File size: 88,574 Bytes
6fa4bc9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
{
    "paper_id": "2020",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T13:05:03.050140Z"
    },
    "title": "TermPortal: A Workbench for Automatic Term Extraction from Icelandic Texts",
    "authors": [
        {
            "first": "Hjalti",
            "middle": [],
            "last": "Dan\u00edelsson",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Iceland",
                "location": {}
            },
            "email": "hjaltid@hi.is"
        },
        {
            "first": "\u00c1g\u00fasta",
            "middle": [],
            "last": "\u00deorbergsd\u00f3ttir",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "The \u00c1rni Magn\u00fasson Institute for Icelandic Studies",
                "location": {}
            },
            "email": "agusta.thorbergsdottir@arnastofnun.is"
        },
        {
            "first": "Stein\u00fe\u00f3r",
            "middle": [],
            "last": "Steingr\u00edmsson",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "The \u00c1rni Magn\u00fasson Institute for Icelandic Studies",
                "location": {}
            },
            "email": "steinthor.steingrimsson@arnastofnun.is"
        },
        {
            "first": "Gunnar",
            "middle": [
                "Thor"
            ],
            "last": "\u00d6rn\u00f3lfsson",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "The \u00c1rni Magn\u00fasson Institute for Icelandic Studies",
                "location": {}
            },
            "email": "gunnar.thor.ornolfsson@arnastofnun.is"
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Automatic term extraction (ATE) from texts is critical for effective terminology work in small speech communities. We present TermPortal, a workbench for terminology work in Iceland, featuring the first ATE system for Icelandic. The tool facilitates standardization in terminology work in Iceland, as it exports data in standard formats in order to streamline gathering and distribution of the material. In the project we focus on the domain of finance in order to do be able to fulfill the needs of an important and large field. We present a comprehensive survey amongst the most prominent organizations in that field, the results of which emphasize the need for a good, up-to-date and accessible termbank and the willingness to use terms in Icelandic. Furthermore we present the ATE tool for Icelandic, which uses a variety of methods and shows great potential with a recall rate of up to 95% and a high C-value, indicating that it competently finds term candidates that are important to the input text.",
    "pdf_parse": {
        "paper_id": "2020",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Automatic term extraction (ATE) from texts is critical for effective terminology work in small speech communities. We present TermPortal, a workbench for terminology work in Iceland, featuring the first ATE system for Icelandic. The tool facilitates standardization in terminology work in Iceland, as it exports data in standard formats in order to streamline gathering and distribution of the material. In the project we focus on the domain of finance in order to do be able to fulfill the needs of an important and large field. We present a comprehensive survey amongst the most prominent organizations in that field, the results of which emphasize the need for a good, up-to-date and accessible termbank and the willingness to use terms in Icelandic. Furthermore we present the ATE tool for Icelandic, which uses a variety of methods and shows great potential with a recall rate of up to 95% and a high C-value, indicating that it competently finds term candidates that are important to the input text.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "Terminology extraction is the task of automatically extracting relevant terms from a given corpus. An up-to-date reliable termbase of systematically collected terms or term candidates from recent texts is of great importance to translators and users of translated texts. Such a tool can be very useful for standardizing vocabulary in specialized fields, which again is crucial for translation work, leading to increased translator productivity and helping to make new texts and translations more coherent and unambiguous. Until now, terminology databases in Iceland have been constructed manually by experts in their subject fields. Termbases in more than 60 different fields have been created and made available online in \u00cd\u00f0or\u00f0abankinn 1 . The Translation Centre of the Ministry for Foreign Affairs has also made their terminology database available online 2 . There are several downsides to manual collection of terminology. New terminology often takes a long time to reach publicly accessible termbases. Some of the collected terms may not see widespread use before being supplanted by newer or better-known ones, but nonetheless linger on in the termbase, increasing the risk of ambiguity unbeknownst to the termbase editors. In certain fields there may also be a lack of experts interested in doing the terminology work, making standardization of terminology even harder. Through the adoption of state-of-the-art methods for the automatic extraction of terms, new terminology can be made available much earlier in publicly accessible termbases, where it can facilitate more effective standardization. Editors can also more easily collect statistics about terminology use and cite real-world usage examples. We present TermPortal, the first build of a workbench for semi-automatic collection of Icelandic terminologies. The workbench includes an automated terminology extraction tool that provides editors of terminologies with lists of new terminology candidates from relevant texts. For our ini-tial version we focus on the acquisition of potential new terms, and the domain of finance. An emphasis on recall over precision allows us to potentially create a corpus with which to conduct future research. Meanwhile, since financial terminology is used in a variety of different texts, there is abundant data on which to try our system -a useful property both for developing our system and for learning about how difficult it is to automatically extract terminology from different texts. There is also a great need for a continually updated termbase in this active field, as was confirmed in a thorough survey conducted at the start of the project. We describe the methodology for the survey and the results in Section 3, while the emphasis on term acquisition over term filtering is noted in Section 4. TermPortal consists of two main pieces of software: One is the TermPortal workbench described in Section 4, which includes an automatic pipeline to extract terminology from media and a web platform where users can create, manage and maintain termbases. The other is the Automatic Term Extraction (ATE) system described in Section 5. It is a central component in the TermPortal workbench, but can also be used in isolation.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "TermPortal is not the first termbase management tool to offer ATE, although it is the first to support Icelandic.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2."
            },
            {
                "text": "Tilde Terminology 3 is a cloud-based terminology extraction and management tool based on the Terminology as a Service (TaaS) project (Gornostay and Vasiljevs, 2014) . It allows users to upload monolingual documents and employs the CollTerm term extraction system (Pinnis et al., 2012) to extract term candidates, as well as offering various methods for automatic identification of translations for the candidates, such as lookup in EuroTermBank (Rirdance, 2006; Gornostaja et al., 2018) and parallel data that Tilde have mined from the web. There are also several multilingual terminology workbenches available on the web. Terminologue 4 is an open-source cloud-based terminology management tool developed by Dublin City University. It is quite flexible and enables users to define the languages used in a given termbase, as well as employing a hierarchical structure for terminology domains. It also supports importing and exporting termbases in TBX format. A multitude of commercial solutions is also available. Among the solutions available are SDL MultiTerm, TermWiki and Termbases.eu. In our work, we sacrifice some of the flexibility provided by workbenches such as Terminologue for the sake of making the process of extracting the terms themselves as straightforward and linear as possible. Much like in Tilde Terminology, we offer ATE as well as lookup in an existing termbank 5 , but do not support term alignment between languages in the current version.",
                "cite_spans": [
                    {
                        "start": 133,
                        "end": 164,
                        "text": "(Gornostay and Vasiljevs, 2014)",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 263,
                        "end": 284,
                        "text": "(Pinnis et al., 2012)",
                        "ref_id": "BIBREF15"
                    },
                    {
                        "start": 445,
                        "end": 461,
                        "text": "(Rirdance, 2006;",
                        "ref_id": "BIBREF16"
                    },
                    {
                        "start": 462,
                        "end": 486,
                        "text": "Gornostaja et al., 2018)",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 636,
                        "end": 637,
                        "text": "4",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "ATE Management",
                "sec_num": "2.1."
            },
            {
                "text": "While there are no studies on automatic extraction specifically for Icelandic, much less a particular domain such as finance, terminology extraction from monolingual corpora is a well-established field applying many different approaches. It can be said that there are two general approaches to automated terminology extraction from monolingual texts: statistical and rule-based. The rule-based methods commonly include tokenization, PoS-tagging, lemmatization, stemming and other common Natural Language Programming (NLP) approaches to linguistic analysis. A number of tools support these approaches for Icelandic texts: Some are multifunctional, such as Reynir (\u00deorsteinsson et al., 2019) , and the IceNLP collection (Loftsson and R\u00f6gnvaldsson, 2007) , while others are specialized in particular tasks: ABLTagger, a new BiLSTM PoS-tagger, has reached 95.17% accuracy for PoS-tagging Icelandic texts with a rich morphosyntactic tagset (Steingr\u00edmsson et al., 2019) ; and a recent lemmatization tool, Nefnir, has shown good results for lemmatizing Icelandic texts (Ing\u00f3lfsd\u00f3ttir et al., 2019) . Some of these tools are employed in our extraction process. In terminology extraction, linguistic features specific to an individual language are commonly used, in particular morphosyntactic information and discourse properties that distinguish noun phrases which are technical terms (Justeson and Katz, 1995) . The statistical methods range from very basic approaches like counting the number of n-grams in a document to using statistical measures and recently word embeddings in neural networks. The use of statistical measures in automated term extraction can be governed by two aspects of terms, termhood and unithood, introduced by Kageura and Umino (1996) . Termhood is considered to be \"the degree to which a stable lexical unit is related to some domain-specific concepts\", a notion closely related to a standard definition of a term. Meanwhile, unithood is \"the degree of strength or stability of syntagmatic combinations and collocations\". Certain methods can exploit both unithttps://clarin.is/en/resources/termbank/ hood and termhood, the C-value introduced by Frantzi et al. (2000) being a common measure. Many successful systems, however, employ a hybrid approach, using linguistic features to limit the search space and then applying statistical filtering. See Vintar (2010) for an example of such an approach. More recently deep learning approaches have been tested, using word embeddings. Zhang et al. (2018) give an example of such a method, using word embeddings to compute 'semantic importance' scores for candidate terms, which are then used to revise the scores of candidate terms computed by a base algorithm using linguistic rules and statistical heuristics. Bilingual extraction is another approach to ATE. In contrast to monolingual extraction which is concerned with identifying terms in a corpus, bilingual extraction primarily deals with aligning terms in different languages. Recently some advances have been made in automatically extracting terms from comparable corpora using deep learning methods (Liu et al., 2018; Heyman et al., 2018) . The general idea in these methods is to project word embeddings in both languages to a shared vector space. In our work, however, we focus on monolingual extraction.",
                "cite_spans": [
                    {
                        "start": 662,
                        "end": 689,
                        "text": "(\u00deorsteinsson et al., 2019)",
                        "ref_id": null
                    },
                    {
                        "start": 718,
                        "end": 751,
                        "text": "(Loftsson and R\u00f6gnvaldsson, 2007)",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 935,
                        "end": 963,
                        "text": "(Steingr\u00edmsson et al., 2019)",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 1062,
                        "end": 1090,
                        "text": "(Ing\u00f3lfsd\u00f3ttir et al., 2019)",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 1377,
                        "end": 1402,
                        "text": "(Justeson and Katz, 1995)",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 1730,
                        "end": 1754,
                        "text": "Kageura and Umino (1996)",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 2166,
                        "end": 2187,
                        "text": "Frantzi et al. (2000)",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 2369,
                        "end": 2382,
                        "text": "Vintar (2010)",
                        "ref_id": "BIBREF19"
                    },
                    {
                        "start": 2499,
                        "end": 2518,
                        "text": "Zhang et al. (2018)",
                        "ref_id": "BIBREF20"
                    },
                    {
                        "start": 3123,
                        "end": 3141,
                        "text": "(Liu et al., 2018;",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 3142,
                        "end": 3162,
                        "text": "Heyman et al., 2018)",
                        "ref_id": "BIBREF7"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Automatic Extraction",
                "sec_num": "2.2."
            },
            {
                "text": "Given the apparent scarcity of up-to-date terminology databases in Iceland, the first part of our project was to examine the views domain specialists hold on terminology. More specifically, we wanted to investigate the perceived value of terminological data, the level of interest in the use, acquisition, and sharing of terminology, the quality of facilities currently employed for storage of term databases, and the level of importance assigned to instant access to current terminological data. In order to be better able to deliver a useful system, we decided to work on only one domain for the first version of our system, the domain of finance. There were several reasons for the choice of this particular domain. While term collections in any domain require regular updates to prevent their obsolescence, Iceland's financial environment has changed extensively in recent times. The introduction of European directives alone has brought a host of new concepts to the field. Extant Icelandic terminology databases and dictionaries now contain a great number of deprecated, obsolete and superseded terms, making it even more difficult to find the correct Icelandic financial terms in what is already a relatively complex field for terminology. By narrowing our focus we are also able to get a comprehensive view of the term usage and needs of a specific group. We therefore commissioned a survey on the subject of terminology within this domain, and submitted it to financial institutions, corporations, and translation agencies.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The Survey",
                "sec_num": "3."
            },
            {
                "text": "The survey included questions on term-related issues, term cataloging, and opinions on terminology and term-related tasks, including the importance of terms in the workplace and the willingness to share collected terminology.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "How the Survey was Formulated",
                "sec_num": "3.1."
            },
            {
                "text": "Very High High Neutral Low Very Low None Importance of term translations 60.9% 21.7% 17.4% 0.0% 0.0% 0.0% Interest in free access to a trustworthy termbank 69.7% 26.1% 0.0% 4.2% 0.0% 0.0% Willingness to share terminology with others through a termbank 50.0% 22.7% 22.7% 0.0% 0.0% 4.5% Willingness to take part in terminology work with others 27.3% 27.3% 36.3% 9.1% 0.0% 0.0% Table 1 : Questions in survey about interest in using and working towards a common termbase. ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 375,
                        "end": 382,
                        "text": "Table 1",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Question",
                "sec_num": null
            },
            {
                "text": "Since the subject matter was so clearly delineated -terminology within a single domain -the survey was only directed at the most prominent organizations in that particular field. For each participating organization, the survey was put to the single representative considered to have the most extensive experience and play the most significant role in that organization's approach to, and policies on, terminology. The number of invited participants was consequently kept fairly low, but was also estimated to represent those organizations that would have the greatest interest in the potential value of terminology within the field, and the most extensive abilities to deploy that terminology in everyday tasks. As a result, their opinions on the subject were considered highly relevant and extremely valuable. Out of the twenty-five invited organizations, twenty-three took part in the survey -a response rate which the survey conductors considered to be quite high and thus likely to result in more reliable survey results. Moreover, the conductors noted that the representatives' extensive experience within their respective organizations was likely to produce informed, thought-out answers that could be trusted to be truthful; even more so since the survey was anonymous and conducted through an intermediary rather than The \u00c1rni Magn\u00fasson Institute for Icelandic Studies directly. The only overt classification of participants was a grouping of answers into the three categories mentioned earlier: institutions, corporations and translation agencies. Participants were asked sixteen questions. The results were decisive, and markedly in the terms' favor. Table 1 shows the responses to four of the questions, those concerning interest in using a common termbase and willingness to take part in building one. The survey participants see definite value in domain terminology with almost everyone in favor of free access to a trustworthy termbank and the majority interested or willing to take part in terminology work. Table 1 also displays a notable downward gradient among the percentage of responses in the Very High column: There is clear interest among participants in having access to highquality terminology, but slightly less so in participating in information sharing with others (including potential competitors), and rather less so in devoting time and manpower of their own to create a terminology collection at all. Of the three types of participating organizations, translation agencies -which tend to have the smallest staff -were the ones with the lowest willingness to share their own data and take on additional work load. This puts The \u00c1rni Magn\u00fasson Institute for Icelandic Studies at an advantage, being an institute whose domain is separate from the survey participants and whose staff includes experts knowledgeable in this field: It indicates that if we were to lay the terminology groundwork by establishing TermPortal, we would have gotten past any major hurdles of cooperation from these participants, and could likely expect a higher willingness in active participation (such as through user testing) during future stages of the tool's development. Access to terminology and term databases was deemed both of clear importance (see Figure 1) and, in its current form, severely lacking (see Figure 2 ). Also, even though the majority of respondents estimated that their staff look up domain terms weekly or more often (see Figure 3 ), most participants responded that no term registration whatsoever was performed within their organization. At the same time, a majority believed the most pressing issue related to terminology in the workplace was that up-to-date terms had not been collected and made available to all (see Figure 2 ). This lack of availability of Icelandic-language collections, both for up-to-date terms and in general, was reinforced when the organizations were asked where their staff would look for assistance with translations of finance terminology. A majority responded that they would ask their coworkers, rather than look to online resources, specialists in the field or any other potential resource. Attitudes toward Icelandic terminology in particular were predominantly positive. As evident from Figure 4 , when asked about their chosen language for terminology, none of the participants explicitly said they preferred non-Icelandic terms and over 90% stated they used Icelandic terminology, either when available or exclusively. These results clearly indicate the importance of easy and open access to up-to-date data. Indeed, the availability of Icelandic terminology may be seen as a vital precondition for clear and efficient communication in each field. To be able to meet the needs of this influential user group, and other professional terminology users, exploring new ways of implementing terminology collection and storage was necessary. We need to look beyond the increasingly dated methods of manual termbase construction and try to simplify the process of preparing, storing, and sharing term glossaries. This will enable specialists in the field to focus on more productive endeavors than basic terminology work and make it easier to centralize that work. Our aim was that all potential users of Icelandic terminology, including field specialists and translators, would be able to spend less time hunting down possible term candidates, and instead could simply edit or approve listed candidates, ensuring greater consistency in terminology use, dissemination, standardization, and translation.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 1661,
                        "end": 1668,
                        "text": "Table 1",
                        "ref_id": null
                    },
                    {
                        "start": 3263,
                        "end": 3272,
                        "text": "Figure 1)",
                        "ref_id": null
                    },
                    {
                        "start": 3321,
                        "end": 3329,
                        "text": "Figure 2",
                        "ref_id": null
                    },
                    {
                        "start": 3453,
                        "end": 3461,
                        "text": "Figure 3",
                        "ref_id": null
                    },
                    {
                        "start": 3753,
                        "end": 3761,
                        "text": "Figure 2",
                        "ref_id": null
                    },
                    {
                        "start": 4255,
                        "end": 4263,
                        "text": "Figure 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Survey Results",
                "sec_num": "3.2."
            },
            {
                "text": "The TermPortal workbench is an online terminology acquisition and management system. Authenticated users can create termbases and upload texts which are subsequently processed by the automatic term extraction (ATE) tool described in Section 5. After candidate terms are extracted, they are displayed alongside the source text. Selecting a term candidate highlights each of its occurrences in the text, allowing the user to quickly see the phrase in context. An example of this is shown in Figure 5 . Furthermore, each occurrence's enveloping sentence is stored, for later use as usage examples. Term candidates have five defined stages: At this stage the user can either accept or reject the term candidate. Rejected term candidates will be hidden from future ATE results for that termbase, but can be viewed separately and recovered. The tool extracts term candidates and makes note of each candidate's occurrences in the source text, highlighting them on command. Users can then choose whether to accept or reject each of the term candidates provided by said tool. Accepted candidates are added to the active termbase, and can be further processed, adding definitions, references to related terms and translations. Lists of fully or partially processed terms can be exported in standard formats such as TBX 6 and CSV 7 , enabling easy integration of those termbases into other systems which conform to those standards. In addition to supporting exportation of termbases, users can share them with other users with varying privileges.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 489,
                        "end": 497,
                        "text": "Figure 5",
                        "ref_id": "FIGREF1"
                    }
                ],
                "eq_spans": [],
                "section": "The TermPortal Workbench",
                "sec_num": "4."
            },
            {
                "text": "\u2022 Automatically extracted \u2022 Rejected \u2022 Manually entered / Accepted \u2022 Reviewed \u2022 Publishable",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The TermPortal Workbench",
                "sec_num": "4."
            },
            {
                "text": "\u2022 Owners, or co-owners have full privileges over the termbase, including giving other users privileges and general termbase administration in addition to managing the terms within the termbase. The user who creates a termbase is by default its owner.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The TermPortal Workbench",
                "sec_num": "4."
            },
            {
                "text": "\u2022 Editors have privileges over terms and texts within the termbase, but not the termbase itself. Their privileges include uploading and processing texts, accepting or rejecting term candidates, and modifying term entries.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The TermPortal Workbench",
                "sec_num": "4."
            },
            {
                "text": "\u2022 Reviewers can supply commentary on terms which have previously been accepted by editors or owners. They also have rights to mark terms as 'reviewed'.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The TermPortal Workbench",
                "sec_num": "4."
            },
            {
                "text": "\u2022 Viewers have view-only rights to the termbase and no edit privileges.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The TermPortal Workbench",
                "sec_num": "4."
            },
            {
                "text": "Until now, no publicly available workbench designed for terminology work in Icelandic has existed, meaning that editors for each domain set their own individual workflows and standards, which can cause difficulty when termbases 6 ISO 30042:2019 7 RFC 4180 are combined and centralized. A standardized work environment for terminology collection will enforce homogeneity in termbase structure between subject areas, facilitating easy termbase compilation. Interactive use of the ATE component turns the complex task of identifying new terms into a sequence of binary questions, greatly simplifying the workflow of termbase editors and potentially increasing productivity.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The TermPortal Workbench",
                "sec_num": "4."
            },
            {
                "text": "The ATE tool lies at the heart of the TermPortal workbench. It accepts input in the form of Icelandic text, processes the text in order to find possible candidates, calculates the candidates' term likelihoods, and outputs a sorted list of those terms it deems most likely to be heretofore unseen terms within a given domain. As noted, this is the first tool of its kind to support Icelandic, and terminology databases have until now been constructed by hand. As a result, our focus was on maximizing the tool's ability to gather potential new terminology and create a sizable initial database suitable for further computerized work and research. Accordingly, term recall was considered to be of primary importance, and was heavily emphasized over precision during the tool's development. Finetuning precision will be part of future work on TermPortal.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Automatic Term Extraction",
                "sec_num": "5."
            },
            {
                "text": "Although the ultimate goal of the ATE tool is to be capable of handling texts from any domain, we initially focus on the financial sector as we do in other parts of the project. This means that we sourced testing data solely from that particular field. The data came in two forms: Randomly selected texts originating from various sources in the financial sector, primarily laws, regulations, reports, and educational materials; and known finance terms listed in the aforementioned terminology database compiled by the Translation Centre of the Ministry for Foreign Affairs (see Section 1). While the random compilation of the general texts -some of which carry confidentiality clauses -makes it impractical to publish them as datasets, all the known finance terms may be accessed through the Ministry's website, which allows content filtering according to subject area. Known  Terms   Total   1  250  250  500  2  500  500  1,000  3  1,000  1,000  2,000  4 2,000 2,000 4,000 In order to evaluate the tool, we created four text files that combined these two types of data. Each file contained one sentence or clause per line and had an equal ratio between lines of random clauses and lines of known terms. The smallest file of 500 lines thus contained 250 random clauses and an additional 250 known terms; and with each test set the file size doubled, as shown in Table 2 . During each test, one of these files served as the program's main input.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 870,
                        "end": 956,
                        "text": "Known  Terms   Total   1  250  250  500  2  500  500  1,000  3  1,000  1,000  2,000  4",
                        "ref_id": "TABREF0"
                    },
                    {
                        "start": 1363,
                        "end": 1370,
                        "text": "Table 2",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Data Preparation",
                "sec_num": "5.1."
            },
            {
                "text": "Alongside that file, we provided the tool with two others: A unique list of just over 2,000 known finance terms, in lemmatized form, whose contents did not overlap with the terms added to the input file, and a list of 280 grammatical category patterns that corresponded to all known financial terms. Each entry in the pattern list contained an ordered sequence of grammatical tags, such as ('a', 'v', 'n'), corresponding to ('adjective', 'verb', 'noun') . In section 5.3 we describe how these patterns are used to identify potential term candidates in the program's input.",
                "cite_spans": [
                    {
                        "start": 424,
                        "end": 453,
                        "text": "('adjective', 'verb', 'noun')",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Test Set Random Clauses",
                "sec_num": null
            },
            {
                "text": "In choosing our methods, we needed to consider certain constraints while trying to provide maximum coverage. The Icelandic language is morphosyntactically rich, with a relatively free sentence word order, high inflectional complexity, and a high ratio of compound words, all of which affect the linguistic aspects of term extraction (Bjarnad\u00f3ttir et al., 2019) . Moreover, while we focused on the financial sector during development, the ATE tool needed to be domain-agnostic by design and be able to run without any prior training, which already eliminates a host of options. Lastly, certain supplementary data, in the form of known terms and stop-words from that domain, might be available at times but could not be a prerequisite. As a result of these factors, we implemented three methods of term extraction, all of which are applied to input that the tool has already lemmatized. The first method is C-value (Frantzi et al., 2000) , modified to include single-word terms (Barr\u00f3n-Cedeno et al., 2009) . This is likely one of the best-known term search methods in existence. It is language-and domain-independent, does not require any information other than the text input itself, and relies on the kind of linguistic preprocessing (i.e. tagging and category filtering) that would likely always be incorporated by any ATE tool when applied to Icelandic texts. The second method, which we term the 'stem ratio', is one we created specifically for this particular project, and is intended to take advantage of the high number of compound words in Icelandic while remaining unaffected by the issue of multiple potential word orders. When applying this method, the ATE tool employs a separate program called Kvistur, which decompounds Icelandic words (Da\u00f0ason and Bjarnad\u00f3ttir, 2015) . Icelandic compounds are morpohologically right-headed (Bjarnad\u00f3ttir, 2017) , so through Kvistur the ATE tool analyzes the morphological structure of the words contained within each term candidate, extracts all rightmost stems, and compares them to all rightmost stems found in known candidates. If the total number of all the stems in the words of a given candidate is A, and the total number of those same stems in the entirety of known candidates is B, the stem ratio for that candidate is A/B. (Candidates with no compound words are simply not assigned stem ratios.) Hence, the more common that a candidate's morphological heads are within known terms, the higher its stem ratio will be. A candidate with a high stem ratio shares a great deal of both morphological structure and meaning with existing terms, and is itself thus likely to be a new term. It should be noted that Icelandic is a fairly complex language; as such, we will constrain our discussion of ATE methods to ways in which this particular project was implemented, since any further details would require a separate chapter unto themselves.",
                "cite_spans": [
                    {
                        "start": 333,
                        "end": 360,
                        "text": "(Bjarnad\u00f3ttir et al., 2019)",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 913,
                        "end": 935,
                        "text": "(Frantzi et al., 2000)",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 976,
                        "end": 1004,
                        "text": "(Barr\u00f3n-Cedeno et al., 2009)",
                        "ref_id": "BIBREF0"
                    },
                    {
                        "start": 1750,
                        "end": 1782,
                        "text": "(Da\u00f0ason and Bjarnad\u00f3ttir, 2015)",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 1839,
                        "end": 1859,
                        "text": "(Bjarnad\u00f3ttir, 2017)",
                        "ref_id": "BIBREF1"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Methods for term extraction",
                "sec_num": "5.2."
            },
            {
                "text": "The third method is Levenshtein-distance, which in our context is the minimum number of single-character edits required to change one string into another. The Levenshtein algorithm is comparatively straightforward, well supported in Python, and has been used or considered for ATE (Nazarenko and Zargayouna, 2009; Droppo and Acero, 2010) and other term-extraction-type projects in the past (Runkler and Bezdek, 2000) . For each candidate, we find the lowest possible Levenshtein-distance between it and any known term. The lower this value is, the more the candidate resembles a known term letter-for-letter, irrespective of factors such as multiple inflections or morphosyntactic structures.",
                "cite_spans": [
                    {
                        "start": 281,
                        "end": 313,
                        "text": "(Nazarenko and Zargayouna, 2009;",
                        "ref_id": "BIBREF13"
                    },
                    {
                        "start": 314,
                        "end": 337,
                        "text": "Droppo and Acero, 2010)",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 390,
                        "end": 416,
                        "text": "(Runkler and Bezdek, 2000)",
                        "ref_id": "BIBREF17"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Methods for term extraction",
                "sec_num": "5.2."
            },
            {
                "text": "Overall, these three methods cover a wide range of possible terms. The C-value finds those candidates that are clearly important to the input itself, in terms of their unithood and termhood. The stem ratio finds any candidatesgenerally lengthy and complex ones -whose composition, structure and meaningful parts bear clear resemblance to existing terms, even when the less-meaningful parts may be completely dissimilar. Lastly, Levenshtein-distance takes a much rawer approach and finds those candidates -here generally ones that are short or contain simple wordswhich simply resemble known ones in terms of spelling, and which might otherwise be overlooked by the stem ratio. It should be noted that since the latter two methods rely on comparisons to the list of known terms, they will not be adversely affected by candidates' low frequencies of occurrence in the input -but they do require an actual list of terms in order to work at all. In addition, variations on all three approaches are certainly possible, but for the most part we decided to refrain from complicating our algorithms until we'd compiled a solid database of terms for further testing; the one exception being a slight change to C-value calculations to account for single-word terms (Barr\u00f3n-Cedeno et al., 2009).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Methods for term extraction",
                "sec_num": "5.2."
            },
            {
                "text": "Linguistic Processing Tool Set size C-value L-distance S-ratio Recall (%) ABLTagger + Nefnir 500 Table 3 : Average values for ATE methods across linguistic processors for all data sets. Threshold values not applied.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 97,
                        "end": 104,
                        "text": "Table 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Methods for term extraction",
                "sec_num": "5.2."
            },
            {
                "text": "The tool is divided into the following four sections, which run sequentially: Preprocessing, linguistic processing, statistical processing, and output. During preprocessing, the tool checks what data is being supplied -the main input and a list of category patterns are mandatory for every activation, while a list of known terms and a separate list of stop words (not used in our tests) are optional -and loads any comparison data into memory. If a list of known terms is included, its contents are expected to be in lemmatized form for comparison purposes, although if need be the ATE tool itself is capable of lemmatizing the list's contents by using the linguistic support programs detailed below. If the tool is provided with such a list it will also, through the aforementioned program Kvistur, compile an additional list containing the compound word heads of all known terms. In the linguistic section, the tool reviews one line at a time, and tokenizes, tags and lemmatizes it. There are two primary ways in which the tagging and lemmatization may be done, decided on by the tool's administrator: Through the Reynir 8 Python package (\u00deorsteinsson et al., 2019) , or through the tagger ABLTagger 9 (Steingr\u00edmsson et al., 2019) and lemmatizer Nefnir 10 (Ing\u00f3lfsd\u00f3ttir et al., 2019) . For a language as morphologically rich as Icelandic, we felt it necessary to have more than one processing option, although it should be stated that our purpose is not to compare the programs themselves -the primary notable difference is that Reynir automatically performs a more exhaustive and thus more time-consuming analysis. At the end of this section, each line has been converted to a sequence of tuples, where each tuple contains a single, now lemmatized word from the phrase, and a corresponding tag for that word's grammatical category. One last function serves as a bridge to the statistical section: Before the ATE tool applies any of the three extraction methods, it compares each tuple sequence against every single entry in the list of grammatical category patterns known to represent known terms. Any continuous part of the sequence that matches a known grammatical word pattern is automatically added to the list of term candidates. The extraction methods -C-value, stem ratio and Levenshteindistance -are then calculated for every entry on that candi-date list.",
                "cite_spans": [
                    {
                        "start": 1141,
                        "end": 1168,
                        "text": "(\u00deorsteinsson et al., 2019)",
                        "ref_id": null
                    },
                    {
                        "start": 1205,
                        "end": 1233,
                        "text": "(Steingr\u00edmsson et al., 2019)",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 1259,
                        "end": 1287,
                        "text": "(Ing\u00f3lfsd\u00f3ttir et al., 2019)",
                        "ref_id": "BIBREF8"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Usage",
                "sec_num": "5.3."
            },
            {
                "text": "In the output section, the tool prepares the list for use by subsequent parts of TermPortal. The tool also includes threshold values that may be set for each of the three methods, in which case every candidate will have to meet at least one of the thresholds (if applicable, since not all methods are necessarily being applied each time) in order to remain on the candidate list at all. ATE programs generally require specialist input when the final term lists are reviewed. As such, these threshold values help keep the output manageable, particularly while the tool's focus is still on recall.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Usage",
                "sec_num": "5.3."
            },
            {
                "text": "In a project of this nature -where the ATE tool will be applied to an input of continuously changing size and content, rather than a predetermined corpus -the primary focus of evaluation is whether the tool demonstrably works against test inputs of, again, varied sizes and content: If it can properly parse the input, compare it against known terms, find the majority of candidates we know to be present, and display sensible statistical values over a spectrum of different inputs. This effectively means we wanted to measure its recall of the candidates we had intentionally inserted. As noted earlier, measuring precision, on the other hand, was not considered a priority at this stage of development. For a similar reason, we do not focus on narrowing the threshold values during this initial run; rather, we expect to continually adjust them once the TermPortal is actively receiving live data and compiling terminology. Instead, we want to see if the values are being applied in a consistent manner during these initial runs. The results from our four datasets were consistent and promising, as may be seen in Table 3 . Therein, we see the results of applying the two linguistic processing methods and subsequently the three statistical processing methods (C-value, Levenshtein-distance and Stem ratio) to the four data sets described in Table 2 (in each case, the terms being measured for recall were removed from the list of known terms that the program used to calculate Levenshtein-distance and Stem ratio). The lowest recall percentage, 80.0%, resulted from the smallest dataset when parsed by ABLTagger and Nefnir, while Reynir had 92.8% on that same dataset. Larger datasets increased overall recall for ABLTagger and Nefnir with both models, reaching 89.35% on the largest set, while Reynir's lowest recall dipped only to 89.0% with that same set. As may be seen, once the amount of input reaches a particular threshold, the recall rates between the two processing options tend to converge.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 1116,
                        "end": 1123,
                        "text": "Table 3",
                        "ref_id": null
                    },
                    {
                        "start": 1344,
                        "end": 1351,
                        "text": "Table 2",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation",
                "sec_num": "5.4."
            },
            {
                "text": "Averages for the values calculated by the statistical methods -C-value, Levenshtein-distance and stem ratio -were assigned to every possible candidate, not merely the ones on our recall list, and were highly consistent across both linguistic options for every dataset. The highest difference in averages was 0.123 for C-value in the 2,000-line set, 0.596 for Levenshtein-distance in the 500-line set, and 0.808 for stem ratio in the 500-line set. Lastly, it should be noted that between them, these linguistic processing tools collectively managed impressive recall.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation",
                "sec_num": "5.4."
            },
            {
                "text": "In fact, out of the 2,000 known terms we inserted into the largest dataset, only 100 failed to be acknowledged at all. Given that many of the financial terms contain complex words that may at times be quite dissimilar from most text that the linguistic programs were trained on or programmed to recognize, a collective 95% recall rate -meaning that in at least one of the two processing options, the words were correctly tokenized, tagged, lemmatized, matched to known grammatical category patterns, and passed on for value calculation -is a highly positive result. As noted earlier, the two options offer differing depths of linguistic processing, with the associated increase in workload and processing time. As such, Reynir is likely to be used more often on shorter texts, particularly if a more thorough approach is required, while ABLTagger and Nefnir are the preferred choice for processing greater volumes of incoming text at a reasonable pace.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation",
                "sec_num": "5.4."
            },
            {
                "text": "The TermPortal is in closed testing. ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Availability and licensing",
                "sec_num": "6."
            },
            {
                "text": "We have presented TermPortal, a workbench for terminology work using an automated term extraction tool, adapted to Icelandic and the domain of finance. The automatic term extraction tool, built for the workbench, shows promising results with a recall rate of up to 95%. The workbench and the ATE tool show great potential in answering the needs of industry, as manifested in a survey we conducted among the most prominent user group, which shows great interest in improving the state of affairs in Icelandic terminology work within the field of finance. We have implemented approaches to term extraction suitable to data at hand. As more data accrues we expect to develop a far more robust test set than the one used for our initial tests. This will permit greater granularity of test results, along with variations such as testing other term ratios than 50/50 in the program's input. Other future work may include using deep learning approaches, such as word embeddings and bilingual extraction where parallel data is available. To improve the workbench, prospective users will be involved in testing and the resulting feedback used",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion and Future Work",
                "sec_num": "7."
            },
            {
                "text": "http://idord.arnastofnun.is 2 http://hugtakasafn.utn.stjr.is",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "term.tilde.com",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "www.terminologue.org 5 The Icelandic Term Bank: https://idord.is,",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "https://pypi.org/project/reynir 9 https://github.com/steinst/ABLTagger 10 https://github.com/jonfd/Nefnir",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "https://termportal.arnastofnun.is to help adapt the system even further to the needs of users. Users are also expected to help test the quality of our term databases, with an eye toward improving the precision with which the ATE tool collects new terms. Furthermore, user testing will yield precision statistics for the ATE tool, enabling us to tweak the parameters of the system to give a good balance of precision and recall.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "An improved automatic term recognition method for Spanish",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Barr\u00f3n-Cedeno",
                        "suffix": ""
                    },
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Sierra",
                        "suffix": ""
                    },
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Drouin",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Ananiadou",
                        "suffix": ""
                    }
                ],
                "year": 2009,
                "venue": "International Conference on Intelligent Text Processing and Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "125--136",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Barr\u00f3n-Cedeno, A., Sierra, G., Drouin, P., and Ananiadou, S. (2009). An improved automatic term recognition method for Spanish. In International Conference on In- telligent Text Processing and Computational Linguistics, pages 125-136. Springer.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Phrasal compounds in modern Icelandic with reference to Icelandic word formation in general",
                "authors": [
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Bjarnad\u00f3ttir",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [
                            "I"
                        ],
                        "last": "Hlynsd\u00f3ttir",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Steingr\u00edmsson",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "Proceedings of the 22nd Nordic Conference on Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "13--48",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Bjarnad\u00f3ttir, K., Hlynsd\u00f3ttir, K. I., and Steingr\u00edmsson, S. (2019). DIM: The Database of Icelandic Morphology. In Proceedings of the 22nd Nordic Conference on Compu- tational Linguistics, NODALIDA 2019, Turku, Finland. Bjarnad\u00f3ttir, K. (2017). Phrasal compounds in modern Icelandic with reference to Icelandic word formation in general. In Further investigations into the nature of phrasal compounding, pages 13-48. Language Science Press, Berlin.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Kvistur: V\u00e9lraen stofnhlutagreining samsettra or\u00f0a",
                "authors": [
                    {
                        "first": "J",
                        "middle": [
                            "F"
                        ],
                        "last": "Da\u00f0ason",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Bjarnad\u00f3ttir",
                        "suffix": ""
                    }
                ],
                "year": 2015,
                "venue": "Or\u00f0 og tunga",
                "volume": "17",
                "issue": "",
                "pages": "115--132",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Da\u00f0ason, J. F. and Bjarnad\u00f3ttir, K. (2015). Kvistur: V\u00e9l- raen stofnhlutagreining samsettra or\u00f0a. Or\u00f0 og tunga, 17:115-132.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Context dependent phonetic string edit distance for automatic speech recognition",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Droppo",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Acero",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "2010 IEEE International Conference on Acoustics, Speech and Signal Processing",
                "volume": "",
                "issue": "",
                "pages": "4358--4361",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Droppo, J. and Acero, A. (2010). Context dependent pho- netic string edit distance for automatic speech recogni- tion. In 2010 IEEE International Conference on Acous- tics, Speech and Signal Processing, pages 4358-4361. IEEE.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Automatic recognition of multi-word terms:. the C-value/NCvalue method",
                "authors": [
                    {
                        "first": "K",
                        "middle": [
                            "T"
                        ],
                        "last": "Frantzi",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Ananiadou",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Mima",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "International Journal on Digital Libraries",
                "volume": "3",
                "issue": "",
                "pages": "115--130",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Frantzi, K. T., Ananiadou, S., and Mima, H. (2000). Auto- matic recognition of multi-word terms:. the C-value/NC- value method. International Journal on Digital Li- braries, 3:115-130.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "eTranslation TermBank: stimulating the collection of terminological resources for automated translation",
                "authors": [
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Gornostaja",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Auksori\u016bt\u0117",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Dahlberg",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Domeij",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Van Dorrestein",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Hallberg",
                        "suffix": ""
                    },
                    {
                        "first": "L",
                        "middle": [],
                        "last": "Henriksen",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Kallas",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Krek",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Lagzdin",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the XVIII EURALEX International Congress",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gornostaja, T., Auksori\u016bt\u0117, A., Dahlberg, S., Domeij, R., van Dorrestein, M., Hallberg, K., Henriksen, L., Kallas, J., Krek, S., Lagzdin , \u0161, A., et al. (2018). eTransla- tion TermBank: stimulating the collection of termino- logical resources for automated translation. In Proceed- ings of the XVIII EURALEX International Congress, EU- RALEX 2018, Ljubljana, Slovenia.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Terminology resources and terminology work benefit from cloud services",
                "authors": [
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Gornostay",
                        "suffix": ""
                    },
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Vasiljevs",
                        "suffix": ""
                    }
                ],
                "year": 2014,
                "venue": "Proceedings of the Ninth International Conference on Language Resources and Evaluation",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gornostay, T. and Vasiljevs, A. (2014). Terminology re- sources and terminology work benefit from cloud ser- vices. In Proceedings of the Ninth International Con- ference on Language Resources and Evaluation, LREC 2014, Reykjavik, Iceland.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "A deep learning approach to bilingual lexicon induction in the biomedical domain",
                "authors": [
                    {
                        "first": "G",
                        "middle": [],
                        "last": "Heyman",
                        "suffix": ""
                    },
                    {
                        "first": "I",
                        "middle": [],
                        "last": "Vulic",
                        "suffix": ""
                    },
                    {
                        "first": "M.-F",
                        "middle": [],
                        "last": "Moens",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "BMC Bioinformatics",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Heyman, G., Vulic, I., and Moens, M.-F. (2018). A deep learning approach to bilingual lexicon induction in the biomedical domain. In BMC Bioinformatics.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Nefnir: A high accuracy lemmatizer for Icelandic",
                "authors": [
                    {
                        "first": "S",
                        "middle": [
                            "L"
                        ],
                        "last": "Ing\u00f3lfsd\u00f3ttir",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Loftsson",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [
                            "F"
                        ],
                        "last": "Da\u00f0ason",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Bjarnad\u00f3ttir",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the 22nd Nordic Conference on Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ing\u00f3lfsd\u00f3ttir, S. L., Loftsson, H., Da\u00f0ason, J. F., and Bjar- nad\u00f3ttir, K. (2019). Nefnir: A high accuracy lemmatizer for Icelandic. In Proceedings of the 22nd Nordic Confer- ence on Computational Linguistics, NODALIDA 2019, Turku, Finland.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Technical terminology: some linguistic properties and an algorithm for identification in text",
                "authors": [
                    {
                        "first": "J",
                        "middle": [
                            "S"
                        ],
                        "last": "Justeson",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [
                            "M"
                        ],
                        "last": "Katz",
                        "suffix": ""
                    }
                ],
                "year": 1995,
                "venue": "Natural Language Engineering",
                "volume": "1",
                "issue": "",
                "pages": "9--27",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Justeson, J. S. and Katz, S. M. (1995). Technical termi- nology: some linguistic properties and an algorithm for identification in text. Natural Language Engineering, 1:9-27.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Methods of automatic term recognition: A review",
                "authors": [
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Kageura",
                        "suffix": ""
                    },
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Umino",
                        "suffix": ""
                    }
                ],
                "year": 1996,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kageura, K. and Umino, B. (1996). Methods of automatic term recognition: A review.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Towards a unified framework for bilingual terminology extraction of single-word and multi-word terms",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    },
                    {
                        "first": "E",
                        "middle": [],
                        "last": "Morin",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [
                            "P"
                        ],
                        "last": "Saldarriaga",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the 27th International Conference on Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Liu, J., Morin, E., and Saldarriaga, S. P. (2018). Towards a unified framework for bilingual terminology extraction of single-word and multi-word terms. In Proceedings of the 27th International Conference on Computational Linguistics, COLING 2018, Santa Fe, New Mexico.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "IceNLP: A Natural Language Processing Toolkit for Icelandic",
                "authors": [
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Loftsson",
                        "suffix": ""
                    },
                    {
                        "first": "E",
                        "middle": [],
                        "last": "R\u00f6gnvaldsson",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "Proceedings of Interspeech -Speech and language technology for less-resourced languages",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Loftsson, H. and R\u00f6gnvaldsson, E. (2007). IceNLP: A Natural Language Processing Toolkit for Icelandic. In Proceedings of Interspeech -Speech and language tech- nology for less-resourced languages, Interspeech 2007, Antwerp, Belgium.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "Evaluating term extraction",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Nazarenko",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Zargayouna",
                        "suffix": ""
                    }
                ],
                "year": 2009,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Nazarenko, A. and Zargayouna, H. (2009). Evaluating term extraction.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "A wide-coverage context-free grammar for Icelandic and an accompanying parsing system",
                "authors": [
                    {
                        "first": "V",
                        "middle": [],
                        "last": "\u00deorsteinsson",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "\u00d3lad\u00f3ttir",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Loftsson",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the International Conference on Recent Advances in Natural Language Processing",
                "volume": "",
                "issue": "",
                "pages": "1397--1404",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "\u00deorsteinsson, V., \u00d3lad\u00f3ttir, H., and Loftsson, H. (2019). A wide-coverage context-free grammar for Icelandic and an accompanying parsing system. In Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2019), pages 1397-1404.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "Term extraction, tagging, and mapping tools for under-resourced languages",
                "authors": [
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Pinnis",
                        "suffix": ""
                    },
                    {
                        "first": "N",
                        "middle": [],
                        "last": "Ljube\u0161ic",
                        "suffix": ""
                    },
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Stefanescu",
                        "suffix": ""
                    },
                    {
                        "first": "I",
                        "middle": [],
                        "last": "Skadina",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Tadic",
                        "suffix": ""
                    },
                    {
                        "first": "T",
                        "middle": [],
                        "last": "Gornostay",
                        "suffix": ""
                    }
                ],
                "year": 2012,
                "venue": "Proceedings of the 10th Conference on Terminology and Knowledge Engineering",
                "volume": "2012",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Pinnis, M., Ljube\u0161ic, N., Stefanescu, D., Skadina, I., Tadic, M., and Gornostay, T. (2012). Term extraction, tagging, and mapping tools for under-resourced languages. In Proceedings of the 10th Conference on Terminology and Knowledge Engineering, TKE 2012, Madrid, Spain.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "Towards Consolidation of European Terminology Resources: Experience and Recommendations from EuroTermBank Project",
                "authors": [
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Rirdance",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Rirdance, S. (2006). Towards Consolidation of European Terminology Resources: Experience and Recommenda- tions from EuroTermBank Project. Tilde.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Automatic keyword extraction with relational clustering and Levenshtein distances",
                "authors": [
                    {
                        "first": "T",
                        "middle": [
                            "A"
                        ],
                        "last": "Runkler",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [
                            "C"
                        ],
                        "last": "Bezdek",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Ninth IEEE International Conference on Fuzzy Systems. FUZZ-IEEE 2000",
                "volume": "2",
                "issue": "",
                "pages": "636--640",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Runkler, T. A. and Bezdek, J. C. (2000). Automatic key- word extraction with relational clustering and Leven- shtein distances. In Ninth IEEE International Confer- ence on Fuzzy Systems. FUZZ-IEEE 2000 (Cat. No. 00CH37063), volume 2, pages 636-640. IEEE.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "Augmenting a BiLSTM tagger with a morphological lexicon and a lexical category identification step",
                "authors": [
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Steingr\u00edmsson",
                        "suffix": ""
                    },
                    {
                        "first": "\u00d6",
                        "middle": [],
                        "last": "K\u00e1rason",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Loftsson",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of Recent Advances in Natural Language Processing",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Steingr\u00edmsson, S., K\u00e1rason, \u00d6., and Loftsson, H. (2019). Augmenting a BiLSTM tagger with a morphological lex- icon and a lexical category identification step. In Pro- ceedings of Recent Advances in Natural Language Pro- cessing, RANLP 2019, Varna, Bulgaria.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "Bilingual term recognition revisited: the bag-of-equivalents term alignment approach and its evaluation",
                "authors": [
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Vintar",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "Terminology",
                "volume": "16",
                "issue": "",
                "pages": "141--158",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Vintar, S. (2010). Bilingual term recognition revisited: the bag-of-equivalents term alignment approach and its eval- uation. Terminology, 16:141-158.",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "SemRe-Rank: Improving automatic term extraction by incorporating semantic relatedness with personalised PageRank",
                "authors": [
                    {
                        "first": "Z",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Gao",
                        "suffix": ""
                    },
                    {
                        "first": "F",
                        "middle": [],
                        "last": "Ciravegna",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "TKDD",
                "volume": "12",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Zhang, Z., Gao, J., and Ciravegna, F. (2018). SemRe- Rank: Improving automatic term extraction by incorpo- rating semantic relatedness with personalised PageRank. TKDD, 12:57:1-57:41.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "type_str": "figure",
                "num": null,
                "text": "How often do participants look up domain terms? I c e l a n d i c I c e l a n d i c , i f f o u What is the language of choice for terminology?",
                "uris": null
            },
            "FIGREF1": {
                "type_str": "figure",
                "num": null,
                "text": "A term candidate highlighted in context in the TermPortal workbench.",
                "uris": null
            },
            "TABREF0": {
                "content": "<table/>",
                "type_str": "table",
                "num": null,
                "text": "The four test sets.",
                "html": null
            },
            "TABREF2": {
                "content": "<table/>",
                "type_str": "table",
                "num": null,
                "text": "It will be open for use for all parties interested in undertaking terminological work in Iceland, running on servers at The \u00c1rni Magn\u00fasson Institute for Icelandic Studies 11 . The ATE tool is available under an open Apache 2.0 license.",
                "html": null
            }
        }
    }
}