJ bracis-28376 77 16 - - PUNCT bracis-28376 77 17 filled fill VERB bracis-28376 77 18 contextualized contextualized ADJ bracis-28376 77 19 sentence sentence NOUN bracis-28376 77 20 chunk chunk NOUN bracis-28376 77 21 ) ) PUNCT bracis-28376 77 22 , , PUNCT bracis-28376 77 23 that that PRON bracis-28376 77 24 avoids avoid VERB bracis-28376 77 25 the the DET bracis-28376 77 26 use use NOUN bracis-28376 77 27 of of ADP bracis-28376 77 28 padding padding NOUN bracis-28376 77 29 tokens token NOUN bracis-28376 77 30 , , PUNCT bracis-28376 77 31 the the DET bracis-28376 77 32 truncation truncation NOUN bracis-28376 77 33 of of ADP bracis-28376 77 34 core core NOUN bracis-28376 77 35 sentences sentence NOUN bracis-28376 77 36 , , PUNCT bracis-28376 77 37 and and CCONJ bracis-28376 77 38 that that SCONJ bracis-28376 77 39 defines define VERB bracis-28376 77 40 at at ADP bracis-28376 77 41 run run NOUN bracis-28376 77 42 time time NOUN bracis-28376 77 43 the the DET bracis-28376 77 44 number number NOUN bracis-28376 77 45 of of ADP bracis-28376 77 46 core core NOUN bracis-28376 77 47 sentences sentence NOUN bracis-28376 77 48 and and CCONJ bracis-28376 77 49 edge edge NOUN bracis-28376 77 50 tokens token NOUN bracis-28376 77 51 in in ADP bracis-28376 77 52 each each DET bracis-28376 77 53 chunk chunk NOUN bracis-28376 77 54 . . PUNCT bracis-28376 78 1 core core NOUN bracis-28376 78 2 sentences sentence NOUN bracis-28376 78 3 are be AUX bracis-28376 78 4 the the DET bracis-28376 78 5 sentences sentence NOUN bracis-28376 78 6 taken take VERB bracis-28376 78 7 into into ADP bracis-28376 78 8 account account NOUN bracis-28376 78 9 by by ADP bracis-28376 78 10 the the DET bracis-28376 78 11 classifier classifier NOUN bracis-28376 78 12 and and CCONJ bracis-28376 78 13 edge edge NOUN bracis-28376 78 14 tokens token NOUN bracis-28376 78 15 are be AUX bracis-28376 78 16 tokens token NOUN bracis-28376 78 17 inserted insert VERB bracis-28376 78 18 at at ADP bracis-28376 78 19 the the DET bracis-28376 78 20 edges edge NOUN bracis-28376 78 21 of of ADP bracis-28376 78 22 the the DET bracis-28376 78 23 chunks chunk NOUN bracis-28376 78 24 as as ADP bracis-28376 78 25 a a DET bracis-28376 78 26 way way NOUN bracis-28376 78 27 to to PART bracis-28376 78 28 provide provide VERB bracis-28376 78 29 context context NOUN bracis-28376 78 30 for for ADP bracis-28376 78 31 the the DET bracis-28376 78 32 first first ADJ bracis-28376 78 33 and and CCONJ bracis-28376 78 34 last last ADJ bracis-28376 78 35 core core ADJ bracis-28376 78 36 sentences sentence NOUN bracis-28376 78 37 ( ( PUNCT bracis-28376 78 38 see see VERB bracis-28376 78 39 fig fig NOUN bracis-28376 78 40 . . PUNCT bracis-28376 78 41     SPACE bracis-28376 78 42 2 2 NUM bracis-28376 78 43 ) ) PUNCT bracis-28376 78 44 . . PUNCT bracis-28376 79 1 the the DET bracis-28376 79 2 procedure procedure NOUN bracis-28376 79 3 requires require VERB bracis-28376 79 4 two two NUM bracis-28376 79 5 hyperparameters hyperparameter NOUN bracis-28376 79 6 : : PUNCT bracis-28376 79 7 \(c_{len}\ \(c_{len}\ NOUN bracis-28376 79 8 ) ) PUNCT bracis-28376 79 9 ( ( PUNCT bracis-28376 79 10 the the DET bracis-28376 79 11 number number NOUN bracis-28376 79 12 of of ADP bracis-28376 79 13 tokens token NOUN bracis-28376 79 14 in in ADP bracis-28376 79 15 a a DET bracis-28376 79 16 chunk chunk NOUN bracis-28376 79 17 ) ) PUNCT bracis-28376 79 18 and and CCONJ bracis-28376 79 19 \(m_{edges}\ \(m_{edges}\ NUM bracis-28376 79 20 ) ) PUNCT bracis-28376 79 21 ( ( PUNCT bracis-28376 79 22 the the DET bracis-28376 79 23 minimum minimum NOUN bracis-28376 79 24 desired desire VERB bracis-28376 79 25 number number NOUN bracis-28376 79 26 of of ADP bracis-28376 79 27 tokens token NOUN bracis-28376 79 28 in in ADP bracis-28376 79 29 the the DET bracis-28376 79 30 edges edge NOUN bracis-28376 79 31 of of ADP bracis-28376 79 32 a a DET bracis-28376 79 33 chunk chunk NOUN bracis-28376 79 34 ) ) PUNCT bracis-28376 79 35 . . PUNCT bracis-28376 80 1 in in ADP bracis-28376 80 2 their their PRON bracis-28376 80 3 setup setup NOUN bracis-28376 80 4     SPACE bracis-28376 80 5 [ [ X bracis-28376 80 6 11 11 NUM bracis-28376 80 7 ] ] PUNCT bracis-28376 80 8 , , PUNCT bracis-28376 80 9 the the DET bracis-28376 80 10 authors author NOUN bracis-28376 80 11 exploit exploit VERB bracis-28376 80 12 a a DET bracis-28376 80 13 dataset dataset NOUN bracis-28376 80 14 with with ADP bracis-28376 80 15 277 277 NUM bracis-28376 80 16 judgments judgment NOUN bracis-28376 80 17 , , PUNCT bracis-28376 80 18 31,865 31,865 NUM bracis-28376 80 19 sentences sentence NOUN bracis-28376 80 20 , , PUNCT bracis-28376 80 21 and and CCONJ bracis-28376 80 22 13 13 NUM bracis-28376 80 23 rhetorical rhetorical ADJ bracis-28376 80 24 roles role NOUN bracis-28376 80 25 . . PUNCT bracis-28376 81 1 they they PRON bracis-28376 81 2 experiment experiment VERB bracis-28376 81 3 with with ADP bracis-28376 81 4 four four NUM bracis-28376 81 5 pre pre ADJ bracis-28376 81 6 - - ADJ bracis-28376 81 7 trained train VERB bracis-28376 81 8 transformers transformer NOUN bracis-28376 81 9 and and CCONJ bracis-28376 81 10 two two NUM bracis-28376 81 11 additional additional ADJ bracis-28376 81 12 approaches approach NOUN bracis-28376 81 13 based base VERB bracis-28376 81 14 on on ADP bracis-28376 81 15 sentence sentence NOUN bracis-28376 81 16 chunks chunk NOUN bracis-28376 81 17 besides besides SCONJ bracis-28376 81 18 the the DET bracis-28376 81 19 dfcsc dfcsc NOUN bracis-28376 81 20 one one NUM bracis-28376 81 21 . . PUNCT bracis-28376 82 1 dfcsc dfcsc NOUN bracis-28376 82 2 - - PUNCT bracis-28376 82 3 based base VERB bracis-28376 82 4 models model NOUN bracis-28376 82 5 perform perform VERB bracis-28376 82 6 the the DET bracis-28376 82 7 best good ADJ bracis-28376 82 8 and and CCONJ bracis-28376 82 9 the the DET bracis-28376 82 10 reported report VERB bracis-28376 82 11 gains gain NOUN bracis-28376 82 12 are be AUX bracis-28376 82 13 respectively respectively ADV bracis-28376 82 14 of of ADP bracis-28376 82 15 3.76 3.76 NUM bracis-28376 82 16 % % NOUN bracis-28376 82 17 and and CCONJ bracis-28376 82 18 8.71 8.71 NUM bracis-28376 82 19 % % NOUN bracis-28376 82 20 in in ADP bracis-28376 82 21 the the DET bracis-28376 82 22 best good ADJ bracis-28376 82 23 and and CCONJ bracis-28376 82 24 worst bad ADJ bracis-28376 82 25 cases case NOUN bracis-28376 82 26 . . PUNCT bracis-28376 83 1 fig fig NOUN bracis-28376 83 2 . . PUNCT bracis-28376 84 1 2 2 X bracis-28376 84 2 . . X bracis-28376 84 3 illustration illustration NOUN bracis-28376 84 4 of of ADP bracis-28376 84 5 a a DET bracis-28376 84 6 dfcsc dfcsc NOUN bracis-28376 84 7 . . PUNCT bracis-28376 85 1 remark remark VERB bracis-28376 85 2 that that SCONJ bracis-28376 85 3 the the DET bracis-28376 85 4 classifier classifier NOUN bracis-28376 85 5 takes take VERB bracis-28376 85 6 into into ADP bracis-28376 85 7 account account NOUN bracis-28376 85 8 only only ADV bracis-28376 85 9 the the DET bracis-28376 85 10 tokens token NOUN bracis-28376 85 11 chosen choose VERB bracis-28376 85 12 as as ADP bracis-28376 85 13 core core NOUN bracis-28376 85 14 sentence sentence NOUN bracis-28376 85 15 representations representation NOUN bracis-28376 85 16 . . PUNCT bracis-28376 86 1 \(s_i\ \(s_i\ NOUN bracis-28376 86 2 ) ) PUNCT bracis-28376 86 3 means mean VERB bracis-28376 86 4 the the DET bracis-28376 86 5 i i PROPN bracis-28376 86 6 - - PUNCT bracis-28376 86 7 th th VERB bracis-28376 86 8 sentence sentence NOUN bracis-28376 86 9 , , PUNCT bracis-28376 86 10 \(y_i\ \(y_i\ PROPN bracis-28376 86 11 ) ) PUNCT bracis-28376 86 12 is be AUX bracis-28376 86 13 the the DET bracis-28376 86 14 label label NOUN bracis-28376 86 15 assigned assign VERB bracis-28376 86 16 to to ADP bracis-28376 86 17 \(s_i\ \(s_i\ NOUN bracis-28376 86 18 ) ) PUNCT bracis-28376 86 19 , , PUNCT bracis-28376 86 20 h h NOUN bracis-28376 86 21 represents represent VERB bracis-28376 86 22 a a DET bracis-28376 86 23 token token ADJ bracis-28376 86 24 embedding embed VERB bracis-28376 86 25 , , PUNCT bracis-28376 86 26 t t PROPN bracis-28376 86 27 represents represent VERB bracis-28376 86 28 a a DET bracis-28376 86 29 token token ADJ bracis-28376 86 30 , , PUNCT bracis-28376 86 31 \(|t_i|\ \(|t_i|\ NUM bracis-28376 86 32 ) ) PUNCT bracis-28376 86 33 is be AUX bracis-28376 86 34 the the DET bracis-28376 86 35 number number NOUN bracis-28376 86 36 of of ADP bracis-28376 86 37 tokens token NOUN bracis-28376 86 38 in in ADP bracis-28376 86 39 \(s_i\ \(s_i\ NOUN bracis-28376 86 40 ) ) PUNCT bracis-28376 86 41 , , PUNCT bracis-28376 87 1 \ \ PROPN bracis-28376 87 2 ( ( PUNCT bracis-28376 87 3 \texttt \texttt PROPN bracis-28376 87 4 { { PUNCT bracis-28376 87 5 < < X bracis-28376 87 6 s>}\ s>}\ ADJ bracis-28376 87 7 ) ) PUNCT bracis-28376 87 8 and and CCONJ bracis-28376 87 9 \(\texttt \(\texttt NUM bracis-28376 87 10 { { PUNCT bracis-28376 87 11 < < X bracis-28376 87 12 /s>}\ /s>}\ X bracis-28376 87 13 ) ) PUNCT bracis-28376 87 14 are be AUX bracis-28376 87 15 special special ADJ bracis-28376 87 16 tokens token NOUN bracis-28376 87 17 . . PUNCT bracis-28376 88 1 full full ADJ bracis-28376 88 2 size size NOUN bracis-28376 88 3 image image NOUN bracis-28376 88 4 3 3 NUM bracis-28376 88 5 experimental experimental ADJ bracis-28376 88 6 setup setup NOUN bracis-28376 88 7 3.1 3.1 NUM bracis-28376 88 8 datasets dataset NOUN bracis-28376 88 9 we we PRON bracis-28376 88 10 employ employ VERB bracis-28376 88 11 two two NUM bracis-28376 88 12 datasets dataset NOUN bracis-28376 88 13 to to PART bracis-28376 88 14 evaluate evaluate VERB bracis-28376 88 15 the the DET bracis-28376 88 16 models model NOUN bracis-28376 88 17 which which PRON bracis-28376 88 18 we we PRON bracis-28376 88 19 refer refer VERB bracis-28376 88 20 to to ADP bracis-28376 88 21 as as ADP bracis-28376 88 22 7 7 NUM bracis-28376 88 23 - - NOUN bracis-28376 88 24 roles role NOUN bracis-28376 88 25 and and CCONJ bracis-28376 88 26 4 4 NUM bracis-28376 88 27 - - NOUN bracis-28376 88 28 roles role NOUN bracis-28376 88 29 . . PUNCT bracis-28376 89 1 both both PRON bracis-28376 89 2 are be AUX bracis-28376 89 3 derivations derivation NOUN bracis-28376 89 4 of of ADP bracis-28376 89 5 the the DET bracis-28376 89 6 original original ADJ bracis-28376 89 7 dataset dataset NOUN bracis-28376 89 8 developed develop VERB bracis-28376 89 9 in in ADP bracis-28376 89 10 [ [ PUNCT bracis-28376 89 11 15]footnote 15]footnote NUM bracis-28376 89 12 1 1 NUM bracis-28376 89 13 that that PRON bracis-28376 89 14 comprises comprise VERB bracis-28376 89 15 100 100 NUM bracis-28376 89 16 judgment judgment NOUN bracis-28376 89 17 documents document NOUN bracis-28376 89 18 written write VERB bracis-28376 89 19 in in ADP bracis-28376 89 20 english english PROPN bracis-28376 89 21 from from ADP bracis-28376 89 22 the the DET bracis-28376 89 23 indian indian ADJ bracis-28376 89 24 legal legal ADJ bracis-28376 89 25 system system NOUN bracis-28376 89 26 split split VERB bracis-28376 89 27 into into ADP bracis-28376 89 28 training training NOUN bracis-28376 89 29 set set NOUN bracis-28376 89 30 ( ( PUNCT bracis-28376 89 31 80 80 NUM bracis-28376 89 32 documents document NOUN bracis-28376 89 33 and and CCONJ bracis-28376 89 34 16,845 16,845 NUM bracis-28376 89 35 sentences sentence NOUN bracis-28376 89 36 ) ) PUNCT bracis-28376 89 37 , , PUNCT bracis-28376 89 38 validation validation NOUN bracis-28376 89 39 set set NOUN bracis-28376 89 40 ( ( PUNCT bracis-28376 89 41 10 10 NUM bracis-28376 89 42 documents document NOUN bracis-28376 89 43 and and CCONJ bracis-28376 89 44 2,142 2,142 NUM bracis-28376 89 45 sentences sentence NOUN bracis-28376 89 46 ) ) PUNCT bracis-28376 89 47 , , PUNCT bracis-28376 89 48 and and CCONJ bracis-28376 89 49 test test NOUN bracis-28376 89 50 set set VERB bracis-28376 89 51 ( ( PUNCT bracis-28376 89 52 10 10 NUM bracis-28376 89 53 documents document NOUN bracis-28376 89 54 and and CCONJ bracis-28376 89 55 2,197 2,197 NUM bracis-28376 89 56 sentences sentence NOUN bracis-28376 89 57 ) ) PUNCT bracis-28376 89 58 . . PUNCT bracis-28376 90 1 a a DET bracis-28376 90 2 team team NOUN bracis-28376 90 3 of of ADP bracis-28376 90 4 legal legal ADJ bracis-28376 90 5 experts expert NOUN bracis-28376 90 6 annotated annotate VERB bracis-28376 90 7 the the DET bracis-28376 90 8 documents document NOUN bracis-28376 90 9 and and CCONJ bracis-28376 90 10 each each DET bracis-28376 90 11 sentence sentence NOUN bracis-28376 90 12 is be AUX bracis-28376 90 13 labeled label VERB bracis-28376 90 14 with with ADP bracis-28376 90 15 one one NUM bracis-28376 90 16 of of ADP bracis-28376 90 17 the the DET bracis-28376 90 18 following follow VERB bracis-28376 90 19 13 13 NUM bracis-28376 90 20 rhetorical rhetorical ADJ bracis-28376 90 21 roles role NOUN bracis-28376 90 22 : : PUNCT bracis-28376 90 23 fact fact NOUN bracis-28376 90 24 ( ( PUNCT bracis-28376 90 25 fac fac PROPN bracis-28376 90 26 ) ) PUNCT bracis-28376 90 27 , , PUNCT bracis-28376 90 28 issues issue NOUN bracis-28376 90 29 ( ( PUNCT bracis-28376 90 30 iss iss PROPN bracis-28376 90 31 ) ) PUNCT bracis-28376 90 32 , , PUNCT bracis-28376 90 33 argument argument NOUN bracis-28376 90 34 petitioner petitioner NOUN bracis-28376 90 35 ( ( PUNCT bracis-28376 90 36 arg arg NOUN bracis-28376 90 37 - - PUNCT bracis-28376 90 38 p p NOUN bracis-28376 90 39 ) ) PUNCT bracis-28376 90 40 , , PUNCT bracis-28376 90 41 argument argument NOUN bracis-28376 90 42 respondent respondent NOUN bracis-28376 90 43 ( ( PUNCT bracis-28376 90 44 arg arg NOUN bracis-28376 90 45 - - PUNCT bracis-28376 90 46 r r NOUN bracis-28376 90 47 ) ) PUNCT bracis-28376 90 48 , , PUNCT bracis-28376 90 49 statute statute NOUN bracis-28376 90 50 ( ( PUNCT bracis-28376 90 51 sta sta NOUN bracis-28376 90 52 ) ) PUNCT bracis-28376 90 53 , , PUNCT bracis-28376 90 54 dissent dissent NOUN bracis-28376 90 55 ( ( PUNCT bracis-28376 90 56 dis dis PROPN bracis-28376 90 57 ) ) PUNCT bracis-28376 90 58 , , PUNCT bracis-28376 90 59 precedent precedent NOUN bracis-28376 90 60 relied rely VERB bracis-28376 90 61 upon upon SCONJ bracis-28376 90 62 ( ( PUNCT bracis-28376 90 63 pre pre VERB bracis-28376 90 64 - - NOUN bracis-28376 90 65 r r NOUN bracis-28376 90 66 ) ) PUNCT bracis-28376 90 67 , , PUNCT bracis-28376 90 68 precedent precedent NOUN bracis-28376 90 69 not not PART bracis-28376 90 70 relied rely VERB bracis-28376 90 71 upon upon SCONJ bracis-28376 90 72 ( ( PUNCT bracis-28376 90 73 pre pre VERB bracis-28376 90 74 - - ADJ bracis-28376 90 75 nr nr PRON bracis-28376 90 76 ) ) PUNCT bracis-28376 90 77 , , PUNCT bracis-28376 90 78 precedent precedent NOUN bracis-28376 90 79 overruled overrule VERB bracis-28376 90 80 ( ( PUNCT bracis-28376 90 81 pre pre NOUN bracis-28376 90 82 - - NOUN bracis-28376 90 83 o o NOUN bracis-28376 90 84 ) ) PUNCT bracis-28376 90 85 , , PUNCT bracis-28376 90 86 ruling rule VERB bracis-28376 90 87 by by ADP bracis-28376 90 88 lower low ADJ bracis-28376 90 89 court court NOUN bracis-28376 90 90 ( ( PUNCT bracis-28376 90 91 rlc rlc PROPN bracis-28376 90 92 ) ) PUNCT bracis-28376 90 93 , , PUNCT bracis-28376 90 94 ruling rule VERB bracis-28376 90 95 by by ADP bracis-28376 90 96 present present ADJ bracis-28376 90 97 court court NOUN bracis-28376 90 98 ( ( PUNCT bracis-28376 90 99 rpc rpc PROPN bracis-28376 90 100 ) ) PUNCT bracis-28376 90 101 , , PUNCT bracis-28376 90 102 ratio ratio NOUN bracis-28376 90 103 of of ADP bracis-28376 90 104 the the DET bracis-28376 90 105 decision decision NOUN bracis-28376 90 106 ( ( PUNCT bracis-28376 90 107 rod rod NOUN bracis-28376 90 108 ) ) PUNCT bracis-28376 90 109 , , PUNCT bracis-28376 90 110 and and CCONJ bracis-28376 90 111 none none NOUN bracis-28376 90 112 ( ( PUNCT bracis-28376 90 113 non non ADJ bracis-28376 90 114 ) ) PUNCT bracis-28376 90 115 . . PUNCT bracis-28376 91 1 we we PRON bracis-28376 91 2 follow follow VERB bracis-28376 91 3 the the DET bracis-28376 91 4 same same ADJ bracis-28376 91 5 procedure procedure NOUN bracis-28376 91 6 from from ADP bracis-28376 91 7 [ [ X bracis-28376 91 8 15 15 NUM bracis-28376 91 9 ] ] PUNCT bracis-28376 91 10 to to PART bracis-28376 91 11 create create VERB bracis-28376 91 12 the the DET bracis-28376 91 13 7 7 NUM bracis-28376 91 14 - - PUNCT bracis-28376 91 15 roles role NOUN bracis-28376 91 16 dataset dataset NOUN bracis-28376 91 17 . . PUNCT bracis-28376 92 1 the the DET bracis-28376 92 2 issue issue NOUN bracis-28376 92 3 labels label NOUN bracis-28376 92 4 are be AUX bracis-28376 92 5 renamed rename VERB bracis-28376 92 6 as as ADP bracis-28376 92 7 fact fact NOUN bracis-28376 92 8 , , PUNCT bracis-28376 92 9 the the DET bracis-28376 92 10 argument argument NOUN bracis-28376 92 11 petitioner petitioner NOUN bracis-28376 92 12 and and CCONJ bracis-28376 92 13 argument argument NOUN bracis-28376 92 14 respondent respondent NOUN bracis-28376 92 15 labels label NOUN bracis-28376 92 16 are be AUX bracis-28376 92 17 renamed rename VERB bracis-28376 92 18 as as ADP bracis-28376 92 19 argument argument NOUN bracis-28376 92 20 ( ( PUNCT bracis-28376 92 21 arg arg NOUN bracis-28376 92 22 ) ) PUNCT bracis-28376 92 23 , , PUNCT bracis-28376 92 24 and and CCONJ bracis-28376 92 25 the the DET bracis-28376 92 26 precedent precedent NOUN bracis-28376 92 27 relied rely VERB bracis-28376 92 28 upon upon SCONJ bracis-28376 92 29 , , PUNCT bracis-28376 92 30 precedent precedent NOUN bracis-28376 92 31 not not PART bracis-28376 92 32 relied rely VERB bracis-28376 92 33 upon upon SCONJ bracis-28376 92 34 and and CCONJ bracis-28376 92 35 precedent precedent NOUN bracis-28376 92 36 overruled overrule VERB bracis-28376 92 37 ( ( PUNCT bracis-28376 92 38 pre pre ADJ bracis-28376 92 39 - - ADJ bracis-28376 92 40 o o NOUN bracis-28376 92 41 ) ) PUNCT bracis-28376 92 42 labels label NOUN bracis-28376 92 43 are be AUX bracis-28376 92 44 renamed rename VERB bracis-28376 92 45 as as ADP bracis-28376 92 46 precedent precedent NOUN bracis-28376 92 47 ( ( PUNCT bracis-28376 92 48 pre pre NOUN bracis-28376 92 49 ) ) PUNCT bracis-28376 92 50 . . PUNCT bracis-28376 93 1 also also ADV bracis-28376 93 2 , , PUNCT bracis-28376 93 3 the the DET bracis-28376 93 4 dissent dissent NOUN bracis-28376 93 5 and and CCONJ bracis-28376 93 6 none none NOUN bracis-28376 93 7 sentences sentence NOUN bracis-28376 93 8 are be AUX bracis-28376 93 9 ignored ignore VERB bracis-28376 93 10 for for ADP bracis-28376 93 11 classification classification NOUN bracis-28376 93 12 purposes purpose NOUN bracis-28376 93 13 . . PUNCT bracis-28376 94 1 although although SCONJ bracis-28376 94 2 , , PUNCT bracis-28376 94 3 some some DET bracis-28376 94 4 models model NOUN bracis-28376 94 5 rely rely VERB bracis-28376 94 6 on on ADP bracis-28376 94 7 document document NOUN bracis-28376 94 8 - - PUNCT bracis-28376 94 9 level level NOUN bracis-28376 94 10 data datum NOUN bracis-28376 94 11 to to PART bracis-28376 94 12 produce produce VERB bracis-28376 94 13 feature feature NOUN bracis-28376 94 14 vectors vector NOUN bracis-28376 94 15 and and CCONJ bracis-28376 94 16 discarding discard VERB bracis-28376 94 17 such such ADJ bracis-28376 94 18 sentences sentence NOUN bracis-28376 94 19 would would AUX bracis-28376 94 20 result result VERB bracis-28376 94 21 in in ADP bracis-28376 94 22 a a DET bracis-28376 94 23 data data NOUN bracis-28376 94 24 loss loss NOUN bracis-28376 94 25 degree degree NOUN bracis-28376 94 26 . . PUNCT bracis-28376 95 1 regarding regard VERB bracis-28376 95 2 chunk chunk NOUN bracis-28376 95 3 - - PUNCT bracis-28376 95 4 based base VERB bracis-28376 95 5 models model NOUN bracis-28376 95 6 , , PUNCT bracis-28376 95 7 dissent dissent NOUN bracis-28376 95 8 and and CCONJ bracis-28376 95 9 none none NOUN bracis-28376 95 10 sentences sentence NOUN bracis-28376 95 11 are be AUX bracis-28376 95 12 part part NOUN bracis-28376 95 13 of of ADP bracis-28376 95 14 the the DET bracis-28376 95 15 model model NOUN bracis-28376 95 16 input input NOUN bracis-28376 95 17 but but CCONJ bracis-28376 95 18 their their PRON bracis-28376 95 19 predicted predict VERB bracis-28376 95 20 labels label NOUN bracis-28376 95 21 are be AUX bracis-28376 95 22 discarded discard VERB bracis-28376 95 23 when when SCONJ bracis-28376 95 24 we we PRON bracis-28376 95 25 compute compute VERB bracis-28376 95 26 the the DET bracis-28376 95 27 loss loss NOUN bracis-28376 95 28 function function NOUN bracis-28376 95 29 value value NOUN bracis-28376 95 30 . . PUNCT bracis-28376 96 1 regarding regard VERB bracis-28376 96 2 pe pe ADV bracis-28376 96 3 - - PUNCT bracis-28376 96 4 based base VERB bracis-28376 96 5 models model NOUN bracis-28376 96 6 , , PUNCT bracis-28376 96 7 such such ADJ bracis-28376 96 8 sentences sentence NOUN bracis-28376 96 9 are be AUX bracis-28376 96 10 not not PART bracis-28376 96 11 part part NOUN bracis-28376 96 12 of of ADP bracis-28376 96 13 the the DET bracis-28376 96 14 model model NOUN bracis-28376 96 15 input input NOUN bracis-28376 96 16 but but CCONJ bracis-28376 96 17 they they PRON bracis-28376 96 18 are be AUX bracis-28376 96 19 exploited exploit VERB bracis-28376 96 20 when when SCONJ bracis-28376 96 21 computing compute VERB bracis-28376 96 22 the the DET bracis-28376 96 23 positional positional ADJ bracis-28376 96 24 embeddings embedding NOUN bracis-28376 96 25 to to PART bracis-28376 96 26 preserve preserve VERB bracis-28376 96 27 the the DET bracis-28376 96 28 actual actual ADJ bracis-28376 96 29 sentence sentence NOUN bracis-28376 96 30 positions position NOUN bracis-28376 96 31 . . PUNCT bracis-28376 97 1 by by ADP bracis-28376 97 2 discarding discard VERB bracis-28376 97 3 the the DET bracis-28376 97 4 dissent dissent NOUN bracis-28376 97 5 and and CCONJ bracis-28376 97 6 none none NOUN bracis-28376 97 7 sentences sentence NOUN bracis-28376 97 8 , , PUNCT bracis-28376 97 9 the the DET bracis-28376 97 10 7 7 NUM bracis-28376 97 11 - - PUNCT bracis-28376 97 12 roles role NOUN bracis-28376 97 13 dataset dataset NOUN bracis-28376 97 14 is be AUX bracis-28376 97 15 composed compose VERB bracis-28376 97 16 of of ADP bracis-28376 97 17 16,264 16,264 NUM bracis-28376 97 18 , , PUNCT bracis-28376 97 19 2,109 2,109 NUM bracis-28376 97 20 , , PUNCT bracis-28376 97 21 and and CCONJ bracis-28376 97 22 2,094 2,094 NUM bracis-28376 97 23 sentences sentence NOUN bracis-28376 97 24 in in ADP bracis-28376 97 25 the the DET bracis-28376 97 26 training training NOUN bracis-28376 97 27 , , PUNCT bracis-28376 97 28 development development NOUN bracis-28376 97 29 , , PUNCT bracis-28376 97 30 and and CCONJ bracis-28376 97 31 test test NOUN bracis-28376 97 32 sets set NOUN bracis-28376 97 33 , , PUNCT bracis-28376 97 34 respectively respectively ADV bracis-28376 97 35 . . PUNCT bracis-28376 97 36 to to PART bracis-28376 97 37 be be AUX bracis-28376 97 38 effective effective ADJ bracis-28376 97 39 , , PUNCT bracis-28376 97 40 pe pe ADJ bracis-28376 97 41 - - PUNCT bracis-28376 97 42 based base VERB bracis-28376 97 43 models model NOUN bracis-28376 97 44 require require VERB bracis-28376 97 45 some some DET bracis-28376 97 46 degree degree NOUN bracis-28376 97 47 of of ADP bracis-28376 97 48 correlation correlation NOUN bracis-28376 97 49 between between ADP bracis-28376 97 50 sentence sentence NOUN bracis-28376 97 51 positions position NOUN bracis-28376 97 52 and and CCONJ bracis-28376 97 53 rhetorical rhetorical ADJ bracis-28376 97 54 roles role NOUN bracis-28376 97 55 [ [ X bracis-28376 97 56 12 12 NUM bracis-28376 97 57 ] ] PUNCT bracis-28376 97 58 . . PUNCT bracis-28376 98 1 thus thus ADV bracis-28376 98 2 , , PUNCT bracis-28376 98 3 we we PRON bracis-28376 98 4 designed design VERB bracis-28376 98 5 the the DET bracis-28376 98 6 4 4 NUM bracis-28376 98 7 - - PUNCT bracis-28376 98 8 roles role NOUN bracis-28376 98 9 dataset dataset NOUN bracis-28376 98 10 to to PART bracis-28376 98 11 perform perform VERB bracis-28376 98 12 a a DET bracis-28376 98 13 fairer fair ADJ bracis-28376 98 14 evaluation evaluation NOUN bracis-28376 98 15 of of ADP bracis-28376 98 16 pe pe ADV bracis-28376 98 17 - - PUNCT bracis-28376 98 18 based base VERB bracis-28376 98 19 models model NOUN bracis-28376 98 20 . . PUNCT bracis-28376 99 1 to to PART bracis-28376 99 2 achieve achieve VERB bracis-28376 99 3 this this DET bracis-28376 99 4 goal goal NOUN bracis-28376 99 5 , , PUNCT bracis-28376 99 6 we we PRON bracis-28376 99 7 modified modify VERB bracis-28376 99 8 the the DET bracis-28376 99 9 original original ADJ bracis-28376 99 10 dataset dataset NOUN bracis-28376 99 11 from from ADP bracis-28376 99 12 [ [ X bracis-28376 99 13 15 15 NUM bracis-28376 99 14 ] ] PUNCT bracis-28376 99 15 by by ADP bracis-28376 99 16 maintaining maintain VERB bracis-28376 99 17 the the DET bracis-28376 99 18 roles role NOUN bracis-28376 99 19 occurring occur VERB bracis-28376 99 20 often often ADV bracis-28376 99 21 at at ADP bracis-28376 99 22 the the DET bracis-28376 99 23 same same ADJ bracis-28376 99 24 positions position NOUN bracis-28376 99 25 ( ( PUNCT bracis-28376 99 26 i.e. i.e. X bracis-28376 99 27 , , PUNCT bracis-28376 99 28 the the DET bracis-28376 99 29 fact fact NOUN bracis-28376 99 30 , , PUNCT bracis-28376 99 31 ruling rule VERB bracis-28376 99 32 by by ADP bracis-28376 99 33 present present ADJ bracis-28376 99 34 court court NOUN bracis-28376 99 35 and and CCONJ bracis-28376 99 36 ratioofthedecision ratioofthedecision NOUN bracis-28376 99 37 ) ) PUNCT bracis-28376 99 38 and and CCONJ bracis-28376 99 39 by by ADP bracis-28376 99 40 renaming rename VERB bracis-28376 99 41 the the DET bracis-28376 99 42 remaining remain VERB bracis-28376 99 43 roles role NOUN bracis-28376 99 44 as as ADP bracis-28376 99 45 other other ADJ bracis-28376 99 46 . . PUNCT bracis-28376 100 1 we we PRON bracis-28376 100 2 do do AUX bracis-28376 100 3 not not PART bracis-28376 100 4 discard discard VERB bracis-28376 100 5 sentences sentence NOUN bracis-28376 100 6 in in ADP bracis-28376 100 7 this this DET bracis-28376 100 8 process process NOUN bracis-28376 100 9 and and CCONJ bracis-28376 100 10 hence hence ADV bracis-28376 100 11 the the DET bracis-28376 100 12 4 4 NUM bracis-28376 100 13 - - PUNCT bracis-28376 100 14 roles role NOUN bracis-28376 100 15 dataset dataset NOUN bracis-28376 100 16 has have VERB bracis-28376 100 17 the the DET bracis-28376 100 18 same same ADJ bracis-28376 100 19 number number NOUN bracis-28376 100 20 of of ADP bracis-28376 100 21 sentences sentence NOUN bracis-28376 100 22 as as ADP bracis-28376 100 23 the the DET bracis-28376 100 24 original original ADJ bracis-28376 100 25 one one NUM bracis-28376 100 26 . . PUNCT bracis-28376 101 1 the the DET bracis-28376 101 2 original original ADJ bracis-28376 101 3 dataset dataset NOUN bracis-28376 101 4 is be AUX bracis-28376 101 5 highly highly ADV bracis-28376 101 6 unbalanced unbalanced ADJ bracis-28376 101 7 and and CCONJ bracis-28376 101 8 this this PRON bracis-28376 101 9 is be AUX bracis-28376 101 10 a a DET bracis-28376 101 11 characteristic characteristic NOUN bracis-28376 101 12 of of ADP bracis-28376 101 13 the the DET bracis-28376 101 14 7 7 NUM bracis-28376 101 15 - - PUNCT bracis-28376 101 16 roles role NOUN bracis-28376 101 17 and and CCONJ bracis-28376 101 18 4 4 NUM bracis-28376 101 19 - - PUNCT bracis-28376 101 20 roles role NOUN bracis-28376 101 21 datasets dataset NOUN bracis-28376 101 22 as as ADV bracis-28376 101 23 well well ADV bracis-28376 101 24 . . PUNCT bracis-28376 102 1 the the DET bracis-28376 102 2 distribution distribution NOUN bracis-28376 102 3 of of ADP bracis-28376 102 4 sentences sentence NOUN bracis-28376 102 5 by by ADP bracis-28376 102 6 rhetorical rhetorical ADJ bracis-28376 102 7 role role NOUN bracis-28376 102 8 in in ADP bracis-28376 102 9 the the DET bracis-28376 102 10 three three NUM bracis-28376 102 11 datasets dataset NOUN bracis-28376 102 12 is be AUX bracis-28376 102 13 shown show VERB bracis-28376 102 14 in in ADP bracis-28376 102 15 fig fig NOUN bracis-28376 102 16 . . PUNCT bracis-28376 102 17     SPACE bracis-28376 103 1 3 3 NUM bracis-28376 103 2 . . X bracis-28376 103 3 fig fig NOUN bracis-28376 103 4 . . PUNCT bracis-28376 104 1 3 3 X bracis-28376 104 2 . . X bracis-28376 104 3 number number NOUN bracis-28376 104 4 of of ADP bracis-28376 104 5 sentences sentence NOUN bracis-28376 104 6 per per ADP bracis-28376 104 7 rhetorical rhetorical ADJ bracis-28376 104 8 role role NOUN bracis-28376 104 9 in in ADP bracis-28376 104 10 the the DET bracis-28376 104 11 original original ADJ bracis-28376 104 12 , , PUNCT bracis-28376 104 13 7 7 NUM bracis-28376 104 14 - - PUNCT bracis-28376 104 15 roles role NOUN bracis-28376 104 16 , , PUNCT bracis-28376 104 17 and and CCONJ bracis-28376 104 18 4 4 NUM bracis-28376 104 19 - - PUNCT bracis-28376 104 20 roles role NOUN bracis-28376 104 21 datasets dataset NOUN bracis-28376 104 22 . . PUNCT bracis-28376 105 1 full full ADJ bracis-28376 105 2 size size NOUN bracis-28376 105 3 image image NOUN bracis-28376 105 4 3.2 3.2 NUM bracis-28376 105 5 models model NOUN bracis-28376 105 6 we we PRON bracis-28376 105 7 implement implement VERB bracis-28376 105 8 models model NOUN bracis-28376 105 9 relying rely VERB bracis-28376 105 10 on on ADP bracis-28376 105 11 the the DET bracis-28376 105 12 approaches approach NOUN bracis-28376 105 13 from from ADP bracis-28376 105 14 [ [ X bracis-28376 105 15 10 10 NUM bracis-28376 105 16 , , PUNCT bracis-28376 105 17 12 12 NUM bracis-28376 105 18 ] ] PUNCT bracis-28376 105 19 , , PUNCT bracis-28376 105 20 and and CCONJ bracis-28376 105 21 [ [ X bracis-28376 105 22 11 11 NUM bracis-28376 105 23 ] ] PUNCT bracis-28376 105 24 . . PUNCT bracis-28376 106 1 for for ADP bracis-28376 106 2 each each DET bracis-28376 106 3 approach approach NOUN bracis-28376 106 4 , , PUNCT bracis-28376 106 5 we we PRON bracis-28376 106 6 experiment experiment VERB bracis-28376 106 7 with with ADP bracis-28376 106 8 two two NUM bracis-28376 106 9 or or CCONJ bracis-28376 106 10 three three NUM bracis-28376 106 11 transformer transformer NOUN bracis-28376 106 12 encoders encoder NOUN bracis-28376 106 13 . . PUNCT bracis-28376 107 1 details detail NOUN bracis-28376 107 2 of of ADP bracis-28376 107 3 our our PRON bracis-28376 107 4 results result NOUN bracis-28376 107 5 and and CCONJ bracis-28376 107 6 source source NOUN bracis-28376 107 7 code code NOUN bracis-28376 107 8 are be AUX bracis-28376 107 9 available available ADJ bracis-28376 107 10 at at ADP bracis-28376 107 11 https://github.com/alexlimatds/bracis_2023 https://github.com/alexlimatds/bracis_2023 PROPN bracis-28376 107 12 . . PUNCT bracis-28376 108 1 mixup mixup NOUN bracis-28376 108 2 - - PUNCT bracis-28376 108 3 a a PRON bracis-28376 108 4 : : PUNCT bracis-28376 108 5 these these PRON bracis-28376 108 6 are be AUX bracis-28376 108 7 models model NOUN bracis-28376 108 8 whose whose DET bracis-28376 108 9 training training NOUN bracis-28376 108 10 includes include VERB bracis-28376 108 11 mixup mixup NOUN bracis-28376 108 12 vectors vector NOUN bracis-28376 108 13 as as SCONJ bracis-28376 108 14 proposed propose VERB bracis-28376 108 15 by by ADP bracis-28376 108 16 [ [ X bracis-28376 108 17 10 10 NUM bracis-28376 108 18 ] ] PUNCT bracis-28376 108 19 . . PUNCT bracis-28376 109 1 at at ADP bracis-28376 109 2 first first ADV bracis-28376 109 3 , , PUNCT bracis-28376 109 4 a a DET bracis-28376 109 5 transformer transformer NOUN bracis-28376 109 6 encoder encoder NOUN bracis-28376 109 7 is be AUX bracis-28376 109 8 fine fine ADV bracis-28376 109 9 - - PUNCT bracis-28376 109 10 tuned tune VERB bracis-28376 109 11 on on ADP bracis-28376 109 12 the the DET bracis-28376 109 13 training training NOUN bracis-28376 109 14 dataset dataset NOUN bracis-28376 109 15 in in ADP bracis-28376 109 16 the the DET bracis-28376 109 17 same same ADJ bracis-28376 109 18 fashion fashion NOUN bracis-28376 109 19 as as SCONJ bracis-28376 109 20 the the DET bracis-28376 109 21 singlesc singlesc NOUN bracis-28376 109 22 approach approach NOUN bracis-28376 109 23 described describe VERB bracis-28376 109 24 ahead ahead ADV bracis-28376 109 25 ( ( PUNCT bracis-28376 109 26 step step VERB bracis-28376 109 27 a a PRON bracis-28376 109 28 in in ADP bracis-28376 109 29 fig fig NOUN bracis-28376 109 30 . . PUNCT bracis-28376 109 31     SPACE bracis-28376 110 1 4 4 NUM bracis-28376 110 2 ) ) PUNCT bracis-28376 110 3 . . PUNCT bracis-28376 111 1 in in ADP bracis-28376 111 2 the the DET bracis-28376 111 3 following following NOUN bracis-28376 111 4 , , PUNCT bracis-28376 111 5 the the DET bracis-28376 111 6 fine fine ADV bracis-28376 111 7 - - PUNCT bracis-28376 111 8 tuned tune VERB bracis-28376 111 9 transformer transformer NOUN bracis-28376 111 10 is be AUX bracis-28376 111 11 employed employ VERB bracis-28376 111 12 to to PART bracis-28376 111 13 yield yield VERB bracis-28376 111 14 sentence sentence NOUN bracis-28376 111 15 embeddings embedding NOUN bracis-28376 111 16 which which PRON bracis-28376 111 17 serve serve VERB bracis-28376 111 18 as as ADP bracis-28376 111 19 source source NOUN bracis-28376 111 20 data datum NOUN bracis-28376 111 21 of of ADP bracis-28376 111 22 the the DET bracis-28376 111 23 mixup mixup NOUN bracis-28376 111 24 procedure procedure NOUN bracis-28376 111 25 ( ( PUNCT bracis-28376 111 26 step step NOUN bracis-28376 111 27 b b NOUN bracis-28376 111 28 in in ADP bracis-28376 111 29 fig fig NOUN bracis-28376 111 30 . . PUNCT bracis-28376 111 31     SPACE bracis-28376 112 1 4 4 NUM bracis-28376 112 2 ) ) PUNCT bracis-28376 112 3 . . PUNCT bracis-28376 113 1 finally finally ADV bracis-28376 113 2 , , PUNCT bracis-28376 113 3 the the DET bracis-28376 113 4 mixup mixup NOUN bracis-28376 113 5 vectors vector NOUN bracis-28376 113 6 and and CCONJ bracis-28376 113 7 the the DET bracis-28376 113 8 sentence sentence NOUN bracis-28376 113 9 embeddings embedding NOUN bracis-28376 113 10 are be AUX bracis-28376 113 11 employed employ VERB bracis-28376 113 12 to to PART bracis-28376 113 13 train train VERB bracis-28376 113 14 a a DET bracis-28376 113 15 mixup mixup NOUN bracis-28376 113 16 - - PUNCT bracis-28376 113 17 a a DET bracis-28376 113 18 model model NOUN bracis-28376 113 19 ( ( PUNCT bracis-28376 113 20 step step NOUN bracis-28376 113 21 c c NOUN bracis-28376 113 22 in in ADP bracis-28376 113 23 fig fig NOUN bracis-28376 113 24 . . PUNCT bracis-28376 113 25     SPACE bracis-28376 113 26 4 4 NUM bracis-28376 113 27 ) ) PUNCT bracis-28376 113 28 . . PUNCT bracis-28376 114 1 the the DET bracis-28376 114 2 training training NOUN bracis-28376 114 3 of of ADP bracis-28376 114 4 a a DET bracis-28376 114 5 mixup mixup NOUN bracis-28376 114 6 - - PUNCT bracis-28376 114 7 a a DET bracis-28376 114 8 model model NOUN bracis-28376 114 9 updates update VERB bracis-28376 114 10 only only ADV bracis-28376 114 11 the the DET bracis-28376 114 12 weights weight NOUN bracis-28376 114 13 of of ADP bracis-28376 114 14 its its PRON bracis-28376 114 15 classification classification NOUN bracis-28376 114 16 layer layer NOUN bracis-28376 114 17 . . PUNCT bracis-28376 115 1 sentences sentence NOUN bracis-28376 115 2 are be AUX bracis-28376 115 3 represented represent VERB bracis-28376 115 4 by by ADP bracis-28376 115 5 the the DET bracis-28376 115 6 last last ADJ bracis-28376 115 7 embedding embed VERB bracis-28376 115 8 vector vector NOUN bracis-28376 115 9 of of ADP bracis-28376 115 10 the the DET bracis-28376 115 11 token token ADJ bracis-28376 115 12 marking mark VERB bracis-28376 115 13 the the DET bracis-28376 115 14 beginning beginning NOUN bracis-28376 115 15 of of ADP bracis-28376 115 16 the the DET bracis-28376 115 17 input input NOUN bracis-28376 115 18 ( ( PUNCT bracis-28376 115 19 [ [ X bracis-28376 115 20 cls cls X bracis-28376 115 21 ] ] PUNCT bracis-28376 115 22 or or CCONJ bracis-28376 115 23 \(\texttt \(\texttt NOUN bracis-28376 115 24 { { PUNCT bracis-28376 115 25 < < X bracis-28376 115 26 s>}\ s>}\ ADJ bracis-28376 115 27 ) ) PUNCT bracis-28376 115 28 ) ) PUNCT bracis-28376 115 29 . . PUNCT bracis-28376 116 1 we we PRON bracis-28376 116 2 employ employ VERB bracis-28376 116 3 incaselaw incaselaw NOUN bracis-28376 116 4 [ [ X bracis-28376 116 5 16 16 NUM bracis-28376 116 6 ] ] PUNCT bracis-28376 116 7 and and CCONJ bracis-28376 116 8 roberta roberta NOUN bracis-28376 116 9 - - PUNCT bracis-28376 116 10 base base NOUN bracis-28376 116 11 [ [ X bracis-28376 116 12 13 13 NUM bracis-28376 116 13 ] ] PUNCT bracis-28376 116 14 as as ADP bracis-28376 116 15 sentence sentence NOUN bracis-28376 116 16 encoders encoder NOUN bracis-28376 116 17 . . PUNCT bracis-28376 117 1 incaselaw incaselaw NOUN bracis-28376 117 2 is be AUX bracis-28376 117 3 a a DET bracis-28376 117 4 bert bert PROPN bracis-28376 117 5 model model NOUN bracis-28376 117 6 pre pre VERB bracis-28376 117 7 - - VERB bracis-28376 117 8 trained train VERB bracis-28376 117 9 with with ADP bracis-28376 117 10 corpora corpora NOUN bracis-28376 117 11 from from ADP bracis-28376 117 12 the the DET bracis-28376 117 13 indian indian ADJ bracis-28376 117 14 legal legal ADJ bracis-28376 117 15 system system NOUN bracis-28376 117 16 . . PUNCT bracis-28376 118 1 the the DET bracis-28376 118 2 generation generation NOUN bracis-28376 118 3 of of ADP bracis-28376 118 4 mixup mixup NOUN bracis-28376 118 5 vectors vector NOUN bracis-28376 118 6 is be AUX bracis-28376 118 7 oriented orient VERB bracis-28376 118 8 to to ADP bracis-28376 118 9 selected select VERB bracis-28376 118 10 labels label NOUN bracis-28376 118 11 and and CCONJ bracis-28376 118 12 an an DET bracis-28376 118 13 augmentation augmentation NOUN bracis-28376 118 14 rate rate NOUN bracis-28376 118 15 \(\gamma \(\gamma NOUN bracis-28376 118 16 \ \ PROPN bracis-28376 118 17 ) ) PUNCT bracis-28376 118 18 . . PUNCT bracis-28376 119 1 for for ADP bracis-28376 119 2 each each DET bracis-28376 119 3 selected select VERB bracis-28376 119 4 label label NOUN bracis-28376 119 5 \(l_i\ \(l_i\ NOUN bracis-28376 119 6 ) ) PUNCT bracis-28376 119 7 , , PUNCT bracis-28376 119 8 we we PRON bracis-28376 119 9 generate generate VERB bracis-28376 119 10 \(n_{l_i \(n_{l_i NOUN bracis-28376 119 11 } } PUNCT bracis-28376 119 12 \cdot \cdot NOUN bracis-28376 119 13 \gamma \gamma NUM bracis-28376 119 14 \ \ NOUN bracis-28376 119 15 ) ) PUNCT bracis-28376 119 16 mixup mixup NOUN bracis-28376 119 17 vectors vector NOUN bracis-28376 119 18 , , PUNCT bracis-28376 119 19 where where SCONJ bracis-28376 119 20 \(n_{l_i}\ \(n_{l_i}\ X bracis-28376 119 21 ) ) PUNCT bracis-28376 119 22 means mean VERB bracis-28376 119 23 the the DET bracis-28376 119 24 number number NOUN bracis-28376 119 25 of of ADP bracis-28376 119 26 training training NOUN bracis-28376 119 27 sentences sentence NOUN bracis-28376 119 28 labeled label VERB bracis-28376 119 29 with with ADP bracis-28376 119 30 \(l_i\ \(l_i\ NOUN bracis-28376 119 31 ) ) PUNCT bracis-28376 119 32 . . PUNCT bracis-28376 120 1 a a DET bracis-28376 120 2 mixup mixup NOUN bracis-28376 120 3 vector vector NOUN bracis-28376 120 4 is be AUX bracis-28376 120 5 generated generate VERB bracis-28376 120 6 from from ADP bracis-28376 120 7 a a DET bracis-28376 120 8 sentence sentence NOUN bracis-28376 120 9 embedding embed VERB bracis-28376 120 10 labeled label VERB bracis-28376 120 11 with with ADP bracis-28376 120 12 \(l_i\ \(l_i\ NOUN bracis-28376 120 13 ) ) PUNCT bracis-28376 120 14 and and CCONJ bracis-28376 120 15 a a DET bracis-28376 120 16 sentence sentence NOUN bracis-28376 120 17 embedding embed VERB bracis-28376 120 18 labeled label VERB bracis-28376 120 19 with with ADP bracis-28376 120 20 \(l_j\ \(l_j\ PROPN bracis-28376 120 21 ) ) PUNCT bracis-28376 120 22 where where SCONJ bracis-28376 120 23 \(i \(i ADJ bracis-28376 120 24 \ne \ne PROPN bracis-28376 120 25 j\ j\ PROPN bracis-28376 120 26 ) ) PUNCT bracis-28376 120 27 ( ( PUNCT bracis-28376 120 28 the the DET bracis-28376 120 29 embeddings embedding NOUN bracis-28376 120 30 are be AUX bracis-28376 120 31 selected select VERB bracis-28376 120 32 at at ADP bracis-28376 120 33 random random ADJ bracis-28376 120 34 ) ) PUNCT bracis-28376 120 35 . . PUNCT bracis-28376 121 1 when when SCONJ bracis-28376 121 2 experimenting experiment VERB bracis-28376 121 3 with with ADP bracis-28376 121 4 the the DET bracis-28376 121 5 7 7 NUM bracis-28376 121 6 - - PUNCT bracis-28376 121 7 labels label NOUN bracis-28376 121 8 dataset dataset NOUN bracis-28376 121 9 , , PUNCT bracis-28376 121 10 we we PRON bracis-28376 121 11 set set VERB bracis-28376 121 12 \(\gamma \(\gamma NOUN bracis-28376 122 1 = = NOUN bracis-28376 122 2 0.5\ 0.5\ NUM bracis-28376 122 3 ) ) PUNCT bracis-28376 122 4 , , PUNCT bracis-28376 122 5 \(\alpha \(\alpha NOUN bracis-28376 122 6 = = ADJ bracis-28376 122 7 0.1\ 0.1\ X bracis-28376 122 8 ) ) PUNCT bracis-28376 122 9 , , PUNCT bracis-28376 122 10 the the DET bracis-28376 122 11 selected select VERB bracis-28376 122 12 labels label NOUN bracis-28376 122 13 are be AUX bracis-28376 122 14 argument argument NOUN bracis-28376 122 15 , , PUNCT bracis-28376 122 16 statute statute NOUN bracis-28376 122 17 , , PUNCT bracis-28376 122 18 precedent precedent NOUN bracis-28376 122 19 , , PUNCT bracis-28376 122 20 rulingbylowercourt rulingbylowercourt ADJ bracis-28376 122 21 , , PUNCT bracis-28376 122 22 and and CCONJ bracis-28376 122 23 rulingbypresentcourt rulingbypresentcourt NOUN bracis-28376 122 24 which which PRON bracis-28376 122 25 results result VERB bracis-28376 122 26 in in ADP bracis-28376 122 27 3,881 3,881 NUM bracis-28376 122 28 synthetic synthetic ADJ bracis-28376 122 29 vectors vector NOUN bracis-28376 122 30 . . PUNCT bracis-28376 123 1 for for ADP bracis-28376 123 2 the the DET bracis-28376 123 3 4 4 NUM bracis-28376 123 4 - - PUNCT bracis-28376 123 5 labels label NOUN bracis-28376 123 6 dataset dataset NOUN bracis-28376 123 7 , , PUNCT bracis-28376 123 8 we we PRON bracis-28376 123 9 set set VERB bracis-28376 123 10 \(\gamma \(\gamma NOUN bracis-28376 124 1 = = NOUN bracis-28376 124 2 0.5\ 0.5\ NUM bracis-28376 124 3 ) ) PUNCT bracis-28376 124 4 , , PUNCT bracis-28376 124 5 \(\alpha \(\alpha NOUN bracis-28376 124 6 = = ADJ bracis-28376 124 7 0.1\ 0.1\ X bracis-28376 124 8 ) ) PUNCT bracis-28376 124 9 , , PUNCT bracis-28376 124 10 the the DET bracis-28376 124 11 selected select VERB bracis-28376 124 12 labels label NOUN bracis-28376 124 13 are be AUX bracis-28376 124 14 fact fact NOUN bracis-28376 124 15 , , PUNCT bracis-28376 124 16 ratio ratio NOUN bracis-28376 124 17 of of ADP bracis-28376 124 18 the the DET bracis-28376 124 19 decision decision NOUN bracis-28376 124 20 , , PUNCT bracis-28376 124 21 and and CCONJ bracis-28376 124 22 ruling rule VERB bracis-28376 124 23 by by ADP bracis-28376 124 24 present present ADJ bracis-28376 124 25 court court NOUN bracis-28376 124 26 which which PRON bracis-28376 124 27 results result VERB bracis-28376 124 28 in in ADP bracis-28376 124 29 4,408 4,408 NUM bracis-28376 124 30 synthetic synthetic ADJ bracis-28376 124 31 vectors vector NOUN bracis-28376 124 32 . . PUNCT bracis-28376 125 1 fig fig NOUN bracis-28376 125 2 . . PUNCT bracis-28376 126 1 4 4 X bracis-28376 126 2 . . X bracis-28376 126 3 steps step NOUN bracis-28376 126 4 performed perform VERB bracis-28376 126 5 to to PART bracis-28376 126 6 train train VERB bracis-28376 126 7 a a DET bracis-28376 126 8 mixup mixup NOUN bracis-28376 126 9 - - PUNCT bracis-28376 126 10 a a DET bracis-28376 126 11 model model NOUN bracis-28376 126 12 . . PUNCT bracis-28376 127 1 step step VERB bracis-28376 127 2 a a DET bracis-28376 127 3 : : PUNCT bracis-28376 127 4 fine fine ADV bracis-28376 127 5 - - PUNCT bracis-28376 127 6 tuning tuning NOUN bracis-28376 127 7 of of ADP bracis-28376 127 8 the the DET bracis-28376 127 9 transformer transformer NOUN bracis-28376 127 10 encoder encoder NOUN bracis-28376 127 11 by by ADP bracis-28376 127 12 training train VERB bracis-28376 127 13 a a DET bracis-28376 127 14 singlesc singlesc NOUN bracis-28376 127 15 model model NOUN bracis-28376 127 16 . . PUNCT bracis-28376 128 1 step step NOUN bracis-28376 128 2 b b NOUN bracis-28376 128 3 : : PUNCT bracis-28376 128 4 generation generation NOUN bracis-28376 128 5 of of ADP bracis-28376 128 6 mixup mixup NOUN bracis-28376 128 7 synthetic synthetic ADJ bracis-28376 128 8 vectors vector NOUN bracis-28376 128 9 from from ADP bracis-28376 128 10 sentence sentence NOUN bracis-28376 128 11 embeddings embedding NOUN bracis-28376 128 12 yielded yield VERB bracis-28376 128 13 by by ADP bracis-28376 128 14 the the DET bracis-28376 128 15 fine fine ADV bracis-28376 128 16 - - PUNCT bracis-28376 128 17 tuned tune VERB bracis-28376 128 18 transformer transformer NOUN bracis-28376 128 19 encoder encoder NOUN bracis-28376 128 20 . . PUNCT bracis-28376 129 1 step step NOUN bracis-28376 129 2 c c NOUN bracis-28376 129 3 : : PUNCT bracis-28376 129 4 employing employ VERB bracis-28376 129 5 sentence sentence NOUN bracis-28376 129 6 embeddings embedding NOUN bracis-28376 129 7 and and CCONJ bracis-28376 129 8 mixup mixup NOUN bracis-28376 129 9 vectors vector NOUN bracis-28376 129 10 to to PART bracis-28376 129 11 train train VERB bracis-28376 129 12 the the DET bracis-28376 129 13 classifier classifier NOUN bracis-28376 129 14 of of ADP bracis-28376 129 15 the the DET bracis-28376 129 16 mixup mixup NOUN bracis-28376 129 17 - - PUNCT bracis-28376 129 18 a a DET bracis-28376 129 19 model model NOUN bracis-28376 129 20 . . PUNCT bracis-28376 130 1 full full ADJ bracis-28376 130 2 size size NOUN bracis-28376 130 3 image image NOUN bracis-28376 130 4 mixup mixup PROPN bracis-28376 130 5 - - PUNCT bracis-28376 130 6 b b NOUN bracis-28376 130 7 : : PUNCT bracis-28376 130 8 these these PRON bracis-28376 130 9 are be AUX bracis-28376 130 10 models model NOUN bracis-28376 130 11 similar similar ADJ bracis-28376 130 12 to to ADP bracis-28376 130 13 the the DET bracis-28376 130 14 mixup mixup NOUN bracis-28376 130 15 - - PUNCT bracis-28376 130 16 a a DET bracis-28376 130 17 ones one NOUN bracis-28376 130 18 but but CCONJ bracis-28376 130 19 without without ADP bracis-28376 130 20 a a DET bracis-28376 130 21 prior prior ADJ bracis-28376 130 22 encoder encoder NOUN bracis-28376 130 23 fine fine ADV bracis-28376 130 24 - - PUNCT bracis-28376 130 25 tuning tune VERB bracis-28376 130 26 step step NOUN bracis-28376 130 27 . . PUNCT bracis-28376 131 1 the the DET bracis-28376 131 2 training training NOUN bracis-28376 131 3 of of ADP bracis-28376 131 4 a a DET bracis-28376 131 5 mixup mixup NOUN bracis-28376 131 6 - - PUNCT bracis-28376 131 7 b b NOUN bracis-28376 131 8 model model NOUN bracis-28376 131 9 updates update VERB bracis-28376 131 10 the the DET bracis-28376 131 11 weights weight NOUN bracis-28376 131 12 of of ADP bracis-28376 131 13 the the DET bracis-28376 131 14 encoder encoder NOUN bracis-28376 131 15 and and CCONJ bracis-28376 131 16 of of ADP bracis-28376 131 17 the the DET bracis-28376 131 18 classification classification NOUN bracis-28376 131 19 layer layer NOUN bracis-28376 131 20 ( ( PUNCT bracis-28376 131 21 i.e. i.e. X bracis-28376 131 22 , , PUNCT bracis-28376 131 23 the the DET bracis-28376 131 24 transformer transformer NOUN bracis-28376 131 25 is be AUX bracis-28376 131 26 fine fine ADV bracis-28376 131 27 - - PUNCT bracis-28376 131 28 tuned tune VERB bracis-28376 131 29 during during ADP bracis-28376 131 30 the the DET bracis-28376 131 31 training training NOUN bracis-28376 131 32 of of ADP bracis-28376 131 33 the the DET bracis-28376 131 34 model model NOUN bracis-28376 131 35 ) ) PUNCT bracis-28376 131 36 . . PUNCT bracis-28376 132 1 since since SCONJ bracis-28376 132 2 the the DET bracis-28376 132 3 sentence sentence NOUN bracis-28376 132 4 embeddings embedding NOUN bracis-28376 132 5 change change VERB bracis-28376 132 6 during during ADP bracis-28376 132 7 the the DET bracis-28376 132 8 fine fine ADV bracis-28376 132 9 - - PUNCT bracis-28376 132 10 tuning tuning NOUN bracis-28376 132 11 , , PUNCT bracis-28376 132 12 the the DET bracis-28376 132 13 mixup mixup NOUN bracis-28376 132 14 procedure procedure NOUN bracis-28376 132 15 yields yield VERB bracis-28376 132 16 a a DET bracis-28376 132 17 new new ADJ bracis-28376 132 18 set set NOUN bracis-28376 132 19 of of ADP bracis-28376 132 20 synthetic synthetic ADJ bracis-28376 132 21 vectors vector NOUN bracis-28376 132 22 at at ADP bracis-28376 132 23 each each DET bracis-28376 132 24 training training NOUN bracis-28376 132 25 epoch epoch NOUN bracis-28376 132 26 as as SCONJ bracis-28376 132 27 shown show VERB bracis-28376 132 28 in in ADP bracis-28376 132 29 fig fig NOUN bracis-28376 132 30 . . PUNCT bracis-28376 132 31     SPACE bracis-28376 133 1 5 5 NUM bracis-28376 133 2 . . PUNCT bracis-28376 134 1 as as ADP bracis-28376 134 2 before before ADV bracis-28376 134 3 , , PUNCT bracis-28376 134 4 we we PRON bracis-28376 134 5 exploit exploit VERB bracis-28376 134 6 incaselaw incaselaw NOUN bracis-28376 134 7 and and CCONJ bracis-28376 134 8 roberta roberta PROPN bracis-28376 134 9 as as ADP bracis-28376 134 10 sentence sentence NOUN bracis-28376 134 11 encoders encoder NOUN bracis-28376 134 12 . . PUNCT bracis-28376 135 1 when when SCONJ bracis-28376 135 2 experimenting experiment VERB bracis-28376 135 3 with with ADP bracis-28376 135 4 the the DET bracis-28376 135 5 7 7 NUM bracis-28376 135 6 - - PUNCT bracis-28376 135 7 labels label NOUN bracis-28376 135 8 dataset dataset NOUN bracis-28376 135 9 , , PUNCT bracis-28376 135 10 we we PRON bracis-28376 135 11 set set VERB bracis-28376 135 12 \(\gamma \(\gamma NOUN bracis-28376 136 1 = = NOUN bracis-28376 136 2 1.0\ 1.0\ NUM bracis-28376 136 3 ) ) PUNCT bracis-28376 136 4 , , PUNCT bracis-28376 137 1 \(\alpha \(\alpha NOUN bracis-28376 137 2 = = ADJ bracis-28376 137 3 0.1\ 0.1\ X bracis-28376 137 4 ) ) PUNCT bracis-28376 137 5 , , PUNCT bracis-28376 137 6 and and CCONJ bracis-28376 137 7 select select VERB bracis-28376 137 8 all all DET bracis-28376 137 9 seven seven NUM bracis-28376 137 10 labels label NOUN bracis-28376 137 11 , , PUNCT bracis-28376 137 12 which which PRON bracis-28376 137 13 results result VERB bracis-28376 137 14 in in ADP bracis-28376 137 15 about about ADV bracis-28376 137 16 14,800 14,800 NUM bracis-28376 137 17 synthetic synthetic ADJ bracis-28376 137 18 vectors vector NOUN bracis-28376 137 19 per per ADP bracis-28376 137 20 training train VERB bracis-28376 137 21 epoch epoch NOUN bracis-28376 137 22 . . PUNCT bracis-28376 138 1 for for ADP bracis-28376 138 2 the the DET bracis-28376 138 3 4 4 NUM bracis-28376 138 4 - - PUNCT bracis-28376 138 5 labels label NOUN bracis-28376 138 6 dataset dataset NOUN bracis-28376 138 7 , , PUNCT bracis-28376 138 8 we we PRON bracis-28376 138 9 set set VERB bracis-28376 138 10 \(\gamma \(\gamma NOUN bracis-28376 139 1 = = NOUN bracis-28376 139 2 1.0\ 1.0\ NUM bracis-28376 139 3 ) ) PUNCT bracis-28376 139 4 , , PUNCT bracis-28376 139 5 \(\alpha \(\alpha NOUN bracis-28376 139 6 = = NOUN bracis-28376 139 7 1.0\ 1.0\ NUM bracis-28376 139 8 ) ) PUNCT bracis-28376 140 1 , , PUNCT bracis-28376 140 2 the the DET bracis-28376 140 3 selected select VERB bracis-28376 140 4 labels label NOUN bracis-28376 140 5 are be AUX bracis-28376 140 6 fact fact NOUN bracis-28376 140 7 , , PUNCT bracis-28376 140 8 ratio ratio NOUN bracis-28376 140 9 of of ADP bracis-28376 140 10 the the DET bracis-28376 140 11 decision decision NOUN bracis-28376 140 12 , , PUNCT bracis-28376 140 13 and and CCONJ bracis-28376 140 14 ruling rule VERB bracis-28376 140 15 by by ADP bracis-28376 140 16 present present ADJ bracis-28376 140 17 court court NOUN bracis-28376 140 18 which which PRON bracis-28376 140 19 results result VERB bracis-28376 140 20 in in ADP bracis-28376 140 21 about about ADP bracis-28376 140 22 8,470 8,470 NUM bracis-28376 140 23 per per ADP bracis-28376 140 24 training train VERB bracis-28376 140 25 epoch epoch NOUN bracis-28376 140 26 . . PUNCT bracis-28376 141 1 fig fig NOUN bracis-28376 141 2 . . PUNCT bracis-28376 142 1 5 5 X bracis-28376 142 2 . . X bracis-28376 142 3 training training NOUN bracis-28376 142 4 procedure procedure NOUN bracis-28376 142 5 of of ADP bracis-28376 142 6 a a DET bracis-28376 142 7 mixup mixup NOUN bracis-28376 142 8 - - PUNCT bracis-28376 142 9 b b NOUN bracis-28376 142 10 model model NOUN bracis-28376 142 11 . . PUNCT bracis-28376 143 1 full full ADJ bracis-28376 143 2 size size NOUN bracis-28376 143 3 image image NOUN bracis-28376 143 4 pe pe PROPN bracis-28376 143 5 - - PUNCT bracis-28376 143 6 s s PART bracis-28376 143 7 : : PUNCT bracis-28376 143 8 these these PRON bracis-28376 143 9 are be AUX bracis-28376 143 10 models model NOUN bracis-28376 143 11 that that PRON bracis-28376 143 12 rely rely VERB bracis-28376 143 13 on on ADP bracis-28376 143 14 the the DET bracis-28376 143 15 sum sum NOUN bracis-28376 143 16 of of ADP bracis-28376 143 17 sentence sentence NOUN bracis-28376 143 18 embeddings embedding NOUN bracis-28376 143 19 and and CCONJ bracis-28376 143 20 positional positional ADJ bracis-28376 143 21 embeddings embedding NOUN bracis-28376 143 22 as as SCONJ bracis-28376 143 23 proposed propose VERB bracis-28376 143 24 in in ADP bracis-28376 143 25 [ [ X bracis-28376 143 26 12 12 NUM bracis-28376 143 27 ] ] PUNCT bracis-28376 143 28 . . PUNCT bracis-28376 144 1 we we PRON bracis-28376 144 2 exploit exploit VERB bracis-28376 144 3 incaselaw incaselaw NOUN bracis-28376 144 4 and and CCONJ bracis-28376 144 5 roberta roberta PROPN bracis-28376 144 6 as as SCONJ bracis-28376 144 7 sentence sentence NOUN bracis-28376 144 8 encoders encoder NOUN bracis-28376 144 9 and and CCONJ bracis-28376 144 10 sentences sentence NOUN bracis-28376 144 11 are be AUX bracis-28376 144 12 encoded encode VERB bracis-28376 144 13 in in ADP bracis-28376 144 14 isolation isolation NOUN bracis-28376 144 15 . . PUNCT bracis-28376 145 1 pe pe PROPN bracis-28376 145 2 - - PUNCT bracis-28376 145 3 c c NOUN bracis-28376 145 4 : : PUNCT bracis-28376 145 5 like like ADP bracis-28376 145 6 pe pe VERB bracis-28376 145 7 - - PUNCT bracis-28376 145 8 s s PART bracis-28376 145 9 models model NOUN bracis-28376 145 10 , , PUNCT bracis-28376 145 11 but but CCONJ bracis-28376 145 12 we we PRON bracis-28376 145 13 employ employ VERB bracis-28376 145 14 concatenation concatenation NOUN bracis-28376 145 15 instead instead ADV bracis-28376 145 16 of of ADP bracis-28376 145 17 sum sum VERB bracis-28376 145 18 to to PART bracis-28376 145 19 combine combine VERB bracis-28376 145 20 sentence sentence NOUN bracis-28376 145 21 embeddings embedding NOUN bracis-28376 145 22 and and CCONJ bracis-28376 145 23 positional positional ADJ bracis-28376 145 24 embeddings embedding NOUN bracis-28376 145 25 . . PUNCT bracis-28376 146 1 dfcsc dfcsc PROPN bracis-28376 146 2 - - PUNCT bracis-28376 146 3 sep sep PROPN bracis-28376 146 4 : : PUNCT bracis-28376 146 5 these these PRON bracis-28376 146 6 are be AUX bracis-28376 146 7 models model NOUN bracis-28376 146 8 based base VERB bracis-28376 146 9 on on ADP bracis-28376 146 10 the the DET bracis-28376 146 11 dfcsc dfcsc NOUN bracis-28376 146 12 approach approach NOUN bracis-28376 146 13 [ [ X bracis-28376 146 14 11 11 NUM bracis-28376 146 15 ] ] PUNCT bracis-28376 146 16 . . PUNCT bracis-28376 147 1 we we PRON bracis-28376 147 2 employ employ VERB bracis-28376 147 3 incaselaw incaselaw NOUN bracis-28376 147 4 , , PUNCT bracis-28376 147 5 roberta roberta NOUN bracis-28376 147 6 - - PUNCT bracis-28376 147 7 base base NOUN bracis-28376 147 8 , , PUNCT bracis-28376 147 9 and and CCONJ bracis-28376 147 10 longformer longformer NOUN bracis-28376 147 11 - - PUNCT bracis-28376 147 12 base base NOUN bracis-28376 147 13 [ [ X bracis-28376 147 14 2 2 NUM bracis-28376 147 15 ] ] PUNCT bracis-28376 147 16 as as ADP bracis-28376 147 17 sentence sentence NOUN bracis-28376 147 18 encoders encoder NOUN bracis-28376 147 19 . . PUNCT bracis-28376 148 1 longformer longformer NOUN bracis-28376 148 2 is be AUX bracis-28376 148 3 a a DET bracis-28376 148 4 model model NOUN bracis-28376 148 5 that that PRON bracis-28376 148 6 relies rely VERB bracis-28376 148 7 on on ADP bracis-28376 148 8 a a DET bracis-28376 148 9 sparse sparse ADJ bracis-28376 148 10 self self NOUN bracis-28376 148 11 - - PUNCT bracis-28376 148 12 attention attention NOUN bracis-28376 148 13 mechanism mechanism NOUN bracis-28376 148 14 , , PUNCT bracis-28376 148 15 which which PRON bracis-28376 148 16 allows allow VERB bracis-28376 148 17 it it PRON bracis-28376 148 18 to to PART bracis-28376 148 19 handle handle VERB bracis-28376 148 20 longer long ADJ bracis-28376 148 21 sentences sentence NOUN bracis-28376 148 22 than than ADP bracis-28376 148 23 the the DET bracis-28376 148 24 bert bert NOUN bracis-28376 148 25 - - PUNCT bracis-28376 148 26 based base VERB bracis-28376 148 27 models model NOUN bracis-28376 148 28 . . PUNCT bracis-28376 149 1 in in ADP bracis-28376 149 2 all all DET bracis-28376 149 3 models model NOUN bracis-28376 149 4 , , PUNCT bracis-28376 149 5 core core NOUN bracis-28376 149 6 sentences sentence NOUN bracis-28376 149 7 are be AUX bracis-28376 149 8 represented represent VERB bracis-28376 149 9 by by ADP bracis-28376 149 10 the the DET bracis-28376 149 11 last last ADJ bracis-28376 149 12 hidden hide VERB bracis-28376 149 13 states state NOUN bracis-28376 149 14 of of ADP bracis-28376 149 15 the the DET bracis-28376 149 16 respective respective ADJ bracis-28376 149 17 separator separator NOUN bracis-28376 149 18 tokens token NOUN bracis-28376 149 19 ( ( PUNCT bracis-28376 149 20 the the PRON bracis-28376 150 1 [ [ X bracis-28376 150 2 sep sep NOUN bracis-28376 150 3 ] ] X bracis-28376 150 4 or or CCONJ bracis-28376 150 5 \(\texttt \(\texttt NOUN bracis-28376 150 6 { { PUNCT bracis-28376 150 7 < < X bracis-28376 150 8 /s /s X bracis-28376 150 9 > > X bracis-28376 150 10 } } PUNCT bracis-28376 150 11 \)token \)token PROPN bracis-28376 150 12 to to ADP bracis-28376 150 13 the the DET bracis-28376 150 14 right right NOUN bracis-28376 150 15 of of ADP bracis-28376 150 16 a a DET bracis-28376 150 17 core core NOUN bracis-28376 150 18 sentence sentence NOUN bracis-28376 150 19 ) ) PUNCT bracis-28376 150 20 . . PUNCT bracis-28376 151 1 we we PRON bracis-28376 151 2 set set VERB bracis-28376 151 3 \(m_{edges}=250\ \(m_{edges}=250\ NOUN bracis-28376 151 4 ) ) PUNCT bracis-28376 151 5 for for ADP bracis-28376 151 6 the the DET bracis-28376 151 7 three three NUM bracis-28376 151 8 models model NOUN bracis-28376 151 9 , , PUNCT bracis-28376 151 10 \(c_{len}=512\ \(c_{len}=512\ NOUN bracis-28376 151 11 ) ) PUNCT bracis-28376 151 12 for for ADP bracis-28376 151 13 the the DET bracis-28376 151 14 models model NOUN bracis-28376 151 15 based base VERB bracis-28376 151 16 on on ADP bracis-28376 151 17 incaselaw incaselaw NOUN bracis-28376 151 18 and and CCONJ bracis-28376 151 19 roberta roberta PROPN bracis-28376 151 20 , , PUNCT bracis-28376 151 21 and and CCONJ bracis-28376 151 22 \(c_{len}=1,024\ \(c_{len}=1,024\ NOUN bracis-28376 151 23 ) ) PUNCT bracis-28376 151 24 for for ADP bracis-28376 151 25 the the DET bracis-28376 151 26 model model NOUN bracis-28376 151 27 based base VERB bracis-28376 151 28 on on ADP bracis-28376 151 29 longformer longformer PROPN bracis-28376 151 30 . . PUNCT bracis-28376 152 1 dfcsc dfcsc NOUN bracis-28376 152 2 - - PUNCT bracis-28376 152 3 cls cls NOUN bracis-28376 152 4 : : PUNCT bracis-28376 152 5 like like ADP bracis-28376 152 6 dfcsc dfcsc NOUN bracis-28376 152 7 - - PUNCT bracis-28376 152 8 sep sep NOUN bracis-28376 152 9 models model NOUN bracis-28376 152 10 , , PUNCT bracis-28376 152 11 but but CCONJ bracis-28376 152 12 each each DET bracis-28376 152 13 core core NOUN bracis-28376 152 14 sentence sentence NOUN bracis-28376 152 15 is be AUX bracis-28376 152 16 represented represent VERB bracis-28376 152 17 by by ADP bracis-28376 152 18 the the DET bracis-28376 152 19 concatenation concatenation NOUN bracis-28376 152 20 of of ADP bracis-28376 152 21 the the DET bracis-28376 152 22 last last ADJ bracis-28376 152 23 hidden hide VERB bracis-28376 152 24 states state NOUN bracis-28376 152 25 regarding regard VERB bracis-28376 152 26 the the DET bracis-28376 152 27 special special ADJ bracis-28376 152 28 token token ADJ bracis-28376 152 29 marking mark VERB bracis-28376 152 30 the the DET bracis-28376 152 31 beginning beginning NOUN bracis-28376 152 32 of of ADP bracis-28376 152 33 the the DET bracis-28376 152 34 chunk chunk NOUN bracis-28376 152 35 ( ( PUNCT bracis-28376 152 36 [ [ X bracis-28376 152 37 cls cls X bracis-28376 152 38 ] ] PUNCT bracis-28376 152 39 or or CCONJ bracis-28376 152 40 \(\texttt \(\texttt NOUN bracis-28376 152 41 { { PUNCT bracis-28376 152 42 < < X bracis-28376 152 43 s>}\ s>}\ ADJ bracis-28376 152 44 ) ) PUNCT bracis-28376 152 45 ) ) PUNCT bracis-28376 152 46 and and CCONJ bracis-28376 152 47 the the DET bracis-28376 152 48 corresponding correspond VERB bracis-28376 152 49 separator separator NOUN bracis-28376 152 50 token token NOUN bracis-28376 152 51 ( ( PUNCT bracis-28376 152 52 [ [ X bracis-28376 152 53 sep sep X bracis-28376 152 54 ] ] X bracis-28376 152 55 or or CCONJ bracis-28376 152 56 \(\texttt \(\texttt NOUN bracis-28376 152 57 { { PUNCT bracis-28376 152 58 < < X bracis-28376 152 59 /s>}\ /s>}\ X bracis-28376 152 60 ) ) PUNCT bracis-28376 152 61 ) ) PUNCT bracis-28376 152 62 . . PUNCT bracis-28376 153 1 we we PRON bracis-28376 153 2 also also ADV bracis-28376 153 3 implement implement VERB bracis-28376 153 4 the the DET bracis-28376 153 5 following follow VERB bracis-28376 153 6 baselines baseline NOUN bracis-28376 153 7 : : PUNCT bracis-28376 153 8 singlesc singlesc NOUN bracis-28376 153 9 : : PUNCT bracis-28376 153 10 these these PRON bracis-28376 153 11 are be AUX bracis-28376 153 12 single single ADJ bracis-28376 153 13 sentence sentence NOUN bracis-28376 153 14 classification classification NOUN bracis-28376 153 15 models model NOUN bracis-28376 153 16 , , PUNCT bracis-28376 153 17 i.e. i.e. X bracis-28376 153 18 models model NOUN bracis-28376 153 19 that that PRON bracis-28376 153 20 do do AUX bracis-28376 153 21 not not PART bracis-28376 153 22 rely rely VERB bracis-28376 153 23 on on ADP bracis-28376 153 24 chunks chunk NOUN bracis-28376 153 25 . . PUNCT bracis-28376 154 1 each each DET bracis-28376 154 2 sentence sentence NOUN bracis-28376 154 3 is be AUX bracis-28376 154 4 fed feed VERB bracis-28376 154 5 to to ADP bracis-28376 154 6 the the DET bracis-28376 154 7 model model NOUN bracis-28376 154 8 one one NUM bracis-28376 154 9 by by ADP bracis-28376 154 10 one one NUM bracis-28376 154 11 and and CCONJ bracis-28376 154 12 is be AUX bracis-28376 154 13 represented represent VERB bracis-28376 154 14 by by ADP bracis-28376 154 15 the the DET bracis-28376 154 16 hidden hidden ADJ bracis-28376 154 17 state state NOUN bracis-28376 154 18 of of ADP bracis-28376 154 19 the the DET bracis-28376 154 20 token token NOUN bracis-28376 154 21 that that PRON bracis-28376 154 22 marks mark VERB bracis-28376 154 23 the the DET bracis-28376 154 24 beginning beginning NOUN bracis-28376 154 25 of of ADP bracis-28376 154 26 the the DET bracis-28376 154 27 input input NOUN bracis-28376 154 28 ( ( PUNCT bracis-28376 154 29 \(\texttt \(\texttt X bracis-28376 154 30 { { PUNCT bracis-28376 154 31 < < X bracis-28376 154 32 s>}\ s>}\ X bracis-28376 154 33 ) ) PUNCT bracis-28376 154 34 or or CCONJ bracis-28376 154 35 [ [ X bracis-28376 154 36 cls cls NOUN bracis-28376 154 37 ] ] PUNCT bracis-28376 154 38 ) ) PUNCT bracis-28376 154 39 . . PUNCT bracis-28376 155 1 singlesc singlesc NOUN bracis-28376 155 2 models model NOUN bracis-28376 155 3 are be AUX bracis-28376 155 4 the the DET bracis-28376 155 5 reference reference NOUN bracis-28376 155 6 baselines baseline NOUN bracis-28376 155 7 of of ADP bracis-28376 155 8 mixup mixup NOUN bracis-28376 155 9 and and CCONJ bracis-28376 155 10 pe pe PROPN bracis-28376 155 11 models model NOUN bracis-28376 155 12 , , PUNCT bracis-28376 155 13 since since SCONJ bracis-28376 155 14 such such ADJ bracis-28376 155 15 approaches approach NOUN bracis-28376 155 16 do do AUX bracis-28376 155 17 not not PART bracis-28376 155 18 consider consider VERB bracis-28376 155 19 context context NOUN bracis-28376 155 20 to to PART bracis-28376 155 21 encode encode VERB bracis-28376 155 22 sentences sentence NOUN bracis-28376 155 23 . . PUNCT bracis-28376 156 1 the the DET bracis-28376 156 2 exploited exploited ADJ bracis-28376 156 3 pre pre ADJ bracis-28376 156 4 - - ADJ bracis-28376 156 5 trained train VERB bracis-28376 156 6 encoders encoder NOUN bracis-28376 156 7 are be AUX bracis-28376 156 8 incaselaw incaselaw NOUN bracis-28376 156 9 and and CCONJ bracis-28376 156 10 roberta roberta NOUN bracis-28376 156 11 - - PUNCT bracis-28376 156 12 base base NOUN bracis-28376 156 13 . . PUNCT bracis-28376 157 1 cohan cohan PROPN bracis-28376 157 2 : : PUNCT bracis-28376 157 3 these these PRON bracis-28376 157 4 are be AUX bracis-28376 157 5 models model NOUN bracis-28376 157 6 that that PRON bracis-28376 157 7 follows follow VERB bracis-28376 157 8 the the DET bracis-28376 157 9 chunk chunk NOUN bracis-28376 157 10 design design NOUN bracis-28376 157 11 from from ADP bracis-28376 157 12 [ [ X bracis-28376 157 13 4 4 NUM bracis-28376 157 14 ] ] PUNCT bracis-28376 157 15 , , PUNCT bracis-28376 157 16 that that ADV bracis-28376 157 17 is is ADV bracis-28376 157 18 , , PUNCT bracis-28376 157 19 chunks chunk NOUN bracis-28376 157 20 that that PRON bracis-28376 157 21 do do AUX bracis-28376 157 22 not not PART bracis-28376 157 23 share share VERB bracis-28376 157 24 sentences sentence NOUN bracis-28376 157 25 or or CCONJ bracis-28376 157 26 tokens token NOUN bracis-28376 157 27 . . PUNCT bracis-28376 158 1 cohan cohan PROPN bracis-28376 158 2 models model NOUN bracis-28376 158 3 are be AUX bracis-28376 158 4 the the DET bracis-28376 158 5 reference reference NOUN bracis-28376 158 6 baselines baseline NOUN bracis-28376 158 7 of of ADP bracis-28376 158 8 dfcsc dfcsc NOUN bracis-28376 158 9 models model NOUN bracis-28376 158 10 since since SCONJ bracis-28376 158 11 such such ADJ bracis-28376 158 12 approaches approach NOUN bracis-28376 158 13 rely rely VERB bracis-28376 158 14 on on ADP bracis-28376 158 15 chunks chunk NOUN bracis-28376 158 16 to to PART bracis-28376 158 17 encode encode VERB bracis-28376 158 18 sentences sentence NOUN bracis-28376 158 19 . . PUNCT bracis-28376 159 1 the the DET bracis-28376 159 2 exploited exploited ADJ bracis-28376 159 3 pre pre ADJ bracis-28376 159 4 - - ADJ bracis-28376 159 5 trained train VERB bracis-28376 159 6 encoders encoder NOUN bracis-28376 159 7 are be AUX bracis-28376 159 8 incaselaw incaselaw NOUN bracis-28376 159 9 , , PUNCT bracis-28376 159 10 roberta roberta PROPN bracis-28376 159 11 , , PUNCT bracis-28376 159 12 and and CCONJ bracis-28376 159 13 longformer longformer NOUN bracis-28376 159 14 . . PUNCT bracis-28376 160 1 models model NOUN bracis-28376 160 2 based base VERB bracis-28376 160 3 on on ADP bracis-28376 160 4 incaselaw incaselaw NOUN bracis-28376 160 5 and and CCONJ bracis-28376 160 6 roberta roberta PROPN bracis-28376 160 7 have have VERB bracis-28376 160 8 the the DET bracis-28376 160 9 following follow VERB bracis-28376 160 10 hyperparameter hyperparameter NOUN bracis-28376 160 11 values value NOUN bracis-28376 160 12 : : PUNCT bracis-28376 160 13 85 85 NUM bracis-28376 160 14 as as ADP bracis-28376 160 15 the the DET bracis-28376 160 16 maximum maximum ADJ bracis-28376 160 17 sentence sentence NOUN bracis-28376 160 18 length length NOUN bracis-28376 160 19 , , PUNCT bracis-28376 160 20 512 512 NUM bracis-28376 160 21 as as ADP bracis-28376 160 22 the the DET bracis-28376 160 23 chunk chunk NOUN bracis-28376 160 24 length length NOUN bracis-28376 160 25 , , PUNCT bracis-28376 160 26 and and CCONJ bracis-28376 160 27 7 7 NUM bracis-28376 160 28 as as ADP bracis-28376 160 29 the the DET bracis-28376 160 30 maximum maximum ADJ bracis-28376 160 31 number number NOUN bracis-28376 160 32 of of ADP bracis-28376 160 33 sentences sentence NOUN bracis-28376 160 34 in in ADP bracis-28376 160 35 a a DET bracis-28376 160 36 chunk chunk NOUN bracis-28376 160 37 . . PUNCT bracis-28376 161 1 the the DET bracis-28376 161 2 model model NOUN bracis-28376 161 3 based base VERB bracis-28376 161 4 on on ADP bracis-28376 161 5 longformer longformer NOUN bracis-28376 161 6 has have VERB bracis-28376 161 7 the the DET bracis-28376 161 8 following follow VERB bracis-28376 161 9 hyperparameter hyperparameter NOUN bracis-28376 161 10 values value NOUN bracis-28376 161 11 : : PUNCT bracis-28376 161 12 85 85 NUM bracis-28376 161 13 as as ADP bracis-28376 161 14 the the DET bracis-28376 161 15 maximum maximum ADJ bracis-28376 161 16 sentence sentence NOUN bracis-28376 161 17 length length NOUN bracis-28376 161 18 , , PUNCT bracis-28376 161 19 1,024 1,024 NUM bracis-28376 161 20 as as ADP bracis-28376 161 21 the the DET bracis-28376 161 22 chunk chunk NOUN bracis-28376 161 23 length length NOUN bracis-28376 161 24 , , PUNCT bracis-28376 161 25 and and CCONJ bracis-28376 161 26 14 14 NUM bracis-28376 161 27 as as ADP bracis-28376 161 28 the the DET bracis-28376 161 29 maximum maximum ADJ bracis-28376 161 30 number number NOUN bracis-28376 161 31 of of ADP bracis-28376 161 32 sentences sentence NOUN bracis-28376 161 33 in in ADP bracis-28376 161 34 a a DET bracis-28376 161 35 chunk chunk NOUN bracis-28376 161 36 . . PUNCT bracis-28376 162 1 each each DET bracis-28376 162 2 sentence sentence NOUN bracis-28376 162 3 is be AUX bracis-28376 162 4 represented represent VERB bracis-28376 162 5 by by ADP bracis-28376 162 6 the the DET bracis-28376 162 7 hidden hidden ADJ bracis-28376 162 8 state state NOUN bracis-28376 162 9 of of ADP bracis-28376 162 10 the the DET bracis-28376 162 11 following follow VERB bracis-28376 162 12 separator separator NOUN bracis-28376 162 13 token token VERB bracis-28376 162 14 . . PUNCT bracis-28376 163 1 all all DET bracis-28376 163 2 models model NOUN bracis-28376 163 3 use use VERB bracis-28376 163 4 a a DET bracis-28376 163 5 single single ADJ bracis-28376 163 6 fully fully ADV bracis-28376 163 7 - - PUNCT bracis-28376 163 8 connected connect VERB bracis-28376 163 9 layer layer NOUN bracis-28376 163 10 as as ADP bracis-28376 163 11 their their PRON bracis-28376 163 12 classifier classifier NOUN bracis-28376 163 13 . . PUNCT bracis-28376 164 1 except except SCONJ bracis-28376 164 2 for for ADP bracis-28376 164 3 the the DET bracis-28376 164 4 mixup mixup NOUN bracis-28376 164 5 - - PUNCT bracis-28376 164 6 a a DET bracis-28376 164 7 models model NOUN bracis-28376 164 8 , , PUNCT bracis-28376 164 9 the the DET bracis-28376 164 10 encoder encoder NOUN bracis-28376 164 11 is be AUX bracis-28376 164 12 fine fine ADV bracis-28376 164 13 - - PUNCT bracis-28376 164 14 tuned tune VERB bracis-28376 164 15 together together ADV bracis-28376 164 16 with with ADP bracis-28376 164 17 the the DET bracis-28376 164 18 classifier classifier NOUN bracis-28376 164 19 during during ADP bracis-28376 164 20 the the DET bracis-28376 164 21 training training NOUN bracis-28376 164 22 of of ADP bracis-28376 164 23 the the DET bracis-28376 164 24 respective respective ADJ bracis-28376 164 25 model model NOUN bracis-28376 164 26 . . PUNCT bracis-28376 165 1 for for ADP bracis-28376 165 2 each each DET bracis-28376 165 3 mixup mixup NOUN bracis-28376 165 4 - - PUNCT bracis-28376 165 5 a a DET bracis-28376 165 6 model model NOUN bracis-28376 165 7 , , PUNCT bracis-28376 165 8 the the DET bracis-28376 165 9 transformer transformer NOUN bracis-28376 165 10 encoder encoder NOUN bracis-28376 165 11 is be AUX bracis-28376 165 12 tuned tune VERB bracis-28376 165 13 with with ADP bracis-28376 165 14 the the DET bracis-28376 165 15 same same ADJ bracis-28376 165 16 hyperparameters hyperparameter NOUN bracis-28376 165 17 as as ADP bracis-28376 165 18 its its PRON bracis-28376 165 19 singlesc singlesc NOUN bracis-28376 165 20 counterpart counterpart NOUN bracis-28376 165 21 ( ( PUNCT bracis-28376 165 22 e.g. e.g. ADV bracis-28376 165 23 , , PUNCT bracis-28376 165 24 singlesc singlesc NOUN bracis-28376 165 25 - - PUNCT bracis-28376 165 26 incaselaw incaselaw NOUN bracis-28376 165 27 and and CCONJ bracis-28376 165 28 mixup mixup NOUN bracis-28376 165 29 - - PUNCT bracis-28376 165 30 a a DET bracis-28376 165 31 - - PUNCT bracis-28376 165 32 incaselaw incaselaw NOUN bracis-28376 165 33 ) ) PUNCT bracis-28376 165 34 . . PUNCT bracis-28376 166 1 for for ADP bracis-28376 166 2 their their PRON bracis-28376 166 3 classifiers classifier NOUN bracis-28376 166 4 , , PUNCT bracis-28376 166 5 we we PRON bracis-28376 166 6 adopt adopt VERB bracis-28376 166 7 200 200 NUM bracis-28376 166 8 training training NOUN bracis-28376 166 9 epochs epoch NOUN bracis-28376 166 10 , , PUNCT bracis-28376 166 11 \(10^{-2}\ \(10^{-2}\ NOUN bracis-28376 166 12 ) ) PUNCT bracis-28376 166 13 as as ADP bracis-28376 166 14 initial initial ADJ bracis-28376 166 15 learning learning NOUN bracis-28376 166 16 rate rate NOUN bracis-28376 166 17 , , PUNCT bracis-28376 166 18 and and CCONJ bracis-28376 166 19 0.2 0.2 NUM bracis-28376 166 20 as as ADP bracis-28376 166 21 dropout dropout NOUN bracis-28376 166 22 rate rate NOUN bracis-28376 166 23 . . PUNCT bracis-28376 167 1 the the DET bracis-28376 167 2 remaining remain VERB bracis-28376 167 3 models model NOUN bracis-28376 167 4 are be AUX bracis-28376 167 5 fine fine ADV bracis-28376 167 6 - - PUNCT bracis-28376 167 7 tuned tune VERB bracis-28376 167 8 for for ADP bracis-28376 167 9 four four NUM bracis-28376 167 10 epochs epoch NOUN bracis-28376 167 11 , , PUNCT bracis-28376 167 12 and and CCONJ bracis-28376 167 13 we we PRON bracis-28376 167 14 adopt adopt VERB bracis-28376 167 15 \(10^{-5}\ \(10^{-5}\ VERB bracis-28376 167 16 ) ) PUNCT bracis-28376 167 17 as as ADP bracis-28376 167 18 initial initial ADJ bracis-28376 167 19 learning learning NOUN bracis-28376 167 20 rate rate NOUN bracis-28376 167 21 and and CCONJ bracis-28376 167 22 0.2 0.2 NUM bracis-28376 167 23 as as ADP bracis-28376 167 24 dropout dropout NOUN bracis-28376 167 25 rate rate NOUN bracis-28376 167 26 of of ADP bracis-28376 167 27 the the DET bracis-28376 167 28 classification classification NOUN bracis-28376 167 29 layer layer NOUN bracis-28376 167 30 . . PUNCT bracis-28376 168 1 the the DET bracis-28376 168 2 pre pre ADJ bracis-28376 168 3 - - ADJ bracis-28376 168 4 trained train VERB bracis-28376 168 5 transformer transformer NOUN bracis-28376 168 6 models model NOUN bracis-28376 168 7 are be AUX bracis-28376 168 8 used use VERB bracis-28376 168 9 with with ADP bracis-28376 168 10 default default NOUN bracis-28376 168 11 parameters parameter NOUN bracis-28376 168 12 , , PUNCT bracis-28376 168 13 except except SCONJ bracis-28376 168 14 for for ADP bracis-28376 168 15 the the DET bracis-28376 168 16 longformer longformer NOUN bracis-28376 168 17 , , PUNCT bracis-28376 168 18 whose whose DET bracis-28376 168 19 global global ADJ bracis-28376 168 20 attention attention NOUN bracis-28376 168 21 is be AUX bracis-28376 168 22 set set VERB bracis-28376 168 23 to to PART bracis-28376 168 24 consider consider VERB bracis-28376 168 25 \(\texttt \(\texttt NOUN bracis-28376 168 26 { { PUNCT bracis-28376 168 27 < < X bracis-28376 168 28 s>}\ s>}\ X bracis-28376 168 29 ) ) PUNCT bracis-28376 168 30 and and CCONJ bracis-28376 168 31 \(\texttt \(\texttt NUM bracis-28376 168 32 { { PUNCT bracis-28376 168 33 < < NOUN bracis-28376 168 34 /s>}\ /s>}\ ADJ bracis-28376 168 35 ) ) PUNCT bracis-28376 168 36 tokens token NOUN bracis-28376 168 37 . . PUNCT bracis-28376 169 1 for for ADP bracis-28376 169 2 all all DET bracis-28376 169 3 models model NOUN bracis-28376 169 4 we we PRON bracis-28376 169 5 employ employ VERB bracis-28376 169 6 the the DET bracis-28376 169 7 cross cross NOUN bracis-28376 169 8 entropy entropy PROPN bracis-28376 169 9 loss loss NOUN bracis-28376 169 10 function function NOUN bracis-28376 169 11 and and CCONJ bracis-28376 169 12 the the DET bracis-28376 169 13 adam adam PROPN bracis-28376 169 14 optimizer optimizer NOUN bracis-28376 169 15 ( ( PUNCT bracis-28376 169 16 \(\beta \(\beta X bracis-28376 169 17 _ _ NOUN bracis-28376 169 18 1=0.9\ 1=0.9\ NUM bracis-28376 169 19 ) ) PUNCT bracis-28376 169 20 , , PUNCT bracis-28376 169 21 \(\beta \(\beta X bracis-28376 169 22 _ _ NOUN bracis-28376 169 23 2=0.999\ 2=0.999\ NUM bracis-28376 169 24 ) ) PUNCT bracis-28376 169 25 , , PUNCT bracis-28376 169 26 \(\epsilon \(\epsilon NOUN bracis-28376 169 27 = = NOUN bracis-28376 169 28 10^{-8}\ 10^{-8}\ NUM bracis-28376 169 29 ) ) PUNCT bracis-28376 169 30 , , PUNCT bracis-28376 169 31 \(weight\_decay=10^{-3}\ \(weight\_decay=10^{-3}\ NUM bracis-28376 169 32 ) ) PUNCT bracis-28376 169 33 ) ) PUNCT bracis-28376 169 34 . . PUNCT bracis-28376 170 1 the the DET bracis-28376 170 2 learning learning NOUN bracis-28376 170 3 rate rate NOUN bracis-28376 170 4 is be AUX bracis-28376 170 5 scheduled schedule VERB bracis-28376 170 6 linearly linearly ADV bracis-28376 170 7 from from ADP bracis-28376 170 8 an an DET bracis-28376 170 9 initial initial ADJ bracis-28376 170 10 value value NOUN bracis-28376 170 11 to to ADP bracis-28376 170 12 zero zero NUM bracis-28376 170 13 , , PUNCT bracis-28376 170 14 and and CCONJ bracis-28376 170 15 there there PRON bracis-28376 170 16 are be VERB bracis-28376 170 17 no no DET bracis-28376 170 18 warm warm ADJ bracis-28376 170 19 - - PUNCT bracis-28376 170 20 up up ADP bracis-28376 170 21 steps step NOUN bracis-28376 170 22 . . PUNCT bracis-28376 171 1 we we PRON bracis-28376 171 2 adopt adopt VERB bracis-28376 171 3 a a DET bracis-28376 171 4 batch batch NOUN bracis-28376 171 5 size size NOUN bracis-28376 171 6 of of ADP bracis-28376 171 7 4 4 NUM bracis-28376 171 8 for for ADP bracis-28376 171 9 longformer longformer NOUN bracis-28376 171 10 - - PUNCT bracis-28376 171 11 based base VERB bracis-28376 171 12 models model NOUN bracis-28376 171 13 and and CCONJ bracis-28376 171 14 a a DET bracis-28376 171 15 batch batch NOUN bracis-28376 171 16 size size NOUN bracis-28376 171 17 of of ADP bracis-28376 171 18 16 16 NUM bracis-28376 171 19 for for ADP bracis-28376 171 20 the the DET bracis-28376 171 21 other other ADJ bracis-28376 171 22 models model NOUN bracis-28376 171 23 . . PUNCT bracis-28376 172 1 python python NOUN bracis-28376 172 2 3.8.13 3.8.13 NUM bracis-28376 172 3 , , PUNCT bracis-28376 172 4 pytorch pytorch NOUN bracis-28376 172 5 1.13.0 1.13.0 NUM bracis-28376 172 6 , , PUNCT bracis-28376 172 7 and and CCONJ bracis-28376 172 8 hugging hug VERB bracis-28376 172 9 face face NOUN bracis-28376 172 10 transformer transformer NOUN bracis-28376 172 11 4.28.1 4.28.1 NUM bracis-28376 172 12 are be AUX bracis-28376 172 13 the the DET bracis-28376 172 14 programming programming NOUN bracis-28376 172 15 language language NOUN bracis-28376 172 16 and and CCONJ bracis-28376 172 17 main main ADJ bracis-28376 172 18 libraries library NOUN bracis-28376 172 19 we we PRON bracis-28376 172 20 use use VERB bracis-28376 172 21 to to PART bracis-28376 172 22 develop develop VERB bracis-28376 172 23 all all DET bracis-28376 172 24 models model NOUN bracis-28376 172 25 . . PUNCT bracis-28376 173 1 hugging hug VERB bracis-28376 173 2 face face NOUN bracis-28376 173 3 is be AUX bracis-28376 173 4 mainly mainly ADV bracis-28376 173 5 used use VERB bracis-28376 173 6 to to PART bracis-28376 173 7 load load VERB bracis-28376 173 8 pre pre ADJ bracis-28376 173 9 - - ADJ bracis-28376 173 10 trained train VERB bracis-28376 173 11 transformer transformer NOUN bracis-28376 173 12 models model NOUN bracis-28376 173 13 . . PUNCT bracis-28376 174 1 a a DET bracis-28376 174 2 single single ADJ bracis-28376 174 3 16 16 NUM bracis-28376 174 4 gb gb NOUN bracis-28376 174 5 tesla tesla PROPN bracis-28376 174 6 v100 v100 PROPN bracis-28376 174 7 sxm2 sxm2 PROPN bracis-28376 174 8 gpu gpu PROPN bracis-28376 174 9 is be AUX bracis-28376 174 10 used use VERB bracis-28376 174 11 in in ADP bracis-28376 174 12 all all DET bracis-28376 174 13 experiments experiment NOUN bracis-28376 174 14 . . PUNCT bracis-28376 175 1 3.3 3.3 NUM bracis-28376 175 2 evaluation evaluation NOUN bracis-28376 175 3 all all DET bracis-28376 175 4 models model NOUN bracis-28376 175 5 are be AUX bracis-28376 175 6 trained train VERB bracis-28376 175 7 on on ADP bracis-28376 175 8 the the DET bracis-28376 175 9 training training NOUN bracis-28376 175 10 set set NOUN bracis-28376 175 11 and and CCONJ bracis-28376 175 12 evaluated evaluate VERB bracis-28376 175 13 on on ADP bracis-28376 175 14 the the DET bracis-28376 175 15 test test NOUN bracis-28376 175 16 set set VERB bracis-28376 175 17 . . PUNCT bracis-28376 176 1 for for ADP bracis-28376 176 2 each each DET bracis-28376 176 3 model model NOUN bracis-28376 176 4 , , PUNCT bracis-28376 176 5 the the DET bracis-28376 176 6 training training NOUN bracis-28376 176 7 and and CCONJ bracis-28376 176 8 evaluation evaluation NOUN bracis-28376 176 9 procedure procedure NOUN bracis-28376 176 10 is be AUX bracis-28376 176 11 repeated repeat VERB bracis-28376 176 12 five five NUM bracis-28376 176 13 times time NOUN bracis-28376 176 14 with with ADP bracis-28376 176 15 a a DET bracis-28376 176 16 different different ADJ bracis-28376 176 17 random random ADJ bracis-28376 176 18 seed seed NOUN bracis-28376 176 19 for for ADP bracis-28376 176 20 each each DET bracis-28376 176 21 repetition repetition NOUN bracis-28376 176 22 ( ( PUNCT bracis-28376 176 23 the the DET bracis-28376 176 24 same same ADJ bracis-28376 176 25 five five NUM bracis-28376 176 26 seeds seed NOUN bracis-28376 176 27 for for ADP bracis-28376 176 28 all all DET bracis-28376 176 29 models model NOUN bracis-28376 176 30 ) ) PUNCT bracis-28376 176 31 . . PUNCT bracis-28376 177 1 the the DET bracis-28376 177 2 evaluation evaluation NOUN bracis-28376 177 3 metric metric NOUN bracis-28376 177 4 is be AUX bracis-28376 177 5 the the DET bracis-28376 177 6 macro macro ADJ bracis-28376 177 7 f1 f1 PROPN bracis-28376 177 8 score score NOUN bracis-28376 177 9 . . PUNCT bracis-28376 178 1 to to PART bracis-28376 178 2 compare compare VERB bracis-28376 178 3 the the DET bracis-28376 178 4 results result NOUN bracis-28376 178 5 obtained obtain VERB bracis-28376 178 6 by by ADP bracis-28376 178 7 the the DET bracis-28376 178 8 implemented implement VERB bracis-28376 178 9 models model NOUN bracis-28376 178 10 , , PUNCT bracis-28376 178 11 we we PRON bracis-28376 178 12 use use VERB bracis-28376 178 13 the the DET bracis-28376 178 14 almost almost ADV bracis-28376 178 15 stochastic stochastic ADJ bracis-28376 178 16 order order NOUN bracis-28376 178 17 ( ( PUNCT bracis-28376 178 18 aso aso NOUN bracis-28376 178 19 ) ) PUNCT bracis-28376 178 20 test test NOUN bracis-28376 179 1 [ [ X bracis-28376 179 2 6 6 NUM bracis-28376 179 3 ] ] PUNCT bracis-28376 179 4 with with ADP bracis-28376 179 5 bonferroni bonferroni NOUN bracis-28376 179 6 correction correction NOUN bracis-28376 179 7 as as SCONJ bracis-28376 179 8 implemented implement VERB bracis-28376 179 9 by by ADP bracis-28376 179 10 [ [ X bracis-28376 179 11 19 19 NUM bracis-28376 179 12 ] ] PUNCT bracis-28376 179 13 . . PUNCT bracis-28376 180 1 aso aso PROPN bracis-28376 180 2 is be AUX bracis-28376 180 3 a a DET bracis-28376 180 4 statistical statistical ADJ bracis-28376 180 5 significance significance NOUN bracis-28376 180 6 test test NOUN bracis-28376 180 7 designed design VERB bracis-28376 180 8 to to PART bracis-28376 180 9 compare compare VERB bracis-28376 180 10 two two NUM bracis-28376 180 11 score score NOUN bracis-28376 180 12 distributions distribution NOUN bracis-28376 180 13 produced produce VERB bracis-28376 180 14 by by ADP bracis-28376 180 15 deep deep ADJ bracis-28376 180 16 learning learning NOUN bracis-28376 180 17 models model NOUN bracis-28376 180 18 . . PUNCT bracis-28376 181 1 the the DET bracis-28376 181 2 test test NOUN bracis-28376 181 3 analyzes analyze VERB bracis-28376 181 4 two two NUM bracis-28376 181 5 distributions distribution NOUN bracis-28376 181 6 and and CCONJ bracis-28376 181 7 quantifies quantify VERB bracis-28376 181 8 the the DET bracis-28376 181 9 amount amount NOUN bracis-28376 181 10 of of ADP bracis-28376 181 11 stochastic stochastic ADJ bracis-28376 181 12 order order NOUN bracis-28376 181 13 violation violation NOUN bracis-28376 181 14 by by ADP bracis-28376 181 15 computing compute VERB bracis-28376 181 16 a a DET bracis-28376 181 17 value value NOUN bracis-28376 181 18 \(\epsilon \(\epsilon NOUN bracis-28376 181 19 _ _ NOUN bracis-28376 181 20 { { PUNCT bracis-28376 181 21 min}\ min}\ NOUN bracis-28376 181 22 ) ) PUNCT bracis-28376 181 23 . . PUNCT bracis-28376 182 1 given give VERB bracis-28376 182 2 the the DET bracis-28376 182 3 scores score NOUN bracis-28376 182 4 over over ADP bracis-28376 182 5 multiple multiple ADJ bracis-28376 182 6 random random ADJ bracis-28376 182 7 seeds seed NOUN bracis-28376 182 8 from from ADP bracis-28376 182 9 two two NUM bracis-28376 182 10 models model NOUN bracis-28376 182 11 a a PRON bracis-28376 182 12 and and CCONJ bracis-28376 182 13 b b NOUN bracis-28376 182 14 , , PUNCT bracis-28376 182 15 a a PRON bracis-28376 182 16 is be AUX bracis-28376 182 17 declared declare VERB bracis-28376 182 18 superior superior ADJ bracis-28376 182 19 if if SCONJ bracis-28376 182 20 \(\epsilon \(\epsilon NOUN bracis-28376 182 21 _ _ PRON bracis-28376 182 22 { { PUNCT bracis-28376 182 23 min min NOUN bracis-28376 182 24 } } PUNCT bracis-28376 182 25 < < X bracis-28376 182 26 \tau \tau PROPN bracis-28376 182 27 \ \ PROPN bracis-28376 182 28 ) ) PUNCT bracis-28376 182 29 , , PUNCT bracis-28376 182 30 where where SCONJ bracis-28376 182 31 \(\tau \(\tau ADV bracis-28376 182 32 \ \ NOUN bracis-28376 182 33 ) ) PUNCT bracis-28376 182 34 is be AUX bracis-28376 182 35 0.5 0.5 NUM bracis-28376 182 36 or or CCONJ bracis-28376 182 37 less less ADJ bracis-28376 182 38 ( ( PUNCT bracis-28376 182 39 or or CCONJ bracis-28376 182 40 more more ADV bracis-28376 182 41 precisely precisely ADV bracis-28376 182 42 , , PUNCT bracis-28376 182 43 we we PRON bracis-28376 182 44 confirm confirm VERB bracis-28376 182 45 that that SCONJ bracis-28376 182 46 a a PRON bracis-28376 182 47 is be AUX bracis-28376 182 48 stochastically stochastically ADV bracis-28376 182 49 dominant dominant ADJ bracis-28376 182 50 over over ADP bracis-28376 182 51 b b NOUN bracis-28376 182 52 in in ADP bracis-28376 182 53 more more ADJ bracis-28376 182 54 cases case NOUN bracis-28376 182 55 than than SCONJ bracis-28376 182 56 vice vice NOUN bracis-28376 182 57 versa versa ADV bracis-28376 182 58 ) ) PUNCT bracis-28376 182 59 . . PUNCT bracis-28376 183 1 we we PRON bracis-28376 183 2 can can AUX bracis-28376 183 3 also also ADV bracis-28376 183 4 interpret interpret VERB bracis-28376 183 5 \(\epsilon \(\epsilon NOUN bracis-28376 183 6 _ _ NOUN bracis-28376 183 7 { { PUNCT bracis-28376 183 8 min}\ min}\ X bracis-28376 183 9 ) ) PUNCT bracis-28376 183 10 as as ADP bracis-28376 183 11 a a DET bracis-28376 183 12 confidence confidence NOUN bracis-28376 183 13 score score NOUN bracis-28376 183 14 : : PUNCT bracis-28376 183 15 the the PRON bracis-28376 183 16 lower low ADJ bracis-28376 183 17 it it PRON bracis-28376 183 18 is be AUX bracis-28376 183 19 , , PUNCT bracis-28376 183 20 the the PRON bracis-28376 183 21 stronger strong ADJ bracis-28376 183 22 the the DET bracis-28376 183 23 evidence evidence NOUN bracis-28376 183 24 that that SCONJ bracis-28376 183 25 a a PRON bracis-28376 183 26 is be AUX bracis-28376 183 27 superior superior ADJ bracis-28376 183 28 to to PART bracis-28376 183 29 b. b. VERB bracis-28376 183 30 in in ADP bracis-28376 183 31 this this DET bracis-28376 183 32 context context NOUN bracis-28376 183 33 , , PUNCT bracis-28376 183 34 and and CCONJ bracis-28376 183 35 following follow VERB bracis-28376 183 36 the the DET bracis-28376 183 37 recommendation recommendation NOUN bracis-28376 183 38 of of ADP bracis-28376 183 39 \(\tau \(\tau PROPN bracis-28376 183 40 = = SYM bracis-28376 183 41 0.2\ 0.2\ ADJ bracis-28376 183 42 ) ) PUNCT bracis-28376 183 43 for for ADP bracis-28376 183 44 multiple multiple ADJ bracis-28376 183 45 comparisons comparison NOUN bracis-28376 183 46 , , PUNCT bracis-28376 183 47 we we PRON bracis-28376 183 48 formulate formulate VERB bracis-28376 183 49 our our PRON bracis-28376 183 50 hypotheses hypothesis NOUN bracis-28376 183 51 as as ADP bracis-28376 183 52 $ $ SYM bracis-28376 183 53 $ $ SYM bracis-28376 183 54 \begin{aligned \begin{aligne VERB bracis-28376 183 55 } } PUNCT bracis-28376 183 56 & & CCONJ bracis-28376 183 57 h_0 h_0 PROPN bracis-28376 183 58 : : PUNCT bracis-28376 183 59 \epsilon \epsilon NOUN bracis-28376 183 60 _ _ PUNCT bracis-28376 183 61 { { PUNCT bracis-28376 183 62 min min NOUN bracis-28376 183 63 } } PUNCT bracis-28376 183 64 \ge \ge PROPN bracis-28376 183 65 0.2 0.2 NUM bracis-28376 183 66 { { PUNCT bracis-28376 183 67 } } PUNCT bracis-28376 183 68 & & CCONJ bracis-28376 183 69 { { PUNCT bracis-28376 183 70 } } PUNCT bracis-28376 183 71 \text \text ADV bracis-28376 183 72 { { PUNCT bracis-28376 183 73 ( ( PUNCT bracis-28376 183 74 { { PUNCT bracis-28376 183 75 a a PRON bracis-28376 183 76 } } PUNCT bracis-28376 183 77 is be AUX bracis-28376 183 78 not not PART bracis-28376 183 79 superior superior ADJ bracis-28376 183 80 to to ADP bracis-28376 183 81 { { PUNCT bracis-28376 183 82 b b NOUN bracis-28376 183 83 } } PUNCT bracis-28376 183 84 ) ) PUNCT bracis-28376 183 85 } } PUNCT bracis-28376 183 86 \\ \\ NOUN bracis-28376 183 87 & & CCONJ bracis-28376 183 88 h_1 h_1 PROPN bracis-28376 183 89 : : PUNCT bracis-28376 183 90 \epsilon \epsilon NOUN bracis-28376 183 91 _ _ PUNCT bracis-28376 183 92 { { PUNCT bracis-28376 183 93 min min NOUN bracis-28376 183 94 } } PUNCT bracis-28376 183 95 < < X bracis-28376 183 96 0.2 0.2 NUM bracis-28376 183 97 { { PUNCT bracis-28376 183 98 } } PUNCT bracis-28376 183 99 & & CCONJ bracis-28376 183 100 { { PUNCT bracis-28376 183 101 } } PUNCT bracis-28376 183 102 \text \text ADV bracis-28376 183 103 { { PUNCT bracis-28376 183 104 ( ( PUNCT bracis-28376 183 105 { { PUNCT bracis-28376 183 106 a a PRON bracis-28376 183 107 } } PUNCT bracis-28376 183 108 is be AUX bracis-28376 183 109 superior superior ADJ bracis-28376 183 110 to to ADP bracis-28376 183 111 { { PUNCT bracis-28376 183 112 b b NOUN bracis-28376 183 113 } } PUNCT bracis-28376 183 114 ) ) PUNCT bracis-28376 183 115 } } PUNCT bracis-28376 183 116 \end{aligned}$$ \end{aligned}$$ ADP bracis-28376 183 117 in in ADP bracis-28376 183 118 our our PRON bracis-28376 183 119 experiments experiment NOUN bracis-28376 183 120 , , PUNCT bracis-28376 183 121 we we PRON bracis-28376 183 122 compare compare VERB bracis-28376 183 123 all all DET bracis-28376 183 124 pairs pair NOUN bracis-28376 183 125 of of ADP bracis-28376 183 126 models model NOUN bracis-28376 183 127 based base VERB bracis-28376 183 128 on on ADP bracis-28376 183 129 five five NUM bracis-28376 183 130 random random ADJ bracis-28376 183 131 seeds seed NOUN bracis-28376 183 132 each each PRON bracis-28376 183 133 using use VERB bracis-28376 183 134 aso aso NOUN bracis-28376 183 135 with with ADP bracis-28376 183 136 a a DET bracis-28376 183 137 confidence confidence NOUN bracis-28376 183 138 level level NOUN bracis-28376 183 139 of of ADP bracis-28376 183 140 \(\alpha \(\alpha NOUN bracis-28376 183 141 = = SYM bracis-28376 183 142 0.05\ 0.05\ NUM bracis-28376 183 143 ) ) PUNCT bracis-28376 183 144 ( ( PUNCT bracis-28376 183 145 before before ADP bracis-28376 183 146 adjusting adjust VERB bracis-28376 183 147 for for ADP bracis-28376 183 148 all all DET bracis-28376 183 149 pairwise pairwise NOUN bracis-28376 183 150 comparisons comparison NOUN bracis-28376 183 151 using use VERB bracis-28376 183 152 the the DET bracis-28376 183 153 bonferroni bonferroni NOUN bracis-28376 183 154 correction correction NOUN bracis-28376 183 155 ) ) PUNCT bracis-28376 183 156 . . PUNCT bracis-28376 184 1 we we PRON bracis-28376 184 2 only only ADV bracis-28376 184 3 compare compare VERB bracis-28376 184 4 scores score NOUN bracis-28376 184 5 obtained obtain VERB bracis-28376 184 6 on on ADP bracis-28376 184 7 the the DET bracis-28376 184 8 same same ADJ bracis-28376 184 9 dataset dataset NOUN bracis-28376 184 10 . . PUNCT bracis-28376 185 1 table table NOUN bracis-28376 185 2 1 1 NUM bracis-28376 185 3 . . PUNCT bracis-28376 186 1 results result NOUN bracis-28376 186 2 achieved achieve VERB bracis-28376 186 3 by by ADP bracis-28376 186 4 the the DET bracis-28376 186 5 baseline baseline NOUN bracis-28376 186 6 models model NOUN bracis-28376 186 7 and and CCONJ bracis-28376 186 8 the the DET bracis-28376 186 9 models model NOUN bracis-28376 186 10 of of ADP bracis-28376 186 11 interest interest NOUN bracis-28376 186 12 . . PUNCT bracis-28376 187 1 the the DET bracis-28376 187 2 f1 f1 PROPN bracis-28376 187 3 scores score NOUN bracis-28376 187 4 are be AUX bracis-28376 187 5 averages average NOUN bracis-28376 187 6 of of ADP bracis-28376 187 7 five five NUM bracis-28376 187 8 evaluations evaluation NOUN bracis-28376 187 9 and and CCONJ bracis-28376 187 10 refer refer VERB bracis-28376 187 11 to to ADP bracis-28376 187 12 the the DET bracis-28376 187 13 best good ADJ bracis-28376 187 14 fine fine ADJ bracis-28376 187 15 - - PUNCT bracis-28376 187 16 tuning tune VERB bracis-28376 187 17 epoch epoch NOUN bracis-28376 187 18 ( ( PUNCT bracis-28376 187 19 i.e. i.e. X bracis-28376 187 20 , , PUNCT bracis-28376 187 21 the the DET bracis-28376 187 22 epoch epoch NOUN bracis-28376 187 23 that that PRON bracis-28376 187 24 gives give VERB bracis-28376 187 25 the the DET bracis-28376 187 26 highest high ADJ bracis-28376 187 27 f1 f1 NOUN bracis-28376 187 28 score score NOUN bracis-28376 187 29 ) ) PUNCT bracis-28376 187 30 . . PUNCT bracis-28376 188 1 the the DET bracis-28376 188 2 epoch epoch PROPN bracis-28376 188 3 column column NOUN bracis-28376 188 4 reports report VERB bracis-28376 188 5 the the DET bracis-28376 188 6 best good ADJ bracis-28376 188 7 fine fine ADJ bracis-28376 188 8 - - PUNCT bracis-28376 188 9 tuning tune VERB bracis-28376 188 10 epoch epoch NOUN bracis-28376 188 11 ( ( PUNCT bracis-28376 188 12 e.g. e.g. ADV bracis-28376 188 13 , , PUNCT bracis-28376 188 14 1/4 1/4 NUM bracis-28376 188 15 indicates indicate VERB bracis-28376 188 16 the the DET bracis-28376 188 17 1st 1st ADJ bracis-28376 188 18 epoch epoch NOUN bracis-28376 188 19 out out ADP bracis-28376 188 20 of of ADP bracis-28376 188 21 a a DET bracis-28376 188 22 total total NOUN bracis-28376 188 23 of of ADP bracis-28376 188 24 4 4 NUM bracis-28376 188 25 epochs epoch NOUN bracis-28376 188 26 ) ) PUNCT bracis-28376 188 27 . . PUNCT bracis-28376 189 1 for for ADP bracis-28376 189 2 mixup mixup NOUN bracis-28376 189 3 - - PUNCT bracis-28376 189 4 a a DET bracis-28376 189 5 models model NOUN bracis-28376 189 6 , , PUNCT bracis-28376 189 7 we we PRON bracis-28376 189 8 report report VERB bracis-28376 189 9 the the DET bracis-28376 189 10 encoder encoder NOUN bracis-28376 189 11 fine fine ADV bracis-28376 189 12 - - PUNCT bracis-28376 189 13 tuning tune VERB bracis-28376 189 14 epoch epoch NOUN bracis-28376 189 15 followed follow VERB bracis-28376 189 16 by by ADP bracis-28376 189 17 the the DET bracis-28376 189 18 mixup mixup NOUN bracis-28376 189 19 - - PUNCT bracis-28376 189 20 a a DET bracis-28376 189 21 classifier classifier NOUN bracis-28376 189 22 training training NOUN bracis-28376 189 23 epochs epoch NOUN bracis-28376 189 24 . . PUNCT bracis-28376 190 1 \(\sigma \(\sigma NOUN bracis-28376 190 2 \ \ NOUN bracis-28376 190 3 ) ) PUNCT bracis-28376 190 4 denotes denote VERB bracis-28376 190 5 the the DET bracis-28376 190 6 standard standard ADJ bracis-28376 190 7 deviation deviation NOUN bracis-28376 190 8 . . PUNCT bracis-28376 191 1 the the DET bracis-28376 191 2 scores score NOUN bracis-28376 191 3 in in ADP bracis-28376 191 4 bold bold ADJ bracis-28376 191 5 are be AUX bracis-28376 191 6 the the DET bracis-28376 191 7 best good ADJ bracis-28376 191 8 for for ADP bracis-28376 191 9 each each DET bracis-28376 191 10 dataset.full dataset.full PROPN bracis-28376 191 11 size size NOUN bracis-28376 191 12 table table NOUN bracis-28376 191 13 4 4 NUM bracis-28376 191 14 results result NOUN bracis-28376 191 15 and and CCONJ bracis-28376 191 16     SPACE bracis-28376 191 17 discussion discussion NOUN bracis-28376 191 18 table table NOUN bracis-28376 191 19     SPACE bracis-28376 191 20 1 1 NUM bracis-28376 191 21 presents present VERB bracis-28376 191 22 the the DET bracis-28376 191 23 results result NOUN bracis-28376 191 24 achieved achieve VERB bracis-28376 191 25 by by ADP bracis-28376 191 26 the the DET bracis-28376 191 27 models model NOUN bracis-28376 191 28 described describe VERB bracis-28376 191 29 in in ADP bracis-28376 191 30 sect sect NOUN bracis-28376 191 31 . . PUNCT bracis-28376 191 32     SPACE bracis-28376 191 33 3 3 NUM bracis-28376 191 34 and and CCONJ bracis-28376 191 35 by by ADP bracis-28376 191 36 the the DET bracis-28376 191 37 mtl mtl PROPN bracis-28376 191 38 - - PUNCT bracis-28376 191 39 lsp lsp PROPN bracis-28376 191 40 - - PUNCT bracis-28376 191 41 bilstm bilstm NOUN bracis-28376 191 42 - - PUNCT bracis-28376 191 43 bert bert NOUN bracis-28376 191 44 model model NOUN bracis-28376 191 45 reported report VERB bracis-28376 191 46 in in ADP bracis-28376 191 47 [ [ X bracis-28376 191 48 15 15 NUM bracis-28376 191 49 ] ] PUNCT bracis-28376 191 50 . . PUNCT bracis-28376 192 1 we we PRON bracis-28376 192 2 confirm confirm VERB bracis-28376 192 3 the the DET bracis-28376 192 4 statistical statistical ADJ bracis-28376 192 5 power power NOUN bracis-28376 192 6 of of ADP bracis-28376 192 7 the the DET bracis-28376 192 8 samples sample NOUN bracis-28376 192 9 ( ( PUNCT bracis-28376 192 10 scores score NOUN bracis-28376 192 11 ) ) PUNCT bracis-28376 192 12 by by ADP bracis-28376 192 13 employing employ VERB bracis-28376 192 14 bootstrap bootstrap NOUN bracis-28376 192 15 power power NOUN bracis-28376 192 16 analysis analysis NOUN bracis-28376 192 17 [ [ X bracis-28376 192 18 22 22 NUM bracis-28376 192 19 ] ] PUNCT bracis-28376 192 20 . . PUNCT bracis-28376 193 1 baselines baseline NOUN bracis-28376 193 2 : : PUNCT bracis-28376 193 3 all all DET bracis-28376 193 4 cohan cohan PROPN bracis-28376 193 5 models model NOUN bracis-28376 193 6 achieve achieve VERB bracis-28376 193 7 higher high ADJ bracis-28376 193 8 scores score NOUN bracis-28376 193 9 than than ADP bracis-28376 193 10 singlesc singlesc NOUN bracis-28376 193 11 ones one NOUN bracis-28376 193 12 and and CCONJ bracis-28376 193 13 the the DET bracis-28376 193 14 differences difference NOUN bracis-28376 193 15 are be AUX bracis-28376 193 16 statistically statistically ADV bracis-28376 193 17 significant significant ADJ bracis-28376 193 18 in in ADP bracis-28376 193 19 both both DET bracis-28376 193 20 datasets dataset NOUN bracis-28376 193 21 ( ( PUNCT bracis-28376 193 22 \(\epsilon \(\epsilon NOUN bracis-28376 193 23 _ _ PRON bracis-28376 193 24 { { PUNCT bracis-28376 193 25 min min NOUN bracis-28376 193 26 } } PUNCT bracis-28376 193 27 \le \le PROPN bracis-28376 193 28 0.05\ 0.05\ NUM bracis-28376 193 29 ) ) PUNCT bracis-28376 193 30 in in ADP bracis-28376 193 31 all all DET bracis-28376 193 32 comparisons comparison NOUN bracis-28376 193 33 ) ) PUNCT bracis-28376 193 34 . . PUNCT bracis-28376 194 1 this this PRON bracis-28376 194 2 is be AUX bracis-28376 194 3 expected expect VERB bracis-28376 194 4 since since SCONJ bracis-28376 194 5 cohan cohan PROPN bracis-28376 194 6 models model NOUN bracis-28376 194 7 exploit exploit VERB bracis-28376 194 8 more more ADJ bracis-28376 194 9 contextual contextual ADJ bracis-28376 194 10 data datum NOUN bracis-28376 194 11 to to PART bracis-28376 194 12 yield yield VERB bracis-28376 194 13 sentence sentence NOUN bracis-28376 194 14 embeddings embedding NOUN bracis-28376 194 15 . . PUNCT bracis-28376 195 1 the the DET bracis-28376 195 2 differences difference NOUN bracis-28376 195 3 between between ADP bracis-28376 195 4 singlesc singlesc NOUN bracis-28376 195 5 - - PUNCT bracis-28376 195 6 incaselaw incaselaw NOUN bracis-28376 195 7 and and CCONJ bracis-28376 195 8 singlesc singlesc PROPN bracis-28376 195 9 - - PUNCT bracis-28376 195 10 roberta roberta PROPN bracis-28376 195 11 models model NOUN bracis-28376 195 12 are be AUX bracis-28376 195 13 significant significant ADJ bracis-28376 195 14 in in ADP bracis-28376 195 15 both both DET bracis-28376 195 16 datasets dataset NOUN bracis-28376 195 17 as as ADV bracis-28376 195 18 well well ADV bracis-28376 195 19 ( ( PUNCT bracis-28376 195 20 \(\epsilon \(\epsilon NOUN bracis-28376 195 21 _ _ PRON bracis-28376 195 22 { { PUNCT bracis-28376 195 23 min min NOUN bracis-28376 195 24 } } PUNCT bracis-28376 195 25 \le \le PROPN bracis-28376 195 26 0.08\ 0.08\ PROPN bracis-28376 195 27 ) ) PUNCT bracis-28376 195 28 ) ) PUNCT bracis-28376 195 29 . . PUNCT bracis-28376 196 1 we we PRON bracis-28376 196 2 also also ADV bracis-28376 196 3 have have VERB bracis-28376 196 4 significant significant ADJ bracis-28376 196 5 differences difference NOUN bracis-28376 196 6 for for ADP bracis-28376 196 7 cohan cohan PROPN bracis-28376 196 8 - - PUNCT bracis-28376 196 9 longformer longformer PROPN bracis-28376 196 10 vs. vs. ADP bracis-28376 196 11 cohan cohan PROPN bracis-28376 196 12 - - PUNCT bracis-28376 196 13 incaselaw incaselaw PROPN bracis-28376 196 14 and and CCONJ bracis-28376 196 15 cohan cohan PROPN bracis-28376 196 16 - - PUNCT bracis-28376 196 17 longformer longformer PROPN bracis-28376 196 18 vs. vs. PROPN bracis-28376 196 19 cohan cohan PROPN bracis-28376 196 20 - - PUNCT bracis-28376 196 21 roberta roberta PROPN bracis-28376 196 22 ( ( PUNCT bracis-28376 196 23 \(\epsilon \(\epsilon NOUN bracis-28376 196 24 _ _ X bracis-28376 196 25 { { PUNCT bracis-28376 196 26 min}=0.01\ min}=0.01\ NOUN bracis-28376 196 27 ) ) PUNCT bracis-28376 196 28 in in ADP bracis-28376 196 29 both both DET bracis-28376 196 30 comparisons comparison NOUN bracis-28376 196 31 ) ) PUNCT bracis-28376 196 32 for for ADP bracis-28376 196 33 the the DET bracis-28376 196 34 7 7 NUM bracis-28376 196 35 - - PUNCT bracis-28376 196 36 roles role NOUN bracis-28376 196 37 dataset dataset NOUN bracis-28376 196 38 , , PUNCT bracis-28376 196 39 and and CCONJ bracis-28376 196 40 for for ADP bracis-28376 196 41 cohan cohan NOUN bracis-28376 196 42 - - PUNCT bracis-28376 196 43 incaselaw incaselaw PROPN bracis-28376 196 44 vs vs ADP bracis-28376 196 45 cohan cohan PROPN bracis-28376 196 46 - - PUNCT bracis-28376 196 47 roberta roberta PROPN bracis-28376 196 48 ( ( PUNCT bracis-28376 196 49 \(\epsilon \(\epsilon NOUN bracis-28376 196 50 _ _ PRON bracis-28376 196 51 { { PUNCT bracis-28376 196 52 min}=0.04\ min}=0.04\ PROPN bracis-28376 196 53 ) ) PUNCT bracis-28376 196 54 ) ) PUNCT bracis-28376 196 55 and and CCONJ bracis-28376 196 56 cohan cohan PROPN bracis-28376 196 57 - - PUNCT bracis-28376 196 58 incaselaw incaselaw PROPN bracis-28376 196 59 vs vs ADP bracis-28376 196 60 cohan cohan PROPN bracis-28376 196 61 - - PUNCT bracis-28376 196 62 longformer longformer NOUN bracis-28376 196 63 ( ( PUNCT bracis-28376 196 64 \(\epsilon \(\epsilon NOUN bracis-28376 196 65 _ _ NOUN bracis-28376 196 66 { { PUNCT bracis-28376 196 67 min}=0.00\ min}=0.00\ NOUN bracis-28376 196 68 ) ) PUNCT bracis-28376 196 69 ) ) PUNCT bracis-28376 196 70 for for ADP bracis-28376 196 71 the the DET bracis-28376 196 72 4 4 NUM bracis-28376 196 73 - - PUNCT bracis-28376 196 74 roles role NOUN bracis-28376 196 75 dataset dataset NOUN bracis-28376 196 76 . . PUNCT bracis-28376 197 1 the the DET bracis-28376 197 2 mtl mtl PROPN bracis-28376 197 3 - - PUNCT bracis-28376 197 4 lsp lsp PROPN bracis-28376 197 5 - - PUNCT bracis-28376 197 6 bilstm bilstm NOUN bracis-28376 197 7 - - PUNCT bracis-28376 197 8 bert bert NOUN bracis-28376 197 9 model model NOUN bracis-28376 197 10 remarkably remarkably ADV bracis-28376 197 11 outperforms outperform VERB bracis-28376 197 12 all all DET bracis-28376 197 13 baselines baseline NOUN bracis-28376 197 14 . . PUNCT bracis-28376 198 1 although although SCONJ bracis-28376 198 2 we we PRON bracis-28376 198 3 can can AUX bracis-28376 198 4 not not PART bracis-28376 198 5 perform perform VERB bracis-28376 198 6 aso aso NOUN bracis-28376 198 7 tests test NOUN bracis-28376 198 8 among among ADP bracis-28376 198 9 this this DET bracis-28376 198 10 model model NOUN bracis-28376 198 11 and and CCONJ bracis-28376 198 12 our our PRON bracis-28376 198 13 baselines baseline NOUN bracis-28376 198 14 because because SCONJ bracis-28376 198 15 the the DET bracis-28376 198 16 scores score NOUN bracis-28376 198 17 of of ADP bracis-28376 198 18 mtl mtl PROPN bracis-28376 198 19 - - PUNCT bracis-28376 198 20 lsp lsp PROPN bracis-28376 198 21 - - PUNCT bracis-28376 198 22 bilstm bilstm NOUN bracis-28376 198 23 - - PUNCT bracis-28376 198 24 bert bert NOUN bracis-28376 198 25 are be AUX bracis-28376 198 26 not not PART bracis-28376 198 27 available available ADJ bracis-28376 198 28 , , PUNCT bracis-28376 198 29 the the DET bracis-28376 198 30 differences difference NOUN bracis-28376 198 31 are be AUX bracis-28376 198 32 likely likely ADV bracis-28376 198 33 significant significant ADJ bracis-28376 198 34 because because SCONJ bracis-28376 198 35 the the DET bracis-28376 198 36 differences difference NOUN bracis-28376 198 37 are be AUX bracis-28376 198 38 large large ADJ bracis-28376 198 39 and and CCONJ bracis-28376 198 40 the the DET bracis-28376 198 41 reported report VERB bracis-28376 198 42 standard standard ADJ bracis-28376 198 43 deviation deviation NOUN bracis-28376 198 44 is be AUX bracis-28376 198 45 small small ADJ bracis-28376 198 46 . . PUNCT bracis-28376 199 1 mixup mixup NOUN bracis-28376 199 2 models model NOUN bracis-28376 199 3 : : PUNCT bracis-28376 199 4 regarding regard VERB bracis-28376 199 5 the the DET bracis-28376 199 6 two two NUM bracis-28376 199 7 datasets dataset NOUN bracis-28376 199 8 , , PUNCT bracis-28376 199 9 the the DET bracis-28376 199 10 score score NOUN bracis-28376 199 11 differences difference NOUN bracis-28376 199 12 between between ADP bracis-28376 199 13 each each DET bracis-28376 199 14 mixup mixup NOUN bracis-28376 199 15 - - PUNCT bracis-28376 199 16 a a NOUN bracis-28376 199 17 or or CCONJ bracis-28376 199 18 mixup mixup NOUN bracis-28376 199 19 - - PUNCT bracis-28376 199 20 b b NOUN bracis-28376 199 21 model model NOUN bracis-28376 199 22 and and CCONJ bracis-28376 199 23 its its PRON bracis-28376 199 24 singlesc singlesc NOUN bracis-28376 199 25 counterpart counterpart NOUN bracis-28376 199 26 ( ( PUNCT bracis-28376 199 27 e.g. e.g. ADV bracis-28376 199 28 , , PUNCT bracis-28376 199 29 mixup mixup NOUN bracis-28376 199 30 - - PUNCT bracis-28376 199 31 a a DET bracis-28376 199 32 - - PUNCT bracis-28376 199 33 incaselaw incaselaw NOUN bracis-28376 199 34 vs. vs. ADP bracis-28376 199 35 singlesc singlesc NOUN bracis-28376 199 36 - - PUNCT bracis-28376 199 37 incaselaw incaselaw NOUN bracis-28376 199 38 ) ) PUNCT bracis-28376 199 39 are be AUX bracis-28376 199 40 really really ADV bracis-28376 199 41 small small ADJ bracis-28376 199 42 , , PUNCT bracis-28376 199 43 and and CCONJ bracis-28376 199 44 according accord VERB bracis-28376 199 45 to to ADP bracis-28376 199 46 the the DET bracis-28376 199 47 aso aso NOUN bracis-28376 199 48 tests test NOUN bracis-28376 199 49 , , PUNCT bracis-28376 199 50 these these DET bracis-28376 199 51 differences difference NOUN bracis-28376 199 52 do do AUX bracis-28376 199 53 not not PART bracis-28376 199 54 have have VERB bracis-28376 199 55 statistical statistical ADJ bracis-28376 199 56 significance significance NOUN bracis-28376 199 57 . . PUNCT bracis-28376 200 1 hence hence ADV bracis-28376 200 2 , , PUNCT bracis-28376 200 3 the the DET bracis-28376 200 4 mixup mixup NOUN bracis-28376 200 5 approaches approach NOUN bracis-28376 200 6 do do AUX bracis-28376 200 7 not not PART bracis-28376 200 8 harm harm VERB bracis-28376 200 9 or or CCONJ bracis-28376 200 10 improve improve VERB bracis-28376 200 11 the the DET bracis-28376 200 12 base base NOUN bracis-28376 200 13 singlesc singlesc NOUN bracis-28376 200 14 models model NOUN bracis-28376 200 15 . . PUNCT bracis-28376 201 1 also also ADV bracis-28376 201 2 , , PUNCT bracis-28376 201 3 there there PRON bracis-28376 201 4 are be VERB bracis-28376 201 5 no no DET bracis-28376 201 6 significant significant ADJ bracis-28376 201 7 differences difference NOUN bracis-28376 201 8 among among ADP bracis-28376 201 9 mixup mixup NOUN bracis-28376 201 10 - - PUNCT bracis-28376 201 11 a a NOUN bracis-28376 201 12 and and CCONJ bracis-28376 201 13 mixup mixup NOUN bracis-28376 201 14 - - PUNCT bracis-28376 201 15 b b NOUN bracis-28376 201 16 models model NOUN bracis-28376 201 17 , , PUNCT bracis-28376 201 18 considering consider VERB bracis-28376 201 19 both both CCONJ bracis-28376 201 20 datasets dataset NOUN bracis-28376 201 21 and and CCONJ bracis-28376 201 22 the the DET bracis-28376 201 23 same same ADJ bracis-28376 201 24 encoder encoder NOUN bracis-28376 201 25 . . PUNCT bracis-28376 202 1 previous previous ADJ bracis-28376 202 2 work work NOUN bracis-28376 202 3 [ [ X bracis-28376 202 4 10 10 NUM bracis-28376 202 5 ] ] PUNCT bracis-28376 202 6 achieves achieve VERB bracis-28376 202 7 performance performance NOUN bracis-28376 202 8 gains gain NOUN bracis-28376 202 9 by by ADP bracis-28376 202 10 exploiting exploit VERB bracis-28376 202 11 mixup mixup NOUN bracis-28376 202 12 vectors vector NOUN bracis-28376 202 13 : : PUNCT bracis-28376 202 14 the the DET bracis-28376 202 15 f1 f1 PROPN bracis-28376 202 16 score score NOUN bracis-28376 202 17 achieved achieve VERB bracis-28376 202 18 by by ADP bracis-28376 202 19 a a DET bracis-28376 202 20 linear linear ADJ bracis-28376 202 21 classifier classifier NOUN bracis-28376 202 22 improves improve VERB bracis-28376 202 23 by by ADP bracis-28376 202 24 4.3 4.3 NUM bracis-28376 202 25 % % NOUN bracis-28376 202 26 in in ADP bracis-28376 202 27 the the DET bracis-28376 202 28 worst bad ADJ bracis-28376 202 29 case case NOUN bracis-28376 202 30 and and CCONJ bracis-28376 202 31 by by ADP bracis-28376 202 32 6.8 6.8 NUM bracis-28376 202 33 % % NOUN bracis-28376 202 34 in in ADP bracis-28376 202 35 the the DET bracis-28376 202 36 best good ADJ bracis-28376 202 37 case case NOUN bracis-28376 202 38 . . PUNCT bracis-28376 203 1 so so ADV bracis-28376 203 2 why why SCONJ bracis-28376 203 3 do do AUX bracis-28376 203 4 our our PRON bracis-28376 203 5 mixup mixup NOUN bracis-28376 203 6 - - PUNCT bracis-28376 203 7 based base VERB bracis-28376 203 8 models model NOUN bracis-28376 203 9 not not PART bracis-28376 203 10 outperform outperform VERB bracis-28376 203 11 their their PRON bracis-28376 203 12 baselines baseline NOUN bracis-28376 203 13 ? ? PUNCT bracis-28376 204 1 this this PRON bracis-28376 204 2 may may AUX bracis-28376 204 3 be be AUX bracis-28376 204 4 due due ADJ bracis-28376 204 5 to to ADP bracis-28376 204 6 the the DET bracis-28376 204 7 size size NOUN bracis-28376 204 8 of of ADP bracis-28376 204 9 the the DET bracis-28376 204 10 datasets dataset NOUN bracis-28376 204 11 . . PUNCT bracis-28376 205 1 in in ADP bracis-28376 205 2 [ [ X bracis-28376 205 3 10 10 NUM bracis-28376 205 4 ] ] PUNCT bracis-28376 205 5 , , PUNCT bracis-28376 205 6 there there PRON bracis-28376 205 7 is be VERB bracis-28376 205 8 little little ADJ bracis-28376 205 9 data datum NOUN bracis-28376 205 10 , , PUNCT bracis-28376 205 11 so so CCONJ bracis-28376 205 12 the the DET bracis-28376 205 13 additional additional ADJ bracis-28376 205 14 data datum NOUN bracis-28376 205 15 provided provide VERB bracis-28376 205 16 by by ADP bracis-28376 205 17 mixup mixup NOUN bracis-28376 205 18 is be AUX bracis-28376 205 19 beneficial beneficial ADJ bracis-28376 205 20 . . PUNCT bracis-28376 206 1 conversely conversely ADV bracis-28376 206 2 , , PUNCT bracis-28376 206 3 the the DET bracis-28376 206 4 dataset dataset NOUN bracis-28376 206 5 of of ADP bracis-28376 206 6 the the DET bracis-28376 206 7 current current ADJ bracis-28376 206 8 work work NOUN bracis-28376 206 9 is be AUX bracis-28376 206 10 larger large ADJ bracis-28376 206 11 , , PUNCT bracis-28376 206 12 leaving leave VERB bracis-28376 206 13 no no DET bracis-28376 206 14 room room NOUN bracis-28376 206 15 for for ADP bracis-28376 206 16 improvements improvement NOUN bracis-28376 206 17 based base VERB bracis-28376 206 18 on on ADP bracis-28376 206 19 data datum NOUN bracis-28376 206 20 augmentation augmentation NOUN bracis-28376 206 21 . . PUNCT bracis-28376 207 1 pe pe PROPN bracis-28376 207 2 models model NOUN bracis-28376 207 3 : : PUNCT bracis-28376 207 4 concerning concern VERB bracis-28376 207 5 the the DET bracis-28376 207 6 7 7 NUM bracis-28376 207 7 - - PUNCT bracis-28376 207 8 roles role NOUN bracis-28376 207 9 dataset dataset NOUN bracis-28376 207 10 , , PUNCT bracis-28376 207 11 the the DET bracis-28376 207 12 pe pe PROPN bracis-28376 207 13 - - PUNCT bracis-28376 207 14 s s NOUN bracis-28376 207 15 models model NOUN bracis-28376 207 16 perform perform VERB bracis-28376 207 17 worse bad ADJ bracis-28376 207 18 than than ADP bracis-28376 207 19 their their PRON bracis-28376 207 20 respective respective ADJ bracis-28376 207 21 singlesc singlesc NOUN bracis-28376 207 22 counterparts counterpart NOUN bracis-28376 207 23 , , PUNCT bracis-28376 207 24 but but CCONJ bracis-28376 207 25 the the DET bracis-28376 207 26 difference difference NOUN bracis-28376 207 27 is be AUX bracis-28376 207 28 only only ADV bracis-28376 207 29 significant significant ADJ bracis-28376 207 30 for for ADP bracis-28376 207 31 singlesc singlesc NOUN bracis-28376 207 32 - - PUNCT bracis-28376 207 33 roberta roberta PROPN bracis-28376 207 34 vs. vs. X bracis-28376 207 35 pe pe PROPN bracis-28376 207 36 - - PUNCT bracis-28376 207 37 s s NOUN bracis-28376 207 38 - - PUNCT bracis-28376 207 39 roberta roberta NOUN bracis-28376 207 40 ( ( PUNCT bracis-28376 207 41 \(\epsilon \(\epsilon NOUN bracis-28376 207 42 _ _ X bracis-28376 207 43 { { PUNCT bracis-28376 207 44 min}=0.03\ min}=0.03\ NOUN bracis-28376 207 45 ) ) PUNCT bracis-28376 207 46 ) ) PUNCT bracis-28376 207 47 . . PUNCT bracis-28376 208 1 this this PRON bracis-28376 208 2 means mean VERB bracis-28376 208 3 that that SCONJ bracis-28376 208 4 the the DET bracis-28376 208 5 pe pe PROPN bracis-28376 208 6 - - PUNCT bracis-28376 208 7 s s PART bracis-28376 208 8 approach approach NOUN bracis-28376 208 9 harms harm VERB bracis-28376 208 10 the the DET bracis-28376 208 11 base base NOUN bracis-28376 208 12 singlesc singlesc PROPN bracis-28376 208 13 - - PUNCT bracis-28376 208 14 roberta roberta PROPN bracis-28376 208 15 model model NOUN bracis-28376 208 16 instead instead ADV bracis-28376 208 17 of of ADP bracis-28376 208 18 improving improve VERB bracis-28376 208 19 it it PRON bracis-28376 208 20 . . PUNCT bracis-28376 209 1 for for ADP bracis-28376 209 2 the the DET bracis-28376 209 3 4 4 NUM bracis-28376 209 4 - - PUNCT bracis-28376 209 5 roles role NOUN bracis-28376 209 6 dataset dataset NOUN bracis-28376 209 7 , , PUNCT bracis-28376 209 8 the the DET bracis-28376 209 9 pe pe PROPN bracis-28376 209 10 - - PUNCT bracis-28376 209 11 s s NOUN bracis-28376 209 12 models model NOUN bracis-28376 209 13 outperform outperform VERB bracis-28376 209 14 their their PRON bracis-28376 209 15 respective respective ADJ bracis-28376 209 16 singlesc singlesc NOUN bracis-28376 209 17 counterparts counterpart NOUN bracis-28376 209 18 , , PUNCT bracis-28376 209 19 but but CCONJ bracis-28376 209 20 the the DET bracis-28376 209 21 differences difference NOUN bracis-28376 209 22 are be AUX bracis-28376 209 23 not not PART bracis-28376 209 24 significant significant ADJ bracis-28376 209 25 . . PUNCT bracis-28376 210 1 also also ADV bracis-28376 210 2 , , PUNCT bracis-28376 210 3 there there PRON bracis-28376 210 4 are be VERB bracis-28376 210 5 no no DET bracis-28376 210 6 significant significant ADJ bracis-28376 210 7 differences difference NOUN bracis-28376 210 8 among among ADP bracis-28376 210 9 the the DET bracis-28376 210 10 pe pe PROPN bracis-28376 210 11 - - PUNCT bracis-28376 210 12 c c NOUN bracis-28376 210 13 models model NOUN bracis-28376 210 14 and and CCONJ bracis-28376 210 15 their their PRON bracis-28376 210 16 respective respective ADJ bracis-28376 210 17 singlesc singlesc NOUN bracis-28376 210 18 counterparts counterpart NOUN bracis-28376 210 19 for for ADP bracis-28376 210 20 the the DET bracis-28376 210 21 two two NUM bracis-28376 210 22 datasets dataset NOUN bracis-28376 210 23 . . PUNCT bracis-28376 211 1 we we PRON bracis-28376 211 2 expected expect VERB bracis-28376 211 3 poor poor ADJ bracis-28376 211 4 results result NOUN bracis-28376 211 5 ( ( PUNCT bracis-28376 211 6 i.e. i.e. X bracis-28376 211 7 , , PUNCT bracis-28376 211 8 no no DET bracis-28376 211 9 gain gain NOUN bracis-28376 211 10 or or CCONJ bracis-28376 211 11 decrease decrease NOUN bracis-28376 211 12 in in ADP bracis-28376 211 13 performance performance NOUN bracis-28376 211 14 ) ) PUNCT bracis-28376 211 15 from from ADP bracis-28376 211 16 the the DET bracis-28376 211 17 pe pe PROPN bracis-28376 211 18 models model NOUN bracis-28376 211 19 with with ADP bracis-28376 211 20 respect respect NOUN bracis-28376 211 21 to to ADP bracis-28376 211 22 the the DET bracis-28376 211 23 7 7 NUM bracis-28376 211 24 - - PUNCT bracis-28376 211 25 roles role NOUN bracis-28376 211 26 dataset dataset NOUN bracis-28376 211 27 , , PUNCT bracis-28376 211 28 since since SCONJ bracis-28376 211 29 most most ADJ bracis-28376 211 30 of of ADP bracis-28376 211 31 its its PRON bracis-28376 211 32 labels label NOUN bracis-28376 211 33 do do AUX bracis-28376 211 34 not not PART bracis-28376 211 35 have have VERB bracis-28376 211 36 a a DET bracis-28376 211 37 clear clear ADJ bracis-28376 211 38 correlation correlation NOUN bracis-28376 211 39 with with ADP bracis-28376 211 40 sentence sentence NOUN bracis-28376 211 41 positions position NOUN bracis-28376 211 42 . . PUNCT bracis-28376 212 1 in in ADP bracis-28376 212 2 the the DET bracis-28376 212 3 case case NOUN bracis-28376 212 4 of of ADP bracis-28376 212 5 performance performance NOUN bracis-28376 212 6 degradation degradation NOUN bracis-28376 212 7 , , PUNCT bracis-28376 212 8 pe pe PROPN bracis-28376 212 9 vectors vector NOUN bracis-28376 212 10 should should AUX bracis-28376 212 11 be be AUX bracis-28376 212 12 working work VERB bracis-28376 212 13 as as ADP bracis-28376 212 14 noise noise NOUN bracis-28376 212 15 data datum NOUN bracis-28376 212 16 . . PUNCT bracis-28376 213 1 conversely conversely ADV bracis-28376 213 2 , , PUNCT bracis-28376 213 3 we we PRON bracis-28376 213 4 expected expect VERB bracis-28376 213 5 pe pe INTJ bracis-28376 213 6 vectors vector NOUN bracis-28376 213 7 to to PART bracis-28376 213 8 provide provide VERB bracis-28376 213 9 gains gain NOUN bracis-28376 213 10 in in ADP bracis-28376 213 11 the the DET bracis-28376 213 12 experiments experiment NOUN bracis-28376 213 13 with with ADP bracis-28376 213 14 the the DET bracis-28376 213 15 4 4 NUM bracis-28376 213 16 - - PUNCT bracis-28376 213 17 roles role NOUN bracis-28376 213 18 dataset dataset NOUN bracis-28376 213 19 , , PUNCT bracis-28376 213 20 which which PRON bracis-28376 213 21 does do AUX bracis-28376 213 22 not not PART bracis-28376 213 23 happen happen VERB bracis-28376 213 24 . . PUNCT bracis-28376 214 1 we we PRON bracis-28376 214 2 believe believe VERB bracis-28376 214 3 that that SCONJ bracis-28376 214 4 the the DET bracis-28376 214 5 positional positional ADJ bracis-28376 214 6 embeddings embedding NOUN bracis-28376 214 7 employed employ VERB bracis-28376 214 8 do do AUX bracis-28376 214 9 not not PART bracis-28376 214 10 represent represent VERB bracis-28376 214 11 positions position NOUN bracis-28376 214 12 in in ADP bracis-28376 214 13 a a DET bracis-28376 214 14 way way NOUN bracis-28376 214 15 that that PRON bracis-28376 214 16 can can AUX bracis-28376 214 17 enrich enrich VERB bracis-28376 214 18 the the DET bracis-28376 214 19 input input NOUN bracis-28376 214 20 data datum NOUN bracis-28376 214 21 of of ADP bracis-28376 214 22 the the DET bracis-28376 214 23 models model NOUN bracis-28376 214 24 . . PUNCT bracis-28376 215 1 dfcsc dfcsc NOUN bracis-28376 215 2 models model NOUN bracis-28376 215 3 : : PUNCT bracis-28376 215 4 regarding regard VERB bracis-28376 215 5 the the DET bracis-28376 215 6 7 7 NUM bracis-28376 215 7 - - PUNCT bracis-28376 215 8 roles role NOUN bracis-28376 215 9 dataset dataset NOUN bracis-28376 215 10 , , PUNCT bracis-28376 215 11 all all DET bracis-28376 215 12 dfcsc dfcsc NOUN bracis-28376 215 13 models model NOUN bracis-28376 215 14 outperform outperform VERB bracis-28376 215 15 their their PRON bracis-28376 215 16 respective respective ADJ bracis-28376 215 17 cohan cohan PROPN bracis-28376 215 18 counterparts counterpart NOUN bracis-28376 215 19 . . PUNCT bracis-28376 216 1 the the DET bracis-28376 216 2 differences difference NOUN bracis-28376 216 3 are be AUX bracis-28376 216 4 significant significant ADJ bracis-28376 216 5 with with ADP bracis-28376 216 6 \(\epsilon \(\epsilon NOUN bracis-28376 216 7 _ _ NOUN bracis-28376 216 8 { { PUNCT bracis-28376 216 9 min}=0.14\ min}=0.14\ NOUN bracis-28376 216 10 ) ) PUNCT bracis-28376 216 11 for for ADP bracis-28376 216 12 dfcsc dfcsc NOUN bracis-28376 216 13 - - PUNCT bracis-28376 216 14 cls cls NOUN bracis-28376 216 15 - - PUNCT bracis-28376 216 16 incaselaw incaselaw NOUN bracis-28376 216 17 vs. vs. ADP bracis-28376 216 18 cohan cohan PROPN bracis-28376 216 19 - - PUNCT bracis-28376 216 20 incaselaw incaselaw PROPN bracis-28376 216 21 , , PUNCT bracis-28376 216 22 and and CCONJ bracis-28376 216 23 \(\epsilon \(\epsilon NOUN bracis-28376 216 24 _ _ NOUN bracis-28376 216 25 { { PUNCT bracis-28376 216 26 min}=0.01\ min}=0.01\ NOUN bracis-28376 216 27 ) ) PUNCT bracis-28376 216 28 for for ADP bracis-28376 216 29 the the DET bracis-28376 216 30 remaining remain VERB bracis-28376 216 31 comparisons comparison NOUN bracis-28376 216 32 . . PUNCT bracis-28376 217 1 for for ADP bracis-28376 217 2 the the DET bracis-28376 217 3 4 4 NUM bracis-28376 217 4 - - PUNCT bracis-28376 217 5 roles role NOUN bracis-28376 217 6 dataset dataset NOUN bracis-28376 217 7 , , PUNCT bracis-28376 217 8 only only ADV bracis-28376 217 9 dfcsc dfcsc PROPN bracis-28376 217 10 - - PUNCT bracis-28376 217 11 sep sep NOUN bracis-28376 217 12 - - PUNCT bracis-28376 217 13 incaselaw incaselaw NOUN bracis-28376 217 14 and and CCONJ bracis-28376 217 15 dfcsc dfcsc NOUN bracis-28376 217 16 - - PUNCT bracis-28376 217 17 cls cls NOUN bracis-28376 217 18 - - PUNCT bracis-28376 217 19 incaselaw incaselaw NOUN bracis-28376 217 20 do do AUX bracis-28376 217 21 not not PART bracis-28376 217 22 outperform outperform VERB bracis-28376 217 23 their their PRON bracis-28376 217 24 respective respective ADJ bracis-28376 217 25 cohan cohan PROPN bracis-28376 217 26 counterparts counterpart NOUN bracis-28376 217 27 . . PUNCT bracis-28376 218 1 the the DET bracis-28376 218 2 aso aso PROPN bracis-28376 218 3 test test NOUN bracis-28376 218 4 results result NOUN bracis-28376 218 5 in in ADP bracis-28376 218 6 \(\epsilon \(\epsilon NOUN bracis-28376 218 7 _ _ NOUN bracis-28376 218 8 { { PUNCT bracis-28376 218 9 min}=0.23\ min}=0.23\ NOUN bracis-28376 218 10 ) ) PUNCT bracis-28376 218 11 for for ADP bracis-28376 218 12 cohan cohan NOUN bracis-28376 218 13 - - PUNCT bracis-28376 218 14 incaselaw incaselaw PROPN bracis-28376 218 15 vs. vs. ADP bracis-28376 218 16 dfcsc dfcsc PROPN bracis-28376 218 17 - - PUNCT bracis-28376 218 18 sep sep PROPN bracis-28376 218 19 - - PUNCT bracis-28376 218 20 incaselaw incaselaw NOUN bracis-28376 218 21 , , PUNCT bracis-28376 218 22 and and CCONJ bracis-28376 218 23 in in ADP bracis-28376 218 24 \(\epsilon \(\epsilon NOUN bracis-28376 218 25 _ _ NOUN bracis-28376 218 26 { { PUNCT bracis-28376 218 27 min}=0.00\ min}=0.00\ NOUN bracis-28376 218 28 ) ) PUNCT bracis-28376 218 29 for for ADP bracis-28376 218 30 cohan cohan NOUN bracis-28376 218 31 - - PUNCT bracis-28376 218 32 incaselaw incaselaw PROPN bracis-28376 218 33 vs. vs. ADP bracis-28376 218 34 dfcsc dfcsc NOUN bracis-28376 218 35 - - PUNCT bracis-28376 218 36 cls cls NOUN bracis-28376 218 37 - - PUNCT bracis-28376 218 38 incaselaw incaselaw NOUN bracis-28376 218 39 , , PUNCT bracis-28376 218 40 meaning mean VERB bracis-28376 218 41 that that SCONJ bracis-28376 218 42 only only ADV bracis-28376 218 43 the the DET bracis-28376 218 44 latter latter ADJ bracis-28376 218 45 comparison comparison NOUN bracis-28376 218 46 is be AUX bracis-28376 218 47 significant significant ADJ bracis-28376 218 48 . . PUNCT bracis-28376 219 1 the the DET bracis-28376 219 2 dfcsc dfcsc NOUN bracis-28376 219 3 models model NOUN bracis-28376 219 4 using use VERB bracis-28376 219 5 roberta roberta PROPN bracis-28376 219 6 and and CCONJ bracis-28376 219 7 longformer longformer PROPN bracis-28376 219 8 encoders encoder NOUN bracis-28376 219 9 achieve achieve VERB bracis-28376 219 10 remarkable remarkable ADJ bracis-28376 219 11 improvements improvement NOUN bracis-28376 219 12 : : PUNCT bracis-28376 219 13 the the DET bracis-28376 219 14 lowest low ADJ bracis-28376 219 15 and and CCONJ bracis-28376 219 16 highest high ADJ bracis-28376 219 17 gains gain NOUN bracis-28376 219 18 are be AUX bracis-28376 219 19 5.9 5.9 NUM bracis-28376 219 20 % % NOUN bracis-28376 219 21 and and CCONJ bracis-28376 219 22 10.4 10.4 NUM bracis-28376 219 23 % % NOUN bracis-28376 219 24 , , PUNCT bracis-28376 219 25 respectively respectively ADV bracis-28376 219 26 . . PUNCT bracis-28376 220 1 this this DET bracis-28376 220 2 behavior behavior NOUN bracis-28376 220 3 is be AUX bracis-28376 220 4 also also ADV bracis-28376 220 5 reported report VERB bracis-28376 220 6 in in ADP bracis-28376 220 7 [ [ X bracis-28376 220 8 11 11 NUM bracis-28376 220 9 ] ] PUNCT bracis-28376 220 10 , , PUNCT bracis-28376 220 11 and and CCONJ bracis-28376 220 12 thus thus ADV bracis-28376 220 13 we we PRON bracis-28376 220 14 consider consider VERB bracis-28376 220 15 this this DET bracis-28376 220 16 additional additional ADJ bracis-28376 220 17 evidence evidence NOUN bracis-28376 220 18 that that SCONJ bracis-28376 220 19 the the DET bracis-28376 220 20 roberta roberta PROPN bracis-28376 220 21 and and CCONJ bracis-28376 220 22 longformer longformer PROPN bracis-28376 220 23 encoders encoder NOUN bracis-28376 220 24 exploit exploit VERB bracis-28376 220 25 contextual contextual ADJ bracis-28376 220 26 data datum NOUN bracis-28376 220 27 in in ADP bracis-28376 220 28 a a DET bracis-28376 220 29 better well ADJ bracis-28376 220 30 way way NOUN bracis-28376 220 31 than than ADP bracis-28376 220 32 the the DET bracis-28376 220 33 incaselaw incaselaw NOUN bracis-28376 220 34 one one NOUN bracis-28376 220 35 ( ( PUNCT bracis-28376 220 36 roberta roberta PROPN bracis-28376 220 37 and and CCONJ bracis-28376 220 38 longformer longformer PROPN bracis-28376 220 39 rely rely VERB bracis-28376 220 40 on on ADP bracis-28376 220 41 the the DET bracis-28376 220 42 same same ADJ bracis-28376 220 43 training training NOUN bracis-28376 220 44 procedure procedure NOUN bracis-28376 220 45 , , PUNCT bracis-28376 220 46 which which PRON bracis-28376 220 47 is be AUX bracis-28376 220 48 different different ADJ bracis-28376 220 49 from from ADP bracis-28376 220 50 the the DET bracis-28376 220 51 one one NOUN bracis-28376 220 52 used use VERB bracis-28376 220 53 for for ADP bracis-28376 220 54 incaselaw incaselaw NOUN bracis-28376 220 55 ) ) PUNCT bracis-28376 220 56 . . PUNCT bracis-28376 221 1 comparing compare VERB bracis-28376 221 2 the the DET bracis-28376 221 3 performance performance NOUN bracis-28376 221 4 of of ADP bracis-28376 221 5 the the DET bracis-28376 221 6 dfcsc dfcsc NOUN bracis-28376 221 7 models model NOUN bracis-28376 221 8 using use VERB bracis-28376 221 9 table table NOUN bracis-28376 221 10     SPACE bracis-28376 221 11 2 2 NUM bracis-28376 221 12 , , PUNCT bracis-28376 221 13 we we PRON bracis-28376 221 14 come come VERB bracis-28376 221 15 to to ADP bracis-28376 221 16 some some DET bracis-28376 221 17 conclusions conclusion NOUN bracis-28376 221 18 . . PUNCT bracis-28376 222 1 first first ADV bracis-28376 222 2 , , PUNCT bracis-28376 222 3 regarding regard VERB bracis-28376 222 4 the the DET bracis-28376 222 5 7 7 NUM bracis-28376 222 6 - - PUNCT bracis-28376 222 7 roles role NOUN bracis-28376 222 8 dataset dataset NOUN bracis-28376 222 9 , , PUNCT bracis-28376 222 10 we we PRON bracis-28376 222 11 see see VERB bracis-28376 222 12 that that SCONJ bracis-28376 222 13 all all DET bracis-28376 222 14 longformer longformer NOUN bracis-28376 222 15 - - PUNCT bracis-28376 222 16 based base VERB bracis-28376 222 17 models model NOUN bracis-28376 222 18 are be AUX bracis-28376 222 19 superior superior ADJ bracis-28376 222 20 to to ADP bracis-28376 222 21 those those PRON bracis-28376 222 22 based base VERB bracis-28376 222 23 on on ADP bracis-28376 222 24 roberta roberta PROPN bracis-28376 222 25 and and CCONJ bracis-28376 222 26 incaselaw incaselaw PROPN bracis-28376 222 27 ; ; PUNCT bracis-28376 222 28 that that SCONJ bracis-28376 222 29 all all DET bracis-28376 222 30 roberta roberta PROPN bracis-28376 222 31 - - PUNCT bracis-28376 222 32 based base VERB bracis-28376 222 33 models model NOUN bracis-28376 222 34 are be AUX bracis-28376 222 35 superior superior ADJ bracis-28376 222 36 to to ADP bracis-28376 222 37 those those PRON bracis-28376 222 38 based base VERB bracis-28376 222 39 on on ADP bracis-28376 222 40 incaselaw incaselaw NOUN bracis-28376 222 41 ; ; PUNCT bracis-28376 222 42 and and CCONJ bracis-28376 222 43 that that SCONJ bracis-28376 222 44 there there PRON bracis-28376 222 45 is be VERB bracis-28376 222 46 no no DET bracis-28376 222 47 significant significant ADJ bracis-28376 222 48 difference difference NOUN bracis-28376 222 49 between between ADP bracis-28376 222 50 the the DET bracis-28376 222 51 dfcsc dfcsc NOUN bracis-28376 222 52 - - PUNCT bracis-28376 222 53 sep sep PROPN bracis-28376 222 54 and and CCONJ bracis-28376 222 55 dfcsc dfcsc PROPN bracis-28376 222 56 - - PUNCT bracis-28376 222 57 cls cls NOUN bracis-28376 222 58 approaches approach NOUN bracis-28376 222 59 when when SCONJ bracis-28376 222 60 we we PRON bracis-28376 222 61 compare compare VERB bracis-28376 222 62 models model NOUN bracis-28376 222 63 that that PRON bracis-28376 222 64 employ employ VERB bracis-28376 222 65 the the DET bracis-28376 222 66 same same ADJ bracis-28376 222 67 encoder encoder NOUN bracis-28376 222 68 . . PUNCT bracis-28376 223 1 regarding regard VERB bracis-28376 223 2 the the DET bracis-28376 223 3 4 4 NUM bracis-28376 223 4 - - PUNCT bracis-28376 223 5 roles role NOUN bracis-28376 223 6 dataset dataset NOUN bracis-28376 223 7 , , PUNCT bracis-28376 223 8 we we PRON bracis-28376 223 9 find find VERB bracis-28376 223 10 that that SCONJ bracis-28376 223 11 dfcsc dfcsc NOUN bracis-28376 223 12 - - PUNCT bracis-28376 223 13 sep sep PROPN bracis-28376 223 14 - - PUNCT bracis-28376 223 15 longformer longformer NOUN bracis-28376 223 16 is be AUX bracis-28376 223 17 superior superior ADJ bracis-28376 223 18 to to ADP bracis-28376 223 19 all all DET bracis-28376 223 20 models model NOUN bracis-28376 223 21 using use VERB bracis-28376 223 22 roberta roberta PROPN bracis-28376 223 23 and and CCONJ bracis-28376 223 24 incaselaw incaselaw PROPN bracis-28376 223 25 ; ; PUNCT bracis-28376 223 26 that that SCONJ bracis-28376 223 27 dfcsc dfcsc NOUN bracis-28376 223 28 - - PUNCT bracis-28376 223 29 cls cls NOUN bracis-28376 223 30 - - PUNCT bracis-28376 223 31 longformer longformer NOUN bracis-28376 223 32 is be AUX bracis-28376 223 33 superior superior ADJ bracis-28376 223 34 to to ADP bracis-28376 223 35 dfcsc dfcsc NOUN bracis-28376 223 36 - - PUNCT bracis-28376 223 37 cls cls NOUN bracis-28376 223 38 - - PUNCT bracis-28376 223 39 roberta roberta PROPN bracis-28376 223 40 , , PUNCT bracis-28376 223 41 dfcsc dfcsc PROPN bracis-28376 223 42 - - PUNCT bracis-28376 223 43 sep sep PROPN bracis-28376 223 44 - - PUNCT bracis-28376 223 45 incaselaw incaselaw NOUN bracis-28376 223 46 , , PUNCT bracis-28376 223 47 and and CCONJ bracis-28376 223 48 dfcsc dfcsc NOUN bracis-28376 223 49 - - PUNCT bracis-28376 223 50 cls cls NOUN bracis-28376 223 51 - - PUNCT bracis-28376 223 52 incaselaw incaselaw NOUN bracis-28376 223 53 ; ; PUNCT bracis-28376 223 54 that that SCONJ bracis-28376 223 55 the the DET bracis-28376 223 56 two two NUM bracis-28376 223 57 roberta roberta NOUN bracis-28376 223 58 - - PUNCT bracis-28376 223 59 based base VERB bracis-28376 223 60 models model NOUN bracis-28376 223 61 are be AUX bracis-28376 223 62 superior superior ADJ bracis-28376 223 63 to to ADP bracis-28376 223 64 those those PRON bracis-28376 223 65 based base VERB bracis-28376 223 66 on on ADP bracis-28376 223 67 incaselaw incaselaw NOUN bracis-28376 223 68 ; ; PUNCT bracis-28376 223 69 and and CCONJ bracis-28376 223 70 that that SCONJ bracis-28376 223 71 there there PRON bracis-28376 223 72 is be VERB bracis-28376 223 73 no no DET bracis-28376 223 74 significant significant ADJ bracis-28376 223 75 difference difference NOUN bracis-28376 223 76 between between ADP bracis-28376 223 77 the the DET bracis-28376 223 78 dfcsc dfcsc NOUN bracis-28376 223 79 - - PUNCT bracis-28376 223 80 sep sep PROPN bracis-28376 223 81 and and CCONJ bracis-28376 223 82 dfcsc dfcsc PROPN bracis-28376 223 83 - - PUNCT bracis-28376 223 84 cls cls NOUN bracis-28376 223 85 approaches approach NOUN bracis-28376 223 86 when when SCONJ bracis-28376 223 87 we we PRON bracis-28376 223 88 compare compare VERB bracis-28376 223 89 models model NOUN bracis-28376 223 90 that that PRON bracis-28376 223 91 employ employ VERB bracis-28376 223 92 the the DET bracis-28376 223 93 same same ADJ bracis-28376 223 94 encoder encoder NOUN bracis-28376 223 95 . . PUNCT bracis-28376 224 1 when when SCONJ bracis-28376 224 2 comparing compare VERB bracis-28376 224 3 dfcsc dfcsc NOUN bracis-28376 224 4 and and CCONJ bracis-28376 224 5 mtl mtl PROPN bracis-28376 224 6 - - PUNCT bracis-28376 224 7 lsp lsp PROPN bracis-28376 224 8 - - PUNCT bracis-28376 224 9 bilstm bilstm NOUN bracis-28376 224 10 - - PUNCT bracis-28376 224 11 bert bert NOUN bracis-28376 224 12 models model NOUN bracis-28376 224 13 , , PUNCT bracis-28376 224 14 we we PRON bracis-28376 224 15 find find VERB bracis-28376 224 16 that that SCONJ bracis-28376 224 17 roberta roberta NOUN bracis-28376 224 18 - - PUNCT bracis-28376 224 19 based base VERB bracis-28376 224 20 models model NOUN bracis-28376 224 21 perform perform VERB bracis-28376 224 22 similarly similarly ADV bracis-28376 224 23 and and CCONJ bracis-28376 224 24 that that SCONJ bracis-28376 224 25 longformer longformer NOUN bracis-28376 224 26 - - PUNCT bracis-28376 224 27 based base VERB bracis-28376 224 28 models model NOUN bracis-28376 224 29 perform perform VERB bracis-28376 224 30 better well ADJ bracis-28376 224 31 . . PUNCT bracis-28376 225 1 table table NOUN bracis-28376 225 2 2 2 NUM bracis-28376 225 3 . . PUNCT bracis-28376 226 1 confidence confidence NOUN bracis-28376 226 2 scores score NOUN bracis-28376 226 3 ( ( PUNCT bracis-28376 226 4 \(\epsilon \(\epsilon NOUN bracis-28376 226 5 _ _ X bracis-28376 226 6 { { PUNCT bracis-28376 226 7 min}\ min}\ PROPN bracis-28376 226 8 ) ) PUNCT bracis-28376 226 9 ) ) PUNCT bracis-28376 226 10 from from ADP bracis-28376 226 11 aso aso NOUN bracis-28376 226 12 tests test NOUN bracis-28376 226 13 comparing compare VERB bracis-28376 226 14 dfcsc dfcsc NOUN bracis-28376 226 15 models model NOUN bracis-28376 226 16 . . PUNCT bracis-28376 227 1 comparisons comparison NOUN bracis-28376 227 2 follow follow VERB bracis-28376 227 3 the the DET bracis-28376 227 4 row row VERB bracis-28376 227 5 - - PUNCT bracis-28376 227 6 to to ADP bracis-28376 227 7 - - PUNCT bracis-28376 227 8 column column NOUN bracis-28376 227 9 direction direction NOUN bracis-28376 227 10 , , PUNCT bracis-28376 227 11 and and CCONJ bracis-28376 227 12 significant significant ADJ bracis-28376 227 13 differences difference NOUN bracis-28376 227 14 ( ( PUNCT bracis-28376 227 15 i.e. i.e. X bracis-28376 227 16 , , PUNCT bracis-28376 227 17 \(\epsilon \(\epsilon NOUN bracis-28376 227 18 _ _ PRON bracis-28376 227 19 { { PUNCT bracis-28376 227 20 min min NOUN bracis-28376 227 21 } } PUNCT bracis-28376 227 22 < < X bracis-28376 227 23 0.2\ 0.2\ ADJ bracis-28376 227 24 ) ) PUNCT bracis-28376 227 25 ) ) PUNCT bracis-28376 227 26 are be AUX bracis-28376 227 27 indicated indicate VERB bracis-28376 227 28 by by ADP bracis-28376 227 29 underlined underline VERB bracis-28376 227 30 values.full values.full PROPN bracis-28376 227 31 size size NOUN bracis-28376 227 32 table table NOUN bracis-28376 227 33 5 5 NUM bracis-28376 227 34 final final ADJ bracis-28376 227 35 remarks remark NOUN bracis-28376 227 36 in in ADP bracis-28376 227 37 this this DET bracis-28376 227 38 work work NOUN bracis-28376 227 39 , , PUNCT bracis-28376 227 40 we we PRON bracis-28376 227 41 compare compare VERB bracis-28376 227 42 approaches approach NOUN bracis-28376 227 43 from from ADP bracis-28376 227 44 recent recent ADJ bracis-28376 227 45 proposals proposal NOUN bracis-28376 227 46 to to PART bracis-28376 227 47 handle handle VERB bracis-28376 227 48 the the DET bracis-28376 227 49 legalrrl legalrrl NOUN bracis-28376 227 50 task task NOUN bracis-28376 227 51 . . PUNCT bracis-28376 228 1 all all DET bracis-28376 228 2 approaches approach NOUN bracis-28376 228 3 rely rely VERB bracis-28376 228 4 on on ADP bracis-28376 228 5 pre pre ADJ bracis-28376 228 6 - - ADJ bracis-28376 228 7 trained train VERB bracis-28376 228 8 transformers transformer NOUN bracis-28376 228 9 to to PART bracis-28376 228 10 encode encode VERB bracis-28376 228 11 sentences sentence NOUN bracis-28376 228 12 and and CCONJ bracis-28376 228 13 employ employ VERB bracis-28376 228 14 different different ADJ bracis-28376 228 15 strategies strategy NOUN bracis-28376 228 16 to to PART bracis-28376 228 17 improve improve VERB bracis-28376 228 18 reference reference NOUN bracis-28376 228 19 baselines baseline NOUN bracis-28376 228 20 . . PUNCT bracis-28376 229 1 models model NOUN bracis-28376 229 2 based base VERB bracis-28376 229 3 on on ADP bracis-28376 229 4 the the DET bracis-28376 229 5 mixup mixup NOUN bracis-28376 229 6 and and CCONJ bracis-28376 229 7 pe pe PROPN bracis-28376 229 8 approaches approach NOUN bracis-28376 229 9 are be AUX bracis-28376 229 10 unable unable ADJ bracis-28376 229 11 to to PART bracis-28376 229 12 overcome overcome VERB bracis-28376 229 13 the the DET bracis-28376 229 14 singlesc singlesc NOUN bracis-28376 229 15 baselines baseline NOUN bracis-28376 229 16 . . PUNCT bracis-28376 230 1 for for ADP bracis-28376 230 2 the the DET bracis-28376 230 3 pe pe PROPN bracis-28376 230 4 models model NOUN bracis-28376 230 5 , , PUNCT bracis-28376 230 6 we we PRON bracis-28376 230 7 conclude conclude VERB bracis-28376 230 8 that that SCONJ bracis-28376 230 9 the the DET bracis-28376 230 10 7 7 NUM bracis-28376 230 11 - - PUNCT bracis-28376 230 12 role role NOUN bracis-28376 230 13 dataset dataset NOUN bracis-28376 230 14 does do AUX bracis-28376 230 15 not not PART bracis-28376 230 16 have have VERB bracis-28376 230 17 a a DET bracis-28376 230 18 sufficient sufficient ADJ bracis-28376 230 19 correlation correlation NOUN bracis-28376 230 20 between between ADP bracis-28376 230 21 labels label NOUN bracis-28376 230 22 and and CCONJ bracis-28376 230 23 sentence sentence NOUN bracis-28376 230 24 positions position NOUN bracis-28376 230 25 , , PUNCT bracis-28376 230 26 which which PRON bracis-28376 230 27 nullifies nullify VERB bracis-28376 230 28 the the DET bracis-28376 230 29 usefulness usefulness NOUN bracis-28376 230 30 of of ADP bracis-28376 230 31 the the DET bracis-28376 230 32 models model NOUN bracis-28376 230 33 . . PUNCT bracis-28376 231 1 for for ADP bracis-28376 231 2 the the DET bracis-28376 231 3 4 4 NUM bracis-28376 231 4 - - PUNCT bracis-28376 231 5 role role NOUN bracis-28376 231 6 dataset dataset NOUN bracis-28376 231 7 , , PUNCT bracis-28376 231 8 we we PRON bracis-28376 231 9 believe believe VERB bracis-28376 231 10 that that SCONJ bracis-28376 231 11 the the DET bracis-28376 231 12 positional positional ADJ bracis-28376 231 13 embeddings embedding NOUN bracis-28376 231 14 employed employ VERB bracis-28376 231 15 are be AUX bracis-28376 231 16 not not PART bracis-28376 231 17 powerful powerful ADJ bracis-28376 231 18 enough enough ADV bracis-28376 231 19 to to PART bracis-28376 231 20 enrich enrich VERB bracis-28376 231 21 the the DET bracis-28376 231 22 input input NOUN bracis-28376 231 23 data datum NOUN bracis-28376 231 24 of of ADP bracis-28376 231 25 the the DET bracis-28376 231 26 models model NOUN bracis-28376 231 27 . . PUNCT bracis-28376 232 1 models model NOUN bracis-28376 232 2 based base VERB bracis-28376 232 3 on on ADP bracis-28376 232 4 the the DET bracis-28376 232 5 dfcsc dfcsc NOUN bracis-28376 232 6 approach approach NOUN bracis-28376 232 7 , , PUNCT bracis-28376 232 8 on on ADP bracis-28376 232 9 the the DET bracis-28376 232 10 other other ADJ bracis-28376 232 11 hand hand NOUN bracis-28376 232 12 , , PUNCT bracis-28376 232 13 show show VERB bracis-28376 232 14 remarkable remarkable ADJ bracis-28376 232 15 performance performance NOUN bracis-28376 232 16 . . PUNCT bracis-28376 233 1 most most ADJ bracis-28376 233 2 of of ADP bracis-28376 233 3 them they PRON bracis-28376 233 4 overcome overcome VERB bracis-28376 233 5 strong strong ADJ bracis-28376 233 6 baselines baseline NOUN bracis-28376 233 7 based base VERB bracis-28376 233 8 on on ADP bracis-28376 233 9 sentence sentence NOUN bracis-28376 233 10 chunks chunk NOUN bracis-28376 233 11 . . PUNCT bracis-28376 234 1 dfcsc dfcsc NOUN bracis-28376 234 2 models model NOUN bracis-28376 234 3 that that PRON bracis-28376 234 4 exploit exploit VERB bracis-28376 234 5 roberta roberta PROPN bracis-28376 234 6 and and CCONJ bracis-28376 234 7 longformer longformer PROPN bracis-28376 234 8 encoders encoder NOUN bracis-28376 234 9 perform perform VERB bracis-28376 234 10 similarly similarly ADV bracis-28376 234 11 or or CCONJ bracis-28376 234 12 better well ADJ bracis-28376 234 13 than than ADP bracis-28376 234 14 a a DET bracis-28376 234 15 model model NOUN bracis-28376 234 16 based base VERB bracis-28376 234 17 on on ADP bracis-28376 234 18 the the DET bracis-28376 234 19 mtl mtl PROPN bracis-28376 234 20 - - PUNCT bracis-28376 234 21 lsp lsp PROPN bracis-28376 234 22 approach approach NOUN bracis-28376 234 23 proposed propose VERB bracis-28376 234 24 in in ADP bracis-28376 234 25 [ [ X bracis-28376 234 26 15 15 NUM bracis-28376 234 27 ] ] PUNCT bracis-28376 234 28 . . PUNCT bracis-28376 235 1 from from ADP bracis-28376 235 2 the the DET bracis-28376 235 3 obtained obtain VERB bracis-28376 235 4 results result NOUN bracis-28376 235 5 , , PUNCT bracis-28376 235 6 we we PRON bracis-28376 235 7 propose propose VERB bracis-28376 235 8 a a DET bracis-28376 235 9 future future ADJ bracis-28376 235 10 work work NOUN bracis-28376 235 11 that that PRON bracis-28376 235 12 combines combine VERB bracis-28376 235 13 mtl mtl PROPN bracis-28376 235 14 - - PUNCT bracis-28376 235 15 lsp lsp PROPN bracis-28376 235 16 and and CCONJ bracis-28376 235 17 dfcsc dfcsc PROPN bracis-28376 235 18 approaches approach NOUN bracis-28376 235 19 , , PUNCT bracis-28376 235 20 since since SCONJ bracis-28376 235 21 the the DET bracis-28376 235 22 mtl mtl PROPN bracis-28376 235 23 - - PUNCT bracis-28376 235 24 lsp lsp PROPN bracis-28376 235 25 - - PUNCT bracis-28376 235 26 bilstm bilstm NOUN bracis-28376 235 27 - - PUNCT bracis-28376 235 28 bert bert NOUN bracis-28376 235 29 model model NOUN bracis-28376 235 30 does do AUX bracis-28376 235 31 not not PART bracis-28376 235 32 rely rely VERB bracis-28376 235 33 on on ADP bracis-28376 235 34 sentence sentence NOUN bracis-28376 235 35 chunks chunk NOUN bracis-28376 235 36 to to PART bracis-28376 235 37 encode encode VERB bracis-28376 235 38 sentences sentence NOUN bracis-28376 235 39 with with ADP bracis-28376 235 40 bert bert PROPN bracis-28376 235 41 . . PUNCT bracis-28376 236 1 such such ADJ bracis-28376 236 2 work work NOUN bracis-28376 236 3 would would AUX bracis-28376 236 4 provide provide VERB bracis-28376 236 5 interesting interesting ADJ bracis-28376 236 6 evidence evidence NOUN bracis-28376 236 7 about about ADP bracis-28376 236 8 the the DET bracis-28376 236 9 ability ability NOUN bracis-28376 236 10 of of ADP bracis-28376 236 11 lsp lsp PROPN bracis-28376 236 12 signals signal NOUN bracis-28376 236 13 to to PART bracis-28376 236 14 boost boost VERB bracis-28376 236 15 sentence sentence NOUN bracis-28376 236 16 chunk chunk NOUN bracis-28376 236 17 - - PUNCT bracis-28376 236 18 based base VERB bracis-28376 236 19 models model NOUN bracis-28376 236 20 . . PUNCT bracis-28376 237 1 we we PRON bracis-28376 237 2 also also ADV bracis-28376 237 3 say say VERB bracis-28376 237 4 that that SCONJ bracis-28376 237 5 the the DET bracis-28376 237 6 sentence sentence NOUN bracis-28376 237 7 classification classification NOUN bracis-28376 237 8 approaches approach NOUN bracis-28376 237 9 evaluated evaluate VERB bracis-28376 237 10 in in ADP bracis-28376 237 11 this this DET bracis-28376 237 12 work work NOUN bracis-28376 237 13 do do AUX bracis-28376 237 14 not not PART bracis-28376 237 15 rely rely VERB bracis-28376 237 16 on on ADP bracis-28376 237 17 knowledge knowledge NOUN bracis-28376 237 18 specific specific ADJ bracis-28376 237 19 to to ADP bracis-28376 237 20 the the DET bracis-28376 237 21 legal legal ADJ bracis-28376 237 22 domain domain NOUN bracis-28376 237 23 . . PUNCT bracis-28376 238 1 therefore therefore ADV bracis-28376 238 2 , , PUNCT bracis-28376 238 3 more more ADJ bracis-28376 238 4 studies study NOUN bracis-28376 238 5 should should AUX bracis-28376 238 6 be be AUX bracis-28376 238 7 conducted conduct VERB bracis-28376 238 8 to to PART bracis-28376 238 9 evaluate evaluate VERB bracis-28376 238 10 the the DET bracis-28376 238 11 efficiency efficiency NOUN bracis-28376 238 12 of of ADP bracis-28376 238 13 such such ADJ bracis-28376 238 14 approaches approach NOUN bracis-28376 238 15 in in ADP bracis-28376 238 16 other other ADJ bracis-28376 238 17 domains domain NOUN bracis-28376 238 18 . . PUNCT bracis-28376 239 1 despite despite SCONJ bracis-28376 239 2 this this PRON bracis-28376 239 3 , , PUNCT bracis-28376 239 4 we we PRON bracis-28376 239 5 believe believe VERB bracis-28376 239 6 that that SCONJ bracis-28376 239 7 incorporating incorporate VERB bracis-28376 239 8 domain domain NOUN bracis-28376 239 9 - - PUNCT bracis-28376 239 10 specific specific ADJ bracis-28376 239 11 knowledge knowledge NOUN bracis-28376 239 12 into into ADP bracis-28376 239 13 models model NOUN bracis-28376 239 14 based base VERB bracis-28376 239 15 on on ADP bracis-28376 239 16 pre pre ADJ bracis-28376 239 17 - - ADJ bracis-28376 239 18 trained train VERB bracis-28376 239 19 transformers transformer NOUN bracis-28376 239 20 is be AUX bracis-28376 239 21 a a DET bracis-28376 239 22 valuable valuable ADJ bracis-28376 239 23 research research NOUN bracis-28376 239 24 direction direction NOUN bracis-28376 239 25 . . PUNCT bracis-28376 240 1 notes note VERB bracis-28376 240 2 1.available 1.available NUM bracis-28376 240 3 at at ADP bracis-28376 240 4 https://github.com/exploration-lab/rhetorical-roles https://github.com/exploration-lab/rhetorical-role NOUN bracis-28376 240 5 . . PUNCT bracis-28376 241 1 references reference NOUN bracis-28376 241 2 aragy aragy VERB bracis-28376 241 3 , , PUNCT bracis-28376 241 4 r. r. PROPN bracis-28376 241 5 , , PUNCT bracis-28376 241 6 fernandes fernandes PROPN bracis-28376 241 7 , , PUNCT bracis-28376 241 8 e.r e.r PROPN bracis-28376 241 9 . . PROPN bracis-28376 241 10 , , PUNCT bracis-28376 241 11 caceres caceres PROPN bracis-28376 241 12 , , PUNCT bracis-28376 241 13 e.n e.n PROPN bracis-28376 241 14 . . PROPN bracis-28376 241 15 : : PUNCT bracis-28376 242 1 rhetorical rhetorical ADJ bracis-28376 242 2 role role NOUN bracis-28376 242 3 identification identification NOUN bracis-28376 242 4 for for ADP bracis-28376 242 5 portuguese portuguese ADJ bracis-28376 242 6 legal legal ADJ bracis-28376 242 7 documents document NOUN bracis-28376 242 8 . . PUNCT bracis-28376 243 1 in in ADP bracis-28376 243 2 : : PUNCT bracis-28376 243 3 britto britto PROPN bracis-28376 243 4 , , PUNCT bracis-28376 243 5 a. a. PROPN bracis-28376 243 6 , , PUNCT bracis-28376 243 7 valdivia valdivia PROPN bracis-28376 243 8 delgado delgado PROPN bracis-28376 243 9 , , PUNCT bracis-28376 243 10 k. k. PROPN bracis-28376 243 11 ( ( PUNCT bracis-28376 243 12 eds eds PROPN bracis-28376 243 13 . . PUNCT bracis-28376 243 14 ) ) PUNCT bracis-28376 243 15 bracis bracis PROPN bracis-28376 243 16 2021 2021 NUM bracis-28376 243 17 . . PUNCT bracis-28376 244 1 lncs lncs PROPN bracis-28376 244 2 ( ( PUNCT bracis-28376 244 3 lnai lnai ADJ bracis-28376 244 4 ) ) PUNCT bracis-28376 244 5 , , PUNCT bracis-28376 244 6 vol vol NOUN bracis-28376 244 7 . . PROPN bracis-28376 244 8 13074 13074 NUM bracis-28376 244 9 , , PUNCT bracis-28376 244 10 pp pp X bracis-28376 244 11 . . PUNCT bracis-28376 245 1 557–571 557–571 NUM bracis-28376 245 2 . . PUNCT bracis-28376 245 3 springer springer NOUN bracis-28376 245 4 , , PUNCT bracis-28376 245 5 cham cham PROPN bracis-28376 245 6 ( ( PUNCT bracis-28376 245 7 2021 2021 NUM bracis-28376 245 8 ) ) PUNCT bracis-28376 245 9 . . PUNCT bracis-28376 246 1 https://doi.org/10.1007/978-3-030-91699-2_38 https://doi.org/10.1007/978-3-030-91699-2_38 PROPN bracis-28376 246 2 chapter chapter NOUN bracis-28376 246 3     SPACE bracis-28376 246 4 google google PROPN bracis-28376 246 5 scholar scholar NOUN bracis-28376 246 6     SPACE bracis-28376 246 7 beltagy beltagy ADJ bracis-28376 246 8 , , PUNCT bracis-28376 246 9 i. i. NOUN bracis-28376 246 10 , , PUNCT bracis-28376 246 11 peters peters PROPN bracis-28376 246 12 , , PUNCT bracis-28376 246 13 m.e m.e PROPN bracis-28376 246 14 . . PROPN bracis-28376 246 15 , , PUNCT bracis-28376 246 16 cohan cohan PROPN bracis-28376 246 17 , , PUNCT bracis-28376 246 18 a. a. NOUN bracis-28376 246 19 : : PUNCT bracis-28376 246 20 longformer longformer NOUN bracis-28376 246 21 : : PUNCT bracis-28376 246 22 the the DET bracis-28376 246 23 long long ADJ bracis-28376 246 24 - - PUNCT bracis-28376 246 25 document document NOUN bracis-28376 246 26 transformer transformer NOUN bracis-28376 246 27 . . PUNCT bracis-28376 247 1 corr corr PROPN bracis-28376 247 2 abs/2004.05150 abs/2004.05150 PROPN bracis-28376 247 3 ( ( PUNCT bracis-28376 247 4 2020 2020 NUM bracis-28376 247 5 ) ) PUNCT bracis-28376 247 6 . . PUNCT bracis-28376 248 1 https://arxiv.org/abs/2004.05150 https://arxiv.org/abs/2004.05150 PROPN bracis-28376 248 2 bhattacharya bhattacharya PROPN bracis-28376 248 3 , , PUNCT bracis-28376 248 4 p. p. PROPN bracis-28376 248 5 , , PUNCT bracis-28376 248 6 paul paul PROPN bracis-28376 248 7 , , PUNCT bracis-28376 248 8 s. s. PROPN bracis-28376 248 9 , , PUNCT bracis-28376 248 10 ghosh ghosh PROPN bracis-28376 248 11 , , PUNCT bracis-28376 248 12 k. k. PROPN bracis-28376 248 13 , , PUNCT bracis-28376 248 14 ghosh ghosh PROPN bracis-28376 248 15 , , PUNCT bracis-28376 248 16 s. s. PROPN bracis-28376 248 17 , , PUNCT bracis-28376 248 18 wyner wyner NOUN bracis-28376 248 19 , , PUNCT bracis-28376 248 20 a. a. NOUN bracis-28376 248 21 : : PUNCT bracis-28376 248 22 deeprhole deeprhole NOUN bracis-28376 248 23 : : PUNCT bracis-28376 248 24 deep deep ADJ bracis-28376 248 25 learning learning NOUN bracis-28376 248 26 for for ADP bracis-28376 248 27 rhetorical rhetorical ADJ bracis-28376 248 28 role role NOUN bracis-28376 248 29 labeling labeling NOUN bracis-28376 248 30 of of ADP bracis-28376 248 31 sentences sentence NOUN bracis-28376 248 32 in in ADP bracis-28376 248 33 legal legal ADJ bracis-28376 248 34 case case NOUN bracis-28376 248 35 documents document NOUN bracis-28376 248 36 . . PUNCT bracis-28376 249 1 artificial artificial ADJ bracis-28376 249 2 intelligence intelligence NOUN bracis-28376 249 3 and and CCONJ bracis-28376 249 4 law law NOUN bracis-28376 249 5 , , PUNCT bracis-28376 249 6 pp pp ADP bracis-28376 249 7 . . PUNCT bracis-28376 250 1 1–38 1–38 PROPN bracis-28376 250 2 ( ( PUNCT bracis-28376 250 3 2021 2021 NUM bracis-28376 250 4 ) ) PUNCT bracis-28376 250 5 google google PROPN bracis-28376 250 6 scholar scholar NOUN bracis-28376 250 7     SPACE bracis-28376 250 8 cohan cohan PROPN bracis-28376 250 9 , , PUNCT bracis-28376 250 10 a. a. PROPN bracis-28376 250 11 , , PUNCT bracis-28376 250 12 beltagy beltagy NUM bracis-28376 250 13 , , PUNCT bracis-28376 250 14 i. i. NOUN bracis-28376 250 15 , , PUNCT bracis-28376 250 16 king king PROPN bracis-28376 250 17 , , PUNCT bracis-28376 250 18 d. d. PROPN bracis-28376 250 19 , , PUNCT bracis-28376 250 20 dalvi dalvi PROPN bracis-28376 250 21 , , PUNCT bracis-28376 250 22 b. b. PROPN bracis-28376 250 23 , , PUNCT bracis-28376 250 24 weld weld PROPN bracis-28376 250 25 , , PUNCT bracis-28376 250 26 d. d. PROPN bracis-28376 250 27 : : PUNCT bracis-28376 250 28 pretrained pretraine VERB bracis-28376 250 29 language language NOUN bracis-28376 250 30 models model NOUN bracis-28376 250 31 for for ADP bracis-28376 250 32 sequential sequential ADJ bracis-28376 250 33 sentence sentence NOUN bracis-28376 250 34 classification classification NOUN bracis-28376 250 35 . . PUNCT bracis-28376 251 1 in in ADP bracis-28376 251 2 : : PUNCT bracis-28376 251 3 proceedings proceeding NOUN bracis-28376 251 4 of of ADP bracis-28376 251 5 the the DET bracis-28376 251 6 2019 2019 NUM bracis-28376 251 7 conference conference NOUN bracis-28376 251 8 on on ADP bracis-28376 251 9 empirical empirical ADJ bracis-28376 251 10 methods method NOUN bracis-28376 251 11 in in ADP bracis-28376 251 12 natural natural ADJ bracis-28376 251 13 language language NOUN bracis-28376 251 14 processing processing NOUN bracis-28376 251 15 and and CCONJ bracis-28376 251 16 the the DET bracis-28376 251 17 9th 9th ADJ bracis-28376 251 18 international international ADJ bracis-28376 251 19 joint joint ADJ bracis-28376 251 20 conference conference NOUN bracis-28376 251 21 on on ADP bracis-28376 251 22 natural natural ADJ bracis-28376 251 23 language language NOUN bracis-28376 251 24 processing processing NOUN bracis-28376 251 25 ( ( PUNCT bracis-28376 251 26 emnlp emnlp NOUN bracis-28376 251 27 - - PUNCT bracis-28376 251 28 ijcnlp ijcnlp NOUN bracis-28376 251 29 ) ) PUNCT bracis-28376 251 30 , , PUNCT bracis-28376 251 31 pp pp PROPN bracis-28376 251 32 . . PUNCT bracis-28376 252 1 3693–3699 3693–3699 NUM bracis-28376 252 2 . . PUNCT bracis-28376 252 3 association association NOUN bracis-28376 252 4 for for ADP bracis-28376 252 5 computational computational ADJ bracis-28376 252 6 linguistics linguistic NOUN bracis-28376 252 7 , , PUNCT bracis-28376 252 8 hong hong PROPN bracis-28376 252 9 kong kong PROPN bracis-28376 252 10 , , PUNCT bracis-28376 252 11 november november PROPN bracis-28376 252 12 2019 2019 NUM bracis-28376 252 13 . . PUNCT bracis-28376 253 1 https://doi.org/10.18653/v1/d19-1383 https://doi.org/10.18653/v1/d19-1383 NOUN bracis-28376 253 2 . . PUNCT bracis-28376 254 1 https://aclanthology.org/d19-1383 https://aclanthology.org/d19-1383 PROPN bracis-28376 254 2 devlin devlin PROPN bracis-28376 254 3 , , PUNCT bracis-28376 254 4 j. j. PROPN bracis-28376 254 5 , , PUNCT bracis-28376 254 6 chang chang PROPN bracis-28376 254 7 , , PUNCT bracis-28376 254 8 m.w m.w PROPN bracis-28376 254 9 . . PROPN bracis-28376 254 10 , , PUNCT bracis-28376 254 11 lee lee PROPN bracis-28376 254 12 , , PUNCT bracis-28376 254 13 k. k. PROPN bracis-28376 254 14 , , PUNCT bracis-28376 254 15 toutanova toutanova PROPN bracis-28376 254 16 , , PUNCT bracis-28376 254 17 k. k. PROPN bracis-28376 254 18 : : PUNCT bracis-28376 255 1 bert bert PROPN bracis-28376 255 2 : : PUNCT bracis-28376 255 3 pre pre ADJ bracis-28376 255 4 - - NOUN bracis-28376 255 5 training training NOUN bracis-28376 255 6 of of ADP bracis-28376 255 7 deep deep ADJ bracis-28376 255 8 bidirectional bidirectional ADJ bracis-28376 255 9 transformers transformer NOUN bracis-28376 255 10 for for ADP bracis-28376 255 11 language language NOUN bracis-28376 255 12 understanding understanding NOUN bracis-28376 255 13 . . PUNCT bracis-28376 256 1 in in ADP bracis-28376 256 2 : : PUNCT bracis-28376 256 3 proceedings proceeding NOUN bracis-28376 256 4 of of ADP bracis-28376 256 5 the the DET bracis-28376 256 6 2019 2019 NUM bracis-28376 256 7 conference conference NOUN bracis-28376 256 8 of of ADP bracis-28376 256 9 the the DET bracis-28376 256 10 north north ADJ bracis-28376 256 11 american american ADJ bracis-28376 256 12 chapter chapter NOUN bracis-28376 256 13 of of ADP bracis-28376 256 14 the the DET bracis-28376 256 15 association association NOUN bracis-28376 256 16 for for ADP bracis-28376 256 17 computational computational ADJ bracis-28376 256 18 linguistics linguistic NOUN bracis-28376 256 19 : : PUNCT bracis-28376 256 20 human human ADJ bracis-28376 256 21 language language NOUN bracis-28376 256 22 technologies technology NOUN bracis-28376 256 23 , , PUNCT bracis-28376 256 24 volume volume NOUN bracis-28376 256 25 1 1 NUM bracis-28376 256 26 ( ( PUNCT bracis-28376 256 27 long long ADJ bracis-28376 256 28 and and CCONJ bracis-28376 256 29 short short ADJ bracis-28376 256 30 papers paper NOUN bracis-28376 256 31 ) ) PUNCT bracis-28376 256 32 , , PUNCT bracis-28376 256 33 pp pp ADV bracis-28376 256 34 . . PUNCT bracis-28376 257 1 4171–4186 4171–4186 NUM bracis-28376 257 2 . . PUNCT bracis-28376 258 1 association association NOUN bracis-28376 258 2 for for ADP bracis-28376 258 3 computational computational ADJ bracis-28376 258 4 linguistics linguistic NOUN bracis-28376 258 5 , , PUNCT bracis-28376 258 6 minneapolis minneapolis PROPN bracis-28376 258 7 , , PUNCT bracis-28376 258 8 minnesota minnesota PROPN bracis-28376 258 9 , , PUNCT bracis-28376 258 10 june june PROPN bracis-28376 258 11 2019 2019 NUM bracis-28376 258 12 . . PUNCT bracis-28376 259 1 https://doi.org/10.18653/v1/n19-1423 https://doi.org/10.18653/v1/n19-1423 NOUN bracis-28376 259 2 . . PUNCT bracis-28376 259 3 https://aclanthology.org/n19-1423 https://aclanthology.org/n19-1423 PROPN bracis-28376 259 4 dror dror PROPN bracis-28376 259 5 , , PUNCT bracis-28376 259 6 r. r. PROPN bracis-28376 259 7 , , PUNCT bracis-28376 259 8 shlomov shlomov PROPN bracis-28376 259 9 , , PUNCT bracis-28376 259 10 s. s. PROPN bracis-28376 259 11 , , PUNCT bracis-28376 259 12 reichart reichart PROPN bracis-28376 259 13 , , PUNCT bracis-28376 259 14 r. r. PROPN bracis-28376 259 15 : : PUNCT bracis-28376 259 16 deep deep ADJ bracis-28376 259 17 dominance dominance NOUN bracis-28376 259 18 how how SCONJ bracis-28376 259 19 to to PART bracis-28376 259 20 properly properly ADV bracis-28376 259 21 compare compare VERB bracis-28376 259 22 deep deep ADJ bracis-28376 259 23 neural neural ADJ bracis-28376 259 24 models model NOUN bracis-28376 259 25 . . PUNCT bracis-28376 260 1 in in ADP bracis-28376 260 2 : : PUNCT bracis-28376 260 3 korhonen korhonen PROPN bracis-28376 260 4 , , PUNCT bracis-28376 260 5 a. a. PROPN bracis-28376 260 6 , , PUNCT bracis-28376 260 7 traum traum PROPN bracis-28376 260 8 , , PUNCT bracis-28376 260 9 d.r d.r PROPN bracis-28376 260 10 . . PROPN bracis-28376 260 11 , , PUNCT bracis-28376 260 12 màrquez màrquez PROPN bracis-28376 260 13 , , PUNCT bracis-28376 260 14 l. l. PROPN bracis-28376 260 15 ( ( PUNCT bracis-28376 260 16 eds eds PROPN bracis-28376 260 17 . . PUNCT bracis-28376 260 18 ) ) PUNCT bracis-28376 260 19 proceedings proceeding NOUN bracis-28376 260 20 of of ADP bracis-28376 260 21 the the DET bracis-28376 260 22 57th 57th ADJ bracis-28376 260 23 conference conference NOUN bracis-28376 260 24 of of ADP bracis-28376 260 25 the the DET bracis-28376 260 26 association association NOUN bracis-28376 260 27 for for ADP bracis-28376 260 28 computational computational ADJ bracis-28376 260 29 linguistics linguistic NOUN bracis-28376 260 30 , , PUNCT bracis-28376 260 31 acl acl PROPN bracis-28376 260 32 2019 2019 NUM bracis-28376 260 33 , , PUNCT bracis-28376 260 34 florence florence NOUN bracis-28376 260 35 , , PUNCT bracis-28376 260 36 italy italy PROPN bracis-28376 260 37 , , PUNCT bracis-28376 260 38 july july PROPN bracis-28376 260 39 28 28 NUM bracis-28376 260 40 – – PUNCT bracis-28376 260 41 august august PROPN bracis-28376 260 42 2 2 NUM bracis-28376 260 43 , , PUNCT bracis-28376 260 44 2019 2019 NUM bracis-28376 260 45 , , PUNCT bracis-28376 260 46 volume volume NOUN bracis-28376 260 47 1 1 NUM bracis-28376 260 48 : : PUNCT bracis-28376 260 49 long long ADJ bracis-28376 260 50 papers paper NOUN bracis-28376 260 51 , , PUNCT bracis-28376 260 52 pp pp ADV bracis-28376 260 53 . . PUNCT bracis-28376 260 54 2773–2785 2773–2785 NUM bracis-28376 260 55 . . PUNCT bracis-28376 261 1 association association NOUN bracis-28376 261 2 for for ADP bracis-28376 261 3 computational computational ADJ bracis-28376 261 4 linguistics linguistic NOUN bracis-28376 261 5 ( ( PUNCT bracis-28376 261 6 2019 2019 NUM bracis-28376 261 7 ) ) PUNCT bracis-28376 261 8 . . PUNCT bracis-28376 262 1 https://doi.org/10.18653/v1/p19-1266 https://doi.org/10.18653/v1/p19-1266 NOUN bracis-28376 262 2 . . PUNCT bracis-28376 263 1 https://doi.org/10.18653/v1/p19-1266 https://doi.org/10.18653/v1/p19-1266 NOUN bracis-28376 263 2 feijo feijo PROPN bracis-28376 263 3 , , PUNCT bracis-28376 263 4 d. d. PROPN bracis-28376 263 5 , , PUNCT bracis-28376 263 6 moreira moreira PROPN bracis-28376 263 7 , , PUNCT bracis-28376 263 8 v. v. ADV bracis-28376 263 9 : : PUNCT bracis-28376 263 10 summarizing summarize VERB bracis-28376 263 11 legal legal ADJ bracis-28376 263 12 rulings ruling NOUN bracis-28376 263 13 : : PUNCT bracis-28376 263 14 comparative comparative ADJ bracis-28376 263 15 experiments experiment NOUN bracis-28376 263 16 . . PUNCT bracis-28376 264 1 in in ADP bracis-28376 264 2 : : PUNCT bracis-28376 264 3 proceedings proceeding NOUN bracis-28376 264 4 of of ADP bracis-28376 264 5 the the DET bracis-28376 264 6 international international ADJ bracis-28376 264 7 conference conference NOUN bracis-28376 264 8 on on ADP bracis-28376 264 9 recent recent ADJ bracis-28376 264 10 advances advance NOUN bracis-28376 264 11 in in ADP bracis-28376 264 12 natural natural ADJ bracis-28376 264 13 language language NOUN bracis-28376 264 14 processing processing NOUN bracis-28376 264 15 ( ( PUNCT bracis-28376 264 16 ranlp ranlp PROPN bracis-28376 264 17 2019 2019 NUM bracis-28376 264 18 ) ) PUNCT bracis-28376 264 19 , , PUNCT bracis-28376 264 20 pp pp ADP bracis-28376 264 21 . . PUNCT bracis-28376 265 1 313–322 313–322 NUM bracis-28376 265 2 . . PUNCT bracis-28376 265 3 incoma incoma PROPN bracis-28376 265 4 ltd ltd PROPN bracis-28376 265 5 . . PROPN bracis-28376 265 6 , , PUNCT bracis-28376 265 7 varna varna PROPN bracis-28376 265 8 , , PUNCT bracis-28376 265 9 bulgaria bulgaria PROPN bracis-28376 265 10 , , PUNCT bracis-28376 265 11 september september PROPN bracis-28376 265 12 2019 2019 NUM bracis-28376 265 13 . . PUNCT bracis-28376 266 1 https://doi.org/10.26615/978-954-452-056-4_036 https://doi.org/10.26615/978-954-452-056-4_036 PROPN bracis-28376 266 2 . . PUNCT bracis-28376 267 1 http://aclanthology.org/r19-1036 http://aclanthology.org/r19-1036 PROPN bracis-28376 267 2 kalamkar kalamkar PROPN bracis-28376 267 3 , , PUNCT bracis-28376 267 4 p. p. PROPN bracis-28376 267 5 , , PUNCT bracis-28376 267 6 et et PROPN bracis-28376 267 7 al al PROPN bracis-28376 267 8 . . PUNCT bracis-28376 267 9 : : PUNCT bracis-28376 268 1 corpus corpu NOUN bracis-28376 268 2 for for ADP bracis-28376 268 3 automatic automatic ADJ bracis-28376 268 4 structuring structuring NOUN bracis-28376 268 5 of of ADP bracis-28376 268 6 legal legal ADJ bracis-28376 268 7 documents document NOUN bracis-28376 268 8 . . PUNCT bracis-28376 269 1 in in ADP bracis-28376 269 2 : : PUNCT bracis-28376 269 3 proceedings proceeding NOUN bracis-28376 269 4 of of ADP bracis-28376 269 5 the the DET bracis-28376 269 6 thirteenth thirteenth ADJ bracis-28376 269 7 language language NOUN bracis-28376 269 8 resources resource NOUN bracis-28376 269 9 and and CCONJ bracis-28376 269 10 evaluation evaluation NOUN bracis-28376 269 11 conference conference NOUN bracis-28376 269 12 , , PUNCT bracis-28376 269 13 pp pp PROPN bracis-28376 269 14 . . PUNCT bracis-28376 269 15 4420–4429 4420–4429 NUM bracis-28376 269 16 . . PUNCT bracis-28376 270 1 european european ADJ bracis-28376 270 2 language language PROPN bracis-28376 270 3 resources resource NOUN bracis-28376 270 4 association association PROPN bracis-28376 270 5 , , PUNCT bracis-28376 270 6 marseille marseille PROPN bracis-28376 270 7 , , PUNCT bracis-28376 270 8 france france PROPN bracis-28376 270 9 , , PUNCT bracis-28376 270 10 june june PROPN bracis-28376 270 11 2022 2022 NUM bracis-28376 270 12 . . PUNCT bracis-28376 271 1 https://aclanthology.org/2022.lrec-1.470 https://aclanthology.org/2022.lrec-1.470 PROPN bracis-28376 271 2 li li PROPN bracis-28376 271 3 , , PUNCT bracis-28376 271 4 d. d. PROPN bracis-28376 271 5 , , PUNCT bracis-28376 271 6 yang yang PROPN bracis-28376 271 7 , , PUNCT bracis-28376 271 8 k. k. PROPN bracis-28376 271 9 , , PUNCT bracis-28376 271 10 zhang zhang PROPN bracis-28376 271 11 , , PUNCT bracis-28376 271 12 l. l. PROPN bracis-28376 271 13 , , PUNCT bracis-28376 271 14 yin yin PROPN bracis-28376 271 15 , , PUNCT bracis-28376 271 16 d. d. PROPN bracis-28376 271 17 , , PUNCT bracis-28376 271 18 peng peng PROPN bracis-28376 271 19 , , PUNCT bracis-28376 271 20 d. d. PROPN bracis-28376 271 21 : : PUNCT bracis-28376 271 22 class class NOUN bracis-28376 271 23 : : PUNCT bracis-28376 271 24 a a DET bracis-28376 271 25 novel novel ADJ bracis-28376 271 26 method method NOUN bracis-28376 271 27 for for ADP bracis-28376 271 28 chinese chinese ADJ bracis-28376 271 29 legal legal ADJ bracis-28376 271 30 judgments judgment NOUN bracis-28376 271 31 summarization summarization NOUN bracis-28376 271 32 . . PUNCT bracis-28376 272 1 in in ADP bracis-28376 272 2 : : PUNCT bracis-28376 272 3 proceedings proceeding NOUN bracis-28376 272 4 of of ADP bracis-28376 272 5 the the DET bracis-28376 272 6 5th 5th ADJ bracis-28376 272 7 international international ADJ bracis-28376 272 8 conference conference NOUN bracis-28376 272 9 on on ADP bracis-28376 272 10 computer computer NOUN bracis-28376 272 11 science science NOUN bracis-28376 272 12 and and CCONJ bracis-28376 272 13 application application NOUN bracis-28376 272 14 engineering engineering NOUN bracis-28376 272 15 . . PUNCT bracis-28376 273 1 csae csae VERB bracis-28376 273 2 2021 2021 NUM bracis-28376 273 3 . . PUNCT bracis-28376 274 1 association association NOUN bracis-28376 274 2 for for ADP bracis-28376 274 3 computing compute VERB bracis-28376 274 4 machinery machinery NOUN bracis-28376 274 5 , , PUNCT bracis-28376 274 6 new new PROPN bracis-28376 274 7 york york PROPN bracis-28376 274 8 ( ( PUNCT bracis-28376 274 9 2021 2021 NUM bracis-28376 274 10 ) ) PUNCT bracis-28376 274 11 . . PUNCT bracis-28376 275 1 https://doi.org/10.1145/3487075.3487161 https://doi.org/10.1145/3487075.3487161 VERB bracis-28376 275 2 de de X bracis-28376 275 3 lima lima PROPN bracis-28376 275 4 , , PUNCT bracis-28376 275 5 a.g a.g PROPN bracis-28376 275 6 . . PROPN bracis-28376 275 7 , , PUNCT bracis-28376 275 8 boughanem boughanem PROPN bracis-28376 275 9 , , PUNCT bracis-28376 275 10 m. m. NOUN bracis-28376 275 11 , , PUNCT bracis-28376 275 12 da da PROPN bracis-28376 275 13 s. s. PROPN bracis-28376 275 14 aranha aranha PROPN bracis-28376 275 15 , , PUNCT bracis-28376 275 16 e.h e.h PROPN bracis-28376 275 17 . . PROPN bracis-28376 275 18 , , PUNCT bracis-28376 275 19 dkaki dkaki PROPN bracis-28376 275 20 , , PUNCT bracis-28376 275 21 t. t. PROPN bracis-28376 275 22 , , PUNCT bracis-28376 275 23 moreno moreno PROPN bracis-28376 275 24 , , PUNCT bracis-28376 275 25 j.g j.g PROPN bracis-28376 275 26 . . PROPN bracis-28376 275 27 : : PUNCT bracis-28376 276 1 exploring explore VERB bracis-28376 276 2 sbert sbert NOUN bracis-28376 276 3 and and CCONJ bracis-28376 276 4 mixup mixup NOUN bracis-28376 276 5 data datum NOUN bracis-28376 276 6 augmentation augmentation NOUN bracis-28376 276 7 in in ADP bracis-28376 276 8 rhetorical rhetorical ADJ bracis-28376 276 9 role role NOUN bracis-28376 276 10 labeling labeling NOUN bracis-28376 276 11 of of ADP bracis-28376 276 12 indian indian ADJ bracis-28376 276 13 legal legal ADJ bracis-28376 276 14 sentences sentence NOUN bracis-28376 276 15 . . PUNCT bracis-28376 277 1 in in ADP bracis-28376 277 2 : : PUNCT bracis-28376 277 3 tamine tamine PROPN bracis-28376 277 4 , , PUNCT bracis-28376 277 5 l. l. PROPN bracis-28376 277 6 , , PUNCT bracis-28376 277 7 amigó amigó NOUN bracis-28376 277 8 , , PUNCT bracis-28376 277 9 e. e. PROPN bracis-28376 277 10 , , PUNCT bracis-28376 277 11 mothe mothe PROPN bracis-28376 277 12 , , PUNCT bracis-28376 277 13 j. j. PROPN bracis-28376 277 14 ( ( PUNCT bracis-28376 277 15 eds eds PROPN bracis-28376 277 16 . . PUNCT bracis-28376 277 17 ) ) PUNCT bracis-28376 277 18 proceedings proceeding NOUN bracis-28376 277 19 of of ADP bracis-28376 277 20 the the DET bracis-28376 277 21 2nd 2nd ADJ bracis-28376 277 22 joint joint ADJ bracis-28376 277 23 conference conference NOUN bracis-28376 277 24 of of ADP bracis-28376 277 25 the the DET bracis-28376 277 26 information information NOUN bracis-28376 277 27 retrieval retrieval NOUN bracis-28376 277 28 communities community NOUN bracis-28376 277 29 in in ADP bracis-28376 277 30 europe europe PROPN bracis-28376 277 31 ( ( PUNCT bracis-28376 277 32 circle circle NOUN bracis-28376 277 33 2022 2022 NUM bracis-28376 277 34 ) ) PUNCT bracis-28376 277 35 , , PUNCT bracis-28376 277 36 samatan samatan ADJ bracis-28376 277 37 , , PUNCT bracis-28376 277 38 gers ger NOUN bracis-28376 277 39 , , PUNCT bracis-28376 277 40 france france PROPN bracis-28376 277 41 , , PUNCT bracis-28376 277 42 july july PROPN bracis-28376 277 43 4–7 4–7 PROPN bracis-28376 277 44 , , PUNCT bracis-28376 277 45 2022 2022 NUM bracis-28376 277 46 . . PUNCT bracis-28376 278 1 ceur ceur NOUN bracis-28376 278 2 workshop workshop NOUN bracis-28376 278 3 proceedings proceeding NOUN bracis-28376 278 4 , , PUNCT bracis-28376 278 5 vol vol NOUN bracis-28376 278 6 . . PROPN bracis-28376 278 7 3178 3178 NUM bracis-28376 278 8 . . PUNCT bracis-28376 279 1 ceur-ws.org ceur-ws.org PROPN bracis-28376 279 2 ( ( PUNCT bracis-28376 279 3 2022 2022 NUM bracis-28376 279 4 ) ) PUNCT bracis-28376 279 5 . . PUNCT bracis-28376 280 1 https://ceur-ws.org/vol-3178/circle_2022_paper_29.pdf https://ceur-ws.org/vol-3178/circle_2022_paper_29.pdf PROPN bracis-28376 280 2 de de X bracis-28376 280 3 lima lima PROPN bracis-28376 280 4 , , PUNCT bracis-28376 280 5 a.g a.g PROPN bracis-28376 280 6 . . PROPN bracis-28376 280 7 , , PUNCT bracis-28376 280 8 moreno moreno PROPN bracis-28376 280 9 , , PUNCT bracis-28376 280 10 j.g j.g PROPN bracis-28376 280 11 . . PROPN bracis-28376 280 12 , , PUNCT bracis-28376 280 13 da da PROPN bracis-28376 280 14 s. s. PROPN bracis-28376 280 15 aranha aranha PROPN bracis-28376 280 16 , , PUNCT bracis-28376 280 17 e.h e.h PROPN bracis-28376 280 18 . . PROPN bracis-28376 280 19 : : PUNCT bracis-28376 281 1 irit_iris_a irit_iris_a NOUN bracis-28376 281 2 at at ADP bracis-28376 281 3 semeval-2023 semeval-2023 ADJ bracis-28376 281 4 task task NOUN bracis-28376 281 5 6 6 NUM bracis-28376 281 6 : : PUNCT bracis-28376 281 7 legal legal ADJ bracis-28376 281 8 rhetorical rhetorical ADJ bracis-28376 281 9 role role NOUN bracis-28376 281 10 labeling labeling NOUN bracis-28376 281 11 supported support VERB bracis-28376 281 12 by by ADP bracis-28376 281 13 dynamic dynamic ADJ bracis-28376 281 14 - - PUNCT bracis-28376 281 15 filled fill VERB bracis-28376 281 16 contextualized contextualized ADJ bracis-28376 281 17 sentence sentence NOUN bracis-28376 281 18 chunks chunk NOUN bracis-28376 281 19 . . PUNCT bracis-28376 282 1 in in ADP bracis-28376 282 2 : : PUNCT bracis-28376 282 3 proceedings proceeding NOUN bracis-28376 282 4 of of ADP bracis-28376 282 5 the the DET bracis-28376 282 6 17th 17th ADJ bracis-28376 282 7 international international ADJ bracis-28376 282 8 workshop workshop NOUN bracis-28376 282 9 on on ADP bracis-28376 282 10 semantic semantic ADJ bracis-28376 282 11 evaluation evaluation NOUN bracis-28376 282 12 ( ( PUNCT bracis-28376 282 13 semeval-2023 semeval-2023 NOUN bracis-28376 282 14 ) ) PUNCT bracis-28376 282 15 . . PUNCT bracis-28376 283 1 association association NOUN bracis-28376 283 2 for for ADP bracis-28376 283 3 computational computational ADJ bracis-28376 283 4 linguistics linguistic NOUN bracis-28376 283 5 , , PUNCT bracis-28376 283 6 toronto toronto PROPN bracis-28376 283 7 , , PUNCT bracis-28376 283 8 canada canada PROPN bracis-28376 283 9 , , PUNCT bracis-28376 283 10 july july PROPN bracis-28376 283 11 2023 2023 NUM bracis-28376 283 12 google google PROPN bracis-28376 283 13 scholar scholar NOUN bracis-28376 283 14     SPACE bracis-28376 283 15 de de X bracis-28376 283 16 lima lima PROPN bracis-28376 283 17 , , PUNCT bracis-28376 283 18 a.g a.g PROPN bracis-28376 283 19 . . PROPN bracis-28376 283 20 , , PUNCT bracis-28376 283 21 moreno moreno PROPN bracis-28376 283 22 , , PUNCT bracis-28376 283 23 j.g j.g PROPN bracis-28376 283 24 . . PROPN bracis-28376 283 25 , , PUNCT bracis-28376 283 26 boughanem boughanem PROPN bracis-28376 283 27 , , PUNCT bracis-28376 283 28 m. m. NOUN bracis-28376 283 29 , , PUNCT bracis-28376 283 30 dkaki dkaki NOUN bracis-28376 283 31 , , PUNCT bracis-28376 283 32 t. t. PROPN bracis-28376 283 33 , , PUNCT bracis-28376 283 34 da da PROPN bracis-28376 283 35 s. s. PROPN bracis-28376 283 36 aranha aranha PROPN bracis-28376 283 37 , , PUNCT bracis-28376 283 38 e.h e.h PROPN bracis-28376 283 39 . . PROPN bracis-28376 283 40 : : PUNCT bracis-28376 283 41 leveraging leverage VERB bracis-28376 283 42 positional positional ADJ bracis-28376 283 43 encoding encoding NOUN bracis-28376 283 44 to to PART bracis-28376 283 45 improve improve VERB bracis-28376 283 46 fact fact NOUN bracis-28376 283 47 identification identification NOUN bracis-28376 283 48 in in ADP bracis-28376 283 49 legal legal ADJ bracis-28376 283 50 documents document NOUN bracis-28376 283 51 . . PUNCT bracis-28376 284 1 in in ADP bracis-28376 284 2 : : PUNCT bracis-28376 284 3 first first ADJ bracis-28376 284 4 international international ADJ bracis-28376 284 5 workshop workshop NOUN bracis-28376 284 6 on on ADP bracis-28376 284 7 legal legal ADJ bracis-28376 284 8 information information NOUN bracis-28376 284 9 retrieval retrieval NOUN bracis-28376 284 10 ( ( PUNCT bracis-28376 284 11 legalir legalir NOUN bracis-28376 284 12 ) ) PUNCT bracis-28376 284 13 at at ADP bracis-28376 284 14 ecir ecir NOUN bracis-28376 284 15 2023 2023 NUM bracis-28376 284 16 , , PUNCT bracis-28376 284 17 pp pp ADJ bracis-28376 284 18 . . PUNCT bracis-28376 284 19 11–13 11–13 NUM bracis-28376 284 20 ( ( PUNCT bracis-28376 284 21 2023 2023 NUM bracis-28376 284 22 ) ) PUNCT bracis-28376 284 23 . . PUNCT bracis-28376 285 1 https://tmr.liacs.nl/legalir/legalir2023_proceedings.pdf https://tmr.liacs.nl/legalir/legalir2023_proceedings.pdf PROPN bracis-28376 285 2 liu liu PROPN bracis-28376 285 3 , , PUNCT bracis-28376 285 4 y. y. PROPN bracis-28376 285 5 , , PUNCT bracis-28376 285 6 et et PROPN bracis-28376 285 7 al al PROPN bracis-28376 285 8 . . PUNCT bracis-28376 285 9 : : PUNCT bracis-28376 286 1 roberta roberta NOUN bracis-28376 286 2 : : PUNCT bracis-28376 286 3 a a DET bracis-28376 286 4 robustly robustly ADV bracis-28376 286 5 optimized optimize VERB bracis-28376 286 6 bert bert NOUN bracis-28376 286 7 pretraining pretraine VERB bracis-28376 286 8 approach approach NOUN bracis-28376 286 9 . . PUNCT bracis-28376 287 1 corr corr PROPN bracis-28376 287 2 abs/1907.11692 abs/1907.11692 PROPN bracis-28376 287 3 ( ( PUNCT bracis-28376 287 4 2019 2019 NUM bracis-28376 287 5 ) ) PUNCT bracis-28376 287 6 . . PUNCT bracis-28376 288 1 https://arxiv.org/abs/1907.11692 https://arxiv.org/abs/1907.11692 PROPN bracis-28376 288 2 ma ma PROPN bracis-28376 288 3 , , PUNCT bracis-28376 288 4 l. l. PROPN bracis-28376 288 5 , , PUNCT bracis-28376 288 6 zhang zhang PROPN bracis-28376 288 7 , , PUNCT bracis-28376 288 8 y. y. PROPN bracis-28376 288 9 , , PUNCT bracis-28376 288 10 wang wang PROPN bracis-28376 288 11 , , PUNCT bracis-28376 288 12 t. t. PROPN bracis-28376 288 13 , , PUNCT bracis-28376 288 14 liu liu PROPN bracis-28376 288 15 , , PUNCT bracis-28376 288 16 x. x. PROPN bracis-28376 288 17 , , PUNCT bracis-28376 288 18 ye ye PROPN bracis-28376 288 19 , , PUNCT bracis-28376 288 20 w. w. PROPN bracis-28376 288 21 , , PUNCT bracis-28376 288 22 sun sun PROPN bracis-28376 288 23 , , PUNCT bracis-28376 288 24 c. c. PROPN bracis-28376 288 25 , , PUNCT bracis-28376 288 26 zhang zhang PROPN bracis-28376 288 27 , , PUNCT bracis-28376 288 28 s. s. PROPN bracis-28376 288 29 : : PUNCT bracis-28376 288 30 legal legal ADJ bracis-28376 288 31 judgment judgment NOUN bracis-28376 288 32 prediction prediction NOUN bracis-28376 288 33 with with ADP bracis-28376 288 34 multi multi ADJ bracis-28376 288 35 - - ADJ bracis-28376 288 36 stage stage NOUN bracis-28376 288 37 case case NOUN bracis-28376 288 38 representation representation NOUN bracis-28376 288 39 learning learning NOUN bracis-28376 288 40 in in ADP bracis-28376 288 41 the the DET bracis-28376 288 42 real real ADJ bracis-28376 288 43 court court NOUN bracis-28376 288 44 setting setting NOUN bracis-28376 288 45 . . PUNCT bracis-28376 289 1 in in ADP bracis-28376 289 2 : : PUNCT bracis-28376 289 3 proceedings proceeding NOUN bracis-28376 289 4 of of ADP bracis-28376 289 5 the the DET bracis-28376 289 6 44th 44th ADJ bracis-28376 289 7 international international PROPN bracis-28376 289 8 acm acm PROPN bracis-28376 289 9 sigir sigir PROPN bracis-28376 289 10 conference conference PROPN bracis-28376 289 11 on on ADP bracis-28376 289 12 research research NOUN bracis-28376 289 13 and and CCONJ bracis-28376 289 14 development development NOUN bracis-28376 289 15 in in ADP bracis-28376 289 16 information information NOUN bracis-28376 289 17 retrieval retrieval NOUN bracis-28376 289 18 , , PUNCT bracis-28376 289 19 pp pp ADP bracis-28376 289 20 . . PUNCT bracis-28376 289 21 993–1002 993–1002 NUM bracis-28376 289 22 . . PUNCT bracis-28376 289 23 sigir sigir PROPN bracis-28376 289 24 ’ ' PUNCT bracis-28376 289 25 21 21 NUM bracis-28376 289 26 . . PUNCT bracis-28376 289 27 association association NOUN bracis-28376 289 28 for for ADP bracis-28376 289 29 computing compute VERB bracis-28376 289 30 machinery machinery NOUN bracis-28376 289 31 , , PUNCT bracis-28376 289 32 new new PROPN bracis-28376 289 33 york york PROPN bracis-28376 289 34 ( ( PUNCT bracis-28376 289 35 2021 2021 NUM bracis-28376 289 36 ) ) PUNCT bracis-28376 289 37 . . PUNCT bracis-28376 290 1 https://doi.org/10.1145/3404835.3462945 https://doi.org/10.1145/3404835.3462945 PROPN bracis-28376 290 2 malik malik PROPN bracis-28376 290 3 , , PUNCT bracis-28376 290 4 v. v. PROPN bracis-28376 290 5 , , PUNCT bracis-28376 290 6 sanjay sanjay PROPN bracis-28376 290 7 , , PUNCT bracis-28376 290 8 r. r. PROPN bracis-28376 290 9 , , PUNCT bracis-28376 290 10 guha guha PROPN bracis-28376 290 11 , , PUNCT bracis-28376 290 12 s.k s.k PROPN bracis-28376 290 13 . . PROPN bracis-28376 290 14 , , PUNCT bracis-28376 290 15 hazarika hazarika NOUN bracis-28376 290 16 , , PUNCT bracis-28376 290 17 a. a. NOUN bracis-28376 290 18 , , PUNCT bracis-28376 290 19 nigam nigam PROPN bracis-28376 290 20 , , PUNCT bracis-28376 290 21 s. s. PROPN bracis-28376 290 22 , , PUNCT bracis-28376 290 23 bhattacharya bhattacharya PROPN bracis-28376 290 24 , , PUNCT bracis-28376 290 25 a. a. PROPN bracis-28376 290 26 , , PUNCT bracis-28376 290 27 modi modi PROPN bracis-28376 290 28 , , PUNCT bracis-28376 290 29 a. a. NOUN bracis-28376 290 30 : : PUNCT bracis-28376 290 31 semantic semantic ADJ bracis-28376 290 32 segmentation segmentation NOUN bracis-28376 290 33 of of ADP bracis-28376 290 34 legal legal ADJ bracis-28376 290 35 documents document NOUN bracis-28376 290 36 via via ADP bracis-28376 290 37 rhetorical rhetorical ADJ bracis-28376 290 38 roles role NOUN bracis-28376 290 39 . . PUNCT bracis-28376 291 1 in in ADP bracis-28376 291 2 : : PUNCT bracis-28376 291 3 proceedings proceeding NOUN bracis-28376 291 4 of of ADP bracis-28376 291 5 the the DET bracis-28376 291 6 natural natural ADJ bracis-28376 291 7 legal legal ADJ bracis-28376 291 8 language language NOUN bracis-28376 291 9 processing processing NOUN bracis-28376 291 10 workshop workshop NOUN bracis-28376 291 11 2022 2022 NUM bracis-28376 291 12 , , PUNCT bracis-28376 291 13 pp pp ADJ bracis-28376 291 14 . . PUNCT bracis-28376 292 1 153–171 153–171 NUM bracis-28376 292 2 . . PUNCT bracis-28376 292 3 association association NOUN bracis-28376 292 4 for for ADP bracis-28376 292 5 computational computational ADJ bracis-28376 292 6 linguistics linguistic NOUN bracis-28376 292 7 , , PUNCT bracis-28376 292 8 abu abu PROPN bracis-28376 292 9 dhabi dhabi PROPN bracis-28376 292 10 , , PUNCT bracis-28376 292 11 united united PROPN bracis-28376 292 12 arab arab PROPN bracis-28376 292 13 emirates emirates PROPN bracis-28376 292 14 ( ( PUNCT bracis-28376 292 15 hybrid hybrid NOUN bracis-28376 292 16 ) ) PUNCT bracis-28376 292 17 , , PUNCT bracis-28376 292 18 december december PROPN bracis-28376 292 19 2022 2022 NUM bracis-28376 292 20 . . PUNCT bracis-28376 293 1 https://aclanthology.org/2022.nllp-1.13 https://aclanthology.org/2022.nllp-1.13 PROPN bracis-28376 293 2 paul paul PROPN bracis-28376 293 3 , , PUNCT bracis-28376 293 4 s. s. PROPN bracis-28376 293 5 , , PUNCT bracis-28376 293 6 mandal mandal PROPN bracis-28376 293 7 , , PUNCT bracis-28376 293 8 a. a. NOUN bracis-28376 293 9 , , PUNCT bracis-28376 293 10 goyal goyal PROPN bracis-28376 293 11 , , PUNCT bracis-28376 293 12 p. p. PROPN bracis-28376 293 13 , , PUNCT bracis-28376 293 14 ghosh ghosh PROPN bracis-28376 293 15 , , PUNCT bracis-28376 293 16 s. s. PROPN bracis-28376 293 17 : : PUNCT bracis-28376 293 18 pre pre ADJ bracis-28376 293 19 - - ADJ bracis-28376 293 20 training training ADJ bracis-28376 293 21 transformers transformer NOUN bracis-28376 293 22 on on ADP bracis-28376 293 23 indian indian ADJ bracis-28376 293 24 legal legal ADJ bracis-28376 293 25 text text NOUN bracis-28376 293 26 . . PUNCT bracis-28376 294 1 corr corr PROPN bracis-28376 294 2 abs/2209.06049 abs/2209.06049 PROPN bracis-28376 294 3 ( ( PUNCT bracis-28376 294 4 2022 2022 NUM bracis-28376 294 5 ) ) PUNCT bracis-28376 294 6 . . PUNCT bracis-28376 295 1 https://doi.org/10.48550/arxiv.2209.06049 https://doi.org/10.48550/arxiv.2209.06049 PRON bracis-28376 295 2 reimers reimer NOUN bracis-28376 295 3 , , PUNCT bracis-28376 295 4 n. n. NOUN bracis-28376 295 5 , , PUNCT bracis-28376 295 6 gurevych gurevych NOUN bracis-28376 295 7 , , PUNCT bracis-28376 295 8 i. i. NOUN bracis-28376 295 9 : : PUNCT bracis-28376 295 10 sentence sentence NOUN bracis-28376 295 11 - - PUNCT bracis-28376 295 12 bert bert PROPN bracis-28376 295 13 : : PUNCT bracis-28376 295 14 sentence sentence NOUN bracis-28376 295 15 embeddings embedding NOUN bracis-28376 295 16 using use VERB bracis-28376 295 17 siamese siamese ADJ bracis-28376 295 18 bert bert NOUN bracis-28376 295 19 - - PUNCT bracis-28376 295 20 networks network NOUN bracis-28376 295 21 . . PUNCT bracis-28376 296 1 in in ADP bracis-28376 296 2 : : PUNCT bracis-28376 296 3 inui inui PROPN bracis-28376 296 4 , , PUNCT bracis-28376 296 5 k. k. PROPN bracis-28376 296 6 , , PUNCT bracis-28376 296 7 jiang jiang PROPN bracis-28376 296 8 , , PUNCT bracis-28376 296 9 j. j. PROPN bracis-28376 296 10 , , PUNCT bracis-28376 296 11 ng ng PROPN bracis-28376 296 12 , , PUNCT bracis-28376 296 13 v. v. PROPN bracis-28376 296 14 , , PUNCT bracis-28376 296 15 wan wan PROPN bracis-28376 296 16 , , PUNCT bracis-28376 296 17 x. x. NOUN bracis-28376 296 18 ( ( PUNCT bracis-28376 296 19 eds ed NOUN bracis-28376 296 20 . . PUNCT bracis-28376 296 21 ) ) PUNCT bracis-28376 296 22 proceedings proceeding NOUN bracis-28376 296 23 of of ADP bracis-28376 296 24 the the DET bracis-28376 296 25 2019 2019 NUM bracis-28376 296 26 conference conference NOUN bracis-28376 296 27 on on ADP bracis-28376 296 28 empirical empirical ADJ bracis-28376 296 29 methods method NOUN bracis-28376 296 30 in in ADP bracis-28376 296 31 natural natural ADJ bracis-28376 296 32 language language NOUN bracis-28376 296 33 processing processing NOUN bracis-28376 296 34 and and CCONJ bracis-28376 296 35 the the DET bracis-28376 296 36 9th 9th ADJ bracis-28376 296 37 international international ADJ bracis-28376 296 38 joint joint ADJ bracis-28376 296 39 conference conference NOUN bracis-28376 296 40 on on ADP bracis-28376 296 41 natural natural ADJ bracis-28376 296 42 language language NOUN bracis-28376 296 43 processing processing NOUN bracis-28376 296 44 , , PUNCT bracis-28376 296 45 emnlp emnlp NOUN bracis-28376 296 46 - - PUNCT bracis-28376 296 47 ijcnlp ijcnlp NOUN bracis-28376 296 48 2019 2019 NUM bracis-28376 296 49 , , PUNCT bracis-28376 296 50 hong hong PROPN bracis-28376 296 51 kong kong PROPN bracis-28376 296 52 , , PUNCT bracis-28376 296 53 china china PROPN bracis-28376 296 54 , , PUNCT bracis-28376 296 55 november november PROPN bracis-28376 296 56 3–7 3–7 NUM bracis-28376 296 57 , , PUNCT bracis-28376 296 58 2019 2019 NUM bracis-28376 296 59 , , PUNCT bracis-28376 296 60 pp pp ADJ bracis-28376 296 61 . . PUNCT bracis-28376 297 1 3980–3990 3980–3990 NUM bracis-28376 297 2 . . PUNCT bracis-28376 297 3 association association NOUN bracis-28376 297 4 for for ADP bracis-28376 297 5 computational computational ADJ bracis-28376 297 6 linguistics linguistic NOUN bracis-28376 297 7 ( ( PUNCT bracis-28376 297 8 2019 2019 NUM bracis-28376 297 9 ) ) PUNCT bracis-28376 297 10 . . PUNCT bracis-28376 298 1 https://doi.org/10.18653/v1/d19-1410 https://doi.org/10.18653/v1/d19-1410 PROPN bracis-28376 298 2 savelka savelka PROPN bracis-28376 298 3 , , PUNCT bracis-28376 298 4 j. j. PROPN bracis-28376 298 5 , , PUNCT bracis-28376 298 6 westermann westermann PROPN bracis-28376 298 7 , , PUNCT bracis-28376 298 8 h. h. PROPN bracis-28376 298 9 , , PUNCT bracis-28376 298 10 benyekhlef benyekhlef NOUN bracis-28376 298 11 , , PUNCT bracis-28376 298 12 k. k. PROPN bracis-28376 298 13 : : PUNCT bracis-28376 298 14 cross cross ADJ bracis-28376 298 15 - - ADJ bracis-28376 298 16 domain domain ADJ bracis-28376 298 17 generalization generalization NOUN bracis-28376 298 18 and and CCONJ bracis-28376 298 19 knowledge knowledge NOUN bracis-28376 298 20 transfer transfer NOUN bracis-28376 298 21 in in ADP bracis-28376 298 22 transformers transformer NOUN bracis-28376 298 23 trained train VERB bracis-28376 298 24 on on ADP bracis-28376 298 25 legal legal ADJ bracis-28376 298 26 data datum NOUN bracis-28376 298 27 . . PUNCT bracis-28376 299 1 in in ADP bracis-28376 299 2 : : PUNCT bracis-28376 299 3 ashley ashley PROPN bracis-28376 299 4 , , PUNCT bracis-28376 299 5 k.d k.d PROPN bracis-28376 299 6 . . PROPN bracis-28376 299 7 , , PUNCT bracis-28376 299 8 atkinson atkinson PROPN bracis-28376 299 9 , , PUNCT bracis-28376 299 10 k. k. PROPN bracis-28376 299 11 , , PUNCT bracis-28376 299 12 branting brante VERB bracis-28376 299 13 , , PUNCT bracis-28376 299 14 l.k l.k PROPN bracis-28376 299 15 . . PROPN bracis-28376 299 16 , , PUNCT bracis-28376 299 17 francesconi francesconi PROPN bracis-28376 299 18 , , PUNCT bracis-28376 299 19 e. e. PROPN bracis-28376 299 20 , , PUNCT bracis-28376 299 21 grabmair grabmair NOUN bracis-28376 299 22 , , PUNCT bracis-28376 299 23 m. m. NOUN bracis-28376 299 24 , , PUNCT bracis-28376 299 25 walker walker PROPN bracis-28376 299 26 , , PUNCT bracis-28376 299 27 v.r v.r PROPN bracis-28376 299 28 . . PROPN bracis-28376 299 29 , , PUNCT bracis-28376 299 30 waltl waltl PROPN bracis-28376 299 31 , , PUNCT bracis-28376 299 32 b. b. PROPN bracis-28376 299 33 , , PUNCT bracis-28376 299 34 wyner wyner PROPN bracis-28376 299 35 , , PUNCT bracis-28376 299 36 a.z a.z PROPN bracis-28376 299 37 . . PROPN bracis-28376 299 38 ( ( PUNCT bracis-28376 299 39 eds ed NOUN bracis-28376 299 40 . . PUNCT bracis-28376 299 41 ) ) PUNCT bracis-28376 299 42 proceedings proceeding NOUN bracis-28376 299 43 of of ADP bracis-28376 299 44 the the DET bracis-28376 299 45 fourth fourth ADJ bracis-28376 299 46 workshop workshop NOUN bracis-28376 299 47 on on ADP bracis-28376 299 48 automated automate VERB bracis-28376 299 49 semantic semantic ADJ bracis-28376 299 50 analysis analysis NOUN bracis-28376 299 51 of of ADP bracis-28376 299 52 information information NOUN bracis-28376 299 53 in in ADP bracis-28376 299 54 legal legal ADJ bracis-28376 299 55 text text NOUN bracis-28376 299 56 held hold VERB bracis-28376 299 57 online online ADV bracis-28376 299 58 in in ADP bracis-28376 299 59 conjunction conjunction NOUN bracis-28376 299 60 with with ADP bracis-28376 299 61 the the DET bracis-28376 299 62 33rd 33rd ADJ bracis-28376 299 63 international international ADJ bracis-28376 299 64 conference conference NOUN bracis-28376 299 65 on on ADP bracis-28376 299 66 legal legal ADJ bracis-28376 299 67 knowledge knowledge NOUN bracis-28376 299 68 and and CCONJ bracis-28376 299 69 information information NOUN bracis-28376 299 70 systems system NOUN bracis-28376 299 71 , , PUNCT bracis-28376 299 72 asail@jurix asail@jurix X bracis-28376 299 73 2020 2020 NUM bracis-28376 299 74 , , PUNCT bracis-28376 299 75 december december PROPN bracis-28376 299 76 9 9 NUM bracis-28376 299 77 , , PUNCT bracis-28376 299 78 2020 2020 NUM bracis-28376 299 79 . . PUNCT bracis-28376 300 1 ceur ceur NOUN bracis-28376 300 2 workshop workshop NOUN bracis-28376 300 3 proceedings proceeding NOUN bracis-28376 300 4 , , PUNCT bracis-28376 300 5 vol vol NOUN bracis-28376 300 6 . . PROPN bracis-28376 300 7 2764 2764 NUM bracis-28376 300 8 . . PUNCT bracis-28376 301 1 ceur-ws.org ceur-ws.org X bracis-28376 301 2 ( ( PUNCT bracis-28376 301 3 2020 2020 NUM bracis-28376 301 4 ) ) PUNCT bracis-28376 301 5 . . PUNCT bracis-28376 302 1 https://ceur-ws.org/vol-2764/paper5.pdf https://ceur-ws.org/vol-2764/paper5.pdf PROPN bracis-28376 302 2 ulmer ulmer PROPN bracis-28376 302 3 , , PUNCT bracis-28376 302 4 d. d. PROPN bracis-28376 302 5 , , PUNCT bracis-28376 302 6 hardmeier hardmeier NOUN bracis-28376 302 7 , , PUNCT bracis-28376 302 8 c. c. PROPN bracis-28376 302 9 , , PUNCT bracis-28376 302 10 frellsen frellsen PROPN bracis-28376 302 11 , , PUNCT bracis-28376 302 12 j. j. PROPN bracis-28376 302 13 : : PUNCT bracis-28376 302 14 deep deep ADJ bracis-28376 302 15 - - PUNCT bracis-28376 302 16 significance significance NOUN bracis-28376 302 17 - - PUNCT bracis-28376 302 18 easy easy ADJ bracis-28376 302 19 and and CCONJ bracis-28376 302 20 meaningful meaningful ADJ bracis-28376 302 21 statistical statistical ADJ bracis-28376 302 22 significance significance NOUN bracis-28376 302 23 testing testing NOUN bracis-28376 302 24 in in ADP bracis-28376 302 25 the the DET bracis-28376 302 26 age age NOUN bracis-28376 302 27 of of ADP bracis-28376 302 28 neural neural ADJ bracis-28376 302 29 networks network NOUN bracis-28376 302 30 . . PUNCT bracis-28376 303 1 arxiv arxiv PROPN bracis-28376 303 2 preprint preprint NOUN bracis-28376 303 3 arxiv:2204.06815 arxiv:2204.06815 NOUN bracis-28376 303 4 ( ( PUNCT bracis-28376 303 5 2022 2022 NUM bracis-28376 303 6 ) ) PUNCT bracis-28376 303 7 vaswani vaswani NOUN bracis-28376 303 8 , , PUNCT bracis-28376 303 9 a. a. NOUN bracis-28376 303 10 , , PUNCT bracis-28376 303 11 et et PROPN bracis-28376 303 12 al al PROPN bracis-28376 303 13 . . PUNCT bracis-28376 303 14 : : PUNCT bracis-28376 304 1 attention attention NOUN bracis-28376 304 2 is be AUX bracis-28376 304 3 all all PRON bracis-28376 304 4 you you PRON bracis-28376 304 5 need need VERB bracis-28376 304 6 . . PUNCT bracis-28376 305 1 in in ADP bracis-28376 305 2 : : PUNCT bracis-28376 305 3 guyon guyon NOUN bracis-28376 305 4 , , PUNCT bracis-28376 305 5 i. i. PROPN bracis-28376 305 6 , , PUNCT bracis-28376 305 7 von von PROPN bracis-28376 305 8 luxburg luxburg PROPN bracis-28376 305 9 , , PUNCT bracis-28376 305 10 u. u. PROPN bracis-28376 305 11 , , PUNCT bracis-28376 305 12 bengio bengio PROPN bracis-28376 305 13 , , PUNCT bracis-28376 305 14 s. s. PROPN bracis-28376 305 15 , , PUNCT bracis-28376 305 16 wallach wallach PROPN bracis-28376 305 17 , , PUNCT bracis-28376 305 18 h.m h.m PROPN bracis-28376 305 19 . . PROPN bracis-28376 305 20 , , PUNCT bracis-28376 305 21 fergus fergus PROPN bracis-28376 305 22 , , PUNCT bracis-28376 305 23 r. r. PROPN bracis-28376 305 24 , , PUNCT bracis-28376 305 25 vishwanathan vishwanathan PROPN bracis-28376 305 26 , , PUNCT bracis-28376 305 27 s.v.n s.v.n PROPN bracis-28376 305 28 . . PROPN bracis-28376 305 29 , , PUNCT bracis-28376 305 30 garnett garnett PROPN bracis-28376 305 31 , , PUNCT bracis-28376 305 32 r. r. PROPN bracis-28376 305 33 ( ( PUNCT bracis-28376 305 34 eds eds PROPN bracis-28376 305 35 . . PUNCT bracis-28376 305 36 ) ) PUNCT bracis-28376 305 37 advances advance NOUN bracis-28376 305 38 in in ADP bracis-28376 305 39 neural neural ADJ bracis-28376 305 40 information information NOUN bracis-28376 305 41 processing processing NOUN bracis-28376 305 42 systems system NOUN bracis-28376 305 43 30 30 NUM bracis-28376 305 44 : : PUNCT bracis-28376 305 45 annual annual ADJ bracis-28376 305 46 conference conference NOUN bracis-28376 305 47 on on ADP bracis-28376 305 48 neural neural ADJ bracis-28376 305 49 information information NOUN bracis-28376 305 50 processing processing NOUN bracis-28376 305 51 systems system NOUN bracis-28376 305 52 2017 2017 NUM bracis-28376 305 53 , , PUNCT bracis-28376 305 54 december december PROPN bracis-28376 305 55 4–9 4–9 NUM bracis-28376 305 56 , , PUNCT bracis-28376 305 57 2017 2017 NUM bracis-28376 305 58 , , PUNCT bracis-28376 305 59 long long ADJ bracis-28376 305 60 beach beach NOUN bracis-28376 305 61 , , PUNCT bracis-28376 305 62 ca ca PROPN bracis-28376 305 63 , , PUNCT bracis-28376 305 64 usa usa PROPN bracis-28376 305 65 , , PUNCT bracis-28376 305 66 pp pp X bracis-28376 305 67 . . PUNCT bracis-28376 306 1 5998–6008 5998–6008 NUM bracis-28376 306 2 ( ( PUNCT bracis-28376 306 3 2017 2017 NUM bracis-28376 306 4 ) ) PUNCT bracis-28376 306 5 . . PUNCT bracis-28376 307 1 https://proceedings.neurips.cc/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-abstract.html https://proceedings.neurips.cc/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-abstract.html PROPN bracis-28376 307 2 wang wang PROPN bracis-28376 307 3 , , PUNCT bracis-28376 307 4 c. c. PROPN bracis-28376 307 5 , , PUNCT bracis-28376 307 6 nulty nulty PROPN bracis-28376 307 7 , , PUNCT bracis-28376 307 8 p. p. NOUN bracis-28376 307 9 , , PUNCT bracis-28376 307 10 lillis lillis PROPN bracis-28376 307 11 , , PUNCT bracis-28376 307 12 d. d. PROPN bracis-28376 307 13 : : PUNCT bracis-28376 307 14 a a DET bracis-28376 307 15 comparative comparative ADJ bracis-28376 307 16 study study NOUN bracis-28376 307 17 on on ADP bracis-28376 307 18 word word NOUN bracis-28376 307 19 embeddings embedding NOUN bracis-28376 307 20 in in ADP bracis-28376 307 21 deep deep ADJ bracis-28376 307 22 learning learning NOUN bracis-28376 307 23 for for ADP bracis-28376 307 24 text text NOUN bracis-28376 307 25 classification classification NOUN bracis-28376 307 26 . . PUNCT bracis-28376 308 1 in in ADP bracis-28376 308 2 : : PUNCT bracis-28376 308 3 proceedings proceeding NOUN bracis-28376 308 4 of of ADP bracis-28376 308 5 the the DET bracis-28376 308 6 4th 4th ADJ bracis-28376 308 7 international international ADJ bracis-28376 308 8 conference conference NOUN bracis-28376 308 9 on on ADP bracis-28376 308 10 natural natural ADJ bracis-28376 308 11 language language NOUN bracis-28376 308 12 processing processing NOUN bracis-28376 308 13 and and CCONJ bracis-28376 308 14 information information NOUN bracis-28376 308 15 retrieval retrieval NOUN bracis-28376 308 16 , , PUNCT bracis-28376 308 17 nlpir nlpir NOUN bracis-28376 308 18 2020 2020 NUM bracis-28376 308 19 , , PUNCT bracis-28376 308 20 pp pp ADV bracis-28376 308 21 . . PUNCT bracis-28376 309 1 37–46 37–46 NUM bracis-28376 309 2 . . PUNCT bracis-28376 309 3 association association NOUN bracis-28376 309 4 for for ADP bracis-28376 309 5 computing compute VERB bracis-28376 309 6 machinery machinery NOUN bracis-28376 309 7 , , PUNCT bracis-28376 309 8 new new ADJ bracis-28376 309 9 york(2021 york(2021 NOUN bracis-28376 309 10 ) ) PUNCT bracis-28376 309 11 . . PUNCT bracis-28376 310 1 https://doi.org/10.1145/3443279.3443304 https://doi.org/10.1145/3443279.3443304 PROPN bracis-28376 310 2 yuan yuan PROPN bracis-28376 310 3 , , PUNCT bracis-28376 310 4 k.h k.h PROPN bracis-28376 310 5 . . PROPN bracis-28376 310 6 , , PUNCT bracis-28376 310 7 hayashi hayashi PROPN bracis-28376 310 8 , , PUNCT bracis-28376 310 9 k. k. PROPN bracis-28376 310 10 : : PUNCT bracis-28376 310 11 bootstrap bootstrap NOUN bracis-28376 310 12 approach approach NOUN bracis-28376 310 13 to to ADP bracis-28376 310 14 inference inference NOUN bracis-28376 310 15 and and CCONJ bracis-28376 310 16 power power NOUN bracis-28376 310 17 analysis analysis NOUN bracis-28376 310 18 based base VERB bracis-28376 310 19 on on ADP bracis-28376 310 20 three three NUM bracis-28376 310 21 test test NOUN bracis-28376 310 22 statistics statistic NOUN bracis-28376 310 23 for for ADP bracis-28376 310 24 covariance covariance NOUN bracis-28376 310 25 structure structure NOUN bracis-28376 310 26 models model NOUN bracis-28376 310 27 . . PUNCT bracis-28376 311 1 br br X bracis-28376 311 2 . . PUNCT bracis-28376 312 1 j. j. PROPN bracis-28376 312 2 math math PROPN bracis-28376 312 3 . . PUNCT bracis-28376 313 1 stat stat PROPN bracis-28376 313 2 . . PUNCT bracis-28376 314 1 psychol psychol NOUN bracis-28376 314 2 . . PUNCT bracis-28376 315 1 56(1 56(1 NUM bracis-28376 315 2 ) ) PUNCT bracis-28376 315 3 , , PUNCT bracis-28376 315 4 93–110 93–110 PROPN bracis-28376 315 5 ( ( PUNCT bracis-28376 315 6 2003 2003 NUM bracis-28376 315 7 ) ) PUNCT bracis-28376 315 8 . . PUNCT bracis-28376 316 1 https://doi.org/10.1348/000711003321645368 https://doi.org/10.1348/000711003321645368 NOUN bracis-28376 316 2 . . PUNCT bracis-28376 317 1 https://bpspsychub.onlinelibrary.wiley.com/doi/abs/10.1348/000711003321645368 https://bpspsychub.onlinelibrary.wiley.com/doi/abs/10.1348/000711003321645368 PROPN bracis-28376 317 2 zhang zhang PROPN bracis-28376 317 3 , , PUNCT bracis-28376 317 4 h. h. PROPN bracis-28376 317 5 , , PUNCT bracis-28376 317 6 cissé cissé NOUN bracis-28376 317 7 , , PUNCT bracis-28376 317 8 m. m. NOUN bracis-28376 317 9 , , PUNCT bracis-28376 317 10 dauphin dauphin PROPN bracis-28376 317 11 , , PUNCT bracis-28376 317 12 y.n y.n PROPN bracis-28376 317 13 . . PROPN bracis-28376 317 14 , , PUNCT bracis-28376 317 15 lopez lopez PROPN bracis-28376 317 16 - - PUNCT bracis-28376 317 17 paz paz NOUN bracis-28376 317 18 , , PUNCT bracis-28376 317 19 d. d. PROPN bracis-28376 317 20 : : PUNCT bracis-28376 317 21 mixup mixup NOUN bracis-28376 317 22 : : PUNCT bracis-28376 317 23 beyond beyond ADP bracis-28376 317 24 empirical empirical ADJ bracis-28376 317 25 risk risk NOUN bracis-28376 317 26 minimization minimization NOUN bracis-28376 317 27 . . PUNCT bracis-28376 318 1 in in ADP bracis-28376 318 2 : : PUNCT bracis-28376 318 3 6th 6th ADJ bracis-28376 318 4 international international ADJ bracis-28376 318 5 conference conference NOUN bracis-28376 318 6 on on ADP bracis-28376 318 7 learning learn VERB bracis-28376 318 8 representations representation NOUN bracis-28376 318 9 , , PUNCT bracis-28376 318 10 iclr iclr NOUN bracis-28376 318 11 2018 2018 NUM bracis-28376 318 12 , , PUNCT bracis-28376 318 13 vancouver vancouver PROPN bracis-28376 318 14 , , PUNCT bracis-28376 318 15 bc bc PROPN bracis-28376 318 16 , , PUNCT bracis-28376 318 17 canada canada PROPN bracis-28376 318 18 , , PUNCT bracis-28376 318 19 april april PROPN bracis-28376 318 20 30 30 NUM bracis-28376 318 21 may may PROPN bracis-28376 318 22 3 3 NUM bracis-28376 318 23 , , PUNCT bracis-28376 318 24 2018 2018 NUM bracis-28376 318 25 , , PUNCT bracis-28376 318 26 conference conference NOUN bracis-28376 318 27 track track NOUN bracis-28376 318 28 proceedings proceeding NOUN bracis-28376 318 29 . . PUNCT bracis-28376 319 1 openreview.net openreview.net X bracis-28376 319 2 ( ( PUNCT bracis-28376 319 3 2018 2018 NUM bracis-28376 319 4 ) ) PUNCT bracis-28376 319 5 . . PUNCT bracis-28376 320 1 https://openreview.net/forum?id=r1ddp1-rb https://openreview.net/forum?id=r1ddp1-rb PROPN bracis-28376 320 2 zhong zhong PROPN bracis-28376 320 3 , , PUNCT bracis-28376 320 4 h. h. PROPN bracis-28376 320 5 , , PUNCT bracis-28376 320 6 xiao xiao PROPN bracis-28376 320 7 , , PUNCT bracis-28376 320 8 c. c. PROPN bracis-28376 320 9 , , PUNCT bracis-28376 320 10 tu tu PROPN bracis-28376 320 11 , , PUNCT bracis-28376 320 12 c. c. PROPN bracis-28376 320 13 , , PUNCT bracis-28376 320 14 zhang zhang PROPN bracis-28376 320 15 , , PUNCT bracis-28376 320 16 t. t. PROPN bracis-28376 320 17 , , PUNCT bracis-28376 320 18 liu liu PROPN bracis-28376 320 19 , , PUNCT bracis-28376 320 20 z. z. PROPN bracis-28376 320 21 , , PUNCT bracis-28376 320 22 sun sun NOUN bracis-28376 320 23 , , PUNCT bracis-28376 320 24 m. m. NOUN bracis-28376 320 25 : : PUNCT bracis-28376 320 26 how how SCONJ bracis-28376 320 27 does do AUX bracis-28376 320 28 nlp nlp NOUN bracis-28376 320 29 benefit benefit VERB bracis-28376 320 30 legal legal ADJ bracis-28376 320 31 system system NOUN bracis-28376 320 32 : : PUNCT bracis-28376 320 33 a a DET bracis-28376 320 34 summary summary NOUN bracis-28376 320 35 of of ADP bracis-28376 320 36 legal legal ADJ bracis-28376 320 37 artificial artificial ADJ bracis-28376 320 38 intelligence intelligence NOUN bracis-28376 320 39 . . PUNCT bracis-28376 321 1 in in ADP bracis-28376 321 2 : : PUNCT bracis-28376 321 3 proceedings proceeding NOUN bracis-28376 321 4 of of ADP bracis-28376 321 5 the the DET bracis-28376 321 6 58th 58th ADJ bracis-28376 321 7 annual annual ADJ bracis-28376 321 8 meeting meeting NOUN bracis-28376 321 9 of of ADP bracis-28376 321 10 the the DET bracis-28376 321 11 association association NOUN bracis-28376 321 12 for for ADP bracis-28376 321 13 computational computational ADJ bracis-28376 321 14 linguistics linguistic NOUN bracis-28376 321 15 . . PUNCT bracis-28376 322 1 pp pp ADJ bracis-28376 322 2 . . PUNCT bracis-28376 323 1 5218–5230 5218–5230 NUM bracis-28376 323 2 . . PUNCT bracis-28376 323 3 association association NOUN bracis-28376 323 4 for for ADP bracis-28376 323 5 computational computational ADJ bracis-28376 323 6 linguistics linguistic NOUN bracis-28376 323 7 , , PUNCT bracis-28376 323 8 online online ADJ bracis-28376 323 9 , , PUNCT bracis-28376 323 10 july july PROPN bracis-28376 323 11 2020 2020 NUM bracis-28376 323 12 . . PUNCT bracis-28376 324 1 https://doi.org/10.18653/v1/2020.acl-main.466 https://doi.org/10.18653/v1/2020.acl-main.466 NOUN bracis-28376 324 2 . . PUNCT bracis-28376 325 1 https://aclanthology.org/2020.acl-main.466 https://aclanthology.org/2020.acl-main.466 PROPN bracis-28376 325 2 download download VERB bracis-28376 325 3 references reference NOUN bracis-28376 325 4 author author NOUN bracis-28376 325 5 information information NOUN bracis-28376 325 6 authors author NOUN bracis-28376 325 7 and and CCONJ bracis-28376 325 8 affiliations affiliation NOUN bracis-28376 325 9 instituto instituto PROPN bracis-28376 325 10 federal federal PROPN bracis-28376 325 11 do do PROPN bracis-28376 325 12 rio rio PROPN bracis-28376 325 13 grande grande PROPN bracis-28376 325 14 do do VERB bracis-28376 325 15 norte norte NOUN bracis-28376 325 16 , , PUNCT bracis-28376 325 17 natal natal NOUN bracis-28376 325 18 , , PUNCT bracis-28376 325 19 brazil brazil PROPN bracis-28376 325 20 alexandre alexandre PROPN bracis-28376 325 21 gomes gomes PROPN bracis-28376 325 22 de de PROPN bracis-28376 325 23 lima lima PROPN bracis-28376 325 24 universidade universidade PROPN bracis-28376 325 25 federal federal PROPN bracis-28376 325 26 do do AUX bracis-28376 325 27 rio rio PROPN bracis-28376 325 28 grande grande PROPN bracis-28376 325 29 do do VERB bracis-28376 325 30 norte norte NOUN bracis-28376 325 31 , , PUNCT bracis-28376 325 32 natal natal NOUN bracis-28376 325 33 , , PUNCT bracis-28376 325 34 brazil brazil PROPN bracis-28376 325 35 alexandre alexandre PROPN bracis-28376 325 36 gomes gomes PROPN bracis-28376 325 37 de de PROPN bracis-28376 325 38 lima lima PROPN bracis-28376 325 39     SPACE bracis-28376 325 40 & & CCONJ bracis-28376 325 41     SPACE bracis-28376 325 42 eduardo eduardo PROPN bracis-28376 325 43 henrique henrique PROPN bracis-28376 325 44 da da PROPN bracis-28376 325 45 s. s. PROPN bracis-28376 325 46 aranha aranha VERB bracis-28376 325 47 institut institut PROPN bracis-28376 325 48 de de X bracis-28376 325 49 recherche recherche X bracis-28376 325 50 en en X bracis-28376 325 51 informatique informatique PROPN bracis-28376 325 52 de de PROPN bracis-28376 325 53 toulouse toulouse NOUN bracis-28376 325 54 , , PUNCT bracis-28376 325 55 umr umr PROPN bracis-28376 325 56 5505 5505 NUM bracis-28376 325 57 , , PUNCT bracis-28376 325 58 cnrs cnrs NOUN bracis-28376 325 59 , , PUNCT bracis-28376 325 60 31000 31000 NUM bracis-28376 325 61 , , PUNCT bracis-28376 325 62 toulouse toulouse NOUN bracis-28376 325 63 , , PUNCT bracis-28376 325 64 france france PROPN bracis-28376 325 65 josé josé PROPN bracis-28376 325 66 g. g. PROPN bracis-28376 325 67 moreno moreno PROPN bracis-28376 325 68 , , PUNCT bracis-28376 325 69     SPACE bracis-28376 325 70 taoufiq taoufiq PROPN bracis-28376 325 71 dkaki dkaki PROPN bracis-28376 325 72     SPACE bracis-28376 325 73 & & CCONJ bracis-28376 325 74     SPACE bracis-28376 325 75 mohand mohand NOUN bracis-28376 325 76 boughanem boughanem PROPN bracis-28376 325 77 authors author NOUN bracis-28376 325 78 alexandre alexandre PROPN bracis-28376 325 79 gomes gomes PROPN bracis-28376 325 80 de de PROPN bracis-28376 325 81 limaview limaview PROPN bracis-28376 325 82 author author NOUN bracis-28376 325 83 publications publication NOUN bracis-28376 325 84 search search NOUN bracis-28376 325 85 author author NOUN bracis-28376 325 86 on on ADP bracis-28376 325 87 : : PUNCT bracis-28376 325 88 pubmed pubmed PROPN bracis-28376 325 89     SPACE bracis-28376 325 90 google google PROPN bracis-28376 325 91 scholar scholar PROPN bracis-28376 325 92 josé josé PROPN bracis-28376 325 93 g. g. PROPN bracis-28376 325 94 morenoview morenoview PROPN bracis-28376 325 95 author author NOUN bracis-28376 325 96 publications publication NOUN bracis-28376 325 97 search search NOUN bracis-28376 325 98 author author NOUN bracis-28376 325 99 on on ADP bracis-28376 325 100 : : PUNCT bracis-28376 325 101 pubmed pubmed PROPN bracis-28376 325 102     SPACE bracis-28376 325 103 google google PROPN bracis-28376 325 104 scholar scholar NOUN bracis-28376 325 105 taoufiq taoufiq PROPN bracis-28376 325 106 dkakiview dkakiview NOUN bracis-28376 325 107 author author NOUN bracis-28376 325 108 publications publication NOUN bracis-28376 325 109 search search NOUN bracis-28376 325 110 author author NOUN bracis-28376 325 111 on on ADP bracis-28376 325 112 : : PUNCT bracis-28376 325 113 pubmed pubmed PROPN bracis-28376 325 114     SPACE bracis-28376 325 115 google google PROPN bracis-28376 325 116 scholar scholar PROPN bracis-28376 325 117 eduardo eduardo PROPN bracis-28376 325 118 henrique henrique PROPN bracis-28376 325 119 da da PROPN bracis-28376 325 120 s. s. PROPN bracis-28376 325 121 aranhaview aranhaview PROPN bracis-28376 325 122 author author NOUN bracis-28376 325 123 publications publication VERB bracis-28376 325 124 search search NOUN bracis-28376 325 125 author author NOUN bracis-28376 325 126 on on ADP bracis-28376 325 127 : : PUNCT bracis-28376 325 128 pubmed pubmed PROPN bracis-28376 325 129     SPACE bracis-28376 325 130 google google PROPN bracis-28376 325 131 scholar scholar NOUN bracis-28376 325 132 mohand mohand NOUN bracis-28376 325 133 boughanemview boughanemview NOUN bracis-28376 325 134 author author NOUN bracis-28376 325 135 publications publication NOUN bracis-28376 325 136 search search NOUN bracis-28376 325 137 author author NOUN bracis-28376 325 138 on on ADP bracis-28376 325 139 : : PUNCT bracis-28376 325 140 pubmed pubmed PROPN bracis-28376 325 141     SPACE bracis-28376 325 142 google google PROPN bracis-28376 325 143 scholar scholar NOUN bracis-28376 325 144 corresponding correspond VERB bracis-28376 325 145 author author NOUN bracis-28376 325 146 correspondence correspondence NOUN bracis-28376 325 147 to to PART bracis-28376 325 148 alexandre alexandre VERB bracis-28376 325 149 gomes gomes PROPN bracis-28376 325 150 de de PROPN bracis-28376 325 151 lima lima PROPN bracis-28376 325 152 . . PUNCT bracis-28376 326 1 editor editor NOUN bracis-28376 326 2 information information NOUN bracis-28376 326 3 editors editor NOUN bracis-28376 326 4 and and CCONJ bracis-28376 326 5 affiliations affiliation NOUN bracis-28376 326 6 federal federal PROPN bracis-28376 326 7 university university PROPN bracis-28376 326 8 of of ADP bracis-28376 326 9 são são PROPN bracis-28376 326 10 carlos carlos PROPN bracis-28376 326 11 , , PUNCT bracis-28376 326 12 são são PROPN bracis-28376 326 13 carlos carlos PROPN bracis-28376 326 14 , , PUNCT bracis-28376 326 15 brazil brazil PROPN bracis-28376 326 16 murilo murilo PROPN bracis-28376 326 17 c. c. PROPN bracis-28376 326 18 naldi naldi PROPN bracis-28376 326 19 centro centro PROPN bracis-28376 326 20 universitario universitario PROPN bracis-28376 326 21 da da PROPN bracis-28376 326 22 fei fei PROPN bracis-28376 326 23 , , PUNCT bracis-28376 326 24 são são PROPN bracis-28376 326 25 bernardo bernardo PROPN bracis-28376 326 26 do do AUX bracis-28376 326 27 campo campo PROPN bracis-28376 326 28 , , PUNCT bracis-28376 326 29 brazil brazil PROPN bracis-28376 326 30 reinaldo reinaldo PROPN bracis-28376 326 31 a. a. PROPN bracis-28376 326 32 c. c. PROPN bracis-28376 326 33 bianchi bianchi PROPN bracis-28376 326 34 rights right NOUN bracis-28376 326 35 and and CCONJ bracis-28376 326 36 permissions permission NOUN bracis-28376 326 37 reprints reprint NOUN bracis-28376 326 38 and and CCONJ bracis-28376 326 39 permissions permission VERB bracis-28376 326 40 copyright copyright NOUN bracis-28376 326 41 information information NOUN bracis-28376 326 42 © © ADP bracis-28376 326 43 2023 2023 NUM bracis-28376 326 44 the the DET bracis-28376 326 45 author(s author(s NOUN bracis-28376 326 46 ) ) PUNCT bracis-28376 326 47 , , PUNCT bracis-28376 326 48 under under ADP bracis-28376 326 49 exclusive exclusive ADJ bracis-28376 326 50 license license NOUN bracis-28376 326 51 to to ADP bracis-28376 326 52 springer springer NOUN bracis-28376 326 53 nature nature PROPN bracis-28376 326 54 switzerland switzerland PROPN bracis-28376 326 55 ag ag PROPN bracis-28376 326 56 about about ADP bracis-28376 326 57 this this DET bracis-28376 326 58 paper paper NOUN bracis-28376 326 59 cite cite VERB bracis-28376 326 60 this this DET bracis-28376 326 61 paper paper NOUN bracis-28376 326 62 de de X bracis-28376 326 63 lima lima PROPN bracis-28376 326 64 , , PUNCT bracis-28376 326 65 a.g a.g PROPN bracis-28376 326 66 . . PROPN bracis-28376 326 67 , , PUNCT bracis-28376 326 68 moreno moreno PROPN bracis-28376 326 69 , , PUNCT bracis-28376 326 70 j.g j.g PROPN bracis-28376 326 71 . . PROPN bracis-28376 326 72 , , PUNCT bracis-28376 326 73 dkaki dkaki PROPN bracis-28376 326 74 , , PUNCT bracis-28376 326 75 t. t. PROPN bracis-28376 326 76 , , PUNCT bracis-28376 326 77 da da PROPN bracis-28376 326 78 s. s. PROPN bracis-28376 326 79 aranha aranha PROPN bracis-28376 326 80 , , PUNCT bracis-28376 326 81 e.h e.h PROPN bracis-28376 326 82 . . PROPN bracis-28376 326 83 , , PUNCT bracis-28376 326 84 boughanem boughanem NOUN bracis-28376 326 85 , , PUNCT bracis-28376 326 86 m. m. NOUN bracis-28376 326 87 ( ( PUNCT bracis-28376 326 88 2023 2023 NUM bracis-28376 326 89 ) ) PUNCT bracis-28376 326 90 . . PUNCT bracis-28376 327 1 evaluating evaluate VERB bracis-28376 327 2 recent recent ADJ bracis-28376 327 3 legal legal ADJ bracis-28376 327 4 rhetorical rhetorical ADJ bracis-28376 327 5 role role NOUN bracis-28376 327 6 labeling labeling NOUN bracis-28376 327 7 approaches approach NOUN bracis-28376 327 8 supported support VERB bracis-28376 327 9 by by ADP bracis-28376 327 10     SPACE bracis-28376 327 11 transformer transformer NOUN bracis-28376 327 12 encoders encoder NOUN bracis-28376 327 13 . . PUNCT bracis-28376 328 1 in in ADP bracis-28376 328 2 : : PUNCT bracis-28376 328 3 naldi naldi PROPN bracis-28376 328 4 , , PUNCT bracis-28376 328 5 m.c m.c PROPN bracis-28376 328 6 . . PROPN bracis-28376 328 7 , , PUNCT bracis-28376 328 8 bianchi bianchi PROPN bracis-28376 328 9 , , PUNCT bracis-28376 328 10 r.a.c r.a.c ADP bracis-28376 328 11 . . PUNCT bracis-28376 328 12 ( ( PUNCT bracis-28376 328 13 eds ed NOUN bracis-28376 328 14 ) ) PUNCT bracis-28376 328 15 intelligent intelligent ADJ bracis-28376 328 16 systems system NOUN bracis-28376 328 17 . . PUNCT bracis-28376 329 1 bracis bracis PROPN bracis-28376 329 2 2023 2023 NUM bracis-28376 329 3 . . PUNCT bracis-28376 330 1 lecture lecture NOUN bracis-28376 330 2 notes note NOUN bracis-28376 330 3 in in ADP bracis-28376 330 4 computer computer NOUN bracis-28376 330 5 science science NOUN bracis-28376 330 6 ( ( PUNCT bracis-28376 330 7 ) ) PUNCT bracis-28376 330 8 , , PUNCT bracis-28376 330 9 vol vol NOUN bracis-28376 330 10 14196 14196 NUM bracis-28376 330 11 . . PUNCT bracis-28376 331 1 springer springer NOUN bracis-28376 331 2 , , PUNCT bracis-28376 331 3 cham cham PROPN bracis-28376 331 4 . . PUNCT bracis-28376 332 1 https://doi.org/10.1007/978-3-031-45389-2_2 https://doi.org/10.1007/978-3-031-45389-2_2 PROPN bracis-28376 332 2 download download NOUN bracis-28376 332 3 citation citation NOUN bracis-28376 332 4 .ris .ris PUNCT bracis-28376 332 5 .enw .enw PROPN bracis-28376 332 6 .bib .bib PUNCT bracis-28376 333 1 doi doi PROPN bracis-28376 333 2 : : PUNCT bracis-28376 333 3 https://doi.org/10.1007/978-3-031-45389-2_2 https://doi.org/10.1007/978-3-031-45389-2_2 PUNCT bracis-28376 333 4 published publish VERB bracis-28376 333 5 : : PUNCT bracis-28376 333 6 12 12 NUM bracis-28376 333 7 october october PROPN bracis-28376 333 8 2023 2023 NUM bracis-28376 333 9 publisher publisher NOUN bracis-28376 333 10 name name NOUN bracis-28376 333 11 : : PUNCT bracis-28376 333 12 springer springer NOUN bracis-28376 333 13 , , PUNCT bracis-28376 333 14 cham cham PROPN bracis-28376 333 15 print print PROPN bracis-28376 333 16 isbn isbn PROPN bracis-28376 333 17 : : PUNCT bracis-28376 333 18 978 978 NUM bracis-28376 333 19 - - SYM bracis-28376 333 20 3 3 NUM bracis-28376 333 21 - - PUNCT bracis-28376 333 22 031 031 NUM bracis-28376 333 23 - - PUNCT bracis-28376 333 24 45388 45388 NUM bracis-28376 333 25 - - SYM bracis-28376 333 26 5 5 NUM bracis-28376 333 27 online online ADJ bracis-28376 333 28 isbn isbn NOUN bracis-28376 333 29 : : PUNCT bracis-28376 333 30 978 978 NUM bracis-28376 333 31 - - SYM bracis-28376 333 32 3 3 NUM bracis-28376 333 33 - - PUNCT bracis-28376 333 34 031 031 NUM bracis-28376 333 35 - - PUNCT bracis-28376 333 36 45389 45389 NUM bracis-28376 333 37 - - SYM bracis-28376 333 38 2 2 NUM bracis-28376 333 39 ebook ebook NOUN bracis-28376 333 40 packages package NOUN bracis-28376 333 41 : : PUNCT bracis-28376 333 42 computer computer NOUN bracis-28376 333 43 sciencecomputer sciencecomputer NOUN bracis-28376 333 44 science science NOUN bracis-28376 333 45 ( ( PUNCT bracis-28376 333 46 r0 r0 NOUN bracis-28376 333 47 ) ) PUNCT bracis-28376 333 48 share share VERB bracis-28376 333 49 this this DET bracis-28376 333 50 paper paper NOUN bracis-28376 333 51 anyone anyone PRON bracis-28376 333 52 you you PRON bracis-28376 333 53 share share VERB bracis-28376 333 54 the the DET bracis-28376 333 55 following follow VERB bracis-28376 333 56 link link NOUN bracis-28376 333 57 with with ADP bracis-28376 333 58 will will AUX bracis-28376 333 59 be be AUX bracis-28376 333 60 able able ADJ bracis-28376 333 61 to to PART bracis-28376 333 62 read read VERB bracis-28376 333 63 this this DET bracis-28376 333 64 content content NOUN bracis-28376 333 65 : : PUNCT bracis-28376 333 66 get get VERB bracis-28376 333 67 shareable shareable ADJ bracis-28376 333 68 linksorry linksorry NOUN bracis-28376 333 69 , , PUNCT bracis-28376 333 70 a a DET bracis-28376 333 71 shareable shareable ADJ bracis-28376 333 72 link link NOUN bracis-28376 333 73 is be AUX bracis-28376 333 74 not not PART bracis-28376 333 75 currently currently ADV bracis-28376 333 76 available available ADJ bracis-28376 333 77 for for ADP bracis-28376 333 78 this this DET bracis-28376 333 79 article article NOUN bracis-28376 333 80 . . PUNCT bracis-28376 334 1 copy copy VERB bracis-28376 334 2 shareable shareable ADJ bracis-28376 334 3 link link NOUN bracis-28376 334 4 to to PART bracis-28376 334 5 clipboard clipboard NOUN bracis-28376 334 6 provided provide VERB bracis-28376 334 7 by by ADP bracis-28376 334 8 the the DET bracis-28376 334 9 springer springer NOUN bracis-28376 334 10 nature nature PROPN bracis-28376 334 11 sharedit sharedit PROPN bracis-28376 334 12 content content NOUN bracis-28376 334 13 - - PUNCT bracis-28376 334 14 sharing share VERB bracis-28376 334 15 initiative initiative NOUN bracis-28376 334 16 keywords keyword VERB bracis-28376 334 17 indian indian ADJ bracis-28376 334 18 legal legal ADJ bracis-28376 334 19 system system NOUN bracis-28376 334 20 sentence sentence NOUN bracis-28376 334 21 classification classification NOUN bracis-28376 334 22 sentence sentence NOUN bracis-28376 334 23 encoding encode VERB bracis-28376 334 24 dfcsc dfcsc NOUN bracis-28376 334 25 publish publish VERB bracis-28376 334 26 with with ADP bracis-28376 334 27 us us PROPN bracis-28376 334 28 policies policy NOUN bracis-28376 334 29 and and CCONJ bracis-28376 334 30 ethics ethic NOUN bracis-28376 334 31 search search NOUN bracis-28376 334 32 search search NOUN bracis-28376 334 33 by by ADP bracis-28376 334 34 keyword keyword NOUN bracis-28376 334 35 or or CCONJ bracis-28376 334 36 author author NOUN bracis-28376 334 37 search search NOUN bracis-28376 334 38 navigation navigation NOUN bracis-28376 334 39 find find VERB bracis-28376 334 40 a a DET bracis-28376 334 41 journal journal NOUN bracis-28376 334 42 publish publish VERB bracis-28376 334 43 with with ADP bracis-28376 334 44 us we PRON bracis-28376 334 45 track track VERB bracis-28376 334 46 your your PRON bracis-28376 334 47 research research NOUN bracis-28376 334 48 discover discover VERB bracis-28376 334 49 content content NOUN bracis-28376 334 50 journals journal NOUN bracis-28376 334 51 a a DET bracis-28376 334 52 - - PUNCT bracis-28376 334 53 z z NOUN bracis-28376 334 54 books book NOUN bracis-28376 334 55 a a DET bracis-28376 334 56 - - PUNCT bracis-28376 334 57 z z NOUN bracis-28376 334 58 publish publish NOUN bracis-28376 334 59 with with ADP bracis-28376 334 60 us us PROPN bracis-28376 334 61 journal journal PROPN bracis-28376 334 62 finder finder PROPN bracis-28376 334 63 publish publish VERB bracis-28376 334 64 your your PRON bracis-28376 334 65 research research NOUN bracis-28376 334 66 language language NOUN bracis-28376 334 67 editing edit VERB bracis-28376 334 68 open open ADJ bracis-28376 334 69 access access NOUN bracis-28376 334 70 publishing publishing NOUN bracis-28376 334 71 products product NOUN bracis-28376 334 72 and and CCONJ bracis-28376 334 73 services service NOUN bracis-28376 334 74 our our PRON bracis-28376 334 75 products product NOUN bracis-28376 334 76 librarians librarian VERB bracis-28376 334 77 societies society NOUN bracis-28376 334 78 partners partner NOUN bracis-28376 334 79 and and CCONJ bracis-28376 334 80 advertisers advertiser NOUN bracis-28376 334 81 our our PRON bracis-28376 334 82 brands brand NOUN bracis-28376 334 83 springer springer NOUN bracis-28376 334 84 nature nature PROPN bracis-28376 334 85 portfolio portfolio PROPN bracis-28376 334 86 bmc bmc PROPN bracis-28376 334 87 palgrave palgrave PROPN bracis-28376 334 88 macmillan macmillan PROPN bracis-28376 334 89 apress apress PROPN bracis-28376 334 90 discover discover VERB bracis-28376 334 91 your your PRON bracis-28376 334 92 privacy privacy NOUN bracis-28376 334 93 choices choice NOUN bracis-28376 334 94 / / SYM bracis-28376 334 95 manage manage NOUN bracis-28376 334 96 cookies cookie NOUN bracis-28376 334 97 your your PRON bracis-28376 334 98 us us PROPN bracis-28376 335 1 state state NOUN bracis-28376 335 2 privacy privacy NOUN bracis-28376 335 3 rights right NOUN bracis-28376 335 4 accessibility accessibility NOUN bracis-28376 335 5 statement statement NOUN bracis-28376 335 6 terms term NOUN bracis-28376 335 7 and and CCONJ bracis-28376 335 8 conditions condition NOUN bracis-28376 335 9 privacy privacy NOUN bracis-28376 335 10 policy policy NOUN bracis-28376 335 11 help help NOUN bracis-28376 335 12 and and CCONJ bracis-28376 335 13 support support VERB bracis-28376 335 14 legal legal ADJ bracis-28376 335 15 notice notice NOUN bracis-28376 335 16 cancel cancel VERB bracis-28376 335 17 contracts contract NOUN bracis-28376 335 18 here here ADV bracis-28376 335 19 129.74.145.123 129.74.145.123 NUM bracis-28376 335 20 hesburgh hesburgh PROPN bracis-28376 335 21 library library PROPN bracis-28376 335 22 er er INTJ bracis-28376 335 23 unit unit NOUN bracis-28376 335 24 ( ( PUNCT bracis-28376 335 25 3005732405 3005732405 NUM bracis-28376 335 26 ) ) PUNCT bracis-28376 335 27 northeast northeast ADJ bracis-28376 335 28 research research NOUN bracis-28376 335 29 libraries library NOUN bracis-28376 335 30 ( ( PUNCT bracis-28376 335 31 nerl nerl PROPN bracis-28376 335 32 ) ) PUNCT bracis-28376 335 33 ( ( PUNCT bracis-28376 335 34 8200828607 8200828607 NUM bracis-28376 335 35 ) ) PUNCT bracis-28376 335 36 nerl nerl VERB bracis-28376 335 37 ta ta X bracis-28376 335 38 account account NOUN bracis-28376 335 39 ( ( PUNCT bracis-28376 335 40 3006206169 3006206169 NUM bracis-28376 335 41 ) ) PUNCT bracis-28376 335 42 university university NOUN bracis-28376 335 43 of of ADP bracis-28376 335 44 notre notre PROPN bracis-28376 335 45 dame dame PROPN bracis-28376 335 46 hesburgh hesburgh PROPN bracis-28376 335 47 library library NOUN bracis-28376 335 48 ( ( PUNCT bracis-28376 335 49 3000184373 3000184373 NUM bracis-28376 335 50 ) ) PUNCT bracis-28376 336 1 © © ADP bracis-28376 336 2 2025 2025 NUM bracis-28376 336 3 springer springer NOUN bracis-28376 336 4 nature nature NOUN