File size: 53,265 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
{
    "paper_id": "2020",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T07:12:50.820460Z"
    },
    "title": "An Analysis of Capsule Networks for Part of Speech Tagging in High-and Low-resource Scenarios *",
    "authors": [
        {
            "first": "Andrew",
            "middle": [],
            "last": "Zupon",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Arizona",
                "location": {}
            },
            "email": "zupon@email.arizona.edu"
        },
        {
            "first": "Faiz",
            "middle": [],
            "last": "Rafique",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Arizona",
                "location": {}
            },
            "email": "faizr@email.arizona.edu"
        },
        {
            "first": "Mihai",
            "middle": [],
            "last": "Surdeanu",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of Arizona",
                "location": {}
            },
            "email": "msurdeanu@email.arizona.edu"
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Neural networks are a common tool in NLP, but it is not always clear which architecture to use for a given task. Different tasks, different languages, and different training conditions can all affect how a neural network will perform. Capsule Networks (CapsNets) are a relatively new architecture in NLP. Due to their novelty, CapsNets are being used more and more in NLP tasks. However, their usefulness is still mostly untested. In this paper, we compare three neural network architectures-LSTM, CNN, and CapsNet-on a part of speech tagging task. We compare these architectures in both high-and low-resource training conditions and find that no architecture consistently performs the best. Our analysis shows that our CapsNet performs nearly as well as a more complex LSTM under certain training conditions, but not others, and that our CapsNet almost always outperforms our CNN. We also find that our CapsNet implementation shows faster prediction times than the LSTM for Scottish Gaelic but not for Spanish, highlighting the effect that the choice of languages can have on the models.",
    "pdf_parse": {
        "paper_id": "2020",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Neural networks are a common tool in NLP, but it is not always clear which architecture to use for a given task. Different tasks, different languages, and different training conditions can all affect how a neural network will perform. Capsule Networks (CapsNets) are a relatively new architecture in NLP. Due to their novelty, CapsNets are being used more and more in NLP tasks. However, their usefulness is still mostly untested. In this paper, we compare three neural network architectures-LSTM, CNN, and CapsNet-on a part of speech tagging task. We compare these architectures in both high-and low-resource training conditions and find that no architecture consistently performs the best. Our analysis shows that our CapsNet performs nearly as well as a more complex LSTM under certain training conditions, but not others, and that our CapsNet almost always outperforms our CNN. We also find that our CapsNet implementation shows faster prediction times than the LSTM for Scottish Gaelic but not for Spanish, highlighting the effect that the choice of languages can have on the models.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "Neural networks have become a common tool in natural language processing (NLP) for many tasks, but are different architectures better suited for different tasks, languages, and/or resources? To try to answer this question, we examine the performance of two common neural network architectures, long short-term memory networks (LSTM) (Greff et al., 2017) and convolutional neural networks (CNN) (LeCun et al., 1989) , against the newer capsule networks (CapsNets), another neural network architecture based on CNNs (Hinton et al., 2011) .",
                "cite_spans": [
                    {
                        "start": 333,
                        "end": 353,
                        "text": "(Greff et al., 2017)",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 394,
                        "end": 414,
                        "text": "(LeCun et al., 1989)",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 514,
                        "end": 535,
                        "text": "(Hinton et al., 2011)",
                        "ref_id": "BIBREF3"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "While LSTMs and CNNs are common in NLP, capsule networks are relatively new to the field. Due to their recency, it's not always clear if or when they are better than other widely used sequence models. This paper investigates the CapsNet architecture in comparison with LSTMs and CNNs. For our analysis, we apply these three architectures to a part of speech (POS) tagging task, on two languages, and using both low-and high-resource scenarios.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Much of the focus of NLP research is on resource-rich languages like English. However, the performance of different models can depend on the linguistic properties of the language under study (Bender, 2009) and the amount of training data available. To compare the performance of these architectures under different training conditions, we look at Spanish-another resource-rich language-and Scottish Gaelic-a low-resource language using different amounts of training data. This comparison is a step in the right direction, but it does have the limitations of comparing neural network architectures implemented in different frameworks and only comparing two languages.",
                "cite_spans": [
                    {
                        "start": 191,
                        "end": 205,
                        "text": "(Bender, 2009)",
                        "ref_id": "BIBREF0"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "The main contribution of this paper is comparing the LSTM, CNN, and CapsNet architectures across different training conditions. Our analysis finds that none of the architectures consistently performs best across training conditions. This illustrates how different languages and training conditions can inform which architecture is best suited for a given NLP task, and that there is no obviously correct answer.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "CapsNets are a relatively new type of neural network. Hinton et al. (2011) introduces the architecture, with modifications by Sabour et al. (2017) (dynamic routing) and Hinton et al. (2018) (EM routing) . A CapsNet is essentially a modified ver-sion of a CNN that trades max pooling for a more data-retentive process called routing by agreement. Instead of the prediction with the highest score getting chosen, the weighted sum of all predictions are considered for classification. Essentially, a Cap-sNet uses convolution to create first round predictions for objects-primary capsules-and then utilizes routing by agreement to predict the presence of higher level objects-secondary capsules.",
                "cite_spans": [
                    {
                        "start": 54,
                        "end": 74,
                        "text": "Hinton et al. (2011)",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 126,
                        "end": 146,
                        "text": "Sabour et al. (2017)",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 169,
                        "end": 189,
                        "text": "Hinton et al. (2018)",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 190,
                        "end": 202,
                        "text": "(EM routing)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "Many implementations of CapsNets are designed for image recognition (Hinton et al., 2011; Sabour et al., 2017; Hinton et al., 2018) . However, the CapsNet architecture is being applied more and more to NLP tasks, including Chinese word segmentation (Li et al., 2018) , and multi-label text classification and question answering (Zhao et al., 2019) . This paper continues this path by investigating how CapsNets compare to other neural network architectures for the task of part of speech tagging.",
                "cite_spans": [
                    {
                        "start": 68,
                        "end": 89,
                        "text": "(Hinton et al., 2011;",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 90,
                        "end": 110,
                        "text": "Sabour et al., 2017;",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 111,
                        "end": 131,
                        "text": "Hinton et al., 2018)",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 249,
                        "end": 266,
                        "text": "(Li et al., 2018)",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 328,
                        "end": 347,
                        "text": "(Zhao et al., 2019)",
                        "ref_id": "BIBREF9"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "Our comparison considers two languages: Spanish 1 and Scottish Gaelic 2 . Spanish is a resource-rich language, being the second most spoken language by number native speakers, fourth most spoken language by total number of speakers, and the third or fourth most widely used language on the internet 3 . Scottish Gaelic is a low-resource language, with 57,375 fluent speakers in Scotland per the 2011 census 4 . The Spanish data come from the UD Spanish AnCora treebank 5 . The Scottish Gaelic data come from the UD ARCOSG treebank 6 . Both corpora use 17 part of speech tag classes.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Data",
                "sec_num": "3"
            },
            {
                "text": "To study how different low-resource conditions affect training, we artificially create training partitions of different sizes. From the original training data (train100), we create partitions consisting of 50% (train50), 10% (train10), and 1% (train1) of the training sentences. The amount of data for each partition is shown in Table 1 for Spanish and Table 2 for Scottish Gaelic. We use FastText word embeddings (Grave et al., 2018) for both Spanish (2,000,000 words) and Scottish Gaelic (14,318 words). The embedding dimension is 300. Table 2 : Number of sentences, tokens, and average sentence length for each partition of Scottish Gaelic. The n in the train partitions corresponds to the amount (percent) of the original data used for training.",
                "cite_spans": [
                    {
                        "start": 414,
                        "end": 434,
                        "text": "(Grave et al., 2018)",
                        "ref_id": "BIBREF1"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 329,
                        "end": 336,
                        "text": "Table 1",
                        "ref_id": "TABREF0"
                    },
                    {
                        "start": 353,
                        "end": 360,
                        "text": "Table 2",
                        "ref_id": null
                    },
                    {
                        "start": 538,
                        "end": 545,
                        "text": "Table 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Data",
                "sec_num": "3"
            },
            {
                "text": "In this section, we describe the implementation details of our CapsNet, CNN, and LSTM methods. Our CapsNet and CNN implementations build on top of Yeung et al.'s implementation 7 , which was kept as close as possible to the architectures described by Sabour et al. (2017) . Importantly, we tried to keep all three models as close to each other as possible in order to make our comparison as faithful as possible. However, certain differences persist for this project-for example, the CapsNet and CNN are implemented in Python using Tensorflow 8 and Keras 9 , whereas the LSTM is implemented in Scala using DyNet. 10 The hyperparameters for our CapsNet and CNN implementations were chosen to be as close as possible to the original implementation. The hyperparameters of the LSTM were chosen to be a reasonable approximation to the CapsNet and CNN models. It is important to note that our comparison does not attempt to compare the best of the best of each architecture.",
                "cite_spans": [
                    {
                        "start": 251,
                        "end": 271,
                        "text": "Sabour et al. (2017)",
                        "ref_id": "BIBREF8"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Approach",
                "sec_num": "4"
            },
            {
                "text": "7 https://github.com/Chucooleg/CapsNet_ for_NER 8 https://www.tensorflow.org/ 9 https://keras.io/ 10 We used the implementation from the processors library (https://github.com/clulab/processors), which relies on DyNet (https://dynet.readthedocs. Table 3 : Spanish Precision, Recall, and F1 scores. The scores are an average of 5 different random seeds and their standard deviation, along with the average training/prediction times of each model.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 246,
                        "end": 253,
                        "text": "Table 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Approach",
                "sec_num": "4"
            },
            {
                "text": "Our CapsNet model has two 1D convolutional layers, two routing by agreement capsule layers and one fully connected layer. Both convolutional layers have 256 channels, a kernel size of 3, and a stride of 1. The primary capsule layer has 160 capsules with 8 dimensions, a kernel size of 3 and stride of 1. There are 17 secondary capsules with dimensions of 16 and 3 dynamic routing passes.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "CapsNet Implementation",
                "sec_num": "4.1"
            },
            {
                "text": "Our CNN model has three 1D convolutional layers, a max pooling layer, and two fully connected layers. The first two convolutional layers are identical to the first two layers of the CapsNet. The third convolutional layer has 128 channels, size of 3 and stride 1. The two feed-forward layers have a size fo 328 and 192. These settings were chosen to make the CNN implementation as comparable as possible to the CapsNet implementation.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "CNN Implementation",
                "sec_num": "4.2"
            },
            {
                "text": "io).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "CNN Implementation",
                "sec_num": "4.2"
            },
            {
                "text": "The LSTM code we used is a reimplementation of the LSTM-CRF approach of Lample et al. (2016) . To make this implementation as similar as possible with the previous two approaches, we: (a) removed the CRF layer, 11 and (b) removed the characterlevel biLSTM encoder from the word embeddings. Thus, the actual LSTM architecture used consists of three layers: (i) an input layer with 300dimensional FastText word embeddings; (ii) one biLSTM intermediate layer, where each LSTM has a hidden state of dimension 128 neurons, and (iii) a linear output layer coupled with a softmax function to output the POS tags.",
                "cite_spans": [
                    {
                        "start": 72,
                        "end": 92,
                        "text": "Lample et al. (2016)",
                        "ref_id": "BIBREF5"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "LSTM Implementation",
                "sec_num": "4.3"
            },
            {
                "text": "In addition to our four training data conditions per language, we evaluate the use of learning the word embeddings during training for all models (\"learn\" vs. \"no learn\"), yielding 24 training conditions per language. We trained all models five times with a different random seed and averaged the results. Each condition trained for 10 epochs, with early stopping after 2 epochs if the loss did not improve. The results are given in Table 3 (Spanish) and Table 4 (Scottish Gaelic). We report Precision, Recall, F1, training time, and prediction time. These results show a few trends:",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 433,
                        "end": 440,
                        "text": "Table 3",
                        "ref_id": null
                    },
                    {
                        "start": 455,
                        "end": 462,
                        "text": "Table 4",
                        "ref_id": "TABREF3"
                    }
                ],
                "eq_spans": [],
                "section": "Results",
                "sec_num": "5"
            },
            {
                "text": "1. The LSTM always outperforms the CapsNet and CNN for Spanish, but the CapsNet and CNN occasionally outperform the LSTM for Scottish Gaelic, whose training dataset is an order of magnitude smaller than the Spanish one.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Results",
                "sec_num": "5"
            },
            {
                "text": "The difference in F1 on the no learn train condition between the Spanish 10% and Scottish Gaelic 100% partitions, which have a comparable number of sentences, is greater for the LSTM (down 9.93%) than the Capsnet (down 2.98%) or CNN (up 3.93%). This suggests that properties of the language, not just the amount of data, play a role in performance.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "2.",
                "sec_num": null
            },
            {
                "text": "3. The LSTM benefits only slightly from using learned embeddings, while both the CapsNet and CNN get a much larger performance boost. We see this in the Spanish 1% condition, where the LSTM F1 improves by 0.01%, but the CapsNet and CNN models improve by 4.82% and 3.39%, respectively. 4. Another obvious difference is in the model training and prediction times. The training time for the CapsNet and CNN is much slower than the LSTM. However, for the Scottish Gaelic case CapsNets are much faster than the LSTM at prediction time. This is an encouraging result, considering that our CapsNet implementation is in Python, whereas the LSTM is implemented in a faster Scala framework.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "2.",
                "sec_num": null
            },
            {
                "text": "Overall, the LSTM performs best in most conditions, but the CapsNet often comes close. The CapsNet also usually outperforms the CNN. These performance differences are potentially offset by faster prediction time, depending on the language. The balance between predictive accuracy, training time, and prediction time can be delicate, especially when looking at low-resource languages. These results suggest that depending on the use case, a Cap-sNet architecture may be preferable to an LSTM, despite the fact that when more resources are available, the LSTM tends to perform the best under the common hyperparameters investigated here.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "2.",
                "sec_num": null
            },
            {
                "text": "We also compared different hyperparameters for the LSTM and CapsNet, which is shown in Table 5 . The values we chose for the LSTM hidden state size and the CapsNet capsule layer kernel size perform the best in nearly all conditions.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 87,
                        "end": 94,
                        "text": "Table 5",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "2.",
                "sec_num": null
            },
            {
                "text": "In this paper, we compare the performance of three neural network architectures-LSTM, CNN, and CapsNet-on part of speech tagging and find that LSTMs are not always better under the common hyperparameters investigated. We examine how the best performing model changes under different high-and low-resource training conditions using Spanish and Scottish Gaelic. We show that the relatively new CapsNet architecture performs nearly as well as the more complex LSTM under certain conditions and outperforms the CNN under most conditions we examined. These results suggest that there is no one obviously clear choice for a model architecture, and that the properties of a language and the amount of training data can affect which architecture performs best. Future work should address the limitations of this paper. Specifically, future effort should consider more training conditions, including other languages; the consistency of these results within groups of similar languages; and making the implementation of these architectures closer, to guarantee the performance differences are due to the architecture and not an artifact of how they were implemented.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion",
                "sec_num": "6"
            },
            {
                "text": "In initial experiments we observed that the CRF layer had a major contribution to other sequence models such as named entity recognition, but no impact on POS tagging.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Linguistically na\u00efve != language independent: Why NLP needs linguistic typology",
                "authors": [
                    {
                        "first": "Emily",
                        "middle": [
                            "M"
                        ],
                        "last": "Bender",
                        "suffix": ""
                    }
                ],
                "year": 2009,
                "venue": "Proceedings of the EACL 2009 Workshop on the Interaction between Linguistics and Computational Linguistics: Virtuous, Vicious or Vacuous?",
                "volume": "",
                "issue": "",
                "pages": "26--32",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Emily M. Bender. 2009. Linguistically na\u00efve != lan- guage independent: Why NLP needs linguistic ty- pology. In Proceedings of the EACL 2009 Workshop on the Interaction between Linguistics and Compu- tational Linguistics: Virtuous, Vicious or Vacuous?, pages 26-32, Athens, Greece. Association for Com- putational Linguistics.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Learning word vectors for 157 languages",
                "authors": [
                    {
                        "first": "Edouard",
                        "middle": [],
                        "last": "Grave",
                        "suffix": ""
                    },
                    {
                        "first": "Piotr",
                        "middle": [],
                        "last": "Bojanowski",
                        "suffix": ""
                    },
                    {
                        "first": "Prakhar",
                        "middle": [],
                        "last": "Gupta",
                        "suffix": ""
                    },
                    {
                        "first": "Armand",
                        "middle": [],
                        "last": "Joulin",
                        "suffix": ""
                    },
                    {
                        "first": "Tomas",
                        "middle": [],
                        "last": "Mikolov",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "Proceedings of the International Conference on Language Resources and Evaluation",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Edouard Grave, Piotr Bojanowski, Prakhar Gupta, Ar- mand Joulin, and Tomas Mikolov. 2018. Learning word vectors for 157 languages. In Proceedings of the International Conference on Language Re- sources and Evaluation (LREC 2018).",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "LSTM: A search space odyssey",
                "authors": [
                    {
                        "first": "Klaus",
                        "middle": [],
                        "last": "Greff",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Rupesh",
                        "suffix": ""
                    },
                    {
                        "first": "Jan",
                        "middle": [],
                        "last": "Srivastava",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Koutn\u00edk",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [],
                        "last": "Bas",
                        "suffix": ""
                    },
                    {
                        "first": "J\u00fcrgen",
                        "middle": [],
                        "last": "Steunebrink",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Schmidhuber",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "IEEE Transactions on Neural Networks and Learning Systems",
                "volume": "28",
                "issue": "10",
                "pages": "2222--2232",
                "other_ids": {
                    "DOI": [
                        "10.1109/TNNLS.2016.2582924"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Klaus Greff, Rupesh K. Srivastava, Jan Koutn\u00edk, Bas R. Steunebrink, and J\u00fcrgen Schmidhuber. 2017. LSTM: A search space odyssey. IEEE Transac- tions on Neural Networks and Learning Systems, 28(10):2222-2232.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Transforming auto-encoders",
                "authors": [
                    {
                        "first": "Geoffrey",
                        "middle": [
                            "E"
                        ],
                        "last": "Hinton",
                        "suffix": ""
                    },
                    {
                        "first": "Alex",
                        "middle": [],
                        "last": "Krizhevsky",
                        "suffix": ""
                    },
                    {
                        "first": "Sida",
                        "middle": [
                            "D"
                        ],
                        "last": "Wang",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "Artificial Neural Networks and Machine Learning",
                "volume": "",
                "issue": "",
                "pages": "44--51",
                "other_ids": {
                    "DOI": [
                        "10.1007/978-3-642-21735-7_6"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Geoffrey E. Hinton, Alex Krizhevsky, and Sida D. Wang. 2011. Transforming auto-encoders. In Artifi- cial Neural Networks and Machine Learning, pages 44-51. Springer.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Matrix capsules with EM routing",
                "authors": [
                    {
                        "first": "Geoffrey",
                        "middle": [
                            "E"
                        ],
                        "last": "Hinton",
                        "suffix": ""
                    },
                    {
                        "first": "Sara",
                        "middle": [],
                        "last": "Sabour",
                        "suffix": ""
                    },
                    {
                        "first": "Nicholas",
                        "middle": [],
                        "last": "Frosst",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "International Conference on Learning Representations",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Geoffrey E. Hinton, Sara Sabour, and Nicholas Frosst. 2018. Matrix capsules with EM routing. In Interna- tional Conference on Learning Representations.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Neural architectures for named entity recognition",
                "authors": [
                    {
                        "first": "Guillaume",
                        "middle": [],
                        "last": "Lample",
                        "suffix": ""
                    },
                    {
                        "first": "Miguel",
                        "middle": [],
                        "last": "Ballesteros",
                        "suffix": ""
                    },
                    {
                        "first": "Sandeep",
                        "middle": [],
                        "last": "Subramanian",
                        "suffix": ""
                    },
                    {
                        "first": "Kazuya",
                        "middle": [],
                        "last": "Kawakami",
                        "suffix": ""
                    },
                    {
                        "first": "Chris",
                        "middle": [],
                        "last": "Dyer",
                        "suffix": ""
                    }
                ],
                "year": 2016,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "arXiv": [
                        "arXiv:1603.01360"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Guillaume Lample, Miguel Ballesteros, Sandeep Sub- ramanian, Kazuya Kawakami, and Chris Dyer. 2016. Neural architectures for named entity recognition. arXiv preprint arXiv:1603.01360.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Backpropagation applied to handwritten zip code recognition",
                "authors": [
                    {
                        "first": "Y",
                        "middle": [],
                        "last": "Lecun",
                        "suffix": ""
                    },
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Boser",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [
                            "S"
                        ],
                        "last": "Denker",
                        "suffix": ""
                    },
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Henderson",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [
                            "E"
                        ],
                        "last": "Howard",
                        "suffix": ""
                    },
                    {
                        "first": "W",
                        "middle": [],
                        "last": "Hubbard",
                        "suffix": ""
                    },
                    {
                        "first": "L",
                        "middle": [
                            "D"
                        ],
                        "last": "",
                        "suffix": ""
                    }
                ],
                "year": 1989,
                "venue": "Neural Computation",
                "volume": "1",
                "issue": "4",
                "pages": "541--551",
                "other_ids": {
                    "DOI": [
                        "10.1162/neco.1989.1.4.541"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Y. LeCun, B. Boser, J. S. Denker, D. Henderson, R. E. Howard, W. Hubbard, and L. D. Jackel. 1989. Back- propagation applied to handwritten zip code recog- nition. Neural Computation, 1(4):541-551.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Capsules based chinese word segmentation for ancient chinese medical books",
                "authors": [
                    {
                        "first": "Si",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    },
                    {
                        "first": "Mingzheng",
                        "middle": [],
                        "last": "Li",
                        "suffix": ""
                    },
                    {
                        "first": "Yajing",
                        "middle": [],
                        "last": "Xu",
                        "suffix": ""
                    },
                    {
                        "first": "Zuyi",
                        "middle": [],
                        "last": "Bao",
                        "suffix": ""
                    },
                    {
                        "first": "Lu",
                        "middle": [],
                        "last": "Fu",
                        "suffix": ""
                    },
                    {
                        "first": "Yan",
                        "middle": [],
                        "last": "Zhu",
                        "suffix": ""
                    }
                ],
                "year": 2018,
                "venue": "IEEE Access",
                "volume": "6",
                "issue": "",
                "pages": "70874--70883",
                "other_ids": {
                    "DOI": [
                        "10.1109/ACCESS.2018.2881280"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Si Li, Mingzheng Li, Yajing Xu, Zuyi Bao, Lu Fu, and Yan Zhu. 2018. Capsules based chinese word seg- mentation for ancient chinese medical books. IEEE Access, 6:70874-70883.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Dynamic routing between capsules",
                "authors": [
                    {
                        "first": "Sara",
                        "middle": [],
                        "last": "Sabour",
                        "suffix": ""
                    },
                    {
                        "first": "Nicholas",
                        "middle": [],
                        "last": "Frosst",
                        "suffix": ""
                    },
                    {
                        "first": "Geoffrey",
                        "middle": [
                            "E"
                        ],
                        "last": "Hinton",
                        "suffix": ""
                    }
                ],
                "year": 2017,
                "venue": "Advances in Neural Information Processing Systems",
                "volume": "30",
                "issue": "",
                "pages": "3856--3866",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Sara Sabour, Nicholas Frosst, and Geoffrey E. Hinton. 2017. Dynamic routing between capsules. In Is- abelle Guyon, Ulrike. Von Luxburg, Samy Bengio, Hanna Wallach, Rob Fergus, S.V.N. Vishwanathan, and Roman Garnett, editors, Advances in Neural In- formation Processing Systems 30, pages 3856-3866. Curran Associates, Inc.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Towards scalable and reliable capsule networks for challenging nlp applications",
                "authors": [
                    {
                        "first": "Wei",
                        "middle": [],
                        "last": "Zhao",
                        "suffix": ""
                    },
                    {
                        "first": "Haiyun",
                        "middle": [],
                        "last": "Peng",
                        "suffix": ""
                    },
                    {
                        "first": "Steffen",
                        "middle": [],
                        "last": "Eger",
                        "suffix": ""
                    },
                    {
                        "first": "Erik",
                        "middle": [],
                        "last": "Cambria",
                        "suffix": ""
                    },
                    {
                        "first": "Min",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    }
                ],
                "year": 2019,
                "venue": "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {
                    "DOI": [
                        "10.18653/v1/p19-1150"
                    ]
                },
                "num": null,
                "urls": [],
                "raw_text": "Wei Zhao, Haiyun Peng, Steffen Eger, Erik Cambria, and Min Yang. 2019. Towards scalable and reliable capsule networks for challenging nlp applications. Proceedings of the 57th Annual Meeting of the As- sociation for Computational Linguistics.",
                "links": null
            }
        },
        "ref_entries": {
            "TABREF0": {
                "html": null,
                "type_str": "table",
                "content": "<table><tr><td>train100</td><td>14,305</td><td>446,144</td><td>31.2</td></tr><tr><td>train50</td><td>7,152</td><td>255,213</td><td>35.7</td></tr><tr><td>train10</td><td>1,430</td><td>43,480</td><td>30.4</td></tr><tr><td>train1</td><td>143</td><td>5,912</td><td>41.3</td></tr><tr><td>dev</td><td>1,654</td><td>52,511</td><td>31.7</td></tr><tr><td>test</td><td>1,721</td><td>52,801</td><td>30.7</td></tr><tr><td colspan=\"4\">Partition Sentences Tokens Avg. Sent. Length</td></tr><tr><td>train100</td><td>1,015</td><td>22,963</td><td>22.6</td></tr><tr><td>train50</td><td>507</td><td>10,870</td><td>21.4</td></tr><tr><td>train10</td><td>101</td><td>1,543</td><td>15.3</td></tr><tr><td>train1</td><td>10</td><td>67</td><td>6.7</td></tr><tr><td>dev</td><td>642</td><td>9,949</td><td>15.5</td></tr><tr><td>test</td><td>536</td><td>9,946</td><td>18.6</td></tr></table>",
                "num": null,
                "text": "Number of sentences, tokens, and average sentence length for each partition of Spanish. The n in the train partitions corresponds to the amount (percent) of the original data used for training."
            },
            "TABREF1": {
                "html": null,
                "type_str": "table",
                "content": "<table><tr><td>Model</td><td>P</td><td>R</td><td>F1</td><td>Train t</td><td>Predict t</td></tr><tr><td/><td/><td/><td>03)</td><td>3,222 s</td><td>165 s</td></tr><tr><td>50% of train, caps, no learn</td><td colspan=\"3\">92.78 (0.45) 93.58 (0.20) 93.18 (0.31)</td><td>5,386 s</td><td>223</td></tr><tr><td>50% of train, cnn, no learn</td><td colspan=\"3\">92.36 (0.59) 93.54 (0.12) 92.95 (0.29)</td><td>3,392 s</td><td>206 s</td></tr><tr><td>50% of train, lstm, no learn</td><td colspan=\"3\">98.31 (0.03) 98.31 (0.03) 98.31 (0.03)</td><td>1,566 s</td><td>175 s</td></tr><tr><td>10% of train, caps, no learn</td><td colspan=\"3\">88.37 (0.73) 89.48 (0.67) 88.92 (0.56)</td><td>3,186 s</td><td>208 s</td></tr><tr><td>10% of train, cnn, no learn</td><td colspan=\"3\">88.23 (0.60) 89.17 (0.45) 88.70 (0.33)</td><td>3,373 s</td><td>191 s</td></tr><tr><td>10% of train, lstm, no learn</td><td colspan=\"3\">96.89 (0.14) 96.89 (0.14) 96.89 (0.14)</td><td>613 s</td><td>170 s</td></tr><tr><td>1% of train, caps, no learn</td><td colspan=\"3\">76.63 (1.11) 80.62 (0.74) 78.56 (0.30)</td><td>3,370 s</td><td>205 s</td></tr><tr><td>1% of train, cnn, no learn</td><td colspan=\"3\">73.96 (2.66) 74.78 (0.68) 74.34 (1.44)</td><td>2,898 s</td><td>187 s</td></tr><tr><td>1% of train, lstm, no learn</td><td colspan=\"3\">91.79 (0.24) 91.79 (0.24) 91.79 (0.24)</td><td>375 s</td><td>162 s</td></tr><tr><td>100% of train, caps, learn</td><td colspan=\"4\">96.30 (0.35) 95.61 (0.19) 96.00 (0.08) 14,223 s</td><td>219 s</td></tr><tr><td>100% of train, cnn, learn</td><td colspan=\"4\">96.01 (0.34) 95.43 (0.15) 95.72 (0.16) 12,794 s</td><td>211 s</td></tr><tr><td>100% of train, lstm, learn</td><td colspan=\"3\">98.43 (0.04) 98.43 (0.04) 98.43 (0.04)</td><td>4,280 s</td><td>172 s</td></tr><tr><td>50% of train, caps, learn</td><td colspan=\"4\">95.59 (0.27) 94.68 (0.16) 95.13 (0.09) 11,571 s</td><td>225 s</td></tr><tr><td>50% of train, cnn, learn</td><td colspan=\"4\">95.36 (0.10) 94.61 (0.08) 94.98 (0.06) 11,318 s</td><td>206 s</td></tr><tr><td>50% of train, lstm, learn</td><td colspan=\"3\">98.17 (0.06) 98.17 (0.06) 98.17 (0.06)</td><td>1,333 s</td><td>171 s</td></tr><tr><td>10% of train, caps, learn</td><td colspan=\"3\">92.45 (0.26) 90.18 (0.32) 91.30 (0.08)</td><td>3,767 s</td><td>209 s</td></tr><tr><td>10% of train, cnn, learn</td><td colspan=\"3\">91.41 (0.42) 89.49 (0.25) 90.44 (0.20)</td><td>3,832 s</td><td>191 s</td></tr><tr><td>10% of train, lstm, learn</td><td colspan=\"3\">96.84 (0.07) 96.84 (0.07) 96.84 (0.07)</td><td>1,157 s</td><td>172 s</td></tr><tr><td>1% of train, caps, learn</td><td colspan=\"3\">84.12 (0.66) 82.65 (0.46) 83.38 (0.24)</td><td>3,452 s</td><td>205 s</td></tr><tr><td>1% of train, cnn, learn</td><td colspan=\"3\">79.71 (0.91) 75.87 (1.01) 77.73 (0.35)</td><td>2,979 s</td><td>188 s</td></tr><tr><td>1% of train, lstm, learn</td><td colspan=\"3\">91.80 (0.21) 91.80 (0.21) 91.80 (0.21)</td><td>440 s</td><td>178 s</td></tr></table>",
                "num": null,
                "text": "100% of train, caps, no learn 93.85 (0.35) 94.47 (0.24) 94.16 (0.23) 9,032 s 218 s 100% of train, cnn, no learn 93.76 (0.40) 94.20 (0.11) 93.98 (0.22) 4,802 s 200 s 100% of train, lstm, no learn 98.54 (0.03) 98.54 (0.03) 98.54 (0."
            },
            "TABREF2": {
                "html": null,
                "type_str": "table",
                "content": "<table><tr><td>Model</td><td>P</td><td>R</td><td>F1</td><td colspan=\"2\">Train t Predict t</td></tr><tr><td/><td/><td/><td>30)</td><td>256 s</td><td>38 s</td></tr><tr><td>50% of train, caps, no learn</td><td colspan=\"3\">75.90 (1.02) 73.98 (0.47) 74.92 (0.54)</td><td>809 s</td><td>18 s</td></tr><tr><td>50% of train, cnn, no learn</td><td colspan=\"3\">71.97 (1.62) 69.33 (0.80) 70.61 (0.47)</td><td>616 s</td><td>15 s</td></tr><tr><td>50% of train, lstm, no learn</td><td colspan=\"3\">75.36 (0.27) 75.36 (0.27) 75.36 (0.27)</td><td>119 s</td><td>37 s</td></tr><tr><td>10% of train, caps, no learn</td><td colspan=\"3\">49.24 (2.78) 37.36 (1.43) 42.44 (1.35)</td><td>641 s</td><td>19 s</td></tr><tr><td>10% of train, cnn, no learn</td><td colspan=\"3\">54.64 (2.16) 50.04 (1.69) 52.18 (0.64)</td><td>555 s</td><td>15 s</td></tr><tr><td>10% of train, lstm, no learn</td><td colspan=\"3\">53.31 (8.11) 53.31 (8.11) 53.31 (8.11)</td><td>86 s</td><td>40 s</td></tr><tr><td>1% of train, caps, no learn</td><td>7.94 (1.40)</td><td>7.82 (1.28)</td><td>7.86 (1.26)</td><td>486 s</td><td>18 s</td></tr><tr><td>1% of train, cnn, no learn</td><td>16.87 (3.86)</td><td>8.42 (3.52)</td><td>10.58 (3.09)</td><td>546 s</td><td>16 s</td></tr><tr><td>1% of train, lstm, no learn</td><td colspan=\"3\">21.37 (3.17) 21.37 (3.17) 21.37 (3.17)</td><td>44 s</td><td>35 s</td></tr><tr><td>100% of train, caps, learn</td><td colspan=\"4\">90.81 (0.35) 87.91 (0.25) 89.34 (0.25) 1,907 s</td><td>19 s</td></tr><tr><td>100% of train, cnn, learn</td><td colspan=\"4\">88.82 (0.80) 85.57 (0.50) 87.17 (0.32) 1,742 s</td><td>16 s</td></tr><tr><td>100% of train, lstm, learn</td><td colspan=\"3\">89.84 (0.15) 89.84 (0.15) 89.84 (0.15)</td><td>317 s</td><td>39 s</td></tr><tr><td>50% of train, caps, learn</td><td colspan=\"4\">85.38 (0.62) 81.63 (0.22) 83.46 (0.32) 1,285 s</td><td>18 s</td></tr><tr><td>50% of train, cnn, learn</td><td colspan=\"4\">82.26 (0.63) 77.95 (0.48) 80.05 (0.54) 1,075 s</td><td>16 s</td></tr><tr><td>50% of train, lstm, learn</td><td colspan=\"3\">83.66 (0.57) 83.66 (0.57) 83.66 (0.57)</td><td>155 s</td><td>39 s</td></tr><tr><td>10% of train, caps, learn</td><td colspan=\"3\">55.16 (1.55) 46.21 (2.04) 50.26 (1.45)</td><td>717 s</td><td>19 s</td></tr><tr><td>10% of train, cnn, learn</td><td colspan=\"3\">57.84 (3.16) 55.13 (1.53) 56.37 (1.26)</td><td>629 s</td><td>15 s</td></tr><tr><td>10% of train, lstm, learn</td><td colspan=\"3\">65.49 (0.67) 65.49 (0.67) 65.49 (0.67)</td><td>101 s</td><td>40 s</td></tr><tr><td>1% of train, caps, learn</td><td>8.31 (1.99)</td><td>8.55 (2.25)</td><td>8.42 (2.10)</td><td>509 s</td><td>19 s</td></tr><tr><td>1% of train, cnn, learn</td><td colspan=\"3\">18.23 (3.47) 11.63 (5.57) 13.63 (5.31)</td><td>544 s</td><td>16 s</td></tr><tr><td>1% of train, lstm, learn</td><td colspan=\"3\">21.37 (3.17) 21.37 (3.17) 21.37 (3.17)</td><td>52 s</td><td>36 s</td></tr></table>",
                "num": null,
                "text": "100% of train, caps, no learn 82.34 (0.82) 80.58 (0.22) 81.45 (0.32) 1,020 s 19 s 100% of train, cnn, no learn 79.40 (1.12) 77.19 (0.96) 78.27 (0.53) 651 s 16 s 100% of train, lstm, no learn 81.86 (0.30) 81.86 (0.30) 81.86 (0."
            },
            "TABREF3": {
                "html": null,
                "type_str": "table",
                "content": "<table><tr><td>LSTM Hidden State Size</td><td colspan=\"4\">Spanish-100 Spanish-1 Scottish Gaelic-100 Scottish Gaelic-1</td></tr><tr><td>64</td><td>98.40</td><td>91.76</td><td>89.46</td><td>18.55</td></tr><tr><td>128</td><td>98.43</td><td>91.80</td><td>89.84</td><td>21.37</td></tr><tr><td>256</td><td>98.42</td><td>91.44</td><td>89.25</td><td>15.02</td></tr><tr><td colspan=\"5\">Capsule Layer Kernel Size Spanish-100 Spanish-1 Scottish Gaelic-100 Scottish Gaelic-1</td></tr><tr><td>3</td><td>96.00</td><td>83.38</td><td>89.34</td><td>8.42</td></tr><tr><td>5</td><td>96.00</td><td>82.69</td><td>89.21</td><td>9.92</td></tr><tr><td>7</td><td>95.99</td><td>82.59</td><td>88.62</td><td>6.06</td></tr></table>",
                "num": null,
                "text": "Scottish Gaelic Precision, Recall, and F1 scores. The scores are an average of 5 different random seeds and their standard deviation, along with the average training/prediction times of each model."
            },
            "TABREF4": {
                "html": null,
                "type_str": "table",
                "content": "<table/>",
                "num": null,
                "text": "F1 scores for different hyperparameter choices for LSTM hidden state size and CapsNet capsule layer kernel size on the Spanish and Scottish Gaelic 100% and 1% learned embeddings training conditions. The hyperparameter values in italics (hidden state size 128 and kernel size 3) are the values chosen for our bigger comparison."
            }
        }
    }
}